Statistical and ANN Modeling of Corrosion Behavior of Austenitic Stainless Steels in Aqueous Environments.

Abstract

This study applies statistical approaches utilizing linear regression and artificial neural networks (ANNs) to predict the corrosion behavior of austenitic stainless steels (316L, 904L, and AL-6XN) under various environmental conditions. The environmental variables considered include temperature (30-90 °C), chloride ion concentration (20-40 g/L), and pH (2-6). Analysis of variance (ANOVA) confirmed that the input variables, including the Pitting Resistance Equivalent Number (PREN ranging from 24 to 45), significantly affect the critical pitting potential. The influence of the variables was ranked in the order: PREN, temperature, pH, and chloride ion concentration. A linear regression model was developed using significant factors and interactions identified at the 95% confidence level, achieving a predictive performance with R² = 0.789 for critical pitting potential. To predict potentiodynamic polarization curves, an ANN based on supervised learning with backpropagation was employed. The ANN model demonstrated a remarkably high predictive performance with R² = 0.972 in complex corrosion environments. The predicted polarization curves reliably estimated electrochemical characteristics such as corrosion current, corrosion potential, and pitting potential. These results provide a valuable tool for predicting and understanding the corrosion behavior of stainless steels, which can aid in corrosion prevention strategies and material selection decisions.