Fault Prediction Of Planetary Gear Based On Functional Data Fitting And Convolution Neural Network

With the increasing automation and intellectualization of mechanical equipment.People's demand for health assessment and management of mechanical equipment has become particularly urgent.The planetary gear is the most important part of rotating machinery.It usually works in a low-speed,heavy-duty environment.Once the failure occurs,it will cause chain reaction,causing huge economic losses and endangering personal safety.Therefore,it is of great significance to carry out predictive maintenance.In this paper,planetary gear is taken as the research object,and fault prediction is studied.The main research contents are as follows:Firstly,a method of functional transformation of original discrete vibration data is proposed.The function data fitting method in the idea of functional data analysis is adopted.Aiming at the problem of data loss in discrete data.Using the idea of functional data analysis to transform discrete data into functions.Replacing discrete data points in functional form.For the normal operation of planetary gear data,fitting with traditional fourier basis function.Fault data that do not apply to traditional basis functions.Establishment of basis function models for different fault locations.The feasibility of the basis function is verified by the experimental data.Function coefficients are extracted as a new sample data set after filtering error parameters.Secondly,establishment of prediction model of planetary gear convolution neural network.Establishing a three-layers deep planetary gear convolution neural network prediction model for massive planetary gear data.To solve the limitation of shallow prediction method only applicable to small batch sample data.To make up for the shortcomings of traditional shallow prediction methods which need artificial feature extraction and feature dimension reduction.Compared with the most representative BP neural network in the shallow prediction model.It is proved that the convolution neural network prediction model of planetary gear has better prediction effect when the sample data is massive.Thirdly,the influence of internal structure of convolution neural network on prediction results is studied.According to the internal structure of the convolution neural network,the dimensions of the input data are different,the size of the convolution kernel is different,the BN layer is present,and the depth of the model structure is different.Analysis of the impact of the above factors on the prediction results.Internal structure parameters of improved convolution neural network prediction model.At last a 7-layer deep convolution neural network prediction model with BN layer and one-dimensional input data form is determined,and the size of convolution core decreases with the increase of layers.Finally,set up planetary gear fault test bench.Designing the test scheme of simulated fault of planetary gear and collecting the data of original time domain signal.Fault prediction based on convolution neural network prediction model with 7 layers of structural depth.Compared with the convolution neural network model with 3 layers of structure depth,the prediction accuracy is improved.