Fatigue life prediction method based on data fusion

Abstract

To address the challenges posed by the time-consuming nature of fatigue test and the scattered nature of test data，it is evident that P-S-N curves derived from small samples with high survival rates lack sufficient accuracy，leading to unreliable predictions of fatigue life. The data fusion method based on the performance-life probability mapping principle is used to fuse small sample fatigue data of different stress levels, and the feasibility of obtaining accurate P-S-N curves by this method is analyzed and evaluated. The results demonstrated that P-S-N curves obtained post-fusion are closer to the P-S-N curve derived from larger sample datasets. This approach effectively enhances both reliability and accuracy in predicting fatigue life while simultaneously reducing the amount of required fatigue tests. A comparative evaluation is conducted on the predictive capabilities for fatigue life before and after fusion using different models; notably，it is found that the three-parameter power function model demonstrates superior predictive ability，whereas when ample fatigue data is available，the prediction capabilities among four models（Basquin S-N model，exponential S-N model，three-parameter power function S-N model（based on lognormal distribution），and three-parameter power function S-N model（based on three-parameter Weibull distribution） exhibit a considerable degree of resemblance.

Keywords

• fatigue
• data augmentation
• stress life model
• life prediction
• model comparison