Research on prediction of welding deformation based on improved GA-BP neural network

Abstract

Abstract Welding deformation prediction can predict the deformation that may occur during the welding process, so that corresponding measures can be taken to control the deformation, so as to improve the manufacturing quality and driving safety of the car body. In this paper, the finite element analysis software is used to simulate the welding process of the load wheel flange plate of the special vehicle body. The load wheel flange plate is one of the main components connecting the load wheel and the body, which can reduce the impact caused by uneven ground or high speed driving, so as to protect the load wheel, axle and body system. In this paper, the finite element analysis model of the load wheel flange is established. Based on the finite element analysis method, the welding deformation data set under different welding voltage, welding current and welding speed is obtained. In this paper, the process parameters of orthogonal experimental design are used for welding, and the deformation of the flange plate of the load wheel after welding is measured by three-dimensional laser scanner. The simulation results of welding deformation are in good agreement with the experimental results, and the relative error is controlled within 9.4%. Therefore, the finite element simulation method in this paper can better reflect the actual welding deformation. In order to improve the efficiency and accuracy of welding deformation prediction, a deformation prediction model based on improved genetic algorithm optimized BP neural network(improved GA-BP)is proposed. The average absolute error MAE and the determination coefficient R2 are used as test parameters to evaluate the accuracy of the established model. The results show that compared with the unimproved BP neural network, the GA-BP model has a 0.04% increase in R2 and a 13.04% decrease in MAE in the performance of the test set, which has a high reference value for the subsequent improvement of welding quality. Finally, the improved GA-BP algorithm model is used to verify the engineering verification of the flange plate of the load wheel obtained by the welding experiment. Based on the improved GA-BP neural network, the error between the predicted value of the welding deformation and the average value of the experimental measurement is within 10%, and the predicted value is consistent with the experimental measurement value. The application value of the improved GA-BP neural network in engineering is verified.