International System Dynamics Conference 2017
Justifying maintenance studying system behavior: a multi-
purpose approach using multi-objective optimization
Gary Linnéusson!, Amos H.C. Ng! and Tehseen Aslam!
‘School of Engineering Science, University of Skévde
gary .linneusson@ his.se
Abstract
Industrial maintenance includes rich internal dynamic complexity on how to deliver value. While
the technical development has provided with applicable solutions in terms of reliability and
condition based monitoring, managing maintenance is still an act of balancing, trying to please the
short-termism from the economic requirements and simultaneously address the necessity of strategic
and long-term thinking. By presenting an analysis to justify maintenance studying system behavior,
this paper exemplifies the contribution of the combined approach of a system dynamics maintenance
performance model and multi-objective optimization. The paper reveals how insights from the
investigation, of the near optimal Pareto-front solutions in the objective space, can be drawn using
visualization of performance of selected parameters. According to our analysis, there is no return
back to the single use of system dynamics; the contribution to the analysis of exploring system
behavior, from applying multi-objective optimization, is extensive. However, for the practical
application, the combined approach is not a replacement - but a complement. Where the
interpretation of the visualized Pareto-fronts strongly benefits from the understanding of the model
dynamics, in which important nonlinearities and delays can be revealed, and thus facilitate on the
selected strategical path for implementation.
Keywords: maintenance performance, strategic development, system dynamics, simulation, multi-
objective optimization
Introduction
Justifying maintenance is not straightforward, if it was, any company would have full control over
tradeoffs between money spent in their maintenance organization and their effect on production
throughput or service to its customers. Any market, where your products or services compete, there is
an upper level for what customers are ready to pay. Having the consequence that for the specific
department there is normally a budget limiting the ambition for maintenance development to support
production with required dependability. It is of interest, from a practical stand point, to better understand
the underlying structures in maintenance, resulting in its system behavior, and to identify the best trade-
off between conflicting objectives, in order to attain strategic development of the maintenance
1
International System Dynamics Conference 2017
performance. Furthermore, as a consequence from increased competition, the improvement potential
becomes harder to gain putting higher demands on future methods for justifying maintenance.
Therefore, this paper presents one approach to economically justify maintenance, focusing on the study
of system behavior, by the combined utilization of a system dynamics (SD) maintenance performance
model and the simulation-based optimization (SBO) approach of multi-objective optimization (MOO).
Where the application of MOO leads to the thorough investigation of the trade-offs between the
conflicting objectives of, for instance, the short-term economic requirements, (Sherwin, 2000), and the
long-term development needs (Repenning and Sterman, 2001), and thus support the maneuver in
systems with different short- and long-run dynamics, addressed in (Rahmandad and Repenning, 2015).
Except a few studies, including ours that investigated the integration of MOO and system dynamics
(SD) models (Aslam, 2013, Duggan, 2008), the use of SBO with SD-models is in general much less
reported. As a matter of fact, the work of Aslam (2013) has exemplified applying MOO on SD-models,
whereas one is the well-known beer game model of (Sterman, 2000) which has shown possible to draw
generalized conclusions through studying the resulting patterns from the extensive amount of different
optimal simulation runs (Aslam, 2013); thus MOO can support the identification of innovative principles
that make up certain patterns of the non-dominated optimal solutions from the SD-model under study.
More concretely, for the model applied in this study, it provides a method for the thorough analysis of
the trade-offs between conflicting objectives, such as availability, maintenance costs, and maintenance
consequential costs.
The applied research work represented by this paper, and the choices of methodologies, has several
purposes:
e Firstly, address the practical problem in automotive industry of attaining sustainability in the
strategic development of maintenance performance. This is the fundamental research
motivation. To support a sustainable development Linnéusson et al. (2015a) calls for a systems
thinking approach to better address maintenance cost modeling; which should include the
visualization of consequential maintenance costs; with the purpose to minimize short-term my-
budget-thinking and support the long-term development of maintenance performance.
e Secondly, by applying the systems thinking approach, the ambition is to introduce thinking
differently, and more holistically, to defeat chronic reactiveness and to contribute to the shift in
mind on the added value from maintenance, brought up in (Linnéusson et al., 2015b), where the
need to build a maintenance SD-model was elaborated on. Because, utterly, what is needed and
sought to support, is to transcend current paradigm (Donella, 1999) of short-termism within the
maintenance context. Even if such endeavor may be considered too ambitious, working in that
direction is considered fundamental in this research.
International System Dynamics Conference 2017
e Thirdly, visualize system behavior, applications investigating maintenance performance, in for
instance (Linnéusson et al., 2017b), have applied SD to analyze such system behavior and
expose possible paths towards proactiveness. The model includes the interaction of maintenance
in production, studying maintenance performance, based on the efficiency of applied pallet of
maintenance methodologies (Tsang, 2002), such as: run-to-failure, preventive maintenance
using fixed intervals, condition-based maintenance using inspections, and condition-based
maintenance using sensors, and the load on equipment in production including feedback to
equipment degradation, inspired by (Ledet and Paich, 1994, Sterman, 2000), and its
corresponding effect to the mean delay of breakdowns. It included cost consequences from
model behavior which explicitly visualize consequential maintenance costs (V orster and De La
Garza, 1990). Furthermore, continuous development based on breakdowns, similar to the
Reliability centered maintenance (RCM) concept was also included.
e Fourthly, increase knowledge elicitation from SD-models, the application using the combined
approach of SD and MOO (Duggan, 2008, Aslam, 2013) enables extensive evaluation of the
decision- and objective space, and their visualization. It has enabled meta-analyses comparing
several scenario’s Pareto-fronts to distinguish characteristics based on starting point in the
proactive maintenance work (Linnéusson et al., 2017a). The many SD-model evaluations in a
MOO study also lead to the merciless verdict on attained internal validation. The reward from
its application is vast information of the pattems between parameters with respect to the
optimization objectives, see for example, the parallel coordinate heat maps in Figure 7 and
Figure 8.
e Fifthly, support the practical improvement of precision in maintenance’ activities towards
proactiveness and higher efficiency. By the application of above mentioned methods,
contributing to the improved evaluation of strategic development, this purpose is supported and
can generate policies on the general level of maintenance performance development.
Hence, the purpose with this study mainly focuses on the fourth point above, however, with the
predecessor points as basis, and, with the aim to deliver value to the fifth purpose.
