A Intelligent Generation Approach of High-quality Initial Solutions for Assembly Sequence Planning of Complex Models

Assembly sequence planning for complex assembly models is often time consuming, and existing methods still have room for improvement in terms of quality. As a main method to solve the assembly sequence planning problem, the efficiency and effect of optimization algorithm are significantly affected by the initial solution. Therefore, an intelligent generation method for assembly sequence planning with high-quality initial solution based on reinforcement learning for complex models is proposed. Firstly, a complex assembly model containing many parts is taken as input, and its initial state is constructed by explosion view. Secondly, considering the part interference, direction consistency and connection stability in the assembly process, the Deep Q-Network (DQN) algorithm is used to generate high-quality assembly sequence, that is, the initial solution. Finally, aiming at reducing assembly direction change, reducing part interference, and improving connection stability, a mathematical planning problem is constructed, and the initial solution generated by DQN is used in the improved optimization algorithm to obtain high-quality optimal solutions, and the high-quality assembly sequence scheme of the model is output. Through experiments on a public data set composed of complex assembly models, the results show that compared with the initial solutions generated by random generation and optimization algorithm, the proposed method can generate high-quality initial solutions, significantly accelerate the convergence speed (by about 50%) and improve the convergence effect of the optimization algorithm (by about 75%), so as to obtain high-quality optimal solutions. In the process of generating high-quality initial solutions, the generation speed of this method is about 700% faster than that of the optimization algorithm, showing a significant speed advantage.