Data integration and workflow automation between AI-supported shoulder surgery customization medical software and manufacturing and distribution infrastructures

This thesis presents an ontology-driven, AI-enhanced Knowledge-Based System (KBS) designed to support knowledge integration, decision-making, and information interoperability within the orthopedic prosthesis manufacturing domain. Conducted in close collaboration with an orthopedic manufacturing company, the research addresses the challenge of managing heterogeneous, distributed, and evolving knowledge within highly regulated medical-industrial environments. After reviewing the state of the art in semantic technologies, Natural Language Processing (NLP), Large Language Models (LLMs), and multimodal systems, the thesis introduces a comprehensive domain analysis that informs the design of a custom ontology and an operational Knowledge Graph (KG). This semantic infrastructure enables computational continuity from pre-operative planning to prosthesis configuration, production, and distribution, and is validated through real-world industrial use cases. To reduce the cost and rigidity of manual knowledge engineering, the thesis proposes a semi-automated KBS population pipeline that leverages LLMs for structured information extraction from technical documentation, supported by expert validation. Further contributions include a hybrid reasoning framework combining embedding-based retrieval with ontology-guided inference, and a multimodal Retrieval-Augmented Generation (RAG) architecture that systematically evaluates how textual, visual, and structured data can be fused to improve retrieval effectiveness, providing broader empirical insights into multimodal fusion strategies for RAG systems. Overall, the thesis delivers an end-to-end framework for knowledge acquisition, representation and multimodal retrieval, advancing intelligent automation in orthopedic manufacturing. At the same time, it demonstrates how such technologies can be responsibly integrated into real industrial settings, highlighting the importance of structured knowledge, human-in-the-loop validation, and hybrid architectures for the reliable adoption of AI in sensitive domains.