Decision Analytics Journal 11 (2024) 100485

Contents lists available at ScienceDirect

Decision Analytics Journal

journal homepage: www.elsevier.com/locate/dajour

A hybrid simheuristic algorithm for solving bi-objective stochastic flexible

job shop scheduling problems
Saman Nessari a , Reza Tavakkoli-Moghaddam a,b ,∗, Hessam Bakhshi-Khaniki a , Ali Bozorgi-Amiri a
a School of Industrial Engineering, College of Engineering, University of Tehran, Tehran, Iran
b
Research Center of Performance and Productivity Analysis, Istinye University, Istanbul, Turkey

ARTICLE INFO ABSTRACT

Dataset link: https://github.com/SamanNsr/Hy The flexible job shop scheduling problem (FJSSP) is a complex optimization challenge that plays a crucial
brid-Simheuristic-BiObj-Stoch-FJSSP role in enhancing productivity and efficiency in modern manufacturing systems, aimed at optimizing the
allocation of jobs to a variable set of machines. This paper introduces an algorithm to tackle the FJSSP by
Keywords:
Flexible job shop scheduling
minimizing makespan and total weighted earliness and tardiness under uncertainty. This hybrid algorithm
Multi-objective stochastic optimization effectively addresses the complexities of stochastic multi-objective optimization by integrating the equilibrium
Simheuristics optimizer (EO) as an initial solutions generator, Non-dominated sorting genetic algorithm II (NSGA-II),
Equilibrium optimizer and simulation techniques. The algorithm’s effectiveness is validated by showcasing specific instances and
Non-dominated sorting genetic algorithm delivering decision results for optimal scheduling across varying levels of uncertainty. Results reveal the
algorithm’s consistent superiority in managing the complexities of stochastic parameters across various problem
scales, achieving lower makespan and improved Pareto front quality compared to existing methods. Particularly
notable is the algorithm’s faster convergence and robust performance, as validated by the statistical Wilcoxon
test, which confirms its reliability and efficacy in handling dynamic scheduling situations. These findings
underscore the algorithm’s potential in providing flexible, robust solutions. The proposed algorithm’s unique
balance of exploitative and explorative capabilities within a simulation framework enables effective handling
of uncertainty in the FJSSP, offering flexibility and customization that is adaptable to various scheduling
environments.

1. Introduction Equilibrium optimizer (EO), a physics-inspired algorithm, achieves

optimal solutions through random updates of particle concentrations.
Efficient production scheduling plays a crucial role in the strategic Guided by equilibrium candidates, EO balances exploration and ex-
planning and management of contemporary manufacturing enterprises. ploitation, effectively avoiding local optima. It outperforms popular
A well-devised scheduling strategy can significantly enhance productiv- meta-heuristic algorithms like PSO, GWO, and genetic algorithm (GA)
ity and optimize resource utilization [1]. A flexible job shop scheduling
in tests on mathematical and engineering problems, making it a concise
problem (FJSSP) is an NP-hard classical scheduling problem [2], where
and competitive optimization solution [12].
machine flexibility, the ability of machines to perform various opera-
In summarizing the points made, the main motivation of this study
tions, plays a pivotal role. This flexibility allows for dynamic adjust-
ment in task assignments, contributing to effective use of resources and is to develop a hybrid simheuristic algorithm combining the EO, NSGA-
enhanced responsiveness to job requirements. Numerous meta-heuristic II, and Monte Carlo simulation to leverage their strengths to tackle
algorithms, including non-dominated sorting genetic algorithm (NSGA the FJSSP. The critical importance of integrating these methods lies in
II) [3], particle swarm optimization (PSO) [4], grey wolf optimizer addressing the dual objectives of makespan and total weighted earliness
(GWO) [5], and simulated annealing (SA) [6] have been introduced in the context of stochastic environments, which has not been con-
to address the challenges of the FJSSP. A job shop scheduling problem currently considered in prior studies and is the focus of this research.
(JSSP) has been influenced by various uncertainties [7], with efforts Conversely, most related works have employed either purely deter-
to address these uncertainties evolving over time. Many uncertainties ministic or less integrative stochastic approaches, whereas this study
can arise due to factors such as varying processing times [8], machine utilizes a simulation to manage uncertainty dynamically, a method
breakdowns [9], unexpected arrival of new jobs [10], and changes in that considers a broader array of stochastic scenarios. The approach
job priorities [11].

∗ Corresponding author at: School of Industrial Engineering, College of Engineering, University of Tehran, Tehran, Iran.
E-mail addresses: [email protected] (S. Nessari), [email protected] (R. Tavakkoli-Moghaddam), [email protected] (H. Bakhshi-Khaniki),
[email protected] (A. Bozorgi-Amiri).

https://doi.org/10.1016/j.dajour.2024.100485
Received 31 December 2023; Received in revised form 24 May 2024; Accepted 27 May 2024
Available online 29 May 2024
2772-6622/© 2024 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC license
(http://creativecommons.org/licenses/by-nc/4.0/).
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

