Research On Wear Status Monitoring Of Milling Cutter Based On Deep Learning

The impeller is a key component of many large-scale equipment such as hydropower and nuclear pumps.Impeller machining quality directly affects the operating state and service life of the equipment.The wear of the milling cutter during the machining process directly affects the machining accuracy of the impeller.Therefore,it is of great significance to study the monitoring method of the wear state of the milling cutter during the machining process.This paper proposes to use the deep learning method to monitor the wear state of the milling cutter.According to the characteristics of time series signals,this paper proposes to improve the deep learning network by adding compressed sensing technology and Gram transform.The main research contents of the thesis are as follows:(1)This paper fully analyzes the research status of domestic and foreign research and the advantages and disadvantages of existing monitoring methods,then analyzes the tool wear mechanism,and finally determines the tool wear monitoring signal and monitoring method.Through theoretical analysis,it is determined that the current signal can reflect the wear state of the tool,and the feasibility of signal acquisition is ensured by constructing a signal acquisition platform.Three different tests were designed,and the tool stage wear state data,tool life data and field experiment data were collected,and three data sets were formed,which laid the foundation for subsequent research.(2)This paper uses the compressed sensing and stack sparse self-encoder to extract the wear state information of the milling cutter in the current RMS signal.First,the frequency domain data of the current signal is compressed using compressed sensing.In order to improve the robustness of the network,Gaussian white noise is added to the observed signal.The compressed data is then input into the stack sparse self-encoding network.The last layer of the network is connected to the softmax classifier,which uses the semi-supervised learning method to train the network and extract the feature information that characterizes the degree of tool wear.(3)In view of the shortcomings of the softmax classifier on the classification of feature data,this paper proposes to use the Gram transform to map the feature data to the two-dimensional space,and to use the residual neural network with deeper layers to further extract the feature information of the tool wear state..Network parameters are trained using supervised learning methods.The last layer of the diagnostic model is connected to the radial basis function to predict the tool wear and the reliability of the tool is evaluated based on the tool wear warning value.(4)Based on LabVIEW,the milling cutter wear condition monitoring platform is built to realize the online evaluation of the tool wear state.The monitoring platform realizes on-line monitoring of tool wear status by calling the trained deep learning network model.Simultaneous monitoring platform with current RMS signal waveform display,storage and simple offline analysis.The system has the advantages of simple operation and easy operation.