Tunnel collapse risk assessment based on multistate fuzzy Bayesian networks

Excessive structural deformation or collapse can lead to heavy casualties and substantial property loss. This paper presents a novel integrated risk assessment method based on multistate fuzzy Bayesian networks integrated with historical data, expert investigations, probability distribution calculations, discrepancy analysis, sensitivity analysis, and decision‐making. A new expert investigation is proposed with small probability intervals, expert weights, confidence index, etc. After gaining expert judgment by expert investigation, Chauvenet's criterion is first introduced in a discrepancy analysis to eliminate outlier data from the expert judgment and obtain a more reliable value. The t distribution and its confidence interval are also adopted to determine the characteristic value of the survey data as a triangular fuzzy number. A conditional probability table of the model is integrated with historical data and prior knowledge through the weight index. Sensitivity analysis is used to identify the critical factors by changing the probability distribution of each factor and observing the related changes in the risk event. The proposed method ensures the accuracy and scientific rigor of the assessment and the diagnosis of a tunnel accident. This method is successfully applied to assess the collapse probability of the Yu Liao Tunnel.