Research On Surface Roughness Prediction Of Ball-end Milling Based On Theoretical Model Driven Knowledge-based Neural Network

In the actual service environment,unqualified surface processing quality can easily lead to wear,fatigue,corrosion,and other defects in advance,which will not only cause great economic losses,but also endanger life safety.The online prediction of the surface roughness of the parts is the basis for achieving processing quality assurance,processing efficiency improvement and controllable cutting process,which has positive practical significance.In this paper,the surface roughness prediction of ball-end milling is studied:Firstly,the research background and significance of ball-end milling surface roughness prediction are summarized,and the research status at home and abroad is summarized based on theoretical modeling method and data-driven method.Due to the lack of process information,high computational cost and the uncertainty of manual adjustment parameters,the prediction accuracy of the theoretical model is low.In addition,due to the lack of theoretical support and the “black box problem”,the application of data-driven methods in practice is also limited.Then,in view of the above problems,this paper first proposes the idea of combining the theoretical model of surface roughness with the data-driven method,that is,Knowledgebased neural network(KBa NN).Its core idea is to use the low-dimensional characteristics of dynamic signals to model the theoretical modeling results twice,which is realized by Knowledge-based neural networks with radial basis functions(KBa NN＿RBF).The modeling process of this method is as follows:1)The surface roughness of ball-end milling is theoretically modeled by using tool parameters and milling parameters,and the theoretical modeling results are used as prior knowledge.2)The low-dimensional features are extracted from the ball-end milling monitoring signals after feature selection using the nearest neighbor component analysis(NCA)algorithm to correct the prediction error caused by only considering the static information in the theoretical model and use it as the input of the model.3)KBa NN＿RBF is established to improve the traditional KBa NN,which improves the ability to deal with nonlinear problems,so that it can better adapt to this work.4)The prior knowledge is embedded into the KBa NN＿RBF model,and the lowdimensional features are used as supplementary information.The theoretical modeling results are re-modeled,and the particle swarm optimization(PSO)algorithm is used to optimize the network structure to achieve accurate prediction of ball-end milling surface roughness.Finally,a multi-condition ball-end milling verification experiment was designed.The results show that KBa NN＿RBF.