-
Notifications
You must be signed in to change notification settings - Fork 1
FDS Road Map
This wiki describes research plans for enhancing the fire model FDS.
FDS and Smokeview were officially released in 2000. However, for two decades prior to 2000, 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 (large-eddy simulation), 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.
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 model to close the LES subgrid stress 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 basic combustion model and the finite volume radiation transport solver, respectively. These improvements were implemented in version 2 (2001). Versions 3 (2002) and 4 (2004) saw gradual improvements in these routines, along with the introduction of parallel processing and various fire-specific features.
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. Version 5 was released in 2007. It included a major overhaul of the input parameters and constructs. Over the next three years improvements were gradually added, and in 2010, work began on FDS 6, which began beta testing in the fall of 2012. FDS 6 was officially released in the fall of 2013.
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:
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 prediction 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.
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.
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.
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 core mass, momentum and energy transport algorithms within FDS are fairly robust and efficient, but up until the introduction of FDS 6, these algorithms exhibited over-shoots and under-shoots of scalar quantities in regions of steep gradients, a condition common to second order accurate finite-difference schemes. A new TVD (Total Variation Diminishing) transport algorithm for the density and species mass fractions has been implemented in FDS 6. The algorithm includes several different types of flux-limiting finite difference schemes that eliminate the excessive over and under-shooting in regions of steep gradients. One drawback of the new schemes, however, are that they require more CPU time per time step, and efforts are on-going for reducing the cost.
There are no plans to give up the low Mach number assumption or the large eddy simulation (LES) approach, but FDS 6 uses a different form of LES than was used in versions 1 through 5. This is referred to as Deardorff's model, but FDS 6 also includes several LES options, including the dynamic Smagorinsky model and the old default constant coefficient Smagorinsky model.
Potential Research Topic: Explore ways to increase the speed and decrease the memory requirements of the new transport schemes. Use the FDS Validation Suite to demonstrate equivalent or improved accuracy of the new algorithms.
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. FDS 6 includes a way to tighten the tolerance of the pressure solution across multiple meshes by iterating the single mesh solutions, but this method is relatively slow.
Potential Research Topic: Improve efficiency of parallel version of FDS on very large computing clusters (100+ processors). Document scalability using conventional parallel computing terminology. Priority: High.
A key component of the FDS algorithm is the projection scheme which enforces the velocity divergence constraint implied by combining the continuity and energy equations. The projection requires the solution of an elliptic PDE and this constraint poses problems for efficient distributed memory parallel computing. We are working in collaboration with Susanne Kilian of hhpberlin to develop a fast global pressure solver. We are also pursuing research in OpenMP parallelization on shared memory machines. This effort is led by Christian Rogsch of the University of Wuppertal. As we move forward both these approaches will be extremely important in establishing a robust algorithm with optimal performance.
Potential Research Topic: Adaptive Mesh Refinement (AMR). Explore strategies for structured grid refinement in a low-Mach LES solver and develop refinement criteria for fire applications. A key aspect of this research will focus on load balancing for parallel calculations. Priority: Medium in the short term; High in the long term.
While working towards FDS 6, it became clear that the mixture fraction framework was becoming a limitation. It relied upon explicit assumptions about the fuel chemistry and the combustion reaction that resulted in inflexibility in the framework. Users and developers were locked into to the assumptions without any easy method of redefining them other than re-writing portions of the input processing and combustion routines. Starting in FDS 6, a new lumped species framework has been developed to enable the user to have greater flexibility in specifying minor product species such as toxicants while still maintaining some level of computational efficiency. The new lumped species framework will be coupled with a more flexible combustion routine to enable easier exploration of reaction schemes.
Current Activities: Led by Jason Floyd of Hughes Associates and Craig Weinschenk of NIST, we are moving towards a more general combustion formulation than previous versions. The method makes use of "lumped" species mixtures that reduce the number of species transport equations that need to be explicitly solved.
The new algorithms have 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.
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. Starting in FDS 6, there is an optional soot deposition model implemented by Jason Floyd and Kris Overholt that uses empirical correlations to predict the rate of soot deposition onto walls.
