Extreme high accuracy prediction and design of Fe-C-Cr-Mn-Si steel using machine learning

Abstract

Solid solution strengthening theory is essential for designing steel with high microhardness. Experimental determination is quite time consuming and costly. It is necessary to develop an alternate approach to rapidly and accurately predict new solid solution strengthening theory for steel. In this study, a data-driven model combining machine learning (ML), firefly optimization algorithm (FA) and conditional generative adversarial networks (CGANs) were proposed to predict solid solution strengthening theory of Fe-C-Cr-Mn-Si steel. Three alloys were fabricated using cladding to validate the predict accuracy of the models. The results show that the trained support vector regression (SVR) model demonstrated the highest prediction precision for microhardness. The coefficient of determination (R2) value increased from 0.85 to 0.89 and root mean square error (RMSE) decreased from 0.39 to 0.31 after introducing the modified solid solution strengthening theory. The experimental validation revealed a minimum error of 1.17% between the predicted value and the experimental value. The investigation provides a valuable method to expedite design of Fe-C-Cr-Mn-Si steel with extreme high accuracy.

Keywords

• Fe-C-Cr-Mn-Si steel
• Machine learning
• Conditional generative adversarial networks
• Solid solution strengthening
• Firefly optimization algorithm