Digital twin model for analyzing deformation and seepage in high earth-rock dams

Abstract

Digital twin technology is vital in hydraulic engineering for real-time visualization, performance analysis, and risk management of water infrastructure systems. This paper proposes a digital twin model for deformation and seepage analysis of high earth-rock dams, integrating deep learning with the Finite Element Method (FEM). Key contributions include a sensor-based monitoring point model with time-variant update and extrapolation capabilities and a point-to-domain model that achieves full-domain monitoring predictions from point-level monitoring by learning the node relationships generated by FEM using neural networks and dynamic monitoring loss functions. Applied to a 186-m dam, the model achieves an average error of 3.17 %, improving deformation prediction accuracy by 19.44 % and simulation accuracy by 64.42 %. This approach facilitates real-time monitoring, predictive analysis, and early warnings, making it a powerful tool for hydraulic engineering safety. Future work will focus on exploring three-dimensional high-precision modeling and advancing data fusion techniques.