Potential Research Topic: Reduced and full-scale experiments quantifying the mass of soot depositing onto surfaces. Experiments should isolate the various deposition mechanisms (thermophoresis, gravitational, and turbulent) to enable independent V&V of the specific mechanisms. Priority: Medium
Potential Research Topic: Algorithms for the oxidation of soot deposited onto surfaces. Priority: Medium
Potential Research Topic: Visualization techniques to shade surfaces based upon the amount of soot deposited. Priority: Low
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.
Potential Research Topic: Identify and perform simulations of full-scale or reduced-scale experiments where the fire self-extinguished or was extinguished by a diluent or suppressing agent. The results could be added to a new chapter on Suppression in the FDS Validation Guide. Priority: High.
Modeling of ignition or re-ignition (deflagrations and backdrafts) is also a long term goal. The new lumped species approach 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.
Potential Research Topic: Modeling backdraft experiments, noting that FDS currently lacks certain necessary physical mechanisms that would need to be developed in cooperation with us. Priority: Medium.
Potential Research Topic: Algorithms for the re-ignition of combustible gasses that have undergone extinction. Currently, FDS will burn fuel if it encounters sufficient oxygen whether or not some form of ignition source is present. Priority: Medium.
Potential Research Topic: Treatment of turbulence/chemistry/radiation interactions. Some potential approaches include: off-line tabulation of reduced-dimensional models, e.g. LEM (linear eddy model), prescribed PDF (probability density function), and direct quadrature method of moments (DQMOM). Note that a simple version of DQMOM is currently implemented in FDS 6. Improvements in this area would be in modeling the subgrid scalar variance (algebraic or transport equation) and using this as a quadrature constraint to set a cell's initial state of mixing. How important is this issue in fire calculations (especially for CO and soot formation)? And how can these methods be incorporated in FDS in a tractable way (see Obey the Speed Limit above). Priority: Low to Medium, depending on how well it can be shown that any of these techniques or methods improves the current, default operation of FDS.
Potential Research Topic: Explore reduced mechanisms for the prediction of CO and soot formation and oxidation (identified as a need by the International Forum of Fire Research Directors (FORUM)). These mechanisms must be computationally efficient and capable of being used for the room and building scale modeling efforts typically undertaking by practicing engineers.
Simo Hostikka of VTT, Finland, developed the basic framework for describing heat conduction and pyrolysis within bounding solids. The algorithm includes 1-D heat transfer through multiple layers of materials, and a generalized pyrolysis algorithm to allow for multiple materials undergoing multiple reactions.
The major hurdle in this area is still the availability of material property data. After 30 years of model evolution, there is now a fairly broad consensus on the basic approach for modeling solids within a fire model, but there remains the major hurdle of measuring the necessary material properties.
Potential Research Topics: With an eye to the emerging trend of standardizing the measurement and interpretation of solid phase thermo-physical properties, demonstrate that we can, or cannot, model the decomposition and burning of various solids and liquids. For classes of solids that do not seem amenable to standardized measurement methods, suggest alternatives and demonstrate practicality and effectiveness. Priority: High.
Potential Research Topic: Create FDS models of standard or common non-standard test methods for the purposes of pyrolysis model validation and parameter estimation. Priority: High.
Potential Research Topic: Generalize the two-flux model based on the Schuster-Schwarzschild approximation for cylindrical and spherical coordinates.
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.
Potential Research Topic: Simulate standard ADD (Actual Delivered Density) and RDD (Required Delivered Density) sprinkler approval tests. Add to FDS Validation Guide chapter on Suppression. Priority: Medium.
Potential Research Topic: Overhaul and simplify the droplet evaporation algorithm in the source code file part.f90. Currently, the heat and mass transfer is calculated droplet by droplet, but it would be more effective to process the droplets en masse rather than one by one. We need to also check the validity of the Claussius-Clapeyron evaporation model at temperatures ranging from ambient to flame. Priority: High.
Potential Research Topic: Improve modeling of dense water spray. Currently, water droplets do not interact each other. Priority: Medium.
FDS has always been tied to structured Cartesian block geometries. Given that the zeroth-order specification of the heat release is often the best that can be expected from a fire model, the block geometries have not typically been viewed as a limiting factor. However, at a minimum, it is often a cumbersome process to input geometric information for something as simple as a cylinder. As we begin to tackle more challenging problems with FDS (fire spread, atmospheric flows) we are gradually being forced to account for second-order physical processes and we therefore require second-order numerical treatments at solid interfaces. For example, to first order the spread of wildfires is dictated by surface winds. In mountainous regions where these fires occur, the terrain is complex and simulations using stair-stepped grids lead to spurious fluctuations in the wind field.
