Distribution of Energy in Large Compartment Fires

Transcription

1 Bachelor of Engineering Thesis Distribution of Energy in Large Compartment Fires Student Name: Benjamin Linnan Student Number: Course Code: CIVL4580 Supervisors: Dr C. Maluk and Prof J. L. Torero Submission Date: 28 October 2016 A thesis submitted in partial fulfilment of the requirements of the Bachelor of Engineering degree in Civil Engineering 0

2 1

3 Benjamin Linnan 62 Lake Clarendon Way Lake Clarendon QLD October 2016 Professor José L. Torero Head of the School of Civil Engineering The University of Queensland St Lucia QLD 4072 Dear Sir, I hereby submit my Thesis titled Distribution of Energy in Large Compartment Fires for consideration as partial fulfilment of the Bachelor of Engineering degree. All the work contained within this Thesis is my original work except where otherwise acknowledged. I understand that this thesis may be made publicly available and reproduced by The University or Queensland unless a limited term embargo on publication has been negotiated with a sponsor. Yours sincerely, Benjamin Linnan Student ID:

4 3

5 ABSTRACT A series of large-scale experiments were conducted in Watford, England in 2013 for the primary reason of investigating the distribution of energy within large compartment fires for structural design purposes. These experiments systematically tested the traditional fire regimes (Regime I and Regime II which correspond to under and overventilated conditions respectively) while assessing a range of potential fire behaviours. The results of the experimental data analysis aim to provide a platform for characterising fires in large spaces. A large compartment was carefully designed to house and withstand the series of experiments. In total 10 experiments were analysed to obtain the energy distribution within the compartment. In order to collect the data required to provide a detailed analysis a set of specialised data collection instruments were used. Thin Skin Calorimeters used to measure the incident heat flux for each surface. Ultimately leading to quantification of heat losses into the surfaces. Thermocouples used to record the temperatures both within the compartment and leaving the compartment. Velocity Probes used to quantify the mass flow of the hot gasses leaving through the opening. Mass Flow Controllers used to measure the amount of heat supplied to the compartment. To control the experiment conditions, the following experimental apparatus was used: Gas burners The source heat supply to the compartment Shutters The apparatus used to systematically control the ventilation conditions during each experiment. The key principle of the experimental data analysis was energy conservation within a control volume. Ultimately, the following losses as a rate of energy transfer (i.e. heat) were accounted for: Heat losses into surfaces Heat losses out of the openings Heat storage within the gas phase i

6 Coupled with a recorded heat supply to the compartment through the use of the mass flow controllers the distribution of energy within the compartment could be quantified. Overall, for the experimental series conducted the largest portion of heat loss was accounted for by the hot gasses leaving the openings of the compartment. The largest heat loss accounted for by a surface was the ceiling which generally received approximately 20% of the heat loss for constant ventilation conditions. However, this increased in magnitude for the variable ventilation experiment simulations peaking at 40%. The other surfaces within the compartment accounted for a significantly smaller portion of the heat losses. Finally, the assessment between ventilation conditions demonstrated that thermal boundary conditions were not necessarily more severe for a Regime I fire as assumed by the original compartment fire framework. ii

7 ACKNOWLEDGEMENTS Firstly, I would like to express my genuine gratitude to my supervisor Dr Cristian Maluk for his relentless support, aid and patience in my progression for this thesis. Without his calm demeanour, continuous availability and open mind my progress in writing and understanding the concepts behind this dissertation would not have been possible. For this I am truly grateful to have been supervised by Dr Maluk. Secondly, I would like to thank Prof José L. Torero for his guidance provided in my decision making throughout the year. This expression of gratitude extends to giving me the opportunity to work with the data produced by the fire testing done in Watford. I wish to also thank Dr Juan P. Hidalgo for his continuous support and encouragement throughout the final stages of my thesis. Dr Hidalgo s feedback has been invaluable. A special thanks is extended to Andy Wong, Pratvi Patel and Tam Do for their efforts with respect to the data analysis prior to my continuation of their work. Without them I would not have had the great platform to build my thesis on and without their help I would not have been able to analyse all of the data available. A special mention goes to Andy Wong for his aid and contribution to my ability to efficiently and swiftly learn MATLAB coding. Finally, I wish to thank my friends and family for their moral support and technical guidance. Without them everything I have achieved would not have been possible. iii

8 Contents Abstract... i Acknowledgements... iii List of Tables... ix List of Figures... x Notation... xv 1 Introduction Background Scope of the research Research Objective Out of Scope Assumptions Outline of the chapters Literature Review Key Definitions Traditional Methods of Estimating Fire Severity Outcomes of Past Work Research Context Description of the Compartment Compartment Location and Dimensions Compartment Floor Compartment Floor Layout Compartment Floor Materials Compartment Walls Compartment Wall Layout Compartment Wall Materials Compartment Roof Compartment Roof Layout Compartment Roof Materials Compartment Opening Compartment Opening Layout Description of the instrumentation iv

9 5.1 Overview of Instrumentation Thermocouples Thin Skin Calorimeters (TSCs) Velocities probes Fire Simulation Gas Burners Description of the Experiments Experiment Overview Fire Modes Ventilation Modes Shutter Rate Modes Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment v

10 7.8.1 Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Fire Mode Ventilation Mode Summary of Events Experiment Summary Experimental Data Analysis MATLAB Raw Data Treatment Velocity Calculation Incident Heat Flux Calculation Energy Conservation in a Control Volume Heat Supply to the Compartment Net Heat Flux Calculation Gas Phase Heat Storage Calculation Heat Loss out of the Opening Sensitivity Analysis Time-step and Material Thickness Sensitivity Analysis Material Composition Sensitivity Analysis Smoothing of qinc Sensitivity Analysis Results and discussion Fully-developed Fire Experiments 1, 4, 7 and Experiment Experiment Experiment Experiment vi

11 9.1.5 Comparison of Fully-developed Fires Growing Fire Experiments 2 and Experiment Experiment Comparison of Growing Fires Travelling Fire Experiments 3, 6, 9 and Experiment Experiment Experiment Experiment Comparison of Traveling Fires Conclusions and Recommendations Key Findings Further Research References Appendices Appendix A Time-Step and Material Thickness Sensitivity Analysis Appendix B Material Composition Sensitivity Analysis Appendix C Experimental Flow Results Experiment Experiment Experiment Experiment Experiment Experiment Experiment Experiment Experiment Appendix D Experimental Heat Flux Results Experiment Experiment Experiment Experiment Experiment Experiment vii

12 Experiment Experiment Experiment viii

13 LIST OF TABLES Table 1: Test instrumentation resources Table 2: Thermocouple locations Table 3: Incident heat flux sensors of interior surfaces Table 4: Thin Skin Calorimeter locations Table 5: Characterisation of fire modes Table 6: Characterisation of ventilation modes Table 7: Characterisation of shutter rate modes Table 8: Noted events during Experiment Table 9: Noted events during Experiment Table 10: Noted events during Experiment Table 11: Noted events during Experiment Table 12: Noted events during Experiment Table 13: Noted events during Experiment Table 14: Noted events during Experiment 7. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter Table 15: Noted events during Experiment 8. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter Table 16: Noted events during Experiment 9. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter Table 17: Noted events during Experiment 10. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter Table 18: Summary of experiments Table 19: Fourier Number results for appropriate materials Table 20: Material properties Table 21: Material structure composition Table 22: Modified material composition for modelling ix

14 LIST OF FIGURES Figure 1: Characteristic difference between Regime I and Regime II fires... 2 Figure 2: Plan view of the large compartment setup withing the Burn Hall (Hidalgo, 2013) Figure 3: Isometric view of the large compartment (south-west view) (Hidalgo, 2013) Figure 4: Compartment Floor Layout Figure 5: Gas Burner Arrangement Figure 6: Compartment Floor Material Composition for the Gas Burner Experiment Series Figure 7: Compartment Wall Material Composition Figure 8: Compartment Roof Material Composition Figure 9: (a) compartment opening layout. (b) shutter system. (c) dimensions and composition of fire shutters Figure 10: Thermocouple locations (Plan view) Figure 11: (a) thermocouple location at the opening (Elevation view). (b) thermocouple location within the compartment (Elevation view) Figure 12: Thin Skin Calorimeter (TSC) Figure 13: Arrangement of roof and floor TSCs Figure 14: Arrangement of back wall TSCs Figure 15: Arrangement of left and right wall TSCs Figure 16: (a) Image of the velocity probes used during the experiments. (b) Arrangement of velocity probes at the openings Figure 17: (a) Gas burner dimensions and labelling. (b) Gas burner pilot Figure 18: Gas burner layout Figure 19: Inverse ventilation factor relationship with the number of openings Figure 20: Over-ventilated conditions arrangement Figure 21: Under-ventilated conditions arrangement Figure 22: Starting position of ventilation mode Figure 23: Systematic removal of shutters for ventilation mode Figure 24: Experiment 1 HRR and target HRR Figure 25: Experiment 2 HRR and target HRR Figure 26: Experiment 3 HRR and target HRR x

