Modeling the Correlation of Composition-Processing-Property for TC11 Titanium Alloy Based on Principal Component Analysis and Artificial Neural Network

Abstract

In the present investigation, the correlation of composition-processing-property for TC11 titanium alloy was established using principal component analysis (PCA) and artificial neural network (ANN) based on the experimental datasets obtained from the forging experiments. During the PCA step, the feature vector is extracted by calculating the eigenvalue of correlation coefficient matrix for training dataset, and the dimension of input variables is reduced from 11 to 6 features. Thus, PCA offers an efficient method to characterize the data with a high degree of dimensionality reduction. During the ANN step, the principal components were chosen as the input parameters and the mechanical properties as the output parameters, including the ultimate tensile strength (\( \upsigma_{\text{b}} \)), yield strength (\( \upsigma_{0.2} \)), elongation (\( \updelta \)), and reduction of area (φ). The training of ANN model was conducted using back-propagation learning algorithm. The results clearly present ideal agreement between the predicted value of PCA-ANN model and experimental value, indicating that the established model is a powerful tool to construct the correlation of composition-processing-property for TC11 titanium alloy. More importantly, the integrated method of PCA and ANN is also able to be utilized as the mechanical property prediction for the other alloys.