Effect of Surface Modification Using GTAW as Heat Source and Cryogenic Treatment on the Surface Hardness and Its Prediction Using Artificial Neural Network

Abstract

High-wear-resisting EN 31 bearing steel has been widely used to make components such as roller bearing, ball bearing, spline shaft, and other components like tiller blades, punches and dies are subjected to severe abrasion to require high surface hardness. To obtain high surface hardness, EN 31 steel is usually surface modified using various methods like conventional heat treatment (591 HV), cryogenic treatment (688 HV) and GMAW. But, there are no studies on surface modification of EN 31 using gas tungsten arc (GTA) heat source followed by cryogenic treatment. To improve the hardness further, surface alloying using gas tungsten arc followed by cryogenic treatment is done in this study. EN 31 steel is surface-hardened by using GTA heat source by varying the welding current, electrode tip angle and shallow and deep cryogenic treatments (SCT & DCT) by varying soaking time and temperature. Microstructures were studied and microhardness was measured. It is found that cryogenic treatment leads to formation of carbide particles in martensite matrix with reduced retained austenite which improves the microhardness from 258 to 898 HV after SCT and 1856 HV for DCT. Further, in this work, a back-propagation artificial neural network (ANN) which uses gradient descent learning algorithm is used to predict the microhardness of EN 31 steel for the entire ranges of parameters used in the experiments. The ANN model is trained and tested using 200 experiments done. The input parameters of the ANN model are 4 variables (welding current, electrode tip angle, cryogenic soaking time and temperature). Using MATLAB, a programme was developed and by varying the transfer function (tansig and logsig) different ANN models are constructed for the prediction of microhardness. This study shows that back-propagation artificial neural network (ANN) which uses gradient descent learning algorithm is very efficient for predicting the microhardness of EN 31 steel.