The Road Traveled

FDS and Smokeview were officially released in 2000. However, for two decades various CFD codes using the basic FDS hydrodynamic framework were developed at NIST by Howard Baum, Ron Rehm and Kevin McGrattan for different applications and for research. In the mid 1990s, many of these different codes were consolidated into what eventually became FDS. Before FDS, the various models were referred to as LES, NIST-LES, LES3D, IFS (Industrial Fire Simulator), and ALOFT (A Large Outdoor Fire Plume Trajectory).

The early NIST LES model described the transport of smoke and hot gases in an enclosure using the Boussinesq approximation, where it is assumed that the density and temperature variations in the flow are relatively small. Much of the early work with this form of the model was devoted to the formulation of the low Mach number form of the Navier-Stokes equations and the development of the basic numerical algorithm. Early validation efforts compared the model with salt water experiments. Charley Fleischmann, then a graduate student at Berkeley, and eventually his student, Jason Clement, validated the hydrodynamic model in FDS by measuring salt water flows using Laser Induced dye Fluorescence (LIF).

Eventually, the Boussinesq approximation was dropped and simulations began to include more fire-specific phenomena, such as fire plumes, ceiling jets, sprinkler activation, warehouse fires, and large oil fires. The early validation efforts were encouraging, but still pointed out the need to improve the hydrodynamic model. To address this, we introduced the Smagorinsky form of large eddy simulation in 1998. This addition improved the stability of the model because of the relatively simple relation between the local strain rate and the turbulent viscosity.

The first official version of FDS, released in 2000, was aimed at large scale simulations of smoke movement from prescribed, well-ventilated fires, ideal for design work where the fire's heat release rate is not predicted by the model, but rather specified by the Authority Having Jurisdiction, or AHJ. Over the next few years, Jason Floyd, then a NIST post-doctoral fellow, and Simo Hostikka of VTT, Finland, as a guest researcher at NIST, developed the mixture fraction combustion model and the finite volume radiation transport solver, respectively, that have been the backbone of FDS ever since. These improvements were implemented in version 2 (2001). Versions 3 (2002) and 4 (2004) saw gradual improvements in these routines, along with the development of multiple-meshes (Kuldeep Prasad) and parallel processing (Chuck Bouldin and Kevin McGrattan).

During the NIST Investigations of the World Trade Center collapse and the Station Nightclub fire, it became fairly obvious what needed to be done with FDS to make it an effective tool for reconstructing fires. Up to that point, FDS had been used by the FPE community for design applications, and to some extent forensic work, but the scope of the Investigations pushed the model to its limits. By 2005, it was clear that FDS was going to need a major overhaul, so we set about creating a new version (FDS 5) that would dramatically increase the flexibility and functionality of the model. The work proceeded along two broad fronts - the gas phase and the solid phase. In short, better combustion and better pyrolysis. Jason Floyd and Kevin McGrattan tackled the gas phase and Simo Hostikka at VTT tackled the solid. Meanwhile, the FDS user community continued working with FDS 4, and we struggled to keep up with the ever-increasing demand. By 2006, it became painfully evident to all of us that we could not keep up with the tremendous growth in support requests, and we were also not able to efficiently merge our various new algorithms into FDS 5. We decided to open up a new front - IT support, both to service user needs and to help us developers work better together. Glenn Forney, via Smokeview, had up to then played the dual roles of computer support and software support, but he was becoming over-burdened with other responsibilities. So at the end of 2006, we hired Bryan Klein to take some of the weight off our shoulders.

Proposing New Construction

Before talking about the Road Ahead, a few words about how we decide on new areas of development. First and foremost, suggestions and requests by the user community are given a high priority, especially if the ideas come from different sectors of fire protection. Also, the individual developers, and the organizations they represent, have a specific research agenda. Wherever they come from, proposals for new features in FDS or for changing existing features must meet a number of criteria before they will be considered, which are explained below:

This Road was Improved?

