Using artificial neural networks for the prediction of dimensional error on inclined surfaces manufactured by ball-end milling

Industrial demand for models and simulation tools that can predict dimensional errors in manufacturing processes is vigorous. One example of these processes is ball-end finishing of inclined surfaces, which is a very complex task, due to the high number of variables that may influence dimensional errors during a cutting process and their different nature. This work firstly analyses the potential of semiempirical models to address the ball-end milling finishing, to conclude that these models are unable to process and to replicate the full range of milling strategies and slope combinations. Secondly, it goes on to analyse the possibilities of artificial neural networks as a means of overcoming this limitation. Two types of neural networks, multilayer perceptron (MLP) and radial basis functions (RBF), are tested. The results show that RBFs predict better than MLPs in all cases, achieving a precision of 1.83 mu m in root mean squared error (RMSE) and a correlation coefficient of 0.897 with a 10 x 10 cross-validation scheme. Their training and tuning times are also 2.5 times shorter in all cases. Finally, the use of 3D figures, generated from the best RBF model, yields interesting industrial results in the field of process engineering.