Estimation of remaining fatigue life under two-step loading based on kernel-extreme learning machine

Remaining fatigue life estimations are not trivial problems for most engineering applications. They could even become quite challenging for general multistep load spectrums, particularly when elastoplastic stresses and strains are involved. In such case, the remaining life predictions could be very sensitive to the chosen damage indicators, leading to complex and non-uniform processes to model non-linear behavior of the materials which exhibit different characteristics for fatigue damage accumulation. To overcome this problem, a data-driven model based upon the kernel-extreme learning machine is proposed to estimate the remaining life of materials under two-step loading. Different from conventional empirical damage models, the proposed model can automatically acquire the optimal mapping relationship from the training samples, which is a quite versatile and flexible method to mathematically describe the indications of fatigue damage mechanism. Moreover, to maintain the basic physical rationality and good measurability, the input variables can also be referenced from conventional damage models. Extensive experimental results of nine materials under two-step loading are collected from the open literature and are used to validate the proposed model. It is shown that the proposed model can provide a much better estimation of the remaining life against the other three conventional models and the standard extreme learning machine (ELM)-based model.