Detection Of Cutting Chatter Based On LGBM And Frequency Distribution Features

Chatter is a kind of self-excited vibration caused by the instability of the cutting system,which will cause changes in the frequency and energy distribution of the cutting signal,resulting in poor processing quality,aggravated tool wear,and reduced machine tool accuracy.Based on the LGBM algorithm that is suitable for massive data processing and the frequency distribution features sensitive to chatter,this paper proposes a chatter recognition technology for turning and milling processing that can be applied to variable cutting parameters at the same time.This paper verifies the accuracy of the chatter recognition technology of turning and milling processing under the cutting parameters in the domain and outside the domain(processing parameters not appearing in the training set,validation set and test set)through experiments,which significantly improves the applicability of the method.This paper first applies the zero-order frequency domain method to solve the stability equation to obtain the relationship between spindle speed,depth of cut and cutting stability.Based on this,the turning experiment of stainless steel long hollow shaft and aluminum block milling experiment are designed.Studying the changes in the spectrum data distribution of the cutting signal during the machining process,the changes in cutting sound and the quality of the processed surface,combined with the tool or workpiece modal frequency obtained from the hammering modal experiment,to determine the chatter identification standard and the relationship between chatter and spectrum data distribution.Based on this relationship,four features that are sensitive to chatter are extracted,and RFE is used for feature selection,and accurate calibration of feature data is achieved.Secondly,this paper selects the LGBM(Light GBM)that can realize data and feature parallelism,and quickly train and test massive data to classify the feature data.Aiming at the problem of the imbalance in the number of chatter samples and normal samples in turning and milling processing,a scheme based on the grid search for the optimal category weight parameters is proposed,which to a certain extent alleviates the disadvantages of the imbalance problem on the model training effect.Finally,this paper verifies the applicability and reliability of the LGBM model through a comparative analysis of the accuracy of the test set and the time-consuming prediction of the multiple models.In the in-domain cutting parameters and the out-of-domain cutting parameters,the experiments of turning stainless steel long hollow shafts and aluminum block milling verify the effectiveness and accuracy of the chatter sensitivity features and chatter detection method proposed in this paper.Regarding the phenomenon that the recognition accuracy of out-of-domain cutting parameters machining chatter is lower than that of in-domain cutting parameters,this paper explains the relevance of the recognition accuracy and its relationship from the perspective of the distribution of feature data in the feature space,which provides a theoretical basis and guidance for the reuse of the model.