Q-Learning-Based Meta-Heuristics for Solving Disassembly Line Balancing Problems

With the growing emphasis on sustainable manufacturing, the recovery and reuse of end-of-life products have become increasingly important. Disassembly plays a critical role in this process by facilitating the recovery of reusable and recyclable components, thereby reducing resource consumption, waste generation, and associated environmental impacts. This work addresses disassembly line balancing problems with a fixed number of workstations for minimizing the maximum completion time (makespan). First, a mathematical model is developed that satisfies the priority and time interference constraints. Second, four meta-heuristics with a Q-learning strategy are proposed, including Q-learning-based variable neighborhood search (QVNS), Q-learning-based artificial bee colony algorithm (QABC), Q-learning-based genetic algorithm (QGA), and Q-learning-based particle swarm optimization (QPSO). In the iterations of meta-heuristics, Q-learning is employed to select competitive local search strategies. Finally, experiments are conducted for solving five large-scale instances, and the proposed algorithms are compared with existing algorithms. The statistical results and comparisons show that the meta-heuristics with Q-learning strategy have competitive performance on the tested instances. Among the proposed algorithms, QPSO achieves the best average performance on the tested instances.