[
Fire modelling is about to take a quantum leap in performance with the development of the FireGrid project. John Holden examines how we may be able
to access fire and building performance in real-time, or faster.
For decades, fire modelling has been a fundamental tool used by fire safety engineers. These fire models have evolved from very simple, conservative equations into very sophisticated models based upon computational fluid dynamics (CFD). Essentially CFD involves solving equations describing the flows, temperatures and pressures across many small finite volumes, thus predicting or estimating the conditions at a point many metres from the fire source. In the modelling of fire dynamics, there is the added complexity of defining and modelling the fire source, as well as considering the initial conditions, boundary conditions and fluid dynamics associated with CFD.
Clearly, such models – while providing a powerful tool for the investigation and research of fire related problems – generally require significant computer processing and equally importantly, experienced people to get meaningful results. When computing resource was relatively scarce and expensive, significant planning and analysis of fire modelling procedures was required to make efficient use of facilities. This tended to ensure that such activities were entrusted only to experts. Now that computer power is widely available, and cheap, the use of CFD techniques has become widespread. While the wider availability of such tools is to be welcomed, more and more fire modelling is being performed without adequate consideration of both the fire science and the numerical aspects of the process, leading to potentially erroneous outcomes.
One of the most important aspects is to ensure that the assumptions made in the model are relevant. This is best done by validating the outcome of the model against actual data from a real fire test. Unfortunately such an approach is not always appropriate because real fires are not like test fires. However, valid extrapolation of a test fire into a real application can be done using an appropriate CFD model – but it must be done with caution. There are many aspects to ensuring that a model is producing accurate results including sensitivity analysis, convergence criteria and reality checks, but these are beyond the scope of this article.
The FireGrid project is developing an alternative strategy. This involves the modelling of fires in buildings as they actually develop, and uses the high performance computer resources available through the ‘Grid’, sometimes colloquially referred to as the internet on steroids! Put simply, the Grid is a term used to describe the resources and infrastructure (often the internet) used to make significant computer resources available to users on-demand. For example, a fire modelling programme run different parts of the modelling process in parallel on a number of computers connected to the Grid. The beauty of this approach is that the computers used to accelerate the modelling process can be anywhere in the world. Each of these computers would deliver the output from their particular calculations back to the original programme, which would combine them all into an overall result for the modelling process. By dividing the task into several parts, running these in parallel on different computers, and re-combining the outputs to create the overall result of the modelling, the time required to complete the entire process is greatly reduced.
The FireGrid system aims to model the likely sequence of events following the outbreak of a fire faster than real- time. In other words, it aims to forecast events before they actually occur, and will use real data from the affected building’s sensors to verify the results of, or make adjustments to, the modelling process. Using such ‘sensor-steered’ modelling will ensure that reliable predictions are achieved throughout the real-time modelling process.
Turbocharged computer power
Most of us are familiar with Moore’s law, Intel co-founder Gordon Moore’s 1965 prediction that the number of transistors that could be integrated into a single computer processor chip would roughly double every 18 months or so. That prediction has been upheld for the last 40 years, meaning that we are fast approaching being able to squeeze a billion tiny transistors onto a single chip. Although more processors generally means more speed, when it comes to processing power, size – or the lack of it – isn’t everything. The speed at which these transistors can operate is the ultimate key to how quickly our computers can crunch numbers. The so-called clock frequency of computer chips has improved over roughly the same time, from a few hundred kiloHertz to over
3 GigaHertz. The result is that today you probably have more computing power on your desktop than was available from a million pound supercomputer in the early 1990s. And it gets better – all this computing power actually comes at a lower cost. According to Intel, if transatlantic flights had followed the same price and performance improvements as PCs, we should now be able to cross the Atlantic ocean in under a second for less than a penny!
It doesn’t end there. While the power of the processor incorporated in the humble PC has grown exponentially, so has the processing power of the supercomputer – which in today’s parlance is a computer utilising several processors that work in parallel to increase the number of calculations that can be performed per second. Here, not only has the speed of their processors increased, but so has the number of processors that can be harnessed to work in parallel and so speed up even the most calculation-intensive applications.
Currently the world’s fastest recorded supercomputer is IBM’s BlueGene/L which contains 128,000 processors and at its peak, is capable of performing 360 trillion floating point calculations per second. To put that in perspective, the IBM computer is around a million times faster than one of today’s really fast desktop PCs. Of course, such computers are one-offs and are generally designed and commissioned for specific functions – such as modelling nuclear reactions or space shuttle engines. However, there are still a huge number of so-called high performance computers (HPCs) available for those of us with more day-to-day requirements, such as fire safety engineering. The availability of low cost, high speed processing power means that, in principle, calculation hungry applications such as CFD fire modelling could now be performed by anyone with a computer.
CFD fire modelling
Fire modelling is one of the most powerful tools available to the fire safety engineer. This is particularly the case when it includes the reaction to the fire of building occupants, as well as the building itself.
The processes and events that occur during a fire are, in most cases, very complex in nature. Their complexity arises from the fact that the physical and chemical processes controlling fire and smoke development – such as turbulence, combustion and radiation – interact with each other and with the building envelope. In order to have confidence in the predictions made by fire modelling programmes, it is essential that the important transport processes controlling fire development are properly understood, and that the key components are clearly identified. Because of the mutual interactions of these processes and their coupling with any enclosure, reduced scale experiments alone are often not sufficient to reproduce full scale features. Mathematical models supported by full-scale experiments offer a practical solution for a better understanding of fire and the fluid dynamics, heat transfer and other phenomena involved.
