A Voiceprint Diagnosis Method for Mechanical Faults of 10kV Circuit Breakers Based on Improved FastICA and Multi-Feature Fusion
The mechanical fault model of 10 kV circuit breakers based on voiceprint features is susceptible to environmental noise,resulting in low recognition accuracy and long recognition time.This article proposes a multi-feature hybrid voiceprint diagnosis method for mechanical faults of 10 kV circuit breakers based on improved FastICA and Bi-LSTM.Firstly,the FastICA algorithm is improved by using Pearson coefficients.The collected sound is separated by noise using the improved FastICA algorithm,and pure 10 kV circuit breaker state voiceprint signals are extracted.Then,the Fourier transform is used to analyze the frequency domain information of the 10 kV circuit breaker in various states.Based on the analysis results,appropriate time-domain,frequency-domain,and acoustic features are selected,and through difference analysis,features with high contribution are selected to form one-dimensional mixed features.Finally,using mixed features as diagnostic criteria,a fault classification model based on Bi-LSTM is established.The results indicate that this method can effectively identify eight common mechanical faults in 10 kV circuit breakers.The recognition accuracy can reach 99.3%,meeting the accuracy and speed requirements of the power grid for electrical equipment fault diagnosis.
国家科技期刊平台
NSTL
万方数据
