Design optimization with sensitivity constraints

One important problem with using optimization models for design is that parameter values, and thus model results, are often uncertain. A general and practicable approach, sensitivity constrained nonlinear programming (SCNLP), is developed for extending nonlinear programming models to include functions of the system sensitivity to changes in parameter values. Using SCNLP, sensitivity-based functions can be constructed that represent approximately many meaningful design criteria, including first-order estimates of the variance, reliability, and robustness. Thus SCNLP can be used to generate alternative designs that are good with respect to traditional objectives, and that incorporate explicitly the concerns about uncertainty in parameter values. A solution procedure is presented, and an implementation is discussed briefly. To illustrate the approach, a simple heat exchanger design SCNLP is formulated and solved. More significant design problems can be considered, and an application of SCNLP to a complex wastewater treatment plant design problem is discussed briefly.