Circuit Breaker Target Sound Signal Detection Method based on VAD and SVDD Algorithms

Currently, the closed-set recognition algorithm is the primary research method for switchgear circuit breaker fault diagnosis. For real-time acquisition of the operation of a long voice signal from a switchgear circuit breaker, a significant number of silent signal segments and non-circuit breaker action sound signals are present; these types of signals will cause false alarms with the closed-set recognition algorithm. As a result, this article proposes a method for extracting the target sound signal by combining the deep support vector date description(SVDD) algorithm with the voice activity detection (VAD) algorithm. Initially, the double threshold endpoint detection algorithm is employed to intercept lengthy sound samples in order to eliminate voiceless segment sound signals. Subsequently, the Deep SVDD algorithm is utilized to train the model to acquire the capability of single classification, thereby excluding abnormal sound signals that do not correspond to circuit breaker actions and diminishing the rate of false alarms.