Advantages and Disadvantages of Fire Modelling Dr Guillermo Rein School of Engineering University of Edinburgh & Imperial College London Dr Guillermo Rein 9 May 2012 Chief Fire Officers Association Annual Conference 2012 Comhdháil Bhliantúil Chumann Phríomh-Oifigigh Dóiteáin 2012
Fire Modelling is ubiquitous On What? Ignition, Flame, Plume, Smoke, Spread, Visibility, Toxicity, Extinction For What? Live safety, Structural behaviour, Performance based Design, Forensic investigations, Risk, When used with caution, very powerful tool Very dangerous when miss-used Fire modelling is now very common for most fire safety calculations
FDS is king Fire Dynamics Simulator (FDS) solves well all important fire mechanisms It is the most commonly used CFD model for fire applications, because: 1. It is Free 2. Its open source nature make it excellent for Research 3. There are hundreds of Papers showing good results This has led to: A critical mass of industry and academic users Approval of many key infrastructure projects by the sole use of FDS The impression that FDS is fully validated
Examples from NIST website with FDS Hamins et al, Characterization of Candle Flames, Journal of Fire Protection Engineering 15, 2005
Examples from NIST website Link to video: http://video.google.com/videoplay?docid=-9024280504374819454#
Examples from NIST website Link to video: http://video.google.com/videoplay?docid=4830080566059919470#
Prediction or Recreation? The previous examples on fire modelling are remarkable But these were conducted after the experiments and after having full access to the experimental data of the phenomena under simulation Did the authors show all their simulations or only a selected set (or just one simulation)? What would be the result if the simulations are conducted before the experiment instead of after? What is the difference between forecast, prediction and recreation? The following slides are the work of The University of Edinburgh investigating these questions
a Priori vs. a Posteriori Has the whole process of fire modelling been validated? Are the results the same if modellers do not have access to the results a priori? Do we really know all the Strengths and Limitations of fire modelling as in realistic scenarios?
Dalmarnock Fires - July 2006 N Fire Abecassis-Empis et al., Characterisation of Dalmarnock Fire Test One, Experimental Thermal and Fluid Science 32 (7), pp. 1334-1343, 2008.
Flat Layout Abecassis-Empis et al., Characterisation of Dalmarnock Fire Test One, Experimental Thermal and Fluid Science 32 (7), pp. 1334-1343, 2008.
Heavily Instrumented 8 Lasers Deflection Gauges 20 Heat Flux Gauges 10 Smoke Detectors 10 CCTV ENLARGE ENLARGE ENLARGE ENLARGE 270 Thermocouple 14 Velocity Probes
Average Compartment Temperature Abecassis-Empis et al., Characterisation of Dalmarnock Fire Test One, Experimental Thermal and Fluid Science 32 (7), pp. 1334-1343, 2008.
Aftermath
Information given to Modelling Teams Detailed geometry (plan and dimensions) Detailed fuel load (dimensions, locations, photographs, descriptions) Ventilation conditions (including breakage of one window) Photographs of set up in the compartment Sofa fire measured in the laboratory
"I always avoid prophesying beforehand because it is much better to prophesy after the event has already taken place" Sir Winston Churchill, circa 1945
Results: Heat Release Rate Rein et al. Fire Safety Journal 44 (4) pp. 590-602, 2009
Diversity of views diversity of Behaviours
Hot Layer Temperature
Hot Layer Height
Dalmarnock Conclusions Real fire frequently faced by F&R Service Large scatter around the measurements (much larger than experimental error) During the growth phase: 20 to 500% error in hot layer temperature. 20 to 800% in local temperatures Inherent difficulties of predicting dynamics Fire modelling vs. fire models
Degrees of Freedom The excess in degrees of freedom Ill-defined and uncertain parameters that cannot be rigorously and uniquely determined lead to errors, doubts, curve fitting and arbitrary value selection. Give me four parameters, and I will draw an elephant for you; with five I will have him raise and lower his trunk and his tail Carl F Gauss (1777 1855)
a Priori vs. a Posteriori a Priori Fire Modelling Safety, Design and Engineering Maximum error a Posteriori Fire Model Model development and Research Minimum error
a Posteriori of Dalmarnock Simulations conducted after having full access to all the measurements using FDSv4 Jahn et al, 9th IAFSS Symp, 2008
Ensemble of possible fires Sofa fire Blanket and sofa fire Medium sofa fire
A Priori vs. A Posteriori Hot Layer Temperature Predictions a priori a posteriori
A Posteriori Modelling A posteriori level of agreement reached with measurements is: 10 to 50% for average hot layer temperature 20 to 200% for local temperatures A priori was: 20 to 500% for average hot layer temperature 20 to 800% for local temperatures Drastic reduction of the uncertainty from a priori to a posteriori after adjusting uncertain parameters All validation studies have been done a posteriori. Is this acceptable?
