Trita-Ee 2007 043

Maintenance Management in Power Systems
A Ph.D. course by Assistant Professor Lina Bertling and Professor Matti Lehtonen July 2007
Kungliga Tekniska Hgskolan School of Electrical Engineering 100 44 Stockholm TRITA-EE 2007:043
Helsinki Univeristy of Technology Power Systems Laboratory 02015 Helsinki
Course material for Maintenance Management in Power Systems 2007
Foreword
This report presents the documentation for the PhD course on Maintenance management in power systems (F2C5060) given at KTH School of Electrical Engineering during the spring 2007. The course is jointly organized by Professor Matti Lehtonen at Helsinki University of Technology (TKK), and Assistant Professor Lina Bertling at KTH. The course was given for the first time in 2005, and was then held at TKK. News for this year are that; the start up day for the course has been extended to include a more thoroughly theory part, the seminar days included several guest lectures, and that the recommended literature include selected material from the 9th International Conference on Probabilistic Methods Applied to Power Systems (PMAPS) which was held at KTH in 2006. Thanks all attendees for creating a successful course!
Lina Bertling Assistant Professor KTH Course responsible Stockholm, July 15, 2007
KTH/School of Electrical Engineering/RCAM
Contents
General 1 Course program and final seminar schedule 2 List of attendees 3 Project report guidelines and list of proposed topics 4 Recommended literature General Course Material 5 Start up theory day including Reliability, maintenance and RCM by Lina Bertling (theme 1 and 2) Diagnostic methods by Nathaniel Taylor (theme 3) 6 Asset Management and applications by Matti Lehtonen (theme 4) Guest lectures during the seminar days Relibility analysis using Fuzzy interval mathematics by Suprakash Gupta (theme 1) Computational Intelligence for Scheduling Power System Maintenance by Ganesh Kumar Venayagamoorthy (theme 4) Diagnostic Based Maintenace of Mechanical Systems using the PSDmethod by Fredrik Stillesj and Niclas Dahlgren (theme 3) Course Project reports 7 Theme 1 Reliability data assessment and reliability modelling Evaluation and development of test systems for analysis of electrical distribution systems Carl-Johan Wallnerstrm Life cycle scheduling for power transformers considering major overhauls and investments Jussi Palola Literature review in reliability assessments incorporating HVDC links by Johan Setrus Distribution system reliability engineering methods Influence of cable system and overhead line differences by Anna Guldbrand Theme 2 Reliability centred maintenance. Banken and RCM A functional evaluation of maintenance management at Svenska Kraftnt by Per Nrman Evaluating the streamlined RCM project at Ringhals by Pia Gustafsson Reliability and cost centered maintenance methods by Julia Nilsson Implementation of production assurance programmes in a production plant: concept and discussion by Javad Barabady Theme 3 Condition monitoring and diagnostics methods. Oil condition monitoring of power generators by Saurabah Kumar Modeling and experimental verification of CC overhead MV networks for PD detection by Murtaza Hashmi
Assessment of on-condition maintenance by Alireza Ahmadi 10 Theme 4 Maintenance scheduling and optimization. Finite stochastic dynamic programming for maintenance optimization by Francois Besnard e-Maintenance impact on availability performance and life support for geographically distributed complex technical systems by Ramin Karim
1. Course program and final seminar schedule
Course program
2006-09-08
Maintenance Management in Power Systems (5 credits)

KTH School of Electrical Engineering and Helsinki University of Technology (TKK), Power systems and high voltage engineering, offer a jointly organized post-graduate course on Maintenance management in power systems. (At KTH the course has the code 2C5060). The course is given every second year alternating between Finland and Sweden. The first course was held in 2005. Documentation from this course can be found in the report data base at http://www.ee.kth.se. Course subjects: The overall topic of maintenance management has been divided into four themes: 1. Reliability data assessment and reliability modelling. 2. Reliability centred maintenance. 3. Condition monitoring and diagnostics methods. 4. Maintenance scheduling and optimization. Course description: The course begins with a full day of lectures covering basic theory for the course content. Individual projects are defined for each participants, within one of the themes for the course. The participants will work on these projects with supervision from the course responsible. A final project report (10-15 pages) shall be written and presented oral at the concluding course seminar. Each participant shall also review and comment on one other project report. A final course material will be presented after the course with all course reports and introductory lectures. More details with instructions for the project report and oral presentations will be handed out at the course start. Course examination: The examination for the course includes the following activities: 1. The individual task shall be presented in a written report, and orally at the course seminar. Approved task will be rewarded 4 credits. 2. Participation at the course days with one introduction day and the three day course seminar. Written comments shall be communicated during the seminar to one other project report. Approved task will be rewarded 1 credit. Course material: Lecture handouts, summary reports, conference and journal papers. A list of recommended literature that is related to the individual tasks will be handed out on the first day of the course. Course report from 2005 that can be found at Course schedule: Registration: by January 30, 2007 to matti.lehtonen@hut.fi lina.bertling@ee.kth.se. The number of participants is limited to 45. or
Course start with introductory lectures and handing out material; February 5 at KTH and February 6 at TKK. Full day 10-16. Submission of draft version of project reports by May 19.
Submission of project reports and presentation slides by June 4. Course seminar; June 11 to 13, 2007. From 10-16 at KTH, Electrical Engineering, Seminarierummet.
Accomodation: Rooms have been reserved at Hotel Arcadia, Krsbrsvgen 1. The hotel is located about 5 minutes walking distance from KTH Campus. For reservation contact; reservation.arcadia@elite.se. Course fee: There is no fee for registered PhD students, other participants are welcome to fee of 500 euros. Course responsible and lecturers: Professor Matti Lehtonen, TKK and Assistant Professor Lina Bertling, KTH. Short biographies and contact information are presented below.
Matti Lehtonen (1959) was with VTT Energy, Espoo, Finland from 1987 to 2003, and since 1999 has been a professor at the Helsinki University of Technology, where he is now head of Power Systems and High Voltage Engineering. Matti Lehtonen received both his Masters and Licentiate degrees in Electrical Engineering from Helsinki University of Technology, in 1984 and 1989 respectively, and the Doctor of Technology degree from Tampere University of Technology in 1992. The main activities of Dr. Lehtonen include power system planning and asset management, power system protection including earth fault problems, harmonic related issues and applications of information technology in distribution systems. Helsinki University of Technology, Power Systems and High Voltage Engineering, P.O.Box 3000, FIN-02015 HUT, Finland, Tel. +358 9 4515484, Fax +358 9 460224, E-mail: Matti.Lehtonen@hut.fi Lina Bertling was born in Stockholm in 1973. She received the Ph.D., and Tech. Lic. degree in Electric power systems in 2002 respectively 1999, and the M.Sc. degree in Systems engineering in 1997, all from KTH - the Royal Institute of Technology, Stockholm, Sweden. She is engaged at KTH School of Electrical Engineering as Assistant Professor. She is leader of the research group on reliability-centered asset management (RCAM), and project leader for a research program at the Swedish Centre of Excellence in Electric Power Systems (EKC) on Maintenance management. She was the general chair of the 9th International conference on probabilistic methods applied to power systems (PMAPS) in Stockholm, in 2006. Her research interests are in; power system maintenance planning and optimization including reliability-centered maintenance (RCM) methods, reliability modeling and assessment for complex systems, and lifetime- and reliability modeling for electrical components. KTH School of Electrical Engineering, 100 44 Stockholm, SWEDEN, Phone: +46 8 790 6508 E-mail: lina.bertling@ee.kth.se, www.ee.kth.se
Course schedule
2007-06-12
2C5060 Maintenance Management in Power Systems

Monday 11/6 Time 10.0010.15 10.1510.45 10.4511.15 11.1512.30 12.3013.00 13.0013.30 13.3014.00 14.0014.30 14.3015.30 Project title Introduction Evaluating the streamlined RCM project at Ringhals Reliability and cost centered maintenance methods LUNCH Implementation of production assurance programmes in a production plant: concept and discussion Finite stochastic dynamic programming for maintenance optimization Evaluation and development of test systems for analysis of electrical distribution systems Coffee break Guest lecture: Suprakash Gupta Banaras Hindu University Tuesday 12/6 Time 9.009.30 9.3010.00 10.0010.15 10.1510.45 10.4511.15 Project title Life cycle scheduling for power transformers considering major overhauls and investments Oil condition monitoring of power generators Break Literature review in reliability assessments incorporating HVDC links e-Maintenance impact on availability performance and life support for geographically distributed complex technical systems Johan Setrus Saurabah Kumar Presenter Jussi Palola Opponent Murtaza Hashmi Javad Barabady Pia Gustafsson Presenter Lina Bertling Pia Gustafsson Julia Nilsson Julia Nilsson Javad Barabady Opponent
Francois Besnard
Carl-Johan Wallnerstrm Anna Guldbrand
Carl Johan Wallnerstrm
Saurabah Kumar
Johan Setrus
Ramin Karim
Per Nrman
11.1512.30 12.3013.00 13.0013.30
LUNCH Modeling and experimental verification of CC overhead MV networks for PD detection Distribution system reliability engineering methods Influence of cable system and overhead line differences Banken and RCM A functional evaluation of maintenance management at Svenska Kraftnt SOCIAL ACTIVITIES & DINNER Murtaza Hashmi Jussi Palola
Anna Guldbrand
Ramin Karim
13.3014.00 14.0018.00
Per Nrman
Alireza Ahmadi
Wednesday 13/6 Time 9.0010.00 10.0011.00 Project title Heuristic maintenance optimization for power distribution systems Guest Lecture: Computational Intelligence for Scheduling Power System Maintenance Coffee break Guest Lecture: A revolutionary method for monitoring electric and mechanic status of equipment used in industrial processes. Assessment of on-condition maintenance Concluding remarks Fredrik Stillesj Niclas Dahlgren PSD Insight AB Alireza Ahmadi Lina Bertling, Matti Lehtonen Francois Besnard Presenter Patrik Hilber KTH G. Kumar Venayagamoorthy University of Missouri-Rolla Opponent
11.0011.15 11.1512.15
12.1512.45 12.4513.00
2. List of attendees
Name Students Javad Barabady Alireza Ahmadi Ramin Karim Anna Guldbrand Johan Setrus Anna Franzn Julia Nilsson Carl-Johan Walnerstrm Andrea Lang Pia Gustafsson Ambika Patra Saurabh Kumar Olov Candell Per Nrman Francois Besnard Jussi Palola Murtaza Hashmi Guest Lecturers Suprakash Gupta Fredrik Stillesj, Niclas Dahlgren Ganesh Kumar Venayagamoorthy Patrik Hilber Nathaniel Taylor Course Assistant Tommie Lindquist Course Leaders Lina Bertling Matti Lehtonen Email Javad.Barabady@ltu.se alireza.ahmadi@ltu.se Ramin.Karim@ltu.se anna.guldbrand@iea.lth.se johan.setreus@ee.kth.se anna.franzen@ee.kth.se julia.nilsson@ee.kth.se cjw@kth.se andrea.lang@ee.kth.se PiaG@ics.kth.se ambika.patra@ltu.se saurabh.kumar@ltu.se Olov.Candell@sat.saabgroup.com PerN@ics.kth.se fbesnard@kth.se jussi.palola@helsinginenergia.fi hashmi@cc.hut.fi Affiliation LTU LTU LTU LTH KTH KTH KTH KTH KTH KTH LTU LTU LTU KTH KTH Helsingin Energia HUT Banaras Hindu University niclas.dahlgren@psdinsight.se ganeshv@umr.edu patrik.hilber@ee.kth.se nathaniel.taylor@ee.kth.se tommie.lindquist@ee.kth.se lina.bertling@ee.kth.se matti.lehtonen@hut.fi PSD Insight AB University of MissouriRolla KTH KTH KTH KTH HUT
3 List of proposed topics and guidelines for the report preparation
Guidelines for the report preparation

The submitted project reports shall be presented following the general guidelines summarized below. Contact information Author, affiliation, email Page setup Use one column format with 2.5 cm for all margins (top, bottom, left, right) Page numbering Page numbers located at center in the bottom of the page Font size Use the font Times New Roman or Helvetica Font size 12 for the body text Font size 14 with capital letters for the title (centered) Font size 12 with italics for Authors, Affiliations and contact data Font size 12 with italics for Figure and Table headings Font size 12 or 10 for table contents Font size 12 bold, capital letters for Headings (left alignment) Font size 12 bold, sentence case for subheadings (left alignment) Line spacing Use single line spacing. Leave two free lines before the headings and one free line after the headings. Paper length The recommended paper length is 10 pages. This is not a strict limitation or requirement, however. Submitting the manuscript The draft report and accompanied presentation slides must be submitted not later than May 19, 2007 in electronic form (*.doc, *.rtf or *.pdf) to lina.bertling@ee.kth.se. The final report shall be submitted by July 9, 2007.
List of proposed topics

Theme 1: Reliability data assessmet and reliability modelling 1. Reliability assessment of surge arresters 2. Failure causes in medium voltage networks 3. Reliability analysis and reliability engineering methods for distribution systems 4. Reliability modeling of HV switchgear 5. Monte Carlo simulation methods for reliability modeling and assessment 6. Markov-models in fault frequency modeling 7. Risk analysis in power systems 8. Probabilistic Safety Analysis (PSA) in power systems 9. Reliability modeling and design for complex power systems 10. Life time modeling and management of transformers Theme 2: Reliability centred maintenance 11. Risks related to the introduction of RCM 12. Reliability- and cost centered maintenance methods 13. Different maintenance strategies and their impact on power system reliability 14. Prioritization of maintenance methods (RCM,CBM,TBM,CM) 15. RCM applications for overhead lines 16. RCM applications in underground power systems 17. RCM applications for power transformers 18. RCM applications for switchgear 19. RCM applications for secondary substations 20. RCM applications for generator systems Theme 3: Condition monitoring and diagnostics methods 21. Condition monitoring of wooden poles 22. Maintenance and condition monitoring on large power transformers 23. Aging phenomena of paper-oil insulation in power transformers 24. On-line monitoring applications for power transformers 25. Tests and diagnostics of insulating oil 26. Diagnostics of dissolved gas in insulating oil of transformers 27. Dielectric diagnostics measurements of transformers and their interpretation 28. PD-measurements for transformers and their interpretation 29. Condition monitoring methods of power cables and their accessories 30. Aging phehomena of cable insulation materials 31. On-line monitoring applications for power cables
32. Condition monitoring of circuit breaker and switchgear 33. Condition monitoring of surge arresters 34.On-line monitoring applications for secondary substations 35.Maintenance and monitoring methods for induction motors 36.Maintenance and monitoring of large generators 37.Diagnostics measurements of large generators 38.On-line monitoring of generators Theme 4: Maintenance scheduling and optimisation 39. Maintenance scheduling for off-shore wind power plants 40. Evaluation of different tools for reliability assessment e.g. RADPOW, NEPLAN, NetBAS etc. 41. Evaluation of using the tool VeFoNet for purpose of maintenance planning, RCM. 42. Evalution of using the tool NEPLAN for the purpose of maintenance planning 43. Compliation of existing functions for RCM in commercial tools like NetBas, NEPLAN. Meldis, Maximo etc. 44. Maintenance optimization techniques for distribution systems 45. Maintenance scheduling and optimization in power plants
In addition, the students are free to propose a subject of their own, as long as the subject is in line to the themes above. It is recommended that, if only possible, the students select a subject which is close to their PhD-project.
4. Recommended literature
Main resources The main source for literature is the IEEE Xplore (with approximately 1.5 million research papers) http://ieeexplore.ieee.org A CD with all articles presented at the 9th International conference on Probabilistic Methods Applied to Power Systems (PMAPS) will be distributed to all course participants. The conference was held at KTH 18-22 June 2006.
Links with reference literature in power industry EPRI www.epri.com Cigr www.cigre.be SINTEF www.sintef.no Elforsk www.elforsk.se SwedEnergy www.svenskenergi.se UiS www.uis.no/research/maintenance_and_asset_management
Example on books on reliability assessment, maintenance and RCM and power system applications Hoyland A., Rausand M., System reliability theory - models and statistical methods, Wiley Series, 2004 Patrick D. T. OConnor: Practical Reliability Engineering, Wiley&Sons Ltd, 2002 Andrew K.S. Jardine and Albert H.C. Tsang: Maintenance, Replacement and Reliability, Taylor and Francis, 2005. Roy Billinton and Ron Allan, Reliability Evaluation of Power Systems Lina Bertling, RCM for Electric Power Distribution Systems, KTH, Doctoral thesis 2002 (includes a reference list with fundamental books on distribution systems and RCM)
Links with reference literature at KTH Theses at KTH (e.g. Bertling 2002 on RCM) http://media.lib.kth.se/kthdiss.asp Research at KTH see annual report http://www.ets.kth.se/annual_report.htm Research at KTH/RCAM see list of publications at http://www.ee.kth.se/rcam Course material and reference material: http://www.ets.kth.se/eek/lina/2C4030.html
Links with reference to different standards IEEE http://standards.ieee.org Military standards e.g. MIL STD ISO http://www.iso.org IEC http://www.iec.ch SIS http://www.sis.se UTEK http://www.utek.se
Start up theory day including

Reliability, maintenance and RCM by Lina Bertling (theme 1 and 2) Diagnostic methods by Nathaniel Taylor (theme 3) Asset Management and applications by Matti Lehtonen (theme 4)
Probabilistic approach for Power System assessment - introductory lecture for PhD course on
maintenance management
Lina Bertling Assistant Professor KTH Electrical Engineering
Course material for PhD course on Maintenance Management in Power Systems, KTH and TKK, 2007.
Maintenance Management in Electric Power Systems Lina Bertling 2007-02-05 1
Introduction Problem
Problem The Electric Power System is not 100% reliable Risk and cost for failure
Modell
Stocastic modells of the Electric Power System
Probability theory
Evaluation
Analytical or simulation method
Input data Tools
Result
Quantitative measures for reliability
Maintenance Management in Electric Power Systems
Lina Bertling
2007-02-05
Introduction Power system

The Electric power system is an extremely complex system due to e.g: large with complex components geographical spread national and international interconnections electrical energy can not be efficient stored unexpected events in one part of the system can lead to large consequences in other parts of the system depends on the economic, social and political climate
Lina Bertling 2007-02-05 3
Introduction Historical changes

Reliability quite new as a concept for technical systems and started with the first world war: number of flight accidents per hour of flight time After second world war focus on planning and construction for generation with a quite stable economics Economic crisis with 1970s oil crisis, the knowledge that nuclear power is not the future energy solution 1980s, and environmental issues De-regulation introduced from 1990s that created new roles and requirements based on economical rather than technical issues
Introduction Main questions

The main function is: to supply customers with electrical energy with high level of reliability at a reasonable cost The level of reliability can be affected by: resources in form of redundant components (to reduce the probability of failure) and spares (to reduce the outage time ) Main questions are therefore to decide: number of spares and level of redundancy, at what cost?
Lina Bertling
2007-02-05
Introduction Main difficulties

The probability of non supply can be reduced by investments in planning, operation and maintenance of the electric power system These investments raise the price of the electrical energy and has to be reflected in the cost of utilities, customer tariffs etc. The conflict between requirements on reliability and economics can lead to difficult decisions
Lina Bertling
2007-02-05
Introduction Criteria
Examples on criteria used to solve this conflict between reliability and economics: Generation capacity = expected maximum load + fixed percentage of expected load Operation capacity = expected load + reserve (=one/more of the largest units) Network capacity n-k criteria where the system should be in function although k units are in failure
Lina Bertling
2007-02-05
Introduction Weaknesses
Weaknesses in the deterministic criteria: a fixed percentage of generation capacity do not guarantee enough reserve capacity all planning of the electric power system is based on prognosis for expected load which always included uncertainty failure rate e.g. for lines depends on several issues like: length, design, geographical, environment etc, and consequently the (n1) criteria can not guarantee a certain level of disturbances
Lina Bertling
2007-02-05
Introduction Basic concepts

Dependability The collective term used to describe the availability performance and its influencing factors: reliability performance, maintainability performance, and maintenance support performance (IEC 300-1) The termination of an items ability to perform a required function (BS 4778)
Failure
Lina Bertling
2007-02-05
Introduction Basic concepts

Example on dependability (reliability) measures* 1. 2. 3. 4. Mean time to failure (MTTF) Failure frequency Probability that a component do not fail within a time (0,t] (survival function) Probability that a component is in function at time t (availability at time t)
* 3=4 for non repairable components
Lina Bertling
2007-02-05
10
Introduction Application areas

Reliability analysis provide decision support: Risk and security analysis Environmental protection Quality Optimization of operation and maintenance Design Validation/Verification
Lina Bertling
2007-02-05
11
Introduction Failures
Example for electrical distribution systems (Bertling 2002)
Failure
DAMAGING FAULT Two models of failure Permanent forced outages
NON-DAMAGING FAULT Two models of restoration
Transient forced outages Automatic switching
Temporary forced outages Manual switching or fuse replacement
Passive event
Active event
Lina Bertling
2007-02-05
12
Introduction Failures
Definitions for failures: Active failure is a failure for a unit that cause the primary protection system to operate (breakers or disconnectors ) closest to the unit Passive failure is a failure that is not active
Lina Bertling
2007-02-05
13
Introduction Models
Stochastic models; Probability density function to describe the failure process The density function is its derivate and presents how the probability is spread Expected value for example mean time to failure Fundamental measures; Probability- (survival-) function R(t) Hazard function z(t) Mean time to failure (MTTF)
Lina Bertling
2007-02-05
14
Introduction Models
Time to failure for a unit imply the time from the unit is put into operation until it fails for the first time. Let t=0 be the start point. A failure occurrence is a random event and the time to failure can therefore be described by a stochastic variable T. The condition for the unit at time t can be presented by the stochastic variable X(t) :
1 if the unit is in function at time t X (t ) = 0 if the unit is in failure at time t
Lina Bertling
2007-02-05
15
Introduction Models
The continuous stochastic variable T with density function f(t) and probability density function F(t) defines:
F (t ) = P(T t ) = f (u )du
0
for t > 0
F(t) is the probability that the unit fails within the time period (0,t]
Lina Bertling
2007-02-05
16
Introduction Models
Example on density function and probability density function
Density function Distribution fucntion
1/m=
FX(x)
fX(x)
x X Exp(m)
Lina Bertling
2007-02-05
17
Introduction Models
A system of units are put in operation at time t = 0 The probability that the system is in function, R, at a later time, t = t0, is called the probability function Assume that the system consist of units: E1,E2, ...,En with known connections and with probability function p1 , p2 , ... ,pn. The probability that E1 fails before time t1 is 1- p1 = q1 o.s.v. Assume that the units fails independently There are two fundamental situations
Lina Bertling
2007-02-05
18
Introduction Models
1. Serial system A system of serial couples units is in function at time t=t0 if and only if all units are in function. According with the multiplication rule for independent events is:
R = p1 p2 ... pn
Lina Bertling
2007-02-05
19
Introduction Models
2. Parallel coupling A system consist of units in parallel and is in function at time point if at least one unit is in function The opposite event is that all unit fails
1 R = q1 q 2 ...q n
R = 1 (1 p1 )(1 p2 )... (1 pn )
R = 1 (1 p1 )(1 p2 )... (1 pn )
Lina Bertling
2007-02-05
20
10
Introduction Models
How is the survival function changing in time?
1 R (t)
X
Lina Bertling
2007-02-05
21
Introduction Models
The survival function R is defined by:
R( t ) = P( T > t ) = 1 P( T t ) = 1 F ( t ) R ( t ) = f ( t ) F ( 0) = 0 F ( ) = 1
Lina Bertling
2007-02-05
22
11
Introduction Models
The expected time to failure can be evaluated as follows:
MTTF = R(t )dt

0
Lina Bertling
2007-02-05
23
Introduction Models
The hazard function is a measure for when a unit fails within a certain time period known that it was is function a certain time
T life time
t+h
Lina Bertling
2007-02-05
24
12
Introduction Models
Relationship between hazard function and the survival function
h0 1 1 P( t < T t + h ) ( F ( t + h ) F (t ) ) 1 P( T t + h | T > t ) = h = h = h P(T > t ) R(t ) F ( t ) f ( t ) R (t ) = = = R(t ) R(t ) R(t )
Lina Bertling
2007-02-05
25
Introduction Models
The survival function and the probability density function can therefore be expressed by the hazard function as:
z (u )du
0 t
R (t ) = e
f (t ) = z (t )e
z (u )du
0
for t > 0
Lina Bertling
2007-02-05
26
13
Introduction Models
Example: Exponential distribution that give a constant hazard rate
T Exp( 1 )
f (t ) = e t
R(t ) = P(T t ) = 1 P (T t ) = 1 F (t )R(t ) = f (t )

R(t ) = e t (t ) = f (t ) e t = t = R( t ) e
Lina Bertling
2007-02-05
27
Introduction Models example

A typical model the bath tube
Hazard function
Operation time
Time
Lina Bertling
2007-02-05
28
14
Introduction Models example
10
A typical model the weibull distribution

Failure rate function X Weibull (m,)
Weibull(1,0.5) Weibull(1,1) Weibull(1,3)
6
(t)
0.5
1.5 t
2.5
Lina Bertling
2007-02-05
29
Introduction Techniques
A technical system can often be described by a set of components or functional block that fulfills certain required functions Different techniques are used to express the structural relation between the system and its components regarding system failures. Example of those are: Failure mode and effect analysis (FMEA) Failure trees Block diagram
Lina Bertling
2007-02-05
30
15
Introduction Component
The component value for the system reliability depends on: where in the system structure the component is placed what reliability the component has There are several methods to evaluate and define a components value in a system e.g. how critical it is for the system availability
Lina Bertling
2007-02-05
31
Introduction Markov
In Markov modeling is the system defined by its states and transitions between different states more than two states possible constant transition rates No memory
Lina Bertling
2007-02-05
32
16
Reliability Centred Maintenance Management for Power Systems - introductory lecture for PhD course on
maintenance management
Lina Bertling Assistant Professor KTH Electrical Engineering
Course material for PhD course on Maintenance Management in Power Systems, KTH and TKK, 2007.
Agenda
Introduction: concepts and scope Method: RCM och RCAM Examples: RCAM for electrical distribution system Examples from ongoing studies Summary: result and challenges for future Other: PMAPS, contact information
Lina Bertling
2007-02-05
Introduction AM
Asset management (AM) is a concept used today for planning and operation of the electrical power system The aim of AM is to handle physical assets in an optimal way in order to fulfil an organisations goal whilst considering risk where: the risk could be defined by the probability of failure occurrence and it consequence the goal could be maximum asset value, maximum benefit or minimal life cycle cost the constraints could be set on revenues or availability for power supply
Introduction MM
There are different possible actions to handle these assets: e.g. acquire, replace or redesign Maintenance management (MM) is here defined as a strategy to handle decisions for these assets and to make right decisions on: what assets to apply actions for what actions to apply how to apply the actions when to apply the actions
Lina Bertling
2007-02-05
Introduction Scope
To make the right decisions there is a need of; condition data failure statistics reliability modelling techniques reliability assessment tools maintenance planning tools systematic techniques for maintenance planning e.g. reliability-centred maintenance (RCM) method This different type of needs are covered within the course subject for this course.
Lina Bertling
2007-02-05
Introduction Maintenance
Equipment maintenance is one of the tools for MM either by Preventive (PM) or Corrective Maintenance (CM). Reliability-Centered Maintenance provides a tool for MM by balancing between CM and PM to reach costeffective maintenance plans
Maintenance Underhll
Preventive Maintenance Frebyggande Underhll
Predetermined Maintenance Frutbestmt Underhll
2007-02-05 6
Corrective Maintenance Avhjlpande Underhll
Condition based Maintenance Tillstndsbaserat Underhll
Lina Bertling
RCM History
Originated in the civil aircraft industry in 1960s with Boeing 747 series US Department of Defence defined RCM with first full description in 1978 (Nowlan) Introduced for Nuclear power industry 1980s by EPRI Introduced for Hydro Power Plants in 1990s e.g. Norway and Sweden (Vattenfall 19972005) Today attempts implementing RCM for planning of transmission- and distribution systems, and wind power parks
Lina Bertling
2007-02-05
RCM Concepts
Reliability-Centered Maintenance is a systematic risk based qualitative method that aims to optimize maintenance achievements The following features define and characterize RCM: 1. preservation of system function, 2. identification of failure modes, 3. prioritizing of function needs, and 4. selection of applicable and effective maintenance tasks. RCM does not add anything new in a technical sense. It is a new working process.
Lina Bertling
2007-02-05
RCM Versions
There are several descriptions and appreciations of the process defining an RCM plan and performing RCM. The fundamental principles are however the same (or should be..). Three fundamental descriptions of RCM have been selected for this presentation: 1. Nowlan step for an initial RCM-plan 2. Smith 1993 seven steps 3. Moubray seven questions
Lina Bertling
2007-02-05
RCM According to Nowlan

Nowlan defines the process to formulates an initial RCM plan, when the data is limited, as follows: 1. partitioning the equipment into object categories in order to identify those items that require intensive study, 2. identify those items that require intensive study, 3. identifying significant items that are those that have essential safety or economic consequences and hidden functions that require scheduled maintenance,
Lina Bertling
2007-02-05
10
RCM According to Nowlan cont.

4. evaluate the maintenance requirements for each significant item and hidden function in terms of the failure consequences and select only those tasks that will satisfy these requirements, 5. identify items for which no applicable or effective task can be found, then either recommend design changes if safety is involved, or assign no scheduled maintenance tasks to these items until further information becomes available,
Lina Bertling
2007-02-05
11
RCM According to Nowlan cont.

6. select conservative initial intervals for each of the included tasks and grouping the tasks in maintenance packages for application, 7. establish an age-exploration program to provide the factual information necessary to revise initial decisions
Lina Bertling
2007-02-05
12
RCM According to Smith

1. System selection and information collection. 2. System boundary definition. 3. System description and functional block diagrams. 4. System functions and functional failures. 5. Failure mode and effects analysis (FMEA). 6. Logic decision tree analysis (LTA). 7. Selection of maintenance tasks.
Lina Bertling
2007-02-05
13
RCM According to Moubray

To analyze the maintenance aspects of a system and its components, the first step is to identify the system items, and which of these that ought to be analyzed. Thereafter the RCM process can be formulated into seven questions
Lina Bertling
2007-02-05
14
RCM According to Moubray

1. What are the functions and performances required? 2. In what ways can each function fail? 3. What causes each functional failure? 4. What are the effects of each failure? 5. What are the consequences of each failure? 6. How can each failure be prevented? 7. How does one proceed if no preventive activity is possible?
RCM Where are we today?

Today RCM is used or considered by an increasing number of electrical power utilities and industries. The introduction of RCM is however, a complicated, resource requiring and a longterm project. For future work it is therefore necessary to ask: What is the benefit of introducing RCM, what does the experience tell, how could the method be further developed, and what role has RCM as a tool in the asset management process?
RCM Developments into RCAM

RCM does not solve the fundamental problem of how the system reliability is impacted by component maintenance With the aim to solve this problem a quantitative method for RCM has been developed at KTH i.e. Reliability-Centered Asset Management RCAM includes a relationship between reliability performance indices and the effect of maintenance measures, outgoing from causes of failures and failure mechanism for components in the Electric Power Systems
(t,PM)
RCAM Overall logic

The figure shows overall logic for the steps in RCAM Paper* provides a presentation of underlying theory and relationships
*A reliability-centered maintenance method for assessing the impact of maintenance in power distribution systems, Bertling, Allan, Eriksson, IEEE Trans. on Power Systems, Vol. 1, 2005.
Stage 1 System reliability analysis 1. Define reliability model and required input data
*
For each: critical component i,
Stage 1: System
2. Identify critical components by reliability analysis Stage 2 Component reliability modelling
3. Identify failure causes by failure mode analysis
4. Define a failure rate model
reliability assessment identify critical components
5. Model effect of PM on reliability
PM method j, and
failure cause k.
Are there more causes of failures ? No
Yes
Yes
Stage 2: component
Are there alternative PM methods ? No 6. Deduce PM plans and evaluate resulting model
reliability modeling and the effect of maintenance (t,PM);
Are there more critical components ? No Stage 3 System reliability cost/benefit analysis 7. Define strategy for PM when, what, how
*
Yes
8. Estimate composite failure rate
Stage 3: System
9. Compare reliability for PM methods and strategies
10. Identify cost-effective PM strategy
reliability assessment and cost analysis

2007-02-05 18
RCAM plan
Lina Bertling
RCAM Example A Summary

Stage 1: Stage 2: Stage 3:
System reliability assessment and identification of critical components

0.35 0.3 0.5
Component reliability modeling (t,PM) outgoing from causes of failures

Water tree length [%]
System analysis implementing maintenance strategies and performing cost analysis

x 10 15
6
Discount Rate zero S

2
0.25 0.2 0.15 0.1 0.05 0 1 2 3 4 5 Case Component Type
0.4 0.3 0.2 0.1 0
Present Value [SEK]
Failure rate [int/yr]
Unavailability [h/yr]
10
(t) [f/yr]
S 5 S1
2 3 4 5 Case Component Type
TCCM
6
TCPM
U /U bd 0
x 10
si
TCPM
rp
2.5
Discount Rate 7%
Energy not supplied [kWh/yr]
400 300 200 100 0

0 5 10
Present Value [SEK]
Outage time [h/int]
2 1.5 1 0.5 0
15
10
5 S 0 TCCM TCPM
si
S2 S2
1
PM
15
20
25
30
S1 TCPM
rp
Time [yr]
Lina Bertling
2007-02-05
19
RCAM Example A stage 1

Sp
c1 220 kV c2 c3 c4 c8 c9 c10 110 kV c5 c6 c14 c15 c49 c50 c51 c52 c56 c58 c30
SW SJ
Birka an urban distribution system in the Stockholm area 220kV11kV Average load point at 11kV
c11 c12 c13 c19 c20 c21 c25 c22 c29 c28 c26 c23 c24 c36 c37 c40 c38 c41 c43 c44 c45 c47 33 kV
c7
c53 c54 c55 c57
c16 c17 c18 c27
11 kV c39 c42 c46 c48
c31
HD
c32 c33 c34 c35
LH11
Lina Bertling
2007-02-05
20
10

Survey of disturbance statistics for the Birka system (1982-1990) and (3980 reports) identified that the: 11kV level was the critical voltage level with aspect to number of failures (35%) and customer outage times 11kV under ground cable contributed to 62% of the interruptions, and 78% of affected customers
A failure rate model for the 11 kV cable could support in finding efficient maintenance strategies for the Birka system- that is a model on the form: (t,PM)
Lina Bertling
2007-02-05
21

Sp
c1 c2 c3 c4 c8 c9 c10 110 kV c5 c6 c14 c15 c49 c50 c51 c52 c56 c58 c30
SW SJ
0.35 0.3 0.5
220 kV
Identification of critical component type

Failure rate [int/yr] Unavailability [h/yr]
1 2 3 4 5 Case Component Type 0.25 0.2 0.15 0.1 0.05 0 0.4 0.3 0.2 0.1 0
c11 c12 c13 c19 c20 c21 c25 c22 c29 c28 c26 c23 c24 c36 c37 c40 c38 c41 c43 c44 c45 c47 33 kV
c7
c53 c54 c55 c57
c16 c17 c18 c27
2.5
Energy not supplied [kWh/yr]

1 2 3 4 5 Case Component Type
400 300 200 100 0
Outage time [h/int]
2 1.5 1 0.5 0
11 kV c39 c42 c46 c48
c31
HD
c32 c33 c34 c35
Reliability performance for a load point
LH11
Lina Bertling
2007-02-05
22
11

Identifying failure causes for the critical component in the system
Cable system
Function: transmit energy
Failure mode of function
Short circuit
Failure event insulation failure Failure cause
Open circuit
conductor failure
Water treeeing
Damage 16 % Digging 9% Sabotage 2% Fabric/ material 14 %
Material/ Method 59 % Lack of Maint. 5% Wrong method/ instruction 15 %
Personnel 12 % Wrong manoevre 5% Wrong montage/ laying 5%
CU & AL contact
Water treeing
Corrosion
Lack of oil
Bending damage
Scratch in insualtion
Lina Bertling
2007-02-05
24

Water trees water penetration through the insulation, early produced (mid-1970s) XLPE insulated cables. Process to relate underlying failure cause to reliability for an XLPE cable effected by water tree
Water-tree growth
Decreased breakdown voltage

Lina Bertling
Increased failure rate

2007-02-05 25

Data used for the failure rate modeling: 1. Disturbance statistics for water treed cables to predict the failure rate (Survey by Swedish utilities) 2. Measurements and modeling of the condition for the cable insulation (water tree length and break down strength) (KTH/Diagnostics group and SINTEF) 3. Results from preventive maintenance actions for cables by rehabilitation method* (SINTEF)
*Injection of silicon-based liquid between the individual wires of the conductor which stops the growth, of the current, water-trees
13

Resulting relationships for an XLPE cable effected by water treeing Time for preventive maintenance
Water tree length [%]
(t) [f/yr]
Ubd/U0
10
tPM
15
20
25
30
Time [yr]
Lina Bertling
2007-02-05
27

Three different maintenance actions are considered for the underground cable component; 1. PM by rehabilitation 2. PM by replacement 3. only corrective maintenance The effect of applying PM at different times are evaluated, and two strategies are selected for demonstration purposes: - S1: PM is applied at years: 9, 11 and 12 and each time are 10% of the 11kV cables effected - S2: PM is applied at years: 9, 11 and 12 and each time are 30% of the 11kV cables effected
14

Resulting composite failure rate functions
Composite Failure Rate [f/yr]
0.25
(t,CM)
(t,PM S ) i 1
Corrective maintenance Preventive maintenance with rehabilitation and strategy S1 or S2
0.2
(t,PMiS2)
0.15
0.1 0 5 10 15 Time [yr] 20 25 30
Lina Bertling
2007-02-05
29

Assume that the component failure rate function for component i can be obtained as a sum of contributions from the different causes of failures of type k.
(t ) = i k (t )
i
Deduce a model for functional relationship between reliability and PM activities j as a function of time.
k =1
i (t , PM ) = ijk (t , PM )
Deduce different plans for applying PM, and evaluate the resulting effect on the component failure rate
j =1 k =1
Lina Bertling
2007-02-05
30
15

The proposed economic analysis relates to the modelling and the evaluation of the following: The cost of restoring a failure CCM f The cost of PM CCM PM The cost of interruption* CCM int Annualized costs and the present values are evaluated The optimal maintenance method, for a utility, is the solution that minimizes the total cost, which is the sum of the above costs.
* includes the customer costs.
Lina Bertling
2007-02-05
31

The cost functions are calculated as follows;
CCM f (t ) = i (t ) c if (1 + d 1 )
i =1 n n t
CPM f (t , S ) = i (t , S ) c if (1 + d 1 ) i =1 t0 t tPM 1 0 n Ci tPM 1 t tPM 2 PMj (T tPM1 + 1) i =1 j S1 CPMPM (t, S ) = # # n i tPMv t T (T tPM1 + 1) CPMj v i =1 j Sv
t
Lpi CCMint (t) = ELpi (t, CM) cint (1 + d1 ) i =1 Lpi CPMint (t, S) = ELpi (t, S ) cint (1 + d1 ) i =1 nlp
nlp
Lina Bertling
2007-02-05
32
16

The total annualized costs in [cost/yr] are calculated:
TCCM (t ) = CCM f (t ) + CCM int (t )
TCPM (t , S ) = CCM f (t , S ) + CCM int (t , S ) + CPM PM (t , S )

The present values are calculated
TCCMPV (t ) =
t =t0
T CCM (t ) PV (t , d )
f 2 t =t0
TCPMPV (t , S ) =
T CPM (t , S ) PV (t , d )
f 2
The cost-effective solution is the maintenance strategy that provides the lowest total cost when comparing the total costs for PM with different sets of S, and with no PM, that is CM.
Lina Bertling
2007-02-05
33

Results from present value analyses of the total costs Economic factors e.g. discount rate make a great impact on the final result
x 10 15
6
Discount Rate zero S2
Present Value [SEK]
10
S 5 S
1
TCCM
6
TCPM
si
TCPMrp
x 10 15
Present Value [SEK]
Discount Rate 7%
10
5 S 0 TCCM TCPM
si
S S2 S1 TCPMrp
Lina Bertling
2007-02-05
34
17
RCAM
Stage 1: Method to identify critical components a step towards optimization
Examples from ongoing work

Stage 2: Reliability modeling for breakers; Statistics (Svk 1999-2003) Test data (ABB) Stage 3: Life Cycle Cost evaluation of using condition monitoring systems for wind power parks
Sp
c1 c2 c3 c4 c8 c9 c10 110 kV c5 c6 c14 c15 c49 c50 c51 c52 c56 c58 c30 c31 c53 c54 c55 c57 c16 c17 c18 c27 c7 c19 c20 c21 c25 c22 c29 c28 c26 c11 c12 c13 c23 c24 c36 c40 c37 c38 c41 c43 c44 c45 c47 33 kV 220 kV
11 kV c42 c39 c46 c48
SJ
HD
c32 c33 c34 c35
S J
LH11
Examples from ongoing PhD projects within RCAM by; Hilber, Lindquist, Nilsson/Lang
RCAM
Objective: min CIC + CCM + CPM CIC [/yr] customer interruption cost cost of corrective maintenance CCM [/yr] cost of preventive maintenance CPM [/yr] Three alternatives for each component:
1. 2. 3. Keep current maintenance level () Improve the maintenance (/2) Decrease the maintenance (2)
An approximate evaluation of the benefit of each individual maintenance alternative is made
Lina Bertling
2007-02-05
36
18
RCAM
c1 c2 c3 c4

Sp
220 kV c8 c9 c10 110 kV
c5 c6 c14 c15 c49 c50 c51 c52 c56 c58 c30 c53 c54 c55 c57 c16 c17 c18 c27 c7 c19 c20 c21
c11 c12 c13 c23 c24 c36 c37 c40 c38 c41 c43 c44 c45 c47 33 kV
c25 c22 c29 c28 c26
Cus tomer node LH11 HD SW
SEK/int. 162360 87400 10400
SEK/kW h 27.6 14.8 64
11 kV c39 c42 c46 c48
c31
Increase maintenance
HD
SJ SJ
c32 c33 c34 c35
Keep maintenance lev. Decrease maintenance

Figure from: RCM for Electric Power Distribution Systems. Figure 10.21. Lina Bertling, KTH
LH11
Lina Bertling
2007-02-05
37
Conclusions and challenges

RCM methods and its extensions like RCAM provides efficient tools for maintenance management It is beneficial to apply maintenance strategies based on the results of quantitative systematic techniques such as RCAM Challenges to overcome implementing these type of methods; Relating maintenance effort and reliability benefit is a very complicated problem Lack of input data in detail and range General resistance using new methods General difficulties due to the new roles of contractor and subcontractor, and organizations with lack of resources and knowledge
Lina Bertling
2007-02-05
38
19
PMAPS2006 at KTH
Probabilistic methods applied to power systems 300 participants, 35 countries, 200 selected papers, 6 workshops, 3 tutorials, 2 plenary sessions, etc. More information at; www.pmaps2006.org
Lina Bertling
2007-02-05
39
Thank you for the attention!

Welcome to contact me for further information and questions!
Lina Bertling, Assistant Professor KTH School of Electrical Engineering Teknikringen 33 100 44 Stockholm, SWEDEN E-mail: lina.bertling@ee.kth.se Phone: +46 8 790 6508 Web: www.ee.kth.se/users/linab, www.ee.kth.se/rcam and www.comp.ee.kth.se
Lina Bertling
2007-02-05
40
20
Diagnostic Methods for Electrical Insulation

With a focus on Stator Insulation and Low Frequency methods
Presented as a part of the Asset Management course.
Nathaniel Taylor Division of Electromagnetic Engineering, KTH 5th (KTH) and 6th (TKK) of February, 2007
x 10 8 10
11 9
6 4 charge, C
C,C, F
2 0 2 4 6 8
10
12
current, A
1.0 kV pk 10.0 kV pk 20.0 kV pk 10

13
0 PD x 20 2 0 1 time, s 2
current, A
0.1 kV pk
0.92 0.86 0.81 0.75 0.69 0.63 0.58 0.52 0.46 0.4 0.35 0.29 0.23 0.17 0.12 0.06 0 90 180 270 phase, degrees 360
2 current, A 0
x 10
50.00Hz 2 current, A PD x 13 0
x 10
5.00Hz
PD x 7 2 0 2 0 PD x 51 2 0 10 time, s 20 x 10
7
2 0 2 x 10
6
0.01 time, s 0.50Hz
0.02
0.1 time, s 0.05Hz
0.2
10
10
10
10 10 frequency, Hz
10
10
The Plan Stator insulation systems

Special case of highly stressed composite insulation.
Dielectric spectroscopy
Continuous-current diagnostic methods
Partial discharge measurement and analysis

Pulsed-current diagnostic methods
Vagaries of diagnostic methods.