15 Figure 27: Experiment 4 HRR and target HRR Figure 28: Experiment 5 HRR and target HRR Figure 29: Experiment 6 HRR and target HRR Figure 30: Experiment 7 HRR and target HRR Figure 31: Experiment 8 HRR and target HRR Figure 32: Experiment 9 HRR and target HRR Figure 33: Experiment 10 HRR and target HRR Figure 34: Opening 3 temperature contour Figure 35: Opening 8 temperature contour Figure 36: Opening 13 temperature contour Figure 37: Conduction correction factor, C, as a function of temperature Figure 38: Element increments for temperature profile calculation Figure 39: Element increments between two different materials Figure 40: Probe Locations at the Openings Figure 41: (a) Typical Case Calculation Method of Energy Leaving the Compartment. (b) Special Case Calculation Method of Energy Leaving the Compartment Figure 42: Sensitivity analysis of the change in time-step and material thickness to 0.25s and 2.5mm respectively Figure 43: Sensitivity analysis of the change in material composition for the back wall Figure 44: Result of smoothing the calculated qinc Figure 45: Temperature profile before and after the smoothing of qinc Figure 46: Change in calculated qnet after smoothing qinc Figure 47: Outflow velocities for each opening during Experiment Figure 48: Inflow velocities for each opening during Experiment Figure 49: Cross-sections take for analysing the temperature contours within the compartment Figure 50: Temperature contours of cross sections 1-1 to at t = 875s for Experiment Figure 51: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 52: Energy distribution recorded for Experiment xi

16 Figure 53: Normalised energy distribution with respect to the energy supply for Experiment Figure 54: Energy distribution recorded for Experiment Figure 55: Normalised energy distribution with respect to the energy supply for Experiment Figure 56: Energy distribution recorded for Experiment Figure 57: Normalised energy distribution with respect to the energy supply for Experiment Figure 58: Energy distribution recorded for Experiment Figure 59: Normalised energy distribution with respect to the energy supply for Experiment Figure 60: Comparison of fully-developed fire results across different ventilation conditions (ceiling and opening heat losses) Figure 61: Comparison of fully-developed fire results across different ventilation conditions (surface heat losses) Figure 62: Energy distribution recorded for Experiment Figure 63: Normalised energy distribution with respect to the energy supply for Experiment Figure 64: Energy distribution recorded for Experiment Figure 65: Normalised energy distribution with respect to the energy supply for Experiment Figure 66: Comparison of growing fire results across different ventilation conditions (ceiling and opening heat losses) Figure 67: Comparison of growing fire results across different ventilation conditions (surface heat losses) Figure 68: Energy distribution recorded for Experiment Figure 69: Normalised energy distribution with respect to the energy supply for Experiment Figure 70: Energy distribution recorded for Experiment Figure 71: Normalised energy distribution with respect to the energy supply for Experiment Figure 72: Energy distribution recorded for Experiment Figure 73: Normalised energy distribution with respect to the energy supply for Experiment xii

17 Figure 74: Energy distribution recorded for Experiment Figure 75: Normalised energy distribution with respect to the energy supply for Experiment Figure 76: Comparison of travelling fire results across different ventilation conditions (ceiling and opening heat losses) Figure 77: Comparison of travelling fire results across different ventilation conditions (surface heat losses) Figure 78: Sensitivity analysis of the change in time-step and material thickness to s and 1.25mm respectively Figure 79: Sensitivity analysis of the change in material composition for the ceiling Figure 80: Sensitivity analysis of the change in material composition for the floor Figure 81: Sensitivity analysis of the change in material composition for the left wall Figure 82: Sensitivity analysis of the change in material composition for the right wall Figure 83: Outflow velocities for each opening during Experiment Figure 84: Inflow velocities for each opening during Experiment Figure 85: Outflow velocities for each opening during Experiment Figure 86: Inflow velocities for each opening during Experiment Figure 87: Outflow velocities for each opening during Experiment Figure 88: Inflow velocities for each opening during Experiment Figure 89: Outflow velocities for each opening during Experiment Figure 90: Inflow velocities for each opening during Experiment Figure 91: Outflow velocities for each opening during Experiment Figure 92: Inflow velocities for each opening during Experiment Figure 93: Outflow velocities for each opening during Experiment Figure 94: Inflow velocities for each opening during Experiment Figure 95: Outflow velocities for each opening during Experiment Figure 96: Inflow velocities for each opening during Experiment Figure 97: Outflow velocities for each opening during Experiment Figure 98: Inflow velocities for each opening during Experiment Figure 99: Outflow velocities for each opening during Experiment Figure 100: Inflow velocities for each opening during Experiment xiii

18 Figure 101: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 102: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 103: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 104: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 105: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 106: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 107: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Figure 108: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 9 (Note*: the scale was increased for this experiment) Figure 109: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 10 (Note*: the scale was increased for this experiment) xiv

19 NOTATION A i = Tributary area A T = Area of the compartment surfaces excluding the floor and opening A O = Area of opening C = Thin skin calorimeter (TSC) conduction correct factor c P = Specific heat capacity F o = Fourier s Number H = Height H c = Heat of combustion h = Tributary height h c = Convective heat transfer coefficient h c_disc = Convective heat transfer coefficient of the TSC disc h c_solid_exposed = Convective heat transfer coefficient of the exposed solid h c_solid_unexposed = Convective heat transfer coefficient of the unexposed solid i = Time increment j = Element increment k = Thermal conductivity L = Characteristic Length M air = Molar mass of air m = Mass m disc = Mass of TSC disc m = Mass flow rate N = Element increment P = Pressure P atm = Atmospheric pressure P r = Prandtl s Number Q = Rate of heat Q gas = Rate of heat in the gasses Q in,propane = Rate of heat supplied by the burnt propane Q loss,opening = Rate of heat loss out of the opening Q loss,surfaces = Rate of heat loss into the surfaces q inc = Incident heat flux q net = Net heat flux R = Ideal gas constant R e = Reynold s Number S disc = Surface are of the TSC disc T = Temperature T amb = Ambient temperature T disc = Temperature of the TSC disc T g = Gas temperature T K = Temperature conversion factor from degrees to kelvin T s = Surface temperature T TC = Temperature measure by the thermocouple t = time V = Volume = Characteristic spread rate V S xv

20 V BO V v w x α disc α solid γ ε disc ε solid ν ρ σ Ф τ = Burnout rate of fuel = Volume flow rate = Velocity = Tributary width = Element width = Absorptivity of the disc = Absorptivity of the solid = McCaffrey probe constant = Emissivity of the disc = Emissivity of the solid = Kinematic viscosity = Density = The sum of = Stefan-Boltzmann s = Inverse ventilation factor = Thermal diffusivity xvi

21 1 INTRODUCTION 1.1 BACKGROUND An internationally significant milestone in terms of fire safety engineering occurred with Ingberg s publication of his works (Ingberg, 1928). This work was mainly focused on the fire severity within a building. To understand the fire severity typical fuel loads were placed within the experimental compartments and assessed until burnout occurred. What followed was the pioneering period of fire safety engineering with extensive works being produced to understand key elements of the fire safety engineering framework. After the early research in the behaviour of fire it was established that there is a strong correlation between compartment fire characteristics and their surroundings such as geometry, ventilation, fuel, etc. all of which combine to produce a very complex reaction in the event of a fire (Drysdale, 2011). With this high level of complexity between the fire dynamics in the compartment and the compartment itself came a move towards simplifying the problem in order to quickly quantify the performance of fire safety designs. To elaborate, based on a predetermined design of a compartment the thermal boundary conditions could be estimated resulting in the knowledge of the potential thermal loads on structural elements which can ultimately be designed for. A large factor in this simplification was the limitation of compartment size and the assumption that the fire was uniform throughout the compartment. Thomas and Heseldon (1972) had demonstrated with limiting the size of the compartment and opening the fire would be controlled by the amount of air that could flow into the compartment, this typified a Regime I fire. Whereas, a fire that was not restricted due to the amount of air that could flow into the compartment and only by the amount of fuel surface area was classified as a Regime II fire (Thomas and Heseldon, 1972). The traditional fire regimes were summarised by Majdalani et al. A Regime I fire is classified as a fully-developed fire controlled by the opening geometry and thus a correlation for the fire and its duration could be established against the ventilation (Majdalani et al., 2015). A key characteristic of a Regime I fire is a uniformly hot compartment while the flow in and out of the compartment is driven by a pressure 1