of conducting the study involves initially using EO to generate a high- can be further divided into population-based and single-based meth-
quality population by optimizing the single objective of makespan, ods, have been extensively explored for solving the FJSSP [32]. GAs
which then sets a solid foundation for NSGA-II to handle the subsequent are among the most widely applied population-based meta-heuristics
bi-objective optimization. The simulation aspect helps in managing for the FJSSP [40–42]. These algorithms employ various encoding
the inherent uncertainty effectively. In this context, the fundamental schemes, crossover operators, and mutation operators to evolve a pop-
innovations of this study are (1) the initial high-quality population ulation of solutions. GAs have successfully been applied to both single-
generation using EO for more effective multi-objective optimization; objective and multi-objective variants of the FJSSP [17,43]. Other
(2) the novel integration of EO, NSGA-II, and simulation to effectively population-based techniques (e.g., PSO [44], ant colony optimization
handle the stochastic nature of the FJSSP; (3) the use of a struc- (ACO) [45], artificial bee colony (ABC) [46], and their variants) have
tured simheuristic approach to provide robust, adaptable scheduling also been proposed to solve the FJSSP and its extensions. In the
solutions suitable for various manufacturing environments. context of single-based methods, SA algorithms, with different cooling
The structure of the paper is as follows. Section 2 presents the schedules and neighborhood structures, have been widely used to
literature review. Sections 3 and 4 present the problem description tackle the FJSSP [47]. Tabu search (TS) and its variants (e.g., re-
and the solution method, respectively. Sections 5 and 6 present the active TS) have also been successfully applied to the FJSSP [48].
findings and the managerial insight, respectively. Section 7 presents the Variable neighborhood search [49], iterated local search (ILS) [50],
conclusion. and their hybridizations with other meta-heuristics have been proposed
to enhance the exploration and exploitation capabilities of the search
2. Literature review process [40].
Recent studies have explored quantum annealing (QA) as a meta-
The FJSSP is an extension of the classical JSSP and is considered heuristic for the FJSSP. Schworm et al. [51] demonstrate the efficiency
an NP-hard combinatorial optimization problem with significant real- of a QA-based approach using quantum hardware for the FJSSP, out-
world applications [2]. Due to its complexity and relevance, numerous performing classical methods and showcasing its potential for advanced
solution approaches have been proposed in the literature, which can manufacturing scheduling problems. However, previous QA applica-
be broadly classified into three categories: exact algorithms, heuristics, tions to the FJSSP have been limited to small problems and single
and meta-heuristics [13]. The FJSSP has important real-world appli- objectives, failing to capture the multi-criteria nature of real-world
cations [14] in various industries, including manufacturing [15–17] scheduling. Schworm et al. [52] filled that gap by proposing a novel
printing [18], the automotive industry [19,20], auto parts remanu- QA-based solving algorithm that can optimize multiple objectives like
facturing [21], and the aerospace industry [22]. Researchers have makespan, workload, and job priorities simultaneously for the FJSSP.
developed specialized solution methods and models to address the Abu-Marrul et al. [53] introduced a simheuristic algorithm merging
specific requirements and constraints encountered in these practical Iterated Greedy with Monte Carlo simulation to tackle a complex
scenarios. Boyer et al. [23] introduce the generalized FJSSP, inspired machine scheduling issue in the oil and gas sector, dealing with un-
by seamless rolled ring manufacturing. It extends the classical FJSSP certainty in stochastic processing times and release dates. A robust
with constraints like machine capacity, time lags, holding times, and JSSP dealing with machine unavailability from maintenance activities
sequence-dependent setups. Jim et al. [24] introduce a model for the is tackled using simheuristic algorithms [54]. Caldeira and Gnanavel-
green FJSSP with time-of-use electricity pricing, aiming to minimize babu [55] proposed a simheuristic to minimize the expected makespan
costs and carbon emissions and maximize customer satisfaction. It in the stochastic FJSSP, outperforming meta-heuristics and offering a
proposes an improved genetic algorithm (IGA) with a four-layer en- solution for addressing uncertainty. Li et al. [56] introduced a novel
coding scheme for chromosomes. This algorithm combines global and simheuristic algorithm for the FJSSP with uncertain processing times,
local search strategies for population initialization, separate crossover outperforming other algorithms under uncertainty.
and mutation operations for different chromosome layers, and selects A bi-objective FJSSP is addressed, aiming to minimize earliness,
dominant individuals to enhance solution quality. tardiness, and risks from uncertain processing times using a simheuris-
Researchers have formulated integer linear programming (ILP) and tic [57]. Castaneda et al. [58] presented a methodology combining
Mixed Integer Linear programming (MILP) models to solve the FJSSP meta-heuristic, simulation, and fuzzy methods to tackle stochastic and
optimally [25–28]. These models aim to optimize objectives such as fuzzy uncertainty in a permutation flow shop scheduling problem
makespan, tardiness, and workload balancing, subject to various con- (FSSP), outperforming regular optimization. Wang et al. [59] intro-
straints. Branch-and-bound techniques have also been employed in duced the hybrid adaptive differential evolution algorithm for fuzzy
conjunction with ILP/MILP models to solve the FJSSP [29–31]. Fur- multi-objective JSSP, demonstrating superiority in solution quality
thermore, constraint programming (CP) has emerged as a promising and stability. Gheisariha et al. [60] developed an enhanced multi-
exact approach for solving the highly complex FJSSP and its many objective harmony search algorithm for a flexible FSSP. It addresses
variants. CP models have demonstrated the ability to optimally solve sequence-dependent setup times, transportation, and rework.
some medium-scale instances and provide feasible solutions for large Lim et al. [61] presented an SA-based hyper-heuristic for solv-
practical cases that are challenging for other exact methods like Mixed ing the complex FJSSP, demonstrating the potential for automatically
Integer Programming (MIP) [13]. However, meta-heuristics still often configuring heuristics in scheduling. Saqlain et al. [62] introduced
outperform pure CP in terms of solution quality, especially in larger a Monte Carlo tree search algorithm for the FJSSP, outperforming
instances, within reasonable computational times [32]. As a result, baselines in minimizing makespan. Three algorithms are proposed
hybrid CP meta-heuristic algorithms that combine the strengths of both for parallel FSSP with stochastic times, offering insights into vari-
paradigms are gaining interest [33]. While tuning CP models remains ability and performance tradeoffs [63]. A simheuristic algorithm is
difficult due to the complex constraints, CP offers a valuable comple- explored for the stochastically distributed assembly permutation FSSP,
ment to meta-heuristic methods for tackling the FJSSP’s combinatorial delivering solutions balancing makespan and risk [64]. Fu et al. [65]
complexity [34,35]. proposed a multi-objective ABC algorithm for a hybrid FSSP model with
Various heuristic methods, including dispatching rules [36], beam stochastic times. Experiments showed it outperforms other algorithms
search [37], and other constructive and improvement heuristics, have in optimizing quality and tardiness.
been developed to obtain good solutions for the FJSSP in reason- According to Table 1, the reviewed articles tackle various schedul-
able computational times [27,38,39]. These heuristics often incorpo- ing problems under uncertainty, including FJSSP and FSSP models,
rate problem-specific knowledge and dispatching rules to guide the using simheuristics and multi-objective optimization algorithms. How-
search process effectively. Moreover, Meta-heuristic approaches, which ever, there is a gap in addressing the bi-objective stochastic FJSSP with

2
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Table 1
Literature review.
Study Manufacturing Stochastic Source of Objectives Simheuristic Method Robustness measures
system uncertainty
Boyer et al. Generalized None Minimize MILP +Constraint –
(2021) flexible job completion time Programming
shop
Fu et al. Hybrid flow ✓ Processing Maximize quality ✓ Multi-objective ABC –
(2021) shop times and minimize +Stochastic
tardiness simulation
Caldeira and Flexible job ✓ Processing Minimize expected ✓ Jaya algorithm Reliability plots and
Gnanavel- shop time makespan +Monte Carlo boxplots
babu simulation
(2021)
Li, Gong and Flexible job ✓ Processing Minimize makespan Reinforcement –
Lu (2022) shop times (fuzzy) and machine learning
workload
Wang, Gao Job shop ✓ Processing Minimize Hybrid differential –
and Pedrycz times and completion time, evolution
(2022) due dates delay time, and
energy use
Castaneda Flow shop ✓ Processing Minimize expected ✓ Meta-heuristic Confidence intervals
et al. (2022) times makespan +Monte Carlo of survival function
+Fuzzy
Souza et al. Job shop ✓ Machine Minimize the ✓ Genetic algorithm Quality and solution
(2022) availability weighted sum of +Simulation robustness
expected makespan optimization
and job time
deviations
Rodríguez- Flexible job ✓ Processing Minimize earliness ✓ NSGA-II +Monte Deterioration
Espinosa shop times and tardiness and Carlo simulation minimization;
et al. (2023) associated risks minimize schedule
deterioration
Abu-Marrul Parallel ✓ Processing Minimize weighted ✓ Iterated Greedy Confidence
et al. (2023) machines times and completion time +Monte Carlo intervals; trade-off
release dates Simulation analysis; re-planning
strategy
Tutumlu and Flexible job None Minimize makespan MIP +Hybrid –
Saraç (2023) shop genetic algorithm
Saqlain, Ali Flexible job ✓ Processing Minimize makespan ✓ Reinforcement –
and Lee shop time and waiting times learning
(2023) Maximize improve
utilization
Lim, Wong Flexible job None Minimize makespan Neighbourhood –
and Chin shop structures
(2023)
Schworm Flexible job None Minimize makespan, Quantum annealing –
et al. (2023) shop tardiness, energy
consumption
Maximize machine
utilization
Schworm Flexible job None Minimize makespan Quantum annealing –
et al. (2024) shop and total workload
Jia et al. Green flexible None Minimize cost Improved genetic –
(2024) job shop (energy cost +order algorithm
delay cost), carbon
emissions
Maximize customer
satisfaction
This study Flexible job ✓ Processing Minimize ✓ EO-NSGA-II Wilcoxon test
shop times makespan, and +Monte Carlo
total weighted simheuristic
earliness and
tardiness

the specific objectives of minimizing makespan and the total weighted To address this gap, the proposed hybrid simheuristic method puts
tardiness and earliness. forward an integration of the EO, NSGA-II, and Monte Carlo simulation.
The papers that come closest examine FJSSP models with makespan, The key innovations include:
earliness, and tardiness objectives [57], as well as a fuzzy FJSSP [61],
but do not directly target the combined makespan and total earliness • Generating a high-quality initial population for the multi-objective
and tardiness objectives for a stochastic FJSSP. In addition, existing algorithm in a hybrid simheuristic algorithm by first optimizing
research lacks comprehensive approaches, particularly in the context of a single objective (i.e., makespan).
large-scale problem instances that leverage combining multiple meta- • Integrating this initial population into a simheuristic NSGA-II to
heuristic approaches with simulation to address the uncertainty and then handle the bi-objectives of makespan and total weighted
this specific problem scenario. earliness, under uncertainty and large-scale problems.

