Machine-learning methods to predict the wetting properties of iron-based composites
Amir Kordijazi, Hathibelagal M. Roshan, Arushi Dhingra, Marco Povolo, Pradeep K. Rohatgi, Michael Nosonovsky
Abstract
The authors used three different methods of statistical data analysis to establish correlations between the water contact angle (CA) on ductile iron and composition, roughness (grit size), elapsed time between sample preparation and CA measurement and droplet size. The three methods are linear regression analysis (LRA), artificial neural network (ANN) model and multivariate polynomial regression analysis. It was established that the size of the water droplet is statistically insignificant, while correlations with the other three parameters were found. Surface roughness is the most important predictor of CA. A low coefficient of determination of the linear regression indicates that the correlation is non-linear. The ANN model showed much stronger predictive potential than LRA. The authors discuss the correlation with the experimental values of the CA and the physical mechanisms behind the observed trends. It is particularly promising that the ANN can be trained to predict the wetting characteristics. The application of machine-learning methods to synthesize new materials and coatings with desired surface properties, such as self-cleaning, is a technology that may become part of the emergent ‘triboinformatics’ field, related to the application of machine-learning methods to surface science and engineering.