Research On Rolling Bearing Fault Diagnosis Method Based On Vibration Characteristics And Convolutional Neural Network

With the continuous improvement of manufacturing level,rotary machinery is gradually developing towards precision,systematization and intelligence.Rollin g bearings are often used in the field of rotating machinery.The complexity of working environment is easy to cause damage to rolling bearings.The health of the bearing directly affects the overall operation of the mechanical system.The rolling bearing produces regular vibration and shock when the fault occurs,which makes the research of rolling bearing fault diagnosis technology based on vibration characteristics has important practical significance.Taking rolling bearings as the research object,this paper studies feature extraction and fault classification technology,and proposes a rolling bearing fa ult diagnosis method based on vibration characteristics and convolutional neural network.The main research contents are as follows:(1)Aiming at the difficulty of feature extraction caused by noise interference of rolling bearings,a rolling bearing fault diagnosis method based on SSA optimization and improved threshold was proposed.Firstly,an improved threshold func tion is proposed and Sparrow Search Algorithm(SSA)is used to optimize its parameters.Then,the optimized threshold function is used to reduce the noise of the rolling bearing vibration signal.Then,the Empi rical Mode Decomposition(EMD)is adopted to decompose the noise reduction signal and put forward a comprehensive evaluation index P,which is taken as the standard to select the decomposition component,and then the feature extraction is realized.Finally,the envelope spectrum of the reconstructed signal is analyzed.Simulation signals and actual bearing signals were used to verify the proposed method.The experimental results show that the proposed method has excellent noise redu ction and feature extract ion capabilities,and can realize more accurat e fault diagnosis.(2)Aiming at the problem that the original signal is indirectly input into the convolutional neural network with high artificial dependence and easy to ignore the time dependence,which leads to the loss of fault information,a rolling bearing fault diagnosis method based on MTF-SPCNN with small samples was proposed.Firstly,one-dimensional time domain signals are transformed into two-dimensional image input networks by Markov Transition Field(MTF)encoding method,and time dependence of signals is preserved.T hen,the Stripe Pooling Module(SPM)is proposed and introduced into the network t o construct the Stripe Pooling Convolutional Neural Networks(SPCNN).Improve the ability of the model to extract remote spatial features and capture long-distance directional information;At the same time,the channel attention-mechanism(SE)is added in front of the maximum pooling layer.The information weights are located there.Finally,the data set of Case Western Reserve University(CWRU)and the MFS data set of our laboratory were used for verification.The experimental results show that the MTF-SPCNN model can still maintain a high accuracy of fault diagnosis under the condition of small samples and varying working conditions.Has good stability and generalization performance.(3)Aiming at the problem that the convolutional neural network of indirect input of original signal is easy to ignore the time dependence,which leads to the loss of some important features,a rolling bearing fault diagnosis method based on MTF and SSCAM-MSCNN under small samples was proposed.Firstly,the Stripe Self-Calibrating Attention Mechanism(SSCAM)was proposed,and the self-calibrating function was added on the basis of the stripe pooling module.On the one hand,the sensitivity of the model in the long distance direction was improved,and on the other hand,the depend ence relationship between channels was established.Extracting global featur es;Then,the SSCAM module is introduced into the Multi-Scale Convolutional Neural Network(MSCNN)to establish the SSCAM-MSCNN model.Finally,the MTF two-dimensional image is used as network input to obtain fault diagnosis results.Two data sources were used for verification.The experimental results show that the SSCAM-MSCNN model has higher fault diagnosis accuracy and generalization performanc e for small samples and variable working condition data.At the same time,Gaussian white noise is added to the data,and the experimental results show that th e proposed method has stronger anti-interference performance.