Sound pattern recognition from multiple sources in an enclosed space

Abstract

In repair workshops or production factories, technical personnel are often required to identify the faulty components or devices in a malfunctioning machine system. To facilitate such fault diagnosis, this research aims at developing a method for identifying those components or devices which, due to their changed sound arising from faults, cause variations in the overall sound field in an enclosed space. To do this, three microphones, hung at the centre and the ends of a rod with a given length, were made to rotate in a plane above the sound sources in order to acquire information on the sound distribution in the enclosure. The statistical properties (auto spectra and cross spectra) of the acquired signals, as functions of sound frequency and rotating angle, were taken as the object for pattern recognition.