A Resequencing Method For Automotive Painting Operations Considering Rework Based On Deep Q Network

With the development of information technology,intelligent manufacturing has become an inevitable trend in the development of manufacturing industry.In the field of automotive manufacturing,in order to meet the diversified market demands,flexible production methods characterized by multiple varieties and small batches are widely used.In the automotive paint shop,there is a high probability that the cars after painting have quality defects and need to be reworked and repainted,which disrupts the original paint plan and reduces the scheduling performance.How to flexibly schedule painting sequences to reduce costs and improve efficiency in the case of rework interference is one of the problems that need to be solved for automotive manufacturing enterprises.This thesis studies the resequencing problem for automotive painting operations with rework.In previous studies,the interference of rework factors was often ignored,and most of them focused on reducing the number of color changes as a single objective,without considering the relationship between the painting sequence and the downstream shop demand sequence.In addition,most of the solution methods were exact algorithms or heuristic algorithms,which lack flexibility for complex dynamic scheduling problems.Therefore,considering the perception and decision-making ability of reinforcement learning to the dynamic environment,a resequencing method for painting operations based on Deep Q Network(DQN)is proposed in this thesis.This method can determine the painting sequence online according to the production situation and achieve the better production goals.The specific work is as follows:First,under the background of practical application,this thesis analyzes the main production processes for cars in the automotive paint shop,and extracts the key processes to clarify the research problem framework.Then,to be more realistic,this thesis defines two evaluation criteria to measure the resequencing effect,one is to reduce the number of color changes,and the other is to reduce the sequence displacement between the painting sequence and the final assembly demand sequence.Aiming at the two objectives,a mathematical model for this problem is established.Next,the problem is transformed into a Markov decision process.According to the production situation,six key features are extracted to define the state,and according to the optimization goals and constraints of the problem,a reasonable instant reward function is designed.Furthermore,an algorithm framework is designed based on the DQN algorithm,which combines the online painting process and the offline learning process to train the agent to learn better strategies,so that the decision maker can dynamically decide the painting sequence online according to the current production situation.Finally,three sets of experiments are designed to verify the convergence and effectiveness of the method,and analyze the influence of different object weights combinations and constraints on the resequencing effect.Compared with the traditional scheduling method that inserts the rework cars based on rules,the optimization method in this thesis can flexibly respond to changes in the environment.When the rework occurs,the subsequent production sequence can be adjusted in time to achieve the optimal or near optimal scheduling in the current state and get a better resequencing effect.The innovation of the resequencing method in this thesis provides a new solution to the resequencing problem for painting operations with rework process.