PSI - Issue 68

Wenqi Liu et al. / Procedia Structural Integrity 68 (2025) 458–464 L. Wenqi et al. / Structural Integrity Procedia 00 (2025) 000–000

462

5

where γ is the scaling parameter, r is the constant term, and d is the degree of polynomial. Using GridSearch CrossValidation method, the best hyperparameters were selected from the hyperparameter range δ in [0.01, 0.1, 1, 10], ω in [0.1, 1, 10,100], γ in [0.001, 0.01, 0.1, 1], d in [2, 3, 4, 5] and r in [0, 0.1, 0.5, 1]. It is definitely that the optimal prediction model can be found through more detailed hyperparameter selection but based on the balance between computational cost and prediction accuracy, limited parameter ranges were determined as above. Under specific hyperparameters, the model is trained using the training data, and the model performance is evaluated using the relative error between the predicted value of the test data and the actual data. The optimal function is controlled by finding the smallest values among the maximum errors of the four focused testing properties. Finally, the kernel function of the SVR algorithm is a 2-degree polynomial function with hyperparameters ( ω =1, γ =1, δ =0.1, r =0.1). The comparison between the experimental and predicted data is presented in Fig. 4.

Fig. 4. Experimental and SVR predicted data comparison. (a) Elastic modulus; (b) Strength; (c) Ductility.

4. Results and discussion 4.1. Temperature and strain rate sensitivity on tensile properties of Ti65

The key tensile properties of Ti65 from varying loading conditions are shown in Fig. 5. The elastic modulus of Ti65 shows a distinctly negative linear correlation with the temperature, and the correlation coefficients are all above 0.98 at different strain rates. Whereas the strain rate has a slightly smaller impact on the elastic modulus with a difference below 7% from 10 -5 s -1 to 10 -2 s -1 . It is indicated that the strength of Ti65 generally shows thermal softening and strain rate hardening behavior, except for 490 °C. In general, the strength of metallic materials is logarithmically dependent on strain rate and exponentially dependent on temperature, which is true for both RT and 180 °C. However, the strength at 490 °C presents a negative strain rate correlation from 10 -4 s -1 to 10 -2 s -1 . Moreover, the serrated flow curves could be observed at 490 °C (Fig.1), referring to the Portevin-Le Chatelier (PLC) effect. The abnormal stress– strain responses in the intermediate temperature and low strain rate ranges are caused by the dynamic strain aging effect. In a specific temperature–strain rate range, the velocity of dislocation slip and the rate of solute atom diffusion become nearly equal, resulting in the repeated pinning of moving dislocations by solute atoms. The increased external force is required to free the pinned dislocation from the solute atoms, like Si and C in Ti alloys, and to continue the slip and maintain the plastic deformation, thus leading to the increased stress response and serrated flow curve (PLC effect). Furthermore, the DSA effect is strongly affected by the temperature, strain rate, and stress states (Liu and Lian (2019)). The increased plastic deformation rate and lower temperature will raise the activated threshold of the DSA effect, i.e. the postponed DSA starting plastic strain value. This is the reason for the weaker strain rate sensitivity on UTS than YS at 180 °C. Lastly, the performance at 650 °C is affected by the creep phenomenon. During this process, the material demonstrates a significantly lower stress level than the regular range due to the activation of new plastic deformation mechanisms at high temperatures, such as dislocation cross-slip and climbing, or diffusion of vacancies along the grain boundaries (Viswanathan et al. (2002), Gollapudi et al. (2008)). The complex temperature and strain rate sensitivity are also observed in the ductility of Ti65. The macroscopic specimen deformation and fracture modes in Fig 2 present an evident temperature dependency. At RT, the non uniform plastic deformation and macroscopic necking are not obvious, and the final fracture is relatively flat. With the increase in temperature, both uniform plastic deformation and strain localization are enhanced, especially for specimens with extremely large plastic deformation at 650 °C and slow loading rates. It is worth noting that the sudden