Proceedings of the 20th World Congress

Proceedings
The of
of the
International
Proceedings 20th
20th World
Federation
the Worldof Congress
Automatic Control
Congress
Proceedings
The of the 20th World Congress Control
The International
Toulouse, France,Federation
International Federation of
of Automatic
July 9-14, 2017 Available
Automatic online at www.sciencedirect.com
Control
The International Federation
Toulouse, of Automatic Control
Toulouse, France,
France, July
July 9-14,
9-14, 2017
2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 9014–9019
Simulation Model of a Quality Control
Simulation
SimulationinModel
Model of
of a
a Quality Control
Control 
QualityIndustry
Laboratory Pharmaceutical
Laboratory in Pharmaceutical
Laboratory in Pharmaceutical Industry
Industry 
∗,∗∗ ∗
Andrea Costigliola ∗,∗∗ Filipe A. P. Ataı́de ∗
Andrea
SusanaCostigliola
Andrea M. Vieira ∗,∗∗
Costigliola ∗∗ Filipe
Filipe
∗,∗∗João M.A.
A. P.
P. Ataı́de
Ataı́de ∗∗ ∗
Andrea
Susana Costigliola
M. Vieira ∗∗ Filipe
∗∗ João M. A.C.
C. P.Sousa
Ataı́de
Sousa ∗∗
∗∗
∗
Susana M. Vieira ∗∗ João M. C. Sousa ∗∗
Susana M. Vieira João M. C. Sousa
∗
∗
Hovione Farmaciência S.A., Lisboa, Portugal (e-mail:
∗ Hovione Farmaciência S.A., Lisboa, Portugal (e-mail:
∗ Hovione Farmaciência
∗∗ Hovione Farmaciência S.A.,
acostigliola@hovione.com,
acostigliola@hovione.com,
Lisboa,
Lisboa, Portugal
Portugal (e-mail:
S.A., fataide@hovione.com)
fataide@hovione.com) (e-mail:
IDMEC, acostigliola@hovione.com,
Instituto Superior
acostigliola@hovione.com, fataide@hovione.com)
Técnico, Universidade
fataide@hovione.com) de Lisboa,
∗∗
∗∗ IDMEC,
∗∗ IDMEC, PortugalInstituto
Instituto
(e-mail:Superior
Superior Técnico, Universidade
Técnico, Universidade de
susana.vieira@tecnico.ulisboa.pt, de Lisboa,
Lisboa,
IDMEC, PortugalInstituto
(e-mail:Superior Técnico, Universidade
susana.vieira@tecnico.ulisboa.pt, de Lisboa,
Portugal (e-mail:
Portugal (e-mail: susana.vieira@tecnico.ulisboa.pt,
jmsousa@tecnico.ulisboa.pt).
susana.vieira@tecnico.ulisboa.pt,
jmsousa@tecnico.ulisboa.pt).
jmsousa@tecnico.ulisboa.pt).
jmsousa@tecnico.ulisboa.pt).
Abstract: Laboratories are critical components in drug manufacturing, and inefficiencies in
Abstract:
Abstract: Laboratories
Laboratories are
are critical
critical components
components in
in drug manufacturing, and inefficiencies in
laboratory management
Abstract:
laboratory Laboratoriescan
management
can have
are have a major
critical impact on
components
a major
major impact in drug
on drug manufacturing,
the overall supply chain
manufacturing,
the model
overall supply chain
and inefficiencies
service
andservice level. The
inefficiencies
level. TheThe
in
in
laboratory
aim of
laboratory this management
paper
management is to can
build
can ahave
have a
Discrete
a majorEvent impact on
Simulation
impact on the
the overall
overall supply
of a
supplyQualitychain
chain service
Control
service level.
laboratory.
level. The
aim
aim of this
of this paper is to
to build a Discrete
Discrete Event Simulation
Simulation model treatment
of aa Quality
Quality Control laboratory.
To of this paper
To achieve
aim achieve this
is
is to build
this objective,
paper objective, build a
a
a generic
a Discrete
generic
Event
framework
Event Simulation
framework for
model
for information of
model treatment
information of a Quality and
and
Control
Control laboratory.
organization
laboratory.
organization
was
was
To
built.
To achieve
In
achieve this
particular, objective,
information
this objective, a generic
a generic framework
coming from
framework for information
different databases
for information treatment
was
treatment and
organized organization
into
and organizationa single was
one
was
built.
built.
that In
In
was particular,
particular,
used as information
information
input to a coming
coming
discrete from
from
event different
different
simulation databases
databases
model. Thewas
was organized
organized
proposed into
into
model aa represents
single
single one
one
built.
that In particular,
was usedwork information
as input
input to coming
aa discrete
discrete fromsimulation
event different databases
model. Thewasproposed
organizedmodel into a represents
single one
that
that was
in detail
was used
the
used as
as flow to
input of a
to a discrete event
quality control
event simulation
laboratorymodel.
simulation and itThe
model. The proposed
is intended
proposed asmodel represents
a support
model representstool
in
in
for detail
detail
planning, the scheduling
the work flow
work flow and of a
of a decision
quality control
quality control
making. laboratory
laboratory
The model and
and
was itvalidated
it is intended
is intended using asreal
as a support
a support
data, andtool
toolit
in
for detail
planning, the scheduling
work flow and of a decision
quality control
making. laboratory
The model and
was itvalidated
is intended using asreal
a support
data, andtoolit
for
proved
for planning,
to
planning, be scheduling
effective
scheduling in and
the decision
estimation
andestimation making.