To address this issue we are experimenting with a new geometry module which creates a solid surface interface based on triangular facets as the primitive geometry element. We are currently collaborating with Global Engineering and Materials, Princeton, New Jersey, to develop a two-way coupling between FDS and a 3D finite-element solver (GEM uses Abaqus, but our interface is not Abaqus-specific). The research needs in this area involve eliminating numerical diffusion of heat and mass in immersed boundary methods (IBM) without resorting to a formal cutcell method. Cutcell methods suffer from problems related to small unstructured cells that are inevitably created when a slanted surface interfaces with the Cartesian mesh. These cells cause time step contraints and stability problems unless they are merged with neighboring cells. This merging process and all code needed for the unstructured geometry engine is inherently messy.
Potential Research Topic: Chimera (overlapping) mesh methods for heat and mass transport near complex surfaces. Priority: High (short term).
Potential Research Topic: Validate radiation model for complex geometry. Priority: Medium (short term), High (long term).
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.
Potential Research Topic: In cooperation with VTT, Finland, better characterize the spray of fine mist and validate FDS for mist suppression. Priority: Medium.
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 20% 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.
One particular challenge of using the gray gas approximation is the need to specify a characteristic path length of radiation. The path length is an important parameter for gases with strongly non-uniform absorption spectra, such as water vapor. If the scenario includes many different geometrical length scales (such as a tunnel), defining a single value for the path length becomes cumbersome.
Potential Research Topic: Examine the means to reduce the ray-effect.
Potential Research Topic: Determine a better way to define the radiation length scale or another way to calculate the 'path-mean' absorption coefficients from the narrow-band data.
FDS is written in Fortran 2003. Both OpenMP and MPI are implemented in a single release executable.
Potential Research Topic: Perform scaling studies to determine the optimum use of OpenMP and MPI over a network or cluster of multi-core, multi-processor nodes.
The International Forum of Fire Research Directors (FORUM) has identified the ability to determine the relationship between aspects of a buildings design and the safety of building occupants as one of the top five research priorities. For whole building analysis, environmental control systems are a major contributor to the movement of toxic combustion products through a building. FDS 6 will be released with a first generation HVAC model as a first step in assessing the impact of HVAC systems on smoke movement. Further development will be needed to account for control system behavior, transport delay of species, combustion within HVAC systems, deposition of aerosols, and to support modeling of in-duct, fire detection systems.
Potential Research Topic: Algorithms for determining the flow loss through filters exposed to soot, simple approaches to model duct heat transfer, computationally efficient approaches to model mixing and combustion processes within HVAC duct networks, accounting for transport delays and mass storage in ducts. Priority: Medium.
Potential Tool Development: User interfaces for developing the inputs to model complex HVAC networks.
In the context of FDS+Evac development we use the Fractional Effective Dose methods presented in the SFPE Handbook of FIRE Protection Engineering (Purser, 2008a). For these computations, we need a capability to define arbitrary toxic species as products of combustion. The concentrations of these species are usually so small, that they need not to be considered in gas transport computation (with possible exceptions).
Since some gases are produced in well-ventilated conditions and some in under-ventilated, it might be reasonable to let the user define if the gas appears in fixed proportion to major carbon carrying species, as discussed in Combustion section above. Assuming the CO model works, this would introduce under-ventilated products.
In FDS6, the FED/FEC algorithm will recognize the following species:
Asphyxiants: CO, HCN, Low O2, CO2
Irritants: HCL, HBr, HF, SO2, NO2, CH2CHO (acrolein), CH2O (formaldehyde), X(user defined)
Potential Research Topic: Yields of these species under different conditions, in relation to CO2 or CO production.
Current Activities: FDS is developed and maintained 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 a Discussion Group hosted by Google Groups and an Issues page hosted by GitHub.
Contributions to FDS and Smokeview can be made by forking the FDS and Smokeview repositories and submitting pull requests to the code developers.
FDS has benefited 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.