Towards Large Scale Collaborative Production Planning in Additive Man...

CMfg is a key enabler of collaborative production (CP) systems, especially in combination with Additive Manufacturing (AM). Recent studies have shown that CP can reduce costs in an AM operation significantly. However, existing solution methods are limited to medium size instances. Nevertheless, such an environment usually leads to large-scale instances as many participants must be matched and scheduled. This study aims to close this research gap by proposing a hybrid Mixed Integer Linear Programming (MILP) - Machine Learning (ML) framework. This approach is inspired by an auction-based framework widely researched in logistics and consists of five steps. In the first step, machines autonomously select jobs from the existing production plan to forward them to the CMFg platform. Then the platform creates promising bundles of the transferred parts. In the bidding step, manufacturing machines autonomously report the marginal costs of the packages. The winner of the bundles is determined via a combinatorial reverse auction, and the costs of the reallocated bundles are shared. In the bidding step, the marginal costs are determined by solving a production planning problem for every bundle and is subsequently the most time-demanding step. Our ML-enriched framework eases this problem by splitting this step into two sub-steps. In the first one, the costs of the bundles are estimated via a supervised ML model. The estimated costs are reported to the auctioneer, requesting an accurate report to the most promising bidders in the second sub-step. In our experiments, we investigate several ML models and demonstrate the most effective one for cost estimation. We also show that our enhanced approach reduces computational time significantly.