Evaluation of collapse possibility of deep foundation pits in metro stations based on multi-state fuzzy Bayesian networks

Many construction risk factors and frequent collapse accidents are associated with deep foundation pits in metro stations, and there are limitations for the traditional methods to conduct risk analysis of such complex systems with multiple states. In this study, a method for evaluation of collapse possibility of deep foundation pit collapse based on multi-state fuzzy Bayesian network is proposed. The multi-state Bayesian network model was constructed via fault tree transformation, and fuzzy numbers were used to describe the fault state and the failure rate of root nodes, which overcomes the problem that the traditional methods cannot consider the influence of intermediate fault states and are difficult to obtain the accurate failure rate. Based on the forward reasoning of Bayesian network, the risk probability of foundation pit collapse can be calculated in two different ways including the fuzzy probability of root nodes and the actual fault state in construction. As a result, a real-time dynamic risk analysis during foundation pit construction can be achieved. Furthermore, the key risk factors can be identified for the guidance of risk control according to the sensitivity analysis results. In addition, the posterior probability of each root node can be obtained by backward reasoning to carry out fault diagnosis and further predict the system state. Two case studies show that the proposed method can scientifically and reasonably evaluate the collapse risk of foundation pit and determine the key risk factors, which can be used as a decision-making tool for safety risk management of foundation pit construction.