Large-scale Wind Turbine Fault Diagnosis Reasearch

With the rapid development of global new energy technology, and thecontinuous increase of the proportion of wind power, the significance ofthe research for large-scale wind turbine fault diagnosis and reliabilityanalysis is becoming more and more evident. Aiming at improving thesystem reliability, wind turbine fault diagnosis methods were regard as themost important research content here. The whole content were focused onthe fault diagnosis methods of the key parts of the wind turbine, whichwere generator, main shaft, brake and gear box. The researched windturbine fault diagnosis methods or working condition prediction methodswere based on the wavelet and its combination with neural network,Kalman filter and Hilbert transformation. All the researched fault diagnosisor working condition prediction methods were designed into thelarge-scale wind turbine fault diagnosis system. The main content can beconcluded in the following four paragraphs.(1) In order to find a new wind turbine generator fault diagnosismethod, the reasons and the courses of different kinds of the generatorfaults were researched into. The generator fault diagnosis method based onwavelet packet decomposition was found effective, which was applied intothe rotor broken bar fault diagnosis. The new generator fault diagnosismethod shows good performance in improving the rotor fault featurerecognition ability.(2) Temperature data of spindle bearing of large wind turbines wastargeted to predict the main shaft bearing working condition. The waveletneural network based on Kalman filter was used to forecast the spindletemperature to predict the main shaft bearing faults, and the brake wearrate was used to predict its working condition. The aim is to maintain the reliability of the main shaft bearing and the safe operation of the brake.(3) In order to find a better gearbox fault diagnosis method, gearboxvibration data were analyzed to find some clues. Firstly, the waveletde-noising algorithm was applied into the original vibration signal. Thenby using the multi-scale wavelet decomposition and reconstruction methodto get the pure vibration, we can observe Hilbert transform spectralenvelope of the signal. Achieve the corresponding signal peak, effectivevalue, average, kurtosis value, to study the relationship between theseparameters and different kinds of failure phenomenon, and to get thequantitative expression of the different types of problems. The new methodwas tested on the simulated wind turbine bench, which showed goodperformance.(4) Overall design scheme for large-scale wind turbine fault diagnosissystem based on the researched fault diagnosis methods or workingcondition prediction methods was given. A wind turbine fault diagnosissystem based on the wind turbine test bench was installed to simply testthe design scheme.