Hybrid BN Approach to Analyzing Risk in Tunnel-Induced Bridge Damage
Yan Zhang,Limao Zhang,Xianguo Wu
DOI: https://doi.org/10.1061/(asce)cf.1943-5509.0001310
2019-01-01
Journal of Performance of Constructed Facilities
Abstract:Tunnels are generally built under trunk roads (where traffic conditions are complex and heavy) in order to share the load of urban road systems. The construction of tunnels has significant effects on the surrounding environment, especially in connection with nearby bridges, since it may cause soil loosening, ground deformation, and bridge pile settlement. Traditional methods, such as fuzzy set, fault tree analysis (FTA), artificial neural network (ANN), are inefficient and inaccurate when it comes to analyzing risk magnitudes or probabilities in a dynamic environment. To fill this gap, this paper, with the aim of predicting, monitoring, and diagnosing risk factors for tunnel-induced damage, provides a new framework that integrates the advantages of rough set (RS), cloud model (CM), and Bayesian network (BN). A RS approach is used to handle the vagueness and uncertainty and reduce redundancies for improving the efficiency of the evaluation process, CM is employed to discretize continuous attributes without changing their distinguishing characteristics, and BN is utilized to train mess data with a reasonable inference mechanism and update the risk magnitude with given evidence. To model the risk network of the tunnel-induced bridge damage, 14 critical influential factors are identified. A real tunnel project that crosses an overpass of expressways in China is used as a case to prove the applicability and feasibility of the proposed approach. Results verify that the proposed hybrid BN approach can be used to predict, supervise, and diagnose the risk before, during, and after tunnel construction. This research contributes to (a) the state of knowledge by providing a novel risk analysis approach that is capable of handling fuzziness, uncertainty, and dynamics in factor characterization; and (b) the state of the practice by providing insights into a better understanding of how to predict, control, and diagnose risks under given observations.