摘要
螺栓松动产生的声音信号往往是非平稳的,其信号特性随时间变化,但现有算法难以准确捕捉松动特征,导致识别敏感度低。为此,展开基于深度卷积神经网络的输电铁塔螺栓松动声纹识别算法研究。对输电铁塔螺栓松动声音信号进行预处理,通过对声音信号进行分帧和加窗处理,确保信号的连续性和平滑过渡。采用小波变换提取松动特征,利用分形盒维数矩阵和相关系数等方法进一步细化特征描述。通过深度卷积神经网络对处理后的声纹数据进行学习,实现松动区域的准确定位和螺栓松动状态的声纹识别。对输电塔各个部位测点实测数据的分析表明,该方法对螺栓松动的识别灵敏度较高,对判断螺栓松动状态有一定的增强。
The sound signal generated by bolts looseness is often non⁃stationary,meaning the characteristics of the signal change over time.However,existing algorithms struggle to accurately capture the loosening characteristics,resulting in low recognition sensitivity.Therefore,Deep Convolutional Neural Network⁃based voiceprint recognition algorithm is proposed for recognition bolts looseness in transmission towers.The sound signal of loose bolts in transmission towers is preprocessed,and the sound signal is divided into frames and windowed to ensure signal continuity and smooth transition.Wavelet transform is used to extract loose features,and methods such as fractal box dimension matrix and correlation coefficient are used to further refine the feature description.By using Deep Convolutional Neural Networks to learn the processed voiceprint data,accurate localization of loose areas and voiceprint recognition of bolt looseness status can be achieved.The analysis of measured data from various parts of the transmission tower show that this method has high sensitivity in identifying bolt looseness and has a certain enhancement in judging the state of bolts looseness.
作者
陈春超
周健
杨家龙
周旭日
张引
CHEN Chunchao;ZHOU Jian;YANG Jialong;ZHOU Xuri;ZHANG Yin(Jiangsu Provincial Power Transmission and Transformation Co.,Ltd.,Nanjing 211102,China;State Grid Yancheng Power Supply Company,Yancheng 224008,China)
出处
《电子设计工程》
2025年第24期168-173,共6页
Electronic Design Engineering
基金
国网江苏省电力有限公司科技项目(J2023155)。
关键词
深度卷积神经网络
输电铁塔
螺栓松动
螺栓声纹识别
Deep Convolutional Neural Network
transmission towers
bolts looseness
bolt voiceprint recognition
