Abstract: Damage and cracks may occur with dam during long operation period or when earthquake happens. It is difficult to diagnose internal damage by conventional methods, which would leave security hazards. A two-step damage detection method for large-scale hydraulic structures was introduced to solve this problem. Taking non-overflow section of Wudu reservoir as an example, combined with vibration parameter identification technique, research on damage identification of gravity dam based on RBF neural networks is carried out. Firstly, this paper verifies the validity of this method by theoretical research and numerical simulation. Then, the result of damage identification is checked by comparing with the dynamic characteristics obtained from the shaking table model test. The results indicate that the method is feasible to identify the position of damage and predict the extent of damage of gravity dam, but further study is still needed to solve practical problems.

Key words: neural networks, gravity dam, damage identification, model test