1.

Introduction
Multi-state production systems (MSS) represent a class of particularly complex and
versatile systems, characterized by their ability to operate at multiple performance levels
instead of simply switching between a fully operational state and a failure state. This ability to
operate in degraded states, where production is reduced but not interrupted, is essential in
many industries where operational continuity is crucial. However, this flexibility introduces
unique management challenges, especially regarding component redundancy and maintenance
strategies.
Redundancy, which involves integrating additional components or operating in parallel, is
a key strategy to enhance the reliability of MSS. It allows the system to continue functioning
even if one or more components fail, thereby increasing the overall system availability.
However, implementing redundancies must be carefully planned, as it can lead to additional
costs in terms of initial investment and energy consumption. On the other hand, maintenance,
whether preventive or corrective, plays a crucial role in keeping system components in good
working condition, preventing unexpected failures, and extending equipment lifespan. A well-
designed maintenance plan can thus reduce production interruptions and associated
emergency repair costs.
Despite the importance of these two aspects, redundancy and maintenance optimization is
often addressed separately in the literature. This separation can lead to inefficiencies and
suboptimal solutions. For example, a system with poorly calibrated redundancy might require
more frequent maintenance, thus increasing operating costs. Conversely, a maintenance plan
developed without considering the system's redundant configuration might not fully leverage
the benefits of this redundancy, leading to unnecessary costs and suboptimal performance.
The objective of this article is to propose an integrated approach for optimizing
redundancy and maintenance in MSS. We aim to design an optimal system by determining
the number of machines operating in parallel for each subsystem, selecting the best version of
each machine from those available on the market, and developing a maintenance plan that
minimizes the total cost. By integrating these elements, our approach overcomes the
inefficiencies associated with separate optimization of redundancy and maintenance,
maximizing system availability and performance while reducing associated costs.
To achieve this, the article is structured as follows: In Section 2, we review the existing
literature on redundancy and maintenance optimization, highlighting integrated approaches
and their advantages over separate methods. Section 3 focuses on modeling the production
system, presenting the notations used, the system configuration, the Universal Moment
Generating Function (UMGF) method applied to multi-state systems, and the detailed
formulation of the optimization problem. Section 4 describes the methodological approach
adopted to solve the problem, with particular emphasis on the optimization algorithm and its
specific adaptations to this context. In Section 5, an illustrative example is presented to
demonstrate the effectiveness of the proposed approach, with a comparative analysis of results
obtained through integrated optimization versus separate optimizations. Finally, Section 6
concludes the article by summarizing the main contributions and suggesting directions for
future work, including improvements to optimization algorithms and their application to other
types of systems.
2. Literature Review
In this section, we present a detailed literature review, divided into five subsections:
Methods for Evaluating Availability and Reliability of Multi-State Systems, Maintenance in
Multi-State Systems, Redundancy Issues in Multi-State Systems and Joint Optimization of
Maintenance and Redundancy.
 2.1. Methods for Evaluating Availability and Reliability of Multi-State Systems
Assessing the performance of multi-state systems is a critical aspect of engineering
applications, requiring advanced methodologies to evaluate reliability and availability.
Several approaches have been proposed in the literature, with one of the most efficient
methods being the universal generating functions (UGF) introduced by [1]. This technique
has been applied in numerous studies to evaluate system reliability. [2] addressed the
multistage expansion problem for multistate series-parallel systems, aiming to minimize the
total investment costs over the study period while satisfying reliability constraints at each
stage. In this study, a genetic algorithm was used as an optimization tool, and UGF was
employed to evaluate the availability of multistate series-parallel systems. [3] introduced
fuzzy universal generating functions (FUGF) for the reliability assessment of multi-state
systems. This approach is particularly valuable for systems with multiple operational states
and complex dependencies. [4] developed a universal generating function-based performance
model for multi-state systems subject to correlated failures. This model offers a systematic
approach for analyzing system performance under correlated failure scenarios. [5] introduced
a fuzzy universal generating function-based method for the reliability evaluation of series
systems with performance sharing between adjacent units under parametric uncertainty. More
recently, [6] proposed a universal generating function-based approach to evaluate the
reliability of project completion time, providing narrow reliability bounds. [7] proposed a
methodology for the reliability evaluation of composite generation and transmission systems
using binary logistic regression and parallel processing. By leveraging logistic regression
techniques and parallel processing, this approach provides valuable insights into the
performance of interconnected systems. Li et al. (2023) conducted a reliability analysis of
cold-standby phased-mission systems based on the GO-FLOW methodology and universal
generating functions. [8] presented a reliability analysis method for complex
electromechanical multi-state systems using universal generating function technique. This
method incorporates fuzzy logic into reliability assessments, addressing uncertainties and
performance variations in series systems.
Other techniques have also been proposed to assess the performance of multi-state
systems. [9] introduced a Monte Carlo simulation approach to the availability assessment of
multi-state systems with operational dependencies. [10] conducted a reliability analysis of
multi-trigger binary systems subject to competing failures. This study provides insights into
the reliability assessment of systems with competing failure modes. [11] presented a
stochastic hybrid automaton model of a multi-state system with aging, focusing on reliability
assessment and design implications. [12] developed a multi-state Markov chain model for
rainfall to be used in the optimal operation of rainwater harvesting systems. This model offers
a systematic approach to optimizing the operation of rainwater harvesting systems based on
multi-state Markov chain analysis. [13] used surrogate modeling to assist in creep-fatigue
reliability assessment in a low-pressure turbine disc considering multi-source uncertainty.
[14] proposed a generic physics-informed neural network-based framework for the reliability
assessment of multi-state systems. By integrating physics-informed models with neural
networks, this framework offers an advanced approach to analyzing system reliability. [15]
introduced a deep neural network (DNN)-based reliability evaluation method for multi-state
series-parallel systems, considering semi-Markov processes. [16] developed an adaptive
hybrid deep learning-based reliability assessment framework for damping track systems
considering multi-random variables. This framework leverages deep learning techniques to
assess the reliability of complex systems under varying operational conditions. [17] focused
on the reliability assessment of multi-state weighted k-out-of-n man-machine systems
considering dependent machine deterioration and human fatigue. This study highlights the
importance of considering human factors and machine dependencies in reliability evaluations
of complex man-machine systems.
In this paper, the Universal Moment Generating Function method is employed to analyze the
reliability of multi-state systems. This choice is based on several key advantages that UMFG offers
over other approaches. Firstly, UMFG is renowned for its accuracy, providing precise results in
availability assessment. Additionally, it enables rapid evaluation of the performance of multi-state
systems, which is crucial for integration into metaheuristics. Finally, UMFG is robust, allowing for the
modeling of complex configurations. [1]
2.2. Maintenance in Multi-State Systems
Many researchers and practitioners have worked on various maintenance policies to prevent
system degradation and improve reliability. In many studies, systems were often considered binary,
meaning they were either fully operational or failed (e.g. [18]; [19]).
With technological evolution and the increasing complexity of systems, researchers become
interested to analyse multi-state systems, offering a more realistic representation of various operational
and degradation states. These systems can transition through a range of intermediate states between
full operation and complete failure, necessitating a more nuanced maintenance approach. In this
context, [20] investigated the integration of noncyclical preventive maintenance scheduling and
production planning for multi-state systems to improve system efficiency. In [21] , the authors
introduced a maintenance model designed for multi-state systems, which include a variety of
performance levels. The model incorporates preventive maintenance actions that influence system
elements age. A genetic algorithm is proposed to identify the most effective maintenance sequence,
ensuring desired system reliability while minimizing costs. [22] developed a multi-stage imperfect
maintenance strategy for multi-state systems with variable user demands, emphasizing the importance
of adaptive maintenance approaches. [23] explored selective maintenance optimization for multi-state
systems, considering stochastically dependent components and stochastic imperfect maintenance
actions. [24] proposed an optimal maintenance strategy for multi-state systems with a single
maintenance capacity and arbitrarily distributed maintenance time, incorporating machine learning
techniques. [25] focused on remaining useful life prediction and predictive maintenance strategies for
multi-state manufacturing systems, highlighting the role of functional dependence in maintenance
approaches. [26] introduced an integrated scheduling and flexible maintenance approach for
deteriorating multi-state single machine systems using a reinforcement learning approach. [27]
explored artificial intelligence-based maintenance decision-making and optimization for multi-state
component systems. [28] investigated the optimization of condition-based maintenance for multi-state
deterioration systems under random shock. [29] developed a Markov Regenerative Process Model for
analyzing the dependability and performance of a two-unit multi-state system under maintenance.
Lastly, [30] conducted a study on the performance efficiency and cost analysis of multi-state systems
experiencing successive damage and maintenance in multiple shock events, highlighting the
implications of shock events on maintenance costs and system performance. Shoorkand et al.
(Shoorkand, Nourelfath, & Hajji, 2024) integrated predictive maintenance with production planning to
improve the operational reliability of multi-state systems, enabling more efficient resource utilization
and reducing unexpected downtimes.
2.3. Redundancy optimization in Multi-State Systems
Numerous researchers have delved into the redundancy allocation problem in multistate
systems, contributing significant advancements to the field. For instance, [31] introduced a
value-driven approach for optimizing reliability-redundancy allocation in multi-state systems.
In 2017, [32] focused on optimal redundancy allocation for maximizing reliability in multi-
state computer networks subject to correlated failures. [33] explored the behavioral study and
availability optimization of a multi-state repairable system with hot redundancy, emphasizing
performance optimization using hot redundancy concepts. [34] addressed optimal redundancy
allocation for multi-state series-parallel systems under epistemic uncertainty, highlighting the
importance of considering uncertainty in redundancy optimization. Further advancements
include [35], who employed an ordinal optimization-based genetic algorithm for redundancy
optimization in multi-state series-parallel systems. [36] concentrated on redundancy allocation
optimization for multi-state systems with hierarchical performance requirements, contributing
to the understanding of optimizing system reliability while minimizing costs. More recently,
[37] conducted a reliability analysis of multi-state balanced systems with a standby
component switching mechanism, emphasizing the importance of maintaining system balance
in redundancy optimization.
2.4. Joint Optimization of Maintenance and Redundancy
In our knowledge, only few works have been published on an integrated maintenance and
redundancy optimization. [38] explored the optimization of redundancy and maintenance for multi-
state series-parallel systems. Their objective is to determine the optimal system structure and
replacement schedule to ensure desired reliability levels while minimizing associated costs. Similarly,
[39] proposed a framework for jointly optimizing imperfect preventive maintenance and redundancy,
the goal is to maximize system availability while considering budget constraints. Their heuristic
approach, based on Markov processes and genetic algorithms, offers a promising method to tackle this
complex challenge. In [40]. In this paper, we propose an optimization model that integrates optimal
redundancy allocation, preventive and corrective maintenance planning, and energy consumption
management. This model aims to reduce the system total cost by simultaneously addressing the
interdependencies between redundancy, maintenance schedules, and energy usage, providing a holistic
approach to system optimization.
The following table (see table 1) provides a comparative analysis of various research works in the
domain of reliability assessment, maintenance optimization, redundancy optimization, and energy
optimization for multi-state systems (MSS). Each reference is categorized based on the methodologies
employed in reliability evaluation and the specific areas of system optimization addressed. This
structured summary helps highlight the diversity of approaches used in optimizing complex systems,
including our contribution to the field, which integrates UMGF with optimization techniques for
maintenance, redundancy, and energy efficiency.
Table 1: Comparative analysis of reliability assessment and optimization methods in multi-state
system
Reference Reliability Evaluation Maintenance Redundancy Energy Description
Method Optimization Optimization Optimizatio
n
[1] UMGF  Introduces the Universal Generating Function for
reliability evaluation.
[2] UMGF ✔️ ✔️  Studies system expansion with availability constraints.
[3] UMGF  Uses fuzzy methods for multi-state system reliability
assessment.
[4] UMGF  Models performance with correlated failures using UGF.
[5] UMGF  Evaluates series systems with performance sharing and
parametric uncertainty.
[6] UMGF  Determines reliability bounds for project completion
time.
[7] Binary Logistic  Evaluates composite generation and transmission
Regression, Parallel systems.
Processing
[8] UMGF  Analyzes complex electromechanical multi-state
systems.
[9] Monte Carlo Simulation  Uses Monte Carlo simulation for availability assessment.
[10] UMGF  Analyzes multi-trigger binary systems.
[11] GO-FLOW  Analyzes cold-standby phased-mission systems.
Methodology, UMGF
[12] Markov Chain Model  Uses a multi-state Markov chain model for rainwater
harvesting systems.
[13] Surrogate Modeling  Evaluates creep-fatigue reliability with surrogate
modeling.
[14] Neural Network  Uses a physics-informed neural network framework for
Framework reliability assessment.
[15] DNN  Uses a DNN-based method for reliability evaluation.
[16] Hybrid Deep Learning  Uses a hybrid deep learning framework for reliability
assessment.
[17] UMGF  Assesses reliability of multi-state weighted k-out-of-n
systems.
[18] Selective Maintenance ✔️  Optimizes imperfect selective maintenance.
Optimization
[19] Binary Adversarial  Detects anomalies in marine diesel engines using Bi-
Autoencoder AAE.
[20] Preventive Maintenance ✔️  Integrates preventive maintenance scheduling and
Scheduling production planning.
[21] Imperfect Preventive ✔️  Optimizes imperfect preventive maintenance.
Maintenance
Optimization
[22] Multi-stage Imperfect ✔️  Uses a multi-stage imperfect maintenance strategy.
Maintenance
[23] Selective Maintenance ✔️  Optimizes selective maintenance considering dependent
Optimization components.

