Размещено на http://www.allbest.ru/

Abstract

While flame spread through uniform fuel-air mixtures has been widely studied in combustion science, there has been relatively little attention given to the study of non-homogenous, or layered, fuel-air mixtures. However, these systems are common occurrences in such cases as terrestrial fuel spills and fuel leaks in both normal and microgravity. Conducting research on layered fuel-air mixtures and understanding the properties of flame propagation has potential implications for fire safety (both on earth and in space), as well as being of fundamental interest. The main objective behind this study is to determine flame speed, flammability regions, stability limits, and the shape of a flame propagating through a free, layered fuel-air mixture, as opposed to flame spread though layered mixtures over a solid surface, which had been previously studied. A free layer eliminates contact between the flame and the floor, which in turn reduces heat transfer and flow field effects. Such a system also simulates a fuel leak in microgravity conditions where the fuel vapor can be distributed by the slow ventilation flows, or a leak in normal gravity where a light fuel can ride in a plume.

The system chosen for study consists of a 79 cm long, roughly 10 cm² flow duct. A heated, porous bronze, fuel emitting airfoil is positioned 10 cm from the inlet along the centerline while a slow stream of air is blown parallel to the airfoil, creating the layered mixture in the laminar wake region. To design the flow duct geometry, a 2-D, multi-species, non-reacting numerical model of the system was developed using the FLUENT CFD software. This model accounts for diffusion and temperature of the fuel, which was ethanol in this study. The model provides a better understanding of the characteristics of the flow in the experimental apparatus, such as predicting velocity profiles, fuel concentration, and an estimated flame shape. Modeling results show that the flammable region in the duct is approximately 1 cm thick. The modeling results were used to position the igniter for the experimental runs, and to choose the inlet velocity and airfoil temperature.

Analytical calculations were also performed to determine the conditions under which a stable, stationary (i.e. non-propagating) flame could exist in the wake of the airfoil. In this configuration, the velocity of the propagating flame is balanced by the convective velocity of the fuel/air mixture. The calculations also show the precise locations in the flow field wherein a stoichiometric fuel/air mixture exists.

Once the geometry was characterized numerically, cold flow and combustion tests were performed. Cold flow testing included smoke tests which visualized the flow to ensure a steady, laminar quality, as well as hotwire anemometer and thermocouple scans to measure velocity and temperature profiles, respectively; all of these agreed with model predictions.

Preliminary experimental results show that it is possible to obtain a propagating flame in a non-uniform free layer with flame spread rates of up to 180 cm/s in flame fixed coordinates. If conditions where optimal, a triple flame structure would form. Image sequences of the side view of the flame spread, along with spread rate, are presented in this thesis.

Contents

1. Introduction

1.1 Motivation

1.2 Literature Review

1.2.1 Free Layers

1.2.2 Floor Or Ceiling Layers

• 1.3 Nasa Layers Project History
• 1.3.1 Experimental
• 1.3.2 Numerical Modeling
• 1.4 Objectives Of The Present Study
• 1.5 Organization Of The Thesis
• 2 Computational Fluid Dynamics Modeling Setup
• 2.1 Geometry Definition
• 2.2 Mesh Generation
• 2.3 Zone/Boundary Sets
• 2.4 Fluent Setup
• 2.4.1 Defining The Models
• 2.4.1.1 Solver
• 2.4.1.2 Species
• 2.4.1.3 Energy
• 2.4.1.4 Viscous
• 2.4.2 Defining The Material Properties
• 2.4.3 Defining The Operating Conditions
• 2.4.4 Defining The Boundary Conditions
• 2.4.5 Executing The Fluent Code
• 3 Modeling Results
• 3.1 Fluent Computational Fluid Dynamics
• 3.1.1 Contour Plots
• 3.1.1.1 Mole Fraction Contours
• 3.1.1.2 Equivalence Ratio Contours
• 3.1.1.3 Velocity Contours
• 3.1.1.4 Temperature Contours
• 3.1.2 X-Y Plots
• 3.1.2.1 Equivalence Ratio Vs. Y-Position
• 3.1.2.2 Velocity Profile Line Plots
• 3.1.3 Surface Integrals
• 3.2 Predicted Stationary Flame Shape/Location
• 3.2.1 Laminar Flame Speed (Uniform Mixtures)
• 3.2.2 Predicted Stationary Flame Location/Shape
• 3.3 Modeling Summary
• 4 Development Of The Experimental Apparatus
• 4.1 Airfoil Style/Design
• 4.1.1 Naca 0012
• 4.1.2 Parallel Plate
• 4.2 Duct Design
• 4.3 Instrumentation And Features
• 4.3.1 Instrumentation
• 4.3.2 Airfoil Internal Heaters
• 4.3.3 Air Inducer
• 4.3.4 Fueling System
• 4.3.5 Ignition System
• 4.4 Testing Conditions
• 5 Experimental Results
• 5.1 Cold Flow Tests
• 5.1.1 Duct Calibration
• 5.1.2 Velocity Scans
• 5.1.3 Temperature Scans
• 5.1.4 Smoke Wire
• 5.2 Combustion Tests
• 5.2.1 Test Matrix
• 5.2.2 Experimental Procedure
• 5.2.3 Fuel Vapor Profile
• 5.2.4 Flame Ignition
• 5.2.5 Flame Structure
• 5.2.6 Flame Spread Rates
• 6 Summary And Conclusions
• 6.1 Summary Of Work To Date
• 6.2 Conclusions And Suggestions
• Appendix A - Mass Diffusivity Calculation
• Appendix B - Mass Fraction Calculation
• Appendix C - Equivalence Ratio Calculation
• Appendix D - Fuel Tube Diameter Calculation
• Appendix E - Hotwire Calibration
• Appendix F - Rotameter Calibration Curve
• Appendix G - Autocad Drawings Of Duct

