Machine Learning for Corrosion Resistance Alloys

Abstract

Determining the ability of corrosion resistance of alloys is of great importance to many applications. While one could call the pitting resistance equivalence number (PREN) a quantitative description of the relative corrosion resistance of stainless steels, it is an empirical value based on the chemical composition of the elements in alloys only, derived by mathematically fitting of experimental alloy data. It provides no scientific insight. In this work we collect a large number of experimental data published in the literature on corrosion of alloys to study the factors that determine the corrosion resistance of materials. A database that includes the alloy composition, electrochemical parameters as well as polarization testing parameters was therefore generated. Machine learning approaches (Lasso and Ridge regression) were used to find the unexpected correlations between aforementioned experimental parameters and the materials' properties. We further use the machine learning to prioritize the features that determine the corrosion resistance and anticipate that this work can be an useful tool for materials design.