Research On Electrical Equipments Fault Detection Based On Audio Feature

Electrical equipment fault detection is a technology that detects the occurrence of abnormal electrical equipments working through real-time health monitoring. It can analyze and judge the electrical equipment fault type and predict the future operation situation. Traditional fault detection methods for electrical equipment are regular maintenance and state maintenance. These methods hav played a prominent role in the prevention of electrical fault. However, these methods are generally achieved through contact detection, and they are not real-time monitoring of electrical equipment working conditions. With the continuous expansion of the scale of the grid and the use of smart devices and instruments, traditional methods can not meet the requirements of the monitoring equipment. So an on-line real-time monitoring and fault diagnosis expert system has been developed.In this paper, we proposed an online electrical equipment fault detection method based on audio feature analysis. The audio signal processing is now widely used in engines and other mechanical equipment fault diagnosis. Experienced staff can work with the sound emitted by the electrical device in the transmission station to determine whether the device is abnormal or not. Therefore, we can design a scheme to monitor the electrical equipments according to the sound signal and it can be used as an effective complement to other electrical equipment monitoring. First we use a microphone array to collect the electrical equipments working sound. Microphone array can form a beam to reduce the noise and reverberation, which can effectively suppress ambient noise to get to the best sound signal. Electrical equipment around the work environment is very complicated, and when we collect the sound, it is inevitably to be mixed with other interfering signals, such as human voices and bird singing sound. These interference sources are statistically independent. This paper describes an independent component analysis (ICA) based on a signal separation method to isolate the collected mixed sound, we can use FastICA algorithm that is based on the largest negentropy to separate the collected sound to several independent source signals. It has the advantage of fast convergence and robust. And then we extracts the characteristic parameters of these isolated sound signals. We choose Mel-frequency cepstral cofficients as the characteristic parameter of the sound signal. Finally we use DTW algorithm to determine whether the electrical equipment is working good or not. The simulation result shows this algorithm can effectively separate the multiple independent source signals. The separation accuracy is above95%for typical sample mixed sounds and the reliability of electrical equipment fault detection system based on audio signal processing is ensured.