Online optimization of automobile painting process parameters based on Q-learning

• 1、College of Management and Economics，Tianjin University，Tianjin 300072

Abstract：To improve the applicability of process parameters in automobile painting process, online optimization of process parameters based on Q-learning was proposed. Considering the factors that affect product quality under actual production conditions, knowledge can be learned from offline data and applied to parameter setting of online production process. First, a prediction model of painting quality based on BP neural network was trained by offline training data. And the actual painting environment was simulated by the model. Then, a process parameter optimization model based on Q-learning was established according to the quality prediction results. The process parameter optimization strategy was obtained through the interaction with the quality prediction model. Finally, the effectiveness and applicability of the proposed model were verified by an example in a large domestic automobile enterprise. The result shows that the MSE of the quality prediction model converges to 3.84×10-5. Q-learning algorithm converges to the optimal strategy. Q table can be used to retrieve the optimal action set under different initial process parameters, and the target process parameters can be achieved by executing several actions from the initial process parameters.

Keywords： Q-learning；Neural Network；Online Process Parameter Optimization；Automobile Painting