ABSTRACT
Title of thesis: STUDYING WILDLAND FIRE SPREAD
USING STATIONARY BURNERS
Daniel Gorham, Master of Science, 2014
Thesis directed by: Professor Michael J. Gollner
Department of Fire Protection Engineering
Experiments were performed using stationary gas burners and liquid fuel-
soaked wicks in order to characterize the flame geometry and buoyant instabilities
present in stationary flames under inclined and forced flow configurations. These
stationary flames revealed behavior similar to spreading wildland fires but offer
an ideal platform to carefully study fundamental wildland fire behaviour such as
reacting-flows instabilities that may play a critical role in fire spread.
Two types of flow conditions were used to perform experiments, a sloped fuel
surface and forced-flow (wind aided) configuration. Small-scale inclined experiments
were performed at the University of Maryland with liquid-fuel soaked wicks and
large-scale experiments at the USDA Forest Service Missoula Fire Sciences Labo-
ratory with a gas burner. These experiments were performed with over a range of
heat-release rates and burner sizes for angles from 0 to 60 degrees from the horizon-
tal. Forced-flow experiments were performed with gas burners in a large-scale wind
tunnel at the Missoula Fire Sciences Laboratory and at the University of Maryland
with a well characterized wind blower. These experiments were performed for a
range of heat-release rates and burner sizes in wind speeds from 0.2 to 3.0 ms−1.
Flame geometry, such as centerline flame length and flame tilt angle were
determined from high-speed side-view images. Flame intermittency and pulsations
close to an inert surface were also measured using high-speed videography and arrays
of micro-thermocouples. A method was developed to track the extension of the flame
close to the surface which would come in direct contact with unburnt fuels ahead of
the fire. These methods showed that the pulsation frequency is complex, however
scaling over several orders of magnitude suggest large-scale coherent structures are
present in the flow. Using peak frequencies measured, the stationary experiments
follow a Strouhal-Froude scaling similar to flame pulsations in spreading fires and
puffing pool fires.
Streamwise streaks were also observed and measured in the flow using high-
speed videography. Streak spacing was observed to be associated with possible
counter-rotating vortex structures in the fire. The spacing for streaks at the base of
the flame for these stationary experiments appeared to be dependent on the incoming
boundary-layer conditions and could possibly scale with the centerline flame length,
similar to flame towers observed in spreading fires.
STUDYING WILDLAND FIRE SPREAD USING STATIONARY
BURNERS
by
Daniel Gorham
Thesis submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Master of Science
2014
Advisory Committee:
Professor Michael J. Gollner, Committee Chair
Professor Arnaud Trouve´
Dr. Mark Finney
c© Copyright by
Daniel Gorham
2014
Table of Contents
List of Figures iv
Nomenclature vii
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Flame Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Flame Pulsation . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.3 Streamwise Streaks . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Experimental Setup 9
2.1 Small-Scale Inclined Experiments . . . . . . . . . . . . . . . . . . . . 9
2.2 Large-scale Inclined . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Missoula Forced-Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 UMD Forced-flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Wind blower design . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Results 30
3.1 Flame Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.1 Inclined Experiments . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.2 Forced-flow Experiments . . . . . . . . . . . . . . . . . . . . . 35
3.2 Flame Pulsation and Intermittency . . . . . . . . . . . . . . . . . . . 40
3.2.1 Inclined Experiments . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 Forced-flow Experiments . . . . . . . . . . . . . . . . . . . . . 50
3.3 Streamwise Streaks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.1 Inclined Experiments . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.2 Forced-flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4 Thermocouple Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
ii
4 Discussion 61
4.1 Strouhal-Froude Flame Pulsation Relationship . . . . . . . . . . . . . 61
4.2 Dimensionless Fire Length and Fire Size . . . . . . . . . . . . . . . . 63
4.3 Flame Location Probability Density . . . . . . . . . . . . . . . . . . . 65
4.4 Go¨rtler Number Streak Formation . . . . . . . . . . . . . . . . . . . . 67
4.5 Boundary Layer Effects on Streak Formation . . . . . . . . . . . . . . 68
5 Conclusion 72
Bibliography 73
iii
List of Figures
2.1 Diagram of experimental setup used for inclined experiments at UMD. 10
2.2 Diagram of the experimental setup for large-scale inclined experi-
ments at the Missoula Fire Sciences Laboratory, (left) topview and
(right) sideview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Picture of experimental setup for large-scale inclined experiments
with modified flame wall. . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Top-view diagram of forced-flow experimental setup performed in the
wind tunnel at the Missoula Fire Lab. . . . . . . . . . . . . . . . . . 18
2.5 Photograph of the forced-flow experimental setup performed in the
wind tunnel at the Missoula Fire Lab. In this view, the wind is
blowing toward the photographer. . . . . . . . . . . . . . . . . . . . . 19
2.6 Diagram of the forced-flow experimental setup at UMD. . . . . . . . 22
2.7 Picture of the forced-flow experimental setup at UMD. . . . . . . . . 23
2.8 Design and components of a uniform blower built at UMD for forced-
flow experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.9 Top and front view diagram of wind-blower with dimensions. . . . . . 26
2.10 Velocity profiles measured at the outlet of the exhaust duct for dif-
ferent blower capacities (%) and free-stream velocities (U , m/s). . . . 28
2.11 Velocity profiles above burner, located 8 cm downstream from the
exhaust outlet are presented for different blower capacities (%) and
free-stream velocities, U (m/s). The insulation board is connected
flush to the blower outlet. . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1 Simple two-dimensional geometry diagram for a wind-blown flame. . . 31
3.2 Flame averaging process based on image luminosity. (a) Still image
from experiment video (b) black-and-white image converted using av-
erage threshold and (c) averaged image with average flame geometry
represented by red pixels. . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Flame length, Lf , versus the slope angle, θ, of large-scale inclined
experiments for different fire sizes. . . . . . . . . . . . . . . . . . . . . 33
3.4 Flame height, Hf , versus the slope angle, θ, of large-scale inclined
experiments for different fire sizes. . . . . . . . . . . . . . . . . . . . . 34
iv
3.5 Average flame images from two inclined experiments, both with burner
depth = 12.7 cm and fire size = 154 kW. (a) The slope angle is 20
degrees and the flame geometry rises sharply above the downstream
surface like a plume (b) The slope angle is 40 degrees and the flame
geometry follows the downstream surface like it is attached. . . . . . 35
3.6 Horizontal flame height, xf , versus the slope angle, θ, of large-scale
inclined experiments for different fire sizes. . . . . . . . . . . . . . . . 36
3.7 Flame tilt angle, φ, versus the slope angle, θ, of large-scale inclined
experiments for different fire sizes. . . . . . . . . . . . . . . . . . . . . 37
3.8 Flame length, Lf , versus wind speed, U . . . . . . . . . . . . . . . . . 38
3.9 Horizontal flame length, xf , versus wind speed, U . . . . . . . . . . . . 39
3.10 Flame height, Hf , versus wind speed, U . . . . . . . . . . . . . . . . . 40
3.11 Flame tilt angle, φ, versus wind speed, U . . . . . . . . . . . . . . . . 41
3.12 Sample flame location downstream of burner, xe, versus time for
large-scale inclined experiment. Heat-release rate = 154 kW, burner
depth = 12.7 cm, slope angle = 30 deg. . . . . . . . . . . . . . . . . . 42
3.13 Sample level-crossing frequency at locations downstream of the burner
for large-scale inclined experiment. Heat-release rate = 154 kW,
burner depth = 12.7 cm, slope angle = 30 deg. . . . . . . . . . . . . . 44
3.14 Sample probability-density function of normalized flame location for
large-scale inclined experiment. Heat-release rate = 154 kW, burner
depth = 12.7 cm, slope angle = 30 deg. . . . . . . . . . . . . . . . . . 45
3.15 Sample fast-fourier transform of flame location signal for large-scale
inclined experiment. Heat-release rate = 154 kW, burner depth =
12.7 cm, slope angle = 30 deg. . . . . . . . . . . . . . . . . . . . . . . 46
3.16 Pulsation frequency, f , versus flame length, Lf for large-scale inclined
experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.17 Pulsation frequency, f , versus heat-release per unit depth of burner,
Q′. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.18 Peak frequency location versus flame length, Lf for large-scale in-
clined experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.19 Peak frequency location versus heat-release per unit depth of burner,
Q′ for large-scale inclined experiments. . . . . . . . . . . . . . . . . . 50
3.20 Peak pulsation frequency, f , versus flame length, Lf from forced-flow
experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.21 Peak frequency location versus flame length, Lf for forced-flow ex-
periments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.22 Peak pulsation frequency, f , versus the wind speed times flame length,
ULf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.23 Light intensity across the width of the burning region for a single
image. The local maximums in the light intensity are marked with
black triangles and represent flame peaks. . . . . . . . . . . . . . . . 55
3.24 Histogram of streak spacing over the length of an experiment. A log-
normal probability density function distribution was fit to the data
and is shown with the bold red line. . . . . . . . . . . . . . . . . . . . 57
v
3.25 Streamwise streak spacing, λ, versus flame length, Lf from large-scale
inclined experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.26 Streamwise streak spacing, λ, versus the flame length, Lf from forced-