The application of MOO enables this paper to explore the different objective space caracteristics for
how two categories of equipment at one production unit, representing equipment with low and high
criticallity, may be most benefitially developed, with respect to the underlying SD-model. The
outcome of the investigation is thus a visualization of the Pareto-front trade-offs between the
investigated conflicting objectives and a set of model parameters, supporting the analysis of system
behavior for the decsion-maker.
International System Dynamics Conference 2017
Multi-objective optimization
Multi-objective optimization (MOO) is a discipline that has been studied since the 1970s. Its application
areas range widely from resource allocation, transportation, and investment decisions to mechanical
engineering, chemical engineering, and automation applications, to name a few. In contrast to single-
objective optimization, in which only one objective function is considered, MOO considers multiple
objective functions simultaneously and seeks to identify a set of optimal solutions which are defined as
Pareto-optimal solutions. A solution is considered to belong to the Pareto-optimal set when there is no
other solution that can improve at least one of the optimization objectives without deteriorating any
other objective. This set of solutions is also known as the Pareto-front when plotted on the objective
space. Figure 1 illustrates the concept of decision and objective space, as well as the domination and
non-domination of solutions in MOO. The search space of a multi-objective optimization problem is
represented by the decision space where the design variables, which are the input parameters, constitute
a set of solutions that are evaluated through a solver, which in this work is mainly a simulation model,
and mapped to the objective space. Thus, a certain solution A with its inherent values of the design
parameters x, and x, is evaluated through the solver which subsequently results in A’ in the objective
space representing the fitness or performance of solution A in terms of the objective functions f, and f2.
Decision space Solver Objective space
min f,
Parameters Bigs : Objectives min fz
(x4, £2.) vminated solunons fi. fa~hi
© Noo.
sine aie
Figure 1. Concept of Non-domination, Decision and Objective Space, from (Aslam, 2013).
The main concept of MOO is to evaluate two or more conflicting objectives against each other and
obtain the Pareto-optimal solutions and the Pareto-front (Basseur et al., 2006). This comparison of the
solutions is executed on the basis of the domination concept in which a solution s,is said to dominate a
solution s, if sis no worse then sz, with respect to all optimization objectives, and where s, is strictly
better than s2 in at least one optimization objective (Deb, 2001).
Applying MOO with SD
The simulation principle for evaluating SD models using MOO follows the general simulation-based
optimizing process as is seen in Figure 2. For this study the “Simulation Model”, which for the
optimization model is regarded as a “Black Box”, has utilized the SD-model included in appendix. The
4
International System Dynamics Conference 2017
SD-model was built in Vensim DSS, and the MOO-simulation model in modeFrontier. The MOO-
simulation model utilizes the NSGA-II algorithm, and the evaluation process activates and executes
multiple runs of the SD-model. Using a double quad-core processor enables eight simultaneous
evaluations, which implies 1.5-2 hours for about 50,000 evaluations.
Optimization
Exogenous 5
Components squint)
Output values Input parameter
Endogenous Boundaries
Components
a a Multi-Objective A
System Simulation Model Tnput/output parameter : Optimal
Environment (Black Box) | “ mapping > | Optimization Model I> cortions Sets
(Optimization Engine)
a re EE
Input values
Relationships Optimization
Objectives
Figure 2. A general simulation-based optimizing process, adapted from (Aslam, 2013).
The applied procedure for the MOO-SD analysis presented in this paper has followed according to
Figure 3. Where Step 1 is the ordinary modelling procedure for the SD-model, which in this case used
a previously developed model presented in (Linnéusson et al., 2017b). Step 2 includes setting up the
optimization model in the optimization engine based on the selected conflicting objectives in the SD-
model, and defines the amount of evaluations. Step 3 includes evaluating the initial results, where
strange results may indicate on the need of SD-model modifications in order to get reasonable output
values. This process harshly exposes any inability to generate a valid answer for all evaluations; a
process following iterations of model improvements to provide a more stable and valid SD-model. Step
4 can be performed when the SD-model can be considered valid enough for its purpose, and provides
with possibility to analyze the results, according to the scatter plots and parallel coordinates presented
later. Step 5 may be applied if the analysis benefits from investigating different points of origin, where
scenarios with different initial conditions may be explored, with purpose to learn from how important
knowledge about present condition before conducting a implementation journey towards a future state,
as examined in (Linnéusson et al., 2017a). Step 6 represents the possible post-analysis of solutions of
interest, however not explored in this paper, where the explored Pareto-front solutions may be further
analyzed utilizing the SD-model again in order to apply the strengths of SD to facilitate the desired
development. Furthermore, in order to conduct the MOO-simulation, in Step 4, the methodology for
SD-MOO presented in (Aslam, 2013) has been used, which in detail describes the steps of decision
space sampling, global objective space search, and local objective space refinement, which leads to the
presentation of optimal solutions.
International System Dynamics Conference 2017
1. Develop the SD-model for the case, problem boundary and validation aspects, according to the
standard procedures (Barlas, 1996, Sterman, 2000)
2. Define the MOO-model, such as, the input and the ing objecti |
3. Test run MOO-model, if needed improve SD-model to enable valid MOO-evaluations, according to
below, and initiate search of Pareto-front
Evaluations MOO-model | mp
Evaluate results in
t respect to
validation
Improve model
= Visualize Pareto-
4. Evaluate results from MOO-scenario, explore Pareto-front solutions
front solutions
| 5. Meta-analysis, compare different MOO-scenarios | > Visuals ae
solutions
6. Select Pareto-front solutions of interest, from MOO-scenarios, for further investigation in the SD-model
to study their dynamic behaviors over time, to support decision-making for the specific strategy
Figure 3. The general procedure for a MOO-SD analysis.
The maintenance performance model
For a full model presentation, see (Linnéusson et al., 2017b), in appendix the complete structure with
corresponding model equations is provided. Figure 4, represents an overview of the model, illustrated
using five general parts.
Production and
Maintenance
Performance
Fi
@
Operation load and Maintenance
risk for failure Development
4) Prows [>—_\ :
vaca iim Holistic Economic
an ae Performance
maintenance in Applied Maintenance Strategies
Preventive Maintenance
Le __| Performance
Equipment Health
Status
Figure 4. Overview of the maintenance performance SD-model.
