AI/ML Modeling for Aluminum Alloy Corrosion Resistance

Corrosion costs the global economy over $2.5 trillion each year, eroding cars, bridges, ships, and aircraft. For decades, developing corrosion-resistant alloys has relied on slow trial-and-error methods. The sheer number of possible combinations makes full testing impossible. We worked with NNL to explore how AI and machine learning (ML) could offer a faster route. ​

​

By using AI and ML to predict and optimize corrosion resistance in aluminum alloys, our hope is to demontrate how scientists can accelerate discovery far beyond traditional trial-and-error methods, identifying compositions that last longer and perform better in demanding environments such as aircraft, ships, and coastal structures. The practical implications are significant: safer and more reliable aerospace and marine systems, reduced maintenance costs, and extended lifespans for critical infrastructure. 

Our team compiled a comprehensive dataset of aluminum alloys from open sources, including corrosion rates under various environmental conditions (like different water chemistries) and detailed alloy compositions. We then standardized this data for use in ML models. Two modeling approaches were explored: a “forward” approach, where the model takes an alloy’s composition plus the environment and predicts the corrosion rate; and an “inverse” approach, where the desired corrosion performance is given and the model suggests what alloy composition could achieve. For the forward problem, the researchers tested three different ML algorithms: a Random Forest, a feed-forward Neural Network, and Gaussian Process Regression (GPR). These models were trained to recognise patterns in how alloy elements and conditions affect corrosion. For the inverse problem, they built a regression model (including ensemble methods like random forests and gradient boosting) to work backwards from corrosion rate targets to recommended alloy ingredient percentages.