1. Introduction

1.1 Motivation

The overwhelming majority of flame studies of gases found in the combustion literature focus on either uniform, premixed systems or completely non-premixed systems (Glassman, 1996). It is well known, for example, that in the absence of flow field or geometrical effects, a flame will propagate through a uniform, premixed fuel-air mixture at a constant speed (called the laminar flame speed) that is a function of the chemical kinetic reaction rate and thermal diffusivity as well as equivalence ratio of the mixture. For most hydrocarbon/air mixtures, the maximum laminar flame speed is on the order of 40 cm/s. Examples of premixed systems include Bunsen burners, gas stoves and internal combustion engines.

In non-premixed systems, the fuel and oxidizer remain on opposite sides of the flame and meet at the flame front. Examples of non-premixed systems include candles, droplets and sprays. These flames are often called diffusion flames because molecular diffusion of fuel and oxidizer is often the controlling parameter.

Many practical systems, however, cannot be classified as either uniform premixed, or completely non-premixed. One such system is a non-uniform, or layered, fuel-air mixture. Although this type of system is a common occurrence in such cases as fuel spills in normal gravity and fuel leaks in microgravity, there has been comparatively little attention given to the study of non-homogenous fuel systems. From the few previous studies, it has been shown that flames spreading through layered fuel-air systems can propagate four times faster than typical laminar flame speeds.

Although they have received relatively little attention in the literature, non-uniform fuel-air mixtures are common occurrences in real-life combustion situations. They are present in fire hazards such as automobile and aircraft crashes, chemical spills, and underground mining situations. Flames spreading through layered systems are known to propagate over fences and even past the end of a fuel spill. Layered systems can also occur on board a spacecraft and therefore can present a realistic danger (Miller, et al. 2000, 2001, 2002).

Conducting research on layered fuel mixtures and understanding the properties of such flame propagation has practical relevance in terms of fire safety, both on earth and in space. A better understanding of flame propagation under these conditions can aid in the design of automobiles and other forms of transportation, chemical plants and storage facilities, and spacecraft/stations all from a safety aspect to prevent the propagation of such flames. A better understanding of non-uniform premixed flame propagation can also result in development of systems that extinguish fires more effectively.

Another important aspect of this research is to gain a fundamental understanding of the underlying physical phenomena of free layer fuel mixtures. These aspects include studying the effects of the fuel concentration gradient as well as the effects of buoyancy on the flame. Specifically, the objective is to understand the flame structure and determine what makes these flames spread so fast, as has been shown in prior research described below.

1.2 Literature Review

During the past three decades several experimental and theoretical studies have been performed on flame propagation through non-uniform premixed gas systems (Kaptein and Hermance, 1976; Feng, et al., 1975, Ishida, 1988 and Miller et al, 2002). Each of these studies has focused on flame spread above a solid surface. This thesis presents the results of a new experimental study aimed at developing and testing an apparatus to study the propagation of flames through free, stratified fuel/air mixtures. A review of the previous literature is provided in the following sections.

1.2.1 Free Layers

Phillips (1965) at the Safety in Mines Research Establishment studied the shape of a flame propagating through a fuel layer with no surface boundaries. This free, stratified fuel/air system was generated using a flow duct with a converging observation window (to stabilize the flame). A stream of pure fuel, methane in this case, was introduced at the entry of the duct near the roof while air was let in the lower part of the chamber. A partition separated the two flows near the entry until they passed through a mixing section. The stream then passed through a flow chamber into the observation region where the flame was ignited. The flame was stabilized if the fuel and air flows were adjusted to the point where the flame speed and the gas velocity were equal. The results showed what became known as a triple flame. Characteristics of a triple flame include a very broadly curved flame front, where the leading edge was centered about the stoichiometric limit. The top half of the curved flame front is a rich premixed flame, and the bottom half a lean premixed flame. A trailing diffusion flame is formed in the wake of the flame front between the two premixed flames. Figure 1 shows an image of a triple flame as seen from the side in one of Phillips' experiments.

Figure 1: Triple flame propagating through a fuel layer with no surface boundaries (Phillips, 1965).

Hirano and coworkers (1980) developed a two dimensional numerical model to simulate the gas movement ahead of a propagating flame. One case that was modeled consisted of a uniform stream of air and fuel through which a flame spreads, surrounded by pure air. This stream was assumed to be in free space. This assumption is important as it eliminates any surface effects from walls, floors, and ceilings. A diagram of the model employed by Hirano and coworkers is shown in Figure 2.

Figure 2: Schematic diagram of flame spread model (Hirano, 1980).

