EFFICIENT METHOD FOR PROBABILISTIC FIRE

SAFETY ENGINEERING
RUBEN VAN COILE
GHENT UNIVERSITY, BELGIUM
Thanks to the advent of Performance Based Design (PBD), the fire safety community is increasingly
looking at probabilistic concepts. These probabilistic concepts allow an explicit evaluation of the safety
level or failure probability associated with each design, thus moving away from current deterministic
acceptance criteria where a design is considered to be “safe” when it meets a (conservative) design
specification for a single design fire (or at best a set of design fires). Considering for example the topic
of external fire spread it is currently unclear which safety level corresponds with the design rules
incorporated in BR 187, nor is it clear which safety level is targeted or which failure probability is
considered acceptable. Nevertheless, it should be clear to all stakeholders that perfect safety does not
exist and that every design inherently has some (low) probability of failure.
Very valuable contributions have been made to advance the field of probabilistic fire safety engineering.
See for example (Van Weyenberge et al., 2015) on the topic of Life Safety in rail tunnels, (Hostikka and
Keski-Rahkonen, 2003) on a probabilistic evaluation of design fires, and (Lange et al., 2014) on a
comprehensive methodology for probabilistic PBD for fire safety of structures. The existing full-
probabilistic fire safety methodologies are however often based on Monte Carlo simulations (MCS).
Since Monte Carlo simulations are dependent on a very large number of model evaluations, the very
large computational requirements associated with these MCS have been a major issue which to a large
extent hampers the application of probabilistic PBD in practice.
To tackle this issue a number of advanced methodologies have been developed, see for example (Guo
and Jeffers, 2014) for a methodology to be applied in Finite Element models for structural fire safety
engineering. For many practical applications it may however be preferred to do a probabilistic analysis
using existing (commercial) tools, making the application of many advanced methodologies difficult.
Alternatively, the distribution type (e.g. normal, lognormal) is often assumed, see for example (Gernay
et al., 2015) where a normal distribution is assumed for the critical temperature in steel columns.
However, as illustrated in (Van Coile et al., 2013) the non-linearity associated with fire exposure can
make it very difficult to make safe a priori assumptions.
Recently a very interesting methodology has been proposed by Zhang (2013) for mathematically
estimating the full probability density function (PDF) fY describing the uncertain model output variable
Y. The methodology uses the principle of Maximum Entropy (ME) for the estimation – which is
considered to be the mathematical correct way to estimate the PDF without introducing biases – and
reduces the computational requirements by applying the Multiplicative Dimensional Reduction Method
(MDRM) and Gaussian interpolation. Consequently, the method is known as ME-MDRM. The
feasibility of the method for computationally expensive problems in the field of structural reliability has
been demonstrated by Balomenos et al. (2015).
The current paper explores the application of ME-MDRM to topics of fire safety engineering, and starts
with introducing the methodology in detail. In order to make a mathematical estimation of the full PDF
describing the uncertain model output Y, ME-MDRM requires only 1 + 4n model evaluations, with n
the number of uncertain input variables xi. Consider as an example a square concrete column with a
width of 300 mm and with 6 variables considered to be stochastic (amongst others the concrete
compressive strength and concrete cover – full details in the full paper). The column is eccentrically
loaded and is exposed from 4-sides to 60 minutes of ISO 834 standard fire (considering prescriptive
requirements).
In order to demonstrate the structural fire resistance of the eccentrically loaded column, a reliability
analysis will be performed and therefore, the PDF describing the maximum eccentric load Pmax (for a
given eccentricity e = 0.05 m) has to be determined. A numerical model has been developed to calculate
the maximum vertical load for a given realization of the input variables xi, but is computationally too
expensive to allow for Monte Carlo simulations (outside of a research environment). Consequently, ME-
MDRM is applied, requiring only 25 model evaluations.
Figure 1 visualizes the PDF obtained by the ME-MDRM together with a histogram of 10000
(computationally expensive) Monte Carlo simulations, as well as a lognormal approximation (LN)
which uses the 10000 MCS as input data for estimating the parameters of the lognormal distribution
(and is thus at a huge advantage compared to the ME-MDRM). Figure 1 clearly shows that for this
example the ME-MDRM has an excellent performance despite the very low computational
requirements. Based on the results given here and other applications given in the full paper, it is
concluded that ME-MDRM is a very promising methodology which can greatly advance the application
of probabilistic fire safety engineering in practice.

Figure 1. ME-MDRM result for PDF describing Pmax at 60 minutes ISO834 for e = 0.05 m, and comparison with
histogram of 10000 MCS, and a lognormal approximation (with parameters based on the MCS).
REFERENCES

Balomenos, G.P., Genikomsou, A.S., Polak, M.A., Pandey, M.D. (2015). Efficient method for probabilistic finite element
analysis with application to reinforced concrete slabs. Engineering Structures, 103, 85-101.
Gernay, T., Elhami Khorasani, N., Garlock, M. (2015). Fragility analysis of steel building in fire. Proc. First Int. Conf. on
Structural Safety under Fire and Blast (CONFAB 2015), Glasgow, 02-04/09, UK, 252-261.
Guo, Q., Jeffers, A.E. (2014). Direct differentiation method for response sensitivity analysis of structures in fire. Engineering
Structures, 77, 172-180.
Hostikka, S., Keski-Rahkonen, O. (2003). Probabilistic simulation of fire scenarios. Nucl. Eng. Des., 224, 301-311.
Lange, D., Devaney, S., Usmani, A. (2014). An application of the PEER PBEE framework to structures in fire. Engineering
Structures, 66, 100-115.
Van Coile, R., Caspeele, R., Taerwe, L. (2013). The mixed lognormal distribution for a more precise assessment of the
reliability of concrete slabs exposed to fire. Proc. of ESREL 2013, Amsterdam, 29/09-02-10, the Netherlands, 2693-2699.
Van Weyenberge, B., Deckers, X., Caspeele, R., Merci, B. (2015). Development of a Risk Assessment Method for Life Safety
in Case of Fire in Rail Tunnels. Fire Technology, Online First.
Zhang, X. (2013). Efficient Computational Methods for Structural Reliability and Global Sensitivity Analyses. Doctoral
dissertation. Waterloo University, Waterloo, Canada.