Machine Learning for metallic coatings modeling & quality prediction

Data-driven AI, based on Machine Learning (ML) methods, can be used for High Entropy Alloys (HEAs) coatings’ modelling and quality prediction. Using experimental or historical data concerning HEAs coatings’ processes & materials, ML can be useful in cases that physics-based models do not exist, are not adequate or are not efficient e.g. due to new composition of alloys, new coatings processes’ parameters, multi-scale modeling, stochastic factors that influence quality, etc. Using ML methods, including Artificial Neural Networks (ANNs), Classification and Regression methods, ML models can be created to predict alloys’ structure (e.g. crystal structure) & coatings properties (e.g. hardness) and for several alloy compositions and temperatures. This ML modeling approach is being applied within the Horizon Europe project M2DESCO, and several ML methods & training algorithms have already been used and tested within the project with satisfactory results, including the Feed-Forward Multilayer Perceptron (FF-MLP) ANN using the Backpropagation training algorithm, and ML Classifiers, like the LBFGS Maximum Entropy Multiclass Classifier and the Light Gradient Boosting Machine (LightGBM) Classifier. Especially the LightGBM multiclass classifier provided the best alloys’ properties prediction results for 2 of the alloys used in the project, which were the Co-Cr-Fe-Mo-Ni and Al-Cr-Mo-Ti-W alloys, and for several compositions (molar fractions) of these alloys.