Fire Safety Journal 71 (2015) 279–286

Contents lists available at ScienceDirect

Fire Safety Journal

http://d0379-71

n Corr10900 E

E-m

journal homepage: www.elsevier.com/locate/firesaf

Self induced buoyant blow off in upward flame spread on thin solidfuels

Michael C. Johnston a,n, James S. T'ien a, Derek E. Muff a, Xiaoyang Zhao a, Sandra L. Olson b,Paul V. Ferkul c

a Case Western Reserve University, Cleveland, OH, USAb NASA Glenn Research Center, Cleveland, OH, USAc National Center for Space Exploration Research, Cleveland, OH, USA

a r t i c l e i n f o

Article history:Received 7 May 2014Received in revised form15 October 2014Accepted 23 November 2014

Keywords:Buoyant blow offMaterial flammability limitsUpward burning limitOne-sided extinctionFlame spreadSIBAL fuel

x.doi.org/10.1016/j.firesaf.2014.11.00712/& 2014 Elsevier Ltd. All rights reserved.

espondence to: Glennan Bldg. MS 418, Caseuclid Avenue, Cleveland, OH 44106, USA.ail address: michael.c.johnston@case.edu (M.C

a b s t r a c t

Upward flame spread experiments were conducted on long thin composite fabric fuels made of 75%cotton and 25% fiberglass of various widths between 2 and 8.8 cm and lengths greater than 1.5 m.Symmetric ignition at the bottom edge of the fuel resulted in two sided upward flame growth initially. Asflame grew to a critical length (15–30 cm depending on sample width) fluctuation or instability of theflame base was observed. For samples 5 cm or less in width, this instability lead to flame blow off on oneside of the sample (can be either side in repeated tests). The remaining flame on the other side wouldquickly shrink in length and spread all the way to the end of the sample with a constant limiting lengthand steady spread rate. Flame blow off from the increased buoyancy induced air velocity (at the flamebase) with increasing flame length is proposed as the mechanism for this interesting phenomenon.Experimental details and the proposed explanation, including sample width effect, are offered in thepaper.

& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The upward flame spread configuration is often used as a me-tric for quantifying the overall flammability of materials andmerits further in depth study of the characteristics and boundariesof flammability. For example, NASA material flammability flightqualification test NASA-STD-6001 Test #1 [15] and UnderwritersLaboratories test UL-94V [23] both utilize upward flame spreadgeometries similar to the ones used in this work. The result ofNASA Test #1 is a simple pass/no-pass criteria based on whetherthe flame damaged region propagates upwards further than 15 cmon a 5 cm�30 cm sample. The results of UL-94V are categorizedbased primarily on duration of burning. Neither of these tests takeinto account detailed mechanisms of flame propagation or ex-tinction and are assumed to be a worst case flammability config-uration based on the fact that gravity tends to accelerate flamespread in the upward direction.

It has been previously shown that sample width can have asignificant effect on the characteristics of upward flame spread,including flame size, heat generation rate, and spread rate

Western Reserve University,

. Johnston).

[6,7,10,14,17,19,21]. It should be clear that sample width may alsoaffect the flame extinction limits despite the fixed width criteria instandardized flammability testing methods. In this work, upwardflame spread tests were conducted in normal gravity using aspecial composite fabric fuel. Several sample widths were used. Anunexpected but very interesting phenomenon, i.e. self-inducedflame extinction when the flame reached a certain length, wasobserved in many of the tests. The observation and a proposedinterpretation are discussed below.

2. Material and methods

The experimental setup is shown schematically in Fig. 1. Thisconfiguration mimics that of the NASA STD-6001 Test #1 and UL-94V. The thin fuel is sandwiched between four parallel stainlesssteel sample holders 0.035″ (0.889 mm) thick�2.5″ (6.35 cm)wide with adjustable exposed sample width of 2-8.8 cm andheight up to 1.8 m. The fuel used in this experiment is unique. It ismade from a simple weave fabric consisting of thread spun with75% cotton and 25% fiberglass strands with an area density of0.01805 g/cm2 and is about 0.31 mm thick. As the cotton burnsaway, the fiberglass component of the thread is left behindmaintaining the fuels structural integrity and shape. This inertmatrix simplifies the burning characteristics of the fuel by

Fig. 1. (a) Experimental setup: video cameras image both front and back of the thinfuel sample, a still image camera translates along the edge of the sample holderwith the flame. A remote camera strobe is located behind the sample pointed at anupward angle (left). (b) Zoomed in front view shows the flame tip, pyrolysis tip, andpyrolysis base. Smoldering of left over fuel can be seen below the pyrolysis base.Thin black stainless steel sample holders can be seen to the left and right of the fuelsample (right).