CFD provides the potential to study the complicated problems encountered in fire safety engineering and the performance assessment of materials and construction products. These could involve single or multi-phase, turbulent fluid flows with or without combustion. It offers a means of optimising fire safety engineered solutions for innovative architectural design using the assumption of a ‘design fire’. The problems encountered in indoor air movement, early detection of smoke, effects of water sprinklers on smoke layers and the eventual dispersion of combustion products within the atmospheric boundary layer can all be examined by CFD. Fire safety practitioners can also use programmes to increase their understanding of the consequences of alternative design choices.
There is a real danger, however, of generating erroneous or misleading results from such modelling if, for example, an inappropriate ‘design fire’ is chosen, or if there is inadequate appreciation of the underlying fire ‘science’ and of the limitations of the numerical methods employed. One way of verifying the results of a fire model is to compare the predictions of the modelling process with the actual events of a real fire. Such confirmation of results would provide confidence in the modelling process, which could be extrapolated to other models relying on similar processes and assumptions. Such an approach is not always practical or possible, however, particularly where the modelling of large scale structures or buildings is involved.
There is now the possibility of developing an alternative strategy, which combines advanced knowledge of fire science and modelling techniques with the unrivalled high performance computer resources now available on the Grid. That is, to forecast the consequences of an actual building fire during its progress, and to use data from the building’s sensors to verify or correct the modelling process. This is the goal of FireGrid.
FireGrid
FireGrid is a three year, GB pound 2.3 million project funded by the Technology Strategy Board and due for completion in April 2009. The FireGrid consortium is led by BRE and includes The University of Edinburgh; Arup; Abaqus UK; Ansys Europe; Xtralis (formerly Vision Fire & Security); the London Fire and Emergency Planning Authority and the Institute of High Performance Computing (A*STAR) in Singapore.
FireGrid offers a new approach to the computer modelling of fires. Engineers are developing a fire modelling capability that will model the growth of a real fire in a building – and its effects on the building and its occupants – to provide results faster than real time. In other words, it will be able to predict, rather than retrospectively model, how a fire will develop and how the building and its occupants will react to the fire before these events actually unfold. This will enable it to provide incident commanders and other emergency responders with timely information on the development of a fire, and early warning of events such as flashover or the collapse of all or part of a building. This will be achieved by using advanced programming techniques, including BRE’s Monte Carlo based CRISP risk assessment and egress packages, CFD codes such as BRE’s JASMINE code, ANSYS-CFX and FDS, structural response packages from Abaqus UK, and fire structure coupling provided by Ansys-CFX. These advanced algorithms will be processed using the high performance computer resources available through the Grid.
Fine tuning
The ability to model the growth of a fire faster than real time brings an additional and important benefit. FireGrid generates data describing the properties of an actual fire, for example, temperatures at specific points in the affected building, the structural response of the building or the reaction of its occupants to the fire – before the events giving rise to those properties have actually occurred. This makes it possible to compare these predictions with actual data generated by the building’s sensors at a time corresponding to that for which the predicted values were calculated. This feedback from the building sensors enables real time checks of the accuracy of the forecasts and, if required, to steer the modelling process, taking into account differences between predicted and actual data values. This data comparison and self-correction mechanism provides robust verification of the modelling process and, importantly, the accuracy of the predictions and other information communicated to the emergency responders.
Conclusion
Our understanding of fire science and modelling is continuously growing, but it is the dramatic improvements in the processing speed and availability of computer resources that is enabling us to implement the knowledge and ideas that we are generating in new ways. The availability of high performance computers, accessed through the Grid, is opening up the possibility of forecasting the likely sequence of events following the outbreak of a fire in a building, and to enhance their accuracy by verifying or steering the CFD modelling that generates these forecasts, using data from the building’s own sensors.
John Holden PhD, BSc, MIET is principal consultant at BRE Global. The author gratefully acknowledges the contribution on CFD fire modelling from professor Suresh Kumar, Professor Geoff Cox and other members of the FireGrid consortium. FireGrid is funded by the Technology Strategy Board along with contributions from project partners. For further information about the FireGrid project visit www.firegrid.org
MAPPING OUT THE GRID
FireGrid is a three year, GB pound 2.3 million project supported by the DTI-led Technology Programme and due for completion in April 2009. The project aims to help the fire and rescue services reduce the number of casualties and damage to property resulting from fires, by providing critical and timely forecasts concerning the likely sequence of events following the outbreak of fire in a building – before such events occur. This will include predictions of how the fire is expected to spread, how the structural integrity of the building will be affected and how the building occupants are likely to react in response to the fire.
It is anticipated that this information will be used by the emergency services to make best-informed decisions regarding their response to the fire. The FireGrid consortium is led by BRE and includes the University of Edinburgh; Arup; Abaqus UK; Anys Europe; Xtralis (formerly Vision Fire & Security); the London Fire and Emergency Planning Authority; and the Institute of High Performance Computing, (A*STAR) in Singapore).
The FireGrid project brings together a number of equipment and programming disciplines including: sensor technologies and networks; data translation, verification and communication; fire modelling (including the reaction of the building and its occupants to the fire); and Grid-enabled high performance computing. Small and large scale fire tests will be used throughout the project to develop and verify data gathering systems, communication networks and modelling algorithms.