A new artificial neural network model for prediction of fatigue strength and yield strength of various steel grades

Abstract

The prediction of mechanical properties of steels traditionally relies on experimental testing and empirical modeling, which are time-consuming and resource-intensive. Recent advances in artificial intelligence (AI) and machine learning (ML) enable efficient estimation of material properties directly from chemical composition data. In this study, two Artificial Neural Network (ANN) models were developed to predict key mechanical properties of steels—fatigue strength and yield strength—using chemical composition as the sole input. Each network employed a feed-forward backpropagation architecture with Bayesian regularization and was trained on a dataset comprising 200 steel grades, including structural, high-strength, and tool steels. The models were evaluated through regression analysis and further validated using experimental data for three representative steels: S355J2 + N, S690QL, and X37CrMoV5-1. The fatigue strength model achieved correlation coefficients of R = 0.895 (training), 0.833 (validation), and 0.893 (testing), with an average prediction error of approximately 9%. The yield strength model reached R = 0.835 (training), 0.829 (validation), and 0.864 (testing), with a mean absolute percentage error of 3.97%. The obtained results demonstrate that ANN-based models can reliably capture the relationship between chemical composition and mechanical properties of steels, supporting preliminary material screening and property estimation. The proposed approach provides a basis for the development of data-driven tools in digital materials engineering.