There were several assumptions made in developing the model. For simplicity, the model assumed the flame propagation to be two-dimensional. Since the fuel mixture immediately before the flame front experiences sudden velocity changes, effects of viscosity on the flow were neglected, resulting in an inviscid model. The flow was modeled as incompressible since the flow is well below the sonic regime. The flow field was simulated as a “surface” where the tip of the flame front was positioned. This “surface” was generated using a line source in the 2D code. Since the flame shape cannot be determined before the model runs, a point source, simulating burned gas behind the flame front, takes into account the thermal expansion of the gas, resulting in the shape of the flame in the model. And while the flow is incompressible, this density of this hot gas, lower than the flow density, is taken into account in the strength of the line source. The model was developed with equations that were derived to represent the flow field, which included stream functions, velocity potentials, and densities.

The model ultimately predicted that the flammable layer just before the flame front expands. This result is due to disturbances from the movement of the flame propagating towards the unburned flammable mixture. Flame speeds were also predicted as a function of the mixture equivalence ratio according to the following relationships

where S is the normal burning velocity, V_f the flame velocity in flame coordinates, and ?_f the equivalence ratio. Hirano and coworkers also found that the maximum flame propagation occurs when ?_f = 1.1. From the results above, this gives a propagation speed of around 4 times that of the laminar flame speed

1.2.2 Floor or Ceiling Layers

The earliest reported work on flame propagation through layered mixtures was done in 1965 by Phillips at the Safety in Mines Research Establishment. Experiments were conducted to characterize the behavior of flames propagating through a layered mixture along a ceiling or roof. This configuration is especially important to the mining industry where gases can collect along the roof a mine thus creating a flammable layered mixture.

This experimental setup consisted of an open-base gallery with a porous roof. 85% Methane mixed with nitrogen (which would not affect the concentration measurements) was allowed to diffuse through the porous roof and was distributed in a uniform manner all along the length of the gallery. The thickness of the flammable layer was governed by the molecular diffusivity of the gas mixture and the period of time during which the mixture was allowed to diffuse prior to ignition. . To ignite the system, a continuous spark igniter was placed at some distance from the roof, varying for each run.

Phillips found that the flame traveled through the mixture at a speed of nearly 183 cm/s, which is approximately 4.5 times faster than the average laminar flame speed of 40 cm/s for methane. Several experiments were done with fuel concentrations ranging from an equivalence ratio of zero to well above the rich flammability limit. Results showed that the thickness of the fuel layer had no effect on the flame speed. However, the flame volume depended on the amount of fuel in the system. It was also noted that, as the roughness of the porous roof increased, the flame speed slowed.

Feng, Lam, and Glassman (1975) studied the behavior patterns of flames through a layered system of methane and air. In their study, the methane was not allowed to diffuse through the air. Rather, it was setup as a combustible layer of homogeneously mixed methane-air on top of a layer of pure air. The experiments began with a rectangular gallery with a removable separator plate that could be set at various heights. Premixed fuel was injected into the top of the gallery, then the separator plate was removed and ignition took place.

The results showed that that the flame speed was related to the thickness of the combustible layer with respect to the gallery height. The flame speed increased as the ratio of the gallery height to the thickness of the combustible layer decreased. As the flame speed increased, the acceleration of the flame decreased. The only steady flame propagation speed was obtained when the gallery was set at its maximum height, which was 22.3 times the size of the combustible fuel layer, approximately 16 in. (40.64 cm). The flame speed was reported as 188 cm/s, very similar to the results of Phillips (1965).

The experiments of Feng and coworkers were performed in conjunction with a variety of analytical models that the group had developed. The first model showed that the flame speed is fastest, around 3 times that of the laminar flame speed, when the gallery's height is infinite. A second model dealt with a gallery with a finite height. This model predicted a steady flame speed for the gallery at its maximum height, which agreed with their experimental results.

Liebman and coworkers (1970) at the US Bureau of Mines studied the propagation of flames through heavier gases concentrated along floors and lighter gases on ceilings. Fuels that were studied included butane, propane, and propylene, at concentrations of 17% and 100%. It should be noted that all of these concentrations are in the fuel rich region. For the tests, the fuel was injected into variable height gallery at floor level. A soap film separator kept the fuel from diffusing into the gallery. When the correct mixture was obtained, the soap film was ruptured and the fuel was allowed to diffuse for a given amount of time, depending on the fuel layer thickness. After the fuel had diffused, a flame was ignited by use of a spark igniter.

Their experiments resulted in a flame spread rate that was approximately 3 times the laminar flame speed. These results were comparable to those of Phillips (1965) and Feng, et al. (1975). However, investigations concluded that the velocity of the flames were dependant upon thickness of the flammable zone and concentration gradients. Liebman and coworkers (1970) noted that the smaller the combustible layer, the slower the flame speed, which contradicted the conclusions of Phillips (1965). Results showed an increase of about 15 cm/s as the flammable layer thickness increased from .1 to .5 in. (.25 to 1.25 cm). These tests also included the use of an interferometer to analyze the fuel vapor concentration before and during ignition. The results showed that the fuel layer was disturbed by the movement of the flame as far as 10 cm in front of the flame. The researchers also noted that flames propagated through regions that are considered below the lean flammability limit for uniform mixtures, which can be directly related to the influence of combustion in closer fuel rich zones.

