Application of machine learning for the classification of corrosion behavior in different environments for material selection of stainless steels

Corrosion behavior prediction of materials in any given environmental condition is important to minimize time-consuming experimental work to avoid failures and catastrophes in industry. Supervised machine learning (ML) techniques are recently explored to predict corrosion behavior. However, there is still a lack of research that proposes a model capable of predicting the corrosion behavior of a wide range of stainless steel grades in varying environments, including acids, bases, and salts. Moreover, conventional experimental approaches are often insufficient in identifying the most influential factors in the corrosion process due to its multivariate and non -linear nature.This study presents the development and evaluation of multiple ML models in predicting the corrosion behavior of different types of stainless steel in varying environments. The prediction performance of four ML algorithms, decision tree (DT), support vector machine (SVM), random forest (RF), and bagging classifier, were compared. Initially, the algorithms were fitted to a dataset based on the type of electrolyte (Dataset No. 1) and then modeled on a modified dataset (Dataset No. 2) in which the types of electrolytes were replaced with their critical ions contributing to corrosion reactions. The Bagging classifier achieved the highest prediction accuracy of 94.4% for Dataset No. 1, while the DT model was the most suitable for Dataset No. 2 with a testing accuracy of 93.95%. The application-driven approach of confusion matrix analysis to select the model's capacity to correctly identify severe and poor corrosion behavior confirmed that Bagging and DT classifiers are the most suitable ML algorithms for predicting corrosion behavior in Dataset No. 1 and No. 2, respectively. Furthermore, the feature importance analysis identified hydrogen and sulfide concentrations in corrosive environments, as well as the sum of the number of alloying elements, as the most influential factors, contributing up to 77.8% to the corrosion behavior. As a result, users of stainless steels can leverage this model to predict the corrosion behavior of specific materials in specific environments, facilitating informed material selection for various applications, without the need of lengthy and costly experiments.