Improved Mel Frequency Cepstral Coefficients for Compressors and Pumps Fault Diagnosis with Deep Learning Models

• DOI: 10.3390/app14051710
• Publication Date: 2024-02-20T12:50:26Z
• Authors: Diego Cabrera; Ruben Medina; Mariela Cerrada; René-Vinicio Sánchez; Edgar Estupiñan; Chuan Li
• Keywords: Technology; 0209 industrial biotechnology; QH301-705.5; T; Physics; QC1-999; long short term memory; 02 engineering and technology; fault diagnosis; multi-stage centrifugal pumps; Engineering (General). Civil engineering (General); Chemistry; convolutional neural networks; mel frequency cepstral coefficients; TA1-2040; Biology (General); reciprocating compressors; QD1-999

Compressors and pumps are machines frequently used in petroleum chemical industries for fluid transportation through flow systems to keep industrial processes running permanently. As their failure can produce costly disruption, developing fault detection diagnosis tools is essential accurately detecting diagnosing faults. This research proposes a bi-dimensional representation of the vibration signal corresponding Mel Frequency Cepstral Coefficients (MFCC) first two derivatives as features. The pseudo-periodic nature signature rotating exploited put forward an efficient accurate patch-wise classification method. approach enables 13 combined types faults multi-stage centrifugal pump 17 reciprocating compressor. Classification performed using Long Short-Term Memory (LSTM) network, bidirectional (BiLSTM) neural Convolutional Neural Network (CNN). Accurate over 99% attained, showing that proposed feature extraction procedure correctly classifies large set simultaneously appearing such machines.