Recognition of facility sound with background noises

The detection of abnormality in a facility is vital for plant operation. The methods by cameras and sensors are traditionally used but not sufficient for abnormal detection in the early stage. On the other hand, human feels the change of surroundings by various sensings such as eyes, noses, and ears. The diagnosis by the sound has the advantage of being able to detect the wide-ranging abnormalities. The purpose of this study is the recognition of various facility sounds with background noise by use of the signal processing. Sounds of 9 facilities in the plant are recorded. The recorded sounds of facilities were preprocessed by applying Fast Fourier Transformation. The features of sounds were extracted and classified by a Neural Network. As a result of the test, sounds of 9 facilities were recognized with over a certainty of about 94[%]. Moreover, the equipment in the factory was diagnosed by applying the recognition system. The abnormality of the turbine was artificially made as an example and the diagnostic result was successful.