CFD-based approximation concepts for aerodynamic design optimization with application to a two-dimensional scramjet vehicle

This dissertation investigates the application of approximation concepts to aerodynamic optimization. Such techniques, which are gaining popularity in structural optimization, offer the potential of providing the accuracy of a high-fidelity "detailed" analysis model at greatly reduced computational cost. This is because the detailed model is used only to "fine-tune" an approximate model which is then used in the optimizer. The test problem treated is the design optimization of a 2-D scramjet vehicle flying at Mach 6.0 at 30 km altitude. The objective function is net thrust. The following approximation concepts are used: the Taylor series approximation to wall pressures and inlet plane flow properties; and Haftka's Global-Local Approximation applied to the same variables. The performance of these techniques is compared to that for optimization using CFD alone. Cost reductions are quantified.; It is shown that modifications must be made to the formulation of the approximation concepts as they are used in structural optimization, due to the changing grid geometries required by the CFD solver. All correction factors for the approximation concept are applied not to the CFD grid points, but to a constant, dense, nondimensionalized "correction point grid", which does not change as the CFD grid changes. It is also shown that, in areas where discontinuous phenomena are not important (such as in the scramjet nozzle), the approximation concepts can be successfully used, after this modification is made. Optimizations of the nozzle region show that all the approximation concepts result in a 68% reduction in the number of calls to the CFD solver.; In regions dominated by shock impingements (such as the forebody/inlet), it is found that approximation concepts applied to point properties cannot be used as they currently are in structural optimization, due to the effects of shock movement during correction factor calculations, and due to artifacts of the CFD solver, such as shock smearing. In fact, even though the CFD and the (uncorrected) approximate models optimize to very nearly the same design, the Taylor series and GLA fail to do so. However, application of the GLA to the integrated objective function (net thrust) with zeroth-order correction factors, is unsuccessful.; To lay the groundwork for future investigation, a method of improving the behavior of the point-property GLA in the presence of shock impingements is developed and tested. This involves using "floating" pre- and post-shock coordinate systems for each wall surface. The result is a dramatic reduction in the erratic behavior of the GLA. This technique may form the basis of a generally-applicable GLA technique for aerodynamic optimization.