基于深度学习的核电站泵类电机故障诊断方法  

Deep learning-based fault diagnosis method for pump motors in nuclear power plants

作　　者：尹文哲 夏虹[1,2] 朱少民 王志超[1,2] 张汲宇 姜莹莹 YIN Wenzhe;XIA Hong;ZHU Shaomin;WANG Zhichao;ZHANG Jiyu;JIANG Yingying(Key Laboratory of Nuclear Safety and Advanced Nuclear Energy Technology(Harbin Engineering University),Ministry of Industry and Information Technology,Harbin 150001,China;Fundamental Science on Nuclear Safety and Simulation Technology Laboratory,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学核安全与先进核能技术工业和信息化部重点实验室,黑龙江哈尔滨150001 [2]哈尔滨工程大学核安全与仿真技术重点学科实验室,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2024年第12期2350-2357,共8页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(U21B2083)。

摘　　要：为了准确识别核电站泵类电机的运行状态以及增强故障诊断系统的鲁棒性,本文提出了一种基于深度学习的故障诊断方法。利用多个传感器采集鼠笼式三相异步电机不同位置处的振动信号对电机状态进行分析,采用变分模态分解和短时傅里叶变换对电机振动信号进行处理,将获取的时频特征输入到深度残差神经网络中,以得到诊断结果,并对比了单传感器和多传感器监测策略的测试结果。结果表明:提出的方法能够准确识别电机故障,且具有良好的鲁棒性,2种监测策略的鲁棒性能差距明显;比较不同的诊断模型与本文方法表明本文方法的鲁棒性能最优。A fault diagnosis method based on deep learning was proposed in this study to accurately identify the operational state of pump motors in nuclear power plants and enhance the robustness of fault diagnosis systems.Vibration signals at various positions of squirrel-cage three-phase asynchronous motors were collected by multiple sensors to analyze the state of the motor.Variational mode decomposition(VMD)and short-time Fourier transform(STFT)were applied to process the vibration signals,extracting time-frequency features.These features were input into a deep residual neural network(ResNet)for fault identification,yielding diagnostic results.The effectiveness of the proposed method was verified through experiments,which included comparing the test results between single-and multi-sensor monitoring strategies.The results demonstrate that the proposed method accurately identifies motor faults and exhibits strong robustness.A notable difference in robust performance is observed between the two monitoring strategies.In addition,different diagnostic models were established and compared with the proposed method,with the comparison showing that the proposed method has the best robust performance.

关 键 词：核电站 电机 故障诊断 深度学习 时频分析 残差神经网络 变分模态分解 短时傅里叶变换 

分 类 号：TK05[动力工程及工程热物理]