flow experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1 Strouhal-Froude scaling for stationary burner experiments performed
with the wind tunnel and wind blower. . . . . . . . . . . . . . . . . . 62
4.2 Flame length, Lf , versus dimensionless fire length, D∗ for large-scale
inclined fires. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Location of peak frequency versus dimensionless fire size, Q∗ for large-
scale inclined experiments. . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Probability density function of normalized flame location for three
large-scale inclined experiments at 40 degrees with a burner depth of
12.7 cm and fire sizes of 38, 77, and 154 kW. . . . . . . . . . . . . . . 66
4.5 Go¨rtler number plotted versus the wind speed for two different meth-
ods of determining the momentum thickness θm. The Blasius momen-
tum thickness comes from the Blasius solution for boundary-layer flow
over a flat plate. The experimental momentum thickness comes from
velocity profiles measured with the hot-wire anemometer. . . . . . . . 68
4.6 Streak spacing, versus the Go¨rtler number calculated with the experimentally-
measured momentum thickness. . . . . . . . . . . . . . . . . . . . . . 69
4.7 The incoming boundary layer has an obvious effects on the formation
of streamwise streaks and properties of the flame. These two pictures
are from experiments using 30 × 12 cm wick, soaked with 120 mL of
heptane fuel, in 1.3 m/s forced-flow. In the picture on the left the
leading edge of the burner is parallel with the blower outlet, and on
the right there is a 28 cm inert surface upstream of the wick. . . . . . 71
vi
Nomenclature
cp Specific heat capacity (kJ/kg-K)
d Burner depth (m)
D∗ Characteristic fire length
f Flame pulsation frequency (Hz)
Fr Froude number
g Acceleration due to gravity, 9.81 (m/s2)
G Go¨rtler number
Hf Flame height (m)
Lf Center-line flame length (m)
Q Heat-release rate (kW)
Q′ Heat-release rate per unit depth, Q/d, (kW/m)
Q∗ Non-dimensional fire size
Re Reynolds number
St Strouhal number
T Temperature (K)
U Free-stream velocity (m/s)
w Burner width (m)
xa Flame attachment length (m)
xe Flame extension length - location in attachment region (m)
xf Horizontal flame length (m)
Greek
δ Boundary layer thickness (m)
φ Flame tilt angle (deg)
θ Slope angle (deg)
θm Momentum thickness (m)
λ Streak spacing (m)
ν Kinematic viscosity
ρ Density (kg/m3)
vii
Chapter 1: Introduction
Wildland fire spread is often described as a function of three fire behavior
components; fuel, weather, and topography. Since fire spread is essentially a series of
ignitions, effects of fuels on spread are determined based on their propensity to ignite
including loading, shape and size, compactness, continuity, and chemical content.
Weather is the most variable component of fire behavior, changing in both time
and space including atmospheric conditions such as temperature, relative humidity,
stability, precipitation and wind speed. In contrast topography is relatively constant
in time and only changes in space, such as elevation, aspect, land features, and
steepness of slope.
There are several components that effect and contribute to fire behaviour. In
the experiments performed here, wind speed and steepness of slope were indepen-
dently investigated. Increasing the wind speed introduces fresh oxygen to burning
fuels and can increase fire intensity as well as influence flame length and geometry.
Increasing slopes can also have drastic effects on fire behavior, increasing intensities
and flame length.
1
1.1 Motivation
Finney et al. [1] recently called for the development of a fundamental theory
for wildfire spread. Current operational models used for predictions of fire spread
are empirically based, while available “physically based” models lack a physical ba-
sis for the assumptions used to describe fuel particle ignition and fire spread. While
computational power has increased, numerical tools for both operational firefight-
ing and long-term predictions continue to rely on semi-empirical fire spread models
primarily developed in the early 1970’s. The lack of advancement in fundamen-
tal understanding of fire spread leads current models to be unreliable. Without a
fundamental theory for fire spread, combustion, fluid dynamics, and heat transfer
processes cannot be reliably applied to develop a truly physically based model.
Recent studies of spreading wind-driven fires in the 3 × 3 m wind tunnel at
the USDA Forest Service, Missoula Fire Science Laboratory have shown coherent
structures that form in the streamwise direction of the flow as well as spanwise
fluctuations that propagate to the downstream edge of the flame zone contributing
to intermittent fuel heating. The highly spatially-uniform fuel beds used in these
experiments allowed for the observation of these structures with more repeatable
results than previous efforts [2]. The results suggest that flame spread in fine fuel
beds is driven by non-steady convective heating and intermittent flame contact on
fuel particles. These heating characteristics were measured using micro thermocou-
ple arrays and high speed video. The traveling flaming region and large experiment
size, however, makes it difficult to carefully study these properties, for instance us-
2
ing a statistical analysis of these features which appear stochastically in the flow. A
technique was therefore needed that could study these newly-observed instabilities
and other general structures of propagating wildfires in a small-scale configuration
that can be utilized over long times.
A stationary, non-spreading fire was chosen as it allows for a thorough sta-
tistical analysis of the flame structure. This technique has been used in the study
of flame spread through the built environment before, however previous work fo-
cused on time-averaged properties of the flame, not intermittent effects important
in spread through fine fuels [3]. Long-duration experiments allow for a large sample
size and more control over variations in experimental parameters, such as decoupling
of the heat-release rate of the fire from flow conditions, unachievable in spreading
fires. The flame zone depth in the direction of fire spread can also be carefully
adjusted via the size of the burner. High speed video and micro-thermocouples are
useful with these fires to reveal and track buoyant instabilities in the fire flow which
resemble those appearing in spatially-uniform fuel beds. The same intermittent
heating observed in the fuel bed experiments were observed here with stationary
burners, but with the ability to collect a larger data set. Observations made in
these small-scale experiments can therefore be related to large-scale observed be-
havior. Other modifications to the flow field, such as the incoming boundary-layer
thickness is also adaptable in these small-scale experiments.
3
1.2 Literature Review
1.2.1 Flame Geometry
Thomas [4] recognized a deficit in the research of fires exposed to the exterior
environment in 1967 and presented some of the first engineering-based experimental
and theoretical work on wildland fire spread. These experiments looked at spreading
fires in both discrete and continuous fuel beds. His work supported the assumption
of radiation dominated heating of unburned fuel ahead of the flame front, leading
to flame spread. Thomas described the flame size as a dimensionless parameter, the
ratio of flame length to burning region depth. Scaling analysis was used to determine
the effect of external winds on flames, with the ratio of inertial forces from the wind
and buoyant forces from the fire expressed as the Froude number
Fr =
U2
gLf
(1.1)
This parameter is expressed in terms of the free-stream velocity U , gravitational
acceleration g, and flame length Lf .
Motivated by a devastating escalator fire in King’s Cross station, London,
several investigators explored flame geometry for stationary burners in inclined
trenches. Atkinson [5] performed experiments in an inclined trench to explore fire-
induced flow for various configurations. He used a 10 x 27.6 cm gas burner built
into a apparatus rotatable from 20 to 90 degrees. For angles less than 20 degrees the
flow was plume dominated, and between 20 and 45 degrees buoyancy-driven flow
4
attached along an inert surface downstream of the burner, representing unignited
fuel in a spreading fire. Smith [6] provides a qualitative relationship between the
tilt angle (φ) of a flame and the slope angle (θ) of an inclined trench (Eq 1.2).
φvertical = 2θ. (1.2)
This relationship is based on a mass balance of air on the upstream and downstream
side of the flame.
Several other investigators [5–9] have explored the effect of side walls on in-
clined fires and their effect on flame geometry, later defined as the trench effect. The
plume geometry in non-trench inclined fires have also been investigated in orienta-
tions from pool to ceiling fires [10, 11].
Oka et al. explored the effects of cross-wind on an unconfined, stationary
fire geometry. Their experiments included the use of a square propane burner with
fire sizes ranging from 0.75 to 16 kW in cross-wind velocities of 0.5 to 4.0 m/s
[12]. Experiments were performed for two surface conditions, “with floor,” where
there was a 1.2 m long flat plate extending from the outlet of the blower to the
burner surface and “without floor,” where this false floor was removed. These
essentially created two inlet flow conditions, one with a developed boundary layer
and another with a uniform incoming flow. The effects of burner aspect ratios on
the flame geometry was also measured, using luminosity from cameras to determine
the average flame shape and two-dimensional thermocouple arrays for isothermal
curves [13]. Based on their results, the authors proposed that the flame tilt angle
5
can be determined using a ratio of mass fluxes, between the upward mass flux from
the fuel and the mass flux of air from the cross-wind. The balance is written as a
ratio of the Froude number and the non-dimensional fire size, Q∗,
Hf
d
=
Fr2/3
Q∗
(1.3)
where Hf is the flame height and d in the burner depth.
1.2.2 Flame Pulsation
A review of pulsations observed in diffusion flames, especially pool fires was
presented by Malalasekera et al. [14]. The authors described the mechanism of flame
oscillations in pool fires as three steps; formation of a torodial vortical structure due
to acceleration of buoyant gases, contraction of the flame near the base of the flame
as the vortex rises, and accumulation of buoyant gas to return to the base flame
structure.
