Automated Fault Detection of Tri-Lobe Compressor Using Wavelet Transformation and Support Vector Machine

In order to diagnose the faults of industrial rotating machines automatically, an expert system is used in this paper. A tri-lobe roots blower compressor is used as a test rig to represent an industrial machine. The proposed method for training the expert system includes: data acquisition, signal processing and intelligent pattern recognition stages. Acceleration signals of healthy and faulty compressor components were acquired in the first stage. The signals were conditioned to be used for the signal processing as the next stage. It is necessary to find pattern recognition criterion of the compressor fault diagnosis. Therefore feature extraction of data was performed as part of the second stage. In the third stage, a support vector machine tool was trained and employed to classify the faults. The proposed procedure was tested and the obtained results showed that this algorithm works very well and it fully classifies the faults automatically.