电工技术学报  2024, Vol. 39 Issue (12): 3840-3854    DOI: 10.19595/j.cnki.1000-6753.tces.230689
电力电子 |
基于短时傅里叶变换和深度网络的模块化多电平换流器子模块IGBT开路故障诊断
朱琴跃, 于逸尘, 占岩文, 李杰, 华润恺
同济大学电子与信息工程学院 上海 201804
IGBT Open-Circuit Fault Diagnosis of Modular Multilevel Converter Sub-Module Based on Short-Time Fourier Transform and Deep Networks
Zhu Qinyue, Yu Yichen, Zhan Yanwen, Li Jie, Hua Runkai
School of Electrical and Information Engineering Tongji University Shanghai 201804 China
全文: PDF (9972 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 针对现有模块化多电平换流器(MMC)子模块故障诊断过程中所需传感器较多、测量干扰较大等问题,提出一种基于深度学习的MMC子模块IGBT开路故障诊断方法。在对MMC子模块开路故障特征进行分析的基础上,利用短时傅里叶变换(STFT)提取桥臂电压信号的谐波分量信息作为故障诊断所需的特征参数。将所得到的特征参数进行处理后构建故障诊断样本,在通过深度置信网络实现故障类型快速检测的基础上,依据不同故障类型,构建多个基于卷积神经网络的故障定位网络,进而实现开路故障的检测与定位。通过129电平的MMC系统仿真模型和降功率的MMC实验系统搭建,对该文所提方法进行了验证。仿真和实验结果表明,所提故障诊断方法可以在减少传感器数量的基础上实现子模块开路故障的诊断,提高系统的可靠性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
朱琴跃
于逸尘
占岩文
李杰
华润恺
关键词 模块化多电平换流器开路故障诊断短时傅里叶变换卷积神经网络    
Abstract:The modular multilevel converter (MMC) is a new type of multilevel converter that features low switch frequency, high output waveform quality, and low consistency requirements for power electronic device switching. Therefore, it has been widely applied in various fields such as flexible DC transmission, power electronic transmission, and rail transit traction control. However, the increase in the number of sub-modules also means an increase in potential failure points. To meet the growing demands for system safety and reliability, timely and effective diagnosis needs to be carried out after system failures. Existing MMC fault diagnosis methods mainly fall into two categories: additional hardware circuit-based and software algorithm-based. Among them, the hardware-based method exhibits excellent diagnostic accuracy and short diagnostic time but may incur additional hardware costs and wiring complexity, while the software-based method does not require additional hardware circuitry and is relatively simple and convenient to implement. However, existing software-based method often heavily depends on the capacitance voltage information of sub-modules during the diagnosis process. The excessive number of voltage sensors not only increases the complexity of system wiring but also adds to the potential failure points of the system.
To this end, a deep learning-based IGBT open-circuit fault diagnosis method for MMC sub-modules using a minimum number of voltage sensors is proposed, which only requires two arm voltage sensors for each phase. Based on the analysis of open-circuit fault characteristics of MMC sub-module IGBTs, the bridge arm voltage is selected as the required electrical parameter for fault diagnosis. The short-time Fourier transform (STFT) is utilized to extract harmonic component information from the bridge arm voltage signal. By arranging frequency domain information at different sampling times, the matrix containing time-frequency domain information is constructed as the diagnostic sample. By combining with the deep belief network (DBN), an IGBT open-circuit fault detection method based on STFT-DBN is proposed, which realizes the rapid diagnosis of the fault type. Meanwhile, to address the problem of relatively low location accuracy of the DBN, a convolutional neural network (CNN) based faulty sub-module location network is established for each fault type after the fault type is identified by the fault detection network. Combining these two methods allows for rapid fault type detection and high-accuracy faulty sub-module location.
A 129-level MMC simulation model was established using Matlab/Simulink, and arm voltage information was collected under normal operation and various fault conditions. By applying the proposed method for data preprocessing, input samples were constructed, and the dataset was divided. Then the fault diagnosis networks were trained and tested separately, and the effectiveness of the proposed method was verified based on simulation data. Finally, the hardware and software of an MMC experimental system were designed in the laboratory environment, and an experimental system with reduced power levels was constructed. The control unit was implemented by a combination of DSP TMS320F28335 and FPGA, and various fault conditions were simulated by changing the drive signals. Arm voltage sensors were used to collect data from the experimental system under different fault conditions. The experimental results show that the proposed method can achieve high-precision and short-time fault diagnosis by using only arm voltage information.
Key wordsModular multilevel converter(MMC)    open-circuit fault diagnosis    short-time Fourier transform    convolutional neural networks   
收稿日期: 2023-05-18     
PACS: TM46  
基金资助:国家自然科学基金资助项目(51777141)
通讯作者: 朱琴跃 女,1970年生,教授,博士生导师,研究方向为电气智能诊断和电力电子变流控制。E-mail:zqymelisa@tongji.edu.cn   
作者简介: 于逸尘 男,1998年生,硕士研究生,研究方向为模块化多电平换流器的故障诊断。E-mail:2130679@tongji.edu.cn
引用本文:   
朱琴跃, 于逸尘, 占岩文, 李杰, 华润恺. 基于短时傅里叶变换和深度网络的模块化多电平换流器子模块IGBT开路故障诊断[J]. 电工技术学报, 2024, 39(12): 3840-3854. Zhu Qinyue, Yu Yichen, Zhan Yanwen, Li Jie, Hua Runkai. IGBT Open-Circuit Fault Diagnosis of Modular Multilevel Converter Sub-Module Based on Short-Time Fourier Transform and Deep Networks. Transactions of China Electrotechnical Society, 2024, 39(12): 3840-3854.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.230689          https://dgjsxb.ces-transaction.com/CN/Y2024/V39/I12/3840