22 difference between the inside and outside of the compartment. The temperatures within the compartment are expected to be high and heat transfer uniform. A Regime II fire is also classified as a fully-developed fire but it is controlled by the fuel surface area as there is large openings allowing for ample air supply resulting in no correlation with the fire and duration against ventilation conditions (Majdalani et al., 2015). As the large opening does not allow for the retention of the hot gasses produced by the fire a uniform hot layer cannot form. Internal temperatures were also theoretically lower as a large portion of the hot gasses could flow out of the compartment and this resulted in a non-uniform heat transfer within the compartment. See Figure 1 for an illustration of the differences between Regime I and Regime II fires. Experimental results for a Regime II fire showed large amounts of scatter and the thermal load produced by a Regime I fire was deemed more severe. This resulted in the compartment fire framework as it is today being designed based on Regime I fire behaviour. Figure 1: Characteristic difference between Regime I and Regime II fires Driven by increased architectural needs and technical advances; large, open floor plan compartments became and keep to become prevalent within the modern built environment (Schumacher, 2011). As a result, there is an increased risk to the occupants of these buildings as key fundamentals of fire safety design are inherently lost. For example, large open spaces may result in longer periods before detection of a fire while compartmentation is also reduced. Compartmentation is the confinement of space using surfaces that are designed to withstand the thermal conditions within a compartment fire until burnout. As a result, the traditional compartment was designed to contain the fire and limit the fire spread through the rest of the building. Moreover, 2

23 less compartmentation has been driven by the need for open spaces of commercial (i.e. office) occupation. As previously highlighted compartmentation is essential in controlling fire spread which has two major effects, property protection and an increase in the available safe egress time (ASET) (Drysdale, 2011). The ASET is essentially the duration of a fire where the conditions within the compartment remain tenable for occupants. Large compartment fires also demonstrate different fire dynamics because of a potential shift from the Regime I to Regime II fire. As a result, the thermal boundary conditions within the compartment will not be as anticipated in the original compartment fire framework. Using current techniques result in a potential large over-estimation of the thermal load required for design. This a direct result of the conservative assumptions adopted by the compartment fire framework. Based on correlations shown by burnout tests the maximum temperature within a compartment can currently be assumed according to the geometry of the compartment. This maximum temperature is assumed homogenous throughout the compartment and ultimately results in the overestimation of the energy distribution within the compartment. Even further, large compartments that demonstrate Regime II fire behaviour were shown to result in a lower maximum temperature in comparison to Regime I fire behaviour and the temperature was not uniform throughout the compartment. Again, this would result in a large overestimation of the potential thermal loads within the compartment. The assumption of a uniform fire within a large compartment has also been criticised. In contrast it has been shown that an accidental fire in a large compartment actually burns over a limited area and spreads across the floor plate (Stern-Gottfried and Rein, 2012). These fires have been termed travelling fires. The travelling fire theory assumes two regions within the compartment, the near field and the far field. The near field is the set region where the fire is assumed to be and contrastingly the far field is the areas away from the fire. These regions are assumed to be constant in size but move within the compartment such that the flame spread rate is equivalent to the burnout rate. A near field temperature of 1200 C and a far field temperature of 300 C is assumed. The use of travelling fires in structural analysis results in the ability to assess an entire range of fire sizes rather than the conventional method to assume a fullydeveloped compartment fire. The travelling size of the fire is arbitrarily picked for 3

24 sensitivity analysis purposes (i.e. to assess for the worst case scenarios). However, the spread rate of the fire is calculated based off of a correlation between the size of the fire and burnout times. Coupled with the assumption of a constant fuel source for the length of the compartment this results in a constant spread rate (i.e. time to burnout is constant for any location within the compartment). Growing fires are another attempt at creating a more accurate design fire for structural design purposes. The same principles of the travelling fire theory are applied. However, the near field region grows in size and as a result the far field region reduces in size. This is essentially assuming that the flame spread rate is faster than the burnout rate. The spreading fire theory also assumes a near field temperature of 1200 C and a far field temperature of 300 C. 1.2 SCOPE OF THE RESEARCH With the use of the current compartment fire framework tools there is a potential area of improvement in the analysis for large compartment fire design. Ideally this would include a much clearer understanding of the fire behaviour seen in Regime II fire types. The current use of a homogenous maximum temperature for the structural design of a compartment is a potential large over-estimation of the actual conditions. This overestimation is potentially magnified with Regime II fire behaviour because of characteristic non-homogenous temperatures within the compartment that were observed. As a result, a major element missing in the current compartment fire framework is an energy balance within the control volume (i.e. the compartment). Conservation of energy is not considered in the currently used design fires and compartment fire framework. Conservation of energy is a foundation principle which demonstrates that energy cannot be created of destroyed, only altered from form to form (Bornakke and Sonntag, 2013). As a result, there is potential to analyse large compartment fires based off the conservation of energy within a control volume allowing for a detailed understanding of the energy distributions within the compartment. 4

25 On a whole all design fire theories are an overestimation if used appropriately. This is accepted in practice because an inherent factor of safety is essentially incorporated into the fire safety design of the building with regards to the design fire used. However, more accurate analysis could be provided with the use of the fundamental concept, conservation of energy Research Objective The primary object of this research is to create a platform for analysing large compartment fires from the results of large-scale fire tests. The large-scale fire tests are to be analysed based off a simple energy balance upholding the conservation of energy within a control volume Out of Scope To limit the level of analysis to an achievable amount of work the following list was considered out of scope: Application of variable thermal conductivity of materials Spatial analysis Radiation correction of thermocouples Identification of amount of energy supplied from each gas burner Assumptions The surface materials were considered to have constant thermal conductivity for the duration of all experiments even with the large changes of temperature. A linear velocity profile was appropriate to quantify the mass flow rate in and out of the compartment. Consequently, the neutral axis was defined as the height at which zero velocity occurred. Constant conditions as measured by all sensors for their respective tributary areas/volumes throughout the analysis. Energy supplied to the compartment through the propane was considered a complete combustion process. All sensors performed at a reasonable accuracy and any broken sensors were identified and treated appropriately. 5

26 1.3 OUTLINE OF THE CHAPTERS This thesis is organised into 12 chapters. Chapter 1 contains background information and the scope of the research. Chapter 2 is a history of the understanding of fire severity and contains example results and conclusions of large-scale fire tests. Chapter 3 highlights the context of the work within the thesis. Chapter 4 is a description of the compartment used to house the experiments. Chapter 5 is a detailed description of the experimental instrumentation used for the analysis of the experiments. Chapter 6 highlights the method of fire simulation used for the experimental analysis. Chapter 7 summarises the characteristic components of each experiment. Chapter 8 is a detailed description of the analysis methods adopted for the interpretation of the experimental results. Chapter 9 is a discussion of the results produced from the experimental analysis. Chapter 10 contains the conclusions and recommendations of the findings from the experimental analysis. The references used for the thesis is presented in Chapter 11. Lastly, the appendix containing additional results from sensitivity analysis conducted is seen in Chapter 12. 6

