A Bayesian Method for Dam Failure Risk Analysis using Causal Loop Diagrams and Word Frequency Analysis

Abstract

Earthen dams operate in complex environments where their safety is often affected by multiple uncertain risks. A Bayesian network (BN) is often used to analyze the dam failure risk, which is an effective tool for this issue as its excellent ability in representing uncertainty and reasoning. The validity of the BN model is strongly dependent on the quality of the sample data, making convincing modeling rationale a challenge, which limits its use. There has been a lack of systematic analysis of the dam failure data of China, which further leads to a lack of in-depth exploration of potential associations between risk factors. In this paper, we established a comprehensive database containing various dam failure cases in China. Herein, historical dam failure statistics are used to develop BN models for risk analysis of earthen dams in China. In order to unleash the value of the historical data, we established a Bayesian network through the Causal Loop Diagrams (CLD) based on the nonlinear causal analysis. We determined the conditional probabilities using Word Frequency Analysis (WFA). By comparing with the Bayesian learning results, the modeling method of BN proposed in our study has apparent advantages. According to the BN model established in this paper, the probabilities of dam failure with three damage modes of seepage damage, overtopping and structural instability are 22.1%, 58.1%, and 7.9%, respectively. In addition, we demonstrated how to perform the inference process of the dam failure path. This will provide helpful information for dam safety practitioners in their decision-making process.