Fig. 2. (a) Unburned SIBAL fabric (left) and (b) inert fiberglass matrix after theflame has passed (right). Ruler notches are 1/32″ (0.79 mm).

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286280

preventing tearing, ripping, and curling of the solids surface aswould happen with other burning materials such as paper. A de-tailed comparison of the burning characteristics of this fuel withother materials is given in Ref. [8].

This custom-made fabric fuel (referred to as SIBAL), namedafter the experiment for which it was originally designed Solid

Inflammability Boundary At Low-speed [4] has been studied in alarge number of careful laboratory scale experiments in a varietyof environmental conditions [4,8,11]. The pre-burned SIBAL fuelcan be seen in Fig. 2a, and after the flame front has passed theremaining inert matrix is shown in Fig. 2b.

The leftover fiberglass matrix has been found to act as a flamearrester since the gaps between the threads are large enough toallow gas to pass through but smaller than the quenching dia-meter of the flame. This allows for the somewhat unique possi-bility of a one-sided flame existing on a thin fabric fuel [11]. Notealso that this fuel sample is sufficiently thin so that, in most ex-periments, it behaves as a thermally-thin specimen.

Fuel ignition was achieved using a 30 cm long 29 gage Kanthalhot wire powered with 3.7 amps (about 62 W) bent into a saw-tooth pattern alternating on the front and back of the fuel surfaceat the free bottom edge of the fuel sample. Ignition power wasremoved when a robust flame was observed.

The burning material is imaged at 30 frames per second withtwo 1080p high-definition video cameras perpendicular to thefront and back surfaces of the fuel. A third high-resolution stillcamera zoomed to the size of the flame views the fuel and sampleholder from the edge and moves along a track parallel to the flamepropagation. The still camera is capable of shooting an 8 frameburst in approximately 1 s. A strobe light located on the oppositeside of the sample holder illuminates the unburned fuel vapor orsmoke and captures the instantaneous smoke field. The stillcamera needs to integrate the light generated from the flame overapproximately 1/30–1/60th of a second in order to record theimage. However, the strobe illuminates the smoke field only forabout 1/1000th of a second to eliminate motion blur of the smoke.

The position of the pyrolysis front and pyrolysis base aretracked with custom imaging software by searching for brightnessthresholds along the centerline of the fuel sample. The threshold

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286 281

value changes between tests depending on illumination fromflame brightness and can affect the tracked pyrolysis position by afew pixels. However, this contributes to an uncertainty of only afew millimeters in absolute pyrolysis position and has almost noeffect on measurement of propagation velocity.

Fig. 3. Pyrolysis positions and length vs. time for (a) 5-cm wide SIBAL fabric, (b) 2-cm wide SIBAL fabric, and (c) 7.5-cm wide SIBAL fabric.

3. Results

Fig. 1b shows the front view of an upward spreading flame overthe SIBAL fabric. Flame and solid images are recorded by the frontvideo camera. Pyrolysis tip and base are identified for furthercomputer processing. One notices that for this thin solid, thepyrolysis base moves with the gas flame base when solid com-bustibles are mostly spent. A small amount of combustible residueremains attached to the fiberglass fibers and undergoes smolder-ing. The percentage of fuel left for smoldering is small, less thanapproximately 10% by mass (measured by microbalance in similartests).

