Third-order spectral techniques for the diagnosis of motor bearing condition using artificial neural networks

This paper addresses the development of a novel condition monitoring procedure for rolling element bearings which involves a combination of signal processing, signal analysis and artificial intelligence methods. Seven approaches based on power spectrum, bispectral and bicoherence vibration analyses are investigated as signal pre-processing techniques for application in the diagnosis of a number of induction motor rolling element bearing conditions. The bearing conditions considered are a normal bearing and bearings with cage and inner and outer race faults. The vibration analysis methods investigated are based on the power spectrum, the bispectrum, the bicoherence, the bispectrum diagonal slice, the bicoherence diagonal slice, the summed bispectrum and the summed bicoherence. Selected features are extracted from the vibration signatures so obtained and these are used as inputs to an artificial neural network trained to identify the bearing conditions. Quadratic phase coupling (QPC), examined using the magnitude of bispectrum and bicoherence and biphase, is shown to be absent from the bearing system and it is therefore concluded that the structure of the bearing vibration signatures results from inter-modulation effects. In order to test the proposed procedure, experimental data from a bearing test rig are used to develop an example diagnostic system. Results show that the bearing conditions examined can be diagnosed with a high success rate, particularly when using the summed bispectrum signatures.