Several investigators have previously explored using a Strouhal-Froude rela-
tionship to scale the flame pulsation frequency of pool fires [15]. The Strouhal
number is a non-dimensional representation of the oscillations in a flow,
St =
fLf
U
(1.4)
where f is the characteristic frequency, Lf is the characteristic length, and U is the
characteristic velocity [16].
More recently, Finney et al. [2,17] used Strouhal-Froude scaling to understand
6
the flame pulsations in laboratory-scale spreading wildfire experiments. They used
a frequency obtained from micro-thermocouples based on level-crossing of the gas-
phase flame temperature. The flame length and free-stream wind speed were used
as the characteristic length and velocity respectively.
1.2.3 Streamwise Streaks
Vortex structures have also been observed in laboratory and field observations
of wildfires [18]. Streamwise streaks in the burned fuel were observed in spreading
fire experiments, and observed flame towers were measured and correlated with
flame length [2]. The structures resemble counter-rotating vortex pairs (Taylor-
Go¨rlter vortices) which are known to form over concave surface [19]. Boundary
layer effects on the vortex formation were initially investigated [20] which lead to a
multi-mechanism instability approach to understanding.
Several investigators observed streamwise streaks in small inclined experiments
[21–23]. In these experiments a metal plate submerged in water was heated to induce
natural convection and slightly inclined. Investigators observed “secondary-flows”
which were seen as streamwise streaks, analogous to Go¨rtler vortices. Streaks present
in wall-bounded turbulent boundary layer flows have also been observed [24].
1.3 Research Objectives
The objective of this thesis is to study flame structure and instabilities on sta-
tionary fires to better understand these features occurring at large scale in the hope
7
of translating that knowledge to the development of a fundamental understanding
of wildland fire spread. Experiments were performed in inclined and forced-flow
configurations, two of the most important conditions in wildfire spread. Several fuel
sources were used, including variations in liquid fuel type and gas burner size.
The flame geometry has been shown to be an important factor for understand-
ing fire spread. Important parameters such as the flame length, flame height and
tilt angle were measured for these experiments. Determining the effects of flow con-
ditions on flame geometry and the potential for flame contact to unignited fuels will
lead to a better understanding of particle heating in wildfire spread.
Flame pulsations ahead of the flame front can also cause intermittent flame
contact and heating of fuel particles, leading to ignition. Direct flame contact on
the fuel particles causes conductive and convective heating and convective cooling,
forming a non-steady particle ignition process for fine, discrete fuels. The frequency
of flame pulsations and the distance ahead of the flame front these pulsations travel
were measured to better understand their underlying mechanisms. Coherent struc-
tures and instabilities observed in the flow are suspected to be responsible for the
pulsating phenomena. Studying this behavior in small-scale experiments is vital to
characterize the mechanics responsible for spreading from small to large scales.
8
Chapter 2: Experimental Setup
2.1 Small-Scale Inclined Experiments
Initial experiments were performed using a small-scale inclined apparatus with
liquid-fuel soaked wicks at the University of Maryland. These experiments were
performed first to determine if using stationary burners and image analysis was a
valid way to observe instabilities. The steady burning rate of these inclined fuel
wicks was also measured to determine the effect of slope angle and burner aspect-
ratio on the burning rate of the fuel.
The experimental apparatus used was adapted from a previous study on in-
clined flame spread over solid fuels [11]. The apparatus consisted of a base plate
with two vertical supports, and aluminum plate connected in a fashion so it is ro-
tatable 360 degrees (Figure 2.1). The rotatable surface was a 22 × 60.5 × 0.3 cm
aluminum plate. Non-combustible insulation board was used as an inert surface
around a fuel-soaked wick. This single sheet of insulation board, 60 × 40 cm was
Superwool brand High-Temperature Board SB with a rated temperature of 1150◦C
and a density of 360 kg/m3.
A slot for the fuel-soaked wick was cut into the insulation board 5 cm from
the bottom and 5 cm from each side (Figure 2.1). This was done both to reduce the
9
Figure 2.1: Diagram of experimental setup used for inclined experiments at UMD.
effects of side air entrainment and downstream of the wick, provide an attachment
surface, acting as the surface where unburned fuels would be during simulated flame
spread.
The same insulation board was used to make the liquid fuel burners. The wicks
measured 30 cm in width and three depths were used: 3, 6, and 12 cm. After the
insulation was cut to size for the desired wick, it was exposed to a propane torch for
10-15 minutes to burn off any residual binders from the manufacturing processing.
It also made the insulation more porous so it could absorb more fuel. The bottom
and sides of the wick were covered with aluminum foil leaving only the top surface
expose, preventing liquid fuel from leaking out of the bottom or sides of the wick
when inclined.
The burning rate was determined based on the mass-loss rate of the burning
fuel. A Metler-Tolledo MS32001L load cell was used to measure the instantaneous
mass of the fuel sample at 1 Hz with a precision of 0.1 grams. The balance was
10
connected to a computer via USB cable and data was logged using Balance Link
software. Various liquid fuels with different properties were used for these exper-
iments to achieve a range of fire sizes including methanol, ethanol, acetone, and
heptane.
The amount of fuel used was based based on the size of the wick and vol-
ume required for total saturation. The volume of fuel used for 3, 6, and 12 cm
deep wicks was 60, 120, and 180 mL respectively. The same amount of fuel was dis-
tributed evenly on the wick surface with a liquid syringe for each of the experimental
configurations.
Side-view images were taken with multiple cameras including a Canon SLR
and a Casio EX-F1. To capture a two-dimensional image of the flame the camera
was mounted on a tripod and rotated to the same angle as the apparatus. Images
were capture continuously with the Canon SLR at 3 frames per second (fps) and in
high speed with the Casio EX-F1 at 300 fps. The SLR images were high-resolution,
at 6000 × 4000 pixels while high speed video was recorded at 528 × 312 pixels.
Top-view images were also taken with both the Canon and Casio cameras.
Before each experiment, any residual soot from previous experiments was
burned off the wick with a propane torch for 3 to 5 minutes. While the wick was
allowed to cool, the apparatus was rotated to the desired angle, which was mea-
sured using a digital inclinometer. The side-view camera was manually focused on
the middle of the top surface of the insulation board. The top-view camera was
setup on a tripod behind apparatus as approximately at 45 degree angle measured
from the surface of the insulation board. The camera was manually focused on the
11
wick and downstream inert surface. Once cool, the fuel wick was fully saturated
with liquid fuel using a syringe, and then placed into its slot in the insulation board.
Mass data began being recorded on the computer connected to the load cell. The
lights were turned off and the cameras turned on. The wick was ignited with a
lighter, being careful not to touch the surface disturbing the mass data.
2.2 Large-scale Inclined
Based on successful observation of large-scale fire features with the small-scale
experiments at UMD, large-scale experiments were performed at the USDA Forest
Service Missoula Fire Sciences Laboratory. Here, a large 1.81 m × 0.61 m flame wall
designed by Jimenez et al. [25] was retrofitted for inclined experiments. A gearbox
and crank were added to the rotation system to allow the wall to be held at an angle
between 0 and 90 degrees (horizontal and vertical). An extension at the top of the
burner was built for the inert attachment surface.
To achieve an even, uniform gas flow at the burner surface, a three layer
diffusion medium was used. An interior layer of 0.06 cm thick fiberglass cloth,
followed by a layer of fiberglass batting, and finally a 2.5 cm thick sheet of ceramic
foam made up the burner surface. The porosity of the top layer of Cotronic ceramic
foam was 17.7 pores per centimeter. The exposed burner surface area was 0.61
meters wide and 1.83 meters long. To produce a line fire, a large section of the
burner was sealed so that gas flowed only out of the exposed ceramic surface. A
1.5 m long steel sheet covered the bottom portion of the burner (Figure 2.2) and
12
Figure 2.2: Diagram of the experimental setup for large-scale inclined
experiments at the Missoula Fire Sciences Laboratory, (left) topview and
(right) sideview.
was sealed around the edges with aluminum tape. The remaining 0.33 m could be
sealed with aluminum foil and insulation to change the exposed burner area. The
burner width was always 0.61 m wide, and experiments were performed at depths
of 6.4, 12.7, and 25.4 cm.
The fuel used in the burner was ethylene gas, chosen to reduce buoyant effects
in the burner plenum because its molecular weight is close to that of air, 28.05 g/mol
versus 28.96 g/mol for ethylene and air respectively. Gas flow into the burner was
controlled with an Omega mass flow controller. Experiments were performed using
ethylene gas flow rates of 39.1, 78.1, 156.3, and 234.6 lpm corresponding to fire sizes
of 38.6, 77.1, 154.3, and 231.6 kW, respectively.