Fig. 3a shows a typical case of flame ignition, growth, and up-ward spread of a 5-cm wide sample of our thin fabric fuel. Theposition of pyrolysis front and burn out front are plotted withrespect to time on the abscissa. In the very early stages, the pyr-olysis tip can be seen to propagate downstream (upwards) at anincreasing rate (curve is concave upward). Due to the physicallythin nature of the fabric used, the amount of solid fuel available topyrolyze is limited and the flame base moves upward when mostof the combustible is consumed. Note that during the initialgrowth stage, the flame tip will accelerate upwards while theflame base lags behind, remaining at the ignition location until thefuel begins to burn out. At some time, depending on conditions,the burn out zone (and therefore the flame base) will propagateupward. If the fuel burnout rate catches up with the pyrolysis frontpropagation rate, a constant flame length is reached and a steadyspread with a constant spread rate results [22]. In many normalgravity upward tests on wide samples, steady spread may not beobserved for the available fuel height. Steady spread with a lim-iting flame length is easier to find for narrow samples, in lowpressure environments, in partial gravity, and in concurrent purelyforced low-velocity flow in microgravity [1,2,3,5,22].

In the present, work one side of the flame is extinguished(blown off) during the flame growth as indicated in Fig. 3a for the5-cm wide sample. This happens at t�20 s when the pyrolysislength is about 35 cm long. The flame remaining on the other sidecontinues burning but quickly shrinks in length. The flameshrinking is due to the lack of flame heat input from the blown offside. Conceptually, it is as if the fuel thickness doubled. Respond-ing to this flame blow off and length shrinking event, the rate ofpyrolysis front propagation decreases followed by a decrease ofthe fuel burnout rate as shown in Fig. 3a (with some time delay).The one-sided flame then reaches a limiting length (approximately20 cm) and spreads steadily all the way to the end of the sample.

The 2-cm wide sample test shown in Fig. 3b exhibits a quali-tatively similar trend as in Fig. 3a for the 5-cm sample. However,the critical pyrolysis length for the one-sided blow off to occur isshorter (�18 cm). Both the initial flame growth rate and the finalone-side steady spread rates are slower than the 5 cm case. Con-sistently, the steady pyrolysis length is also shorter (aproximately12 cm).

For the 7.5-cm wide sample, Fig. 3c shows both the prolysisfront growth rate and the pyrolysis length increase with timecontinuously over the entire sample. There is no one-sided ex-tinction and steady spread rates are not observed. The flamespread is in the growth phase for the entire duration of the test.Flame tracking was terminated when the flame tip reached the topof the sample. Tests with an 8.8 cm wide sample show a similar

trend.These upward tests have been repeated many times. Since the

one-side extinction was observed for the narrower samples (2–5 cm widths), much care was taken to eliminate non-symmetriesin the experimental configuration and in the environment.Nevertheless, repeated tests show that extinguishment can occuron either side of the sample. After ignition, the initial flame

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286282

growth is observed to be very symmetric on both sides of thesample. One-sided extinction occurs only when the flame reachesa critical length (e.g. �18 cm for the 2 cm wide sample and�35 cm for the 5 cm wide sample). As an example, the 5 cm widecase was repeated over 30 times, about ten of which had ex-tremely symmetric ignitions. Full two sided propagation was onlyobserved three times on this sample size. The flame extinction istherefore not due to an ignition anomaly nor non-symmetry in theexperimental setup. We believe that the observed extinction is atrue physical phenomenon in buoyant flames. The next section isdevoted to the explanation.

4. Discussion

4.1. Proposed mechanism of self-induced flame extinction

Why does one side of the flame extinguish itself when it be-comes too long? And why does the shorter and weaker flame onthe other side of the sample remain?

Before going into our interpretation, let us examine a detailedsequence of photographs of the flame growth, extinction, spreadevents. Fig. 4 shows the edge-view pictures of the 5 cm widesample case which was shown in Fig. 3a. The camera translateswith the flame in consecutive images to keep the region of interestin view. The relative time stamps are shown at the top of eachphoto.

In Fig. 4a, a well established two sided flame is already present.The strobe light can be seen in the bottom of the frame, but as thecamera translates upward it will leave the field of view. In frame b,the flame base on the right hand side of the fuel sample is be-ginning to retreat downstream. Frame c shows the right hand sideflame base continuing to retreat but unburned fuel pyrolyzate(visible as white smoke) continues to leave the fuel surface. Framed shows the right hand side flame nearing extinction, the flame ispropagating into a region where the fuel surface is not adequatelypreheated, and the retreating flame has blown off.

