A New Cooling-Rate-Dependent Machine Learning Feature for the Design of Thermally Sprayed High-Entropy Alloys

Highly accurate machine learning (ML) approaches rely heavily on the quality of data and the design features that are used as inputs to the model. The applicability of these methods for phase formation predictions is questionable when it comes to the design of thermally sprayed high-entropy alloy (HEA) coatings using gas or water atomized powders as feedstock material. Phase formation from liquid state depends on the cooling rate during atomization which is several orders of magnitude higher when compared to arc-melted as-cast HEAs. In addition, during plasma spray the powder melts in the flame and re-solidifies under different cooling rates during deposition. To our knowledge, almost all ML algorithms are based on available datasets constructed from relatively low cooling rate processes such as arc melting and suction casting. A new approach is needed to broaden the applicability of ML algorithms to rapid solidification manufacturing processes similar to gas and water atomization by making use of existing data and theoretical models. In this study the authors introduce a cooling-rate-dependent design feature that can lead to accurate predictions of the HEA powder phase formation and the subsequent phases found in the spray coated materials. The model is validated experimentally and also by comparing the predictions with existing coating related data in the literature.