Machine learning-driven stability analysis of eco-friendly superhydrophobic graphene-based coatings on copper substrate

This study inspects the integration of machine learning (ML) techniques with materials science to develop durable, eco-friendly superhydrophobic (SHP) graphene-based coatings for copper. We employed various ML and regression models, including XGBoost, polynomial regression models, Random Forest (RF), K-Nearest Neighbours (KNN), and Support Vector Regression (SVR), to predict the stability of the contact angle (CA) under different stress conditions, such as NaCl immersion, abrasion cycles, tape peeling tests, sand impact, and open-air exposure. Our findings demonstrate that ensemble learning models, particularly XGBoost and Random Forest, outperform traditional regression techniques by effectively capturing nonlinear dependencies between stress parameters and CA retention. Higher-order polynomial regression models also exhibit strong predictive accuracy, making them well-suited for conditions where CA follows a well-defined trend. In contrast, SVR and KNN show limited generalization due to their sensitivity to hyperparameter selection and local interpolation effects, leading to weaker performance in datasets with high variability. ML-based algorithms predict CA values for tested coatings at longer term with respect to experimental tests, and underlined the beneficial effect of graphene incorporation in the coatings to extend the service life and preserve superhydrophobicity, overall reflecting the material's resilience under mechanical stress. The study highlights the importance of advanced predictive models, such as higher-degree polynomial regression and XGBoost, in capturing the complex relationships between variables influencing coating stability. Additionally, the integration of these models significantly accelerates the design and analysis process by reducing the reliance on time-consuming experimental testing.