decision making.of The
performance model
The model was
parameters validated
was validated such using
as, system
using real data,
throughput,
real throughput, and
data, and it it
proved
proved
equipment to
to be
be effective
effective
usage rate,in
in the
the
system estimation of
of
responsiveness performance
performance
and tasks parameters
parameters
processing such
such as,
as,
times. system
system throughput,
Furthermore, the
proved
equipment to be effective
usage was in the
rate,tested
system estimation of
responsiveness performance
and parameters
tasks processing
processing such as, system
times. how throughput,
Furthermore, the
equipment
simulation model
equipment usage
usage rate,
rate, system
system withresponsiveness
an alternative and
responsiveness and tasks
scheduling
tasks policy to evaluate
processing times.
times. Furthermore,
modifications
Furthermore, the
the
simulation
simulation model
model was
was tested
tested with
with an
an alternative scheduling policy to evaluate how modifications
on the system
simulation
on the systemmodel may
may wasimprove
tested its
improve
its
with an alternative
alternative scheduling
performance.
performance. scheduling policy
policy to to evaluate
evaluate how how modifications
modifications
on
on the
the system
system may
may improve
improve its
its performance.
performance.
© 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Computer Simulation, Discrete Event Systems, Modeling, Quality Control,
Keywords:
Keywords: Computer
Scheduling, Computer
Pharmaceutical Simulation,
Simulation,Industry Discrete
Discrete EventEvent Systems,
Systems, Modeling,
Modeling, Quality Quality Control,
Control,
Keywords:
Scheduling, Computer
Pharmaceutical Simulation,Industry Discrete Event Systems, Modeling, Quality Control,
Scheduling, Pharmaceutical
Scheduling, Pharmaceutical Industry Industry
1. INTRODUCTION ity. In this scenario, quality control laboratories play an
1.
1. INTRODUCTION
INTRODUCTION ity.
ity. In
In thisrole
scenario,
in thequality control laboratories
process.play an
1. INTRODUCTION In this
important
ity.
important
scenario,
thisrole
scenario,
in the
quality
quality
drug
control
control laboratories
drug manufacturing
manufacturing laboratories
process.
play
Labo-
play
Labo-
an
an
In the last decades, drug manufacturing evolved driven by ratory important
important role
management
role in
in the
the is drug
a
drug manufacturing
complex task
manufacturing which process.
involves
process. Labo-
Labo-re-
In
In the
the last
external last decades,
decades,forces,
economic drug
drug manufacturing
manufacturing
patents expiration evolved
and driven
evolved driven
increased by ratory
by ratory
sources management
management
(personnel and is aaequipment)
is complex task
complex task whichand
which
planning involves
involves
schedul-re-
re-
In the last decades, drug manufacturing evolved driven by ratory
sources management
(personnel is aequipment)
and complex task whichand
planning involves
schedul-re-
external
external
competition.economic
economic
In order forces,
forces, patents expiration
patents expiration
to maintain expiration
their competitiveand increased
and increased
increased sources
ing, (personnel
analysis and
prioritization, equipment)
results
advan- sources (personnel and equipment) planning and schedul- planning
evaluation and andschedul-
docu-
external
competition.economic
In forces, patents and ing, analysis prioritization, results evaluation and docu-
competition. In order
tage, pharmaceutical
competition. In
to
to maintain
order companies
order to
their
their competitive
maintainorganized
maintain their competitive
themselves
competitive
advan-
advan-
advan-
ing,
into mentation.
ing, analysis
analysis prioritization,
prioritization, results results evaluation
evaluation and and docu-
docu-
tage,
tage,
complexpharmaceutical
pharmaceutical
organizations companies
companies
(supply organized
organized
chain and themselves
themselves
contract into
into
manu- mentation.
mentation.
tage, pharmaceutical companies organized themselves into mentation.
Quality control laboratories are critical components in
complex
facturing organizations
complex organizations
networks), and (supply
startedchain
(supply to beand
chain and
more contract
contract
concerned manu-
manu- on Quality control laboratories are critical components in
complex
facturing organizations
networks), and (supply
started chain
to be and
more contract
concerned manu- on Quality
drug
Quality control
manufacturing laboratories
control laboratories and are
inefficienciescritical
are critical in components
quality
components controlin
in
facturing
achieving
facturing networks),
operational
networks), and
and started
excellence
started to
to be
through
be more
morethe concerned
optimization
concerned on
on drug
drug
may manufacturing
manufacturing
delay obtaining and
and
results,inefficiencies
inefficiencies
affect in
in
negatively quality
quality
their control
control
quality
achieving
achieving
of the operational
operational
processes involvedexcellence
excellence
in drug through
through the
the
manufacturing: optimization
optimization
chemi- drug
may manufacturing
delay obtaining and
results,inefficiencies in quality control
achieving operational excellence through the optimization may can
and delay obtaining
have a major impactaffect
results, affect
on thenegatively
negatively their
their quality
their quality
overall supply chain
of
calthe
of
of the
the
processes
processes
processes
processes
involved
involvedsupply
modeling,
involved
in
in drug
in drug
drug of manufacturing:
manufacturing:
raw materials, logistic
manufacturing: chemi- may
chemi-
chemi- and
and
service
delay
can
can
obtaining
have
have
level. aa major
The major
results,
impact
impact
situation
affect
can on
on
negatively
the
the
be overall
overall
magnified supply
supply
in
quality
the chain
chain
case
cal
cal
cal
processes
processes
operations,
processes
modeling,
modeling,
quality
modeling,control,supply
supply
supplyetc. ofof raw
of raw materials,
raw materials, logistic and
materials, logistic
logistic service
service
canlevel.
