Analysis of signal processing for abnormal acoustic diagnostics of rotational machine (Extraction of inter frame peak frequency fluctuation)

It is important to detect abnormalities at an early stage to efficiently maintain industrial machines. Acoustic sensors, i.e., microphones, have an advantage in that they do not need to be in direct contact with the point of diagnostic. For using acoustic sensors, however, we must choose the best acoustic feature extraction method. The power spectrum method is often used for predictive diagnosis. Though, the differences between normal and abnormal power spectrum is small. Therefore, we focused on pitch changes to distinguish failure signals by calculating the inter frame peak frequency fluctuation with relative standard deviation (RSD). We evaluated its fluctuation by simulating motor failure, and found that the differential fundamental peak-frequency’s RSD is more than 10%. We confirmed the value that was not be detected during normal operation mode. And then, we concluded that using fluctuation in peak-frequency is suitable for early abnormal detection.