Available  online  at  www.sciencedirect.

com  
Available  online  at  www.sciencedirect.com  
ScienceDirect  
Available ScienceDirect  
online
Procedia
Available atonline
CIRPwww.sciencedirect.com
00 (2019) 000–000
at www.sciencedirect.com
www.elsevier.com/locate/procedia
ScienceDirect
Procedia CIRP 00 (2019) 000–000
ScienceDirect www.elsevier.com/locate/procedia

Procedia CIRP 00 (2017)

Procedia 000–000
CIRP 81 (2019) 300–305
www.elsevier.com/locate/procedia
52nd CIRP Conference on Manufacturing Systems
52nd CIRP Conference on Manufacturing Systems
A Cyber-Physical Failure Management System for Smart Factories
A Cyber-Physical Failure
28th CIRP DesignManagement System
Conference, May 2018, forFrance
Nantes, Smart Factories
Matthias Schneider *, Dominik Lucke , Thomas Adolf
a, a,b a

A new methodology
a
Institute forto
Fraunhofer Matthias
b
analyze
Schneider
Manufacturing
a,
*,the
Engineering functional
Dominik
and Lucke
Automation
a,b
and
, Thomas
IPA, Nobelstraßephysical
Adolf
12, 70569
a
architecture
Stuttgart, Germanyof
Hochschule Reutlingen, ESB Business School, Alteburgstr. 150, 72762 Reutlingen, Germany
existing products a
for an
* Corresponding author. Tel.: +49-711-970-1658.
b
assembly
E-mail address:
Hochschule Reutlingen,
oriented product family identification
Fraunhofer Institute for Manufacturing Engineering and Automation IPA, Nobelstraße 12, 70569 Stuttgart, Germany
matthias.schneider2@ipa.fraunhofer.de
ESB Business School, Alteburgstr. 150, 72762 Reutlingen, Germany

* Corresponding author. Tel.: +49-711-970-1658. E-mail address: matthias.schneider2@ipa.fraunhofer.de

Paul Stief *, Jean-Yves Dantan, Alain Etienne, Ali Siadat
Abstract École Nationale Supérieure d’Arts et Métiers, Arts et Métiers ParisTech, LCFC EA 4495, 4 Rue Augustin Fresnel, Metz 57078, France
Abstract
In smart factories, maintenance is still an important aspect to safeguard the performance of their production. Especially in case of failures of
* Corresponding author. Tel.: +33 3 87 37 54 30; E-mail address: paul.stief@ensam.eu
machine components diagnosis is a time-consuming task. This paper presents an approach for a cyber-physical failure management system, which
In smart factories, maintenance is still an important aspect to safeguard the performance of their production. Especially in case of failures of
uses information from machines such as programmable logic controller or sensor data and IT systems to support the diagnosis and repairing
machine components diagnosis is a time-consuming task. This paper presents an approach for a cyber-physical failure management system, which
process. Key element is a model combining the different information sources to detect deviations and to determine a probable failed component.
uses information from machines such as programmable logic controller or sensor data and IT systems to support the diagnosis and repairing
Furthermore, the approach is prototypically implemented for leakage detection in compressed air networks.
Abstract
process. Key element is a model combining the different information sources to detect deviations and to determine a probable failed component.
Furthermore, the approach is prototypically implemented for leakage detection in compressed air networks.
In©today’s
2019 The Authors.
business Published by
environment, theElsevier Ltd. This
trend towards is anproduct
more open access
varietyarticle under the CC BY-NC-ND
and customization is unbroken.license
Due to this development, the need of
© 2019 The Authors. Published by Elsevier
(http://creativecommons.org/licenses/by-nc-nd/3.0/)
agile and reconfigurable production systems Ltd.
emerged to cope with various products and product families. To design and optimize production
© 2019 The
This is an openAuthors.
access Published by
article under Elsevier Ltd.
the scientific
CC BY-NC-ND This is an open
license access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/)
Peer-review
systems as under
well as responsibility
to choose the of the
optimal product committee
matches, of the
product52nd CIRPmethods
analysis Conferenceare on Manufacturing
needed. Indeed, Systems.
most of the known methods aim to
(http://creativecommons.org/licenses/by-nc-nd/3.0/)
Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems.
