Evolutionary spectral based classification of machine vibrations

In this paper, we present a method for time-frequency analysis and classification of electrical machine vibrations based on evolutionary spectrum. Time-frequency analysis provides a means for identifying changes in the vibrations produced by machines. In this work, vibrations generated by different electrical machines are recorded by using accelerometers during operation. Three categories of these vibrations are then defined based on their evolutionary spectra: 1) the normal operation, 2) small level of failure, and 3) high level of failure. A method is presented for automatic detection and classification of abnormality in these machine vibrations using their joint time-frequency moments and neural networks. Simulation results are given to illustrate our algorithm.