| A strategy has been developed for performing constrained design optimization involving large-scale simulations or experiments. The strategy significantly reduces problems which have existed when attempting a deterministic sensitivity analysis. Statistical response surface methodology (RSM) is used to estimate parameter sensitivities and to de-couple the search and analysis phases of optimization. The benefits result from two characteristics of RSM: (1) a priori selection of analysis cases provides increased flexibility in performing the analysis, and (2) "smoothing" of the sensitivity information occurs when developing a least-squares regression model, thus reducing the effects of computational or measurement noise. The net result is an expansion of the class of design problems to which modern optimization methods may be applied. |
PDF Full Text Request