The Study On The Prediction Of Multi-axis Milling’s Surface Roughness Based On The BP Neural Network

With the development of processing technology, the processing technology of multi-axis milling has been widely used in developed countries. However, it is still in its infancy in China. As the machinery industry develops rapidly, how to obtain a high quality surface has become a dilemma for researchers.In this paper, how to control the multi-axis milling’s surface roughness is deeply studied, and the theory of BP artificial neural networks is used to build mathematical models for the purpose of predicting the surface roughness. It provides a new idea of the improvement of the surface quality and the optimization of machining parameters in the multi-axis milling areas, which has important theoretical and practical values. In this paper, the prediction of surface roughness of milling aluminum alloy 6061 using multi-axis CNC machining center is mainly studied. The detailed content is as follows:(1) The processing mechanism of milling is analyzed,andthe Multi-axis face milling of aluminum alloy 6061 is carried out based on the analysis of the factors affecting the surface roughness. Complete randomalized design is processed aimed at exploring the influence of the effects of various factors affecting surface roughness, and providing theoretical and experimental basis for better prediction of surface roughness.(2) By introducing BP artificial neural network theory to the field of multi-axis milling, the influences of cutting speed Vc, feed quantityfz, cutting depthapand Spacing h to the surface roughness in face milling are researched. The forecasting surface roughness network model of multi-axis milling is established, say, the input layer,the hidden layer and the output layer. By dynamically adjusting the number of nodes in the hidden layer, the final structure of the 4-9-1 network is determined. Aimed at the key shortcomings of the BP network,say,the slow convergence and prone to converging to local minima, the weights and thresholds are optimized to achieve a better convergence effect by using the genetic algorithm. Through training and verifying network model, the prediction accuracy and convergence ability of the model is finally confrimed to have reached the preset requirements.