Using approximations to accelerate engineering design optimization
Virginia J. Torczon, Michael W. Trosset · 7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization · 1998
Optimization problems that arise in engineering design are often characterized by several features that hinder the use of standard nonlinear optimization techniques. Foremost among these features is that the functions used to dene the engineering optimization problem usually require the solution of dierential equations, a process which is itself computationally intensive. Within a standard nonlinear optimization algorithm, the solution of these dierential equations is required for each iteration of the algorithm. To mitigate such expense, an attractive alternative is to replace the computationally intensive objective with a less expensive surrogate. In conformance with engineering practice, we draw a crucial distinction between surrogate models and surrogate approximations. Surrogate models are auxiliary simulations that are less physically faithful, but also less computationally expensive, than the expensive simulation that is regarded as "truth." An instructive example is the use of an equivalent-plate analysis method in lieu of a nite element analysis, e.g. to analyze a wing-box of a high-speed civil transport. Surrogate models exist independently of the expensive simulation and can provide new information about the physical phenomenon of interest without requiring additional runs