The extinguished side of the fuel sample will not reignite due to

Fig. 4. One sided extinguishment from the edge view is shown. Gravity is oriented paral(RHS) flame begins retreating downstream, (c) RHS flame is continuing to retreat while phand side continues to escape, (f) the left hand side flame shrinks due to the reduced hepropagate to the end of the sample. (For interpretation of the references to color in thi

the inert mesh left behind which acts as a flame arrester. The bluecolored smoke seen near the bottom of the photos are the pro-ducts of smoldering fuel residue. It is unknownwhether the actualsmoke is blue (different compositions compared to the whitesmoke) or if it is artificial color cast created by the strobe light(�5200 K color temperature). Frame e shows the large amount offuel vapor which continues to enter the gas phase, but remainsunburned. The absence of flame heat feedback from the ex-tinguished side shortens the flame on the remaining side. One mayregard this as equivalent to an increase of sample thickness (sincethe flame is only on one side). The shortened flame shows goodstability with no indication of being near a blow-off limit and isable to spread upward all the way to the end of the sample. Soagain, why does the longer (seemingly more robust) flame extin-guish, but the shorter (weaker) flame remain?

To answer this question, we first examine the extinction me-chanism of a diffusion flame, specifically for a spreading soliddiffusion flame in concurrent flow. Fig. 5(a) illustrates severalhypothetical flammability boundaries using ambient oxygen per-centage as the ordinate and flow velocity at the flame base (theflame stabilization zone) as the abscissa. Different flammabilityboundaries represent different sample widths and it is expectedthat narrower samples will have a smaller flammability domaindue to three dimensional aerodynamics effects such as lateral heatloss, lateral fuel vapor escape (due to diffusion) [17], and lateralcold air entrainment [16,17,21]. Each boundary consists of twobranches: a high velocity blow off branch and a low-velocityquenching branch [3]. In a fixed ambient oxygen environment,quenching occurs when the oxygen supply rate becomes too lowand the weak low-intensity flame loses a large percentage of en-ergy due to radiation and conduction. This is an active area ofresearch interest in microgravity combustion. On the other end,high-velocity extinction is a flame stabilization problem. When theair velocity near the flame stabilization zone becomes too large,the flow residence time in the reaction initiation zone becomestoo small (or in nondimensional terms, the Damkohler numberbased on the stabilization zone size is too small) and the flamecannot be stabilized. The reaction zone is therefore blown off

lel to the sample length. (a) A two sided flame is established, (b) the right hand sideyrolysis is maintained, (d) RHS flame blows off, (e) unburned fuel vapor on the rightat flux to the solid, and (g) the left hand side flame reaches a steady length and wills figure, the reader is referred to the web version of this article.)

Fig. 5. (a) Qualitative flammability map bounded by quenching and blow off limitsfor the 5 cm wide sample. Qualitative blow off branches are drawn for 2 cm and7.5 cm wide samples (left). (b) Qualitative curve of the characteristic entrained airvelocity in the flame stabilization zone plotted with respect to time (right).

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286 283

downstream. In this extinction mode, the near limit flame has ahigh intensity since the air velocity and oxygen supply rate arehigh. Note that the relevant air velocity used to characterize ex-tinction is at the flame stabilization zone where fuel vapor firstmeets the upstream oxygen. In the upward spreading flame con-figuration, this occurs at the flame base. In upward flame spread,the velocity at the base is induced by gravity acting upon the entireflame and thermal plume and thus the buoyant velocity magni-tude at the base is affected by the size of the flame. Within certainlimits, it is expected that a longer flame will induce a largeraverage velocity at the base (see Appendix for computations tosupport this argument). A purely forced system, in contrast, has

