Interpretable ML reveals physics phase changes in material corrosion resistance

bravo_abad · x · 2026-08-20

Kei Masuhara and coauthors use interpretable ML to study corrosion resistance in steam-coated aluminum. Addressing the challenge of correlated variables (thickness, morphology, etc.), they train a random forest model on physically motivated descriptors (R²=0.67). SHAP analysis identifies surface morphology and film thickness as dominant variables. Accumulated Local Effects (ALE) reveal something deeper: the role of morphology changes as the coating develops. In early-growth films, morphology reflects protective surfaces, but this shifts in later stages, successfully detecting a change in the governing physics.

Original post →

More from Research

Research channel →