Wind Turbine Blade Fault Audio Monitoring Method Based on Convolutional Generative Adversarial Network

Abstract

The invention relates to a method for audio monitoring of faults of blades of a wind generation set on the basis of a convolutional generation defense network. The method comprises the steps that a blade rotation audio signal acquired by a sound acquisition device is divided, power spectrum matrixes of all the sets of blades are analyzed through principal component analysis, the generation defensenetwork is constructed, and characteristic square matrixes of the blades are alternately trained by means of anoGAN through the deep convolutional generation defense network; after a defense model isgenerated through training of data of one set of blades, the characteristic square matrixes of the remaining sets of blades are input to an abnormity detector, and the input to-be-tested data are subjected to iteration training again as a first part of model output; the fitting degree of the input data relative to original training data is calculated, and the final model loss is used as a fittingerror; and traversing is carried out for error calculation, so that a difference of each set of blades relative to the remaining sets of blades is obtained. The method is high in efficiency, low in cost, free of manual operation and suitable for fault monitoring of the wind generation set.

Description

technical field [0001] The present application relates to an audio monitoring method for faults of wind turbine blades based on convolutional generative confrontation network, which is applicable to the technical field of fault monitoring of wind turbines. Background technique [0002] Wind turbines operate under alternating loads for a long time, and the blades are prone to cracks, corrosion and other failures, which not only reduce the service life of the blades, but also affect the capture of wind energy by the wind turbine. Safe and healthy operation is of great significance. At present, the methods used for fault monitoring of fan blades include image recognition, thermal imaging, laser detection, and vibration signal feature recognition. [0003] The Chinese patent application number 201910603546.9 proposes an audio signal detection method based on the effect of contour coefficient optimization K-means clustering, which distinguishes faulty frames and non-faulty frame...

AI Technical Summary

Problems solved by technology

Binary clustering is difficult to accurately detect faults with weak performance characteristics in audio. In addition, manual judgment is required for the existence of periodic features, which is difficult to achieve automatic and efficient detection

[0004] The Chinese patent with the application number CN201710641430.5 also proposes a method of using a sound collection device to diagnose faults through audio, mainly intercepting the characteristic frequency and comparing the characteristic frequency to diagnose the fault. This method is similar to the method based on vibration signals Fault diagnosis, but due to the different propagation modes of audio and vibration signals, and the characteristics of being susceptible to noise interference, the difference in analysis methods is determined

If the audio signal is simply analyzed in the frequency domain, it is difficult to achieve the detection accuracy of the traditional vibration signal using the same method, and the periodic characteristics accompanying the blade sweeping are lost

Images