Using multiple regression analysis to predict directionally solidified TiAl mechanical property

Abstract

The mechanical properties of TiAl alloy prepared by directional solidification were predicted through a machine learning algorithm model. The composition, input power, and pulling speed were designated as input variables as representative factors influencing mechanical properties, and multiple linear regression analysis was conducted by collecting data obtained from the literature. In this study, the R2 value of the tensile strength prediction result was 0.7159, elongation was 0.8459, nanoindentation hardness was 0.7573, and interlamellar spacing was 0.9674. As the R2 value of the elongation obtained through the analysis was higher than the R2 value of the tensile strength, it was confirmed that the elongation had a closer relationship with the input variables (composition, input power, pulling speed) than the tensile strength. By adding the elongation to the tensile strength as an input variable, it was observed that the R2 value was further increased. The tensile test prediction results were divided into four groups: The group with the lowest residual value (predicted value-actual value) was designated as group A, and the group with the largest residual value was designated as group D. When comparing the values of group A and group D, more overpredictions occurred in group A, while more underpredictions occurred in group D. Using the residuals and R2 values, the cause of the well-prediction was studied, and through this, the relationship between the mechanical properties and the microstructure was quantitatively investigated.