have The
level.
a major
The
impact
situation
situation can
can
onbethemagnified
be
overall supply
magnified in
in the
the
chain
case
operations,
operations, quality
quality control,
control, etc.
etc.
of
of
a
service
a
contract
level. The
contract
manufacturing
situation can
manufacturing
organization,
be magnified
organization,
that
that the case
produces
in produces
case
operations,
Pharmaceutical quality control, etc.
companies operate in one of the most goods of
of aa contract
under the
contract manufacturing
brand of its organization,
manufacturing clients and therefore
organization, that
that produces
has to
produces
Pharmaceutical
competitive and companies operate in
in one of
of the
the most goodswith
goods under
under the brand
the brand of ofits
of its clients
clients and and materials.
therefore has
therefore has to
Pharmaceutical
Pharmaceutical companies
regulated markets,
companies operate and
operate in one
compliance
one of the with deal
most
most goods
deal under
with
a large
a the brand
large
number
number of ofitsprojects
clients and
projects and therefore has to
materials. to
competitive
competitive
Good and
and
Manufacturing regulated
regulated
Practicesmarkets,
markets,(GMP) and
and compliance
compliance
and Good with
with
Labo- deal
deal with
with a
a large
large number
number of
of projects
projects and
and materials.
materials.
competitive
Good and regulated markets, and and compliance with Given the high mix of products and tests it is impor-
ratory Manufacturing
Good
Good Manufacturing
Practices (GLP)
Manufacturing
Practices
Practices
is mandatory
Practices
(GMP)
(GMP)for
(GMP) and
and
Good
Good Labo-
marketing
Good Labo-
any tant
Labo- Giventothe
Given the high mix
high mix of products
of products
strategiesandand tests
fortests it is
it is impor-
ratory
ratory
drug. Practices
Practices
Those (GLP)
(GLP)
regulations is
is mandatory
mandatory
were introduced for
for marketing
marketing
to ensure any
any
that Given
tant tothedevelop
high mix
develop
effective
of products
effective strategies and for tests it is impor-
laboratory
laboratory
man-
impor-
man-
ratory Practices (GLP) is mandatory for marketing any tant
agement.to develop
Laboratory effective strategies
information for laboratory
management man-
systems
drug.
every Those
drug.
drug. Those
Those
regulations
regulations
pharmaceutical
regulations
were
were introduced
product
were introduced
meets safety
introduced
to
to
ensure
to and
ensure
ensure that tant
that
quality
that agement.
agement.
(LIMS)
to develop
Laboratory
Laboratory
used in
effective strategiesmanagement
information
information
pharmaceutical
for laboratory
management
industry often lack
man-
systems
systems
of fea-
every
every pharmaceutical
pharmaceutical
requirements in a systematicproduct
product meets
meets
fashion. safety
safety and
and quality
quality agement.
(LIMS) Laboratory
used in information
pharmaceutical management
industry often lack systems
of fea-
every pharmaceutical
requirements product meets safety and quality (LIMS)
tures
(LIMS) used
essential in
used in(i.e. pharmaceutical
(i.e. information
pharmaceutical industry
on often
processing
industry lack
times,
oftentimes, of fea-
work
lack ofwork
fea-
requirements in
requirements in a
in a systematic
a systematic fashion.
systematic fashion.
fashion. tures
tures
flow) essential
essential
for planning, (i.e. information
information
scheduling on
on
and processing
processing
stock times,
management. work
During its development life cycle, a drug must be con- tures essential (i.e. information on processing times, work
During its
its development life
life cycle, aa drug must be con-
con- flow)flow) for
for planning, scheduling and stock management.
During
During its development
stantly monitored
developmentwith laboratory
life cycle,tests
cycle, drug
a drug must its
to assess
must be qual-
be con- flow) for planning,
As pointed planning, scheduling and
out byscheduling
Juran and
stock
stock management.
andGodfrey management.
