Risk of human fatality in building fires: A decision tool using Bayesian networks
Introduction
The complexity of our society is continuously increasing. Advanced technology allows the accommodation of a large population with increasing demands on goods and mobility in the very small space that the Netherlands provide. Therefore, the available space is used at maximum. As an example, preparations are made to build a roof over several kilometres of a 10 lane highway – that carries dangerous goods – and to build offices and may be even houses on top of it. But this intense use of space does not go without a price. There is an increased potential for an accident to become a large-scale disaster. For example, an explosion of a track carrying dangerous goods on the highway mentioned above may end in a large number of causalities among the people living or working in the buildings on top of the highway. Although such an accident remains a rare event, its consequences can reach a large extent. Therefore, the authorities would like to know the consequences of such an accident and to prepare for intervention, before the accident happens.
The people in charge of taking decisions in the design phase of such a complex project need a tool that helps them to choose among the alternative designs that one which ensures with a certain probability the smallest damage. Given the fact that solutions for the large demand on space are innovative designs, the outcome of a possible accident and in particular a fire in such a building cannot be estimated based on past experience and statistical data. Moreover, prescriptive codes cannot be applied to these innovative designs. Therefore, new methods to test the level of safety of people inside buildings are needed. These methods should take into consideration all uncertain conditions in which a future possible fire could take place and, therefore, should be based on computer simulations.
There is a large range of models that simulate the evacuation of a building, from simple models that simulate only the movement of people within the building, to very complex models that attempt to incorporate human behaviour [1], [2]. They are used in order to decide on the structure of the building, the position of the exits, the size of the doors, corridors, and staircases. They can estimate the time needed to evacuate the whole building or only some parts of the building. They help also to find bottlenecks of a building regarding evacuation (where people may be trapped, where queues can be formed, etc.). However, they cannot consider the conditions outside buildings, for example, how neighbourhood of buildings, weather conditions, or intervention of fire fighting services can influence evacuation, and, implicitly, the outcome of a fire in terms of number of deaths. As an example, one of the tunnels in the High Speed Railway Line is designed for quick evacuation of passengers from the tunnel but further investigations, with the tunnel already finished, show that there is not enough space at ground level to accommodate the fleeing crowd, let alone vehicles and equipment of emergency services. This example shows the need to include in models not only people evacuation, but also fire development and rescue services’ actions, taking into account the characteristics of people and structure, location, and the external factors. For this particular case, if all these factors are included into the Bayesian belief net (BBN) model, one may set values for the available safe place and may obtain that the number of people at risk is high, or that the probability to have a high-consequence fire is high.
The goal of the model presented in this paper is to put together not only people and their behaviour during evacuation, but also fire fighters’ actions, structure of the building, and characteristics of the building and the environment, in an overall model. Model results are more useful in a comparative sense rather than in an absolute sense. Using this tool, more alternatives can be compared with each other, with the actual level of safety, the desired level of safety, or with existing codes and procedures. The model could be used to analyze the “what-if” scenarios, as well as the low probability–high-consequence scenarios.
The model proposed in this paper is based on the Bayesian belief net approach, a probabilistic method that can accommodate the complexity of the system under analysis. In Section 2 of the paper, the approach chosen to reach the goal of the work is presented. This section is a short summary of general characteristics, advantages and disadvantages of this method, comparisons with methods used before, and a short presentation of attempts to use the BBN approach in the field of fire safety. The model for percentage of deaths in a fire is presented in Section 3. Three phases of building up the BBN model for fire safety are presented here. First, the process of building the graphical structure of the network and of quantifying it is succinctly described. The last part of the section presents some example of analysis and results that can be obtained using this method in order to give an idea about applicability of the BBN approach for the estimation of probability distribution of percentage of deaths in case of fire. The last section of the paper presents the conclusions and gives directions about the future work to be done.