3
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

• Considering stochastic processing time in the context of the FJSSP 𝐶𝑖𝑗 ≤ 𝑆𝑡𝑖,𝑗+1 , ∀𝑖, 𝑗 = (1, … , 𝑛 − 1) (8)
simheuristics. ′ ′
𝐶𝑖𝑗 − 𝐶𝑖′ ,𝑗 ′ ≥ 𝑃 𝑡𝑖𝑗 𝑋𝑖𝑗𝑘 𝑋𝑖′ 𝑗 ′ 𝑘 , ∀𝑖, 𝑗, 𝑖 , 𝑗 , 𝑘 (9)
• Using statistical Wilcoxon tests to assess the robustness of the
𝑀
algorithm across different scenarios. ∑𝑖𝑗
𝑋𝑖𝑗𝑘 = 1, ∀𝑖, 𝑗 (10)
𝑘=1
In essence, our approach uniquely combines single and multi-objective
optimization, leverages their strengths through a hybrid simheuristic The FJSSP is formulated as a mathematical model with the objec-
framework, and targets the open challenge of a bi-objective stochastic tives of minimizing the total weighted tardiness and earliness (1), as
FJSSP to advance the state-of-the-art. The robustness analysis further well as minimizing the makespan (2). The processing time 𝑃 𝑡𝑖𝑗 for each
evaluates the consistency of results across different contexts. operation j of job i on machine m is defined based on the difference
between the completion time (Cij) and the start time (St 𝑖𝑗 ) (3). These
3. Problem description completion times and machine assignments are decision variables.
The total earliness 𝐸𝑖 and tardiness 𝑇𝑖 for each job i with due date
The FJSSP extends the classical JSSP by allowing each operation 𝑑𝑖 are calculated based on the differences with the job completion time
to be processed on a set of capable machines rather than a single 𝐶𝑖𝑗 , which is constrained to be after all its operations are complete.
designated machine. The FJSSP addresses the optimal arrangement These feed into the weighted objective function for total earliness and
of n jobs across m machines within a flexible manufacturing system, tardiness (4), (5). The makespan 𝐶𝑚𝑎𝑥 depends on the maximum 𝐶𝑖𝑗
aiming to achieve objectives such as reducing makespan and managing (6). Precedence constraints are enforced between operations of each job
job earliness and tardiness [66]. Each job i consists of a sequence of based on start times 𝑆𝑡𝑖𝑗 (7), (8). Non-overlap constraints prevent over-
operation 𝑂𝑖𝑗 that must be processed sequentially, following specific lapping execution on the same machine m based on processing periods
precedence constraints. Notably, each operation 𝑂𝑖𝑗 within job i has a (9). Finally, machine assignment constraints allocate each operation j
subset 𝑀𝑖𝑗 of m machines on which it can be processed. This means of job i to exactly one of its eligible machines 𝑀𝑖𝑗 (9).
that for each operation, a decision must be made not only regarding In essence, the mathematical model aims to leverage the flexi-
the sequence of operations on each machine but also which specific bilities in routing operations across machines to minimize weighted
machine from the available subset 𝑀𝑖𝑗 should be assigned to pro- earliness and tardiness and makespan, while ensuring precedence satis-
cess that operation. The FJSSP introduces additional flexibility in the faction and non-overlap during schedule construction. The complexity
scheduling process by allowing operations to be assigned to different warrants meta-heuristic solution approaches.
capable machines, potentially leading to better resource utilization and
shorter makespans. However, this increased flexibility also makes the 3.2. Stochastic processing times
problem more complex to solve optimally, as the search space for
possible solutions is significantly larger compared to the classical JSSP.
In scheduling problems like the FJSSP, processing times are tra-
ditionally assumed to be deterministic, meaning they are known pre-
3.1. Problem formulation cisely. However, in actual manufacturing settings, processing times of-
ten exhibit variability due to factors such as machine breakdowns, oper-
The notations used in the proposed mathematical model are sum- ator skill levels, material availability, and unforeseen delays. Stochastic
marized as follows: processing times introduce randomness or uncertainty into schedul-
n Total number of jobs ing. Instead of fixed, predetermined processing times, these times are
m Total number of machines treated as random variables. This implies that the duration needed
𝑂𝑖𝑗 Operation j of job i to complete an operation on a machine is not precisely predictable
𝑀𝑖𝑗 Subsets of machines for 𝑂𝑖𝑗 but follows a probability distribution. Modeling processing times as
𝐶𝑚𝑎𝑥 Makespan stochastic variables provides a more realistic portrayal of the dynamic
TWTE Total weighted tardiness and earliness nature of manufacturing processes. Common probability distributions
𝑊 𝑇𝑖 Tardiness weight of job i used for this purpose include the normal, exponential, uniform, and
𝑊 𝐸𝑖 Earliness weight of job i lognormal distributions, among others. The selection of a distribution
𝑇𝑖 Tardiness of job i depends on the specific characteristics of the manufacturing process
𝐸𝑖 Earliness of job i and the type of uncertainty involved.
𝑆𝑡𝑖𝑗 Starting time of processing of 𝑂𝑖𝑗 The stochastic processing times 𝑆𝑃 𝑡𝑖𝑗 are made to follow a log-
𝑃 𝑡𝑖𝑗 Processing time of 𝑂𝑖𝑗 normal distribution (11) with variance which UL denotes uncertainty
𝐶𝑖𝑗 Completion time of 𝑂𝑖𝑗 level with three values of 0.1, 0.25, and 0.5 (12). The Log-Normal
𝑑𝑖 Due date of job i distribution is characterized by two parameters, 𝜇 (13) and 𝜎 (14), cor-
𝑋𝑖𝑗𝑘 1 if machine k is selected for operation 𝑂𝑖𝑗 ; 0, otherwise responding to each operation. The lognormal distribution is suitable for
modeling positive random variables such as processing times. By defin-
∑ ing the variance through the uncertainty level UL parameter, different
𝑀𝑖𝑛𝑇 𝑊 𝑇 𝐸 = 𝑊 𝑇𝑖 𝑇𝑖 + 𝑊 𝐸𝑖 𝐸𝑖 (1) scenarios are generated from low to high variability. The processing
𝑖
time for each operation then becomes a lognormal random variable
𝑀𝑖𝑛𝐶𝑚𝑎𝑥 (2) based on its 𝜇 and 𝜎 sampled from the distributions corresponding to
s.t. the UL.
∑ Using this stochastic modeling approach, the processing times take
𝐶𝑖𝑗 − 𝑆𝑡𝑖𝑗 = 𝑃 𝑡𝑖𝑗 𝑋𝑖𝑗𝑘 , ∀𝑖, 𝑗 (3)
𝑘∈𝑀𝑖𝑗
on a range of values across simulations as opposed to fixed deter-
ministic quantities. The FJSSP is executed repeatedly to evaluate the
𝑇𝑖 ≥ 𝑚𝑎𝑥0, 𝐶𝑖𝑗 − 𝑑𝑖 , ∀𝑖, 𝑗 (4) objectives under these randomized processing time scenarios. The per-
𝐸𝑖 ≥ 𝑚𝑎𝑥0, 𝑑𝑖 − 𝐶𝑖𝑗 , ∀𝑖, 𝑗 (5) formance and robustness of scheduling solutions can then be analyzed
under controlled settings of uncertainty. Employing variance as a driver
𝐶𝑚𝑎𝑥 ≥ 𝑆𝑡𝑖𝑗 + 𝑃 𝑡𝑖𝑗 , ∀𝑖, 𝑗 (6)
offers flexibility in problem complexity and enables studying scheduler
𝐶𝑖𝑗 − 𝐶𝑖,𝑗−1 ≥ 𝑃 𝑡𝑖𝑗 𝑋𝑖𝑗𝑘 , ∀𝑖, 𝑗 = (2, … , 𝑛), 𝑘 (7) behavior and tradeoffs at different stochasticity levels.

4
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Fig. 1. General framework of the EO-sim-NSGA-II.

In effect, the integration of stochastic and simulation modules 4. Methodology

provides levers to systematically vary the degree of uncertainty in pro-
cessing times. This facilitates a robustness assessment of scheduling so- An optimization framework is proposed that synergistically com-
lutions across different stochastic problem environments for the FJSSP. bines a hybrid meta-heuristic algorithm with a simheuristic approach.
This framework leverages the explorative strengths of meta-heuristics
𝑆𝑃 𝑡𝑖𝑗 = 𝐸[𝑆𝑃 𝑡𝑖𝑗 ] (11)
and the probabilistic modeling capability of simulations to effectively
tackle complex, real-world optimization problems. As shown in Fig. 1,
𝑉 𝑎𝑟[𝑆𝑃 𝑡𝑖𝑗 ] = 𝑈 𝐿.𝐸[𝑆𝑃 𝑡𝑖𝑗 ] (12) the framework comprises two key components working in tandem.
( )
1 𝑉 𝑎𝑟[𝑆𝑃 𝑡𝑖𝑗 ] 4.1. Sim-NSGA-II
𝜇𝑖𝑗 = ln 𝐸[𝑆𝑃 𝑡𝑖𝑗 ] −
ln 1 + (13)
2 𝐸[𝑆𝑃 𝑡𝑖𝑗 ]2
√
|√ ( )| Simheuristic represents an innovative optimization approach that
|√ 𝑉 𝑎𝑟[𝑆𝑃 𝑡𝑖𝑗 ] ||
|
𝜎𝑖𝑗 = |√ln 1 + | (14) seamlessly integrates simulation and meta-heuristic methods. Instead
| 𝐸[𝑆𝑃 𝑡𝑖𝑗 ]2 ||
| of relying on a deterministic way, a simheuristic harnesses simulations
| |

