Overdue service of thermal power units has become a trend

but fatigue crack of turbine rotor steel seriously affects the operation safety of steam turbine units. Due to the lack of the fatigue crack growth(FCG) test data of rotor steel

and large computation cost for stochastic model modeling and solution

the estimation of fatigue crack remaining useful life(RUL) is currently insufficient. On the basis of fatigue crack growth tests and analysis on its random models

a modified Gaussian membership information expanded(GMIE) sample domain method is proposed to generate virtual samples based on mega trend diffusion(MTD). Meanwhile

an extreme machine learning(ELM) neural network combined with the expective regression(ER) model is used to predict the RUL of fatigue crack propagation. The RUL of fatigue crack propagation under a specific cycle is calculated. By comparing the results with the RUL probability density function(PDF) curve and fatigue crack propagation curve of the existing numerical analysis methods

it shows that mean absolute percentage error(δ_(MAPE)) is 2.78%

which verifies the effectiveness of the proposed method and provides robust support for safe operation of the turbine rotor systems.