27 2 LITERATURE REVIEW 2.1 KEY DEFINITIONS The very important definitions of fire severity and fire-resistance ratings were defined by Ingberg early in his research (Ingberg, 1942). The fire severity term was used as a measure of the intensity and duration of a fire. It was also expressed in terms of a time exposure equivalent to a standard fire test. Originally fire severity was quantified on fuel loads only, however, new techniques were adapted with further research. These techniques are highlighted in the following review. Fire-resistance rating is defined as the length of time that a structural element performs at a satisfactory level in a standard fire test. These definitions have been carried through the modern era of fire safety engineering. 2.2 TRADITIONAL METHODS OF ESTIMATING FIRE SEVERITY Kunio Kawagoe and Takashi Sekine, pioneers in fire safety research produced a series of reports titled, Estimation of Fire Temperature-time Curve in Rooms (Kawagoe and Sekine, 1964). The studies were conducted with the intent to have application with actual buildings in Japan during the 1960 s. These reports formed an analysis of the fundamental heat balance equation imbedded with an opening factor for various experimental compartments and openings to calculate a maximum flame temperature. The heat balance equation is an energy balance within the compartment, but this was used to quantify a maximum temperature to be used for the design of the entire compartment (i.e. a conservative approach). Due to the design of the typical Japanese buildings in this era use of an opening factor was appropriate as compartments were likely to be within the characteristic under-ventilated fire scenario. Full-scale experiments were conducted and compared to the expected results calculated from the heat balance equation mentioned above. It was concluded that the predicted fire temperature rises approximately agreed with the actual temperature rises. Surveys were conducted to gather an understanding of typical fire loads depending on the use of different concrete buildings and typical key dimensions (i.e. floor area, window area and window height). Theoretically larger opening factors resulted in larger temperatures within the compartment but shorter fire durations, it was noted that larger opening factors may result in temperatures greater than 1,000 C. 7

28 Whereas, larger fuel loads affected the duration of the fire and the amount of energy within the compartment. Once enough information was acquired the buildings were classified based on an opening factor and durations were quantified for average and above average fuel loads (i.e. 50 and 100kg/m 2 respectively). In order to design for structural capacity, the Japanese Industrial Standard Curve (JIS Curve) was used to test structural elements in the same manner as the Standard Fire test. That is, based on the classification of the building and the expected fuel load a theoretical fire temperature-time curve could be calculated. In order to test the structural performance in this fire scenario structural elements were subject to the JIS Curve for an equivalent JIS testing time. This equivalent time was calculated as the time the element needs to be subject to a JIS test in order to apply the same amount of fire severity to the element. As a result, the thermal energy that the tested element is exposed to is equivalent between the burnout tests and the JIS Curve allowing for systematic studies of structural performance. This concept was originally introduced by Ingberg in order to correlate full-scale fire test results to provide an answer to the level of fire-resistance construction required to contain a fire to the enclosure (Ingberg, 1928). Harmathy in 1972 produced a two-part paper which endeavoured to present all of the available information on compartment fires at the time (Harmathy, 1972). Of particular interest were the comments on fire severity and the heat balance for a burning compartment. In terms of fire severity Harmathy suggested the validity of using a time equivalent approach to analyse structural capacity of elements is unreasonable because the temperature of the flame and duration of exposure is not equally important. Harmathy s comments were directly relatable to Kawagoe s method of determining fire endurance requirements using equivalent fire severities (Kawagoe and Sekine, 1964). Instead it was suggested that three important parameters be used as the measure of fire severity. The three parameters are listed below: The duration of the fully developed fire; The effective heat flux ; and The average gas temperature (spatial and temporal). 8

29 The fire severity can only be examined by an analysis of the fundamental heat balance. However, in contrast to the traditional analysis of the heat balance equation it was used to estimate the effective heat flux rather than the gas temperature. As both parameters are required for solving the heat balance an iterative approach is adopted by an initial guess of the gas temperature to solve for an effective heat flux then the effective heat flux is used to calculate the gas temperature to check for validity and to be used for the next iteration. This was repeated until sufficient accuracy of the calculated parameters was achieved. Harmathy conducted an analysis of experimental data based off of the three parameters influenced by different ventilation to measure fire severity. Firstly, an important note was made in relation to a very large amount of energy contained in the fuel left the compartment through the opening (ranging from 50 to 90%). The largest effective heat flux occurred at the critical value of the ventilation parameter (i.e. between under and over-ventilated conditions). It was assessed that over-ventilated fires resulted in the shortest fully-developed fire periods and was not affected by changes in ventilation or fuel load. However, in under-ventilation conditions the fullydeveloped fire duration was observed to increase rapidly with decreasing ventilation. This scenario also resulted in a decreased effective heat flux but this beneficial effect was outweighed by the increase in duration and resulted in an overall higher level of fire severity. Maximum gas temperatures were observed in under-ventilated scenarios and depended largely on the fire load and ventilation factor. With the outlined results in terms of the three highlighted parameters it was concluded that the exact opposite for the desired design was the most effective in terms of structural capacity. That is, lower fire severity levels were anticipated by the design of an over-ventilated compartment rather than under-ventilated. Harmathy and Mehaffey, in 1983, produced a paper which was a review of various post-flashover models over a 25 year period (Harmathy and Mehaffey, 1983). Throughout the paper the characterization of a fires destructive potential was addressed (fire severity). It was stated that the derivation of information that can form a basis for assessing the structural and thermal performance of the compartment boundaries is potentially the most important aspect of post-flashover compartment fire modelling. However, incorporating any fire severity into a model is typically much too 9

30 complex. As a result, models have been used mainly to provide information which is to be inputted into a separate analysis to assess structural performance. This position is historically where different directions have been taken. As mentioned previously Kawagoe (Kawagoe and Sekine, 1964) did assessment of fire severity with the temperature history of the fire gases. Whereas, Harmathy (Harmathy, 1972) suggested that fire severity parameters; the duration of the fully developed fire; the effective heat flux ; and the average gas temperatures would be more appropriate to characterize fire severity. However, even more detailed investigation by Harmathy (Harmathy, 1980) demonstrated that a single parameter, the heat load was sufficient in defining the fire severity. The heat load is defined as the time integral of the penetration heat flux factored by the inverse thermal inertia of the compartment boundaries. The major result of the heat load parameter was that it could be normalised and yield a unique descriptor for fire severity regardless of the bounding material. However, there were some limitation with respect to bounding materials such as unprotected steel and aluminium. Harmathy and Mehaffey also reviewed the standard fire resistance test (Harmathy, 1981). It was noted that with detailed analysis the normalized heat load was essentially a function of the duration of the standard fire tests if they were performed in highly efficient furnaces. However, as standard fire tests are conducted in furnaces not typically with this high level of efficiency consideration is needed for secondary factors such as furnace size, lining materials, combustion products and so on. Harmathy (Harmathy, 1981) demonstrated that the normalized heat load theorem is not strictly correct due to 5 different common construction materials being exposed to a standard fire test having 5 slightly different normalized heat load versus fire duration curves. If this theory was strictly correct then all 5 specimens would have yielded a single curve. However, it was suggested that the normalized heat load theory was an acceptable approximation. Further comparisons of the same material in different furnaces went on to demonstrate slight variations of the normalized heat load versus fire duration curves. Overall, all theories that have been derived in the past have been to understand maximum fire severities within a compartment using the fundamental heat balance. 10

31 However, this is not accurate in terms of energy conservation as these maximum conditions does not necessarily occur for the duration of the fire and in the entire compartment. 2.3 OUTCOMES OF PAST WORK Buc produced a report highlighting the small details of fire tests that need to be understood in order to validate results (Buc, 2008). Standard test methods pose potential limitations because important fire parameters may not provide enough versatility to capture effects of exposure to different environments. As a result, there is a large potential for results that look good, however, they may be misleading and could potentially promote insufficient design and increase of hazard. If standard tests are proved to be unable of modelling an appropriate scenario custom testing may be required. Large scale testing provides the most accurate test results but generate a cost issue. Most fire tests are done to a small or intermediate scale and require a large level of awareness with difficulties in extrapolating data to large scale events. In 1939 a series of fire test were conducted at the National Bureau of Standards (Rodak and Ingberg, 1967). A one-story fire resistive building was used in fire scenarios to generate data on the intensity and duration of fires for residential occupancy buildings with variable fuel loads. Thermocouples were used to measure a distribution of temperatures throughout the test building for the duration of the experiment for their use in time-temperature curves. Ingberg s traditional method of calculating fire severity was adopted for the analysis of these experiments as highlighted previously (Ingberg, 1928). It was concluded that this use of the time equivalent fire method for assessing fire severity and fire-resistance rating for a burnout test in an enclosure needed further study. In 2007 Welch et al. produced a paper titled BRE large compartment fire tests Characterising post-flashover fires for model validation (Welch et al., 2007). This paper was primarily produced to validate models and assess the accuracy of measured data from large-scale fire tests. However, there was also interest expressed in the equivalent radiative flux within the test compartment. The radiative heat flux was used to characterise the level of thermal exposure of structural components. To measure the heat fluxes special designed steel billets were installed across the compartment 11

