Motor current signal analysis using a modified bispectrum for machine fault diagnosis

This paper presents the use of the induction motor current to identify and quantify common faults within a two-stage reciprocating compressor. The theoretical basis is studied to understand current signal characteristics when the motor undertakes a varying load under faulty conditions. Although conventional bispectrum representation of current signal allows the inclusion of phase information and the elimination of Gaussian noise, it produces unstable results due to random phase variation of the sideband components in the current signal. A modified bispectrum based on the amplitude modulation feature of the current signal is thus proposed to combine both lower sidebands and higher sidebands simultaneously and hence describe the current signal more accurately. Based on this new bispectrum a more effective diagnostic feature namely normalised bispectral peak is developed for fault classification. In association with the kurtosis of the raw current signal, the bispectrum feature gives rise to reliable fault classification results. In particular, the low feature values can differentiate the belt looseness from other fault cases and discharge valve leakage and intercooler leakage can be separated easily using two linear classifiers. This work provides a novel approach to the analysis of stator current for the diagnosis of motor drive faults from downstream driving equipment.