Final Remarks CFD is a cost effective and powerful tool but potentially misleading Parameter values used can be as important as the mathematical model used Fire modelling is one decade behind empirical knowledge What to ask from a fire modelling study 1. Sensitivity to other parameter values? 2. Can results be confirmed by alternative means? 3. Validated model & modeller for similar scenarios? 4. Ask for 3 rd party review from experts
Thanks! Villemard, 1910, National Library of France Paleofuture: prediction made in 1900 of the fire-fighting of the year 2000
A Simple yet Meaningful Fire Scenario Cubic enclosure of sides 20 m long Scenario related to smoke movement and life safety in atria Pool fires in the range from 1 to 3 MW (measured mass loss rate) Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
20-m Cubic Enclosure Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
Grid effects vs. Plume Theory 1.3 MW fire 2.3 MW fire Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
2006 Murcia Fire Tests in a 20-m cube Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
height of 4.5 m Experiments vs. Modelling: Plume Temperature height of 8.5 m for a 1.3 MW fire height of 12.5 m height of 20 m Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
Licensed, 2011 Thanks!
CFD is king If used with caution, very powerful tool But potentially dangerous if miss-used Manuals contain most of the theory behind the model and a very large section on validation Only a handful of pages mention limitations of the model Be aware that important limitations do exits in the art of fire modelling 3
Fuel Load Mixed livingroom/office space Fuel load is ~ 32 kg/m 2 of equivalent wood Test set-up designed for robustness and high repeatability
Video
Compartment Temperature Stern-Gottfried et al., Fire Safety Journal 45, pp. 249 261, 2010. doi:10.1016/j.firesaf.2010.03.007
Local Temperatures
Local Heat Flux to Wall (vs time)
Local Wall Temperatures (vs. time)
Tests One and Two: Repeatability
Grid Dependency Jahn et al, Fire Safety Science 9, pp. 1341-1352, 2008. http://hdl.handle.net/1842/2696
Local Temperature Predictions
Grid effects vs. Plume Theory for a 1.3 MW fire Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
Experiments vs. Modelling: Temperature near the walls height of 15 m height of 10 m for a 1.3 MW fire height of 5 m Gutiérrez-Montes, Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827 1839, 2009
Fire Modelling vs. Fire Models There are many papers addressing the validation of fire models Different models (FDS, SmartFire, CFX, FLUENT, CFAST, ) Different scenarios Focus on the mathematical engines Validations are done a posteriori This is of great value for research and development but introduce a natural bias
The need for Round-Robin Studies In 2006, Edinburgh organized a Round-Robin study of fire modelling using the large-scale tests conducted in Dalmarnock. International pool of experts independently provide a priori predictions of Dalmarnock Fire Test One using a common set of information describing the scenario.
Dalmarnock Round Robin, 2006 Validation and comparison to experiments is a hot topic since 1980. This is an ongoing debate. As the field advances, progressively more mature topics are discussed a Priori vs. a Posteriori Edinburgh conducted in 2006 fire tests in a real high-rise building, Dalmarnock Test One involved a typical residential fire scenario Round-robin teams independently simulated the test a priori using a common high-level detail description of the scenario
Possible Outcomes: a priori discussions A B C Variables shown here: HRR, Smoke layer, Wall temperature and heat fluxes
Simulations 10 Submitted simulations: 8 Field Models (FDS v4) and 2 Zone models (CFAST v6) NOTE: teams were asked to forecast as accurately as possible and not to use safety factors or applied it to design purposes Rein et al. Round-Robin Study of a priori Modelling Predictions of The Dalmarnock Fire Test One, Fire Safety Journal 44 (4) pp. 590-602, 2009
Analysis of Assumptions/Strategies General classification of input files yields these groups: Means to input/predict the HRR: 2 simulations used fully prescribed HRR (***) 7 simulations used partially prescribed HRR (**) 1 simulations used fully predicted HRR (*) Means to input the ignition source: 3 simulations used provided sofa HRR but extrapolated it (**) 5 simulations did not used provided sofa HHR but other (**) 1 simulation used provided sofa HRR as measured (*)
Lessons and Recommendations Fire predictions work well away from the flame and in simple geometries (where most fire models have been calibrated) Main source of scatter is the excess in degrees of freedom (specially properties and parameters) Predictions of fire growth do not provide good results Best practice is that fire growth is not be predicted by the model In the meantime, this is a great opportunity for further research Rein et al. Fire Safety Journal 44 (4) pp. 590-602, 2009
Conclusions Large prediction scatter around measurements (>>experimental error) Growth phase: 20 to 500% error in avg. temperature and 20 to 800% in local temperature Main source of scatter is the excess in degrees of freedom Predictions of fire growth does not provide good results Complex modelling scenario but faced frequently by fire services Great opportunity for further research
Application examples Application of FDS in large compartments to study smoke movement The scenario can be compared to analytical solutions, thus allowing for an informed grid selection Also, experiments are available to the same scenario so validation and checking for order of magnitude is possible
What to ask of a CFD study 1. Grid independence study? Time step independence study? 2. Boundary independence study? 3. Sensitivity to Parameters? 4. The results have been confirmed by alternative means (calculation and/or experiments)? 5. Validation of the code and users in similar scenarios?