Dynamic Bayesian network model for the safety risk evaluation of a diversion tunnel structure

A diversion tunnel is an important part of a water conservancy project. Many factors influence the safety of a diversion tunnel structure, and the risk situation of these factors changes with time during the operating period. Analysis and evaluation of the safety of a diversion tunnel structure are important for ensuring its normal operation. However, the influence factors are complex, and the detection and evaluation of structural safety remain challenging.

In this paper, a dynamic Bayesian network model for the safety evaluation of a diversion tunnel structure was established. First, a three-level influencing index system of tunnel structure safety was determined through literature research and expert consultation, combined with the world's current tunnel safety standards. The index system included 7 aspects and 26 specific indices, such as crack length, crack width, and pH value. The risk situation of each index was divided into five levels (from A to E), with each level corresponding to a specific risk probability and risk value, aiming to quantify the risk of the diversion tunnel structure. Second, index weights were assigned through expert consultation, and the conditional probability was determined based on the fuzzy analytic hierarchy process. Finally, the prior probability was obtained through the inspection results of intelligent robots, and the transfer probability was determined according to the exponential distribution hypothesis of tunnel life. The time slice interval was set as 1 year, and the safety situation and future development trend of the diversion tunnel structural risk were calculated. In addition, by setting the overall risk level of the tunnel structure, the most likely risk probability distribution of each index was obtained through backpropagation.

The model was applied to the structural safety evaluation of the diversion tunnel of a hydropower station in China, and the assessment results showed that: (1) According to forward inference, the overall risk value of the diversion tunnel was 0.230, which was very low, but lining cracks and lining spalling were structural safety problems that need attention. The evaluation results of the model were consistent with the engineering judgment. (2) The prediction of the development trend of structural risk indicated that this risk increased to 0.800 after approximately 40 years, requiring remedial action. (3) The backpropagation of risk revealed that different safety indices should receive attention in different safety periods of diversion tunnel operation. The risk influencing the degree of the lining spalling and operating environment risk was higher when the diversion tunnel was in a relatively safe state, but when the diversion tunnel was in a relatively dangerous state, the lining deformation, lining crack, and material deterioration were the main risk factors.

The proposed dynamic Bayesian network model performs with good accuracy and practicability for the risk assessment of a diversion tunnel structure. Furthermore, the model can predict the development trend of the structural risk and identify the key influencing index, which is important for diversion tunnel operation and maintenance.