Weld Bead Geometry Prediction of Additive Manufacturing Based on Neural Network

Abstract

Due to the large number of welding process parameters and the model output parameters, the prediction result of weld bead geometry using neural network in wire and arc welding test is not ideal. Therefore, multi-population genetic algorithm is introduced to optimize the initial weights and threshold of neural network in this paper, and this can avoid premature convergence. The performance between back propagation (BP) neural network and improved BP neural network prediction model are compared based on the designed verification samples, and the results show that the improved BP neural network model has higher accuracy. In addition, the improved BP neural network model is used to simulate the weld process, and the influence of the weld process parameters on weld bead geometry is revealed.