There is a substantial cost in terms of time and effort to develop and test new source code. Proposed changes must make that effort worth it. In general this occurs by demonstrating that the proposed changes will demonstrably improve the predicton of FDS for a wide range of fire protection applications. For some computed quantities this is a very high bar. For example in well-ventilated fires FDS has been shown to make predictions of far-field gas temperatures and species concentrations within experimental uncertainty. This is an area where it would be difficult to demonstrate significant improvement, especially in light of the results of a recent US NRC V&V exercise (NUREG 1824) in which it was shown that FDS, for certain predicted quantities, is within the uncertainty of the experiments against which it was compared.

Obey the Speed Limit

One of the main advantages of FDS over other CFD models is its fast computational speed and relatively modest requirements in terms of computational hardware. FDS was developed for practicing fire protection engineers who typically cannot afford to have a computation take months to complete. Even if the application is not necessarily practical, like a combustion research application, we still insist that any change to FDS must be evaluated for its impact on computational speed. If the end result of a proposed change is to greatly increase the computational time with little added benefit, such a change will not be considered; however, a proposed change that has little benefit in terms of predictive outcome but does reduce computational time is likely to be considered. Along with this is the development concept that routines related to a particular phenomenon should consume computer time in proportion to the impact of that phenomenon. For example, radiation from a fire plume typically accounts for around a third of the energy released by the fire. Thus, the various routines for radiation heat transfer should only consume about a third of the computer time. The basis for this is that if FDS spends a lot of time computing phenomena that have little impact on the overall uncertainty of the computation, then that time would be better spent improving the computation of a more critical phenomenon.

No Roads to Nowhere

We strive to maintain a high degree of flexibility within the FDS source code. This allows developers to experiment with adding new features or changing how existing features are done without having to overhaul large portions of the source code. A proposed change that restricts this flexibility is not desired.

Along with developmental flexibility is the broad applicability of FDS. Proposed changes that would restrict the use of FDS to specific types of fire protection problems are not desired.

No Limited Access Highways

One reason for the success of FDS is the ability for a fire protection engineer with modest training to use FDS for traditional applications such as smoke control, detector layout, etc. While there are and always will be applications that require a high level of knowledge of FDS, its numerics, and the physics and chemistry behind fire, the developers do not wish that for traditional applications. Changes to the code that would make it difficult for a typical user to use for traditional applications are not desired.

The Road Ahead

Combustion

Current Activities: Led by Jason Floyd of Hughes Associates, we have developed a more general combustion formulation than previous versions. It still uses the mixture fraction approach to streamline the solution of the species transport equations, but now the mixture fraction is decomposed into 2, 3 and possibly 4 components representing fuel, products of incomplete combustion, and products of complete combustion. In this way, soot and CO growth/destruction can be modeled whereas in the past combustion was always assumed instantaneous and complete. The new algorithm has been validated against experiments ranging from Kermit Smyth's Wolfhard-Parker burner to Craig Beyler's under-ventilated hood series to Bill Pitts' reduced scale enclosure tests. Work continues along these lines with the reduced and full-scale under-ventilated compartment experiments of Johnsson and Bundy at NIST. Details of the model can be found in the FDS Technical Reference Guide.

Future Activities: The new CO production algorithm gets released this autumn, and over the next few years we will monitor its progress. It takes years for any new algorithm to truly sink in with the users. They are always skeptical at first, and it will take a series of independent reviews by interested FDS enthusiasts to really shake out the new features. In the meantime, we plan to tie soot growth/destruction in with the current CO algorithms. Exactly how hasn’t been determined yet. The only thing that is clear is how the potential algorithm will be incorporated into the mixture fraction framework.

