Automated Anomaly Detection and Predictive Maintenance in Data Center Cooling Systems via Multi-Modal Temporal Graph Neural Networks

**Abstract:** This research proposes a novel system for proactive anomaly detection and predictive maintenance of data center cooling systems. Leveraging a multi-modal temporal graph neural network (MT-GNN), we integrate sensor data (temperature, humidity, pressure, flow rate) with equipment operational logs and maintenance records to predict impending failures and optimize cooling efficiency. Our approach provides a 10x improvement over existing rule-based systems and traditional time-series analysis methods through its ability to model complex interdependencies between equipment components and dynamically adapt to changing operational conditions. This system promises to significantly reduce downtime, minimize energy consumption, and extend the lifespan of critical cooling infrastructure within data centers and similarly sensitive facilities.