Introduction

Ceiling pockets have become a common architectural feature prompting the need for sprinkler protection criteria. This report establishes benchmarks for ceiling pockets where the absence of sprinklers does not decrease the level of protection provided by the sprinkler system.  An abbreviated version of this report was submitted with the proposal for NFPA 13 resulting in criteria being defined for ceiling pockets.  

Overall Approach

It was determined that in the absence of sprinklers in the ceiling pocket, the remaining sprinklers at the lower ceiling elevation must continue to provide adequate coverage for the entire floor area including the floor area under the ceiling pocket. In other words, the sprinklers at the lower ceiling elevation cannot exceed their maximum allowable spacing. This requirement limits one dimension of the ceiling pocket to some value less than the maximum sprinkler spacing. It was also determined that in order for the sprinklers at the lower ceiling elevation to provide an acceptable level of fire protection for the entire area, the activation times must not be excessively delayed due to the ceiling pocket. The sprinklers must activate in the same time frame as produced by standard response sprinklers in a room without a ceiling pocket. The goal of this undertaking is arriving at dimensions for a ceiling pocket that would not result in excessively delayed activation times and allow for adequate floor coverage by the sprinklers at the lower ceiling elevation.

Description of Fire Model

To evaluate the question of what effect a ceiling pocket has on sprinkler activation times, we utilized a fire modeling program.  The older zone based fire models do not have the ability to evaluate situations in such fine detail.  In order to solve this problem, we utilized the latest simulator in fire modeling based on computational fluid dynamics.  It is the Fire Dynamics Simulator (FDS) which divides the room into thousands of cells and models the impact of a fire between all cells (see Figure 1). FDS was released January 2000 by the National Institute of Standards and Technology, a division of the US Department of Commerce. (See Figure 1)  The software program actually incorporates two programs.  One is Smokeview, which allows the user to see a graphical representation of fire conditions with respect to all 3 axis and throughout the evaluated time period (See Figure 2).  The other program is the fire simulator program called Fire Dynamics Simulator (FDS), which is a compiled program with no interface of its own, unlike other fire models such as FPETOOL.  Instead, the user must write a data file in source code (See Figure 3).  This file contains all physical parameters for the room including obstructions and dimensions, fire source data, data capture guidelines, and other miscellaneous parameters. On PC's, this is accomplished through MS-DOS and writing a file with a ".data" extension.  Once the source code is written, it is fed into the FDS executable program that converts the data for viewing in Smokeview. Additionally, some data such as sprinkler heat temperatures at different time steps are given for use in spreadsheets. 

Although FDS is far more accurate and refined than the older zone fire models, it takes significantly longer to complete a simulation.  We performed approximately 60 simulations with each taking between 4 to 8 hours (with a 500 MHz processor). 

Description of Room

Two basic rooms were used in the modeling. The first room had a lower ceiling elevation of 3m (9.8ft), width of 15m (49.2ft), and length of 20m (65.6ft). The second room had a lower ceiling elevation of 4.5m (14.8ft), width of 12m (39.4ft), and length of 18m (59.1ft).  The dimensions are shown first in metric since FDS uses it.

To simulate a large open room, which is a more conservative arrangement, walls were omitted from the rooms. In other words, the length and width of the room only establishes a ceiling area for the model. With this arrangement, once heat passes the sprinklers and reaches the ceiling area boundary, it essentially disappears and no longer effects the room. This is similar to a long corridor or a large office area. By contrast, creating walls in the room allows heat to accumulate and build an upper gas layer. Excluding the impact from the upper layer provides conservative results and allows the evaluation to be applicable to all situations and not just smaller rooms.  For the 2nd case of the 4.5m ceiling, the room was shortened because it was realized that the room doesn't need to extend much further than the sprinklers to yield accurate results. This allowed for faster computation time, especially since the 2nd case had a higher ceiling and needed more grid cells.

