Daniela Hanea

Analysis of the Schiphol Cell Complex fire using a Bayesian belief net based model

In the night of the 26 and 27 October 2005, a fire broke out in the K-Wing of the Schiphol Cell Complex near Amsterdam. Eleven of 43 occupants of this wing died due to smoke inhalation. The Dutch Safety Board analysed the fire and released a report 1 year later. This article presents how a probabilistic model based on Bayesian networks can be used to analyse such a fire. The paper emphasises the usefulness of the model for this analysis. In additional it discusses the applicability for prioritisation of the recommendations such as those posed by the investigation board for the improvements of fire safety in special buildings. The big advantage of the model is that it can be used not only for fire analyses after accidents, but also prior to the accident, for example in the design phase of the building, to estimate the outcome of a possible fire given different possible scenarios. This contribution shows that if such a model was used before the fire occurred the number of fatalities would have not come as a surprise, since the model predicts a larger percentage of people dying than happened in the real fire.

Quantitative and qualitative analysis of the expert and non-expert opinion in fire risk in buildings

Expert judgment procedure is a method very often used in the area of risk assessments of complex systems or processes to fill in quantitative data. Although it has been proved to be a very reliable source of information when no other data are available, the choice of experts is always questioned. When the available data are limited, the seed questions cover only partially the domains of expertise, which may cause problems. Expertise is assessed not covering the full object of study but only those topics for which seed questions can be formulated. The commonly used quantitative analysis of an expert judgment exercise is combined with a qualitative analysis. The latter adds more insights to the relation between the assessors field and statistical knowledge and their performance in an expert judgment. In addition the qualitative analysis identifies different types of seed questions. Three groups of assessors with different levels of statistical and domain knowledge are studied. The quantitative analysis shows no differences between field experts and non-experts and no differences between having advanced statistical knowledge or not. The qualitative analysis supports these findings. In addition it is found that especially technical questions are answered with larger intervals. Precaution is required when using seed questions for which the real value can be calculated, which was the case for one of the seed questions.

Estimating the Statistical Distribution of Human Damage Produced by a Fire in a Building Using Bayesian Belief Nets

The complexity of the cities’ layout and other public spaces, together with the large number of people involved leads to increased strain on the resources of emergency responders. An accident, such as a fire, remains a rare event so it is difficult for those in charge of preparing for an emergency and deciding on the acceptability of risk to get a picture of such an event. The interest of all emergency response agencies is to minimize the impact of disaster events on the entities of interest, which include first of all the human population. For this, there is need for a tool that helps the decision makers estimate the distribution of the fire outcome, given different information about the environment in which the fire takes place. This paper discusses the possibility of using continuous Bayesian belief nets for the study of the factors that influence the risk to which the people involved in a building fire are exposed, and how these factors influence the risk. The big advantage of Bayesian belief net approach is that it can model uncertain events. The distribution of the variables of interest can be easily updated given information about some of the other variables. Moreover, the intuitive visual representation of the problem at hand can help people to understand complex systems or processes, like a fire in a building. In this study, the approach is tested for a small example and the results are analyzed. The possibility of extending this method to a more complex model is discussed.Copyright