Friday, 10 August 2012

Forecasting fire dynamics: tomorrow's infrastructure protection


Imagine a technology able to forecast fires. It would lead to a paradigm shift in the response to emergencies and provide the Fire Services with essential information about the ongoing blaze with some lead time (i.e. seconds or minutes ahead of the event). It would also allow for the future of infrastructure protection to be implemented in smart buildings.

316 s after ignition. could this be forecasted ahead of time? [Rein 2012]
But despite advances in the understanding of fire dynamics over the past decades and despite the advances in computational capacity, our ability to predict the behaviour of fires in general and building fires in particular remains very limited. The state-of-the-art of computational fire dynamics is not fast or accurate enough to provide valid forecasts on time...

Paleofuture: forecast made in 1900 of the fire-fighting in the year 2000.
By Villemard, 1910, National Library of France
But we found a way to solve this problem. In a recently finished PhD thesis and set of published  papers, we show the technology is possible. We  propose to use sensor measurements of the ongoing fire to steer and accelerate computer simulations. This takes advantage of the concept of data assimilation (similar to what meteorologists do to forecast the weather).

Our method consists on combining a simplified spread mechanism with a fire model, and use sensor data to find the fire parameters that dominate the spread. This way, the model automatically recovers information lost by approximations in the physics, chemistry and the maths.

Concept of data assimilation and the sensor steering of model predictions [Cowlard et al 2010]
 A series of compartment fire cases haven been studied this way, and we investigated two different fire models. First a simple two-zone model, and then a state-of-the-art computational fluid dynamics (CFD) model.
 
For the simple two-zone forecast model, the firepower and the growth rate were estimated correctly up to 30 s ahead of the event: the model was faster than the fire. This was the very first time a fire forecast technology was demonstrated and the first time positive lead times were reached. The results show that the simple model is able to deliver fast and useful information about the ongoing fire thanks to the sensor data. This initial work demonstrated that the new methodology is effective, and allowed us to move to the next level of complexity. 


Computational domain of the fire
compartment [Jahn et al 2012]
For the CFD forecast model, we use a coarse grid that provides short computation times. Spatially resolved forecasts were obtained in reasonable time. It is even possible to estimate the growth rates of several different spreading fires simultaneously. Although actual positive lead times were not reached here with CFD, it is shown that the use of relatively coarse grid size in the forward model significantly accelerates the assimilation (up to 100 times faster) without loss of forecast accuracy. Actual positive lead times with CFD are possible by reducing the computational time by at least another order of magnitude in the near future using high performance computing techniques.


Dalmarnock Fire Test One conducted on July 25th.
Our latest bit on the topic was a test case using the measurement data  from a real fire. We forecasted in near real time the Dalmarnock Test One, conducted in 2006 inside the 3.5 x 4.7 x 2.4 m living room in a high rise building in the city of Glasgow. It was possible to find a good fit between the observations and the forecast using CFD.


The results are a fundamental step towards the development of forecast technologies able to lead the fire emergency response. The work opens the door to forecasting fire dynamics, but it is an on-going research topic.

We are happy that the work has been featured in the media and  people is being exposed to this novel idea:

Our research resources on the topic (in reverse chronological order):

1) Jahn, Rein and Torero (2012), Forecasting fire dynamics using inverse
Computational Fluid Dynamics and Tangent Linearisation, Advances in
Engineering Software 47 (1), pp. 114-126. doi:10.1016/j.advengsoft.2011.12.005

2) Rein (2012), Plenary Keynote: Numerical forecasting of fire dynamics: tomorrow's infrastructure protection - Young Investigators Conference of the European Community on Computational Methods in Applied Sciences (ECCOMAS), Aveiro. See below.

3) Jahn, Rein and Torero (2011), Forecasting Fire Growth using an Inverse CFD Modelling Approach in a Real-Scale Fire Test, Fire Safety Science 10, pp 1349-1358, doi:10.3801/IAFSS.FSS.10-1349 

4) Jahn, Rein and Torero (2011), Forecasting Fire Growth using an Inverse Zone Modelling Approach, Fire Safety Journal 46, pp. 81–88. doi:10.1016/j.firesaf.2010.10.001. Paper shortlisted for 2010 Lloyd's Science of Risk Prize.

5) Jahn (2010), Inverse Modelling to Forecast Enclosure Fire Dynamics, PhD Thesis, School of Engineering, University of Edinburgh.

6) Cowlard, Jahn, Empis, Rein and Torero (2010), Sensor Assisted Fire Fighting, Fire Technology 46 (3), doi:10.1007/s10694-008-0069-1


Numerical forecasting of fire dynamics (Plenary YIC ECCOMAS)