Rapid dust concentration prediction using sparse sensors during tunnel construction

The drilling–blasting method is a widely used and efficient excavation technique in tunnel construction. However, it generates significant dust pollution, rendering comprehensively and rapidly assessing the distribution of dust concentration during tunnel construction crucial. Existing sensor-based dust-monitoring technologies provide only sparse dust data, while computational fluid dynamics (CFD) simulations are computationally intensive and time-consuming. This study introduces a data-driven approach that integrates principal component analysis (PCA) and Gaussian process regression (GPR) to decompose and reconstruct dust concentration fields, overcoming the limitations of conventional techniques. The approach reconstructs the global dust concentration field on a two-dimensional tunnel plane using sparse sensor measurements, eliminating the need for repeated, time-consuming numerical simulations. Key factors influencing reconstruction accuracy were investigated. The results demonstrate that appropriately increasing the number of PCA modes and sensors improves accuracy, while optimizing sensor placement significantly enhances performance. Specifically, a high-precision reconstruction of the dust concentration field was achieved with an optimized arrangement of five sensors. Additionally, a comparison of different stages in the dust migration process revealed that the method achieved the highest reconstruction accuracy during the steady diffusion stage, highlighting the significant influence of airflow on reconstruction accuracy. Finally, the method was applied to predict dust concentration in a tunnel under construction. The relative error was 19.79%, while the computation time was merely 0.28% of that required for CFD calculations, demonstrating its efficiency and reliability.