analyze a product
Peer-review or one product
under responsibility family on the
of theIndustrie physical
scientific level. Different product
committee of cyber-physical
the 52nd CIRP families, however, may differ largely
Conference on Manufacturing Systems. in terms of the number and
Keywords:
nature manufacturing;
of components. smart
This dactory;
fact impedes 4.0; maintenance;
an efficient comparison and choice system
of appropriate product family combinations for the production
system. A new
Keywords: methodology
manufacturing; smartisdactory;
proposed to analyze
Industrie existing products
4.0; maintenance; in viewsystem
cyber-physical of their functional and physical architecture. The aim is to cluster
these products in new assembly oriented product families for the optimization of existing assembly lines and the creation of future reconfigurable
assembly systems. Based on Datum Flow Chain, the physical structure of the products is analyzed. Functional subassemblies are identified, and
a1.  Introduction
Even in future smart factories, unplanned machine
functional analysis is performed. Moreover, a hybrid functional and physical architecture graph (HyFPAG) is the output which depicts the
similarity between product families by providing design support to both, production breakdowns
Even insystemare still possible
smartandas
planners it is not
product economic
designers. An to monitor
illustrative
1.  Introduction future factories, unplanned machine
The of
example recent years trends
a nail-clipper is usedinto factories
explain thetowards extensive Anevery
proposedanmethodology. component
industrial case studywith
on physical
two productsensors.
families Then
of a rapid
steering
breakdowns are still possible as it is not economic to monitor reaction
columns of
application
thyssenkrupp of lean
Presta principles,
France is then towards
carried more
out to complex
give a firstmachines
industrial and
evaluationrepair
of theis required
proposed to minimize
approach. the time to repair
every component with physical sensors. Then a rapid reaction and finally
The recent years trends in factories towards an extensive
©and2017equipment
The Authors.andprinciples,
the wideby
Published introduction
Elseviermore
B.V. ofcomplex
IT systems led to the downtime
and of the machine.
repair is required to minimizeTherefore,
the time failure management
to repair and finally
application of lean towards machines
closer linked
Peer-review underprocess steps,
responsibility also
of the acrosscommittee
scientific companies.of the An
28th CIRPapproaches with supporting
Design Conference 2018. IT-tools such as ticketing systems
the downtime of the machine. Therefore, failure management
and equipment and the wide introduction of IT systems led to
unplanned breakdown ofsteps,
a machine can have far-reaching or automaticwith
approaches monitoring
supporting and IT-tools
notification suchsystems or knowledge
as ticketing systems
closer
Keywords: linked
Assembly; process
Design method; also
Family across companies.
identification An
consequences in the entire value adding chain. A result of this management systems exist since years.
or automatic monitoring and notification systems or knowledge But especially the
unplanned breakdown of a machine can have far-reaching
development is high failure cost. The importance of diagnosis still can be a time-consuming
management systems exist since years. But especially the task. To shorten the
consequences in the entire value adding chain. A result of this
maintenance increases, as its main objective is to ensure the diagnosis times one key element is to provide
diagnosis still can be a time-consuming task. To shorten the the maintenance
development is high failure cost. The importance of
1.operability
Introduction ofincreases,
a factoryasat its a minimum of cost. Maintenance workers
of with
the product
diagnosis timesrelevant
range
one key and information. is to In
characteristics
element this context,
manufactured
provide relevant
and/or
the maintenance
maintenance main objective is to ensure the
must evolve in order to safeguard the performance of a factory. information
assembled
workers with can
in this be the machine
system.information.
relevant documentation,
In this context, the main
In this maintenance
challenge
context, relevantin
operability of a factory at a minimum of cost. Maintenance
Cyber-physical systems (CPS) theare performance
becoming
in themore
history including beinspections, failures and repairs, similar
Due
must to in
evolve theorder fast development
to safeguard of aand
domain moreof
factory. modelling
information andcananalysis is now not
the machine only to copemaintenance
documentation, with single
established in factories failure
history situations and repair instructions of
andsame machines or
communication
Cyber-physical an and
andsystems ongoingoffer
(CPS) are the
trend opportunity
becomingof digitization
more and to makeand
more products, a limited product
including inspections, rangefailures
or existing product
repairs, families,
similar
factories smart and to dissolve the rigid hierarchies in component types. The most important information is the
digitalization,
established inmanufacturing
factories and enterprises are facing important
offer the opportunity to make but also situations
failure to be able and to analyze
repair and to compare
instructions products
of same to define
machines or
production into a flexible production network. possible damaged
types. component and its location. To solve is this
challenges
factories smart in today’s and tomarket dissolve environments:
the rigid This production
a continuing
hierarchies in new product families.
