PSI - Issue 68

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

459

2

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

1. Introduction Titanium alloy Ti65 (Ti-Al-Sn-Zr-Mo-Si-Nb-Ta-W) is a near-α titanium alloy to be utilized as disks or blades in aero-engine at 600 ℃ ∼ 650 ℃. At high temperatures, Ti65 has excellent tensile strength, creep resistance, thermal stability, specific strength, and high toughness (Zhang et al. (2020)). The plastic deformation behavior and mechanical properties of titanium alloy are sensitive to temperature, strain rate, plastic strain and the interaction of each other (Li et al. (2022)). Especially with elevated temperatures, titanium alloys undergo strain hardening, dynamic strain aging (DSA), creep or even dynamic recovery (DRV) and dynamic recrystallization (DRX) during tensile deformation, resulting in the non-linear stress–strain curves (Balasubramanian and Anand (2002)). It has been proved that the flow stress increase during the hardening process could be attributed to the strain hardening and precipitation hardening of near-α titanium alloy (Zhao et al. (2020)). DSA effect also brings abnormal thermal hardening behavior and negative strain rate sensitivity. While the flow stress decrease could be caused by deformation heating, creep, DRV/DRX, and strain localization, etc. (Peng et al. (2014)). A number of constitutive models with temperature and strain rate dependence have been established, which could give accurate predictions of the nonlinear relationship between stress and strain. It will be critical for performance evaluation and safe design of titanium alloys (Savaedi et al. (2022)). At present, scholars commonly use two types of constitutive models: empirical models (such as the Arrhenius model, Johnson–Cook (JC) model) and machine learning (ML) methods. Modified Arrhenius-type constitutive models with strain compensated using the polynomial equation or exponential equation have been established to predict the complex deformation behavior of Ti alloys (TC4 (Imran et al. (2022)) and Ti65 (Zhu et al. (2024)) at temperatures from 750 °C to 900 °C and strain rates of 10 -4 –10 -1 s −1 . A modified JC constitutive model has been established to accurately predict the flow stress of TA15 titanium alloy at temperatures ranging from 750 °C to 950 °C and strain rates of 10 -3 –1 s −1 (Ji et al. (2020)). Besides, the application of machine learning techniques has substantially enhanced the forecasting of the flow behavior. The advanced ML models including support vector regression (SVR), artificial neural network, generalized regression neural network, and random forest were trained to predict the flow behavior of Fe2Ni2CrAl1.2 multi-principle element alloys under the temperatures of 800–1100 °C and the strain rates of 10 -3 –1 s -1 (Qiao et al. (2024)). The predictive capability of ML algorithms was estimated, and the SVR model normally performed better with limited data size. The deformation behavior of Ti−13Nb−13Zr alloy at temperatures ranging from 650 °C to 900 °C and strain rates of 10 -2 –10 s −1 was investigated by Shi et al. (2019). The support vector regression model could accurately track the highly nonlinear flow behaviors with the correlation coefficient R -values always larger than 0.9999. In this study, the Johnson–Cook model and support vector regression model are comparatively investigated to predict the tensile properties of a Ti65 alloy at temperatures ranging from 25 °C to 650 °C and strain rates ranging from 10 -5 s -1 to 10 -2 s -1 . 2. Experiments The material used in this investigation is Ti65 alloy. Smooth round bar specimens with outline dimensions of ∅ 7 mm × 71 mm and gauge dimensions of ∅ 5 mm × 25 mm were taken from the heavy plate along the circle direction. U niaxial tensile tests were performed at varying temperatures and strain rates according to the standard GB/T 228.2-2015 to characterize the flow behaviors of the investigated material. The specimens were heated by a resistive high-temperature furnace to the targeted temperatures and held for 15 minutes before the tensile testing. To start the tensile testing, the specimen of Ti65 was gripped at each end and stretched along its length direction at a constant crosshead velocity in the universal tensile test machine until it fractured. The ambient temperatures were set as 25 °C, 180 °C, 490 °C, and 650 °C. The chosen crosshead velocities were 0.015, 0.15, and 15 mm/min, corresponding to the uniform strain rate of 10 -5 s -1 , 10 -4 s -1 , and 10 -2 s -1 before necking. The force and displacement were measured during loading using a load cell and an extensometer. From the measured force–displacement data, both engineering stress–strain curves (Fig.1 b) and flow curves (Fig.1 c) can be calculated. 2-3 parallel tests were performed for each loading condition and the average values were accounted for the focused tensile properties, such as Young’s modulus (E), yield strength (YS, R p0.2 ), ultimate tensile strength (UTS, R m ), and fracture elongation (A f ). The extensometer was removed at the engineering strain of 5% during the testing. At that point, the test was paused while the displacement remained constant, and then the test was resumed after the extensometer was removed.