Enhancing Model-Based System Architecting Through Knowledge-Based Design Space Exploration

Abstract

System architecting is a critical stage of the system lifecycle, where many potentially costly mistakes may be made. MBSE has been widely applied to modelling a system and its architecture. However, less attention has been given to the detail of design exploration and design decisions being
made at this stage of system development. As such, a set of research questions have been formulated, considering: the current state of MBSE wrt. design space exploration, techniques that could be used or developed to improve design exploration, and benefits or limitations of emphasizing system architecture design exploration within MBSE processes.

In responding to these questions, this thesis has reviewed current MBSE environments and their current issues wrt. to design exploration. In doing so, 21 requirements were identified for better addressing design space exploration for system architecting using MBSE. Potential techniques that can complement typical MBSE system architecting practice were then investigated. In response to the 21 identified requirements, a knowledge-based method for guiding systems architecting, DRAGONS, was developed. This was the primary contribution of the thesis. This approach drew on adaptable knowledge bases that can be trained on existing system models and is decoupled from the design space exploration stage. This allowed the system architect the final choice of how a design solution should be selected. This method included an ontology with 5 fundamental types of architecture element (Requirement, Function, Component, Mode and Parameter) and 6 relationship types (Parent, Satisfy, Assigned to, Grouped to, Dependency and Interface). However, a major feature of this method was that its knowledge base, and underlying ontology, could be refined and expanded by eliciting knowledge from existing system architectures.

This method was applied to a range of earth observation spacecraft for validation, using a set of metrics to provide an in-depth understanding of its utility. The dataset used for this testing included 1,674 nodes and 10,170 relationships, sourced from 17 designs and 7 requirement specifications. The completed knowledge base developed for DRAGONS during this process included 304 different types and 6,329 relationships.

Furthermore, a system optimisation study was completed in full using this method, and finally, it was coupled with an LLM to investigate AI methods for system architecting. Overall, it was found that knowledge-based techniques could be used to effectively guide a system architect towards good quality architectures. Being decoupled from the design space exploration makes these methods flexible across different design problems types, though not without some difficulties for parameter-based problems. This approach offers opportunities for automation that speed up the design process and allow deeper exploration of designs, but human input is still essential to achieve high quality designs.

Date of Award	30 Sept 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Chris M Snider (Supervisor) & Lucy Berthoud (Supervisor)