Research On Wear Data Analysis And Reliability Modeling Of Milling Tool

Tool is a key part in the process of machine tool cutting process,the surface machining quality of parts is closely related to the state of the tool.It will seriously affect the quality of cutting manufacture processing when the tools seriously wear down.It is of great significance to analyze tool degradation failure data,monitor tool working state in cutting process to make full use of tool performance,reduce machining cost of CNC machine tool and improve machining quality of workpiece.In this paper,tool wear data is collected offline,and reliability modeling and analysis of tool wear data are firstly carried out,then tool wear state classification model is proposed by using online sensor data combined with feature extraction method.Finally,tool wear prediction model is further proposed based on the classification model.Generally,based on the tool wear data and sensor data generated during machining,a variety of tool wear state monitoring methods are proposed.The main research contents are summarized as follows:Firstly,the competitive failure model and the tool reliability model under multiple working conditions are established.Because of two failure forms in the process of tool failure,Weibull distribution is used to construct the cutting tool of sudden failure model,the Gamma process is used to construct tool wear failure model,and the tool failure of competition model is built.In addition,the reliability of the cutting tool under the condition of the working condition of single model is constructed.Considering the diversity of the working conditions of cutting tool,the Taylor tool life equation is blended into the shape parameter of Gamma process,and the objective function is obtained by maximum likelihood estimation method,the model parameters are obtained by genetic algorithm,then the condition of multi-conditions of cutting tool reliability model is obtained.Second,tool wear state classification model based on multi-domain features and residual network is constructed.The tool wear mechanism is analyzed,and the trend of sensor signal in the tool wear monitoring experiment is drawn.According to the wear mechanism,the wear value of the tool state monitoring experiment is divided corresponding with wear stages.From the time domain,frequency domain and time-frequency domain respectively,some features are extracted including mean,root mean square value,variance,peak-to-peak value,Kurtosis coefficient,kurtosis coefficient,spectral skewness coefficient,spectral kurtosis coefficient and wavelet energy.The nine features are reconstructed into grayscale images.Tool wear state recognition networks based on convolutional neural network and residual network are proposed.The recognition accuracy of convolutional network is 96.8%,and that of residual network is97.8%.The results show that the residual network is effective in tool state classification.Thirdly,the prediction model of tool wear value based on adaptive feature extraction method and long and short-term memory network is constructed.Firstly,the frequency domain data of the sensor data in the process of cutting tools is obtained by fast Fourier transform,and the data of the low frequency part is intercepted to reduce the input data volume of the model.According to the unsupervised training features of sparse autoencoder,the stack sparse autoencoder is proposed to adaptively extract the features of sensor signals in the frequency domain,and a prediction model based on long and short-term memory network is constructed.Then,the prediction model is combined with the attention mechanism.The model results are evaluated by means of mean square error,mean absolute error,Pearson correlation coefficient and mean absolute percentage error,and the feasibility and practicability of the proposed tool wear prediction model are demonstrated.Based on tool wear and sensor data of the machining process,three tool wear condition monitoring methods have been proposed.It contains the reliability model based on statistical method,the state classification model based on data driven and the wear prediction model based on deep learning.It provides a theoretical basis for tool changing decision and machining error compensation.And it is of great significance to improve machining quality and reduce production cost.