component The It can
most be observed
important that classical existing
information the
network isintoformed by production
CPS and network. makes This numerous new localization
possible challenge,
damaged one idea is toitsuse a partial Todigital twin
tendency
production towards a reduction
flexible of product development times and
production product families arecomponent
regrouped inand function location.
of clients orsolve this
features.
information sources available. [1][2] Different concepts and of the machine
localization or
challenge, equipment
one idea that detects
is tofamilies during
use a partial operation
digital twin
shortened
network productis formed lifecycles.
by CPS In addition,
and makes there is an increasing
numerous new However, assembly oriented product are hardly to find.
frameworks sources for smart factories already exist. [3][4][5] deviations
of On
thethe with a minimized
machine numberdetects of physical sensors, as
information
demand of customization, available.
being[1][2]
at the Different
same timeconceptsin a globaland productorfamily equipmentlevel, that
products differ during
mainly operation
in two
Therefore, a promising approach is the extensive application of they are
deviations expensive or
with a minimized not applicable
number due to harsh environments.
of physical and sensors, as
frameworks with
competition for competitors
smart factories all over already exist.This
the world. [3][4][5]
trend, main characteristics: (i) the number of components (ii) the
integrated digital tools approach
and CPS is tothe
leverage further potentials But
theyonly through the
arecomponents
expensive combination
or(e.g. applicableofdue
not mechanical, thetoinformation provision
harsh electronical).
environments.
Therefore,
which a promising
is inducing the development extensive
from macro application
to micro of type of electrical,
of the digitalization of manufacturing. [6][7] and analysis
ButClassical
only further
through the potentials
combination forof maintenance
themainly
informationcan be raised.
integrated
markets, digitalintools
results and CPSlot
diminished to leverage
sizes duefurther potentials
to augmenting methodologies considering single provision
products
of the digitalization
product of manufacturing.
varieties (high-volume [6][7] production) [1].
to low-volume andsolitary,
or analysis already
further potentials for maintenance
existing product families can be raised.
analyze the
2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
To cope with this augmenting variety
(http://creativecommons.org/licenses/by-nc-nd/3.0/) as well as to be able to product structure on a physical level (components level) which
identify
2212-8271 possible
Peer-review ©under The optimization
2019responsibility
Authors. ofPublished potentials
by Elsevier
the scientific in This
Ltd.
committee ofthe existing
theis52nd
an open
CIRPaccess causes on difficulties
article under
Conference the CC BY-NC-ND
Manufacturing regarding
Systems. license an efficient definition and
(http://creativecommons.org/licenses/by-nc-nd/3.0/)
production system, it is important to have a precise knowledge comparison of different product families. Addressing this
Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems.
2212-8271 © 2019 The Authors. Published by Elsevier Ltd.
This is an©open
2212-8271 2017access article Published
The Authors. under theby CC BY-NC-ND
Elsevier B.V. license (http://creativecommons.org/licenses/by-nc-nd/3.0/)
Peer-review
Peer-review under
under responsibility
responsibility of scientific
of the the scientific committee
committee of the of theCIRP
28th 52ndDesign
CIRPConference
Conference2018.
on Manufacturing Systems.
10.1016/j.procir.2019.03.052
 Matthias Schneider et al. / Procedia CIRP 81 (2019) 300–305
2 Matthias Schneider et al. / Procedia CIRP 00 (2019) 000–000
Apart from the effort for development of application specific
diagnosis functions the integration of the different models and
systems cost a lot of time and effort. Also, the IT-Systems must
be capable to process and consider the permanently growing
amount of collected information from different sources. All
these challenges demand new integrated cyber-physical failure
management systems that simplify and speed up the application
specific implementation.
This paper presents an approach for a cyber-physical failure
management system, capable to use information from
machines such as programable logic controllers (PLC), sensor
data of CPS and IT systems to support the diagnosis and
repairing process. Core element is an information model
creating at runtime a partial digital twin, standardizing the
information representation. In this partial digital twin for
purpose of failure management the different information
sources are combined to detect deviations and to determine the
location of a probable failed component. Key element on the
technical level is a highly scalable search engine which is
capable to process the large growing amount of information
performantly.
