Full Title - Summary of the Paper on Development of an Integrated Risk Assessment Method to Quantify the Life Safety Risk in Buildings in Case of Fire 

By Bart Van Weyenberge,^1,2 Xavier Deckers,^1,2 Robby Caspeele³ and Bart Merci^1
1Ghent University – UGent, Dept. Flow, Heat and Combustion Mechanics, Belgium
²Engineered Solutions, Ghent, Belgium
³Ghent University – UGent, Dept. Structural Engineering, Belgium

Introduction

This article is a summary of a recently published article in Fire Technology about the development of an integrated probabilistic risk assessment methodology to quantify the actual fire safety level for occupants in buildings with challenging designs.¹ The goal of the method is to objectify the safety level of both the prescriptive- and performance-based designs (PBD).

The method provides a solution for the restrictive scope of prescriptive-based codes,^2-3 on the one hand and the lack of transparency of input parameters and output safety factors in PBD methods on the other hand. To increase the possibility of innovation and reduce the ambiguity, PBD codes entail defining prescribed scenarios and parameters to be implemented in the analysis.^4-5 However, no unbiased method has been developed yet to include the effect of the uncertainty of these input parameters with respect to the obtained safety factor from ASET/RSET (available/required safe egress time) analyses.⁶ Therefore, the proposed quantitative risk assessment (QRA) method, consistently takes parameter uncertainties and the reliability of safety systems into account through probabilistic distributions that have direct impact on the obtained safety factor.

Even though some efforts have been made to develop quantitative risk analysis models, those models are only validated for simple cases and still require large amounts of computational power and probabilistic techniques that focus on single parts of the model (e.g., on smoke spread). The proposed methodology for this research enables analyzing complex building configurations in an automated way, using advanced simulation techniques to cope with the large amount of data involved in response predictions, which otherwise are not possible through the simplified method.

Framework

The main framework of the proposed fire risk analysis model consists of seven steps similar to the SFPE Engineering Guide to Performance-Based Fire Protection.^7-9 The method provides a tool to deal with the deterministic and probabilistic part of the framework.

Once the problem is clearly specified, from building configuration and boundary conditions to the goals and objectives, the performance criteria are developed accordingly. Next, the fire safety design is developed based on the main principles of best practices in fire safety design rules.⁸

The next step is conducting the deterministic and probabilistic analysis (illustrated by a case study). The results should be compared to pre-defined performance criteria. If the results meet the risk criteria, the final design can be accepted. If not, the fire safety design should be modified, by either changing the configuration or adding more safety measures, and the process should be repeated.

Case Study

The proposed model applies to a multi-purpose commercial building with a total surface area of 25,000 m². See Figure 1 for a 3D view.

The building is considered to be occupied by customers (90%) who are not familiar with the building and staff (10%) who are familiar with the building. The objective of the analysis is to analyze life safety in case of fire and the pre-defined risk criteria to quantify the risk level. Three types of risk-based performance criteria are defined, i.e., the probability of a fatality over a period of time, the individual and the societal risk criteria.

This case study analyzed three developed fire safety designs. In the first option, all the floors are considered as a single compartment and the design consists of a smoke and heat control (SHC) and sprinkler system. Every floor is divided into two SHC zones, with a fixed smokescreen in between. An ASET/RSET analysis evaluates the performance of the design.

The second option involves the design of a complete prescriptive solution, implementing a sprinkler system, with compartmented floor levels and every floor divided into compartments smaller than 2,500 m². The compartments are connected by three self-closing doors in case of fire.

In the third option applies the same fire safety design as option 1. However, no SHC system is implemented. This shows the effect of the SHC system.

Next is performing the probabilistic and deterministic analysis. First, the main representative design fire scenarios are defined based on floor levels, locations, compartment types, etc. Then the most-important input variables are chosen.¹⁰ Figure 2 presents the cumulative density distributions (CDFs) for four sensitive input parameters for smoke spread and for evacuation in Figure 3.

The CDFs are taken from literature, depending on the types of occupants, fire load, etc. The grey hatched areas are the domains analyzed in the probabilistic analysis. The figures show that most of the standards and guidelines use fixed parameters, which are mostly on the conservative side (right part of the figures) of the distributions.

The chosen discrete parameters generates a bow-tie model.¹ The bow-tie technique requires forming fault trees at the left side and event trees at the right side of a particular event (e.g., start of fire, detection and sprinkler activation).

The next step involves choosing a response surface model (RSM) to analyze the remaining variables. The basic concept of RSM is to approximate the responses in the analyzed domain of input possibilities for a specific model without relying on the physics of the system. The analysis applies the Polynomial Chaos Expansion (PCE) method because of its higher accuracy when less-irregular patterns are observed (e.g., far from the fire).¹¹

After defining the RSM, the sampling points for the variables to execute the deterministic analyses are chosen. First, the response surface, which predicts the global field of responses, is generated based on a limited set of these sampling points. Hereafter, a sampling pattern to represent the whole field of possibilities is defined by Design of Experiments (DoE). For the example treated here, a full factorial design is selected. For every variable, a minimum of three values are chosen.¹¹

Next, the process uses multiple sub-models such as the smoke spread, evacuation and toxicity model to conduct the deterministic evaluation of the support points. These models each provide intermediate results and have interactions as shown in Van Weyenberge, Deckers, Caspeele and Merci. The input parameters are divided into primary and secondary parameters, with significant and minor effects on the sub-model outputs, respectively.