Some closing comments
Machine Construction
Stator insulation
slot end
Rotor insulation Mechanical (bearings)

3
Stator Insulation
Un Un Un 1 kV = form-wound 5 kV = corona prevention (grading) 30 kV in spite of ratings 11500 MW!
1111111111111 0000000000000 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111
stator iron conductors turn insulation main insulation semicon layer slot wedge
Strand
Turn
Main
picture: Siemens & vonRoll Isola

4
Relevant Constructional Variations

Insulation binding material (between the mica pieces): bitumen high loss, migrates polyester resin epoxy resin Hydro- or turbo-generator: speed aects diameter/length, which aects ratio of endlength/slot-length and propagation of PD signals. Cooling and coolant: indirect: heat comes out of windings through the insulation direct: a coolant (typ. water) is used within the windings air/hydrogen: air has lower EBD and heat capacity, allows O3 generation by PD, and assists oxidation of organic insulation Corona-prevention (semiconductor layers) or not?
Stator Insulation Defects
Manifestation: Delamination, cavities Electrical Treeing Worn slot semiconductor Stress: Thermal, Electrical Ambient, Mechanical Cause: primary, secondary?
6
Worn end semiconductor Contamination on end-windings
Diagnostic Methods in Common Use

Many methods dont depend on electrical connection to the windings, e.g. detect PD by measuring RF emissions, chemicals, sound, light, or use mechanical measurements and visual inspection, PD measurement by couplers. Others more relevant to this work: o-line, with galvanic connection. Note that o-line usually implies signicant dierences from normal stresses. On-lineO-line: expenses, realistic versus controlled conditions. Measurement (Physical State) P(failure) Complex system. Dierent phenomena can look similar in a particular measurement. Cannot expect any method to give no false positives or negatives.
7
Electrical Diagnostics in Common Use for Stator Insulation

Insulation Resistance (IR): DC with quite HV, i.e. a form of time-domain spectroscopy with stepped voltage). This is also known as megging, and the derived quantity Polarisation Index (PI) is popular. Capacitance and Loss, possibly including Tip-Up: generally power frequency (50 Hz), with several voltage steps typically up to the rated line voltage per phase, i.e. overvoltage factor of 3 Partial Discharge (PD) measurement: again, generally power frequency, often as phase resolved pattern.
Dielectric Spectroscopy (DS)

Dielectric Materials and Systems Dielectric Measurements Developments with varied driving amplitude; non-linearity Power-Equipment Applications of DS
Dielectric Materials
Polarisation: bounded movement of charge, changing charge-distribution increases supplied charge to electrodes (if xed V) many mechanisms (e , natural dipole, ionic . . . ) some mechanisms very fast, lumped as , an increased 0 other mechanisms have considerable dynamics: represent as such with polarisation function f (t) or susceptibility ( ):
P (t) = 0
0
f ( )E (t )d
V
free space
P ( ) = 0 ( )E ( ) ( ) = ( ) i ( ) . . . = F {f (t)}

10
Dielectric Spectroscopy
Dielectric Spectroscopy = Dielectric Response at varied t or .
Time-Domain (TD) measure f (t), (plus conduction and possibly prompt response): e.g. Step-Response, e.g. Polarisation-Depolarisation Currents (PDC) Ramp-Response Return Voltage Measurement Many frequencies are measured at once, but noise rejection is low. Frequency-Domain (FD) measure ( ) (plus conduction and prompt response), usually by applying a (usually) sinusoidal driving voltage V ( ) and measuring the resultant current I ( ) for various . C0 I ( ) ( ) i ( ) = C ( ) iC ( ) = iV ( ) 0 i h h i ( ) i ( ) = 0 ( ) + ( 0 ) + 0 i 0 ( ) + We favour FD: ease of PD measurement, form of non-linearity.
11
Practical DS Measurements
Basic Measurement sees: Free-space capacitance Polarisation current Bulk conduction Surface conduction (Fringing, material case) Guarding: remove surface conduction and fringing. Removal of conduction current is harder: PDC time-domain methods, or analysis of frequencydomain results within assumptions of linearity. On actual equipment, one might be less interested in the material properties than in the total current.
h A
material material A
surface
12
HV-(FD)DS System
HV amplifier
30kV, ~50mA
Vref
Electrometers FB(v)
divider
Measurement Object
Vmeas
FB(i)
v
guard electrode measurement electrode
Imeas
ADC, DSP, computer
i
instrument earth
A much simplied block diagram, showing the measurement and guard electrodes.
13
Non-linearity measured with FD-DS

Non-linearity:
out in
relation is dependent on amplitude.
Currents at frequencies other than the fundamental i non-linear or supply voltage not sinusoidal. Frequency-representation allows even small non-linearity to be seen. Frequency-representation still allows a lot of directly interpretable information about the form of the time-domain distortion (odd/even, sine/cosine). C and C are values calculated on the fundamental components: C maintains its signicance of power loss.
K K
s (t) = Re
n=0
|S n |e
i(nt+Sn )
=
n=0
(An cos(nt) + Bn sin(nt))
14
Applications of HV-DS
Why HV? Non-linearity Higher current (SNR) Development of HV-FDDS eld-test system Water-trees in XLPE cables (picture) Machines: usually in time (TD)
Megging? (Qualies? Inverse Step Response!) Ramp test A little recent research interest, TD & FD End-windings often guarded! (lab, not eld)
ETK Aesthetic Obligation
15
Partial Discharge measurement and analysis

Partial discharges (PDs) PD detection and analysis Eect of frequency on PDs PDs and PD measurement in stator insulation
16
Partial Discharges (PDs)

A discharge that fails to bridge the space between the electrodes that are applying the eld; it therefore cannot develop into a disruptive discharge. For example:
divergent field barrier
Presence of PD can suggest a problem in an insulation system; detection methods may allow a distinction to be seen between dierent sources of PD activity. PD may cause insulation degradation by thermal, chemical, electrical and radiation eects.
17
PD Detection
Many detectable consequences, for example . . . released chemicals, e.g. Ozone (O3 ) remnant chemical eects, e.g. powder on surfaces sound, ultrasound visible light, UV light RF emission HF and total (integrated) currents in supply
Some PD detection methods give an averaged idea of PD activity, some give information about each PD event (pulse). Some PD detection methods localise the PD sources.
18
PD Charge & Calibration

PD charge > measured charge Qapparent < Qactual Qapparent energy in PD (xed V) Calibration: relate measured Qapparent to a known injection at the terminals. Frequency-response (spectrum of pulse) can aect measured charge. DS methods can (but neednt) capture the whole measured charge.
Qapparent = V Vcavity
Vcavity Qactual V
Qactual
Simplistic relation of actual and measured PD charge, assuming cavity voltage not changed.
19
VF-PRPD System
HV amplifier
filter coupling capacitor 200pF
Vref sync PRP

DAP and computer
Measurement object
PRPD system
preamplifier alternative placements of measuring impedance
Measuring impedance is typically in test-object earth for lab, or in coupling capacitor earth for eld tests of earthed objects. In our division at KTH, all but the PR-PD system is home-brew.
20
Phase-resolved PD: Simple Example

Qmax PD counts (#cycles)
1
charge channels
1 1 2
0
2 1 3
Qmax 0
90
180 phase channels
270
360
For each cycle, for each phase-channel (x), any measured PD charge, Q (y ) increments the count in that phase-amplitude (x-y ) point.
21
Phase-resolved PD: Example Plot

x 10
9
0.46 0.4 5
charge, C
0.35 0.29
0.23 0.17
0.12 0.06 0 90 180

phase, degrees
270
360
Intense cluster at bottom is typical for cavities; here there are also many, widely distributed, large charges delamination?
22
Phase-resolved PD: View as Current

x 10
9
0.46 0.4 5
charge, C
x 10
I(t)
0.35
current, A
0.5
0.29 0 0.23 0.17 5 0.12 0.06 0 90 180

phase, degrees
0.5
270
360
1 0
0.5
1
time, s
1.5
x 10
| I |, A
I(f): Polar form
sine
,A
4 2 0 2 4
x 10
I(f): Rectangular form (sin/cos)
0 x 10
7
harmonic order, n
harmonic order, n
180 90 0 90 180 0 1 2 3 4 5 6 7
harmonic order, n
phase, degrees
4 2 0 2 4
cosine
,A
harmonic order, n
23
Applications of PD Measurement
Oil-paper insulation bushings transformers paper-insulated cables and accessories Generator windings: well-established Busbars, GIS, outdoors (e.g. UV cameras)
24
PD with Variable Frequency Excitation

Sources of dependence frequencycavity relaxation of voltage across the cavity due to conduction of current through surface on the cavitys walls. material relaxation of (space-charge) eld in material, due to bulk and bulk . statistical delay due to the (random) occurence of a suitable initiating ionisation in the cavity volume v . Consider e.g. large and small material and surface , or both = 0.
bulk bulk A surface h
25
Example: PD Frequency-Dependence
Number of PD pulses from a single cylindrical cavity. Note the greater distinction between dierent cavities when at low frequency.
10 8 6
10 7 1.5
10 8 6
4 2 0 -2 10
4 2 0 -2 10
10
-1
10 f
10
10
10
-1
10 f
10
10
26
Frequency-Dependence of Stator PD
All three general-case eects material , cavity , statistical may be involved in the many sizes and shapes of cavity and delamination that can all be present together within the insulation. Around the end-windings: PD on the surface, damaged stress-grading, conductive contamination; PD between end-windings. Spreading of earth potential increases the stress in the insulation material under the stress-grading more PD, from this part of insulation too? At extreme low frequency, end-winding stress-grading may have earth potential up to its end surface PD?
27
PD Measurement on Stator Insulation

discharge between endwindings worn endsemicon 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 (end) 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 1111 111 000 111 0000 000 111 000 111 0000 000 1111 111 000 111 0000 000 1111 111 000 111 0000 1111 000 111 000 111 0000 000 1111 111 000 111 0000 000 1111 111 000 111 0000 1111 (slot) 000 111 000 111 0000 000 1111 111 000 111 0000 1111 000 111 000 111 0000 000 1111 111 000 111 0000 000 1111 111 000 111 0000 1111 000 111 000 111 0000 000 1111 111 000 111 0000 1111 worn slot semicon internal cavities/delaminations
Several very dierent sources: dierent magnitude to t between detection threshold and full-scale Large size: many simultaneous PDs: some lost in dead-time? PD propagation: stator is large, contains magnetic material but also the unshielded end-windings; measuring system frequency-response important.
28
Advantages: HV-FDDS or VF-PRPDA

FD-DS is here extended with the independent (controlled) variable of voltage amplitude, V , and with the independent (measured) variable of the low harmonic spectrum. excite non-linearities in the test-object distinguish non-linear components in the current PRPDA is here extended with the independent variable frequency, f . get a better idea of location and type of PD sources reduce driving voltage source if able to stay fn
29
Simultaneous VF-PRPDA and HV-FDDS

Both methods require a quite expensive HV amplier, driven by the controlling computer. Both methods may take a long time, when doing LF sweeps at several voltages! Both methods give complementary information, and it may even be good to compare measurements from the same measurement time if PD currents measured by the two systems are to be compared. = combined system, measuring simultaneously, gets more information, more directly comparable information, takes no more time than using just one of the methods, and costs less than the sum of the separate systems.
30
Simultaneous Measurement: Usefulness

Simultaneous Measurement even without a strong link between DS and PD measurements may be useful in itself: more clues to the state of the insulation, without extra measurement time. If PD current seen by integrating charges measured by PD system is very similar to the PD current measured in the DS system, the PD part of the DS measurement could be removed. If not, then there are two measures of PD, which may be a useful complement.
31
Difficulties in Diagnosis
32
Field-Test: DS on Whole Windings

Large-diameter hydro-generator. Sn = 10 MVA, Un = 6.3 kV About 30 years old. Diagnostic testing before and after some maintenance work.
33
Whole Winding: Insulation Resistance

IR (megging) at 5 kV. 3 occasions. Measure u, v, w, uvw.
1500
Expect 5000 M at 600 s Large dierence between phases
"resistance", MOhm
1000 v w
x : 1a + : 1b o : 2
u 500
uvw value tracks worst of u, v, w Large dierence between the three measurement occasions 1a,1b,2!
uvw
0 0
100
200
300 time, s
400
500
600
34
Whole Windings: DS with LV & LF

10 10 10 10
5
Occasion 1b.
C uvw C u, v, w
capacitance, F
Measurement: peak
50
C uvw
8
C similar for u,v,w LF C w rises

10
2
w rising C u, v, w
10 2 10
10
10 10 frequency, Hz
LF C uvw rises too
35
Whole Windings: DS with HV

5.75 x 10
7
2.8 2.6
x 10
C (0.316 Hz), F
5.74
C (0.316 Hz), F
u 2.4 2.2 2 1.8 0 v w
5.73 u 5.72 v w 5.71 0 2 4 6
applied peak voltage, kV
applied peak voltage, kV
(Note: a dierent occasion (2) from the LV&LF measurement (1b) ). Too little time or current to go much lower or higher in f . A large dierence in loss is seen in u, with HV.
36
Final Points
Many dierent types of diagnostic measurement for (stator) insulation systems. Little unanimity as to most suitable, useful measurements, even for a particular apparatus and failure mode. Results can often be ambiguous. Use of multiple methods may help to give better estimation of probability and mode of potential failure. On-line monitoring is increasingly popular (better systems available? regulatory market pressure?). O-line (during maintenance) measurements generally less rigidly performed than decades ago; variable interval, consideration of cost of measurements.
37
Corner Stones of Network Asset Management

Maintenance management in power systems
Helsinki University of Technology
Matti Lehtonen
KTHKTH-TKK course 2007
Network Asset Management

Planning methods -long term & expansion -replacements -optimal network structure
Load capacity and loading levels - normal state - reserve capacity - load forecast
Maintenance - life time control - condition monitoring - refurbishments - network value
Load capacity management
Normal state load capacity defines the aging and utilisation degree of power components. components. Reserve state load capacity defines the risks. risks. Present utilisation degree defines the timing of new major investments. investments. It is possible to increase utilisation degree and reduce risks if reliable information about network and component states. states.
Key factor: factor: upup-toto-date information about load flows and network state Cable networks and power transformers
thermal limits are decisive load currents & dynamic thermal time constants knowing the installation environment is essential
Overhead lines
voltage drop is the limiting factor key factor is the voltage monitoring and control decisive factor voltage level at the customer
4
RealReal-time network state estimation
combines network measurements, measurements, kWhkWh-meter data and load models => reliable onon-line information about network state data base about loads and component temperatures forecast function to support network operation
20 kV 0,4 kV
110 kV
P,Q,I
U P,Q,I P,Q,I P,Q,I I,T
T P,Q,I U P,Q,I,U U P,Q,I,U,T
MITTAUS LASKETTU
KTHKTH-TKK course 2007 5
Real time state estimation
Voltage monitoring and control according to EN 50160 voltage as received by customers HVHV- and LVLV-voltage drop estimated together
In overhead systems voltage drop is decisive to load capacity!
0,04 talvi Jnnitteen osuus koko vuoden havainnoista 0,035 0,03 0,025 0,02 0,015 0,01 0,005 0 208 211 214 217 220 223 226 229 232 235 238 241 244 247 Jnnite [V] kevt kes syys
Maintenance management
Condition monitoring
inspections: what, when, how much ? knowledge about remaining life time
Condition of network areas / sections

maintain or rebuild ? timing of refurbishment actions
Management of network value

key factor is the maintenance strategy
7
Inspections & condition classes

Example:
Condition grade
4 3,5
Life-time model of wood poles timing of 1. inspection estimation of remaining life
3 2,5 2 1,5 1 0,5 0 0 5 10 15 20 Age Mean value Deviation 25 30 35 40 45
Optimising inspections
RCM-example:
inspection of wood poles Key factor is the estimated condition class (OC-outage cost)
class > 3.5 => red colour class > 2.5 & OC > 5 kEur => violet class > 1.5 & OC > 10 kEur => yellow OC > 200 kEur => blue
Optimising maintenance strategy

STRATEGIC DECISIONS: Reliability centered maintenance RCM Condition based maintenance CBM Time based mainteance TBM Corrective maintenance CM
Risk of outage/ loading degree
TBM
RCM CBM
CM
Condition/ remaining lifetime
10
Optimising maintenance strategy

Optimal maintenance: Preventive maintenance v.s. Corrective maintenance v.s. Costs of faults and outages
Costs of corrective maintenance Costs of outages Total costs Optimal volume of preventive maintenance
Target area
Costs of preventive maintenance
11
Management of network value

ASSET MANAGEMENT STRATEGY:
Number of observations 600
500
Network value estimate - technical state now - forecast (5-10 yr) - reports to authorities Plans and budgets for replacements Optimising the maintenance of different components and network parts
400
300
200
100
0 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
Installation year
Grade 0 Grade 0 (u) Grade 1 Grade 1 (u) Grade 2 Grade 2 (u) Grade 3 Grade 3 (u) Grade 4 Grade 4 (u)
12
Network planning methods
Time spans of long term planning

transmission system 1010-30 yrs subtransmission system 6 20 yrs primary substations 6 20 yrs distribution system topology 6 20 yrs distribution feered 4 12 yrs distribution laterals 1 4 yrs secondary substations 0.5 2 yrs
The role of long term planning is to guide short term investment decisions so that the development of network structure is sustainable. sustainable.
13
Example of planning challenges: Siting of primary substations
To find the optimal solution one should make right guess guess about the other substations which will be built later. later. In the worst case a lot of extra costs due to rearrangements of MVMV-lines. lines. Build new substations or reinforce the existing ones. ones. Optimun number of substations ?. Long term plan: plan: the number and location of primary substations. substations. Shorter term planning: planning: timing. timing.
14
Requirements for successful planning
Good input data: knowledge of loads and careful spatial load forecasts Daily investment decisions are checked against long term plan, plan, which is updated every 5th year. year. Careful optimisation of primary substation number and locations. locations. Degree of uncertainty in loads and external factors is high => several scenarios analysed Ranking of the investments in increasing supply reliability using incremental cost/ cost/benefit analysis. analysis.
15
On the optimal strategies of condition monitoring and maintenance allocation in distribution systems
Matti Lehtonen
Helsinki University of Technology
Maintenance strategies in power system level

CM, Corrective maintenance fix or replace in case of fault TBM,Time-based maintenance traditional fixed intervals, safety dictated targets CBM, Condition based maintenance, actions based on state of the component, estimated or measured RCM, Reliability centered maintenance, actions selected according to the state and function of the component => optimisation task !
Risk of outage/ loading degree
TBM
RCM CBM
CM
Condition/ remaining lifetime
RCM at the component level What are the failure modes and their effects (FMEA analysis) How maintenance actions can affect the above What is the cost efficiency of maintenance activities Critical issue is the quality of data:
deterioration models fault frequency dependence on component state data from field: dianostics, inspections
Example 1: Power transformers
DP-number
1200
1000
800 DP DP 600 Overhaul effect T04 400
200
0 0 5 10 15 20 25 Age in years 30 35 40 45 50
Example 1 continued
Decay in paper DP means increased failure probability Optimal time for overhaul, when sum of outage costs, maintenance, overhaul and investment at minimum :
( COC + CM + COH + CINV ) = min ! where COC = T P OC
Optimal time for replacement when difference in outage and maintenance costs exceeds the investment costs:
COC + CM > CINV where COC = T P OC
Example 2: Wood poles

4 3,5 3 Condition grade 2,5 2 1,5 1 0,5 0 0 5 10 15 20 Age Mean value Deviation 25 30 35 40 45
Example 2 continued
The expected outage costs of a pole: COC = T P OC
where T is outage time, P power interrupted, OC per unit outage cost
RCM-classification poles for inspection

the poles with highest expected outage costs inspected first the challenge is to express as a function of pole condition the same logic can be applied to pole change: when difference in expected outage costs > investment, then change a new pole.
Conclusion
The critical data is the component fault frequency as a function of the age and maintenance actions For successful optimum this information should be precise and reliable The data should be verified by fault statistics and condition inspections
Guest lectures during the seminar days

Relibility analysis using Fuzzy interval mathematics by Suprakash Gupta (theme 1) Computational Intelligence for Scheduling Power System Maintenance by Ganesh Kumar Venayagamoorthy (theme 4) Diagnostic Based Maintenace of Mechanical Systems using the PSD-method by Fredrik Stillesj and Niclas Dahlgren (theme 3)
Relibility analysis using Fuzzy interval mathematics

Dr. Suprakash Gupta
Reader, Department of Mining Engineering, Institute of Technology, Banaras Hindu University, Varanasi, India. Email:- suprakash_gupta@yahoo.co.in
Introduction Concept of fuzzy interval mathematics Application of fuzzy interval mathematics in reliability Estimation of reliabiliy parameters from experts subjective judgements Algorithm Case study Results Conclussion
INTRODUCTION
Fault Tree Analysis (FTA) is a widely used method for analysing systems failure logic and calculating overall reliability. FTA is the translation of a physical system into a structured logic diagram. In conventional FTA, the failure probabilities of basic events are treated as exact values. In the absence of adequate data for ststistical inference, it may be necessary to work with rough estimate of probabilities from the subjective assessment of experts. Probabilistic objective events estimation using statistical analysis of data Subjective events - estimation using fuzzy interval arithmetic concept on the subjective judjements of experts.
Fuzzy interval mathematics and its use in reliability

Fuzzy sets a set of ordered pairs that accomodates various degrees of membership values of its elements on the real continuous interval [0, 1]. Symbolically, level set a number of crisp sets called level sets or simply cut may be associated with fuzzy set . All of the elements of cut belong to the fuzzy set at least to the degree and are expressed as ~
~(x ) } A = { x X A where [ 0,1]
~ ~ (x)) x X} A = {(x, A
cut of convex fuzzy set - for a convex fuzzy set , ~ ( ) its cut is denoted by A = [a 1 , a (2 ) ]
Experts subjective assessment and fuzzy numbers subjective information is imprecise and can be modelled by a fuzzy number. Fuzzy numbers can be represented by a membership function. Four main arithmetic operations on two fuzzy numbers, described by cuts are as follows
A trapezoidal fuzzy number can be defined by a quadruple (a1,a2,a3,a4) with a1a2a3a4 and its membership function is
Lingustic values of reliability data linguistic variables are used to deal many real-world situations that cannot be defined exclusively using a conventional quantitative technique due to their vagueness. The values of linguistic variables are given in common language and they constitute a term set of a linguistic variable. The term set of the linguistic variable failure interval (TF) used here is
The membership functions of fuzzy variables of the term set TF were developed on the basis of the experts perception of the subjective terms as represented below
All the experts were asked to comment on their perception of all the seven subjective terms in a range (x,y) of failure interval. The values of the quadruple (a,b,c,d) of the trapezoidal membership function were determined from the experts response i.e. a = minimum of all xij, b = average of all xij, c = average of all yij and d = maximum of all yij.
The membership functions for the linguistic values of the failure interval in trapezoidal fuzzy numbers are as follows
Method of aggregating the fuzzy numbers

Let the quadruplet Aij = (aij, bij, cij, dij) represent the trapezoidal fuzzy numbers where i = 1, 2, , m and j = 1, 2, ., n are the linguistic expressions given to event i, by expert j. The expression for aggregating the
judgements of n experts in a fuzzy number is
Where Mi = average fuzzy number of event i and wj = is the weight of the judgement of jth expert.
Defuzzification method of fuzzy output Estimation of reliability parameters from experts judgement is a fuzzy number and required defuzzyfication as the reliability calculation require a crisp value. This was done by cetre of area (COA). If a fuzzy set out(x) can be described in the interval [x1, x2] then the COA defuzzification x* of out(x) can be defined as
Importance ranking of basic events

Birnmaum Importance - The birnbaum factor of a component is calculated by determining overall system probability with the basic event out and then calculating the overall system probability with the same basic event in.
I iB ( t ) = Fs ,1.0 Fs ,0.0
Fussell-Vesely Correction factor Birnbaum factor multiplied by its probability of failure
IiF V ( t ) = IiB ( t ) q i ( t )
Proposed methodology
Algorithm
Construct the FT Classify the basic events in two groups Determine reliability parameters for first group Collect experts openion regarding second group Convert linguistic events into fuzzy numbers Aggregate experts openion Defuzzify the fuzzy output Calculate system reliability Rank the basic events for future course of action
Case study
Systems FT consists of 51 basic events. Adequate data were available for 35 events and scale () and shape () parameters of Weibull distribution were estimated. MTTF of 16 events were estimated using experts openion. Overall variation of system reliability with time were studied. Ranking of components was done at t = 10 hours by Birnbaum factor and Fussell-Vesely factor.
Discussion of results
Statistical analysis of TTF data reveals a few components are presently in their worn-out period while some shows a decreasing failure rate. 50% of the components are in their chance failure region. The characteristic life of the system is approximately 10 hours. The track carrier has the highest impact on system reliability as confirmed by the rank list. Line pan, flight bar, chain link also contribute much to the system reliability.
10
Conclussions
Proposed model can be used for reliability analysis of a system in the case of inefficient maintenance reporting or insufficient data. It can be used to find the weak links in the system for design as well as quality improvement. The outcome of this type of analysis can be useful tool for management to frame their replacement and maintenance policy. The hybrid approach fulfils the gap between available and required information. It ensures that even with limited amount of documented data a basic analysis for the performance of a fleet of equipment can be carried out.
Thank you
11
Computational Intelligence for Scheduling Power System Maintenance

G. Kumar Venayagamoorthy, PhD
Associate Professor of Electrical and Computer Engineering & Director of Real-Time Power and Intelligent Systems Laboratory University of Missouri-Rolla, USA http://www.umr.edu/~ganeshv www.ece.umr.edu/RTPIS gkumar@ieee.org Presented at 2C5060, Maintenance Management in Power Systems, Royal Institute of Technology, Stockholm, Sweden June 13, 2007
Acknowledgement
Financial support from the following organizations is gratefully acknowledged for the work presented herein:
National Science Foundation, USA Office of Naval Research, USA University of Missouri Research Board.
Outline
Introduction to Computational Intelligence (CI) Generator Maintenance Scheduling (GMS) Potentials of Advanced CI for Dynamic Power System Maintenance Scheduling
Intelligence

Ability to comprehend, to understand and profit from experience, to interpret intelligence, having the capacity for thought and reason (especially, to a higher degree). Creativity, skill, consciousness, emotion and intuition.
Computational Intelligence (1)

Computational Intelligence (CI) is the study of adaptive mechanisms to enable or facilitate intelligent behavior (intelligence) in complex and changing environments. These mechanisms include paradigms (AI) that exhibit an ability to learn or adapt to new situations, to generalize, abstract, discover and associate.
Computational Intelligence (2)
Computational Intelligence (CI) can be defined

as the computational models and tools of intelligence capable of taking raw numerical sensory data directly, processing them by exploiting the representational parallelism and pipelining the problem, generating reliable & timely responses and withstanding high fault tolerance.
AI vs. CI
Traditional AI has studied the possibilities of creating programs and machines that can rival human abilities in some tasks by means of heuristics, encapsulated knowledge, symbolic logic and many other methods.
In contrast, modern CI has explored the potential for creating intelligent machines by modeling the behaviors and mechanisms that underlie biologically intelligent organisms.
Computational Intelligence Paradigms

http:// www.ece.umr.edu/RTPIS
Neuro-Swarm Systems
Neural & Immune Networks
Swarm Intelligence
Neuro-Genetic Systems
Evolutionary-Swarm Systems
Neuro-Fuzzy Systems
Fuzzy-PSO Systems
Evolutionary Computing
Fuzzy-GA Systems
Fuzzy Systems
Neural Networks
A neural network can be defined as a massively parallel distributed processor made up of simple processing units, which has the natural propensity for strong experiential knowledge and making it available for use. The neural network resembles the brain in two aspects
Knowledge is acquired by the network from its environment through a learning process. Interneuron connection strengths, known as synaptic weights, are used to store acquired knowledge.
Artificial Immune Systems
Artificial Immune Systems (AIS) are biologically inspired models for immunization of engineering systems. The pioneering task of AIS is to detect and eliminate non-self materials, called antigens such as virus or cancer cells. The AIS also plays a great role to maintain its own system against dynamically changing environment. The immune systems thus aim at providing a new methodology suitable for dynamic problems dealing with unknown/hostile environments.

Evolutionary Computing has as its objective the model of natural evolution, where the main concept is survival of the fittest: the weak must die, the elites move to the next level. In natural evolution, survival is achieved through reproduction. Offspring, reproduced from two parents, contain genetic material of both parents hopefully the best characteristics of each parent. Those individuals that inherit the bad characteristics are weak and lose the battle to survive. In some bird species, one hatchling manages to get more food, gets stronger, and at the end kicks out all its siblings from the nest to die. GAs, GP, EP, ES
Swarm Intelligence
Swarm intelligence originated from the study of colonies (ants, bees, termites) or swarms of social organisms flock of birds, school of fish. Studies of the social behavior of organisms (individuals) in swarms prompted the design of very efficient optimization and clustering algorithms. SI is an innovative distributed intelligent paradigm for solving optimization problems.
Fuzzy Systems

Traditional set theory requires elements to be either part of a set or not. Similarly, binary-valued logic requires the values of parameters to be either 0 or 1, with similar constraints on the outcome of an inferencing process. Fuzzy sets and fuzzy logic allow what is referred to as approximate reasoning. With fuzzy sets, an element belongs to a set to a certain degree of certainty. Fuzzy logic allows reasoning with these uncertain facts to infer new facts, with a degree of certainty associated with each fact. In a sense, fuzzy sets and logic allow the modeling of common sense.
1. Evolution is the process of adaptation with the aim of improving the survival capabilities through processes such as natural selection, survival of the fittest, reproduction, mutation, competition and symbiosis. 2. Evolution is an optimization process. 3. EC is a field of CI which models the processes of natural evolution.
Main Steps in Evolutionary Computation
1. Encoding of solutions to the problem as a chromosome. 2. Fitness function evaluation of individual strength. 3. Initialization of the initial populations. 4. Selection operators. 5. Reproduction operators.
General Evolutionary Algorithm

1. Initialize a population of N individuals. 2. While no convergence: Evaluate the fitness of each individual in the population Perform cross-over select 2 individuals produce offspring Perform mutation select one individual mutate Select the new generation Evolve the next generation. Convergence is reached when: max. generations is exceeded, acceptable fitness is evolved, fitness does not change significantly over past x generations.
Evolutionary Computing vs. Classical Optimization

1. No-Free-Lunch theorem [Wolpert and Macready 1996] states that cannot exist any algorithm for solving all problems that is on average superior to any other algorithm. 2. Thus, the motivation for research into new optimization especially EC. 3. Classical optimization algorithms are very successful for linear, quadratic, strongly convex, unimodal problems. 4. EAs are more efficient for discontinuous, nondifferentiable, mutlimodal and noisy problems. 5. COs use deterministic rules to move from one point to other in the search space while ECs use probabilistic transition rules.
Evolutionary Computing vs. Classical Optimization

1. COs use sequential search while EAs use parallel search. 2. COs use derivative information, using first order or second order, of the search space to guide the path to the optimum. 3. ECs use no derivative information. Only fitness values of individuals are used to guide the search.
Simple Flowchart for Genetic Algorithm
Main Steps in Genetic Algorithm
Uniform Crossover
Parent 1
Parent 2
Mask
1 0 1 0 0 1 0 0 0 1
Offspring 1
Offspring 2
Random Mutation
1. Bit positions are chosen randomly and the corresponding bit values negated. 2. Pseudocode algorithm assume mutation rate pm For each i = 1, , I: (a) Compute a random value ~ U(0,1). (b) If pm then complement bit.
Before Mutation
1 0 1 0 0 1 0 0 0 1
Mutation points After Mutation

0 0 1 0 1 1 0 1 0 0
Applications of Evolutionary Computation

1. Planning, for example, routing optimization and scheduling GMS. 2. Design filters, neural network architectures and structural optimization. 3. Control controllers for process and plants. 4. Classification and clustering. 5. Function approximation and time series modeling. 6. Composing music. 7. Data mining. 8. Many others!
A huge swarm of red-billed queleas returns to the communal roost at dusk, Okavango Delta, Botswana.
Swarm Intelligence (SI) is the property of a system whereby the collective behaviors of (unsophisticated) agents interacting locally with their environment cause coherent functional global patterns to emerge.
School of Fish/ Flock of Birds

and the thousands off fishes moved as a huge beast, piercing the water. They appeared united, inexorably bound to a common fate. How comes this unity? Anonymous, 17th century
The motion of a flock of birds is one of natures delights.
Flock of Birds
A flock exhibits many contrasts. It is made up of discrete birds yet overall motion seems fluid. It is simple in concept yet is so visually complex, it seems randomly arrayed and yet is magnificently synchronized. Creates a strong impression of intentional centralized control. All evidence indicates that flock motion must be merely the aggregate result of the actions of individual animals, each acting solely on the basis of its own local perception of the world [Reynolds, 1987]. Birds draft behind each other. Birds that are in front, or on the outside, eventually move to the rear, or the inside, of the flock so that the wind-breaking is shared equally. This leads to an oval shaped flock.
Particle Swarm Optimization (PSO)

A concept applicable to optimizing nonlinear functions Has roots in artificial life and evolutionary computation Developed by Kennedy and Eberhart (1995) Key points Simple in concept Easy to implement Computationally efficient Effective on a variety of problems
Swarm Topology Ring (lbest version)

In this model, the particles do not take into account all of the other particles but only their neighbors. This corresponds to a limited perception of the outside world. This is actually more accurate than assuming total perception. For herds and schools of fish, especially in murky waters, there is a limited perceptual range. Even for birds, some of the birds will be occluded by others.
Swarm Topology Star (gbest version)
This results in the entire swarm particles trying to converge on one spot.
Particle Swarm Optimization

The system initially has a population of random solutions called particles Each particle has random velocity and memory that keeps track of previous best position and corresponding fitness The previous best value of the particle position is called the pbest It has another value called gbest, which is the best value of all the pbest positions in the swarm Basic concept of PSO lies in accelerating each particle towards its pbest and the gbest locations at each time step In local PSO, the gbest is changed to lbest where lbest is the best value of all the particles in local neighborhood.
PSO Equations
Xk+1 Vmod t (Gbest-Xk)
s (Pbestk-Xk) Xk r Vini X
The velocity of the particles is given as follows
Vid = w Vid + c1 rand1 ( Pbestid X id ) + c2 rand 2 (Gbestid X id )

The position vector of the particles is changed as follows
X id = X id + Vid
Further Details of PSO

Performance of each particle measured according to a predefined fitness function. Inertia weight influences tradeoff between global and local exploration. Good approach is to reduce inertia weight during run (i.e., from 0.9 to 0.4 over 1000 generations) Usually set c1 and c2 to 2 Usually set maximum velocity to dynamic range of variable.
Optimal Generator Maintenance Scheduling
GMS Problem
Generally, there are two main categories of objective functions in GMS - reliability and economic cost. This study applies the reliability criteria of leveling reserve generation for the entire period of study. This can be realized by minimizing the sum of squares of the reserve over the entire operational planning period. The problem has a series of unit and system constraints to be satisfied. The constraints include the following: Maintenance window and sequence constraints Crew and resource constraints Load and reliability constraints
1 if unit i starts maintenance in period t X it = otherwise 0
GMS Problem
Define
1 if unit i starts maintenanc e in period t X it = otherwise 0
The objective is to minimize the sum of squares of the reserve generation given by (1).
2 Min P X P L it ik ik t X t i i I k S it t it
(1)
GMS Problem
subject to the maintenance window constraint
it
= 1
tT i
(2)
M ik AMt
the crew constraint
X
iTt kSit
ik
(3)
and the load constraint
P X
it i iIt kSit
ik
P ik Lt
(4)
Penalty cost given by (5) is added to the objective function ((1)) if the schedule cannot meet the load or the crew and resource constraints (5) Penalty Cost
Discrete PSO
Discrete Particle Swarm Optimization (DPSO) allows discrete steps in velocity and thus in position. The new velocity and position for each particle is determined according to the velocity and position update equations given by (6) and (7).
Vi ( k ) = round ( w Vi ( k 1) + c1 rand 1 ( Pb X i ( k 1)) + c 2 rand 2 ( Pg X i ( k 1)))

(6) (7)
X i (k ) = X i (k 1) + Vi (k )
Flowchart for GMS with MDPSO

Start Initialize a population of particles
Evaluate the objective function given in ((1))
DPSO
Update pbest gbest and
Update velocity and position of each particle
X i (k ) = X i (k 1) + Vi (k )
EA
No
Is rand < mutation rate ? Yes Perform mutation using ((8))
X n = Pg n + 0 .5 randn () Pg n
No
Is termination condition satisfied ? Yes End
GMS with MDPSO

Data for the 13 units test system
Unit 1 2 3 4 5 6 7 8 9 10 11 12 13 Capacity (MW) 555 180 180 640 640 276 140 90 76 94 39 188 52
Maintenance duration (Weeks) Manpower required (per week)
10
10+10 +5+5+ 5+5+3
15+15
20
15+15 +15
15+15 +15
3+2+2+2 +2+2+2 +2+2+3
10+10 +5+5
20
15+15
10+10 +10 +10
15+15
15+15
10+10 +10
The maintenance outages for the generating units are scheduled to minimize the sum of squares of reserves and satisfy the: maintenance window constraint each unit must be maintained exactly once every 26 weeks without interruption load and spinning reserve - systems peak load including 6.5% spinning reserve is 2500MW crew constraints there is only 40 crew available each week for the maintenance work.
Results
To implement the DPSO and MDPSO, a population size of 30 particles was chosen to provide sufficient diversity into the population taking into account the dimensionality and complexity of the problem. This population size ensured that the domain is examined in full, and on the other hand it would mean increasing the running time.
Results
Three different DPSO/MDPSO parameters applied and 10000 iterations executed.
DPSO Mean number of iterations and standard deviation (x 106) 1.083 w=0.8, c1=2, c2=2 Linearly decreasing w (0.9 to 0.4), c1=2, c2=2 Constriction factor w=0.729 c1=1.49, c2=1.49 1.096 Ranking MDPSO Mean number of iterations and standard deviation (x 106) 1.073 Ranking
0.10 0.09
0.12
0.091
1.093
0.105
0.107
1.092
1.090
Best generator maintenance schedules obtained by DPSO and MDPSO

Week no. 1 2 3 4 5 6 7 8 9 10 11 12 13
Results
6 6, 7 6, 7
Generating units scheduled for maintenance
DPSO
2, 6, 7 6, 13
2, 6, 7 3, 6
3, 6, 13 6, 12
6, 13 6, 11, 12
6, 10 6, 10, 11
6, 10 6, 10
6, 10, 11 6, 7, 10
MDPSO
2, 6, 13
2, 6, 13
Week no.
14
15
16
17
18
19
20
21
22
23
24
25
26
Generating units scheduled for maintenance
DPSO
10, 11 7, 10
12
12
1, 8
1, 9
1, 9 1, 9
MDPSO
7, 8
1,9
Results
3200
MDPSO
Generation (MW)
3000 2800 2600 2400 0
DPSO System capacity Load constraint
10 15 20 25 Maintenance period (Weeks) 40
30
MDPSO
30 Crew 20 10 0 0 5 10 15 20
DPSO Crew available
25
30
Maintenance period (weeks)
Results
Convergence performance graph of a typical DPSO and MDPSO for w=0.8 and c1=c2=2. The results show that DPSO is likely to be stuck in a local optimum whereas MDPSO has a higher chance of escape from the local attraction.
x 10 3 2.8 2.6 Gbest fitness 2.4 2.2 2 1.8 1.6 1.4 1.2 1 0 50 100 150 200 250 300 350 400 450 500 MDPSO DPSO 6
Iterations
Results
The mean iterations for both the MDPSO and the DPSO when all constraints are satisfied. These results are obtained over five trials. DPSO converged prematurely than the MDPSO. The MDPSO avoided the premature convergence by making the mutated particles enter other region to continue searching for better global optimum solution.
Iterations when all constraints are satisfied DPSO Mean number of iterations and standard deviation (x 103) 0.04 w=0.8, c1=2, c2=2 Linearly decreasing w (0.9 to 0.4), c1=2, c2=2 Constriction factor [29] w=0.729 c1=1.49, c2=1.49 Ranking MDPSO Mean number of iterations and standard deviation (x 103) 0.768 1.732 Ranking
0.127
0.12
1 2
0.868
1 2
0.078
1.032 1.269
0.829
0.828
2.732
Results
Percentage of feasible optimal maintenance schedules obtained after 5000 and 10000 iterations. The result shows the efficiency and better performance of MDPSO over the DPSO. Both algorithms performed best for w=0.8 and c1=c2=2 and worst for the constriction factor based PDO for this GMS problem.
DPSO (Per cent of feasible optimal schedules) 5000 Iterations 43% w=0.8, c1=2, c2=2 Linearly decreasing w (0.9 to 0.4), c1=2, c2=2 Constriction factor [29] w=0.729 c1=1.49, c2=1.49 43% 53% 10000 Iterations 58%
MDPSO (per cent of feasible optimal schedules) 5000 Iterations 58% 10000 Iterations 79%
57%
66%
37%
50%
50%
62%
Optimal Dynamic Scheduling
Dynamic Programming
The basic concept of all forms of dynamic programming can be summarized as follows: Model of Utility Reality (F) Function (U) Dynamic Programming Secondary Utility (J)
Dynamic Programming
Bellmans equation of dynamic programming
J(t) =
k=0
U(t + k)
k
Approximate dynamic programming is obtained using a neural network called the Critic network to approximate the J - function.
Adaptive Critic Designs (ACDs)

The Adaptive critic designs have the potential of replicating critical aspects of human intelligence: - ability to cope with a large number of variables in parallel, in real time, in a noisy nonlinear nonstationary environment. The ACDs show a family of promising methods to solve optimal control problems. The origins of ACDs are ideas synthesized from dynamic programming, reinforcement learning and backpropagation.
Adaptive critic designs may be defined as designs that attempt to approximate dynamic programming in the general case. There are different ACDs varying in both complexity and power. All ACDs employ neural networks as modules of design.
Adaptive Critic Designs (ACDs)
1) 2) 3) 4)
Types of primary reinforcement: Explicit targets for system outputs are provided at every step. Explicit differentiable cost as a function of system variables is provided at every step. A graded cost is provided at each step but explicit relationship with system states is not given. Ungraded reinforcement is provided when appropriate, e.g. binary outcome at the end of a game.
ACDs as Supervised and Reinforcement Learning
ACDs are supervised learning systems in the cases of (1) & (2). They are reinforcement learning systems in the cases of (3) & (4). Critic may be thought of as a transformer of cases (3) & (4) to case (2).
Adaptive Critic Designs

A family of adaptive critic designs was proposed by Werbos in 1977 as a new optimization technique combining concepts of reinforcement learning and approximate dynamic programming. The adaptive critic method determines optimal control laws for a system by successively adapting two neural networks, namely an Action neural network (which dispenses control signals) and a Critic neural network (which learns the desired performance index for some function associated with the performance index). These two neural networks approximate the HamiltonJacobi-Bellman equation associated with optimal control theory.
A Hierarchy of Designs
Heuristic Dynamic Programming (HDP) Dual Heuristic Programming (DHP) Globalized Dual Heuristic Programming (GDHP)
Critic Choices
Heuristic Dynamic Programming (HDP), adapts a Critic network whose output is an approximation of J(R(t)). The temporal difference method of Barto, Sutton and Anderson turns out to be a special case of HDP. Dual Heuristic Programming (DHP), adapts a Critic network whose outputs represent the derivatives of J(R(t)). Globalized DHP (GDHP), adapts a Critic network whose output is an approximation of J(R(t)), but adapts it so as to minimize errors in the implied derivatives of J (as well as J itself, with some weighting). GDHP tries to combine the best of HDP and DHP.
Evaluation Function critic (HDP & DHP)

R(t) HDP J(t)
R(t)
DHP
J (t ) (t ) = R (t )
GDHP: Double Output Solution
R(t)
GDHP-style Critic
J(t) dJ(t)/dR(t)
Forms of ACDs
Action dependent Adaptive Critics - ADHDP, ADDHP & ADGDHP
R(t)
Action
u(t)
Critic
J(t)
Model dependent Adaptive Critics - HDP, DHP & GDHP

R(t) u(t) R(t+1)
Action
Model
Critic
J(t+1)
Adaptive Critic Designs
S Ferrari, et al, Model-Based Adaptive Critic Designs, in the Handbook of Learning and Approximate Dynamic Programming, Edited by J Si, A G Barto, W B Powell, D Wunsch, Wiley-IEEE press, July 2004, ISBN: 0-471-66054-X.
ADHDP Network Model Independent
Critic Neural Network

J ( Y ( t )) =
Y (t + 1) Y (t ) Y (t 1)
Target = J( Y(t+1)) + U( Y(t))
^ ^ ^
k =0
kU ( Y ( t + k ))
J( Y(t+1)) U( Y(t))
CRITIC Neural Network
Y (t ) Y (t 1) Y (t 2)
^ ^
J( Y(t)) + error
ADHDP with Fuzzy Action Network
Adaptive Critic Designs - HDP

Y(t)
TDL Yref PLANT Y(t)
ACTION Neural Network
A(t)
J(t) 1
TDL
J (t ) A(t )
Y (t )
TDL MODEL Neural Network
J (t ) Y (t )
^
[A(t-1), A(t-2), A(t-3)] [ Y(t-1), Y(t-2), Y(t-3)]
General Training Procedure

Then the Action network is trained further while keeping the Critic network weights fixed. This process of training the Critic and the Action one after the other, is repeated until an acceptable performance is reached. It is assumed that there is no concurrent adaptation of the pretrained Model network, and WC is initialized to small random values.
Coordination Between LTS and STS

The long-term scheduling (LTS) problem represents fuel allocation and budgeting, emission and production costing. The objective of short-term scheduling (STS) is to minimize the operation cost over hourly, daily or weekly horizon. STS is required to meet fuel supply and consumption limits, unit commitment requirements, transmission security constraints and hourly system demands. The third layer is the real-time economic dispatch with an objective of optimal dispatch for committed units to meet system requirements in the real-time operation.
1-2 years weekly increments
weeks hourly increments
Dynamic Scheduling Algorithm
HDP - LTS
Y(t)
TDL Yref Power System Y(t)
Dynamic Scheduler for LTS
A(t)
CRITIC LTS & STS
J(t) 1
TDL
J (t ) A(t )
Y (t )
TDL
J (t ) Y (t )
^
[A(t-1), A(t-2), A(t-3)] [ Y(t-1), Y(t-2), Y(t-3)]
LTS MODEL
HDP - STS
Y(t)
TDL Yref Power System Y(t)
Dynamic Scheduler for STS
A(t)
CRITIC STS
J(t) 1
TDL
J (t ) A(t )
Y (t )
TDL
J (t ) Y (t )
^
[A(t-1), A(t-2), A(t-3)] [ Y(t-1), Y(t-2), Y(t-3)]
STS MODEL
Thank You!
G. Kumar Venayagamoorthy, PhD Associate Professor of Electrical and Computer Engineering & Director of Real-Time Power and Intelligent Systems Laboratory University of MissouriRolla, USA
http://www.umr.edu/~ganeshv www.ece.umr.edu/RTPIS gkumar@ieee.org
Diagnostic Based Maintenace of Mechanical Systems using the PSD-method

Seminar at Department of Electrical Systems 13/6 2007
Condition Based Maintenace
Production
Stop time Stop Total stop time
Maintenace when needed
Planned maintenace Stop Total stop time Condition based maintenance Stop
State of the art techniques

Vibration measurements:certified technique; standard in industry Shock pulse monitoring Maintenace management systems Industrial IT-solutions
The PSDmethod:Theory
We use the induction motor as a power sensor A change in motor speed (slip) gives a change in motor torque and therefore a change of applied power Analysis of motor power in time and frequency domain
Object under test Bearing Bearing Applied energy Motor Gear wheel Delivered work
PSD Collector
PSD-method:pump example
PSD-method: Generator example
Parameter history
Demo
PSD technologies in products

Today
In our product PSD Collector
Heat and electricity production plants General Industries
OEM
Tomorrow
PSD Collector V2.0 OEM
Integrated Licenses
The day after tomorrow

Consumer products?
Course Project reports
Theme 1 Reliability data assessment and reliability modelling

Evaluation and development of test systems for analysis of electrical distribution systems Carl-Johan Wallnerstrm Life cycle scheduling for power transformers considering major overhauls and investments Jussi Palola Literature review in reliability assessments incorporating HVDC links by Johan Setrus Distribution system reliability engineering methods Influence of cable system and overhead line differences by Anna Guldbrand
Evaluation and Development of Test Systems for Analysis of Electrical Distribution Systems
A feasibility study within an IEEE PES Task Force on Distribution System Reliability
Draft July 5, 2007
Carl Johan Wallnerstrm
Evaluation and Development of Test Systems for Analysis of Electrical Distribution Systems, DRAFT 5/7-07 C. J. Wallnerstrm, KTH Electrical Engineering, Stockholm, Sweden
Abstract
This report is within the beginning of an IEEE Power Engineering Society (PES) Task Force with the overall goal of developing international test systems. The future work on the Task Force will be discussed during General Meeting 2007 in Tampa, Florida (see section 6.2) and this work will give several inputs to the overall project. First of benefits of developing and using international test systems are discussed in the report, such as the possibilities of comparing research results. The electrical power industry has gone trough fast worldwide changes (re-regulations, new techniques etc.), which both increase the need to doing analysis on distribution systems and to investigate the needs of updating existing test systems (alternative create complete new). Some important aspects when creating or developing international test system are discussed in the report. A grate focus is on new circumstances since the previous test systems were constructed. Another discussed aspect is the problem of handle differences between distribution systems (e.g. between countries) when creating general international test system. A test system could both be well defined or act as a core system which could be supplemented by the users with additional data and modification depending of the purpose. One common used solution to handle differences between distribution systems is to have an overall high voltage system (e.g. transmission level), which has many sub-systems with different characteristics (urban, rural, industrial etc.). Traditional, the focus of reliability analysis was on generation and transmission, and the first international test systems did not have any data of the distribution level. However, several studies have shown that most of the customer outages depend on failures at the distribution level, and the focus on distribution system analysis has increased. Four existing test system have been exemplified and shortly described in the report. Two of them have been published in IEEE transactions: IEEE Reliability Test System (RTS) and RBTS. The first of them is larger, but only partly defined at the distribution level. Both of them have sub systems with different kinds of customers and characteristics. Finally, two computer programs for reliability analysis of distribution systems, RADPOW and NEPLAN, are introduced. The test systems in this report have party been implemented in the two programs in earlier studies.
Contents
Abstract ...................................................................................................................................... 2 Contents...................................................................................................................................... 3 1 Introduction ........................................................................................................................ 4 1.1 Goal ............................................................................................................................ 4 1.2 Background ................................................................................................................ 4 1.3 Scope and outline ....................................................................................................... 4 2 The benefits of international test systems .......................................................................... 5 3 Aspects on present and future electrical power systems .................................................... 6 3.1 Maintenance management.......................................................................................... 6 3.2 Re-regulations, new laws and changed attitude ......................................................... 6 3.3 Comparison between different countries.................................................................... 6 3.4 New techniques and ways of building........................................................................ 8 3.5 Common mode failures .............................................................................................. 8 3.6 Other aspects .............................................................................................................. 8 4 Examples of test systems available .................................................................................... 9 4.1 IEEE Reliability Test System (RTS).......................................................................... 9 4.2 The RBTS................................................................................................................. 11 4.3 Existing Test Systems in NEPLAN ......................................................................... 12 4.4 Birka test system ...................................................................................................... 13 5 Examples of software available........................................................................................ 14 5.1 NEPLAN .................................................................................................................. 14 5.2 RADPOW................................................................................................................. 14 6 Conclusions ...................................................................................................................... 15 7 Literature .......................................................................................................................... 16
1 Introduction
This project report is written as a part of the examination in the post-graduate course F2C5060 on Maintenance Management in Power Systems. The course is given as collaboration between KTH School of Electrical Engineering and Helsinki University of Technology (TKK) Power systems and high voltage engineering, with Professor Matti Lehtonen (TKK) and Assistant Professor Lina Bertling (KTH) as examiners.
1.1 Goal
This work is within the beginning of an IEEE Power Engineering Society (PES) Task Force on Distribution System Reliability. The goals are to describe the benefits of using international test systems, describe aspects of present and future electrical distribution systems, and finally give an overview of some existing test systems. The descriptions of different aspects are motivated by the possible of new important circumstances for the electrical distribution industry since the previous test systems were construated. The overall long-term goals of the entire task force are to review existing available test systems and take one of following decisions (or a combination of theese) of how to continue: 1. Keep one or several existing test system(s) unchanged 2. Develop (an) existing test system(s). 3. Create new test system(s). Regardless of the continuation, the task force will continue working on making one or some test systems more easily available and advertised, and motivate the benefits of using test systems.
1.2 Background
Since the 70s, there have been IEEE projects of developing international electrical test systems for reliability assessment [1]. Traditional, the focus was on generation and transmission [2], and the first test systems did not include data at the distribution level. However, several studies have shown that most of the customer outages depend on failures at the distribution level [2]-[8]. This fact, together with a higher focus of customer outages costs since the worldwide trend of re-regulations, have increased the interest in reliability assessment for distribution system planning, e.g. maintenance management. During the IEEE General Meeting (GM) in San Francisco 14 June 2005, Lina Bertling leads the effort with this task force included members of the RRPA (Reliability, Risk and Probability Applications SubCommittee) and other interested persons and organizations e.g. CIRED, SwedEnergy [9]. During IEEE PES GM June 2007, Tampa, Florida, the project will be discussed in an open meeting (see section 6.2). This report is one of the contributions to bring up questions to discuss. The meeting will then become important to decide the contents of future work.
1.3 Scope and outline

In Chapter 2, an approach of describing the benefits of international test systems is given, which will motivate the project and give ideas of how test systems could be used. Chapter 3 shortly describes some aspects of electrical distribution system that perhaps can be useful to take into consideration when international test systems are developed. This report did not take a position of the best way of handle these aspects, but gives sometimes some alternatives. Chapter 4 and 5 give a short brief of existing test systems and software available. 4
2 The benefits of international test systems

Good test systems could be valuable for e.g. the development and evaluation of methods in maintenance management, but the possible usages of test systems are enormous. Models of technical systems with adequate simplifications are needed in several situations, both in academic and industry projects. These models could be created with respect to a specific real system, e.g. in distribution system planning of an electrical distribution system. Another alternative is to create a model, which symbolize a general system with typical characteristics. A test system is simply a model, which could be useful by different users and analysis. A test system could be based on a real system (e.g. Birka [8]) or be based on general compromises and assumptions. Test systems, in particular international with many users and possible application, must be developed carefully to satisfy every possible aspect. There is, however, a great work to create a good test system. Hence, one obvious benefit is that the use of a pre-defined test systems, if possible, save a lot of work compared with creating an own model for each analysis or study. If no test system satisfy every aspect, it could still be easier to modify an existing system compared with creating a complete new. For academic projects, one great benefit is the possibility to more easily compare research results, e.g. comparison of different reliability methods. Official defined test systems are also easy to use as references in e.g. academic publications; if an own model is used instead, all used data must be published (which could become several additional pages), since all academic work must be possible to repeat. Another benefit is the possibility to compare different software, e.g. RADPOW and NEPLAN. The work itself, of developing international test system, could give valuable discussions and exchanges between countries, researchers and companies. The exchange of ideas and knowledge of e.g. building models and differences between countries, are also an important part of the international work on test systems, as well as the concrete development.
3 Aspects on present and future electrical power systems

3.1 Maintenance management
One common assumption (e.g. in NEPLAN) is to not take out a component for preventive management when it have a redundant alternative in failure condition. Most reliability data are presented as mean values. To optimize maintenance plans, failure rates dependent on the time since the component was out for preventive management or replaced could be valuable. The problem is that methods often should be easy to use (especially methods used by the industry) and most of the existing methods and software are based on constant frequencies There is also a problem to get proper input data for this kind of analysis. Other input data which could be useful in maintenance management are for example: cost of preventive maintenance or monitoring; cost of reparation and fixed costs associated with maintenance.
3.2 Re-regulations, new laws and changed attitude

3.2.1 Re-regulations different regulating regimes The regulation in a country gives different outage costs and different incentives of building electrical distribution systems. An international test system could be useful to compare different regulations [13], evaluate or develop a specific regulation. 3.2.2 New aspects of outage costs The outage cost could differ between different countries, depending on e.g. the regulation and the cost of employees. The cost could be seen as a cost for the customers or for the electrical distribution system operator (DSO). There have been several studies [14] of the cost for the customer, and several methods of estimate this cost. The problem of having the customer interruption cost including in an international test system is that customers probably not experience the same cost in every country. Sometimes these studies are used as an input to different regulations, which make them indirectly to a cost for the DSO as well. Other examples of costs for the DSOs associated with interruptions are repair costs and a lower goodwill. 3.2.3 Risk management Risk management of electrical distribution systems includes several aspects, such as reliability, safety and pollution. To develop methods for risk management in electrical distribution systems, test systems could be used. In Sweden a new law [12] (since 2006) makes annual risk and vulnerability analysis compulsory for every DSO. 3.2.4 Other changed economic requirements Re-regulations and more private operators have increased the need to be economic efficient, but on other hand, the modern society is more and more dependent of reliable distribution which means more costs. However, it is difficult to know how the economic requirements will change in the future, but the DSOs have to be more carefully in the economic planning.
3.3 Comparison between different countries

3.3.1 Voltage levels and the amount of sub systems The voltage levels could be different in different electrical power systems. There are also sometimes differences within the same country or even in the same system. RBT is an 6
example of a test system, divided into two parts, with different voltages. For examples have Sweden different structures in different parts of the country, depending of the history, e.g. 220/70/20/0.4 kV or 130/40/10/0.4 kV. Test systems are often only defined with some of the voltage levels included. However, to include data for all voltage levels could be problematic; the transmission system is connected to several regional distribution systems, which each distributes electricity to several local distribution systems, which distributes to several lowvoltage systems each. Even small transmission systems, could therefore give a great number of different sub-systems. A test system could be constructed different to handle the great amount of sub systems: By only include one, or a few voltage levels (such as RBT). Give data for only a small selection of sub systems, such as the development of RTS made in [2] or develop o a few test systems for each voltage levels, e.g. urban or rural By create an electrical system of a smaller area but with data of more voltage levels, e.g. RBTS compared with RBT. 3.3.2 Different traditions of building electrical distribution systems There are different traditions of building (design and used techniques) electrical distribution system between, but also depending on when the system was built and sometimes between different DSOs in the same country. This report did not investigate these differences, but a comparative study could be good as future work within the task force. 3.3.3 Different geographical condition The geographical conditions have a great impact of the distribution systems. Earlier international test systems, such as RTS [1] [2] and RBTS [3] [4] partly handle this by construct the sub-systems so they complement each other. Some are typical urban, rural, industry or different mixtures of these; some have local generation etc. But it is not only different geographical human structure that affects the characteristic of e.g. reliability data. One example is that landscapes with much trees more often gives outages. Another example is that terrain with much stones impact the reliability (and even the possibility) of underground cables. The climate, such as wind, frost and snow, also affect the reliability. The variation of sub-systems has to be balanced with the complexity of a test system. 3.3.4 Develop international test system(s) useful for several countries This report did not take a position on the best why of handle this, but presents and discusses some alternatives: Alternative 1: A core system which could be supplemented by the users with additional data and modification depending of the purpose. The advantage of this solution is the flexibility and that only one international test system is needed. The disadvantage is that is could be difficult to compare the results. Alternative 2: A system defined in detail, used by everyone. Easy to compare obtained results from different countries, but the test system could differ from the real systems, which must be taken into consideration when the results are analyzed. Alternative 3: Several different test systems defined in detailed. More work to develop and only sometimes possible to compare with other analyzes, but on other hand could this solution give test systems which, in average, are more similar with real distribution systems. Alternative 4: Like e.g. RBTS [3] [4], i.e. have many sub-systems with different characteristics. 7
3.4 New techniques and ways of building

