LLM-Enhanced Intent-Aware for Proactive Decision Support Services in Industrial Activities

The growing complexity of industrial systems demands a transition from passive monitoring to proactive decision support, a shift that hinges on advanced intelligent perception. However, most systems remain confined to brittle, rule-based logic, which operates on fixed symptom-to-action mappings and thus cannot perceive the underlying, context-dependent operational intent. This perceptual gap is particularly detrimental especially in fault diagnosis, where this inability to adapt leads to frequent misdiagnoses and costly downtime. To overcome this limitation, this paper introduces the Intent-Aware Enhancement Framework (IAEF)1, which replaces static rules by actively creating and reasoning over a dynamic causal model. This is achieved through two core method. The Information Theory-guided Causal Graph Revision (ITCGR) algorithm provides the foundation by constructing a reliable causal model, uniquely leveraging information-theoretic metrics to guide an LLM for verifiable revisions on sparse data. Building on this model, the Multi-scale Adaptive Path Reasoning (MAPR) method then infers the true operational intent, employing a novel adaptive fusion model to robustly navigate complex inference chains. Experimental validation in a real-world case demonstrates the framework’s ability to accurately diagnose fault root causes under varying conditions, a task where traditional systems fail. The proposed approach significantly outperforms baselines, providing a foundational methodology for advancing industrial intelligence.