A Data-Driven Method Based on Hybrid Feature Fusion for Wind Turbine Blade Imbalance Fault Diagnosis

Wind turbine blade imbalance faults (WTBIFs) significantly impact wind turbine safety. Current research primarily focuses on diagnosing WTBIFs using traditional signal processing and time-frequency analysis methods, but lacks in-depth investigation into root cause localization. To address the issue, this paper proposes a data-driven method integrating hybrid features. Firstly, a parallel connection architecture combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) is developed to complementarily extract deep time series features from nacelle vibration acceleration signals, which are fused with traditional handcrafted features. Then, A multi kernel extreme learning machine (MKELM) is employed for fault classification by exploring nonlinear mappings between WTBIFs and features. Experimental results on actual wind farm operational records confirm that the proposed outperforms advanced approaches in key metrics including Accuracy (95.45%), Precision (96.92%), Recall (92.12%) and F1-score (94.26%).