Space-filling experiment designs for branching and nested factors via dynamic concept-subspace allocation

Defense test and evaluation often explores complex systems through simulation modeling. These experiments are costly, time-consuming, or computationally intensive. Consequently, experiment designs must support limited executions while preserving broad coverage of multidimensional input spaces. In addition, applying space-filling experiment designs is complicated by optional and conditional inputs, such as in system-of-systems studies with optional subsystems and policies. These cases induce variable-dimension hierarchical design spaces with multiple feasible concept subspaces or distinct sets of nested factors that are conditionally relevant based on branching factors’ values. This research proposes and evaluates algorithms to design space-filling experiments that recognize the hierarchical relevance conditions in these design spaces. We extend space-filling metric computations for hierarchical design spaces and propose metrics to quantify distinct active branching factor combinations, concept subspace, and coverage. In numeric experiments, we evaluate the resulting designs and find that the proposed algorithms create designs with a better maximin design-point distance compared with traditional space-filling designs that ignore the conditional relevance of factors. In a system-of-systems case study, we execute the designs and assess the results for output space-filling properties, finding the proposed algorithms produce statistically significant improvements in output-space-filling metrics relative to random designs and traditional Latin hypercube designs.