Multi-way machine learning estimation of hardness of high-carbon martensitic stainless steels depending on the influences of austenitizing temperature, quenching media, and annealing temperature

Abstract

In recent years, parallel to the new industrial trends, the applicability of novel artificial neural network systems and machine learning techniques in the manufacturing and metallurgy sectors has risen due to the increasing competitiveness among steel, pipeline, and construction firms. This work focuses on estimating the hardness of high-carbon martensitic stainless steels depending on the heat treatment media, austenitizing temperature (AT), and secondary annealing temperature using different machine-learning methodologies for the first time in the literature. The attained outcomes indicated that the highest average hardness level was found in the medium-level tempering temperature and low AT in brine. As the tempering temperatures rose to the upper limits, measured hardness results diminished in both quenching media. Besides, according to all fold types and error metrics, the random forest (RF) machine-learning model was the strongest approach to estimate the final hardness values of the heat-treated samples whereas the Bayesian ridge (BR) was the poorest way.