(1999), quality
stantly
stantly monitored
monitored with
with laboratory
laboratory tests
tests to
to assess
assess its
its qual-
qual- As pointed out by Juran and Godfrey (1999), quality
stantly monitored with laboratory
 This work is funded by Portuguese Funds through the tests to assess its qual- As
As pointed
control can
pointed out
be
out by
seen
by Juran
as
Juran a and
recursive
and Godfrey
process
Godfrey (1999),
composed
(1999), quality
by
quality
 control
control
three can
can
main be
be
tasks: seen
seen as
as
quality a
a recursive
recursive
planning, process
process
control composed
composed
and improve- by
by
 This
FCT
 This - work is
is funded
Foundation
work funded for byby Portuguese
Science Funds
Funds through
and Technology
Portuguese throughunder the
the control
three can tasks:
main be seen as a planning,
quality recursive process control composed
and improve- by
FCT
the
This - work
FCT project
is
Foundation
- Foundation
funded for by Portuguese
Science
(UID/EMS/50022/2013)
for Science and and and Funds
Technology
Technology
through
iDecision4Care
the
under
under
three
ment.
three main
Thus,
main tasks:
quality
tasks: quality
control
quality planning,
is not
planning, onlycontrol
control and
responsible
and improve-
for
improve-the
FCT
the - Foundation
project for Science and and
(UID/EMS/50022/2013)
(IF/00833/2014/CP1238/CT0002). S. M. and
Technology
Vieira iDecision4Care
under
acknowledges
ment.
ment.
executionThus,
Thus, quality
of quality control
qualitycontrol tests but is not
is not
alsoonly
only responsible
forresponsible
the development for the
for the
the
the
project
project
(UID/EMS/50022/2013)
(UID/EMS/50022/2013) and
iDecision4Care
iDecision4Care ment.
executionThus, of quality
quality control
tests is not
but alsoonlyfor responsible
the for the
development
(IF/00833/2014/CP1238/CT0002).
the support by Program Investigador
(IF/00833/2014/CP1238/CT0002). S.
S. FCTM. Vieira
M. (IF/00833/2014)acknowledges
Vieira acknowledges from execution
and the
execution of quality
improvement
of quality tests tests
of but
methods also
but also to for
be the
used
forused development
in the control
the indevelopment
(IF/00833/2014/CP1238/CT0002). S. FCTM. (IF/00833/2014)
Vieira acknowledges and the improvement of methods to be the control
the
the support
FCT, by
by Program
co-funded
support Investigador
by the European
Program Investigador SocialFCT Fund (ESF) throughfrom
(IF/00833/2014) the
from and
and the
process.
the improvement
Dedicated optimization
improvement of
of methods
methods to
to be
methods
be used
for in
used in the
the control
planning and
control
the
FCT, support by
co-funded Program
by the Investigador
European SocialFCT Fund(IF/00833/2014)
(ESF) through from
the process. Dedicated optimization methods for planning and
Operational
FCT, co-funded
FCT, co-funded
Program
by the Human
European
by the Human
European
Potential
Social Fund
Social Fund
(POPH).(ESF)The
(ESF)The
authors
through
through
the
the
process.
scheduling,
process. Dedicated
in
Dedicated addition optimization
with
optimization a methods
generic
methods for
framework planning
for planningfor and
infor-
and
Operational
acknowledge the
Operational Program
supportHuman
Program of Hovione Potential (POPH).
Farmaciência
Potential (POPH). authors
S.A.The authors scheduling,
scheduling, in addition
in addition with aa generic
with generic framework
framework for infor-
for infor-
Operational
acknowledge Program
the support Human
of Hovione Potential (POPH).
Farmaciência
acknowledge the support of Hovione Farmaciência S.A. S.A. The authors scheduling, in addition with a generic framework for infor-
acknowledge the support of Hovione Farmaciência S.A.
Copyright
2405-8963 ©© 2017,
2017 IFAC 9348Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Copyright
Peer review© 2017
©under IFAC
2017 responsibility
IFAC of International Federation of 9348
9348Control.
Automatic
Copyright © 2017 IFAC 9348
10.1016/j.ifacol.2017.08.1582
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Andrea Costigliola et al. / IFAC PapersOnLine 50-1 (2017) 9014–9019 9015
mation management can improve laboratory performances
and resource usage.
1.1 Related Work
In the pharmaceutical industry, laboratory management is
often done based on supervisors’ experience, rather than
with validated data driven models. Literature is short of
examples of generic optimization frameworks applicable to
pharmaceutical quality control laboratory management.
Klaessens et al. (1988) developed a simulation model that
consists of classes of objects (sample, sample generator,
planner, analyst and instruments) and applies concepts of
queuing theory to describe a laboratory organization. The
authors proposed two planning strategies: centralized (the
workload of a day is planned by the expert planner) and
decentralized (the analysts determine which samples are
analyzed).
Schäfer (2004) proposes a set of terms and definitions
that can be used to model all components interacting
on the workbench (i.e. samples, instruments, sensors, re-
sults, information systems) and provides a description
of a schematic scheduling process applicable in quality
control laboratories. His approach comprises the following
steps: process description, information treatment, work-
plan generation, scheduling generation, execution, instru-
mental control and data storage.
Maslaton (2012) presents a two steps process towards the
optimization of a quality control laboratory. The first step
is related with resource planning, while the latter refers
to the scheduling of the laboratory. The author proposes
to organize information on products and raw materials
into groups, based on products similarities, define types
and number of tests for each group and estimate analysis
processing time for each test. Once the correct amount
of resources has been determined the second step of the
optimization process is the scheduling of the analyses.
To assign the best available resource to a sample/test, clas-
sical optimization algorithms can be used effectively. Boyd
and Savory (2001) applied a genetic algorithm for labora-
tory personnel scheduling. Their program maximizes the
value of a fitness function that measures how well a given
work shift scheduling of analysts and their skills matches
a set of work tasks. Similarly, Dudnikov et al. (2012)
developed a laboratory analysis planning system to sched-
ule analyses on the available resources. Their algorithm
considers two types of analyses (periodic and unplanned ),
and creates a 24 hours scheduling considering only periodic
samples that is rearranged in case of unplanned samples.