5
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

to evaluate solutions, thereby adeptly accommodating the intricate and Algorithm 2: Monte Carlo Simulation
probabilistic nature inherent in real-world problem-solving. Input: Number of simulations S, distribution D
As depicted in Algorithm 1, at the core of this framework, the NSGA- Output: Estimated value
II algorithm orchestrates the exploration of solution spaces, employing sum ← 0
its adaptive mechanisms to intelligently navigate through potential for s = 1 to S do
solutions. The NSGA-II, a widely adopted multi-objective evolutionary Generate a random sample x from distribution D Calculate the
algorithm, utilizes techniques such as nondominated sorting, crowding value f (x)
distance, and elitist selection to uncover the Pareto frontier. It operates sum ← sum + f (x)
by maintaining a population of candidate solutions and iteratively end for
evolving this population through genetic operations such as selection, average ← sum/S
crossover, and mutation. Central to its efficacy is the concept of Pareto Return average
dominance, wherein one solution is deemed superior to another if it
excels in at least one objective without worsening any other objective. 4.2. EO-sim-NSGA-II
Simultaneously, the Monte Carlo simulation injects a probabilistic
element into the optimization process. It systematically generates and A two-stage hybrid simheuristic algorithm is proposed for the res-
analyzes numerous scenarios by sampling from probability distribu- olution of the multi-objective optimization problem. In the first stage,
tions, thereby generating diverse data points that comprehensively the problem is reformulated into a single objective, employing the EO,
assess solution quality under varying conditions. As depicted in Algo- as demonstrated in Algorithm 3. Initialization of a random solution
rithm 2, Monte Carlo simulation, which is a costly and computationally population by EO initiates an evolution process driven by equilibrium
intensive technique, relies on random sampling to yield numerical forces of attraction and repulsion. This process steers the population
results for complex problems fraught with uncertainty. By iterative towards diverse optima, reflecting trade-offs between exploration and
sampling from probability distributions representing uncertain input exploitation.
parameters, Monte Carlo simulation furnishes estimates of the prob-
ability distribution of output variables, enabling decision-makers to Algorithm 3: InitializePopulation: Equilibrium Optimizer
evaluate the robustness and reliability of different solutions under Input: Population size N, number of iterations T
uncertainty. Output: Optimized solution
Algorithm 1 serves as a blueprint, outlining the procedural steps Initialize Random population P with N individuals Evaluate fitness
that illustrate the collaboration within the simheuristic framework. To of individuals in P
execute Algorithm 1 effectively, the configuration of four key param- for t = 1 to T do
eters is required, namely the population size for the meta-heuristic Update the position of each individual in P
algorithm, the designated number of generations for evolutionary pro- Evaluate the fitness of individuals in P
cesses, as well as the mutation rate and crossover rate. The solutions Sort individuals in P based on fitness
of the deterministic model undergo a short simulation in a stochas- Calculate the mean position of the top k individuals in P
tic model, enabling the swift evaluation of its fitness concerning the Calculate the standard deviation of the positions of the top k
defined objectives. This preliminary phase serves as a rapid screening individuals in P
process, swiftly gauging the potential of solutions within a constrained for each individual p in P do
timeframe. Following this, the final Pareto frontier advances to the final Update the position of p using the Equilibrium Equation
step, undergoing an extensive long simulation, allowing for a deeper end for
understanding of solution performance and robustness. Evaluate the fitness of individuals in P
Sort individuals in P based on fitness
Algorithm 1: SIM-NSGA-II
Update the equilibrium factor based on the fitness
Input: Population size N, number of generations G, mutation rate improvement
mr, crossover rate cr Update the step size based on the fitness improvement
Output: Pareto front approximation Apply the boundary constraints to the positions of individuals
P ← InitializePopulation(N ); in P
ShortSimulation(pop); end for
for g = 1 to G do Return the best individual in P
Create an empty offspring population Q
while |𝑄| < 𝑁 do The set of equilibrated single-objective solutions produced by EO
Select parents from P using binary tournament selection serves as the initial population for the NSGA-II in the second stage. By
Perform crossover and mutation to create offspring utilizing the EO solutions as a seed, sim-NSGA-II commences its search
Add offspring to Q from pre-optimized starting points. This EO-Sim-NSGA-II framework
end while leverages the single-objective exploration of EO in the initial stage to
𝑄′ ← 𝑅𝑒𝑝𝑎𝑖𝑟(𝑄) provide well-initialized starting points for the focused multi-objective
ShortSimulation(𝑄′ ); search conducted by NSGA-II in the subsequent stage.
Merge P and 𝑄′ into a combined population R Moreover, in various real-world scenarios, multiple objectives are
Perform non-dominated sorting on R often interconnected or interdependent, implying that improvements in
Assign crowding distance to individuals in front of R one objective can have positive or negative effects on other objectives.
Select the best N individuals from R based on crowding These relationships can vary across different regions or solutions within
distance the search space. Consequently, while the overall objectives may not
Update P with the selected individuals globally conflict, trade-offs and conflicts between objectives may arise
Increment generation counter g at specific points or local optima within the problem space. Thereby,
end for the diverse solutions generated by EO for one objective contribute to
for s ∈ P do increased diversity within the Pareto front. Detailed implementation
LongSimulation(s); specifics of both algorithms are provided later. Performance evaluations
end for against standalone applications on multi-objective benchmark functions
Return Pareto front approximation from P validate the hybrid approach’s capacity to enhance convergence.

6
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Fig. 2. FJSSP representation coding.

4.3. FJSSP representation Table 2

Parameters configuration for running experi-
ment.
In a scenario involving the FJSSP with J jobs and M machines, the
Parameters Value
coding representation is structured with J x M variables that assign
sequence operations of each job to their eligible machines. Specifically, Population size 100
Crossover rate 0.8
each of the J jobs contains multiple operations that will only be
Mutation rate 0.2
processed once on each machine. EO iteration 100
As an illustration, Fig. 2 depicts an example with 3 jobs and 3 Total iteration 400
machines, it can be observed that this encoding denotes a permutation Long simulation iteration 10 000
that sequences each job exactly once on each machine according to the Short simulation iteration 20
Runs per experiment 30
job-machine sequence. By encapsulating complete scheduling decisions
in a compact format, this representation facilitates convenient manipu-
lation by the optimization algorithm to improve objectives. The vector
is decoded into feasible schedules satisfying constraints. 5.1. Experiments with three uncertainty levels for the stochastic parameters

5. Computational experiments The first experiment compares the EO-Sim-NSGA-II against the Sim-
NSGA-II using the deterministic NSGA-II across low, medium, and high
The computational experiments were conducted on a benchmark set uncertainty levels. The uncertainty is modeled by setting the variance
consisting of 9 instances, each varying in dimension, by considering parameter UL to 0.5, 1.0, and 2.0 for the three scenarios respectively.
parameters for running the experiments from Table 2, to comprehen- Each instance executes 30 times with the best outcome chosen for anal-
sively assess the performance of the EO-SIM-NSGA-II algorithm. The ysis. For each run, the algorithms take the obtained Pareto frontiers,
implementation of the simheuristic was in Python programming lan- consider the best value of each objective as the best result, and use
guage. These experiments were executed on a hardware configuration
the metrics as per Eq. (1). Also, the mean number of Pareto-optimal
featuring an Apple Silicon M1 Pro chip with 16 GB of RAM, operating
solutions (NPS) and mean CPU times are examined. Table 3 summarizes
on macOS.
the comparative results grouped by job and machine combinations for
For each dataset instance, the EO-SIM-NSGA-II algorithm was ex-
the variance settings.
ecuted, generating optimization results that showcase its efficacy in
Table 3 clearly illustrates that, in most instances, the EO-Sim-NSGA-
solving diverse problem dimensions. Subsequently, to provide a thor-
II outperformed the Sim-NSGA-II, exhibiting a narrower gap compared
ough analysis, convergence charts were constructed to illustrate the
to the best deterministic objectives discovered by the NSGA-II. Remark-
algorithm’s convergence behavior across various datasets. These charts
ably, certain runs reveal negative gaps, indicating solutions exceeding
offer insights into the convergence speed and stability of the algorithm
the deterministic performance despite uncertainty. This underscores
across different problem instances. Moreover, to ascertain the robust-
the efficacy of integrating EO and NSGA-II to balance exploration and
ness and statistical significance of the obtained results, a Wilcoxon
test was conducted. The results obtained from these computational exploitation. Moreover, the EO-Sim-NSGA-II consistently excelled in
experiments serve to elucidate the strengths and efficacy of the EO- discovering a greater number of high-quality Pareto solutions, show-
SIM-NSGA-II algorithm. The ensuing discussion will delve into the casing its proficiency in exploring diverse Pareto frontiers. Notably, the
insights gleaned from these experiments, providing a comprehensive computational efficiency of the EO-Sim-NSGA-II, measured in CPU time
understanding of the algorithm’s performance, its strengths, and areas for the same total epoch, surpassed that of the Sim-NSGA-II. This sug-
for potential improvement. gests that the EO-Sim-NSGA-II not only achieves superior solutions but
A key metric used in the comparative analysis is the percent gap (𝛥o ) also does so more swiftly and with reduced computational demands.
in best objective values between the deterministic (𝑓𝑑 ) and stochas- Also, Figs. 3a and 3b illustrate sample scheduling solutions of ex-
tic solutions (𝑓𝑠 ), quantified as Eq. (15). A positive gap percentage periments E1 and E4, obtained by the EO-sim-NSGA-II algorithm, These
indicates the stochastic solution is worse than the deterministic case. Gantt charts offer valuable visualizations that highlight the algorithm’s
ability to tackle problems. The inclusion of specific scheduling instances
𝑓𝑠 − 𝑓𝑑 as figures provides examples that demonstrate the algorithm’s capabil-
𝛥𝑜 = (15) ities, enhancing the comprehension of its performance. Such targeted
𝑓𝑑