The side-view camera used for these experiments was a Phantom model high-
13
speed camera with a fish-eye lens recording at 120 fps at a resolution of 1920 × 1200
pixels. Due to the length of the flame wall, both the burner and inert surface were
very high so that the side-view camera needed to be mounted on a tripod capable
of reaching 9 feet. The camera was rotated on the tripod to be at the same angle as
the burner and inert surface on the apparatus. The camera was focused on the top
edge of the inert surface. The fish-eye lens allowed for a wide range to be captured
past the leading edge of the burner, however corrections were made via comparison
with ruler measurements to minimize and distortion effects from the lens. Side-view
videos were only recorded for a 10 second duration for each test using the Phantom
camera because of memory limitations.
The top-view camera was a Casio EX-FH25 recording at 120 fps at a resolution
of 640 × 480 pixels. In order to maintain a consistent viewing angle on the flame
while the flame wall was rotated to higher slope angle, the tripod was mounted
to the bottom of the flame wall. This allowed for the top-view camera to remain
focused on the same portion of the flame without making adjustments when the
flame wall was rotated.
An array of micro-thermocouples were used to measure temperature fluctu-
ations downstream of the burner. The thermocouples were Omega brand K-type,
with bead diameter of 12 microns. The thermocouple array consisted of 52 probes
mounted in a wire holder 1 cm above the inert surface, with the first thermocouple
5 cm downstream of the burner and subsequent thermocouples spaced 2 cm down-
stream. A National Instruments data logger was used to record temperature signals
from all thermocouples at 120 Hz.
14
Figure 2.3: Picture of experimental setup for large-scale inclined exper-
iments with modified flame wall.
15
A picture of the experimental setup during a test is shown in Figure 2.3. The
wall is shown rotated to a desired slope angle, measured by a digital angle finder.
A side-view camera was placed on its tripod and focused on the top surface of the
inert insulation board. The top-view camera was mounted on the tripod, which
maintained an orientation angle with the burner. Before tests, the burner was first
purged with nitrogen gas and then ethylene fuel was injected through the mass flow
controller. Once the flame was ignited with a propane torch the nitrogen was turned
off and the plenum allowed enough time to be cleared of any remaining nitrogen. The
lights in the combustion chamber were turned off and the cameras began recording,
as well as the thermocouple data acquisition. After each experiment the apparatus
was rotated back to the horizontal position and the burner purged with nitrogen to
burn any remaining fuel.
2.3 Missoula Forced-Flow
While working as a visiting researcher at the fire lab in Missoula, initial forced-
flow experiments with a stationary burner were performed in the same wind tunnel
used for large spreading fire experiments. The initial prototype for the flame wall,
a 30 × 25 cm sintered metal burner was used as the burner, and an apparatus for
holding the ceramic insulation board was built flush with the burner surface.
The wind tunnel used had a cross-sectional area of 3 × 3 meters. The wind
speed was measured with a velocity probe in the centerline of the wind tunnel, and
experiments were performed at wind speeds of 0.22, 0.44, 0.67, 0.89, 1.11, and 1.34
16
m/s.
Cotronics high-temperature ceramic insulation board was placed flush with
the burner surface, providing an inert surface for experiments and limiting side-air
entrainment. The top surface of the burner and inert insulation was approximately
30 cm above the floor of the wind tunnel, putting it directly into the free-stream
flow. A diagram of the experimental setup is show in Figure 2.4. Propane gas was
used in the burner for all forced-flow experiments. The fuel flow rate was measured
with a volumetric flow meter and experiments were conducted at flow rates of 9, 13,
and 18 lpm corresponding to fire sizes of 7.5, 10.9, and 15.1 kW.
The same Phantom camera described earlier was used to record side-view
video. The camera was mounted on a tripod outside of the wind tunnel, level with
the top of the burner and inert surface of the apparatus. The camera had a clear
line-of-sight of the apparatus in the wind tunnel through an observation window.
Despite the tripod shown in the photograph, top-view images were not captured for
the wind tunnel experiments at Missoula because of difficulties avoiding disturbing
the flow when mounting the camera. A picture of the experimental setup in the
wind tunnel is shown in Figure 2.5.
Forced-flow experiments were performed in the Missoula wind tunnel over the
course of a week, with all of experiments for one fire size completed straight-through
(all completed in the same day). The experiments at 4.3 kW were completed first,
and then the apparatus had to be removed from the wind tunnel for other experi-
ments. The next time the apparatus was put in the wind tunnel, the experiments
for fire sizes 6.3 and 8.7 kW were completed in succession without removing the
17
Figure 2.4: Top-view diagram of forced-flow experimental setup per-
formed in the wind tunnel at the Missoula Fire Lab.
18
Figure 2.5: Photograph of the forced-flow experimental setup performed
in the wind tunnel at the Missoula Fire Lab. In this view, the wind is
blowing toward the photographer.
19
apparatus from the wind tunnel again.
2.4 UMD Forced-flow
After successfully acquiring results using the stationary burner in the Missoula
wind tunnel, smaller, more detailed forced-flow experiments were performed at the
University of Maryland using both liquid and gaseous fuels with a specially-built
uniform velocity wind blower for combustion experiments. Design and characteristic
of the wind blower are discussed in Section 2.4.1. Some initial experiments were
performed with heptane soaked fuel wicks to determine the effect of the incoming
boundary layer across the flat surface while later, sand burners were designed and
used for testing with gaseous fuels.
The forced-flow experiments with the liquid fuel wicks used the same apparatus
as in small-scale inclined experiments, but with the apparatus parallel with the floor
(a slope angle of 0◦). Two wick sizes were used, 6 and 12 cm deep in the flow
direction, both 30 cm wide. The wicks were fully saturated with heptane in the
same manner described previously.
Several sand burners were designed and built, including two iterations and
three different burner dimensions. The first burner was made with 1/32 inch (0.07
cm) thick galvanized steel with a burner surface of 25.4 × 4.54 cm. The burner was
6.5 cm tall with a 2 cm tall plenum, and 4.5 cm of fine sand. Propane fuel entered a
1/4 inch (0.6 cm) NPT bulkhead fitting on the bottom side of the burner and filled
the plenum. The sand was supported above the plenum with a mesh screen which
20
also acted as a diffuser to evenly distribute mass flux of gas fuel from the top of the
burner.
After intial testing with the fuel wicks and the first generation sand burner,
two design additional sand burners were designed and built. Both burners were
made of 1/8 inch (0.3 cm) thick stainless steel, bent and and welded to form the
burner walls. The burners were both 8 cm tall and 25 cm wide, measured on the
inside. One burner was 5 cm deep in the flow direction while the other was 10.6
cm. A 1/4 inch (0.6 cm) NPT connector was bored through the bottom side of the
burner and welded in place. The plenum on both burners was approximately 2 cm
and the same sand was use to diffuse the gas, 6 cm deep on these burners.
All three sand burners were operated with propane fuel. The mass flow rate of
propane gas was controlled with an Alicat MCR-100SLPM-D mass flow controller.
Experiments were conducted with fuel flow rates of 4.9, 6.6, 10, and 15 lpm cor-
responding to fire sizes of 2.3, 3.2, 4.8,and 7.2 kW, respectively. A diagram of the
experimental apparatus is shown in Figure 2.6. The free-stream velocity was con-
trolled by the power of the blower. Experiments were performed at wind speeds
ranging from 0.6 to 2.7 m/s.
Two side-view cameras were used for these experiments. High-speed and high-
definition video was taken with a Casio EX-F1 digital camera. The high-speed video
was taken at 300 fps at 512 × 384 pixel resolution. The same camera was used
for high-definition video at 60 frames-per-second at 1920 × 1080 pixel resolution.
A Nikon D7100 DSLR camera was also used for continuous shooting of side-view
images at 6 frames-per-second. At high-resolution, top-view camera images were
21
Figure 2.6: Diagram of the forced-flow experimental setup at UMD.
22
Figure 2.7: Picture of the forced-flow experimental setup at UMD.
23
taken with the previously-described Canon SLR.
A picture of the experimental setup is shown in Figure 2.7. The apparatus was
setup at the outlet of the blower exhaust and the propane cylinder and mass-flow
controller connected to the burner. A side-view camera was setup to be focused on
the top of the inert surface and the top-view camera focused on the burner and the
downstream surface.
To begin testing the blower was plugged in and turned on to the desired ve-
locity, which was measured in the free-stream with an Omega hot wire anemometer.
The fuel flow rate out of the gas burner was set on the mass-flow controller and the
burner ignited with a propane torch. The lights in the room were shut off and cam-
eras began recording. Experiments were run for 30 seconds after which the cameras
were shut off and exhaust hood was turned on.
2.4.1 Wind blower design
Forced-flow combustion experiments are difficult to conduct in traditional wind
tunnels for several reasons. In recirculating wind tunnels the makeup air is contam-
inated with by-products of combustion and can effect the combustion efficiency at
the fire source. Many large wind tunnels are constructed of non-fire resistant mate-
rials and would become overwhelmed by moderately sized fires in the test chamber.
A solution was to design and build a wind blower, where a uniform-velocity flow
field can be blown onto a fire and exhausted in a fire hood similar to ambient fire
experiments.
24
Figure 2.8: Design and components of a uniform blower built at UMD
for forced-flow experiments.
