Characterizing small fatigue crack propagation behavior is essential for assessing structural integrity, and refining fatigue life models by incorporating critical factors like crack closure effect. The integration of scanning electron microscopy (SEM) with digital image correlation (DIC) techniques offers significant advantages for in situ microscale observation and analysis, enabling full-field and non-contact deformation measurement. In this study, the small fatigue crack growth behavior of nickel-based superalloy GH2036 is investigated through in situ SEM monitoring of single-edge notched tensile specimens. The deformation field near the crack tip is quantified using SEM-DIC, based on which a previously proposed parameter termed the crack opening ratio (COR) is introduced and employed to evaluate crack closure effect at the microscale for the first time. The crack closure level, specifically the crack opening load ratio, determined via COR shows good agreement with results from the stress intensity factor (K)-based approach. This consistency underscores the reliability of the COR methodology for microscale fatigue characterization, providing a robust framework to advance understanding of crack closure mechanisms and refine fatigue life prediction models.

Amid the global energy transition, the rapid growth of wind turbine deployment has highlighted the need for accurate fatigue load prediction to support structural design and ensure operational reliability. This study proposes a neural network-based method for estimating fatigue loads at critical locations of large wind turbines. Wind speed, turbulence intensity, and yaw angle were used as input features, while the damage equivalent loads at the blade root, tower base, and yaw bearing served as prediction targets. A dataset comprising 2139 operating conditions was constructed, and two predictive models—an artificial neural network (ANN) and a Bayesian neural network (BNN)—were developed and evaluated using standard error metrics. The results show that the BNN consistently achieves lower prediction errors and higher goodness-of-fit values than the ANN across all outputs, demonstrating improved accuracy and stability. The BNN model attained excellent predictive performance, with an overall coefficient of determination (R²) of 0.9998, a root mean square error (RMSE) of 0.012, and a mean absolute percentage error (MAPE) of only 0.1877%. These findings indicate that probabilistic neural networks hold strong potential for enhancing fatigue load prediction and can provide valuable support for wind turbine structural assessment, design optimization, and active yaw control strategies.