Fuzzy reinforcement learning approach for supporting ubiquitous manufacturing facility selection in drone manufacturing
Yu-Cheng Wang, Toly Chen, Chi-Wei Lin
Abstract
Abstract Ubiquitous manufacturing (UM) systems of 3D printing facilities can be established to meet the explosive growth demand for drones, which has rarely been discussed in the past. To plan the operations in the UM system, a fuzzy reinforcement learning approach is proposed in this study. The fuzzy reinforcement learning approach employes the Q-learning algorithm to accelerate the prioritizing of criteria for 3D printing facility selection. Subsequent fuzzy technique for order preference by similarity to the ideal solution (FTOPSIS) application compares various 3D printing facilities to ensure that drone orders are correctly assigned to suitable UM partners. The fuzzy reinforcement learning approach has been applied in a case where twenty drone manufacturers sought help from multiple 3D printing facilities to manufacture drone parts using a UM system. According to the experimental results, the most critical criteria for evaluating the suitability of a 3D printing facility was “average product quality”, followed by “consistency with technology requirements”. In addition, by excluding 3D printing facilities already occupied by some drone part orders, the loads on the 3D printing facilities in the UM system could be balanced. Furthermore, in contrast to the genetic algorithm (GA) employed in previous research, the fuzzy reinforcement learning method was model-free and particularly efficient in unpredictable or random settings.