Multi‐Model Study of Fatigue Life Prediction for Hydraulic Concrete

ABSTRACT To solve the difficult problem of fatigue life prediction of hydraulic concrete under complex stresses, this work conducts fatigue tests at stress levels of 0.75, 0.80, and 0.85, combined with multi‐model comparisons. The experimental study reveals that cyclic loading induced nonlinear hysteresis and a three‐stage failure process: microcrack initiation, stable propagation, and macrofracture. The probability model based on Weibull distribution quantified the stress sensitivity of fatigue life through P – S – N curves. The GM(1,1) model was enhanced with dynamic time weighting and normalization, achieving Level 1 accuracy under small samples. The DT‐ANN cascade model, trained with deterministic perturbation, delivered strong predictive performance. The three models together constitute the prototype of the multi‐level method system of “probabilistic assessment–small sample analysis–intelligent prediction”, which provides important theoretical support and technical means for improving the durability design and safety assessment of hydraulic concrete and is of great significance for guaranteeing the long‐term safe operation of structures.