Proximal Policy Optimization Algorithm for Multi-objective Disassembly Line Balancing Problems

As more and more end-of-life products are accumulated over time, there is an urgent need for their recycling. Disassembly is a key step to do so. In order to improve the operational efficiency of disassembly lines, a disassembly line balance problem (DLBP) has drawn many researchers’ attention. There are multiple factors that affect disassembly quality and efficiency, e.g., workstation allocation and disassembly revenue. This work addresses a multi-objective DLBP. We consider three objectives: maximizing the net profit of disassembly, minimizing the maximal gap of working time among workstations, and minimizing the risk of performing dangerous disassembly tasks. An improved proximal policy optimization is proposed for the multi-objective DLBP. Five real-world products are used to test its effectiveness and feasibility. Experimental results verify the strength of the algorithm by comparing it with an Actor-Critic algorithm.