the velocity magnitude at the stabilization zone which is con-trolled by the upstream condition and will not change with theflame length. In normal earth gravity, the buoyant induced flowvelocity at the flame base may become large enough so the blowoff extinction conditions are reached. To make sure that flameextinction mode in normal gravity is indeed from blow off and notfrom quenching, the following analysis is made. By balancinggravity acceleration and fluid inertia, gravity induced velocity u is�(gL)1/2, where g is gravitational level and L is the pertinentlength. Choosing L as the length of the flame stabilization zone,L�α/u, where α is the thermal diffusivity of air, this yields u�(αg)1/3.Depending on what temperature α is evaluated, u can be estimatedbetween 8 and 18 cm/s. Since this flow speed estimate is based onthe smallest pertinent length in the flame, it is the minimum buoyantinduced velocity in the flame and occurs at the flame base. Thecomputed flow velocity at the lowest oxygen point (the dividingpoint between blow off and quenching branches, the bottom of theU-shaped curve in Fig. 5) is about 5 cm/s [3]. Therefore, flame ex-tinction mechanism in this free convection upward burning config-uration is by blowoff. Although, the extinction mode is blowoff, theextinction limit is close to the lowest oxygen point along theflammability boundary, i.e. near the bottom of the the U-shaped inFig. 5a. Note that the lowest oxygen point (referred to as the fun-damental oxygen limit in Ref. [18], is the merging point between thequenching and the blowoff branch. At this point, the extinctionmechanisms of residence time and heat loss are of comparable im-portance. So even when the extinction is on the blowoff side, there iseffect of heat loss as qualitatively illustrated in Fig. 5a for the differentsample widths. We expect that a narrower sample will have asmaller flammable range as supported by related experimental evi-dence on flammable limits in variable ambient pressure for upwardflame spread [9] and downward flame spread [6]. Coupled with theprevious statement that a longer flame will induce a larger flowvelocity at the flame base is sufficient to explain the observed self-induced extinction phenomena in upward spread, to be detailednext.

For 21% oxygen, the flame blowoff velocity boundaries for threesample widths are illustrated in Fig. 5(b) with time as the ordinate.The burning history of three samples are qualitatively traced. Forthe 5 cm wide sample, trace A starts with a small induced velocityat the flame base. As the flame lengthens, the buoyant velocityincreases and eventually exceeds the critical blowoff velocity, atwhich time the flame on one side is blown off (marked by an “x” inthe graph). The absence of the flame on one side leads to a shorterflame on the remaining side (roughly half the length). The buoyantinduced velocity on this remaining side reduces at the base justenough to stave off blowoff and the one-sided flame can persist(the fact that a shorter flame induces a lower flow velocity at theflame base will be discussed in Appendix). The shorter one-sidedflame reaches a limiting length for steady upward spread.

Trace B for the 2 cm case in Fig. 5(b) is similar to trace A with asmaller critical blowoff velocity and shorterflame. For the 7.5-cmsample, trace C shows that the blowoff velocity limit is larger andthe buoyant induced velocity at the base does not cross over thelimit. The flame continues to grow until reaching the end of thesample.

4.2. Comments on the statistical nature of flame extincion processes

Why does the flame blows off on one side of the sample but noton both sides simultaneously? We believe this has to do with thestability of near-limit flames. As many experimenters investigatingextinction can attest, flames close to the extinction limits are verysensitive to disturbances. The closer the flame is to the limit, thesmaller the disturbace needed to trigger an extinction. Theoreti-cally, extinction limits are neutral stable points in stability analysis

Fig. 6. A 5-cm wide sample which does not extinguish. Despite the constant flame size on the reverse side (not shown), the front side flame nearly blows off but is able torecover.

Fig. 7. The flame base (highlighted with a solid yellow line) changes shape significantly during propagation just before blowff occurs. (For interpretation of the references tocolor in this figure legend, the reader is referred to the web version of this article.)

