An Optimal Design of Radial Basis Function Neural Network Coupled with Genetic Algorithm for Weld Bead Geometry Prediction

This study investigates the prediction of weld bead geometry of MIG-CO2 welded butt joints made of EN-3A mild steel. To precisely construct correlations between input parameters and output parameters, the study combines Artificial Neural Networks (ANN) with a Genetic Algorithm (GA). Taguchi’s L25 partial-factorial design is used in the experiments to collect data that will be used to train a Radial Basis Function Neural Network (RBFNN) model. A real-coded GA is used to fine-tune the connection weights of the RBFNN considering experimental results. The study presents a thorough parametric analysis that uses Python programming to identify the best neural network architecture. In a comparison investigation, the suggested GA-RBFNN model predicts weld bead geometry accuracy better than traditional models like regression and Backpropagation Neural Network (BPNN).