1.2 Objectives
The objective of this work is to build a generic framework
for modeling and simulating a quality control laboratory
of a pharmaceutical company. In particular, the modeling
stage involves the study of the system and the design
of an integrated database that contains all the relevant
information that can be used for planning and scheduling.
Moreover, it uses concepts of Discrete Event System (DES)
and Petri net graphs theory for modeling purposes. See
Cassandras and Lafortune (2009) and Law (2015).
9349
A model of the real system is simulated using the academic
license of Simio simulation software (version 8.132). The
developed model will serve as basis for planning, schedul-
ing and decision making in quality control laboratories.
2. QUALITY CONTROL LABORATORY OVERVIEW
Hovione Farmaciência S.A. is an international company
operating in the contract manufacturing business sector.
Quality control represents an important part of the drug
manufacturing process. With thousands of samples to
be analyzed every year, laboratories management can be
difficult and inefficiencies at quality control level may
affect negatively the overall supply chain service level.
Among several quality control laboratories, this study
focuses on a particular one, that is located in the Por-
tuguese facility of Loures (Lisbon). It is furnished with 70
equipment of 6 different types and employs 20 analysts
working on 3 work shifts. The six classes of equipment are
related with particular analysis techniques: High Perfor-
mance Liquid Chromatography (HPLC), Gas Chromatog-
raphy (GC), Particle Size Analysis (PSA), Karl Fischer
Titration, Differential Scanning Calorimetry (DSC) and
X-Ray Powder Diffraction (XRPD).
Typical tasks in pharmaceutical quality control laborato-
ries are analytical methods development and validation,
tests on final products, raw materials, in-process control
and stability analyses. Analytical methods must be de-
signed, developed and validated with the goal to rapidly
test a specific compound.
2.1 Laboratory Staff and Shifts
The quality control lab considered in this study is open
24 hours a day, 7 days a week. A reason for this is the
need to have always analysts available to support the
manufacturing process with in-process control tests. Other
types of tests (i.e. stability, development, validation) are
performed 5 days a week, from Monday to Friday. A total
of twenty analysts have been considered: six analysts work
on 12 hours rotating shifts, four analysts on two different
8 hours shifts, and ten analysts work 5 days a week on a
fix 8 hours shift. See table 1.
Table 1. Analysts Work Shifts.
Work Days a Hours Analysts
Shift week shifts on Shift
Shift 1 7
8:00-20:00
2
20:00-8:00
8:00-17:00
Shift 2 5 2
17:00-24:00
Shift 3 5 8:00-17:00 10
2.2 Analysis Work flow
Every day a high number of samples enters the laboratory
to be analyzed. According to the analytical technique
needed to analyze a particular sample, it flows to the
appropriate group of machines and it is scheduled to one of
the available machines. Independently from the analytical
technique used, an analysis starts with the preparation
of solutions and materials, and with the setup of the
Proceedings of the 20th IFAC World Congress
9016
Toulouse, France, July 9-14, 2017 Andrea Costigliola et al. / IFAC PapersOnLine 50-1 (2017) 9014–9019

equipment. Next two steps involve the preparation of to gather information on tasks processing times and to
the sample and the verification of the equipment (system divide equipment “hands-on” and “hands-off” tasks. With
suitability). This step is needed to check if the equipment this purpose, it was asked to analysts to collect data
is suitable to run a specific test. Finally, the sample is during a period of one month using a form model. All this
analyzed and analysis data are processed by an analyst information was organized into a unique database, whose
(see figure 1). UML representation is shown in figure 2.
Work Bench
Sample Entity LIMS database Analytical Method
PK Analysis PK Sample Number PK Analytical Method id
System Sample
Sample
Preparation Preparation FK Sample Number Project Run Time
Analyst Analyst
Release Date FK Analytical Method Number System
Solutions
Due Date FK Instrument
Equipment Number Analysis
Location Solutions

Equipment System Identifier Number System

Analysis Injections
Setup Suitability Received date
Analyst Number Analysis
Completed date Injections

Analyst
Instruments
Data
Processing PK Instrument

Analyst Equipment Type

Fig. 1. Main tasks of a quality control test. Fig. 2. UML representation of the designed database.

3. QC LABORATORY MODEL DESIGN 3.2 Input Analysis

The goal of this study is to evaluate the performances of The database structure presented allows to easily access
the quality control laboratory and to propose useful tools to relevant information that can be used for modeling. In
for planning and scheduling. The proposed methodology a simulation model it is important to define probability
considers analysts and equipment as processing units (or distributions for input in order to simulate the random-
resources). Jobs and activities are defined by the sample ness existing in a real system. With this purpose, LIMS
and by the particular analytical method needed to test database offered good quality information to estimate
it. Thus, samples can be considered as the input of the samples arrival rate, and number of analyses for each
system. In the simulation software paradigm a sample can analytical technique. LIMS provides an arrival table where
be defined as an entity that flows through the system each entry represents the couple analysis/sample and can
and need to perform a number of activities at particular be considered as input to the system. In table 2 it is shown
working stations. the percentage of analyses per analytical technique.
