Multi-Agent Reinforcement Learning for Real-Time Contamination Source Localization in Cleanrooms

**Abstract:** This paper introduces a novel approach to real-time contamination source localization in cleanrooms employing a multi-agent reinforcement learning (MARL) framework. Traditional methods rely on manual inspection or static sensor networks, lacking responsiveness to dynamic contamination patterns. Our system utilizes a swarm of mobile air quality sensors, each equipped with a reinforcement learning agent, to autonomously explore the cleanroom and dynamically learn optimal search strategies for pinpointing contamination sources. The framework combines computationally efficient particle filtering with a novel reward function that prioritizes source convergence and minimal disturbance to cleanroom operations. Results demonstrate a 37% reduction in source localization time compared to existing methods and a 92% accuracy in identifying contamination points. This system offers a scalable and adaptive solution for enhancing cleanroom hygiene and minimizing production losses.