Selective disassembly planning considering process capability and component quality utilizing reinforcement learning

Disassembly is a crucial process for achieving circular products, enabling function recovery, material reuse, and recycling. Disassembly planning is complex due to epistemic uncertainty associated with each unique product's conditions, i.e., quality and aleatoric uncertainty about the capabilities of available resources and processes, and the cost benefits of associated operations impede planning. Therefore, the disassembly is intended to result in keeping the maximum value for the disassembled units of the product. In selective disassembly, the specification of the units of the product to be disassembled is acquired, leaving the rest of the product intact. The benefit of selective disassembly is to minimize waste during dismantling and maximize the reuse of the disassembled components for economic and ecological sustainability. The challenges in disassembly sequence planning include product complexity, operational and technological process capabilities, and the lack of information regarding the product architecture. For this complex planning task, limited studies have been performed on incorporating process capabilities with respect to the operations resources for selective disassembly planning. In this paper, an approach for optimal sequence planning of the selective disassembly process is put forward, taking into account multiple constraints, i.e., quality, time, and process capability. The intelligent planning approach takes advantage of a reinforcement learning model to handle the complexity of the planning problem. The approach has been implemented and tested on an industrial reference assembly. The result shows that the complex task of selective disassembly planning can be efficiently performed utilizing the proposed approach.