3.1 Data Processing Table 2. Percentage of analyses per technique.
Technique HPLC GC PSA KF DSC XRPD
In the section above, the main components of the system Percentage 40.7% 25.7% 18.8% 7.1% 4.0% 3.7%
have been identified: analysis work flow, entities (samples)
and processing units (analysts and equipment). The next Many of the processing times in the model are stochastic
step is to estimate entities arrival rate, process logics and in nature, while others are deterministic. Equipment run
processing times. With this purpose, a study of the system time and number of preparations have been retrieved from
and of the available data was performed. Information was analytical methods reports, and thus are assumed to be
distributed on different databases, and in some cases it deterministic. Processing times for tasks performed by the
was stored in document repositories in unstructured form. analysts (i.e. equipment setup, sample preparation, data
Thus, it was necessary to extract and organize the relevant processing) were estimated fitting probability distributions
information coming from these data sources. to data gathered with the time study. Processing times
Four different data sources have been considered. The first distribution used in simulation are summarized in table 3.
one is the Laboratory Information Management System Table 3. Processing times distributions. Time
(LIMS) that contains information on all the analyses per- unit is minutes.
formed, such as number of samples processed, analytical
Equip. Sample Equip. Run Time Data
methods used, equipment used, release date and completed Prep. Setup Process
date. However, LIMS has no information about processing HPLC T (5, 40, 100) T (25, 50, 65) Det. T (2, 5, 10)
times and process logics. Another source of information GC T (5, 10, 35) T (2, 5, 30) Det. T (10, 20, 45)
is the document mangament system (DMS) that contains PSA T (5, 7.5, 21) U (1, 32) T (5, 10, 30) T (2, 6, 10)
reports with the description of all the analytical methods. KF T (1, 6, 25) T (4, 8, 16) T (2, 5, 56) T (10, 20, 30)
Given the unstructured form of the data, an information DSC T (5, 10, 30) T (1, 3, 12) U (12, 180) T (4, 9, 16)
extraction algorithm (IE) was implemented to analyze XRPD T (5, 10, 20) T (1, 5, 20) T (11, 25, 203) T (2, 7, 30)
text and extract information on: analytical method name,
number of samples to be prepared, system suitability and Given the empirical nature of the collected data, three
analysis run time. Furthermore, a time study was designed types of probability distribution are used: deterministic

9350
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Andrea Costigliola et al. / IFAC PapersOnLine 50-1 (2017) 9014–9019 9017
(Det.), triangular pdf T (min, mode, max) and uniform pdf
U (min, max).
3.3 Discrete Event Simulation Model
The simulation model was developed under the following
assumptions:
(1) The couple sample/analysis has been considered as a
single entity. Every sample entity is processed using
only one machine;
(2) The system suitability has a validity of 24 hours. The
system suitability does not need to be re-executed
within 24 hours, unless a new sample using a different
analytical method is scheduled to the same machine;
(3) Analysts can perform every type of work regardless
the work shift they are allocated to. In the real system
every work shift is allocated only to few operation
types (i.e. in-process control, stability, development).
The quality control simulation model was built using Simio
simulation software, and it is composed by four types of
objects: samples, machines, analysts and work location
nodes.
Samples have been modeled as input entities that travel
through the system to perform certain tasks. Each sample
has properties that define the type of equipment needed
to process it, and additional information about run time
and number of preparations.
A machine can be seen as a series of processes needed
to perform an analysis (system preparation, system suit-
ability, sample preparation, analysis and data processing).
These tasks have to be performed at different physical
location (bench, machine or computer location). The spa-
tial component has been added to those processes as an
attribute (work location node) that specifies the physical
place in the laboratory where an analyst has to go to
perform the requested task.
An analyst has been modeled as a secondary resource of
the system, and his/her presence is necessary to perform
specific tasks. As referred, analysts work in shifts and their
availability in the laboratory varies according to the work
shifts defined in table 1.
Generic Equipment Model A generic machine model has
been implemented to simulate the analysis process. This
model is valid for every analytical technique regardless
of the equipment used. Thus, the analysis process was
modeled using the concepts of discrete event systems
(DES) theory and it was represented using timed Petri
nets formalism.
A timed Petri net is a bipartite graph with two types of
nodes (places and transitions) and a set of weighted arcs
that links places to transitions and transitions to places.
Places contains information about the state of a Petri
net. This state is expressed by the marking function that
assigns a number of tokens to each place of the Petri net. A
transition is usually associated with an event that can be
asynchronous or time dependent. An event can “happen”
only if the relative transition is enabled. See Cassandras
and Lafortune (2009).
9351
The timed Petri net used to design the generic equipment
model (figure 3) is characterized by a set of places P , that
represents the five analysis tasks described above, more
some auxiliary places needed to represent queues and to
avoid impossible system dynamics (i.e. perform the system
suitability and run the analysis at same time). When a
sample enters the system it waits in a queue and it is
scheduled according to a policy that tends to minimize the
maximum lateness (calculated as the difference between
the estimated completion time and the due date). This
scheduling policy is known as maximum lateness first ).
See Pinedo (2015).
Input_buffer
t1: new_sample
System_preparation
v2: preparation_time
System_wait
Analysis_preparation
t3: system_start
System_suitability
2
v4: max(system_time, analysis_prep_time)
Analysis_run
2
v5: analysis_time
n_Analysis_Done
System_free Data_processing v6: data_processig_time
Fig. 3. Discrete Event System model of a generic equip-
ment using timed Petri net.