Kaptein and Hermance also studied the behaviors of flames propagating through a layered fuel-air system (1976). Their experimental apparatus was a 240 cm x 25 cm x 8cm open trough with plexiglass walls. The bottom of the trough was a wire mesh screen, which supported 100 micron glass beads. The apparatus was lowered into a fuel tray just enough to wet the bottom of the glass beads. Fuel was pulled to the top of the glass bead layer by capillary action and diffused vertically into the trough. The thickness of the layer depended upon the diffusion time. As the diffusion time increased, the layer thickness increased. Different fuels studied were benzene, hexane, heptane, and methyl alcohol. A hotwire was positioned at one end of the trough at approximately the stoichiometric mixture level.

Results of their study concluded that flames propagated through a layered mixture at velocities of 2 m/s to 4 m/s. The propagation speed depended upon which fuel was used as well as the thickness of the flammable layer. For each fuel, flame speed increased as the flammable layer thickness increased. Hexane produced the slowest propagation velocity at 180 cm/s a flame spreading through methyl alcohol propagated at the fastest, reaching 431 cm/s. As in previous examples of prior research presented above, the flame structure proved to be that of a triple flame, with a rich wing, a lean wing, and a trailing diffusion flame along the centerline.

1.3 NASA Layers Project History

Research on the gravitational effects on flames spreading through layered fuel-air mixtures formed by evaporating liquids is currently being conducted at NASA Glenn Research Center and Rowan University (Miller, et al, 2000, 2001, 2002). The “NASA Layers Project” has been ongoing since 1996. Various alcohols have been tested under normal and microgravity conditions. To date, the focus has been on cases where the mixture is at stoichiometric conditions or where it is fuel lean. Computer models are being developed at Rowan University (Marchese, 2000) while experiments are conducted at NASA Glenn Research Center.

1.3.1 Experimental

The emphasis up to now has been on quiescent tests wherein flames spread along the bottom of a gallery that contains the fuel. The gallery is 80 cm (31 inches) long and has a 10 cm (4 inch) square cross section, and can be used in experiments conducted both in normal and microgravity. The experiment consists of one aluminum and one Lexan sidewall, with a removable Lexan top. The base of this original Layers gallery is a fuel tray covered with porous bronze plate. This apparatus is shown in Figure 3.

Figure 3: NASA Layers porous plate floor apparatus (Miller, et al., 2002).

Before the experiments, fuel is poured though the porous bronze frit into the temperature controlled fuel tray. A cover is then placed over the frit. Any extraneous fuel vapors are exhausted from the duct with a fan. When the tray reaches its sought operating temperature, an actuator slides the cover off the frit. The fuel is left to diffuse for a certain amount of time, between 5 and 60 seconds. This diffusion time controls the thickness of the layer and is controlled by a timer relay. When the time expires, the hot wire igniter is automatically fired. As the flame spreads, cameras and an interferometer record data. Results from these tests have shown that the flame speed is a function of the temperature of the fuel as well as the diffusion time. However, this is more of an effect of the maximum fuel concentration in the layers rather than the actual layer thickness, which is controlled by the temperature. Figure 4 is an image of a flame spreading through a layered propanol-air mixture at 27 є C.

Figure 4: Flame spread through a layered propanol/air mixture above a 27єC porous solid surface in normal gravity (Miller, et al., 2002).

Other conclusions of this work are related to with the effect of the microgravity environment. Before this project, there had been no studies of flame spread through non-uniform mixtures in microgravity conditions. The 2.2 second Drop Tower at NASA Glenn Research Center was used to obtain microgravity conditions. Results showed a number of phenomena due to the effects of buoyancy. The height of the flame was shown to be higher in microgravity than in normal gravity. Also, the flame spread rate was higher in microgravity conditions. In some cases, flames spread at rates as much as 80% faster in microgravity than in normal gravity (Miller, et al., 2002).

1.3.2 Numerical Modeling

A numerical model was developed in concurrence to the experiments performed with the porous plate apparatus. Its purpose served as a means of predicting the outcome of experimental results, taking into account effects of such properties as diffusion time, chemical kinetics of the system, as well as the surface effects stemming from the contact with the porous bronze. The model can also be used to investigate quantities not measured in the experiment, such as the velocity or temperature fields.

This model was adapted from a model originally developed by Schiller and coworkers (1996) to model flame spread across a liquid fuel surface as opposed to a solid boundary. As detailed in the previous reference, the numerical model uses the SIMPLEC algorithm (Pantankar, 1980) and a hybrid differencing scheme to solve the gas-phase continuity, species, energy, x-y momentum equations and the liquid phase energy and x-y momentum equations.

In the work conducted at Rowan University prior to the present thesis, the effects of gravity on flame propagation through layered premixed gas mixtures were examined by simulating ignition and flame spread across propanol/air, methanol/air and ethanol/air mixtures at various initial pool temperatures in the superflash regime at normal gravity and at microgravity. To date, propanol/air results have been studied in the most detail. Propanol/air was selected because the predictions of the model with this fuel agreed best with experiments that were done with subflash pools (Schiller, et al., 1996). Ethanol has also been modeled, but the results have not been compared with experiments.

As shown in Figure 5, to simulate the experimental rig currently in use at NASA Glenn, the liquid tray was modeled as an 80 cm liquid surface with a fuel depth of 2 mm. The height of the gas phase above the liquid pool was 10 cm. The gas phase was modeled as closed at the ignition end of the domain and open at the top and right hand sides of the domain. The rectangular numerical domain used in this study consisted of 112 grid points in the x-direction, with 82 grid points in the gas phase y-direction and 32 grid points in the liquid phase y-direction.

