A comparative study on the performance of artificial neural networks and regression models in modeling the heat source model parameters in GTA welding

Abstract Welding is one the most important joining methods Analytical and FEM techniques are commonly employed to model various welding processes. The heat source model is a key part of welding simulation. Proper determination of heat source parameters is one of the main factors in the accuracy of the welding simulation. In this study, artificial neural networks and regression modeling have been employed to establish the relationships between welding input variables and the parameters for the Goldak heat source model. The 27 data needed for modeling has been gathered based on full factorial design. While ANN slightly outperforms regression, both ANN and second order regression functions have good agreements with actual experiments. The approach presented here may be used to accurately specify heat source parameters for any given set of welding process variables.