Research On Milling Cutter Wear Monitoring Model Based On Feature Fusion

In the process of NC machine tool machining,tool wear will have a direct impact on the machining accuracy and surface roughness of the workpiece,but also indirectly affect the operating efficiency of the machine tool.Therefore,efficient and accurate monitoring of tool wear state and timely warning before tool failure are of great significance to improve workpiece machining accuracy,machine tool operation efficiency and reduce production cost.In this paper,milling cutter machining as the research object,combined with sensor signal,milling cutter wear monitoring research.This paper is summarized as follows:Firstly,the research significance of milling cutter wear monitoring is expounded.Starting from the tool wear monitoring method,wear state recognition and wear quantity prediction,the research status at home and abroad is summarized and analyzed,and the shortcomings of the current related research are put forward.The indirect monitoring method which takes cutting force,vibration and acoustic emission signals as monitoring signals is determined,and the overall framework of this study is designed.Then,the form and process of milling cutter wear are briefly described,the experimental conditions and collection information of the data set used in this paper are introduced in detail,and the original data are preprocessed to divide the different states of milling cutter wear,so as to prepare the data for the following chapters.Secondly,the time domain analysis,frequency domain analysis and wavelet packet decomposition were used to extract the time domain,frequency domain and Baud signs related to tool wear,and a total of 147 tool wear features were extracted.In order to further obtain the optimal feature vector with less invalid values and low redundancy,PCA was used to reduce the feature dimension of tool wear features,and 43 optimal features with cumulative contribution rate greater than 95% were retained.In order to avoid the possible problems of poor classification model learning and poor classification accuracy caused by the unbalanced distribution of sample categories,use the Borderline-SMOTE technique to balance the sample data of the best characteristics acquired.Then,in order to optimize the parameters of SVM,WOA was introduced,and a WOA-SVM based tool wear state recognition model was proposed.In order to verify the influence of the quantity balance of a few class samples on the performance of the classification model,the optimal feature vector and the balanced sample data were taken as input respectively to verify the accuracy of the model.The experimental results show that WOA-SVM model has better performance in the data after sample balancing processing,and the classification accuracy is higher than that of the data without sample balancing processing.Moreover,the WOA-SVM model was also compared with other classic models,from the perspective of classification accuracy of the model and parameter optimization time,fully proved the superiority of BSMOTE-WOA-SVM model.Finally,build a composite neural network based Conv LSTM-Att amount of tool wear prediction model.With preprocessed data as input,1D-CNN was used to extract depth features from original data and filter invalid features.LSTM was used to learn the temporal relationship of filtered tool wear features,and the feature vector with higher sensitivity to tool wear was obtained.Finally,Attention mechanism was introduced to learn the contribution weights of different feature vectors to tool wear prediction,strengthen the grasping ability of the model to key information,and thus improve the prediction accuracy of the model.Through experiments,compared with other models,the superiority of the Conv LSTM-Att tool wear prediction model is verified from the perspective of the error of each model.