The ceiling was left with its default setting of being an adiabatic surface. In actuality, a ceiling will absorb heat. However, two conditions validate using this property.  First, all results are being compared to the same condition, so the time difference between activations should remain the same even if the ceiling properties were changed.  Second, there are any numbers of materials of which a ceiling could be constructed, and we could not consider all of them. One ceiling surface condition was selected for all simulation runs, which maintained consistency in activation time differences.

Description of Pockets

Pockets with varying dimensions were evaluated. In each case, the ceiling pocket was positioned in the center of room with the long length dimension running along the longer "y" room dimension. Maximum and minimum limits were:

Max width: 3.67m (12.0 ft)               Min width: 1.0m  (3.3 ft)

Max length: 14m (45.9ft)                 Min length: 5.0m (16.45ft)

Max depth: 5m (16.4ft)                   Min depth: 0.2m   (0.7 ft)

Other Parameters of the Model

An ambient temperature of 21C (70(F) was selected for all simulations. Sprinklers were actually set up to act as heat detectors because a sprinkler designation in the FDS program would discharge water, which would influence the other sprinkler/detector activation time. It is important to note that heat detectors and sprinklers behave exactly the same with regard to  response/activation time.

Grid cells for the 3m ceiling were 0.33x0.33x0.17m (1.1x1.1x0.6ft).  This results in a total number of grid cells of at least 81,000 (depending on pocket depth).  Results were compared against a grid doubled in resolution along the x & y axis.  This produced a grid of cube size 0.17x0.17x0.17 and 324,000 cells in the model.  This produced resulting activation times 7 seconds less than the coarser grid.  Finally the resolution was again increased by 75% to cells 0.15x0.16x0.11m for a total of 562,500 cells.  However, the time was only one second different (1 second less) than the previous grid resolution. It would appear that the 0.17x0.17x0.17 grid is the critical grid resolution where results are grid independent. However, CPU time for this grid was triple the main coarse grid that was used for all the models. CPU time for these finer resolutions was not feasible.  The trade-off for using the coarse grid is to introduce a +/- 10 second margin of error for grid resolution inaccuracies. 

The 4.5m ceiling used a finer grid of 0.2x0.2x0.17 to help eliminate grid inaccuracies while not hogging CPU time. The estimated margin of error for the 4.5m ceiling because of grid dependency is +/- 5 seconds. The re-calibrated room used a grid of 0.2x0.2x0.17 (although sometimes 0.1 in the z direction for greater accuracy when needed) and should have an error factor of +/-7 seconds. 

A margin of error was included in the comparison of activation times.

Description of Fire

This endeavor sought to address ceiling pockets in light hazard occupancies only. One of the most common and severe fire scenarios in this occupancy involves a burning couch (or other upholstered furniture). Data from NIST for a peak heat release rate and growth factor was utilized. A typical couch burns at a peak HRR of 1645 KW/m2 with a medium growth curve. A fire starting at 0 MW at t=0 and ramped up according to a standard HRR=(t2 was entered into the source code. The couch had dimensions of 2x1m (long dimension parallel to long dimension of room & pocket) and was 1 m high. The fire was placed in the center of the pocket with its 2m axis parallel to the y dimension.  This avoided the fuel package extending beyond the edge of the ceiling pocket and ensured the entire fire plume was captured within the ceiling pocket. 

Part way through the evaluation the dimension of the fire source was modified to 1x1m.  The change was made because sprinkler activation was occurring prior to the time required to reach full involvement of the couch surface.  By assigning a larger surface area to the initial HRR, the fire plume characteristics were being affected.  All other fire parameters were left at default values.  Additionally, two runs were performed with a more realistic HRR curve -- quick ramp up times with a rapid decline. The comparative results were not significantly affected by the more realistic fire curve.

Description of Sprinklers

Sprinklers were placed midway in the room in the "y" dimension, aligned as close as possible to the center axis of the fire. They were placed at varying distances from the edge of the pocket, ranging from 0.15m (0.5ft) to 1.0m (3.3ft). However, a majority of the runs were performed with the sprinkler at 0.67m (2.2ft) from the edge. Sprinklers were set for activation temperature of 68.3C (155F) and most were set to QR specifications (RTI of 50). A select few were set to standard response RTI's of 177. Sprinklers were normally 4.4m apart (14.5 ft).  With the fire located in the middle of the pocket, it allowed sprinklers on both sides of the pocket to be located symmetrically.  See Figures 4A and 4B for sprinkler locations (shown as dots) and fuel/pocket orientation.

