Hybrid Model Based On Feature Selection And Transfer-learning For Workpiece Geometric Accuracy Prediction Of Five-axis Flank Milling

Geometric accuracy prediction is a crucial part of error compensation,which has important guiding significance for improving machining quality.At present,the mechanism of the complex five-axis flank milling system under the coupling of multiple factors is not clear,and the traditional prediction models based on the analysis theories such as milling force and forming mechanism are limited.For high-precision geometric accuracy prediction,this paper deduced the variation rule of geometric accuracy,and further proposed a data-driven hybrid model based on feature selection and transfer-learning.The main research contents are as follows:The milling cutter error model of the effective milling radius considering the tool runout and the force deformation of the tool was introduced.Combined with the characteristics of five-axis flank milling,the variation rule of the workpiece geometric accuracy was deduced—along the tool feed direction,the geometric accuracy at different contour lines on the same workpiece has a similar variation rule,and its reliability was verified by machining experiments.Based on feature selection and transfer learning methods,a hybrid model for highprecision geometric accuracy prediction was proposed.The model performed feature selection through recursive feature elimination with cross validation to obtain parameters suitable for model training;transfer-learning was performed through Tradaboost to transfer the geometric accuracy data with similar variation rule.The model took the machining parameters and the geometric features of the workpiece as input,and the geometric accuracy of the workpiece as the output.The model was trained with a small amount of data,so that it can learn the mapping relationship between input and output independently to realize the prediction of geometric accuracy.Taking the "S" test piece as the research object,and the point contour error as the geometric accuracy evaluation index,the optimization and verification of the hybrid model were carried out.Through the comparison of multiple sets of experiments,the influences of the two hyperparameters,the number of measurement points and the way of point distribution in the target domain,on the prediction accuracy were explored.Finally,the appropriate hyperparameters were selected to optimize the model and improve its prediction accuracy.Through comparative experiments with the prediction model without feature selection and the support vector regression model with feature selection,the influence of feature selection and machine learning model on prediction accuracy was further explored.The experimental results showed that the prediction accuracy of the hybrid model proposed in this paper was improved by 40.7% and 65.3%,respectively,which proved that the feature selection method and transfer-learning proposed in this paper are reliable.The model had good prediction accuracy and generalization ability under the condition of reducing the data sample size requirement,and improved the geometric accuracy evaluation efficiency.In this paper,for the prediction of the geometric accuracy of the five-axis flank milling workpiece,a complete set of geometric accuracy rapid evaluation methods including the analysis of the geometric accuracy variation rule,the selection of influencing factors,and the prediction of point contour error based on transfer-learning were formed.The reliability of the method was verified by experiments.