Section snippets
Method
In recent years, fire regulations in the Netherlands have tended to change from prescriptive codes to performance-based regulations. This change of principles makes possible more flexible and innovative designs and cost-effective structures. However, it also increases the number of studies that involve risk analysis by demand of authorities, who want to be assured that the solutions chosen are acceptable. This section of the paper presents three methods that are used in risk analysis in general
Model for the percentage of deaths in building fires
This section presents the three steps that have been followed through building the model for human damage produced by fire in a public building and the type of results that can be produced using the BBN method. Given the large structure of the resulting network and the goal of the paper to introduce the application of the BBN method in the field of fire safety, as well as to show the capabilities and usefulness of the method in this field, a more reduced structure of the network is used as an
Conclusions
In current society with a rapid rate of development, with new and modern architecture that involves more and more people, it seems to be very clear that there is need for a comprehensive model that can be used to estimate the extent of a possible fire in a building. The model should take into consideration all the factors involved in a complex system such as fire in a building, and especially interactions between them.
The use of BBNs in general and in particular in the field of fire safety is
Acknowledgements
The authors would like to thank Anca Hanea and Roger Cooke from the Department of Applied Mathematics (DIAM), Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, The Netherlands, for their valuable comments on earlier drafts of this paper.
References (34)
Risk analysis and fire safety engineering
Fire Safety Journal
(1998)Bayesian model-based diagnosis
International Journal of Approximate Reasoning
(2001)- et al.
Making decisions: using Bayesian nets and MCDA
Knowledge-Based Systems
(2001) - et al.
Eliciting conditional and unconditional rank correlations from conditional probabilities
Reliability Engineering and System Safety
(2008) - et al.
A review of the methodologies used in the computer simulation of evacuation from the built environment
Building and Environment
(1999) - E. Kuligowski, The evaluation of performance-based design process for hotel building: the comparison of two egress...
- et al.
An updated international survey of computer models for fire and smoke
Journal of Fire Protection Engineering
(2003) - et al.
Fault trees and sequence dependencies
Fire safety management and risk assessment
Facilities
(1999)- et al.
Fire safety design based on calculations: uncertainty analysis and safety verification
Fire Safety Journal
(1996)
Construction and local computation aspects of network belief functions (with discussion)
Bayesian belief networks: odds and ends
The Computer Journal
An Introduction to Bayesian Networks
Bayesian Networks and Decision Graphs
Experts in Uncertainty
Cited by (69)
Performance investigation of fire protection and intervention strategies: Artificial neural network-based assessment framework
2021, Journal of Building EngineeringOptimal firefighting to prevent domino effects: Methodologies based on dynamic influence diagram and mathematical programming
2021, Reliability Engineering and System SafetyA combined physics of failure and Bayesian network reliability analysis method for complex electronic systems
2021, Process Safety and Environmental ProtectionUncertainty-based decision-making in fire safety: Analyzing the alternatives
2020, Journal of Loss Prevention in the Process IndustriesCitation Excerpt :BN are based in Bayes theorem, using known probabilities of basic elements to estimate the conditional probabilities of dependent elements. This is a powerful use of known probabilities to model unknown probabilities in complex systems such as estimating the dynamic operational risk assessment in a chemical process setting (Barua et al., 2016) or estimating the risk of human fatality in building fires (Hanea and Ale, 2009). As this method is based on conditional probability theory, its weaknesses are similar.
Modeling risks in dependent systems: A Copula-Bayesian approach
2019, Reliability Engineering and System SafetyCitation Excerpt :However, it is not suitable for the complex system due to its strict operation conditions: (1) The database should be complete; (2) Nodes could not have too many possible states; and (3) Child nodes could only own fewer parent nodes. If these conditions are not satisfied, a number of probabilities assigned to child nodes will experience exponentially increase with the parent nodes number, resulting in exponential execution times and complexity [47,48]. Meanwhile, it is also impossible to populate large conditional probability tables (CPT) and satisfy child marginal constraints.
Systematic review of human and organizational risks for probabilistic risk analysis in high-rise buildings
2019, Reliability Engineering and System Safety