Surface roughness prediction in micromilling using neural networks and Taguchi´s design of experiments

The present research is to analyze the effects of spindle speed (n), feedrate (f) and axial depth of cut (ap) on average surface roughness parameters (R_a) in the micromilling operation. Compared with the conventional milling operation, the non-linearity of micromilling is more obviously, because of the minimum chip thickness, tool radial error motion and workpiece inhomogeneous inherent. In this work, the experimental design adopts the Taguchi´s approach to acquire enough training information with minimal experiment number. Based on the experimental results, a neural network model is developed, trained and used to predict the bottom surface roughness in the micromilling operation. Finally, the effects of each machining parameter and the interaction effects of each two-parameter combination to R_a are analyzed in detail.