Research On Wear Assessment Method Of Milling Tool Based On Multi-source Information Fusion

In the process of machining,the wear and failure of cutting tools is the main cause of CNC machine failure.Accurately identifying the wear conditions of tools and replacing failed tools in time can not only help increase the trouble-free working time of CNC machine tools,but also extend the tool life.Affected by the processing environment and processing parameters,a single monitoring signal often cannot effectively reflect all the information about tool wear,which will lead to insufficient generalization ability of the tool wear assessment model based on a single signal.It is of great significance to study the tool wear assessment model which integrates multitype monitoring signals for the efficient management of tool life of CNC machine tools.In this paper,the tool wear formed during the milling process is taken as the research object.By preprocessing and analyzing the multi-source monitoring signals of tool wear,this research first proposes a multi-source information fusion method based on hybrid neural network.Then,this paper studies the modeling and optimization method of tool wear assessment.The main research work is as follows:(1)Tool wear mechanism and preprocessing of multi-source signals.Based on the study of the causes and forms of tool wear in the process of machining,the characteristics of various signals suitable for tool wear monitoring are analyzed.Considering the influence of different wear conditions on the monitoring signals,a multi-source monitoring signal composed of vibration,current and acoustic emission is selected.The data generated from the wear experiment of milling tool under multiple working conditions are sorted out,and the problems of data abnormality and data missing in the data set are processed and analyzed.(2)Modeling of milling tool wear assessment based on hybrid neural network model.Based on the experimental data of tool wear monitored by multiple sensors,a hybrid model of tool wear assessment is constructed based on the combination of convolutional neural network(CNN)and long short term memory(LSTM).The hybrid model is composed of CNN and LSTM in series,where CNN is used as the feature extraction layer of the model.The feature extraction of multi-source signals is achieved through a combination of one-dimensional convolution and pooling operations.Then,the extracted abstract features are reconstructed according to the temporal relationship.Finally,the reconstructed features are input into the LSTM to complete the classification and assessment of the current wear conditions of the tool.And this research verifies the effectiveness of this hybrid model by comparing with the traditional multi-source information fusion method based on manual feature extraction.(3)Research on optimization method of assessment model based on data augmentation.Since the data set constructed by the wear experiment of milling tool is unbalanced,this will lead to poor generalization of the assessment model.To solve this problem,the characteristics of various data augmentation technologies are analyzed and compared,and a multivariable time series data generation algorithm based on Dynamic Time Warping(DTW)is selected.On this basis,an optimized weighted softDTW Barycenter Averaging(Soft-DBA)method is proposed to enhance the data set.In this method,soft-DTW is used as the similarity measure between samples,and the new samples are synthesized by selecting multiple samples from the same class and assigning weights according to the relative distance.And this research uses the enhanced data set is used to train the model and compare the optimization effect of the model to verify the effectiveness of the method.