When designing complex systems, choices related to the system architecture, the description of the functions and components of a system, greatly influence to which extent design goals can be achieved. The architecture design space, the set of all possible architectures for a given design problem, can be extremely large due to the combinatorial nature of architectural choices. Additionally, the integration of innovative technologies for future systems requires the application of multidisciplinary, simulation-based evaluation. These two challenges are addressed by System Architecture Optimization (SAO): the combination of numerical optimization algorithms with multidisciplinary, simulation-based evaluation, to explore architecture design spaces without requiring the evaluation of all architectures.

A function-based method is developed for modeling SAO problems. The Architecture Design Space Graph (ADSG) is developed to represent function-to-component allocation, component characterization, and component connection choices. Algorithms are developed to automatically encode an ADSG as a numerical optimization problem, and to decode generated design vectors into architecture instances. The modeling method is made available through a web-based GUI application.

SAO problems are solved using evolutionary and Bayesian Optimization (BO) algorithms. A new hierarchical sampling algorithm is developed to prevent under- or over-sampling regions in the design space when building the initial Design of Experiments (DoE). Investigations are performed into correction algorithms and into exposing information about design space hierarchy. For BO, a strategy is developed to deal with hidden constraints stemming from simulation failures. It is shown that both evolutionary and BO algorithms can solve SAO problems, however BO can do so with 92% less function evaluations as demonstrated for a realistic SAO problem.

Multidisciplinary, simulation-based evaluation in large, cross-organizational systems engineering projects is enabled by leveraging collaborative Multidisciplinary Design Analysis and Optimization (MDAO). Methods are developed for automatically modifying the behaviour of a collaborative MDAO workflow for each evaluated system architecture, for propagating all information about the system architecture to the Central Data Schema (CDS) as used in collaborative MDAO, and for executing the architecture generation process in the computational environment where the workflow is executed.