Sparse Bayesian learning beamforming combined with short-time Fourier transform for fault detection of wind turbine blades

Abstract

Wind energy is one of the important renewable energy resources. The wind turbines need to be checked every now and then to enhance security. The rotor blade of the wind turbine can be damaged due to long-term running in harsh environments and complex vibrations resulting in the crack of blade. For detection of the blade faults, the sparse Bayesian learning (SBL) beamforming (BF) is implemented to the acoustic data received by a microphone array on the ground for signal enhancement. The direction of arrival of the abnormal sound can be estimated with high resolution meanwhile interferences, such as noise emitted by cooling fans, can be suppressed by the low sidelobes provided by the SBL-BF. After the Short-Time Fourier Transform (STFT) is carried out over the enhanced signals, it is seen from the time-frequency spectrum that the abnormal sound appears with an approximate 6-s cycle. By detecting the cyclical characteristics, one can decide whether there is or not the blade fault. A real-time processing system for detection of the blade faults underwent a number of tests in a coastal plain, a hilly area, and the Xinjiang Plateau. The test results have demonstrated the effectiveness of the blade fault detection framework.