Prediction of Microstructural and Mechanical Properties of Steel Welds with Artificial Neural Networks

Abstract The mechanical properties of the weld metals are dependent on the alloying elements and the microstructure of the weld metal. Neural network analysis is widely applicable in various fields, aiming to enhance efficiency thorough analysis. The application of artificial intelligence techniques for the rapid and accurate determination of physical properties offers a significant time, cost and labor advantage in industrial production processes due to the time consuming and costly nature of traditional methods. This study was aimed to investigate the relationships between structural and microstructural properties against alloying elements in SMAW weld metal using neural networks and to analyze them through this innovative methodology. In this study, the Levenberg-Marquardt algorithm is utilized to predict the physical properties of weld metal through artificial neural networks approach using 94 sets of weld metal composition and microstructural properties. The success rates of the modelling were found to be 93.14% for acicular ferrite, 95.92% for hardness, 94.17% for yield strength, and 96.32% for ultimate tensile strength. It is also feasible to make reverse predictions of weld metal composition in order to predict weld metal properties such as hardness, yield strength, acicular ferrite percentage and ultimate tensile strength within a range of alloying elements percentages, with a reasonable degree of accuracy.