Wind Turbine Gearbox Fault Diagnosis Method Research Based On Covariance Matrix Manifold

Owing to the serious energy crisis and environmental problems, wind energy as a green pollution-free energy has already got the attention of the world. With the expansion of the installed capacity, the timely discovery and maintenance of wind turbine fault become more and more important. Gear box has the high failure rate among the components in wind turbines and the vibration signal analysis is the most common technology in its fault diagnosis. Gearbox vibration signals of wind turbines have complicated compositions, strong background noise and obvious non-stationary characteristics. The new method research suited to non-stationary and non-gaussian characteristics of vibration signals is the key to improve the level of fault diagnosis technology. Therefore, it is of great practical significance to ensure normal safe operation of wind turbines that the feature-extraction, fault-detection and fault-classification methods research of wind turbine gearbox vibration signals is carried out.Aiming at the disadvantages of the traditional vibration signal analysis methods, such as the only single-channel signal extraction and the inability to extract the correlation between channels and structure information from a multichannel vibration signal simultaneously, this paper is based on the vibration test of wind turbine gear box and puts forward a gearbox fault diagnosis method based on the covariance matrix manifold analysis of the multichannel vibration signal.The main research content is as follows:Firstly, the basic theories and methods of wind turbine gearbox fault diagnosis are summarized. The basic components of wind turbines are elaborated. Overseas and domestic research statuses are analyzed. And the vibration signal feature extraction methods are introduced. With the analysis of multichannel vibration signals, the method based on multi-channel vibration signal covariance matrix of the manifold is presented.Secondly, in view of the connection and structure feature extraction problems of multichannel vibration signals of wind turbine gearbox, the ellipsoid visualization method of the multichannel vibration signal based on covariance matrix manifold representation is put forward. This method converts the multichannel vibration time series to the covariance matrix series, which can effectively express the associated structural information between channels. By the singular value decomposition, covariance matrix can be converted to the ellipsoid and quaternion. On this basis, the visualization of gearbox vibration states can be realized and then the anisotropy analysis is carried out, which is helpful to understand the characteristics of vibration signal and to troubleshoot the problems.Thirdly, in view of non-stationary and non-gaussian characteristics of gearbox vibration signal, a fault detection and localization method based on covariance matrix and Riemannian manifold distance is put forward through representing the multichannel vibration signals with covariance matrix manifold. This method applies covariance matrix manifold as descriptor and Riemannian distance as a similarity measurement, and combines with statistical process control diagram to realize the gearbox fault detection and localization. Through test and analysis, this method not only can effectively detect the correlation between the multichannel vibration signals, but also has performance advantages on fault detection accuracy rate and algorithm complexity.Finally, aiming at the wind turbine gearbox fault classification problems, this paper proposes a multi-scale manifold entropy fault classification method based on covariance matrix manifold representation of vibration signals and traditional multivariate multi-scale entropy algorithm. In this method, the traditional multivariate multi-scale entropy algorithm is applied to the covariance matrix manifold entropy, which overcomes the traditional problems such as the large amount of calculation and the difficulty in quantifying the correlation between channels and realizes the effective extraction of the complexity and correlation characteristics of the gearbox vibration signals. Through the failure data verification, the results indicate that this method can effectively reduce the computational complexity and avoid simultaneously high-frequency interference in traditional time-frequency analysis, which has higher classification accuracy.