Fire Probablistic Risk Assessment Challenges: Fire Modeling
– By Dr. Martin G. Plys, Vice President, Waste Technology; Fauske & Associates, LLC –
Introduction
Fauske & Associates was invited to prepare presentations for the PWR Owners Group Third International Technical Workshop in London, England, April 28-29, 2009. We provided a presentation on Fire Probablistic Risk Assessment (PRA) Challenges for Nuclear Power Plants related to our areas of current customer support and expertise, Cable Selection/Circuit Analysis and Fire Modeling. This article provides selected lessons learned from our presentation on fire modeling challenges. This article is based on direct experience for fire modeling in both BWR’s and PWR’s, and the experience applies equally to U.S. and foreign installations.
From the U.S. perspective, fire PRA methodology is outlined in a key document created jointly by EPRI and NRC, EPRI-1011989, NUREG/CR-6850. Tasks 8 and 11 pertain to fire modeling, and cover scoping and detailed modeling respectively. Challenges in applying the fire modeling methodology are discussed here following the order of work flow in practical application.
Scenario Definition
Defining a fire modeling scenario means selecting an initial ignition source, targets that could be damaged or catch fire, the geometry of the room (or rooms) and details of relationships between ignition sources, targets, and other obstacles, the geometry of flow paths, the configuration of doors and vents that may vary with time, and the influence of engineered systems such as forced flows and fire suppression.
Scenario selection involves a priori decisions that may drive the selection of appropriate software, and may drive whether or not a useful answer can be obtained. Selecting the appropriate ignition source and its characteristics (see scenario model and inputs below) can be a challenge even though the list of ignition sources is usually defined by a walk-down. For example, it is desirable to limit the number of scenarios that must be evaluated, by finding appropriate bounding scenarios or scenarios that represent the threshold between success and failure. If targets cannot survive a limiting ignition source (combustible load or location), then scoping modeling can be used to efficiently identify the threshold combination of load and location.
Electrical cabinets are often arranged in rows that are linked together in their interior because they share a power bus; this is typical in relay and switchgear rooms. The target of interest is usually a single rack location in a cabinet that provides a control or power function. In such a case, the limiting ignition source is the nearest linked cabinet with non-trivial combustible load (mainly insulation). A fire in an adjacent cabinet is typically not survivable; a fire several cabinets away could propagate toward the target but perhaps this would take sufficient time that manual suppression could be credited. Scoping modeling could be used to show whether or not a fire in nearby cabinet rows could directly fail the target, and detailed modeling could be used to show the time to damage. Sometimes the target cabinet receives power from a different cabinet the same room. That second cabinet is typically linked together with other power supplies and a local transformer. In such a case, the appropriate ignition source to consider is a cabinet that affects the power supply.
Fire Modeling Software Selection
Three basic levels of modeling software are used in practice, as described next. The essence of the challenge is that no single method is perfect, so that in practice all three approaches are usually followed in order of complexity for a given application.
1. Scoping models, which are simple equations usually rendered in a spreadsheet. These models are fast to apply and their results are easy to explain. Scoping is used to rule out scenarios that result in obvious success or failure, and focus on borderline scenarios.
2. Zone models, which consider “tank and tube” geometry with a “smoky layer” of combustion products and soot that overlies a “lower layer” of air. Zone models vary in their complexity. For example, both CFAST (available from NIST) and FATE (developed by FAI) are zone models, but FATE has a much broader scope and can consider multi-dimensional heat conduction in targets, and also considers aerosol behavior, so it can track the transport and deposition of radionuclides and toxic species. Zone models are valuable when the boundary conditions reliably allow prediction of smoky layer elevation and temperature, for example, in a room with well-defined flow paths to other rooms. Zone models are also the most practical to analyze migration of smoke and contamination; for example, we have run FATE models to analyze the pressure distribution and aerosol behavior in facilities with over one hundred rooms.
3. CFD (computational fluid dynamics) models, which solve for detailed distributions of temperature and composition in multiple dimensions. The most practical CFD option in the U.S. is FDS developed by NIST. The advantage of using FDS is to obtain good resolution in the vicinity of a fire, although usually at the expense of significant computation time. Tedious attention to detail is required for CFD application because boundary conditions and grid size can significantly alter results.
Scenario Model and Inputs
A scenario model and inputs means definition of details of the heat release rate (HRR) history from the combustible load, appropriate peak HRR, and growth and decay characteristics, definition of how the geometries of the room, fire, targets, and other obstacles are rendered, definition of pyrolysis characteristics that influence the potential for propagation, and definition of performance of engineered systems.
Some challenges related to the HRR are:
• Limited data are available for electrical cabinet fires. The 98% peak HRR value used in the U.S. is about a factor of two lower than an experimental result obtained in Europe and presented to the post-SMiRT conference on fire safety in nuclear power plants and installations in 2007.
• Cabinet peak HRR data cannot be easily or reliably scaled by obvious factors such as height, volume, surface area, etc. based upon reported information.
• Motor HRR values recommended for use in the U.S. are based upon cabinet data and hence are even more unreliable.
• For cables, the pyrolysis rate law is usually not available. In special cases, we have been able to find TGA/DTA data that allow us to construct Arrhenius parameters. Without a reliable rate law, it is not possible to predict the spreading rate of a fire.
In general, the difficulty with uncertainty in the HRR is overcome by examining a range of peak HRRs, but preserving the combustible mass. This strategy works for detailed modeling because success is usually the result of the ability to survive a transient with limited overall energy input. The pyrolysis rate law issue has been overcome as stated above when data allow an Arrhenius curve to be constructed, and in addition, lower and upper bound rate laws can be used to quantify any uncertainty and sensitivity of results to these parameters.
Some challenges related to room and target inputs are:
• CFD model results can be sensitive o the volume, shape, and placement of interior items in a room. Emphasis should be placed on getting the correct free volume and the volume versus elevation relationship, and on correct elevations of obstacles that partition rooms, beam pockets, and flow paths.
• Target temperature response is generally difficult to model in terms of a detailed interior temperature distribution due to the complicated geometry, such as in a multiple conductor cable. For cables, data and models can be tuned so that the temperature inside the outer jacket can be assigned a value corresponding to failure.
Interpretation of Results
The key challenge to apply fire modeling results to draw conclusions is simply, how do you know if the results are correct, conservative, or non-conservative? This challenge is met through the experience of the analyst, and familiarity with experimental data that are used to validate the fire models. The analyst must understand the similarities and differences between experimental configurations and scenarios, and those plant configurations and scenarios that are being modeled. As an example, the failure criterion for a cable is a threshold temperature above which failure is assumed and below which survivability is assumed. If the cable is located in a place where a strong temperature gradient is predicted to exist, then clearly the precise location of the cable, or the precise slope of the gradient, can both influence the result of the fail/survive criterion. On the bottom line, in-depth experience is required, and ideally the analyst will have written and validated his/her own models, because this provides a basis for knowing the strengths and weaknesses of the various available software.
Fire Risk Services Team Experience
The preceding example challenges in fire modeling are derived from direct experience of FAI and the Westinghouse Fire Risk Services Team (FRST). The FRST can provide a full range of services related to fire modeling. Levels of attack the FRST can provide include modeling to support a full fire PRA, modeling to support safe shutdown analysis, a “triage plan” to determine where fire modeling would certainly resolve issues, could potentially resolve issues, or cannot realistically provide benefit, or a management plan to determine realistic costs and schedule associated with such undertakings given a plant-specific state of information. For further information please contact Dr. Martin G. Plys at plys@fauske.com.