The application of machine learning (ML) has accelerated the process of developing novel materials for a wide variety of applications. ML algorithms are used to learn from patterns in past data and use it to predict future events. In a new study, IISc researchers led by Abhishek Kumar Singh, Associate Professor at the Materials Research Centre, have developed highly accurate ML models to predict the Vickers hardness of nickel and cobalt-based superalloys.
Superalloys are materials possessing excellent mechanical strength and creep resistance at high temperatures, high surface stability and high corrosion resistance. Due to these properties, they are heavily employed in the aerospace, marine, chemical and petrochemical industries.
Top row: Images representing the 2-point correlations and the variance captured by the principal components obtained using electron micrograph. Bottom row: Results of the machine learning model developed using 2-point correlations and compositions.
To develop the ML model, a database was generated initially, comprising of the microstructures, compositions and Vickers hardness of several cobalt and nickel-based superalloys. SEM microstructures were denoised and thresholded to obtain binary microstructure. The binary microstructure was then used to calculate 2-point correlations. 2-point correlations are statistically derived parameters, which give the probabilities of finding different phases in microstructure at a specified distance. Further, principal component analysis (PCA) was performed on these correlations to select the most dominant correlations. These PCA-derived correlations along with composition of the superalloys are used as descriptors for building the ML models.
The approach developed in this study can be generalised for any material property, making it highly useful and adaptive.
N. Khatavkar, S. Swetlana, A. K. Singh, Accelerated prediction of Vickers hardness of Co- and Ni-based superalloys from microstructure and composition using advanced image processing techniques and machine learning, Acta Materialia, In press, (2020)