Quality Control laboratory Model The described model
has been implemented using Simio simulation software. It
was created a dedicated library containing the following
objects:
• Equipment: implementation of the generic equip-
ment model presented above;
• HPLCs: liquid chromatography machines group;
• GCs: gas chromatography machines group;
• KFs: Karl Fischer titration machines group.
These objects have been used to build a complete model of
the quality control laboratory system. As referred, when
a sample is received in the laboratory it flows through the
system towards the group of machines that can perform
the test required. The scheduling is made within the
equipment group and the sample is allocated to one of the
available machines. To replicate the behavior of the real
system and to evaluate its performances, every analysis
has been scheduled to the same equipment that was used
to run the test in the real laboratory within the considered
time period. This choice allowed an easier validation of the
developed model.
Proceedings of the 20th IFAC World Congress
9018
Toulouse, France, July 9-14, 2017 Andrea Costigliola et al. / IFAC PapersOnLine 50-1 (2017) 9014–9019
To check the correctness of the developed model and to
visualize the scheduling, a 3D graphical model of the
laboratory has been implemented and it is shown in figure
4.
Fig. 4. Quality Control laboratory model: Simio 3D view.
4. SIMULATION RESULTS
A simulation study aims to measure the performance
of a real system and to evaluate how modifications on
the system can affect its performance. In this study, the
following aspects were considered:
• System throughput. Number of analyses per-
formed over the total number of analyses required;
• Equipment usage rate. It is defined by the equip-
ment operating time (EOT) over the total available
time (TAT): U sage% = EOTT AT ;
• System responsiveness. Defined as the average
time needed to process an analysis with the system
in a steady state. Also referred to as sample time in
system;
• Tasks processing times. Estimate of run time for
each analysis task.
The discrete event simulation model was validated using a
one year period data (referring to year 2015). Statistical in-
dependent replications of the simulation model, using the
same input parameters, were run to retrieve information
on the behavior of the model.
4.1 Model Validation and Performances
In order to verify the validity and accuracy of the devel-
oped simulation model, two metrics where used: system
throughput and equipment usage rate.
Considering the system throughput defined above, the
simulation model was able to perform all the analyses
that where input to the system with no delay (see table
4). From this point of view, the model was effective to
Table 4. System throughput.
Equip.
Number of analyses performed
Accuracy
Real data Model
HPLC 5968 (of 6099) 6079 (of 6099)
GC 3899 (of 3910) 3910 (of 3910)
PSA 3010 (of 3014) 3014 (of 3014)
KF 1223 (of 1240) 1240 (of 1240)
DSC 571 (of 581) 581 (of 581)
XRPD 471 (of 479) 479 (of 479)
replicate the history. However, this metric is not conclusive
since it only takes into account the total number of
samples processed without considering the time needed
to process them. To assess the validity of the developed
simulation model, we considered the equipment usage rate
that could be easily retrieved from the dedicated software
of each equipment. Thus, the usage rate obtained with
the simulation model has been compared with real data.
In figure 5, it is shown a comparison of the usage rate of
the most representative HPLC machines, between real and
simulated data.
Simulation Model
Real Data
Usage Rate %
HPLC3 HPLC4 HPLC5 HPLC8 HPLC18 HPLC20 HPLC22
Fig. 5. HPLC usage rate(%) comparison between simula-
tion and real data.
In average, the difference between real data and simulated
data is within 1.5%. Similar values have been obtained
considering other types of machines.
In a quality control laboratory, efficiency depends not only
on the quality of the results but also on the time they are
delivered. An important metric to evaluate the respon-
siveness of the system is the amount of time a sample
spends in the laboratory before a result is generated. The
average time in system for a sample is shown in table 5.
This values take into account processing times and waiting
times. The standard deviation is calculated among differ-
ent simulation runs, and it does not represent variability
in samples time in system. Given the lack of data, it is
Table 5. Average time in system for each
sample type.
Sample Type Average Time in System [h]
HPLC 31 ± 1
GC 10.0 ± 0.8
PSA 4.0 ± 0.5
KF 3.0 ± 0.5
DSC 5.0 ± 0.8
XRPD 6.0 ± 0.6
difficult to validate this result. Also, this result does not
take into account the difference nature of a sample (i.e.
in-process control, validation, stability, development). For
instance, the average time in system for in-process control
samples is 2 hours, while a sample with low priority (i.e.
raw material) can be processed even after a week. This
result may be representative considering a sample with
98.2% medium priority and the system in a steady state.
99.8%
99.9% Processing times were input to the simulation model, and
98.7% as a recall they were estimated considering deterministic
98.3% and stochastic times. Table 6 shows average processing
98.4% times obtained running the simulation model.
As expected, these results agree with the input distri-
butions and with the analytical method reports. Sys-
tem Preparation and Analysis Preparation relate with the
9352
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Andrea Costigliola et al. / IFAC PapersOnLine 50-1 (2017) 9014–9019 9019

Table 6. Simulation model average processing 5. CONCLUSIONS

times. Time unite is minutes.