A uniform velocity wind blower was designed and constructed at the University
of Maryland to conduct these forced-flow fire experiments. The wind-blower design
consists of a centrifugal fan that pressurizes a large plenum and then contracts the air
through a mesh screen and flow-straightening honeycomb before traveling through
a 1.3 m long straight exhaust duct shown in Fig. 2.8.
The centrifugal fan was an EBM Papst Nautilair combustion blower controlled
via pulse-width modulation capable of max flow rate of 1200 cfm. The pulse-width
modulation allowed the blower speed to be controlled without causing oscillations
in the flow. The plenum box, converging section, and exhaust duct were all con-
structed out of of 1/2 inch (1.27 cm) thick plywood sheets. The dimensions of these
25
Figure 2.9: Top and front view diagram of wind-blower with dimensions.
components are shown in Figure 2.9. The plenum box was built as a 5-sided cube
permanently connected, with the 6th side being the front where the converging sec-
tion is connected and can be removed to transport the wind-blower and access the
inside. The front side is connected by cinching it with angle brackets and sealing the
connection with foam tape. The converging section was connected to the plenum
box with angle brackets and sealed with silicon sealant. The final piece of the blower
was the exhaust duct. The duct was connect to the converging section and sealed
with duct tape. Approximately 5 cm into the exhaust duct was a mesh screen to
promote mixing of the flow, followed by a 1/2 inch (1.27 cm) thick honeycomb flow
straightener. The flow finally traveled approximately 0.9 m through the duct, reach-
ing the exhaust exit, where a nearly uniform-laminar flow field was blown over a
test region.
The flow field at the outlet of the wind blower was measured with a Dantec
26
Dynamics hot-wire anemometer system. The hot-wire was connected to a two-
dimensional Velmex traverse system to characterize the flow field. A vertical cen-
terline velocity profile was measured with a spatial resolution of 1 mm for four
different. At each point the velocity was sampled at 50 kHz for 10 seconds. The
velocity profiles at the outlet of the exhaust duct are shown in Figure 2.10 for free-
stream velocities of 0.56, 0.99, 1.55, and 2.07 m/s. The boundary-layer thickness
for all four of the velocities was approximately 27 mm. Velocity profiles were also
measured at distances downstream of the exhaust outlet. Figure 2.11 shows the
velocity profiles at the leading edge of the burner and 8 cm downstream from the
exhaust outlet. The boundary layer thickness was found to increase as the velocity
increases, opposite to the Blasius boundary layer theory for a flat plate, something
that will continue to be investigated further.
27
Figure 2.10: Velocity profiles measured at the outlet of the exhaust duct
for different blower capacities (%) and free-stream velocities (U , m/s).
28
Figure 2.11: Velocity profiles above burner, located 8 cm downstream
from the exhaust outlet are presented for different blower capacities (%)
and free-stream velocities, U (m/s). The insulation board is connected
flush to the blower outlet.
29
Chapter 3: Results
3.1 Flame Geometry
Knowledge of the mean flame geometry, such as flame length, tilt angle, etc.
is useful for understanding fire behavior and can be easily obtained from station-
ary fire experiments. Experiments were performed for both horizontal forced-flow
and inclined configurations for both liquid and gas fuel burners. Two-dimensional
flame geometry measurements were taken from the side-view images of experiments.
Flame geometry measurements included the center-line flame length (Lf ), flame
height (Hf ), horizontal flame length (xf ), flame tilt angle (φ), and the flame attach-
ment length (xa) (Figure 3.1). Measurements were taken based on the average flame
tip, which is defined as the furthest downstream point, that when connected to the
mid-burner point, approximately cut the flame in half. The flame tip location and
other relevant parameters were measured via MATLAB’s user-input interface.
To determine the distinction between flame and non-flame in recorded im-
ages, a representative light-intensity threshold was used to convert the images, to
binary black and white images where white pixels represented flame and black rep-
resented non-flame. A MATLAB function thresh tool was used to determine the
light-intensity threshold level for images from an experiment. Once the experimen-
30
Figure 3.1: Simple two-dimensional geometry diagram for a wind-blown flame.
tal video was loaded into MATLAB, the thresh tool function was run on a random
image and the threshold level determined from user input. The level was selected
based on the highest value with no error in interpreting flame from non-flame. The
average of five threshold levels from random images in a single test were averaged
and this value was used for all images in the experiment.
Thresholding converted the gray-scale images to binary, black-and-white im-
ages, with 1 indicating flame and 0 indicating no flame. The average image is the
sum of all the binary matrices, divided by the number of images. This average image
contains the same number of pixels as all of the individual images, with each having
a value between 0 and 1, essentially representing the percentage of time the flame
was at that pixel location. With this “flame probability” image, the average flame
shape based on a 50% probability, following the methodology used by Audoin et al.
for pool fires [26]. A range of pixel values was used, 0.48 to 0.50, to determine the
average flame location. The binary average image was converted in a RGB image,
and those pixels that fell within the range of 0.48 to 0.50 were colored red to denote
31
Figure 3.2: Flame averaging process based on image luminosity. (a) Still
image from experiment video (b) black-and-white image converted using
average threshold and (c) averaged image with average flame geometry
represented by red pixels.
the average flame geometry (Figure 3.2).
Flame height and horizontal flame length were vertical and horizontal mea-
surements respectively from the downstream edge of the burner to the flame tip.
The flame length was a straight line between the mid-burner point and flame tip.
The flame tilt angle, φ, is defined as the angle between the centerline flame length
and the fuel surface. It can be determined by finding the inverse tangent of the
flame height, Hf and the horizontal flame length, xf plus half the burner depth,
d/2,
φ = arctan
(
Hf
xf + d/2
)
. (3.1)
It can also be important to understand the length along the surface a flame ex-
tends forward of the pyrolyzing zone, making up the pre-heating length for convectively-
spreading fires. Flame attachment, xa, is defined as the horizontal flame length in
the region of interest close to the surface.
32
Figure 3.3: Flame length, Lf , versus the slope angle, θ, of large-scale
inclined experiments for different fire sizes.
3.1.1 Inclined Experiments
Measurements of flame geometry from large-scale inclined experiments are
shown below. Several experiments were performed for each set of experimental
conditions (burner depth, fire size, and slope angle) the results were averaged for
that data point. Figure 3.3 shows how the centerline flame length, Lf , increases
fairly linearly with increasing slope angle of the experimental apparatus. The flame
length attaches to the surface as it approaches the vertical orientation and flow is
similar to wall fire configurations. Longer flame lengths increase the flame extension
and convective heating of unburned fuels.
Figure 3.4 shows how the flame height, Hf decreases as the slope angle in-
33
Figure 3.4: Flame height, Hf , versus the slope angle, θ, of large-scale
inclined experiments for different fire sizes.
creases. This is expected as the flame attaches more to the surface as the slope
moves toward a vertical wall fire. For all fire sizes, flame height is approximately
the same for 10 and 20 degree slope angles, but decreases at slope angles above 30
degrees. This is in agreement with the transition between a plume fire to an at-
tached fire. Figure 3.5 shows the average flame geometry of two large-scale inclined
experiments, both with a burner depth of 12.7 cm and fire size of 154 kW. The
image on the left is at a slope angle of 20 degrees where the flame forms a plume
like shape rising sharply from the downstream surface, and the image on the right
is at a slope angle of 40 degrees, where the flame geometry is more attached to the
downstream surface.
34
Figure 3.5: Average flame images from two inclined experiments, both
with burner depth = 12.7 cm and fire size = 154 kW. (a) The slope angle
is 20 degrees and the flame geometry rises sharply above the downstream
surface like a plume (b) The slope angle is 40 degrees and the flame
geometry follows the downstream surface like it is attached.
Figure 3.6 shows the horizontal flame length, xf , which increases as the slope
angle increases, similar to Lf . The horizontal flame length influences radiative
heating of unburned fuels ahead of the flame front.
Figure 3.7 shows the flame angle, φ, which decreases as the slope angle in-
creases. An exponential relationship between the flame angle and the slope angle
could be a factor of the transition from a plume fire to an attached flame. The slope
angle also increases slightly as the fire size increases for the same slope angle, so for
larger fires the flame angle is less sensitive to the slope angle.
3.1.2 Forced-flow Experiments
Results from flame geometry measurements of forced-flow experiments from
both the Missoula wind tunnel and the UMD wind blower are shown in Figures 3.8 -
3.11. This data is compiled from experiments performed in the Missoula wind tunnel
35
Figure 3.6: Horizontal flame height, xf , versus the slope angle, θ, of
large-scale inclined experiments for different fire sizes.
36
Figure 3.7: Flame tilt angle, φ, versus the slope angle, θ, of large-scale
inclined experiments for different fire sizes.
37
Figure 3.8: Flame length, Lf , versus wind speed, U .
and at UMD with the wind blower. The Missoula wind tunnel experiments used
fire sizes of 17, 25, and 35 kW/m, and a burner depth of 24 cm. The experiments
performed at UMD with the wind blower also appear to be segregated based on the
burner used, with the heat-release rate per unit length of 45 and 68 kW/m were
performed with the second iteration of the sand burner design, with a burner depth
of 10 cm. The remaining data points, for fire sizes of 48, 65, 100, and 138 kW/m
were performed with the initial sand burner with a depth of 4.54 cm.
