Reduced interference distributions for the detection and classification of outside sound source acoustic emissions

The problem of detecting and classifying outside sound sources (i.e. diesel engines, rotors, etc.) by their acoustic emissions is applicable to a number of industries. The key to successful detection and classification is selecting harmonically related narrowband peaks in the frequency domain. These harmonically related narrowband features an directly attributed to physical phenomena (i.e. engine firing rates and rotor RPM). The spectral signatures of these sources are non-stationary and require the use of time-frequency analysis techniques to achieve high resolution spectral estimates. This paper demonstrates the use of time-frequency distributions of Cohen's class (1989) on real data for a classification problem. Reduced interference distributions (RIDs), developed by Williams (1989) and others, are successfully applied resulting in high resolution spectral estimates.