[24] Optimal Maintenance ✔️  Determines optimal maintenance strategy with single

Strategy capacity and distributed maintenance time.
[25] Predictive Maintenance ✔️  Predicts remaining useful life and uses predictive
Strategies maintenance strategies.
[26] Reinforcement ✔️  Uses reinforcement learning for integrated scheduling
 Learning for and flexible maintenance.
Maintenance
[27] AI-based Maintenance ✔️  Uses AI for maintenance decision-making and
Decision Making optimization.
[28] Condition-based ✔️  Optimizes condition-based maintenance under random
Maintenance shock.
Optimization
[29] Markov Regenerative  Uses a Markov regenerative process model for system
Process Model dependability.
[30] Performance ✔️  Analyzes performance efficiency and cost with
Efficiency and Cost successive damage.
Analysis
[31] Reliability-Redundancy ✔️  Optimizes reliability-redundancy allocation for multi-
Allocation state systems.
[32] Redundancy ✔️  Allocates redundancy to maximize multi-state computer
Allocation network reliability.
[33] Availability ✔️  Optimizes availability of a multi-state repairable system
Optimization with Hot with hot redundancy.
Redundancy
[34] Redundancy Allocation ✔️  Allocates redundancy under epistemic uncertainty.
[35] Redundancy ✔️  Optimizes redundancy using ordinal optimization-based-
Optimization genetic algorithm.
[36] Redundancy Allocation ✔️  Allocates redundancy with hierarchical performance
requirements.
[37] Reliability Analysis  Analyzes reliability of balanced systems with standby
with Standby components.
Components
[38] Joint ✔️ ✔️  Optimizes both redundancy and maintenance.
Redundancy and
Maintenance
Optimization
[39] Joint Redundancy and ✔️ ✔️  Optimizes redundancy and imperfect preventive
Preventive Maintenance maintenance.
Optimization
[40] Joint ✔️ ✔️  Optimizes redundancy and maintenance strategy.
Redundancy and
Maintenance Strategy
Optimization
OUR UMGF ✔️ ✔️ ✔️  Optimization of Redundancy, Maintenance, and Energy
WORK
 3. Problem formulation

Before formulating the problem, Table 2 presents the key notations used in the calculation
of energy costs and the optimization problem for multi-state systems. These notations define
the variables and parameters related to energy consumption, total cost minimization, system
structure, and reliability constraints.

Table 2: Summary of Notations for Energy Cost Calculation and System Optimization

Symbol Description
SSi Subsystem i, where i=1,2,…,n
n Total number of subsystems in the system
Ni Number of components in subsystem SSi
Cij Component j in subsystem i
kij Number of possible states for component Cij
Gij Set of performance states for component Cij
gijj’ Performance level corresponding to state j′ of component Cij
Xij(t) Random variable representing the performance of component Cij at time t
Pij Set of probabilities associated with the performance states of component Cij
Ψ(⋅) Function combining individual component performances to obtain the overall system
performance
Uij(Z) Moment generating function for component Cij
Upar(Z) Moment generating function for a subsystem in parallel
Userie(Z) Moment generating function for a subsystem in series
δ(Upar(Z),w Operator to calculate the probability that performance exceeds threshold w
)

τj(t) Effective age of component j at time t

tji Time of the i-th preventive maintenance action on component j

ϵi Age reduction coefficient after the iii-th preventive maintenance action
w Performance threshold
Ctotal Total cost (acquisition + maintenance + energy)
R(t) System reliability at time t
T System lifetime

R* Required reliability
3.1. System description
In this article, the system under study is a series-parallel production system. Multi-state
systems represent a class of systems capable of operating at different performance levels,
offering greater flexibility than traditional binary systems. Unlike systems that can only be in
two states (fully operational or complete failure), an MSS can operate in several intermediate
states, corresponding to various levels of degraded performance. This capability allows for
maintaining some level of productivity, even when components are partially failing, which is
crucial in industries where operational continuity is a priority.
An MSS is composed of several subsystems denoted as SSi for i=1, 2,...,n, where n represents
the total number of subsystems within the system. Each subsystem, in turn, consists of a set of
components, with Ni components for each subsystem SSi. The number Ni of components can
vary from one subsystem to another (see Figure 1).

Figure 1 : components of MSS

The components of these subsystems are also multi-state. Each component Cij, where i
denotes the subsystem and j=1,2,...,Ni is the index of the component within the subsystem SSi,
can adopt multiple performance states. These states are described by a set Gij={gij1,gij2,...,gijkij},
where kij represents the number of possible states for component Cij, and each gijj′corresponds
to a specific performance level in state j′.
 State 0 (complete failure): The component is out of service.
 State kij (perfect operation): The component operates at full capacity.
 Intermediate states: The component operates at reduced capacity.
Thus, individual components in the MSS can function in degraded states, affecting the
overall system performance. For example, a compone nt such as a motor or turbine might
operate at reduced capacity in case of partial malfunction rather than stopping completely.
The performance of components can vary over time due to factors such as wear and tear,
maintenance, or changing operating conditions. At any time, t≥0, the performance of a
component Cij is modeled by a random variable Xij(t) taking values in the set Gij. This modeling
represents the dynamic nature of multi-state systems, where a component's performance can
fluctuate based on its state at a given time.
The probabilities associated with each performance state gijj′ of component Cij are grouped in a
set Pij={pij1,pij2,...,pijkij}, where each pijj′ represents the probability that Cij is in state gijj′ at time t,
pijj′=P(Xij(t)=gijj′).
Thus, MSS components are not only capable of functioning in different states, but they also
have a certain probability of being in each of these states at any given time.
The overall performance of an MSS is a combination of the performances of its individual
components. For a system composed of n subsystems and Ni components per subsystem, the
overall performance rate at time t, denoted Y(t), depends on the individual performances of
the components in each subsystem. This overall performance is calculated by a function Ψ,
which combines the individual performances of the components:
Ψ(X11(t),X12(t),...,XnNn(t)),where Xij(t) represents the performance of component Cij at time t.
The function Ψ varies depending on the system configuration.
  Series configuration: The overall system performance is limited by the component
with the lowest performance. A critical fault in a component in series can affect the
entire subsystem.
 Parallel configuration: The system can continue to function even if some components
are in failure. Redundancy allows the system to tolerate failures while maintaining
some production capacity.
Multi-State Systems (MSS) find applications in various fields where component
availability and operational flexibility are crucial. Here are a few examples:
 Energy Plants: Generators can operate at varying capacities depending on their
performance state. An MSS models the overall energy production considering the
different states of the generators.
 Telecommunication Systems: Communication equipment can be operational at
different capacities or out of service. An MSS helps model network availability based
on the performance of the equipment.
 Transport Systems: In a transportation network, vehicles can be fully operational,
partially failing (reduced capacity), or out of service. An MSS models the overall
network capacity by considering the various states of the vehicles.
An MSS captures the complexity of systems where components can operate at multiple
levels of degraded performance. By integrating these different states, MSS allows for a more
realistic assessment of availability and performance in critical systems. Additionally,
managing subsystems and modeling the probabilities associated with each performance state
offers increased flexibility to maximize productivity while reducing failure-related risks.
3.2. UMGF Method
The Universal Moment Generating Function (UMGF) [1] is a mathematical tool used to
evaluate the reliability and performance of complex systems, particularly those with multiple
performance levels, known as multi-state systems. It utilizes moment generating functions to
represent and combine the different performance levels of system components. This allows
for precise and efficient modelling of the system's overall performance based on the
individual performance of its components.
The UMGF is an extended concept of the moment generating function typically used in
statistics to summarize the properties of probability distributions. In the context of multi-state
systems, the UMGF captures all information about a component's performance levels in a
single mathematical function.
For a component Cij in a subsystem, the moment generating function Uij(Z) is defined as:
k ij

Uij(Z)=∑ pij j × Z
gijj'
'
'
j
(1)
where:
 pij j : Probability that component Cij is in state j′.
'