Figure 3.8 shows how the centerline flame length, Lf , increases as the wind
speed increases. There appears to be a grouping for the value of the flame length
based on the burner depth. Experiments with the largest burner depth, 25 cm, are
grouped relatively close for larger flame lengths.
38
Figure 3.9: Horizontal flame length, xf , versus wind speed, U .
Figure 3.9 shows the horizontal flame length, xf , increases as the wind speed
increases and Figure 3.10 shows the flame height, Hf versus the ambient wind speed.
As the free-stream velocity increase, a nearly exponential decay in the flame height is
observed. We would expect the flame height data to show a transition at some point
from a plume-type fire to an attached (boundary-layer) flame, like in the inclined
experiments, but there does not seem to be any distinguishable transition in this
data.
Figure 3.11 shows how the flame tilt angle, φ decreases with increased wind
speed. Similar to flame length measurements, the flame angle measurements are
grouped based on the burner depth. There is a slight exponential relationship be-
tween the flame angle and the wind speed as an increased momentum flux from the
39
Figure 3.10: Flame height, Hf , versus wind speed, U .
incoming wind speed “pushes” the flame closer to attach to the surface.
3.2 Flame Pulsation and Intermittency
The downstream flame location in stationary experiments was determined us-
ing side-view high-speed video and an array of thermocouples ahead of the burner.
A region starting at the downstream edge of the burner and extending fully down-
stream, 1 cm above the surface was determined as the region of interest where
extension of the flame would relate to flame attachment and fuel particle heating.
The flame extension downstream of the burning region is referred to as flame loca-
tion or flame location, and denoted symbolically as xe. This location fluctuated in
40
Figure 3.11: Flame tilt angle, φ, versus wind speed, U .
time as it was observed that the flame would pulse downstream, bursting into what
would be the unburned fuel region in a spreading fire experiment.
Video analysis was performed using a MATLAB script and the VideoReader
function to import the experimental video and decompose it into individual images.
Flame distinction was made using the same threshold determined for mean flame
geometry. Images were converted to black and white using this threshold level. The
previously-described region-of-interest was defined from the downstream edge of the
burner surface (user-input) to the end of the image in the downstream direction,
with a height above the surface of 1 cm. Flame location, xe, was then determined
in this region-of-interest by tracking the furthest-most downstream tip of the flame.
This location fluctuated in time, its downstream location shown for example in
41
Figure 3.12: Sample flame location downstream of burner, xe, versus
time for large-scale inclined experiment. Heat-release rate = 154 kW,
burner depth = 12.7 cm, slope angle = 30 deg.
Figure 3.12.
A method for determining a level-crossing rate was used on the flame location
signal to determine the frequency of flame pulsations at points downstream of the
burner. The crossing rate was determined for locations downstream of the burner in
1 cm increments and converted to a frequency based on the frame rate of the video.
First, the index of all points in the flame location signal that were greater than
the crossing location were determined. Then the indexes where the flame locations
transition from below to above the crossing location were marked. A level-crossing
42
was only considered for one direction; therefore only when the flame was previously
not at a location and then at the next time step it was counted a crossing. This
is somewhat analogous to the variable-interval time averaging (VITA) technique
use to study turbulent flow fluctuations, described elsewhere [27]. The crossing
rate was then determined to be the number of flame location transitions divided
by the number of frames in the video analyzed, and the frequency determined by
multiplying the crossing rate by the frame rate of the video.
The frequency for different locations downstream were determined to look
find the location of peak frequency (intermittency) (Figure 3.13). When a single
frequency was needed for analysis, such as Strouhal-Froude scaling, the peak fre-
quency was used.
The stochastic nature of the flame locations indicates there are a range of
spectral-phenomena occurring. In order to compare the general variance a normal
probability density function of the normalized flame location was determined. The
flame location is normalized by dividing each individual flame location by the mean
flame location. This was done to be able to compare the fluctuations in flame
location for different fires sizes and slope/wind conditions. This density function
captures how far the flame extends downstream beyond the mean flame location.
Figure 3.14 shows and example where the wing of the PDF denotes the probability
a flame will extend beyond the mean flame location.
A fast-fourier transform (FFT) from the time domain to the frequency domain
can also be used to determine the critical frequency of the flame oscillations. Figure
3.15 shows a typical FFT plot of the flame location signal and it is clear that
43
Figure 3.13: Sample level-crossing frequency at locations downstream of
the burner for large-scale inclined experiment. Heat-release rate = 154
kW, burner depth = 12.7 cm, slope angle = 30 deg.
44
Figure 3.14: Sample probability-density function of normalized flame
location for large-scale inclined experiment. Heat-release rate = 154
kW, burner depth = 12.7 cm, slope angle = 30 deg.
45
Figure 3.15: Sample fast-fourier transform of flame location signal for
large-scale inclined experiment. Heat-release rate = 154 kW, burner
depth = 12.7 cm, slope angle = 30 deg.
no dominant frequency is captured with this technique. No distinct but a range of
frequencies appear in the signal. Similar to turbulent phenomena, stochastic process
over a range of frequencies occur. This suggestions that that flame pulsation does
not follow a sine-wave oscillation and more advanced spectral analyzes are required.
3.2.1 Inclined Experiments
Pulsation analysis results for large-scale inclined experiments are shown in Fig-
ures 3.16 - 3.19. Figure 3.16 shows how the peak level-crossing frequency increases
46
Figure 3.16: Pulsation frequency, f , versus flame length, Lf for large-
scale inclined experiments.
slightly with the flame length and decreases slightly as the fire sizes decreases. A
slight linear relationship between the peak frequency and the flame length for dif-
ferent heat-release rates suggests that the relationship between the frequency and
flame length changes based on the fire size. The frequency is more sensitive for
smaller fire sizes and less for larger fires, suggesting there is a dependence on plume
or attached flame geometry.
Figure 3.19 shows the peak frequency plotted against the heat-release rate
divided by the burner depth for different slope angles. Not much correlation is
observed. The downstream distance with the observed peak frequency increases
with both the flame length and fire size (Figure 3.18).
47
Figure 3.17: Pulsation frequency, f , versus heat-release per unit depth
of burner, Q′.
48
Figure 3.18: Peak frequency location versus flame length, Lf for large-
scale inclined experiments.
49
Figure 3.19: Peak frequency location versus heat-release per unit depth
of burner, Q′ for large-scale inclined experiments.
Figure 3.19 shows how the relationship between the peak frequency location
and the heat-release divided by the burner depth increases in proportionality as the
slope angle increases. This is similar to the relationship between the flame height
and the slope angle at different fire sizes, the transition from a plume-fire to an
attached fire occurs above 30 degrees, and this is where the peak frequency location
increases more rapidly for larger fires.
3.2.2 Forced-flow Experiments
Pulsation analysis results for forced-flow experiments performed both in the
Missoula wind tunnel and at UMD with the wind blower are shown below. Fig-
50
Figure 3.20: Peak pulsation frequency, f , versus flame length, Lf from
forced-flow experiments.
ure 3.20 shows the peak frequency versus flame length for various heat-release rates
divided by the depth of the burner. There is a slightly positive relationship between
the flame length and peak frequency but no real trend with the heat-release per unit
length.
The downstream distance of the peak frequency also increases as the flame
length increase (Figure 3.21). For higher wind speeds the flame is tilted closer to
the surface and the downstream reach of the flame increased. This means the flames
are more likely to reach further into an burned fuel bed.
The relationship between the peak frequency location and flame length changes
when the flame length is multiplied by the wind speed (Figure 3.22). The quantity
51
Figure 3.21: Peak frequency location versus flame length, Lf for forced-
flow experiments.
52
Figure 3.22: Peak pulsation frequency, f , versus the wind speed times
flame length, ULf .
of the flame length multiplied by the wind speed is a parameter that is attempting
to capture effect of the wind pushing the flame closer to the downstream surface
and the flame length which attaches to the surface. The relationship becomes more
linear, but there is a clear distinction between experiments performed in the wind
tunnel and ones performed with the wind blower (in Montana and Maryland, re-
spectively). The incoming boundary-layer profile may cause this deviation between
experiments, however the velocity profile above the burner in the Missoula wind
tunnel is unknown.
53
3.3 Streamwise Streaks
Streaks in the flame were observed over stationary burners when wind was
applied in the streamwise direction. These streaks are believed to be manifestations
of peaks and troughs in the flame with associated increases or decreases luminosity
caused by counter-rotating streamwise vortex pairs. These appear to be similar to
previously described as Go¨rtler vortices which occur over curved boundary-layers
or on heated inclined plates [19]. The streak represents the upwash region at the
intersection of two vortices, and so the space between the streaks contains a pair of
streamwise vortices. The space between coherent streaks, λ, was determined using
image analysis from the top view images of the experiments.
Image analysis was performed in MATLAB with the same readin process de-
scribed for the flame pulsation analysis. The burner width was used in the images
to set a scale for converting from pixels to length in centimeters. A crop box re-
gion was selected at the base of the flame where the streaks first formed across the
width of the burner. The depth of the cropbox in the flow direction was set to 1 cm
based on the scale set by the burner width. Each image was converted to gray-scale
and cropped to the same area. The pixel value in each column of the cropbox was
averaged and a light intensity signal across the width of the burner was produced.
