Integrated indicator graph network for industrial process state evaluation

Abstract

The evaluation of the operational status of industrial processes is challenging due to multiple, often conflicting, performance dimensions, such as product quality, energy consumption, and material loss. A significant limitation of existing methods is the failure to capture that the interrelationships between process variables can differ substantially for each distinct performance indicator, or the reliance on a single, fixed process topology. This paper proposes a novel multi‐graph fusion framework for comprehensive status evaluation. First, this paper constructs a multiple indicator‐related quality‐graph‐network, each designed to capture the unique structural information relevant to a single indicator. Second, to integrate this differentiated information, this paper introduces a hybrid fusion method. This method intelligently revises node features by combining supervised revision (based on model reliability analysis) and unsupervised revision (based on node importance analysis), generating a comprehensive graph representation. Finally, a comprehensive evaluation model is built upon the fused graph features to assess the overall operational status. The effectiveness and feasibility of the proposed method are verified through application to the Tennessee Eastman benchmark process.