Modeling and Prediction Method for Young’s Moduli of Ti Alloys Based on Residual Muti-layer Perceptron

Abstract

Accurate Young’s modulus models of β-type Ti alloys can provide a convenient approach to developing Ti alloy, especially non-toxic and biocompatible medical materials. Data-driven approaches can significantly reduce the difficulty of modeling Young’s modulus of Ti alloy and build reliable models by relying on the large amount of available historical data. Therefore, a deep learning model using multi-layer perceptron with residual connection, namely Res-MLP, is designed to establish the Young’s modulus model of Ti alloy according to its element content or composition. Benchmark models are selected for performance comparison, of which performances metrics, including MAE, MAPE, RMSE, and MARNE, are 9.83, 15.05%, 14.86, and 10.57%, respectively. Therefore, the Res-MLP has predictive ability. Compared to SVR, XGBoost, RF, BPNN, CNN, and MLP models, Res-MLP achieves better prediction performance and precision. Moreover, the bootstrapping algorithm is used to expand the sample size. Through a comparative analysis of the predictive performance of Res-MLP before and after dataset expansion, it is demonstrated that data augmentation methods can effectively enhance predictive capabilities. Consequently, the model proposed in this study can provide an effective and efficient data mining tool developing medical Ti alloy materials.