Author(s): Kang Liu; Zhaowei Liu; Yongcan Chen; Haoran Wang
Keywords: Diversion tunnel; Safety evaluation; Analytic Hierarchy Process; Bayesian network
Abstract: The safety evaluation of diversion tunnel structure is the basic process during its operation. However, due to the multiple factors and imperfect engineering data, the present comprehensive evaluation methods mostly rely heavily on the judgment of expert experience. This paper develops a safety evaluation system of diversion tunnel structure based on dynamic update Bayesian network. Firstly, a three-layer index system with a five-level safety grade classification standard is proposed by referring to several worldwide safety standards of tunnels, which includes six aspects and thirteen specific indicators, such as crack length and width, PH, spalling diameter and so on. Secondly, the subjective evaluation model of diversion tunnel structure safety is developed. The relative importance of each indicator is judged by experts in a consultation meeting, and based on these judgement data, the weight of each index is caculated by using he mixed weight method of Analytic Hierarchy Process(AHP) and Entropy Weight Method. By using the subjective evaluation model, dozens of assumed scenarios are obtained to remedy the shortage of practical scenario data. Meanwhile, the practical scenarios are collected from public literatures and un-public engineering material. Finally, the evaluation system is initially developed after the Bayesian network is trained by using both assumed and practical scenario. The new-developed evaluation system is featured with its ability of being updated by the future engineering inspection and evaluation data. This model was used to evaluate the structure safety of Chinese monkey rock hydropower station diversion tunnel, and the evaluation results showed that the diversion tunnel overall risk was very low, but the material deterioration, lining crack, lining spalling need to be focused. Model assessment result is consistent with the engineering judgment, indicated that the model has good accuracy and practicability. This study provides a possible idea for engineering safety evaluation with the absence of measured data.