Compressor blade crack fault detection method

Abstract

The invention discloses a compressor blade crack fault detection method, which comprises the following steps of: collecting acoustic emission signals of two channels of an air outlet of a compressor, and dividing the acoustic emission signals into training samples and test samples; extracting an acoustic emission feature, a time domain feature, a frequency domain feature and a spectrum centroid energy migration feature of the training sample; performing feature selection by using a mixed feature selection method, and determining an optimal feature subset; establishing a two-channel training sample and a test sample feature subset; and combining the two channel training sample feature subsets, training by using a long short-term memory neural network, and finally carrying out blade crack fault classification and detection on a test sample by using the trained long short-term memory neural network to realize crack fault detection of the compressor blade. The method is simple and easy to implement, compared with other existing fault feature and crack detection technologies, the spectral centroid energy migration feature capable of effectively reflecting the blade fault feature can be established, and mixed feature selection and compressor blade crack fault detection are achieved.

Description

technical field [0001] The invention relates to the technical field of acoustic emission signal analysis of rotating machinery, in particular to a method for fault detection of compressor blade cracks. Background technique [0002] Compressors are widely used in petrochemical, electric power and other fields. Blades, as the core components, are prone to crack failures under the action of centrifugal force, friction and unstable airflow loads, which affect the normal operation of the entire compressor. Therefore, timely detection of blade crack failure is of great significance to ensure the safe and stable operation of the compressor. In addition, the acoustic emission signal measured in actual engineering is doped with strong background noise, which causes the crack fault information to be buried in the extracted features, which makes the crack fault detection of compressor blades extremely complicated and difficult. Therefore, how to effectively extract the features reflec...

AI Technical Summary

Problems solved by technology

[0003] Traditional acoustic emission signatures (including energy and amplitude) are very sensitive to cracks, but may not be able to accurately detect blade cracks due to the presence of strong noise

Common features such as time domain, frequency domain and acoustic emission can reflect blade crack fault information to a certain extent, but they are prone to errors under the interference of noise

And extracting too many features causes time-consuming problems, which directly affects the detection efficiency

In addition, once there is a conflict or oversaturation between the characteristics, it will affect the accuracy of fault detection

Images