32 and calibrated in accordance to O Connor et al. (O'Connor et al., 1997). These heat flux sensors were considered capable of characterising heat fluxes in severe thermal environments. Radiative heat fluxes were only considered as it was noted that once compartment temperatures exceeded 1,000 C the convective heat flux usually contribute to less than 10% of the total heat flux. This percentage reduced further once the surface temperature of the billet was in equilibrium with the gas-phase conditions. As a result, it was assumed that steel billet measure radiative heat fluxes only It was stated that the heating potential fires is generally analysed in terms of gas temperatures only, and often thermal exposure is assumed to be uniform, or at best split into two layers. It was concluded that strong variations in the gas temperature distribution and large instantaneous differences of radiative heat fluxes were measured and the more useful measure of thermal exposure is radiative heat flux. It was also stated that the distribution of the larger radiative heat fluxes was notably different from the location of the larger gas temperatures. As a result, the individual measure of heat fluxes was deemed as necessary and most critical in the assessment of fire severity on a compartment in large-scale fire tests. 12

33 3 RESEARCH CONTEXT In 2013 a series of large-scale experiments were conducted in Watford, England by a team of academics and technicians. The compartment that housed the experiments allowed for the systematic control of the ventilation conditions to assess the behaviour of both Regime I and Regime II fires within a large compartment. Additionally, a range of fire behaviours were simulated in a manner that was repeatable for the comparison between experimental conditions. The primary goal of the experiments that were conducted was to understand the distribution of energy throughout each experiment. To understand this a comprehensive analysis of the data acquired as a result of the experimental series was conducted. This analysis essentially quantified the distribution of energy within the compartment using a fundamental approach, conservation of energy within a control volume. In the proceeding chapters the experimental setup and the data analysis are highlighted in detail. 13

34 4 DESCRIPTION OF THE COMPARTMENT 4.1 COMPARTMENT LOCATION AND DIMENSIONS The large compartment fire experiments were conducted in the Burn Hall of the BRE facilities in Watford of Hertfordshire, England. Figure 2 below demonstrates the arrangement of the Burn Hall during the accommodation of these large compartment fire tests with the key dimensions shown. With the geometrical restraints of the Burn Hall the available area for the large experimental compartment was approximately 20m in length and 6m in width Drainage system Area for the data loggers Control room Wiring acces Drainage system Figure 2: Plan view of the large compartment setup withing the Burn Hall (Hidalgo, 2013) The compartment was constructed in the specified area on a staging system 1 metre clear of the ground allowing for easy access. There are three walls that enclosed the compartment, a ceiling and an overhang. All walls were 2m in height whilst the back wall was 18m long and the left and right walls were 5m long. The overhang extended the entire length of the compartment for a height of 0.5m. Fifteen 1.1 metre wide openings separated by a 0.1m vertical element were present along the front side of the compartment coupled with a shutter system which was to vary the ventilation conditions systematically based on the requirements of the experiments. Succeeding sections provide more detailed descriptions of the components of the compartment. Figure 3 below provides an isometric overview of the compartments dimensions and makeup. 14

35 Figure 3: Isometric view of the large compartment (south-west view) (Hidalgo, 2013) 4.2 COMPARTMENT FLOOR Compartment Floor Layout The compartment floor was built with a dual purpose such that it could host two fundamentally different experiments. The initial series of tests comprised of the floor retrofitted with multiple gas burners. The second series of tests held wood cribs, however, these were not assessed within this dissertation. The floor was designed with an internal platform to initially hold the gas burners. After the gas burner series of experiments this internal platform could be replaced with the setup for the wood crib experiments. Figure 4 illustrates the dimensions of the internal (solid blue) and external (red dot hatching) floor components. 15

36 Figure 4: Compartment Floor Layout For the gas burner series of experiments twelve gas burners were used. The burners were aligned such that they provided a heat source to equal area within the compartment. Figure 5 is representative of the gas burner arrangement and illustrates important dimensions. Figure 5: Gas Burner Arrangement Compartment Floor Materials The raised floor was supported by a steel frame and steel tubing legs with an 18mm thick Birch plywood top. The supporting system was capable of withstanding a 5 kn/m 2 load. A steel sheet 0.5mm thick was laid above the supporting system to provide a suitable area to attach hooks that held the thermocouple trees in place. This was also reproduced in the roof such that the opposite end of the thermocouple tree could be held in place. For the gas burner experiments 140mm thick ROCKWOOL FLEXI was used as the final layer except in the location of a burner where a layer of 25mm thick 16

37 ROCKWOOL FLEXI was used instead. Figure 6 shows the compartment floor material composition. Figure 6: Compartment Floor Material Composition for the Gas Burner Experiment Series 4.3 COMPARTMENT WALLS Compartment Wall Layout The compartment consisted of three walls and a large open section as shown in Figure 3. The left and right walls were 5m wide and 3m high, whilst the back wall was 18m long and 2m high. The same materials and arrangements were used for all three walls of the compartment Compartment Wall Materials The compartment walls were assembled using timber studs and steel C-sections. The following list summarises the materials used in the assembly: Standard Plasterboard (15mm) Timber studs, with available space filled by ROCKWOOL FLEXI (140mm). Steel C-sections, with available space filled by ROCKWOOL FLEXI (50mm). Aircrete (50mm), 310mm in length by 215mm high. The above list is representative of the material assemble from the outer face of the wall to the inner exposed face. Figure 7 illustrates the composition of the compartment walls. 17

38 Figure 7: Compartment Wall Material Composition Aircrete was used as the surface that was exposed to the experimental conditions such that there was a minimal amount of material replacement required. The minimisation of material replacement served two purposes; less work was required for the duration of the experiments and more protection was provided to the instruments within the compartment. The timber studs were used to provide load bearing capacity as they supported the ceiling weight transferring the load as a column structure. The steel C- section provided an insulated gap between the timber studs and the Aircrete. This was essential because heat transfer from the hot gases within the compartment to the timber studs was limited. As a result, the timber studs load bearing capacity was not significantly affected. 4.4 COMPARTMENT ROOF Compartment Roof Layout The roof of the compartment was slightly longer and wider than the supporting walls of the compartment. The length of the compartment was approximately 18.4m while the width was approximately 5.4m Compartment Roof Materials The compartment roof was assembled using the same timber stud and steel C-section system as the compartment walls. The following list summarises the materials used in the assembly: Standard Plasterboard (15mm) Timber studs, with available space filled by ROCKWOOL FLEXI (140mm). Steel C-sections, with available space filled by ROCKWOOL FLEXI (50mm). 18

39 Standard Plasterboard (15mm) Black Steel Sheet (0.5mm) ROCKWOOL BEAMCLAD (25mm). This list represents the material composition from the outer non-exposed face to the inner exposed face. Figure 8 illustrates the composition of the compartment walls. Figure 8: Compartment Roof Material Composition The black sheet metal in the roofing system was utilised for multiple uses. These uses are listed below: Minimising damage to the plasterboard; Support for the ROCKWOOL BEAMCLAD through the use of welding pins; and An attachment system for the thermocouple trees as highlighted in the design of the compartment floor. 4.5 COMPARTMENT OPENING Compartment Opening Layout The front face of the compartment had a total of 15 openings. The openings were each 1.1m wide and separated by a 0.1m vertical element. The openings were allocated a numbering system which was Opening 1-15 from the left to right when facing the front of the compartment. This arrangement is illustrated in Figure 9 (a). In order to vary the level of ventilation for the compartment both during and between different experiments a shutter system was built. The shutters were made of a steel framing system which supported a layer of ROCKWOOL FIREPRO (50mm). The shutters had steel wheels attached to the steel frames and a guide rail was made for both above and below the 19

40 opening such that the shutter could slide along the face on the compartment. The shutter system is demonstrated in Figure 9 (b) and (c). The opening also included an overhang of 0.5m leaving 1.5m of clear area for the opening. The overhang spanned the length of the compartment (18m) and can be seen in Figure 9 (a). The overhang also had the same material composition as the roof. (a) (b) (c) Figure 9: (a) compartment opening layout. (b) shutter system. (c) dimensions and composition of fire shutters. 20