3.4.1 More common with cables instead of overhead lines In Sweden, as an example, the replacement of overhead lines with underground cables has rapidly increased, even at locations, e.g. rural systems, that traditional only have only overhead lines. New laws, regulations [12][13] and the opinion after a severe storm (Gudrun January 2005) have give these incentives. Today, it is impossible to know the consequences of these extremely fast changes of the Swedish distribution systems. There could be risks, but also a chance of better reliability. The situation is probably the same in other countries and could be valuable to further investigate. 3.4.2 Distributed generation RBTS has, for example, one sub-system located near generation [3], but no generation is situated at 33-kV level or below. The amount of distributed generation has increased world wide and will probably continue to increase, e.g. with a greater amount of wind power [15]. The interest of research activities and different analyses of distributed generation and its consequences have also increased a lot [15]. Therefore this aspect should be considered if new test systems are developed. 3.4.3 HVDC The consequences of include HVDC in the RTS have been analyzed [6], but none of the test systems included in this report, have any HVDC in the original case. Perhaps it is enough to additional add HVDC when needed, but this has to be considered. 3.4.4 Structure of new electrical power systems New techniques could make the construction of the distribution systems of the future easier, but also with unknown characteristics and risks. Also the components of the future could differ a lot compared with the situation of today. The techniques of today must be compared with the characteristics of the previous test systems, and some kind of estimation of the future could perhaps be done. However, the components in a distribution system have a long life time, so any changes will take relativity long, compared with other technical systems.
3.5 Common mode failures

A Common mode failure is a failure rate for one external event, when more than one component breaks, e.g. when several overhead lines become broken depending on extreme weather. Traditional reliability analysis methods did not handle these kinds of phenomenon, but e.g. in NEPLAN this possibility is included [10]. Without common mode failure, there is a risk that the system indices become too optimistic, since many failures not are independent from each other, especially in case of extreme events, such as Gudrun. Already in the development of RTS, the importance of this phenomenon was discussed [2], but never included.
3.6 Other aspects

This report discusses some aspects, but of course there are a lot more. It is important to decide a sufficient complexity and amount of aspects to take into consideration when a test system is developed. Some important suggestions of future work discussed in [2] which not have been already mentioned in this report are e.g.: Further expansion (load growth) and use of different load duration curves.
4 Examples of test systems available

4.1 IEEE Reliability Test System (RTS)
In 1979, a test system was published by IEEE Subcommittee on the Application of Probability Methods, referred to as the IEEE Reliability Test System (RTS) [1]. This system was made to e.g. compare different reliability methods at the generation and/or transmission level. Originally, the idea was a system which could be use as a frame system, with possibilities to individually add parameters, modifications and improvements [2]. The original version has data only for the transmission level (see Figure 1), which is divided into two parts separated by transformers (north 230 kV and south 130 kV). The system has, at this level, 24 buses (10 load buses) and an installed capacity of 3405 MW (peak load 2850 MW) [2] [10]. Until 1988, several developments of RBS were published for generation and transmission and 1988 data of the subsystems (distribution levels) for two of the load busses was given in [2].
Figure 1 Overall structure of the RTS test system at the transmission level (230 and 138 kV) [2][10]
The choice of the two load busses was made so they complement each other with different characteristics [2]. The voltage levels at the distribution part are 33, 11 and 0.4 kV respectively. Systems at 11 kV and below are radial or operate as radial. The overall structures of the two distribution systems (Bus 1 and 9) are shown in Figure 2 and Figure 3. It is easy to understand why only two of ten regional distribution systems in RTS are well defined since each regional system distributes electrical energy to several local systems each. This means, that only this two well-defined Busses in RTS, includes data of reliability and structure of about 2 regional (Bus 1 and 9) + 50 distribution (local systems connected to Bus 1 and 9) systems. These data are given in a 200 pages dissertation [2], mostly containing pictures of all 50 subsystems together with tables with data. An example of one (of about 50 defined) local distribution system is shown in Figure 4. 9
4.1.1
Bus 1
Figure 2 Overall structure of the distribution system of Bus 1 in RTS [2]
Bus 1 is situated near a power station and has a peak load of 108 MW. This system mostly distributes to small users, i.e. residential, small users, government and commercial. 4.1.2 Bus 9 Bus 9 is not situated near a power station and has a peak load of 175 MW. Compared with bus 1 there are more large users. The customer categories this bus distributes to are: residential, large users (industries etc.), commercial, office and farm.
Figure 3 Overall structure of the distribution system of Bus 9 in RTS [2]
One of the sub systems of Bus 9 is chosen as an example, which gives an idea of the entire size of RTS (see Figure 4).
10
Figure 4 One example the structure for one of about 50 sub-systems in Bus 1 and 9, defined in [2] (RTS, Bus 9, Sub 32)
4.2 The RBTS

RBTS, developed during the 90s in university of Saskatchewan by e.g. Roy Billinton, is smaller than RTS, but on the other hand is every load bus defined at the distribution level [3] [4]. The test system has two generations, six busses, five load busses, and an installed capacity of 240 MW and a peak load of 185 MW. RBTS has five voltage levels (230/138/33/11 kV). There are 18 288 customers, divided into several categories: Residential, Commercial, Government and ins., Small users, Large users and Farms. [3] [4] .
Figure 5 The overall structure of RBTS at the transmission level [3] [4]
11
In 1991 detailed data of load bus 2 and 4 and its sub systems was published [3] and in 1996 data of the load bus 3, 5 and 6 was done [4]. Figure 5 shows the overall structure of the RBTS and Table 1 gives an overview of the characteristics of the five different load busses.
Table 1 Overall description of the load busses in the RBTS [3] [4]
Number 2 3 4 5 6
# cust. 1 908 5 805 4 779 2 858 2 938
Description Typical urban and close to generation, 20 MW peak load Typical industrial and other large customer, 85 MW peak load Typical urban, 40 MW peak load Typical urban, 20 MW peak load Typical rural system, 20 MW peak load
4.3 Existing Test Systems in NEPLAN

NEPLAN has several pre-defined test systems, but just one (in the demo version) includes reliability data. The test system have one generation node which feed the 110 kV level and two transformers separate this part with a 10 kV level, see Figure 6.
Figure 6 A pre-defined test system in NEPLAN
12
The systems has 15 load busses (2 at 110 kV and 5 at 10 kV), 7 317 customers (6 000 at the 110 kV level) and a peak load of about 36.6 MW. The test system both includes underground cables and overhead lines (14 % at 10 kV and 0 % at110 kV, displayed with broken lines). Components with a failure rate in this example are circuit breakers and load switches (different kinds of failures); transformers, lines and busses. Additional reliability data included are common mode failures and interruption costs.
4.4 Birka test system

This test system was developed by Lina Bertling, KTH. In contrast to the other examples of test systems in this report, Birka system (220/110/33/11 kV) is a model of a real distribution system (situated in Stockholm, Sweden). The system has one supply point and three load busses; one load buss (33 kV) feed the underground, the other at 33 kV represent several subsystems distributing electrical energy to 23 400 customers and the load bus at 11 kV represent a mean sub system of this area with 447 customers. There are reliability data for every component in the model, including different kind of failures for the breakers. [8] The system is easy (because of its size) to handle in different analyses and have been used several times within the research and education in the RCAM group at school of Electrical Engineering, KTH.
Figure 7 The structure of the Swedish Birka System [8]
13
5 Examples of software available

In this chapter, two programs used by the RCAM group shortly are described. Another example of software is RELNET, which was used during the development of RTS in 1988 [2].
5.1 NEPLAN
NEPLAN is developed by BCP, a company in Switzerland, in collaboration with e.g. ABB. During the end of the 80s, the first version was done, but NEPLAN is continuous improved and upgraded. NEPLAN handle a great amount of different analyses, there reliability analysis is just one of many possible. The software could take many aspects of reliability into consideration such as common mode failures and interruption costs. [11] It is a commercial program, but a demo version (a limited number of load points allowed) is available for free download at the Internet. One benefit is the graphical interface (see e.g. Figure 6), but like many other commercial programs the algorithms are not published, which sometimes are a weakness when it is used in research. Modified versions of the RBTS (transmission level), RTS (transmission level) and Birka were implemented in NEPLAN as a part of a master thesis [10].
5.2 RADPOW
RADPOW is developed at KTH, school of Electrical Engineering, the RCAM research group, by e.g. Lina Bertling. The algorithms used by RADPOW are, unlike NEPLAN, published [8] which has a great value in the research and the education. It has no limitations of the size of the systems. RADPOW is under development and will be improved, for example, a simulation tool was recently included [16]. The distribution system of Bus 2 in RBTS is implemented in RADPOW, used in the undergraduate course on Reliability assessment of electrical power system in 2005. Some studies have compared the results from RADPOW with NEPLAN (e.g. [11] [10]) which show that the programs give similar, but not identical reliability system indicates for the same system, i.e. RADPOW has some different algorithms compared to NEPLAN
14
6 Closure
6.1 Conclusions and future work
There are several test systems of electrical distribution available and at least two of them are international published in IEEE transactions. Despite all previous defined test systems, needs to further investigate development of test systems has been identified and are motivated by the possible of new important circumstances for the electrical distribution industry since the previous test systems were construated. This report describes some of these aspects, together with other circumstances to take into consideration when developing new test systems. Examples of four different test systems are given, all with different characteristics and size. The great benefits of international test systems are motivated in this report and an increased interest of developing international test systems shows on this fact. The work of developing international test systems is organized as an IEEE PES Task Force on Distribution System Reliability. This work is within the Task Force at an early level to give different inputs to discuss and solve. During IEEE PES GM June 2007, Tampa, Florida, the project will be discussed in an open meeting (see section 6.2). This report is one of the contributions to bring up different questions and aspects to discuss. The results from the meeting will then become input to decide the contents of future work within the task force. The future work will include work on making one or some test systems more easy available and advertised, and motivate the benefits of using test systems and if motivated create new test systems (or develop existing).
6.2 Some first comments with regard to the meeting in Tampa June 25
This report was presented in the Maintenance Management Course before GM 2007 in Tampa and refers to a meeting within the conference as a future activity. However, deadline of final changes in the report is set to about one working-week after GM. It was a great attendance at the meeting, e.g. chairman of the Task Force (TF) Dr. Lina Bertling and the originator of RBTS Professor Roy Billinton was present. Even people that not could join the meeting have shown interest in taking part of the TF. It is too early in the work of follow-up the meeting to include results in this MM-course version of the report. The author will however give some short personal comments from the meeting in this section. Several of the participants of the meeting had the opinion that it is good to first of all investigate the possibilities of using and developing both RBTS and RTS or only one of them. One comment was also that RBTS should be described much more in detail in future versions of the report. A first step is to evaluate these systems further and examine more concrete why test systems are not used so much today by the industry and researchers. Preferably, the TF work could be done with a closer collaboration with the industry than before, to better investigate the real requirements of a test system. This will give an input of how to expand and develop e.g. RBTS. Distributed generation is one example of an aspect that was proposed to be included in future versions. The overall goal could then be to have a well define updated version of one or two existing test system, easily available at e.g. a web-page to be used by several companies and universities.
15
7 Literature
[1] IEEE Subcommittee report. IEEE Reliability Test System, IEEE Transaction PAS-98 pp2047-2054, 1979. [2] Kwok Sang So, Reliability Evaluation of Distribution Systems, A dissertation submitted to the Victoria University of Manchester for the degree of Master of Science, November 1988. [3] Allan R. N., Billinton R., Sjarief I., Goel L. and So K. S., A Reliability Test System for Educational Purposes Basic Distribution System Data and Results, IEEE Transaction Power System, Vol. 6, No. 2, May 1991, pp. 813-820. [4] Billinton R. and Jonnavithula S., A Test System Teaching Overall Power System Reliability Assessment, IEEE Transaction Power System, Vol. 11, Issue 4, Nov. 1996 pp1670 1676. [5] Billinton R. and Allan R.N., Reliability Evaluation of Power Systems, New York, US (2nd Edition, Plenum) 1996. [6] Billinton R. and A. Sankarakrishnan, Adequacy Assessment of Composite Power Systems with HVDC Links Using Monte Carlo Simulations, IEEE Transaction Power System, Vol. 9, Issue 3, Aug. 1994 [7] Hoyland A. and Rausand M., "System reliability theory - models and statistical methods", Wiley Series 2004. [8] Bertling L., "Reliability-centered maintenance for Electric power distribution systems", Doctoral thesis, Royal Institute of Technology (KTH), School of Electrical Engineering, ISBN 91-7283-345-9, Stockholm 2002. [9] Conference material during IEEE GM San Francisco 14 June 2005 [10] Mousavi Gargari S., Reliability Assessment of Complex Power Systems and the Use of NEPLAN Tool, Master of Science Thesis, Stockholm, Sweden June 2006 [11] Wallnerstrm C. J., A comparative study of reliability assessment models for electrical distribution systems with special focus on the Network Performance Assessment Model, Master of Science Thesis, Stockholm, Sweden 2005 [12] The Swedish Law, 2006, Ellag (1997:857) [13] Johan Setrus, Carl Johan Wallnerstrm and Lina Bertling, A Comparative Study of Regulation Policies for Interruption of Supply of Electrical Distribution Systems in Sweden and UK, CIRED2007, 21-24 may [14] Converter R. Billinton, CIGRE Task Force 38.06.01, Methods to Consider Customer Interruption Costs in Power Systems Analysis, August 2001 [15] CIRED 2007, Tutorial 4, RECENT DEVELOPMENTS IN DISTRIBUTED GENERATION [16] Johan Setrus, Development of a Simulation Module for the Reliability Computer Program RADPOW, Master of Science Thesis, Stockholm, Sweden June 2006
16
LIFE CYCLE SCHEDULING FOR POWER TRANSFORMERS CONSIDERING MAJOR OVERHAULS AND INVESTMENTS
Jussi Palola Helen Electricity Network Ltd jussi.palola@helsinginenergia.fi Deterministic decision-making have been used widely to schedule power transformer major overhauls and new unit investments. Usually it works out fine with a sufficient reliability marginal, but it can be wasteful and sometimes risky. Probabilistic decision analysis can aid to find significant unused potential with more precise decisions concerning transformer investments, major overhaul timing and condition monitoring benefits. Dynamic scenario assessment model presented here is developed to give more practical tools for network asset management in the Finlands capital, Helsinki. 1 DYNAMIC SCENARIO ASSESSMENT FOR THE POWER TRANSFORMERS
Transformer life time concept dealt here is three dimensional: technical, economical and strategic lifetime. The aim is to build dynamic investment planning model which combines technical and economic aspects together in order to make more precise strategic decisions. Base for the transformer investment planning is built on nine factors:
Figure 1. Dynamic Scenario assessment model for the power transformer investments.
Dynamic scenario assessment is build to model robustly over twenty year long investment scenarios. The most important factors are loading prediction, transformer reliability functions, customer analysis and finally above all is expert judgement assessing initial data correctness and model suitability for the specific case - also with intuition. 1.1 Loading Predictions
Future investments are highly related on load growth, especially new substation building. There is various ways to construct a loading prediction, in this report we only look into it briefly: model is based on planned city zoning, consumption characteristics and its development in the future. In this manner loading prediction is made for every existing and planned power transformer in Helsinki till the year 2030. 1.1.1 Future City Planning Helsinki city planning department makes design for the future zoning which makes base for the consumption prediction. Next figure illustrates zoning plan for the next four years: orange colour denotes residential, yellow combined commercial and residential domain and blue commercial premises.
Figure 2 Helsinki city zoning for the years 2006-2010
Specific consumption measures for different customer types are recorded to the customer information system. Next thing is to evaluate changes in the consumption characteristics for the coming years: in Helsinki statistical data clearly shows that cooling demand has and is still increasing. Loading prediction sketch is made with these characteristic measures with future city development plan.
1.1.2 Loading Predictions for Every Power Transformer Outcome predicts load for every power transformer till the year 2030 in the dynamic scenario assessment model. Next figure shows the peak load prediction in the winter and in the summer conditions in one of the twenty substations in Helsinki. Winter peak load prediction is counted to describe loading when daily average temperature is -27 Celsius degrees. Also medium voltage network reserves (explained in paragraph 1.3) and substation loading capability (boxes) is modeled.
Kamppi
120 100 Power [MVA] 80 60 40 20 0 2005 2008 2011 2014 2017 2020 2023 2026 2029 M3 M2 M1 winter peak summer peak reserve capacity
Figure 3 Peak load prediction 2007-2030.
Year 2004 loading measurements are used to make reference curves for every substation loading. This model combines substation reference curve and peak load prediction together and outcome gives default hourly load for every present and planned substation for the next 23 years.
Kamppi substation loading (reference year 2004) 55 50 45 40 35 30 25 20 15 10 5 0 1752 3504 Hour 5256
Power (MVA)
loading
7008
8760
Figure 4 Reference loading curve from the Kamppi substation (2004).
1.2
Reliability model
The latest transformer fault rate research in Finland has been made in the year 2005 by Fortum Service. Transformers were used in various locations from industry to power plants and for the most part in the electricity transmission and distribution. Altogether 413 transformers were taken into account and major overhaul was made to 136 unit after 26,6 years operation time in average. For the most part data consist of 110 kV transformers with only one quarter were from 220 kV and 400 kV voltage levels. The most common transformer power size was 31-60 MVA with 104 unit altogether. /4/
T r a n s fo r m e r fa ilu r e r a te a n d tim e in o p e ra tio n
0 .0 6
w ith o u t m a jo r o ve r h a u l
0 .0 5
a f te r m a jo r o ve r h a u l
Failure Rate (failure/year)
0 .0 4
0 .0 3
0 .0 2
0 .0 1
0 0 10 20 30 40
T im e (y e a r s )
Figure 5
Power transformer failure rate function on the basis of time in operation with and without major overhaul and also the difference between those functions. /4/
First version of exponential fault rate function is developed on the basis of the statistical analysis presented above /6/. Statistical fault rate information is ranging up to 45 years in operation, but the information at the end of the time scale is not so reliable, because only few transformers is used fully after 40 years in operation. Usually aged transformers are for back up and they are used just occasionally so the fault statistics show that fault rate is decreasing which is not the case in real life. Exponential fault rate function:
case adjustable constants and C MO (t ) denotes major overhaul effect.
(t) = a e bt + c C MO (t )
where a,b and c are
One clear thing appeared from the historical diagnostic measurements is that cooling system effect is a vast to the transformer lifetime expectation /1/. Here default division to an OFAF (oil forced air forced) and ONAF (oil natural air forced) cooled transformers is applied. Historically OFAF-cooled transformers have been in the higher temperature and
therefore paper insulation degree of polymerisation (DP) has decreased more rapidly compared to ONAF-cooled transformers /5/. Statistically after 25 years in operation paper insulation DP-value is decreased to 600 inside OFAF-cooled transformer when the same value for ONAF-cooled transformer is 900. Next step is to adjust transformer fault rate function constants case by case with the best information available. In the following case expert judgment is applied on the basis of condition information available.
Table 1 Failure rate function constants for different types of power transformers
(t)
a b c
OFAF - normal 0.01 0.035 0.013
OFAF poor 0.01 0.04 0.015
ONAF - normal 0.01 0.03 0.015
In the figure below statistical information for the fault rate with major overhaul effect is illustrated and on the top of that exponent fault rate function is adjusted. The major overhaul effect is of at the end of the life time: statistical average failure rate is 54% lower in the last 20 years if the major overhaul is done. Major overhaul effect CMO is carefully described here with the constant 0,011 drop to the fault rate function after major overhaul is done.
0.1 0.09
Failure Rate (failure/year)
without major overhaul with major ovarhaul
0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 0 10 20 30 40 50 60
Time (years)
Figure 6 Power transformer fault rate with the fitted exponential function for OFAF-transformer (where major overhaul is done after 23 years in operation).
1.2.1 Condition monitoring effect Reliability can be improved with the on-line condition monitoring in our experiences in Helsinki Energy. Typically transformer loading and temperatures are monitored and in special cases on-line gas analyzer is used to monitor the most critical units. Overload, overheating, cooling system failure indicator and oil level alarm is implemented on every
unit. Uncertainty factors increase when there are no measurement results and monitoring in the transformer: life-time prediction is much more inexact and in this case: probability density function is widely distributed. Life-time prediction can be more accurate with the information received from the condition monitoring: probability density function is narrower with monitoring and it gives possibility use components more efficiently. /6/
Figure 7
The idea of predictive probability density function for transformer lifetime: with condition monitoring and without it. /3 [applied]/
Also problems which come along with the new technique needs to be taken in account also, such as shorter lifetime of integrated electrical circuits and reliability of the remote control information systems. In this model on-line condition monitoring is presumed to prevent 33 % from interruptions when gas-in-oil sensor is installed, so the fault rate functions are following:
Without monitoring 1 (x ) = 0,01 e 0.035 x + 0,013 0.011

With monitoring MON (x ) = 0,01 e 0.035 x + 0,013 0.011 * (1 0,33) .
1.3 Network Properties
Usually there are two power transformers in a substation and size is varying from 30 to 40 MVA. Medium voltage reserves are needed especially when the transformer failure occurs: if the load is greater than remaining transformer overloading capacity, then the MV-network reserves is needed to cover up the load which exceeds remaining transformer capacity. Allowed transformer overloading capability in Helsinki Energy is 20%. In the left figure medium voltage reserves are modelled for Kamppi (Km) substation. Interpretation example: numbers (2+22 MVA) between substations (Km) and (Pv): 2 MVA presents transferable MV-load from (Km) to (Pv) substation and second number 22 MVA, means all the transferable power via MV-network from (Pv) to (Km) substation. Medium voltage reserves are planned to implement in this model in the future, this part is not yet fully operational.
1.4
Customer Criticality Analysis for the Power Transformers
Two factors are needed to define risk: probability and consequences of the failure. Customer criticality analysis is covering up the latter one. There are different demands related to electrical supply reliability from the customers. In Helsinki there are 43 power transformers altogether and newest research results for customer interruption cost is applied to assess outage costs for each unit. Next example figure shows customers types in Kamppi substation area in downtown Helsinki. /2/
/kW (60 minutes) 6,25 25,2 21,6 42,7
Customer type residential public sector industry commerce
Figure 9 Customer types in Kamppi substation area in downtown Helsinki and Finnish median value for 60 minute outage cost per kW.
Commercial sector is naturally located to downtowns and commercial consumers value outage costs almost seven times higher compared to residential consumers. /2/ 1.5 Energy and Material Price Expectations
Transformer is a long-term investment and transformer losses present significant deal in the life-cycle costs. Transformer price and losses are related together: higher capital costs can reduce losses and vice versa. Lifetime cost for the loss kilowatt is usually calculated when transformer is about to be purchased with the current energy price. One way to develop optimisation in this sense is to predict long-term transformer load curve and loss energy price. Here is one example where long-term energy price is predicted to increase 1,5% in a year after 2008. Also long-term price for transformer manufacturing materials can be taken in account (like copper price) if usable information is available.
Long-term Energy Price Prediction for Losses
COPPER PRICE 16 14 Price (FIM / kg) Nominal Price
65 55 45 35 2006 2011 2016 2021 2026
/M W h
12 10 8 6 4 1980/01 y = 0.0122x + 8.072
1985/05
1990/09
1996/01
2001/05
DYNAMIC SCENARIO ASSESSMENT
Current transformer investment program 2007- 2030 in Helsinki Energy is based on deterministic decision-making. Now it is time to implement dynamic scenario assessment model in a practice comparing current investment program to alternative investment program. Current investment program Average transformer renewal age is 42,2 years and current program includes existing 43 transformers and 12 planned units. Major overhaul is planned for almost every transformer after 23,8 years in operation average. Alternative investment program Alternative program is for the same transformers as above, only with the different renewal schedule: average renewal age is 43,8 years and major overhaul timing after 24,2 years in operation. Nine ONAF-cooled transformers are used four years longer in average and four OFAF-transformers are used 2 years less. 2.1 Simulation Result
Overall costs for the current investment program are 55 M with 4 % net present value. Reliability level stays the same (violet line) and operational costs are rising up due to higher energy price expectation (brown line).
Current Investment Program Scenario Investments (Present Net Value 4%) CIC Load Loss Expenses No Load Loss Expenses Total Operational Cost Time Based Maintenance OVERALL COSTS 17 183 850 8 468 714 7 274 677 6 831 818 14 106 495 1 469 481 55 335 035
Figure 11 Simulation results for the current investment program.
Next figure (12) evaluates investment program where online condition monitoring is installed on 23 units. Online monitoring is presumed to prevent 33 % of the outages. Average power transformer renewal age in the following program is 1,6 years higher, that is why investment costs are 2,4 M lower compared to current (figure 11) program.
Alternative Investment Program Scenario Investments (Present Net Value 4%) CIC Load Loss Expenses No Load Loss Expenses Total Operational Cost Condition Monitoring investments Time Based Maintenance Cost OVERALL COSTS 14 774 448 7 677 104 7 274 677 6 999 255 14 273 933 300 000 1 469 481 52 768 899
Figure 12 Simulation results for the alternative investment program with condition monitoring equipment.
Four unit renewals are made earlier and nine later in alternative investment program compared to current one. Still the reliability level is little bit better: about 0,8 M for the years 2007-2030. Condition monitoring investment here is justified in this case because the investments are relevantly lower (approximately 200 k). 2.2 Model Uncertainties
Statistical reliability information is quite narrow for the moment because only 413 power transformers are included. Next important thing would be analyse failure modes specifically for the more realistic model. Power transformer long term loading behaviour is related to general market behaviour so loading predictions needs to be updated from time to time. Energy and material price expectation is very rough, it only tries to follow wider general trend: like energy price is going up due to long term political restrictions and increasing consumption. 2.3 Economical Significance
Robust deterministic rules of thumb leads to success in most cases, but the network assets can be used more case sensitively with whole lifecycle in mind, including quality aspects as well. Long term investment planning tool gives network asset manager a possibility to make strategically pertinent decisions. Sophistically used increased computational capacity, sufficient maintenance programs and developing condition monitoring equipment can decrease network companys overall costs approximately 5 20 % with better quality in service according to my intuition. 3 CONCLUSIONS
Probabilistic methods can give relevant information to investment or maintenance program decision-making. Quality aspects, such as reliability, can be assessed more
case sensitively comparing to deterministic decision-making. Dynamic scenario model includes many uncertainties, but still it is better to analyse possible scenarios beforehand than analyse future economics if price levels stays the same. Model is as good as the basic reliability and economic information behind the model so the following thing is to develop reliability models and price scenarios. REFERENCES 1. Helen Electricity Network Ltd information system data, scada-system measurements, diagnostic analysis results and reliability statistics, internal reports. Antti Silvast, Pirjo Heine, Matti Lehtonen, Kimmo Kivikko, Antti Mkinen, Pertti Jrventausta, Cost of Interruption for the Electricity Network customers in Finland, (in Finnish) Helsinki and Tampere Universities of Technology, December 2005. E. Mytyri, Pulkkinen Urho, K. Simola, Application of Stochastic Filtering for Lifetime Prediction, Technical Research Centre of Finland, Reliability Engineering and System Safety 91, Science Direct, Finland 2003. Takala Otso, Maintenance Effects on 110-400 kV Power Transformers Reliability in Finland, customer survey report, (in Finnish) Fortum 2005. Susa, Dejan; Palola, Jussi; Lehtonen, Matti; Hyvarinen, Markku, Temperature Rises in an OFAF Transformer at OFAN Cooling Mode in Service, IEEE Transactions on Power Delivery, 2005. Vol. 20, nro 4, pp. 2517-2525. Michael Schwan, Klaus Schilling, Uwe Zickler and Armin Schnettler, Component Reliability Prognosis in Asset Management Methods, 9th International Conference on Probabilistic Methods Applied to Power Systems KTH, Stockholm, Sweden June 11-15, 2006
2.
3.
4.
5.
6.
10
Literature Review in Reliability Assessments Incorporating HVDC Links

Johan Setrus KTH School of Electrical Engineering, Stockholm, Sweden johan.setreus@ee.kth.se July 10, 2007
Contents
1 2 Introduction HVDC System Overview 2.1 Terminology . . . . . . . . . . . . . . . 2.2 Motivation for a HVDC system solution 2.3 Drawbacks of the HVDC solution . . . 2.4 Applications for the HVDC solution . . 2.5 Two general types of converters . . . . 2.6 HVDC Congurations . . . . . . . . . 2.7 Components in an HVDC station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 2 2 5 6 8 8 9
Reliability assessment of the HVDC system 12 3.1 Annual reliability performance for existing HVDC stations . . . . . . . . . . 12 3.2 Literature review of reliability assessments of the HVDC technique . . . . . . 13 Closure 17
Introduction
Power transmission utilized with direct current (DC) is not a new idea. The rst commercial generated electricity by Thomas Alva Edison in the 1880s was direct current (DC) at low voltage levels [12]. The rst electricity transmission system was also DC, but because of the low voltage the electric energy had to be generated close to its consumers to avoid too large losses. The alternating current (AC) system by Nikola Tesla made it possible to easy transform the voltage to higher voltage levels, suitable for transmission of electric power over long distances. The generation of power was no longer needed to be close to its customers; instead it could be placed where the source of energy was located. For this reason the winner in the "war of currents" in the late 1880s was the AC technique and it has been the dominating technique for power transmission ever since. The history may have look different if there had been a technique earlier that was able to step up the DC voltage to a higher level, like the transformer does with AC. In the year of 1901 such component made it possible when Hewitts mercury-vapour rectier was presented. This high voltage valve made it possible, at least on the paper, to transmit DC power at high voltage levels and thereby long distances. The high voltage direct current (HVDC) technique was born. The development of the mercury arc valves in the 1930s improved the technique and in 1945 a commercial HVDC system in Berlin was commissioned. This system was never put into operation [25]. In 1954 the rst commercial HVDC transmission was put into operation with a 96 km sea cable, 20 MW, between the Sweden mainland and the island of Gotland. Many mercury-arc-based transmission systems were to follow all around the world. The development of power electronics and semiconductors in the late 1960s led to the thyristor based valve technique, rst tested in the Gotland transmission 1967, and later introduced in larger scale in Canada 1972 with a rating of 320 MW. Today the thyristor based line commutated current source converter (CSC) technique is used in the majority of the HVDC transmissions in the world [29]. During the late 1990s the development of semiconductors for power electronics, such as IGBTs and GTOs, had reach the point where its ratings made it possible to be used for voltage source converters (VSC). The rst commercial VSC based HVDC transmission was rst commissioned in 1999 on the island of Gotland with an underground cable of 50 MW. Recent year of ever increasing development of the semiconductors has improved the VSC technology and the largest commissioned HVDC transmission using this technique is now the Estlink, connecting Estonia to Finland with a rating of 350 MW and 150 kV DC voltage [27]. Development of higher power and voltage ratings (up to 1,100 MW and 300 kV) is going on [26]. The objective of this report is to present the HVDC technique and to investigate what has been done within the reliability domain in this area. First an overview of the technique is presented and then the two distinct technologies within HVDC, CSC (HVDC Classic) and VSC (HVDC Light) are compared. The advantages and disadvantages of HVDC is discussed. This is followed by a component and conguration study of typical HVDC transmission setups. A literature review of what has been done regarding reliability assessments and HVDC is presented. The main question is what has been published in order to investigate how the overall system reliability changes when the system is penetrated with HVDC links.
HVDC System Overview
The high voltage direct current (HVDC) technology is used in the transmission system to transmit electric power over long distances by cable or overhead lines. It is used to connect two alternating current (AC) systems or nodes having the same or different frequency. Figure 1 shows a schematic picture of the basic principle of transferring electric energy from one AC system or node to another, in any direction. The system consists of three blocks; the two converter stations and the DC line. Within each station block there are several components involved in the conversion of AC to DC and vice versa.
HVDC System Converter station 1

AC1
DC line
Converter station 2
AC2
Figure 1: Schematic picture of the overall system perspective of a general HVDC system, transferring electric energy from one AC system or node to the other, in any direction.
There are basically two different techniques used for the converters that transform AC to DC and vice versa in an HVDC system. These are the Current Source Converters (CSC) and the Voltage Source Converters (VSC). An HVDC system adopting the CSC technique is also referred to as Classical HVDC or simply HVDC. An HVDC system with VSC is referred to as HVDC Light (ABB) or sometimes HVDC Plus (SIEMENS).
2.1 Terminology
In order to clarify the terminology in this report an approach to dene these has been made below. A schematic single line diagram with some of these terms is shown in Figure 5. The HVDC system contains the converter stations in the system and the transmission lines or cables. The HVDC converter station or HVDC station includes the AC switchyard, the converter transformer system, the converters and the auxiliary subsystems in the station. The valve bridge includes a number of valves in the valve hall. The valves consists of a number of thyristors or transistors, depending on technique, that are connected in series in order to manage the high voltage. The Valve hall is the building that houses the valve bridges in the converter station. The converter transformer system (CTS) includes the converter transformer arrangement connected to the valves.
2.2 Motivation for a HVDC system solution

The reasons and motivations for choosing the HVDC solution instead of the traditional HVAC are many and are dependent of a number of aspects. The most important of these are presented in the following sections. 2
2.2.1
Active power transmission capability
The power transmission capability in a DC line is higher than in a comparable AC line, assuming that they have the same insulation requirements and peak voltage levels. To show this, assume two transmission links, one operating with alternating current and the other operating with direct current. The AC system consists of two three phase lines, which gives six conductors in total. The DC system consists of three bipolar congurations, later described in Section 2.6.3. This type of HVDC conguration has one positive and one negative DC conductor, which also gives six conductors in total. Both systems are assumed to have the same types of conductors with equal cross section area. The applied symmetric rms phase-tophase voltage for the AC line is Vac and for the DC line Vdc , pole-to-ground voltage. Since the bipolar conguration has a positive (+Vdc ) and negative (Vdc ) conductor, the phases-tophase voltage is 2Vdc . For the three phase AC line the active power that can be transmitted is calculated by [24]: P3ac = 3Vac Iac cos (1)
, where Iac is the phase current and the angle between the current and voltage. The active power transmitted through the two three phase lines is 2P3ac . If the same current is assumed to be transmitted in the AC and DC lines, with a cos = 1 for the AC case, the ratio between the active power for the lines is calculated by: Pdc_total 3 2Vdc Idc = P3ac_total 2 3Vac Iac
(2)
If it is assumed that the DC line can be stressed to the same peak voltage level as for the AC line, Vdc can be formulated as [7]: Vac 2 3
Vdc =
(3)
E.g. from rms phase-to-phase to pole-to-ground. This is a rather conservative assumption according to [7], which states that the conductor can be stressed more than this in the DC case. This means that Equation
2 Idc 3 2 Vac Pdc_total 3 = = 2 1.41 P3ac_total 2 3Vac Iac
(4)
This means that a DC conductor can transmit at least 1.41 more power compared to an AC. However, a number of additional parameters such as the corona1 and skin effects2 makes this ratio even larger. Furthermore, for cables the dielectric losses is much smaller and this makes the current capability larger for DC cables. According to [7] this means that the ratio will be in the range of 1.5 to 2.0 for DC cables.
1 Corona denition: "Self-sustained electrical discharges in which the eld-intensied ionization is localized over only a portion of the distance between the electrodes. This may occur either because of eld nonuniformities such as are present at electrode edges, points or wires, or over nonuniformly conducting surfaces, or because the voltage across one of two or more dielectrics in series between the electrodes does not attain the minimum critical value required to support self-sustained discharges in this dielectric." [5] 2 Skin effect denition: The tendency of alternating current to ow near the surface of a conductor.
2.2.2
Transmission cost evaluation
The AC transmission line tends to be more economical for shorter distances than HVDC. This highly depends on the large startup cost for the converter station at both ends of the HVDC transmission line. But since the line cost per kilometer is less for HVDC, considering both the investment and a capitalization of the total energy losses, the HVDC solution may be more economical for longer distances. This applies especially when large amount of bulk power is going to be transmitted from one area to another. For the AC solution extra costs are added to longer lines at large distances as compensation equipment for reactive power is needed, see Section 2.2.4. Figure 2 illustrates the total costs for the two different techniques as it depends on the transmission distance.
Costs Total HVAC cost
Total HVDC cost
Transmission distance (km) Break-even distance Overhead lines: 400 700 km Cables: 25 50 km
Figure 2: Illustration for how the total cost for the HVAC and HVDC depends on the transmission distance [28]. The break-even distance, where the HVDC solution becomes more benecial than HVAC, highly depends on the land conditions and if a cable is going to be used.
2.2.3
Stability and enhanced control
The active power ow in an HVDC system can be controlled in any direction by its operator. The speed of change in power ow highly depends on the specication of the HVDC system and the ability for the AC systems to withstand a fast change. If the HVDC system is interconnected with nodes in a larger AC system, the HVDC can provide the overall system with enhanced overall system stability [15]. A proper control algorithm and modulation of the active power in link can damp oscillations that may occur after disturbances in the system. Depending on its application and design the HVDC system can assist the AC system to damp power oscillation between interconnected systems after disturbances. The HVDC then act as a "rewall" between the two systems. This can also be useful when the systems contain machines oscillating towards each other by various reasons [15]. 2.2.4 Absence of line compensation
Voltage stability limits is one major constraint when designing traditional AC transmission lines over long distances. This is due to the fact of the signicant inductance such line will have. The effects of the inductance has to be compensated along the AC line. The same 4
problem applies in long cables where the large capacitance has to be dealt with. If the line or cable instead is fed with a DC current the effects of the inductance or capacitance is irrelevant since it is dependent of the current frequency, which in the DC case is zero [4]. The compensation along AC cables to maintain the voltage prole is sometimes hard to implement and this specially applies in undersea cables where such solution is highly unpractical. In this case the HVDC solution is the only economical and practical solution and this is therefore one of the main application for the technique. 2.2.5 Environmental aspects
One of the reasons for choosing the HVDC solution may be its low impact on the surrounding environment. The environmental aspects of an HVDC transmission line can be divided into four categories [15]: Visual impact and space requirements - An HVDC transmission with an over head line requires less space per MW than the traditional AC solution and reduces thereby the visual impact by the towers. If a cable is used the only visual impact is the converter stations. However, the size of these stations in comparison with a traditional AC switchyard may have a larger visual impact that has be dealt with. Electric and magnetic elds - The magnetic eld produced by a DC line is stationary while it in the AC case is alternating which can cause inducing body currents. This results in less restrictions for the magnetic eld in the HVDC line [7]. The electric eld is less severe in the DC line compared to the AC line since there is no steady-state displacement current in the DC case [15]. Radio interference - The harmonics caused by the switching process in the HVDC technique cause disturbances in the kHz and MHz region. An appropriate shielding of the valves minimize this problem and makes the radio interference of an HVDC line lower than with a comparable AC [13] [15]. Radio interference is on the other hand normally a minor problem in the transmission system. Audible noise - The converter stations may cause more complaints by the fact that the audible noise levels normally are higher than comparative AC solutions [15]. However, in general this issue can be dealt with by the use of low-noise components or enclosure of specic components such as e.g. the transformers. 2.2.6 Black start capability
For classical HVDC the receiving end of the link needs to be energized in order to start the transmission of power. This is a major drawback when the transmission system needs to be restored after a blackout. HVDC Light with the voltage source converters, on the other hand, produces the AC voltage and frequency itself and does not need generation at each end to work. It thereby has the advantageous capability to restore areas that lack generation after a blackout.
2.3
Drawbacks of the HVDC solution
There are a number of major drawbacks for the HVDC technique that as to be taken into the consideration when choosing between this solution or the more traditional AC solution.
2.3.1
High cost of converters
The startup cost for a transmission line with the HVDC technique is higher compared to a traditional AC line, due to the converter stations that are needed in the sending and receiving end of the line. For this reason the total cost may be higher than the traditional AC. This is also illustrated in Figure 2. 2.3.2 Harmonics
The switching operation of the converters in the HVDC system adds current and voltage harmonics on the AC and DC side respectively. The harmonics are both multiples of the fundamental AC frequency (characteristic harmonics) and non-characteristic harmonics [15]. These have to be eliminated in order to avoid voltage distortion, overheating and additional power losses generally. At the DC side the harmonics are damped with a DC lter, particularly to avoid interference with telephone system and or other types of communication channels [13]. At the AC side the lters normally consists of a high pass lter, and two or more lower order lters that damp the characteristic frequencies of e.g. the 11th and 13th order. The lter design highly depends on the specic AC system impedance that are present in the node [13]. For HVDC Light the requirement for lters is smaller compared to HVDC Classic. Because of the high switching frequency that are used in the converters in HVDC Light, only high pass lters are required on the AC side [15]. At the DC side a capacitor is needed to smooth the DC voltage. 2.3.3 Reactive power consumption
For HVDC Classic the converters consume a large amount of reactive power from their AC connection. During normal conditions the reactive power requirement is around 50% of the active power being transmitted [15]. For this reason equipment are added to compensate this reactive power demand. This is partly done by the harmonic lters, but generally additional shunt capacitors are needed at the converter stations [15]. For HVDC Light no extra reactive power compensation equipment is required since the converters are able to generate or consume active and reactive power independently. 2.3.4 Complex controlling
An HVDC system needs an adequate and complex control system in order to interact with the connected AC system. How the line behaves in order of failures or fast changes is very much dependent off the digital electronics and control algorithms that have been implemented [15]. 2.3.5 Losses in converter station
During the past decade the energy losses in the converter stations has been reduced with the development in the valve and station technology [28]. For classic HVDC the losses are around 0.6% of the rated HVDC transmission capacity per converter station [26]. For HVDC Light the same value is around 1.6% [26]. The main contribution to these values is the losses in the valves.
2.4 Applications for the HVDC solution

The abilities and properties of the HVDC solution have made it suitable for use in the following four main applications [28].
2.4.1
Underground or underwater cables
The charging current in cables being fed with AC makes transmission of energy with HVAC difcult for long distances. In order to keep the voltage levels and losses within reasonable limits, the HVAC solution requires reactive power compensation equipment along the cable, such as series capacitors. Such equipment both adds cost to the line, and is in some cases not possible to implement. One example is overseas transmission, where it is difcult to install compensation equipment along the cable at the sea bed. This is not needed in the HVDC solution as mentioned in Section 2.2.4. Two examples of this type off application is the Swepol link interconnecting Sweden to Poland with an undersea cable (600 MW) [29], and the Gotland link connecting a wind power park on the south part of the island to the north with HVDC Light and an underground cable (50 MW) [29]. 2.4.2 Long distance bulk power transmission
Bulk power transmission over large distances, which exceeds the breakeven distance, is ideal for the HVDC transmission solution. As the worlds energy resources normally are decentralized from the ever increasing energy consumption, this HVDC application is specially interesting area for the future. Higher power transmission in fewer lines and over long distances is possible with the HVDC transmission than with the traditional AC solution [14]. Two examples of this type of HVDC transmission is the Itaipu in Brazil (10258 MW) [29] and the Three Gorges in China (6000 MW) [29], both connecting large hydropower resources to populated areas [26]. 2.4.3 Asynchronous interconnection
The HVDC technique can be used to connect two asynchronous power systems with the same or different frequency for more economic or reliable operation. The interconnection is benecial for both systems and act as a buffer between them. In case of cascading failures in one of the systems, the interconnection will serve as a "rewall" between the systems, preventing the propagation of outages to continue into the connected system [14]. If the HVDC system is built only for the purpose of its ability to perform asynchronous interconnection, the set up is referred to a back-to-back conguration. One example of this application is the four main asynchronous AC systems in North America, being interconnected to each other with twelve back-to-back links [28]. 2.4.4 Stabilization in a larger power system
HVDC transmission can be used within a large interconnected AC system to improve the control of power ow from one part in the system to another and thereby avoid large cascading failures or even blackouts. The stability can also be improved, since the link provides a damping torque, as described in Section 2.2.3. The introduction of this sort of application in the existing AC system requires a great knowledge in the existing AC system in order to determine the right location and specications how the control algorithm within the HVDC is constructed. One example of this type of interconnected HVDC system in a larger synchronous AC system is the Chandrapur-Padghe link (1500 MW) [29] in India. This link has been built to stabilize the system and to increase the power ow capabilities in the particular part of the system.
2.5 Two general types of converters

The HVDC converter station converts the electrical energy from AC to DC or vice versa. This conversion can be performed by one of the two following fundamental types of converters: Current Source Converter (CSC) or the Voltage Source Converter (VSC). The CSC converter is implemented in classical HVDC and the VSC in HVDC Light. The CSC converter was rst utilized with the mercury-arc valves from the early 1950s to the mid 1970s [28]. According to reference [29] there is still six installations in the world that are up and running with this technique. With the development of semiconductors and especially the thyristors in the late 1960s this component could be used for the conversion in CSC converters [28]. Today this is the technique used in the most HVDC installations in the world. During the late 1990s the further development of the semiconductors, such as IGBTs (transistor) and GTOs (thyristor), had reach the point where its ratings made it possible to be used for voltage source converters (VSC) [28]. There are currently ve commercial installations commissioned in the world and this technique is not as proven and mature as HVDC classical [29]. 2.5.1 Comparison of the two types
The general advantages and drawbacks for the HVDC technique in general has already been covered in Section 2.2 and 2.3. In these previous sections some of the differences between classical HVDC and HVDC Light have been covered. A brief comparison of the two converter types is made below. The ratings for HVDC Light installations are not as high as for classical HVDC, but it has a number of advantages and features that makes it interesting. These and other properties are summarized in the following list [27] [28] [15]: The active and reactive power consumption or generation can be controlled independent of each other. No reactive compensation equipment is needed. It is possible, but not common, to eliminate the converter transformer system. The link can be used to start up weak AC networks, after e.g. larger blackouts that lack generation. It does not have a minimum power level that can be transferred; the active power can be controlled between 0 and 100 %. It requires only small AC lters for harmonics elimination.
2.6 HVDC Congurations

Depending on the application and location a number of different HVDC station congurations can be identied. In this report these congurations are presented for classical HVDC with CSC converters, but similar types exists for HVDC Light. For HVDC Light the inclusion of the transformer component is not obligatory, but normally used in order to optimize the performance of the installation. [27]. Figure 3 shows the single line symbols used in the HVDC congurations.
Valve Bridge CSC
Valve Bridge VSC
Converter Transformer
.
Figure 3: Single line symbols for the converter types and converter transformer for the representation of the HVDC system.
2.6.1
Back-to-back conguration
Figure 4(a) shows a block diagram of the back-to-back conguration with a 12-pulse conguration. In this conguration the HVDC stations are located at the same site, with no transmission line between, interconnecting two AC systems. The frequency may be the same or different and the AC system asynchronous. 2.6.2 Monopolar conguration
A monopolar conguration with a 12-pulse conguration is shown in Figure 4(b). The conguration is mainly used in undersea transmissions [27]. This consist of two HVDC stations connected with an single conductor line having a positive or negative polarity. The return current is either returned via the ground or a small metallic conductor. 2.6.3 Bipolar conguration
The bipolar conguration can be seen as two monopolar congurations in parallel, with a positive and negative pole. Figure 4(c) shows a block diagram of a 12-pulse bipolar conguration. The bipolar is commonly used in bulk power transmission lines with overhead lines [27]. If one pole is suffering a failure, it is still possible to run the HVDC system in a derated state with one pole and with a earth return current. If the cable at the failed poled is still intact, this can be used as a metallic current return in order to minimize the losses.
2.7
Components in an HVDC station
The setup in the two converter stations are normally more or less exactly the same, with some possible differences related to the properties of the connected AC system. Figure 5 shows a typical block diagram, including the main components, of a monopolar HVDC converter station with a 12-pulse conguration. The rst block summarizes the components in the AC switchyard which includes circuit breakers, lters and in some cases equipment for reactive power compensation depending on the converter technique. The second block, the converter transformer, is normally implemented with either 1 three-phase or 3 single-phase transformers. The name converter transformer arises from the slightly different design and requirements these types of transformer has compared to a conventional transformer. It serves as a galvanic isolator between the AC and the DC side and transforms the voltage to an appropriate and optimum level for the converters that converts AC to DC. The DC is then ltered before entering the cable or line depending on application.
(a)
AC1
AC2
(b)
AC1
AC2
(c)
AC1
AC2
.
Figure 4: Three basic HVDC congurations with different arrangement of the converter: (a) Back-toback conguration; (b) Monopolar conguration and (c) Bipolar conguration.
10
Figure 5: Single line diagram illustrating the main components in a monopolar HVDC converter station.
11
Reliability assessment of the HVDC system
The actual need of a reliability assessment of an HVDC system is an essential question one may ask before starting an assessment of a system. What can be done in such assessment and why is it necessary? As an approach to answer this question one can conclude that failures of components and systems always will occur randomly, and as a part of this the failures or outage durations will vary from one year to another. In order to handle or learn more about these random events most utilities or companies record details of these events and these data can then be processed to produce a set of reliability measures for the system. A typical measure may be the number of incidents or hours of interruption for a certain component, load point or entire system. Different standards, which specify how these measures should be performed and collected, may be used in order to compare the well being of a specic system to other assessed systems. This property is one valuable reason for the need of a reliability assessment in order to assess the availability of the system. Other valuable properties may be e.g. to (i) identify weak areas in the system in order to perform modications, to (ii) compare different preventive maintenance plans and their responses to the system reliability, to (iii) ensure that the available resources are used in the most optimum way. This list of valuable reasons for performing reliability assessments can be more extensive depending on the area of interest, but one can conclude from these examples of application that there is a need of these quantitative reliability measures or indices. The reliability measures that mentioned above, are statistical indices assessed from historical data of the performance for the station. The same basic principles for the reliability indices are used if the future behavior of the system is going to be predicted with probabilistic models and techniques. It is important to distinguish between these two reliability measures as they arise either from a statistical or probabilistic assessment. The rst part of this chapter, Section 3.1, includes a literature review of the sources and organizations that collects the annual statistical performance of the HVDC installations in the world. This section is followed by a literature review of what has been done in the probabilistic approach of modeling the HVDC system separately, and, imbedded in a composite AC system.
3.1 Annual reliability performance for existing HVDC stations

3.1.1 Guide for the evaluation of reliability IEEE Standard 1240-2000 [10] is a guide for the evaluation of the reliability of HVDC converter stations. The purpose of this guide by IEEE is to help the improvement of the reliability of HVDC converter stations by promoting basics concepts of reliability, availability and maintainability (RAM) in all phases of the stations life cycle. A number of denitions related to HVDC reliability is dened in the document. The guide also includes the CIGR document "Protocol for Reporting the Operational Performance of HVDC Transmission Systems" [6] which species how the reliability recording procedure of HVDC stations should be performed. 3.1.2 HVDC projects listing
A list of the commissioned HVDC projects in the world is provided by the IEEE Transmission and Distribution committee [29]. The list includes commercial HVDC stations in use, retired from serviced, planned and under construction, with its rating, location and year commissioned. Table 1 summarizes the list from IEEE. The list also reveals that the absolute majority of the stations are of the type HVDC classic. HVDC Light stands for ve of the
12
commissioned stations. Of the projects that are planned or under construction only one out of 57 is HVDC Light.
Table 1: Summary of the IEEE Transmission and Distribution committees list of the number of HVDC installations in the world [29].
Commissioned 117
Under construction 6
Planned 51
3.1.3
Availability recordings of HVDC stations in world
The CIGR working group B4.04 collects reliability recordings from a number of HVDC systems in the world. The objective of the sharing of data is that it could be useful in the planning, construction and operation of new and existing projects [10]. The rst data were collected in 1968, including four installations. In the year of 2001 32 HVDC systems reported their reliability data to the group. The data includes general information regarding the station such as a system description, main circuit data and a simplied single line diagram. This information is reported and summarized in a Compendium available from the working group. Operational performance data is collected annually from each installation and summarized in a report every second year with the title "A Survey of the Reliability of HVDC Systems Throughout the World" [9]. How the data should be recorded by the stations operators is specied in the document "Protocol for Reporting the Operational Performance of HVDC Transmission Systems" [10] [6]. This standardize procedure makes it possible to compare historical accumulated data from several systems in order to judge the performance of a specic systems. The data includes a number of reliability measures for the HVDC system such as e.g. the energy availability and energy utilization. It also includes forced outage data for six different component categories within the system, with the number of failures and equivalent outage time in hours.
3.2 Literature review of reliability assessments of the HVDC technique

