Research On Theory And Methodology For Fault Diagnosis Of Rolling Element Bearings In Wind Turbine Gearbox

With the energy shortage and environment pollution becoming more and more serious, one of effective ways to solve the problemes is to develop wind energy. While maintaining the high-speed development of wind power industry, it is worthwhile to pay more attention to the safety operation of wind turbine. The bearings in gearbox have the longest downtime compared to the other components. So the research on its failure has important theoretical value and engineering significance. The object of study in this paper is rolling element bearings of wind turbine gearbox. The fault diagnosis of bearings is important, which can reduce the cost of wind plant and promote the operation efficiency of wind turbines.The major contributions of this thesis are summarized as follows:1. A new approach based on mathematical morphological filter and wavelet is presented, which is to identify bearing faults via vibration while the faults is still in an incipient stage. First, this approach considers the removal of random pulse and white noise, and then a model-based vibration is established for decreasing the blindness and increasing the accuracy of fault diagnosis. Based on the above model, we can deduce that there is a significant increase in the energy of fault characteristic band, when single-point defects occur on a bearing surface. Consequently, the characteristic fault frequencies which are separated from fault characteristic band are utilized to obtain the style and location of defect.2. In order to solve the disadvantage of the dependence of empirical value of existing IMF stop criterion, a new stop criterion based on kurtosis is proposed. Given the sensitivity of kurtosis to pulses, it can reveal the distribution of amplitude. At the same time, this criterion has considered the definition of IMF, the index of orthogonality and completeness so that the IMF component has less error and better orthogonality.3. A new technique based on the optimized EMD and divergence is proposed. It can diagnose the location and degree of defects. The fault signatures of bearings are extracted firstly. Then the J-divergences and KL-divergence between test sample and standard sample are computed. Finally, the divergence and its variation can determine the location and the severity degree of fault, respectively.4. A new approach based on the optimized EMD and adaptive K-means clustering is proposed. It can diagnose the location and degree of defects. The fault signatures of bearings are extracted firstly by optimized EMD. The Principal Component Analysis and Holtelling T2-statistic are applied to improve the weaknesses of K-means clustering, that are, the determination of cluster number and cluster center is difficult and there exists the influence of isolated points. So the adaptive K-means clustering is proposed. This approach can adaptively identify the data sets, depending on the location and the severity level of fault.5. Now the available system which can monitor the key operating parameters of all the important parts of wind turbine is few, and fault diagnosis in the existing system is weak. An integrated system is designed, which can combine the on-line fault warning and fault diagnosis. The hardware and software implementations of system are illustrated, including system structure, system function and integrated system characters. The principle of fault warning system and the steps of fault diagnosis are descriped, emphasizing on the function and implementation of self-learning.