孙英淳,唐斌,蔡先阳.基于倒谱–对称点图谱–卷积神经网络的内燃机增压器滚动轴承故障诊断[J].内燃机工程,2023,44(6):69-76.
基于倒谱–对称点图谱–卷积神经网络的内燃机增压器滚动轴承故障诊断
Fault Diagnosis of Rolling Bearings in the Supercharger of An Internal Combustion Engine Based on Cepstrum–Symmetrized Dot Pattern–Covolution Neural Network
DOI：10.13949/j.cnki.nrjgc.2023.06.009
关键词:滚动轴承  故障诊断  倒谱  对称点图谱  卷积神经网络
Key Words:rolling bearing  fault diagnosis  cepstrum  symmetrized dot pattern （SDP）  convolution neural network（CNN）
基金项目:中央高校基本科研项目（DUT18LAB15）
作者单位E-mail
孙英淳* 大连理工大学 内燃机研究所，大连 116024 syc19971110@mail.dlut.edu.cn 
唐斌* 大连理工大学 内燃机研究所，大连 116024 btang@dlut.edu.cn 
蔡先阳 大连理工大学 内燃机研究所，大连 116024 xianyangcai@163.com 
摘要点击次数: 415
全文下载次数: 256
摘要:针对内燃机增压器滚动轴承振动信号易受噪声影响、故障特征微弱的问题，提出了一种基于倒谱（cepstrum）–对称点图谱（symmetrized dot pattern， SDP）–卷积神经网络（convolution neural network， CNN）的智能故障诊断方法。通过倒谱对原始信号进行故障特征提取，获取能够反映滚动轴承故障类型的特征向量。然后应用对称点图谱方法将一维倒谱数据映射到极坐标空间，并进行灰度化处理得到SDP特征灰度图，将特征图导入到卷积神经网络进行特征挖掘和故障识别。最后通过滚动轴承外滚道、内滚道和滚动体出现损伤的故障试验，构建了9类故障状态原始信号，验证了基于倒谱–SDP–CNN的智能故障诊断方法。结果表明：倒谱–SDP–CNN方法具有运算简便、快捷、受噪声影响较小等优点，对试验测试集的故障识别准确率达到97.5%，可以较为准确地诊断增压器滚动轴承的故障状态和严重程度。
Abstract:An intelligent fault diagnosis method based on cepstrum–symmetrized dot pattern（SDP）–convolution neural network（CNN） was proposed to solve the problems that the rolling bearing vibration signal of the supercharger of an internal combustion engine is easily affected by noise and the fault features are weak. The fault features of the original signal were extracted by using the cepstrum method to obtain enough feature vectors that can reflect the type of rolling bearing faults. Then， the SDP method was applied to map the one-dimensional cepstrum data to polar coordinate space and grayscale them to generate SDP feature grayscale maps， and the feature map was imported into CNN for feature mining and fault identification. After the failure test of the damaged outer raceway， inner raceway and rolling elements of the rolling bearings， nine sets of fault status raw signals were utilized to verify the intelligent fault diagnosis method based on cepstrum–SDP–CNN. The results show that the cepstrum–SDP–CNN method is simple， fast and less affected by noise， and can effectively diagnose supercharger rolling bearing faults. The diagnosis recognition accuracy rate for the test sets was 97.5%， and the proposed method can accurately determine the fault status and severity of rolling bearings.
查看全文  HTML   查看/发表评论