The approach is prototypically implemented to detect
failures in the form of leakages and operating interrupts in a
compressed air network. The share of energy costs in
generation can amount to up to 80% of operating costs.
Compressed air is thus the most expensive energy source in the
industrial environment and offers great potential for savings
measures. [8][9] Leakage detection and repairing in a
compressed air network is still a problem. Between 5% up to
50% of the total compressed air, consumption can be caused by
leakages [8].
The concept of a cyber-physical failure management
system raises some challenges such as the digital representation
of a factory, the collection and analysis of large amounts of
information from different sources as well as the detection and
Factory
Management
Maintenance
Worker
Approaches to
efficiency
increase
Cost
overview
Notification
about failures
Analysis
Failure information Reasoning
(e.g. failure causes,
failure solutions)
Failure Historical information
Management
Manual
failure report
Virtual
Notification Sensors
about failure
Machine
Operator
Fig. 1. Concept of the cyber-physical failure management system.
301
management of failures during running production. The focus
of this paper is on the architecture and the information model
of the failure management system as basis for further analysis
methods. The failure detection which is in the focus of our
future work uses a graph-based approach and model to
calculate and compare the current air flow at the measuring
point based on typical consumptions values of the machines
and components behind. By combining the typical
consumptions values with their current readable PLC status
values, a near real time calculation can be achieved and used
for a deviation detection and failure localization.
2.  Related Work
Failure management deals with the recognition, prevention
and reaction to failures. The goal is to minimize the cost caused
by the failure itself as well as the cost for situation-specific
measures to eliminate the failure. Since the attempt to eliminate
all failures is usually neither practicable nor economically
sensible, it is important to find the optimum between the two
cost categories and to prioritize failures. A failure management
system can therefore rather be an extra module to computerized
maintenance planning and control systems (CMMS). Since the
systems meanwhile offer more and more functions, a clear
demarcation between the individual systems is not always
possible or reasonable. [10]
Failure management systems are often also implemented in
the form of ticket systems. Such systems are called issue
tracking systems or helpdesk systems such as zendesk [11] or
freshdesk [12]. With these systems, the focus is on the
management of failures. Each failure or request is stored in the
form of a ticket that is uniquely identifiable and contains related
information. [13] As these systems are designed for service
desks and not directly for maintenance, they do not have
production related analysis or diagnosis capabilities.
Further
information sources
Production information
(e.g. orders)
Manually captured
information (e.g. counter states)
Information
Machine models,
Consumption profiles,
Information
Machine documentation,
Data Pool
Maintenance history,
...
Machine
Sensor
value
Sensor values,
Operating states, CPS
PLC data,
...
PLC
Production
Machine
302 Matthias Schneider et al. / Procedia CIRP 81 (2019) 300–305
Matthias Schneider et al. / Procedia CIRP 00 (2019) 000–000
For failure detection numerous methods in manufacturing
systems (e.g. signal processing, machine learning or trend
analysis) exist which mainly focus on single manufacturing
systems or technical processes and their directly related sensor
data. [14] This kind of failure detection systems often do not
use indirect information sources like production IT-systems.
3.  Concept
A core element of the cyber-physical failure management
system is an information model which combines the numerous
information sources in smart factories. This enables the
creation of a virtual representation of the factory in the form of
a partial digital twin. With this virtual representation of the
factory it is possible to detect deviations and determine failures
and thus support the maintenance workers in their work with
failure relevant information.
The concept of the cyber-physical failure management
system is depicted in Figure 1. Different information sources
such as machine models or operation profiles represent the
smart factory virtually. Beside static information such as
machine documentation or repair instructions, this includes
also dynamic information like actual sensor values or operating
states of machines, production orders, or maintenance histories.
All information is stored in an information pool, which forms
the basis for detection of failures. The failure management
system analyses the information pool and attempts to detect
failures in factories automatically. If a failure is detected, the
probability of localization is determined. Detected failures are
stored and an associated maintenance requests is generated.