The current three parameter mixture fraction model could be extended to four to account for soot and CO production and transport and local flame extinction. Additional partitionings of the mixture fraction could be added as well; however, there is a point of diminishing returns. The present advantage of using a mixture fraction approach is to track the movement of a number of gas species using a smaller number of scalar variables at the expense of having to use state relationships to obtain species mass fractions. As more mixture fraction variables are added, the net savings in fewer species transport equations vs the expense of state relationships will diminish.

The long term goal of FDS is to track the major carbon-carrying species directly (soot, CO, CO₂, and unburned fuel). Minor species could potentially be tied to the production of the major ones, or empirically determined. However it is done, we do not want to compromise speed and accuracy. As with the current CO production algorithm in FDS 5, we will give the user the option of including more detailed combustion chemistry in fire scenarios where it is warranted.

Recent large scale experiments conducted at NIST under US NRC sponsorship included measurements of smoke concentration that were substantially lower than predictions of both FDS and CFAST, the NIST zone fire model (US NRC, NUREG-1824). A possible explanation has been suggested by researchers at Hughes Associates, who have noted that significant amounts of smoke can be deposited on walls, an effect that is usually neglected by fire models. The Hughes team have not finished their experiments, but if their findings are true, it might mean that FDS will have to track soot/smoke explicitly.

Modeling gas phase suppression (flame extinction) is also a long term goal. FDS 5 has a simple, empirically based model of flame extinction based on the concept of the lower oxygen limit applied locally, grid cell by grid cell. Also available is the same concept for the fuel stream using the lower flammability limit Validation work is on-going to test the simple model, and it is planned to also look at gaseous suppression agents and water mist.

Modeling of ignition or re-ignition (deflagrations and backdrafts) is also a long term goal. The new approach to partitioning the mixture fraction will make possible development into these areas. It is noted that the low-Mach number limitations of FDS means that computations on flows that approach a detonation event will not be possible.

Pyrolysis and the Solid Phase

Current Activity: Led by Simo Hostikka of VTT, Finland, we have rewritten the FDS solid phase algorithms to provide a better a description of heat transfer and pyrolysis. First, we added the functionality of doing 1-D heat transfer through multiple layers of materials. This was an obvious improvement to make, given our experience with insulated steel in the WTC Investigation, and polyurethane foam covered plywood in the RI Nightclub Investigation. Next, we generalized the pyrolysis algorithm to allow for multiple materials undergoing multiple reactions, similar to the approach taken by Kashiwagi et al. in the NASA microgravity program. Work will continue in FY 2008 to verify and validate the new solid phase model, which has been implemented in FDS 5. This effort will be complemented by an external thrust (in cooperation with the SFPE) to develop an engineering guide for obtaining material properties for fire models. It is planned that this external effort will be headed up by new grantees and supported by an SFPE task group. Internally, the project led by Greg Linteris, "Experimental Data for Sub-Grid Modeling," will support this effort. Also, the new project headed by Dan Madrzykowski, "Fire Plume/Wall Interaction," will provide validation data to check the new pyrolysis algorithm.

All of this effort is driving towards better predictions of flame spread and fire growth. It has been noted by many FDS users that the overly simplistic description of solid materials in FDS versions 1-4 has limited our ability to predict flame spread.

Future Activities: Fire models developed by NIST have typically included a "database" of properties for common building materials and potential fuels. These properties are sometimes based on bench-scale measurements done here at NIST, sometimes they are merely plucked from handbooks. For properties like thermal conductivity, specific heat, and density of non-combustibles, this may be OK in the sense that the results of our validation work with US NRC has shown that the model predictions are less sensitive to wall lining properties than many other inputs. However, for properties associated with pyrolysis, we need to do better. One of the biggest mistakes we've made in developing FDS has been to release a "database" of various gas and solid phase properties. Those objecting to the misuse of FDS need only look to this database. Originally, we intended the database to "get the user started" by providing a "template" for developing his/her own properties. Given the scarcity of data for real materials, however, users have simply taken the path of least resistance and rely on the database, using the closest approximation to their materials. Worse, they cite the use of "NIST Standard Data" This is hardly anything of the kind.

