Development of Vickers hardness prediction models via microstructural analysis and machine learning

Abstract

Superalloys are high-temperature materials with outstanding strength and resistance to corrosion. A prior knowledge about its hardness is essential for development of new superalloys for its applications in aeronautics and power industries. Determining the hardness of a material with experiments is usually a destructive process. In this study, using structural, compositional and processing condition parameters as descriptors, machine learning (ML) models are developed to predict Vickers hardness. We employed image processing tools, which extract structural descriptors such as volume fraction, area, perimeter and aspect ratio of the phases in the microstructures. Using the structural features in combination with elemental and processing information as features, a Gaussian process regression model for the prediction of Vickers hardness of superalloys is developed. The model gives an unprecedented accuracy with a minimum root mean square error of 0.15. The descriptors provide insights into structure–property relationships, which are important for designing superalloys with improved Vickers hardness. The proposed method for extracting features from microstructures and combining them with elemental and processing information can be extended to develop ML model for prediction of other mechanical properties of superalloys.