Prediction Model For Surface Roughness By Ultrasonic Vibration Milling Optical Glass Materials Based On Artificial Neural Networks

Optical glass, due to its great high brittleness, high hardness and low fracturetoughness, is a typical difficult-to-cut material. It is difficult to achieve high-efficientand precision machining by traditional processing method. Milling process involveshigher machining-efficient and shorter processing cycle than grinding process whichwas usually used in machining optical glass, but with higher cutting force and worsemachining quality. It can be solved by combining ultrasonic vibration and traditionmilling process, which can achieve fine surface quality with high machining efficientand short processing cycle, by decreasing cutting force and cutting heat.Surface roughness is one of the most important quality factors of optical glass’smachined surface. The ultrasonic vibration milling parameters can be optimized throughsurface roughness predicting before real machining. In this thesis, the influence factorsin milling process are analyzed and the empirical formula applied to ultrasonic vibrationmilling is derived from tradition milling process.The prediction of surface roughness by ultrasonic vibration milling optical glasswas discussed. Several single factor experiments and two groups of multi-factororthogonal experiments have been carried on by ultrasonic vibration machining ZK1optical glass material in the XH714B vertical machining center, to study the influenceof the influencing factors on the surface roughness. In this thesis, these influencingfactors include machining parameters such as spindle speed, feed per tooth and millingdepth, ultrasonic vibration parameters such as frequency and amplitude and tool radius.In this thesis, the influence rule of joint action of influencing factors on the surfaceroughness by ultrasonic vibration milling optical glass is studied through the two groupsof multi-factor orthogonal experiments. And prediction model for the surface roughnessis established based on the experimental results. The artificial neural networks theory isintroduced in this paper as the theoretical basis of the prediction model, and theinstruction of hidden mode of BP neural networks is improved by the LMBP algorithmwhich combined with ACBP. Meantime; the prediction model based non-linearregression theory is established as well. And the prediction model with higher predictionaccuracy can be found through comparing the approximation error and generalization error of the both models.