7
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Fig. 3. Gantt chart of one of the solutions of two experiments of the EO-sim-NSGA-II.

Table 3
Results summary for each algorithm.
Uncertainty level i j-m Sim-NSGA-II EO-Sim-NSGA-II
𝛥𝑜1 𝛥𝑜2 NPS CPU time 𝛥𝑜1 𝛥𝑜2 NPS CPU time
Low E1 10–5 0.072617 0.306785 10 18.02444 0.053791 0.158251 12 16.43649
E2 15–5 0.116044 0.185706 11 23.25542 0.050021 0.149527 16 20.40536
E3 20–5 0.085316 0.319801 15 28.21675 0.059555 0.213556 13 25.85585
E4 10–10 0.062024 0.06986 2 26.64939 0.069579 0.385338 5 22.41676
E5 15–10 0.112312 0.242765 2 35.77363 0.025074 0.000431 12 31.90378
E6 20–10 0.048216 0.040687 4 45.2828 −0.02503 −0.20761 4 40.49214
E7 30–10 0.135357 0.145341 2 64.99246 0.027714 −0.01672 6 59.17984
E8 15–15 0.207685 0.29455 2 48.57303 0.093616 0.00891 5 40.60772
E9 20–15 0.050795 0.11633 3 62.10068 0.020341 −0.15765 7 56.48539
Medium E10 10–5 0.505516 0.381171 1 18.50975 0.471782 0.196241 3 15.05491
E11 15–5 0.519203 0.893717 7 23.81621 0.443966 0.666539 16 20.35887
E12 20–5 0.464871 0.921781 12 28.67281 0.421311 0.585453 9 25.76782
E13 10–10 0.517445 1.397383 6 26.83027 0.388874 1.394518 7 22.52793
(continued on next page)

8
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Table 3 (continued).
Uncertainty level i j-m Sim-NSGA-II EO-Sim-NSGA-II
E14 15–10 0.459242 0.29012 2 36.09009 0.368225 0.253576 6 32.49778
E15 20–10 0.501342 1.029642 1 45.68245 0.282518 0.562286 4 40.535
E16 30–10 0.561429 0.621553 2 65.78497 0.482571 0.469576 3 59.44553
E17 15–15 0.548119 0.226847 1 48.09727 0.535778 0.073819 3 40.63923
E18 20–15 0.471488 0.246217 2 61.80097 0.375723 0.128152 1 56.77659
High E19 10–5 1.046676 1.190311 6 18.69963 0.89024 0.920006 7 14.733
E20 15–5 0.915215 0.895387 4 23.657481 0.891501 0.659131 15 20.28575
E21 20–5 0.942701 1.320942 7 28.5997 0.92256 1.104665 14 26.37614
E22 10–10 0.921795 1.583894 3 26.999842 0.906593 1.141633 9 22.37123
E23 15–10 0.91458 0.768772 1 35.862 0.858763 0.482745 2 30.98958
E24 20–10 0.937112 0.688554 9 46.17402 0.737709 0.385937 2 40.36287
E25 30–10 1.117048 1.083773 5 64.65103 1.036762 0.961248 7 59.73793
E26 15–15 1.016709 1.276376 2 48.68396 1.009566 0.851142 2 40.76972
E27 20–15 1.003926 0.648607 2 62.8053 0.904132 0.483014 5 56.23009

Fig. 4. Pareto frontier solutions of running in each experiment.

illustrations grounded in actual output data strengthen understanding superiority of the EO-sim-NSGA-II algorithm over the sim-NSGA-II
and build further credibility in the algorithm’s operational performance algorithm is evident. Through this analysis, it is observed that the EO-
within complex flexible manufacturing systems. sim-NSGA-II variant outperforms not only the sim-NSGA-II algorithm
Also, Figs. 4a to 4c visually represent the Pareto frontiers across but also the deterministic results. This is attributed to the exploration
small, medium, and large dataset instances of experiments related to of deterministic initial solutions by the EO for the simheuristic algo-
the EO-sim-NSGA-II algorithm, serving as illustrations to comprehend rithm, which sometimes results in negative gaps. Further substantiating
better the problem. The trade-off curves depicted offer a glimpse into the evaluations, convergence analyses were conducted across 30 runs
the algorithm’s ability to balance the minimization of makespan and for each of the 27 experiments, tracking the best global solution in
total earliness and tardiness for problems with different sizes. each iteration. The mean makespan was computed per iteration by
Fig. 6 encapsulates the information from Table 3 in the form of aggregating across runs, excluding the second objective, as EO focuses
distributions representing the gaps between the mean solutions found solely on makespan. Fig. 5 plots the convergence trends on 9 experi-
for two objectives across 30 runs for 27 experiments. This yields ments with low-level uncertainty and different instance sizes, varying
a total of 810 samples, comparing them with their corresponding from small to large. The EO-Sim-NSGA-II displays faster convergence,
deterministic values. These distributions are presented as boxplots, achieving lower makespan faster across problem sizes. This highlights
illustrating the gaps between sim-NSGA-II and EO-sim-NSGA-II across the power of hybridizing multi-objective search and rapid knowledge
all benchmark experiments under varying levels of uncertainty. The building. Furthermore, the greater gaps in large instance experiments

9
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Fig. 5. Convergence charts of experiments.

Fig. 6. Boxplot analysis of uncertainty levels.

showcase the effectiveness of hybridizing simheuristic over the simple significant differences between the two algorithms. The positive sign
simheuristic (i.e., sim-NSGA-II). indicates the superiority of EO-sim-NSGA-II over Sim-NSGA-II, while
the negative sign indicates the vice versa. The equal sign indicates that
5.2. Robustness analysis there is no significant difference. The results reveal that the EO-Sim-
NSGA-II’s solutions demonstrate statistically superior optimization over
A Wilcoxon statistical test (5% significance level) validates the Sim-NSGA-II. This underscores EO-Sim-NSGA-II’s robustness in deliv-
robustness and significance of the performance of the EO-Sim-NSGA-II’s ering high-quality solutions despite problem variations. The Wilcoxon
gains over the Sim-NSGA-II across different experiments. By comparing test provides statistical validation of the EO-Sim-NSGA-II’s reliability
the best objective values in Pareto solutions from 30 runs of each advantages and aptness for complex optimization, reinforcing its com-
experiment, it evaluated algorithm consistency and reliability. The out- petitiveness. In essence, comprehensive analytical assessments cement
comes in Table 4 highlight scenarios with p-values < 0.05, indicating algorithm robustness, informing managerial selections for stochastic