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286284

(e.g.turning point in a response curve) which explains their sen-sitivity to disturbances [20]. This sensitivity leads to a band ofenvironmental conditions that define the limit. Within the band,there is a probability distribution of the likelyhood of an extinctionevent. This probablistic nature also applies spatially. A seeminglysymmetric flame can see the disappearance of symmetry as the‘limit’ is approached. A well-known example case is the blowoffexperiment of a premixed bunsen burner flame from a circulartube. As the mixture feeding velocity is gradually increased nearblowoff, it is often observed that a portion of the flame base islifted first (i.e. local blowoff) before the entire flame blows off.Symmetry is often broken at the premixed flame flashback limitwhere one side of the flame enters into the tube first [12]. In thisexperiment, before blowoff we observe that the flame base firstbecomes unstable. This is manifested as the unsteady retreat andflashback of a portion of the base. This motion is both spatial andtime varying (i.e. irregular) before the entire flame blows off onone side. Figs. 6 and 7 illustrate the flame base fluctuations in twosperate tests: one leads to blowoff and the other does not. Fig. 6shows the front view at various times for a 5 cm wide case whichdid not extinguish. It can be seen that the flame on the front goesthrough the sequence of blowing downstream and shrinking untilaround the 1.6 s mark. In this particular case, due to the smallersize of the flame generating less entrained air velocity, the flame isable to recover. The flame base propagates back upstream to thepyrolysis base where fuel is still being vaporized from the flame onthe backside. Fig. 7 shows a similar test under identical conditionswhere blow off does occur. The flame bases before blowoff arehighlighted over a short duration (the actual images are very faintnear the flame base and therefore colored in yellow). One sees thatboth the shape and the position of the flame base fluctuate in-dicative of the closeness to the blowoff stability limit. Because ofthis random feature of the flame base, symmetry on two sides ofthe sample do not exist as blowoff is approached despite the verysymmetric experimental conditions. In all the tests we conducted

for this fuel, one side of the flame will go out first (with no pre-ferred side). When one side of the flame is blown off, the re-maining flame immediately benefits by a reduction in buoyantflow. It is no longer near the blow off boundary and can bestabilized.

The probabilistic nature of the events also manifests in twoother ways. First, the limiting sample width for flame blowoff inthis experiment is not an exact number. The 5-cm case has beenconducted many times, mostly exhibiting one sided blow off. Ofthe 10 most symmetric ignition and growth phases, 7 of thesehave one-sided blowoff, for the other three cases the flames pro-pagate all the way to the top of the sample. The location whereblowoffs first occur also vary. This makes us conclude that the5 cm wide sample (in air and in earth gravity) is a near limit case.Twelve 2 cm wide and four 3.5 cm wide samples tested all showone-sided blowoff. Although the test number is small, one sidedblow off occurs early indicating the narrow samples are away fromthe probablistic range. There are two 7.5 cm and 8.8 cm samplestested which all show a strong growing flame until reaching theend of the sample. The flame base in the wider samples tend tooscillate upstream and downstream, but do not show an indicationthat they will blow off.

5. Conclusions

A very interesting flame extinction mode has been found inupward spread over a solid fuel. One side of the upward spreadingflame blows off when it becomes too large. An explanation is of-ferred based on increased buoyant induced velocity at the flamebase stabilization zone. Although we report the experimentalfindings only in one special type of solid sample in this paper, webelieve it may be a more general near-limit phenomenon whichcould occur on other sample materials.

Based on our present understanding of flame stability and

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286 285

extinction as a function of flow velocity, it is proposed that a largerbuoyant induced velocity near the flame stabilization zone (at thebase of the flame) may cause blowoff as the flame length increases.To test this hypothesis, numerical simulations have been per-formed (see Appendix). The simulations verify that in a gravita-tional field, a longer flame induces a larger average velocity nearthe flame base region. However, the velocity increase due to thelonger flame is modest. Therefore, the observed blow off phe-nomenon may be most readily observed in near limit conditionswhere the flame is especially sensitive to small flow changes. Onthe other hand, assuming the proposed mechanism is valid, thisself-induced bouyant blowoff is not limited to thin samples. Weexpect thick solids could behave in a similar manner with com-plete flame extinguishment. However, this case is more difficult todetect and distinguish from other extinguishment modes since a

Fig. A1. streamline plot for visible flame length of 9.57 cm (at 2.88 s) vs. 39.7 cm(at 5.14 s). Visible flame is defined using the fuel reaction rate contour value of10�4 g/cm3/s.

Fig. A2. Close up view of the flame stabilization zone for the two flame lengths. Shown onimaginary control volume with the upstream face y1�y2 is shown here and explained

two-sided to one-sided flame transition is not possible with athermally thick solid.

The end result of a large seemingly robust flame extinguishingitself in the upward configuration could have wider implicationswhen considering material qualification from a flammabilitystandpoint. Although flame spread may be unstable in the upwardmode, the material may still be very flammable in otherconfigurations.

