摘要
针对振动信号或红外图像等单类型传感器信息难以准确表征机械设备的健康状态,存在诊断不确定性的问题,提出基于改进卷积神经网络(CNN)的多源异构信息数据级融合诊断方法。首先采用变分模态分解(VMD)和希尔伯特变换(HT)方法将振动信号处理成与红外图像同维的时频图像,并将其与红外图像进行数据级融合,得到多通道融合信号,然后将该信号输入到多通道卷积神经网络中进行训练以构建融合诊断模型,最后通过转子系统故障诊断实例验证了所提方法的正确性。研究结果表明:CNN、SAE和DBN等特征学习方法可以自适应逐层提取特征,提取的特征包含更多有用的诊断信息,优于人工提取特征;多源异构数据对不同故障类型敏感性不同,具有互补性,融合诊断正确率更高,可实现对设备健康类型更精准的判断;多源信息融合诊断方法可以很好地保留原始输入信息,对实验室转子系统故障实现精确诊断;该方法可以减少数据需求量,在小样本背景下具有良好的诊断性能;该方法对噪声敏感性低,在噪声环境下具有较好的鲁棒性和抗噪性。研究结果可为旋转机械的故障诊断提供一定的参考。
To address the difficulty of single-type sensor information such as vibration signals or infrared images to accurately reflect the health status of mechanical equipment,and the problem of diagnosis uncertainty,a multi-source heterogeneous information data-level fusion diagnosis method based on improved convolutional neural network(CNN)is proposed.Firstly,the vibration signal is processed into a time-frequency image with the same dimension as the infrared image by using the variational modal decomposition(VMD)and Hilbert transform(HT)methods,and the data-level fusion is carried out with the infrared image to obtain a multi-channel fusion signal.And then the signal is input into a multi-channel CNN for training to build a fusion diagnosis model.Finally,the correctness of the proposed method is verified through a rotor system fault diagnosis example.The research results show that:CNN,SAE,DBN and other feature learning methods can adaptively extract features layer by layer.The extracted features contain more useful diagnostic information,which is better than manually extracted features.The multi-source heterogeneous data has different sensitivity to different fault types,which is complementary,and has higher fusion diagnosis accuracy,and allows more accurate judgment of equipment health type.The method can well retain the original input information,and provide accurate diagnosis of laboratory rotor system failure.This method can reduce data requirements,providing good diagnostic performance in a small sample background.It has low sensitivity to noise,and can achieve better robustness and noise resistance in a noisy environment.The research results can provide a certain reference for the fault diagnosis research of rotating machinery.
作者
段礼祥
李涛
唐瑜
杨家林
刘伟
Duan Lixiang;Li Tao;Tang Yu;Yang Jialin;Liu Wei(College of Safety and Ocean Engineering,China University of Petroleum(Beijing);PetroChina Tarim Oilfield Company)
出处
《石油机械》
北大核心
2021年第2期60-67,80,共9页
China Petroleum Machinery
基金
国家重点研发计划项目“原油天然气储罐及附属管道、辅助设施检测检验技术研究”(2017YFC0805803)
国家自然科学基金资助项目“基于迁移学习的往复压缩机故障诊断机制及预测预警模型研究”(51674277)
中石油战略合作科技专项“海外长输油气管道灾害监测预警及动力设施诊断技术研究”(ZLZX2020-05-02)。
关键词
机械故障
数据级融合
故障诊断
卷积神经网络
多源异构信息
mechanical failure
data-level fusion
fault diagnosis
convolutional neural network
multi-source heterogeneous information
