5. Discussions
Based on the results of the above four tests, it is shown that the proposed model has improved the accuracy of predicting the remaining life of fatigue loading to a considerable extent compared to the fatigue driving stress model (K-R model). Miner’s rule has a simple application in the form that fatigue damage at different levels can be summed linearly, but fatigue failure in many metallic materials often exhibits highly nonlinear damage behavior. The fatigue driving force model (K-R model) improves on the inherent flaws of the linear damage accumulation rule. It is a nonlinear model that requires only the S-N curve of the material. In addition, it does not require many material property parameters and takes into account the effects of the loading sequence. The K-R model has improved predictions compared to Miner’s rule by having a sum of damage values greater than 1 in the low-high loading sequence and less than 1 in the high-low loading sequence. However, it was found that loading interactions are also a part of the load history and the K-R model does not consider loading interactions. Zhu’s model considers loading interactions by adding the ratio of the two loading levels, while Li’s model considers load interactions by adding a pre-circulation factor. However, it was found that the improvement in the predictive capability of the two modified K-R models relative to the original model was limited, probably due to the complexity of the inherent variable amplitude loading damage evolution, and the two modified models did not provide an explanation in terms of fatigue driving stress evolution. In this paper, based on the driving stress evolution, a new driving stress evolution curve is obtained by taking into account the load interactions, so that the original evolution curve is shifted. Then, the fatigue driving stress equivalence was carried out to obtain a new fatigue residual life prediction model. In particular, compared with Miner’s rule, K-R model, Zhu’s model, and Li’s model, the proposed model has similar prediction ability under multi-stage loading, but for two-stage loading, the proposed model has significantly higher prediction accuracy.