So, starting in FDS 5, the database has gone away. Users will be expected to provide ALL of their own data for a given simulation. And they will undoubtedly need to provide references for the AHJs as to the source of the data. We expect that this will not be met with overwhelming enthusiasm, but as V&V documentation becomes more important, it is inevitable that the property data will be subject to the same scrutiny as the model itself. ASTM E 1355 even calls for it, and it is bound to come up again as we continue to validate more complicated fire phenomena.

The goal of this effort would be an engineering practice guide. The goal is NOT necessarily the development of a new pyrolysis model or new test methods, but rather a balanced description of a burning object along with the corresponding mathematical framework. Balanced means that the complexity of the mathematical equations is consistent with the experiments needed to obtain the empirical parameters. If, for example, the pyrolysis model assumes that the burning rate of a material is linearly proportional to the incoming heat flux divided by a "heat of gasification," then the document should contain this simple equation and the appropriate method for obtaining the effective heat of gasification. In some sense, the heat of gasification is defined by the simple equation. FDS has actually moved beyond this simple model to include the formation of char, multiple layers of materials, and so on. But more complexity typically requires more input parameters; the key is that these parameters should be rigorously based on a mathematical framework. The proposed document would not necessarily focus on any one model. Instead, there would be a hierarchy of pyrolysis models with a description of the necessary means to experimentally determine the input parameters.

There have been numerous pyrolysis models developed over the past 30 years by people like di Blasi, Kashiwagi, Atreya, Fernandez-Pello, Quintiere, and others. FDS has borrowed from many of these. A weakness with the various models is that there is little consideration for providing input values for materials other than PMMA or Douglas Fir. The validity of the models has typically not been validated for most materials. The models almost always seem tailor-made for a particular material, and can almost never be used for real applications. There was no pyrolysis model of polyurethane foam on top of wood for the Station Nightclub Investigation. Nor was there a model of the workstation panel for the WTC. A general-purpose description of pyrolysis and a set of experiments to get the necessary inputs is needed. Unfortunately, the fire and combustion literature is replete with overly detailed models of very specific material configurations and phenomena, with a disregard for practical, general-purpose applications. We should not develop complex, multi-dimensional models of the solid phase without a solid foundation for getting material properties. Unlike most ASTM standards, we're not necessarily talking about developing new standard test methods. Rather, we want to re-assess what available now, especially in light of what is needed for application of CFD fire models to actual fire scenarios. As the FDS "database" goes away, users will need guidance.

Droplets, Particles, and the Dispersed Second Phase

Code development in the areas of spray dynamics, spray heat transfer, and other areas related to dispersed second phases has been modest in comparison to the gas and solid phase efforts. Future areas of work include improved tracking of droplets and particles, improved radiative heat transfer to and from droplets and particles, and improved interaction between droplets and particles and the gas phase. These area would encompass the ability to have combusting particles (fire brands) and to better account for surface wetting, surface suppression, and surface penetration by droplets.

The Active Fire Protection Systems

In the 2006 workshop of The International Forum of Fire Research Directors (FORUM), the improvement of the ability to predict the impact of active fire protection systems on fire growth and fate of combustion products was identified as the most important research topic of the fire research community. Many of the future developments in FDS will focus on this topic. The issues affecting the modeling of fire suppression systems have already been addressed in the sections concerning combustion (modeling gas phase suppression) and Droplets, Particles and Dispersed Second Phase (description of water sprays). The modeling of water mist systems in particular should be addressed because they are rapidly becoming common. In addition to the basic physical submodels of FDS, improvements are needed in the description of the suppression systems. For example, the simultaneous discharge from several sprinkler or water mist nozzles have effects on the pressure of the pipe system, and therefore on the mass flow of suppressant.

