Research on parameter identification of fracture model for titanium alloy under wide stress triaxiality based on machine learning

Abstract

The abilities to describe the fracture behavior and calibrate the relevant parameters are essential factors in evaluating ductile fracture criteria of titanium alloys. In this study, 14 different shapes and notched specimens were designed for uniaxial tensile and compression experiments to characterize their ductile fracture behaviors. Based on the analysis of plastic behavior and fracture mechanism, a mixed hardening model, the Von Mises yield criterion and DF2016 fracture criterion were established, respectively. A parameter-identification method based on machine learning was proposed to improve the parameter calibration of the ductile fracture model. The results showed that the DF2016 fracture model accurately predicted the damage initiation and fracture process of the forged TC4 titanium alloy during the forming process. The machine-learning method avoided extracting different stress state evolution processes and large amounts of data from the numerical model of the calibrated specimens. The combination of the semi-coupled fracture model and parameter-identification method provides a new method that alleviates the difficulty of balancing parameter calibration and the ability to characterize the ductile fracture criteria.