GPT-Manual Generator : A knowledge transformation architecture for maintenance procedure externalisation

This paper introduces GPT-Manual Generator, a novel knowledge transformation architecture that addresses the critical challenge of externalising tacit maintenance expertise into structured documentation. Traditional methods of creating maintenance manuals are resource-intensive, inconsistent, and often fail to capture crucial procedural details. Our system transforms video and audio recordings of maintenance tasks into structured, step-by-step manuals through six key technical contributions: temporally aware processing, automated content enhancement, multi-stage verification, adaptive timestamp-image matching, knowledge standardisation, and automated documentation generation. Through a comparative evaluation against four state-of-the-art language models (GPT-4, GPT-4o, GPT-4.1, and GPT-4.5), we demonstrate the system's superior performance across multiple metrics, including keyword F1 score (74% average versus 41-51% for baseline models), precision, recall, and semantic similarity. Results validate that specialised architectures for maintenance knowledge transformation offer significant advantages over general-purpose models. The GPT-Manual Generator establishes a new technical foundation for multimodal knowledge transformation that bridges the gap between tacit expertise and explicit documentation, addressing critical challenges in maintenance knowledge management while reducing resource requirements and improving documentation accuracy.