Study on high cycle fatigue behaviours and modelling of cast aluminium alloy at elevated temperatures

• Published in: Engineering failure analysis 2025-02, Vol.168, p.109031, Article 109031
• Main Authors: Liang, Yongsen, Zuo, Zhengxing, Wang, Jundiao, Zhao, Chengzhang, Ren, Peirong, Huang, Weiqing
• Format: Article
• Language: English
• Subjects: Cast aluminium alloy; Elevated temperature; High cycle fatigue; Modelling
• ISSN: 1350-6307
• DOI: 10.1016/j.engfailanal.2024.109031

[Display omitted] •The coefficient of variation in fatigue life decreases with increasing load at all temperatures.•Cracks nucleate from pores at RT and 200 ℃, while cracks nucleate from various sites at 300 ℃.•Temperature and load have a comparable impact on fatigue life uncertainty, whereas defect size exhibits a weaker correlation.•Fatigue life prediction models were established based on LEFM, ANN and BN.•Probabilistic inference is performed using BN, achieving a balance in predictiveness and interpretability. High cycle fatigue (HCF) behaviours of cast aluminium alloys at room temperature (RT) and elevated temperatures are examined in this study. The results show that fatigue life significantly decreases at 300 ℃ compared to RT, while the decline at 200 ℃ is relatively small. The coefficient of variation (CV) in fatigue life increases with decreasing load levels at all temperatures. The fracture analysis reveals that pores are primary sites for crack nucleation at both RT and 200 ℃, while cracks also nucleate from hard particles and slip bands at 300 ℃, with the presence of multiple nucleation sites. Correlation analysis shows that the temperature and load level have a comparable impact on the uncertainty in fatigue life, while the effect of crack nucleation defect size is lower than temperature and load level. Fatigue life models were established using linear elastic fracture mechanic (LEFM) and artificial neural network (ANN), respectively, with the former showing lower accuracy due to its inability to capture non-linear material softening at 300 ℃, and the latter demonstrating higher performance with the majority of predictions falling within ± 2.5X scatter bands. Finally, the omni-direction inference between cause and effect is performed using the Bayesian network (BN), which can predict fatigue life distribution interval probabilistically, achieving a balance in predictive and explanatory purposes well.