Research On Milling Cutter Wear State Recognition And Wear Prediction Method Based On Deep Learning

As an important part of the manufacturing industry,milling field has very high machining efficiency and accuracy,and the tool is an important executive component in the processing field.The wear of the tool has a great impact on the machining accuracy and efficiency of the workpiece.The working efficiency and machining quality of machine tools are closely related to the tool state.If the tool state can not be mastered in time,it is easy to cause impact or waste of resources.Therefore,accurate identification of tool wear state and accurate prediction of tool wear amount are the key problems to realize tool reliability machining.In this paper,the milling cutter is taken as the research object,and the learning model based on deep learning is constructed for the tasks of tool wear state recognition and tool wear prediction.Through the analysis of data characteristics,data features are extracted,so as to improve the accuracy and accuracy of tool wear state recognition and wear prediction.The main works of this paper are summarized as follows:1.In view of the large amount of real data and data redundancy,this paper preprocesses the degraded data,including outliers,noise reduction and down sampling;The data set of state recognition and wear prediction is constructed for the processed data.For tool state recognition,this paper uses ensemble empirical mode decomposition(EEMD)to construct the time-frequency diagram of the data,and introduces the idea of fast spectral kurtosis diagram to select the optimal modal component for the situation of many modal components and difficult selection;For tool wear prediction,the preprocessed data are extracted in time domain,frequency domain and time-frequency domain.The features are screened and fused by Spearman and KPCA methods.Finally,the low latitude and strong correlation features are obtained,which can improve the reliability of the subsequent tool wear prediction model.2.In view of the problems of large amount of data and complex interference signal in industrial data acquisition,which lead to the complexity of tool wear state recognition and low recognition accuracy,this paper constructs the eemdfkacnn network model.Through ensemble empirical mode decomposition(EMD),the time-frequency characteristics of fault signals are decomposed from the collected vibration signals under different conditions,and the optimal modal components are selected by using fast spectral kurtosis diagram to generate timefrequency diagram by HHT transformation;The time-frequency diagram is input into the network model constructed in this paper for learning,and the tool wear state is identified,and compared with the traditional identification method.Experiments show that the proposed method has good performance.3.In order to improve the accuracy of tool wear prediction,the traditional machine learning model has low prediction accuracy.In this paper,the tool wear is predicted based on the Bi LSTM model,and the data features are re optimized by using the signal segmentation method.The constructed feature data is input into the Bi LSTM model designed in this paper to predict the tool wear.Through the comparison of various models and algorithms,the Bi LSTM model designed in this paper has a certain application prospect in the prediction problem.