Prediction Modelling Of Surface Roughness Of Ti Alloy Milling

Surface roughness is one of the most important targets measuring quality of machined surface. The machining data can be optioned through predicting surface roughness, so as to increase labor productivity and consume cost and induce laboring intensity before real machining. Geometric model and regression analysis and neural network surface methodology were establish.The surface roughness of titanium alloy TC4 milling experiments were investigated, comparing the simulation results with the experimental results.Also studied the influence of cutting edge shape, tool titling direction, titling angle, feeding mode, run-out of spindle and periodic axial slip,etc.on machined surface roughness in the case of ball-end mill machining. Applying the coordinate transition principle and matrix calculation rule the track function of any point in workspiece coordinate is set up during milling procedure of multi-axis about spherical-shaped milling cutter.Considering cutting speed and feed rates and radial depth-of-cut, regression analysis and neural network surface methodology were adopted to build a prediction model of surface roughness. Through the verification of the built prediction model of surface roughness, result shows that the predietion accuracy and generalization of the model are very high and the model can conveniental predict the effects of milling parameters on surface roughness of machined surface , which contributes to accurately understand the variation law of quality of machined surface following milling parameters and provides the foundation for properly selecting cutting parameters and controlling quality.Three factors and three levels orthogonal milling experiments of titanium alloy TC4 were run,range of milling surface roughness experiments results were obtained. And also training samples were provided and to accomplish model's vertificaition; meanwhile, analyzing the accuracy and feasibility of the established model.Comparing with the experimental results reveals that the model method developed in this paper is efficient.