EVALUATION OF THERMAL RADIATION MODELS FOR FIRE SPREAD BETWEEN OBJECTS Fire and Evacuation Modelling Technical Conference Baltimore, Maryland 15-16 August 2011 Rob Fleury Arup, Sydney, Australia Michael Spearpoint University of Canterbury, New Zealand Charles Fleischmann University of Canterbury, New Zealand

CONTEXT AND MOTIVATION Zone model BRANZFIRE currently being upgraded to include risk-based modelling Quantitative Risk Assessment tool Monte Carlo sampling to address inherent uncertainty in design fires Will include a radiation and ignition sub-model Range of outputs – probabilistic outcomes Project by Building Research Authority New Zealand (BRANZ) and the University of Canterbury, NZ. Funding from Foundation for Research Science and Technology

AIM Evaluate the performance of thermal radiation models Will help determine if or when secondary objects ignite due to thermal radiation Direct radiation from flames only Work by others

APPROACH Experimental Comparison with theoretical models Radiant heat flux measurements taken around gas burner Comparison with theoretical models Shokri & Beyler correlation Point source model Shokri & Beyler detailed method Mudan method Dayan and Tien method Rectangular planar model The main purpose of the experimental work was to provide a comprehensive set of radiative heat flux data to be used for comparison with the thermal radiation models

EXPERIMENTAL Carried out at UC. For controllability and repeatability, a simple propane gas burner was used to model a burning object. This, however, represents a limitation in the experimental work. Burner aspect ratio. Burner angle. Location of heat flux gauges. And orientations. Range of heat release rates. In all, over 3000 independent measurements of radiant heat flux were taken, each one averaged over a 2min logging period.

FLAME HEIGHT Heskestad Thomas ImageStream Produced by ImageStream Reference ImageStream (more info in paper). This work presents interesting opportunities in characterising the flame; however was chosen not to be investigated in great detail in this work, other than determining the mean flame height. Produced by ImageStream

COMPARISON OF MODELS

COMPARISON OF MODELS 29% 221% 103% 58% 47% 41% Re-explain different burner aspect ratios. Point source model on average 29% percentage error from experimental results. Point source equally accurate for square and rectangular fire sources.

COMPARISON OF MODELS Average percentage error from experimental results for different target orientations Vertical Horizontal Shokri and Beyler Correlation 99% N/A Point Source Model 18% 76% Shokri and Beyler Detailed Method 50% 89% Mudan Method 224% 205% Dayan & Tien Method 35% 71% Rectangular Planar Model 40% None of the models were particularly strong at predicting the heat flux to horizontal targets. For vert targets, point source most accurate by considerable margin.

COMPARISON OF MODELS Average percentage error from experimental results for different radiant heat fluxes <5 kW/m² 5-10 kW/m² >10 kW/m² Shokri and Beyler Correlation 126% 48% 29% Point Source Model 17% 31% Shokri and Beyler Detailed Method 46% 54% 49% Mudan Method 205% 240% 238% Dayan & Tien Method 36% 35% Rectangular Planar Model 44% 40% 41% The rather common advice of restricting the use of the point source model to applications where the radiant heat flux is 5 kW/m² or less was found to be irrelevant in this application

COMPARISON OF MODELS Average percentage error from experimental results for different target positions Central Offset Shokri and Beyler Correlation 101% 97% Point Source Model 19% 17% Shokri and Beyler Detailed Method 48% 52% Mudan Method 215% 234% Dayan & Tien Method 36% 33% Rectangular Planar Model 45% 34% Re-explain what the different target positions were, noting that the gauge orientation is still facing the fire.

LIMITATIONS TO RESULTS Single object fires within compartments Propane gas vs real objects (e.g. furniture) Heat release rates 100 – 300 kW Effective fire diameter: max 0.6 m Input parameters Most of the models more applicable to fires >1m in diameter. Talk about input parameters and how there was limited opportunity to explore these. Therefore, largely fixed. Overall, it cannot be said with certainty that the findings of this work extend to any application outside the conditions of the experimental programme. One can, however, make reasonable assumptions to extend the results to fires of a similar fuel and order of magnitude as those tested in this work

Ease of Implementation SELECTING A MODEL Two important factors for selecting model for BRANZFIRE: Accuracy Ease of Implementation 1. Point Source Model 2. Dayan and Tien Method Point source most accurate based on the experimental results obtained. Also the most robust, able to deal with a range of conditions. Ease of implementation. Also, fewer inputs required from user. For design purposes, many geometrical aspects may not be known, therefore the more complex models do not lend themselves well to computational models.

CONCLUSIONS Generally, Point Source and Dayan & Tien models provide best match to experimental data Point Source model (surprisingly) proved to be the most robust, in terms of dealing with different scenarios Point source model relatively straight-forward to implement into two-zone model Point Source model will form part of the radiation and ignition sub-model within BRANZFIRE

Thank you