Development of Vickers hardness prediction models via microstructural quantification and machine learning approaches

Hardness is an important property in superalloys for high-temperature applications. In this talk, I will discuss about data-driven approaches to predict the Vickers hardness in Co- and Ni-based superalloys using machine learning (ML). Conventional and advanced image processing tools are implemented to quantify the microstructural variations with composition and processing conditions. Two different and noble image processing methods are implemented to quantify the scanning electron microscopy (SEM) images of Co- and Ni-based superalloys into descriptors for ML models. The conventional approach extracts geometrical features such as volume fraction, area, and perimeter of the phases from the microstructures. Whereas, the advanced approach uses statistics based 2-point correlations and principal component analysis (PCA) to quantify the microstructural variations. These microstructural descriptors combined with alloy compositions and processing conditions are used to develop Gaussian process regression (GPR) models to predict Vickers hardness. Both the methods, reveal a very good prediction of Vickers hardness with a higher R² greater than 95% and lower rmse less than 0.16 HV. Further analysis of the model presents numerous in-sights into structure-property relationships, which will be also discussed. The ML models developed can be generalized for any mechanical property of interest and can be utilized for accelerated development of new generation of high temperature superalloys.