AUTONOMOUS FIRE RISK PREDICTION AND ADAPTIVE SUPPRESSION IN SERVER ROOMS USING AI AND DIGITAL TWINS

Abstract Server rooms contain mission-critical IT hardware with an exposed risk of fire hazard. Conventional suppression systems like FM200 or CO₂ rely on stationary NFPA standards with no dynamic adaptability to changing fire scenarios. This paper introduces an AI-driven autonomous system with IoT sensors, deep learning, and digital twins to transform server room fire protection. The system caters to three main tasks: prediction of fire spread, adaptive suppression, and verification. For simulation of fire spread, the system utilizes an LSTM-Transformer hybrid model that simulates fire spread five times faster than conventional Computational Fluid Dynamics (CFD) simulation. The adaptive suppression module utilizes reinforcement learning (RL) to fine-tune FM200 discharge rates to achieve maximum utilization of agents and minimize agent usage by 20% compared to NFPA 2001 recommended practices. Lastly, the system is tested in a NVIDIA Omniverse digital twin simulation environment and achieves 98% suppression performance against NFPA 2001 standards. These technologies' integration overcomes static suppression system weaknesses and computationally intensive CFD simulation, providing an efficient, scalable, and dynamic solution to server room fire safety.