Smart IoT Leak Forecasting for Energy-Efficient Liquid-Cooled AI Data Centers

Krishna Chaitanya Sunkara and Rambabu Konakanchi present a proof-of-concept smart IoT monitoring system for probabilistic leak forecasting and instant detection in liquid-cooled AI data centers.

• Main announcement/action: The paper demonstrates a prototype combining LSTM neural networks for probabilistic leak forecasting and Random Forest classifiers for instant detection, reporting 96.5% detection accuracy and 87% forecasting accuracy at 90% probability within ±30-minute windows; the system forecasts leaks 2–4 hours ahead and detects sudden events within 1 minute. The implementation stack uses MQTT streaming, InfluxDB storage, and Streamlit dashboards, and the authors estimate prevention of ~1,500 kWh annual energy waste for a 47-rack facility based on synthetic experiments aligned with ASHRAE 2021 standards.

• Background and other details: Results are based on synthetic-only validation (no operational deployment yet); manuscript is in IEEE conference format (7 pages, 6 figures), submitted to arXiv on 25 Dec 2025 (arXiv:2512.21801) with DOI via DataCite pending; the paper lists authors and provides full-text PDF, HTML experimental view, and TeX source for reproducibility.