LAISA: Large-language Agent for Industrial State Assessment

Industrial state assessment is essential for manufac-turing, energy production, and infrastructure monitoring. Deep learning approaches face challenges including high data labeling costs and limited model generalization. This paper presents LAISA (Large-language Agent for Industrial State Assessment), a novel approach leveraging Large Language Models (LLMs) as intelligent agents for industrial system state assessment. LAISA divides assessment into individual variable analysis and system-level eval-uation. The LLM agent assesses each variable based on contextual information, then evaluates overall system status by analyzing dependencies between variables. Through direct embedding of industrial domain knowledge in specialized prompt templates, LAISA enables zero-shot assessment without requiring additional training or hardware upgrades. Our experiments on industrial datasets demonstrate that LAISA achieves significant accuracy under zero-shot, unsupervised conditions, with GPT - 3.5- Tu rbo outperforming LLaMa- 3: 708 across all evaluation metrics. This work introduces a practical software-defined intelligence solution for industrial applications and extends the application domain of LLM agents beyond traditional natural language processing tasks.