A Resequencing Method For Automobile Painting With Rework Based On Reinforcement Learning

In the wake of rapid growth of information technology and artificial intelligence,the automobile manufacturing industry is also transforming towards intelligence and informatization.For the purpose of catering to the private demands of customers,automobile manufacturing changed from large-scale production to flexible production.In the automobile mixed-flow production system,the production requirements and key performance indicators of each shop are different.As upstream and downstream shops,the paint shop and the assembly shop have different sequence preferences.At the same time,the rework in the paint shop will also affect the sequencing results.Therefore,when determining the painting sequence,how to meet the upstream and downstream production needs and responding flexibly to rework disturbances have become the difficult points.In order to solve the above-mentioned difficult points,this paper summarizes the previous studies.For most studies,only the color changing was considered.They did not deal with rework interference and secondary painting.At the same time,they mostly used precise algorithms and heuristic algorithms.These methods could not dynamically adjust painting schedules and other issues in real time.Therefore,this paper uses the deep reinforcement learning algorithm Actor-Critic,which has significant advantages in sequential decision making.It proposes a resequencing method for automotive in paint shop with rework based on reinforcement learning.First,building a double objective mathematical model.Based on the actual production process of the automobile paint shop,this paper abstracts and refines the problem.Then,it analyzes the sequence changes in the decision-making process and rework process,and clarifies the problem boundary,objectives and constraints.The mathematical model is established with the objectives of minimizing the number of color changes and the deviation between the painting sequence and the assembly demand sequence.Then,designing the Actor-Critic algorithm according to the mathematical model.This paper expresses the problem with a Markov decision process by defining the deviation between the painting sequence and the assembly demand sequence,and constructs the painting environment in the algorithm to meet the needs of the reworking.After that,designing the algorithm structure.Finally,conducting experiments to confirm the effectiveness of the Actor-Critic algorithm in this paper.This paper designs comparative experiments to evaluate the pros and cons of different algorithms.When not considering rework,compare with DQN and genetic algorithm.When considering rework,compare with DQN and rule-based direct insertion method.At the same time,the influencing factors of the algorithm in this paper are analyzed.After that,this paper verifies Actor-Critic algorithm can achieve better resequencing effect regardless of whether rework is considered.The experimental results show that when rework is not considered,the effect of the algorithm in this paper is increased by 4.7% compared to DQN.And when the rework is considered,the effect of the algorithm in this paper is increased by 9.6% compared to DQN.The results display that the method has more significant advantages in all respects.Considering the rework disturbance and the influence of the secondary painting during the resequencing,the method in this paper can dynamically adjust the follow-up painting plan to meet the production needs.At the same time,the strategy-based reinforcement learning algorithm Actor-Critic was introduced to solve the problem of a resequencing method for automotive in paint shop,and it achieves better results.It provides a new solution to the car resequencing problem in paint shop.The use of artificial intelligence methods to solve the resequencing problem has great significance for enterprises to realize intelligent manufacturing.It can improve the decision-making efficiency and the quality of production scheduling.It also provides new possibilities for intelligent information production.