SGSE Night November 20: AI4MBSE

Why Go Alone If Better Together: The Roles of LLM, NLP, and Ontology in MBSE for Generating Patterns of System Architecture in Mission-Critical Systems

Abstract
Large Language Models (LLMs) offer new opportunities for automating system architecture modeling, but their “black box” nature limits explainability, traceability, and regulatory compliance—particularly in mission-critical domains such as aerospace, defense, automotive, and energy. This idea highlights that true value emerges not from LLMs alone, but from their integration with ontologies and Natural Language Processing (NLP) within Model-Based Systems Engineering (MBSE). Ontologies provide the semantic backbone that grounds LLM outputs in verifiable contents and regulatory constraints, while NLP extracts and translates unstructured/semi-structured specifications into structured elements, specifically the requirements. We propose a hybrid framework in which SysML agents serve as both a contextual model and a constraint within a multi-agent architecture. This ensures requirements, structures, behaviors, and parameters remain traceable and reviewable. By combining the power, the approach transforms the Gen AI into trusted copilots for MBSE, enabling pattern-based model generation that balances automation, traceability, and compliance.