Predicting the modulus of elasticity of biocompatible titanium alloys using machine learning

Abstract

Titanium alloys are widely employed in various fields, particularly in biomedical engineering, due to their mechanical and corrosion resistance properties combined with good biocompatibility. The modulus of elasticity is a distinguishing feature of this group of materials compared to others used for similar purposes. A database of approximately 238 titanium alloys free of toxic elements was compiled for this study. The influence of different factors (such as alloy element proportions, density, and specific heat) on the modulus of elasticity was predicted using four methods: support vector machine, XGBoost, Neural Network, and Random Forest. The Random Forest mean absolute error (MAE) of 7.33 GPa, falls within the range of experimentally obtained absolute errors in the literature (up to about 11 GPa). A strong correlation (R2 = 0.72) was observed between experimental and predicted data. Lastly, specific alloying element regions were identified for the modulus of elasticity, which can be used to design new biocompatible titanium alloys in the future.