This enables the responsible maintenance worker to be
automatically notified and provided with relevant failure
information. Relevant failure information is e.g. the probable
failed component, the probability of localization of the failed
component or, if similar failures have already occurred in the
past, suggested solutions for repair of the failure. After the
failure has been repaired, the maintenance worker returns
information about the cause of the failure and how to repair it
back to the failure management system. This results in a
feedback loop for information which on the one site flows into
a bigger information basis for improved failure detection and
on the other site the support for maintenance workers can be
improved for already known failures or similar failure types. In
addition to automatic failure detection, failures can also be
reported manually by machine operators or the maintenance
workers. In this case additional context information can be
provided directly by the reporter.
3.1.  Information model
The information model enables the combination of different
information sources. All information that is represented in the
system is described by this information model. It provides the
basis for the virtual representation of a factory and for the
analysis to detect and locate failures. The information model is
divided into several sub-models:
3
•   Factory Structure: This sub-model mainly contains
static information which describe the real structure of a
factory. This means that the physical objects (e.g.
machines, supply components, sensors, CPS) as well as
their relationships to each other are modeled.
•   Dynamic Factory Information: The sub-model
contains the dynamic factory information (e.g. sensor
values or machine states). It refers to the (static) structure
described in the sub-model Factory Structure and thus
enables a virtual representation of the current state of the
factory.
•   Failure Management: This sub-model manages all
information relevant to failures. This includes failure
reports and their derived maintenance requests and
reports about failure elimination. By linking to
maintenance assets, failure information is linked to
factory information, which enables the provision of
failure relevant context information.
•   User Management: The data of the users of the system
are stored here. Different user groups can be
administered with their capabilities. This enables an
automated assignment and notification of failure types to
a certain group of maintenance workers (e.g. electricians,
facility managers, machine specialists). The
authentication and rights management of the users are
also stored here.
•   Configuration: Since the system is designed to be
generic and targets different types of factories, the
system can be configured accordingly and adapted to the
circumstances of the factory where it is used.
In the information model physical objects such as PLCs,
CPS, machines or smart sensors are represented by an abstract
description containing describing information and state
variables. The location is one of the most important state
variables of an object, since it is the basis for the localization
of failures. A geographical addressing based on longitude and
latitude is used for the location. This makes it possible to
describe the location of an object in the real world uniquely and
globally valid. Thus, objects can be easily localized even across
company locations and the analysis can be extended by
geographical search queries. Relations between the objects
create a virtual representation of the real factory. In addition to
the relations between the objects, an object can also be related
to a maintenance asset, which links the object with the actual
failure management.
The direct integration of CPS and other IT-Systems as well
as their consideration in the data model are an important point
of the cyber-physical failure management system. CPS also
have a certain intelligence that enables them to monitor
themselves and to provide self-diagnostic capabilities. This
means that not all monitoring and analysis functions have to be
provided by a central system. The failure management can be
done in such a way that the failure detection capabilities of CPS
are managed and orchestrated as well as supplemented with
comprehensive analysis functions of the failure management
system.
 Matthias Schneider et al. / Procedia CIRP 81 (2019) 300–305
4 Matthias Schneider et al. / Procedia CIRP 00 (2019) 000–000
3.2.  Basic functionalities
The overall function of the cyber-physical failure
management system is provided by several subfunctions.
Functional relationships exist between these subfunctions. In
summary, the system consists of the following subfunctions:
•   Communication and Integration: The integration of
the shop floor and production IT-systems enables the
failure management system to capture and store factory
information. Thus, the virtual representation of the
factory can always be kept up to date.
•   Virtual Sensors: With virtual sensors several
information sources can be combined to a virtual sensor.
This allows e.g. production orders or other information
from IT systems to be represented as virtual sensors. On
the other side, real sensor values can be combined or
further processed. Thus, the structure of the real factory
can be represented in more detail, which improves the
later analysis.
•   Information Pool: The information pool stores all
information and enables to create a virtual representation
of a factory as well as its historical information. It is the
basis for the analysis and allows to store data of failures
and to link the failures with corresponding factory
information.
•   Failure Management: Provides management
functionality for failures over the entire life cycle of a
failure and handles all information related to failures.
•   Analysis: This functional unit analyses the stored
information and attempts to identify new failures and
correlations to already occurred failures. The analysis
uses the virtual representation of the factory. This
includes consumption profiles and operating states of
Database
Integration
Service
Search
Engine
Fileserver
REST Websocket
Enterprise Resource Planing,
Maintenance workers, Manufacturing Execution
machine operators System, ...
