Wind turbine blade fault detection based on graph Fourier transform and deep learning

Abstract

Wind turbine blades play an important role in harnessing wind power to generate electricity. And they are susceptible to damage due to fatigue loads and exposure to harsh operating environments. Thus, early warning of the damage of wind turbine blades is vital for reducing maintenance costs but in face of challenge of weak signal detection. A spatial-temporal joint processing framework is proposed based on combination of graph Fourier transform and deep learning for detection of wind turbine blade faults. The microphone array processing is utilized to enhance the weak abnormal signal emitted by the damaged wind turbine blades. Then the enhanced signal is projected from time domain into graph domain by graph Fourier transform. And short time Graph Fourier transform (STGFT) features and graph Mel filter banks (Gbank) features are extracted from graph domain. Finally, an advanced deep neural network is designed to extract deep semantic features from the graph signal. The experimental results have validated the effectiveness of the deep learning based fault detection methodology.