41 5 DESCRIPTION OF THE INSTRUMENTATION 5.1 OVERVIEW OF INSTRUMENTATION To conduct such large scale testing in a successful and accurate manner an appropriate level of data collection was required. In order to capture the data required for extensive analysis over 2,000 sensors were used for the series of experiments. Table 1 highlights the quantities and types of test instrumentation used. Table 1: Test instrumentation resources Parameter Measured Gas-phase temperature Incident heat flux Gas flow velocity Sensor Type Thermocouple Thin skin calorimeter (TSC) Bi-Directional Velocity Probe Location in the Number of compartment sensors Interior gas phase 1,624 Openings 75 Exterior near openings 165 Interior surfaces 165 Exterior façade 94 Exterior facing openings 15 Openings 30 O2, CO2, CO Gas Sampling Interior Gas Phase 5 Obscuration Obscuration Meters Interior Gas Phase 5 Propane flow Mass Flow Controller Exterior 2 Mass loss Load Cells Exterior 8 Thermal imaging Thermal Imaging Camera Exterior 1 Video imaging Video Cameras Interior and Exterior ~6 (variable) It is important to note that the above table highlights all of the instrumentation used which included instrumentation for other sub-experiments. The following sections will highlight the instrumentation that was specifically used in the analysis of the design fire experiments. 5.2 THERMOCOUPLES Type K thermocouples (1.5mm diameter bead size) which are known as the most common general-purpose thermocouple were used in the experiments. The thermocouples were used to quantify spatial and temporal temperatures within the 21

42 compartment. In order to achieve this level of quantification an appropriate density of thermocouples was required. To provide this density within the compartment there were 7 rows of 29 thermocouple trees, each tree also consisted of 8 thermocouples in the z-direction. In total, there were 203 thermocouple trees which equated to 1,624 thermocouples inside of the compartment. In addition to the thermocouples inside of the compartment, there was another row of 15 thermocouples at each of the openings centrelines. In contrast to the thermocouple trees within the compartment there were 5 thermocouples per tree due to the reduced height at the opening. Table 2 list the locations of the thermocouples while Figure 10 and Figure 11 (a) and (b) show the locations of the thermocouples. Table 2: Thermocouple locations Location within compartment y-direction location, length (m) x-direction location, depth (m) z-direction location, height (m) Inside the 0.6, 1.2, 1.8, 2.4, 3.0, 3.6, compartment 4.2, 4.8, 5.4, 6.0, 6.6, 7.2, 0.4, 1.1, 1.8, 0.3, 0.6, 0.9, 1.2, 7.8, 8.4, 9.0, 9.6, 10.2, 2.5, 3.2, , 1.6, 1.8 and 10.8, 11.4, 12.0, 12.6, and , 13.8, 14.4, 15.0, 15.6, 16.2, 16.8, and 17.4 Compartment 0.6, 1.8, 3.0, 4.2, 5.4, 6.6, 0.18, 0.43, 0.68, opening 7.8, 9.0, 11.4, 12.6, 13.8, and , 16.2 and 17.4 Note*: All locations of thermocouples have a potential error of ± 50mm. Figure 10: Thermocouple locations (Plan view) 22

43 (a) Figure 11: (a) thermocouple location at the opening (Elevation view). (b) thermocouple location within the compartment (Elevation view). 5.3 THIN SKIN CALORIMETERS (TSCS) The instrumentation used to measure the incident heat flux was TSCs. These TSCs were custom designed and calibrated in accordance with Hidalgo et al. (Hidalgo et al., 2015). The TSCs consisted of a 10mm diameter, 0.5mm thick 304b stainless steel metallic disc. A Type KX thermocouple was welded to the centre of the back face which was not exposed to the fire. The metallic disc was lodged into an 80mm diameter, 50mm thick Ceraboard disc and the thermocouple wires protruded through to the rear of the Ceraboard. The Ceraboard disc was placed into cored holes in the compartment such that the face of the TSC was left sitting flush with the surface of the compartment which was exposed to the incident heat flux. Figure 12 illustrates the typical TSC that was used. (b) 23

44 Figure 12: Thin Skin Calorimeter (TSC) Thin Skin Calorimeters were located on every internal surface of the compartment except the overhang. The overhang was assumed to receive the same incident heat flux as the section of the roof that was closest to the overhang. There were a total of 165 TSC sensors within the compartment, Table 3 highlights their distribution across interior surfaces. Table 3: Incident heat flux sensors of interior surfaces Parameter Measured Incident heat flux Sensor Type Thin skin calorimeter (TSC) Location in the compartment Number of sensors Roof 15 Floor 15 Back wall 45 Left wall 45 Right wall 45 The TSCs were distributed in 3 rows of 15 for both the roof and the floor. The arrangement of the TSCs for these surfaces are illustrated in Figure 13. The back wall surface also had a distribution of 3 rows of 15 TSCs. However, the arrangement was different to the floor and roof due to the change of geometry. The back wall TSC layout is shown in Figure 14. Finally, the left and right walls were allocated 3 rows of 5 TSCs due to the even more restricted geometry which is illustrated in Figure 15. A summary of the TSC locations is provided in Table 4. 24

45 Figure 13: Arrangement of roof and floor TSCs Figure 14: Arrangement of back wall TSCs Figure 15: Arrangement of left and right wall TSCs 25

46 Table 4: Thin Skin Calorimeter locations Location within y-direction location, compartment length (m) Floor and Ceiling 0.6, 1.8, 3.0, 4.2, 5.4, 6.6, 7.8, 9.0, 10.2, 11.4, 12.6, 13.8, 14.4, 15.0, 16.2 and , 1.8, 3.0, 4.2, 5.4, 6.6, 7.8, 9.0, 10.2, 11.4, 12.6, Back Wall 13.8, 14.4, 15.0, 16.2 and 17.4 Left and Right 0.0 and 18.0 Wall x-direction location, depth (m) z-direction location, height (m) 1.1, 2.5 and and , 1.1 and , 1.7, 2.5, 3.3 and , 1.1 and 1.8 There were several more TSCs distributed across the compartment other than the internal surfaces that were listed previously. There were mainly additional TSCs on the exterior of the opening and the facades above the opening. However, these were not critical to the analysis. 5.4 VELOCITIES PROBES At each of the 15 openings approximate centrelines there were 2 bi-directional velocity probes (McCaffrey and Heskestab, 1976). The probes were offset 5cm from the true centreline of each opening so any obscured results caused by the opening thermocouple trees could be avoided. The first probe was located at approximately 220mm above the bottom of the opening whereas the second was located at approximately 1230mm from the bottom of the centreline of each opening. These sensors were allocated their positions such that an inflow and outflow velocity could be quantified respectively. The probes were attached to differential pressure transducers with an accuracy of ±1%. Figure 16 (a) and (b) illustrate a typical velocity probe used and the locations of the velocity probes respectively. 26

47 (a) Figure 16: (a) Image of the velocity probes used during the experiments. (b) Arrangement of velocity probes at the openings. (b) 27

48 6 FIRE SIMULATION 6.1 GAS BURNERS As previously discussed there were 6 pairs of gas burners evenly spaced and installed in the floor system of the compartment. The fuel for the gas burners was propane which was supplied to each burner by two mass flow controllers and had a maximum flow capacity of 3.2kg/min. Additionally, there was fine mass controller used to supply a maximum flow of 0.64kg/min of propane to the pilots. A square metal box with 0.5m sides and 0.35m tall encased the gas burners. Gas was supplied to the burners through the sides that faced the centre of the compartment. The metal box was filled with gravel such that the supply of propane was not localised to a portion of the burner and would spread out across the metal box as it filtered through the gravel. The pilot was also located on the side of the metal box that faced the centre of the compartment. Figure 17 (a) is illustrative of the gas burners and Figure 17 (b) demonstrates a working pilot. (a) Figure 17: (a) Gas burner dimensions and labelling. (b) Gas burner pilot. (b) To help coordinate the analysis the burners were labelled in pairs from 1-6 which corresponded to the right to left side of the compartment. Each pair was also distinguished between being labelled a or b which corresponded to burners in the rear or front row of the compartment respectively. Figure 18 is representative of the gas burner layout. 28