Acknowledgments

This research was initially funded by a grant from NASA (Dr.Gary Ruff, technical monitor) and concluded with a grant from theUnderwriters Laboratories (Dr. Pravinray Gandhi, technicalmonitor).

Appendix. : Supporting results of induced flow field by com-bustion model

One of the key elements in the interpretation of the observedflame extinction event on one side of the fuel sample is that the(average) flow velocity just upstream of the flame base becomestoo large for the flame to be stabilized. Although flame blow off isa well-known phenomenon, self-induced blow off when the flamelength grows too long has not been reported as far as we know. Itis important to verify that a longer flame in a gravitational fieldindeed induces a larger velocity at the flame base. The most directmethod for verification is through experiment. However, mea-suring the detailed buoyant flow field in the flame stabilizationzone is very challenging. A neutrally-buoyant seeding and a non-intrusive method to introduce the seeding are required. So instead,a numerical simulation is used to support our argument.

In the numerical simulation, an upward spreading flame over a2-cm wide sample is solved. The numerical model is modifiedfrom a three-dimensional transient combustion code previouslyused for solid ignition [13]. With the sole purpose of studying theflow field, the rate constants of the gas-phase reaction kinetics arespecified to be large enough so that no flame extinction isexpected.

the plots are streamlines, velocity vectors, grid cells, and reaction rate contours. Anin the text.

Fig. A3. Air mass flux per unit depth (defined as ∫ ρ′ =m g sudy [ /cm/ ]yy12

, y1¼0,y2¼2 mm, integrated for center-cut plane, at 2 mm upstream of maximum reac-tion location) vs. flame length.

M.C. Johnston et al. / Fire Safety Journal 71 (2015) 279–286286

Fig. A1 shows the flow streamlines along the center symmetryplane for a growing flame at two time steps. At 2.88 s, the flamelength (defined by the fuel vapor reaction rate¼10�4 g/cm3/s) is9.57 cm and at 5.14 s, it is 39.4 cm. One sees that there are sub-stantial differences of the flow patterns near the bottom parts ofthe flames. The longer flame draws more flow from a wider up-stream area at the bottom region. This is intuitively correct andshows that a substantial part of the buoyant flame cannot betreated by boundary layer type of analyses. Fig. A2 is the close upof the flow vectors near the flame base for the two flame sizes.One sees a larger velocity profile just upstream of the flame sta-bilization zone for the longer flame. To make it more quantitative,a control volume face is drawn upstream of the flame-base sta-bilization zone. The face is labeled y1�y2 and is located 2 mmupstream of the maximum reaction point and 2 mm in height(about the size of the flame stabilization zone). The air mass flux(per unit depth) entering into the control volume through y1�y2is computed and presented in Fig. A3 as a function of flame length.Qualitatively, one can think about the flame stabilization zone as areactor and the mass flux as being proportional to the averageincoming flow velocity. Fig. A3 shows that as flame length in-creases, the incoming mass flux increases. The percentage of in-crease (over the 9.57 cm flame case) is also shown on the righthand side of Fig. A3. Although the percentage increase is modest,for near-limit situations the blow off boundary can be crossed.

The above simulation is not to compare the experimental re-sults quantitatively as only one sample width is used and with un-calibrated reaction kinetics.The computational results presentedabove is merely to check and verify our postulate that a longerflame indeed induces a greater buoyant velocity at the flame base.It should be noted that this trend is purely based on considerationsfrom fluid mechanics and heat transfer. For example, we havetested this on a vertical plate heated with two different section

lengths. The qualitative trend is the same as the above flame si-mulation, i.e. longer heated section produces a larger induced flowrate at the lower edge of the heated section.

References

[1] L. Chu, C.H. Chen, J.S. T'ien, Upward propagation over paper samples, in:Proceedings of the ASME Paper 81-WA/HT-42, 1981.

[2] I.I. Feier, H.Y. Shih, K.V. Sacksteder, J.S. T’ien, Upward flame spread over thinsolids in partial gravity, Proc. Combust. Inst. 29 (2) (2002) 2569–2577.

