基于声谱图与改进残差网络的断路器故障诊断研究  

Research on Fault Diagnosis of Circuit Breakers Based on Spectrograms and Improved Residual Networks

作　　者：曹健 陈杉杉 包锡军 从乐瑶 周菁 CAO Jian;CHEN Shanshan;BAO Xijun;CONG Leyao;ZHOU Jing(Wuxi Power Supply Branch,State Grid Jiangsu Electric Power Co.,Ltd.,Wuxi 214061,China)

机构地区：[1]国网江苏省电力有限公司无锡供电分公司,江苏无锡214061

出　　处：《机械与电子》2025年第3期9-15,共7页Machinery & Electronics

基　　金：国网无锡供电公司科技项目(SGJSWX00KJJS2312307)。

摘　　要：针对传统开关柜故障监测装置效果不显著,难以及时发现设备运行时的缺陷,以及现有声纹识别故障诊断方法应用于断路器故障诊断中存在识别准确率低的问题,提出了基于声谱图特征的SE ResNet深度学习神经网络识别模型。首先对声音信号进行声谱图(Spectrogram)特征提取,保留了声音信号的细节,提取出音频信号中的瞬态特性和非稳态特性。随后通过识别模型进行故障诊断。通过实验,针对不同模型以及不同特征提取方式进行比较,验证了所提方法的有效性和优越性。The efficacy of conventional switchgear fault monitoring devices is negligible,posing challenges in rapidly identifying flaws during equipment operation.The current methods used for diagnosing faults in circuit breakers using voiceprint recognition have a low level of accuracy in identifying the faults.This research presents a recognition model for deep learning neural networks called SE ResNet.The model is based on spectrogram characteristics.Initially,spectrogram features are derived from the sound signals,retaining the intricacies and capturing the transient and non constant attributes of the audio signals.Afterwards,fault diagnosis is conducted using the recognition model.The proposed strategy is proven to be successful and superior through experimental validation that compares various models and feature extraction approaches.

关 键 词：断路器 声纹识别 声谱图特征 

分 类 号：TM561[电气工程—电器]