Localization Method of Multi-Source Abnormal Sound Signals of Power Equipment Based on GMUSIC

The safety and stability of power equipment in the substation are critical for the power grid. Whether the equipment in an abnormal state could be analyzed by the change of audible sound characteristics which contains abundant operation information. Furthermore, the localization of abnormal sound sources can determine whether the equipment part has a potential defect. In practical applications, because of the requirements for accuracy and real-time performance, large arrays and small snapshots are most used as the detection system. It leads to the failure of the traditional localization method, especially for multiple sound sources with a similar angle. However, there are usually multiple sound sources in substations when detecting abnormal sound for power equipment. Therefore, a sound source localization method based on a uniform planar matrix and GMUSIC algorithm was proposed. The sample covariance matrix and corresponding spatial spectrum of raw sound data are built, and then the source can be located by searching the spectral peak of the spatial spectrum. The field tests indicate that the proposed method can resolve adjacent sound sources with fewer snapshots, and the localization success rate was highly improved.