The following section consists of a literature review of papers within the area of reliability assessments of HVDC. The main part of these handles the reliability evaluation of the HVDC station. Papers evaluating the overall system perspective of introducing HVDC in a composite AC system are not so many but these have on the other hand been analyzed more in detail. 3.2.1 Reliability assessment of the HVDC station
Several papers have been published in the area of evaluating the reliability of the HVDC station as a single system. In all the studied papers the single line diagram of an existing or test HVDC installation is transferred to a reliability block diagram with a number of key components. The single line diagrams are similar to the one presented earlier in Figure 5. An example of the reliability block diagrams presented in the papers are illustrated in Figure 6. Each block is associated with a failure rate and a repair time. The block are in some cases modeled further with e.g. a spare parts or redundancy in some of the components. The different states of these components are illustrated with state space diagrams. Modeling of bipolar HVDC In the studied papers references [8, 17, 22, 23] have been modeling the bipolar HVDC system. The reliability models in these four papers are similar to each other and can be generally 13
AC1
HVDC System
Busbars+ switchgear Compensation Filters Converter Transformer Converter DC line Converter Transformer Converter DC line
Symmetrical
Figure 6: An example of a reliability model for an HVDC system.
illustrated as the block diagram model in Figure 6. The difference between them is the complexity in the modeling of the components. In [8] there is e.g. a spare converter transformer available and in [17] the lter banks are modeled in several capacity states. Reference [8] gives a good overview of the algorithm used in the reliability assessment and describes the adapted theory in detail. Modeling of general HVDC systems with an arbitrary number of poles More general papers for the reliability assessment of HVDC system is presented in references [1821]. These models are not restricted to the bipolar HVDC arrangement. An arbitrary value of N different poles (multipolar) can be included and evaluated by these models. An example of multipolar arrangements is when two or more bipolar congurations are used in long distance bulk power transmissions. References [1820] includes the operating states Healty, Marginal and At risk to determine the reliability of the station. Reference [16] introduce a reliability model for assessments of multiterminal HVDC systems. 3.2.2 Reliability assessment of composite power systems containing HVDC system
Three different papers covering the overall system perspective of HVDC systems embedded in AC systems has been studied. For the modeling of the HVDC stations the papers uses models described in the HVDC station models in Section 3.2.1.
14
Incorporation of a DC link in a composite system adequacy assessment - composite system analysis, R. Billinton and D.S. Ahluwalia [2] This paper describes a reliability analysis of the IEEE-RTS test system, modied with an HVDC link included to the system. The link, transferring energy from a remote generation source, replaces a generation unit in one of the buses in the system. The reliability model for the HVDC link, representing the EEL river bipolar system (320 MW), is described in reference [3]. Each component in this system model has an assumed failure rate and repair time. The output results is a table with the probability for each capacity state in MW, the average failure frequency per year and the mean duration in hours. This HVDC model is then included in the RTS model and a comparison between the basic case for the RTS system and the modied is then performed. A second case, were the capacity in all the HVDC links components is doubled (640 MW), is also evaluated. The evaluation of the entire system reliability is performed by the program COMREL which uses the enumeration approach with a simple network ow method. The results from the modied system shows on a reduction for the systems reliability. The results is expected as the connected HVDC system replaces a generation unit with less components, failure rate and outage time. The authors comment the results as follows in reference [1]: "The reduction in reliability is not a reection on DC systems but on the fact that the generating station at bus #22 in the original system is now considered to be at a remote location and requires lengthy transmission to connect it to the system." Adequacy Assessment of Composite Power Systems With HVDC Links Using Monte Carlo Simulation, R. Billinton and A. Sankarakrishnan [1] This paper is a development of the paper described above. In addition to the earlier inclusion of the remotely connected HVDC link in IEEE-RTS a second case is also investigated. In this case a double circuit AC line is replaced with an equivalent rated bipolar HVDC line. The reliability computer program MECORE, based on the Monte Carlo simulation technique, is used in the evaluation of the reliability indices. A DC ow algorithm is used for the composite system evaluation at contingencies. The results shows that the replacement of the particular double circuit AC transmission line does not have a signicant effect on the overall system reliability. Only a small reduction in reliability is noted. The authors states that the impact could be different for other systems having different size. The quality part of this paper is the discussion from the peer review person at the end of the paper. J.R. Ochoa from ABB has made several comments on the results which the authors of the paper tries to answer (the text has been shortened and modied but the general context is the same): Question: Modeling and evaluating the behavior of the HVDC Link with a DC load ow is highly inaccurate even in steady-state analysis. The Link has e.g. lter banks at both sides whose behavior cannot be appropriate modeled using a simple DC load ow. The same applies for low voltage conditions at either ends of the link. Why did not the authors used a more accurate AC load ow evaluation method, even if the speed of execution becomes longer? Answer: The HVDC Link model has been evaluated separately and is not in the composite analysis of the entire system. In this case an AC/DC load ow approach could be used in the Monte Carlo simulation, but it would signicantly slow down the process with its 10000 samples. The authors of the paper does not expect that the result would be signicantly different with a AC load ow in the evaluation of this system. Question: What is the different between the AC line model and the HVDC model? Answer: No direct answer by the authors. 15
Question: The result shows that the IEEE-RTS is less reliable with the HVDC link than with the AC double line.These results are surprising given the general experience with the operation of HVDC links. Answer: The authors answer that the objective was not to compare the reliability for AC and HVDC in composite systems. The aim was mainly to demonstrate a technique to perform reliability analysis in a AC-HVDC composite system. The discussion above is interesting in two perspectives: (i) Would the results be different if a AC load ow technique were used in the Monte Carlo simulation? (ii) If a comparison between AC and HVDC is going to be performed how does one design these in order to get a fair assessment? Transmission system reliability evaluation incorporating HVDC links and FACTS devices, M. Fotuhi-Firuzabad, R. Billinton and S.O Faried [11] This paper describes a technique to perform AC/DC system reliability analysis in composite system where an HVDC link and a thyristor controlled series capacitor (TCSC) is present. The TCSC is a device that can be connected in parallel to an AC line in order to adjust the transmission infeed impedance and thereby increase the transmission system capacity. The TCSC is a device that is within the concepts of Flexible AC Transmission Systems (FACTS). The studied transmission system model consists of two small areas A and B, both having load points and generation units. In one of the cases in the paper a bipolar HVDC and a TCSC is connected in parallel between the two areas.A reliability block diagram model for both the HVDC link and the TCSC is presented and evaluated. The results from each model are the probability for the different capacity performance level in %. When the two models are combined the performance level in % for the parallel system is evaluated. The reliability system indices are then evaluated with a contingency enumeration technique for the entire system. The intention of the paper is not to evaluate the benets of replacing an AC line with an HVDC, but to illustrate a technique to perform reliability analysis.
16
Closure
This report has presented an overview of the HVDC transmission technique. The advantages and drawbacks have been discussed and a comparison between the two main techniques HVDC Classic and HVDC Light has been presented. The main congurations and components in a basic HVDC system has been shown. A literature review in reliability assessments of HVDC systems has been made. One of the conclusion from this study is that most of the published papers in this area includes the reliability assessment and evaluation of the HVDC station reliability between the sending a receiving node. Very few assessments have been performed in the area of HVDC systems incorporated in larger composite AC systems in order to evaluate its impact on the entire system reliability. When such assessments has been published and performed these have a number of weaknesses: (i) the presentation of the differences of the HVDC model compared to the AC model is hard to follow. This is necessary in order to evaluate how much each part of the models contributes to the overall system reliability. (ii) The reliability data for the components in the HVDC model is normally just assumed and not realistic. (iii) The reliability techniques for the evaluation of the entire system reliability does not include AC load ow techniques that may reveal different result than simpler techniques used in the studied papers. Today the computer resources are more sufcient for the inclusion of such techniques in the reliability calculations. Maybe this could be a breakthrough in the assessment on how the overall system reliability changes when the system is penetrated with HVDC links?
17
References
[1] R. Billinton, A. Sankarakrishnan. Adequacy Assessment of Composite Power Systems With HVDC Links Using Monte Carlo Simulation. IEEE Transactions on Power Systems, 1994. [2] R. Billinton, D.S. Ahluwalia. Incorporation of a DC link in a composite system adequacy assessment - composite system analysis. IEE Proceedings-C, May 1992. [3] R. Billinton, D.S. Ahluwalia. Incorporation of a DC link in a composite system adequacy assessment - DC system modelling. IEE Proceedings-C, May 1992. [4] A. Von Meier. Electric Power Systems: A Conceptual Introduction. IEEE computer society press, 2006. ISBN 0471178594. [5] AIEE Working Group on Denitions. Denition of Terms Related to Corona. IEEE Transactions on Power Apparatus and Systems, 1963. [6] CIGR Study Committee 14-DC Links, #14-97 (WG 04-21). Protocol for Reporting the Operational Performance of HVDC Transmission Systems. Technical report, CIGR, 1997. [7] P. Fischer de Toledo. Feasibility of HVDC for City Infeed. Thesis for the degree of licentiate, Department of Electrical Engineering, KTH, Stockholm, Sweden, 2003. TRITAETS-2003-11. [8] G. Desrochers, S. Lefebire, M. Blanchard, B. Rioux. A Method for Reliability Evaluation of D.C. Links. CIGRE Report, 14(05) 1988 session, 1988. [9] I. Vancers, D.J. Christofersen, A. Leirbukt, M.G. Bennett. A Survey of the Reliability of HVDC Systems Throughout the World During 2001-2002. CIGRE Working Group B4.04, 2004. [10] IEEE Working Group I7, Reliability of HVDC Converter Stations, in the High Voltage Power Electronics Stations Subcommittee for the IEEE-PES Substations Committee. IEEE Standard 1240-2000, IEEE Guide for the evaluation of the Reliability of HVDC Converter Stations. Technical report, IEEE, 2000. ISBN 0-7381-2509-9. [11] M. Fotuhi-Firuzabad, R. Billinton, S.O Faried. Transmission system reliability evaluation incorporating HVDC links and FACTS devices. Power Engineering Society Summer Meeting, 2001. IEEE, 2001. [12] R. Moran. Executioners current Thomas Edison, George Westinghouse, and the invention of the electric chair, Chapter: The Battle of the Currents. New York : Vintage Books, 2003. ISBN 037572446X. [13] N. Mohan, T.M. Undeland, W.P. Robbins. Power Electronics. John Willey and Sons, Inc, 2003. ISBN 0-471-42908-2. [14] P. Pourbeik, M. Bahrman, E. John, W. Wong. Modern countermeasurus to blackouts. Power and Energy Magazine, IEEE, 2006. [15] J. Paulinder. Operation and Control of HVDC links embedded in AC systems. Thesis for the degree of licentiate, Chalmers University of Technology, Gteborg Sweden, 2003. [16] R. Billinton, M. Fotuhi-Firuzabad, S.O. Faried. Reliability evaluation of hybrid multiterminal HVDC subtransmission systems. IEE Proceedings-Generation Transmission Distribution, 149(5), September 2002. 18
[17] R. Billinton, M. Fotuhi-Firuzabad, S.O. Faried, S. Aboreshaid. Composite system reliability evaluation incorporating an HVDC link and a static synchronous series compensator. Proceedings of the 2002 IEEE Canadian Conference on Electrical & computer Engineering, 2002. [18] R. Billinton, S. Aboreshaid, M. Fotuhi-Firuzabad. Well-Being Analysis for HVDC Transmission Systems. IEEE Transactions on Power Systems, 1996. [19] R. Billinton, S. Aboreshaid, M. Fotuhi-Firuzabad. Diagnosing the Health of Bulk Generation and HVDC Transmission Systems. IEEE Transactions on Power Systems, November 1997. [20] R. Billinton, S. Aboreshaid, M. Fotuhi-Firuzabad. A hybrid deterministic / probabilistic technique for HVDC system reliability evaluation. Proceedings of the 1997 IEEE Conference on Communications, Power and Computing WESCANEXPII, 1997. [21] R. Billinton, S. Aboreshaid, M. Fotuhi-Firuzabad. An approach for reliability evaluation of hvdc transmission systems using a well-being framework. Energy Management and Power Delivery, 1998. Proceedings of EMPD 98., 1998. [22] S. Kumganty. Effect of HVDC Component Enhancement on the Overall System Reliability Performance. IEEE Transactions on Power Delivery, 9(1), January 1994. [23] S. Kumganty. Hvdc transmission system models for power system rellablllty evaluation. IEEE Wescanex 95 Proceedings, 1995. [24] L. Sder. Statisk Analys av Elsystem. Institutionen fr elkraftteknik, KTH, Stockholm, Sweden, 2001. [25] SIEMENS. High voltage direct current transmission - proven technology for power exchange. Brochure from SIEMENS, Source http://www.siemens.com, Mars 2007. [26] The ABB Group. ABB Website. http://www.abb.com (last visit July 2007), Mars 2007. [27] V.G. Agelidis, G.D. Demetriades, N. Flourentzou. Recent advances in high-voltage direct current power transmission systems. Proceedings of National Power Electronics Conference of China, Xian, September 23-26 2006. [28] V.K. Sood. HVDC and Facts Controllers. Kluwer Acaddemic Publisher, 2004. ISBN 1-4020-7890-0. [29] Working Group on HVDC and FACTS Bibliography and Records. HVDC Projects Listing. IEEE Transmission and Distribution Committee, 2006. DC and Flexible AC Transmission Subcommittee.
19
Reliability Engineering Methods for Distribution Systems

Influence of differences between cable system and overhead lines
Anna Guldbrand Lund University anna.guldbrand@iea.lth.se
Abstract
One way to improve distribution network reliability is to replace traditional overhead lines with underground cables. Several Swedish distribution companies plan to, during a near future, replace large amount of rural overhead lines by underground cable. The main purpose of this report is to study how extensive replacement of overhead lines by underground cable influences effective use of failure statistics. Does cable system failure characteristics justify the financial means required for a more systemised reliability work? In addition the report describes how the replacement might influence efficient use off different condition assessment methods. To fully utilize the investments required to replace the overhead lines by distribution cables, network owners will have to adjust their reliability engineering methods to suit the new cable networks. In this report different condition assessment methods as well as improved failure statistics for cable systems are considered. The report is based on information from project reports and scientific papers. In addition a number of Swedish distribution network owners have been asked to share their opinion of reliability engineering work of today and the future. The methods used for distribution system condition assessment change as overhead lines are replaced by underground cable. Visual inspections can no longer be used and diagnostic methods are instead introduced. The diagnostic methods are costly to perform. To become frequently used they must prove to be efficient enough to justify the financial means required. New, improved, diagnostic methods are under development and since most cable failures are related to component properties the use of diagnostic methods is likely to increase. Statistics show that a majority of the cable system failures are ageing failures. The ageing failures do not to any large extent depend on environmental factors but on component properties. This implies that component lifetime standard deviation decreases as component data are related to the statistics, yet failure statistics is at present not related to component data. A majority of the network owners contributing to this report agree that the nature of cable failures, except excavation failures, makes it interesting and useful to related cable data to failure statistics and to share the statistics with other network owners. Several Swedish distribution network owners are in the process of installing new program for network analysis. One challenge is to decide which component data that shall be related to the failure statistics. Operational age, maintenance history, manufacturer and year of manufacturer are four manageable factors which influence on failure statistics is already established but yet not thoroughly evaluated. In addition the method of cable excavation is suggested as a fifth interesting factor related to cable system reliability.
1 Introduction
1.1 Increasing importance of reliability
The importance of distribution system reliability is increasing. During the last ten years the increase in cost of power interruptions in the Nordic countries, has exceeded the increase of consumer price index [1]. In addition, the storm Gudrun experience made the Swedish parliament, in December 2005, to pass a law regulating costumer compensation at power outages. The law entitles consumers compensations up to 300 % of their estimated yearly network tariff. One way to increase the distribution network reliability is to replace traditional overhead lines with underground cables. Several Swedish distribution network owners plan to, during a near future, replace large amounts of rural overhead lines by underground cables. Urban distribution networks do already, to a large extent, consist of underground cable. The cablification will cost the distribution owners, and in the end the consumers, a significant amount of money. To make the most of the investments, the network owners will have to adjust their reliability engineering methods to suit the new cable networks.
1.2 Outline of the report

In this report the methods of reliability engineering in network with an extensive amount of cable is compared to the methods of reliability engineering in traditional overhead line networks. Actual networks differences and the reliability methods used at present are described. Further, possible future differences in the reliability engineering methods and effective use of statistics are discussed. The key questions are: How does extensive replacement of overhead lines by underground cable influences effective use of failure statistics? Does cable system failure characteristics justify the financial means required for a more systemised reliability work? How does the replacement of overhead lines with underground cable influence effective use off different condition assessment methods? The report is based on information from project reports and scientific papers. In addition a number of Swedish distribution network owners have been asked to share their opinion of reliability engineering work of today and the future. In this report the questions are sometimes referred to as the survey. The distribution companies where all financing an Elforsk project studying consequences to the distribution network when replacing overhead lines by underground cable [2]. They are therefore considered to have interest in work related to cable networks. The questions are given in Appendix A.
1.3 Reliability engineering

Generally, distribution network owners recognise the importance of reliable power supply to the consumers. The opinions on how systemised, or academic, the distribution system reliability work shall be are however divided. One representative of a distribution company contributing to this thesis expressed his view on registration and storage of failure statistics as something like I do not think one shall care to much about data storage. Problems concerning bad components will be recognized anyway. This is not an expression of lack of interest in the reliability of the power supply, but an expression of scepticism towards the academic reliability methods demanding much information and computer work.
This report focuses on how the introduction of more cable in distribution networks might increase the incentive of more systemised reliability work. An additional advantage connected to the introduction of systemised reliability work, not further mentioned in the report, is the possibility to transfer experience from ageing personnel. A large part of the distribution network workforce is getting close to retirement and the companies risk losing valuable experience when senior staff retires.
2 Background
Reliability is often related to in terms of failure rate and outage time. Cigr defines reliability in the context of power system as A general concept encompassing all the measures of the ability to deliver electricity to all points of the utilization within acceptable standards and in the amount desired [3]. Individual condition monitoring of all equipment installed in distribution systems would be time consuming and costly. Effective statistical methods is therefore of high importance for distribution system reliability.
2.1 Influence of component reliability on system reliability

The general power item reliability concept, i.e. Availability, can bee seen as a combination of three factors; Reliability of a piece of equipment or a part of the system, Maintainability, which is the possibility to detect failures and to reach and restore the components and the Maintenance support or Supportability, i.e. spare parts, maintenance equipment and the ability of the maintenance staff. The availability concept and parameters of importance are illustrated in Figure 1 [4][5]. All three areas are affected when underground cables replace overhead lines. Mounting Preventive Maintenance Manufacturing Reliability Construction
Fault Detection
Fault Localization Manageability Accessibility
Maintainability
Availability
Maintenance Instruments/ support Equipment Personnel Documentation Spare supply

Figure 1, The availability is a combination of reliability, maintainability and maintenance support, all three governed by different parameters
80 % of the failures in distribution network are related to the electrical components, i.e. overhead lines, cable systems, secondary substations or medium voltage switchgear stations [6]. These components are made up of different parts of which all have a probability to fail. Cable system faults are not only faults on the cables but also on joints and terminations. In addition to the condition of individual components, network topology and environmental factors influence the ability of the system to performance a required function.
2.2 The reliability engineering process

One approach to the reliability engineering is to divide the process into four basic steps; past system behaviour, reliability calculation methods, calculation of reliability indices and prognosis of future system [3]. It is mainly the activities in step one, the collecting of data in order to create models of outages and failures, that differ between networks with an extensive amount of cable and traditional overhead line networks. The failure rates of different components, calculated in step one, are used in the subsequent steps of the engineering process.
3 Data collection and the idea about statistics

3.1 Reliability simulation programs
Modern asset management methods require information about component reliability. Most methods, such as reliability centred maintenance, RCM, depend on reliability simulation programs. The input variables required by the programs are general statistic quantities as failure rate and outage time [7][8]. To be able to provide the input variables users must have some knowledge about the reliability of the individual components in the distribution system. The programs are constructed to calculate load point reliability and system reliability indices. Standardised system reliability indices such as SAIDI, System Average Interruption Duration Index, SAIFI, System Average Interruption Frequency Index, and CAIDI, Customer Average Interruption Duration Index, are comparable for all type of network structures. More than one of the companies taking part in the survey for this report, are in the process of installing new computer program for network analysis. They emphasise the importance of computer programs that are easy to use and can handle the large amount of information in an efficient way.
3.2 Specific component statistics

Access to reliable statistics is a key factor for distribution network performance. One area that is emphasized in the Nordic OPAL project is the importance of relating failure statistics to network components [1]. The failure rate and outage time of different components varies considerably depending on factors like operational age and manufacturer. In reliability simulation programs failure rate is coupled to a specific piece of equipment. Provided that the owners have the necessary information about the network components, it will be possible to simulate the reliability using statistics that is specific for their network components. It might however be difficult to find enough data to generate reliably detailed statistics with several factors involved. One additional reason to relate component data to failure statistics is to be able to identify problematic components. To be able to calculate the failure rate of a specific manufacturer or age of the components it is also necessary to keep record of healthy network components since a relative measure as the failure rate is based on a total number of components.
3.3 Present report of failure statistics

Not all distribution network owners believe in the use of detailed failure registration in order to perform reliability analysis. National failure statistics has been questioned since they show the national average failure rate. Different environmental factors such as climate and landscape are not discernible in the statistics. There are however two registration system to which most Swedish distribution companies at present report failure statistics. The EMI, 4
Energy Market Inspectorate, is part of the Swedish Energy Agency. EMI demands annual reports from all Swedish local and regional network owners. The reports include some failure statistics. The statistics are not for individual components but refers to an entire division. The statistics includes interruptions with duration of at least three minutes. Interruption frequency and average interruption time due to failures in the own network is reported as well as interruption frequency and average interruption time due to failures in overlaying networks [9]. Since the statistics does not refer to components or cause of failure but to the systems, it is not very useful for simulating reliability in computer programs. It seems rather to be an instrument to compare the performance of different network owners. The Swedish power distribution trade association Svensk Energi (Swedenenergy) keeps a more detailed record of failure statistics of their members. The statistic is compiled in a system called DARWin. 112 of 180 Swedish distribution network companies reported failure statistics to DARWin during 2005. These companies deliver power to 91 % of the Swedish power consumers [10]. The statistics are detailed in terms of cause of failure, duration and number of costumers affected. The category of the faulted component, i.e. cable, transformer or overhead line, is reported. Any further information about the component, as age or manufacturer, is however not included in the report. Some of the distribution companies answering the survey for this report keep their data registration and storage exactly according to what EMI and Swedenenergy demand. A few of the companies register the faulted component if the component is considered to be the main cause of the failure. They do however not yet use this information to differentiate any failure rates in reliability analyses.
4 Differences of cable and overhead lines failures

Component failures can be characterized as teething failures, random failures or ageing failures relating to the well-known bathtub curve. Teething failures are normally found by onsite test and handled when the components are first connected to the system [6]. Random failures are caused by external factors such as weather conditions or excavator work. The random failure rate is constant over time. Ageing failures are caused by electrical, thermal, mechanical and environmental stress. The ageing failure rate increases over time. One major difference between overhead lines and underground cables is the insulation material. Air insulates overhead lines from earth. Since the insulation is not a part of the overhead lines the conductors are exposed to environmental factors. Modern underground XLPE cables are insulated by polyethylene, while older cables are paper and oil insulated. The ageing of the insulation material in underground cables is a major contributing factor to cable system failures. Degradation of the insulation material often results in partial discharges, PD. If measures are not taken to prevent failure, the partial discharges will eventually cause insulation breakdown.
4.1 Fault Localization

The fault localization time depends on a number of factors. Among these are the localizing method, level of distribution automation, size of the network and whether it is a cable system or an overhead line system. Visible components generally make the localization of the fault easier and faster. Most cable failures, except excavation work, are not visible. The fault location time is a major reason the outage time is much longer in overhead network than in cable systems [11]. In addition, after the fault has been localized, it is complicated to reach the fault and replace or repair the cables. This is particularly critical during the winter when the ground is frozen.
5 Condition assessment
5.1 From inspections to diagnostic methods
Individual condition monitoring of every piece of equipment installed in distribution systems would, as mentioned above, be time consuming and costly. Condition based preventive maintenance will however become increasingly important as the distribution network reliability demands increase. Condition assessment covers both diagnostic methods and inspection. Diagnostic methods, involving measurement equipment, are referred to as objective assessment while inspections depends on the experience of the inspector and are referred to as subjective assessment [4]. As the amount of cable increases in the distribution networks, the objective assessment is likely to become increasingly popular, compared to the subjective methods. There are technical as well as financial reasons to explain this translation. More than half of the failures in cable systems are ageing failures [12]. These failures depend on the condition of the components. The number of failures can therefore be decreased by diagnostic methods that find the damaged components before the failure occurs. Random failures such as excavator work or lightening, on the other hand, are caused by external factors and will not be prevented by diagnostics methods. The random failures dominating the overhead line failure statistics do, to a large extent, depend on the landscape surrounding the conductors. Inspection of the vegetation, animals and lightning related equipment close to the conductors can, to a certain extent, prevent random failures of overhead lines [13]. Cables are located underground and inspection of the cables is therefore not possible during operation. Costs associated to network failures increase as the failure duration or component replacement time increase. Cable systems are more reliable than overhead lines systems, but the average duration of a failure in a cable system is considerably longer than a failure in an overhead system [11]. Consequently the incentive of the distribution network owners to prevent a cable failure is large and the expenses related to the diagnostics tests can be justified.
5.2 Condition assessment methods

5.2.1 Inspection
All distribution companies contributing with information to this work perform visual inspection of their overhead lines and inspect the wooden poles for damage caused by decay. The line inspections are either performed from air, using a helicopter, or from ground using four-wheelers and, during wintertime, snowmobiles. The inspectors look for trees that threat to fall into and break the conductors, and corrosion. At least one of the companies uses video cameras during the inspections. The films can later be watched several times to analyse the systems. Acoustical inspections can be used for corona and surface discharges [14]. Acoustical methods are also usable to inspect cable joints. Traditionally, network companies have a deterministic approach to the inspections; the results are compared to defined limits. If the results exceed the limits, actions are taken. Trees that grow closer to the conductors than what is stated in the regulations will, for example, be cut down. An alternative approach is inspection based condition rating [13]. The idea is to give the network components ratings that reflect the importance of the components. The ratings combined with results of the inspections, are to be used in reliability analysis. This probabilistic method would demand extensive computational resources, quite contrary to the
easy applicable deterministic method. Underground cables cannot be visually inspected while in operation. After a breakdown the faulted cable can however be inspected to increase the understanding of insulation defects that cause failures.
5.2.2 Diagnostic methods

The increasing interest for diagnostic methods is reflected in numerous project and articles relating to the field. As mentioned in Chapter 4 the insulation degradation process in many cases leads to partial discharges before the actual insulation breakdown. Since the partial discharges can be detected it is possible to act on the degradation before failures occur. There are insulation defects, such as formation of water trees, that cannot in time be detected by PD diagnostics. The partial discharges will not be detected until after the water tree has developed in to an electrical tree, which in many cases is too late to prevent failure [12]. Water treeing is a common cause of failure in the Nordic countries. Resources are therefore put into the development of new methods to detect water trees at early stage [4]. Diagnostic methods are either online or offline. Online methods are used on cables in operation while offline methods require the cables to be disconnected from the system and energized from an external source. The water tree detection methods available today are offline methods generally referred to as dielectric spectroscopy, DS. Dielectric spectroscopy, as tan or Doubles method, is applied to a cable radial, whose losses and capacitance are measured and compared to the values of a new cable. Since treeing can be a local phenomenon the fact that dielectric spectroscopy analyses the entire cable radial is a disadvantage of the method [14]. Online diagnosis requires permanent connection to the network [16]. This is costly but opens for the possibility to detect treeing at an early stage. One of the methods under development, intended to become an online diagnostic method, is time domain reflectometry, TDR, [14]. By injecting a pulse to the cable and measuring reflections along the cable it will be possible to locate the treeing. Both PD and DS measurements are crucial for cable system reliability. Simultaneous measurements carried out with the same instrument would make the diagnostic more efficient. A combined system for PD and DS measurements is at present under development at the Royal Institution of Technology [17]. Cables are generally tested at commissioning while diagnostic methods for cable condition assessment are still considered too expensive to use on regular basis by most companies in the survey. One of the companies has, at two occasions, used dielectric spectroscopy on old cables that are of particular importance to the reliability of the system. The test results have differed between the occasions and are therefore considered slightly confusing by the company. One diagnostic method that in contrast to partial discharge detection and dielectric spectroscopy is widespread is the use of thermo cameras. The cameras are used to take pictures of for example cable joints in order to find areas with extensive losses.
5.2.3 Improved after-laying test

Routine tests after manufacturing is standard for high voltage cables. The cables are tested for AC voltage to assure there are faultless as they leave the manufacturer. Despite this, component related cable failures occur in the distribution systems. The conclusion must be that the cables, and joints, are either damaged during the transportation, the installation or after they have been installed. To reduce expensive failures and unscheduled outages the cable systems are therefore tested on-site, after installation. The purpose of the test is to discover damages before they lead to failure. According to several persons involved in the 7
power industry, improvement of the after-laying test carried out today is a key aspect to increase the cable system reliability. For financial reasons it is not possible to use high voltage AC tests to detect damages on site. Alternative waveforms are therefore used. Among them is the damped AC method, DAC, also called oscillating voltage waves, OVW. The equipment necessary to perform this test is small enough to be transported to the site in a car or minivan. The DAC test has a high sensitivity of PD detection and recognition of electrical treeing [18]. In addition to afterlaying test similar testing could be used after approximately two years of operation in order to detect damages caused by rocks in the soil, or other mechanical stress the cables might be exposed to after the installation. The DAC test has to be carried out off line, which might make distribution network owners hesitant towards the technique. Since the testing carried out today is considered insufficient it is however important for the individual companies and associations as Svensk Energi (Swedenenergy) to discuss and improve the after-laying tests.
6 Reliability influence of component factors

The cause of failure on cable systems and overhead lines differs considerably. Figure 2 and Figure 3 illustrate the cause of failure in cable systems and overhead lines. According to the Swedenenergy DARWin statistics [19], 35% of cable system failures are reported as material related failures, i.e. ageing failures. In addition, the component ageing is of significant relevance to the load related failures, i.e. overload, reconnection of load and fuse melting, that are reported to cause 11.5 % of cable system failures. 29 % of the cable system failures are due to external factors such as excavation work or sabotage. The dominant part, at least 70 %, of the overhead line failures is random failures due to environmental factors [20].
Cable system 40 30 20 10 0 Material/Method Personnel Damage Load related Other/Unknown %
Figure 2, Failure statistics of cable systems 2005 [19]

Overhead line 60 40 % 20 0 Tree related Weather related Miscellaneous Unclasified
Figure 3, Failure statistics of overhead line systems [20]
A majority of the companies participating in the survey agrees that the nature of cable failures, except failures due to excavation work, makes it interesting and useful to related cable data to failure statistics. Distribution network companies that are positive towards ambitious registration and systemised reliability work as well as companies that are not, emphasis the importance to recognize the experience of operators and other personnel.
6.1 Excavation work

One major reason the companies in the survey do not systematically relate component data to failure statistics is that they experience excavation work to bee the dominant source of failure. These are random failures, which are not related to component factors and cannot be prevented by diagnostic methods. The company view differs from the DARWin statistics according to which, 26.5 % of the cable failures are due to excavation work. Since failures due to excavation work do not depend so much on environmental but on human factors, it is possible to bring down the number of failures. As mentioned above urban distribution networks do already, to a large extent, consist of underground cable. Most cable practices are therefore developed to suit urban areas. One example is the indication of underground cables. Surface indication in urban areas might result in a large quantity of untraceable markings and consequently underground cables are at present indicated by yellow marking tape above the cables but below ground. In rural areas the conditions are quite different; the cables are distanced from each other and surface markings would not, to the same extent as in urban areas, interfere with other societal functions. Surface indication of rural underground cables is probably necessary to decrease the number of cable failures due to excavation work.
6.2 Cable systems

Data collection and systemized reliability work take a lot of effort from the distribution network companies. It is therefore important to be able to justify the component factors that are to be included in failure statistic reports. The failure statistics imply that the importance of relating component data to failure statistics is higher for cable system than for overhead lines. In addition, since regional differences is of less importance to the cable statistics, it becomes more interesting to share failure statistics with network owners in other parts of the country. Cable systems in rural areas differ from cable systems in urban areas. New, rural specific factors not yet experienced will therefore influence the power system reliability as the rural cable system development continues. It is difficult to for see exactly what the challenges, in terms of reliability, will be. During the years to come factors as the soil type, cable access and infrastructure will most likely prove to influence the failure rate and outage time of distribution systems. Four component specific factors; operational age, maintenance history, manufacturer and year of manufacture, is mentioned as having influence on reliability in several different reports and papers, among them [1] and [21]. A fifth factor, the method of cable excavation, has been mentioned in the company survey and is in this report considered of relevance to the component reliability. There are additional interesting reliability influences such as operational stress. It must however be manageable to keep a record of the influencing factor for it to be of any statistical use. In this report the influence of operational stress is not considered manageable and therefore not further treated.
6.2.1 Component operational age

The operational age of the component has an obvious influence on the component 9
reliability. As mentioned above 59 % of the cable system failures are age related. Consequently, the cable system failure rate increases considerably as the age of the system components increases. Component age is comprehensive and the most important component factor to be used in cable system failure statistics. Age dependency will decrease the standard deviation of the cable system failure rate considerably. A majority of the distribution network owners taking part in the survey claim to register their components. The extent of the registration varies but all companies that keep a register include age, which implies that the importance of component age is recognised. The age dependence is related to both environmental factors and manufacturing. It reflects the average component age dependence. Since the age dependence of the components often differs from the average age dependence, additional factors are useful to further improve the reliability of cable system statistics, i.e. decrease the failure rate or component lifetime standard deviation.
6.2.2 Manufacturer and year of manufacturing

The failure rate and failure rate age dependence may differ considerable between different manufacturers as illustrated in Figure 4 [1]. The figure includes approximate failure rates based on actual statistics.
8 6 4 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Age Figure 4, Component age dependence: XLPE failure rate per 100 km cable, two different manufacturers Manufacturer 1 Manufacturer 2
In addition to produce reliable failure statistics, one major reason to relate component data to failure statistics is, as mentioned in Chapter 3.2, to be able to identify problematic components. Early XLPE Cables, for instance, increased failure rate in the Nordic distribution networks during the eighties. It was the manufacturing technology that caused rapid ageing on the cables. Most of these cables are now replaced but similar manufacture related problems are not unlikely to happen again. The forced cablification of Swedish rural distribution networks has led to large demand of distribution cable. The increased paste in which the manufacturing now takes place might influence the quality of the cables. It is therefore additionally important to keep a record of the distribution cables manufactured at present. In the future it might be possible to notice a difference in the failure rate of cables manufactured during these years. According to the company survey carried out, it is not common to relate manufacturer and manufacturing year to failure statistics. Not until it becomes normal to store this data, the influence of the parameters on failure statistics can be thoroughly evaluated and the level of influence will appear. Figure 5 illustrates cable system failure rate manufacture year dependence. The dependence differs between different cable manufacturers. The rates are approximate but based on actual statistics.
Failures per 100 km
10
Manufacturer 1 Manufacturer 2 Failures per 100 km
4 3 2 1 0
1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 Manufacture year
Figure 5, Manufacture year dependence: XLPE failure rate per 100 km cable, two different manufacturers, Manufacturer 1 data 1978-1988, Manufacturer 2 data 1970-1981
6.2.3 Maintenance history

Average failure rates do not reflect the impact of maintenance and are therefore not useful for reliability centred maintenance [13]. Maintenance history must be included in both cable systems and overhead lines failure statistics, for the statistics to be used in maintenance management.
6.2.4 Method of cable excavation

The present, forced cablification of the Swedish rural distribution network influences the method of cable excavation. Since the speed of the excavation is important, plows are widely used to get the cables in the ground. It is not possible to control the interface between the surrounding material and cables that are plowed in to the ground; objects such as small rocks in the ground might end up very close to the cable and gradually destroy the cable insulation. It is useful for future reliability work, to study whether or not the excavation method influence the cable failure rate.
6.3 Overhead lines

Overhead line failure rate depends, to a large extent, on the numbers of trees in the distribution area. The failure rate used for reliability calculations can be adjusted to reflect this. It is however not necessary for the companies to relate the failure statistics to the number of trees in the area [22]. Instead mathematical models have been developed to convert the average overhead line failure rate to a tree dependent overhead line failure rate. In addition the geographical location, average temperature and annual rainfall are relevant factors to overhead line failure statistics.
6.4 Improvement of failure rate standard deviation

The national failure statistics represents a geographical average. The cable system average failure rate is comparable nationwide. The failure rate standard deviation does therefore not increase when the statistics are combined to an average national failure rate, Figure 6: 1. A small standard deviation is to prefer since this corresponds to actual values that do not differ much from the average value, i.e. the failures are easier to predict. When the failures statistics are related to component data the average rate is differentiated, i.e. there will be a number of different rates. The standard derivation from the specific rates decrease compared to the 11
standard derivation from one average rate, Figure 6: 2. 2,

Component data related to failure statistics Improvement of standard deviation
1, Regional
statistics combined to a national average
Number of different failure rates
Component lifetime standard deviation
Figure 6, Schematic picture of national cable system failure statistics: National co-operation and failure statistics related to component data considerably improve the standard deviation of estimated lifetime of cable system
For cable systems personnel and financial means required for data collection and statistic reports can be justified since the work will result is more accurate statistics, i.e. decreased standard deviation. Figure 7 schematically illustrates that for overhead lines the situation is quite different.
2, Component data
and environmental factors related to failure statistics Improvement of standard deviation
1, Regional
statistics combined to a national average
Number of different failure rates
Component lifetime standard deviation
Figure 7, Schematic picture of national overhead line system failure statistics: National co-operation and failure statistics related to component data do not, to the same extent as in the case of cable systems, improve the standard deviation of estimated lifetime of overhead lines
The average failure rate depends on regional factors. The standard deviation will therefore increase when the statistics are combined to an average national failure rate, Figure 7: 1. When component specific and regional factors are related to the statistics the number of different failure rates increase and the standard derivation from the specific rates decrease, Figure 7: 2. The result might not be a very large improvement, in terms of standard deviation, compared to the original regional statistics. The required personnel and financial means are therefore harder to justify.
12
7 Conclusion
The dominant cause of failure differs between overhead line systems and cable systems. Statistics show that a majority of the cable system failures are ageing failures while a majority of overhead line failures are random failures. The ageing failures do not to any large extent depend on environmental factors but are related to component data and are, in contrast to random failures, possible to prevent by the use of diagnostic methods. These are however costly to perform and most companies claim they cannot justify the financial means required. New, improved, diagnostic methods are under development and since visual inspections, widely use for overhead line systems, cannot be used on cables, the use of diagnostic methods is likely to increase. The characteristics of cable failures imply that component lifetime standard deviation decreases as component data are related to the statistics. In addition the statistics are similar nation wide. None of the companies participating in the survey systematically relate component data to failure statistics at present. One major reason for this is that they, in opposite to the statistics used in this report, experience excavation work to bee the dominant source of failure. A majority of the companies agrees however that the characteristics of cable failures, except excavation work failures, makes it interesting and useful to related cable data to failure statistics and to share the statistics with other network owners. Some of the distribution network owners do not believe that the result justifies the required means and will probably not relate their cable failures to component data unless EMI and Swedenenergy demands this information. The cable damages caused by excavation work does depend on human factors. Excavation work damage on rural cables could probably be decrease by surface cable marking. Ageing failures are caused by cable damages and mechanical stress during transportation, installation and operation. Part of the ageing failures and hence costly unscheduled outages might be prevented by improved after-laying tests such as DAC. Several companies are in the process of installing new program for network analysis. They emphasise the importance of computer programs that are easy to use and can handle the large amount of information in an efficient way. One challenge is to decide which component data that shall be related to failure statistics. Operational age, maintenance history, manufacturer and year of manufacturing are four manageable factors which influence on failure statistics is already established but yet not thoroughly evaluated. In addition the method of cable excavation is suggested as a fifth interesting factor related to cable system reliability. Since it might influence the cable system reliability the high pace in which overhead lines are now replaced by underground cables, further increase the interest of component data registration.
13
References
[1] Heggest J., Gjerde O., Mork R.K., Edfast J., Jensen M.M. and Tapper M.: Bedre utnyttelse av feil- og avbruddsdata, (Better utilization of failure and interruption data), (In Norwegian), Technical Report A6010, SINTEF Energiforskning AS, Trondheim, 2004. Ntkonsekvenser vid kablifiering av luftledningsnt, (Consequences to the distribution network when replacing overhead lines by underground cable), (In Swedish), Project report, Elforsk AB, Stockholm, 2006. Power system reliability analysis application guide, Brochure 026, CIGRE WG 03 of SC 38, 1987. Frslag p utvecklingsprojekt Diagnostiska metoder fr tillstndskontroll av transmissions- och distributionsnt, (Project proposal Diagnostic methods for transmission and distribution network condition assessment), (In Swedish), Report 97:16B, Preliminary study, Elforsk AB, Stockholm, 1997. IEC multilingual dictionary, Sixth edition, 2005 Zhang X., Gockenbach E., Wasserberg V. and Borsi H.: Estimation of the lifetime of the electrical components in distribution networks, IEEE Transactions on Power Delivery, vol. 22, no. 1, pp. 515-522, January 2007. CYME International T & D, Software brochure, 2005. http://www.cyme.com/software/cymdistram/, SKM System Analysis, Software brochure. http://www.skm.com/products_distribution_reliability.shtml Handbok: rsrapport fr elntsverksamheten, (In Swedish), Manual for annual power network report, Energy Market Inspectorate, Stockholm, 2007.
[2]
[3] [4]
[5] [6]
[7] [8] [9]
[10] Indikationer fr frsrjningstrygghet, (Indicators for security of supply), (In Swedish), ER 2007:4, ISSN 1403-1892, Swedish Energy Agency, Eskilstuna, 2007. [11] Kabellgning af distributionsnet Fase 3. Forsyningssikkerhed, (Distribution system cablification Phase 3. Reliability), (In Danish), Fordelingsnetkomiten DEFU, Komitrapport 89, 1992 [12] Gulski E., Smit J.J. and Wester F.J.: PD knowledge rules for insulation condition assessment of distribution power cables, IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 2, pp. 223-239, April 2005. [13] Brown R.E.: Failure rate modelling using equipment inspection data, IEEE Power Engineering Society General Meeting 2004, pp. 693-700, 2004. [14] Henriksen M.: High voltage engineering: Insulation techniques, ageing mechanisms, Course compendium, Chapter 10, DTU, 2006. 14
[15] Dubickas V., Edin H. and Papazyan R.: Cable diagnostics with on-voltage time domain reflectometry, NORDAC 2006. [16] Grape U.: Hur ska vi lokalisera fel i morgondagens landsbygdsnt?, (Fault localization in the rural networks of tomorrow), Elforskdagen, Stockholm, 2006. [17] Taylor N.: Diagnostic methods for electrical insulation, Presentation at Asset management course, KTH, Stockholm, 2007. [18] Brettschneider S., Lemke E., Hinkle J.L. and Schneider M.: Recent field experience in PD assessment of power cables using oscillating voltage waveforms, Conference record of the 2002 IEEE international symposium on electrical insulation, Boston, 2002. [19] Cable system failure statistics 2005, Swedenenergy, 2007. [20] Hilber P.: Component reliability importance indices for maintenance optimization of electrical networks, Licentiate thesis, KTH, Stockholm, 2005. [21] Zickler U. et. al.: Asset management in distribution systems considering new knowledge on component reliability, PSCC 2005, Lige, August 2005. [22] Gilligan S.R.: A method for estimating the reliability of distribution circuits, IEEE Transactions on Power Delivery, vol. 7, no. 2, pp. 694-698, April 1992.
15
Appendix A Distribution company survey

The appendix includes the original questions written in Swedish as well as a translation to English. 1. Uppgifter om elnt Ntspnning Lngd Andel kabel
Ange total lngd och andel kabel om ni inte har mjlighet att dela upp informationen p olika ntspnningar 2. Diagnostik Undersks kablarna och luftledningarnas skick regelbundet? Hur grs detta? Vilka metoder anvnds (exempelvis inspektion)? Vilken utrustning anvnds? Vem gr det? Hur bestmmer man vilka komponenter/ntdelar som ska underskas? Betona grna skillnader mellan luftledning och nergrvd kabel. 3. Registrering av komponenter Vilken information finns registrerad om ntens komponenter? (Tillverkare, Tillverkningsr, Annan information) 4. Insamling av data vid fel Vilken information registreras och lagras vid fel p luftledningar eller jordkabel? Lagras information om komponenten (Tillverkare, Tillverkningsr, Annan information)? Lagras information om feltyp? Lagras information om avbrottstid (vid avbrott)? Lagras ngon annan information? 5. Analys av data Grs tillfrlitlighetsanalyser/berkningar? Vilka parametrar anvnds i dessa analyser/berkningar (felfrekvens, avbrottsfrekvens, avbrottstid)? Anpassas vrdet p dessa parametrar p ngot stt till yttre eller komponentspecifika faktorer?
16
Exempelvis: Anvnds olika felfrekvens beroende p kabelns lder/tillverkare/marktyp? Anvnds olika felfrekvens beroende p luftledningens lder/trdtthet? 6. Tnkbara frndringar Det r inte omjligt att orsakerna till fel p kabelledningar i strre omfattning n fel p luftledningar beror p komponentspecifika egenskaper. Fel p luftledningar skulle d i strre omfattning bero p yttre faktorer som typ av landskap och klimat. Om detta skulle vara fallet gller fr kabel jmfrt med luftledning Det r viktigare att knyta komponentdata till felstatistik Det blir intressantare att dela statistik med andra fretag eftersom geografiska skillnader och klimat spelar mindre roll Vilken komponentinformation anser ni att det r rimligt att registrera och lagra vid fel? Komponentdata (Tillverkare, Tillverkningsr, Annan information)? Typ av fel? Avbrottstid? Annan information? 7. Intuition Finns det ngon allmn knsla om vad som r viktigt och mindre viktigt nr det gller tillfrlitlighetsarbete med luftledning och kabel? Vad som helst!
17
1. Information about the network Nominal voltage Total length Cable length
State total length and cable length for the entire network if it is not possible to divide into different voltage levels. 2. Diagnostics Is the state of cables and overhead lines examined regularly? How is the diagnostics performed? What methods are used (for instance inspection)? What equipment is used? Who carry out the diagnostics? How is it decided which parts of the system that shall be examined? Emphasise any differences between overhead lines and underground cable. 3. Component registration What network component information is registered? (Manufacturer, Year of manufacturing, Other information) 4. Failure data collection What information is registered and stored in case of failure on a cable or overhead line? Is any component data stored? (Manufacturer, Year of manufacturing, Other information)? Cause of failure? Outage time? Any other information? 5. Data analysis Are any reliability analysis/calculations carried out? What parameters are used for these analysis/calculations (failure rate, frequency of interruption, outage time)? Is the rate of these parameters somehow adjusted to environmental or component specific factors? For instance: Does the cable failure rate depend on the operational age, manufacturer or soil type? Does the overhead line failure rate depend on conductor age or number of trees in the area? 18
6. Possible changes It is not impossible that cable failures to a larger extent than overhead line failures depend on component specific factors. Fault on overhead lines would instead to a larger extent depend on environmental factors such as the landscape or the climate. If this would be the case it implies that for cable systems compared to overhead lines It becomes more important to relate component specific data to failure statistics It becomes more interesting to share statistics with companies in other parts of the country since geographical differences and climate are of less importance to failures What data do you consider reasonable or useful to register and store in case of failure? Component data? (Manufacturer, Year of manufacturing, Other information)? Cause of failure? Outage time? Any other information? 7. Intuition Do you experience anything else as important or less important when it comes to reliability engineering work in cable and overhead lines systems? Anything!
19
Theme 2 Reliability centred maintenance.

