Monitoring Research On The Process Stability And Tool Wear In Milling Titanium Alloy Thin-walled Parts

Materials of aviation parts,especially engine parts,have strict requirements on hardness,strength,toughness,lightweight and high temperature resistance.Therefore,titanium alloy thinwalled structure is widely used in aviation parts.However,it has the characteristics of difficult machining,low stiffness,easy to stick to the cutter and so on,and the complex and changeable dynamic characteristics of thin-walled parts during milling.Therefore,chatter and severe tool wear and high cutting temperature are easy to occur in the process of milling,which seriously affects the milling accuracy,surface quality and efficiency of titanium alloy thin-walled parts.In practice,adopting conservative process parameters can effectively avoid chatter,but it seriously restricts the machining efficiency.Especially,when designing milling process for new part geometry,it needs repeated experiments,which is inefficient,time-consuming and dependent on experience of workers,and can not meet the requirements of high precision,high efficiency and intelligent machining.For this reason,in this paper,the author took titanium alloy thin-walled parts as the research object and aimed to realize the stability of the milling process and the monitoring of tool wear for milling processing process.By establishing the milling force model and chatter prediction model of titanium alloy thin-walled parts,the online monitoring method of machining chatter is explored,and the monitoring and prediction model of tool wear is constructed,which lays a theoretical foundation for building a process monitoring system for milling titanium alloy thin-walled parts.The main research work of this thesis is introduced as follows.(1)A prediction model of cutting force and chatter is established considering the influence of tool runout and workpiece deformation on the milling process of titanium alloy thin-walled parts.The influence of tool runout and workpiece deformation is comprehensively considered in the milling force of flat-end milling cutter.First,the geometric parameter model of milling process is established by using cutter hypocycloid motion.Then,the actual cutting model of titanium alloy thin-walled parts is deduced and the thickness model of milling undeformed cutting is obtained.On this basis,combining the average force method to solve the milling force coefficient,the milling force model is established by the linear cutting force method,and the accuracy of the milling force model is verified by experiments and results analysis.Then,according to the proposed milling force model,a system dynamic model of undeformed cutting thickness,considering both tool runout and workpiece deformation,is established.The stability of milling is analyzed by frequency domain method based on 2-DOF dynamic model and considering the relative transfer function of tool-workpiece machining system,and the boundary conditions of machining stability are obtained.After that,experiments and simulations are carried out to verify the reliability of the stability analysis,and the effects of different machining parameters on the machining stability are compared and analyzed.(2)Based on the dynamic characteristics of tool-workpiece machining system and WPT,the chatter identification of stable and chatter milling conditions are carried out.The original signal is decomposed by WPT,and in order to effectively remove noise,the sub-signal containing chatter information is extracted for signal recombination.Afterwards the relationship between the dynamic characteristics of spindle system and chatter identification is established.The results show that the reconstructed signal containing the natural frequency of the tool-workpiece machining system can effectively represent the original signal and effectively identify chatter.In order to extract the chatter frequency band and calculate the chatter index more accurately,a chatter monitoring method is proposed,which combines WPT and optimized variational mode decomposition(O-VMD),then extract the multi-scale permutation entropy(MPE)as an indicator to judge the processing state.The raw signal is decomposed by WPT,and the first signal reconstruction is carried out by calculating the energy features.The reconstructed signal includes the sufficient chatter information.Then,particle swarm optimization is used to optimize the decomposition parameters of VMD.Based on the optimized VMD,the reconstructed signal is decomposed,and then the energy entropy features of each IMF after decomposition are calculated,and the effectiveness of extracting chatter frequency bands is verified using Hilbert spectrum.Finally,the calculated MPE is used as a chatter feature to verify the effectiveness of the proposed method for identifying chatter by comparing and analyzing the entropy values of three different machining states(stable milling,slight chatter,and severe chatter).(3)Based on the built-in current sensor of machine tool,the monitoring method of tool wear is studied.In feature extraction,to fully consider the changing rule of tool wear,statistical features with the same changing trend as tool wear are selected.When selecting features,in order to fully reflect the predictability of tool wear,the predictive index of features is quantified.An unsupervised learning method of the Stacked Denoising Autoencoder(SDA)is introduced to realize automatic feature dimensionality reduction.On account of the constructed multi-layer deep neural network model(DNN),and combined with principal component analysis(PCA)and t-SNE(T-distributed neighborhood embedding),the original signal is analyzed for secondary dimensionality reduction.On this basis,DNN network consisting of four DAs is adopted as the classification model of tool wear state.In order to classify the tool wear state,a Softmax layer is added at the output end of DNN.The labeled data of the original fine-tune the DNN parameters for the first time through back propagation.Then,taking the features of primary dimensionality reduction and secondary dimensionality reduction as new labeled data,the parameters of DNN network are deeply fine-tuned twice by back propagation.It is shown that the classification accuracy of data sets is improved compared with other methods.(4)Research on prediction method of tool remaining useful life(RUL)by combining convolutional neural network(CNN)and bidirectional long-term and short-term memory circulation neural network(BLSTM).Mathematical statistical features and time-frequency features constitute the feature set of the original data.Mathematical statistical features,such as maximum,average and kurtosis,and time-frequency features,such as energy spectrum and kurtosis,are extracted from the data in each sampling interval.In this foundation,the method of normalization to the interval of(0,1)is used to preprocess the feature set.CNN is used to make further feature selection for further effective feature extraction.Input the result of feature selection into BLSTM model,and build full connection layer and linear regression layer to predict and analyze the tool RUL.The prediction effect is evaluated by the mean absolute error(MAE)and root mean square error(RMSE)of the data.It turns out that this model has the characteristics of high prediction accuracy without manual feature selection and prior knowledge reserve.There are 89 figures,25 tables and 145 references in this paper.