The structure in Production and maintenance performance part defines the availability as a consequence
of the current equipment reliability, defined in the structure found in the Equipment health status box,
6
International System Dynamics Conference 2017
together with staffing for unplanned or planned maintenance repairs and their respective productivity,
similar to structure in (Ledet and Paich, 1994). Thus, the better equipment health status is, the breakdown
frequency decreases, and availability increases, however, the higher availability is it also leads to a
higher operational load on equipment which implies higher risk for a failure. The structure in Equipment
health status part defines the aggregated equipment reliability as a consequence of the accumulated
defects, generated by the operation load and collateral damage from breakdowns, and their elimination
through repairs, inspired by the structure of Equipment defects presented in (Sterman, 2000). Based on
the level of Preventive maintenance performance, and the ratio between planned and unplanned repairs,
it results in the certain defect elimination, where planned maintenance has the more efficient approach
to defect elimination. The planned maintenance, is based on the level of applied maintenance
methodology, divided between preventive maintenance using fixed interval, and condition-based
maintenance using manual inspections, or sensors; which in tum results in different efficiency to detect
defects based on which of these three methodologies that are applied. And also, includes the planning
and scheduling capabilities, similar to structure in (Ledet and Paich, 1994), together with a throttle
limited by the pressure to produce on behalf of preventive maintenance, if availability is under its goal
value. The model also includes a structure for a Maintenance development process which defines the
maintenance performance development pace based on policies, resources, delays, work pressure, and
work progress of transforming information of why breakdowns occurred into root-cause
countermeasures, represented in the model by new preventive maintenance activities. The structure
describing the Holistic economic performance box includes, for example, the calculation of total
maintenance costs as a consequence of the production and maintenance performance, including direct
e costs, and cc ial mai e costs from breakdowns, using a simple principle
found in (Wireman, 2004) where the maintenance costs and downtime costs ratio have been empirically
considered on the range from 1:2 to 1:14. The Applied Maintenance Strategies diamond in Figure 4
represent where possible policies and strategies for development interact with the model for this study.
Validation has considered the normal techniques in SD, such as the process according to (Barlas, 1996),
with direct structure tests, structure-oriented behavior tests, and behavior pattem tests. Inputs to
modelling have covered the studies of procedures of the industrial partners and relevant literature. Thus,
the overall model behavior has, to some extent, been considered justified, also including the testing of
assumptions with help of industrial maintenance experts. Furthermore, the application of MOO, in
respect to model validation, is very powerful. Any error in the model will be identified by the evaluation
of so many solutions, thus MOO identifies any weak spots leading to anomalies. In this study, it has had
the effect of improving model equations in order to correct erroneous behavior, adding parameters, as
well as, some new structure.
International System Dynamics Conference 2017
MOO simulation scenarios
The maintenance performance SD-model can be applied for different studies using the MOO-
technique, for instance, comparing different categories of companies at three different states of applied
maintenance methodology, as is presented in (Linnéusson et al., 2017a). This study, however, is an
example where MOO is applied to investigate applicable strategy for two sets of equipment at one
production unit, with different characteristics regarding downtime costs. In a structured maintenance
organization equipment is normally divided into different categories of criticality, where the
consequences from a breakdown in respect to downtime, quality, safety, cost, etc., have been analyzed,
placing equipment into its category of criticallity. This categorization is then used as input value to the
preparation of the maintenance planning for the certain piece of equipment, considering activities such
as: preventive maintenance using fixed intervals (PMfi), condition based maintenance using
inspections (CB Mi), and condition based maintenance using sensors (CBMs). Thus the output should
be a set of maintenance activities that will prevent failure to the required level of the specific category
of criticality. In the real setting there may be several categories of equipment, not just two as in this
example, as provided due to space limitation. Furthermore, in the presented case, for simplicity, in
respect to model comparison only one parameter is chaged between the two scenarios, which is the
repair cost ratio for a planned and unplanned repair. The downtime cost, varies depending on the
consequences in production, if the stop causes quality issues, or increased damage requiring exchange
of more parts than the one causing the breakdown in the first place, thus it is represented by the
criticality of the certain pice of equipment. Therefore, the scenarios for this study are accordingly:
e Scenario S1 includes the equipment of lower criticality, with a cost ratio of 1:4 between the repair cost
for a planned and an unplanned repair.
e Scenario S2, includes the equipment of higher criticality, with a cost ratio of 1:12 between the repair cost
for a planned and an unplanned repair.
The MOO-scenarios apply the input parameters, and ranges, according to Table 1 below. The input
parameters are selected based on their expected effects to attain a proactive behavior in maintenance
in the SD-model, using a time horizon of 10 years. The same initial conditions are applied, using a
Run-To-Failure (RTF) strategy for 50% of the equipment, and the other 50% use PMfi.
Table 1. Input parameter data for the MOO-scenarios
Input 2 Range: | Step:
numberRepairMen 4-50 1
numberMai: i 0- 1
fractionPMiFromRCA O-1 0.05
fractionCBMiFromRCA help 0-1 0.05
goalFractionCBMoverPM 0-1 0.05
i al 4-52 2
goalCB Msensors 0-500 25
International System Dynamics Conference 2017
Each MOO-scenario evaluates the multi-criteria trade-offs bewteen maximized availability, minimized
e cost, and minimized cosequential mai e costs. Hence, the MOO investigation in
this paper explores how to strategically address maintenance activities at one production unit having
two categories of criticallity.
Results and Analysis
The optimization has, for each scenario, been run for at least 50,000 evaluations, following the
methodology for SD-MOO developed by (Aslam, 2013). As previously described the performed MOO
considers three objectives, which suitably can be displayed using a 3D-scatter plot. However, a 3D-
scatter plot can be hard to interpret using a 2D-paper. Therefore the Figure 5 reveals three perspectives
of the same resulting plots, according to their axes. It means, that the left plot reveals all three
objectives in one view. The second plot reveals the trade-off curve between the two objectives
availability and the consequential maintenance cost; which are clearly different for the two scenarios.
The third plot reveals the trade-off curve between the two objectives availability and the maintenance
cost; which clearly shows that scenario S1 and S2 follow near the same trade-off curve on these
objectives. Looking at the middle coordinate set, the S2 is the curve with higher consequential
maintenance cost, and it exhibits a considerable behavior of lower cost as availability increases.
z Se, i
0.97 ly, 0.97 “xe
& <
PB 0.95, 0.95 ><
a JZ joe 2 catagories
| Seema
Bon on 8 Seam?
RK oe 089 &
j
% ow os &
& :
“Vy anki
Min_maintenanceConsequentialCost (Y Axis)
Figure 5. 3D-scatter plots of the same graph, from left: the 3D xyz-axis perspective, the xy-axis perspective, xz-axis perspective.