The first RSM gets the outputs of the smoke spread sub-model and generates input data in terms of visibility, toxicity and temperature components for the evacuation model. Performing the consequence analysis provides the output from the evacuation model. This model will determine the consequences for each occupant in terms of injury or fatality.

The last step uses limit state design to perform the reliability analysis; it is coupled to Fractional Effective Dose (FED) to calculate the individual risk in terms of the probability of fatality.

The next step is to visualize the societal risk by using an FN-curve.

This case study analyzes the suggested sample set for smoke spread by FDS version 6.1.1.¹² After evaluation of the support points, the RSM is generated and the DoE is determined to obtain a uniform representation of the analyzed domain. For each scenario, 100 samples are generated and simulated. The results are used as inputs for the evacuation analysis.

Similar to the smoke spread model, a DoE is chosen based on full factorial design for the evacuation analysis. For every variable, a minimum of four values are chosen and generated in an analogy to the smoke spread sub-model to address the irregularity of the responses. The samples are analyzed with a 3D-evacuation model. After evaluation of the support points, the evacuation RSM is generated.

Similarly to the smoke spread sub-model, the DoE for the evacuation results are generated and evaluated. The output is used as input for the consequence analysis sub-model. A sampling pattern is generated for the probabilistic consequence analysis. Combining toxicity and heat (convective and radiation) effects in a Fractional Incapacitation Dose (FID).^13-14 quantifies the consequences for life safety. This value determines whether a person will manage to escape or become incapacitated.

The results are calculated in terms of a failure probability for every specific scenario and combined into a total failure probability per design. In additional, the individual and the societal risk are calculated for the three fire safety designs.

Results

For the results on the individual and societal risk, refer to the full paper.¹ Figure 4 depicts the societal risk results, together with the acceptable limit obtained from the Netherlands. Based on the absolute criteria, it can be decided that the three designs can be selected. However, it should be pointed out that the strength of the method is in relative comparison between the fire safety designs.

This way, the assumptions made in the design will have a lower impact.

When the accuracy of parameter inputs increases, the updated input parameters can be implemented directly in the model and more emphasis can be on fire safety design by the absolute risk criteria instead of relative comparison.

The right-hand graph shows visualization of the equivalent risk level for two designs, option 1 and 2. To obtain equivalency in the fire safety level, modifications can be made to increase the reliability of the safety systems so the frequencies reduce or improve the SHC design to reduce the presence of toxic and hot gases in the incident compartment. This way reduces the consequences, instead of the frequencies.

Concluding Comments

Due to the significant reduction of the computational demand, by means of the optimized probabilistic techniques, the method seems feasible for use by the private industry. Some of the main limitations of the model are simplifications made with respect to the input parameters and modeling. Only partial validation of the model has been performed to date, so more validation work is required.

Acknowledgments

The authors would like to thank the Flanders Innovation & Entrepreneurship (VLAIO) for supporting project number 130857 for this research.

---

References

¹Van Weyenberge, B., Deckers, X., Caspeele, R., and Merci, B. 2018. Development of an Integrated Risk Assessment Method to Quantify the Life Safety Risk in Buildings in Case of Fire. Fire Technology. doi: 10.1007/s10694-018-0763-6.

²Meacham, B.J. 2000. International experience in the development and use of performance-based fire safety design methods: Evolution, current situation and thoughts for the future. Fire Safety Science, Sixth International Symposium, 59–76.

³Hadjisophocleous, G., and Fu, Z. 2004. Literature Review of Fire Risk Assessment Methodologies. International Journal of Engineering Performance-based Fire Codes, 6:28–45.

⁴Alvarez, A., Meacham, B.J., Dembsey, N., and Thomas, J. 2013. Twenty years of performance-based fire protection design: challenges faced and a look ahead. Journal of Fire Protection Engineering, 23:249–276. doi: 10.1177/1042391513484911.

⁵Fleischmann CM (2011) Is Prescription the Future of Performance-Based Design? Christchurch, New Zealand: Fire Safety Science Tenth International Symposium, 77–94.

⁶Kong, D., Lu, S., Frantzich, H., and Lo, S.M. 2013. A Method for Linking Safety Factor to the Target Probability of Failure in Fire Safety Engineering. Journal of Civil Engineering Management. 19:S212–S221. doi: 10.3846/13923730.2013.802718.

⁷2007. SFPE Engineering Guide to Performance-Based Fire Protection.

⁸SFPE. 2016. SFPE Handbook of Fire Protection Engineering, fifth edition. doi: 10.1007/978-1-4939-2565-0.

⁹T. B. 2002. Risk-informed, performance-based industrial fire protection. USA: TVP.

¹⁰Van Weyenberge, B. 2013. Development of a risk assessment methodology for fire in rail tunnels. Ghent University.

¹¹Van Weyenberge, B., Criel, P., and Deckers, X., et al. 2017. Response surface modelling in quantitative risk analysis for life safety in case of fire. Fire Safety Journal. doi: https://doi.org/10.1016/j.firesaf.2017.03.020.

¹²McGrattan, K., McDermott, R., and Floyd, J., et al. 2012. Computational fluid dynamics modelling of fire. Int J Comut Fluid Dyn 26:349–361.

¹³Van Weyenberge, B., Deckers, X., Caspeele, R., and Merci, B. 2015. Development of a Risk Assessment Method for Life Safety in Case of Fire in Rail Tunnels. Fire Technology, 1465–1479. doi: 10.1007/s10694-015-0469-y.

¹⁴Purser, D.A., Maynard, R.L., and Wakefield, J.C. 2016. Toxicology, Survival and Health Hazards of Combustion Products. Cambridge, UK: RSC.