Atmospheric corrosion rate prediction of low-alloy steel using machine learning models

Abstract Corrosion mitigation is one of the indispensable needs in many industries and is currently being pursued by various methods like surface modification, corrosion inhibitor addition, and cathodic protection systems. Corrosion rate prediction can help in designing alloys with an optimized composition of materials such that it has a lower corrosion rate in the environment where they are exposed. Corrosion rate prediction can also help the manufacturers to plan the replacement of the sample used in advance. Machine learning, which is the science of making machines learn without being explicitly programmed and without using pre-determined equations, can help overcome challenges in predicting corrosion of various materials under a variety of environmental conditions. In this paper, three machine learning algorithms namely Support Vector Regression, Multiple Linear Regression, and Random Forest Regression are used to develop a Hybrid model to predict the corrosion rate of materials. Feature reduction is performed after feature importance calculation using Random Forest Regression model. The accuracy of the developed models were calculated using r 2 scores as an evaluation metric for different train-test split ratios. The input data for various conditions such as open, sheltered, coastal. Etc. are fed to the model and the performance was evaluated. The results show that the proposed Hybrid model outperforms the other baseline approaches with an accuracy of 91.46%, for predicting corrosion rate of materials.