49 Figure 18: Gas burner layout 29

50 7 DESCRIPTION OF THE EXPERIMENTS 7.1 EXPERIMENT OVERVIEW Ten experiments were conducted which were designed to simulate specific scenarios. The experiments systematically varied both ventilation and fire behaviour. As a result, examination of the fire dynamics within the compartment for each characteristic ventilation and fire behaviour scenario was possible Fire Modes Fire behaviour within a large compartment can be very complex and largely variable. There are several possibilities of occurrence with respect to the size of the fire. The fire could be fully-developed and yet restrained to a localised portion of the entire compartments geometry. It is also possible for the fire to be fully-developed and fully engulfing the entire compartment. Furthermore, there is intuitively the possibility that the fire could be somewhere between the two aforementioned scenarios. As a result, three fire modes were considered. These fire modes were described using characteristic spread rates, VS, and the burnout rate of fuel, VBO. The three fire modes are described below: Mode 1: this mode simulated a fire that had an instantaneous spread rate such that the entire compartment was engulfed instantly. The burnout of the fire corresponded to the end of the experiment. This mode was most directly relatable to a Regime II fire (Majdalani et al., 2015). Mode 2: this mode simulated a fire spread of greater magnitude than the burnout rate. As a result, the fire simulated a growing fire as it characteristically advanced through the compartment. Mode 3: this mode simulated a fire spread of equivalent magnitude than the burnout rate. This resulted in a simulation of a fire that advanced through the compartment with a constant size. This mode was most relatable to a travelling fire (Stern-Gottfried and Rein, 2012). The above fire modes are summarised in Table 5. 30

51 Table 5: Characterisation of fire modes Fire Mode Mode 1 Mode 2 Mode 3 Characterisation VS VS > VBO VS VBO Other fire modes were capable to be simulated with the burner system provided. A fourth fire mode was considered which was characterised by a fire spread rate slower than the burnout rate. However, this mode was not utilised in the experimental series Ventilation Modes As highlighted previously a shutter system was incorporated into the design of the compartment. As a result, the level of ventilation for each experiment was easily varied such that both fundamental regimes (Regime I and II) could be achieved (Majdalani et al., 2015). Initial experiments were subject to constant ventilation conditions of both over and under-ventilated conditions. These ventilation conditions were termed Mode 1 and Mode 2 ventilation conditions respectively. To measure the level of ventilation in the compartment the inverse ventilation factor was calculated for each ventilation scenario (Thomas and Heselden, 1972). The inverse ventilation factor is calculated by the following equation: Ф = A T /A O H 1/2 (1) Where A T is the area of the compartment surfaces excluding the floor and opening (m 2 ), A O is the area of the opening (m 2 ) and H is the height of the opening (m). Regime II conditions occur with an inverse ventilation factor up to approximately 10 and Regime I conditions occur approximately above 10 (Drysdale, 2011). Overventilated conditions were achieved when all 15 openings were uncovered which produced an inverse ventilation factor of approximately 4 corresponding to Regime II conditions. Contrastingly, under-ventilated conditions were achieved when only 3 evenly spaced openings were uncovered (Openings 3, 8 and 13) which produced an inverse ventilation factor of approximately 23 corresponding to Regime I conditions. Figure 19 illustrates the different inverse ventilation factors for each ventilation scenario while Figure 20 and Figure 29 illustrate the over and under-ventilation conditions. 31

52 φ (m -1/2 ) Inverse Ventilation Factor vs Openings Number of Openings (-) Figure 19: Inverse ventilation factor relationship with the number of openings Figure 20: Over-ventilated conditions arrangement Figure 21: Under-ventilated conditions arrangement With the robust design of the shutter system both constant and variable ventilation conditions were able to be produced for each unique experiment. To explore the transitional phase between Regime I and II variable ventilation conditions were used. This also had a side-effect of window breakage simulation which is commonly assessed in industry. To conduct the experiments which explored the transitional phase between regimes the first 3 openings from the right side of the compartment were originally uncovered (Openings 13-15). As the experiment commenced the level of ventilation was increased by rolling all shutters simultaneously to the left, the shutter 32

53 in position 12 moved to position 11, the shutter in position 11 moved to position 10 and so on until the shutter at position 1 was removed from the compartment. This process was repeated until the experiment concluded and the variable ventilation mode was named Mode 3. Figure 22 and Figure 23 illustrates the Mode 3 ventilation conditions while Table 6 summarises the ventilation modes. Figure 22: Starting position of ventilation mode 3 Figure 23: Systematic removal of shutters for ventilation mode 3 Table 6: Characterisation of ventilation modes Ventilation Mode Openings Uncovered Inverse Ventilation Factor (m -1/2 ) Mode 1 All 4.1 Mode 2 3, 8 and Mode initially and potentially Shutter Rate Modes As mentioned in the previous section variable ventilation conditions were achieved throughout four of the experiments. To induce these variable ventilation conditions there were two shutter rate modes considered. The first mode consisted of shutter removal every minute, whereas the second mode required a shutter removal every 5 minutes. Table 7 summarises the two shutter rate modes used. 33

54 Table 7: Characterisation of shutter rate modes Shutter Rate Mode Mode 1 Mode 2 Removal Rate 1 per minute 1 per 5 minutes 7.2 EXPERIMENT Fire Mode The fire that was simulated in Experiment 1 was a Mode 1 fire. As highlighted in Table 5 a Mode 1 fire represented a fully-developed fire with an instantaneous spread rate. To produce this instantaneous spread rate all burners were lit simultaneously. Due to functionality issues with the dispersion of propane between burners simultaneous lighting of the gas burners did not systematically occur. However, once all burners were lit the experiment was classified as started. Functionality issues also occurred with the pilot system used to promote the ignition of gas burners. To replicate the effect of the pilot small pots that contained approximately 30ml of Heptane were placed in the corner of each gas burner and ignited. Figure 24 illustrates the actual heat release rate (HRR) recorded and the target HRR for Experiment 1. Figure 24: Experiment 1 HRR and target HRR 34

55 7.2.2 Ventilation Mode The level of ventilation for Experiment 1 was equivalent to Mode 1 ventilation. Mode 1 ventilation was summarised in Table 6 and represented an over-ventilated compartment with all openings uncovered. With all openings uncovered the experiment was characterised with an inverse ventilation factor of approximately 4.1 which fell within the bounds of a Regime II fire (Thomas and Heselden, 1972). Figure 20 illustrates the ventilation conditions achieved Summary of Events A summer of the noted events that occurred during Experiment 1 are highlighted in Table 8. Table 8: Noted events during Experiment 1 Event Local Time Relative to Time Zero (s) Start Lighting Pilots 12:04: All Pilots Lit 12:08: Open Propane Supply to Burners 12:14: Start of HRR Plot (CREF) 12:14: Only Burners 2B & 3B Not Lit 12:16:00-70 All Burners Lit 12:17: % of Max HRR 12:17:54 44 Begin HRR Change 12:22: % of Max HRR 12:23: Begin HRR Change 12:27: % of Max HRR 12:29: Begin HRR Change 12:34: % of Max HRR 12:35: Begin HRR Change 12:39: % of Max HRR 12:39: Close Propane Supply to Burners 12:42:

56 7.3 EXPERIMENT Fire Mode The fire that was simulated in Experiment 2 was a Mode 2 fire. As highlighted in Table 5 a Mode 2 fire resembled a fire with a spread rate faster than a burnout rate. As a result, a growing fire was simulated. This was achieved in the experimental conditions by adding pairs of burners every 2.5 minutes. The burner ignition started with burner pair 1 (far right of the compartment) and burner pairs 2-6 were sequentially added. Some issues with the with the propane supply affected the ability to reach all target HRR during the experiment. The actual HRR recorded and the target HRR can be seen in Figure 25 Figure 25: Experiment 2 HRR and target HRR Ventilation Mode The ventilation conditions of Experiment 1 were maintained for Experiment 2. As a result, over-ventilation conditions were achieved through the Mode 1 ventilation mode. 36