If we are interested in the total maintenance cost, a 2D-scatter plot, as in Figure 6, may be easier to
interpret, where the mai e cost and the cc e cost are summarized on x-axis,
and is compared to its trade-off to availability on y-axis. The comparison between the two scenarios,
using Figure 6, indicates that solutions on the higher range of availability reach lower total costs. It also
reveals that the two scenarios clearly distinguishes in performance, and that scenario S2 have much
higher total costs and that they are on a larger spread. It means that the exploration of optimization
results make known that equipment treated in $2, represented by equipment with higher criticality,
clearly have high potential for high availability solutions to a radically lower total maintenance cost than
to those with lower availability. An analysis on this level, may also indicate that the maintenance
International System Dynamics Conference 2017
organization should prioritize on attaining the development suggested by the S2 solutions, and perhaps
wait with the equipment included in the S1-analysis.
g
@ Scenario
‘@ Scenario?
vax Ayatabity
g
&
.
$%
os9
099 if
osr i
-
os
t
‘iboo00—~—«ao0000 —~—~=«OGOD——~—«oOD ‘00000 ~~ Y000000 ~—~—«Y00000 ——~4200000
7700000 ‘900000
‘maintenanceTotalCost
Figure 6. 2D-scatter plot over trade-off objectives.
Parallel coordinate heat map analysis
To better understand the results in the scatter plots we can also present the results utilizing parallel
coordinate heat maps, which visually display the performance of selected variables, as is seen in Figures
Figure 7-Figure 8. The scales are normalized between the scenarios. Any of the parameters in a Vensim
model can be modelled to be an output parameter for analyzing the results, here some parameters of
interest are represented. By comparing S1 and S2 results, in Figure 7 and Figure 8 respectively, common
and distinguishing patterns can he identified, enriching the analysis to include in a future strategy for
maintenance development. For instance, the best performing solutions in both $1 and $2 follows a
similar pattern on all output parameters. However, solutions just below top performers on availability,
on about availability of 0.96 represented by orange lines, we can see diverging patterns where for S1
amount of maintenance engineers are some more, result on MTTF (mean time to failure) is better,
breakdown rate lower, but takedown rate about same as for $2. And, for the last three parameters the
policies for applied maintenance methodologies (PMfi, CBMi, and CBMs) are needed to be selected
differently to attain the optimal solution. Thus, by applying the parallel coordinates it enables generating
further quantitative knowledge of the patterns of behavior, generated by the SD-model, in respect to its
trade-off solutions.
The presented analysis reveals it possible to identify the specific strategy to apply for both equipment
sets, but also considering $1 and S2 equipment together, where the results clearly indicates that
equipment in scenario S2 should be at main focus due to the much larger leverage on cost performance
from improved proactive behavior in the maintenance function. Such an conclusion may seem obvious,
10
Intemational System Dynamics Conference 2017
however the parallel coordinate heat maps support the differentiation of separate runs, each representing
a behavior graph in the SD-model, which may require improvements that may be considered being more
or less easy to accomplish in the implementation.
1Max_Avail. 3 Min_maintConCost SnoRepairmen —_7 inspint QbreakdownRate 11 PMfi
E iz) co Es) Es) a
‘180900
8 MTTF 10 takedownRate 12. CBMi
SnoRepairmen —7 inspint QbreakdownRate 11 PMI
a 9 z:
| 4
m
Max_Availabilty
eacoool 4 : y 091
2 Min_maintCost
6 noMaintEng 8 MTTF 10 takedownRate 12 CBMi
4 maintTotCost
Figure 8. Parallel coordinate heat map over $2.
Figures 7 and 8 visualize the specific solutions represented by the lines through all parameters in the
parallel coordinate plot. It enables fast overview of how the different solutions perform in respect to the
11
International System Dynamics Conference 2017
selected parameters. These plots exhibits the generalized patterns of, for instance, that to attain the higher
availability solutions for both S1 and $2 it requires more repairmen, but remarkably, S2 solutions
presents less repairmen for the top performing solutions and a lower MTTF average. However, it is also
seen that for the top performing solutions, in both $1 and S2, despite higher direct maintenance cost it
may be beneficial to apply more repairmen due to the resulting lower maintenance consequence cost,
likely, as a consequence from a more proactive behavior in the SD-model.
Discussion and R dations for M
According to the results and analysis it is clear that the more critical equipment has the larger financial
potential from improved maintenance performance. It must be understood clearly, that the solutions
presented in Figure 5 to 8 are only represented by those solutions that are the best trade-off between the
three objectives of maximized availability to the minimized maintenance cost, and the minimized
consequence cost from the performed maintenance. It means that the MOO-analysis explores multiple
SD-model solutions, and selects those on the Pareto-front, and exhibits these. The applied SD-model is
a model that considers the balancing of the proactive versus the reactively performed maintenance.
Therefore, any solution in the plots is the optimal trade-off for the given availability performance that
the SD-model possibly can express. This paper focuses on illustrating the contribution of applying MOO
to the underlying SD-model, while the SD-model itself is not so deeply reviewed within this piece of
paper, this can be further read in (Linnéusson et al., 2017b, Linnéusson et al., 2016).
In order to discuss recommendations to a specific maintenance organization more information to the
decision making will be considered. However, as for the contribution of this study it can be pointed out
that, for those equipment considered more critical where consequences of breakdowns are larger, as in
S2, there is a clear benefit with respect to total maintenance cost to prioritize management of these
equipment. And in respect to selecting key performance indicators, that can guide towards the desired
future state for $1 and S2 together, it should also be considered ok to perform on a poorer level on the
equipment included in the S1 scenario. While equipment included in the S2-analysis are ok to spoil with
higher support even if the direct cost benefit analysis may be hard to motivate. At the same time it means
that the results from a study like this can explore the possible path for a strategy for the production line
at hand. This study has not got into the resulting plots from the specific SD-model experiments, where
the time delays until efforts pay back are reviewed. This would be the next step for management, to
select solutions of interest from the S1 and S2 scenarios and explore their specific behaviors in the SD-
model, in order to justify the required time delays until the expected and desired effects are attained.
Such analysis could be used as a discussion base to draw up their specific strategy for the complete line,
and how to specifically treat the equipment in S1 and $2 respectively.
