Fault Diagnosis Method Of Rotating Machinery Based On Improved Deep Network

Rotating machinery is a key component of mechanical equipment and plays an important role in people’s production of life.In order to avoid the loss of life and property caused by mechanical equipment failure,it is very important to carry out fault diagnosis,condition monitoring and health management of rotating machinery.In recent years,using deep learning methods to solve image recognition problems has gradually become a research hotspot.Inspired by this,the paper constructs a fault diagnosis model by classifying the time-frequency diagram of vibration signals with deep learning algorithms.The paper studies the theoretical basis and practical application of two deep learning algorithms,Convolutional Neural Network(CNN)and Principal Component Analysis Network(PCAnet).Focus on improving PCAnet’s derivative algotithms:Kernal Principal Component Analysis Network(KPCAnet)and Multilinear Principal Analysis Component Network(MPCAnet).Two improved algorithms are proposed,namely Hybrid Kernal Principal Component Analysis Network(HKPCAnet)and Mutilinear Kernal Principal Component Analysis Network(MKPCAnet),they have been successfully applied in the field of rotating machinery fault diagnosis.The content of the paper is carried out from the following aspects:(1)PCAnet was introduced into the field of rotating machinery fault diagnosis,and multiple experimental data sets were used to compare and verify the diagnosis effect,and the following conclusions were drawn: Compared with CNN,PCAnet is more suitable for dealing with small sample faults that samples are difficult to obtain in engineering practice.In the case of fewer layers and smaller input image size,PCAnet has higher classification accuracy and shows better robustness and generalization ability,and the training time is also greatly shortened compared to CNN.This is because PCAnet is simplified compar ed to the CNN network,does not require back-propagation,and has greatly reduced training parameters.(2)KPCAnet was introduced into the field of rotating machinery fault diagnosis.In view of the defect that KPCAnet’s single kernel function cannot take into account the learning ability and generalization ability at the same time,an improved model,HKPCAnet was proposed.Through the verification of multiple sets of experimental data sets,the following conclusions are drawn:when different small sample data sets and different numbers of samples are used as training sets,the hybrid kernal principal component analysis network shows the best classification performance,and the diagnosis results are more stable.It has good generalization performance and robustness.This is because the improved model combines the advantages of the global kernel function and the local kernel function,and has strong generalization ability and learning ability at the same time,which effectively improves the classification perfo rmance of the single-kernel PCA network.(3)The tensor extension of PCAnet,MPCAnet,is applied to the field of fault diagnosis.MPCAnet has the problem of poor nonlinear fitting ability and poor feature clustering ability.So on the basis of MPCAnet,kernal transform is introduced.An improved multi-linear principal component analysis network is proposed,which increase the degree of difference between the training sample,further enhance the generalization ability and classification accuracy when dealing with non-linear data.It is proved that this method has high robustness in different fault diagnosis data sets of rolling bearing and can accurately identify various faults of rolling bearing.