 Gijj’: Performance level associated with state j′ of component Cij. This represents the
quantitative measure of performance when the component is in state j′. Performance
levels are often numerical values representing different operating efficiencies.
 Kij: indicates how many different performance levels the component can have.
The moment generating function encapsulates all information about a component's
possible performance in a single mathematical expression. This function is crucial for
calculating the overall performance of the system by combining the performances of different
components.
 The application of UMGF varies depending on the subsystem configuration. The main
configurations are parallel and series subsystems. The application of UMGF differs
significantly between these two types.
In a parallel subsystem, components work together to ensure the subsystem's overall
performance. The total performance of the subsystem is obtained by combining the
performances of individual components. The total moment generating function for a parallel
subsystem is calculated as the product of the moment generating functions of the individual
components:
N
Upar(Z)=∑ U i (Z )
i=1
(2)
where:
 Ui(Z): Moment generating function of the i-th component in the subsystem. Each
component has a distinct moment generating function reflecting its own performance
levels.
 N: Total number of components in the parallel subsystem. This term indicates how
many components are present in the parallel configuration.
Multiplying these moment generating functions reflects how each component contributes
to the subsystem's overall performance. By combining these functions, we get a
comprehensive view of the subsystem's total performance.
In a series subsystem, the overall performance depends on the component with the lowest
performance because all components must function correctly for the subsystem to be
operational. The total moment generating function for a series subsystem is obtained by using
the convolution of the moment generating functions of the individual components:
Userie(Z)=min(U1(Z),U2(Z),…,UN(Z)) (3)
Convolution of moment generating functions in a series configuration shows how the
failure of one component impacts the entire subsystem. This modelling accurately reflects the
performance of the system in series configurations.
Reliability calculation using UMGF involves several steps. Reliability measures the
probability that the subsystem's overall performance exceeds a certain threshold.
For a parallel subsystem, reliability is determined using the total moment generating
function and evaluating the probability that the performance is greater than or equal to a
threshold w. Reliability is calculated as follows:
Pr(Y(t)≥w) =δ(Upar(Z),w) (4)
where:
 δ(Upar(Z), w): Calculation operator that determines the probability associated with
states where the performance is greater than or equal to the threshold w. This usually
involves summing probabilities for performance levels that meet the condition.
For a series subsystem, reliability is calculated by evaluating the probability that the
total performance exceeds the threshold www. This is done using similar methods with the
moment generating function and applying similar calculations to determine the probability
that the performance is above the threshold.
For example, consider a subsystem consisting of two machines in parallel. Each machine can
have two performance states:
 Machine 1:
State 0 (Failure): Probability = 0.2, Performance = 0
State 1 (Operational): Probability = 0.8, Performance = 1
 Machine 2:
State 0 (Failure): Probability = 0.3, Performance = 0
State 1 (Operational): Probability = 0.7, Performance = 1
 We will use this data to calculate the moment generating function and evaluate the
subsystem's reliability.
For each machine, we calculate the moment generating function:
Machine 1:
U1(Z)=0.2⋅Z0+0.8⋅Z1=0.2+0.8Z
Machine 2:
U2(Z)=0.3⋅Z0+0.7⋅Z1=0.3+0.7Z
The total moment generating function for the two machines in parallel is the product of their
moment generating functions:
Upar(Z)=U1(Z)×U2(Z)=(0.2+0.8Z)×(0.3+0.7Z)
Upar(Z)=(0.2×0.3) +(0.2×0.7Z)+(0.8Z×0.3)+(0.8Z×0.7Z)
Upar(Z)=0.06+0.14Z+0.24Z+0.56Z2
For a performance threshold of 1, we want to determine the probability that the parallel
system has a performance of at least this threshold. Using the moment generating function, we
calculate the probability associated with this performance:
Pr(Y≥1) =1−Coefficient of Z0 in Upar(Z)=1−0.06=0.94
Thus, the reliability of the parallel system, i.e., the probability that the performance is at
least 1, is 0.94 or 94%.
The UMGF is a powerful tool for evaluating the reliability and performance of multi-state
systems. By modelling the individual performance levels of components and combining these
performances at the system level, UMGF provides a precise and efficient means of assessing
overall system performance. The methods described allow for modelling both parallel and
series configurations, offering critical insights for optimizing industrial systems.
3.3. The maintenance plan applied to the system
This section details the maintenance plan implemented to optimize the reliability and
availability of a series-parallel production system while minimizing costs. This plan includes
two main types of interventions: preventive maintenance and corrective maintenance.
3.3.1 Preventive Maintenance
Preventive maintenance (PM) involves proactively intervening on system components
before they fail. This helps maintain high reliability and prevent unplanned shutdowns. PM
actions are triggered either periodically or when the reliability of a component reaches a
critical threshold, denoted R*. PM actions include several types of operations, such as:
 Part Replacement: Replacing a worn component before it fails. For example, replacing
an air filter every 1000 hours to prevent clogging.
 Cleaning: Removing accumulated debris to ensure proper operation. For example,
cleaning the nozzles of an industrial printer every month.
 Adjustment: Adjusting a component's parameters to optimize its operation, such as
recalibrating a temperature sensor.
Age Reduction in Preventive Maintenance
Age reduction models the effect of preventive maintenance actions on the state of a
component. A PM action can be viewed as a partial rejuvenation of the component, delaying
the onset of failures. According to the proportional age reduction model, the effective age of
the component after a PM action is given by the following formula:
τj(t)= τj+(tji) +(t-tji) for tji<t<tji+1 (where 0≤i≤n)
(5)
with
τj+(tji) = ϵi + τj(tji−1)
where:
 τj(t) represents the age of the component at time t,
 ϵi is the age reduction coefficient after the i-th preventive maintenance action,
 tji is the time at which the i-th PM action occurs.
For example, suppose a compressor has an initial age of 2000 hours. After a PM (deep
cleaning) at 1000 hours with a coefficient ϵ=0.5, the effective age of the compressor at 1200
hours becomes:
tj(1200)=2000+0,5⋅(1200−1000)=2100 hours
This means the PM has extended the effective lifespan of the compressor.
3.3.2. Corrective Maintenance
Corrective maintenance occurs when a failure happens. The repairs associated with this
maintenance have a cost, which depends on the component’s failure rate. The expected total
cost of minimal repairs is given by:

CMj=cj⋅∫ h j ( x ) dx
T

0
(6)
where:
 Cj is the repair cost,
 T is the system's lifetime,
 hj(x) is the hazard function of component j.
For example, consider a machine with a repair cost of 400 units and a failure rate of 0.05 per
hour, over a lifetime of 1000 hours, the expected total repair cost is:
CM=400×0,05×1000=20000 monetary units
Hazard Function and Reliability
The hazard function hj(t) describes the failure rate of a component at a time t, while
reliability rj(t) represents the probability that a component operates without failure up to a
certain time. It is calculated by:
− ∫ τ (t )h j(x) dx
rj(t)=e τj
+¿ (t )
¿
ji
(6)
For example, if the hazard function is constant at 0.01, the reliability of a component between
200 and 300 hours is calculated as:
300

− ∫ 0.01 dx
rj(t)= e 200 =exp (−1) ≈0,3679
This means that at 300 hours, there is approximately a 36.79% chance that the component is
still functioning properly.
3.4. Problem Formulation
The objective is to minimize the total cost of a production system considering several
critical elements: the cost of acquiring machines, the cost associated with the maintenance
plan applied to the system, and the energy cost required for its operation. Additionally, it is
essential to maintain the system's reliability above a certain critical threshold.
Ctotal=Cacquisition+Cmaintenance+Cénergie
(7)
 Machine Acquisition Cost:
The cost of acquiring machines is the total sum of the purchase costs for all machines
needed in the multi-state system.
 N Ni

Cacquisition = ∑ ∑ caij (8)

i=1 j=1
ca ij: acquisition cost of machine j in sub-system i
 Maintenance Cost:
The total maintenance cost of a multi-state production system consists of two main
elements: the cost of preventive maintenance and the cost of corrective maintenance.
Preventive Maintenance Cost (CP)
The cost of preventive maintenance is calculated based on the preventive maintenance
actions performed on the system over its lifetime. Each preventive action v i is associated with
a specific cost cp ( v i).
NA
CP=∑ cp(¿ v i)¿
i=1
(9)
Corrective Maintenance Cost (CM)
The cost of corrective maintenance corresponds to the cost of minimal repairs performed
after failures. Each component j of the system has a minimal repair cost cj, which depends on
the failure frequency nj and the hazard function hj(t).
J nj
CM=∑ cj ∑ [τ j(t ji +1)−H (t ji )] (10)
i=1 i=0
where:
 J is the total number of components in the system,
 Cj is the minimal repair cost of component j,
 nj is the number of corrective actions for component j,
 H is the hazard function.
The total maintenance cost is the sum of the preventive maintenance cost and the
corrective maintenance cost: CTM=CP+CM
Energy Cost:
Optimizing maintenance and redundancy in multi-state systems goes beyond mere
reliability and operational availability; it also influences energy consumption. The age and
condition of system components directly impact its energy efficiency.
Proper management of preventive maintenance and redundancy can extend the useful life
of equipment, reducing the need for frequent replacements and costly upgrades. This can help
decrease the energy consumption associated with manufacturing new components and
disposing of obsolete equipment.
Moreover, well-planned preventive maintenance can identify and correct operational
inefficiencies that may contribute to excessive energy consumption. For example, by
detecting and repairing leaks, excessive friction, or imbalances in components, systems can
operate more efficiently, reducing their energy footprint.
Furthermore, judiciously designed redundancy can enable more efficient use of energy
resources by allowing systems to operate at optimal load levels. For example, by
automatically redistributing the workload among redundant components based on their
condition and performance, systems can minimize energy overconsumption resulting from
inefficient resource use.
 In summary, optimizing maintenance and redundancy in multi-state systems can
significantly impact energy consumption, contributing to the overall sustainability and
operational efficiency of industrial infrastructure.
E(t)=a+b×t (for t<T)
E(t) = a + bT + c×(t - T) (for t≥T)
(11)
With :
 a: Initial energy consumption (for a new machine)
 b: Normal increase rate
 c: Accelerated increase rate
 T: Threshold age where acceleration begins
To obtain the total energy cost of a machine over a period T, we start by calculating the
total energy consumption. The period T is divided into several sub-periods of length O. The
energy consumption for each sub-period [t, t+O] is calculated using the trapezoidal rule
applied to the linear function E(t) = a + bt. The consumption for a sub-period is thus:
O∗[E ( t ) + E ( t+O )]
Consumption= 2
(12)
By summing the energy consumption of all sub-periods, the total energy consumption over
the period T is obtained:
T
Total consumption= ∑ O ׿ ¿ ¿ (13)
t =0
For example, if T is 10 hours and the period is divided into sub-periods of 2 hours, the
consumption for each 2-hour interval will be calculated and summed to get the total energy
consumption.
Once the total energy consumption is determined, simply multiply this total by the unit energy
cost Cu, expressed in monetary units per unit of energy. The formula for the total cost is:
Cenergy = (Total consumption) *Cu
(14)
The optimization problem can be then formulated as follows:
Minimize Ctotal=Cacquisition+Cmaintenance+Cenergy
N Ni NA J nj
With Ctotal=∑ ∑ caij+∑ cp(¿ v i)¿+∑ cj ∑ [τ j(t ji +1)−H (t ji )]
i=1 j=1 i=1 i=1 i=0
N Ni T
+∑ ∑ ∑ O× ¿ ¿ ¿ ¿ ¿ (15)
i=1 k=1 t =0

R(t)≥R*, ∀t∈ [0, T]