Fig. 2. System architecture of the cyber-physical failure management system.
303
machines. By combining it with other information
sources such as planned and currently produced orders,
the normal state of a factory can be determined and
deviations such as failures can be detected. Additional
information such as wear profiles or maintenance
histories allows to determine a probability for the
localisation of a detected failure.
•   User Interface: Provides the interface for the user to the
system.
Further important elements of the concept are the provision
of failure relevant context information and the feedback of
failure related information into the failure management system.
If a failure is detected, the maintenance worker can be
supported in diagnosis and repair by providing failure relevant
context information. Context information are e.g. the probable
location of the failure, possible causes, required spare parts,
operating manuals or experiences with similar types of failures.
Moreover, experience values are important for maintenance
issues. In many cases, failure patterns of wear parts occur again
and again in (in)regular time intervals, or different machines
have similar failure patterns because they are based on the same
components. Through information such as the possible cause
of a failure or suggested solutions, the maintenance worker can
be better supported in the event of new failures with similar
failure patterns. Thus, the time required for eliminating the
failure and finally the downtime of the machine can be reduced.
4.  Implementation
The functional units described in the concept are
implemented prototypically in Java by a service-oriented
architecture. The system architecture is shown in Fig. 2. Each
service provides a RESTful API to access the provided
Failure
Sensor Detected
Analysis Analysis Management
data failures
Service 1 Service ... Service
Failure Management System Service
(Service Composition)
Integration Middleware
MQTT OPC UA ...
CPS
PLC
Machine Machine Sensors and actuators
Sensors and actuators
304 Matthias Schneider et al. / Procedia CIRP 81 (2019) 300–305
Matthias Schneider et al. / Procedia CIRP 00 (2019) 000–000 5

functions of the service. The service-oriented approach enables engine which is highly scalable and provides additional search
a high degree of modularity and flexibility. Thus, the system and analysis functionalities. Contextual information such as
can be easily extended by new services or existing services can operation instructions, manuals or repair instructions are often
be replaced with existing functions of external maintenance available as digital documents. Usually, these documents are
management systems. The cyber-physical failure management accessible anyway trough a file server of the company and
system consists of the following services: therefore they don’t have to be stored redundantly in the
database. Only a corresponding link to the document on the file
•   Integration Service: The Integration Service collects
server is stored in the database. The prototype implementation
information from CPS and production IT-systems. At the
uses Nextcloud, an open source file share platform, as file
same time, it enables the creation of virtual sensors. Virtual
server.
sensors can be used to aggregate information or generate
For the integration of PLCs, other IT-systems or smart
new sensor values through mathematical operations. For
sensors the Manufacturing Services Bus (MSB) [15] is used as
virtual sensors, new sensor values are automatically
an integration middleware. The MSB supports various
calculated as soon as corresponding information is
communication protocols such as REST, Websocket, MQTT
received. New information are collected via an integration
and OPC UA as well as manufacturer-specific PLC
middleware. The middleware supports several
communication protocols. In addition, the MSB, offers
communication protocols such as REST, Websocket, OPC
adapters for Enterprise Resource Planning Systems and
UA or MQTT. This enables a low effort integration of shop
Manufacturing Execution Systems, enabling integration of
floor components (e.g. PLCs) and their associated sensors
production IT systems as further information sources.
and actuators as well as the integration of production IT
The user interface enables the maintenance workers to
systems.
interact with the maintenance management. The maintenance
•   Analysis Service: The Integration Service notifies the
workers are notified about new failures and get provided with
Analysis Service about updates in the virtual representation
information about the failures. Once a failure has been
of the factory. This makes it possible to react to changed
eliminated, the maintenance workers can return information
sensor values or machine states, so that the failure analysis
(e.g. about failure causes and possible failure solutions) into the
is always based on the current factory state. On this basis,
failure management system via the user interface. An
an attempt is made to automatically detect and locate
application on a mobile device is best suited to support
failures in the factory. The localization accuracy depends
maintenance workers in their daily work. The mobile device
strongly on the available information. The failure detection
itself is a CPS whose sensors can be used to capture context
approach described in the concept was implemented
information for manual failure reports. For the feedback of
prototypically in the form of a graph-based analysis
information into the system, it is important that the information
algorithm. Therefore, the virtual representation of the
is in a structured form so that it can be further processed
factory structure is modelled as a graph. Each node of the
automatically and considered for further analysis. To achieve
graph is linked to the respective information from the
this, the maintenance worker is supported in the mobile
information pool. When the virtual representation changes,
application by a guided feedback process. Common failure
the affected factory part can be identified directly as a sub-
information (e.g. cause of failure, suggestions for failure
graph. A detected failure can be limited directly to such a
solution or required spare parts) is pre-classified and only must
sub-graph. Based on this sub-graph the probability of
be selected from a catalog.
