Design and Implementation of a Knowledge-Based Assistance System for Smart Failure Management in Manufacturing SMEs Using Large Language Models

Abstract

Manufacturing small and medium-sized enterprises (SMEs) are under increasing pressure to detect, analyse, and eliminate quality-related failures faster and more systematically, while operating with limited personnel, fragmented information structures, and heterogeneous IT environments. Although standards such as ISO 9001 require structured corrective action and documented organizational learning, practical failure management in many SMEs remains reactive, media-discontinuous, and weakly connected to reusable knowledge. This paper presents the design and prototypical implementation of a modular, knowledge-based assistance system that combines established quality engineering methods with natural language processing (NLP), machine learning, and large language models (LLMs) to support the entire problem-solving cycle. The research follows a design science research approach. First, requirements were derived from normative sources, the state of research, and an empirical industry survey with 104 valid company responses. The resulting requirements were structured into seven functional domains. On this basis, a modular reference architecture was developed that integrates structured failure capture, historical document analysis, method-guided problem solving, an explainable knowledge base, and feedback-driven learning loops. The prototype was implemented as a containerized full-stack web application using open-source technologies, including Flask, PostgreSQL, Docker, HTML/CSS/Bootstrap/JavaScript, and Llama 3- and Rasa-based conversational services. Transformer-based subcomponents for root-cause classification and guided 5-Why questioning complement the LLM-supported retrieval-augmented generation (RAG) assistant. The system was prototypically validated using historical failure documentation from manufacturing case studies. Evaluation results indicate improvements in knowledge accessibility, reduction in analysis time, increased consistency in root cause identification, and enhanced standardization of corrective action documentation. The findings suggest that AI-enhanced assistance systems can significantly strengthen organizational learning capabilities in SMEs, provided that they are embedded within structured quality management frameworks. The paper contributes to research in digital quality management by (1) providing a structured requirement-based reference architecture for AI-supported failure management systems, (2) proposing a systematic mapping between data mining methods and problem-solving phases, and (3) demonstrating a practical integration approach for LLMs in industrial quality environments. It bridges the gap between classical quality engineering and modern AI-based knowledge systems, offering a scalable pathway toward intelligent, learning-oriented failure management in manufacturing.

Keywords: Smart Failure Management, Knowledge-based Assistance System, Large Language Models (LLMs), Retrieval-augmented Generation (RAG)

DOI: 10.54941/ahfe1007784