PSI - Issue 68

Wenqi Liu et al. / Procedia Structural Integrity 68 (2025) 458–464

463

6

L. Wenqi et al. / Structural Integrity Procedia 00 (2025) 000–000

slant fracture of 490 °C/QS specimens is caused by the DSA effect. From Fig. 5 d, it is also clear that the fracture elongation increases monotonically with increasing temperature at a fixed rate. The increase in temperature promotes the thermal activation process of atoms and reduces the resistance to dislocation motion, resulting in the improved plastic deformation capacity of Ti65. Besides, at 650 °C, Ti65 performs fracture elongation of approximately 40% and section shrinkage up to 70% due to the activation of additional plastic deformation mechanisms. The strain rate influence on the ductility is non-monotonic at different temperatures. The fracture elongation presents the positive strain rate sensitivity at RT, and the dominant contribution comes from the post-necking elongation. While at 650 °C, the ductility remains decreased with the strain rate reduction, as a higher strain rate counteracts the internal thermal activation effect and restricts creep behavior. Besides, at 180 °C and 490 °C, the fracture elongation tends to increase firstly and then decrease as the strain rate rises, which is related to the DSA effect at certain temperatures and strain rates. Consequently, the DSA and creep effects in the investigated Ti65 result in non-linear and non-monotonic temperature and strain rate sensitivity of its tensile properties. In addition, the cross-interactions among temperature, strain rate, and plastic strain also raise the challenge in the strength and ductility prediction of Ti65.

Fig. 5. Tensile properties of Ti65 at different temperatures and strain rates. (a) Elastic modulus; (b) Strength; (c) Ductility.

4.2. Predication on tensile properties of Ti65 at evaluated temperature and strain rate As aforementioned, key tensile properties under 650 °C/10 -2 s -1 condition were predicted by JC and SVR models. The results and corresponding errors compared to the experimental values are listed in Table 2. Only strength data could be calculated in the JC model. The JC model could well describe the strain hardening of materials in low temperature and strain rate intervals. However, with the appearance of DSA and creep, the classical exponential thermal softening and logarithmic strain rate equations are no longer suitable for describing the strain hardening behavior at high temperatures for Ti65. Meanwhile, the cross-effects from temperature, strain rate, and strain evolution are missing in the JC model, finally resulting in more than 15% predicted derivation of UTS. While the SVR algorithm with excellent data tracking ability performs an error of only 2.24%. Furthermore, with a constant hyperparameter set for the investigated material, the SVR model could also forecast the elastic modulus and ductility. Even with the intricate temperature and rate dependency on the ductility of Ti65, the deviation from the SVR prediction on the final elongation is only 1.13%. It demonstrates the powerful learning and prediction capability of the SVR algorithm with a limited database. SVR introduces a certain degree of tolerance into the model by introducing slack variables and loss functions, thus making it more robust to noise. During training, only those data points that are located near the support vector (i.e., the boundary or error band) will affect the final solution of the model, which makes SVR computationally efficient.

Table 2. Prediction of tensile properties at evaluated temperature and strain rate of Ti65. Exp_RT/QS Exp_650 °C/10 -2 s -1

Prediction_JC Error_JC, % Prediction_SVR Error_SVR, %

Young’s modulus (GPa) Yield strength (MPa)

124.9 927.0 1007

85.75 526.0 630.5 17.90

-

-

87.98 531.5 644.6 18.10

2.60% 1.04% 2.24% 1.13%

487.7 529.8

7.29% 15.98%

Ultimate tensile strength (MPa)

Fracture elongation (%)

3.70

-

-