Valves are the key parts of the reciprocating compressor and faults frequently occur because of its working for a long time and in bad environment. It is extremely difficult to extract feature and establish failure identification model due to the non-stationarity and nonlinearity of the compressor valve vibration signal. According to the difficult diagnosis of large reciprocating compressor valves, the information entropy with good fault-tolerance ability was extracted as the feature parameter, which outlines the overall statistical characteristic of the signal. The extracted wavelet packet entropy was used as input vector to construct the decision function. For solving the defect of traditional classificatory with many samples, a valve wear failure classifier based on Support Vector Machine (SVM) is proposed, The new SVM classifier can be trained in a few samples rapidly to recognize several kinds of new faults. The experimental result demonstrates it was effective of the model in the non-stationarity signal feature extraction and nonlinearity pattern classification with a few samples and the recognizing correct rate increased more greatly compared with traditional BP method.
[1]
Tao Shen,et al.
EXTRACTING INFORMATION ENTROPY FEATURES FOR ROTATING MACHINERY VIBRATION SIGNALS
,
2001
.
[2]
Patrick Haffner,et al.
Support vector machines for histogram-based image classification
,
1999,
IEEE Trans. Neural Networks.
[3]
Yajuan Liu,et al.
Rotating Machinery Fault Diagnosis Based on Support Vector Machine
,
2010,
2010 International Conference on Intelligent Computing and Cognitive Informatics.
[4]
Xu Min-qiang.
Method to Diagnose the Valve Faults of Reciprocating Compressors by Correcting the Indicating Diagrams
,
2009
.
[5]
Zhou Ping.
Fault diagnosis of valve train based on EMD and neural network
,
2008
.
[6]
Chih-Jen Lin,et al.
A comparison of methods for multiclass support vector machines
,
2002,
IEEE Trans. Neural Networks.
[7]
Guo Bao-sheng.
Fault Diagnosis Study of Leaking Gas of Valves Based on Vibration Signal
,
2007
.