基于MWKCNN与迁移学习融合的风机轴承故障诊断方法  

Fault Diagnosis Method ofWind Turbine Bearing Based on MWKCNN-transfer Learning Fusion

作　　者：宁兆秋 张东[2,3] 方文墨 孙志强 孙明 徐继文[4] 杨巍[4] 郑伟[4] 单春海 NING Zhaoqiu;ZHANG Dong;FANG Wenmo;SUN Zhiqiang;SUN Ming;XU Jiwen;YANG Wei;ZHENG Wei;SHAN Chunhai(School of Electric Power,Shenyang Institute of Engineering,Shenyang 110136;Institute of Engineering and Technology,Shenyang Institute of Engineering,Shenyang 110136;Institute of carbon peaking and carbon neutrality,Shenyang Institute of Engineering,Shenyang 110136;Shenyang Aircraft Industry(Group)Co.,Ltd.,Shenyang 110134,Liaoning Province)

机构地区：[1]沈阳工程学院电力学院,辽宁沈阳110136 [2]沈阳工程学院工程技术研究院,辽宁沈阳110136 [3]沈阳工程学院碳达峰碳中和研究院,辽宁沈阳110136 [4]沈阳飞机工业(集团)有限公司,辽宁沈阳110034

出　　处：《沈阳工程学院学报（自然科学版）》2024年第3期7-14,共8页Journal of Shenyang Institute of Engineering:Natural Science

基　　金：辽宁省科技厅创新能力提升联合基金(2022NLTS1601);沈阳市中青年科技创新人才支持计划(RC210143)。

摘　　要：针对实际风电场的风机运行工况多变、数据完备性缺失导致的风机轴承故障诊断精度降低的问题,提出一种基于多重宽核卷积神经网络(multiple wide kernel convolutional neural networks,MWKCNN)与迁移学习融合的风机轴承故障诊断方法。首先,在源域训练MWKCNN风机轴承故障诊断模型;其次,根据3个目标域与源域的相似度,利用基于模型微调的迁移学习方法对源域的MWKCNN模型结构进行调整,并且用实际的轴承数据集进行验证;最后,通过仿真实验进行验证。结果表明:MWKCNN模型对源域的风机轴承故障诊断的精度达到了99.48%;在3种数据完备性缺失的目标域,对风机轴承故障诊断的精度均达到了94%以上;相比于其他模型的迁移效果,MWKCNN模型对轴承振动信号故障特征的挖掘能力更强。Aiming at the problem that the fault diagnosis accuracy of wind turbine bearings is reduced due to the variable operating conditions of wind turbines in actual wind farms and the lack of data completeness,this paper proposes a wind turbine bearing fault diagnosis method based on Multiple Wide Kernel Convolutional Neural Networks and transfer learning fusion.Firstly,MWKCNN wind turbine bearing fault diagnosis model is trained in source domain.Secondly,according to the similarity between the three target domains and the source domain,the MWKCNN model structure of the source domain is adjusted by the transfer learning method based on model fine-tuning,and verified by the actual bearing data set.The simulation results show that the fault diagnosis accuracy of the MWKCNN model in the source domain reaches 99.48%,and the fault diagnosis accuracy of the wind turbine bearing reaches more than 94%in the three target domains with missing data completeness.Compared with other model migration effects,the MWKCNN model has stronger bearing vibration signal fault feature mining capabilities.

关 键 词：多重宽核卷积神经网络 风机轴承 故障诊断 迁移学习 变工况数据量缺失 下采样损失 

分 类 号：TM315[电气工程—电机] TH165.3[机械工程—机械制造及自动化]