PSI - Issue 68

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

461

4

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

presents an exponential tendency according to the JC model. Due to the creep effect at 650 °C, the maximum stress point occurred before the true plastic strain of 0.01 and true stress values are unreliable after strain localization. Therefore, engineering stresses at the engineering plastic strain of 0.01 were extracted and normalized to calibrate the thermal softening effect, as shown in Fig. 3 e-f. All parameters of the Johnson–Cook model are listed in Table 1. With the temperate and strain rate coefficients in the JC model, the strength values at evaluated temperatures and strain rates could be predicted based on the experimental data from RT/QS loading. The prediction performance mainly relied on the equation’s form and parameter fitting quality. The predicted YS and UTS are listed in Table 2.

n , - ̇ ! , "# C , - T m , K T 0 , K m , - 0.037 10 -5 0.015 1941 298 0.73

Table 1. Parameters of Johnson–Cook model. A , MPa B , MPa

100

1043

Fig. 3. Johnson–Cook model parameter calibration process. (a-b) Strain hardening parameters; (c-d) strain rate sensitivity parameter; (e-f) temperature sensitivity parameter. 3.2. Support vector regression model The basic idea of the SVR algorithm is to find a hyperplane that minimizes the distance of all data to this hyperplane. SVR creates a ‘spacing band’ on both sides of the linear function with a spacing of tolerated deviation δ , and the loss of all samples falling into the spacing band will not be considered. The algorithm finally derives an optimized model by minimizing the total loss and maximizing the spacing. The nonlinear problem could be converted into the approximately linear regression problem by the transformation of the kernel function. To predict the tensile properties of Ti65, the input layers in this study are the experimental temperatures and strain rates, and the outputs are the key tensile property parameters, i.e. E, YS, UTS, and A f . The testing data is the properties under 650 °C/0.01 s -1 loading condition, and all other conditions serve as the training databases. The main hyperparameters include tolerated deviation δ , weight of penalty term ω , kernel function type, and corresponding kernel parameters. The polynomial function is employed as the kernel function with the expression of:

) (

)

(

!

(2)

! % % & ! = +

" $ % %

"

#

"

#