Data-Driven Machine Learning Analysis of Nonlinear Environmental Drivers of Marine-to-Inland Atmospheric Corrosion in Aluminum Alloys

Machine learning (ML) offers a powerful approach to analyze the complex, nonlinear interactions in industrial and marine environments where atmospheric conditions involving moisture and chemical pollutants accelerate degradation on different alloys. An accurate understanding of what affects corrosion rates is essential for enhancing the durability and maintenance of aluminum-based infrastructure. Thus, this study applies ML to analyze the corrosion rates of aluminum alloys (AA) 1060, AA 2024, AA 6061, and AA 7075 based on experimental data collected from diverse environments worldwide, with many sites in Hawaii. A CatBoost gradient boosting model was developed incorporating key environmental variables, including average time of wetness and deposition rates of ions such as F−, Cl−, Br−, , , , Li+, Na+, , K+, Mg2+, and Ca2+. Model performance was optimized using the Optuna framework for hyperparameter tuning. The model effectively captured nonlinear relationships in the corrosion behavior of aluminum alloys, revealing the influence of various atmospheric parameters and material properties on corrosion rate variability. Thus, these findings underscore the potential of ML to advance data-driven corrosion modeling, offering insights relevant to material selection, performance evaluation, and experimental design strategies aimed at developing a predictive model and providing insight into lifecycle management in uncontrolled atmospheric environments.