Subject to

to ensure that the system's reliability at each instant t, denoted R(t), remains greater than or
equal to a minimum threshold R*, set according to performance and safety requirements.
4. Methodological Approach
4.1. Detailed Description of NLTA
The Non-Linear Threshold Accepting (NLTA) [41] method is a heuristic designed to
tackle complex combinatorial optimization problems, particularly in the management of
Multi-State Systems. Unlike traditional methods, NLTA employs a non-linear acceptance
mechanism that adapts the acceptance threshold for candidate solutions throughout the
optimization process. This mechanism facilitates a thorough exploration of the solution space
while minimizing the risk of getting trapped in local minima. This non-linear mechanism
allows the acceptance threshold to adjust over iterations, promoting a transition from broad
exploration to more targeted exploitation of promising solutions. Initially, a high threshold
enables the acceptance of diverse solutions, while the threshold progressively decreases,
making the algorithm more selective. The Non-Linear Acceptance Threshold is given by:
1
V out ¿ 2
w
V¿ 1+( )
w0
(14)
In this equation:
 Vout is the output voltage, representing the acceptance level of solutions.
 Vin is the input voltage, reflecting the initial conditions of the system.
 w is the current acceptance threshold, and w0 is the initial threshold.
The Non-Linear Threshold Accepting algorithm effectively balances exploration and
exploitation, while also avoiding local minima and adapting to multiple objectives and
constraints. At the start of the process, NLTA adopts an expansive strategy, accepting a
variety of solutions, even suboptimal ones. This exploration phase is essential for navigating
the solution space, potentially uncovering globally optimal solutions. As the process advances
and the acceptance threshold tightens, NLTA gradually shifts its focus toward refining the
most promising solutions, which enhances the efficiency of the optimization.
Furthermore, NLTA's non-linear acceptance threshold allows it to occasionally accept less
optimal solutions, a feature that is particularly crucial in combinatorial optimization problems,
where the algorithm can otherwise become stuck in local minima. By accepting these
suboptimal solutions, NLTA ensures diversity in the search process, facilitating the
exploration of new regions in the solution space.
Finally, NLTA is designed to be highly flexible, making it capable of incorporating a wide
range of constraints that are specific to multi-state systems (MSS). These constraints may
include budget limitations, reliability requirements, and minimum performance thresholds.
This adaptability allows NLTA to handle complex optimization scenarios where multiple
objectives must be addressed simultaneously, making it particularly well-suited for multi-
objective optimization problems.
Justification of NLTA for MSS Optimization
Given the complexity of the problem, which is NP-hard, the application of a heuristic
approach is essential. Multi-state systems (MSS) involve numerous decision variables, such
as the number of machines per subsystem, the available machine versions, and various
maintenance strategies. This wide range of parameters complicates the optimization process,
as it requires consideration of a multitude of possible combinations. Additionally, decisions
made in one subsystem often affect others, creating complex interdependencies. For instance,
choosing a specific machine version can influence the maintenance strategy for the entire
system, adding another layer of complexity to the problem.
Exact methods, such as linear programming, quickly reach their limits when dealing with
such complex systems. Due to the combinatorial explosion of the solution space, these
methods become impractical for large-scale MSS, which limits their applicability in real-
world scenarios. Moreover, traditional methods often struggle to escape local minima, making
it difficult to reach the global optimum and resulting in suboptimal outcomes for complex
systems.
To overcome these limitations, we have chosen to apply the Non-Linear Threshold
Accepting (NLTA) algorithm, which has proven effective in solving numerous complex
optimization problems. This algorithm offers several advantages. Firstly, it provides
significant flexibility by allowing a broader exploration of the solution space through the
acceptance of initially suboptimal solutions. This feature is crucial for discovering innovative
solutions. Secondly, NLTA’s non-linear acceptance threshold helps to avoid local minima by
permitting temporary deviations from optimal solutions, thus fostering a diverse and dynamic
search strategy. Finally, the NLTA algorithm is highly adaptable, capable of integrating
various constraints and objectives. This makes it particularly effective for optimizing MSS,
where trade-offs often need to be made between conflicting objectives.
The non-linear acceptance mechanism, based on the RC filter model, adjusts the threshold
according to the algorithm's progress. This model ensures a gradual transition between
exploration and exploitation, allowing the algorithm to adapt to the dynamics of the solution
space.
NLTA is a powerful and adaptable method for optimizing Multi-State Systems,
demonstrating superior performance compared to classical methods. Its unique approach
enables it to efficiently explore the solution space while effectively managing complex
interdependencies and constraints that are characteristic of MSS. The flexibility and
adaptability of NLTA have been underscored by Nahas and Nourelfath [41,42], who
highlighted its capacity to accommodate various optimization scenarios and objectives.
Furthermore, subsequent studies by Nahas et al. [43] have validated its effectiveness,
showcasing how NLTA can address intricate optimization challenges, such as balancing cost,
reliability, and performance in multi-state systems. This robustness positions NLTA as a
critical tool for researchers and practitioners aiming to enhance system performance in
complex operational environments.
4.2 NLTA for Multi-State System Optimization
The adaptation of the Neighborhood Local Search Algorithm (NLTA) to Multi-State
Systems is based on two essential steps: the generation of the initial solution and the
generation of neighboring solutions. These steps are designed to maximize the algorithm's
efficiency while considering the specific characteristics of MSS.
I. Generation of the Initial Solution
Step 1. Select randomly the number ni of machines for each subsystem i, respecting the
constraint 1≤ni≤M, where M is the maximum capacity of the subsystem.
Step 2. Select randomly a machine version from the available market options. Each version is
characterized by:
- g: The machine's productivity.
- cm: The minimal repair cost.
- ca: The acquisition cost of each machine version.
- Weibull Parameters: h: Shape parameter (or shape index) of the Weibull distribution; λ:
Scale parameter of the Weibull distribution; η: Location parameter of the Weibull
distribution.
- Energy Cost: a: Energy cost per unit time before the machine reaches its lifetime T; b:
Additional energy cost per unit time after the machine reaches its lifetime T; c: Further
increased cost of energy per unit of time when the machine operates beyond its lifespan
T.
Step 3. Preventive Maintenance Plan: for each machine Cij Select randomly NA preventive
actions. Each maintenance action is characterized by:
- Age Reduction AR: This indicates how much the maintenance action extends the
machine’s useful life by reducing its age.
- Cost C: This represents the financial resources required to perform the maintenance action.
This step integrates maintenance considerations into the initial solution generation,
ensuring that maintenance is considered to optimize the system's overall performance.
 1. Divide the total period T into sub-periods of length O. Recalculate the system's
reliability at each interval ttt. If the reliability falls below the required minimum, apply
a preventive maintenance action randomly from the available actions for each
machine.
2. Evaluate the total cost of the configuration by combining acquisition, maintenance,
and energy costs according to the model's specific formula.
3. Compare the obtained reliability with the required minimum reliability. If the solution
does not meet this constraint, repeat the generation process until a feasible solution is
found.
II. Generation of Neighboring Solutions
After establishing the initial solution, the algorithm explores the search space by
generating neighboring solutions. This is made by slightly modifying the actual solution.
Three types of modifications are made. The first type of modification involves randomly
adding or removing a machine from the system. The second type of modification consists of
randomly adding an additional machine. The third type of modification focuses on altering a
single action within the maintenance plan, ensuring that the overall maintenance strategy is
adjusted without drastically changing the entire [Link] procedure is as follows:
1. Choose randomly a type of modification k
2. If k=1: Adding or Removing a Machine
Step 2.1. Select randomly a subsystem
Step 2.2. Select randomly a binary l:
- If l=1: A new machine is added. In this case, a preventive maintenance action is
randomly selected for this machine and placed at the last position m of the list of actions
applied to the system.
- If l=0: A machine is removed. All the maintenance actions associated with this machine
are removed and new maintenance actions assigned to the existing machines are
randomly chosen from the list of the maintenance actions initially generated.
3. k=2: Replacing a Machine. In this case, a randomly existing machine is replaced with a
different version. The maintenance actions of the old version of the machine are replaced by
new actions randomly chosen from the list of the maintenance actions initially generated.
4. k=3: Modifying the Maintenance Plan. In this case, a maintenance action is randomly
selected and removed, and a new maintenance action is randomly selected from the list of the
maintenance actions initially generated.
5. After each modification, recalculate the total cost and reliability of the newly generated
neighboring solution.
NLTA Acceptance Condition
The NLTA algorithm applies an acceptance condition to evaluate whether the
generated neighboring solution S′ should be accepted or rejected. The acceptance
condition is defined as follows:
1
2
Condition= w
1+( )
w0
Acceptance Criteria:
The neighboring solution S′ is accepted if one of the following conditions is met:
 Reliability: The reliability of S′ meets the required MSS system standards.
 Cost:
o If the total cost of S′ is less than or equal to that of the current solution S, then
accept S ' .
 CS'
o Otherwise, calculate the cost ratio . If this ratio exceeds the acceptance
CS
1
2
condition w , accept S ' .
1+( )
w0
Otherwise:
 Reject the neighboring solution: If none of the above conditions are met, the previous
solution SSS is retained, and the algorithm does not switch to the new solution.
This approach allows the algorithm to accept potentially suboptimal solutions in terms
of cost when the acceptance threshold www permits, promoting exploration of the
solution space and avoiding local minima.
III. Updating the Acceptance Threshold
The parameter www is initialized with the value w0 before starting the iterations. The
threshold www is gradually reduced by Δw after each iteration, adjusting the rigor of solution
acceptance, shifting from a more liberal exploration at the beginning to a more targeted
exploitation of promising solutions.
IV. Stopping Criterion
The process of generating and evaluating neighboring solutions is repeated until a
stopping criterion is met. The stopping criterion used is:
 Maximum number of iterations: Stop the algorithm after a predefined number of
iterations.
5. Exemple illustratif
In this example, we analyze a multi-state system consisting of four subsystems, each
capable of hosting up to five machines operating in parallel, with seven different versions
available for each machine. These machines are defined by several technical and economic
parameters, including their productivity, costs (acquisition and repair), and reliability
parameters modeled by the Weibull distribution. Additionally, a preventive maintenance plan
is implemented, offering several actions designed to extend the lifespan of the machines while
optimizing operational costs.
The system is designed for a lifespan of 25 years, with scheduled maintenance interventions
every 1.5 months. The client’s productivity demand is set at 0.8, while the minimum required
reliability for the system is 0.9. Therefore, the primary challenge is to ensure that the system
remains productive and reliable throughout its lifespan while minimizing the associated repair
and maintenance costs.
The main objective of this example is to design an optimal multi-state system by
making the following decisions:
 Determine the number of machines in parallel for each subsystem: This decision will
enhance redundancy and thus improve the overall reliability of the system. However,
it is crucial to balance the acquisition and energy costs, which increase with the
number of machines.
 Select the optimal versions of the machines available for each subsystem: Each
version differs in terms of cost, productivity, reliability, and energy consumption. The
goal is to choose versions that maximize both productivity and reliability while
minimizing operational costs.
 Develop an optimal preventive maintenance plan: This plan should include regular
actions aimed at extending the lifespan of the machines and preventing costly failures.
It involves selecting the most effective and cost-efficient maintenance actions for each
machine, considering the effects of reducing the machines' age and the costs of
interventions.
 The ultimate goal is to minimize the total cost of the system, which includes the cost of
acquiring machines, repair costs, and energy costs, while adhering to the reliability and
productivity constraints set by the client.
Table 1 outlines the technical and economic parameters for each machine across the
four subsystems of the multi-state system. For each machine and version, key information
includes productivity (g), acquisition cost (ca), minimum repair cost (cm), and reliability
parameters modeled by the Weibull distribution: shape parameter (h0), scale parameter (λ),
and location parameter (η). This data is essential for optimizing performance and costs based
on productivity and reliability requirements.
Table 1: Technical and Economic Parameters of Machines
Repair Purchase
Component Version g λ η h0 Cost Cost a b
1, 0, 1,0
1 1 0,063 1,9 0 1,6 5300 9 1
1, 1,0
2 3 0,066 1,9 0 1,6 5200 1 3
1,0
3 1 0,065 1,8 0 2,2 5400 1 1
1, 1,
1
4 1 0,062 1,9 0,00005 1,5 5500 1 0,9
1, 0,9
5 2 0,07 1,8 0,00005 1 5100 1 8
1, 0,
6 1 0,07 1,9 0,00005 1,6 5300 9 0,9
0, 0,9
7 9 0,06 1,9 0,00005 1,5 5000 1 3
1, 1,0
1 2 0,065 1,9 0,00005 2 5200 1 2
0, 0,9
2 9 0,061 1,9 0 1,6 5300 1 7
0, 0, 0,9
3 9 0,065 2 0,00005 1,5 5200 9 2
1,
2
4 2 0,066 1,8 0 2 5200 1 0,9
0, 0,9
5 1 0,067 1,8 0 1,6 5400 9 7
0, 0, 0,9
6 9 0,063 1,8 0,00005 1,5 5100 9 4
1,
7 3 0,065 1,9 0 2 5100 1 1
3 1,0
1 1 0,068 1,9 0,00005 1,6 5400 1 4
1, 0,9
2 2 0,061 1,9 0,00005 0,9 5800 1 6
0, 0,9
3 9 0,07 1,9 0 1,8 5000 1 7
0, 0,9
4 1 0,069 2 0,00005 1 5900 9 3
1, 0, 1,0
5 1 0,069 2 0 1,1 5100 9 2
6 0, 0,062 2 0,00005 0,7 5500 1 1
9
 1, 1,0
7 2 0,062 1,9 0,00005 1,7 5500 1 3
1, 1,0
1 3 0,065 1,9 0 1,4 5200 1 4
1, 0,9
2 3 0,069 2 0,00005 2,1 5600 1 6
1, 0,9
3 2 0,069 2 0 1,9 5300 1 9
1, 1,0
4
4 1 0,065 2 0,00005 0,6 5300 1 1
0, 0,9
5 1 0,068 1,8 0,00005 1,9 5700 9 6
0, 0, 1,0
6 9 0,064 1,9 0,00005 0,7 5300 9 4
0, 0,9
7 9 0,069 1,9 0 2 5500 1 8
Table 2 summarizes the preventive maintenance actions available for each machine.
Each machine can perform three types of actions, detailing intervention costs and age
reduction coefficients, which measure the effectiveness of each action in extending machine
lifespan. This information is crucial for developing an optimal maintenance strategy that
reduces costs while enhancing reliability.
Table 2: Preventive Maintenance Actions
Action Componen Age
Number t version Reduction PM Cost
1 1 1 0.7 13.8
2 1 1 0.1 8.9
3 1 1 0.4 6.6
4 1 2 0.9 2.7
5 1 2 0.3 8.0
6 1 2 0.0 18.8
7 1 3 0.9 7.9
8 1 3 0.5 10.0
9 1 3 0.7 17.5
10 1 4 0.1 4.9
11 1 4 0.2 18.3
12 1 4 0.6 12.9
13 1 5 0.5 2.7
14 1 5 0.1 9.0
15 1 5 0.2 3.8
16 1 6 0.7 7.7
17 1 6 0.0 9.1
18 1 6 0.2 11.9
19 1 7 0.3 4.4
20 1 7 0.7 10.6
21 1 7 0.8 5.8
22 2 1 0.6 2.7
23 2 1 0.3 7.6
24 2 1 0.9 7.2
25 2 2 0.1 14.4
26 2 2 0.0 18.7
27 2 2 0.7 6.2
28 2 3 0.5 14.7
29 2 3 0.0 16.1
30 2 3 0.1 15.5
31 2 4 0.4 10.1
32 2 4 0.5 12.7
33 2 4 0.1 3.2
34 2 5 0.6 10.4
35 2 5 0.4 17.3
36 2 5 0.3 11.7
37 2 6 0.2 17.0
38 2 6 0.1 3.2
39 2 6 1.0 13.0
40 2 7 1.0 14.0
41 2 7 0.9 5.9
42 2 7 0.4 4.8
43 3 1 0.7 14.7
44 3 1 0.6 4.2
45 3 1 0.2 6.1
46 3 2 0.9 6.0
47 3 2 0.9 16.2
48 3 2 0.1 7.1
49 3 3 0.3 10.8
50 3 3 0.5 6.4
51 3 3 0.8 4.5
52 3 4 0.8 7.8
53 3 4 0.6 3.5
54 3 4 0.3 5.2
55 3 5 0.8 15.6
56 3 5 0.3 5.0
57 3 5 0.5 19.0
58 3 6 1.0 8.1
59 3 6 0.1 17.3
60 3 6 0.4 3.6
61 3 7 0.9 12.7
62 3 7 0.3 8.9
63 3 7 0.5 7.1
64 4 1 0.8 2.9
65 4 1 0.6 12.8
66 4 1 1.0 16.2
67 4 2 0.1 5.5
68 4 2 0.1 10.6
69 4 2 0.4 15.5
70 4 3 0.3 14.8
71 4 3 0.6 8.9
72 4 3 0.1 2.5
73 4 4 0.2 9.6
74 4 4 0.8 10.0
 75 4 4 0.1 3.0
76 4 5 1.0 8.7
77 4 5 0.0 3.4
78 4 5 0.5 8.7
79 4 6 0.2 17.1
80 4 6 0.4 6.7
81 4 6 0.8 2.3
82 4 7 0 11.2
83 4 7 0.1 7.2
84 4 7 0.0 8.1
Example of a Multi-State System
In this example, we examine a multi-state system composed of several subsystems
connected in series. Each subsystem consists of machines of specific versions, and a detailed
maintenance plan is applied to optimize the system's performance and reliability while
minimizing costs.
System Topology
The multi-state system under study consists of four subsystems connected in series.
Each subsystem contains machines of different versions (Figure 2), represented by the
following vector:
 T = [[7], [6, 6], [5, 3], [1]]
Each subsystem contains one or more machines with specific versions. For instance,
subsystem 1 contains a machine of version 7, while subsystem 2 has two machines of version
6.

