Machine learning prediction of the mechanical properties of γ-TiAl alloys produced using random forest regression model

Abstract

The mechanical properties of a directionally solidified (DS) TiAl alloy were predicted through a random forest regression (RFR) machine learning algorithm. The prediction results were evaluated using the R2 value. As a result, the R2 values of prediction for the tensile strength, elongation, nanoindentation hardness, and interlamellar space were 0.9336, 0.9902, 0.8104, and 0.9810, respectively. To observe the correlation between the microstructure and mechanical properties, RFR prediction was conducted with a double input variable. This yielded a R2 value of tensile strength of 0.9856, which was higher than the tensile strength derived with a single input variable. The R2 value of nanoindentation hardness increased to 0.9902, which was higher than the nanoindentation hardness value with a single input variable. Through the use of the double input variables, the relationships among tensile strength and elongation, nanoindentation hardness, and interlamellar space were observed. Using feature importance, which could not be obtained in a previous study using the MLR algorithm, it was possible to determine which input variable had the most efficiency with respect to the output variable. Based on these research results, the speed and accuracy of new alloy development specifically in the design and processing, can be increased. In addition, metallurgical research on the relationship between the interlamellar space and mechanical properties was conducted, and the relationship was verified through the results of machine learning training.

Keywords

• Directionally solidified TiAl alloy
• Microstructure control
• Tensile strength
• Interlamellar space
• Prediction
• Random forest regression