International System Dynamics Conference 2017
Conclusions
Technically this paper presents a multi-objective optimization (MOO) analysis of a system dynamics
(SD) model. Two MOO+SD scenarios are explored. The application area is industrial maintenance,
where there exists short-term and long-term procedures to support production through the delivered
dependability from maintenance performance, here specifically equipment availability.
Studying the optimization results, they provide a rich visual quantification of the near optimal trade-off
solutions between the conflicting objectives of maximizing availability, minimizing maintenance cost,
and minimizing consequential maintenance cost, for two different sets of equipment with different
criticality. A pplying a SD-model of the dynamics between short-term and long-term feedback, it enables
investigating trade-offs that consider the long-term development of maintenance towards a more
proactive behavior. However, the application of MOO to a SD-model adds the dimension of
simultaneously evaluating multiple objectives, and the visual presentation of multiple solutions on the
optimal trade-off between objectives, strongly supporting analysis and the decision making process. As
is given by the presented analysis, where two sets of equipment which differ in criticality, in respect to
the consequential downtime cost from breakdowns, it enables identifying the specific strategy to apply
to the specific equipment set of S1 or S2, but also considering them together, where the results clearly
reveals that equipment in scenario S2 should be at main focus due to the much larger leverage on cost
performance from improved proactive behavior in the maintenance function.
Atleast as for the results presented in this paper, applying MOO to a SD-model provides the conclusion
of that there is no return back to the single use of system dynamics; since the contribution to the analysis
of exploring system behavior, from applying multi-objective optimization, is extensive. However, for
the practical application, the combined approach of MOO+SD should not be a replacement to the SD-
analysis - but should be its complement. Since the interpretation of the visualized Pareto-fronts strongly
benefits from the understanding of the underlying model dynamics, in which important nonlinearities
and delays can be revealed; critical for the facilitation of the selected strategical path forimplementation.
Future work
According to the presented purpose with the research work to support the practical improvement of
precision in maintenance’ activities towards proactiveness and higher efficiency, the application of
MOO+SD contributes to the improved evaluation of strategic development, and can generate policies
on the general level of maintenance performance development. However, the feedback to the practical
implementation perspective, from the higher level strategic development, is also considered key in this
work. Where the combination with the operational level, is considered to benefit from including discrete-
event simulation (DES). Hence, the work reported in this paper represents the foundation into such
stretched analysis, with potential to inquire for the activities that support the investigated path forward.
Combining SD with DES is emerging and has been promoted due to its ability to dramatically increase
13
International System Dynamics Conference 2017
the size of scenario landscape, and exchange of strengths between the two approaches, such as feedback
into DES and details into SD (Sasdad et al., 2014). In other words, future work will investigate a hybrid
approach that considers both the short-term, urgent maintenance tasks planning and improvement of
long-term strategic planning, by combining DES with SD.
Hence, by integrating the above-said approaches together, future work will consider the proposal of an
integrated simulation-based optimization (SBO) framework that can offer the potential to address
industrial maintenance problems that stretch the interface between strategic and operational levels.
Firstly, key leverage processes from the holistic, organizational maintenance behavior perspective can
be identified, using MOO+SD, and presented as input information into a DES model of the production
line, guiding on operational level execution in order to obtain best implementation effects. Secondly,
the connection between strategy and operational level may require that the optimization criteria in a
DES model need to be adapted to the findings on the strategical level obtained in the SD model. Hence,
an overall feedback can be established between the strategic and operational level, contributing to more
precise efforts and empowering maintenance to form its own strategic planning, to a larger extent,
instead of adapting to happenstance. Overall speaking, on a theoretical level, the framework introduces
a methodology for addressing industrial maintenance from a holistic perspective. On a practical level,
the SBO framework can endow maintenance to get in charge of its own optimal planning, instead of
reacting and follow other requirements set by production or poorly defined priorities of activities.
Acknowledgements
This work is partially financed by Knowledge Foundation (KKS), Sweden, through the IPSI Research
School. The authors gratefully acknowledge their provision of the research funding and the supports of
industrial partners, including Volvo Car Corporation and Volvo Group Trucks Operations.
References
ASLAM, T. 2013. Analysis of manufacturing supply chains using system dynamics and multi-objective
optimization. PhD, University of Skévde.
BARLAS, Y. 1996. Formal aspects of model validity and validation in system dynamics. System Dynamics
Review, 12, 183-210.
BASSEUR, M., TALBI, E.-G., NEBRO, A. & ALBA, E. 2006. Metaheuristics for Multiobjective Combinatorial
Optimization Problems: Review and recent issues. INRIA.
DEB, K. 2001. Multi-Objective Optimization using Evolutionary Algorithms, Wiley.
DONELLA, M. 1999. Leverage Points: Places to Intervene in a System. Available:
http://donellameadows.org/wp-content/userfiles/Leverage_Points.pdf [Accessed 20170223].
DUGGAN, J. 2008. Using System Dynamics and Multiple Objective Optimization to Support Policy
Analysis for Complex Systems. /n: QUDRAT-ULLAH, H., SPECTOR, J. M. & DAVIDSEN, P. I. (eds.)
Complex Decision Making: Theory and Practice. Berlin, Heidelberg: Springer Berlin Heidelberg.
LEDET, W. & PAICH, M. The Manufacturing Game. Proceedings from the Goal/QPC Conference 1994
Boston, MA.
LINNEUSSON, G., GALAR, D. & WICKELGREN, M. 2015a. In Need for Better Maintenance Cost Modelling
to Support the Partnership with Manufacturing. In: KUMAR, U., AHMADI, A., VERMA, A. K. &
14
International System Dynamics Conference 2017
VARDE, P. (eds.) Current Trends in Reliability, Availability, Maintainability and Safety: An
Industry Perspective. Switzerland: Springer International Publishing.
LINNEUSSON, G., GALAR, D. & WICKELGREN, M. A path forward: Systems thinking maintenance as part
of shift in mind on added value. In: LAHDELMA, S. & PALOKANGAS, K., eds. International
Conference on Maintenance, Condition Monitoring and Diagnostics, and Maintenance
Performance Measurement and Management, 2015b Oulu, Finland. Pohto.
LINNEUSSON, G., NG, A. H. C. & ASLAM, T. Investigating Maintenance Performance: A Simulation Study.
7th Swedish Production Symposium, 2016 Lund, Sweden.
LINNEUSSON, G., NG, A. H. C. & ASLAM, T. 2017a. Identifying a maintenance strategy to attain
proactiveness: using multi-objective optimisation to investigate a system dynamics
maintenance performance model. in submission process.