[3] P.V. Ferkul, J.S. T’ien, A model of low-speed concurrent flow flame spread overa thin fuel, Combust. Sci. Technol. 99 (4–6) (1994) 345–370.

[4] P. Ferkul, J. Kleinhenz, H. Shih, R. Pettegrew, K. Sacksteder, J. T’ien, Solid fuelcombustion experiments in microgravity using a continuous fuel dispenserand related numerical simulations, Microgravity Sci. Technol. 15 (2) (2004)3–12.

[5] P. Ferkul, S. Olson, M. Johnston, J. T’ien, Flammability aspects of fabric in op-posed and concurrent air flow in microgravity, in: Proceedings of 8th U.S.National Combustion Meeting, Utah, 2013 (pp. Paper # 070HE-0218).

[6] A.F. Frey, J.S. T’ien, Near-limit flame spread over paper samples, Combust.Flame 26 (1976) 257–267.

[7] L.K. Honda, P.D. Ronney, Mechanisms of concurrent-flow flame spread oversolid fuel beds, Proc. Combust. Inst. 28 (2000) 2793–2801.

[8] J. Kleinhenz, The Flame Spread and Extinction Characteristics of Cotton-Fi-berglass Fabric. Case Western Reserve University, M.S. Thesis, Case WesternReserve University, Cleveland, 2002.

[9] J. Kleinhenz, J. T’ien, Combustion of nomex III fabric in potential space habitatatmospheres: cyclic flame spread phenomenon, Combust. Sci. Technol. 179(2007) 2153–2169.

[10] J. Kleinhenz, Z.-G. Yuan, An experimental study of upward burning over longsolid fuels: facility development and comparison, NASA, 2011.

[11] J. Kleinhenz, P. Ferkul, R. Pettegrew, K. Sacksteder, J. T’ien, One-sided flamespread phenomena of thermally thin composite cotton/fiberglass fabric, FireMater. 29 (2005) 23–37.

[12] B. Lewis, G. von Elbe, In Combustion, Flames and Explosions of Gases, Aca-demic Press, New York (1961) 231.

[13] Y.T. Liao, J.S. T’ien, A numerical simulation of transient ignition and ignitionlimit of a composite solid by a localized radiant source, Combust. Theor.Model. 17 (6) (2013) 1096–1124.

[14] W.E. Mell, T. Kashiwagi, Effects of finite sample width on transition and flamespread in microgravity, Proc. Combust. Inst. 28 (2000) 2785–2792.

[15] National Aeronautics and Space Administration, Flammability, Odor, Off-gassing, and Compatibility Requirements and Test Procedures for Materials inEnvironments that Support Combustion. NASA, 1998.

[16] Y. Pizzo, J.L. Consalvi, P. Querre, M. Coutin, B. Porterie, Width effects on theearly stage of upward flame spread over PMMA slabs: experimental ob-servations, Fire Saf. J. 44 (2009) 407–414.

[17] A.S. Rangwala, S.G. Buckley, J.L. Torero, Upward flame spread on a verticallyoriented fuel surface: the effect of finite width, Proc. Combust. Inst. 31 (2)(2007) 2607–2615.

[18] J.L. Rhatigan, H. Bedir, J.S. T’ien, Gas-phase radiative effects on the burning andextinction of solid fuel, Combust. Flame (1998) 231–241.

[19] H.Y. Shih, J.S. T’ien, A three-dimensional model of steady flame spread over athin solid in low-speed concurrent flows, Combust. Theor. Model. 7 (4) (2003)677–704.

[20] J.S. T’ien, The effects of perturbations on the flammability limits, Combust. Sci.Technol. 7 (4) (1973) 185.

[21] K.C. Tsai, Width effect on upward flame spread, Fire Saf. J. 44 (7) (2009)962–967.

[22] Y.-T. Tseng, J.S. T’ien, Limiting length, steady spread, and nongrowing flames inconcurrent flow over solids, J. Heat Transf. 132 (9) (2010) 091201.

[23] Underwriters Laboratories, UL 94, the Standard for Safety of Flammability ofPlastic Materials for Parts in Devices and Appliances testing, Underwriter'sLaboratory, Chicago, 1996.