Accelerated prediction of Vickers hardness of Co- and Ni-based superalloys from microstructure and composition using advanced image processing techniques and machine learning

• Published in: Acta materialia Vol. 196; pp. 295 - 303
• Main Authors: Khatavkar, Nikhil, Swetlana, Sucheta, Singh, Abhishek Kumar
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2020
• Subjects: 2-point correlations; Image processing; Machine learning; Principal component analysis; Superalloys; Vickers hardness; Vickers hardness; Image processing; Superalloys; 2-point correlations; Machine learning; Principal component analysis
• ISSN: 1359-6454; 1873-2453
• DOI: 10.1016/j.actamat.2020.06.042

•Establishing structure-property linkages in Ni- and Co- based superalloys.•Advanced image processing techniques employed to segregate different phases of superalloys with high accuracy.•Feature generation from scanning electron microscope images using statically derived n-point correlations.•Development of highly accurate machine learning based models to predict Vickers hardness of superalloys using compositions and n-point correlations.•Identification of important elements and microstructural characteristics for enhancement of Vickers hardness. [Display omitted] Superalloys constitute an important class of materials that are heavily employed in turbines of aircraft engines and power plants. Vickers hardness is an important mechanical property for selection of a material. In this work, we develop an alternate approach, which uses the microstructures to estimate the hardness of a Co- and Ni- based superalloys. Advanced image processing techniques coupled with data-driven machine learning (ML) are used to predict the Vickers hardness of these superalloys. Complex image derived properties such as 2-point correlations and compositions of superalloys are utilized as a feature to develop highly accurate ML model. The ML model trained through Gaussian process regression (GPR) using microstructure and compositional features show unprecedented accuracy with root mean square error (RMSE) of 0.14 and R2 of 0.98. Further analysis of the model is done to establish a relationship between the Vickers hardness with microstructural and compositional parameters. Addition of certain compounds such as iron and titanium can in general lead to increase in Vickers hardness, while addition of elements such as aluminium, tantalum and hafnium negatively affect the Vickers hardness. Most importantly, the developed ML model is trained on experimental data, as opposed to simulated data, making our approach directly applicable for accurate prediction of Vickers hardness.