To simulate the experiments, the model was initially run for a specified time period (e.g. 10 seconds) without introducing the ignition source. During this period, a time step of 5 ms was used. This allowed the fuel to vaporize at the pool surface and diffuse into the gas phase, setting up initial conditions that were consistent with experiments. The output from the non-reacting case was then used as an input to the reacting case. For the reacting case, a time step of .05 ms is used.

Figure 5: Schematic diagram of transient, two-dimensional flame spread model (Miller, et. al. 2002).

A summary and comparison of flame spread rate results predicted from the model and the experimental runs are presented below in Figure 6.

Figure 6: Comparison of numerical predictions and experimental measurements of flame spread through non-uniform mixtures in normal and reduced gravity (Miller, et al. 2002).

A generally good agreement between the model and the experimental runs can be seen in Figure 5. The model slightly under-predicted the flame spread rate for the 35єC cases, and over-predicted for 27єC cases. Aside from flame spread rates, the model also predicted flame height and fuel vapor concentrations, all of which agreed with experimental runs (Miller, et al., 2002). The model also agrees well with experiments in terms of the qualitative flame shape (See Figure 4).

In brief summary of the porous plate model and experimental results, the flame spread rate was shown to be faster in microgravity conditions than normal gravity and the flame heights were larger in microgravity. The model did not predict much of a difference in the unburned fuel/air mixture upstream of the propagating flame between normal and microgravity conditions. Therefore, Miller and coworkers concluded that buoyancy effects on the flow field, rather than the concentrations of the unburned fuel/air mixture, are the cause of the increased spread rate.

1.4 Objectives of the present study

Since 1996, Miller and coworkers have studied in detail the gravitational influences on flame spread through non-uniform mixtures. Each of the studies (both numerical and experimental) has employed the geometrical configuration described in the previous section in which the flame propagated above a porous fuel source. The main objective of the research study described in this thesis is to analyze the characteristic behavior of a flame propagating through a free, stratified fuel-air mixture.

Development of an apparatus to study flame propagation though a free, stratified layer is important on many accounts. Firstly, it eliminates surface contact between the flame and the floor, which in turn reduces heat transfer effects as well as effects on the flow field. Also, a free stream layer better approximates a fuel leak in microgravity conditions, where a stream of fuel is accumulating and be carried by very slow ventilation flows through surrounding air. Finally, such an apparatus provides an opportunity to stabilize a stationary flame. A stable stationary flame would yield an opportunity to perform additional quantitative flame diagnostics that are not possible with a propagating flame. A schematic diagram of the free-stream non-uniform layers apparatus that was designed, built and tested in the present study is shown in Figure 7. In this apparatus, a laminar air stream flows over a porous bronze airfoil that is supplied with a liquid fuel such as ethanol, propanol or methanol. This system results in a wake behind the airfoil that contains a laminar, non-uniform fuel/air mixture.

Figure 7: Schematic diagram of new technique to study flame spread through free stratified fuel/air mixtures (Hovermann, 2002).

To design this experiment, a 2-D, non-reacting CFD model of the system was developed using the commercially available FLUENT CFD code. This model accounts for diffusion and temperature of the fuels used. The model was developed to give a better understanding of the flow characteristics in the experimental apparatus, such as velocity profiles, fuel concentration, and even an estimated flame shape. Analytical calculations were also performed to determine the conditions under which a stable, stationary (i.e. non-propagating) flame could exist in the wake of the airfoil. In this configuration, the velocity of the propagating flame is balanced by the local convective velocity of the fuel/air mixture. The calculations show the precise locations in the flow field wherein a stoichiometric fuel/air mixture exists.

Once the geometry was characterized numerically and found to be reasonable, the next step was to build the new experimental apparatus. The apparatus consists of a 79 cm long, roughly 10 cm square flow duct. A heated, porous bronze, fuel emitting airfoil is positioned 10 cm from the inlet along the centerline while a slow stream of air is blown parallel to the airfoil, creating the layered mixture in the laminar wake region.

After the apparatus was built, cold flow tests were performed. Cold flow testing included smoke tests which visualized the flow to ensure a steady, laminar quality, as well as hotwire anemometer and thermocouple scans to measure velocity and temperature profiles, respectively; all of these agreed with model predictions.

After completion of the cold flow tests, ignition and combustion experiments were performed. Image sequences of side and top views of propagating flames were obtained. Preliminary experimental results show that it is possible to obtain a propagating flame in a non-uniform free layer with flame spread rates of up to 180 cm/s in flame fixed coordinates.

Using this apparatus, the hope is to form and, under certain conditions, stabilize a non-uniform, premixed flame away from the influence of solid boundaries. In doing so, the goal is to determine flammability regions, stability limits, and flame shape for flames in flowing, non-uniform mixtures in normal and microgravity; and finally, to measure the flame spread velocity as a function of fuel distribution and compare the results to uniform premixed flames.

1.5 Organization of the Thesis

This thesis presents the results of an experimental study aimed at developing and testing a new apparatus to study the propagation of flames through free, stratified fuel/air mixtures. Chapter 2 details the development of a computational model of the fluid dynamics of the experimental apparatus that was developed using the commercial CFD software FLUENT and mesh generation software Gambit. The model was used to design the experimental flow duct and to determine the optimum location(s) for ignition and the locations at which a stable non-propagating flame is possible. Chapter 3 details the CFD modeling results, including contour plots of species, temperature and velocity, as well as the buoyancy effects seen in the modeling.

