Deep learning assisted prediction of microstructure and wear behaviour in plasma nitrided 316 L stainless steel

This study investigates the wear resistance enhancement technique of AISI 316 L stainless steel through plasma nitriding and introduces a hybrid experimental-computational framework that employs Deep Learning (DL) for wear prediction and microstructure classification. The plasma nitridation at a controlled 420 °C created a nitrogen-rich gradient layer that significantly improved the surface properties. Optical microstructural analysis, and also hardness measurements showed that a nitrogen enriched expanded austenite, the so called S- phase layer formed. The layer supports the surface hardness increasing, from 397.7 HV to 634.9 HV. Due to this, nitrogen supersaturation, inside the austenitic lattice. The characterization of the microstructural layers performed by either conventional or advanced techniques showed no presence of dangerous carbide or nitride precipitates in the S-phase which contributed to its hardness. CNN could discriminate between untreated and retreated microstructure images with a classification accuracy of 98.3%. Besides, wear loss estimation was done using Multilayer Perceptrons (MLP) and Long Short-Term Memory (LSTM) yielding R² scores of 0.974 and 0.982 respectively. The results illustrate the synergy of advanced surface modification techniques and AI-based analysis in providing a robust and scalable solution for process optimization, performance forecasting, and real-time diagnostics in tribological applications.