Uncertainty Management in Optimal Disassembly Planning Through Learning-Based Strategies

点击翻译按钮获取中文摘要

AbstractThis grant provides funding for the development of an analytical framework able to represent, analyze and eventually manage the extensive levels of uncertainty inherent in the emerging reverse logistics processes. The particular application area will be the optimal disassembly planning (ODP) problem, which constitutes one of the core activities in any reverse logistics process network. Specific steps in the proposed research program are: (i) the development of a formal representation of the decision making underlying the disassembly process, that will be able to explicitly characterize the involved uncertainties; (ii) the characterization of the optimal disassembly plan in the face of the aforementioned uncertainties; and (iii) the development of computationally effective and efficient algorithms able to analyze the impact of the process uncertainties and to eventually derive an optimized disassembly plan through observation of the process behavior. The representational framework sought in Step (i) will be based on the Petri net modeling framework, which is one of the standard frameworks for modeling discrete event system dynamics. Steps (ii) and (iii) will be based on Dynamic Programming theory and some emerging variants of it known as "Reinforcement Learning" algorithms. The effectiveness and efficiency of the derived techniques will be assessed and demonstrated through application on a number of prototypical case studies to be obtained from the relevant literature.If successful, the results of this research will lead to a better understanding of the uncertainties involved in the ODP problem, and they will provide a methodological framework for explicitly dealing with these uncertainties. In this way, they will further enhance the relevance of the undertaken optimization approaches to the particular ODP problem context, since they will enable the adaptation of the decision making process to the prevailing operational and economic conditions, and they will eventually lead to more environmentally benign and economically ludicrous disassembly plans. At the same time, the proposed research will enrich the field of computational learning theory since, by taking advantage of the particular problem structure, it is expected to offer a number of new implementation insights and techniques for recently emerged algorithms. Finally, the proposed research program and the dissemination of the derived results will increase the awareness of the manufacturing and logistics community regarding (i) the role of the uncertainty in contemporary production systems, and (ii) the availability of formal methods and tools to effectively deal with this uncertainty.