Research on Recurrent Neural Network Based Crack Opening Prediction of Concrete Dam

Abstract

The concrete dam can prevent flooding events and generate a vast amount of electricity, and it is a critical national infrastructure. However, it is easy to get cracked, and cracks usually pose significant potential threats to the safety of the concrete dam. Many researchers have done much research on dam crack protection and explored various rules to protect the concrete dam from cracks. However, the complex and irregular distribution of cracks make this task a very challenging research issue. In this paper, the feature importance of crack influencing factors is firstly analyzed. Then, the Recurrent Neural Network (RNN) is introduced for dam crack modeling. Next, the crack width of the Longyangxia Dam is modeled and tested by using the Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM). Finally, experimental results show that our proposed RNN-based method can effectively predict the crack change of the concrete dam.