Meshing noise effect in design of experiments using computer experiments

This work is intended to show the influence of grid length and meshing technique on the empirical modeling of current distribution in an industrial electroplating reactor. This study confirms the interest of usual DOES for computer experiments. Any 2D mesh generator induced, in this sensitive case, a significant noise representing only less than 5 per cent of the response. The 'experimental error' obeys a normal distribution and the associated replicate SDs represents 20 per cent of the global residual standard deviation. The geometry seems also to influence the corresponding noise. If the current density uniformity could be considered as a severe test, it is obvious that the noise generated by meshing would be amplified for 3D grids that will be in common use in future years. Copyright (C) 2002 John Wiley Sons, Ltd.