Research On Acoustical Feature Extraction In Turbine Blades Health Monitoring

Blade is the key component of large wind turbines for wind energy acquisition.Due to long-term operation in the harsh natural environment and the coupling effect of various complex stresses,blades are prone to cause various security risks,resulting in significant economic losses.So it is urgent to develop the health monitoring technology of wind turbine blades.Because the maintenance mode of regular patrol inspection of wind farm is inefficient and poor in real time,it is difficult to meet the demand for informationized maintenance of intelligent wind farm at present.Although a variety of blade fault detection techniques have been developed,it has not been widely used because of some shortcomings.Therefore,this paper proposes a non-contact online remote acoustic health monitoring system for turbine blades and has an in-depth study on the acoustical feature extraction method of wind turbine blades.The specific research contents and results are as follows:Firstly,the architecture of the remote online acoustical monitoring system for wind turbine blades is designed,which contains acoustic signal acquisition module,front-end data processing module,the communication module and the monitoring and management center.Then the core functions of each module are introduced in detail.Secondly,the method of acoustical feature extraction and adaptive optimization of wind turbine blades based on octave and PCA is established.After using processing method to remove the complex background noise in the acquired acoustic signals,the 1/6 octave energy ratio is extracted as the initial feature according to the frequency domain feature of the acoustic signals.In the meantime,to solve the problem of high dimension of feature space,a feature dimension adaptive optimization method based on PCA is introduced to get the low dimensional acoustic feature,and SVM is applied to evaluate the effectiveness of acoustic feature.The experimental results show that the proposed method can extract the effective blades' acoustical feature.Finally,aiming at the problem of generalization ability and nonlinear representation ability of artificial feature extraction method,an acoustical feature extraction method based on an stacked autoencoder is proposed.The time-frequency data is used to structure the input set,and the Denoising and Dropout methods are introduced to solve model's over fitting.In addition,the PSO algorithm is used to find the optimal hyper parameters of the SDAE model.The experimental results show that the PSO-SDAE algorithm can extract the effective blades' acoustical feature automatically,and has a wide application prospect.