基于Mel时频谱-卷积神经网络的变压器铁芯夹件松动故障声纹模式识别

Voiceprint Recognition of Transformer Core Clamp Looseness Fault by Mel-spectrum and Convolutional Neural Network

铁芯的振动噪声与其紧固状况密切相关,对变压器铁芯松动故障进行声纹特性分析和故障诊断研究是实现电力变压器在线监测的重要内容。提出了一种基于Mel时频谱-卷积神经网络的变压器铁芯松动声纹识别方法。首先提出了基于Mel时频谱的噪声样本处理方法;然后搭建了铁芯夹件松动故障模型,对铁芯在不同松动程度与不同松动模式下的噪声信号进行研究,分析了噪声信号的频域特征;最后构建了Mel时频谱-卷积神经网络变压器铁芯松动声纹识别模型,以Mel时频谱降维处理后的噪声数据,作为深度学习的数据集,实现了铁芯松动故障的准确识别。可为变压器铁芯夹件松动故障诊断以及电网主设备的数据深度挖掘提供参考。

As transformer core clamp looseness leads to vibration noise,it is important to conduct online monitoring based on the analysis of its voiceprint characteristics and fault diagnosis.This paper presents a voiceprint recognition method of transformer core looseness based on Mel-spectrum and convolutional neural network(CNN).Firstly,we processed noise samples by Mel-spectrum.Then,we studied the noise signals of transformer core under different looseness degrees and modes through a self-designed transformer core looseness model,and analyzed their frequency domain characteristics.Finally,we constructed a transformer core looseness voiceprint recognition model by using Mel-spectrum for acoustic data dimension reduction and CNN for deep learning from the processed dataset,accurately identifying the transformer core looseness faults.The research can provide a reference for the fault diagnosis of transformer core clamping looseness and the deep data mining of the power grid.