Stability-Guided Graph Neural Clustering for Industrial-Scale Manufacturing Systems

Abstract

Industrial manufacturing systems increasingly rely on graph-structured data to represent complex interactions among machines, processes, and production units. Identifying meaningful community structures in such graphs is critical for system monitoring, optimization, and fault isolation. However, practical industrial environments impose strict constraints, including limited computational resources, communication bottlenecks, and non-ideal operating conditions, which challenge existing graph neural network (GNN)–based clustering methods. In this work, we propose a resource-aware graph neural clustering framework tailored for industrial manufacturing networks. The proposed approach explicitly incorporates computational and structural constraints into the clustering process, enabling efficient and stable community identification under realistic deployment settings. By integrating constraint-aware representation learning with clustering objectives, the framework balances clustering quality with resource efficiency, avoiding excessive model complexity and unstable assignments. Extensive experiments on industrial-scale and synthetic manufacturing network datasets demonstrate that the proposed method achieves competitive or superior clustering performance compared to existing GNN-based baselines, while significantly reducing computational overhead and improving robustness under constrained conditions. These results highlight the potential of resource-aware graph neural clustering as a practical and scalable solution for real-world industrial manufacturing systems.