LINNEUSSON, G., NG, A. H. C. & ASLAM, T. 2017b. Investigating the strategic development of
maintenance and its effects on production performance and consequential cost using system
dynamics. in submission process.
RAHMANDAD, H. & REPENNING, N. 2015. Capability erosion dynamics. Strategic Management Journal.
REPENNING, N. P. & STERMAN, J. D. 2001. Nobody Ever Gets Credit for Fixing Problems that Never
Happened: Creating and Sustaining Process Improvement. California Management Review, 43,
64-88.
SASDAD, R., MCDONNEL, G., VIANA, J., DESAI, S. M., HARPER, P. & BRAILSFORD, S. 2014. Hybrid
modelling case studies. /n: BRAILSFORD, S. & DANGERFIELD, B. (eds.) Discrete-Event Simulation
and System Dynamics for Management Decision Making. First ed.: John Wiley & Sons.
SHERWIN, D. 2000. A review of overall models for maintenance management. Journal of Quality in
Maintenance Engineering, 6, 138-164.
STERMAN, J. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World, Boston,
Irwin McGraw-Hill.
TSANG, A. H. C. 2002. Strategic dimensions of maintenance management. Journal of Quality in
Maintenance Engineering, 8, 7-39.
WIREMAN, T. 2004. Benchmarking best practices in maintenance management, New York, Industrial
Press.
VORSTER, M. C. & DE LA GARZA, J. M. 1990. Consequential Equipment Costs Associated with Lack of
Availability and Downtime. Journal of Construction Engineering and Management, 116, 656-
669.
Intemational System Dynamics Conference 2017
Appendix Model structure
NOLLONGOUd
NISONVINNOAdd FONVNAINIVIA
40 INSINdOTIANG ADALVYIS
16
International System Dynamics Conference 2017
Appendix Model equations
identified! i = i ions * fi tealthOverPossibleD efects *
quality of inspections ~ equipment/W eek
=M repairs* defect elimi per repair, Equi Jealth /repairDelay) ~
defects/W. eek
initLevelofInspPlans= 0.001 ~ Dmnl
initLibraryofCBMs=0.001 ~ Dmnl
descretionary Jnspections- IF THEN ELSE (ratioLatePlannedW 0>=0. 1, 0, IF THEN ELSE(capacity inspections>=
2*( delay), MIN( tT olnspect / inspection delay, capacity inspections),
SIN aecigmanetsepeet / inspection delay , capacity inspections))) ~ equipment/W eek
start PMwo=IF THEN ELSE(ratioLatePlannedWO>=0.1, 0, PMreplacementBacklog/delay plan PMwo) ~
equipment/W eek
due to ateWO ~ equipment/W eek
breakdownsLateW O= riskLateW O*EquipmentInFullFunctionality/W eek 2 equipment/Week
ratioLatePlannedW O= = Flenned] akedowns/Sumk Mpreparations ~ Dmnl
ity / delay + ateWO ~
equipment/Week
riskLateW O= ratioLatePlannedW O/riskFactorReductionD ueT oPMwork ~ Dmal
fractionInitPMfi= 0.5 oo Dmnl
fractionNormalHealthStatus= 0.7 ~ Dmnl
resourcesScheduledRepairs= 0.2*numberRepairMen ~ people
numberRepairMen= 10 ~ people
resourcesUnscheduledRepairs= 0.8*numberRepairMen ~ people
average CBM interval= avgMaintIntCBMs / (fracti i HealthOverPossibleD efect:
/fractionNormalHealthStatus) ~ Week
defectElimi irs= MIN( repairs* defect ination per repair, i tealth/ repairDelay) ~
defects/W eek
avgMaintIntCB Ms= 104 ~ weeks
repairDelay= 1 ~ Week
initLevelofPMfi= fractionInitPMfi ~~ Dmnl
ithC. INTEG (newCBM. , initial library of CBMs) ~ equipment
initial library of CBMs=initLibraryofCBMs*number of equipment ~ equipment
collateral damage= breakdownRate*probability collateral damage* possible defects per equipment
~ defects/W eek
probability collateral damage= 0.25 ~ Dmnl
wear and tear operations= tbl probability wear and tear Ean ge(EqpmtA ge)* probability wear and tear *
MIN(EquipmentHealth, 9000) ~ defects/W eek
tblContributionMarginOverA vailability( [(0,0)-(1,1)],(0,1),(0.7,1),(1,0.85)) ~ Dmnl
Not used: EqpmtA ge= (initEqpmtA ge*Week+Time)/Week ~ Dmal
max contribution margin per week= 600000 ~ $/Week
Net contribution margin production= A vailability * max contribution margin per week *
tblC ontributionMarginOverA vailability(A vailability) ~ $/Week
Not used: tbl probability wear and tear EqpmtA ge( [(0,0)-(20,4)],(0,1),(10,1)) ~ Dmnl
Not used: initEqpmté ge= 0 ~ weeks
MaintCostOver i: ‘ost/Net margin ~ Dmal
maintenanceC ost=cost man hours + cost breakdowns + cost takedowns +investCBMs ~ $/Week
priceCBMs= 10000 ~ $/equipment
investCBMs= newCBMsensors*priceCBMs ~ $/Week
CBMifaktorRiskReduction= 0.8 e Dmnl
PM work= Equi WithCBMi: ionPlans* CB MifaktorR i i NithCBM:
WithPM PMfaktor ~
goal PM number of equi, goalPMwork ~ equipment
delayOldPMremoval= 26 ~ weeks
International System Dynamics Conference 2017
newPMpreparations= MIN((goal PM M PM ion, PM ion*PM
preparation release) - IF THEN ELSE(goal PM i SumPM i i NithP
delayOldPMremoval, 0) ~ equipment/W eek
goalPMwork=1 ~ Dmnl
CBMs= Equi ithCB M i oe Dmnl
initRatioEquipmentHealth= 0.7737. ~ Dmnl
PMtotal= SumP of ~ Dmnl
CBMi= (Equi! (T olnspect: ithCBM i a Dmnal
initial value of Hidden defects= ini i Tealth* possible defects per equi ~
defects/equipment
PMfaktorRiskReduction= 0.5 we Dmnl
inspection delay=2 ~ Week
PMfi= (Equi NithP ‘PM: M i ~ Dmal
decisionD elayRoleReactiveT oProactive= 4 ~ weeks
roleToProactive=IF THEN ELSE(resUnsRep<1, 0, IF THEN ELSE(resUnsRep>4:A ND :usage reactive staff<0.75, 1, IF
