Research On Prediction Method Of Milling Tool Remain Use Life Based On Transfer Learning

As a key device in manufacturing product processing,the health status of milling cutter will affect the machining accuracy and quality of workpiece to a certain extent.If it is still used after damage,it may damage the machine tool and machining workpiece,resulting in economic losses and even casualties.Therefore,it is particularly important to predict its remain use life and change the cutter in time.As a kind of machine learning method,deep learning is widely used in the field of tool mechanical life prediction because of its good applicability.In the actual processing and production,the changes of processing conditions,tool models,tool materials and other factors will lead to the failure of the original model on the premise of unchanged processing conditions under the new processing conditions,and it is time-consuming and laborious to mark the life information of new tool monitoring data artificially,so the idea of transfer learning is introduced to solve the above problems.Combined with the transfer learning method,this thesis puts forward the research topic of "prediction method of milling cutter remain use life based on transfer learning".Using the transfer prediction model to predict the remain use life of the milling cutter,predict the failure time of the milling cutter in advance and change the cutter in time can effectively reduce the consumption of human and material resources in machining operation and maintenance,which is of great significance to the actual production and processing.To sum up,the main work and achievements of this thesis are as follows:(1)In order to characterize the performance degradation trend of milling cutter and improve the prediction accuracy,a feature representation and prediction model based on deep full convolution neural network is proposed.Firstly,a feature extraction model composed of four convolution layers is constructed to extract the performance degradation trend features contained in the signal;then the life prediction is realized by using the prediction module;finally,the effectiveness of this method is verified by experiments.(2)In the processing scenario of cross working conditions on the same platform,the data distribution is different due to the different degradation trend between the two working conditions,and there will be differences in the distribution of data under the same working condition,which will lead to the decline of prediction accuracy.A transfer prediction model based on dynamic benchmark is proposed.Firstly,the dynamic benchmark which can represent the commonness of source domain is extracted by full convolution variational autoencoder;then,the maximum mean discrepancy method is used to reduce the difference between each source domain individual and target domain individual relative to the dynamic benchmark to achieve domain adaptation;then input the adapted features into the prediction module and output the prediction results;Finally,the effectiveness of this method is verified by experiments.(3)Aiming at the problem that the difference between data is further expanded and the prediction accuracy is difficult to improve in the cross platform and cross working condition processing scenario,a transfer prediction model based on parallel dual channel full convolution neural network is proposed.Firstly,the source domain feature extraction module and the target domain feature extraction module are used to extract the source domain features and target domain features respectively,so as to avoid the confusion of features;then the maximum mean discrepancy method is used to reduce the distribution difference of the characteristics of the two domains;then the two domain features are input into the source domain full connection prediction module and the target domain full connection prediction module respectively,and the average absolute error is used as the measurement criterion to realize the transfer of the prediction module,and then the prediction results are output;finally,the effectiveness of this method is verified by experiments.(4)By integrating the above research results,a transfer prediction system of milling cutter remain use life is designed and developed.After pre-training and testing of historical data,the system can predict the remain life of milling cutters online.By visualizing the prediction curve and demarcating the alarm threshold line,it can provide reference for reducing the risk of accidents by changing cutters in advance.