Machine Learning Modeling of Hardness–Depth Distribution of Nitrided Steels

Nitriding is an essential thermochemical surface treatment that enhances fatigue, wear, and corrosion resistance of steels. This study aims to develop predictive models for hardness and hardened layer thickness of nitrided steels using machine learning techniques. A comprehensive database comprising over twenty thousand hardness values from various steel compositions and nitriding parameters was constructed. Several machine learning algorithms were assessed, leading to the selection of deep neural networks (DNN) for their superior performance. Two predictive models were developed: one incorporating input features such as alloy composition, nitriding temperature, time, and depth; and another based on physically informed parameters like nitrogen diffusion distance and nitrided layer thickness. The latter model demonstrated higher accuracy and applicability, accurately reproducing experimental hardness distributions under diverse alloy and processing conditions. Shapley additive explanations (SHAP) analysis revealed significant contributions from individual alloying elements, highlighting how nitride-forming elements enhance surface strength while high nitriding temperatures can decrease hardness. Additionally, the model elucidated carbon’s suppressive effect on Cr-nitride formation and the time dependence of layer growth. This hybrid, data-driven, and physics-informed framework presents an explainable and versatile tool for optimizing the surface properties of nitrided steels.