Machine Learning-Based Decentralized Collaborative Production Plannin...

Theremarkable success of sharing economy (SE) platforms in the B2C sector has ledto increased attention from both practitioners and researchers seeking to applythese models to the B2B sector. Studies have already demonstrated, combinedwith cloud manufacturing (CMfg) and Additive Manufacturing (AM), SE frameworkscan considerably reduce costs, while increasing the responsiveness of a supplychain. CMfg usually connects many participants who want to collaborativelyshare resources or exchange jobs effectively, however, and, as a result, thereis a need to design an efficient resource allocation mechanism. Promisingapproaches for this environment are decentralized auction frameworks, used toexchange jobs to decrease production costs of operations efficiently andeffectively. In these approaches, agents need to determine the bids offorwarded jobs offered by an auctioneer, then finding the optimum joballocation via a combinatorial auction. In conventional approaches, bids aredetermined by solving numerous production planning problems with commercialsolvers or heuristics. Both approaches lack suitability for a large-scale CMfgplatform, however, as they may find inefficient solutions or are tootime-consuming. This study closes this research gap by applying supervisedmachine learning (ML) models to estimate production costs and report bids,effectively. We investigate the effectiveness of over 40 ML models and comparethe estimated objective value to the exact solution and a well established heuristicfor a single machine AM production planning problem. We then demonstrate theeffectiveness of the most accurate ML model on a decentralized framework, basedon a truthful combinatorial reverse auction. This framework allows machines toautonomously exchange jobs over a CMfg platform to reduce overall andindividual production planning costs. Our enhanced approach significantlydecreases computational time, while delivering efficient results.