Full-field state prediction of high arch dam structure with enhanced fidelity based on digital twin and sensor perception optimization

High arch dams are complex structures subjected to strong nonlinear behavior and extreme loading conditions, where achieving high-accuracy prediction of real-time full-field structural physical states remains challenging due to sparse monitoring data and insufficient guidance for sensor placement. To address this issue, this study proposes a sensor perception optimization-driven digital twin framework that integrates numerical simulation with deep transfer learning. To improve sensor layout design, modal observability is adopted to evaluate the sensitivity of monitoring points to structural responses, while the condition number is used to assess reconstruction stability. In addition, a hybrid optimization strategy combining orthogonal triangular decomposition for candidate sensor screening and greedy iterative selection for layout refinement is developed to integrate heterogeneous monitoring technologies. A case study of a prototype high arch dam under normal, extreme, and seismic conditions demonstrates that the proposed method significantly improves the reconstruction accuracy of deformation and stress fields, enhances sensing robustness, and effectively captures nonlinear and damage-evolving structural behaviors. The proposed framework shows strong potential for intelligent monitoring and safety assessment of high arch dams.