THEN ELSE(resUnsRep<4:A ND: usage reactive staff <0.5, 1, 0))) / decisionDelayRoleReactiveToProactive ~
people/W eek
roleT: IF THEN ELSE( ‘1, 0, IF THEN ELSE( 4:A ND:usage pI ive staff<0.75, 1, IF
THEN ELSE( :A ND: usage pl five staff<0.5, 1, 0))) / decisionDelayRole
~ people/W eek
max Capacity repairs= resL Di frequency on capacity
~ equipment/Week
max capacity schedued repairs= resSchRep* productivity scheduled repairs ~ equipment/Week
resUnsRep= INTEG (roleToReacti eT oProactive, resourcesL i ~ people
decisionDelayRole= 12 ~ weeks
INTEG (roleToP leToReactive, i ~ people
fractionCBMiFromRCA help= 0.45 ~ Dmnl
fractionCBMsFromRCA = 1-fractionCBMiFromRCA-fractionPMiFromRCA ~ Dmnl
delaytime breakdown report= 1 ~ weeks
max capacity implement CBM inspections= (max capacity PM fion) * ity PM to CBM
~ info/Week
max capacity implementing CBM sensors= (max capacity i CBM i fions-CBM ion) *
CBM tosensor =~ info/Week
usage engineers= IF THEN ELSE(max capacity implementing CBM sensors=0, 1, ZIDZ( CBMsensorPreparation, max
capacity implementing CBM sensors)) ~ Dmnl
quality of inspections= 1 ~ Dmnl
cost cost per stop’ ~ $/Week
effect breakdown frequency on capacity= tbl breakdown frequency and stop effect( wnRate/ normal rate)
~ Dmnl
U i INTEG repairs, 0.378*number of equipment) se
equipment
MTTF= Equi i ~ Week
i ionality=INTEG i pail
0.622*number of equipment) ~ equipment
pressure to produce= MIN(MAX (1, goal availability/A vailability) , 4) sad Dmnl
consequential breakdown costs= 12*cost per stop * breakdownRate ~ $/Week
a 4:1 in scenario S1, and 12:1 in scenario S2
planned repairs= NithP i ~ equipment/W eek
PMworkOrder= start PMwo ~ equipment/W eek
maintenance budget= 100000 ~ $/Week
cost man hours= man hour cost per Week*sumStaff ~ $/Week
RCA measureT MIN( nalysisRCAWIP/delay RCA , max capacity RCA) ~
info/Week
diffCostOverB udget= maintenance budget -maintenanceCost a $/Week
Avai ty i i of ~ Dmnl
International System Dynamics Conference 2017
delay plan PMwo= time to plan PMwo /(MIN(fractionPMwork , 0.5)*2) a Week
time to plan PMwo= 2: ~ Week
ti Defective equi delay planning defective equipment work order ~
equipment/W eek
delay planning defective equipment work order= time to plan corrective actions /(MIN(fractionPMwork , 0.5)*2) ~
Week
time to plan corrective actions= 1 ~ Week
breakdown report done=Breakdown reports Backlog / delaytime breakdown report ~ info/Week
newCBM: IF THEN ELSE(goalCB. NithCBM: , MIN(CB. P: *PM
ion release, MIN((goalC. i WithCBM: delay convert to CBM sensors, MAX(0,
ithCBM ionPlans/delay convert to CBM sensors))) , 0) ~ equipment/W eek
poy fraction report per THEN ELSE(number: i :AND:
0) ~ info/equipment
usage reactive staff=ZIDZ(unscheduled repairs, max capacity unscheduled repairs) ~ Dmnl
defectCreation= —_operations+collateral damage ~ defects/Week
A=INTEG (RCA measureT CB C. P i M 1)
~ info
AccCompanyResults= INTEG ( profit or lost,0) oe $
NetProfit=Net contribution margin production:= maintenanceT otalCost ~ $/Week
sumStaff=numberM ai irs+ resourcesU i od
people
profit or lost= NetProfit ~ $/Week
usage p) staff=ZID. i i i capacity i i ~ Dmnl
AccMaintBudgetMargin=INTEG (diffCostOverBudget, 0) ~ $
PMpreparation=MIN(ImplementedRCA *fractionPMiFromRCA /delay PM preparation , max capacity PM preparations)
~ info/Week
numberMaintenanceEngineers= 3 ~ people
productivity PM preparations= 0.5 ~ Dmnl
productivity engineers RCA analysis and PM preparations=10 ~ info/(W eek*people)
max capacity PM preparations= (max capacity RCA - RCA ity PM
info/Week
max capacity RCA =number' i i) ivity engineers RCA analysis and PM preparations ~
info/Weel
capitalInSparePartInventory =(spare part per equipment breakdown strategy* (number of equipment-SumPM preparations) +
spare part per equipment takedown strategy* ((1-fractionC BMoverPM) + 0.5*fractionC BMoverPM) * SumPMpreparations) *
cost per spare part ~ $
planned i i i ithCBMi val ~ i Week
spare part per equipment takedown strategy=2 ~ Dmnl
spare part per equipment breakdown strategy= 5 ~ Dmnl
delay RCA= time to implement/(MIN(fractionPMwork , 0.8)*2) ~ weeks
tbl breakdown frequency and stop effect([(0,0)-(4,9)],(0,1),(1,1),(2,3),(4,9)) ~ Dmnl
normal breakdown rate=18 ~ equipment/W eek
analytic iti i TealthOverPossibleD efects ~ Dmal
BreakdownA nalysisRCA WIP= INTEG (RCA UsefulData-RCA countermeasureToBreakdown, 0) ~ info
productivity CBM to sensor=0.5 ~ Dmnl
breakdown report demand= unscheduled repairs*policy fraction report per breakdown = ~ info/Week
Breskdown reports Backlog= INTEG ( breakdown Teport demand: breakdown report done, 0) ~ info
‘otalCost= ‘ost ~ $/Week
fraction available data RCA = useful info in reports * analytic capabilities ~ Dmnl
goalFractionCB MoverPM= 0.3 ~ Dmnl
CBMpreparation= MIN(ImplementedRCA* fractionCB MiFromRCA /delay convert to CBM , max capacity implement CBM
inspections) ~ info,
CBMsensorP ion= MIN( A*fractionCB. A/delay convert to CBM sensors , max capacity
implementing CBM sensors) ~ info/W eek