localisation can be determined. The analysis capabilities of
the system will be extended by further analysis services in
5.  Evaluation
future work.
•   Failure Management Service: The Failure Management
Compressed air is an important and at the same time one of
Service takes care of the management of existing failures.
the most expensive energy sources in factories, so the detection
A failure is managed over the entire life cycle, from its
and elimination of leakages offers great potential for savings.
detection until its elimination. In this way, information
Therefore, the failure management system was prototypically
(e.g. for repair of the failure and its possible causes) can be
implemented and evaluated for the detection of failures and
fed back into the system and the information base for
leakages in compressed air networks.
analysis can be continuously extended. The Analysis
The system was tested on a model machine that simulates a
Service automatically reports detected failures and
real compressed air network in an industrial production
provides context information. The reporting of failures is
environment. A simple production process is simulated by this
done via the RESTful API of the service. This can also be
model machine. The model machine uses industry standard
used to report failures manually by maintenance workers
compressed air sensors and PLCs with integrated OPC UA
or machine operators.
servers. Leakages and machine failures can be generated
manually. The structure of the model machine is modelled in
The large number of CPS in smart factories, the huge
the failure management system as a production system in a
amount of information as well as the wide variety of
factory. Via OPC UA the information of the PLCs is captured,
information sources sets high requirements for the database
so that a virtual representation containing current sensor values
management system (DBMS) in terms of scalability and
and operating states can be created. The model of the plant
analysis capabilities. Therefore, Elasticsearch is used as
includes consumption profiles for different operating states.
DMBS. Elasticsearch is a document-based NoSQL search
 Matthias Schneider et al. / Procedia CIRP 81 (2019) 300–305
6 Matthias Schneider et al. / Procedia CIRP 00 (2019) 000–000
Based on the virtual representation failures in the form of
leakages or operating interrupts are detected automatically
trough deviations to the expected state. Further context
information (e.g. age of components, wear profiles or historical
failure data) can be used to determine the probability of
localization for the detected failure.
Furthermore, the repairing process was tested with the
mobile application on an Android device. The main users of
this application are maintenance workers. Once a failure has
been detected, a corresponding notification is displayed on the
mobile devices of the responsible maintenance workers. Using
the implemented app, a maintenance worker can retrieve
information about the detected failure as well as context
information for the repair. After repairing the maintenance
worker feeds information (e.g. about the failure cause, needed
spare parts) back to the failure management system.
The evaluation has shown some potential for improvements
of the failure management system which should be considered
in future work:
•   Modelling the factory structure is very time-consuming.
This is currently done textually and can be significantly
improved by providing graphical modeling tools.
•   For the analysis corresponding models of the machines are
needed, which describe the normal state of the machine
(e.g. consumption profiles for power or compressed air).
The generation of the models should be further automated
and created automatically through a teach-in phase.
•   In complex factories, analysis is confronted with large
amounts of information. In order to manage these amounts
of information, the analysis can be extended with machine
learning methods such as neuronal networks. This means
that even large amounts of information can be processed,
and the quality of failure detection and localization can be
continuously improved by training an appropriate model.
•   In order to further support the maintenance workers in
troubleshooting, a recommendation system based on the
feedback of the maintenance workers can be introduced. In
this way, different solutions for failures can be collected
and evaluated. In the long term, the best solution strategy
can be determined and suggested automatically.
•   The model used for the virtual representation of the factory
is currently specific for the use case of failure detection
and management. It is planned to be extended by further
standards (e.g. Industrie 4.0 Administration Shell) to
enable use cases in other fields such as process
optimization, quality or planning.