Figure 3.23 is a sample of the light intensity across the width of the burner. The
local peak in the signal is determined and marked as black triangles on the plot.
These points represent streaks in the flame, otherwise seen as flame peaks in the
streamwise flow.
54
Figure 3.23: Light intensity across the width of the burning region for
a single image. The local maximums in the light intensity are marked
with black triangles and represent flame peaks.
55
The average spacing for each image was determined and all images in an ex-
perimental set analyzed. The streak spacing distribution was plotted as a histogram
to show the distribution over the length of the experiment (Figure 3.24). The streak
spacings were found to follow a log-normal distribution and so a probability density
function was fit with the data. This follows analysis for streak-like structures found
in other turbulent boundary-layer flows [28], where despite sinuous meandering of
the streaks they maintain constant mean spacings. The streak spacing for an exper-
iment was taken from the peak of the log-normal function from the streak spacing
distribution.
3.3.1 Inclined Experiments
The streak spacing for all inclined experiments versus the flame length is shown
in Figure 3.25. The streak spacing does not change much with the flame length,
but decreases slightly as the flame length increases. The arithmetic mean of the
streak spacing for all experiments is λmean = 1.88 cm with a standard deviation
of σ = 0.16. This is a variance of less than 8% for the range of experiments; it
appears that the controlling parameter in the streamwise vortice formation and
spacing was not captured in these experiments, however they were able to be more
reliably produced and statistically analyzed then for spreading experiments
56
Figure 3.24: Histogram of streak spacing over the length of an experi-
ment. A log-normal probability density function distribution was fit to
the data and is shown with the bold red line.
57
Figure 3.25: Streamwise streak spacing, λ, versus flame length, Lf from
large-scale inclined experiments.
58
Figure 3.26: Streamwise streak spacing, λ, versus the flame length, Lf
from forced-flow experiments.
3.3.2 Forced-flow
Streak spacing for forced flow experiments were compared with the flame
length in Figure 3.26. A slight decrease in the spacing was observed as the flame
length increases but the change in was very slight, the average streak spacing for
all experiments is λmean = 1.77 cm with a standard deviation of σ = 0.24. Like the
inclined results for streak spacing, the variance in streak spacing over the range of
experiments performed here is very low.
59
3.4 Thermocouple Data
Thermocouple data was collected for some of the large-scale inclined experi-
ments. Many of the thermocouples broke during the experiments because of the high
temperature of the ethylene flame. The metal holder used to hold the thermocouples
1 cm above the surface warped from the constant heating of the flame. This caused
the thermocouples to break or move significantly out of the region of interest. Due
to unreliable data recorded, comparisons with video data are not presented here.
60
Chapter 4: Discussion
4.1 Strouhal-Froude Flame Pulsation Relationship
Figure 4.1 a Strouhal-Froude relationship, which is based on forced-flow exper-
iments. The Strouhal and Froude number were previously defined in equations 1.4
and 1.1, and the relationship between the two is
St = 2Fr−0.5. (4.1)
The value of the exponent is slightly above the value of -0.43 found in scaling for
larger-scale wind-driven spreading fire experiments [29] and the same for diameter
scaling of puffing in circular pool fires [30]. While the flame length may not be the
most fundamental quantity for scaling analysis because of its coupling to experi-
mental conditions, it is useful as most wildland fire spread analyses are correlated
to this quantity [31].
The agreement in this scaling between stationary and spreading fire experi-
ments suggests that the methods of the experiments performed here can be used to
study the same phenomena in spreading fires. The Strouhal number captures the
pulsation of the flame close to the fuels causing intermittent contact, and the Froude
61
Figure 4.1: Strouhal-Froude scaling for stationary burner experiments
performed with the wind tunnel and wind blower.
number represents the balance of forces from the wind and buoyancy of the flame.
It could be possible to use this scaling to develop a relationship between the pulsing
frequency for a range of wind conditions and fire sizes. A spread rate could possibly
be determined based on the frequency and distance ahead of the flame front. In
future studies, less reliance on the flame length as a length scale is desired, however
it still provides a useful and practical representation of the fire size to scale buoyant
effects.
62
4.2 Dimensionless Fire Length and Fire Size
A dimensionless length scale is often used in the fire research community to
describe the buoyant source strength of fire [32]. D∗,
D∗ =
[
Q
ρcpTg
]2/5
(4.2)
is given by the ratio of the fire heat release-rate Q to the ambient air conditions ρcpT
(density, specific heat, and temperature) and the square root of g, the acceleration
due to gravity, all raised to the 2/5 power (Eq 4.2). The gravity term contributes
to the buoyant force of the fire, and should act parallel to the line of action of the
flame. To incorporate the slope angle of inclined fires, a rotated gravity vector term
parallel to the inclined surface is used, given by the acceleration due to gravity time
the sine of the slope angle:
ginclined = g sin(θ). (4.3)
Flame lengths of pool fires, for instance, can be scaled with the dimensionless
fire length, and so with an adjusted gravity vector (Eq 4.3 they should be able to
as well.
D∗ =
[
Q
ρcpTgincline
]2/5
(4.4)
The dimensionless fire length using the inclined gravity component (Eq 4.4)
can be used to correlate the flame length of the large-scale inclined fire experiments
63
(Figure 4.2). The flame length increases linearly with D∗ by a factor of 1.2 and based
on the definition of this parameter it suggests that the flame length is dependent
on the source fire conditions and slope angle. An attempt was made to capture the
burner depth parameter in the flame length relationship as well, but unfortunately
using a D∗ equation with the heat-release divided by the burner depth did not
correlate well.
Figure 4.2: Flame length, Lf , versus dimensionless fire length, D∗ for
large-scale inclined fires.
Another common fire scaling parameter is the dimensionless fire size, Q∗, given
by the dimensionless fire length D∗ and a physical length scale, both raised to the
5/2 power. Often times the characteristic length scale is a plume dimension, so here
we use the flame length, Lf (Eq 4.5),
64
Figure 4.3: Location of peak frequency versus dimensionless fire size, Q∗
for large-scale inclined experiments.
Q∗ =
(D∗)5/2
(Lf )
5/2 . (4.5)
Figure 4.3 shows the location of the peak level-crossing frequency versus this
dimensionless fire size parameter. The relationship shows the pulsation distance
decreasing as the dimensionless fire size increases.
4.3 Flame Location Probability Density
The density function of the normalized flame location shows when the flame
extended or “pulsed” further downstream than the average flame location. The far
regions of the function can be considered “rare events” when the flame pulses much
65
Figure 4.4: Probability density function of normalized flame location for
three large-scale inclined experiments at 40 degrees with a burner depth
of 12.7 cm and fire sizes of 38, 77, and 154 kW.
further downstream of the burning region, into unburned fuels. The probability of
these rare events increase for the same wind or slope conditions but for increase fire
size (heat-release rate, Q), as shown in Figure 4.4. Correlations of increasing flame
extension frequency with slope angle, wind speed, and others parameters were poor
compared to the strong correlation of fire size and increased extension frequency.
This is perhaps because larger coherent structures develop in more intense turbulent
fires, which can similarly cause increased flame spread rates.
66
4.4 Go¨rtler Number Streak Formation
The Go¨rtler number
G =
Uθm
ν
(
θm
R
)1/2
(4.6)
is a dimensionless numbered that is used to determine the when Gortler vortices
will form in boundary layers over curved surfaces [19]. This dimensionless number
was developed to describe streak formation over a concave surface where θm is the
momentum thickness of the incoming boundary layer and R is the radius of curvature
of the wall. Go¨rtler instabilities have been shown to develop at G greater that 0.3.
The centripetal force is critical in the formation and so is captured by the radius of
curvature. For the experiments performed in this thesis the buoyant force may be
thought of as the centripetal force and so the radius is replaced by the flame length
for the length scale,
G =
Uθm
ν
(
θm
Lf
)1/2
. (4.7)
Figure 4.5 shows the Go¨rtler number for forced-flow experiments conducted
with the wind blower at UMD, where the momentum thickness was determined
using a hot-wire anemometer. All values in this velocity range exceed the predicted
onset value which is given for the original formation of the Go¨rtler number with
centripetal force.
Figure 4.6 shows the streak spacing as a function of the Go¨rtler number and
67
Figure 4.5: Go¨rtler number plotted versus the wind speed for two differ-
ent methods of determining the momentum thickness θm. The Blasius
momentum thickness comes from the Blasius solution for boundary-layer
flow over a flat plate. The experimental momentum thickness comes from
velocity profiles measured with the hot-wire anemometer.
shows a good relationship for correlations of the streak spacing (given the limited
range of values). A relationship between the flame length and large-scale streaks
has been shown for spreading experiments, but an understanding based on the
boundary-layer conditions is needed for these vortex structures.
