LLM-Assisted Semantic Alignment and Integration in Collaborative Model-Based Systems Engineering Using SysML v2

Abstract: Cross-organizational collaboration in Model-Based Systems Engineering (MBSE) faces many challenges in achieving semantic alignment across independently developed system models. SysML v2 introduces enhanced structural modularity and formal semantics, offering a stronger foundation for interoperable modeling. Meanwhile, GPT-based LLMs provide new capabilities for assisting model understanding and integration. This paper proposes a structured, prompt-driven approach for LLM-assisted semantic alignment of SysML v2 models. The core contribution lies in the iterative development of an alignment approach and interaction prompts, incorporating model extraction, semantic matching, and verification. The approach leverages SysML v2 constructs such as alias, import, and metadata extensions to support traceable, soft alignment integration. It is demonstrated with a GPT-based LLM through an example of a measurement system. Benefits and limitations are discussed.

• The paper presents a structured LLM-driven methodology for aligning independently developed SysML v2 models in collaborative MBSE.
• It introduces an iterative, prompt-driven process that includes syntax verification, semantic extraction, and traceability checks.
• The approach demonstrates improved model integration efficiency and consistent outputs with minimal human intervention.

LLM-Assisted Semantic Alignment for Collaborative MBSE with SysML v2

The paper "LLM-Assisted Semantic Alignment and Integration in Collaborative Model-Based Systems Engineering Using SysML v2" (2508.16181) proposes a structured approach leveraging LLMs to facilitate semantic alignment and integration of independently developed models in Model-Based Systems Engineering (MBSE), using the SysML v2 framework. This framework aims to enhance semantic interoperability and traceability across cross-organizational models by using LLM capabilities.

Introduction

Model-Based Systems Engineering (MBSE) seeks to improve systems engineering processes by formally modeling system elements and their interactions. Semantic alignment in collaborative MBSE, especially across different organizational entities, remains challenging due to differing modeling practices and tools. SysML v2 improves upon its predecessor by offering modularity and formal semantics, providing a potential solution for enhancing model interoperability. However, integrating systems with disparate semantics requires innovative approaches.

The development of LLMs, particularly those based on architectures like GPT, provides possibilities for aiding the understanding and integration of SysML models. By employing a structured, prompt-driven methodology, the paper offers an approach to refine the semantic alignment through iterative model extraction, matching, and verification processes. It utilizes SysML v2's constructs such as alias, import, and metadata extensions to achieve soft alignment integration.

Approach Conception

Challenges in Integration

The integration leverages the distinct benefits of LLMs in natural language processing to assist in overcoming several challenges:

1. Model Alignment Efficiency: Addressing semantic ambiguity and hierarchy mapping complexities to automate semantic extraction from disparate system models.
2. Output Validity and Consistency: Ensuring reliable and repeatable outputs from LLMs while addressing issues such as non-deterministic results and hallucinations.
3. Output Traceability: Facilitating traceability through input-output mapping records and confidence annotations.

Semantic Integration Concepts

The paper outlines three concepts for integration:

• Unified Modeling: Although beneficial theoretically, practical application is limited by organizational modeling needs.
• Transformation-Based Integration: Automatable model conversions across methodologies are constrained by the absence of standardized processes for SysML v2.
• Soft Alignment: Supports mapping models without structural modifications, utilizing extensible libraries of SysML constructs, making it an optimal approach for iterative and cross-organizational integrations.

Iterative Refinement

The approach is refined through multiple iterations, employing agile test principles and Design Research Methodology (DRM). Experiments demonstrate an improvement in output consistency and traceability, with structured prompts guiding the LLM through stages of syntax verification, semantic extraction, and alignment generation.

Resulting Processes and Implementation

Proposed Integration Approach

The approach systematically combines LLMs with SysML to generate traceable alignment results. Essential components include:

• Additive Modeling: Creating new alignment packages using the SysML v2 'import' mechanism ensures preservation of original model structures.
• Staged Process: Structured sequences that integrate LLM-guided steps, requiring human verification at each stage.
• Mapping Verification: Enhanced semantic consistency checks supported by structured prompts and user-defined semantic libraries.

  Figure 1: LLM-assisted SysML v2 Model Alignment Process

Process Overview

The method divides tasks into stages, ensuring process transparency through user confirmations and iterative refinements:

• Preparation and Syntax Confirmation
• Model Element Summarization
• Match Candidate Suggestion
• Mapping Verification
• Aligned Package Generation
• Consistency Check
• Export and Documentation

  Figure 2: Excerpt of Alignment Extension Library

Prompt-Driven Realization

Implementing the approach focuses on a structured prompt design that includes alignment via alias and extensions without altering original structures. Outputs include traceable JSON formats that contain confidence scores to aid interpretation and verification.

Figure 3: Example Model Alignment Results

Verification and Applicability

Initial tests showed that minimal human intervention rectified semantic deviations early in the prompt-driven workflow, enabling stable and consistent outputs. Verification with example model alignments highlighted the potential for broader scalability and application in complex industrial scenarios.

Discussion

While promising, the approach faces limitations regarding semantic-level alignment and user prompt configuration. Future work should incorporate ontology-driven enhancements for semantic constraints and explore RESTful API integration for greater toolchain applicability.

Despite these areas for improvement, the structured prompt-driven methodology lays a strong foundation for LLM-assisted SysML v2 semantic alignments, setting the stage for future research to deepen integration challenges and solutions in MBSE environments.

Conclusion and Future Work

This paper introduces a robust approach combining LLM assistance with SysML v2 for semantic alignment in MBSE, demonstrating feasibility through structured processes and semantic extension libraries. As LLMs continue to evolve, the proposed system prompt utilization and iterative refinement suggest promising pathways for enhancing collaborative model alignments and traceability in engineering applications. Future endeavors should focus on deepening semantic understanding, improving tooling integration, and scaling to larger industrial scenarios.