With the widespread application of steel-concrete hybrid wind turbine towers, concrete fatigue design has become a challenge in engineering design and a primary factor affecting costs. Concrete fatigue is influenced by many interrelated nonlinear factors, and there is no consensus on the quantitative measurement of these. To develop more accurate methods for describing the S-N curves of concrete materials and numerical calculation approaches, this study employs a backpropagation neural network (BPNN) to predict concrete fatigue behavior using an expanded experimental database of 558 constant-amplitude fatigue test results. On the test set, the BPNN predictions achieve a coefficient of determination (R²) of 0.926, a root mean square error (RMSE) of 0.352, and a mean absolute error (MAE) of 0.244, significantly outperforming existing code-based methods (NEN 6723: R² = 0.411, RMSE=1.894, MAE = 1.640; EN 1992: R² = 0.410, RMSE=1.312, MAE = 1.085; fib Model Code 2010: R² = 0.056, RMSE=5.755, MAE = 2.011). Additionally, a 1:5 scale specimen based on an engineering prototype was designed and tested under variable-amplitude high-cycle fatigue loading. Three mean stress correction methods (Goodman, Gerber, and R-ratio) are evaluated in fe-safe to determine the most suitable approach for concrete fatigue analysis. The observed fatigue damage evolution follows a three-stage process—stress redistribution, steady damage accumulation, and rapid failure—which the BPNN-based finite element model accurately captures, achieving an average stiffness error of 9.43% compared to experimental measurements. Therefore, the BPNN method proposed can be applied to the calculation of concrete S-N curves when there is sufficient data support, and the fatigue performance of complexly stressed structures can be effectively designed using fe-safe.
Huang et al. (Wed,) studied this question.