Figure 2: Example of a Multi-State System

Maintenance Plan
The maintenance plan applied to this system is represented by the following vector:
 PM = [[65, 1], [65, 1], [21, 1], [55, 1], [65, 1], [19, 1], [39, 1], [19, 1], [64, 1], [37, 1],
[65, 1], [64, 1], [38, 1], [38, 2], [21, 1], [65, 1], [20, 1], [55, 1], [64, 1], [19, 1], [21, 1],
[55, 1], [39, 1], [38, 2], [37, 2], [37, 1], [66, 1], [38, 1], [64, 1], [38, 1], [21, 1], [39, 2],
[37, 1], [56, 1], [37, 2], [19, 1], [64, 1], [39, 1], [38, 1], [20, 1], [55, 1], [20, 1], [37, 1],
[66, 1], [65, 1], [38, 1], [39, 1], [38, 1], [65, 1], [39, 1], [20, 1], [65, 1], [39, 1], [37, 1],
[21, 1], [55, 1], [57, 1], [56, 1], [38, 1], [65, 1]]
Each entry represents a maintenance action (first value) and the targeted machine in the
subsystem (second value). For example, [39, 2] means that maintenance action 39 is applied
to the second machine of version 6 in subsystem 2. This maintenance plan is designed to
reduce machine wear, prevent failures, and extend the system’s lifespan.
Reliability and Cost Calculation
The system's reliability, according to the UMGF method, after applying this
maintenance plan, was estimated at 0.94, exceeding the minimum reliability constraint set at
0.9. This ensures that the system can operate reliably over an extended period while
maintaining high performance.
The total cost of the system, including acquisition, maintenance, and energy costs, is
44,417.57 CAD. This approach effectively balances system reliability with overall costs,
ensuring joint optimization of both the system's topology and the maintenance plan.
This model highlights the importance of strategic resource management to achieve an
optimal trade-off between reliability and cost in a complex production system.
In optimizing our Multi-State System, it is crucial to adjust certain parameters to ensure
optimal performance of the meta-heuristic algorithms used, particularly the Nonlinear
Threshold Accepting (NLTA) algorithm. The key parameters to define include the variation
rate Δω, the initial threshold ω0=400, and the acceptance threshold ω=20.
To optimize the process, these parameters were systematically tested by varying their
values while keeping the others constant. This process was conducted through ten simulations
for each configuration, allowing for the collection of statistical data on the average
performance of the algorithms. In the case of the MSS, specific instances were selected for
parameter tuning. For example, for each pair of values corresponding to the number of
machines and the system's lifespan, the data from the first instance were used to adjust the
parameters. Once these values were established, they were applied to other variations of the
instances, thereby avoiding any risk of overfitting. All algorithms were implemented in
Python on a computer equipped with a 2.8 GHz processor.
Analysis of Optimization Models
In this study, we compare two optimization approaches applied to a multi-state
production system. These approaches, the joint optimization model and the separate
optimization model, aim to minimize the system's total cost by considering several factors,
including machine acquisition, maintenance, energy costs, and the system's reliability.
Reliability is a crucial constraint to ensure the system maintains adequate availability despite
potential failures. This comparison highlights the strengths and limitations of each approach
in optimizing both the system's performance and overall costs.
Joint Optimization Model
The joint optimization model is an approach that simultaneously addresses the three
main aspects of the system: machine acquisition costs, maintenance costs, and energy costs,
while ensuring that the required reliability threshold is met. By optimizing these three
dimensions together, we aim to find a comprehensive solution that minimizes costs while
ensuring reliable operational performance.
The joint approach allows for evaluating how each decision (such as machine selection
and maintenance planning) impacts the overall system, leading to more strategic decisions.
For example, choosing a machine with a slightly higher purchase cost but more energy
efficiency and long-term reliability can reduce total system costs.
For this joint optimization, we used the NLTA algorithm. This algorithm was executed
200,000 times and repeated 10 times to ensure the robustness and accuracy of the results
obtained. Each run of the algorithm produces a configuration of machines and a maintenance
plan that respects the reliability constraints while minimizing costs.
The NLTA algorithm allowed us to:
 Determine the number of machines to be installed in parallel in each subsystem.
 Select the version of each machine from the available options on the market,
considering their cost, reliability, and energy consumption.
 Optimize the preventive maintenance plan to reduce downtime due to failures while
maintaining reliability above the required threshold.
Results:
The table 3 represents the 10 executions of the NTLL algorithm for the joint optimization model.
Table 3: Ten Executions of the NTLL Algorithm for the Joint Optimization Model
Execution Topology Maintenance Plan Total Cost
Number
1 [[7, 7], [4, 6], [3, 3], [1, 3]] [[19, 2], [19, 1], [49, 1], [50, 2], 61755.422
[33, 1], [19, 1], [72, 1], [33, 1], [49,
1], [19, 2], [19, 1], [50, 2], [72, 1],
[50, 1], [72, 1], [50, 2], [19, 2], [33,
1], [19, 1], [33, 1], [50, 1], [33, 1],
[19, 2], [72, 1], [50, 2], [33, 1], [72,
1]]

2 [[7, 7], [4, 6], [3, 3], [1, 3]] [[19, 1], [50, 2], [19, 2], [51, 1], 61754.671
[72, 1], [50, 2], [19, 2], [50, 1], [19,
1], [33, 1], [50, 1], [50, 2], [33, 1],
[19, 1], [72, 1], [50, 2], [19, 2], [72,
1], [50, 1], [19, 2], [33, 1], [50, 2],
[72, 1], [19, 1], [33, 1], [72, 1], [50,
1], [33, 1], [72, 1]]
3 [[7, 7], [6, 4], [3, 3], [3, 3]] [[72, 1], [50, 1], [33, 1], [33, 1], 61751.655
[72, 1], [51, 1], [38, 1], [50, 2], [72,
2], [19, 1], [33, 1], [19, 1], [50, 1],
[19, 1], [72, 1], [33, 1], [19, 2], [51,
1], [72, 1], [49, 1], [33, 1], [72, 1],
[72, 2], [19, 1], [51, 1], [33, 1], [51,
1]]
4 [[7, 7], [4, 6], [3, 3], [1, 3]] [[50, 2], [72, 1], [19, 1], [50, 1], 61752.225
[19, 2], [33, 1], [51, 1], [50, 2], [72,
1], [19, 1], [50, 1], [72, 1], [19, 1],
[51, 1], [50, 2], [19, 2], [33, 1], [50,
2], [19, 2], [33, 1], [50, 1], [72, 1],
[19, 1], [51, 2], [33, 1], [50, 1], [72,
1], [72, 1], [33, 1]]
5 [[5], [4, 6], [3, 3], [1, 3]] [[64, 1], [72, 1], [72, 1], [72, 1], 51202.636
[72, 1], [72, 1], [72, 1], [72, 1], [72,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [72, 1], [72, 1], [72, 1], [72,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [72, 1], [72, 1], [72, 1], [72,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [72, 1], [72, 1], [72, 1], [72,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [72, 1], [72, 1], [72, 1], [64,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [72, 1], [72, 1], [72, 1], [72,
1], [72, 1], [72, 1], [72, 1], [72, 1],
[72, 1], [33, 1]]