6.  Conclusion
In this paper, we present an approach how the potentials of
CPS in smart factories can be used for failure management. For
this purpose, an information model that models the structure of
factories has been designed. By the numerous information
sources, which both the CPS and production IT systems offer,
a virtual representation of factories can be generated. This
forms the basis to detect deviations and to determine possible
failed components in factories.
Based on a service-oriented architecture the concept was
prototypically implemented. The evaluation with a model
305
machine operated with compressed air showed that it is
possible to automatically detect failures in the form of leakages
or operating interruptions during operation of machines and to
support maintenance workers with failure relevant information.
The evaluation showed that the modelling of factory structures
should be supported by graphical tools. The maintenance
worker can also be better supported by a suggestion-based
feedback system when troubleshooting failures. Furthermore,
the evaluation will be carried out in a larger production
scenario. In this context additional failure detection methods
(e.g. machine learning) for the analysis of large amounts of
information will be considered.
Acknowledgement
This research was done within the research project Center
for Cyber-Physical Systems (S-TEC ZCPS) at Fraunhofer
Institute for Manufacturing and Automation Engineering
funded by the Ministry of Economic Affairs, Labour and
Housing Baden-Württemberg.
References
[1] Kagermann H, Wahlster W, Helbig J. Recommendations for implementing
the strategic initiative INDUSTRIE 4.0. Final report of the Industrie 4.0
Working Group: acatech; 2013.
[2] Geisberger E, Broy M. agendaCPS. Berlin, Heidelberg: Springer Berlin
Heidelberg; 2012.
[3] Radziwon A, Bilberg A, Bogers M, Madsen ES. The Smart Factory.
Exploring Adaptive and Flexible Manufacturing Solutions. Procedia
Engineering 2014;69. p. 1184–1190.
[4] Shrouf F, Ordieres J, Miragliotta G. Smart factories in Industry 4.0. A
review of the concept and of energy management approached in
production based on the Internet of Things paradigm. International
Journal of Modern Manufacturing Technologies 2015;VII-1. p. 697-701.
[5] Wang S, Wan J, Di Li, Zhang C. Implementing Smart Factory of Industrie
4.0. An Outlook. International Journal of Distributed Sensor Networks
2016;12-1.
[6] Lucke D, Adolf T, Le T, Bérenguer C, Christien J, Sieg-Zieba S, Haug R.
Integrated dynamic predictive maintenance planning with advanced
deterioration and remaining useful lifetime estimation models. In:
Podofillini L, Sudret B, Stojadinovic B, Zio E, Kröger W, editors. Safety
and Reliability of Complex Engineered Systems: CRC Press; 2015. p.
2261–2269.
[7] acatech, editor. Smart Maintenance für Smart Factories. Mit intelligenter
Instandhaltung die Industrie 4.0 vorantreiben. München: Herbert Utz
Verlag; 2015.
[8] Zimmermann F, Köse E, Reichart M, Weckmann S, Sauer A. Industrie. In:
Sauer A, Bauernhansl T, editors. Energieeffizienz in Deutschland - eine
Metastudie, Berlin: Springer Vieweg; 2016. p. 77–172.
[9] Sauer A, Bauernhansl T, editors. Energieeffizienz in Deutschland - eine
Metastudie. Analyse und Empfehlungen. 2nd ed. Berlin: Springer
Vieweg; 2016.
[10] Fischäder H. Störungsmanagement in netzwerkförmigen
Produktionssystemen. In: Schneider H, Haupt R, editors. Schriften zum
Produktionsmanagement, Wiesbaden: Deutscher Universitäts-Verlag;
2007.
[11] zendesk. 2018. [Online]. Available: https://www.zendesk.com. [Accessed
05.12.2018].
[12] freshdesk. 2018. [Online]. Available: https://freshdesk.com. [Accessed
05.12.2018].
[13] Vincent J, Spier R, Rolsky D, Chamberlain D, Foley R. RT Essentials.
Sebastopol: O'Reilly; 2007.
[14] Huang HH, Wang B. Related work on machine monitoring and
diagnostics. In: Lee J, Wang B, editors. Computer-aided Maintenance,
Boston, MA: Springer US; 1999. p. 41–58.
[15] Schel D, Henkel C, Stock D, Meyer O, Rauhöft G, Einberger P, Stöhr M,
Daxer MA, Seidelmann J. Manufacturing Service Bus. An
Implementation. Procedia CIRP 2018;67. p. 179–184.