Findings

To determine an acceptable activation time delay, we compared the response time of quick response sprinklers in a flat ceiling to the response time of standard response sprinklers in a flat ceiling and used that time difference as the acceptable delay. Then, we compared a room with a pocketed ceiling to a room with a flat ceiling (set at the lower ceiling elevation of the pocketed room). This directly indicates what the delay would have been without a pocket.

For earlier cases, we used the time difference between QR and SR from DETACT.  The 3m ceiling (t is 82 seconds.  The 4m ceiling with the 2x1 fire (t is 70 seconds, and the 1x1 fire (t is 83 seconds.  Margin of error ranges from 10-20 seconds, so about 20 seconds should be subtracted from the (t for a safe margin of error and safety factor.  As an additional safety factor, the activation time is actually the average for both of the symmetrically located sprinklers on each side of the pocket.  See Table 1 for the activation time for different evaluation parameters.

Unusual Findings

One result should be highlighted.  When the ceiling pocket has a depth in the range of 2-7 feet, there is an inverse time relationship.  It would be expected that as the depth is increased, the activation time would likewise increase. This holds true for pockets up to about 3 feet deep. However, at about 3 feet the activation time peaks and actually decreases until ~7ft. After 7 feet, activation time again increases.  We tested to a pocket depth of 13 feet and the most demanding scenario was the 3 foot deep pocket. This occurrence is attributed to the dynamics of the hot gases in the fire plume. As the plume rises into the pocket, it does not simply stratify and calmly fill the pocket as is commonly thought. The velocity and momentum of the gases are such that the plume "travels" out to and down the outer boundaries of the pocket. Once the hot plume reaches the bottom of the pocket it encounters the cooler air not yet affected. At depths less than 3 ft, the plume rolls outside of the pocket.  After 3 ft, the plume rolls inward and back toward the center of the pocket. At this point, a distinct boundary layer is formed and acts somewhat like a ceiling (a ceiling at the same elevation as the lower ceiling) which serves to send portions of the plume out beyond the pocket to the surrounding sprinklers in the lower ceiling. This is illustrated by Figures 5A and 5B for both shallow and deep pockets.  Again, this occurs when the depth of the ceiling pocket ranges from 3ft to 7ft.

Summary / Conclusions

Based on the findings, we concluded that a pocket with the following dimensions is acceptable:

Width - The maximum spacing of the lower ceiling sprinklers

Length - 8m  (26.2 feet)

Height - 4m  (13.1 ft)

 

Download Table 1.pdf » (44kb - must have Adobe Acrobat Reader.)

About the author:

Roland Huggins, P.E.

Roland Huggins is the Vice President of Engineering and Technical Services for the American Fire Sprinkler Association.  He graduated Cum Laude from the University of Maryland and is registered in Fire Protection Engineering.  He is currently on the following NFPA committees: NFPA 13 Correlating Committee, 13 Discharge Criteria, 22, 25, 230, 291, 5000 (Building Code) Correlating Committee, and 5000 Industrial, Storage and Miscellaneous Occupancies.  National activities include:  Board Member of the NFPA Building Fire Safety Systems section, on the  NFPA Research Foundation research advisory council for suppression systems and a variety of their research projects, on the UL Standards Technical Panel responsible for eight UL test standards, on the NICET committee that developed the Testing and Inspection Program, and on the ASSE working committees for test standards 1013 and 1015 (RPZ and DCV backflow preventers).  SFPE activities include being an officer of the local chapter and participation on the task groups that developed the Performance Based Design Guide and the Sprinkler Design Course for Engineers. He has delivered many presentations (regularly at NFPA conventions), written multiple articles on sprinkler systems, and participated in editing the NFPA Inspection Manual and NFPA Sprinkler Handbook.