Research On Fault Diagnosis Algorithms For Rolling Bearing Based On Convolutional Neural Network

With the rapid development of science and technology and the proposal of Industry 4.0,mechanical equipment is moving towards more large-scale and more precise directions.At the same time,there are higher requirements for safety,stability and intelligence in the process of running mechanical equipment.Rolling bearings as one of the important components for the process of connection and power transmission for mechanical equipment,the health of rolling bearings depends on the normal operation of the overall equipment.Therefore,it is important to carry out fast and accurate online fault diagnosis of rolling bearings.However,traditional fault diagnosis methods rely too much on expert experience and deep learning-based fault diagnosis methods have disadvantages such as large model size and long model training time.The analysis and processing for the one-dimensional time sequence of rolling bearing is used to be the main technical route in this thesis,and combined with the convolutional neural network to achieve fault diagnosis research for rolling bearings.The main research contents of this thesis include the following four aspects.Ⅰ.Research on rolling bearing fault diagnosis algorithm based on GASF and Resnet50Aiming at the weak point of the traditional fault diagnosis method combining one-dimensional signal processing method and machine learning,cannot extract the deep features of the signal,which depends on expert experience.A rolling bearing fault diagnosis model based on Gramian Angular Summation Fields(GASF)and Resnet50 is proposed.Firstly,the signal is converted into GASF time series to realize the high-dimensional expression for the signal;secondly,combines with the deep residual network Resnet50 to achieve the extraction and recognition for the signal High-dimensional characteristics.Taking the Case Western Reserve University(CWRU)bearings data as the verification data,comparing with the different combination for Gramian Angular Difference Fileds,Markov Transition Fields,and Recurrence Plot conversion model with the Resnet50 network.The experimental results show that the combination of GASF and Resnet50 network proposed in this thesis has the best diagnostic effect and the diagnostic accuracy is 99.62%.Ⅱ.Research on rolling bearing fault diagnosis algorithm based on BI-TCP convolutional neural networkAiming at the shortcomings for the diagnosis of rolling bearings based on deep learning,which method need so long time to training model and too large space to save the model.A rolling bearing fault diagnosis method is proposed based on the bimodal input and two channel parallel(BI-TCP)Convolutional Neural Network.Firstly,the vibration signal is converted into the corresponding GASF matrix and Euclidean distance matrix,and secondly,the matrix is constructed in a cross-recombination manner as a dual-modal matrix as the input to the network.The network structure is designed as a BI-TCP convolutional neural network,and the independent extraction of different modal features in the input is achieved by designing the convolutional kernel parameters of the first layer network for different channels.Finally,feature fusion is performed in the last layer of the network to achieve fault diagnosis of rolling bearings.The proposed method is compared with CWRU bearing data as the validation object,and the input signal structure of the proposed model and the network structure of the proposed model are changed respectively.The experimental results show that the diagnostic accuracy of the proposed BI-TCP convolutional neural network is 99.75%,and the number of parameters of the proposed model is significantly reduced compared with that of the residual network Resnet50,the model size is only 2.6MB.Ⅲ.Rolling bearing fault simulation platform construction and algorithm example verificationCompleted the design of the rolling bearing fault simulation platform structure and hardware selection,build the bearing fault simulation platform.By replacing the bearings at the load end of the simulation platform,the simulation of different types of bearing faults is realised,and the acquisition of different types of fault signals is completed by combining with Tracer DAQ data acquisition software.The data collected is used as the validation object to verify the two fault diagnosis algorithms proposed in the thesis.The validation results show that both algorithms can effectively extract and identify the characteristics of different types of fault signals and achieve fault diagnosis of bearings.Ⅳ.Online fault diagnosis system design and implementation for rolling bearings based on PyQT5Based on the proposed rolling bearing fault diagnosis algorithms and the rolling bearing fault simulation platform built in this thesis,the online fault diagnosis of rolling bearings is designed through the interface program development tool PyQT5,to achieve real-time acquisition of vibration signals and related time domain analysis of the fault simulation platform,the data with abnormal time domain indicators are then identified through the proposed algorithm to complete the final fault type.In summary,a rolling bearing fault diagnosis algorithm based on GASF and Resnet50 and a rolling bearing fault diagnosis algorithm based on BI-TCP convolutional neural network are proposed.Good experimental results were achieved by the proposed algorithm on the CWRU bearing dataset.Secondly,the case verification of the algorithm proposed in this thesis is completed on the rolling bearing fault simulation platform built in the laboratory.Finally,the design and implementation of the online fault diagnosis system is completed by combining the proposed algorithm with the rolling bearing fault simulation experimental platform built by PyQT5.