High-dimensional extracting chirplet transform for fault diagnosis of wind turbines

Under actual working conditions, vibration signals of wind turbines often have non-proportional frequency components, especially cross-frequencies, which presents additional challenges in fault diagnosis of wind turbines. Therefore, the high-dimensional extracting chirplet transform (HDECT) is proposed to solve the above problems. In the HDECT, a new dimension is first defined based on time-varying window length and chirprate, and then the local maximum extracting operator (LMEO) is constructed based on time, frequency, and new dimension to reassign the energy of the velocity synchronous linear chirplet transform (VSLCT). Specifically, using the LMEO, only the time-frequency energy at the position of the instantaneous frequency (IF) ridge is retained, effectively eliminating interference from other time-frequency energy sources, and the algorithm’s ability to resist noise is improved. The HDECT is verified by analyzing the simulation signal, the vibration signal of the wind turbine planetary gearbox, the bearing signal, and the sound signal of killer whales. The analysis results show that the HDECT can accurately characterize signals with non-proportional fundamental frequencies, especially crossed IFs, so it can accurately identify wind turbine faults.