Deep-Q-network-enhanced aquila-equilibrium hyper-heuristic algorithm for preventive maintenance integrated disassembly line balancing involving worker redeployment

Abstract

The increasing production and replacement rates of modern commodities demand efficient recovery of end-of-life (EOL) products. This study promotes green and sustainable remanufacturing by investigating semi-automated disassembly lines. Preventive maintenance (PM), rarely studied in disassembly lines, is integrated with the line balancing problem and worker redeployment scheduling to construct a stable and reliable disassembly process. Additionally, this study considers the diversity of disassembly robots in PM scheduling, including their operating speed, energy consumption, and maintenance requirements, to enhance line efficiency and reduce carbon emissions. A mixed-integer programming model is proposed for the Preventive Maintenance-integrated Semi-Automated Disassembly Line Balancing Problem (PM-SADLBP), and is verified by an exact Epsilon-Constraint method. To solve this NP-hard problem, a Deep-Q-Network-enhanced Aquila-Equilibrium Hyper-Heuristic algorithm (DN-AEHH) is developed. A Hybrid Adaptive-length Triple-layer Real Encoding Approach and Self-repairing Decoding Mechanism are tailored to create an effective mapping between continuous solution space and discrete balancing and scheduling plans. Numerical experiments demonstrate that DN-AEHH outperforms five state-of-the-art algorithms across multiple problem scales, with a dominant rate of 83.01%. Additionally, managerial application shows 7.72% improvement in energy efficiency and 36.04% reduction in weighted cycle time with DN-AEHH to optimize PM. These findings provide practical guidance for line establishment and maintenance, supporting decision-making for managers with diverse preferences and operational contexts.