Construction of a Composition-Deformation Mechanism-Property Prediction Model to Simultaneously Improve the Strength and Ductility of Β-Titanium Alloys Via Machine Learning

Abstract

Maximizing solid-solution hardening while incorporating deformation twinning is crucial for simultaneously enhancing their strength and ductility of β-type titanium alloys. This study proposes an integrated composition design framework (ICDF) that combines a deformation mechanism machine learning model with a yield strength machine learning model. This framework enables precise control of strengthening and deformation mechanism, effectively addressing the strength-ductility trade-off challenge of β-type titanium alloys. First, using the key alloy factor screening method, the key alloy factor-deformation mechanism prediction model and key alloy factor-yield strength prediction model were developed. Then, five kinds of new β-type titanium alloys with excellent comprehensive properties were designed by combining the two models. The new alloy Ti-5Cr-3Mo-1.5Fe exhibited a tensile strength of 1030 MPa, a yield strength of 920 MPa, and an elongation of 28%. Compared with the commonly used commercial β-type titanium alloy Ti-5Al-5Mo-5V-3Cr (AMS 4983), the strength-plastic product of the new alloy increased by 71.7%, while the total alloying element content decreased by 47.2%. The new alloy exhibits intriguing deformation twinning behaviors, including stress-induced {332} <113> multi-level twinning and {332} <113> cross-twinning, which, in conjunction with the high solid-solution hardening, results in simultaneous enhancement of strength and ductility. This work provides a novel approach for the rapid design of high performance and low alloying titanium alloys through constructing multi-task models between alloy composition, deformation mechanism and resultant properties.

Keywords: β-type Ti alloys, machine learning, Deformation mechanism, Yield strength