| Explosions remain a common form of attack because they inflict a high level of destruction, on both structures and on the communal psyche, for detonating a relatively low cost device. Buildings are susceptible to such attacks because standard design considerations such as weather and vibration do not, in general, also account for blast loads. The potential damage from an explosion may be accurately calculated using computational fluid dynamics to solve the partial differential equations of the underlying physical models. In addition, the design of more effective structures to mitigate the damage from these blasts may be formulated as a minimization problem for a well-defined cost function, typically some combination of pressure and impulse on a particular portion of the building.;Adjoint methods allow the gradients relating cost function to design variables to be calculated with only O(1) unsteady flow runs per design cycle. The present thesis will extend the capabilities of the adjoint method and then use those developments to produce an original design that solves a complicated engineering problem.;First, a series of segmented interpolation schemes for performing adjoint runs is developed including recursive equidistant spacing with data compression to reduce the otherwise severe storage requirements of Jacobian reconstruction. In addition, a scheduled binomial spacing scheme that offers the potential for even greater savings is developed and tuned for minimal overhead. Comparison analysis for time and data storage focuses on a double-wall blast problem, an exemplar for this class of engineering design problems. Theoretical results also predict how the performance of these schemes will scale for ever larger problems.;Then, the use of the adjoint method is demonstrated by solving a complex design problem in step-by-step detail. A thorough system of visualizing gradient information and translating that information into modifications of the design geometry is developed. Also, the use of porous media is considered, various theoretical models compared, and their implications analyzed. Otherwise non-intuitive features suggested by the adjoint ultimately lead to a high-quality design for an original vestibule that mitigates damage from a large explosion. Verification runs are performed on an entire series of designs and for multiple blast scenarios to confirm our results. |
