Fault diagnosis on hermetic compressors based on sound measurements

Abstract

A fault identification study is made to identify five common faults in hermetic compressors manufactured in a large plant. Sound power level is used as raw data. Sound measurements were made in a room where microphones were located at different places of a virtual hemi-sphere, designed according to international standards. Obtained data is analyzed using the artificial neural networks method, where the multilayer perceptron model is used. Two different analysis approaches are carried out. In the first approach, only the summary data that emanated from the information coming from all microphones are used. In the second approach, all data coming from all microphones are used. The results indicate that the first approach is partially successful and the second is successful.