Reinforcement learning fusion algorithm for assembly sequence plannings

Assembly sequence planning can essentially be understood as a problem of ordering product components. An optimal assembly sequence can shorten production cycles and reduce manufacturing costs. In this study, we propose a reinforcement learning algorithm—Greedy-QS—that combines Q-learning and Sarsa algorithms to address the assembly sequence planning problem. The approach involves first abstracting the assembly product into several boolean matrixbased mathematical models, which reduces the complexity of the assembly model and enhances the algorithm's operability for assembly sequence planning. The Greedy-QS algorithm is then employed to find the optimal solution. Experimental results demonstrate that the Greedy-QS algorithm converges faster and achieves higher reward values for the optimal assembly sequence compared to Q-SARSA algorithms, effectively combining the advantages of both Qlearning and Sarsa.