Development of an Algorithm Based on an Artificial Neural Network to Predict the Thermal Profile of Weld Joints of Dissimilar High Strength Steels Using GMAW Process

Abstract

In developing an accurate modelling technique of thermal profile parameters when welding High-strength steel, an algorithm based on an artificial neural network (ANN) for predicting cooling time using Gas Metal Arc Welding (GMAW) was set up. The neural network developed has a 4-20-1 architecture with the input parameters of the voltage of the station ( U ), welding current intensity ( I ), welding speed ( V ), and heat input ( Q ) with the output parameter Cooling time ( ∆t8/5 ). A protocol has been developed with the MATLAB R2020a software containing three neural networks. The goal was to determine the neural network that has the lowest root mean square error (MSE). The results showed that the first system produced an MSE of 1.295 × 10 −3 and a regression R = 0.995 with a Relative error of 0 for 8 of the initial 14 data. The second system produced an MSE of 4.278 × 10 −3 with a regression R = 0.978 with 11/15 showing an error of 0. Finally, the third system, consisting of associated experimental data to the analytical data produced an MSE of 2.506 × 10−3 with a regression R = 0.972 with a slight difference between the input data and predicted data on all 29 points. The results obtained by the first two systems are satisfactory and developed neural networks can be found reliable for predicting cooling times of welded joints of steel high-strength.