Banken and RCM A functional evaluation of maintenance management at Svenska Kraftnt by Per Nrman Evaluating the streamlined RCM project at Ringhals by Pia Gustafsson Reliability and cost centered maintenance methods by Julia Nilsson Implementation of production assurance programmes in a production plant: concept and discussion by Javad Barabady
Banken and RCM

A functional evaluation of a maintenance management at Svenska Kraftnt Per Nrman Industrial Information and Control Systems Royal Institute of Techonology June 2007
Kungliga Tekniska Hgskolan Industriella informations- och styrsystem Adress: Osquldas vg 12 7tr, SE100 44 Stockholm, Tel: 08-790 68 22, Fax: 08-790 68 39, E-post: pern@ics.kth.se, Web: www.ics.kth.se
1 (23)
Table of Contents
Banken and RCM...........................................................................................................................1 Banken and RCM functional suitability...................................................................................3 1. Background............................................................................................................................3 2. Goals with the study.............................................................................................................4 3. Method ...................................................................................................................................4 4. Theory evaluatory frameworks........................................................................................4 4.1. An overview of the functional reference model......................................................4 4.2. Reliability Centered Maintenance according to Moubray......................................6 4.3. Functional requirements on asset management systems for RCM ......................8 5. Data collection ....................................................................................................................10 6. Banken..................................................................................................................................10 6.1. Functionality Plant Design .......................................................................................10 6.2. Functionality Maintenance module.........................................................................10 6.3. Functionality: Document management module....................................................10 6.4. Functionality: Warehouse .........................................................................................11 7. Analysis.................................................................................................................................11 8. Results...................................................................................................................................12 9. Conclusions..........................................................................................................................12 10. References .......................................................................................................................13 Appendix A functional reference model for asset management within electrical distribution....................................................................................................................................14
Kungliga Tekniska hgskolan Industriella informations- och styrsystem
2 (23)
Adress: Osquldas vg 12 7tr, SE100 44 Stockholm, Tel: 08-790 68 22, Fax: 08-790 68 39, E-post: pern@ics.kth.se, Web: www.ics.kth.se
Banken and RCM functional suitability

This report describes the method and theory behind the evaluation of the maintenance management system Banken currently employed at Svenska Kraftnt to support maintenance management operations of their electrical transmission network.
1. Background
The deregulation of the power industry introduced a competitive pressure on utilities to conduct their business operations more efficiently. One of the most obvious ways to increase competitiveness is through improved asset management, i.e. the combined management of operations, maintenance, re-investment and investment of assets at utilities. The power industry has thus far used few cutting-edge techniques and methods for asset management. Rather the focus has been mainly the traditional time-based periodic inspection and maintenance activities, with little or no systematic thought given to how the separate actions impact the performance of the technical systems, and other business operations. In industries external to the power industry, however, new and ambitious approaches towards asset management have emerged. An example of this is Reliability Centered Maintenance (RCM), which stresses a holistic approach to maintenance. RCM comes in many forms or flavors but a common denominator is a system perspective, where maintenance management plans are strictly related to the function of the overall system. Rather than attempting to achieve excellence with respect to maintenance of isolated system components or subsystems, RCM proposes that no preventive maintenance and inspection activities should be undertaken unless it can be shown that they increase the overall reliability of the system. Although RCM as a method and maintenance philosophy purports to be general and transgress industry boundaries, RCM has made little progress in the electrical power industry and particularly in the transmission and distribution business. A possible explanation for this lies in poor data quality. Irrespective of choice of RCM-method, it requires a comprehensive systems understanding, and a far-reaching analysis of the function and possible failure modes of not only the overall system but also its constituent parts. This naturally emphasizes data quality as a critical success factor. If the whereabouts of many system components is unknown or uncertain, the RCM analysis will be poorer. Today, many large transmission and distribution have a bad understanding of their networks, and poor data concerning network component specifics. The introduction of RCM is very much linked to the use of information technology, since the amount of data that needs to be managed when managing as complex a system as for instance a distribution network by far exceeds human memory capacity. It is the intention of this report to evaluate the functional suitability of a modern maintenance management with respect to RCM-capability. Functional suitability refers to an information systems ability to meet a set of functional requirements, i.e. to perform a number of tasks to support business operations. [10]
3 (23)
2. Goals with the study

This study intends to investigate whether or not the maintenance management system Banken currently in use at Svenska Kraftnt offers the functions needed should Svenska Kraftnt decide to introduce the RCM-concept as their maintenance and asset management strategy. During the course of this investigation the following goals will be met: To develop and describe a generic functional requirements specification for Reliability Centered Maintenance To investigate the functional content of Banken at SvK To determine whether or not the functional content of Banken meets the requirements of the generic functional requirements specification.
3. Method
The angle of attack chosen in this investigation is to start with a generic functional specification called a functional reference model for asset management applications already developed in another project [1]. The RCM-specific functions are a subset of the functional reference model. Thus, the first step consists of delimiting the functional reference model using RCM-knowledge based on Moubray. This approach is a slightly refined version of the one found in [2]. The data collection at Svenska Kraftnt was restricted to one occasion where the system responsible demonstrated the functional capabilities of Banken.
4. Theory evaluatory frameworks

The evaluatory framework is a subset of the functional reference model developed in an earlier project. The functional reference model contains four levels of abstraction, to give an appreciation of the size and scope of the model, the top level contains four functions and the fourth and lowest contains approximately 400 functions. The model was developed based on an international standard [11] and validated in a comprehensive case study. The model has also been used to evaluate functional content in asset management applications at Vattenfall, and has as such proven itself to be useful. For more details regarding these evaluations see [3][4][5][6][7]
4.1. An overview of the functional reference model

This chapter will present the two top levels of the functional reference model.[6] See Figure 1 below. Due to space-constraints, the functional reference model cannot be presented in its entirety here, for a more detailed view see Appendix A.
4 (23)
Figure 1: The business functions and business sub-functions of the functional reference model. Going through the business functions and business sub-functions presented in Figure 1 above from left to right, we start with the business function entitled External to DMS. This business function contains functions exclusively utilized in DMSapplications but of a general, less business-domain specific nature. Three business subfunctions from this business function were included in the model. These are the Supply Chain and logistics sub-function dealing with supply of among other things spare parts and containing functions to manage warehouses and materials, the Premises sub-function dealing with some aspects of customer and company-owned property issues and the Document Management business sub-function dealing with distributing and modifying business critical documents of various kinds. The Records and Asset Management business function is, together with Maintenance and Construction (see below) one of the most important for asset management. There are three business subfunctions within it. The first is Substation and Network Inventory which contains functions for storing and displaying relevant data from the asset repository database. The Geographical Inventory business subfunction contains functions for displaying the electrical network on geographical maps. The Asset Investment Planning sub-function contains functions necessary for doing long-term planning of investment in new assets. This includes doing LCC-analyses and some measure of network calculations. The Maintenance and Construction business function contains five business subfunctions. The Maintenance and Inspection sub-function contains functions for the efficient execution of the routine preventive and corrective maintenance and inspection activities performed at an electric utility. The Construction and Design sub-function deals with the issues related to building various kinds of facilities, primarily the work-related issues. Work Scheduling is a sub-function used for detailed planning of work activities. Field Recording and Design is a sub-function dealing
5 (23)
with providing field crews with applications that are designed for out-of-office use. This entails real-time communications, the ability to access the asset repository from a remote location and so on. Work Dispatch is a business sub-function intended for centralized supervision of dispatched field crews. The business function Customer Support with its only sub-function Trouble Call Management was included in the functional reference model, as the contact with customers has some bearing on maintenance activities. Especially the corrective maintenance associated with customer outages. The present lack of sensors in lowvoltage distribution networks makes customer reports critical for assessments of network status, and consequently provides maintenance planners with relevant information. The last business function Operational Planning and Optimization and its subfunction Switch Action Scheduling/Work Scheduling, contains functions for the co-ordination of the de-energization of electrical equipment and planned maintenance activities.
4.2. Reliability Centered Maintenance according to Moubray

The view on RCM adopted in this project is that of Moubray, who has written a highly cited book on the subject [9]. Moubrays approach to RCM boils down to seven questions which must be answered for each asset or component of the system being subjected to RCM. The terms component and asset will be used interchangeably throughout this report. 4.2.1 What is the functional requirements of the asset, and what are the performance requirements? This question pinpoints the function of the asset. Although the answer might seem straightforward, it is often not. Describing the function of an asset is rarely limited to describing a certain type of operation (for instance performing the calculation 3 times 3), but also entails implicit constraints on that operation (being able to perform the calculation within 3 seconds for instance). These other constraints can be termed the performance of the component, or, using IT language, non-functional properties of components. 4.2.2 What are the functional failures of the asset? The idea behind preventive maintenance is to counteract failures of assets. Any systematic maintenance planning therefore must take into account the ways in which the assets may fail to achieve their functional goals as specified in the previous question. The present question answers the way in which any asset may fail to fulfill its function. For instance, a calculator may fail in reaching its functional specifications if it computes too slowly, or if its computations yield imprecise or inaccurate results (computing 3 times 3 to be 11 for instance).
6 (23)
4.2.3 What causes the failures? Having determined what kind of functional failures there are, it is important to get to the bottom of what causes these failures. To be able to answer this question, a technical analysis of the asset is in order. The answer to this question is on the form of a number of failure modes, i.e. causes of failures. For example, the failure of calculating 3 times 3 accurately may be due to faulty batteries of a pocket calculator. 4.2.4 What are the effects of the failure This question is often answered in conjunction with the preceding question, and its objective is to discern the importance of the failure modes. This question does not address what happens on the system-level, this is covered in the next question. Instead this question concerns the assets behavior in itself whenever a failure mode occurs. For instance, faulty batteries of the pocket calculator manifest themselves in higher response time to input data, which may be used as an indication of that particular failure mode. 4.2.5 What are the consequences of a failure? After painstakingly having gone through all the components of the system and discovered failure modes and effects, the next part of the analysis consists of analyzing the impacts these failures have on the overall system. According to Moubray there are four different types of consequences: 1. Hidden failure consequences; these refer to non-observable consequences of a failure, which may not be dangerous in themselves, but which in combination with other failures may have a very negative impact on the system. Ffor instance battery degradation may reduce performance of calculator in an imperceptible way but nevertheless cause user irritation. 2. Saftey and environmental consequences; a failure may be potentially harmful to either humans or the surrounding environment. For instance, a short circuit in the pocket calculator may cause severe burn wounds. 3. Operational consequences; a failure may have an adverse impact on the smooth execution of the companys business process. If for example the pocket calculator computes prices incorrectly, the margins of the business may be lower than intended. 4. Non-operational consequences; these are all the other consequences of a failure, meaning that the only thing that has to be done is to repair the asset. If there for instance is a dent in the calculator metal casing, this has no consequence as described above, but must be mended all the same. 4.2.6 What can be done to prevent an asset failure? This question concerns what preventive actions may be taken to reduce the likelihood of, or mitigate the consequences of the failure. There are three types of maintenance actions that may be undertaken with respect to this question:
7 (23)
Condition based maintenance actions which involves monitoring the condition of the asset and only perform maintenance whenever the condition is below pre-defined levels. Periodic replacement of assets are performed at predetermined time intervals
Periodic refurbishment of assets at predetermined time intervals. Which of these actions that are rational is dependant upon the nature of the asset and the consequences a failure may have. For instance, if a calculator is known to last for two years and then start degrading rapidly, it is rational to exchange it after two years. 4.2.7 What can be done if there is no suitable preventive action? In the event of there being no suitable maintenance activity to correct a failure, some other kind of action must be taken. If the consequence of a failure is serious, rebuilding the system to avoid these consequences is a viable option. For instance, if it is determined that the pocket calculator is impossible to maintain and indeed essential to the safe conduction of business operations, it may be best to discontinue the use of the calculator and instead invest in a state-of-the-art cash registry with calculating functionality.
4.3. Functional requirements on asset management systems for RCM

Based on work previously conducted by this author described in [2], functional requirements have been elicited to suit the work processes described in Moubrays seven questions as described above. These functional requirements have been translated to the maintenance process of the electric distribution companies and subsequently mapped to the functional reference model. The letter E and associated numbers that are found within parenthesis below refer to the specific number of the abstract component of the functional reference model. 4.3.1 First question: determining function of assets To establish what function an asset has two functions must be present, both having to do with storing and displaying data from the asset repository. (E11) Equipment characteristics. (E12) Connectivity model. 4.3.2 Second question: Determining component failures This question requires the same functions as the previous question. 4.3.3 Third question: Finding component failure modes To find the failure modes, a number of sources must be consulted. Firstly, it is important to know the maintenance and failure history as well as the inspection results of the component in question, as well as information in documents provided by the
8 (23)
supplier of the component. Consequently, the following abstract components must be found in an information system supporting RCM: (E43) Asset maintenance history (E44) Asset failure history (E42) Manage inspection Readings (E11) Equipment characteristics
(E10) Document management 4.3.4 Fourth question: Failure effects The failure effects of a failure mode the same functions as were used to answer the previous question are needed. 4.3.5 Fifth question: Failure consequences Failure consequences have an impact on the system level, meaning that functions giving a system view are needed, as well as decision support functions able to aggregate several measures into decision support. The following functions are applicable: (E12) Connectivity model (E34) Contingency analysis (E23) Risk Management (E24) Environmental management
(E25) Decision support 4.3.6 Sixth Question: Maintenance Planning The maintenance planning require knowledge about previous maintenance activities as well as of the specifics of each of the assets in the network. The functions needed for this are: (E18) Life Cycle Planning (E19) Reliability Centered Analysis (E27) Maintain Work Triggers (E39) Maintenance Program Management (E23) Risk Management (E24) Environmental Management
(E25) Decision Support 4.3.7 Seventh question: Managing non-maintainable risks Risks that cannot be avoided through preventive maintenance activities must be managed through investments and new construction. This requires functions not implemented in an asset management system, consequently they excluded in this study.
9 (23)
5. Data collection
This chapter briefly describes data collection at Svenska Kraftnt. The data collection consisted of a total of six hours of demonstrations and interviews were Bankens functional characteristics was given much attention.
6. Banken
The system Banken is composed of four overarching components. One component called Plant Design with the Master asset repository as well as functions for altering the contents of the asset repository. This component is linked to the Maintenance component which contains functions for maintenance planning and follow-up. The maintenance component is integrated with a warehouse component with the capacity to display warehouse status in terms of stock levels and automatically warn whenever the levels are below predetermined limits. All objects and work orders in the Banken system may be associated with documents of different kinds, drawings, inspection protocols and so on.
6.1. Functionality Plant Design

Plant design has the ability to display equipment characteristics as well as the connectivity model of the network can be found here. It is possible to query the database for information regarding asset failure history.
6.2. Functionality Maintenance module

There is functionality for maintenance program management. Preventive maintenance tasks can be scheduled in time and issued as work orders to contractors. Furthermore, there is functionality to assign criticality for each component type, thereby allowing risk management. All issued work orders may be traced through this module, which makes it easy to show asset maintenance history for all components. There is a possibility to set periodicity of maintenance tasks in this module, which corresponds to maintaining work triggers.
6.3. Functionality: Document management module

The document management module provides document management functionality, which is to say that it allows objects of the asset repository to be connected to associated documents, manuals, drawings etc. Document management also entails version management to prevent duplication of important documents, and this is fully supported in the Banken Document management module.
10 (23)
6.4. Functionality: Warehouse

The warehouse component contains functionality for checking warehouse levels and similar, which is very important from an overall maintenance management process perspective, but hardly relevant from an RCM analysis point of view.
7. Analysis
The analysis is summarized in the Table 1 below, showing which modules of Banken supporting which of the different functions derived above.
Table 1: The functional suitability of Banken with respect to Reliability Centred Maintance.
Function/component (E19) Reliability Centered Analysis (E27) Maintain Work Triggers (E39) Maintenance Program Management (E18) Life Planning (E12) model Cycle
Plant design
Maintenance
Warehouse
Document managenent
X X
Connectivity
(E34) Contingency analysis E23) Management Risk X
(E24) Environmental management (E25) support Decision X X X X X X X
(E43) Asset maintenance history (E44) Asset failure history (E42) Manage inspection Readings (E11) Equipment
11 (23)
characteristics (E10) Document management X
8. Results
The analysis of the Banken functionality revealed that there were several functions missing, making RCM difficult. Out of the 15 functions deemed necessary for sufficient RCM-support, 6 were not implemented in the system. It should however be pointed out that with a broadened unit of analysis, i.e. an extended investigation incorporating more systems at Svenska Kraftnt, functions such as life-cycle planning could very well turn out to be implemented in some other system in use at Svenska Kraftnt. Overall, Banken is a reasonably well-functioning maintenance management system, especially with regard to day-to-day operations where support is needed for amongst other things work order initation and follow-up.
9. Conclusions
This document has described an investigation of Svenska Kraftnts maintenance management system Banken with respect to its functional capability for RCM-analysis. The investigation has revealed that Banken lacks some functions, but is a fairly good system overall, with
12 (23)
10. References
[1] Nrman, P., A functional reference model for work and maintenance applications within asset management at Vattenfall, Master thesis, XR-EE-ICS_2006:012, School of Electrical Engineering, KTH, 2006 [2] Bertling, L., Tillfrlitlighetsanalys fr elkraftssystem, A-ETS-EEK-0501, School of Electrical Engineering, KTH, June 2005 pp. 417-516 [3] Andersson, A., A Method for Evaluating Business Value - A Case Study on Work and Maintenance Applications within Asset Management at Vattenfall, Master thesis, XR-EE-ICS_2006:015, School of Electrical Engineering, KTH, 2006 [4] Johnson, M., A Method for Measuring Technical Quality of IT Systems - A Case Study on Work and Maintenance Applications within Asset Management at Vattenfall AB, Master thesis, XR-EE-ICS 2006:016, School of Electrical Engineering, KTH, 2006 [5] Nrman, P., et al, Validation and Refinement of an Asset Management Subset of the IEC 61968-1 Interface Reference Model, Proceedings of the IEEE Power Systems Conference and Exposition, 2006 [6] Nrman, P., A Functional Reference Model For Asset Management Applications Based on IEC 61968-1, Proceedings of the Nordic Distribution and Asset Management Conference, 2006 [7] Gammelgrd, M., et al, Architecture Scenario Analysis Estimating the Credibility of the Results, to appear in Proceedings of the Seventeenth Annual International Symposium of The International Council on Systems Engineering, 2007 [8] Gammelgrd, M., Evaluation of business value of IT-system scenarios a case study at a large north European power company, Proceedings of the Portland International Conference on Management of Engineering and Technology PICMET, 2007 [9] Moubray, J., Reliability-centred Maintenance, Butterworth Heinemann, 1991 [10] International Organization for Standardization. ISO/IEC 9126-1 International Standard - Software Engineering Product Quality - Part 1: Quality Model, ISO/IEC, 2001. [11] IEC Technical Committee 57, Working Group 14, IEC 61968-1 Application integration at electric utilities System Interfaces for distribution management Part 1: Interface Architecture and General Requirements, International Electrotechnical Committee, 2003
13 (23)
Appendix A functional reference model for asset management within electrical distribution
This appendix contains the three top levels of the functional reference model for asset management used in this evaluation are described below. Descriptions of abstract components
Function to facilitate the procurement of services from contractors [E4] This function keeps track of legal and other issues regarding written contracts. Contracts may deal with material goods or services. [E4] Function to manage warehouse logistics [E4] Function to manage of materials in store and materials needed for construction and maintenance. [E4][V3] Function to record and display adresses of customers and adresses of premises owned by the utility [E10-14] Function to keep track of which substation feeds which node in the electrical network, this is especially important when locating causes of outages. [E10-14] Information about the meter, not the data in the meter. [E10-14] This function deals with the legal aspects of running and extending an electrical network. This includes, but is not limited to, agreements with land owners on whose property the lines are built. [E4] Function to manage all aspects of owning and operating real estate. [E1014]
Business Function
E1 External to DMS
Business SubFunction
Supply Chain and Logistics
Abstract Component
Procurement
E2
Contract Manangement Warehouse Logistics Materials Management
E3
E4
E5
Premises (PRM)
Address
E6
Source substation
E7
Meter information
E8
Right of ways, easements, grants
E9
Real estate management
14 (23)
E10
Document Management
Document Management [1]
Function to facilitate an easy management of the various types of documents needed in the daily operations of a power company. This includes the easy access, distribution and changing of documents. [E10-14] A function to present characteristics of all objects in the asset repository. "Characteristics" refer to a set of properties, some of which may change over time, the combined set of all these properties are refered to as the equipments "condition". [E10-14], [E10,11][V5] Complete description of all kinds of electrical connections within the network. [E10,11] Function to show a more detailed view (electrical and otherwise) of substations within the network. [E10-14] Function to store and display information about the communications equipment installed in the network. [E10,E11] Function to display the network on a geographical map. [E10,11] An application with functionality to show geographic ("planimetric and/or topographic") maps which together with the network display function can be show the networks spatial distribution. [E10,11] This function stores and implements the utility's maintenance strategy on the asset investment planning for the utility. Guidelines outlined in the strategy regarding what triggers investments and similar things are operationalized in this function. This function does a LCCanalysis for different investment alternatives. [E6]
E11
Records and management
asset
Substation network (EINV)
and inventory
Equipment Characteristics
E12
Connectivity Model Substation Display
E13
E14
Telecontrol Database
Geographical Inventory (GINV)
E15
Network Displays
E16
Cartographic Maps
E17
Asset Investment Planning (AIP)
Maintenance Strategy
E18
Life Cycle Planning
15 (23)
E19
Reliability Centered Analysis
This function calculates the reliability of the network in its current and future state. The analysis uses previous information about the reliability of the components in the network together with probabilistic methods, and aggregates the reliability from the component level to the network level. [E6] Function to store and display standard design solutions when planning new investments. [E6] Function to give an overall picture of an assets performance. Performance refers to dynamic characteristics, such as the condition of a certain component and the quality of operation, rather than static properties such as year built etc.[E6][V7] Function to calculate required performance indices for the reliability of the network.[E1,E2] Risk management means the process or function to handle probabilistic occurences (with negative consequenses). It involves assigning monetary value to consequences of events which occur with estimated probabilities. In this business sub-function "Risks" refer to probabilities of events that are harmful to humans or property. Also, this function imports environmental risks from "Environmental Management" and operational risks from "Reliability Centered Analysis" to provide an overview over all risks. Function to manage environmental issues of a local/global nature such as avoiding oil spills or preventing hazardous materials to contaminate nature. [E6][V6]
E20
Engineering and design standards
E21
Performance Measurements
E22
Performance Indices
E23
Risk management
E24
Enviromental management
16 (23)
E25
Decision Support
Function to collect and analyse all available relevant data (regarding risks, environment, financial situation and so on) and based on this prioritize investment alternatives. [E6][V6] Function to allocate budgetet funds for investment projects (short term) and to budget funds for future investment projects (long term) [E6] Work triggers decide when to act. A trigger can be some level of performance of an asset which when it is too low automatically triggers a maintenance activity. This function ensures that these levels are in sync with the overall maintenance plans and that they are set optimally.[V7][V2] Function to group assets according to preferred criteria (geographic location, functional specification etc.) This is done to get an overview and to simplify investment planning. [E6] Function to store, analyse and display previous failures of components and component types. [E6][E10,11] Function to break down the total financial performance of the business unit down to plants, to systems to components. If such a breakdown is possible it serves as valuable input to asset investment planning. [E6] Functions which contains information regarding the temperature limits of all assets. The temperature limit on lines determines the ammount of energy they can transfer. [E10-14].
E26
Budget Allocation
E27
Maintain Work Triggers
E28
Asset Maintenance Groups (lists)
E29
Asset Failure History
E30
Asset Financial Performance
E31
Thermal ratings of network equipment and lines
17 (23)
E32
Load forecast
A load forecast function predicts the system load. The forecast is done with a time-frame of about a year ahead. To do this historic data of load is used in conjunction with probabilistic methods and predicted extensions of the network, due to new loads. [E8,9] Function to calculate the voltages of all nodes in the network, and the power flow through different lines. Also known as load flow analysis.[E8,9] This function studies the effect of a failure of a system component. This information can be used to determine how to restore power to customers who are affected by an outage. Also this function is used to analyse effects of loss of generation and transmission facilities. This analysis, which is done on a regular basis, determines the resulting voltages and power flows given some contingencies. This function needs data from a load flow application.[E8,9] Function to do a short circuit analysis. A short circuit analysis is used to determine the short-circuit currents given failure modes in various parts of the system. This is used to protect equipment from being overloaded and to help set fuse breakers. [E8,9] Function to calculate the optimal power flow with respect to energy losses and switching/design scenarios. [E8,9] Function to calculate the energy losses in the network. [E8,9] Function to keep track of feeder profiles which means the voltages in different nodes of the network. The voltage intervals these nodes belong to and so on. [E8,9]
E33
Power flows computation
E34
Contingency analysis
E35
Short circuit analysis
E36
Optimal power flow Energy Loss calculations
E37
E38
Feeder voltage profiles
18 (23)
E39
Maintenance and construction
maintenance and inspection (MAI)
maintenance program management
Function to manage the maintenance program. A maintenance program is an operationalized maintenance strategy, or maintenance plans on a long- and short term basis. In the maintenance program intervals inspection and preventive maintenance activities are broken down and specified. Also included in this function is a decision support function, coordinating the maintenance activities with the network extension planning and other investment projects. [E10,11] Work triggers decide when to act. A trigger can be some level of performance of an asset which when it is too low automatically triggers a maintenance activity. This function ensures that these levels are in sync with the overall maintenance plans and that they are set optimally.[V7][V2] Function to group assets according to preferred criteria (geographic location, functional specification etc.) This is done to get an overview and to simplify investment planning. [E6] Function to handle the data collected from inspections and other sources of information (sensors etc.) [E10,11] Archives notes on earlier maintenance activities on specific assets and/or groups of assets. [E10,11] Function to store, analyse and display previous failures of components and component types. [E6][E10,11] Function to track and follow up issued work orders for maintenance and inspection, with respect to both the technical and financial aspects.[E10,E11]
E40
Maintain Work Triggers
E41
Asset Maintenance Groups (lists)
E42
manage inspection readings Asset Maintenance history Asset Failure History
E43
E44
E45
work order status tracking
19 (23)
E46
work order closing
Function to support the closing of a work order. This involves both updating the asset repository as well as the technical and financial follow up. [E10,11] Function to manage the financial issues of maintenance and inspection, involves managing payment of invoices etc. [E10,11] Function to manage the process of refurbishing components in the network, including the necessary changes in the asset repository, the replacement of refurbished parts in their functional locations, and the financial issues involved (due to the changed value of the network). [E10-14] Function to plan the process of inspection of the facilities. [E10,11] Function to create a work order. [E7] Function to create a work order. [E7] Function to facilitate the creation of the work design needed to perform required work. [E7] Function to estimate cost of doing work. [E7] Function to plan work other than projects. This includes making the choice of who is to perform the work (contractor or in-house crews). [E15] Function to track and follow up issued work orders for maintenance and inspection, with respect to both the technical and financial aspects.[E10,E11] Function to support the closing of a work order. This involves both updating the asset repository as well as the technical and financial follow up. [E10,11]
E47
financial control
E48
Refurbishment processing (1)
E49
Inspection [1] work initation

Construction and design (CON)
E50 E51
work initation work design work cost estimation work flow management
E52
E53
E54
E55
work order status tracking
E56
work order closing
20 (23)
E57
financial control
Function to manage the financial issues of maintenance and inspection, involves managing payment of invoices etc. [E10,11] Function for easy planning and design of projects[E7] Function to store data about newly installed equipment in the asset repository. It also deals with the procedure of making the installed equipment operational and ready to use. [E10-14] Function to manage the process of scrapping phased out equipment. This includes both financial and environmental aspects.[E10-14] Function manage the process of phasing out equipment from the network. [E10-14] A function to make detailed plans of the tasks that are part of the work order.[E7] Function to keep track of work crews and their activities. This includes keeping track of current and future activitites, knowledge training and so forth. [E7] Function to handle all equipment necessary to perform the requested work. This could involve items such as tools and vehicles. [E10,11] Function to coordinate material supply for construction work. [E7] Permits are needed to do work on some instances. For instance, work on some voltage levels demands that the crew has an adequate training and some substation are not to be entered because of security implications. [E7]
E58
Project planning and scoping (1)
E59
Phase-in Equipment (1)
E60
Asset Scrapping (1)
E61
Phase out equipment (1)

work scheduling (SCHD)
E62
work task planning
E63
crew management
E64
Equipment management
E65
Material coordination
E66
Permit management
21 (23)
E67
Field recording and design (FRD)
Field design
When an asset is refurbished or replaced some of the planning is made on site "in the field". This functions allows the crew to record the design, or parts of the desing out of office. [E10-14] A function to store and transmit data collected during field inspections. [E3] Function to manage work crews time entries, i.e the time they have spent on actual work. Function to keep track of delivered material, in the field and in stores. [E7] Function to keep track of equipment in the field. [E1014] Function to track what takes place in the field, i.e crew movements and so on.[E7] Real time communication. [E7] Function to incorporate weather forecasts, and actual weather into the planning and dispatching of work crews. [E7] Function to process outage calls from customers. [E1,E2] Function to process calls from customers related to power quality issues such as voltage dips and so on. Also included in this function is the management of trouble calls related to other issues related to flaws in the network, not falling under either the cathegories "outage" or "power quality". (E.g. a leaning pole). [E1014] Function to support the automatic or manual notification of those affected by planned outages [E1,E2] Function to support the necessary media communications during an outage.
E68
Field inspection results
E69
Crew time entry
E70
Actual materials Actual Equipment [E10-14]

Work dispatch (DSP)
E71
E72
Field status tracking Real time communication Weather monitoring
E73
E74
E75
Customer Support
Trouble Call Management
Outage calls
E76
Power Quality
E77
Planned outage notifications Media communication s
E78
22 (23)
E79
Restoration projectio/confir mation
Function to estimate time needed for the restoration of power to affected customers and to notify these customers of projected restoration times. [E1,E2] Record kept of all outages. Often required by authorities. [E1,E2] Most work on electrical equipment cannot be done when power is on. Therefore a function to schedule and coordinate when power is to be off and when to do work is needed. "Clearance" is the same thing as giving someone permission to work on deenergized cables. Also entails notifying operations control when work is performed on energized cables. [E1], [E10-14] A function to determine what the consequenses of deenergizing a line would be in terms of number of affected customers. This analysis enables the work planner to schedule work on lines with as little impact as possible on the companies customers, and to notify those affected about the outage. [E1]
E80
Outage History
E81
Operational Planning and optimisation
Switch action scheduling/work scheduling
Release/Cleara nce remote switch command scheduling
E82
Customer outage analysis and information
23 (23)
Evaluating the Streamlined RCM project at Ringhals

Pia Gustafsson Department of Industrial information and control systems Royal Institute of Technology (KTH) Stockholm Sweden
Abstract
This study describes the project of introducing Streamlined Reliability Centred Maintenance, SRCM, at Ringhals. A framework for assessing the project is presented and an evaluation of the project has been done with respect to how well the set up goals have been met. The result has been that the soft goals, i.e. concerning organisational motivation and culture of work has been reached, as well as the goals of fulfilling the governmental demands while neither cost related nor performance related benefits have been identified.
Table of content
Evaluating the Streamlined RCM project at Ringhals ............................................................... 1 Abstract ...................................................................................................................................... 1 1. Introduction ........................................................................................................................ 1 1.1. Outline of the report ................................................................................................... 1 1.2. The SCRM process..................................................................................................... 2 1.3. Criticality.................................................................................................................... 2 1.4. Difference between SRCM and RCM........................................................................ 3 2. Structure of the study ......................................................................................................... 4 2.1.1. Documentation ................................................................................................... 4 2.1.2. Interviews ........................................................................................................... 5 3. Benefits assessment framework ......................................................................................... 6 3.1. Cost benefits........................................................................................................... 6 3.2. Performance benefits.............................................................................................. 7 3.3. Quality benefits ...................................................................................................... 7 4. Realised benefits ............................................................................................................ 8 4.1. Cost benefits........................................................................................................... 8 4.2. Performance benefits.............................................................................................. 9 4.3. Quality benefits ...................................................................................................... 9 5. Analysis and discussion ................................................................................................... 10 5.1. Cost benefits............................................................................................................. 10 5.2. Success factors ......................................................................................................... 10
1.
Introduction
Several organisations within the Swedish power company Vattenfall AB (Vattenfall) have tried to introduce Reliability Centred Maintenance (RCM) during the years with various results. The nuclear power plant Ringhals has four reactors and has to follow Statens Krnkraftsinspektions Frfattningssamling (SKIFS) which are the governmental documents that regulate the operation of nuclear power plants in Sweden. 1997 Ringhals analysed their processes and identified a need for better decision support for choosing the right maintenance operations and frequency of them to be able to support the observance of the regulations. RCM was seen as a possibility to solve these problems and optimise the maintenance costs. A pilot project was started 2000 to evaluate different variations of the RCM methodology. The pilot project tried two different implementations of RCM, of which one was the Streamlined RCM (SRCM), on seven systems at one of the four reactors. The pilot project resulted in a suggestion of implementing SRCM at Ringhals [4]. In 2003 the nuclear power plant Ringhals started with the introduction of SRCM, which is a commercialized form of RCM that doesnt require as much time and effort to introduce as a regular RCM approach. When a project like this is initiated there are often several things that are supposed to come out of the project. Some of the wanted goals are attained and some are not while other goals that were not thought of from the beginning might be fulfilled. This case study aims at assessing the benefits that the SRCM project has brought with. The overall goal with this project is to give a fair idea of which changes the introduction of SRCM has led to at Ringhals. This case study will start with clarifying which the original goals were and how well these have been fulfilled. It will also look if there are other theoretical benefits with the project that might have been fulfilled, and hopefully it will be a help to highlight areas where the goals havent been met and suggest possible approaches to solve this.
1.1.
Outline of the report
The next chapter starts with an introduction to RCM and SRCM, how the processes are implemented and what the differences between them are. Chapter 3 describes the structure of the case study; how the framework for the analysis has been created and the empirical sources that have been used. Chapter 4 presents the framework and chapter 5 presents how well the benefits have been fulfilled in the project. Chapter 6 contains analysis and discussion, where the future of SRCM and possible success factors are mentioned.
Streamlined Reliability Centred Maintenance Reliability Centred Maintenance, RCM, is a process used to determine the maintenance requirements of any physical asset in its operating context (Moubray 1991). The RCM process consists of seven questions that should be answered. These all deal with identifying what the purpose of a system is, what the consequences would be if a failure occurs and what the reason could be for such a failure. Based on this information the most suitable maintenance program will be created.
1.2.
The SCRM process
Introducing SRCM starts out with an analytical SRCM process where all the systems are analysed from a functional perspective. This SRCM process should then lead to changes in the Maintenance program, MP. For a fully successful SRCM implementation, SRCM analyses should be made continuously for new systems or system changes. The SRCM process used at Ringhals is described in Figure 1. The process starts with data collection of the system that is being analysed and general planning of the analysis process. The documents required for the analysis are documents for normal operation, alarms, rounds, and schemes for the system, flowcharts and logic. In step 2, component data such as component ID and component name is being imported from the SAP system to the SRCM Workstation. In step 3, the system functions are defined with help of the system description, and in the following step, its correspondent failure functions. The analysis leader set the criticality to the failure modes, and these are agreed upon on a project meeting where the maintenance engineers are present.
Figure 1: The SRCM process [SRCM Work flow]
If necessary, the criticality analysis is revised in step 7. The current maintenance program is imported from the SAP program and in step 9; the changes to the maintenance program are suggested for the critical functions. In step 10 the current and the suggested maintenance program are compared. The changes are discussed during a project meeting and agreed upon or revised before the final report is written, step 12.
1.3.
Criticality
The fifth step of the SRCM process is criticality analysis. This means that all failure modes are categorized as being either critical or not, there are no levels of criticality. A failure mode is said to be critical if it in case of a failure would lead to anything of the following: Affecting the function of fast stop of the reactor [Gransson] Resulting in reduced effect or efficiency. Resulting in environmental infringement Resulting in negative safety, violation of STF regulations Resulting in person damage
Resulting insignificant damage >100 SEK Resulting in negative affect on the radiology environment Resulting in fire Only the critical failure modes are analysed in the SRCM process.
1.4.
Difference between SRCM and RCM
SRCM is said to be a more time efficient method then RCM without lowing the quality of the analyses. Besides from the criticality analysis, that decides which systems that will be analysed, streamlined RCM differs from regular RCM in which failure modes are included. In RCM multiple failure criteria decide which modes that should be analysed, while in SRCM, only single failure criteria are considered. Theres one exception to this, and it is failure of security components since their functionality is only triggered when another component/function has failed.
2.
Structure of the study
To make sure that no benefits are missed out on, two major areas have been studied; the benefits that were expected by the organisation to be realized, and the benefits that are possible to obtain according to literature within the area and spokespersons for the SRCM/RCM method. The expected benefits of the project are supposed to complement the list of theoretical benefits to create a reference list to check of the realised benefits against. Before the SRCM process started out, a pilot project was carried out to assess the theoretical benefits of a future implementation of SRCM. The results from that study have been of great importance in this chapter, but also the impressions and thoughts of the project members as they entered the project. Figure 2 shows that the theoretical benefits come from theoretical studies and the expected benefits come from empirical studies.
Figure 2: Different benefits and how they are related.
It is supposed that the realised benefits should be a subset of the possible and expected benefits. The expected benefits and the theoretical benefits will be studied first and will create the assessment framework for which benefits that have been realised. The benefits that arouse from literature discuss RCM and not SRCM, but there is reason to believe that SRCM is supposed to bring these benefits as well.
2.1.1.
Documentation
The empirical sources used for this study are interviews and documents. The documents that have been used are shown in Table 1. The right hand column shows which kind of benefits the document has contributed to.
[1] Final report SCRM R1-421 Marcus Ekelundh 1857266/3.0 [2] Living Program DRAFT [3] Implementering av resultat frn SRCM Hans Gransson 1851066/3.0 [4] SRCM Task Force Olle Erixon 1714929/2.0 [5] Projekt specifikation Infrande av SRCM inom Ringhalsgruppen Hans Gransson 1734265/5.0 [6] Framstegsrapport SRCM 2006 Hans Gransson 1927252/1.1 [7] SRCM Handbook Realised benefits Realised benefits Realised benefits Expected benefits Expected benefits
Realised benefits
Gran Eriksson 1750956/3.0 [8] SRCM Workstation Advanced Training Workshop ERIN version 4.0 Table 1: Documents used for empirical data.
Realised benefits Realised benefits
2.1.2.
Interviews
Ten persons have been interviewed for approximately one to two hour each. The interviews have been open circling around what the expectations were before the project, what the effects have been so far and important areas for the future project. For the people involved in the pilot study, the focus has been on the expectations on the project, and the people working with the project at present have been asked questions about. Four interviewees contributed mostly to the expected benefits, and six of them mostly to identify which the realised benefits were. A list of the interviewees without their names is presented in Appendix B.
3.
Benefits assessment framework
The assessment framework is made of the theoretical and the expected benefits. The assessment of the introduction of RCM at Ringhals is based on how well these benefits have been reached. The benefits are arranged after three main areas; cost related benefits, performance related benefits and quality related benefits. The quality benefits include both higher service quality (better maintenance) and higher organisational quality. The benefits that are part of the framework are shown in Figure 3:
Theoretical & Expected Benefits
Cost related benefits
Performance related benefits
Cost efficiency of maintenance
Lower costs of new investments Quality related benefits
Improved availability of functions
Improved efficiency of functions
Yield of functions
Organisationa l quality
Service quality
Governmental demands
Staff knowledge
Organisationa l culture
Quality of maintenance plan
Safety and environmental integrity
Maintenance program should be documented.
State surveillance during operation.
Knowledge about functions
Knowledge of different failure modes
Unified work culture
Improved team working
Fewer nonplanned measures
Traceability of the maintenance plan
Knowledge of operational work
Knowledge of criticality of components
Figure 3: Theoretical benefits viewed as three main areas.
The benefits are described more in detail below.
3.1.
Cost benefits
One cost benefit of using RCM is that it will lead to a more cost efficient maintenance work. (Moubray) gives an example of routine maintenance workload that has been reduced to 50% when using RCM. Cost efficient maintenance is also mentioned by three of the interviewees. The savings that are expected are because of less maintenance would be needed [11][12] and that the resources, in terms of staff, are used in a more efficient way [15]. This is in some way 6
contradictory to what the pilot study [4] suggested, since it stressed that the direct monetary benefits would not be substantial.
RCM could also lead to lower costs of new investments; this would follow by a more reliable maintenance which leads to less need for new investments according to Moubray.
3.2.
Performance benefits
Performance is defined by three factors by (Moubray); availability, efficiency and yield. The availability is a measure of the uptime of the system, while efficiency measures the rate the asset is working. The yield determines the output of the asset. All of these areas can be approved by RCM according to Moubray. One of the interviewees expressed a possible benefit in a more efficient use of the personnel resources when the deeper knowledge that comes together with the SRCM analyses can make right staff group perform the right kind of measurements [15].
3.3.
Quality benefits
Organisational quality RCM also affects many quality attributes such as the staff knowledge about system functions [Moubray] [12][15], their failure modes [Moubray][12], knowledge about operational work [13][14] and criticality of components [12][13][14][15]. Another knowledge related issue is that a large part of the mechanical staff that soon will get retired within the next 5-10 years and by documenting and justifying the routines, their knowledge will not disappear [12]. One benefit that has been stressed by the interviewees is that the SRCM analysis process would improve the organisational culture. One of the participants in the pilot project pointed out that increasing the collaboration between the operational staff and the maintenance staff was one of the objectives that were stressed after the pilot was done [11][12][13][14][15].This can also unify the work culture between the two groups [11][12] and improve the motivation of the staff [11][13][14]. Service quality Governmental demands are an important expected benefit since that was the reason why the SRCM project was initiated at first hand [8][13][14][15]. The governmental demands include having a clear documentation that is traceable of how and why the maintenance plan looks like [11][12]. Other quality benefits are that the safety (Moubray) and environmental integrity can rise (Moubray), and the non-planned measures will be fewer (Moubray) [15].. The quality of the maintenance plan and the maintenance work is of course important, and this has been identified by many [8][9][12][15].
4.
Realised benefits
The realised benefits have been based on interviews, and documents produced during the SRCM process. Here, the benefits from the assessment framework are presented with respect to how they actually fell out, but first a short description of what the project has lead to so far is presented. Sixteen analysis reports from Ringhals 1 were studied and concluded that 2100 changes were to be made in the maintenance plans, in average 123 changes per systems. This might sound like a lot of changes, but in reality 787 changes were labelled as RETAIN while 498 were labelled ADD. The changes that were added were mostly maintenance tasks that already were performed, but not mentioned in the maintenance plan or where it now was established that the responsibility was at either the maintenance or the operational staff. 44 changes were increased with respect to the periodicity and 207 changes pointed at a decrease. Only 7 maintenance tasks were deleted.
1000 900 800 700 600 500 400 300 200 100 0 Tot. ADD RETAIN MODIFIED increased MODIFIED decreased DELETE
Figure 4: The characteristics of changes in the maintenance plan.
It is interesting to notice that 57 of the changes that added a maintenance task were so called run to failure were the maintenance is reduced to none.
4.1.
Cost benefits
All of the interviewees stress that there have been no cost benefits because of the SRCM analyses [16]. This is because the maintenance intervals have not changed significantly after the analyses. One explanation that is offered is that nuclear power stations have historically had very high reliability and safety demands and hence put a lot of effort at maintainability [13][14]. Another interviewee claims that SRCM is very expensive and this reduces the possible cost benefits [16]. One of the respondents suggests that less corrective maintenance might lead to lower costs, but has no evidence for this [11].
4.2.
Performance benefits
The same answers as were given for the cost benefits were given for performance. The nuclear power plants have had high availability historically. Therefore, changes due to the RCM project have been small and few [13][14][16]. .
4.3.
Quality benefits
Organisational quality Staff knowledge The interviewees agreed upon that the knowledge about the system functionality, its critical components and failure modes][13][14]. Organisational culture The interviewees agreed that the SRCM analysis process has improved the organisational culture [13][14]. Service quality Governmental demands are an important expected benefit since that was the reason why the SRCM project was initiated at first hand [8][13][14][15]. The governmental demands include having a clear documentation that is traceable of how and why the maintenance plan looks like [11][12]. Other quality benefits are that the safety (Moubray) and environmental integrity can rise (Moubray), and the non-planned measures will be fewer (Moubray) [15]. The quality of the maintenance plan and the maintenance work is of course important, and this has been identified by many [8][9][12][15].
5.
Analysis and discussion
Before the SRCM process started, a pilot project was carried out to assess the theoretical benefits of a future implementation of SRCM. The results of this project have had a great impact on which the expected benefits of SRCM were.
5.1.
Cost benefits
In the report for the pilot project [4], it says that the strictly economical benefits achieved in the pilot do not solely suggest an implementation of SRCM at Ringhals. In the interviews, this is confirmed and stressed by one of the participants of the pilot project. A possible explanation is given by one of the interviewees; Ringhals has traditionally had very high efficiency in their production in comparison with other nuclear power stations and hence they didnt have big costs to cut down on. The interviewee also pointed out that other organisations may be able to motivate an introduction of SRCM because of economically benefits only. Otherwise, nothing is mentioned about increasing the function availability or efficiency. This might be related to that the nuclear power industry in general and Ringhals power plants in particular has had a fairly high availability historically
5.2.
Success factors
To get the most benefits out of the SRCM project, it needs to be continuously updated. Ringhals calls this the living program [10]. The changes that have been identified in the The project of analysing the systems at the four blocks was not completely finished when the study was made. Table 2 shows how many of the systems at respective block (R1, R2, R3) that have been analysed, that are ongoing and that are remaining to be analysed.
R1 R2 R3 R4 Remaining Ongoing 9 10 22 13 12 13 12 13 Table 2: Project status Finished 30 9 19 18
One opinion that were mentioned by several participants of the SRCM analyses was that the analyses tended to be more efficient and of better quality when the analysis leader had good knowledge of the systems then when the leader is a consultant [13][14]. References Moubray, J. (1991) Reliability-centred maintenance, Butterworth Heinemann, Oxford, England Wireman, T. (2005) Developing performance indicators for Managing Maintenance, Industrial Press, New York, USA
10
Appendix A: Questionnaire
Control/strategy Common criteria for criticality for all units Higher competence for maintenance- and drift personnel Efficiency Optimized maintenance strategy Common vocabulary and understanding between maintenance and drift Focus at maintenance of security-, STF related equipment Findings Personnel/organisation Higher competence Better cooperation Generation shift Modern maintenance system Focus at TBU Foundation for advanced TBU Governmental demands Maintenance program should be documented Analysis of availability of competence Felfunctioner, their probability and consequences State surveillance during drift Cost benefits Cost-effectiveness of maintenance Lower costs of new investments Performance benefits Improved availability of functions Improved efficiency of functions Yield of functions Quality benefits Knowledge about functions Knowledge of different failure modes Improved team working Higher motivated staff Unified work culture Safety and environmental integrity Fewer non-planned measures
11
Appendix B: List of interviewees

[9] [10] [11] [12] [13] [14] [15] [16] [17] [18] XX XX XX XX XX XX XX XX XX XX SRCM SRCM pilot project mangager SRCM SRCM operational staff SRCM operational staff Realised benefits Expected benefits Expected benefits Realised benefits Expected benefits Expected benefits Realised benefits
12
Reliability and Cost Centered Maintenance Methods
Nuclear Power and Reliability Centered Maintenance
(RCM)
Julia Nilsson
julia.nilsson@ee.kth.se
Project in the course 2C5060 Maintenance Management in Power Systems July 9, 2007
Abstract
This is a summarizing report with aim to describe the nuclear power plant Forsmark from a maintenance perspective, and also to describe how Reliability Centered Maintenance, (RCM) is used in nuclear power today. Theory about maintenance and RCM is described in the opening. The maintenance organization at Forsmark is described and the useage of RCM within nuclear power today is presented. Several attempts have been made where RCM is used within nuclear power in Sweden and Finland. At Oskarshamn nuclear power plant for example RCManalyzes are carried out on systems that at system follow-up and trend analysis appeared to have a need for improvements of the maintenance program. At Ringhals RCM-analyzes should be carried out for all safety systems and on some system important for operation of facilities. At TVO in Oikiluoto in Finland a method is used where components and not the systems functions are in focus. All the analyzed nuclear power plants use or have as goal to use their own adjusted RCM-model. At Forsmark for larger systems a method called REM (Reversed Maintenance ) has been adjusted, that today not is in use.
Contents
1 Introduction
1.1 1.2 1.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Questions at Issue . . . . . . . . . . . . . . . . . . . . . . . . . .
3
3 3 3
2 Terminology 3 Theory
3.1 3.2 3.3 4.1 4.2 Preventive Maintenance . . . . . . . . . . . . . . . . . . . . . . . Corrective Maintenance . . . . . . . . . . . . . . . . . . . . . . . Reliability Centered Maintenance, RCM . . . . . . . . . . . . . . Maintenance Organization at Forsmark . . . . . . . . . . . . . . 4.1.1 Maintenance Division . . . . . . . . . . . . . . . . . . . . 4.1.2 Maintenance Process . . . . . . . . . . . . . . . . . . . . . Usage of RCM within Nuclear Power . . . . . . . . . . . . . . . . 4.2.1 RCM at Oskarshamn . . . . . . . . . . . . . . . . . . . . . 4.2.2 RCM at Forsmark . . . . . . . . . . . . . . . . . . . . . . 4.2.3 RCM within the Ringhals Group . . . . . . . . . . . . . . 4.2.4 RCM at Teollisuuden Voima Oy, TVO, in Oikiluoto, Finland Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 6
6 6 7
4 Application and Analysis
10
10 10 10 12 12 13 13 14
5 Closure
5.1 5.2
16
16 16
1 Introduction
Reliability Centered Maintenance (RCM) [1], is a structured approach that focus on reliability aspects when determining maintenance plans, that is to nd a balance between preventive and corrective maintenance. Reliability Centered Asset Management (RCAM) [2] is a development of RCM into a quantitative approach that focus on relating preventive maintenance to total maintenance cost and system reliability. In several countries there has been pointed to that there with the deregulation of the electricity market is a risk that the nuclear power companies focus on minimizing costs for operation and maintenance, and that there because of this is risk for safety. RCM and RCAM here comes in as good tools to balance between the right measures for cost eectiveness and operational safety.
1.1 Background
This report is a project in the course Maintenance Management in Power Systems. It should be investigated how RCM is used within the nuclear power industry in Sweden and Finland today, and especially how Forsmark use maintenance management. Forsmark has been identied as a suitable place for a case study. At Forsmark improvements of the maintenance program are carried out constantly. The aim is to reach a cost eective maintenance with retained availability. Improvements and adjustments are today implemented in other ways but the aim is to introduce RCM.
1.2 Objective
The objective of this report is to see what has been made when it comes to nuclear power and RCM. A comparing study of how RCM is used at dierent nuclear power plants in Sweden and Finland has been made. An analysis of how the maintenance organization looks and how dierent maintenance concepts are used at Forsmark has been carried out.
1.3 Questions at Issue

The maintenance organization should be investigated carefully. Other elds of interest is what costs most to maintain and where the largest risks are. Where can RCM and RCAM be used and where can these methods make a dierence? An important part of the RCAM method is to reconnect with maintenance data. The analyzing tool BI-Cycle has been developed to be able to analyze large amounts of data that have been collected during several years for a large amount of components. Maintenance data can be reconnected with BI-Cycle and components with the largest maintenance cost can be pointed at.
2 Terminology
Terms and abbreviations of dierent concepts are here presented shortly.
RCM
RCM stands for Reliability Centered Maintenance. RCM gives a systematic method to balance between preventive and corrective maintenance, and to chose right preventive maintenance action for right component at the right time to reach the most cost ecient solution.
RCAM
A development of RCM is RCAM, Reliability Centered Asset Management. The method intend to relate closer to the impact of maintenance to the cost and reliability of the system.
LCC
LCC stands for Life Cycle Cost. The method is a economical calculating method and intend to calculate the total cost for a system during its life length. A goal is to minimize the total Life Cycle Cost.
FMEA
FMEA stands for Failure Mode and Eect Analysis. The method was introduced by the aircraft industry and the manufacturer of Boeing in the year of 1957. The aim with the method is to nd all the ways that a system can fail. Three questions should be answered, these are: 1. What failures can occur? 2. What are the eects of these failures? 3. What are the causes of these failures?
PMEA
PMEA stands for Preventive Maintenance Eect Analysis.
MTBF
MTBF stands for Mean Time Between Failure.
LTA
LTA stands for Logic Tree Analysis.
REM
REM stands for Reversed Maintenance and is a method that has been used earlier at large reviews of large systems at Forsmark.
SRCM
SRCM stands for Streamlined RCM and is a method that is used at Ringhals and earlier at Barsebck.
3 Theory
Essential concepts regarding maintenance are presented by way of introduction with a connection to Swedish standard and the standard Dependability Management.
Maintenance
There is not one single correct solution of how the concept of maintenance should be used. In this work it is divided into preventive and corective maintenance. Other examples on denitions of maintenance is presented by Endrenyi et. al. [3], where maintenance is divided into, among other things, scheduled and predictive maintenance. According to Swedish standard [4] maintenance is a combination of all technical and administrative actions during the life cycle of an item intended to retain it in or restore it to a state in which it can perform a required function.
3.1 Preventive Maintenance

Preventive maintenance is maintenance carried out at predetermined intervals or according to prescribed criteria and intended to to reduce the probability of failure or the degradation of the functioning of an item [4]. Preventive maintenance means that the maintenance is planned and periodical. Maintenance is carried out at even intervals to prevent that failures occur. Preventive maintenance can exist of exchange, samples, inspections or rounds. Within nuclear power it is common that a lot of the maintenance is controlled by the authority (SKI), that species demands that must be fullled to secure the facilities security [5]. One does not want to perform more preventive maintenance that is needed, as this takes a lot of resources. Therefore one must work on optimizing maintenance by not having to short intervals between preventive maintenance measures. To perform maintenance to often can lead to that the component is maintained until it brakes. This means that a components failure intensity increases as its function is temporary or lasting after a preventive maintenance measure. Sometimes it is better to not perform preventive maintenance on a functioning component. To determine intervals for preventive maintenance access to historical data is needed. By storing data about failures that occurs in data bases, intervals can be optimized to minimize the amount of failures in relation to the cost for preventive maintenance.
3.2 Corrective Maintenance

Maintenance carried out after fault recognition and intended to put an item into a state in which it can perform a required function [4]. No maintenance is carried out as long as the component is at function. When a component has failed it is exchanged or repaired. A component that is repaired can be repaired to as good as new or in the state that it was in before it failed [5].
Sometimes components are used deliberately until they fail, so called run to failure. The reason can be that it costs more to perform preventive maintenance than not to, which often is right for components that are simple or easy to exchange. A component that deliberately is driven until failure can be categorized as planned corrective maintenance. This is not uncommon in nuclear power. A component that is driven until failure does not always have to be attended to immediately. For these components the maintenance measure is planned to a later stage, e.g. review.
3.3 Reliability Centered Maintenance, RCM

Reliability Centered Maintenance, RCM, gives a systematic method to balance between preventive and corrective maintenance, and to chose the right preventive maintenance activity for right component at the right time and to the right cost ecient solution [1]. RCM has its origin in the civil aircraft industry and Boeing 747. The rst description came by Nowlan in 1978. RCM was introduced in nuclear power in 1980. RCM is known by: obtaining system function, identifying failure modes, prioritize functions and to chose ecient maintenance measures.
Reliability Centered Asset Management, RCAM

Reliability Centered Asset Management, RCAM, is a development of RCM into a quantitative method [2]. The goal is to relate preventive maintenance to total maintenance cost and the systems reliability. The method is developed from RCM principles and intend to closer relate the impact of maintenance on a systems cost and reliability. Quantitative methods are used to at component level see the eects of preventive maintenance on the system. A central feature of the method is that the analysis goes from system level to component level and back to system level. To analyze maintenance aspects of a system and its components, the rst step is to identify the systems units, and which of these that ought to be analyzed.
RCM according to Moubray

According to Moubray seven questions shall be answered when the system units have been identied and which of these that ought to be analyzed [6]: 1. Which are the functions and performances needed? 2. In what way can each function fail? 3. What causes each functional failure? 4. Which are the eects of each failure? 5. Which are the consequences of each failure? 6. How can each failure be prevented? 7. How does one proceed if no preventive measure is possible?
In the rst point it is stressed that when the function is specied it is important to indicate a special level that the unit shall receive. Functions are what an asset is expected to perform but can also be anything that an asset has to obey to as e.g. color or form. Functions are divided into primary and secondary functions. Primary functions describe the main purpose for the asset. Secondary functions describes further functions that the asset should have such as color and safety aspects [6]. It is then indicated in which ways the unit cannot cope with the demands. The origin of the failure is later described and here one must be careful so that this is analyzed on the right level. To much detail can make the process long and costly, while a to low level of detail can make the process meaningless. Then the course of events at a failure be described. Among other things physical or environmental and how to restore equipment should be in the description. The consequences of a failure can be divided into three categorizes: 1. Safety and environmental consequences 2. Operational consequences 3. Non operational consequences If a person can be injured or if the failure can do that an environmental law is broken the consequence is classed as a safety or environmental consequence. Operation consequences aect costs that concern production and operation. Non operational consequences only gives costs as operation. Then a decision tree is used to decide which maintenance that should be implemented. Depending on the consequence classication of the failure and what type of maintenance that is used the optimal maintenance strategy is determined.
The Standard Dependability Management