10
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

Table 4 have substantiated the EO-Sim-NSGA-II’s consistent outperformance

p-values of the Wilcoxon test EO-sim-NSGA-II vs Sim-NSGA-II with 5% significance.
against benchmarks, cementing accuracy, reliability, and robustness.
Uncertainty level j-m 𝑂1 - p-value 𝑅1 𝑂2 - p-value 𝑅2
The algorithm has displayed sophisticated learning capabilities in nav-
Low E1 3.45E−02 + 4.66E−03 + igating complexity from dynamism and stochasticity across problem
E2 5.59E−05 + 3.74E−03 +
scales. Convergence studies have reaffirmed the hybrid’s competen-
E3 2.77E−03 + 2.34E−02 +
E4 3.13E−04a – 2.13E−01a = cies in achieving high-quality solutions faster, validating the method-
E5 1.58E−03 + 1.06E−02 + ological innovations. Key contributions including the uniqueness of
E6 4.41E−05 + 4.60E−04 + the hybrid technique, holistic uncertainty-based analysis, and multi-
E7 1.82E−05 + 2.09E−04 +
faceted performance assessments significantly advance the domains of
E8 4.97E−05 + 5.05E−04 +
E9 1.46E−03 + 1.06E−02 + optimization, simulation, and manufacturing research. The integrated
Medium E10 1.06E−05 + 1.86E−09 + approach offers flexibility and customizability based on scheduling
E11 6.15E−08 + 1.86E−09 + environments. In essence, this work has successfully engineered and
E12 1.64E−07 + 1.86E−09 +
demonstrated an elite, resilient solution approach for conquering the
E13 3.73E−09 + 8.03E−02a =
E14 2.61E−08 + 2.55E−07 +
onerous challenges of unpredictable multi-objective optimization in
E15 1.86E−09 + 1.86E−09 + next-generation smart factories. By scientifically harnessing algorithmic
E16 1.99E−06 + 1.86E−09 + power, it delivers transformative capabilities for managing manufactur-
E17 7.32E−02a = 3.73E−09 + ing complexity. The insights propel further opportunities in embedding
E18 7.99E−06 + 2.37E−05 +
High E19 2.21E−02 + 1.82E−05 +
intelligence within industrial systems for the navigation of multifaceted
E20 9.61E−02a = 2.61E−08 + dynamics.
E21 7.67E−02a = 2.09E−04 + However, it is important to acknowledge certain limitations in
E22 1.58E−03 + 3.15E−07 + our proposed approach. First of all, while the integration of hybrid
E23 5.59E−05 + 1.86E−09 +
E24 6.15E−08 + 1.86E−09 +
simheuristics offers promise in addressing the bi-objective stochastic
E25 1.23E−04 + 3.24E−06 + FJSSP, the effectiveness of the approach may vary depending on the
E26 5.02E−02a = 5.59E−09 + specific characteristics and complexities of the scheduling instances, it
E27 9.52E−04 + 4.18E−04 + needs to be tested in other problem domains. Moreover, the perfor-
a
The worst value. mance of the proposed method could be influenced by the choice of
parameter settings and the quality of the initial population generated
during the optimization process, so it needs to be studied in a way that
scheduling scenarios. Embedding such validations facilitates sound de- can be tuned for different problems. Finally, the proposed algorithm’s
cisions when navigating the intricacies of unpredictable optimization applicability to real-world manufacturing environments may require
landscapes. additional adaptation to account for practical constraints and dynamics,
which were not fully explored in this study. Future research efforts
6. Managerial insights should thus focus on mitigating these limitations and conducting exten-
sive empirical evaluations to validate the applicability and effectiveness
Managing a stochastic bi-objective FJSSP entails substantial com- of the proposed approach across diverse problem domains and scales.
plexity from uncertainties that hamper optimization. Balancing mini-
mized makespan against reduced earliness and tardiness becomes ex- Declaration of competing interest
ceedingly challenging under unpredictability. Effectively navigating
this landscape necessitates managerial cognizance of inherent trade-
The authors of this research entitled ‘‘A Hybrid Simheuristic Algo-
offs and tailored algorithm selection adept at handling stochasticity.
rithm for Solving Bi-Objective Stochastic Flexible Job Shop Scheduling
Comprehensive analytical assessments, including convergence and sta-
Problems’’ certify that there is no any affiliation with or involvement in
tistical appraisals like the Wilcoxon test, inform insightful scheduling
any organization or entity with financial interest (e.g., honoraria; ed-
strategies that reliably attain optimality across stochastic scenarios.
ucational grants; participation in speakers’ bureaus; membership, em-
Additionally, embracing stochasticity underscores the criticality of
ployment, consultancies, stock ownership, or other equity interest; and
adaptive resilience. Developing scheduling policies not only flexible but
expert testimony or patent-licensing arrangements), or non-financial
adaptable to random variations empowers agile responses to unfore-
interest (e.g., personal or professional relationships, affiliations, knowl-
seen events. Embedding adaptability fortifies the system’s capability
edge or beliefs) in the subject matter or materials discussed in this
to maintain optimal performance despite volatility. This entails con-
tinuously improving job completion times and managing earliness and manuscript.
tardiness deviations. The essence of managerial insight involves lever-
aging analytical insights and adaptive policies to foster an environment Availability of data and material
driving robust optimization across stochasticity. By combining algorith-
mic and strategic competencies, managers can aptly balance production The data and code that support the findings of this study are
objectives with resilience to uncertainties. available at https://github.com/SamanNsr/Hybrid-Simheuristic-BiObj-
Stoch-FJSSP.
7. Conclusion
References
This paper has introduced a pioneering EO-Sim-NSGA-II algorithm
that synergizes the explorative prowess of meta-heuristics with the [1] J. Xie, X. Li, L. Gao, L. Gui, A new neighbourhood structure for job shop
exploitative focus of simheuristics for addressing a stochastic multi- scheduling problems, Int. J. Prod. Res. 61 (2023) 2147–2161, http://dx.doi.org/
objective FJSSP. Through an intricate integration of the EO and NSGA- 10.1080/00207543.2022.2060772.
II embedded within a Monte Carlo simulation framework, this method [2] J. Xie, L. Gao, K. Peng, X. Li, H. Li, Review on flexible job shop scheduling, IET
Collab. Intell. Manuf. 1 (2019) 67–77, http://dx.doi.org/10.1049/iet-cim.2018.
balances single and multi-objective searches to offer a potent solution
0009.
tackling the combined minimization of makespan and total weighted [3] M. Rabiee, M. Zandieh, P. Ramezani, Bi-objective partial flexible job shop
earliness and tardiness amid uncertainty. Extensive computational ex- scheduling problem: NSGA-II, NRGA, MOGA and PAES approaches, Int. J. Prod.
periments and statistical testing based on controlled variance injection Res. 50 (2012) 7327–7342, http://dx.doi.org/10.1080/00207543.2011.648280.