6 [[7, 7], [4, 6], [3, 3], [1, 3]] [[72, 1], [19, 2], [51, 1], [50, 2], 61751.706
[19, 2], [19, 1], [50, 1], [33, 1], [50,
2], [72, 1], [50, 1], [19, 1], [33, 1],
 [19, 2], [51, 1], [50, 2], [72, 1], [33,
1], [33, 1], [50, 1], [19, 1], [50, 2],
[50, 1], [19, 2], [72, 1], [33, 1], [72,
1], [33, 1]]
7 [[7, 7], [4, 6], [3, 3], [1, 3]] [[50, 1], [19, 2], [33, 1], [50, 2], 61751.837
[19, 1], [72, 1], [50, 1], [19, 2], [50,
2], [19, 1], [72, 1], [33, 1], [33, 1],
[50, 1], [38, 1], [50, 1], [19, 1], [50,
2], [72, 1], [19, 1], [51, 2], [19, 2],
[50, 1], [72, 1], [33, 1], [72, 1]]
8 [[7, 7], [4, 6], [3, 3], [1, 3]] [[50, 1], [50, 2], [72, 1], [19, 1], 61751.119
[19, 2], [33, 1], [19, 2], [50, 1], [19,
1], [50, 2], [33, 1], [50, 1], [72, 1],
[72, 1], [19, 2], [49, 1], [50, 2], [19,
1
], [33, 1], [50, 1], [72, 1], [19, 1],
[50, 2], [33, 1], [19, 2], [72, 1], [72,
1], [33, 1]]
9 [[7, 7], [4], [3], [1, 3]] [[72, 1], [51, 1], [64, 1], [64, 1], 41461.150
[64, 1], [72, 1], [64, 1], [72, 1], [64,
1], [64, 1], [72, 1], [64, 1], [64, 1],
[72, 1], [64, 1], [72, 1], [72, 1], [64,
1], [64, 1], [64, 1], [64, 1], [64, 1],
[64, 1], [72, 1], [64, 1], [72, 1], [64,
1], [64, 1], [72, 1], [64, 1], [72, 1],
[64, 1], [64, 1], [64, 1], [64, 1], [64,
1], [64, 1], [72, 1], [71, 1], [64, 1],
[64, 1], [64, 1], [64, 1], [72, 1], [64,
1], [64, 1], [64, 1], [72, 1], [64, 1],
[64, 1], [72, 1], [64, 1], [72, 1], [64,
1], [64, 1], [72, 1], [72, 1], [64, 1],
[64, 1], [19, 1]]
10 [[7], [6, 4], [3], [3, 1]] [[38, 1], [51, 1], [75, 1], [19, 2], [38, 41848.561
1], [50, 2], [38, 1], [75, 1], [19, 1], [50,
1], [19, 2], [38, 1], [75, 1], [50, 2], [38,
1], [75, 1], [19, 1], [51, 2], [38, 1], [51,
1], [75, 1], [38, 1], [19, 2], [75, 1], [51,
1], [51, 2], [38, 1]]

 The best solution among the 10 obtained executions:

 Topology: [[7, 7], [4], [3], [1, 3]]
 Maintenance plan: [[72, 1], [51, 1], [64, 1], [64, 1], [64, 1], [72, 1], [64, 1],
[72, 1], [64, 1], [64, 1], [72, 1], [64, 1], [64, 1], [72, 1], [64, 1], [72, 1], [72, 1],
[64, 1], [64, 1], [64, 1], [64, 1], [64, 1], [64, 1], [72, 1], [64, 1], [72, 1], [64, 1],
[64, 1], [72, 1], [64, 1], [72, 1], [64, 1], [64, 1], [64, 1], [64, 1], [64, 1], [64, 1],
[72, 1], [71, 1], [64, 1], [64, 1], [64, 1], [64, 1], [72, 1], [64, 1], [64, 1], [64, 1],
[72, 1], [64, 1], [64, 1], [72, 1], [64, 1], [72, 1], [64, 1], [64, 1], [72, 1], [72, 1],
[64, 1], [64, 1], [19, 1]]
 Total cost: 41461.150
Separate Optimization Model
 The separate optimization model differs from the joint approach by dividing the
optimization into three distinct phases: optimizing machine acquisition costs, followed by
optimizing the maintenance plan, and then adding energy costs. This approach allows each
aspect of the system to be handled independently, but it can lead to suboptimal total costs, as
decisions made in one phase may limit choices in subsequent phases.
 Phase 1: Redundancy Optimization
In this phase, we focus exclusively on determining the number of machines and their
versions to be installed in parallel. The NLTA algorithm is again used to explore different
machine configurations, optimizing only for acquisition costs. The goal is to minimize the
purchase cost of the machines while ensuring a certain level of redundancy to guarantee
system reliability.
The algorithm explores various combinations of:
 Number of machines installed in parallel in each subsystem.
 Machine versions available on the market, each with different costs, energy
consumption, and reliability levels.
Results:
Table 4 presents the results of ten executions of the NTLL algorithm for the separate optimization
model, concentrating on acquisition costs only. It highlights the different system topologies,
maintenance plans, and the corresponding total costs, providing insights into the impact of acquisition
costs on system performance.
Table 4: Ten Executions of the NTLL Algorithm for the Separate Optimization Model
Based on Acquisition Cost Only
Execution Topology Maintenance Plan Total Cost
Number
1 [[7, 7], [6, 6], [3, 3], [1, 1]] [[49, 1], [65, 2], [37, 1], [49, 1], [39, 1],
[39, 1], [65, 2], [21, 1], [20, 1], [21, 1], 40600
[39, 1], [21, 1], [37, 2], [64, 2], [38, 2],
[37, 1], [39, 1], [38, 2], [51, 1], [20, 2],
[65, 2], [49, 1], [21, 1], [21, 1], [51, 1],
[37, 2], [66, 2], [39, 1], [39, 1], [37, 2],
[38, 1], [65, 2], [21, 2], [64, 2], [37, 2],
[38, 2], [38, 2], [39, 1], [50, 1], [49, 1],
[39, 2], [19, 2], [38, 1], [64, 2], [20, 1],
[64, 2], [51, 1], [38, 1], [20, 2], [38, 1],
[37, 1], [39, 2], [38, 2], [64, 2], [64, 2],
[65, 1], [39, 2], [51, 1]]
2 [[7, 7], [6, 6], [3, 3], [1]] [[50, 1], [20, 2], [51, 1], [20, 1], [19, 2], 35400
[21, 1], [50, 1], [51, 1], [49, 2], [51, 1],
[50, 1], [38, 1], [50, 1], [50, 1], [21, 1],
[19, 1], [49, 1], [21, 2], [39, 1], [20, 2],
[51, 1], [51, 2], [51, 1], [39, 1], [21, 2],
[19, 2], [66, 1], [20, 1], [51, 1], [50, 1],
[21, 2], [19, 1], [39, 1], [19, 1], [39, 1],
[51, 1], [49, 1], [51, 1], [38, 1], [21, 1],
[50, 1], [19, 2], [49, 1], [38, 1], [19, 2],
[37, 1], [37, 1], [38, 1], [19, 1], [49, 1],
[21, 2], [20, 2], [20, 2], [49, 1], [21, 1],
[20, 1], [37, 1], [19, 1], [51, 1], [65, 1]]
3 [[7, 7], [6], [3, 3], [1, 1]] [[66, 1], [51, 1], [51, 1], [19, 1], [49, 1], 35500
[51, 2], [51, 1], [39, 1], [49, 2], [37, 1],
[19, 1], [51, 1], [20, 1], [19, 1], [65, 1],
[64, 2], [51, 2], [39, 1], [66, 2], [50, 2],
[19, 1], [20, 1], [66, 1], [21, 1], [39, 1],
 [51, 2], [51, 1], [21, 1], [64, 1], [51, 2],
[19, 1], [50, 2], [65, 2], [50, 1], [20, 1],
[51, 2], [19, 1], [66, 2], [51, 1], [65, 1],
[64, 2], [21, 1], [21, 1], [66, 2], [66, 1],
[65, 1], [50, 1], [19, 1], [38, 1], [66, 2],
[51, 2], [20, 1], [21, 1], [65, 1], [38, 1],
[20, 1], [64, 2], [64, 2], [21, 1], [50, 2]]
4 [[7, 7], [6, 6], [3, 3], [1, 1]] [[37, 2], [37, 2], [20, 1], [21, 1], [19, 1], 40600
[38, 2], [38, 1], [39, 2], [49, 2], [50, 2],
[66, 1], [19, 1], [49, 2], [49, 2], [19, 2],
[21, 2], [37, 2], [49, 2], [50, 2], [66, 1],
[20, 1], [38, 1], [21, 2], [21, 1], [39, 2],
[49, 1], [39, 2], [39, 2], [66, 1], [39, 2],
[19, 1], [65, 1], [38, 2], [20, 1], [21, 1],
[66, 1], [37, 2], [50, 2], [49, 2], [21, 1],
[49, 2], [39, 1], [65, 1], [20, 1], [37, 2],
[50, 2], [19, 2], [66, 1], [37, 1], [39, 2],
[51, 2], [37, 1], [50, 2], [50, 2], [49, 2],
[50, 2], [66, 1], [20, 1], [65, 1], [66, 1]]
5 [[7, 7], [6, 7], [3, 3], [1, 1]] [[19, 1], [19, 2], [49, 1], [65, 1], [49, 1], 40600
[41, 1], [21, 2], [39, 1], [66, 1], [39, 1],
[38, 1], [66, 1], [65, 2], [39, 1], [20, 2],
[50, 1], [66, 1], [65, 1], [64, 1], [20, 1],
[19, 1], [39, 1], [65, 2], [64, 1], [65, 2],
[20, 2], [39, 1], [64, 1], [49, 1], [20, 1],
[38, 1], [64, 1], [19, 2], [49, 1], [65, 2],
[64, 2], [66, 2], [50, 1], [20, 2], [19, 1],
[66, 1], [21, 2], [39, 1], [66, 1], [49, 1],
[49, 1], [65, 1], [41, 1]]
6 [[7, 7], [6, 6], [3, 3], [1, 1]] [[21, 1], [19, 1], [66, 2], [20, 2], [20, 2], 40600
[64, 1], [49, 2], [38, 1], [51, 2], [21, 1],
[20, 1], [38, 1], [20, 1], [65, 1], [51, 1],
[50, 2], [66, 1], [49, 1], [19, 1], [19, 2],
[64, 1], [66, 2], [21, 1], [21, 1], [51, 1],
[39, 1], [21, 1], [19, 2], [50, 1], [38, 1],
[19, 1], [20, 2], [20, 2], [66, 2], [19, 1],
[49, 2], [20, 1], [19, 1], [66, 2], [49, 1],
[21, 1], [49, 1], [39, 1], [51, 1], [20, 1],
[21, 1], [19, 1], [49, 1], [39, 1], [65, 2],
[37, 1], [50, 1], [20, 1], [65, 2], [39, 1],
[19, 1], [66, 1], [51, 2], [64, 1], [20, 1]]
7 [[7], [6, 7], [3, 3], [1]] [[42, 1], [40, 1], [40, 1], [40, 1], [50, 1], 30400
[49, 1], [40, 1], [50, 1], [51, 1], [65, 1],
[40, 1], [65, 1], [40, 1], [66, 1], [51, 1],
[64, 1], [65, 1], [49, 1], [65, 1], [40, 1],
[49, 1], [51, 1], [40, 1], [40, 1], [42, 1],
[49, 1], [40, 1], [41, 1], [49, 1], [50, 2],
[40, 1], [66, 1], [49, 1], [66, 1], [51, 1],
[40, 1], [50, 2], [51, 1], [40, 1], [42, 1],
[41, 1], [41, 1], [64, 1], [50, 1], [64, 1],
[51, 2], [41, 1], [42, 1], [50, 1], [66, 1],
[51, 2], [41, 1], [65, 1], [41, 1], [42, 1],
[64, 1], [64, 1], [50, 2], [49, 2], [21, 1]]
8 [[7, 7], [6, 7], [3, 3], [1, 1]] [[41, 1], [50, 2], [20, 2], [51, 2], [41, 1], 40600
[39, 1], [19, 2], [65, 1], [20, 2], [19, 1],
 [40, 1], [51, 2], [20, 2], [21, 2], [66, 1],
[66, 1], [51, 2], [41, 1], [65, 1], [49, 2],
[40, 1], [51, 2], [39, 1], [41, 1], [42, 1],
[51, 2], [21, 2], [50, 2], [65, 1], [39, 1],
[40, 1], [37, 1], [38, 1], [65, 1], [49, 2],
[64, 1], [19, 2], [40, 1], [51, 1], [51, 2],
[50, 2], [50, 2], [40, 1], [38, 1], [41, 1],
[50, 2], [41, 1], [50, 2], [20, 2], [21, 2],
[49, 2], [21, 2], [41, 1], [49, 2], [40, 1],
[19, 2], [42, 1], [51, 1], [37, 1], [66, 1]]
9 [[7, 7], [6, 6], [3, 3], [1, 1]] [[37, 1], [20, 1], [39, 1], [39, 2], [21, 2], 40600
[50, 1], [38, 1], [39, 2], [51, 1], [49, 2],
[21, 1], [65, 1], [64, 1], [66, 1], [66, 1],
[50, 1], [64, 1], [37, 1], [65, 1], [64, 1],
[21, 1], [66, 1], [50, 1], [65, 1], [21, 1],
[20, 1], [38, 1], [39, 2], [21, 1], [38, 1],
[19, 1], [65, 1], [50, 1], [21, 1], [65, 1],
[19, 1], [38, 2], [49, 1], [51, 2], [38, 1],
[21, 1], [65, 1], [20, 1], [21, 1], [39, 1],
[20, 1], [51, 1], [66, 1], [49, 1], [39, 2],
[21, 1], [49, 1], [65, 1], [19, 2], [19, 1],
[65, 1], [64, 1], [64, 1], [66, 1], [38, 1]]
10 [[7], [6, 7], [3, 3], [1]] [[42, 1], [40, 1], [40, 1], [40, 1], [50, 1], 30400
[49, 1], [40, 1], [50, 1], [51, 1], [65, 1],
[40, 1], [65, 1], [40, 1], [66, 1], [51, 1],
[64, 1], [65, 1], [49, 1], [65, 1], [40, 1],
[49, 1], [51, 1], [40, 1], [40, 1], [42, 1],
[49, 1], [40, 1], [41, 1], [49, 1], [50, 2],
[40, 1], [66, 1], [49, 1], [66, 1], [51, 1],
[40, 1], [50, 2], [51, 1], [40, 1], [42, 1],
[41, 1], [41, 1], [64, 1], [50, 1], [64, 1],
[51, 2], [41, 1], [42, 1], [50, 1], [66, 1],
[51, 2], [41, 1], [65, 1], [41, 1], [42, 1],
[64, 1], [64, 1], [50, 2], [49, 2], [21, 1]]
 The best solution among the 10 obtained executions:

 Topology: [[7], [6, 7], [3, 3], [1]]

 Maintenance Plan: [[42, 1], [40, 1], [40, 1], [40, 1], [50, 1], [49, 1], [40, 1],
[50, 1], [51, 1], [65, 1], [40, 1], [65, 1], [40, 1], [66, 1], [51, 1], [64, 1], [65, 1],
[49, 1], [65, 1], [40, 1], [49, 1], [51, 1], [40, 1], [40, 1], [42, 1], [49, 1], [40, 1],
[41, 1], [49, 1], [50, 2], [40, 1], [66, 1], [49, 1], [66, 1], [51, 1], [40, 1], [50, 2],
[51, 1], [40, 1], [42, 1], [41, 1], [41, 1], [64, 1], [50, 1], [64, 1], [51, 2], [41, 1],
[42, 1], [50, 1], [66, 1], [51, 2], [41, 1], [65, 1], [41, 1], [42, 1], [64, 1], [64, 1],
[50, 2], [49, 2], [21, 1]]
 Total cost: 30,400
 Phase 2: Fixing the Topology and Maintenance Plan Optimization
Once the number of machines and their versions are determined, the system's topology
is fixed. At this stage, we optimize the maintenance plan. Since the system's configuration is
now fixed, it is no longer possible to modify the number or versions of the machines. The
goal of this phase is to maximize the system's lifetime by minimizing failures through an
appropriate preventive maintenance plan.
The maintenance plan is tailored to the machines already selected and aims to:
 Reduce the risk of failures while maintaining reasonable maintenance costs.
  Ensure that the system's reliability continues to meet the required threshold throughout
its operational life.
However, fixing the topology at this stage may limit the system's flexibility in the long
term. If the system's needs change or operational conditions evolve, it will be difficult to
adapt maintenance decisions without reinvesting in new machines.
Results:
Total cost = acquisition cost + maintenance cost = 30,400 + 562.575 = 30,962.575
 Phase 3: Adding Energy Costs
The third and final phase involves calculating energy costs based on the system's final
configuration. The energy cost is based on the energy consumption of the machines selected
during the first phase. At this stage, it is no longer possible to change the machine
configuration, and energy costs are simply added to the acquisition and maintenance costs to
obtain the total cost.
This approach may lead to higher-than-expected energy costs because the machine
selection did not take energy criteria into account from the beginning, unlike the joint
optimization model.
Results:
Total cost = acquisition cost + energy cost + maintenance cost = 30,400 + 14,446.986 + 562.575 =
45,409.561
Comparison of Models
Results:
Table 5 presents the various costs associated with both the joint and separate optimization
models. It includes the acquisition cost, energy cost, maintenance cost, and the total cost for
each model, allowing for a comparative analysis of the financial implications of each
optimization approach.
Table 5: Costs Associated with Joint and Separate Optimization Models
Acquisition energy cost maintenance cost Total cost
Cost
Joint 30700 10563.523 197.627 41461.150
Optimization
Separate 30400 14446.986 562.575 45409.561
Optimization

Table 6 presents the topology and maintenance plan for each optimization model, showcasing
the configurations of the multi-state systems and the maintenance actions for both the joint
and separate approaches.
Table 6: Topology and Maintenance Plan for Each Model
Topology Maintenance Plan Total cost
Joint [[7, 7], [4], [3], [1, 3]] [[72, 1], [51, 1], [64, 1], [64, 1], [64, 41461.150
Optimization 1], [72, 1], [64, 1], [72, 1], [64, 1], [64,
1], [72, 1], [64, 1], [64, 1], [72, 1], [64,
1], [72, 1], [72, 1], [64, 1], [64, 1], [64,
1], [64, 1], [64, 1], [64, 1], [72, 1], [64,
1], [72, 1], [64, 1], [64, 1], [72, 1], [64,
1], [72, 1], [64, 1], [64, 1], [64, 1], [64,
1], [64, 1], [64, 1], [72, 1], [71, 1], [64,
1], [64, 1], [64, 1], [64, 1], [72, 1], [64,
1], [64, 1], [64, 1], [72, 1], [64, 1], [64,
1], [72, 1], [64, 1], [72, 1], [64, 1], [64,
1], [72, 1], [72, 1], [64, 1], [64, 1], [19,
 1]]
Separate [[7], [6, 7], [3, 3], [1]] [[42, 1], [40, 1], [40, 1], [40, 1], [50, 45409.561
Optimization 1], [49, 1], [40, 1], [50, 1], [51, 1], [65,
1], [40, 1], [65, 1], [40, 1], [66, 1], [51,
1], [64, 1], [65, 1], [49, 1], [65, 1], [40,
1], [49, 1], [51, 1], [40, 1], [40, 1], [42,
1], [49, 1], [40, 1], [41, 1], [49, 1], [50,
2], [40, 1], [66, 1], [49, 1], [66, 1], [51,
1], [40, 1], [50, 2], [51, 1], [40, 1], [42,
1], [41, 1], [41, 1], [64, 1], [50, 1], [64,
1], [51, 2], [41, 1], [42, 1], [50, 1], [66,
1], [51, 2], [41, 1], [65, 1], [41, 1], [42,
1], [64, 1], [64, 1], [50, 2], [49, 2], [21,
1]]

 CT (Joint Optimization) < CT (Separate Optimization)

When comparing the results of both approaches, several key points emerge:
 Joint Model: This approach offers more comprehensive optimization by
simultaneously considering acquisition, maintenance, and energy. This allows
for achieving lower total costs and better reliability since each decision is made
considering its impact on the entire system.
 Separate Model: While this approach helps minimize short-term acquisition
costs, it tends to be suboptimal when maintenance and energy costs are added
later. The lack of a holistic view limits the ability to find effective trade-offs
between these different cost factors.
The analysis shows that the joint optimization model is more advantageous when it comes to
minimizing total costs while respecting the reliability constraint. By integrating all aspects of
the system (acquisition, maintenance, energy) from the beginning, this approach allows for
more strategic and effective decision-making. On the other hand, the separate optimization,
while easier to implement, often leads to less efficient solutions in the long term due to its
fragmented approach.
[Link]
This article presents an innovative approach to the joint optimization of multi-state
systems by integrating redundancy, maintenance, and energy costs into a single global
optimization process. Unlike traditional methods where redundancy and maintenance are
addressed separately, our integrated approach enables more informed and strategic decision-
making, maximizing system performance while minimizing total cost. This joint optimization
is particularly relevant for MSS, which are characterized by multiple performance levels and
increased complexity due to transitions between operational states.
One of the key contributions of this article is the application and adaptation of the
NLTA algorithm to multi-state systems. This algorithm, used to explore and optimize
different system configurations, has proven effective in terms of robustness and accuracy. By
running multiple iterations of the algorithm, we were able to identify the optimal combination
of redundancy and maintenance plans that minimizes overall costs while ensuring the required
reliability level for the MSS under study. Integrating this method into the MSS framework has
resulted in a significant improvement in managing system reliability and overall performance.
Additionally, the joint optimization of MSS has allowed us to incorporate critical
aspects such as energy costs, which are often overlooked in traditional approaches. Including
this parameter has strengthened the relevance of our approach for industrial applications,
where reducing energy costs is increasingly becoming a priority. By accounting for various
costs simultaneously, we demonstrated that our model not only enhances the long-term
performance of the system but also optimizes its operational expenses.
In terms of future perspectives, several avenues for further research and improvement
can be explored. First, it would be interesting to investigate the application of the NLTA
algorithm to other types of complex systems, particularly in fields such as critical
infrastructure or smart power grids. Additionally, algorithmic improvements could be
considered, such as combining NLTA with other optimization techniques like evolutionary
algorithms or neural networks to allow for broader and faster exploration of solution spaces.
On another front, integrating smart sensors and real-time data could enhance the
effectiveness of the optimization process by enabling continuous adaptation of maintenance
and redundancy plans in response to evolving operational conditions in MSS. This could pave
the way for dynamic optimization, allowing systems to proactively adjust to failures or
performance drops before they occur.
Moreover, studying the effects of uncertainty and random degradation in the model
could also be a future research direction. This would make the model more robust against
unexpected changes and variations in operating conditions. Given the sensitivity of MSS to
such changes, such an extension would provide a more flexible and realistic approach suited
to industrial realities.
In conclusion, the joint optimization approach proposed in this article represents a
significant advancement in the management and optimization of multi-state systems. By
integrating redundancy, maintenance, and energy considerations, this methodology provides a
comprehensive solution for reducing costs and improving MSS reliability while remaining
adaptable to changing requirements and operational constraints. Future work should focus on
enhancing the algorithm and adapting this framework to a broader range of systems to
maximize its impact in the industrial sector.