According to [7] RCM is an approach to establish a preventive maintenance program that eciently shall allow that necessary security is attained for equipment and structure. Nine steps are mentioned: 1. Dene system limits 2. Dene functions for each system 3. Identify critical components 4. Identify relevant functional failure causes 5. Predict the eects and the probability for these failures 6. Use decision trees to categorize eects of failures 7. Identify eective maintenance methods that include he initial maintenance program 8. Possible new design o the process 9. Establishment of a dynamic maintenance program, where dynamic means an update of the program with help of experience data 8
These steps gain as result a maintenance method with aim to:
Keep up the function when it comes to security Keep up natural security and reliability levels Optimize availability Gain necessary information for improvement where natural reliability has shown to be insucient To reach these goals at a minimized Life Cycle Cost, including maintenance costs and costs for remaining failures Gain necessary information to maintain a dynamic maintenance program.
4 Application and Analysis

4.1 Maintenance Organization at Forsmark
In Forsmark 17% of the electricity in Sweden is produced at the three production units of Forsmark; F1, F2 och F3. The company has about 750 employees of which about 180 persons are working at the maintenance unit [8].
4.1.1 Maintenance Division

Forsmark has since the year 2000 a common maintenance unit, FM, that has functional responsibility for the company common maintenance process. The unit performs operative maintenance and assembling activity at the three production facilities. Functional responsibility means to make sure that a company common area works and is developing within the frame for given conditions, which means to develop and suggest changes of maintenance and methods. The orderer, the production unit, shall then, based on decision criteria from FM, prioritize maintenance measures. FM should work as support for the operation management in technical questions and perform labor in a union way that is quality secured according to valid instructions. The planning, FMP, hold together contracts and orderings and leads the internal labor within FM to based on demand from the orderer compile measure program and activity plans that correspond to the orderers demands and wishes. Union responsibility means a responsibility for a ordered support measure but also for continuing experience follow-up, analyzes and advice in the union area. Union responsible unit is the companies professional internal supplier.
4.1.2 Maintenance Process

The maintenance process at Forsmark is divided into two parts: maintenance management and operational maintenance. There is a obvious dierence of the process and the organization that works in the process. Operative maintenance include valuing, planning and implementing with the task to valuing and carry out identied maintenance requirements. Maintenance management includes management, analysis and method development, its task is to be responsible for that there are ambitions strategies, goals and a spoken responsibility division [8]. The step maintenance management consists of the functions management and analysis and method development [9]. Based on given and assumed conditions the management main task is to, with advanced planning and in a systematic way lead, develop and report the activity. The managements work is characterized by a whole sight on facility function, facility worth and own result. The production units, and the unit for common service that is responsible for maintenance of the facilities, denes the demands on the supplier of maintenance measures from the guidelines that are given by the management of Forsmark. The result is an agreement about the maintenance work that is supposed to be
10
planned, resource secured and followed up. The maintenance management operates a long-term and strategic activity, where the time perspective for matter managing usually is expressed in weeks, months or years. The step operative maintenance consists of the functions valuing, planning and implementing. Based on given conditions, measure program and the facilities technical condition is the operative maintenances main task to value and implement identied maintenance need. The work is characterized by knowing about details concerning component function and qualied system- and process knowing. Planning and judgement of necessary resources for the labors implementation. Compared to maintenance management the activity at the operative maintenance is more event ruled. The maintenance organization has as support for its labor a own developed, computer based maintenance system, FENIX, that is based on the facility directory. The system contains labor orders and safety references and make sure that samples occur with the right intervals. The analyze tool BI-Cycle is used for trend analysis [8].
Maintenance Concepts at Forsmark

Within the maintenance at Forsmark several specic concepts are used regarding maintenance measures and maintenance types [9]. The two main measures are preventive and corrective maintenance and have been dened earlier in this work according to Swedish standard. Preventive maintenance is at Forsmark divided into four maintenance types; condition control, condition based maintenance, time based operation and authority controlled maintenance. Corrective maintenance is divided into two sub types; planned and unplanned corrective maintenance. For every maintenance type a sample of dierent maintenance methods are used based on control measures and improvement measures.
Preventive maintenance is dened as a planned maintenance measure performed with purpose to prevent functional obstructions or faults.
Condition control is a maintenance type implemented in purpose to clarify the technical condition. Condition based maintenance is a maintenance type implemented from the technical conditions point of view. Time based operation is a calendar controlled maintenance type that aims to prevent faults/obstacles. Authority controlled maintenance is a maintenance type controlled by authority regulation.
Corrective maintenance is a planned or unplanned maintenance measure caused by functional obstacle or faults.
Planned corrective maintenance is a maintenance type that intend to restore fault functioning equipment to a useful state, where the fault function is accepted or prevented.
11
Unplanned corrective maintenance is a maintenance type that intend to restore fault functioning equipment to a usable state, where the fault function is unwished or un prevented.
4.2 Usage of RCM within Nuclear Power

RCM in Sweden and Finland
Several nuclear power companies within the Nordic countries are using dierent types of RCM as maintenance management. In [5] RCM at dierent nuclear power plants is presented.
4.2.1 RCM at Oskarshamn

At Oskarshamn (OKG) need adjusted RCM-analyzes are performed. This means that RCM-analyzes are performed at systems that at systems follow-up and trend analysis have shown to have a need for improvement of the maintenance program. The improvement need can depend on that the amount of failures have increased, that the system have a large amount of corrective maintenance, that the system is expensive to maintain, that the system has a large amount of corrective maintenance in relation to preventive maintenance or if the system causes high costs at operation hold-up [5]. The RCM-process at OKG consists of four steps: 1. Fault and maintenance history This step aims to use the experience that are collected as historical data of fault reports, maintenance measures and facility condition. 2. Priority of objects for maintenance analysis In this step objects that need a deeper maintenance analysis are prioritized. The objects are classied in three levels: 0,1 and 2. The criteria that the objects are classied for are reactor safety, operation avalilability, working environment and maintenance cost. 3. Identifying change requirement To identify the requirement of changes in the maintenance program Failure Mode and Eect Analysis (FMEA) is performed. Also Preventive Maintenance Eect Analysis (PMEA) is performed with help of analysis tools. Calculations of MTBF-values are carried out with help of BI-Cycle, after that eventual failures in the fault reports are corected. MTBF stands for Mean Time Between Failure. 4. Improvement of current maintenance program When a requirement of changing the current maintenance program has been identied, Logic Tree Analysis (LTA), can be used to identify suitable maintenance measures. When suitable maintenance measures have been identied the components fault characteristic is determined. For objects with increasing fault intensity maintenance intervals are determined depending on the chosen maintenance measure. Possible maintenance measures are periodical preventive maintenance, condition controll or periodical trial.
12
4.2.2 RCM at Forsmark

At Forsmark constant improvement of the maintenance program is in progress. The aim is to attain a cost ecient maintenance with retained availability. Improvement and modications are today carried out in other ways but the aim is to implement RCM. Reassess of a maintenance measure is done according to rm routines that are a part of the process steps valuing, analysis and method development, described in chapter 4.1. At the modication work methods that have its origin in RCM are used. For more extensive modications of systems or larger functions a method that looks like RCM called REM (Reversed Maintenance) has been adjusted, that not are in use today. The REM-analysis starts with the current maintenance program and historical corrective maintenance. At the analysis the following criteria for a system or a function should be valued:
Maintenance cost Operation availability Safety and environmental aect Systems size
4.2.3 RCM within the Ringhals Group

The Ringhals group started in 2003 a large project that is expected to be going on for four to ve years. In this project RCM-analyzes should be performed for all safety systems and for some systems that are important for the operation of the facilities. The most important with the project is to gain a documented maintenance system that describes the optimal maintenance measures for each object, and actual measures and intervals where consideration is taken for authority demands, modication times and connection to other objects. In this way knowledge is secured for future generation changes. A goal with analyzing is to decrease costs for maintenance, but since it in some cases require a increased maintenance to secure function it is not sure that costs initially will decrease for each system. Analyzes are performed according to the method SRCM. The rst step in the SRCM-process is to chose system for analyzes. The starting point is that the system has a safety related function. After that the extent of the analysis is extended with the systems that e.g. have high costs for corrective or preventive maintenance. In a SRCM-analysis some assumptions are made about the system to make the analysis easier. It is common that some components are excluded from the analysis because they are considered to have a small impact on the operation. The Ringhals group has chosen to include all components in the analyzes. Besides this it is normally assumed that the systems in ows are in place and always available. For the chosen systems information and knowledge about the systems are collected in step two by interviews with operation and maintenance personnel and by current documentation of the system. The documentation can consist of ow charts, drawings and maintenance history.
13
Step three consists of dening functions and functional failures, that is make a FMEA. In this analysis equipment, fault condition and failure eects are identied. Thereby how critical each fault is is decided. The critical evaluation criteria are that the fault leads to:
scram decreased power production or availability that a limit value for technical specication is exceeded increased safety risks for the personnel major injury the limit for environmental discharge is exceeded that the public is aected by radiation risk and re.
In step four critical activities are chosen with Logic Tree Analysis (LTA), that aims to identify which preventive maintenance measures that should be applied. Interval for activities are determined by calculating MTBF or by identifying the point where faults can be detected for a condition monitored system. In the next step non-critical components are decided. Non-critical evaluating criteria should decided if: 1. costs of repair or exchange of components that are run to failure becomes high, 2. fault of the component leads to fault in other important components, 3. a simple preventive maintenance measure prevents that the components availability decreases, 4. potential security risks for the personnel increases if the component is run to failure. The last steps consists of bringing the preventive maintenance program for the system and compare to the program that the SRCM-analysis led to. Then the project group gives the analysis to concerned personnel in the maintenance organization that nally decides how the new maintenance program should be implemented.
4.2.4 RCM at Teollisuuden Voima Oy, TVO, in Oikiluoto, Finland

TVO:s maintenance process consists of their own developed method for maintenance analyzes. The method is built on RCM but focus in the analyzes are at components and not the systems functions. The reason to this choice is that the analyzes should be implemented by the maintenance personnel that have good knowledge about the facilities components. At TVO it is important to document the experience that is with the personnel out in the facilities. TVO:s method for maintenance analyzes consists of three steps: 14
1. classication of components 2. choice and motivation of maintenance measures 3. overhaul of the amount of spare parts in store. In TVO:s computerized maintenance system there are a maintenance module. Components are classied and categorized in four dierent priority groups with help of a decision tree. Priority 1 is given to components that always needs to work for the operation or safety. Priority 2 is given to components that aects operation or safety, but that not demands that the electricity production stop. Priority three is given to components that are expensive to repair but that not are important for operation or safety. Priority four is given to the components that not has a need for preventive maintenance measures, that is they do not aect safety or operation and are simple or easy to repair. The priority is managed by which function that the component should fulll. The step to chose maintenance measure is built on the FMEA that is carried out in the previous step. The failure modes are analyzed and if possible data is produce about MTBF, repair times and so on. Next step is to produce maintenance measures for the identied failure modes. It is here pointed that every maintenance measure must prevent a failure. After that it is analyzed if there are anything that can be improved by e.g. condition monitoring. Finally a decision is taken about chosen maintenance program for studied component.
15
5 Closure
5.1 Conclusions
Maintenance concepts at Forsmark can be connected to Swedish standard as it is divided into preventive and corrective maintenance. Preventive maintenance is divided into e.g. condition control and condition based maintenance as in Swedish standard but at Forsmark even authority controlled maintenance occurs. Corrective maintenance is divided into planned and unplanned corrective maintenance. Several attempts have been made where RCM is used within nuclear power in Sweden and Finland. At Oskarshamn nuclear power plant for example RCManalyzes are carried out on systems that at system follow-up and trend analysis appeared to have a need for improvements of the maintenance program. At Ringhals RCM-analyzes should be carried out for all safety systems and on some system important for operation of facilities. At TVO in Oikiluoto in Finland a method is used where components and not the systems functions are in focus. All the analyzed nuclear power plants use or have as goal to use their own adjusted RCM-model. At Forsmark for larger systems a method called REM (Reversed Maintenance ) has been adjusted, that today not is in use.
5.2 Future Work

To further investigate how RCM is used within nuclear power today is an interesting topic for future work. To take a closer look at a special component or subsystem and perform a LCC-analysis for that special component or subsystem also is of great interest since the concepts of LCC and RCM has great impact on each other according to e.g. the standard Dependability Management.
16
References
[1] L. Bertling. Funktionsskerhetsinriktat underhll, reliability centred maintenance, tillfrlitlighetsanalys fr elkraftssystem fr vattenfall eldistribution. Lecture in the course 2C4530 A-ETS/EEK-0504, 2005-10-24. [2] L. Bertling R.N. Allan and R. Eriksson. A reliability-centred asset maintenance method for assessing the impact of maintenance in power distribution systems. IEEE Trans. Power Syst., 20(1):7582, Feb 2005. [3] J. Endrenyi et al. Ieee/pes task force on impact of maintenance strategy on reliability. "the present status of maintenance strategies and the impact of maintenance on reliability". IEEE Trans. on Power Systems, 16(4):638646, November 2001. [4] Maintenance Terminology, Swedish Standard SS-EN 13306. SIS Frlag AB, Stockholm, 2001. [5] S. Landqvist. Jmfrelse av underhllsmodeller vid de nordiska krnkraftverken. Master's thesis, KTH, 2004. [6] J. Moubray. RCM II. Butterworth-Heinemann, Oxford 1995. [7] IEC. Dependability Management 60300-3-11 Application Guide-Reliability centred maintenance, rst edition, 03 1999. [8] L. Kecklund. Underhllsstrategier och skerhet p en avreglerad elmarknad. Rapport 2004:40, SKI, 2004. [9] J. Lundvall. Instruction for the maintenance process. Forsmark.
17
Implementation of production assurance programmes in a production plant: concept and discussion

Javad Barabady
Lule University of Technology, Sweden
Abstract Recently, there has been a high interest in utilization of production assurance programmes in various industries. The programme describes the production assurance activities necessary to fulfill the objectives, how they will be carried out, by whom and when. The activities provide input to decisions regarding concept, design, manufacturing, construction, installation, operation and maintenance. Such programmes are petroleum oriented and play a significant role to support the decision-making process. It is a challenge to utilize such programmes in other areas of applications. The purpose of this research study is to provide an overview of production assurance programme which can be useful for application areas other than petroleum industry especially for power plant industries as well as distribution of power. The paper reviews and discusses the most existing knowledge related to the production assurance programmes. 1. Introduction During the last few years, the production assurance programme (PAP) has experienced a faster development, and it increasingly plays an important role to support the decision-making processes for manager and engineers. Production assurance is a term used to describe how a system is capable of meeting demand for deliveries or performance. The production assurance concept is built on several other concepts such as availability, reliability, maintainability, supportability, deliverability and so on. Some of these factors as well as their relationships are illustrated in Figure 1.
Availability (Item) Availability (system)
Consequence of item failure
Configuration Utilities etc.
Production Availability
Consequence for production Capacity Demand etc.
Deliverability
Reliability
Design Tolerances Design margins Quality control Operating Conditions etc.
Maintainability
Organization Resources Tools Spares Accessibility Modularization etc
Supportabillity
Logistics Spare parts Tools Personnel Equipment etc.
Compensation
Storage Substitution etc.
Uptime
Downtime
Figure 1: Illustration of relationship between some production assurance terms (adapted from NORSOK Z-016)
Such programmes are based on production assurance (also referred to as regularity, or more specifically as production availability) concept. Those programmes are well-known and have been applied by the Norwegian oil and gas industry for many years (Aven, 1987; Rausand, 1998; Signoret, 1998; Kawauchi & Rausand, 2002; Enrico et. al., 2006). A key document for these programmes is Norsok Z-016 (1998), the standard established and used by the Norwegian oil and gas industry as a guideline for PAP. It covers the analysis of reliability and maintenance of components, systems and operations associated with exploration drilling, exploitation, processing
and transport of petroleum resources. However, there is not much literature available regarding this subject for other industries. There are work covering reliability programmes for different industries, but not specifically addressing production and production availability. We refer to Knowles et al. (1995), Klinger et al. (1992), Hagen (2006), Pecht et al (2002), Guthrie et al,(1990) and Ke & Hwang (1997). We would also like to draw attention to the standards IEEE Std 9331999, NASA-STD-8729.1 (1998), IEEE Std 1332-1998, ISO/CD 20815-2005, IAEA-TECDOC1264 (2001), IEC 60300-1 (1993), IEC 60300-2 (1995) and IEC 60300-3-1 (1991). Their main focus is the process of ensuring a reliable product. We believe that the concept of PAP is useful and can be applied in other industries. However, It is in our view that the concept of PAP has not yet been sufficiently developed. Therefore, to transfer the concept of PAP from Norwegian oil and gas industry to other areas of applications first we have to clarify it. The purpose of this research study is to review and discuss existing knowledge of production assurance concepts in order to provide an overview of different issues of PAP; it could be useful to apply in new areas of applications such as mining, hydropower, and railway. This paper is explanatory in nature. The method used for this paper is mainly literature study. Most of the information and data have been taken from relevant articles and books. In addition, conversations with experts have been carried out. The discussion is endeavoring to answer questions such as what is the concept of PAP?, what are the main elements of PAP?, what types of performance criteria and importance measures to be used?, how to define and use those performance criteria?, What is the concept of production assurance programmes? and etc. The rest of the paper is organized as follows. In Section 2, we briefly review the basic concept of production assurance. The purpose of the presentation is not to give a complete overview of the theories, but to introduce the reader to relationship of the concept with dependability concept. We cannot just refer to textbooks or other papers, as the points we would like to make are not addressed in the literature in a way that we can utilise. In section 3, we will clarify different elements of PAP. We discuss different issues related to each elements of PAP using some examples form the offshore industry as illustrations. Section 4 is presented a methodology for implementation and execution of PAP in a production plant. In section 5, we discuss and conclude the paper. 2. Definition and concept of production assurance For a production plant the concept of system availability does not give information about the capacity and production rate. For example, let us assume that the system availability is 90% for a specific period, but the system capacity and the production rate may be less than desired. This could for example be caused by process bottlenecks, reduced efficiency due to aging, reduced effectiveness due use of old technology, ineffective process organization, etc. As a result the production availability performance goals may not be achieved. This means that the concept of system availability alone is not a good performance measure. We believe the most appropriate measure for the performance of a production plant is production availability performance. In the following the underlying concepts of production assurance performance are discussed. 2.1. Reliability performance The formal definition of reliability according to IEV (191-02-06) is the ability of an item to perform required function under given conditions for a given time interval. As the required function may vary under different modes of operation a number of reliabilities for a certain item may be considered. Furthermore, the given conditions may vary under the different operating modes due to various factors such as operating environment (temperature, corrosiveness,
dust, etc.). Hence, it is essential to identify various foreseeable conditions and operating modes, as well as item (system, equipment, component, etc.) use and misuse in the requirements specification phase of system design. In operation phase the reliability of a system can be improved through modification of system including operational and maintenance programs. Proper equipment maintenance and operation can assist in ensuring that the design in reliability performance is achieved and avoid failures as well as reduce the cost. The reliability performance characteristics can be designed during the design phase and aim is to prevent the occurrence of failures as far as possible. 2.2. Maintainability performance Maintainability formally is defined as the ability of an item under given conditions of use, to be retained in, or restored to, a state in which it can perform a required function, when maintenance is performed under given conditions and using stated procedures and resources (IEV (191-02-07). It is noted that maintainability may be used as a measure of maintainability performance. It should be noted that the procedures and resources, as well as the conditions the maintenance is performed under plays an important role in retaining or restoring the item to a satisfactory state. Often systems are designed, built and tested in an environment with a comfortable temperature and good lighting. However, in real life maintenance is often performed in a harsh and remote environment, e.g. bad weather conditions, geographic distance and infrastructure (Gao and Markeset, 2007). Thus, the operating situation needs to be taken into account when designing for good maintainability. 2.3. Maintenance support performance Maintenance support performance is defined as (IEV 191-02-08) the ability of a maintenance organization, under given conditions to provide upon demand the resources required to maintain an item, under a given maintenance policy. Defining and developing maintenance procedures, procurement of maintenance tools and facilities, logistic administration, documentation, and development and training programmes for maintenance personnel are some of the essential features of a maintenance support system. Furthermore, for complex, advanced and integrated production systems one often need external support from for example the original equipment manufacturer to provide expert assistance, field services, spare parts and tools, as well as training of operation and maintenance personnel, etc. Thus we can see that maintenance support performance is part of the wider concept of product support, including support to the product as well as support to the client (Markeset, 2003). Reliability, maintainability, and maintenance support performance together constitute the basis for availability performance. 2.4. Availability performance The formal definition of availability performance is the ability of an item to be in a state to perform a required function under given conditions at a given instant of time or over a given time interval, assuming that the required external resources are provided (IEV 191-02-05). There are various measures of availability performance defined in literature, e.g. instantaneous availability, asymptotic availability, asymptotic mean availability, and mean availability. We refer you Barlow & Proschan (1975), and IEV (2006). All of these measures are based on the function X(t), which denotes the status of a system at time t. For example the instant availability at time t (or point availability) is defined by: A(t ) = P( X (t ) = 1) (1)
This is the probability that the system is operational at time t. Since it is very difficult to obtain an explicit expression for A(t), other measures of availability have been proposed. One of these the most frequently used availability measures is the Steady-state availability or limiting availability which is defined the mean of the instantaneous availability under steady-state condition over a given time interval (IEV, 2005-10-08) express as: A = Lim A(t ) (2)
t
This quantity is the probability that the system will be available after it has been run for a long time, and is a significant measure of performance of a system. Often steady state availability is also defined depending on whether the waiting time or the preventive maintenance times are included in or excluded from the calculation. Therefore, depending on the definitions of uptime and downtime, there are three different forms of the steady-state availability, namely inherent availability, the achieved availability and the operational availability. We refer you to Ebling (1997); and Blanchard et. al (1995) for further discussion. 2.5. Production assurance performance Production assurance is a term used to describe how a system is capable of meeting demand for deliveries or performance. Production availability, deliverability or other appropriate measures can be used to express production assurance. Deliverability is defined as the ratio of deliveries to planned deliveries over a specified period of time, when the effect of compensating elements such as substitution from other producers and downstream buffer storage is included (NORSOK Z016). The production assurance performance of a production plant may be measured in many different ways. Consider a production plant system, the production rate at time t is denoted by D(t) which can be defined as the number of items e.g. barrels of oil that are produced per time unite at time t. The planned production rate for such production plant is denoted by D0(t). In this section, we present some alternative which can be used to express production assurance performance for such production plant using existing literature Aven (1987), NORSOK Z-016 (1998), Rausand (1998), and ISO/Cd 20815 (2005). a) The first alternative for the production assurance performance is the throughput availability, A1(t1, t2), which is mean actual production in (t1, t2) divided by planned production in(t1, t2). It can be formulated as:
t2
A1 (t1 , t 2 ) =
Mean actual production in (t1 , t 2 ) Planned production in (t1 , t 2 )
t1 t2 t1
E ( D(t ))dt
(3)
D0 (t )dt
Throughput availability is also referred as throughput capacity which is calculated by the throughput volume divided by a fixed reference throughput volume for an item, subsystem, or system. Different alternative can be used as a fixed reference throughput e.g. the design capacity of the item, subsystem, or system. We refer to ISO/CD 20815 (2005) for discussion. If for example in a mining company, A1(t1, t2) = 1, this mean we expect to produce 100% of the planned production in the interval (t1, t2). If the production is stopped in part of the interval, we may be able to achieve 100% by increasing the production beyond the plan level in the rest of time interval. It should be noted that an optimal buffer inventory level must be built up to guarantee the continuous supply during stopping time in order to satisfy the demand. Therefore, throughput availability in some cases may be greater than 1.0, which means that we expect to produce more than planned production. 4
b) The second alternative for the production assurance performance is the demand availability which is mean production of time in (t1, t2) where D(t)>D0(t). It can be defined as: A2 (t1 , t 2 ) = 1 2 P(D(t ) D0 (t ) )dt t 2 t1 t1
t
(4)
The demand availability expresses the expected proportion of time the throughput exceeds the demand. If for example for a power plant, A2(t1, t2)=0.98, this mean we are meeting the planned production ( i.e. D(t)>D0(t) ) in 98% of the time. It gives no information about the production in the remaining 2% and the same result is obtained for A2 when the production is totally shut down as when the production is slightly less then D0(t). c) The third alternative for production assurance performance is to use the on-stream availability. The on-stream availability, A3, equals the expected proportion of time the throughput is greater than zero, i.e. t 1 2 A3 (t1 , t 2 ) = P(D(t ) > 0)dt (5) t 2 t1 t1 d) The fourth alternative for the production assurance performance is the production availability at level c at time t, Ac(t). It can be defined as the probability that the throughput is at least at level c at time t, i.e., Ac(t) = Pr{D(t) C} (6) Relative to a level c, we can define an availability measure as the (expected) proportion of time the system state is at least c. e) The fifth alternative for the production assurance performance is the average system availability, if the production plant system is restored an as good as new condition after each failure, which equals the mean proportion of time the system is functioning, can be written as
AO = MTBM MTBM + MDT
(5)
where the mean time between maintenance operations (MTBM) includes both unscheduled and preventive maintenance, and MDT is the mean maintenance down time and includes maintenance time, logistics delay time, and administrative delay time. From above-mentioned alternative, the most commonly used is the throughput availability (throughput capacity). However, the most adequate measure for production assurance performance depends on actual situation and type of production plant. Prediction of production rate is main step for calculation of production assurance performance. It influence by capacity performance and system availability performance. Throughput capacity distribution (PDC) can be used as a quantitative measure of the production assurance performance which is a probability distribution of the possible throughput capacity level for component, subsystem of system. Let us look at an example. Consider a simple numerical example of a dehydrator unit of an offshore installation producing oil and gas; it consists of three parallel dehydrators, which are illustrated in Figure 2.
Dehydrator (DEH-A) Dehydrator (DEH-B) Dehydrator (DEH-C)
Figure 2. System configuration of the dehydrator unit Each dehydrator is assumed to have a throughput capacity of either full or 0. Therefore, the possible capacity performance levels for dehydrator unit are thus 3c, 2c, c, and 0. Note that c is the capacity performance of each dehydrator. The Markov diagram for three components in parallel may be as figure 3. The necessary input data, i.e. the failure rate and repair rate have taken from the OREDA handbook (1997). The data presented in table 3.
Figure3. Markov diagram for dehydrator unit which is included three identical and independent dehydrators, each with capacity performance level C.
Table 3. Failure rate, repair rates, and capacity performance for all components
Name of unit Failure rate Repair rate HPS-A HPS-B HPS-C 8.9110 8.9110 8.9110
-5 -5 -5
Capacity
2 10 6 Sm /d
3 3 3
Design capacity Capacity performance

2.5 10 6 Sm /d
3 3 3
2.5410 2.5410 2.5410
-3 -3 -3
80 % 80 % 80 %
2 10 6 Sm /d 2 10 6 Sm /d
2.5 10 6 Sm /d 2.5 10 6 Sm /d
The planned production rate for DEH unit is 6 10 6 Sm3/d. Based on the above-mentioned information the throughput availability, demand availability and on-stream availability can be calculated. In this example the throughput capacity has a discrete distribution. The result of calculation is given in Table 4.
Table 4. The PDC for the dehydrator unite obtained by Markov model Throughput capacity level (%) 0 33 67 100 Probability
3.95 10 4 2.38 10 2 7.6 10 2 9.08 10 1
The throughput availability, demand availability and on-stream availability of DEH unit are 95.8%, 90.8%, and 99,5% respectively. It can be concluded that the planned production rate is met in 90.8% of period and 9.2% of period system work with less than planned production rate or it failed.
3. Production assurance programme (PAP) This section presents and discusses different elements of PAP based on the existing literature. The objectives of PAP all related to optimum overall economy of the equipment during a systems life cycle phases, reduction of health, safety and environmental risks, quality improvement, as well as considerations related to human factors and customer satisfaction. A PAP is influenced by the policies of the organization, the product being developed, and organizationally-unique practices. These policies cause the PAP to vary from organisation to organization and from product to product. However, there are some common features, and these will be examined in the following. As there is not a formal definition of a PAP, first we start with definition of a PAP: A PAP is a formal management system, which assures the collection of important information about plant performance throughout each phases of its life, and directs the use of this information in the implementation of analytical and management processes which are specifically designed to meet two specific objectives: Confirm that the plant is expected to meet, or continues to meet, each of the performance goals such as production availability, deliverability, technical integrity and reliability to reduce the total business risks and to meet customer demands and profit goals. Facilitate identification of opportunities and cost effective ways to implement and execute improvement actions needed to enhance production availability performance, reduce risks and uncertainties, and improve profits, effectiveness, efficiency and productivity. The PAP will focus on the expenditure of resources available for performance improvement in areas, where the economic return is largest, or to optimize the design and operation of the plant. A typical production availability programme can be expected to have followings four broad functional elements: Goals, performance criteria and requirements, Management program and implementing procedures, Analytical tools and investigative methods, and Information management. These elements discusses in the following section. i) Goals and performance criteria: The PAP requires the definition and assignment of a broad set of high level plant goals and performance criteria which can be used for comparison with actual or predicted plant performance. The goals and criteria may be either deterministic or probabilistic. We will return to this issue in Section 3. For a recent and general reference on how to define appropriate goals and criteria, see Aven et. al. (2006) and the references therein. See also Hjorteland & Aven (2003) and Hjorteland et al. (2006). ii) Management and implementing procedures: Management is a continuous activity throughout all phases of an engineering project. The following two management tasks need to be performed to ensure that the PAP becomes an effective tool: i) Develop, maintain, and implement a PAP plan, expressing what to do, how to do it, and who is to do it. ii) Establish and maintain a PAP review process: The PAP reviews should be conducted by experts who are not directly involved in the project team. According to NORSOK Z-016 important tasks of management are to monitor the overall production availability level, manage reliability and maintainability of critical components and continuous identification of the need for production availability activities. A further objective of production availability management is to contribute with practical technical or operational recommendations. In order to fulfil these objectives, technical and operational means may be used
during design or operation to change the production availability level. Production availability management must include surveillance of project activities and decisions which may have an undesired effect on production availability. iii) Analytical tools and investigative methods, Production availability analyses: The question is how the system will perform in the future from a production availability point of view. Analysis of the system and modelling will increase the knowledge and hopefully reduce uncertainties. Production availability analyses provide a basis for decisions concerning choice of solutions and measures to achieve an optimum economy within the given constraints. The analyses must be performed at a point in time when sufficient details are available. However, results must be presented in time for input to the decision process (NORSOK Z-016, 2003). A PAP uses a set of investigative and analytical and/or simulation methods and approaches to calculate and maximize the reliability, availability and capacity of important subsystems or components. Examples of such methods and approaches are given in the coming section. iv) Information management: The effectiveness of the PAP depends on the quality and accessibility of the information used to provide the feedback to the management systems about how well the system is performing and where to obtain improvements. Collecting production availability data is costly and this effort needs to be balanced against the intended uses and benefits. A typical data collection process may consist of collecting data from different sources into one database where type and format of data are pre-defined (ISO/DIS 14224). It is referred to Markeset and Kumar (2003) for some key factors influencing the management of reliability, availability and maintainability data and information systems. 4. Implementation and execution of production assurance programmes A PAP program needs to be implemented and executed in such a way that it provides for an effective basis for management to make decisions. A methodology need to be established to ensure that the production plant has a cohesive and cost-effective PAP, and should reflect production availability performance requirements as well as activities defined in view of actual needs with respect to personnel, resources, budget, interfaces, milestones, and access to data and information. The basis feature of the PAP implementing and execution procedure is illustrated in Figure 4. The methodology is based on three main tasks, namely: Establishing PAP requirements; Providing input to the design processes; and Monitoring PAP achievements and publish periodic report.
When the PAP requirements for the production plant are established based on the goals and criteria (Task 1), the PAP should address them by providing input to design process and operation (Task 2). A primary design is developed to meet these PAP requirements. For the design process, this means using a variety of production availability-related design techniques or tools (e.g. such as Redundancy and Diversity, Modularity and Diagnostics, and Reliability vs. Maintainability Trade-off Studies) to help establishing a design that meets the qualitative and quantitative PAP requirements. This design is then evaluated to see if the PAP requirements for the production plant are being met and also to identify critical functions and design elements (Task 3). If the results of Task 3 indicate that the PAP requirements for the production plant will not be reached, the management must decide whether the PAP requirements for the production plant are attainable. If the requirements are attainable the design should be revised in an iterative process until the PAP requirements are met. If the PAP requirements are not attainable, they should be revised. If these requirements cannot be revised in a satisfactory way, then the goals should be either revised or
cancelled. In the design process, different alternatives should be generated and their performance evaluated, the aim being to balance the pros and cons of the alternatives (Aven et al. 2006).
Figure 4. Assessment of PAP program (Barabady and Aven 2007) For the operation phase, this means having programs in place, such as a Spare-Part Program, and a Preventive Maintenance Program, to help ensure that the required production availability performance is achieved. Preparation for operation will be activities which will be run in design phase and other phases. The following activities shall be performed: Prepare maintenance programme utilising reliability and regularity knowledge Prepare spare parts programme Prepare plans and systems for regularity and reliability data collection.
In operation phase, the production plant should be monitored and evaluated (task 3). If the results show that the requirements will not be met and the PAP requirements will be attainable, the design or operation and maintenance strategies should be revised and re-evaluated again. In the operation phase, observations of performance of the installation should be made to evaluate the need for measures, modifications, and improvements.
4. Discussion and conclusions The paper indicates that the ccombinations of capacity performance and dependability concept can be used to describe production assurance performance. Therefore, the production assurance performance gives a more comprehensive measure of a production plants real performance compare to system availability performance. The most adequate measure of production assurance is depending on actual situation and type of production plant. However, the most commonly used is the throughput availability. The section 2 of the paper indicates that i) the throughput availability is useful alternative for production assurance performance in those production plant which produce to buffer e.g. a mine production system; ii) the demand availability is relevant alternative for production assurance performance when we are judging the production assurance performance just in time production, e.g. a gas production system; and iii) if the production system is restored an as good as new condition after each failure, the average system availability performance can be used as a good performance measures of the production plant. It is relevant only to distinguish between full production availability (100%) and no production availability (0%). In describing the production assurance of a given system it is necessary to specify four things: the component failure rate; the repair rate or maintenance process; component capacity performance; and system configuration which describes how the components are functionally connected and the roles of operation. Hence, before the discussion on the production assurance model is meaningful the above four characteristics must be explored. In the PAP, the starting point is the definition of goals and requirement. Goals are more internally driven and may or may not be met. Requirements, on the other hand, are more specific and are customer driven. Goals can be modified to satisfy requirements. It should be noted that the goals have different dimensions, which may be in conflict. Thus, an overall evaluation needs to be made balancing all the dimensions to obtain a good decision. It is recommended to consider following criteria for establishing goals: i) Attainability: Goals should be set at levels reasonably attainable within the available time span. Large goals over long periods should be avoided to maintain interest and commitment. Sub-goals over shorter periods are more attainable and more cost effective. ii) Measurability: Goals provide standards against which performance may be assessed and, therefore, should be selected for suitability and defined in a way that enables measurement. iii) Supportability: Support and resources must be available at the time they are needed to achieve goals. Transforming overall PAP goals to specific requirements is complicated. It is a challenge to identify proper requirements. The paper indicates that the focus should be on establishing PAP requirements for those functions which are important for achieving the goals, and then on translating these requirements to the systems designed to accomplish these functions. For example, consider an offshore installation producing oil and gas, there are some functions e.g., the gas, separated by the separation unit from the water and oil coming from the well, is first compressed by two turbo-compressors (TCs), then dehydrated through a triethylene glycol (TEG) unit and finally exported. Based on this function the sub-level requirement can be defined as: the separation unit and the triethylene glycol (TEG) unit shall have a minimum probability at a demand (e.g. 3 6 10 6 Sm /d) equal to x. Instead of a sharp level of requirements, ranges of probability may be used. For the x %, requirements to be meaningful, it must not be seen as a sharp line; we should always look for alternatives and then evaluate their performances and depending on the situation we may accept different levels of requirements. This principle can be applied for all functions and by this the high-level requirements (which could be formulated as confidence in meeting the goals) are ensured. In general, the PAP must focus on the functions or specific tasks and missions 10
of design elements instead of the design elements themselves. This is because a particular design element can perform several functions and the consequences of the failure of one function are likely to be quite different from the consequences of the failure of another function. The question is how the system will perform in the future from production assurance point of view and how uncertainty can be reduced using the concept of PAP? It is important to ensure that the PAP information is adequate for decision-making and that the information is provided in a costeffective and timely manner. It is a challenge how to treat identified uncertainties when conducting production assurance analyses for a production plant. The uncertainty in the predictions made during the initial design stages would be relatively high because of the lack of design detail. As the design becomes more detailed, relevant generic data can be used to obtain more accurate predictions. Finally, as sufficient testing and operating data become available, the uncertainties in the predictions will be further reduced. It should be also noted that the effectiveness of a PAP depends very much upon the quality, accessibility and fidelity of the information used to provide the feedback to the management systems about how well it is performing, and where to make improvements. The experience from the operational phase of a production plant shall be transferred to parties involved in the design phase in order to stimulate improvements in design of new similar production plant. This includes a review of assumptions made for the predictions in the design phase in comparison to actual conditions experienced during operation, including operational principles and maintenance logistics. It seems that the production assurance concept is related to the concept of Overall Equipment Efficiency (OEE) and Overall Line Efficiency (OLE) which are frequently used in component and equipment manufacturing companies. However, the concept of production assurance performance is broader and applicable to a wide range of manufacture and production process. Hence, it should be clarified by doing a comparison between them. As a final result, establishing the PAP will help ensure that the production plant has a cohesive and cost-effective production assurance programme, reflecting a production availability process and activities defined in view of the actual needs, available personnel resources, budget framework, interfaces, milestones, and access to data and general information. 4.1. Application of PAP for power system Making correct decision in a dynamic business environment is a major challenge for the process plant engineers and mangers. The dynamic business environment is characterized by short term and long term uncertainties in business process combined with short term focus on meeting customers and shareholder requirements. From other point of view, customers are interested in purchasing product as per specification, in right quantity and in right quality at the specified time. In order to meet customer requirements, we need production processes and deliver the product as per specifications. The power plant systems are large, complex, and integrated. Failures occur more or less frequently. The consequences of failure are high cost of maintenance, possible loss of production, and exposure to accidents, etc. Such situation necessitates a need for tools and engineering (and business) solutions to minimise the total business risk through assurance of world class performance of the production plants. Therefore, there is a visible and significant need to analyses production assurance and uncertainty analysis and management in a power plant. The study indicates that the production assurance programme can be used as a tool for meeting these challenges in a power plant. Using the concept of PAP for a power plant hopefully provides better predictions of regularity and production availability both for existing infrastructure and future developments. The usefulness of
11
such programmes can be manifold, both in engineering design and in the operational phase: production plant: Compare different alternatives with respect to production availability and deliverability. Identify relevant cost parameters Predict production levels for financial planning, cost/benefit evaluations and sales contract negotiations. Identify critical items requiring spare part attention and special requirement, and form input to spare part planning. Evaluate maintenance and operating strategies to see their effects on field performance. Identify system bottlenecks, vulnerabilities and components with unnecessary over-capacity. Evaluate the availability effects of system modifications, e.g. equipment redundancy and capacity modifications.
Analysing and modelling of production assurance of a power plant will increase the knowledge and hopefully reduce uncertainties and would be helpful for making decision about the process by a decision-maker. The decision maker facing some decision points in a production plant. These decision points include problem and opportunities, related to production availability, cost as well as uncertainties and safety. Having identified the main decision points, adequate decision alternatives need to be generated and assessed, relating to whether or not to execute an activity, alternative maintenance policies, etc. Focus must be on decision making in situations which are involving high risks and large uncertainties, and such decision-making is difficult as it is hard to predict what would be the consequences (outcomes) of the decisions. Therefore, in order to improve production availability and reduce uncertainties emphasis should be on the generation of different alternatives and analysis of such alternative in order to provide decision support, balancing all the pros and cons of the alternatives.
References
Aven, T. 1987. Availability evaluation of oil/gas production and transportation systems, Reliability Engineering 18(1): 35-44. Aven, T. and Hjortrland, A. (2003)A predictive Bayesian approach to multistate reliability analysis, International Journal of Reliability, Quality and Safety Engineering, v 10, n 3, September, 2003, p 221-234 Aven, T., Vinnem, J.E. & Red, W. 2006. On the use of goals, quantitative criteria and requirements in safety management. Risk management 8:118-132. Barabady, J. and Aven, T. 2007. Implementation of production availability programmes in engineering projects , ESREL 2007. Barabady, J. and Markeset, T. 2007. Review and discussion of production assurance programmes, submitted for publication in an international Journal. Barlow, R.E., Proschan, F., Statistical theory of reliability and life testing probability models, Holt, Rinehart and Winston, Inc, 1975 Blanchard, B. S., Verma, D. and Peterson, E.L. (1995). Maintainability: a key to effective serviceability and maintenance management, John Wiley & Sons, Inc., New York. Ebeling, C.E. (1997), An introduction to reliability and maintainability engineering, Mc Graw-Hill, New York Enrico, Z., Baraldi, P. & Edoardo, P. 2006. Assessment of the availability of an offshore installation by Monte Carlo simulation. International Journal of Pressure Vessels and Piping 83(4): 312-320 Gao, X. and Markeset, T. 2007. Design for production assurance considering influence factors, ESREL 2007.
12
Guthrie, V.H., Farquharson, J.A., Bonnett, R.W. & Bjoro, E.F. 1990. Guidelines for integrating RAM consideration into an engineering project. IEEE Transactions on Reliability 39(2): 133-139 Hagen, W.F. 2006. Effects of a reliability program on machine tool reliability, Proceedings Annual Reliability and Maintainability Symposium, Page(s):481-485 Hjorteland, A. & Aven, T. 2003. Reliability analysis and observables, Published in proceedings of the 14th European Safety and Reliability Conference (ESREL 2003) Vol. 1, ISBN 90 5809 551 7, Maastricht. Hjorteland, A., Aven, T. and steb, R. 2006. Uncertainty treatment in production assurance analyses throughout the various phases of a project, Reliability Engineering and System Safety, to appear. IAEA-TECDOC-1264. 2001. Reliability assurance programme guidebook for advanced light water reactors. printed by the IAEA in Austria. IEC 60300-3-10 (2001). International Standard IEC 60300-3-10: Dependability management Part 3-10: Application guide Maintainability. IEC, Geneva. IEC 60706-2 (2006). International Standard IEC 60706-2: Maintainability of equipment Part 2: Maintainability requirements and studies during the design and development phase, 2nd ed., IEC, Geneva. IEC 60706-3 (2006). International Standard IEC 60706-3: Maintainability of equipment Part 3: Verification and collection, analysis and presentation of data, 2nd ed., IEC, Geneva. IEEE Std 1332-1998, IEEE Standard Reliability Program for the Development and Production of Electronic Systems and Equipment, New York. IEEE Std 933-1999. IEEE Guide for the Definition of Reliability Program Plans for Nuclear Power Generating Stations. IEV 191 (2007). International Electrotechnical Vocabulary (IEV) Online, Chapter 191: Dependability and quality of service. http://std.iec.ch/iec60050 (accessed January 2007). ISO/Cd 20815, Petroleum, petrochemical and natural gas industries-Production assurance and reliability management ISO/DIS 14224. 2004. Petroleum, petrochemical and industries- Collection and exchange of reliability and maintenance data for equipment. Kawauchi, Y. & Rausand, M. 2002. A new approach to production regularity/availability assessment in the oil and chemical industries Reliability Engineering & System Safety 75( 3): 379-388 Ke, H-Y. & Hwang, C-P. 1997. Reliability programme management based on ISO 9000, International Journal of Quality & Reliability Management. 14(3): 309-318. Klinger, D.J., Saraidaridis, C.I. & Vanderbei, K.S. 1992. Reliability program management: Today and tomorrow, Proceedings Annual Reliability and Maintainability Symposium, Page(s): 85-92 Knowles, I., Malhotra, A., Stadterman, T.J. & Munamarty, R. 1995. Framework for a Dual-use reliability program standard, Proceedings Annual Reliability and Maintainability Symposium, Page(s):102 105 Markeset, T. & Kumar, U. 2003. Integration of RAMS and risk analysis in product design and development work processes: A case study. Journal of Quality in Maintenance Engineering. 9(4): 393-410 Markeset, T. 2003. Dimensioning of Product Support - Issues, Challenges, and Opportunities, Doctoral Thesis No 6, University of Stavanger, ISBN 82-7644-197-1, ISSN 1502-3877 NASA-STD-8729.1. 1998. Planning, developing and managing an effective reliability and maintainability (R&M) program. NORSOK Standard Z-016. 2003. Regularity management and reliability technology. OREDA. 1997. Offshore reliability data. 3rd ed. Trondheim, Norway: SINTEF industrial Management Pecht, M., Das, D. & Ramakrishnan, A. 2002. The IEEE standards on reliability program and reliability prediction methods for electronic equipment, Microelectronics Reliability 42: 1259-1266. Rausand, M. 1998. Introduction to reliability engineering, In: C.G. Soares, Editor, Risk and reliability in marine technology: 371381. Rotterdam: Balkema. Signoret, J-P. 1998. Availability of petroleum installation by using Markov processes and Petri net modelling, In: C.G. Soares, Editor, Risk and reliability in marine technology: 455472. Rotterdam: Balkema.
13
Theme 3 Condition monitoring and diagnostics methods.