11
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

[4] M. Nouiri, A. Bekrar, A. Jemai, S. Niar, A.C. Ammari, An effective and distributed [28] G.A. Kasapidis, S. Dauzère-Pérès, D.C. Paraskevopoulos, P.P. Repoussis, C.D.
particle swarm optimization algorithm for flexible job-shop scheduling problem, Tarantilis, On the Multiresource Flexible Job-Shop Scheduling Problem with
J. Intell. Manuf. 29 (2018) 603–615, http://dx.doi.org/10.1007/s10845-015- Arbitrary Precedence Graphs, vol. 32, 2023, pp. 2322–2330, http://dx.doi.org/
1039-3. 10.1111/POMS.13977.
[5] C. Lin, Z. Cao, M. Zhou, Learning-based grey wolf optimizer for stochastic [29] A. Ahmadi-Javid, M. Haghi, P. Hooshangi-Tabrizi, Integrated job-shop scheduling
flexible job shop scheduling, IEEE Trans. Autom. Sci. Eng. 19 (2022) 3659–3671, in an FMS with heterogeneous transporters: MILP formulation, constraint pro-
http://dx.doi.org/10.1109/TASE.2021.3129439. gramming, and branch-and-bound, Int. J. Prod. Res. (2023) http://dx.doi.org/
[6] M.A. Cruz-Chávez, M.G. Martínez-Rangel, M.H. Cruz-Rosales, Accelerated simu- 10.1080/00207543.2023.2230489.
lated annealing algorithm applied to the flexible job shop scheduling problem, [30] J. Ahn, H.J. Kim, A branch and bound algorithm for scheduling of flexible
Int. Trans. Oper. Res. 24 (2017) 1119–1137, http://dx.doi.org/10.1111/itor. manufacturing systems, IEEE Trans. Autom. Sci. Eng. (2023) http://dx.doi.org/
12195. 10.1109/TASE.2023.3296087.
[7] P.B. Luh, Chen Dong, L.S. Thakur, An effective approach for job-shop scheduling [31] C. Juvin, L. Houssin, P. Lopez, Logic-based benders decomposition for the
with uncertain processing requirements, IEEE Trans. Robot. Autom. 15 (1999) preemptive flexible job-shop scheduling problem, Comput. Oper. Res. 152 (2023)
328–339, http://dx.doi.org/10.1109/70.760354. 106156, http://dx.doi.org/10.1016/J.COR.2023.106156.
[8] Lei De-Ming, Xiong He-Jing, Job shop scheduling with stochastic processing
[32] M. Schlenkrich, S.N. Parragh, Solving large scale industrial production scheduling
time through genetic algorithm, in: 2008 International Conference on Machine
problems with complex constraints: an overview of the state-of-the-art, Procedia
Learning and Cybernetics, IEEE, 2008, pp. 941–946, http://dx.doi.org/10.1109/
Comput. Sci. 217 (2023) 1028–1037, http://dx.doi.org/10.1016/J.PROCS.2022.
ICMLC.2008.4620540.
12.301.
[9] E. Ahmadi, M. Zandieh, M. Farrokh, S.M. Emami, A multi objective optimization
[33] G. Ziadlou, S. Emami, E. Asadi-Gangraj, Network configuration distributed
approach for flexible job shop scheduling problem under random machine
production scheduling problem: A constraint programming approach, Comput.
breakdown by evolutionary algorithms, Comput. Oper. Res. 73 (2016) 56–66,
Ind. Eng. 188 (2024) 109916, http://dx.doi.org/10.1016/J.CIE.2024.109916.
http://dx.doi.org/10.1016/j.cor.2016.03.009.
[34] D. Müller, D. Kress, Filter-and-fan approaches for scheduling flexible job shops
[10] D. Rahmani, M. Heydari, Robust and stable flow shop scheduling with unex-
under workforce constraints, Int. J. Prod. Res. 60 (2022) 4743–4765, http:
pected arrivals of new jobs and uncertain processing times, J. Manuf. Syst. 33
//dx.doi.org/10.1080/00207543.2021.1937745.
(2014) 84–92, http://dx.doi.org/10.1016/j.jmsy.2013.03.004.
[11] D. Golenko-Ginzburg, S. Kesler, Z. Landsman, Industrial job-shop scheduling with [35] V. Boyer, J. Vallikavungal, X. Cantú Rodríguez, M.A. Salazar-Aguilar, The
random operations and different priorities, Int. J. Prod. Econ. 40 (1995) 185–195, generalized flexible job shop scheduling problem, Comput. Ind. Eng. 160 (2021)
http://dx.doi.org/10.1016/0925-5273(95)00078-8. 107542, http://dx.doi.org/10.1016/J.CIE.2021.107542.
[12] A. Faramarzi, M. Heidarinejad, B. Stephens, S. Mirjalili, Equilibrium optimizer: [36] M. Thenarasu, K. Rameshkumar, M. Di Mascolo, S.P. Anbuudayasankar, Multi-
A novel optimization algorithm, Knowl. Based Syst. 191 (2020) 105190, http: criteria scheduling of realistic flexible job shop: a novel approach for integrating
//dx.doi.org/10.1016/j.knosys.2019.105190. simulation modelling and multi-criteria decision making, Int. J. Prod. Res. 62
[13] S. Dauzère-Pérès, J. Ding, L. Shen, K. Tamssaouet, The flexible job shop (2024) 336–358, http://dx.doi.org/10.1080/00207543.2023.2238084.
scheduling problem: A review, European J. Oper. Res. 314 (2024) 409–432, [37] K. Hadj Salem, V. Jost, Y. Kieffer, L. Libralesso, S. Mancini, Minimizing makespan
http://dx.doi.org/10.1016/J.EJOR.2023.05.017. under data prefetching constraints for embedded vision systems: a study of
[14] C. Destouet, H. Tlahig, B. Bettayeb, B. Mazari, Flexible job shop scheduling optimization methods and their performance, Oper. Res. 22 (2022) 1639–1673,
problem under Industry 5.0: A survey on human reintegration, environmental http://dx.doi.org/10.1007/S12351-021-00647-0/METRICS.
consideration and resilience improvement, J. Manuf. Syst. 67 (2023) 155–173, [38] K. Huang, W. Gong, C. Lu, An enhanced memetic algorithm with hierarchical
http://dx.doi.org/10.1016/J.JMSY.2023.01.004. heuristic neighborhood search for type-2 green fuzzy flexible job shop schedul-
[15] L.M. Steinbacher, D. Rippel, P. Schulze, A.K. Rohde, M. Freitag, Quality-based ing, Eng. Appl. Artif. Intell. 130 (2024) 107762, http://dx.doi.org/10.1016/J.
scheduling for a flexible job shop, J. Manuf. Syst. 70 (2023) 202–216, http: ENGAPPAI.2023.107762.
//dx.doi.org/10.1016/J.JMSY.2023.07.005. [39] K.C.W. Lim, L.P. Wong, J.F. Chin, Hyper-heuristic for flexible job shop scheduling
[16] Q. Gong, J. Li, Z. Jiang, Y. Wang, A hierarchical integration scheduling method problem with stochastic job arrivals, Manuf. Lett. 36 (2023) 5–8, http://dx.doi.
for flexible job shop with green lot splitting, Eng. Appl. Artif. Intell. 129 (2024) org/10.1016/J.MFGLET.2022.12.009.
107595, http://dx.doi.org/10.1016/J.ENGAPPAI.2023.107595. [40] B. Tutumlu, T. Saraç, A MIP model and a hybrid genetic algorithm for flexible
[17] H. Tang, Y. Xiao, W. Zhang, D. Lei, J. Wang, T. Xu, A DQL-NSGA-III algorithm job-shop scheduling problem with job-splitting, Comput. Oper. Res. 155 (2023)
for solving the flexible job shop dynamic scheduling problem, Expert Syst. Appl. 106222, http://dx.doi.org/10.1016/J.COR.2023.106222.
237 (2024) 121723, http://dx.doi.org/10.1016/J.ESWA.2023.121723. [41] L. Meng, W. Cheng, B. Zhang, W. Zou, W. Fang, P. Duan, An improved genetic
[18] W.T. Lunardi, E.G. Birgin, D.P. Ronconi, H. Voos, Metaheuristics for the online algorithm for solving the multi-AGV flexible job shop scheduling problem,
printing shop scheduling problem, European J. Oper. Res. 293 (2021) 419–441, Sensors 23 (2023) 3815, http://dx.doi.org/10.3390/S23083815.
http://dx.doi.org/10.1016/J.EJOR.2020.12.021. [42] M. Liu, J. Lv, S. Du, Y. Deng, X. Shen, Y. Zhou, Multi-resource constrained flex-
[19] Q. Gao, F. Gu, L. Li, J. Guo, A framework of cloud–edge collaborated digital twin ible job shop scheduling problem with fixture-pallet combinatorial optimisation,
for flexible job shop scheduling with conflict-free routing, Robot. Comput. Integr. Comput. Ind. Eng. 188 (2024) 109903, http://dx.doi.org/10.1016/J.CIE.2024.
Manuf. 86 (2024) 102672, http://dx.doi.org/10.1016/J.RCIM.2023.102672. 109903.
[20] M.M. Wocker, F.F. Ostermeier, T. Wanninger, R. Zwinkau, J. Deuse, Flexible job
[43] Y. Tian, Z. Gao, L. Zhang, Y. Chen, T. Wang, A multi-objective optimization
shop scheduling with preventive maintenance consideration, J. Intell. Manuf.
method for flexible job shop scheduling considering cutting-tool degradation with
(2023) 1–23, http://dx.doi.org/10.1007/S10845-023-02114-3/TABLES/3.
energy-saving measures, Mathematics 11 (2023) 324, http://dx.doi.org/10.3390/
[21] W. Zhang, Y. Zheng, R. Ahmad, An energy-efficient multi-objective integrated
MATH11020324.
process planning and scheduling for a flexible job-shop-type remanufacturing
[44] J. Shi, M. Chen, Y. Ma, F. Qiao, A new boredom-aware dual-resource constrained
system, Adv. Eng. Inform. 56 (2023) 102010, http://dx.doi.org/10.1016/J.AEI.
flexible job shop scheduling problem using a two-stage multi-objective particle
2023.102010.
swarm optimization algorithm, Inf. Sci. (N. Y.) 643 (2023) 119141, http://dx.
[22] Z. Tian, X. Jiang, W. Liu, Z. Li, Dynamic energy-efficient scheduling of multi-
doi.org/10.1016/J.INS.2023.119141.
variety and small batch flexible job-shop: A case study for the aerospace industry,
[45] S. Yan, G. Zhang, J. Sun, W. Zhang, S. Yan, G. Zhang, et al., An improved
Comput. Ind. Eng. 178 (2023) 109111, http://dx.doi.org/10.1016/J.CIE.2023.
ant colony optimization for solving the flexible job shop scheduling problem
109111.
with multiple time constraints, Math. Biosci. Eng. 20 (2023) 7519–7547, http:
[23] V. Boyer, J. Vallikavungal, X. Cantú Rodríguez, M.A. Salazar-Aguilar, The
//dx.doi.org/10.3934/MBE.2023325.
generalized flexible job shop scheduling problem, Comput. Ind. Eng. 160 (2021)
107542, http://dx.doi.org/10.1016/J.CIE.2021.107542. [46] Y. Li, C. Liao, L. Wang, Y. Xiao, Y. Cao, S. Guo, A reinforcement learning-
[24] S. Jia, Y. Yang, S. Li, S. Wang, A. Li, W. Cai, et al., The green flexible artificial bee colony algorithm for flexible job-shop scheduling problem with lot
job-shop scheduling problem considering cost, carbon emissions, and customer streaming, Appl. Soft. Comput. 146 (2023) 110658, http://dx.doi.org/10.1016/
satisfaction under time-of-use electricity pricing, Sustainability 16 (2024) 2443, J.ASOC.2023.110658.
http://dx.doi.org/10.3390/SU16062443. [47] Z. Zhang, Y. Fu, K. Gao, H. Zhang, L. Wang, A cooperative evolutionary
[25] L. Meng, B. Zhang, K. Gao, P. Duan, An MILP model for energy-conscious algorithm with simulated annealing for integrated scheduling of distributed
flexible job shop problem with transportation and sequence-dependent setup flexible job shops and distribution, Swarm Evol. Comput. 85 (2024) 101467,
times, Sustainability 15 (2022) 776, http://dx.doi.org/10.3390/SU15010776. http://dx.doi.org/10.1016/J.SWEVO.2023.101467.
[26] J. Tang, G. Gong, N. Peng, K. Zhu, D. Huang, Q. Luo, An effective memetic [48] J. Xie, X. Li, L. Gao, L. Gui, A hybrid genetic tabu search algorithm for
algorithm for distributed flexible job shop scheduling problem considering distributed flexible job shop scheduling problems, J. Manuf. Syst. 71 (2023)
integrated sequencing flexibility, Expert Syst. Appl. 242 (2024) 122734, http: 82–94, http://dx.doi.org/10.1016/J.JMSY.2023.09.002.
//dx.doi.org/10.1016/J.ESWA.2023.122734. [49] K. Sun, D. Zheng, H. Song, Z. Cheng, X. Lang, W. Yuan, et al., Hybrid genetic
[27] B. Ji, S. Zhang, S.S. Yu, B. Zhang, Mathematical modeling and a novel heuristic algorithm with variable neighborhood search for flexible job shop scheduling
method for flexible job-shop batch scheduling problem with incompatible jobs, problem in a machining system, Expert Syst. Appl. 215 (2023) 119359, http:
Sustainability 15 (2023) 1954, http://dx.doi.org/10.3390/SU15031954. //dx.doi.org/10.1016/J.ESWA.2022.119359.

