Pollutants in tunnels actually fade away based on distance rather than time, a discovery that could cut ventilation energy use by 57 percent.
April 25, 2026
Original Paper
Physics-Driven AI for Robust Particulate Transport Prediction in Tunnels: Integrating Neural Denoising With Symbolic Regression
SSRN · 6637911
The Takeaway
Engineers have traditionally used an exponential model to guess how toxic dust and particles clear out of enclosed spaces. This new physics-driven AI revealed that the decay actually follows a much simpler inverse-distance law. By switching to this more accurate model, tunnel operators can run their massive fans far more efficiently without sacrificing air safety. This change resulted in a 57.8 percent reduction in energy consumption during testing. This finding proves that even our most basic assumptions about how air moves in tunnels have been significantly wasting power for years.
From the abstract
Real-time prediction of particulate concentrations in tunnel construction is essential for worker safety but remains challenging due to the high cost of CFD and severe sensor noise. This study proposes a physics-driven AI framework that integrates neural denoising with physics-guided symbolic regression to achieve robust and interpretable predictions. A teacher–student strategy is adopted, where a neural network reconstructs the latent physical manifold from noisy data, and symbolic regression i