Deep Learning-Driven Distributed Acoustic Sensing for Real-Time Flow Dynamics Monitoring in Large-Diameter Aqueducts

Large-diameter gravity aqueducts play a crucial role in urban water supply, industrial transport, and irrigation, yet they are often subjected to complex flow conditions that can compromise both efficiency and structural integrity. Accurate and real-time flow monitoring is essential for optimizing hydraulic performance and ensuring infrastructure safety. However, conventional techniques like ultrasonic sensing are limited in providing continuous, real-time data on flow states. This study introduces the use of Distributed Acoustic Sensing (DAS) technology for monitoring flow dynamics along a 6 km segment of the 113.1 km Pearl River Delta Water Resources Allocation Project. To handle the large volumes of DAS data, we developed DAS-Water CNN, a convolutional neural network designed to interpret low-frequency acoustic signals and classify different water flow states and velocities. This method enables distributed, real-time monitoring, significantly improving the intelligence and efficiency of urban water management. Our findings demonstrate that DAS, combined with advanced AI algorithms, accurately identifies flow patterns, locations, and velocities, leading to enhanced operational efficiency, reduced maintenance costs, and valuable data support for the advancement of smart water supply systems.