Risk Analysis in Tunnel Construction with Bayesian Networks Using Mutual Information for Safety Policy Decisions

Tunnel construction is affected from its origins by different types of uncertainties responsible for innumerable safety risks. This problem has been addressed constantly during the last times achieving positive results, but the complex work scenarios and the common variability of the construction processes prevent putting an end to this problem. For this reason, this study presents an alternative methodology for safety prioritization in tunnel construction gaining relevant information hitherto unknown which can be crucial for policy making in infrastructure projects. The method proposed consists on the Bayesian analysis of data from occupational accidents recorded during the construction of tunnels in the last years. For this purpose, the model variables are rigorously estimated from expert judgement supported by the analysis of data from previous projects. Once the bayesian model is built, the dependencies among the variables are examined using the mutual information. The results obtained from the mutual information analysis allow to detect the main risks responsible for the occurrence of accidents and how they interact. Afterwards, a simplified Bayesian model with the most relevant risk factors affecting safety is built. Through the bayesian inference process, this condensed and validated model facilitates the exploration of significant contributions for safety policy decisions in tunnel construction. Overall, the results obtained provide a deep insight about the most influential factors on which should be focus the efforts to reduce accidents. Several safety risk factors are further influenced by human and organizational factors, whose effect can be reduced in advance. The mechanism of risk migration was better understood when analysing the interaction between the variables in the Bayesian model. In general, the accurate simplification of the model network demonstrated to be a powerful tool to comprehend the uncertainty associated to complex problems.