Analysis of tunnel accidents by using Bayesian networks

Accidents in tunnels and other underground structures often lead to serious consequences, more serious in general than would have been the case in open air. Tunnels often have to pass through mountainous regions with a lack of redundant detour routes. Additionally to the serious direct consequences in the tunnel, large societal consequences are associated with the temporary closure of such life lines. For this reason careful consideration of risk and safety in tunnels is necessary.

Introduction

Accidents are normally caused by a combination of several factors. In the case of accidents in tunnels these factors could be related to characteristics of the tunnel, (e.g. the length of the tunnel, the illumination, the portal height, the curviness, the longitudinal gradient), they could be related to the characteristics of the traffic (e.g. the AADT, the HGV, the average speed) or these factors could be related to characteristics of the drivers and their vehicles involved.

Research project

In the present project a Bayesian Network is developed based on a database containing accident rates, tunnel characteristics and traffic characteristics from 126 Swiss road tunnels. The accident rates and the traffic characteristics are measured over 5 years. The causal relationships leading to accidents in tunnels are assessed directly based on this data and the conditional probabilities of the network are calibrated by using the so called EM learning algorithm. It is demonstrated how new information, in form of data or in form of expert opinion can be utilized to update the conditional probability tables in the network. The predictive ability of the network is compared with the predictions of a conventional regression model (with parameters fitted to the same database). The Bayesian Network is illustrated in the Figure.

The project is the basis for an ongoing research activity related to the development of a generic risk assessment methodology for road tunnels.

Bayesian network for modelling of accidents and consequences; depending on tunnel and traffic characteristics.

Prof.Dr. Michael H. Faber, This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Dr. Jochen Köhler, This e-mail address is being protected from spambots. You need JavaScript enabled to view it

Matthias Schubert, This e-mail address is being protected from spambots. You need JavaScript enabled to view it

ETH Zurich,

Department of Civil, Environmental and Geomatic Engineering, Institute of Structural Engineering

Chair of Risk and Safety, http://www.ibk.ethz.ch/fa/index_EN