A compressor fault diagnosis method based on parameter identification

Abstract

The invention discloses a compressor fault diagnosis method based on parameter identification, which comprises the following steps: first, the original compressor fault signal is decomposed to obtain a series of components; and then each component is separately obtained from the original compressor fault signal The K-L divergence, mutual information and correlation coefficient values ​​between the components are used as the three-dimensional values ​​of the comprehensive index of the component; then the comprehensive index is used as the feature vector of each component, and the feature vectors of all components are formed into a set, and the set The elements are clustered hierarchically, so that the real components and false components are clustered into two categories; finally, the false components are eliminated. The invention has the advantages that: based on the invention, the analysis error of the signal decomposition method can be reduced, and a fault diagnosis feature threshold library can be constructed.

Description

technical field [0001] The invention relates to the technical field of time-frequency domain analysis of vibration response of rotating machinery, in particular to a parameter identification-based compressor fault diagnosis method. Background technique [0002] The deterioration mechanism of aerodynamic characteristics of a gas turbine compressor is relatively complex, and there are many possible causes of failure. In actual production, in order to accurately locate the fault excitation source, it is necessary to conduct in-depth processing and analysis of the fault parameters. In the process of extracting fault parameter features and building a real-time diagnosis platform, the error of the signal processing method often has a very significant impact on the diagnosis results. [0003] Integrated Empirical Mode Decomposition (EEMD), Hilbert Vibration Decomposition (HVD) and other signal decomposition methods have been widely used in various fields, but these methods have th...

AI Technical Summary

Problems solved by technology

These identification indicators need to distinguish between real and false components by pre-setting thresholds, which cannot realize automatic identification of true and false components; at the same time, the set thresholds are only a summary of the law of the test samples, lacking theoretical support, and are prone to one-sidedness and nature of the test sample set. The impact of a single indicator on the misdiagnosis rate is difficult to promote

The above indicators often cannot fully reflect the internal relationship between the components and the original signal during use

Images