Prediction of laser-welded deformation using artificial neural networks

When predicting welding deformation of the laser-manufactured vehicles and aerospaces, analytical solutions or empirical formulas are not usually accessible in complex problems. Based on the inherent deformation method, a machine learning (ML) approach for predicting welding deformation of welded structures is proposed based on an artificial neural network (ANN). This method is a promising substitute for analytical, empirical, and finite element (FE) solutions due to its accuracy, easy-to-use, efficiency, and universality. First, the outputs of the ANN are determined via dimensionless analysis and comparison of numerical results, which are dimensionally independent. Then, based on the inherent deformation method, the training and validation sets of the ANN are generated through an elastic finite element analysis. At last, the structure of the ANN is determined by analyzing the ANN prediction accuracy with different hidden layers, numbers of neurons, and activation functions. The results show that the ML solutions are in good agreement with the FE results, verifying the effectiveness and generalization ability of the proposed method.