Minimization of total setup time on a bottleneck machine using metaheuristic methods

In today's highly competitive industrial environment, improving the efficiency, flexibility, and cost-effectiveness of production systems has become imperative for companies aiming to achieve sustainable growth and operational excellence. Among the critical factors influencing overall manufacturing performance, the optimization of production scheduling and setup times plays a pivotal role. Setup times, which refer to the time required for machines to switch from producing one product to another, directly impact machine utilization rates, production throughput, and operating costs. Minimizing these times ensures more stable and predictable manufacturing processes and contributes to maximizing resource efficiency. Moreover, industries seeking to adopt lean manufacturing principles and just-in-time strategies consider setup time optimization as a crucial enabler for reducing waste and improving responsiveness. This master's thesis investigates the potential of metaheuristic optimization algorithms in minimizing setup times arising from product transitions in a tire manufacturing line. The specific problem under study emerges from the variability in setup times based on different mold and compound combinations required for each type of tire. In many manufacturing environments, such transition-related setup times are often overlooked in sequencing decisions, which results in production bottlenecks, idle times, and increased operational costs. To address this issue, a combinatorial optimization approach has been adopted, wherein the product sequence on a bottleneck machine is optimized to minimize total setup durations. Such sequencing problems are known to be NP-hard due to the factorial growth in the number of possible permutations, making heuristic and metaheuristic methods indispensable for obtaining feasible solutions in reasonable timeframes. Three well-established metaheuristic algorithms were selected for this purpose: Tabu Search, Simulated Annealing and Particle Swarm Optimization. These methods are known for their effectiveness in solving complex, large-scale problems with vast solution spaces, such as scheduling, routing, and sequencing. Unlike conventional optimization techniques that often struggle with combinatorial complexity and local optima, metaheuristic algorithms can explore diverse regions of the solution space and converge toward near-optimal solutions within a reasonable computational time. The characteristics of each algorithm offer different strengths: Tabu Search uses adaptive memory to avoid cycles, Simulated Annealing uses a probabilistic mechanism to escape local optima, and Particle Swarm Optimization leverages swarm intelligence principles to dynamically update potential solutions based on individual and collective experiences. Each algorithm was implemented using real-world production data obtained from a tire factory. The dataset included daily production plans, setup time matrices between different tire types based on mold and compound transitions, and historical supplier-based sequencing outcomes. The optimization objective was to find the best product sequence for each day over a ten-day horizon that minimized total setup times on the critical bottleneck machine. A comparative analysis was performed by calculating the total setup time resulting from each algorithm's output and comparing it to the baseline sequence currently used by the factory, which is based on supplier delivery order. To ensure fairness in comparison, all algorithms were calibrated using consistent stopping criteria and executed over multiple runs to capture average performance metrics. The results of the experimental study were highly promising. The Simulated Annealing algorithm outperformed the other two approaches by achieving a 13.96% reduction in average setup time compared to the supplier-based sequence. Particle Swarm Optimization followed closely with a 12.83% improvement, and Tabu Search provided an 10.71% reduction. These improvements suggest that even small enhancements in sequencing can translate into significant operational benefits, including shorter lead times, higher throughput, and reduced labor and energy costs. Additionally, graphical analysis showed that the Simulated Annealing algorithm consistently achieved the best performance on seven out of ten days, further validating its robustness in practical scenarios. To determine whether the observed differences between the metaheuristic approaches and the traditional method were statistically significant, the non-parametric Mann-Whitney U test was applied. The test results confirmed that the reductions in setup time obtained through the metaheuristic algorithms were statistically significant at a 95% confidence level (p < 0.05). Notably, although Simulated Annealing yielded the best average performance, the difference in performance between Simulated Annealing and Particle Swarm Optimization was not statistically significant, indicating comparable levels of optimization capability between these two algorithms. However, both performed significantly better than the Tabu Search algorithm, which, while effective, was more sensitive to initial conditions and neighborhood structures. The findings of this research provide strong evidence supporting the applicability of metaheuristic algorithms in real-world production scheduling problems. In particular, their integration into decision support systems and production planning software offers manufacturers the opportunity to dynamically generate optimized schedules that adapt to changing production requirements and constraints. Furthermore, these algorithms are particularly well-suited for scenarios involving sequence-dependent setup times, complex product configurations, and high variability in demand. Given the evolving nature of Industry 4.0, the ability of metaheuristics to accommodate real-time data inputs and operate in uncertain environments makes them highly relevant for smart manufacturing systems. In addition to the operational advantages demonstrated, this study contributes to the broader academic literature on manufacturing optimization by presenting a comparative analysis of three prominent metaheuristic techniques under a real-world case study. The research highlights not only the practical benefits of each algorithm but also the importance of selecting context-appropriate methods based on problem characteristics such as solution space size, setup time variability, and required computational efficiency. For instance, while Simulated Annealing is advantageous in navigating rugged solution landscapes, Particle Swarm Optimization may be more suitable for problems with smoother fitness functions and well-defined global optima. In conclusion, this thesis demonstrates that metaheuristic-based scheduling solutions, particularly Simulated Annealing and Particle Swarm Optimization, can significantly improve setup time performance in tire manufacturing. Given their robust nature and adaptability, these algorithms offer valuable tools for production engineers and operations managers striving for leaner and more responsive manufacturing systems. Future work may explore hybrid algorithmic approaches or real-time applications integrated with manufacturing execution systems to further enhance decision-making and production efficiency in dynamic industrial environments. Moreover, incorporating machine learning models to predict setup times and integrating multi-objective optimization to balance trade-offs between setup time, energy consumption, and delivery deadlines may offer even more comprehensive solutions. As the manufacturing sector continues to digitize, the synergistic use of artificial intelligence and metaheuristic optimization will likely shape the next frontier in production scheduling and resource management.