文献类型：期刊文献

中文题名：基于多模态特征融合的汽轮发电机转子故障诊断

英文题名：Turbo-Generator Rotor Fault Diagnosis Based on Multimodal Feature Fusion

作者：彭爽 杨仁增 毛先胤

第一作者：彭爽

机构：[1]贵州大学电气工程学院,贵州贵阳550025;[2]贵州理工学院贵州省电力大数据重点实验室,贵州贵阳550003;[3]贵州电网有限责任公司电力科学研究院,贵州贵阳550002

第一机构：贵州大学电气工程学院,贵州贵阳550025

年份：2026

卷号：39

期号：5

起止页码：72-79

中文期刊名：电子科技

外文期刊名：Electronic Science and Technology

基金：贵州省科技基金(20181068)。

语种：中文

中文关键词：汽轮发电机;变分模态分解;时序卷积网络;算术优化算法;峭度准则;振动信号;转子故障诊断;多模态特征

外文关键词：turbo-generator;variational mode decomposition;temporal convolutional network;arithmetic optimization algorithm;kurtosis criterion;vibration signal;rotor fault diagnosis;multimodal features

摘要：针对汽轮发电机振动信号易受噪声干扰产生的故障诊断识别难度问题,文中提出了一种基于变分模态分解(Variational Mode Decomposition,VMD)与时序卷积网络(Temporal Convolutional Network,TCN)的智能故障诊断方法。通过算术优化算法确定VMD的模态分量个数k值和惩罚因子α的最优组合,从而利用优化后的VMD将发电机振动数据分解得到多个模态分量,并根据峭度准则进行信号重构。将重构后的信号输入TCN进行特征学习,从而有效识别和诊断转子故障。以一台600 MW汽轮发电机故障前后转子振动数据为样本进行测试,结果表明所提方法对发电机故障的识别准确率为98.13%,与其他智能故障诊断模型相比提高了4~8百分点。
In view of the problem of difficulty in fault diagnosis and identification caused by the vibration signal of steam turbine generators being easily disturbed by noise,an intelligent fault diagnosis method based on VMD(Variational Mode Decomposition)and TCN(Temporal Convolutional Network)is proposed.The optimal combination of the number k value of modal components and the penalty factorαof VMD is determined through the arithmetic optimization algorithm.Thus,the generator vibration data is decomposed by the optimized VMD to obtain multiple modal components,and the signal reconstruction is carried out according to the kurtosis criterion.The reconstructed signal is input into the TCN for feature learning,thereby effectively identifying and diagnosing rotor faults.The rotor vibration data before and after the failure of a 600 MW steam turbine generator are tested as samples.The results show that the proposed method has an accuracy rate of 98.13%in identifying generator faults,which is 4 to 8 percentage points higher than that of other intelligent fault diagnosis models.