System Architecture Design Space Exploration: Integration with Computational Environments and Efficient Optimization

Kurzfassung

System Architecture Optimization (SAO) enables automatically exploring combinatorial system architecture design spaces, which can reduce bias and enable more architectures to be considered in early design phases. This paper presents the Architecture Design Space Graph (ADSG), a directed graph for modeling architecture design spaces, and encoding them as optimization problems to be solved by optimization algorithms. The ADSG is built on top of the Design Space Graph (DSG), which models hierarchical design spaces using selection and connection choices, where selection choices define which nodes are selected in architecture instances, and connection choices represent source-to-target connection problems. Selection and connection choice encoders are introduced that enable full enumeration of valid design vectors, and ensure any source-to-target connection problem can be encoded such that optimization algorithms can effectively search the design space. The ADSG extends the DSG for use in system architecting, by defining nodes such as functions, components and ports, which enables function-based architecture definition. The ADSG can be modeled using ADORE: a Python tool with a web-based GUI that allows connecting to performance evaluation code through Python-based and file-based interfaces, and connecting to open-source architecture optimization algorithms. The presented method is demonstrated by three application cases, each demonstrating different evaluation and optimization aspects: a multi-stage launch vehicle, a guidance, navigation and control system, and a jet engine architecture. It is shown that SAO problems formulated using ADORE perform as well, if not better, than manually-defined SAO problems, without requiring the user to be an expert in formulating optimization problems.