Plant equipment diagnosis by sound processing

Abstract

This paper describes abnormal sound detection in plant equipment. Their target was to detect steam leakages, which is one of the most important indications of power plant failure in its early stages. Fourier analysis, which is a conventional method for the sound processing, cannot detect small abnormal sounds mixed with large normal sounds. The authors focus on the characteristics of adaptive digital filters (ADF), which enhance a small signal in a large background noise, and propose an abnormal sound detection method by comparing the properties between normal and abnormal filters produced by the ADF process. Finally, they show the efficiency of this method by applying it to sounds recorded in a power plant. The results of its application to sound source detection are also reported.