12
S. Nessari, R. Tavakkoli-Moghaddam, H. Bakhshi-Khaniki et al. Decision Analytics Journal 11 (2024) 100485

[50] W. Shao, Z. Shao, D. Pi, Lot sizing and scheduling problem in distributed [59] G.-G. Wang, D. Gao, W. Pedrycz, Solving multiobjective fuzzy job-shop schedul-
heterogeneous hybrid flow shop and learning-driven iterated local search al- ing problem by a hybrid adaptive differential evolution algorithm, IEEE
gorithm, IEEE Trans. Autom. Sci. Eng. (2023) http://dx.doi.org/10.1109/TASE. Trans. Ind. Inform. 18 (2022) 8519–8528, http://dx.doi.org/10.1109/TII.2022.
2023.3326301. 3165636.
[51] P. Schworm, X. Wu, M. Glatt, J.C. Aurich, Solving flexible job shop scheduling [60] E. Gheisariha, M. Tavana, F. Jolai, M. Rabiee, A simulation–optimization model
problems in manufacturing with Quantum Annealing, Prod. Eng. 17 (2023)
for solving flexible flow shop scheduling problems with rework and transporta-
105–115, http://dx.doi.org/10.1007/S11740-022-01145-8/TABLES/6.
tion, Math. Comput. Simulation 180 (2021) 152–178, http://dx.doi.org/10.1016/
[52] P. Schworm, X. Wu, M. Klar, M. Glatt, J.C. Aurich, Multi-objective Quantum
J.MATCOM.2020.08.019.
Annealing approach for solving flexible job shop scheduling in manufacturing,
J. Manuf. Syst. 72 (2024) 142–153, http://dx.doi.org/10.1016/J.JMSY.2023.11. [61] K.C.W. Lim, L.-P. Wong, J.F. Chin, Simulated-annealing-based hyper-heuristic for
015. flexible job-shop scheduling, Eng. Optim. 55 (2023) 1635–1651, http://dx.doi.
[53] V. Abu-Marrul, R. Martinelli, S. Hamacher, I. Gribkovskaia, Simheuristic al- org/10.1080/0305215X.2022.2106477.
gorithm for a stochastic parallel machine scheduling problem with periodic [62] M. Saqlain, S. Ali, J.Y. Lee, A Monte-Carlo tree search algorithm for the flexible
re-planning assessment, Ann. Oper. Res. 320 (2023) 547–572, http://dx.doi.org/ job-shop scheduling in manufacturing systems, Flex. Serv. Manuf. J. 35 (2023)
10.1007/s10479-022-04534-5. 548–571, http://dx.doi.org/10.1007/s10696-021-09437-4.
[54] R.L.C. Souza, A. Ghasemi, A. Saif, A. Gharaei, Robust job-shop scheduling under [63] E.M. Gonzalez-Neira, D. Ferone, S. Hatami, A.A. Juan, A biased-randomized
deterministic and stochastic unavailability constraints due to preventive and simheuristic for the distributed assembly permutation flowshop problem with
corrective maintenance, Comput. Ind. Eng. 168 (2022) 108130, http://dx.doi.
stochastic processing times, Simul. Model. Pract. Theory 79 (2017) 23–36,
org/10.1016/j.cie.2022.108130.
http://dx.doi.org/10.1016/j.simpat.2017.09.001.
[55] R.H. Caldeira, A. Gnanavelbabu, A simheuristic approach for the flexible job
[64] S. Hatami, L. Calvet, V. Fernández-Viagas, J.M. Framiñán, A.A. Juan, A
shop scheduling problem with stochastic processing times, Simulation 97 (2021)
215–236, http://dx.doi.org/10.1177/0037549720968891. simheuristic algorithm to set up starting times in the stochastic parallel flowshop
[56] R. Li, W. Gong, C. Lu, A reinforcement learning based RMOEA/D for bi-objective problem, Simul. Model. Pract. Theory 86 (2018) 55–71, http://dx.doi.org/10.
fuzzy flexible job shop scheduling, Expert Syst. Appl. 203 (2022) 117380, 1016/j.simpat.2018.04.005.
http://dx.doi.org/10.1016/j.eswa.2022.117380. [65] Y. Fu, H. Wang, J. Wang, X. Pu, Multiobjective modeling and optimization for
[57] C.A. Rodríguez-Espinosa, E.M. González-Neira, G.M. Zambrano-Rey, A simheuris- scheduling a stochastic hybrid flow shop with maximizing processing quality and
tic approach using the NSGA-II to solve a bi-objective stochastic flexible job minimizing total Tardiness, IEEE Syst. J. 15 (2021) 4696–4707, http://dx.doi.
shop problem, J. Simul. (2023) 1–25, http://dx.doi.org/10.1080/17477778.2023. org/10.1109/JSYST.2020.3014093.
2231877. [66] Y. Zhou, J.J. Yang, L.Y. Zheng, Hyper-heuristic coevolution of machine assign-
[58] J. Castaneda, X. Martin, M. Ammouriova, J. Panadero, A. Juan, A fuzzy ment and job sequencing rules for multi-objective dynamic flexible job shop
simheuristic for the permutation flow shop problem under stochastic and
scheduling, IEEE Access 7 (2019) 68–88, http://dx.doi.org/10.1109/ACCESS.
fuzzy uncertainty, Mathematics 10 (2022) 1760, http://dx.doi.org/10.3390/
2018.2883802.
math10101760.

13