AI-based Requirement Processing to Support Validation of Future Railway Mobile Communication

The Future Railway Mobile Communication System (FRMCS) represents the next-generation communication standard for European railways, aiming at ensuring interoperability, safety, and digital resilience across onboard and trackside systems. This research presents the FRMCS Requirement Processing System, an AI-based framework that has been developed to automate the extraction and structuring of technical requirements from complex and multi-source specifications. The first step is the faithful extraction of the requirement along with its important information, which includes the type of requirement and the requirement identifier. Subsequently, each requirement is processed and classified using a graph-based approach that takes contextual elements into account. Using the RAG-Anything framework, a comprehensive knowledge graph is built from multimodal document inputs (i.e., text, tables, and figures). The resulting architecture of the framework supports hybrid query modes that combine vector similarity search and graph traversal, enabling the model to retrieve contextual information across specification layers before requirement classification. Output data are generated as structured files containing FRMCS component and function metadata. System evaluation was conducted with human supervision of the output of the framework to ensure a reliable comparison of requirement identification and metadata extraction results. The framework was developed after an initial exploratory phase based on a human-in-the-loop extraction approach utilizing ChatGPT, a conversational chatbot based on a large language model (LLM). This work highlights a novel synergy between AI, prompt engineering, and systems validation, offering a replicable blueprint for the development of future standards such as FRMCS. The approach proved very useful in industrial practice in order to support the design of specific labs for FRMCS validation.

Keywords: Software engineering, design, train control systems, verification and validation, testing, generative AI..