摘要
针对直升机附件齿轮箱在有限多工况条件下故障特征提取难度大、识别准确率低等问题,提出一种结合变分模态分解(variationalmodedecomposition,简称VMD)与多尺度卷积神经网络(multi-scaleconvolutionalneural netwo,简称MCNN)的故障诊断方法。首先,对直升机附件齿轮箱进行地面实验和信号采集,对原始信号进行滤波、降噪等预处理;其次,利用VMD将信号分解为若干个固有模态(intrinsic mode functions,简称IMF),依据齿轮副频率特性对分解模态进行重构与归一化,增强微弱的高频故障特征;最后,将重构信号的每个分量视作不同尺度,经多尺度卷积神经网络进行多尺度特征提取并融合,由指数归一化分类器给出识别的故障类别。实验结果表明,所提方法能够有效增强信号故障特征,挖掘多工况条件下信号的差异性与同一性,在直升机附件齿轮箱振动故障诊断中平均准确率为97.25%。
Aiming at the problems of difficulty in fault feature extraction and low recognition accuracy of helicopter accessory gearbox under limited variable working conditions,a fault diagnosis method is proposed com‑bining variational mode decomposition(VMD)and multi-scale convolution neural network(MCNN).Firstly,the helicopter accessory gearbox is tested on the ground and sampled,and the original signal is preprocessed by filtering and noise reduction.Secondly,the VMD decomposition signal is used as several intrinsic mode functions(IMF)to reconstruct and normalize the decomposition modes according to the frequency characteristics of the gear meshing ground,so as to enhance the weak high-frequency fault characteristics.Finally,each component of the reconstructed signal is regarded as a different scale,and multi-scale features are extracted and fused by MCNN.The identified fault category is given by softmax classifier.The test results show that the proposed method can effectively enhance the signal fault characteristics,excavate the difference and identity of signals under multiple working conditions.In the vibration fault diagnosis of helicopter accessory gearbox,the average accuracy rate is 97.25%.
作者
万安平
龚志鹏
王景霖
单添敏
何家波
WAN Anping;GONG Zhipeng;WANG Jinglin;SHAN Tianmin;HE Jiabo(Department of Mechanical Engineering,Zhejiang University City College Hangzhou,310015,China;Aviation Key Laboratory of Science and Technology on Fault Diagnosis Health Management Shanghai,201601,China)
出处
《振动．测试与诊断》
EI
CSCD
北大核心
2024年第2期246-252,406,共8页
Journal of Vibration,Measurement & Diagnosis
基金
航空科学基金资助项目(20183333001)
国家自然科学基金资助项目(52372420)
中国博士后基金特别资助项目(2018T110587)
浙大城市学院科研培育基金资助项目(J-202220)。
关键词
变分模态分解
多尺度卷积网络
振动故障诊断
附件齿轮箱
variational modal decomposition
multi-scale convolution network
vibration fault diagnosis
accessory gear box
作者简介
第一作者:万安平,男,1983年11月生,博士、副教授。主要研究方向为复杂装备健康管理及维修决策等。E-mail:wanap@hzcu.edu.cn;通信作者:何家波,男,1986年2月生,博士、助理研究员。主要研究方向为复杂设备故障诊断与智能制造。E-mail:jiabohe@zju.edu.cn。