Chapter 4 describes the development of the experimental apparatus used to create a free, stratified fuel/air mixture. The apparatus uses a porous airfoil to inject fuel into a laminar flow duct that uses a Coanda air inducer. Instrumentation includes thermocouples allowing for measurement of fuel stream and airfoil surface temperatures, a hotwire anemometer for velocity scans, smoke wire for flow visualization, as well as color video cameras to record flame spread tests.

Chapter 5 details the experimental tests run to date, including cold flow and combustions tests. Cold flow testing, in which experiments are compared directly with the computational fluid dynamics modeling results of Chapter 3, included velocity measurements using hot wire anemometry, temperature measurements and smoke wire tests. The results conclusively show that the experimental configuration successfully creates a symmetric, low velocity, laminar, stratified fuel/air mixture. Prior to ignition, fuel vapor profiles were qualitatively measured using a Rainbow Schlieren system. A series of combustion experiments were conducted and flame spread rates were measured. Preliminary experimental results show that it is possible to obtain a propagating flame in a non-uniform free layer with flame spread rates of up to 180 cm/s in flame fixed coordinates. Image sequences of the side view of the flame spread, along with spread rates, are presented in Chapter 5.

Chapter 6 provides conclusions and suggestions for future work on both numerical and experimental aspects of this research.

2 Computational Fluid Dynamics Modeling Setup

To get a better understanding of the initial mixing conditions and flow characteristics of the fuel systems used in the experiments, a computational fluid dynamics (CFD) model was developed using commercially available software. In this case, FLUENT versions 5 and later 6 were used to model the flow, fuel concentration, and temperature for laminar flow over a NACA 0012 airfoil shape within a two-dimensional duct.

The goal of the modeling effort was to predict the inlet velocity and airfoil temperature that would produce a flammable mixture within the duct, and determine the extent of that region. It should be noted that this model developed does not account for chemical reactions and heat release from a flame, and is used to predict flow conditions upstream of the flame and/or prior to introduction of the ignition source.

2.1 Geometry Definition

When using FLUENT to model fluid flow, one must first use a separate mesh generation package, or pre-processor, in order to set up the proper geometry. For this study, Gambit 2.0 was used to create the mesh. Given the dimensions of the duct along with 48 total x-y coordinates of the airfoil (Figure 8), the geometry was entered as vertices into Gambit. From here, the vertices were connected to create the edges of the 2-D model. The next step in Gambit is to take the edges and create faces. Once the vertices, edges, and faces are created, actual meshing process can begin.



Figure 8: NACA 0012 airfoil schematic used to generate grid points to enter into Gambit.

2.2 Mesh Generation

To set up the mesh, the first step is to place nodes (points where the grid lines of the mesh connect) on the edges. This process is done by specifying constant interval spacing between the nodes, which provides a uniform mesh, or a gradually increasing/decreasing spacing, which provides a non-uniform mesh with a finer resolution across a certain area, such as along the centerline of the geometry. When the nodes are created, one can then generate the actual mesh along the faces. A few different options for mesh generation are available within Gambit, including those consisting of triangular elements or quadrilateral elements. After trying various combinations of meshing for the experimental geometry, a uniform mesh of 12000 quadrilateral elements was chosen because of the relatively simple, 2-D planar geometry (Figure 9).

Figure 9: Computational grid for FLUENT CFD modeling.

2.3 Zone/Boundary Sets

After meshing, boundary zones are created on the geometry. These zones are used later by FLUENT to specify the boundary conditions. For this study, the top and bottom of the duct along with the edges of the airfoil were specified as separate zones called “walls.” A “wall” is defined as a surface that is assumed to be solid that no fluid can flow through. The front of the duct was specified as a “velocity inlet,” and the rear of the duct was specified as a “pressure outlet.” A “velocity inlet” is used to define the velocity and scalar properties of the flow at inlet boundaries and a “pressure outlet” is used to define the static pressure at flow outlets. It is also noted that the zone types (wall, velocity inlet, etc.) can be changed within FLUENT as well, as long as zones are defined. Once the mesh and zones are created, the mesh is then imported into FLUENT.

2.4 FLUENT Setup

The first steps taken after importing the mesh geometry into FLUENT involve checking the mesh/grid for errors. Checking the grid assures that all zones are present and all dimensions are correct. It is also important to check the volume and make sure that it is not negative. If the volume is shown as negative, there is a problem with the grid, since volume cannot be negative.

The grid can also be displayed to ensure that the mesh generation is qualitatively reasonable (See Figure 9). When the grid is checked completely and free of errors, a scale and units can be assigned. Since Gambit inputs the coordinates as non-dimensional numbers, the grid can be scaled however one chooses. For this study, the grid was created in inches, then scaled to centimeters. The maximum and minimum values for the x and y directions are given in the scaling window. Since the front tip of the airfoil was set as the origin (0, 0) when drawn in Gambit, the minimum x value was -10 cm (-3.937 inches) with a minimum y value of -5.3975 cm (-2.125 inches). Once the grid was set, the solver and boundary conditions of the system were then set and cases were run and analyzed.

2.4.1 Defining the Models

To run the cases, the model properties must be set. Model properties include the internal FLUENT solver type, number and types of species to be used in the model, the species/material fluid and thermal properties, as well as model operating conditions and grid boundary conditions. The following settings were used to create the model in FLUENT.

