Abstract
Many of today’s industrial firms seek the continuous and systematic reduction of variability as a primary engineering goal across key production dimensions. Robust parameter design (RPD) is often considered among the most important methods for achieving these ends. Focused on statistical modelling and numerical optimisation strategies, most researchers typically assume processes possess moderate to low variability, which facilitates the use of ordinary least squares (OLS) regression. Realistically, however, industrial processes often exhibit high variability. In such cases, many of the modelling assumptions underpinning OLS methods do not hold. Consequently, the results and recommendations provided to decision-makers using these could generate suboptimal modifications to processes and products. This paper proposes an alternative method for dealing with high process variability. Specifically, using the coefficient of variation to identify influential sources of variability between design points, the proposed method advocates removing these sources and then applying optimal design theory to rebalance the experimental framework. Thereafter, RPD optimisation schemes may be applied to obtain more precise optimal operating conditions with less variability and bias. A numerical example combined with Monte Carlo simulation is used to illustrate the proposed procedure.