Comparison of search algorithms for assessing airblast effects

Three spatial search algorithms to assess the maximum possible range-to- effect as a function of blast origin are compared. The cost function considered is the number of windows destroyed due to a blast. The 3-D flowflelds are computed using an Euler solver on optimized unstructured grids with 1-3 million elements. Efficient parallel execution of many scenarios on a network of computers ensures answers with reasonable computational times. The three methods compared comprise: a) a standard genetic optimization technique, b) a Kriging model, and c) a Finite Element approximation approach. The results indicate that the two latter models outperform the standard genetic optimization technique by a considerable margin, enabling an accurate search for locations of maximum possible blast damage with far fewer cost function evaluations. In some cases, the number of cost function evaluations (i.e. 3-D Euler runs) can be reduced by an order of magnitude. Moreover, both the Kriging model and the Finite Element approximation approach allow for reliability bounds, thereby enabling proper stopping criteria.