2.4.1.1 Solver

Solver options include Segregated and Coupled, along with sub-options under each solver such as steady/unsteady and implicit/explicit. The Segregated solver is recommended for slow, laminar flows, while the coupled solver is recommended for turbulent flow. For this study, the options chosen were:

· Segregated

· Steady, and

· 2-D.

2.4.1.2 Species

In the species settings, one can select the number of different species to be analyzed in the simulation and add each species to the database. For this study, the options chosen were:

· Multiple Species,

· Ethyl alcohol-air mixture, and

· Multicomponent diffusion

2.4.1.3 Energy

Enabling energy in the solver is needed for the incompressible ideal gas assumption. Accordingly, the option chosen here was: enable energy.

2.4.1.4 Viscous

The viscous model option gives the user the choice between different turbulence models such as k-epsilon, Spalart-Reynolds, and Eddy Dissipation, as well as a laminar model and inviscid model. For the model and experiment, the goal was to keep the flow laminar to avoid unwanted mixing of the fuel. Accordingly, for the viscous model option, a laminar flow model was employed.

2.4.2 Defining the Material Properties

This section of the input contains the options for the materials chosen as the working fluid. For this case, the working fluid is the ethanol-air mixture. Properties that can be specified in this section are density, specific heat, and thermal conductivity. For this study, the following options were chosen:

· density (air) - incompressible ideal gas [kg/m³]

· specific heat (air)- constant: 1000 [J/kg-K] (default value)

· thermal conductivity (air)- constant: .0454 [w/m-K] (default value)

· viscosity (air)- constant: 1.72e-05 [kg/m-s] (default value)

· mass diffusivity (ethanol into air) - constant dilute

approximation: 1.38e-05 m²/s

As detailed in Appendix A, the mass diffusivity for ethanol/air was calculated using Chapman-Enskog theory (Bird, Stewart and Lightfoot, 1960).

2.4.3 Defining the Operating Conditions

The operating conditions include gravity and pressure. Gravity can be entered in values of m/s² in x and y components. Operating pressure is also set in this section. In this study, the duct was modeled for microgravity experiments and normal gravity experiments. In the normal gravity modeling, computations were performed with the gravity vector either parallel (1-gX) or perpendicular (1-gY) to the duct since the experimental apparatus is capable of operating both horizontally or vertically. Accordingly, the following options were chosen for this study:

0-g cases

x: 0 m²/s

y: 0 m²/s

pressure: 101325 Pa

1-gX cases

x: 9.81 m²/s

y: 0 m²/s

pressure: 101325 Pa

1-gY cases

x: 0 m²/s

y: -9.81 m²/s

pressure: 101325 Pa

2.4.4 Defining the Boundary Conditions

Proper specification of the boundary conditions is a vital step in accurately modeling a fluid flow system such as the experimental system under consideration in this thesis. In FLUENT, boundary conditions must be specified at each surface defined in the mesh generation process described in Section 2.3. Specifically, information about the velocity, temperature and species mass fractions must be specified at each surface. For surfaces that have been defined as “walls,” properties can be set to include certain mass fractions of species along a wall, as well as the thermal conditions by specifying temperature, heat flux, radiation, or convection, or a combination. For surfaces that have been defined as “velocity inlets.” input specifications include mass fraction of species and fluid velocity magnitude and component flow direction. For surfaces that have been defined as “pressure outlet” surfaces, the sole input specification is a pressure value. For the modeling performed in this study, the boundary conditions are summarized in Table 1. Once all the models, operating conditions, and boundary conditions are specified, the FLUENT code can be executed.

Table 1: Boundary condition specification for FLUENT modeling of free stratified layer apparatus.

Zone	Type	Boundary Conditions
Airfoil	Wall	Species boundary condition - C2H5OH Specified mass fraction: .394 (See Appendix B for related calculations) Temperature: 323 K, constant No Slip
Duct bottom	Wall	Species boundary condition: all species zero mass fraction Temperature: 300 K No Slip
Duct top	Wall	Species boundary condition: all species zero mass fraction Temperature: 300 K No Slip
Duct front	Velocity inlet	Species boundary condition: O2 Specified mass fraction: .233 (FLUENT assumes remaining mass fraction to be N2. CO2, H2O set to zero Temperature: 300 K, constant Velocity magnitude and direction Velocity: constant 5, 10, 20, 40 cm/s (depending on case) x component: 1 (unit vector direction) y component: 0
Duct rear	Pressure Outlet	Outlet pressure: 101325 Pa

2.4.5 Executing the FLUENT Code

Each case must be initialized before the FLUENT code begins iterating toward a converged solution. Initializing the case essentially provides an initial guess for the first iteration of the solution. In the initialization process, the user must specify which zones will be provided with initial conditions. For the modeling performed in this study the option chosen was to compute from all zones. The final initialization step is for the user to enter the maximum number of iterations, after which the simulation begins. For the modeling performed in this study, the number of iterations ranged between 100 and 1000 depending on the case being run and how long it took to converge.

Eight different model properties were monitored by FLUENT's solver and checked for convergence. This criterion requires that the scaled residuals decrease to 10 ^-3 for all equations except the energy equation, for which the criterion is 10 ^-6. At the end of each solver iteration, the residual sum for each of the conserved variables is computed and stored, thus recording the convergence history.

