Prediction Of Remaining Using Life Of Rolling Bearing Based On EEMD And MCNN-SE-GRU

In this paper,rolling bearings which are common in rotating machinery with high failure rate are taken as the research object,and the Remaining Useful Life(RUL)prediction problem is studied by using data-driven method.In order to solve the problems that extraction of the degradation information of the monitored vibration signals is incomplete and the prediction accuracy of RUL needs to be improved in previous studies.In order to solve the problems that extraction of the degradation information of the monitored vibration signals is incomplete and the prediction accuracy of RUL needs to be improved in previous studies.In this paper,a method combining Ensemble Empirical Mode Decomposition(EEMD)feature extraction and MCNN-SE-GRU deep learning model with multi-scale feature mining,attention mechanism and time sequence correlation was proposed to predict the rolling bearing RUL.And its effectiveness is verified by open source experimental data.The main research contents of this paper are as follows:(1)EEMD decomposition and feature selection based on signal energy ratio.The time-frequency domain signal processing method EEMD was used to decompose the vibration signals of bearings according to the samples in the experiment and obtained the Intrinsic Mode Function(IMF)containing different frequency bands.Since EEMD is an adaptive decomposition algorithm,that is,the number of IMF components is not fixed,in order to control the number of feature inputs and ensure that sufficient information is retained in IMF components of different scales decomposed by EEMD,the method of IMF signal energy ratio is adopted to select the degenerated features.(2)Construction of MCNN-SE-GRU model and RUL prediction.Firstly,in order to enhance the feature extraction ability on bearing degradation signals and the generalization ability of CNN,a Multiscale Convolutional Neural Network(MCNN)was used to expand the CNN perceptual domain.MCNN uses convolutional neural networks with different sizes of convolutional kernels to form different branches for depth feature mining,and splices the obtained multi-scale feature images into a feature image to pass down.Secondly,in order to differentiate the importance of the features obtained from the different branches,the attention mechanism between the branch feature graphics is implemented using the Squeest-and-Excitation(SE)attention module.It is necessary to enhance the high value features and suppress the low value features.Thirdly,the Gated control Recurrent Unit(GRU)module was used to explore the temporal relationship and output it through the Fully Connected Network(FCN).Finally,the moving average method is used to smooth the output results to remove excessive local fluctuations and realize the RUL prediction of rolling bearings.In this paper,the validity and accuracy of the proposed model are verified by the experimental data of accelerated life of rolling bearings.(3)Development of rolling bearing RUL prediction software based on Python and Py QT5.A rolling bearing intelligent life prediction software system was developed using Py QT5 framework and Python programming language.The feature extraction method and RUL prediction algorithm studied in this paper were integrated into the system in a modular way.The system was mainly divided into data processing module,signal analysis module and RUL prediction module.Used for vibration data signal analysis and bearing RUL prediction research.