Voiceprint Recognition of Transformer Fault Based on Blind Source Separation and Convolutional Neural Network

Min Li, Zhan Huamao, Qiu Annan

202111 citationsDOI

Abstract

The transformer operation sound contains the transformer state information, hence voiceprint recognition technology can be used for transformer fault detection. The accuracy of voiceprint recognition is affected by the interference sound in substation. A method based on blind source separation and convolutional neural network is presented in this paper for transformer fault diagnosis by voiceprint recognition. Different types of sounds in substation are collected and their time domain and frequency domain characteristics are analyzed. A database containing interference sound and transformer fault sound has been built. The blind source separation algorithm is used to separate the interference sound and fault sound. The mel spectrum of the sound is extracted to train the convolutional neural network. Convolutional neural network is used for voiceprint recognition. The result shows that the blind source separation algorithm can effectively separate the interference sound and the fault sound, and the accuracy of voiceprint recognition reaches 98.89%.

Topics & Concepts

Computer scienceConvolutional neural networkTransformerBlind signal separationArtificial intelligencePattern recognition (psychology)Speech recognitionArtificial neural networkEngineeringElectrical engineeringComputer networkVoltageChannel (broadcasting)Blind Source Separation TechniquesAdvanced Algorithms and ApplicationsMachine Fault Diagnosis Techniques