The quality control laboratory work flow can be described
Equip. System System Analysis Analysis Data
Prep. Suit. Prep. Proc.
as a discrete event system that changes its state according
HPLC 207 ± 14 227 ± 11 96 ± 3 85 ± 1 59 ± 2 to samples arrivals and timed operations defined on sam-
GC 67 ± 5 152 ± 5 48 ± 1 62 ± 1 39 ± 1 ples. Timed Petri net resulted in a powerful modeling tool
PSA n.a. n.a. 31 ± 1 32 ± 1 16 ± 1 that allowed an easy implementation of the QC laboratory
KF n.a. n.a. 45 ± 3 30 ± 2 29 ± 2 model using a simulation software with 3D modeling capa-
DSC n.a. n.a. 47 ± 2 101 ± 3 24 ± 2 bilities. In order to make easy modifications to the model
XRPD n.a. n.a. 46 ± 1 88 ± 3 30 ± 1 and to easily access to relevant information (i.e. processing
times, arrival rates, etc.), input data was organized into a
Sample Preparation entry in table 3. In particular, the time single database structure.
needed to prepare the system or the analysis is obtained
Simulations of the developed model showed that it is
multiplying the Sample Preparation time of table 3 and
appropriate to mimic the real system and it can be used
the number of samples/solutions needed to run the test.
to estimate performance parameters and as a tool for
planning, scheduling and decision making. Furthermore,
4.2 Simulation with alternative scheduling policy simulation shows that modification of the scheduling pol-
icy may improve system performance and balance the
The simulation model allows to understand how modifi- loading of the machines. However, employees represent an
cations on the real system may affect its performances. important part of the work flow and scheduling effective-
With this purpose, it was tested a modified version of ness depends heavily upon the number of analysts present
the model using a dedicated heuristic scheduling policy in the laboratory. On the other hand, in the simulated
that takes into account two factors: equipment loading and scenarios the number of equipment was appropriate to run
last analytical method used. Preferably, a sample will be all the analyses even considering a 50% of work overload.
allocated to equipment that are already suitable to run The developed simulation model can be applied to a
the required analytical method. However, if an equipment wide range of laboratory organizations and classes of
suitable to run the right analytical method has already instruments. It can be used to simulate different scenarios
a high number of samples in queue (≥5), it will prefer a and as a support tool for decision making (i.e. planning
machine with a shorter queue. and scheduling). The proposed case study highlighted the
With this configuration, two scenarios were tested. In the importance of the estimation of the correct number of
first scenario, the number of samples input to the system resources (machines and equipment) and the effectiveness
is the same of the considered time period. In scenario 2, of dedicated heuristics for analyses scheduling. Suggestions
the number of samples is increased of 50%. In table 7 it to improve the efficiency of the simulated lab are the
is shown the comparison between the two models (Real development of dedicated algorithms for planning and
Model schedules analyses according to real data, while scheduling that rely on the database structure proposed
Modified Model uses the new scheduling policy). in this work.

REFERENCES
Table 7. Average time in system for each
sample type: comparison between models. Boyd, J.C. and Savory, J. (2001). Genetic algorithm for
scheduling of laboratory personnel. Clinical Chemistry,
Average time in System [h] 47(1), 118–123.
Sample Scenario 1 Scenario 2
Cassandras, C.G. and Lafortune, S. (2009). Introduction
Type Real M. Modified M. Real M. Modified M.
to discrete event systems. Springer Science & Business
HPLC 31±1 15±1 57±1 69±8
GC 10.0±0.9 7.0±0.9 29±1 52±10 Media.
PSA 4±1 4.0±0.1 10±1 152±42 Dudnikov, E.E., Sidorenko, A., and Khobotov, E.N.
KF 3.0±0.2 3.0±0.2 3.7±1 18±2 (2012). Laboratory analysis planning system for in-
DSC 5.0±0.2 7.0±0.2 10.0±0.7 237±42 dustrial enterprises. Automation and Remote Control,
XRPD 6±1 8.0±0.3 12±2 312±41 73(11), 1910–1915.
Juran, J.M. and Godfrey, A.B. (1999). Quality handbook.
In the first scenario, the modified model is able to perform McGraw-Hill.
HPLC samples two times faster than the real model. Klaessens, J., Saris, T., Vandeginste, B., and Kateman, G.
Improvement in system responsiveness can be observed (1988). Expert system for knowledge-based modelling of
also on Gas Chromatography samples. For other classes of analytical laboratories as a tool for laboratory manage-
equipment the two models behave in a similar manner. The ment. Journal of Chemometrics, 2(1), 49–65.
improvement observed is given by the fact that the new Law, M.A. (2015). Simulation modeling and analysis.
scheduling policy tries to take advantage of more machines McGraw-Hill, 5th edition.
avoiding long queues. In the second scenario, the modified Maslaton, R. (2012). Resource Scheduling in QC Labora-
model behaves worse than the real one. A reason for this is tories. Pharmaceutical Engineering, 32(6), 1–6.
that there are not enough analysts in the system to process Pinedo, M. (2015). Scheduling. Springer, 5th edition.
all the samples allocated to more machines. Many samples Schäfer, R. (2004). Concepts for dynamic scheduling in the
suffer from starvation and can wait a long time before an laboratory. Journal of the Association for Laboratory
analyst is available to process them. Automation, 9(6), 382–397.

9353