基于换档全过程振动强度的变压器CM型有载分接开关机械故障诊断方法  

A Mechanical Fault Diagnosis Method of Transformer CM-type On-load Tap Changer Based on the Vibration Strength of the Overall Tap Changing Process

作　　者：梁玄鸿 王有元[1] 古洪瑞 LIANG Xuanhong;WANG Youyuan;GU Hongrui(State Key Laboratory of Power Transmission Equipment Technology(School of Electrical Engineering,Chongqing University),Shapingba District,Chongqing 400044,China)

机构地区：[1]输变电装备技术全国重点实验室(重庆大学电气工程学院),重庆市沙坪坝区400044

出　　处：《中国电机工程学报》2024年第10期4145-4156,I0034,共13页Proceedings of the CSEE

基　　金：国家自然科学基金(重点项目)(51637004)。

摘　　要：变压器有载分接开关(on-load top changer,OLTC)的主要故障类型是机械故障,现有大多数研究仅诊断切换开关故障,难以辨识影响换档全过程的传动机构故障。为准确诊断切换开关与传动机构故障,该文提出一种基于换档全过程振动强度的OLTC机械故障诊断方法。首先,将多通道切换开关振动爆发数据转换为时域波形图输入改进的卷积神经网络(convolutional neural network,CNN),以获取池化层特征。然后,提出换档全过程振动强度特征,将换档全过程振动信号划分为多个区间,统计各区间中幅值超过阈值的点数,以表征各时间段平均振动强度。最后,提出一种新的特征处理方法改变以上两种特征的相对大小,并融合两种特征训练分类器诊断机械故障类型。实例分析表明:相比于现有OLTC机械故障诊断方法,所提方法能有效辨识传动机构故障,进一步提升对切换开关故障的诊断精度,具有较强的鲁棒性与泛用性,可为OLTC机械故障诊断研究提供新的思路。The main fault type of power transformer on-load tap changer(OLTC)is mechanical fault.Most recent studies only diagnose diverter switch faults,and have difficulty in identifying the transmission mechanism faults affecting the whole process of tap changing.To diagnose the diverter switch and transmission mechanism faults more accurately,a mechanical fault diagnosis method of on-load tap changer based on the vibration strength of the overall tap changing process is proposed in this paper.First,multi-channel vibration burst data during the diverter switch operation is transformed into time-domain waveform images and input into improved convolutional neural network(CNN)to obtain pooling layer features.Then,the vibration strength feature of the overall tap changing process is proposed.The vibration signal of overall tap changing process is divided into multiple intervals,and the number of points in each interval whose amplitude exceeds the threshold is counted to characterize the average vibration strength in the interval.Finally,a novel feature processing method is proposed to change the relative size of the above two features,and the features are fused to train classifier for mechanical fault diagnosis.The case analysis shows that compared with the existing OLTC mechanical fault diagnostic methods,the proposed method can effectively identify transmission mechanism faults,further improve the diagnostic accuracy of diverter switch faults,and achieve strong robustness and generalizability,which provides a new idea for OLTC mechanical fault diagnosis.

关 键 词：有载分接开关 机械故障 故障诊断 卷积神经网络 特征提取 特征处理 

分 类 号：TM403[电气工程—电器]