Oil condition monitoring of power generators by Saurabah Kumar Modeling and experimental verification of CC overhead MV networks for PD detection by Murtaza Hashmi Assessment of on-condition maintenance by Alireza Ahmadi
OIL CONDITION MONITORING OF POWER GENERATORS

Saurabh Kumar (saurabh.kumar@ltu.se) Division of Operation and Maintenance Engineering Lule University of Technology, Lule
INTRODUCTION
Condition monitoring is a vital process to know the condition of the equipment or machinery. It helps in planning of maintenance activities in an effective way based on the condition of equipment/machinery. Oil analysis is a condition monitoring technique that has been in use for more than 50 years and has proved to be a highly effective management tool for health monitoring of lubricated machineries and the lubricants themselves. The operating environment causes physical degradation of the machines and chemical degradation of the lubricants used in those machines over a period of time. In addition to this, lubricant/oil is contaminated by internal and external factors which change the properties of oil. Undesirable changes in the oil property may affect performance and may lead to failure of the mechanical components. It is, therefore, essential to have a periodic monitoring of the lubricant properties. The lubricant analysis is therefore an accepted tool in the industry for optimization of lubricant uses as well as condition monitoring of equipment. The lubricating oil consists of base oil and a number of additives which are added in different compositions to the base oil, to meet various performance requirements. Analyzing the changes in physical and chemical properties of oil or that of its constituents help in estimating the extent of degradation in properties such as viscosity, acidity, water content, degree of oxidation etc. and provide important information on the condition of the oil as well as that of the machine. This information about the extent of degradation in the oil becomes a deciding factor for changing of the oil and charging of fresh oil into the machine. Apart from oil analysis, other condition monitoring techniques available to today's maintenance professionals include vibration monitoring, thermography, ultra-sonic analysis, etc.
CONDITION MONITORING TECHNIQUES

WDA (Wear Debris Analysis): Wear of metallic parts of a machine produces a lot of minute particles, which are carried by the lubricant. These small metal particles can give information about the machine elements that are wearing, and can be detected by various methods, for example, Spectroscopy. Determination of larger particles can be done using optical or electronic microscopy, or ferrography. Ferrography: This is a form of in-depth WDA where the morphology of individual wear particles is studied and characterised with the aid of a microscope. This type of analysis is generally covered under oil analysis. Vibration monitoring: This involves the measurement of oscillation of a body, machine or part of a machine around a fixed point; vibration is due to an excitation force that causes motion. These readings are 1
subject to a wide variety of mathematical analyses in order to detect such problems as mass unbalance, bent shafts, misalignment, mechanical looseness, belt drive problems, journal bearing wear, gear wear and AC induction motor problems. Thermography: This technique entails the use of infrared thermography whereby temperatures of a wide variety of targets can be measured remotely and without contact. This is accomplished by measuring the infrared energy radiating from the surface of the target and converting the measurement to an equivalent surface temperature. Thermography is used to monitor wherever excess heat can be generated when a component is defective, for example a bearing. Ultra Sonic Inspection: Acoustic emission or transmission and reflection of ultra high frequency sonic waves are used to detect and locate material defects. This technique is particularly useful for finding subsurface defects. Dye penetration: Cracks and surface irregularities can be easily picked up using a dye penetrant. In this technique a fluorescent dye is sprayed onto the surface to be examined, a developer or fixer is added and the component is examined under visible or ultraviolet light to highlight the presence of any surface flaws. Surface defects can also be detected by eddy currents, which occur as a result of electromagnetic induction. Radiography: Radiography uses X-rays in very much the same way that a medical X-ray would reveal a broken bone, to determine if any internal flaw or weakness exists within a mechanical component. Not all techniques can be used on all types of equipment. Some are highly complementary while in some cases others are not applicable at all. Oil analysis: This technique involves the removal of a small sample of oil from an operating piece of machinery. The oil is subjected to a number of physical and chemical tests that determine the health of the machine, the health of the oil and the level of contaminants in the oil. Oil analysis reveals the general health of oil. The oils physical and chemical properties along with the properties of the additives used in the oil can be measured and trended to guide decisions about if and when there should be an oil change or the oil should be regenerated with suitable additives. Oil analysis also identifies when the incorrect oil has been added to a system. Abnormal degradation in oil can be determined by analyzing if the degradation is oxidative, hydrolytic, or from another root cause. In addition to simple oil change decisions, oil analysis supports decisions to change oil base-stock or additive formulation or control the environment in which the oil operates. Machines cannot run efficiently without healthy lubrication, making these decisions imperative for machine reliability. The acidity in oil shows whether the oil is oxidized as a result of operation at high temperature, if there is a high percentage of moisture, or whether the oil has been in service for too long. The viscosity of the oil is a very important parameter and must be in conformity with the requirements of the machine. The alkalinity or the loss of alkalinity of the oil proves that the oil is in contact with inorganic acids such as sulphuric acid or nitric acid. It may be noted that particle analysis is done to get information about the condition of machine elements
that are wearing. To know the condition of the lubricating oils, we perform various tests of viscosity, acidity, spectroscopy, etc. The present work focuses to estimate the changes upon use of oil in two power generators by viscosity measurements, pH measurements, FTIR (Fourier Transform InfraRed) spectroscopic analysis and UV-Visible (Ultra Violet and Visible rays) spectrophotometric analysis.
THE CASE STUDY

Engines of two standby power generators (of the same manufacturer) of 250 KVA each, working in the same environment and under identical load fluctuation conditions were selected for the study. The engine oil tank capacity was 40 liters and the oil used was multigrade (20W40). As per original manufacturers recommendation, the oil change cycle should take place at every 250 hours. But after preliminary study and analysis the change cycle of the oil was extended to 350 hours. Engine oil samples of volume 100ml were collected from the two generator engines. Collection of a lesser volume of oil sample would not have been sufficient for performing the various tests. A greater volume of oil sample may require extra topping of oil as the oil level in the tank may get reduced. Drawing of samples from mid layer of the oil tank was ensured by using a hand pump. This precaution was taken due to the fact that the top layer of the oil often contains floating sludge and the bottom layer contains metal debris which, if taken, may not represent the entire oil present in the oil tank. Oil samples were collected at random intervals of working hours so that minute changes in its properties may be analysed.
ANALYSIS OF OIL CONDITION

Oils undergo destructive changes in property when subjected to oxygen, combustion gasses and high temperatures. The following properties were analyzed to monitor the extent of degradation occurring due to temperature variation, oxidation, nitration and sulphate formation. Viscosity Analysis The most important property of oil is its viscosity. It is defined as the oil's resistance to flow at a specified temperature. It is a measure of the oil's thickness; thick oil has a high viscosity while thin oil has a low viscosity. A fluids resistance to flow is known as kinematic viscosity. Kinematic viscosity is measured in centistokes (cSt). Kinematic Viscosity () = 1 Centistoke (cSt) = 1 mm2 /s It is important to note that as the temperature increases, viscosity decreases, so one must always state the temperature at which viscosity is measured, otherwise the reading will be meaningless (see Leugner, 2000 for details). Two industry standards are used when measuring kinematic viscosity, namely 40C and 100C. The type of oil under consideration and its properties determine which temperature should be employed, although 40C is used most commonly. Dynamic viscosity (sometimes called absolute viscosity) is a measurement of a fluid's resistance to shear at a specified temperature. Dynamic viscosity is measured in centipoises. 3
Dynamic viscosity () = 1 centipoise (cP) = 1 mPa.S The two viscosity measurements are related to each other by the density of the fluid: / = The viscosity of engine oil was measured in centistokes (cst) at 400C using portable Visgage. This viscometer works on the principle of falling sphere method. The viscosity was measured relative to the standard reference oil. Ten readings were taken for each oil sample and the most concurrent reading was selected. pH Measurement pH is a measure of the degree of the acidity or the alkalinity of a solution as measured on a scale (pH scale) of 0 to 14. The midpoint of 7.0 on the pH scale represents neutrality, i.e., a "neutral" solution is neither acidic nor alkaline. Numbers below 7.0 indicate acidity and numbers greater than 7.0 indicate alkalinity. It is important to understand that pH is a measure of intensity, and not capacity; i.e., pH indicates the intensity of alkalinity in the same way temperature tells how hot something is - but not how much heat the substance carries. One can test the pH of a solution using indicator paper or indicator liquid. An indicator is a chemical which will change colour as the pH changes. There are quite a few indicators in use in the laboratory. This is why each type of indicator paper comes with its own colour chart. The pH scale is logarithmic which means that moving on either way on the pH scale results in a 10 fold increase in the degree of alkalinity or acidity. Products at the two extremes (less than pH 1 or greater than pH 13) are extremely oppressive and corrosive. Examples include sulphuric and hydrochloric acid on the acid end, and caustic soda on the alkaline end. The pH scale was developed because the concentration of a solution can vary a lot over time. A pH scale is the easiest way to measure the change in acidity of the oil. The pH was measured by comparing the colour of the pH paper after dipping it in oil with the colour chart provided with the pH paper. Multirange pH indicator papers used were of S.D. Fine Chemicals Limited, Mumbai, India (Product no - 41087). Fourier Transform Infra-Red (FTIR) Spectroscopic Analysis The use of infrared spectroscopy for condition monitoring of oil-lubricated components, breakdown products and contaminants has not been widely used in the past, although infrared studies of lubricating oils themselves have been performed for a number of years. The reason for this is that older dispersive infrared spectrometers would take several minutes to generate a spectrum of the used oil and then considerable additional time would be needed to reduce and interpret spectral data (Geach, 1996). New technology has yielded the Fourier Transform Infrared (FTIR) Spectrometer with a micro computer. This combination in modern equipment produces infrared spectra in seconds and reduces spectral data in minutes. With the help of fourier transform in FTIR Spectrometer, it becomes very easy to obtain a solution of complex equations of the spectra. These equations are then solved within a few seconds in the micro computer with the aid of inbuilt software. FTIR analysis has become a powerful, practical analytical tool for used oil analysis (Mukherjee, et. al., 2000). By FTIR analysis, the organic components present in the lubricating oil are found out. The samples collected at different intervals of time are analyzed using this tool, the change in structure of organic components is
obtained and compared with those of fresh oil samples. Thus the extent of degradation in organic compounds present in the oil is found out. To extract information from infrared spectra of used oils, a basic knowledge of the processes involved in oil degradation is required. The lubricant in a generator engine operates in a very hostile environment and high temperatures. The lubricant is dispersed over a large surface area where it is exposed to chemically reactive byproducts of the combustion process. In addition to this, the oil is exposed to sources of internal and external contamination. Oil degradation by chemical change The chemical degradation of a generator engine oil may be defined by a number of processes: Oxidation At elevated temperatures, oil exposed to oxygen from the air, will oxidize (chemically combine with oxygen) to form a variety of compounds. The majority of these are carbonyl containing compounds (C=O) such as Esters, Ketones and Carboxylic acids. Some of these compounds are dissolved by the oil, or remain suspended owing to dispersive additives in the oil (see Geach, 1996 for details). Carboxylic acids contribute to the acidity of the oil and deplete its basic reserve as neutralization takes place. The net effect of prolonged oxidation is that chemically, the oil becomes acidic causing corrosion, while physically an increase in viscosity occurs. The increase in viscosity however, may be masked by other factors such as fuel dilution. Nitration Nitration is another form of oxidation. Nitration results from the reaction of oil components with nitrogen oxides (NO, NO2 and N2O4) that are produced from the oxidation of atmospheric nitrogen during the combustion process. In addition to causing oil thickening, nitration products are the major cause of the build-up of varnish or lacquer (Kemp, 1991). Sulphate Formation The various oxides of sulfur and water, both of which are combustion byproducts, react together to form sulphuric acid. This acid is neutralized by the basic reserve in the additive package of the oil and normally results in the formation of metallic sulfates (Geach, 1996). An indication of the progress of neutralization is obtained by monitoring the build-up of metallic sulfates in the oil. Soot Index The soot index is a measure of the level of partially burned fuel particles (soot) in the oil. The rate at which soot is deposited in the oil is dependent on engine design and operating conditions. An increase in the soot index will indicate poor combustion of oil and overextended filter change period. Water and Glycol Water and Glycol may be detected at relatively low levels by FTIR. The presence of glycol and water or glycol alone would indicate a coolant leak. Water alone does not necessarily indicate a coolant problem, as traces of water could result from condensation, if an oil sample has been taken from a cold engine (Smith, 1999). Oil degradation by contamination Detrimental contamination may occur from both external and internal sources.
Internal Contamination This mainly comprises soot and metallic particulates (wear metals). Both these contaminants occur on a continuous basis and will build up to undesirable levels at which time the lubricant should be discarded. The rate at which unacceptable levels of contaminant are reached will depend on the condition of the engine and its operating conditions. External Contamination The main contaminants in this category are unburnt fuel, coolant and dirt. These contaminants result from malfunctions and failures in the fuel, cooling and air intake systems respectively. It is essential that these contaminants are detected early if catastrophic failures are to be avoided. Background of Infrared Analysis Infrared radiation is part of the make-up of the electromagnetic spectrum and covers electromagnetic waves with wavelengths between 0.00008 cm and 0.04 cm (Figure 1).
Figure 1: The Electromagnetic Spectrum [Adapted from Geach, 1996] Electromagnetic waves are waves that have both electric and magnetic component. Well known examples of electromagnetic waves include X-rays, visible light, microwaves and radio waves. These waves travel at the same velocity (speed of light) but differ in the wavelength and frequency bands used to describe them. Figure 1 shows the whole range of electromagnetic waves arranged in order of increasing wavelength and depicts what is known as the electromagnetic spectrum. Chemists have adopted a more convenient method of describing infrared radiation, in that it is described in terms of the number of waves that occur per centimeter. This number is called the wavenumber and is actually a measure of wave frequency. Wavenumber (cm-1) = 1/wavelength (cm) Infrared analysis only uses a portion of the IR spectrum known as "Mid-range infrared". It is defined as infrared waves having wave numbers between 4000 and 400 cm-1 (Geach, 1996)
Molecular vibrations and infrared absorption The chemical bonds within organic molecules are in a state of continual vibration, with bonds stretching and contracting as well as bending relative to one another. When an infrared beam falls on a molecule, waves of specific frequencies (wave numbers) are absorbed from the beam by the molecule, and results in changes in the molecular vibrations pattern. The actual frequency of the waves absorbed depends on the types of bonds present in the molecules structure. Different types of bonds within the same molecule would absorb different frequency bands while several identical bonds would absorb the same frequency bands and give rise to stronger absorptions. For example a C=O and a C-H bond in the same molecule would be expected to yield at least two different absorption bands while several C-H bonds in the same molecule would contribute to at least a single relatively strong absorption band. Chemical bonds within a molecule are therefore said to exhibit characteristic infrared absorptions (see, Kemp, 1991 for details). A record of the frequencies at which infrared absorption takes place for an organic compound is a characteristic property of the compound and is called its Infrared Spectrum. An infrared spectrum of a compound will reveal information about molecular structure as the existence of specific groups of atoms may be confirmed from the presence of their characteristic absorptions. The instrument used to record infrared spectra is called an Infrared Spectrometer. The output of spectra from a spectrometer is generally in the form of a plot of Absorbance vs. Wave number, however most instruments also show output spectra in the form of Percentage Transmittance vs. Wave number. Percentage Transmittance is the logarithmic reciprocal of Absorbance. - log ( T/ T ) = A , where T = Intensity of transmitted beam T = Intensity of incidence beam and A = Absorbance.
The FTIR spectra of the oil sample were recorded on Perkin Elmer FTIR-model spectrum 2000, using specular reflectance accessories. In each case the oil sample was spread over a copper foil and the reflectance spectrum was recorded. Ultraviolet and Visible range Spectrophotometric Analysis Ultraviolet and visible (UV-Visible) absorption spectroscopy is the measurement of the attenuation of a beam of light after it passes through a sample or after reflection from a sample surface (Dyer, 1965). Measurements can be at a single wavelength or over an extended spectral range. Ultraviolet and visible lights are energetic enough to promote outer electrons to higher energy levels, and UV-Visible spectroscopy is usually applied to molecules or inorganic complexes in solution. The UV-Visible spectra have broad features that are of limited use for sample identification but are very useful for quantitative measurements. The UV-Visible spectral range is approximately 190 to 900 nm, as defined by the working range of typical commercial UV-Visible spectrophotometers. The light source is usually a deuterium discharge lamp for UV measurements and a tungsten-halogen lamp for visible and NIR (Near Infra-red region) measurements (Tissue, 1996). The instruments automatically swap lamps when scanning between the UV and visible regions. The wavelengths of these continuous light sources are typically dispersed by a holographic grating in a single or double monochromator or spectrograph. Spectrometer designs and optical components are optimized to reject stray light, which is one of the limiting factors in quantitative transmittance measurements. Most commercial UV-Visible absorption spectrometers use one of three
overall optical designs: a fixed or scanning spectrometer with a single light beam and sample holder as shown in Figure 2, a scanning spectrometer with dual light beams and dual sample holders for simultaneous measurement of test sample with respect to a known sample, or a non-scanning spectrometer with an array detector for simultaneous measurement of multiple wavelengths. In single-beam and dual-beam spectrometers, the light from a lamp is dispersed before reaching the sample cell. In an array-detector instrument, all wavelengths pass through the sample and the dispersing element is between the sample and the array detector.
Figure 2: Circuit diagram of a UV-Visible Spectrophotometer [Adapted from Tissue, 1996] UV-Visible analysis can be used to indicate the presence of chromophores (colour producing group) and distinguish between conjugated and nonconjugated carbonyl compounds. This analysis is very important in the present context as it may throw light upon viability of optical sensor. Percentage transmittance of the engine oil was measured using UV-Visible Spectrophotometer, ECIL India Limited-UV5704 SS model. The sources of radiations were, Deuterium lamp for UV region and Tungsten-Halogen lamp for visible and near infrared region of the spectrum. In each case an aliquot was prepared by making 0.5% (v/v) solution of lube oil in petroleum ether (Boiling range: 800C- 1200C). The readings were taken in percentage transmittance and in absorbance mode by placing the test sample in a quartz cuvette. Blank solution, used as a reference contained only petroleum ether. No external dye was used.
RESULTS AND DISCUSSION

Result of viscosity measurement is shown in Figures 3 and 4. The viscosity measurement indicates decrease in viscosity of the oil on use. In general, the decrease in viscosity of a lubricant is most probably due to loss of shear stability of the various components especially that of polymeric components. This loss here indicates degradation of polymers mostly used for VI improvers. Viscosity decrease can also be due to dilution but in this case, probability can be ruled out since both the engine oils show a similar trend, which is, decrease in viscosity on use. These indicate thermal instability of the viscosity improvers or of the antiwear / corrosion inhibitors (Kumar, et. al., 2005).
125 119
120
Viscosity (cst)
Viscosity (cst)
119
115 105 102 99 95 0 100 200 300 400
118 116 114 0 25 50 75 116 115 114.5 100
105
Working Hours
Working Hours
Figure 3: Viscosity Vs Working Hours (Generator-1)
Figure 4: Viscosity Vs Working Hours (Generator-2)
pH of the samples are shown in Figures 5 and 6 and Table 1. The trend in change in pH indicates accumulation of acidic compounds in the oil. This may be due to oxidation of organic compounds or that of sulphur containing compounds. Formation of carboxylic group containing compounds is not indicated in the FTIR spectra of the sample (Figures 7 and 8) so it is possible that the pH is decreasing due to formation of inorganic acids or may be due to sulphoxy compounds or SH group containing compounds.
8.5 8.25 8 7.5
8 8.5
8.25
pH
7 6.5 6 0
pH
7.5 7
7.25 7
6.25 100 200 300
6.25
6.5
6.75 75 100
400
25
50
Working Hours
Working Hours
Figure 5:
pH Vs Working (Generator -1)
Hours
Figure 6:
pH Vs Working (Generator
Hours -2)
Table 1: Analysis of pH value for Generator 1 &2 Generator 1 Working Hours 0 89.4 309.6 319.1 pH Value 8.5-8.0 7.0 6.5-6.0 6.5-6.0
Generator 2 pH Value 8.5-8.0 7.5-7.0 7.0 7.0-6.5
Working Hours 0 48.6 57.9 80.1
The spectra of samples are shown on Figures 7 & 8.

97.3 90 80 89.4HRS FRESH OIL
% Transmittance
70 60 50 40 30 20 10 1.5 4000 3000 2000 1500 1000 500 370.0 309.6HRS
Wave No. (cm-1)
Figure 7: Percentage IR Transmittance Vs Wave Number for Generator 1 The spectra of all the samples of used oil show a similarity with that of the fresh oil. It shows that even after using the oil for more than 300 hours, there has not been any appreciable degradation in the base stock. However, as significant change was noticed at around 975 cm-1, the intensity of the peak diminishes upon use. The peak was no longer observable in the spectrum of the oil sample which has been used for more than 300 hours. Normally this region is assigned for sulphoxides, S=O stretching, OH bending and CO stretching of primary alcohols. Sulphur compounds are used in lube oils as oxidation and corrosion inhibitors as well as for antiwear agent. Sometimes the metallic dispersants may also contain sulphonates. Diminishing intensity of the peak in this region, which is around 975 cm-1, with use indicates loss of antiwear additive or, corrosion inhibitors.
99.0 90 80 FRESH OIL
% Transmittance
70 60 50 40 30 20 11.1 4000.0 3000 2000 1500 1000 500 57.9 HRS 48.6 HRS 80.1HRS
Wave No. (cm-1)
Figure 8: Percentage IR Transmittance Vs Wave Number for Generator 2
10
As shown in Figure 9, the UV visible spectra did not show any characteristic peak of any chromophore (color producing group), like ArN=NAr, C=O and NO2 groups which normally appear in between 165nm to 280nm. Any characteristic peak of a conjugated C=C peak which normally appears at around 200nm was also absent. All the samples showed a small absorption peak at around 400nm, which is indicative of hydrocarbon chromophore with small amount of polycyclic aromatic hydrocarbons (Kemp, 1991). The UV visible spectra indicated absence of any major oxidized product in the oil samples even after prolonged use.
Figure 9: Graph of percentage Transmittance vs Wavelength of the oil samples taken from the Engine of Generator 1. Various test results show that with use, the base stock of the oil does not show any appreciable degradation even for oil used for more than 300hrs. The changes observed in viscosity, pH and the FTIR spectra indicate that degradation of additives is taking place in the lube oil. The additives are used for giving desirable properties to the oil base stock, for example, right viscosity, oxidation stability, corrosion inhibition, etc. Some of the viscosity improvers are thio-esters or petroleum sulphonic acids. The corrosion inhibitors and antiwear additives normally used are molybdenum disulphide, thio compounds or phenol disulphides. The results obtained from pH measurements and FTIR spectroscopy substantiates the findings that in the present case, depletion of sulphur containing additives has taken place. Oxidation of the base stock has not taken place; this can be confirmed from the UV visible spectra of the sample.
CONCLUSION
From the various tests conducted, it may be concluded that: i. ii. There is no appreciable change in the quality of the base stock even after prolonged use, up to 319 hrs, which extends up to 28% of the usual life of engine oil. In the present study, it was found that the pH of the oil drops to the acidic region, if the oil is used for more than 300hrs. This may cause damage to the components by corrosion or abrasion.
11
iii.
Significant depletion of additives especially corrosion inhibitors, VI improvers and antiwear agents have taken place during use.
The present oil sample and the existing operating conditions warrant need to modify the additives used which should be more stable towards oxidation, withstand temperature, should not cause rise in acidity of the lubricants. It may be inferred from the present study that, by addition of calculated amount of the additives which have depleted may make oil fit to be used again after desired filtration. Thus reducing or eliminating the need for oil change, saving environment from the waste oil dumps and also reducing the operational cost by a considerable extent. By proper inspection and by monitoring the condition of the power generator lubricating oil, the oil can be used for an extended life.
Reference
Dyer, J. R. (1965) Applications of Absorption Spectroscopy of Organic Compounds, Prentice Hall Inc., USA, pp 4-21. Geach A., Wear Check Africa, 1996, (http://www.wearcheck.com/literature/techdoc/WZA002.htm) Access Date: 15-04-2007. Kemp, W. (1991) Organic Spectroscopy, Third Edition, Macmillian Press Ltd. Kumar, S., Mishra, N.M. and Mukherjee, P.S. (2005) Additives Depletion and Engine Oil Condition - A Case Study, Industrial Lubrication and Tribology, Volume 57, Number 2, pp. 69 72. Leugner, L. (2000) Practical Handbook of Machinery Lubrication, Maintenance Technology Internation. Mukherjee,P.S., Sinha,A.N. and De,A.. (2000) Rull Assessment by FTIR-a case study on HEMM in Indian Mines, Industrial Lubrication and Tribology, Volume 52, pp. 61-66. Smith, B.C. (1999) Infrared Spectral Interpretation, CRC Press Ltd. Tissue B.M. (1996) Ultraviolet and Visible Absorption Spectroscopy, http://elchem.kaist.ac.kr/vt/chem-ed/spec/uv-vis/uv-vis.htm Access Date: 14-05-2007
12
Modeling and Experimental Verification of CC Overhead MV Networks for PD Detection
Report submitted for the course work

MAINTENANCE MANAGEMENT IN POWER SYSTEMS
G. Murtaza Hashmi
Power Systems and High Voltage Engineering Helsinki University of Technology (TKK) Otakaari 5, 02015, Espoo, Finland murtaza.hashmi@tkk.fi
http://powersystems.tkk.fi/eng/contacts.html
______________________________________________________________________________
Abstract
Partial discharge (PD) measurements are universally accepted as an effective and reliable tool for on-line condition monitoring of high voltage equipment. A relatively new application is the monitoring of fallen trees on covered-conductor (CC) overhead distribution lines. In this paper, theoretical model of the CC line is presented and the frequency-dependent distributed transmission line (TL) parameters, wave propagation characteristics, and characteristic impedance are calculated. The derived wave propagation characteristics are compared with those obtained from the time domain reflectometry (TDR) measurements to validate the theoretical model of the CC line. The TDR measuring system is simulated in EMTP/ATP (electromagnetic transient program/alternate transient program) as real time verification. The measurement results show that the attenuation and the propagation speed are frequency dependent and these values become higher as the frequency of a signal goes up. An accurate determination of the TL characteristics for practical lines will be used as a design aid for the modeling of PD measuring system in EMTP/ATP; thus deciding the optimal locations of the sensors in getting reliable PD detection results.
______________________________________________________________________________ Maintenance Management in Power Systems 2
______________________________________________________________________________
1.
History of CC lines
The MV distribution system in Finland has been constructed using mainly bare overhead lines. The commonest voltage is 20 kV. In normal circumstances, such lines behave reliably, although they demand a lot of space. The space required for double and multiple line structures is especially broad. Exceptional situations produce difficulties for bare overhead lines, e.g. the clashes due to fall of accumulated snow cause short-circuits and damage to the conductors. Trees falling on the lines cause damage to the structures and short-circuits between phases as well as earth faults [1]. In the early 1960s, American utilities began replacing bare conductors in MV electricity distribution networks with covered ones. This was also being done at almost the same time in Australia [2]. The covering materials used were polyvinyl chloride (PVC), high density polyethylene (HDPE), and nylon. The lifetime of these lines was very limited and the general level of interest in them was quite low. In the area where a conductor was fixed to the insulator, it was necessary to remove insulation from the conductor, which caused very intensive corrosion. This fact eventually led to termination of their use. In the early 1970s, CCs became attractive again, especially with their use in Finland because this technique offered increased reliability in adverse weather conditions, satisfied ecological considerations, and also proved to be costeffective for the complete life cycle. Utilities in Sweden and Norway soon adopted this system, which has since spread to utilities throughout Europe, the United States, Australia, and Japan. The use of CC lines using an aluminum alloy conductor covered with a sheath of cross-linked polyethylene (XLPE) or HDPE has been expanding throughout the world ever since. The high voltage distribution lines with CCs were originally introduced in the Nordic countries due to their good applicability to lines passing through forest areas and due to their better operating safety. With such a lot of forest throughout the Scandinavian countries, the problem could have been one of forest fires caused by trees falling onto a bare overhead supply line or unnecessary tree felling to create wayleaves [3]. In the late 1980s, the research was started in Finland into the suitability of covered line for forest areas; however, it was more likely to be the quality of the electricity supply that was the main reason for that research. Consumers in Scandinavia, especially industrial consumers, do not accept loss of supply as lightly as perhaps the British do! The Finnish research and more recent work carried out by ABB, coupled with a developing history of general usage in the field have given very positive results during the last few years. Experience in both laboratory and practical situations in Finland, Sweden, and Norway appears to confirm the advantages of MV covered overhead lines compared with conventional bare conductors [3]. In Europe, the first CC lines went up Finland in the 1970s (known as SAX and later also as PAS), followed by Sweden and Norway (known as BLX) in the late 1980s, and in the UK in the mid 1990s. At the present time, there is more than 8500 km circuit length of PAS conductors in Finland; their share of new MV network construction totaling around 80%. Sweden has more than 9500 km circuit length of BLX conductors; their share of new MV network construction as high as 80%. Norway has more than 4000 km CC line, while, in UK, over 2000 km line have
______________________________________________________________________________ now been erected. Several other European and South American countries are also showing great interest in CC overhead lines.
2.
Objectives and challenges posed by CC lines
The use of CC overhead lines in the distribution system started with the need of decreasing the number of faults caused by the falling trees as well as reducing the expenses with tree clearances and the maintenance and system life span [4]. The statistics confirm that the number of faults dropped to 0.9 per year per 100 km compared to 4.5 per year per 100 km for lines with bare conductors [5]. The general objectives of the CC technology can be described as follows [6]: The negative impact of power lines to biodiversity can be reduced using CC lines. The studies show that power lines have harmful impact on biodiversity because they divide artificially the surrounding nature. Using CC lines, the wayleaves (corridors) of the power lines can be constructed significantly narrower. The safety of power lines improves with CC lines. The fact is that a number of people die in HV accidents because of direct contact to HV conductors. Using CC lines, a greater part of these accidents can be avoided. In addition, many birds die every year by colliding with power lines. A number of these fatal collisions can also be avoided using this technique. With CC lines, old lines can be upgraded without constructing a new transmission line. For instance, when there is a need to increase the voltage, it can be done only by changing the old conductors to covered ones without major changes in structures etc. This can save environment notably. Power lines are surrounded by electric and magnetic fields. These fields possibly impact on health concerns people all over the world. Low magnetic fields in the vicinity of the line are one of the greatest advantages using CC lines.
In addition to above mentioned objectives, the advantages of using CC lines are as follows [7]: No faults or troubles with snow or hoar frost; No interruptions by dropping branches; No faults with touching conductors by ice-shedding; Smaller lanes through forests, and reimbursement for owners is smaller; The clearing of the lane from growing trees is more seldom. The big trees at the border to the forest can remain in their place, therefore, a lot of savings are possible; CC lines are a cheaper alternative to underground cable, especially in difficult terrain
However, these kinds of conductors pose some problems, such as: PDs are produced due to falling/leaning trees on the surface of CC lines. The falling trees may produce knife traces on the conductor surface due to difference in wind pressures, which may aggravate the situation of initiating PDs, resulting in rupture of the conductor insulation. The falling trees cannot be detected with normal protection relays;
______________________________________________________________________________ The results (electric, mechanical, and thermal tests) of the researches performed in the USA can not be applied to Europe and Asia due to climatic differences [4]; The different dielectric constants of the material employed, generating electric field concentration and consequently the possibility of the corona effects; The sensibility to ultra-violet radiations; The susceptibility to thermo mechanical effects, causing cracks.
3.
Motivation for CC modeling
It has already been shown that using Rogowski coil PD sensors, on-line high frequency PD measurements can be utilized for the monitoring of fallen trees on the CC overhead distribution lines [8]. Extracting the features of PD from the measurements to detect and locate such faults on a complicated TL network is a challenging task. The challenge for on-line PD measurements is to find the optimal locations for these sensors with respect to their sensitivity, interference level, signal distinction, and universal applicability [9]. To work with these design aspects of a sensor and in order to relate a measured signal to specific PD amplitude, the wave propagation characteristics of the CC line should be accurately determined. Attenuation is an important parameter in order to estimate the maximum length of the line that can be monitored with a PD sensor. The measurement of the attenuation can give an idea about the length of the conductor at which the PD signal dies, so that the sensor location can be assured to get the required signals. Therefore, attenuation of the PD pulse is an important consideration while deciding the number of sensors and their positioning. In this way, CC lines will be more reliable and the costs related to visual inspection work will also be reduced. The modeling of the measuring system in a time domain digital simulator program, such as EMTP/ATP, poses many challenges. One of the main difficulties is the correct representation of the transmission line (TL) dealing with transient simulation studies [10]. The TL parameters are function of the line length and the per unit parameters vary with frequency. In this paper, CC line is approximated as a two-wire TL and its characteristics (TL parameters, wave propagation characteristics, and characteristic impedance) are derived. The frequency-dependent distributed TL parameters, which are needed to accurately characterize lossy lines, are derived based on the radial geometry and the electrical properties of the CC material. The wave propagation characteristics and characteristic impedance of the CC line are determined based on the calculated TL parameters. Time domain reflectometry (TDR) is a measuring technique to accurately determine the wave propagation characteristics of a cable or a conductor [11]-[13]. The calculated wave propagation characteristics are compared with those obtained from TDR measurements for experimental verification of the theoretical model of the CC line. Moreover, the TDR measuring system is simulated in EMTP/ATP environment for real time verification of the measurements.
______________________________________________________________________________
4.
Transmission line analysis
In the following analysis, the live CC line and the ground return path are used as transmission channel and are approximated as a close form of the two-wire TL. According to [14], the twowire TL must be a pair of parallel conducting wires separated by a uniform distance. Based on the above considerations, the single phase CC line with ground return is regarded as a distributed parameter network, where voltages and currents can vary in magnitude and phase over its length. Therefore, it can be described by circuit parameters that are distributed over its length. A differential length z of a TL is described by its distributed parameters R, L, C, and G as shown in Fig. 1. R defines the resistance per unit length for both conductors (in /m), L defines the inductance per unit length for the both conductors (in H/m), G is the conductance per unit length (in S/m), and C is the capacitance per unit length (in F/m). The quantities v(z, t) and v(z+z, t) denote the instantaneous voltages at locations z and z+z respectively. Similarly, i(z, t) and i(z+z, t) denote the instantaneous currents at the respective locations. In the circuit of Fig. 1 (b), applying Kirchhoffs voltage and current laws respectively, the following two equations can be obtained as [8]:
v( z , t ) Rz i ( z , t ) Lz i ( z , t ) v ( z + z , t ) = 0 t
v ( z + z , t ) i ( z + z , t ) = 0 t
(1)
i ( z , t ) Gz v( z + z, t ) Cz
(2)
Fig. 1. (a) The voltage and current definitions of two-wire TL (b) The equivalent lumped-element circuit
______________________________________________________________________________ If v and i are expressed in phasor form, i.e. v( z , t ) = Re V ( Z ) e jwt and i ( z , t ) = Re I ( z ) e jwt and when z0, the time harmonic line equations can be derived from (1) and (2) as: dV ( z ) = ( R + j L ) I ( z ) dz dI ( z ) = (G + jC )V ( z ) dz (3) (4)
The coupled time harmonic transmission line equations can be combined to solve for V ( z ) and I ( z ) as:
d 2V ( z ) = 2V ( z ) 2 dz d 2 I ( z) = 2 I ( z) dz 2 where (5) (6)
( ) = ( ) + j ( ) = ( R + jL) (G + jC )
(7)
is the propagation constant whose real and imaginary parts, and , are the attenuation constant (Np/m) and phase constant (rad/m) of the line respectively. is the angular velocity (rad/sec), where = 2f , and f is the frequency (Hz) of the propagated signal. The complex propagation constant is also given as:
( ) = Z Y
(8)
where Z = R + jL is the series impedance per unit length and Y = G + jC is the shunt admittance per unit length of the line. The characteristic impedance of the line is given as:
Z0 =
R + jL Z = G + jC Y
(9)
It is clear from (7) and (9) that and Z0 are the characteristic properties of a line whether or not the line is infinitely long. They depend on R, L, G, C, and but not on the line length [15].
______________________________________________________________________________
5.
Theoretical modeling of CC line
When a signal travels along the CC line segment of a certain length, it will be subjected to attenuation losses which depend on the physical properties of the CC material and the height of the conductor above the ground. For a MV line with a propagation constant , these losses are fully determined by the TL parameters (i.e. R, L, G, and C) and the frequency of the propagated signal . The cross-sectional view of the CC under investigation is shown in Fig. 2. As the line is kept at a height of 7 cm above the ground level in TDR measurement, therefore, the same height is considered in theoretical modeling. The model of the CC line is shown in Fig. 3 [16]. The line model per unit length consists of series impedance Z and shunt admittance Y. The aluminum conductor and the ground return path constitutes Z. The internal impedance ( Z i ) of the aluminum conductor including the skin effect is given as [16], [17]: 1 2RC
Zi =
j 0
al
(10)
where 0 is the permeability of the free space and al is the conductivity of the aluminum material. The external impedance ( Z e ) due to geometrical inductance of the CC line is given as:
Ze =
j 0 RCC j 0 D ln R + 2 ln R 2 C CC
(11)
Several expressions for the ground return path impedance have been presented by many researchers, however, one of the simplest forms for ground return path impedance is given by the following logarithmic function [18]:
1+ g D j 0 ln D 2 g
Zg =
(12)
where g =
j 0 ( g + j o rg ) ; g is the ground conductivity (0.001 S/m), 0 is the
permittivity of the free space, and rg is the relative permittivity of the ground (taken as 10 in this case). The total series impedance of the CC line is given as:
Z = Zi + Ze + Z g
(13)
The shunt admittance in the model is made up of the dielectric losses and the capacitances of the conductor and air insulations as:
______________________________________________________________________________
Fig. 2. Cross-sectional view of the CC line (in laboratory tests)
Fig. 3. Equivalent electrical model of the CC line
Y12 = G12 + jC12 Y23 = G 23 + jC 23
(14) (15)
The parameters G12=C12tan, G23=2air/ln(D/RCC), C12=2c/ln(RCC/RC), and C23=20/ln(D/RCC) are given. The loss tangent tan for the CC insulation is taken as 0.0003. The permittivity of the CC material c is equal to 0rc, where 0 is the permittivity of free space and rc is the relative permittivity of the CC polyethylene insulation taken as 2.26. air is the conductivity of the air. The total shunt admittance of the CC line is given as: 1
1 1 + Y Y 12 23
Y=
(16)
______________________________________________________________________________ Using (11-16), it is possible to calculate the frequency-dependent TL parameters. The propagation constant and characteristic impedance of the CC line can be calculated using (8) and (9) respectively. The frequency-dependent attenuation (in dB/m) can be calculated as:
( ) = 8.686 real[ ( )]
The propagation velocity of the signal can be determined using phase constant as:
v=
(17)
(18)
The aforementioned equations can be used to derive TL characteristics from the theoretical model. This model is verified experimentally by taking TDR measurements on the CC line sample of a certain length.
6.
TDR measuring set-up
The TDR measuring set-up consists of: (a) device under test (DUT) i.e. CC line and (b) measuring system (see Fig. 4). The height of the CC line above the ground level affects the TDR response. The minimum height of the CC line above the ground level (7 cm) is preferred because TDR measurements are free from reflections and surrounding noise at this level [13]. In practice, a narrow electric pulse is applied to the CC line and the incident and reflected waves are measured by means of a digital oscilloscope at point M. The measured data is transferred to the computing system (laptop) connected to the oscilloscope through the general purpose interface bus (GPIB), where analysis is performed for the extraction of wave propagation characteristics using MATLAB [13].
Fig. 4. Schematic drawing of the TDR measuring set-up [13]
______________________________________________________________________________
7. 6.1.
Theoretical model verification Model parameters verification
The theoretical model is verified experimentally using TDR measurements and the wave propagation characteristics using these techniques are compared. The CC is assumed to be placed at the same height above the ground level while extracting its propagation characteristic from the both techniques. The attenuation curve obtained theoretically from the CC geometry data is plotted against the attenuation curve obtained through TDR measurements as shown in Fig. 5. It is revealed that attenuation curves have a good match until several MHz (see Fig. 6), however, the attenuation curve obtained from TDR measurements suffer in noise at higher frequencies due to measuring limitations, already explained in reference [13].
7 6 Attenuation (dB/m) 5 4 3 2 1 0 -1
Theoretical modeling TDR measurements
10
10 10 Frequency (Hz)
10
10
Fig. 5. Comparison of attenuation curves obtained from different techniques
0.12 0.1 Attenuation (dB/m) 0.08 0.06 0.04 0.02 0 -0.02
10
10 Frequency (Hz)
Fig. 6. Attenuation curves enlarged in the lower frequency region
______________________________________________________________________________
350 Velocity of propagation (m/s) 300 250 200 150 100 50 10
5
10
10 Frequency (Hz)
10
10
Fig. 7. Comparison of propagation velocity curves obtained from different techniques
The propagation velocity curves obtained from the theoretical model and TDR measurements are shown in Fig. 7. It is clear that propagation velocity curves have a good match at lower frequencies, however, the propagation speed curve obtained from TDR measurements suffer in noise at higher frequencies as already explained in reference [13].
6.2.
Real time verification using EMTP/ATP
In the above sub-section, theoretical model of the CC line has been verified using TDR measurements. The accuracy of the theoretical calculation of the CC line characteristics is assured by studying the time domain model of the TDR measuring system in EMTP/ATP simulation environment. The ATPDraw is used as a graphical interface and the corresponding network of the TDR measuring system (depicted in Fig. 4) is illustrated in Fig. 8.
Fig. 8. ATP Draw circuit for the TDR measuring system
The TLs are represented using distributed parameters Clark model. The coaxial cables are considered as lossless lines having zero resistance and 50 characteristic impedance. The calculated CC line characteristics using theoretical model are used in simulation as: resistance, 3.5 /m; characteristic impedance, 160 [19]; and propagation velocity, 155 m/s. The
______________________________________________________________________________ transient response of the CC line for TDR measurements and its ATP simulation response are shown in Fig. 9.
3 2.5 2 1.5 1 0.5 0 0.2 0.4 0.6 0.8 1 Time (s) 1.2
TDR measurements ATP simulations
Amplitude (V)
1.4
1.6
Fig. 9. Transient response verification for TDR measurements
The experimental and the simulated time domain amplitudes and phases of the reflected pulses from the open end of the CC line have a good match. It verifies that attenuation and propagation velocity have same values in both responses. It shows that TDR measuring system has successfully been modeled, which proves the accuracy of the measuring system. However, the reflected pulses obtained from TDR measurements are distorted. This pulse distortion results as the incident and the reflected pulses travel unequal lengths of line. Therefore, the shapes of the waves will differ as a result of the frequency-dependent attenuation.
8.
Results and discussion
The CC lines in MV networks are located at different heights above the ground level (e.g. 10, 12, and 15 m) in different regions, keeping in view the environmental conditions prevailed. Moreover, the conductor height depends upon the design of the poles on which they are mounted. The frequency-dependent TL parameters can be derived from the theoretical model for the practical CC lines located at different heights above ground level. The frequency-dependent TL parameters for practical lines are shown in Fig. 10. It is revealed that R and G are frequency-dependent; however, L and C are more or less independent of the frequency of the signals propagated. The increase in R and G is due to skin effect which is dominant at higher frequencies. R, G, and C decrease as the height of the conductor is increased, however, L increases by increasing the conductor height. The effect of conductor height upon the value of G is not significant as can be seen in case of other line parameters. The dependency of the wave propagation characteristics and characteristic impedance upon the height of the conductor D above ground level is shown in Fig. 11. The attenuation and propagation velocity are frequency-dependent and these parameters increase by increasing the frequency of the propagated signals. The characteristic impedance is more or less constant at all signal frequencies. ______________________________________________________________________________ Maintenance Management in Power Systems 13
______________________________________________________________________________
(a) 2.5 D=10 m D=12 m D=15 m
2 Resistance ( /m)
1.5
0.5
0 4 10
10
10 10 Frequency (Hz) (b)
10
10
1.66 1.64 Inductance (H/m) 1.62 1.6 1.58 1.56 1.54 1.52 1.5 4 10
D=10 m D=12 m D=15 m
10
10 10 Frequency (Hz) (c)
10
10
0.35 0.3 Conductance (S/m) 0.25 0.2 0.15 0.1 0.05 0 D=10 m D=12 m D=15 m
100
200
300
400 500 600 Frequency (MHz) (d)
700
800
900
1000
8 D=10 m D=12 m D=15 m
7.8 Capacitance (pF/m)
7.6
7.4
7.2
7 4 10
10
10
10
Figure 10. The per-unit length TL parameters for CC overhead distribution lines located at different heights; (a) resistance, (b) inductance, (c) conductance, and (d) capacitance
______________________________________________________________________________
(a) 0.025 D=10 m D=12 m D=15 m
0.02 Attenuation (dB/m)
0.015
0.01
0.005
0 4 10
10
10 10 Frequency (Hz) (b)
10
10
300 D=10 m D=12 m D=15 m
Propagation velocity (m/m)
290
280
270
260
250 4 10
10
10 10 Frequency (Hz) (c)
10
10
500 490 480 470 460 450 440 4 10 D=10 m D=12 m D=15 m
Characterisitc impedance ( )
10
10
10
Figure11. Wave propagation analysis of MV overhead CC line at different heights; (a) attenuation, (b) propagation velocity, and (c) characteristic impedance
The attenuation decreases by increasing the conductor height above the ground level, and conversely, propagation velocity and characteristic impedance increase by increasing the conductor height. These results have a good match with the values given in reference [20], which also verify the accuracy of the proposed theoretical model. In the laboratory, the PDs produced due to leaning trees on 20kV CC line are of magnitude from 3-10nC [8]. This amount could be of several hundred nC in the real situation where many trees fall on the line due to the heavy storms. The minimum detectable magnitude of the PD using Rogowski coil sensor is 100pC [8]. Therefore, the measured value of the attenuation suggests that PDs caused by leaning a tree on the CC line could be detected by Rogowski coil sensor at a distance of 3-4km from the fault location. However, the detectable range of the distance could be
______________________________________________________________________________ longer if the amount of PDs produced will increase due to falling of many trees in heavy storms on practical CC lines having lower attenuation than the measured value. The results drawn from Fig. 11 are useful for the wave propagation analysis of PD signals produced due to falling trees on the CC line. The PD signals travel along the CC line and are detected by a sensor e.g. Rogowski coil. The attenuation of the CC line can be used as a design aid for the sensor location to get the required signal, so that it can be processed for detecting falling trees on it. The frequency-dependent CC line wave propagation characteristics can be used to model the line in EMTP/ATP simulation environment for further investigation of such fault location in an integrated distribution network.
9.
Conclusions
The theoretical modeling of the CC line based on its geometry has been presented using two wire TL theory and its frequency-dependent line characteristics have successfully been derived. It is revealed that the model does not only give correct values of the wave propagation characteristics at lower frequencies, but it also works well at higher frequencies, where TDR measurements fail to extract these characteristics due to measuring limitations. The frequency-dependent TL characteristics are extracted from the proposed theoretical model for the practical CC lines in MV networks located at different heights above the ground level. The attenuation of the CC line can be determined for a specific frequency band at which the PD signals produced due to falling trees propagate along the line. Depending upon the severity of fault, this quantity can assure the placing of a sensor to detect the falling trees on the CC line. In this way, the safety and reliability of the CC lines will improve; as well as resulting in reduced costs for inspection work The wave propagation characteristics obtained from the theoretical model can be used for the accurate modeling of the CC line in EMTP/ATP simulation environment. The simulation of the PD measuring system, including CC line and Rogowski coil PD sensor, will help deciding the number and positioning of the sensors to monitor the falling trees over a specific length of the CC overhead distribution lines.
10.
References
[1] H. Lehtinen, I. Lehtinen, and A. Hinkkuri, Research on Covered Medium-Voltage Overhead Lines in Finland, CIRED 1989, pp.179-184. [2] Igor Ticar, Joe Pihler, Oszkr Br, and Kurt Preis, Partial discharges in insulation of medium voltage systems, The international Journal for Computation and Mathematics in Electrical and Electronics Engineering, Vol. 20, No.2, 2001, pp.473-481. [3] Bob Hart, HV overhead line-Scandinavian experience Power Engineering Journal, June 1994, pp.119-123. [4] W. Pinheiro, et. al.," Study of Polymeric Material Characteristics on Covered Conductors used in Brazil", 1996 IEEE Annual Report-Conference on Electrical Insulation and Dielectric Phenomena, San Francisco, October 20-23, 1996, pp.416-419.
______________________________________________________________________________
[5] Accessories for Medium Voltage Distribution Networks-Ensto Overhead, a report by Utility Networks, Ensto Sekko Oy, Finland, 2006. [6] Joni Brunnsberg, Second Forest SAX project report, September 2003, http://www.forest-sax.com/. [7] W. Panosch, K. Schongrundner, K. Kominek,20 kV Overhead Lines with Covered Conductors, CIRED 2001, IEE Conference Publication No. 482, 18-21 June 2001. [8] G. Murtaza Hashmi, Mikael Nordman, and Matti Lehtonen, A Partial Discharge Detection Concept for Wireless Sensors in Covered-Conductors Distribution System, Proceedings of Europe's premier conference on electrical insulation (INSUCON 2006), Birmingham, UK, May 24-26, 2006. [9] P. C. J. M. van der Wielen , J. Veen, and P. A. A. F. Wouters, Evaluation of Different Types of Sensors and Their Positioning for On-line PD Detection and Localisation in Distribution Cables, Nordic Insulation Symposium, Tampere, Finland, June 11-13, 2003, pp.367-374. [10] H. W. Dommel, Electromagnetic Transients Program - Rule Book, Oregon, 1984. [11] G. Mugala and R. Eriksson, High frequency characteristics of a shielded medium voltage XLPE cable, in CEIDP Annu. Rep., 2001, pp. 132-136. [12] Ruslan Papazyan and Roland Eriksson, Calibration for Time Domain Propagation Constant Measurements on Power Cables, IEEE Transaction on Instrumentation and Measurement, Vol. 52, No. 2, April 2003. [13] G. Murtaza Hashmi, Mikael Nordman, and Matti Lehtonen, Determination of the Wave Propagation Characteristics for Partial Discharge Monitoring in Covered-Conductor Overhead Distribution Networks, Proceedings of the Modern Electric Power Systems conference (MEPS06), Wroclaw, Poland, September 6-8, 2006. [14] D. K. Cheng, Fundamental of Engineering Electromagnetics, Addison-Wesley Publishing, ISBN 0-201-56611-7. [15] H. Meng, S. Chen, Y. L. Guan, C. L. Law, P. L. So, E. Gunawan, and T. T. Lie, A Transmission Line Model for High-Frequency Power Line Communication Channel, Proceedings of International Conference on Power System Technology (PowerCon 2002), Kunming, China, October 13-17, 2002, pp.1290-1295. [16] G. C. Stone, S. A. Boggs, Propagation of Partial Discharge Pulses in Shielded Power Cable, Annual report of the conference on electrical insulation and dielectric phenomenon, National Academy of Science, Washington, DC, 1982, pp. 275-280. [17] G. Mugala, R. Eriksson, U. Gfvert, and P. Pettersson, Measurement Technique for High Frequency Characterization of Semi-Conducting Materials in Extruded Cables, IEEE Transaction on Dielectrics and Electrical Insulation, Vol. 11, No. 3, June 2004. [18] F. Rachidi, C. A. Nucci, and M. Ianoz, Transient Analysis of Multiconductor Lines Above a Lossy Ground, IEEE Transaction on Power Delivery, Vol. 14, No. 1, January 1999. [19] A. Ametani, Y. Kasai, J. Sawada, A. Mochizuki, and T. Yamada, Frequency-dependent impedance of vertical conductors and a multiconductor tower model , IEE Proceedings on Generation, Transmission, and Distribution, Vol. 141, No. 4, July 1994. [20] Waldemar Skomudek " Computer Analysis of Overvoltage Hazard due to Lightning Discharges in Medium Voltage Overhead Lines with Covered Conductors", Journal of Electrical Engineering,, Vol. 55, No. 5-6, pp.161-164,2004.
A review of On-Condition maintenance

