Solution approaches for multi-objective sustainable job shop scheduling problem

In recent decades, sustainability has emerged as one of the foremost global challenges in industrial development. Success is no longer measured solely by cost reduction or productivity improvement; it now also requires the efficient use of resources and the minimization of waste. Effective resource management has become an increasing priority in modern manufacturing, where companies must balance operational efficiency with environmental sustainability. With growing global energy pressures, optimizing production processes while reducing energy consumption has become an urgent challenge. In response to these challenges, sustainable production scheduling has emerged as a key tool to integrate operational efficiency with environmental considerations. This approach not only focuses on economic performance but also aims to build production systems that align with the holistic requirements of sustainability. This thesis addresses a multi-objective sustainable job shop scheduling problem, in which each job consists of multiple operations that can be executed on machines operating at variable speeds. Unlike traditional approaches, this study integrates two energy-saving strategies: machine on/off control and speed scaling. When the idle time between consecutive operations on the same machine exceeds a predetermined threshold, the machine is turned off to reduce energy consumption. Otherwise, it remains in standby mode, consuming significantly less energy and avoiding excessive on/off cycles. Speed scaling strategy, meanwhile, is employed as a versatile tool to help meet the jobs' due dates, any tardiness lead to customer dissatisfaction. Operating machines at higher speeds reduces tardiness penalties but increases energy consumption, and vice versa. Accordingly, the problem involves two conflicting objectives: minimizing total energy consumption and minimizing total tardiness. To address this bi-objective problem, a mixed-integer linear programming (MILP) model was developed, and the augmented ε-constraint (Augmecon) method was applied to obtain exact Pareto-optimal solutions. However, as the problem size increases, the Augmecon method fails to find Pareto-optimal solutions within a reasonable time. Therefore, two specially designed metaheuristic algorithms are proposed: the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and the Multi-Objective Grey Wolf Optimizer (MOGWO). A total of 50 problem instances were generated and categorized into small, medium, and large sizes based on the number of operations. Of these, 12 small instances were randomly generated for this study, while 38 medium and large instances were adapted from well-known benchmark datasets frequently used in the literature. The performance of the proposed algorithms was evaluated using several widely adopted metrics in the literature, including Hypervolume (HV), Quality Metric (QM), CPU time, and the Number of Non-Dominated Solutions (NDS). The MOGWO algorithm outperforms NSGA-II in terms of solution quality (HV and QM) and the number of non-dominated solutions, particularly for small and some medium-sized problems. In contrast, the Augmecon method requires excessive computational effort as the problem size increases. Although it was able to generate Pareto solutions for 8 problem instances, it fails to obtain payoff matrices within the specified time limits for 4 out of 12 instances. NSGA-II, on the other hand, demonstrates high computational efficiency, particularly for large-sized problems; however, its performance in terms of solution quality declines. Interestingly, in larger problems, NSGA-II occasionally outperforms MOGWO in terms of HV and QM. In summary, MOGWO generally produces more diverse and higher-quality Pareto fronts, whereas NSGA-II offers faster convergence and lower computational time, especially in large-sized problem instances. Furthermore, the impact of due-date tightness on algorithmic performance is examined by evaluating three settings: tight, medium, and loose due dates. The results indicate that hypervolume values increases for both algorithms as due dates are relaxed (from tight to loose). Accordingly, solutions obtained under medium due dates outperform those associated with tight due dates, while solutions corresponding to loose due dates surpass those obtained under medium settings. Moreover, the MOGWO generates a larger number of non-dominated solutions under more relaxed due dates, whereas NSGA-II exhibits more fluctuating performance and shows instability in its results.