Design of low-density nickel-based superalloy for turbine blade applications using machine learning techniques

Abstract

New low-density nickel-base alloy has been designed using various classical machine learning algorithms such as linear regression, decision tree, support vector machine, random forest, naïve bayes, and artificial neural network (ANN) approaches for aerospace applications; it has the potential to enhance the thrust to-earth weight ratio significantly over existing manufacturing alloys. An alloying strategy was applied to achieve higher temperature microstructural stability and density reduction. The decision Tree model has been selected for the predicted density of new alloys based on itstest accuracy of 89.5 % and 0.057 RMSE cross-validations because it has better accuracy than other models. The k-NN classification model is chosen for phase prediction based on its trained, test, F1, and confusion matrix scores of 1,0.922,0.85 and low false values, respectively, compared with other models. Three alloys have been selected with low density with gamma and gamma prime phases from room temperature to high temperature. The density of all three alloys is 7.74, 7.76, and 7.78 g/cm3 . This newly designed low-density alloy density is comparatively lower than all six generations of single-crystal nickel-based superalloys. Keywords: Nickel-based superalloys, Chemical composition, Machine learning algorithms, and Thermo-Calc.