Fault Diagnosis Of Wind Turbine Gearbox Based On Improved K-means Clustering Algorithm

With the continuous development of economy,environmental problems have become increasingly prominent.In recent years,in order to improve the environmental pollution and take the road of sustainable development,China has carried out a lot of development and research on wind power resources.Wind turbine can stably convert wind energy into electric energy,and its internal gear box is an important structure to improve the running speed and power quality of wind turbine.However,it is also prone to failure,which will reduce the utilization rate of wind power and increase the operation and maintenance cost.Most losses can be avoided if the unit operation can be accurately grasped,and faults can be found and maintained in time at an early stage.Therefore,the research on gearbox fault diagnosis is significant.In this paper,taking the gearbox as the research object,through analyzing its vibration signal and combining with the pattern recognition algorithm,the fault diagnosis of the working state of the wind turbine generator set is carried out.Aiming at the unstable and nonlinear problems of the vibration signal,the set empirical mode decomposition method is used to extract the feature vector,and the obtained feature vector is quantized by using the approximate entropy principle to make it more recognizable and relevant;Aiming at the problem that the traditional K-means clustering algorithm is easy to fall into local optimum,the part icle swarm optimization algorithm is introduced to improve it,which improves the optimization ability of the algorithm.The specific research contents of this paper are as follows:(1)The operation principle of gearbox of wind turbine is studied,the com mon faults and fault diagnosis methods are summarized,and the existing problems and solutions of gearbox are emphatically expounded.(2)The vibration signals are analyzed and processed.Combining the shannon theory with the actual situation,the sampling frequency of the extracted vibration signal is determined.Comparative tests are carried out on the collected vibration signals by empirical mode decomposition method and ensemble empirical mode decomposition method respectively,and the superiority of en semble empirical mode decomposition method is verified,and the feature vector is preliminarily determined by this method.(3)Quantitative analysis of eigenvector results is carried out.Because the result of the intrinsic mode function is average and the value is small,which is not conducive to quantitative analysis,approximate entropy analysis is introduced to quantify the extracted feature vectors,and make the vector correlation stronger and the experimental results more accurate.(4)Combining particle swarm optimization with pattern recognition for fault diagnosis,the traditional K-means clustering algorithm is improved,and the global optimization,convergence speed and classification accuracy are improved accordingly.In addition,the algorithm flow is simplified,the feature vector is no longer required to be reduced in dimension,and the multi-dimensional vector can be directly pattern recognized.