International System Dynamics Conference 2017
convertPMToCBM= MIN(MAX (0, i MoverPM:
fractionCB MoverPM WithPM ions)/ delay convert to CBM). CBMpreparation* PM preparation release)
~ equipment/W eek
fractionPMiFromRCA=0.5 ~ Dmnl
tbl pressure to close gap( [(0,0)-(100000,1)],(0,1),(1,1),(5,0.9),(10,0.7),(20,0.5),(100,0.2),(100000,0)) ~ Dmnl
time to implement= 13 ~ Week
useful info in reports= tbl pressure to close gap(Breakdown reports Backlog*pressure per breakdown report) ~ Dmnl
delay PM preparation=13 ~~ Week
PM preparation release=l ~ equipment/info
i ithPM INTEG (newP tart PM’ MToCBM-pl d repairs, initial
library of PM preparations) ~ equipment
pressure per breakdown report= 1 ~ l/finfo
goalCBMsensors= 25 ~ equipment
RCAUsefulData= breakdown report done*fraction available data RCA ~ info/Week
fractionCBMiFromRCA= __(1-fractionPMi A )* fractionCB Mi ‘Ahelp ~ Dmnl
productivity PM toCBM= 0.1 ~ Dmnl
SumPM i PM: tT olnspect: i ithCBMi ‘lans+
i ithC. i ithPM i < equipment
capital cost spare part inventory = interest rate spare part inventory/W eek * capi artInventory a
$/Week
interest rate spare part inventory= 0.4 ~ Dmnl
cost per spare part=2000 ~ $/equipment
cost per stop= 1.25*cost per spare part ~ $/equipment
cost cost per stop” ~ $/Week
man hour cost per Week= 2400 ~ $/(Person* W eek)
Week= 52 ~ Week
i i costs + capital cost spare part inventory ~ $/Week
fractionCB MoverPM = (Equi tT olnspect: WithCBMi ans i ‘ithCBMsensors) /
(PM i tT olnspect+ Equil WithCBMi: ionPlans+
i ithC. ithPM i ~ Dmnl
fractionPMwork= PM work/number of equipment ~ Dmnl
pressure scheduling delay= (delay Scheduling takedowns* pressure to produce) os Week
riskFactor {ealthOverPossibleD efects * (risk delayed work/
riskFactorReductionDueToPMwork) ~ Dmnl
riskFactorReductionD ueT oPMwork= tbl reduced risk due to PM work(fractionPMwork) ~ Dmnl
takedown rate p= IF THEN eee ee ented > limit takedown rate*number of equipment/pressure to produce,
0, PlannedT: pI delay) ~ Week
PlannedT: INTEG PMworkOrd due to unperformed takedowns- takedown
rate p - breakdowns due to unperformed takedowns, 4) ~ equipment
operations= Availability * wear and tear operations ~ defects/W eek
capacity inspections= MAX (max capacity schedued repairs - scheduled repairs , 0) * productivity inspections ~
equipment/Week
PMbacklog= PM: (Tolnspect+Defective equi, PlannedTakedowns ~
equipment
tbl reduced risk due to PM work([(0,0.8)-(1,2)],(0,1),(0.3,1.05),(0.6,1.4),(0.75,1.9),(1,2)) ~ Dmnl
defect elimination per repair= MA X (max fixed defects per repair* fracti i JealthOverPossibleDefects, 1) ~
defects/equipment
delayBreakdowns= tbl risk effect on reliability (riskFactorB reakdowns) * average reliability ~ Week
fractionEquipmenttiedthOverPosibleD efects= i Jealth/(number of equi possible defects per equipment)
~ m
tbl risk effect on reliability(((0,0)-(2.1,4)],(0,4),(0.3,3.6),(0.38,3),(0.45,1.5),(0.5,1),(0.65,0.72),(1.05,0.36),(2.1,0.1)) ~
Dmnl
productivity inspections= 2*0.8 ~~ Dmnl
tisk delayed work=2 ~ Dmnl
initial library of inspection plans= initLevelofInspPlans * number of equipment ~ equipment
20
International System Dynamics Conference 2017
initial library of PM preparations= initLevelofPMfi* number of equipment ~ equipment
limit takedown rate= 0.05 ~ Dmnl
productivity scheduled repairs= 36* 0.2 ~ equipment/(W eek*people)
productivity unscheduled repairs= 9*05 ~ equipment/(W eek*people)
average reliability=52 ~ Week
scheduled repairs=MIN(ScheduledMaintenance/delay scheduled repairs , max capacity schedued repairs) ~
equipment/Week
repairs= MIN(L i repairs , max capacity unscheduled repairs)
equipment/W eek
probability wear and tear= 0.015 ~ 1/Week
identifiedL i B NithCB ge CBM interval ~
equipment/Week
identified defective equi identifiedD: i i identifiedI
equipment/Week
number of machines in line= 20 ~ machines
number of equipment= number of machines in line* equipment per machine ~ equipment
equipment per machine=25 ~ equipment/machine
delay scheduled repairs= 0.05~ Week
delay unscheduled repairs= 0.1 ~ Week
max fixed defects per repair= 8 ~ defects/equipment
possible defects per equipment= 20 ~ defects/equipment
goal availability= 0.9 ~ Dmnl
takedownRate= takedownratep ~ equipment/W eek
delay scheduling takedowns=1 ~ Week
fixedInterval= 52 oa Week
EquipmentW ithCB MinspectionPlans= INTEG (convertPMToCBM-newC. planned i
inspections, initial library of inspection plans) ~ equipment
delay convert to CBM= 26 ~ Week
delay convert to CBM sensors= 52 ~ Week
PM INTEG (planned repairs-start PMwo, 0) ~ equipment
inspectionInterval= 4 ~ Week
Tealth= INTEG (defectCreation-defectEli: M -defectE. initial value of Hidden
defects*number of equipment) me defects
EquipmentT olnspect= INTEG (planned inspections-descretionary inspections, 0) ~ equipment
Defective equipment= INTEG (identified defective equi rrective 1) ~ i
INTEG ( heduled repairs, 0) ~ equipment
FINAL TIME =520 ~ Week
INITIAL TIME =0 ~ Week
SAVEPER =13 ~ Week [0,?]
TIME STEP =0.015625 ~ Week [0,?]
21