Bibliography

[1] I. Ushakov, "A universal generating function," Soviet Journal Computer Systems Science, p. 37–
49., 1986.
[2] G. Levitin, "Multistate series–parallel system expansion scheduling subject to availability
constraints," IEEE Transactions on Reliability, vol. 49, no. 1, p. 71–79, 2000.

[3] Y. Ding and . A. Lisnianski, "Fuzzy universal generating functions for multi-state system
reliability assessment," Fuzzy Sets and Systems, vol. 159, no. 3, pp. 307-324, 2008.

[4] . B. Jafary and L. ,Fiondella , "A universal generating function-based multi-state system
performance model subject to correlated failures," Reliability Engineering & System Safety, vol.
152, pp. 16-27, 2016.

[5] S. Qiu and X. Ming, "A fuzzy universal generating function-based method for the reliability
evaluation of series systems with performance sharing between adjacent units under parametric
uncertainty," Fuzzy Sets and Systems, vol. 424, pp. 155-169, 2021.

[6] M. Babaei and A. Rashidi-baqhi, "Universal generating function -based narrow reliability bounds
to evaluate reliability of project completion time," Reliability Engineering & System Safety, vol.
240, 2022.

[7] F. S. Campos , F. A. Assis , d. S. A. M. Leite , J. C. Alex , A. . M. Rodolfo and A. O. S. Marco ,

"Reliability evaluation of composite generation and transmission systems via binary logistic
regression and parallel processing," International Journal of Electrical Power & Energy
Systems, vol. 142, 2022.

[8] w. Xia, Y. Wang, Y. Hao, Z. He, K. Yan and F. Zhao, "Reliability analysis for complex
electromechanical multi-state systems utilizing universal generating function techniques,"
Reliability Engineering & System Safety, vol. 244, 2024.

[9] E. Zio , M. Marella and L. Podofillini , "A Monte Carlo simulation approach to the availability
assessment of multi-state systems with operational dependencies," Reliability Engineering &
System Safety, vol. 92, no. 7, pp. 871-882, 2007.

[10] C. Wang, L. Xing and G. Levitin, "Reliability analysis of multi-trigger binary systems subject to
competing failures," Reliability Engineering & System Safety, vol. 111, pp. 9-17, 2013.

[11] F. Chiacchio , D. D’Urso , G. Manno and L. Compagno , "Reliability analysis of cold-standby

phased-mission system based on GO-FLOW methodology and the universal generating
function," Reliability Engineering & System Safety, vol. 223, pp. 1-13, 2016.

[12] C. Nop , R. M. Fadhil and . K. Unami, "A multi-state Markov chain model for rainfall to be used
in optimal operation of rainwater harvesting systems," Journal of Cleaner Production, vol. 285,
p. 124912, 2021.

[13] R.-Z. Wang, H.-H. Gu, Y. Liu, H. Miura, X.-C. Zhang and S.-T. Tu, "Surrogate-modeling-
assisted creep-fatigue reliability assessment in a low-pressure turbine disc considering multi-
source uncertainty," Reliability Engineering & System Safety, vol. 240, 2023.

[14] T. Zhou, X. Zhang, E. L. Droguett and A. Mosleh, "A generic physicsinformed neural network-
based framework for reliability assessment of multi-state systems," Reliability Engineering &
System Safety, vol. 229, 2023.

[15] Y. Bo , M. Bao , Y. Ding and Y. Hu , "A DNN-based reliability evaluation method for multi-
state series-parallel systems considering semi-Markov process," 2024.

[16] F. Cheng and H. Liu, "An adaptive hybrid deep learning-based reliability assessment framework
for damping track system considering multi-random variables," Mechanical Systems and Signal
Processing, vol. 208, 2024.

[17] H. Che, "Reliability assessment of multi-state weighted k-out-of-n man-machine systems

 considering dependent machine deterioration and human fatigue," Reliability Engineering &
Systems Safety, p. 246, 2024.

[18] C. Su, K. Huang and Z. Wen, "Multi-objective imperfect selective maintenance optimization for
series-parallel systems with stochastic mission duration," in Proceedings of the Institution of
Mechanical Engineers Part O Journal of Risk and Reliability, 2022.

[19] P. Zhang , L. Chaozhe , . X. Huanyun, . Z. , Yongjiu, W. Kai , Z. Yuewen and S. Peiting , "Bi-
AAE: A binary adversarial autoencoder deep neural network model for anomaly detection in
system-levels marine diesel engines," Ocean Engineering, vol. 302, 2024.

[20] M.-C. Fitouhi and N. Mustapha , "Integrating noncyclical preventive maintenance scheduling
and production planning for multi-state systems," Reliability Engineering & System Safety, vol.
121, pp. 175-186, 2014.

[21] G. Levitin and A. Lisnianski, "Optimization of imperfect preventive maintenance for multi-state
systems," Reliability Engineering and System Safety, vol. 67, p. 193–203, 2000.

[22] W. Dong , L. Sifeng , J. B. Suk and L. Yu , "A multi-stage imperfect maintenance strategy for
multi-state systems with variable user demands," Computers & Industrial Engineering, vol. 145,
2020.

[23] A. F. Shahraki , O. P. Yadav and . C. Vogiatzis, "Selective maintenance optimization for multi-
state systems considering stochastically dependent components and stochastic imperfect
maintenance actions," Reliability Engineering & System Safety, vol. 196, p. 106738, 2020.

[24] Y. Chen, Y. Liub and T. Jiang, "Optimal maintenance strategy for multi-state systems with single
maintenance capacity and arbitrarily distributed maintenance time," Reliability Engineering &
System Safety, vol. 211, p. 107576, 2021.

[25] X. Han, Z. Wang, M. Xie, Y. He, Y. Li and W. Wang, "Remaining useful life prediction and
predictive maintenance strategies for multi-state manufacturing systems considering functional
dependence," Reliability Engineering & System Safety, vol. 210, 2021.

[26] H. Wang , Y. Qi and . Z. Shuzhu, "Integrated scheduling and flexible maintenance in

deteriorating multi-state single machine system using a reinforcement learning approach,"
Advanced Engineering Informatics, vol. 49, 2021.

[27] V.-T. Nguyen , D. Phuc , V. Alexandre and I. Benoit , "Artificial-intelligence-based maintenance

decision-making and optimization for multi-state component systems," Reliability Engineering
& System Safety, vol. 228, 2022.

[28] Y. Cao, J. Luo and W. Dong, "Optimization of condition-based maintenance for multi-state
deterioration systems under random shock," Reliability Engineering & System Safety, vol. 115,
pp. 80-99, 2023.

[29] V. P. Koutras , "Markov Regenerative Process Model for the Dependability and Performance of
a Two-Unit Multi-State System under Maintenance," Reliability Engineering & System Safety,
vol. 238, 2023.

[30] H. Dui, Y. Lu, Z. Gao and L. Xing, "Performance efficiency and cost analysis of multi-state
systems with successive damage and maintenance in multiple shock events," Reliability
Engineering & System Safety, vol. 238, p. 109403, 2023.

[31] H. Khorshidi, I. Gunawan and M. Ibrahim, "A value-driven approach for optimizing reliability-
redundancy allocation problem in multi-state weighted k-out-of-n system," Khorshidi, H.,
Gunawan, I., & Ibrahim, M. (2016). A value-driven approach for optimizing reliability-
redundancy allocation proJournal of Manufacturing Systems, vol. 40, pp. 54-62, 2016.
[32] . C.-T. Yeh and L. Fiondella, "Optimal redundancy allocation to maximize multi-state computer
network reliability subject to correlated failures," Reliability Engineering & System Safety, vol.
166, pp. 138-150, 2017.

[33] A. Kumar, V. Kumar and V. Modgil, "Behavioral study and availability optimization of a multi-
state repairable system with hot redundancy," Kumar, A., Kumar, V., & Modgil, V. (2019).
Behavioral study and availability optimization of a multi-state repairable systemInternational
Journal of Quality & Reliability Management, vol. 36, no. 3, pp. 314-330, 2019.

[34] M.-X. Sun, Y.-F. Li and E. Zio, "On the optimal redundancy allocation for multi-state series–
parallel systems under epistemic uncertainty," Reliability Engineering & System Safety, vol. 192,
2019.

[35] Y. Hu, Y. Ding and Z. Zeng, "Redundancy optimization for multi-state series-parallel systems
using ordinal optimization-based-genetic algorithm," in Hu, Y., Ding, Y., & Zeng, Z. (2021).
Redundancy optimization for multi-state series-parallel systems using ordinal opProceedings of
the Institution of Mechanical Engineers Part O Journal of Risk and Reliability, 236(1), 2021.

[36] J. Li, G. Wu, H. Zhou and H. Chen, "Redundancy allocation optimization for multi-state system
with hierarchical performance requirements," Li, J., Wu, G., Zhou, H., & Chen, H. (2022).
Redundancy allocation optimization for multi-state system with hierarcJournal of Risk and
Reliability, Vols. Li, J., Wu, G., Zhou, H., & Chen, H. (2022). Redundancy allocation
optimization for multi-state system with hierarchical perf237, no. 6, pp. 1031-1047, 2022.

[37] X. Zhao, C. Wang and S. Wang, "Reliability analysis of multi-state balanced systems with
standby components switching mechanism," Reliability Engineering & Systems Safety, vol. 242,
p. 109774, 2024.

[38] G. Levitin and A. Lisnianski, "Joint redundancy and maintenance optimization for multistate
series–parallel systems," Reliability Engineering & System Safety, vol. 64, no. 1, pp. 33-42, 1999.

[39] M. Nourelfath, E. Châtelet and N. Nahas, "Joint redundancy and imperfect preventive
maintenance optimization for series–parallel multi-state degraded systems," Reliability
Engineering and System Safety, vol. 103, p. 51, 2012.

[40] Y. Liu, H.-Z. Huang, Z. Wang, Y. Li and Y.-J. Yang, "A Joint Redundancy and Imperfect
Maintenance Strategy Optimization for Multi-State Systems," Reliability, IEEE Transactions on,
vol. 62, pp. 368-378, 2013.