In this study the use of automatic differentiation techniques significantly reduces the number of analysis calls and the CPU (Computer Processing Unit) time required for robust optimum design. Traditionally, robust optimum design procedures have relied on finite differencing techniques to eavaluate sensitivities for use in gradient-based optimization. Robust objective functions and robust constraints are typically formulated as compound functions that invoke the original objective or constraint many times for each robust evaluation. In this research, automatic differentiation techniques are used to avoid the costly finite differencing of robust objective functions and robust constraints. Sensitivities calculated using automatic differentiation are exact and therefore enhance performance. Two new robust optimization extensions are developed to evaluate the utility of using automatic differentiation in robust optimization. One is a sensitivity-based procedure, and the other makes use of experimental design techniques. These two new robust optimization extensions are successfully implemented in application to an aircraft concept sizing test problem.
Automatic differentiation in robust optimization
Automatische Differentiation bei der Optimalauslegung
AIAA Journal ; 35 , 6 ; 1072-1079
1997
8 Seiten, 12 Bilder, 1 Tabelle, 24 Quellen
Aufsatz (Zeitschrift)
Englisch
Automatic Differentiation in Robust Optimization
Online Contents | 1997
|Automatic Differentiation in Robust Optimization
AIAA | 1997
|Automatic differentiation in robust optimization
AIAA | 1996
|Automatic Differentiation in Robust Optimization
British Library Conference Proceedings | 1996
|