Research On Milling Stability And On-Line Chatter Monitoring Of Thin-Walled Parts

Thin wall parts have many thin wall features and are light in weight.They have a wide range of application scenarios.The process from blank to thin wall parts requires the removal of a large amount of materials.The relative stiffness of the formed parts is low,and they are prone to chatter during finishing,which can not reach the design precision of the parts.This problem has also become one of the important issues for researchers and industrial personnel to study.In this paper,the modeling and prediction of stability and the on-line monitoring of chatter are studied.Firstly,considering the structural characteristics of thin-walled parts,according to the principle of modal superposition,the multi-modal dynamic model of thin-walled parts is established.Based on the full discrete method,the established dynamic model is solved.The collected acceleration signal is transformed by FFT,and the accuracy of the obtained lobe diagram is analyzed by combining the natural frequency obtained by modal experiment.The stability boundary obtained by this method can select the cutting parameters to achieve chatter-free cutting of thin-walled parts and ensure the machining accuracy.Secondly,in the aspect of chatter feature extraction,this paper proposes an optimal selection method of decomposition parameters and an energy entropy feature extraction method based on the energy agglomeration characteristics when chatter occurs.According to the given maximum envelope kurtosis,the data parameters of the variational mode are optimized and selected,and then the energy entropy is calculated according to the decomposed data.Finally,the experimental results show that the method can effectively distinguish the chatter frequency band after processing the signal to be processed,which is more conducive to the extraction of energy entropy chatter characteristics.The change of energy entropy is a description of the change of signal energy distribution,which is beneficial to the detection of chatter.Aiming at the problem of chatter state recognition,this paper proposes an efficient on-line chatter monitoring model based on supervised learning and unsupervised learning.Due to the continuity of the machining process,the online monitoring model also needs to input new data.Therefore,based on incremental learning,incremental K-means and incremental SVM strategies are proposed to reduce the model deviation caused by sample label errors during incremental learning,so that the online monitoring model is closely integrated with the current cutting process data in real time.The energy entropy feature is used to verify the clustering efficiency,multisensor signal fusion and IL-KM-SVM model.The results show that the clustering efficiency of IL-KM is 20.16 % higher than that of K-means,and the recognition accuracy of IL-KM-SVM model is 94.7 %,which is 2 % higher than that of IL-SVM.The average recognition time of each sample is 0.16 ms,laying a solid foundation for subsequent online monitoring.Finally,combined with the previous stability prediction model,chatter feature extraction and recognition model,the monitoring platform is developed by Lab VIEW platform.The monitoring system can realize online data acquisition,storage,data analysis,stability prediction and online monitoring of chatter state.The results show that the error of monitoring results meets the requirements and has application value.