Fault Diagnosis Method Of Rolling Bearing And Planetary Gear Box Based On Convolution Neural Network

The rolling bearing and planetary gearbox are the key parts of the mechanical equipment,and their faults are closely related to the normal operation of the equipment.With the developing of high efficiency and intelligence of production,the requirement for intelligence of equipment fault diagnosis is improving,especially the key parts such as rolling bearing and planetary gear box.However,the existing fault diagnosis methods of rolling bearing and planetary gear box have the following deficiencies.Firstly,the most diagnosis methods are based on knowledge related to signal processing.This needs the plentiful mathematical knowledge and the profound professional knowledge as the basic,so the operability is not strong in the actual production.Secondly,the current research on fault diagnosis is only limited to the diagnosis of the fault location,and there is little diagnosis of the fault degree,bearing load and other attributes.At first,for the deficiencies of the above methods for fault diagnosis of rolling bearings and planetary gearboxes,this paper builds a convolutional neural network model based on deep learning knowledge,and performs qualitative diagnosis of rolling bearing faults on the CWRU dataset,and achieves a good accuracy of 99.79%.This method directly uses one-dimensional vibration signal as the input of the network,and uses the powerful learning ability of the convolutional neural network to adaptively extract and learn the characteristics of the vibration signal,which overcomes the difficulty of extracting the characteristic information of the vibration signal by the traditional method.Then,based on qualitative diagnosis of rolling bearings,a quantitative fault diagnosis method for rolling bearing based on multi-attribute convolutional neural network model is proposed.By constructing a multi-attribute convolutional neural network model,the corresponding fault location,fault degree and bearing load of the rolling bearing are diagnosed on the CWRU dataset and the laboratory dataset.And the comprehensive accuracy of each attribute is 89.74% and 96.3% respectively.This method can view not only the accuracy of each attribute,but also the diagnostic results of any combination of the various attributes.Finally,on the basis of the qualitative and quantitative diagnosis of rolling bearing,the fault tree structure,the working conditions parallel structure and theMSCNN model are proposed for the case of the planetary gearbox structure and the complex fault signal propagation path resulting in difficult to diagnose.The fault tree structure can handle all kinds of complex fault types in a unified manner,and can also check the diagnostic effect of each node.The parallel structures can handle variable conditions and predict speed and load of rolling bearing.Based on the multi-attribute convolutional neural network model,the MSCNN model uses big convolution kernel,large area pooled,large steps,and adds dual paths in the network.This not only prevents the network from being too deep to train but also takes care of the problem of insufficient extraction of the network’s shallow features.At the end,in the laboratory planetary gearbox vibration dataset,the complex fault diagnosis of the planetary gearbox under variable conditions was achieved,and the diagnostic accuracy of the planetary gearbox components was 97%.The fault diagnosis method of the rolling bearing and planetary gearbox based on the convolution neural network is simple,the generalization performance is well,the operability is strong and the diagnosis effect is great.