Research On Generalization Of Prediction Model Parameters And Uncertainty Characterization In Milling Stability

With the demand of high-precision and high-efficiency machining,unstable cutting has become one of the important factors restricting the machining efficiency and precision of products.In order to ensure high efficiency and high precision machining,it is necessary to select the optimal combinations of stable milling parameters through the milling stability lobe diagram.The tool tip Frequency response function(FRF)are one of the important inputs for predicting chatter and calculating milling stability lobe diagram.In the actual machining,the tool tip FRF are affected by many factors,it not only depends on the spindle speed but also position-dependent of machine tool.At the same time,the different tool-holder combinations of same machine tool and different machine tools of the same type also eventually affect the tool tip FRF,thus affecting the stability lobe diagram in the milling process.But beyond that,machining is a dynamic process and chatter behavior depends on a number of uncertain factors.Therefore,the stability lobe diagram of milling processes must be comprehensively described or identified from multi-dimensional factors according to the types of machine tools,the positions of machine tool,spindle speeds,different tool-holder combinations,uncertainty and other factors,so as to meet the application in complex industrial field.This work focuses on the generalization of prediction model of tool tip dynamics and uncertainty characterization of its stability in milling process.The main content of this thesis is as follows:A more efficient “inverse stability solution” method is proposed to identify the tool tip dynamics in milling process and the corresponding stability lobe diagram is calculated.The average cutting force method is used to calibrate the radial and tangential cutting force coefficients accurately based on the classical two degree of freedom milling stability model.A method of the Spindle Speed Ramp-up is used to carry out the chatter experiment to efficiently obtain the chatter frequency and the corresponding axial depth of cut under different spindle speeds quickly.A heuristic particle swarm optimization algorithm is proposed,which can realize the reverse identification of tool tip dynamics based on the obtained chatter frequency and the depth of cut limit more efficiently,the proposed identification method is more effective than the previous methods.Finally,the stability lobe diagram in milling process is verified by milling chatter experiment,which proves that the prediction result is more accurate than that calculated by dynamics of tool tip under static condition.A transfer learning method based on conditional distribution adaptation is proposed to predict tool tip dynamics for different tool-holder combinations,which is position-dependent and speed-dependent.For a large five axis beam moving gantry machining center,considering the spatial correlation and variability of the tool tip dynamics that position-dependent and speed-dependent,the dynamics prediction model of the source tool tip is developed by using Kriging method.In order to improve the generalization of the data prediction model of tool tip dynamics for different tool-holder combinations,the source tool data can be used to assist the training of the target tool prediction model because of the target tools are few labels.Considering the condition invariant components with high discriminative power,in order to discover more useful conditional transferable components,a location-scale transformation across domains is used to find more useful conditional transferable components in the source tool,so a conditional distribution adaptive transfer learning method is proposed to predict the tool tip dynamics of the target tools with position-dependent and speed-dependent.Finally,the stability in milling of target tool predicted by transfer learning is verified by experiment.Taking five VMC850 E machining centers as the research object,the differences of tool tip dynamics of different machine tools are analyzed under the same conditions.In order to enhance generalization capability of the data prediction model of tool tip dynamics of different individual machine tools,for the same tool-holder combinations as the source machine tool,a tool tip dynamics offset model is developed based on the smoothing assumption between source machine tool and target machine tools.A transfer learning method based on model offset is proposed to predict tool tip dynamics of the target machine tools with position-dependent and speed-dependent.For the target machine tool with different tool-holder combinations from the source machine tool,a joint distribution weight adaptation transfer learning method is proposed to predict tool tip dynamics of the target machine tools with position-dependent and speed-dependent.Finally,the stability in milling of target machine tools predicted by transfer learning is verified by experiment.Considering the influence of uncertainty factors in milling process,a Bayesian and improved affine invariant Markov chain Monte Carlo method is proposed to quantify the uncertainty of the stability lobe diagram in milling process.A two degree of freedom model is developed considering the influence of uncertainty factors in milling process.An improved Morris global sensitivity analysis method is proposed to determine the input parameters that have the most significant effect on the stability lobe diagram in milling process.According to the influential input parameters determined by the global sensitivity analysis,the posterior distributions of all influential input parameters are calculated by the continuously updated chatter frequency and limit axial depth of cut using a Bayesian inference framework,which is built based on an improved Goodman and Weare Markov chain Monte Carlo(GWMCMC)algorithm.According to the posterior distribution of the influential input parameters,the uncertainties are propagated in the milling process stability model through Monte Carlo(MC)sampling technology,so as to calculate in the milling process and its probability distribution of uncertain interval.Finally,a variety of uncertain factors are designed in milling to verify the stability lobe diagram and its uncertainty boundary.