Knowledge-aware cell formation in matrix-structured manufacturing systems via large and small model synergistic methods

Matrix-structured manufacturing systems (MMS) represent a promising paradigm for achieving resilient and human-centric production. Yet, their cell formation method (CFM) under explicit operational constraints remains underexplored. Current knowledge-aware clustering methods focus mainly on semantic similarity, overlooking practical constraints and lacking closed-loop integration between extracted knowledge and CFM decisions, thereby limiting MMS implementation in practice. Therefore, this study proposes a knowledge-aware framework for CFM in MMS, tailored to the semantic and physical constraints under complex manufacturing environment. First, a process knowledge graph is constructed through hybrid extraction that combines lightweight taggers with a locally deployed large language model (LLM), enabling low-cost parsing of heterogeneous, multilingual, symbol-dense process documents. Second, the knowledge-aware process clustering method generates compact process clusters under resource requirements and human–machine collaboration (HMC) constraints, serving as the basis for cell division. Finally, an enhanced NSGA-II algorithm optimizes the physical MMS cell layout, improving cell utilization while reducing logistics costs. Case studies indicate that the hybrid extraction pipeline approaches frontier LLM extraction quality at lower cost. Building on this knowledge, the proposed CFM architecture delivers substantial gains in cell utilization and reductions in logistics costs, thereby providing a cost-efficient pathway for practical MMS implementation in complex manufacturing environments.