Table 2 is a list of variables and their respective convergence criteria (Note: CO₂ and H₂O appear as species contained in air).

Table 2: Variables and convergence criteria for FLUENT simulation of free layers apparatus.

Variable	Convergence Criterion
Continuity	0.001
X-velocity	0.001
Y-velocity	0.001
Energy	1e-06
C2H5OH	0.001
O2	0.001
CO2	0.001
H2O	0.001

If the solution converges, the results can be analyzed. If the solution does not converge within the given number of iterations, one can request additional iterations or check the results given at that point to determine whether additional iterations will converge toward a physical solution.

3 Modeling results

3.1 FLUENT Computational Fluid Dynamics

Using the FLUENT CFD model described in Chapter 2, a series of 27 simulations were executed. The initial simulation matrix included cases for inflow velocities of 10 cm/s, 20 cm/s, and 40 cm/s, with runs in 0-g, 1-g (-y direction), and 1-g (+x direction) for a total of nine different cases initially. Once the actual experimental testing began, more model simulations were executed, including 1-g (-y direction) cases at 25 cm/s, and well as cases at 25 and 40 cm/s using all air and no fuel flowing through the duct. The latter cases were performed to simulate conditions in the duct for cold flow tests. Table 3 contains a matrix of all FLUENT cases executed to date.

Table 3: Simulation matrix for FLUENT modeling of free layer apparatus.

Run Number	Geometry	X-velocity inlet (cm/s)	Species	T airfoil (K)	gx	gy	Convergence
001	NACA 0012 Straight Duct	10 cm/s	Ethanol/air	323 K	0	0	Yes
002	NACA 0012 Straight Duct	10 cm/s	Ethanol/air	323 K	0	-1	Yes
003	NACA 0012 Straight Duct	10 cm/s	Ethanol/air	323 K	1	0	Yes
004	NACA 0012 Straight Duct	20 cm/s	Ethanol/air	323 K	0	0	Yes
005	NACA 0012 Straight Duct	20 cm/s	Ethanol/air	323 K	0	-1	Yes
006	NACA 0012 Straight Duct	20 cm/s	Ethanol/air	323 K	1	0	Yes
007	NACA 0012 Straight Duct	40 cm/s	Ethanol/air	323 K	0	0	Yes
008	NACA 0012 Straight Duct	40 cm/s	Ethanol/air	323 K	0	-1	Yes
009	NACA 0012 Straight Duct	40 cm/s	Ethanol/air	323 K	1	0	Yes
010	NACA 0012 Straight Duct	25 cm/s	Ethanol/air	323 K	0	-1	Yes
011	NACA 0012 Straight Duct	25 cm/s	Air	293 K	0	0	Yes
012	NACA 0012 Straight Duct	25 cm/s	Air	323	0	-1	Yes
013	NACA 0012 Straight Duct	40 cm/s	Air	323	0	-1	Yes
014	NACA 0012 Straight Duct	40 cm/s	Ethanol/air	338	0	-1	Yes
015	NACA 0012 Straight Duct	80 cm/s	Ethanol/air	323	0	0	Yes
016	NACA 0012 Straight Duct	1 cm/s	Ethanol/air	323	0	0	Yes
017	NACA 0012 Straight Duct	1 cm/s	Ethanol/air	300	0	-1	No
018	NACA 0012 Straight Duct	1 cm/s	Ethanol/air	323	0	-1	No
019	NACA 0012 Straight Duct	5 cm/s	Ethanol/air	323	0	-1	No
020	NACA 0012 Straight Duct	5 cm/s	Ethanol/air	323	1	0	No
021	NACA 0012 Straight Duct	10 cm/s	Ethanol/air	323	-1	0	Yes
022	NACA 0012 Diverging Duct	10 cm/s	Ethanol/air	323	0	0	Yes
023	NACA 0012 Diverging Duct	10 cm/s	Ethanol/air	323	0	-1	Yes
024	NACA 0012 Diverging Duct	20 cm/s	Ethanol/air	323	0	0	Yes
025	NACA 0012 Diverging Duct	20 cm/s	Ethanol/air	323	0	-1	Yes
026	NACA 0012 Diverging Duct	50 cm/s	Ethanol/air	323	0	0	Yes
027	NACA 0012 Diverging Duct	80 cm/s	Ethanol/air	323	0	0	Yes

After iterations converge (and even before so), it is possible to analyze many of the results calculated by FLUENT (although the reported results of cases that have not converged do not represent a physical solution). The results that are reported by FLUENT include velocity vectors, surface integrals (areas, integrals, mass flow rates, and weighted flow rates), volume integrals, flux reports, force reports, path lines, particle tracks, and contour and x-y plots of various system variables. System variables include many values such as pressure, density, velocity, species, properties, wall fluxes, and residuals. Within these plotting parameters are subcategories of each variable (i.e. static pressure, total pressure, absolute pressure, velocity magnitude, stream function, radial velocity, etc.).

The results that were used extensively for design and analysis of the experimental apparatus include contour plots of fuel mole fraction, velocity contours, velocity vectors and density contours. In addition to the variables that are output directly by FLUENT, it is possible to do a variety of post processing using the FLUENT post processor. For the present study, additional post processing included equivalence ratio contour plots and equivalence ratio X-Y plots. Each of these results is summarized in detail below.