SysV2-Repair: Towards LLM-Based repair of SysML®v2 models

Abstract

The adoption of SysML® v2 in Model-Based Systems Engineering (MBSE) creates a growing need for automated support to detect and repair errors in textual system models. While recent work has primarily focused on model generation and error detection, the automated repair of SysML® v2 models remains largely unexplored. In this work, we will propose SysV2-Repair, a localized LLM-based repair framework that combines compiler-guided syntactic validation, semantic consistency checking, Retrieval-Augmented Generation (RAG), and targeted patch generation. Rather than regenerating complete models, the framework will constrain repairs to the localized regions identified by the validation stages. We will evaluate the proposed approach on a benchmark of 83 SysML® v2 models with injected syntactic and semantic errors and compare it against non-localized repair strategies. This work aims to advance reliable and fully automated repair of SysML® v2 models for AI-assisted MBSE.