Fault Feature Extraction Method of Reciprocating Compressor Valve Based on SPA-MF

• Published in: IEEE access Vol. 10; p. 1
• Main Authors: Li, Ying, Pan, Yunjie, Ba, Peng, Wu, Shihu, Chen, Jiawen
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Decomposition; Fault diagnosis; Feature extraction; Fractal analysis; Fractals; Least squares method; MFDFA; Reciprocating compressor valve; Reciprocating compressors; Search algorithms; Signal analysis; smooth prior analysis; SPA-MF; Support vector machines; Trends; Valves; Vibrations; Wavelet analysis
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3226512

Aiming at the problem that the traditional multifractal detrended fluctuation analysis (MFDFA) using the least squares method to fit the trend term is prone to overfitting and takes a long time, this paper proposes a new non-stationary signal analysis method-smoothed prior analysis multifractal (SPA-MF).Firstly, the time sequence data is adaptively decomposed by smooth prior analysis (SPA) to eliminate the local trends of sequence data at different scales, and then the multifractal analysis is performed on the detrended data obtained by the decomposition. At the same time, the sparrow search algorithm (SSA) is used to optimize the parameter of the SPA, so as to eliminate the trend item data more accurately. Through the simulation signal which composed of the BMS signal and noise signal, the feasibility of SPA-MF for feature extraction is proved. Finally, SPA-MF is applied to extract the features of the reciprocating compressor valve vibration signal, and the extracted reciprocating compressor valve features are input into support vector machine (SVM) for classification and recognition. Through the analysis of the experimental results, it can be seen that the recognition rate of the valve features obtained by the traditional MFDFA method is only 87.5%, and the recognition rate of the SPA-MF method proposed in this paper reaches 96.87%, and the time spent on feature extraction using SPA-MF is only about 36% of that of MFDFA method, which proves the SPA-MF method is a feature extraction method with high accuracy and effectiveness.