System architecting is concerned with exploring the tradespace of early, high-level, system design decisions with a holistic, value-centric view. In the last few years, several tools and methods have been developed to support the system architecting process, focusing on the representation of an architecture as a set of interrelated decisions. These tools are best suited for applications that focus on breadth - i.e., enumerating a large and representative part of the architectural tradespace -as opposed to depth - modeling fidelity. However, some problems in system architecting require good modeling depth in order to provide useful results. In some cases, a very large body of expert knowledge is required. Current tools are not designed to handle such large bodies of knowledge because they lack scalability and traceability. As the size of the knowledge base increases, it becomes harder: a) to modify existing knowledge or add new knowledge; b) to trace the results of the tool to the model assumptions or knowledge base. This thesis proposes a holistic framework for architecture tradespace exploration of large complex systems that require a large body of expert knowledge. It physically separates the different bodies of knowledge required to solve a system architecting problem (i.e., knowledge about the domain, knowledge about the class of optimization or search problem, knowledge about the particular instance of problem) by using a rule-based expert system. It provides a generic population-based heuristic algorithm for search, which can be augmented with rules that encode knowledge about the domain, or about the optimization problem or class of problems. It identifies five major classes of system architecting problems from the perspective of optimization and search, and provides rules to enumerate architectures and search through the architectural tradespace of each class. A methodology is also defined to assess the value of an architecture using a rule-based approach. This methodology is based on a decomposition of stakeholder needs into requirements and a systematic comparison between system requirements and system capabilities using the rules engine. The framework is applied to the domain of Earth observing satellite systems (EOSS). Three EOSS are studied in depth: the NASA Earth Observing System, the NRC Earth Science Decadal Survey, and the Iridium GEOscan program. The ability of the framework to produce useful results is shown, and specific insights and recommendations are drawn.
展开▼
