Information Extraction and Knowledge Modeling for Disassembly Processes

Disassembly is an important process step of sustainable manufacturing and circular economy strategies. As the cost of disassembly can rise quickly due to its complexity, disassembly planning is necessary to control these costs. An efficient disassembly planning pipeline starts with an efficient knowledge model. However, current disassembly knowledge models often suffer from rigid data structures, insufficient specification of required information, and fragmented data sources. To address these limitations, we propose a five-step concept that combines the analysis of structured information with the analysis of unstructured text-based information by the use of Large Language Models (LLMs). The foundation of this concept is a basic schema for a knowledge graph to accurately represent product structures. This graph is enriched by extracting and interpreting additional process-specific data from a maintenance manual using LLMs. A proposed interactive LLM-based interface allows users to dynamically query, validate, and extend the resulting knowledge graph in real-time. Preliminary implementations demonstrate the feasibility of this approach, bridging the gap between static state-of-the-art disassembly planning models and abstract ontological frameworks.