Research On Monitoring Of Milling Chatter And Tool Wear Condition For Thin-walled Workpiece

Thin-walled parts are light weight and have good mechanical properties,which can be applied in various industrial fields.But its weak rigidity and mode time-varying characteristics in milling process lead to chatter easily in the cutting process,thus reducing the surface quality and causing severe tool wear.Meanwhile,the thin-walled parts are mostly titanium alloy,high temperature alloy and other alloy materials.The hard to cut characteristics of these materials accelerate the tool wear during the process,and then affect the machining accuracy.Therefore,it is very important to study the monitoring methods of the milling chatter and tool wear state of thin-walled parts,which is of great significance to ensure the milling stability and improve the milling efficiency of thin-walled parts.This paper focuses on the research on the chatter and tool wear monitoring technology of thin-walled parts milling,and proposes a new method to monitor chatter and tool wear by using a variety of sensors,which improves the robustness of the system,so as to reduce the misjudgment of machining state.On this basis,the multi sensor signal characteristics are studied and analyzed,and the high-efficiency chatter and tool wear monitoring models are established respectively,which provides the research foundation for the stable processing of thin-walled parts and the development of industrial intelligence.Firstly,aiming at the chatter identification of thin-walled parts milling,based on the processing characteristics of thin-walled parts,the sound,acceleration and milling force are analyzed and determined as monitoring signals of chatter and tool wear state.The experimental platform is built according to the required signals,including processing flat,sensor,data acquisition device and other equipment.Then,based on the time-domain characteristics of sound signal,frequency domain characteristics of milling force signal and cutting process parameters,a comprehensive and accurate method is proposed to judge the machining state of a certain time period.The surface morphology of the workpiece after machining is verified,which proves the effectiveness of the method,which lays a foundation for the efficient establishment of sample label.Aiming at the wear problem in the cutting process,the paper systematically expounds the tool wear mechanism,and gives the classification and corresponding standards of the tool wear state according to the wear problem of milling tool studied in this paper.Secondly,the paper decomposes the multi sensing signals of sound,acceleration and milling force obtained by the milling chatter experiment of thin-walled parts,establishes the correlation between energy entropy and chatter state based on IMF,and identifies the chatter state of the feature samples based on the MPGA xgboost model.The chatter state of the feature samples is compared with the traditional machine learning model,The advantages of the MPGA xgboost model are verified.The results show that VMD decomposition can effectively separate the chatter frequency band in the signal,and the chatter state can be represented by the energy entropy of IMF.MPGA can optimize the parameters of xbboost effectively.The accuracy of xgboot model is 93%after optimization,and 8%higher than that of the UN optimized one.Compared with different algorithm models,xbboost model has a high accuracy in chatter state recognition,and its prediction accuracy is 5%and 12%higher than SVM and BPNN respectively.Finally,the high dimension feature extraction of the sound,acceleration and milling force signal data in the thin-walled parts stable milling process is carried out,including 117 features in time domain,frequency domain and wavelet packet node energy.Through SVM RFE,the low dimension features are selected and 10 optimal features are selected,Then,the PNN model is used to train and forecast the optimal feature samples,and the prediction results of SVM and BPNN model are compared.The results show that the correlation between the three signals and the wear state of the tool is:milling force>acceleration>sound.The prediction accuracy of PNN model for feature samples after feature selection is 17%higher than that of original feature samples,and the judgment of initial wear and medium wear state is more accurate.Through the comparison of different algorithm models,PNN model has a high accuracy in tool wear state recognition,and its prediction accuracy is 9%and 19%higher than SVM and BPNN respectively.