Acoustical Diagnosis Of Wind Turbine Blade Based On SVDD

Blades are the key components for wind turbines to obtain wind energy.Due to long-term operation in harsh environments,the blade is prone to crack,blister,corrosion and other faults on surface,which brings in significant economic loss.It is urgent to carry out on-line diagnosis for wind turbine blade,which is of great significance to extend the working life and reduce the maintenance cost.In view of the shortcomings of existing diagnostic,techniques such as vibration,acoustic emission and fiber grating,this paper proposes a non-contact acoustic detection method based on support vector data description(SVDD).The specific research contents and results are as follows:Firstly,through the analysis of wind turbine blade failure mechanism and acoustic signal characteristics,acoustic feature extraction method based on wavelet decomposition is proposed.Taking the energy ratio of decomposed sub-bands as feature,characterizing the spectral variation characteristics of acoustic signals generated by faulty blade.The principal component analysis(PCA)method is introduced to eliminate the redundant information in the original feature vectors and reduce the time complexity of the algorithm.The simulation results show that the method can effectively extract the acoustic characteristics of the blade.Secondly,for the problem of imbalance distribution of samples in blade diagnosis,the one-class classifier model based on SVDD is established.According to the basic principle of SVDD,the initial classification model is constructed by using the normal sample set.The improved particle swarm optimization(PSO)algorithm is used to find the optimal kernel parameter,and the fitness function is constructed by the recognition accuracy of normal and faulty sample sets.The simulation results show that the method can achieve effective blades health diagnosis in the case of limited samples.Finally,an incremental learning method based on Karush-Kuhn-Tucker(KKT)condition and hypersphere coordinate mapping is proposed for improving SVDD diagnostic model.Using the original KKT condition to screen key samples from the incremental sample set,and then the clustering degree of the screened sample sets evaluated according to the Hopkins statistic.For sample sets those present irregular distribution and contain a large number of samples,a boundary extraction algorithm using multi-center hypersphere coordinate is proposed,further optimizes the incremental sample set.The simulation results show that the proposed method can effectively reduce the training time of the algorithm with high recognition accuracy.