Artificial neural network based calibration of Goldak heat source parameters in tandem plasma transferred arc process using finite element analysis

Abstract The paper proposes a novel approach for determining multiple heat source parameters to achieve the necessary accuracy in the numerical prediction for a parallel tandem plasma transferred arc process. An artificial neural network (ANN) approach as a supervised learning–based artificial intelligence (AI) was used to model the complex relationship between heat distribution and tandem/dual heat source parameters based on finite element analysis (FEA). This study presents the first ANN-assisted numerical simulation for multiple heat source optimization of the tandem plasma transferred arc process. The model demonstrated high accuracy in predicting and calibrating the tandem heat source parameters, based on systematic simulation trials of the tandem plasma transferred arc welding. The results of the tandem heat source determined by ANN-assisted FEA were verified using experimental data on melt pool size and substrate distortion. This verification considered metrics such as penetration width, penetration depth, melt pool length, melt pool half-width, and the absolute distortion (measured as the difference between minimum and maximum displacement), with relative errors of 7.0%, 7.6%, 0.4%, 1.9%, and 9.7%, respectively.