Beta transus prediction of titanium alloys through integration of artificial neural network and multifactor dimensionality reduction analyses

Abstract

To predict beta transus of titanium alloys, artificial neural network (ANN) and multiple linear regression (MLR) models were developed based on the alloy composition. Mo, V, Zr, Cr, Fe, Al, Si and O were the principle determinants of beta transus. The ‘r²’ (92.0% vs. 90.7%) and mean predicted error [training (1.4% vs. 2.8%) and testing (2% vs. 2.4%)] pattern in ANN and MLR models suggest superior performance of ANN model. Multifactor dimensionality reduction analysis showed interactions among Al, O and Cr, which were confirmed by the ANN model. The positive association of beta transus with aluminium equivalent and inverse association with molybdenum equivalent was demonstrated.