Learning-Based Disassembly Process Planner for Uncertainty Management

As product lifecycles are getting shorter and shorter, manufacturers are facing a great deal of economic and political pressure to reclaim and recycle their obsolete products. Disassembly, as one of the natural solutions, is of increasing importance in material and product recovery. However, this process is fraught with a high level of uncertainty (e.g., variations in product structure and condition, and human factors). The development of an effective modeling and management tool for such involved factors is critical in moving disassembly toward a more efficient and automated regime. This paper builds upon our previous work to undertake this problem. More specifically, a fuzzy Petri net model is introduced to explicitly represent the dynamics inherent in disassembly. Instead of presuming the pertinent data in the model is already known, a self-adaptive disassembly process planner and associated computationally effective algorithms are designed in a way to: 1) accumulate the past experience of predicting such data and, at the same time, 2) exploit the ldquoknowledgerdquo captured in the data to choose the best disassembly plan and improve the overall disassembly performance. To ensure the robustness of the learning procedure, variable memory length is further introduced. The proposed methodology and algorithms are illustrated through the disassembly of a batch of personal computers in a prototypical disassembly system.