Radiation

FDS version 1 used heat emitting particles to represent a fire. Radiation transport consisted of Monte-Carlo ray tracing from the particles to surfaces, essentially painting the radiative fraction of the fire on surfaces. Hot surfaces and hot gas layers were not emitters.

Early in the development of FDS 2 it was recognized that FDS was being used for conditions where the participation of surfaces and gases was important, and thus, a new radiation model was needed. The model that was eventually developed consisted of using RADCAL (added by Jason Floyd while a NIST post-doc) to generate look-up tables of absorptivity and a Finite Volume Method radiation transport model (developed by Simo Hostikka of VTT). This approach consumes around 25 % of the computational time and for the simpler versions of the combustion model generally performs well.

Ideally, we would prefer to compute absorptivity during the actual simulation rather than attempting to use look-up tables which do not readily support the needs of the more complex combustion models being developed. However, current validation efforts indicate that errors resulting from the use of the look-up tables are not large (indeed it is not clear if the errors are any larger than those in the experimental data we use for validation) so any new approach to generating absorptivities or radiation source terms should not consume significantly more time than currently used by the radiation transport. For example, RADCAL could be called for each grid cell each time radiation transport is computed, but that would consume a tremendous quantity of resources.

The Basic Flow Solver

The core mass, momentum and energy transport algorithms within FDS are fairly robust and efficient, and there are no plans to change them. This includes the Smagorinsky form of LES, the basic pressure-velocity coupling, and the low Mach number assumption. However, we still need to improve the coupling of the pressure solver across mesh boundaries in a multi-mesh (often called "multi-block") simulation. Discontinuities in the pressure solution across these boundaries is currently the biggest speed bump to broader, more reliable use of the multi-mesh approach. The hope is to determine a computationally efficient way to perform this coupling that still permits the use of the direct pressure solver (CRAYFISHPAK) which is one of the primary reasons why FDS is fast.

IT Issues and User Support

Current Activities: Led by Bryan Klein of NIST, FDS 5 is being developed using fairly common software development tools. The source code and manuals are stored in an external repository and can be accessed by all members of the team. User support is provided via our old mailing list, and a new Discussion Forum hosted by GoogleGroups and and Issue Tracker hosted by GoogleCode. In short, a typical FDS user merely goes to a website and enters a question or bug report, and we all can answer the question or track the bug until it is fixed. This has dramatically improved our ability to help users, as before it was simply one or two people answering emails.

FDS has always exploited the availability of cheap, fast computers. Now is no different, but the challenges are growing. No longer do we merely provide a simple executable for a single Windows-based PC. We now have single CPU and parallel versions of FDS that run on virtually any operating system. Keeping up with changes in hardware and software keep Bryan Klein busy almost full-time. The FDS users are becoming more sophisticated in their use of parallel computing, CAD software, and so on. There are currently two graphical user interfaces (GUIs) for FDS - one on the market and one in development. We need to stay abreast of these activities or we might start to see a proliferation of old versions of various pieces of software. It is absolutely essential that the user community keep up to date with the latest versions of FDS, because we cannot support multiple versions.

Going Our Way?

FDS has benefitted from the contributions of many people from various sectors of the fire protection community. We encourage collaboration, even though direct funding from NIST and other sources is very limited. Graduate students interested in working with FDS should contact us before embarking on a project so that we can discuss whether or not it meets some of the criterion for improvement that we set out above. Some of the potholes students have encountered in the past are:

• working with old versions of the source code
• trying to improve parts of FDS that validation work has shown to be working well (e.g. US NRC and EPRI, NUREG-1824)
• obsessing over turbulence models
• developing unnecessarily detailed heat transfer and boundary layer algorithms

The Discussion Group and Issue Tracker should be the first line of communication between someone working on FDS and the user community. We ourselves spend a significant amount of time everyday participating the discussions, answering questions, fixing bugs, and so on. Any student who is working with FDS should be monitoring this traffic, and participating as well.