Prediction of laser-spot-weld shape by numerical analysis and neural network

The finite element method (FEM) and neural network were applied for predicting the bead shape in laser spot welding of type 304 thin stainless steel sheets. The parameters of pulsed Nd:YAG laser spot welding such as pulse energy, pulse duration, sheet metal thickness, and gap between sheets were varied for various experiments and numerical simulations. The penetration depth and nugget size of spot welds measured for specimens without gap were compared with the calculated results to verify the proposed finite element model. Sheet metal thickness, gap size, and bead shape of the workpiece without gap were selected as the input variables for the back-propagation learning algorithm of the neural network, while the bead shape of the workpiece with and without gap was considered as its output variable. Various combinations of stainless steel sheet metal thickness were considered to calculate the laser-spot-weld bead shape of the workpiece without gap, which was then used as the input variable of neural network to predict the bead shape for various gap sizes. This combined model of finite element analysis and neural network could be effectively applied for the prediction of bead shapes of laser spot welds, because the numerical analysis of laser spot welding for the workpiece with gap between two sheets is highly limited.