Improving Wind Turbine Gearbox Reliability: A Hybrid Deep Learning Approach Using Bispectrum Image Analysis

Wind turbine gearboxes are widely used in wind power plants, whose operating conditions are characterized by fluctuating wind speeds and loads, introducing noise and complexity into vibration signals, complicating accurate fault diagnosis in wind turbine components. This study explores the potential of bispectrum image analysis combined with Convolutional Neural Network (CNN), CNN-Long Short-Term Memory (LSTM), and CNN-Bidirectional LSTM networks for enhanced fault detection and predictive maintenance for wind turbine systems. Bispectrum images are used to identify nonlinearities, harmonics, and interactions among various frequency components, thereby enhancing the robustness and accuracy of fault diagnosis. Results demonstrate high accuracy ¿98% for all three models. However, CNN-BiLSTM consistently achieves the highest classification accuracy, exhibiting superior generalization capabilities and minimal misclassification. This superior performance is attributed to the CNN-BiLSTM’s ability to discern bidirectional temporal patterns within the bispectrum image data, effectively capturing the complex dynamics of gearbox failures. This research demonstrates the promise of bi-spectrum analysis as an effective method for detecting abnormalities and trends that may signify emerging defects, facilitating timely maintenance and averting catastrophic failures.