In today’s development environment, characterized by an increasing number of requirements and growing pressure to accelerate innovation cycles within Industry 4.0, the effective contextualization of heterogeneous information and the interoperable connection of data silos is becoming a key challenge. Conventional product lifecycle management (PLM) systems often prove to be inflexible and organized in rigid structures. This leads to fragmentation of information and semantic divergence between different data silos. The use of generative artificial intelligence (AI) represents a promising approach to improve information processing and contextualization with help of ontologies because it can create new content, interpret existing data, and manage data objects flexibly. However, generative AI only achieves reliable results if tasks are clearly defined and precisely tailored to the capabilities of the large language model (LLM) in use. Therefore, for complex processing tasks, multi-agent systems (MAS) are ideal because they allow different agents to act autonomously and break down tasks into smaller, manageable units. This article presents a methodology called LLM-based MAS for product development (LaMAS4PD) enabling the systematic ontology-driven structuring of heterogeneous engineering information with help of LLMs. To contextualize datapoints the methodology describes a procedure for the systematic design of an organizational structure with roles for an agent organization. The use of organized MAS helps to systematically integrate data silos and operationalize AI technologies in context of product development. Therefore, specialized LLM-based agents are used in the architecture for information extraction, semantic search, and synthesis to automate the conversion of technical documentation into a unified, ontology-based knowledge graph. By means of Proof of Concept (PoC) implementation the methodology is applied and afterwards evaluated using domain-relevant benchmarks. The results demonstrate the potential of the systematic use of agent-based LLM systems to reduce manual effort, improve data quality, and seamlessly contextualize and connect engineering knowledge in Industry 4.0 environments.
Recker et al. (Thu,) studied this question.