Alireza Ahmadi Lulea University of Technology Division of operation and maintenance engineering
Abstract:
The need for reducing the uncertainty of maintenance activity, providing a more safe operation, increasing system availability and reduction of maintenance costs, has led to an increasing interest for developing of alternative maintenance policies. A maintenance policy which has the ability to help us to use the most of the useful life of the items, without interfering regular operation, and overcomes many of the disadvantages of the previous maintenance policies (i.e. time based and failure based), without increasing the risk of failure. This type of maintenance is known as On-Condition Maintenance. The main aim of this report is to give a general idea about the on-condition maintenance and to contribute for clarification of this maintenance policy. The discussion will include life curve, degradation process, and applicability and effectiveness criteria. Furthermore, a method, which has developed by NAVAIR RCM steering committee for estimating task intervals, will be discussed also.
Introduction
For years, it was believed that, based on the accepted wear and tear model of failure, every item had a fixed age at which either complete overhaul, or discard was necessary to ensure safety and operating reliability. Based on these concepts, there was a widespread belief that all failures could be prevented by age-based overhaul. Therefore, Time-Base Maintenance became the norm for Preventive Maintenance. This kind of approach motivated the indiscriminate use of overhaul or preventive replacement for all items included in a Preventive Maintenance program. (Tsang, 1995) In addition, the Federal Aviation Administration (FAA) in the US was concerned that the reliability of some engines had not been improved by changing either the type or the frequency of overhaul. The available data indicated that although the frequency of some failures had been reduced, many more had remained unchanged or actually increased. These findings could not be explained by using the accepted model of failure. Based on analysis of failure data, it was found that the probability of failure did not necessarily increase with operating age. Since there is seldom any strong relationship between the items operating time and failure probability, many failure modes did not benefit from agebased actions or shop visits. Therefore, the traditional time based policies were found to be ineffective for controlling the failure rate of many items (processonline, 2007). In fact, prescribing age-based actions for non-age-related failures had an adverse effect on the availability and reliability of the items, because it contradicted the actual failure characteristic and degradation processes. As a result, serviceable items were unnecessarily removed from service and sent to shops for overhaul or other maintenance measures. This caused maintenance costs to increase rapidly without equivalent improvements in reliability. The need for the provision of the safety, increased system availability, and reduced maintenance costs have led to an increasing interest in development of alternative maintenance policies. Based on the researches conducted, it has found that, even though many failure modes are not age-related, it is most often possible to identify and recognize the
conditions that indicate an imminent functional failure. Hence, it is possible to apply OnCondition measures, i.e. scheduled inspections, to detect a potential failure condition. This is a method to cope with the failure process proactively, and thereby avoiding the occurrence of functional failure, or its consequences. This maintenance policy has overcome many of the disadvantages of the previous maintenance policies (i.e. corrective and time based maintenance) and has proved its ability to extend the operating life of a system, without increasing the risk of failure that is On-Condition Maintenance.
On-Condition Maintenance
Based on several actuarial analyses, it has been discovered that very few failure modes conform to any age related failure patterns. For example, the outcomes of reliability research conducted by United Airlines in the 1960s showed that no more than 11 percent of the components would have benefited by a limited operating age or scheduled overhaul. In fact, after early period failure, 89 percent of the components had constant failure rate, without any wear age zone. Due to a combination of variations in applied stress and increasing complexity, deterioration is not always proportional to the applied stress, and stress is not always applied consistently (Moubray, 1997); (Nowlan & Heap, 1978). On the other hand, the component may remain as good as new for many years, but as time goes, both constant and stochastic strain will degrade the component, which will lose its original strength. At one point, the component will be exposed to larger stress than its remaining strength, and a failure will occurs (Heggest, et. al 2007). This may happen by incorrect operation like, starting up a machine too quickly without warm up, incorrect assembly like improper alignment of jointing, or some other reasons. In all of these examples when the items enter service it is not possible to predict when the failure will occur. For this reason, such failures are considered as random (Moubray, 1997). In reality, there is often little relationship between equipments operating age (or utilization units) and the probability of getting into the failed state. Even though many failure modes are not age related, it is most often possible to identify and recognize the conditions that indicates that a functional failure is about to occur or that is in the process of occurrence (NAVAIR 0025-400) . Most of the mechanical components do not fail suddenly, but give some sign of the fact that they are about to fail. The amount of warning given by different potential failures varies from microseconds to decades. Longer warning intervals mean greater maintenance task intervals. (AWB 02-1 Issue 1, 27 November 2001) On-condition maintenance can be defined as a scheduled inspection that is designed to detect a potential failures condition long before their occurrence, so that action can be taken to prevent the functional failure or to avoid its consequences. (Moubray, 1997); (Nowlan & Heap, 1978); (MIL-STD-2173). This type of maintenance uses inspection methods or condition monitoring techniques to determine either a problem exists in equipment , how serious the problem is, and how long the equipment can run before failure. The other means of OC maintenance is also to detect and identify specific components (e.g. gears, shafts, bearing) in the equipment that are degrading, and to determine the root cause of the problem( Mobley, 1990). In fact, it is a type of maintenance carried out in response to a significant deterioration in a unit as indicated by a change in the monitoring parameters of the units condition or performances. This means that, the main reason to perform any maintenance action is the change in the condition or performance of the item. Since the time to perform any maintenance action also is determined by monitoring the actual state of the system, its performance, or condition parameters, therefore, the item, which are
inspected, is allowed to be left in service on the condition, as long as they continue to meet specified performance standards until a potential failure is detected (Moubray, 1997). In other words, these tasks required repair or removal of specific components on the condition when they do not meet specified performance standards. On-condition maintenance is also known as Condition Based Maintenance (CBM) and Condition Directed Maintenance (CDM) (Moubray, 1997); (Tsang, 1995), because the need for corrective or consequence avoiding action, is based on the assessment of the condition of the item. In fact, CBM and CDM are on-condition tasks, which are designed to monitor the condition of an item in order to detect incipient failure modes (Kumar, et. al, 2000). Here, the decision maker has to consider the technical performance limits of the component, failure mechanism and degradation process, as well as the failure rate to judge whether immediate corrective action is required or if it is profitable to do some more inspection or to consider some issues, and extend the operational life of the component or system. It is also known as predictive maintenance because we are trying to predict whether, and possibly when, the item is going to fail based on its present behavior and condition (Moubray, 1997). The option of OC maintenance is appropriate on the condition that: Failure prevention is not feasible or how it can be achieved is not yet known, as in case where the event leading to failure occurs in a predominantly random manner In case where hazard rate is high and the impact of failure i.e. its consequences is too much to bear
On-condition tasks discriminate between the items (units, components...) which require maintenance to prevent a functional failure, and those that will probably meet required performance and could survive to the next inspection. In some cases, the time of the next inspection depends on the result of previous inspection. If the inspection exposes that the system is in early stage of degradation, the next inspection interval could be adjusted (i.e. reduced);(Welte, et.al, 2006) so that the degradation process being under control and maximum life length could be achieve. Therefore, each serviceable items remains in service and will be inspected or tested at regular intervals until its failure resistance falls bellow a defined level. This discrimination permits all items to realize most of their useful lives. (Nowlan & Heap, 1978) In aviation industries, the process of "on-condition" maintenance is applied to items on which a determination of their continued airworthiness can be made by visual inspection, measurements, tests or other means without disassembly inspection or overhaul. The condition of an item is monitored either continuously or at specified periods. The item's performance is compared to an appropriate standard to determine if it can continue in service. (AWB 02-1 Issue 1, 27 November 2001). Depending on the operating context of the item, warning of incipient failure enables the users of an asset to reduce or avoid consequences in a number of ways: (Moubray, 1997) Down time: corrective action can be planned at a time, which does not disrupt operations. 2. Repair costs: user may be able to take action to eliminate the secondary damage, which would be cause of another failure. It avoids also the premature removal of items that are still in satisfactory condition. Furthermore, the cost of correcting potential failure is often far less than the cost of correcting functional failures. This not only
1.
reduces the associated cost of repair, but also reduces the related downtime required for corrective action. 3. Safety: warning of failure provides time either to shut down the item before the situation becomes dangerous, or to move people out of harms way. On-condition task could be in a form of a scheduled inspection or examination that is designed to monitor the performance and/or the condition of an item in order to detect incipient failure. It can vary from visual inspection to more advance inspections using a variety of condition monitoring tools (Kumar, et. al, 2000). The four major categories of oncondition maintenance are: 1-condition monitoring techniques which uses specific equipment to monitor the condition of other equipment, 2-inspection techniques based on the human sense, 3-Primary effects monitoring techniques, which entail the intelligent use of the existing gauges and 4-process monitoring equipment and techniques based on variations in product quality. (Moubray, 1997). At the end of this part, it is important to note that sometime it is believed that "on-condition" means fit and forget, or do not do anything until a failure occurs. The above interpretations of "on-condition" maintenance may cause operational surprises, which could not only prove very costly, but also put at risk the safety. "On-condition" maintenance means an inspection/functional check that determines an item's performance and may result in the removal of an item before it fails in service. It is not a philosophy of fit until failure or fit and forgets. (AWB 02-1 Issue 1, 27 November 2001)
P-F Curve
As mentioned above, the point in the failure process at which it is possible to detect whether the failure is occurring or is about to occur is known as potential failure. Potential failure point can be defined also as the point in which the deterioration in condition pr performance can be detected (Kumar, et. al, 2000). By evidence of potential failure, it is possible to find out whether any failure has started to take some actions to prevent it to fail completely or to avoid the failure consequences if possible. In practice, task interval must be selected so that the detection of potential failure is possible before functional failure. (Moubray, 1997) Figure.2 shows a typical life or degradation curve for a single failure process. As shown, the item may remain as good as new for a long time. As time goes, at one point degradation starts to occur (point D) and continuous to deteriorate usually at an accelerating rate (Kumar, et. al, 2000), which is not necessarily related to age. Deterioration continues to point P, at which a potential failure can be detected. If it is not managed both constant and stochastic strain will be exposed to larger stress than its remaining strength, and a functional failure will occur (point F) (Welte, et. al., 2006). P-F interval has considerable variation from minutes to months or more and depends on the types of failure mechanism and speed of deterioration (Kumar, et. al, 2000). If a potential failure is detected between point P and F, it may be possible to take action to prevent or to avoid the consequences of the functional failures. The tasks designed to detect potential failures are known as on-condition tasks. (Moubray, 1997)
(n-1)th
3rd
1st
4th
2nd
Figure 2: P-F Curve
In addition to the potential failure, we need to consider the amount of time (or the number of stress cycles) which elapses, between the point at which it becomes detectable, and the point where it deteriorates into a functional failure. This interval is known as the P-F interval. The P-F interval tells us how often on-condition tasks must be performed. If we want to detect the potential failure before it becomes a functional failure, the interval between checks must be less than the P-F interval. (Moubray, 1997) The P-F interval is known as lead-time to failure (Rausand, 1998, cited by Kumar, et. al, 2000)), or failure development period. The longer the P-F interval, the more time one has to make a good decision and plan action(Kumar, et. al, 2000) .Note that if an on-condition task is done at intervals which are longer than the P-F interval, there is a chance that we will miss the failure altogether. On the other side, if we do the task at too small fraction of the P-F interval, we will waste resources on the inspection process (Moubray, 1997). In fact, the selected interval must provide a reasonable amount of time to take corrective action. By definition, the minimum interval likely to elapse between the discovery of potential failures and the occurrence of the functional failure is known as Nett P-F interval (Moubray, 1997). Another important definition in life curve study in on-condition maintenance is the delay time concept. The delay time of a defect is the time lapse from when a defect could first be detected upon inspection, until the time when its repair can be delayed no longer because of unacceptable consequences (Christer, 1982).
P-F Interval
Once potential and functional failure points have been defined, the P-F interval can be determined, which is the most challenging aspect of evaluating an on-condition task. The P-F interval can be defined by examining the change in trend of operational performance or the change in deterioration mechanisms of condition parameter (Kumar, et.al, 2000). This could be done also by examining test data, operational data, or based on expert judgement. (MILSTD-2173) The method used to determine P-F interval depends on the nature of the failure modes. For example, for failure modes with safety effect, the P-F interval is usually determined through laboratory testing or by analytical method. For example, the P-F interval for structural fatigue can be determined by fracture mechanics analysis. Fracture mechanics analysis predicts the
nth
number of cycles required to "grow" a crack from initiation, to its critical crack length, that is functional failure. (NAVAIR 00-25-400). When real test or operational data is not available to set up the P-F interval the following default actions are listed in order of recommended use and each requires adequate justification (MIL-STD-2173): 1. Using real data from similar equipment 2. If an identical on condition task currently exists, the existing task interval can be used as default. 3. When method 1 and 2 are not available, use a good estimate based on experience. Consider the properties of material, location of the item, operational stress, etc. provide rational for developing the estimate and give reasons for not following methods 1 and 2. When it is difficult to reveal the exact life curve, or it is expensive, a simplification may be appropriate for a certain type of component. One of the interesting simplified models has been introduced by handbook of condition monitoring, published by the Norwegian Electricity Industry (EBL), which is based on expert judgement. According to that, the continuous degradation of the component is simplified by dividing its life into four states. The states description is given in table 1. The item states are ranged from as good as new in state K=1, till critical situation, in state K=4 at which maintenance action must be taken immediately in order to avoid state K=5, that component completely fails and can not provide any services.(Heggset, et.al, 2007, )
Table 1: Technical condition states (Heggset, et.al, 2007)
In addition to this general state specification, more descriptions that are detailed are given in the handbook for different failure modes of all main components in a hydropower plant. Thus, the maintenance personnel have a guideline for the interpretation of different inspections and measurement results in order to define the condition of the component according to the five state- scale. Failure is always assumed to occur when there is a transition from state 4 to state 5 as indicated in figure 3. (Heggset, et.al, 2007, ) The length of the four main states has an element of uncertainty, which can be represented by a probability distribution. Preferably, the estimation of suitable probability distributions that describe the length of the four main states is based on analysis of reliability data and real observation. In case where the reliability data is rare, or there is not thrustable information, the best solution to model states interval is using expert judgement. (Heggset, et.al, 2007, )
Figure 3: Technical condition levels and life curve (Welte, T. Markove.)
For an on condition task to be applicable, the task interval must be reasonably consistent. (MIL-STD-2173) (Kumar, et. al, 2000)
Applicability criteria
Regardless of the method that we use to develop a scheduled maintenance program, the key question is whether, the selected maintenance task is able to fulfill its objectives or not. Therefore, maintenance task selection needs to have overriding criteria to recognize the fulfillment of these objectives. Designing to meet economic requirements and achieving competitive operation in private and in public sector organization depends on prudently balancing what is technically feasible and what is economically acceptable. Unfortunately, there is no short cut method available to reach this balance between technical and economical feasibility. Thus, methods of engineering economy analysis should be used to provide results that will help to attain an acceptable balance. (Sulivan, Wicks & Luxhoj, 2006) In order to achieve above-mentioned requirement, developing a scheduled maintenance program consists of determining which preventive task are both applicable (technically feasible) and effective (worth doing). The applicability of a task depends on the reliability knowledge (Rausand & Van, 1998), failure characteristics of the item and solely on the type of the task (MIL-STD-2173); (Nowlan & Heap, 1978). In general, a PM task is applicable when it can accomplish one of the following three objectives for performing PM (Rausand & Hoyland, 2004), (Smith & Hinchcliffe, 2004): 1. Prevent a failure, 2. Detect onset of a failure, 3. Discover hidden failure. There are also situations where economics and/or technical imitations can dictate a decision to do nothing in PM program definition. This option is called as Run-To-Failure (RTF) or No PM action (Smith & Hinchcliffe, 2004). Therefore, an applicable maintenance task must satisfy the requirements of the type of failure to restore its initial performance capability. In other words, a task is applicable if it eliminate a failure or at least reduce the probability of occurrence to an acceptable level (Hoch, 1990. Cited by Rausand & Van, 1998) or reduce the impact of failures. (Rausand & Van, 1998)
An on-condition inspection is applicable only if the failure characteristics of the item make it possible to define a potential failure condition in which detection of reduced failure resistance also would be possible. Moreover, a reasonably consistent and large enough age interval between the time of potential failure and the time of functional failure must exists (Moubray, 1997); (MIL-STD-2173); (Nowlan & Heap, 1978). On the other hand, if the Nett P-F interval associated with the minimum interval is long enough for suitable action to be taken to deal with the consequences of the failure, then the on condition task is technically feasible. Otherwise, it is not possible to establish a meaningful task interval, and the task in question should be again abandoned in favor of some other way of dealing with the failure. (Moubray, 1997) To develop an on-condition task, an identifiable and measurable degradation characteristic is required as an indicator of the items resistance to failure. For a specific failure mode, we may define different kind of degradation characteristics that can be used to reveal a potential failure condition. For example, several valid potential failure indicators for the failure mode wear could be specified as (NAVAIR 00-25-400): the amount of material lost through wear the level of vibration induced by a worn segment of a rotating component the intensity of heat generated by friction associated with wear.
After deciding which degradation characteristic that is most suitable to use as an indicator, we have to identify the potential to functional failure interval (P-F interval), and the measurement techniques. According to NAVAIR 00-25-400, in order to define an on condition task the following questions must be answered: 1. What degradation characteristic will be used as an indicator of the items failure resistance? 2. What will be defined as functional failure? 3. What will be defined as potential failure? 4. What is the potential failure - functional failure interval (P-F interval), and how consistent is it? 5. Can a task interval be developed that makes the task technically feasible? There are mainly five criteria, which an on-condition task must satisfy to be applicable i.e. technically feasible (Rausand & Hoyland, 2004); (MIL-STD-2173); (Nowlan & Heap, 1978); (Moubray, 1997): 1. It must be possible to detect reduce failure resistance for a specified failure mode. 2. A measurable parameter, which correlates with the onset of failure, could be identified. (Tsang, 1995) 3. It has to be possible also to determine a threshold value i.e. warning limits, for that parameter when action may be taken before full failure occurs (Tsang, 1995) 4. It must be possible to define a clear potential failure condition that can be detected by an explicit task. 5. There must be a reasonably consistent age interval between the time of potential failure and the time of functional failure (the P-F interval)
6. It would be practical to monitor the item at an intervals less than the P-F interval 7. The net P-F interval should be long enough to be able to take a proper action, to reduce or eliminate the consequences of the functional failure
Effectiveness criteria
Normally, answering to the applicability questions is quite clear-cut. This might be done through some research, lab test, or by trend analysis. However, answering to whether the task fulfills the objective in accordance with the effectiveness criteria is a more complex issue. After the applicable task is chosen, the effectiveness of that task in preventing the failure consequences must be determined. A PM task must provide a reduced expected loss related to personnel injuries, environmental damage, production loss, and/or material damage to be effective (Rausand & Hoyland, 2004). Effectiveness of a task is a measure of the result of the maintenance task objectives, which is dependant on the failure consequences (MIL-STD2173); (Nowlan & Heap, 1978). In other words, the maintenance tasks effectiveness is a measure of how well the task accomplishes the intended purpose and if it is worth doing (Hoch, 1990. Cited by Rausand & Van, 1998). The RCM process simply classifies all failure consequences in the following four groups: 1. 2. 3. 4. Safety & environmental consequences Operational consequences Non operational (economical) consequences Hidden failure consequences
The definition of economical consequences will vary from one operating organization to another, but in most cases, it includes any functional failure that has a direct effect on the operational capability or involves a failure mode with unusually high repair costs. (Nowlan & Heap, 1978) In general, the following criteria must be satisfied to make an on- condition tasks effective: Failures with safety/environmental consequences: the effectiveness criteria require that the task reduce the probability of critical failures to an acceptable level (MIL-STD2173). In addition, it will be effective (worth doing) if it can be relied on to give enough warning of the failure to ensure that action can be taken in time to avoid the safety or environmental consequences. (Moubray, 1997) Failure with hidden safety consequences: an on condition tasks intended to prevent a hidden failure should reduce the risk of multiple failures to an acceptably level. (Moubray, 1997) (MIL-STD-2173) If the failure does not involve safety (i.e. operational and economical consequences), the task must be cost effective, so over a period, the cost of doing the on condition tasks must be less that the cost of not doing it. (MIL-STD-2173); (Moubray, 1997); (Nowlan & Heap, 1978)
On-Condition task interval

The task interval is normally a fraction of the time between potential failure and functional failure. While item with economic consequences requires one inspection task, during the P-F interval, safety critical items may require many more inspections within the interval to ensure detection of the potential failure. The task interval should be adjusted by the ability of the task, to detect the potential failures. (MIL-STD-2173) If confidence in the inspection is low, more inspection might be performed to meet the effectiveness criteria. Confidence is based on the type of inspection, accuracy of the test equipment, and the skill level of the personnel performing the task. Task interval determination is a challenging issue as it depends on several factors such as acceptable probability of failure, probability of detecting failure during inspections, probability of failure Occurrence between inspection interval and quality of data. As it seems, many important issues must be taken into consideration during interval estimation. Therefore, it is strongly recommend to be sure that any mathematical or statistical methodology is consistent with available data and assumptions. No method or formula should be used unless there is a clear understanding of their derivations and that appropriate data or assumptions made are valid. (NAVAIR 00-25-400) NAVAIR RCM Steering Committee has been developed the following methods for estimating task intervals for the various types of RCM tasks. These methods have been reviewed and are endorsed by the NAVAIR RCM Steering Committee for use in applications where appropriate. This is just an example and it is the responsibility of the user to determine the appropriateness of a given method to a specific application.
Task Interval for safety consequences

On-condition task intervals are usually based on some fraction of P-F interval. One method of determining an on-condition task interval is based on the premise that any inspection will not be 100% effective. When acceptable probability of failure is the basis for an on-condition task interval (such as for safety and environmental failure modes), it will require some number of inspections over the P-F interval to ensure that the probability of failure is reduced to an acceptable level. (NAVAIR 00-25-400) The goal of a task with safety related consequences is to develop a task interval that will reduce the risk of failure to an acceptable level. In other words, it must reduce the probability of experiencing a failure to an acceptable level. According to (NAVAIR 00-25-400), this is the same as saying that the goal is to reduce the actual probability of failure (Pact) to the acceptable probability of failure (Pacc).
Pact = (1 )
Equation 1
Where: Pact = actual probability of failure = probability of detection during one inspection n = number of inspections during PF
This equation says that the actual probability of failure is equal to the probability of not detecting an existing potential failure (1 - ) in n inspections. As the aim is to reduce the actual probability of failure (Pact) to acceptable probability of failure (Pacc), we can interchange Pact to Pacc in the above Equation and solved it for n. Result would be:
n=
ln( p acc ) ln(1 )
Equation 2
Where: n = number of inspections during the PF interval = probability of detecting failure in one inspection Pacc = acceptable probability of failure Finally, we can calculate task interval by dividing P-F interval to the number of inspection:
t=
Where t = task interval PF = the potential to functional failure interval
PF n
Equation 3
Some consideration should be taken in to account: It should be considered that we should not use MTBF as the value of the P-F Interval, which are completely different values. 2. Pacc is calculated base on the assumption that the potential failure exists (100%). This is a fully conservative assumption and encourages the use of highly reliable inspection techniques. (NAVAIR 00-25-400) 3. The above equation should be used carefully since it is built on the assumption that each potential failure prevents one functional failure. (NAVAIR 00-25-400)
1.
If a more precise interval were desired, the actual failure rate and failure mechanism and distribution would need to be considered in more detail.
Task Interval for non-safety consequences

One method for determining the number of inspections, n, in the potential failure to functional failure interval for failure modes with non-safety consequences is to use a cost optimization formula such as the following: (NAVAIR 00-25-400)
MTBF Ci PF (C npm C pf )Ln(1 ) n= Ln(1 )
Amenable Where: PF = Potential failure to functional failure interval CI = Cost of one preventive maintenance task = (Direct Maintenance Man Hours (DMMH) for inspection) (Labor Cost per hour) + Consumable cost Cpf = Cost of correcting one potential failure = (DMMH to correct potential failure) (Labor Cost per hour) + Spares and Material costs Cnpm = Cost of not doing preventive maintenance = Ccm + Copc Ccm = Cost of corrective maintenance= (Direct Maintenance Man Hours (DMMH) for repair) (Labor Cost per hour) + (Spares and material costs) Copc = Costs due to operational impact (if established). (If not easily defined, one method to quantify this impact is to divide lost operational time due to unscheduled repair of functional failures by total planned operational time (e.g., divide unscheduled downtime hours by total planned available hours) and multiply by the acquisition cost of the end item.)
= Probability of detecting a potential failure with the proposed on-condition task, assuming
the potential failure exists. It should be noted that if Cnpm is equal to Cpf, there is no benefit in discovering potential failure before functional failure. (NAVAIR 00-25-400) The above equation could be interpreted as bellow (NAVAIR 00-25-400): For n 1, it is not cost effective to perform an on-condition task to address the failure mode under evaluation. For n 1, divide the PF by n to determine the appropriate task interval. For 0 < n < 1, the minimum cost effective task interval is obtained by setting n = 1 and PF becomes the task interval. (NAVAIR 00-25-400)
As it seems, there are challenges on OC task development such as defining acceptable probability of failure, assessing the probability of detecting failure by engineers during inspections and estimating the probability of failure occurrence between inspection intervals.
Conclusion:
OC maintenance uses indicators to assess the actual status of the equipment to predict the actual time to failure or loss of efficiency of the equipment so that the user is able to cope with the failure process proactively. In addition, getting proper information about the failure process, makes it possible for better planning and managing the consequences of failure. OnCondition maintenance concept plays an important role in defining maintenance needs of majority of systems and equipments as it increases the availability, hence using the most of the useful life of the equipment and reducing the associated cost of maintenance actions.
Thus, it is one of the preferred approaches to preventive maintenance in Reliability-Cantered Maintenance methodology. In this report, we discussed about issues that forced the industries to change their ideas of traditional time based maintenance, the concept of OC maintenance including life curve, failure process, its applicability, and effectiveness. We also presented a method, which has offered by MIL standard for defining OC task intervals. Task interval determination is a challenging issue as it depends on several factors. These factors include defining the life curve, acceptable probability of failure and probability of detecting failure by engineers during inspections. The majority of these challenges could be managed with relevant, valid, and reliable data. Therefore, data management is a vital task that must be consider for task development. References: AWB, On-condition maintenance, 02-1 Issue 1, 27 November 2001, Civil aviation of Australian government, access at: http://www.casa.gov.au/airworth/awb/02/001.htm Bertling, L., Reliability centered maintenance for electric power distribution systems, Ph.D. thesis, KTH, Stockholm, Sweden, 2002. Christer, AH. 1982, Modeling inspection policies for building maintenance, Journal of the operational Research Society, 33,723-732. EBL: Condition monitoring handbooks: Generator/ Turbine/ Waterway, cooling water and drainage equipment/ Control system, (four online books, in Norwegian), EBL, 1993-2005, Oslo, Norway. Heggest, J., Christensen, J.,Solvang, E., Wellet,T., Bakken, K., 2007, Assessment of remaining lifetime and failure probability for network components- A oractical approach, 19th International Conference on Electricity Distribution, Vienna, Paper no: 0661 Kobbaky, KAH. 1995. Condition Based Maintenance tools and decision-making. Journal of Quality in Main-tenance Engineering 1(3): 3-17. Kobbaky, KAH., Percy, DF., & Fawzi, BB. 1997. Small data set and preventive maintenance modeling. Journal of Quality in Maintenance Engineering 1(2): 136-140. Kumar, U.D., Crocker, J., Kenezevic, J., El-Haram, M., 2000. Reliability, Maintenance and Logistic Support. Kluwer Academic Publishers: US MILSTD-2173: Reliability Centered Maintenance. Washington D.C.: Department of Defense. Mobley, R.K., An Introduction to Predictive Maintenance, Van Nostrand Reinhold, New York, NY, 1990. Moubray, J. 1997. Reliability Centered Maintenance. Oxford: Butterworth-Heinemann. Nowlan, F.S. & Heap, H.F. 1978. Reliability Centered Maintenance. Springfield: National Techni-cal Information Service (NTIS). Rausand, M., Van. J. 1998. Reliability Centered Maintenance. Risk Reliability in Marin Technology, Balkeman, Holland. Smith, A.M. & Hinchcliffe, G.R. 2004. Reliability Centered Maintenance. Oxford: Elsevier. Sulivan, W.G., Wicks, E.M, Luxhoj, J., 2006. Engineering Economy. Thirteen ed. Pearson: New jersey Tsang, A. 1995. Condition Based Maintenance tools and decision-making. Journal of Quality in Main-tenance Engineering 1(3): 3-17. Tsang, A., Jardine, A., & Leung, B. 2006. Data management for CBM optimization. Journal of Quality in Maintenance Engineering 12(1): 37-51. Welte, T; Vatn, J; Heggset, J., 2006, Markov State Model for Optimization of Maintenance and Renewal of Hydro Power Components, Proceedings PMAPS 2006, Stockholm, Sweden.
10
Theme 4 Maintenance scheduling and optimization.

Finite stochastic dynamic programming for maintenance optimization by Francois Besnard e-Maintenance impact on availability performance and life support for geographically distributed complex technical systems by Ramin Karim
Finite Stochastic Dynamic Programming for Maintenance Optimization Maintenance Management in Power Systems
FRANCOIS BESNARD
Course by Assistant Professor Lina Bertling and Professor Matti Lehtonen
Contents
Contents 1 Introduction 1.1 1.2 1.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
III 1 1 2 2 3 3 6 6 7 9 9 17 21 22 23
2 Finite Horizon Stochastic Dynamic Programming 2.1 2.2 2.3 2.4 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimality Equation and Value Iteration algorithm . . . . . . . . . . The curse of dimensionality . . . . . . . . . . . . . . . . . . . . . . . Ideas for a Maintenance Optimization Model . . . . . . . . . . . . .
3 A Proposed Finite Horizon Replacement Model 3.1 3.2 3.3 3.4 One-Component Model . . . . . . . . . . . . . . . . . . . . . . . . .
Multi-Component model . . . . . . . . . . . . . . . . . . . . . . . . . Possible Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography
III
Abstract
This report is an introduction to an optimization approach, Stochastic Dynamic Programming, and its possible application to solve a maintenance problem. An introduction to the theory of Stochastic Dynamic Programming for nite horizon models is given with problem formulation and the value iteration method to solve the models. A maintenance model is proposed to illustrate the possible use of this approach. Some ideas and possible extensions of the model are discussed. Finite horizon Stochastic Dynamic Programming is interesting because it can integrate explicitely the stochastic behavior of failure. The main disadvantage of this approach is the time complexity of the method. In consequence, the number of components must be limited for the model to be tractable.
Chapter 1
Introduction
1.1 Background
The market and competition laws are introduced among power system companies due to the restructuration and deregulation of modern power system. The generating companies, as well as transmission and distribution system operators aim to minimize their costs. Maintenance costs can be a signicant part of the total costs. The pressure to reduce the maintenance budget leads to a need for ecient maintenance. Maintenance cost be divided into Corrective Maintenance (CM) and Preventive Maintenance (PM). CM means that an asset is maintained once an unscheduled functionnal failure occurs. CM can imply high costs for unsupplied energy, interruption, possible deterioration of the system, human risks or environment consequences etc. PM is employed to reduce the risk of unexpected failure. Time Based Maintenance (TBM) is used for the most critical components and Condition Based Maintenance (CBM) for the components that are worth and not too expensive to monitore. These maintenance actions have a cost for unsupplied energy, inspection, repair, replacement etc. An ecient maintenance should balance the corrective and preventive maintenance to minimize the total costs of maintenance. The probability of a functionnal failure for a component is stochastic. The probability depends on the state of component resulting from the history of the component (age, intensity of use, external stress (such as weather), maintenance actions, human errors and construction errors). 1
Stochastic Dynamic Programming (SDP) models are optimization models that integrate explicitely stochastic behaviors. This feature makes the models interesting and was the starting idea of this work.
1.2
Outline
Chapter 2 is an introduction to nite horizon stochastic dynamic programming (FHSDP) models. The formulation of problems and a method for solving the models are discussed. In Chapter 3, a replacement optimization model is proposed. The model is based on the FHSDP approach.
1.3
Remarks
This report is composed of 2 Chapters from a master thesis report [1]. The report focus on Stochastic Dynamic Programming models in general (for example innite horizon models are discussed). An introduction to deterministic dynamic programming is proposed in Chapter 4 in [1] that could be interesting to understand the SDP theroy for nite horizons. Moreover, a litterature study of application of SDP for maintenance optimization has been done in Chapter 8 in [1]. An interesting article for nite horizon SDP is [3] which describe a model to solve a nite horizon generating units maintenance scheduling. Few nite horizon SDP models have been proposed in the litterature while innite horizon models have been used extensively for condition monitoring maintenance. To learn more about Dynamic Programming, the book Markov Decision Processes: Discrete Stochastic Dynamic Programming [2] is recommanded.
Chapter 2
Finite Horizon Stochastic Dynamic Programming

This chapter is an introduction to the theory of Stochastic Dynamic Programming for nite horizon models. The proposed model in Chapter 3 is based on this chapter. For more details and examples, the book Markov Decision Processes: Discrete Stochastic Dynamic Programming [2] is recommended.
2.1
2.1.1
Problem Formulation
Notations for nite horizon SDP models
Numbers N NX NU Number of stages Number of state variables Number of control variables
Variables k i j u Stage State at the current stage State or possible state for the next stage Decision variable
State and Control Space Xk Uk (i) Uk State at stage k Decision action at stage k Optimal decision action at stage k for state i
Probability and Cost functions Ck (i, u, j ) C (i, u, j ) CN (i) fk (i, u, ) Pk (j, u, i) P (j, u, i) (i) Jk Sets X k U k (i) State space at stage k Decision Space at stage k for state i Cost function Cost function if the system is stationary Terminal cost for state i Dynamic function Transition probability function Transition probability function for stationary systems Optimal cost-to-go function from stage k to N starting at state i
2.1.2
Model Description
Dynamic Programming deals with multi-stage or sequential decisions problems. At each decision epoch, the decision maker (also called agent or controller in dierent contexts) observes the state of a system. It is assumed in this thesis that the system is perfectly observable. An action is decided based on this state. This action will result in an immediate cost (or reward) and inuence the evolution of the system. The aim of DP is to minimize (or maximize) the cumulative cost (respectively income) resulting of a sequence of decisions. Stochastic Dynamic Programming can be used to model systems whose evolution depends only on the current state, decision made and on a stochastic variable that describes the stochastic behavior of the system. A Stochastic Dynamic Programming model can be formulated as below: State Space A variable k {0, ..., N } represents the dierent stages of the problem. In general it corresponds to a time variable. The state of the system is characterized by a variable i = Xk of variables. The possible states are represented by a set of admissible states that can depends on k, Xk X k .
Decision Space At each decision epoch, the decision maker must choose an action u = Uk among a set of admissible actions. This set can depend on the state of the system and on the stage, u U k (i). Dynamic of the system and Transition Probability On the contrary with the deterministic case, the state transition does not depend only on the control used but also on a disturbance = k (i, u) Xk+1 = fk (Xk , Uk , ) The eect of the disturbance can be expressed with transition probabilities. The transition probabilities dene the probability that the state of the system at stage k+1 is j if the state and control are i and u at the stage k. These probabilities can depend also on the stage. Pk (j, u, i) = P (Xk+1 = j | Xk = i, Uk = u) If the system is stationary (time-invariant) the dynamic function f does not depends on time and the notation for the probability function can be simplied: P (j, u, i) = P (Xk+1 = j | Xk = i, Uk = u) In this case, one refers to a markov decision process. If a control u is xed for each possible state of the model, then the probability transition can be represented by a markov model. (See Chapter 3 for an example) Cost Function A cost is associated to each possible transition (i,j) and action u. The costs can also depend on the stage. Ck (j, u, i) = Ck (xk+1 = j, uk = u, xk = i) If the transition (i,j) occurs at stage k when the decision is u, then a cost C k (j, u, i) is given. If the cost function is stationary then the notation is simplied by C (i, u, j ). A terminal cost CN (i) can be used to penalize deviation from a desired terminal state. Objective function The objective is to determine the sequence of decision that optimize the expected cumulative cost (cost-to-go function) J (X0 ) where X0 is the initial state of the system:
N 1
J (X0 ) =
(Xk ) Uk U k
min
E {CN (XN ) +
k=0
Ck (Xk+1 , Uk , Xk )}
subject to Xk+1 = fk (Xk , Uk , k (Xk , Uk )) 5
2.2
Optimality Equation and Value Iteration algorithm
The optimality equation for stochastic nite horizon DP is:

Jk (i) = min E {Ck (i, u) + Jk +1 (fk (i, u, ))} uU (i) k
(2.1)
This equation dene a condition for a cost-to-go function of a state i in stage k to be optimal. The equation can be re-written using the probability transitions:
Jk (i) = min uU (i) k Pk (i, u, j ) (Ck (i, u, j ) + Jk +1 (j )) j X k+1
(2.2)
The value iteration algorithm for SDP problems is directly based on equation 2.2. The algorithm starts from the last stage. By backward-recursions it determines at each stage the optimal decision for each state of the system.
JN (i) Jk (i) Uk (i)
= = =
CN (i) i X N (Initialisation of the algorithm) While k 0 do

uUk (i)
min
Pk (i, u, j ) [Ck (i, u, j ) + Jk +1 (j )] j X k+1
i X k i X N
arg min
uUk (i)
Pk (i, u, j ) [Ck (i, u, j ) + Jk +1 (j )] j X k+1
k k1 The recursion nishes when the stage 0 is reached.
2.3
The curse of dimensionality
Consider a nite horizon stochastic dynamic problem with N stages NX states variables, the size of the set for each state variable is S NU control variables, the size of the set for each control variable is A The time complexity of the algorithm is O(N V 2NX ANU ). The complexity of the problem increases exponentionally with the size of the problem (number of state or decision variables). This characteristic of SDP is called the curse of dimensionality. 6
2.4
Ideas for a Maintenance Optimization Model
In this section, possible state variables for a maintenance models based on SDP are discussed.
2.4.1
Age and Deterioration States
The failure probability of components is often modelled as a function of time. A possible state variable for the component is its age. To be precise, the age of the component should be discretized according to the stage duration. If the lifetime of a component is very long it can lead to a very large state space. The time horizon can be considered to reduce the number of states. If a state variable can not reach certain states during the planned horizon, these states can be neglected. If a component, subcomponent or part of a system can be inspected or monitored, dierent levels of deterioration can be used as a state variable. In practice, both age and deterioration state variables could be used complementary. Of course maintenance states should be considered in both cases. It could be possible to have dierent types of failure states, as major failure and minor failures. Minor failures could be cleared by repair while for a major failure a component should be replace.
2.4.2
Forecasts
Some measurements or forecasts can sometime estimate the disturbance a system is or can be subject to. The reliability of the forecasts should be carefully considered. Deterministic information could be used to adapt the nite horizon model on their horizon of validity. It would also be possible to generate dierent scenarios from forcasts, solve the problem for the dierent scenarios and get some conclusions from the dierent solutions. Another way of using forecasting models is to include them in the maintenance problem formulation by adding a specic variable. It will reduce the uncertainties but in return increase the complexity. The proposed model in Chapter 3 gives an example of how to integrate a forecasting model in an electricity scenario. Another factor that could be interesting to forecast is the load. Indeed the production must always be in balance with the generation. Also if there is no consumption, some generation units are stopped. This time can be used for the maintenance of the power plant. Weather forecasting could also be interesting in some cases. For example the power 7
generated by wind farms depends on the wind strength, and maintenance action on oshore wind farms are possible only in case of good weather. For these two reasons, wind forecasting could be interesting for optimizing maintenance actions of oshore wind farms.
2.4.3
Time Lags
An important assumption of a DP model is that the dynamic of the system only depends on the actual state of the system (and possibly on the time if the system dynamic is not stationary). This condition of loss of memory is very strong and unrealistic in some cases. It is sometimes possible (if the system dynamic depends on few precedent states) to overcome this assumption. Variables are added in the DP model to keep in memory the precedent states that can be visited. The computational price is once again very high. For example, in the context of maintenance, it would be interesting to know the deterioration level of an asset at the precedent stage. It would give informations about the dynamic of the deterioration process.
Chapter 3
A Proposed Finite Horizon Replacement Model

A nite horizon SDP replacement model is proposed in this chapter. The model assumes a nite time horizon and discrete decision epochs. The system in consideration is a power generating unit. An interesting feature of the model is the integration of the electricity price as a state variable. Another is the possibility of opportunistic maintenance i.e if one component fails, it is possible to do preventive maintenance on another component that is still working. The proposed model is rst presented for one component and is then generalized to multi-components. Both these models can be solved using the value iteration algorithm.
3.1
3.1.1
One-Component Model
Idea of the Model
In this chapter, an age replacement model based on nite horizon dynamic programming is proposed. The model is rst described for one component for an easier understanding of its principle. The price of electricity was considered as an important factor that could inuence the maintenance decision. Indeed if the electricity price is high, it can be protable to operate the system and wait for lower prices. If a high electricity price is expected in a close future, it could be interesting to 9
do maintenance immediately to be operational later and avoid maintenance in a protable period. The idea was considered for the model. The electricity price was included as a state variable. The variable consider dierent electricity scenario, for example high, medium and low prices. For each scenario, the electricity price vary with a period of a year. There can be transitions from one scenario to another depending on the period of the year. In the scandinavian countries, a large part of the electricity is based on hydropower. The electricity price is in consequence highly inuenced by the weather. If the weather is warm and dry the hydro-storage will be low and the electricity price for the rest of the year may be high. On the opposite, a cold and rainy season may result in low electricity price for the rest of the year. This observation could be used to assume the electricity scenario to be transiant during the summer and stable during the rest of the year, typically interpreted as dry year or wet year. This assumption could be used as a base for modelling the transition for the electricity state.
3.1.2
Notations for the Proposed Model
Numbers NE NW NPM N CM Costs CE (s, k ) CI CP M C CM C N (i) Variables i1 i2 j1 j2 Component state at the current stage Electricity state at the current stage Possible component state for the next stage Possible electricity state for the next stage Electricity cost at stage k for the electricity state s Cost per stage for interruption Cost per stage of Preventive maintenance Cost per stage of Corrective maintenance Terminal cost if the component is in state i Number Number Number Number of of of of electricity scenario working state for the component preventive maintenance state for one component corrective maintenance state for one component
State and Control Space 10
x1 k x2 k
Component state at stage k Electricity state at stage k
Probability function (t) (i) Sets x 2 U (i)

1
Failure rate of the component at age t Failure rate of the component in state Wi
Component state space Electricity state space Decision space for state i
States notations W. P M. CM. Working state Preventive maintenance state Corrective maintenance state
3.1.3
Assumptions
The time span of the problem is T. It is divided into N stages of length Ts such that T = N Ts . The maintenance decision are made sequentially at each stage k=0,1,...,N-1. The failure rate of the component over the time is assumed perfectly known. This function is denoted (t). If the component fails during stage k, corrective maintenance is undertaken for N CM stages with a cost of C CM per stage. It is possible at each stage to decide to replace the component to prevent corrective maintenance. The time of preventive replacement is N P M stages with a cost of C P M per stage. If the system is not working, a cost for interruption C I per stage is considered. The average production of the generating unit is G kW. It means that if the unit is not in preventive maintenance or failure, G Ts kWh are produced during the stage. (Ts in hours) NE possible electricity price scenarios are considered. The prices are supposed xed during a stage (equal to the price at the beginning of scenario). For scenario s, the electricity price per kWh is noted CE (s, k ), k=0,1,...,N-1. It is possible that the electricity price "switch" from one scenario to another one during the time span. The probability of transition at each stage is assumed known. 11
A terminal cost (for stage N) can be used to penalize the terminal stage condition. The manpower is assumed unlimited. Spare parts are not considered.
3.1.4
3.1.4.1
Model Description
State Space
The state vector Xk is composed of two states variables, x1 k for the state of the 2 component (its age) and xk for the electricity scenario. NX = 2 The state of the system is thus represented by a vector as in (3.1): Xk = x1 k x1 1 , x2 2 x x k k x2 k (3.1)
x1 is the set of possible states for the component and x2 the set of possible electricity scenarios. Component state The status of the component (its age) at each stage is represented by one state variable x1 k . There are three types of possible states for the variable. Normal state (W), when the component is working, corrective maintenance (CM) states if the component is in maintenance due to failure and preventive maintenance (PM) states. The meaning of a state is that the component has been in the corresponing condition during the last stage. For example, if the component is in a state PM, it means that during the last stage it has undertaken preventive maintenance. The number of CM and PM states for the component corresponds respectively to N CM and N P M . To limit the size of the state space, it is necessary to limit the number of states W. It can be assumed that when (t) reaches a xed limit max = (Tmax ), preventive maintenance is always made. Another possibility is to assume that i (t) stays constant when age Tmax is reached. This approach was used. The corresponding number of W states is N W = Tmax /Ts or the closest integer in both cases.
12
CM2
CM1
1 Ts (0) Ts (1) Ts (2) Ts (3) Ts (4)
W0
(1 Ts (0))
W1
(1 Ts (1))
W2
(1 Ts (2))
W3
(1 Ts (3))
W4 (1 Ts
1 1
P M1 Figure 3.1: Example of Markov Decision Process for one component with N CM = 3, N P M = 2, N W = 4. Solid line: u=0, Dashed Line: u=1
Figure 3.1 shows an example of graphical representation of the MDP model for one x1 = {W , ..., W , P M , CM , CM }. The State component. In this example, x1 0 1 1 2 4 k W0 is used to represent a new component. P M2 and CM3 are both represented with this state. More generally, x = {W0 , ..., WN W , P M1 , ..., P MN P M 1 , CM1 , ..., CMN CM 1 }.
1
13
Electricity scenario state Electricity scenarios are associated with one state variable x2 k . There are NE possible states for this variable, each state corresponding to one possible electricity scenario. x2 = {S , ..., S } The electricity price of the scenario S at stage k is given x2 1 Ne k by the electricity price function CE (S, k ). Figure 3.2 shows an example for three possibles scenarios. The example considers three electricity scenarios correspond to high, medium and low electricity prices (respectively dry, normal and wet year). The weather during the season inuence the water reserve in a country as Sweden. Hydropower is a large part of the electricity generation in Sweden. Moreover this is a cheap source of energy. In consequence, if there is a low water reserve, more expensive source of energy are needed and the electricity price is higher.
Electricity Prices SEK/MWh
500
450 400 350

Scenario 3 1/3 Scenario 2 1/3
300
1/3
250
Scenario 1
200 k-1 k k+1 Stage
Figure 3.2: Example of electricity scenarios, NE = 3
14
3.1.4.2
Decision Space
At each stage, the decision maker can decide, if the component is not in maintenance, to do preventive maintenance or not depending on the state X of the system. Uk = 0 no preventive maintenance Uk = 1 preventive maintenance The decision space depends only on the component state i1 . U (i) = {0, 1} if i1 {W1 , ..., WN W } else
3.1.4.3
Transition Probabilities
The two state variables are independant. Moreover only the electricity state transitions depend on the stage. Consequently, P (Xk+1 = j | Uk = u, Xk = i)
1 2 2 1 1 2 2 = P (x1 k+1 = j , xk+1 = j | uk = u, xk = i , x = i ) 1 1 1 2 2 2 2 = P (x1 k+1 = j | uk = u, xk = i ) P (xk+1 = j | xk = i )
= P (j 1 , u, i1 ) Pk (j 2 , i2 ) Component state transition probability At each stage k, if the state of the component is Wq the failure rate is assumed constant during the time of the stage and equal to (Wq ) = (q Ts ). The transition probability for the component state is stationary. It can be represented as a Markov decision process as in the example in Figure 3.1. The transition porbabilities not equal to zero are summarized in table 3.1. Note that if N P M = 1 or N CM = 1 then P M1 respectively CM1 correspond to W0 . Electricity State The transition probabilities of the electricity state, Pk (j 2 , i2 ) are not stationary. They can change from stage to stage. 3.1.4.3 with 3.3 give an example of transition probabilities for the electricity scenarios on a 12 stages horizon. In this example, 1, P2 Pk (j 2 , i2 ) can take three dierent values dened by the transition matrices PE E 3 2 2 or PE . i is represented by the rows of the matrices and j by the column. 15
i1 Wq q {0, ..., N W 1} Wq q {0, ..., N W 1} WN W WN W Wq q {0, ..., N W } P Mq q {1, ..., N P M 2} P MN P M 1 CMq q {1, ..., N CM 2} CMN CM 1
u 0 0 0 0 1
j1 Wq+1 CM1 WN W CM1 P M1 P Mq+1 W0 CMq+1 W0
P (j 1 , u, i1 ) 1 (Wq ) (Wq ) 1 (WN W ) (WN W ) 1 1 1 1 1
Table 3.1: Transition probabilities 1 0 0 1 P E = 0 1 0 0 0 1 Stage(k) Pk (j 2 , i2 ) 0 1 PE

Table 3.2: Example of transition matrix for electricity scenarios 1 1 PE 2 1 PE 3 3 PE 4 3 PE 5 2 PE 6 2 PE 7 2 PE 8 3 PE 9 1 PE 10 1 PE
1/3 1/3 1/3 2 PE = 1/3 1/3 1/3 1/3 1/3 1/3
0.6 0.2 0.2 3 PE = 0.2 0.6 0.2 0.2 0.2 0.6 11 1 PE ... ...
Table 3.3: Example of transition probabilities on a 12 stages horizon
3.1.4.4
Cost Function
The costs associated to the possible transitions can be of dierent kinds: Reward for electricity generation= G Ts CE (i2 , k ) (depends on the electricity scenario state i2 and the stage k). Cost for maintenance, CCM or CP M . Cost for interruption, CI . Moreover, a terminal cost noted CN could be used to penalized deviations from required state at the end of time horizon. This option and its consequences was not studied in this work. The transition cost are summarized in Table 3.4. Notice that i2 is a state variable. A possible terminal cost is dened by C N (i) for each possible terminal state C N (i) for the component.
16
i1 Wq q {0, ..., N W 1} Wq q {0, ..., N W 1} WN W WN W Wq P Mq q {1, ..., N P M 2} P MN P M 1 CMq q {1, ..., N CM 2} CMN CM 1
u 0 0 0 0 1
j1 Wq+1 CM1 WN W CM1 P M1 P Mq+1 W0 CMq+1 W0
Ck (j, u, i) G Ts Cel (i2 , k ) C I + C CM G Ts CE (i2 , k ) C I + C CM CI + CP M CI + CP M CI + CP M C I + C CM C I + C CM
Table 3.4: Transition costs
3.2
Multi-Component model
In this section, the model presented in Section 9.1 is extended to multi-components systems.
3.2.1
Idea of the Model
The motivation for a multi-component model is to consider possible opportunistic maintenance. It is sometimes possible to do maintenance on dierent parts of the system at opportunistic times. For example if the system fails, it could be protable to do maintenance on some components of the system that are still working but should be maintained soon. This could be very interesting if the interruption cost is high or if the structure needed for the maintenance is very high. In wind power for example, for certain maintenance actions, an helicopter or a boat can be necessary. The price for their rent can be very high and it could be protable to group the maintenance of dierent wind turbines at the same time.
3.2.2
Notations for the Proposed Model
Numbers NC W Nc P Nc M CM Nc Number Number Number Number of of of of component working state for component c Preventive Maintenance state for component c Corrective Maintenance state for component c 17
Costs
PM Cc CM Cc N (i) Cc
Cost per stage of Preventive Maintenance for component c Cost per stage of Corrective Maintenance for component c Terminal cost if the component c is in state i
Variables ic , c {1, ..., NC } iNC +1 j c , c {1, ..., NC } j NC +1 uc , c {1, ..., NC } State of component c at the actual stage State for the electricity at the actual stage State of component c for the next stage State for the electricity for the next stage Decision variable for component c
State and Control Space xc k , c {1, ..., NC } xc C +1 xN k c uk State of the component c at stage k A component state Electricity state at stage k Maintenance for component c at stage k
Probability functions c (i) Sets x N +1 x C c u (ic )

c
Failure probability function for component c
State space for component c Electricity state space Decision space for component c in state ic
3.2.3
Assumptions
The system is composed of NC components in series. If one component fails, the whole system fails. The failure rate of each component over the time is assumed perfectly known. This function is noted c (t) for component c {1, ..., NC }. If component c fails during stage k, corrective maintenance is undertaken for CM stages with a cost of C CM per stage. Nc c It is possible at each stage to decide to replace a component to prevent corrective maintenance. The time of preventive replacement for component n is P M stages with a cost of C P M per stage. Nc c 18
An interruption cost C I is consider whatever the maintenance is done on the system. The average production of the generating unit is G kW. If none of the component of the unit is in preventive maintenance or failure, G Ts kWh is produced during the stage. (Ts in hours)
N can be used to penalize the terminal stage condition for A terminal cost Cc component c.
3.2.4
3.2.4.1
Model Description
State Space
The state of the system can be represented by a vector as in (3.2). x1 k . . .

c xN k
Nc +1 xk
Xk =
(3.2)
xc k , c {1, ..., NC } represent the state of component c.

c +1 xN represents the electricity state. k
Component Space The number of CM and PM states for component c corresponds respectively to CM and N P M . The number of W states for each component c, N W , is decided in Nc c c the same way that for one component. The state space related to the component c is noted x .
x = {W , ..., W xc W , P M 1, ..., P MN P M 1 , CM 1, ..., CMN CM 1 } 0 Nc k c c
c c
Electricity Space Same as in Section 3.1.
3.2.4.2
Decision Space
At each stage, the decision maker must decide for each component, that is not in maintenance, to do preventive maintenance or do nothing depending on the state of the system. 19
uc k = 0 no preventive maintenance on component n c uk = 1 preventive maintenance on component n The decision variables constitute a decision vector: u1 k u2 k Uk = . . .

(3.3)
c uN k
The decision space for each decision variable can be dened by: c {1, ..., Nc }, u (ic ) =
c
{0, 1} if ic {W0 , ..., WNc W} else
3.2.4.3
Transition Probability
The state variables xc are independent of the electricity state xNc +1 . Consequently, P (Xk+1 = j | Uk = U, Xk = i) = P ((j 1 , ..., j NC ), (u1 , ..., uNC ), (i1 , ..., iNC )) P (j NC +1 , j NC +1 ) (3.4) (3.5)
The probabilities transition of the electricity states, P (j NC +1 , iNC +1 ) , are similar to the one-component model. They can be dened at each stage k by a transition matrices as in the example of Section 8.1. Component states transitions The state variables xc are not independent of each other. Indeed, if one component fails or is in maintenance, the components are not ageing since the system is not working. In consequence, dierent cases must be considered. Case 1 If all the component are working, no maintenance is done, the propability transition of the whole system is the product of the probability transition of each component considered independently.
c If c {1, ..., NC }, yk {W1 , ..., WNn W }, NC
P ((j 1 , ..., j NC ), 0, (i1 , ..., iNC )) =

c=1
P (ic , 0, j c )
Case 2 20
If one of the component is in maintenance or the decision of preventive maintenance is:

NC
P ((j , ..., j
NC
), (u , ..., u
NC
), (i , ..., i
NC
)) =
n=1
Pc
with P c =
c c c c W} P (j , 1, i ) if u = 1 or i {W1 , ..., WNc

c c 1 if ic {W0 , ..., WNc W 1 } and i = j
0 else
3.2.4.4
Cost Function
As for the transition probabilities, there are 2 cases: Case 1 If all the components are working, no maintenance is decided and no failure happens, a reward for the electricity produced is obtained.
c , If c {1, ..., NC }, yk {W1 , ..., WNn W}
C ((j 1 , ..., j NC ), 0, (i1 , ..., iNC )) = G Ts CE (iNC +1 , k ) Case 2 When the system is in maintenance or fails during the stage, an interruption cost C I is considered as well as the sum of all the maintenance actions.
NC
C ((j 1 , ..., j NC ), (u1 , ..., uNC ), (i1 , ..., iNC )) = C (I ) +

c=1
Cc
with C c =
c CM if ic {CM , ..., CM CM } or j = CM1 Cc 1 Nc

c 0 else
n = PM C P M if ic {P M1 , ..., P MNc P M } or j 1
3.3
Possible Extensions
The model could be extended in several directions. The following list summarizes some ideas on issues that could impact on the model: Manpower. It would be interesting to limit the number of maintenance actions possible to do at the same time. A solution would be to consider a global decision space and not individual decision space for each component state variable. 21
Include other types of maintenance actions. In the model, replacement was the only maintenance action possible. In reality there are a lot of possible maintenance actions, such as minor repair, major repair etc. They could be modelled by adding possible maintenance decisions in the model. Time to repair is non deterministic. So that it is possible to model a stochastic reparation time by adding probabilities transition for the maintenance states. Use of deterioration states. If monitoring or inspection of some components are possible, deterioration state variables could be included in the model. Other forecasting states. It could be interesting to add other forecasting state information such as weather and/or load states.
3.4
Conclusion
Finite horizon models based on stochastic dynamic programming could be interesting for short-term planning of maintenance. The approach is limited to systems of a few components due to the curse of dimensionality. The time complexity of the value iteration algorithm increase exponentionnaly with the number of components in the system.
22
Bibliography
[1] Franois Besnard. On stochastic dynamic programming and its application to maintenance. Masters thesis, Royal Institute of Technology (KTH), June 2007. [2] Martin L. Puterman. Markov Decision Processes: Discrete Stochastic Dynamic Programming. John Wiley & Sons, Inc., 1994. [3] A. Rajabi-Ghahnavie and M. Fotuhi-Firuzabad. Application of markov decision process in generating units maintenance scheduling. In Probabilistic Methods Applied to Power Systems, 2006. PMAPS 2006. International Conference on, pages 16, 2006.
23

Trita-Ee 2007 043

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Trita-Ee 2007 043

Hochgeladen von

Copyright:

Verfügbare Formate

Maintenance Management in Power Systems

Helsinki Univeristy of Technology Power Systems Laboratory 02015 Helsinki

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

1. Course program and final seminar schedule

KTH/School of Electrical Engineering/RCAM

Maintenance Management in Power Systems (5 credits)

2C5060 Maintenance Management in Power Systems

Carl-Johan Wallnerstrm Anna Guldbrand

Carl Johan Wallnerstrm

11.1512.30 12.3013.00 13.0013.30

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

3 List of proposed topics and guidelines for the report preparation

Guidelines for the report preparation

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

List of proposed topics

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

KTH/School of Electrical Engineering/RCAM

Course material for Maintenance Management in Power Systems 2007

Start up theory day including

KTH/School of Electrical Engineering/RCAM

Stocastic modells of the Electric Power System

Analytical or simulation method

Input data Tools

Quantitative measures for reliability

Maintenance Management in Electric Power Systems

Introduction Power system

Maintenance Management in Electric Power Systems

Introduction Historical changes

Maintenance Management in Electric Power Systems

Introduction Main questions

Maintenance Management in Electric Power Systems

Introduction Main difficulties

Maintenance Management in Electric Power Systems

Maintenance Management in Electric Power Systems

Maintenance Management in Electric Power Systems

Introduction Basic concepts

Maintenance Management in Electric Power Systems

Introduction Basic concepts

* 3=4 for non repairable components

Maintenance Management in Electric Power Systems

Introduction Application areas

Maintenance Management in Electric Power Systems

DAMAGING FAULT Two models of failure Permanent forced outages

NON-DAMAGING FAULT Two models of restoration

Transient forced outages Automatic switching

Temporary forced outages Manual switching or fuse replacement

Maintenance Management in Electric Power Systems

Maintenance Management in Electric Power Systems