A LangGraph-Based Agentic DQN Framework for Green Predictive Maintenance in IIoT Edge Environments

The rapid evolution of the Industrial Internet of Things (IIoT) has significantly advanced predictive maintenance (PdM); however, deploying complex AI models on resource-constrained edge devices remains a major challenge due to high energy consumption. This research proposes a novel, energy-efficient agentic reinforcement learning framework designed specifically for industrial edge environments. Utilizing the LangGraph architecture, we developed a three-tier agentic system consisting of specialized Planner, Optimizer, and Executor agents. At the core is a Deep Q-Network (DQN) optimized for power-aware decision-making, evaluated on the AI4I 2020 Predictive Maintenance dataset. The framework achieves a simulated 19.6% reduction in estimated energy consumption while maintaining a robust F1-score of 0.84, bridging the gap between high-accuracy diagnostics and sustainable Green AI practices. An earlier version of this work was presented at the ISPEC Global Interdisciplinary Science & Research Congress (ISPEC-GISRC), held on April 27 - May 3, 2026, Rome, Italy, under the title "Energy-Efficient Agentic Reinforcement Learning Framework for Predictive Maintenance in Resource-Constrained Edge Environments".