4.5 Boundary Layer Effects on Streak Formation
One of the benefits of the small-scale forced-flow experiments at UMD was the
use of a fully characterized wind blower. The wind blower was calibrated with a hot-
wire anemometer as described previously. Using the hot-wire it was determined that
68
Figure 4.6: Streak spacing, versus the Go¨rtler number calculated with
the experimentally-measured momentum thickness.
the turbulence intensity of the flow was less than 3% at all points in the flow. With
these calibrations the properties of the hydrodynamic boundary layer are known
and can be used to understanding observations of streaks in the flame.
The incoming boundary layer across the burner was changed by the length of
the flat boundary layer development surface upstream of the burner. For a short
boundary layer development length, the top surface of the apparatus was raised
approximately 2.54 cm above the bottom surface of the blower exhaust duct. This
raised the insulation board out of the boundary layer that developed in the duct.
A long boundary layer development length was achieved by making the top of the
insulation board flush with the outlet of the blower exhaust duct and making the
69
connect seamless with aluminum tape.
Figure 4.7 shows the effects of the boundary layer development length on the
formation of streamwise streaks above the burner surface. These pictures are from
experiments with a heptane fuel soaked wicks with the same initial amount of fuel
(120 mL). The difference between these two experiments is the boundary layer de-
velopment length, xBLD, which is the length of flat surface upsteam of the burner.
The difference between the two cases is visually obvious; with no boundary layer
development length the uniform flow generated flame with no boundary layer devel-
opment has almost no observable streaks, with a flatter, more laminar appearance
while the case with a boundary layer development length forms coherent streaks.
Analysis of the two configurations showed that the boundary-layer configuration
results in higher burning rates (0.8 g/s vs. 0.3 g/s), and higher peak frequencies
than the uniform-flow configuration (7.3 Hz vs. 3.5 Hz). Therefore, the formation
of streaks appears to be important in understanding downstream heating and thus
fire spread.
Streak merging downstream of the burner formed into flame towers can be
observed in top-view images. A successful method for tracking the merging of
the streaks downstream was not determined, but based on visual observations the
streamwise streaks clearly converge to form the flame towers. This may be associ-
ated with spanwise fluctuations of streaks at the base of the flame.
70
Figure 4.7: The incoming boundary layer has an obvious effects on the
formation of streamwise streaks and properties of the flame. These two
pictures are from experiments using 30 × 12 cm wick, soaked with 120
mL of heptane fuel, in 1.3 m/s forced-flow. In the picture on the left the
leading edge of the burner is parallel with the blower outlet, and on the
right there is a 28 cm inert surface upstream of the wick.
71
Chapter 5: Conclusion
This study has presented a framework for the use of stationary gas burners to
study spreading wildland fire behavior, enabling measurements and analysis of key
parameters related to mean flame geometry and temporal fluctuations that play a
role in flame spread. Video and micro-thermocouple techniques used to accomplish
this are shown, as well as preliminary scaling analysis. Recently-discovered instabil-
ities and flow structures are shown to be re-created in steady gas burners, allowing
for more detailed studies, including statistical analysis made possible by the long
time scales stationary flames can create. Correlations for the mean flame geometry,
such as the flame tilt angle and flame length can also be obtained by these studies
with appropriate nondimensional scaling.
72
Bibliography
[1] Mark A Finney, Jack D Cohen, Sara S McAllister, and W Matt Jolly. On the
need for a theory of wildland fire spread. International journal of wildland fire,
22(1):25–36, 2013.
[2] Mark A Finney, Jason Forthofer, Isaac C Grenfell, Brittany A Adam, Nelson K
Akafuah, Kozo Saito, et al. A study of flame spread in engineered cardboard
fuelbeds: Part i: Correlations and observations. 2013.
[3] L Orloff, J De Ris, and GH Markstein. Upward turbulent fire spread and burn-
ing of fuel surface. In Symposium (International) on Combustion, volume 15,
pages 183–192. Elsevier, 1975.
[4] P.H. Thomas. Some aspects of growht and spread of fire in the open. Vol.
40:pp. 139–164, 1967.
[5] Drysdale D.D. Wu Y. Atkinson, G.T. Fire driven flow in an inclined trench.
Vol. 25:pp. 141–158, 1995.
[6] DA Smith. Measurements of flame length and flame angle in an inclined trench.
Fire safety journal, 18(3):231–244, 1992.
[7] DD Drysdale and AJR Macmillan. Flame spread on inclined surfaces. Fire
Safety Journal, 18(3):245–254, 1992.
[8] PJ Woodburn and DD Drysdale. Fires in inclined trenches: the dependence
of the critical angle on the trench and burner geometry. Fire safety journal,
31(2):143–164, 1998.
[9] Ying Wu, HJ Xing, and G Atkinson. Interaction of fire plume with inclined
surface. Fire Safety Journal, 35(4):391–403, 2000.
[10] J De Ris and L Orloff. The role of buoyancy direction and radiation in turbulent
diffusion flames on surfaces. In Symposium (International) on Combustion,
volume 15, pages 175–182. Elsevier, 1975.
73
[11] MJ Gollner, X Huang, J Cobian, AS Rangwala, and FA Williams. Experimental
study of upward flame spread of an inclined fuel surface. Proceedings of the
Combustion Institute, 34(2):2531–2538, 2013.
[12] Y Oka. Modelling of unconfined flame tilt in cross-winds int. assoc. of fire safety
and science. In Proceedings of the 6 th International Symposium on Fire, pages
pp. 1101–1112, 2000.
[13] Yasushi Oka, Osami Sugawa, Tomohiko Imamura, and Yoshiyuki Matsubara.
Effects of cross-winds to apparent flame height and tilt angle from several kinds
of fire source. In Proceedings of the 7th International Symposium on Fire,
International Association of Fire Safety and Science, pages pp. 915–926, 2002.
[14] WMG Malalasekera, Henk K Versteeg, and K Gilchrist. A review of research
and an experimental study on the pulsation of buoyant diffusion flames and
pool fires. Fire and materials, 20(6):261–271, 1996.
[15] Hiroyuki Sato, Kenji Amagai, and Masataka Arai. Scale modeling of puffing
frequencies in pool fires related with froude number. In Progress in Scale Mod-
eling, pages 133–147. Springer, 2008.
[16] Anthony Hamins, Takashi Kashiwagi, and Robert R Buch. Characteristics of
pool fire burning. ASTM special technical publication, 1284:15–41, 1996.
[17] Brittany A Adam, Nelson K Akafuah, Mark Finney, Jason Forthofer, Kozo
Saito, et al. A study of flame spread in engineered cardboard fuelbeds: Part ii:
Scaling law approach. 2013.
[18] Jason M Forthofer and Scott L Goodrick. Review of vortices in wildland fire.
Journal of Combustion, 2011, 2011.
[19] William S Saric. Gortler vortices. Annual Review of Fluid Mechanics,
26(1):379–409, 1994.
[20] P Hall. The linear development of go¨rtler vortices in growing boundary layers.
Journal of Fluid Mechanics, 130:41–58, 1983.
[21] EM Sparrow and RB Husar. Longitudinal vortices in natural convection flow
on inclined plates. Journal of Fluid Mechanics, 37(02):251–255, 1969.
[22] Kenzo Kitamura and Fumiyoshi Kimura. Heat transfer and fluid flow of natural
convection adjacent to upward-facing horizontal plates. International journal
of heat and mass transfer, 38(17):3149–3159, 1995.
[23] Peter Jeschke and Hans Beer. Longitudinal vortices in a laminar natural con-
vection boundary layer flow on an inclined flat plate and their influence on heat
transfer. Journal of Fluid Mechanics, 432:313–339, 2001.
74
[24] James M Wallace. Highlights from 50 years of turbulent boundary layer re-
search. Journal of Turbulence, (13), 2012.
[25] Dan Jimenez, Mark A Finney, Jack Cohen, et al. Design and construction of
gas-fed burners for laboratory studies of flame structure. Notes, 2010.
[26] L Audouin, G Kolb, JL Torero, and JM Most. Average centreline temperatures
of a buoyant pool fire obtained by image processing of video recordings. Fire
Safety Journal, 24(2):167–187, 1995.
[27] RF Blackwelder and RE Kaplan. On the wall structure of the turbulent bound-
ary layer. Journal of Fluid Mechanics, 76(1):89–112, 1976.
[28] John F Foss and James M Wallace. The measurement of vorticity in tran-
sitional and fully developed turbulent flows. In Advances in fluid mechanics
measurements, pages 263–321. Springer, 1989.
[29] Cohen Jack D Forthofer Jason A McAllister Sara S Adam Brittany A Akafuah
Nelson K English Justin Saito Kozo Gorham Daniel J Gollner Michael J Finney,
Mark A. Experimental evidence of buoyancy controlled flame spread in wildland
fires. 2014.
[30] Baki M Cetegen and Tarek A Ahmed. Experiments on the periodic instability
of buoyant plumes and pool fires. Combustion and Flame, 93(1):157–184, 1993.
[31] GM Byram. Combustion of forest fuels. in forest fire: control and use.(ed. kp
davis) pp. 61–89, 1959.
[32] James G Quintiere. Scaling applications in fire research. Fire Safety Journal,
15(1):3–29, 1989.
75