Icing diagnosis method of wind turbine blade based on mechanism and data driving

Blade icing is a major cause of power losses and structural degradation in wind turbines. Accurate diagnosis of wind turbine blade (WTB) icing is essential for ensuring efficient operation, optimizing, maintenance strategies, and the extending turbine lifespan. However, current diagnosis methods often struggle with data imbalance, small sample sizes, and low accuracy. To address these issues, this study proposes a WTB icing diagnosis method that integrates mechanism-based modelling with data-driven techniques. Specifically, a mechanism model grounded in wind turbine operational principles is constructed to simulate blade icing scenarios. The resulting data are used to generate a balanced dataset for icing diagnosis. To enhance the diagnostic accuracy, a diagnosis model based on sparrow search algorithm and capsule network (SSA-CapsNet) is developed. The model's performance under varying sample sizes is thoroughly examined and benchmarked against a convolutional neural network (CNN) model. The results demonstrate that the SSA-CapsNet model is more advantageous even in small-sample contexts. Additional validation using actual operational data and comparisons with several mainstream models further confirm the validity of the proposed approach. Overall, this method provides a promising technical solution for real-time blade icing diagnosis in operational wind farms, which helps improve their operation and maintenance efficiency.