57 7.3.3 Summary of Events A summary of the noted events for Experiment 2 is shown in Table 9. Table 9: Noted events during Experiment 2 Event Local Time Relative to Time Zero (s) Start Lighting Pilots 14:45: All Pilots On 14:52: Burners 1a and 1b On 14:55: Burner 1b Out 14:56: Burner 1a Shut Down 14:57: Burners 1a and 1b On 15:00:29 0 HRR = 0.35 MW 15:00:49 20 Begin HRR Increase 15:03: Burners 2a and 2b On 15:03: HRR = 0.82 MW 15:03: Begin HRR Increase 15:05: Burners 3a On 15:05: Burners 3b On 15:05: HRR = 1.25 MW 15:06: Begin HRR Increase 15:08: Burners 4a On 15:08: Burners 4b On 15:08: Burner 1b Out 15:09: HRR = 1.5 MW 15:09: Burner 1b On 15:09: Burner 1b Out 15:09: Burner 1b On 15:10: Begin HRR Increase 15:10: Burners 5a and 5b On 15:10: HRR = 1.7 MW 15:11: Begin HRR Increase 15:12: Burners 6b On 15:13: HRR = 1.65 MW 15:13: Burners 6a On 15:13: MW < HRR < 1.8 MW 15:14: Propane Supply Off 15:16:

58 7.4 EXPERIMENT Fire Mode The fire that was simulated in Experiment 3 was a Mode 3 fire. As highlighted in Table 5 a Mode 3 fire simulated a fire with a spread rate approximately equivalent to the burnout rate which represented a travelling fire. This was achieved in the experimental conditions by adding and extinguishing pairs of burners every 2.5 minutes. The burner ignition started with burner pair 1 (far right of the compartment). Before burner pair 2 was ignited pair 1 was extinguished by cutting the supply of propane. This pattern repeated until each burner pair was both ignited and extinguished. The HRR of the gas burners coupled with the target HRR are shown in Figure 26 Figure 26: Experiment 3 HRR and target HRR Ventilation Mode The ventilation conditions of Experiment 1 and 2 were again maintained for Experiment 3. Furthermore, over-ventilation conditions were achieved through the Mode 1 ventilation mode. 38

59 7.4.3 Summary of Events Table 10 indicates all of the noted events during Experiment 3. Table 10: Noted events during Experiment 3 Event Local Time Relative to Time Zero (s) Start Lighting Pilots 14:10: All Pilots On 14:20: Burners 1a and 1b On 14:30:52 0 Burners 1a and 1b Off 14:33: Burners 2a and 2b On 14:33: Burners 2a and 2b Off 14:35: Burners 3a and 3b On 14:35: Burners 3a and 3b Off 14:38: Burners 4a and 4b On 14:38: Burners 4a and 4b Off 14:40: Burners 5a and 5b On 14:40: Burners 5a and 5b Off 14:43: Burners 6a and 6b On 14:43: Burners 6a and 6b Off 14:45: EXPERIMENT Fire Mode The fire that was simulated in Experiment 4 was a Mode 1 fire. A Mode 1 fire represented a fully-developed fire with an instantaneous spread rate. The burners were lit as simultaneous as possible with the start of the experiment characterised as the time to which all burners were ignited. The fire mode was the same as Experiment 1 but a different HRR was achieved. Experiment 4 reached the target HRR immediately in contrast to Experiment 1 which reached the target HRR in stages. Similar to Experiments 1 and 2 faults with the propane supply the system did not allow for the desired constant HRR to be produced. Figure 27 is illustrative of the actual HRR recorded and the target HRR for Experiment 4. 39

60 Figure 27: Experiment 4 HRR and target HRR Ventilation Mode The ventilation conditions of the first 3 experiments were altered for the next 3 experiments. As a result, Experiment 4 had constant under-ventilation conditions simulated through the Mode 2 ventilation mode as described in Table 6. The underventilated conditions resulted in an inverse ventilation factor of approximately 23.3 which fell within the bounds of a Regime I fire (Thomas and Heselden, 1972). See Figure 21 for an illustration of the ventilation conditions achieved. 40

61 7.5.3 Summary of Events Table 11 indicates noted events during Experiment 4. Table 11: Noted events during Experiment 4 Event Local Time Relative to Time Zero (s) Begin Ignition of Pilots 09:41: All Pilots On 09:55: Open Propane Supply to Burners 10:04:55-65 All Burners On 10:06:00 0 Close Propane Supply to Burners 10:22: All Burners Off 10:22: EXPERIMENT Fire Mode The fire that was simulated in Experiment 5 was a Mode 2 fire. A Mode 2 fire was a fire with a spread rate faster than a burnout rate simulating a growing fire (see Table 5). This was the same fire simulated in Experiment 2 and the same experimental method was adopted (see section 7.3.1). It should be noted that the recording of the gas supply to the compartment did not start until approximately 300 seconds into the experiment. This is highlighted in Figure 28. Figure 28: Experiment 5 HRR and target HRR 41

62 7.6.2 Ventilation Mode The ventilation conditions of Experiment 4 were repeated for Experiment 5. As a result, under-ventilation conditions were simulated with the Mode 2 ventilation mode Summary of Events The noted events of Experiment 5 are highlighted in Table 12. Table 12: Noted events during Experiment 5 Event Local Time Relative to Time Zero (s) Begin Ignition of Pilots 14:52: All Pilots On 15:02: Burner 1A On 15:09:58 0 Burner 1B On 15:10:01 3 Burner 2A On 15:12: Burner 2B On 15:13: Burner 3A On 15:15: Burner 3B On 15:15: Burner 4A On 15:17: Burner 4B On 15:18: Burner 5A On 15:20: Burner 5B On 15:20: Burner 6A On 15:22: Burner 6B On 15:23: Propane Supply Cut 15:27: EXPERIMENT Fire Mode Experiment 6 was conducted under the same fire mode as Experiment 3, Mode 3 (see 7.4.1). This fire mode was representative of a travelling fire where the spread of the fire was approximately equivalent to the burnout rate. Again, the same experimental method used in Experiment 3 was replicated for Experiment 6. The HRR and the target HRR for Experiment 6 is shown in Figure

63 Figure 29: Experiment 6 HRR and target HRR Ventilation Mode The last experiment of the series 4-6 was conducted under the same ventilation conditions as the first two experiments (4 and 5). Hence, the Mode 2 ventilation mode was used to simulate under-ventilation conditions Summary of Events Table 13 shows all of the noted events during Experiment 6. Table 13: Noted events during Experiment 6 Event Local Time Relative to Time Zero (s) Start Lighting Pilots 16:57: All Pilots On 17:07: Burners 1a and 1b On 17:21:44 0 Burners 1a and 1b Off 17:24: Burners 2a and 2b On 17:24: Burners 2a and 2b Off 17:26: Burners 3a and 3b On 17:26: Burners 3a and 3b Off 17:29:

64 Event Local Time Relative to Time Zero (s) Burners 4a and 4b On 17:29: Burners 4a and 4b Off 17:31: Burners 5a and 5b On 17:31: Burners 5a and 5b Off 17:34: Burners 6a and 6b On 17:34: Burners 6a and 6b Off 17:36: EXPERIMENT Fire Mode The fire mode for the next two experiments (i.e. Experiments 7 and 8) was the same. A Mode 1 fire was simulated corresponding to the fully-developed fire with the instantaneous spread rate. The experimental conditions used to simulate this fire was the same as Experiments 1 and 4 (i.e. all burners lit at once). The HRR achieved in Experiment 7 is highlighted in Figure 30. Figure 30: Experiment 7 HRR and target HRR 44

65 7.8.2 Ventilation Mode In contrast to the first 6 experiments in the series, Experiment 7 was conducted with variable ventilation conditions (i.e. the Mode 3 ventilation mode). As defined in Table 6 the ventilation conditions start with Openings uncovered. As the experiment progressed a shutter was removed at shutter rate Mode 1 (i.e. 1 per minute as defined in Table 7) until all shutters were removed. Thus, the ventilation conditions started as under-ventilated and throughout the duration of the experiment transitioned to overventilated conditions. It is important to note that shutter removal times at the start of the experiment were larger due to the larger number of shutters to be moved Summary of Events Table 14 indicates noted events during Experiment 7. Table 14: Noted events during Experiment 7. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter. Event Local Time Relative to Time Zero (s) Start Lighting Pilots 16:47: All Pilots Lit 16:57: Start Lighting Burners 17:03:53-37 All Burners Lit 17:04:30 0 Open Shutter 12 17:05:30 60 [+30s] Burner 2a Out 17:06: Open Shutter 11 17:06: [+36s] Open Shutter 10 17:07: [+14s] Burner 1b Out 17:07: Open Shutter 9 17:08: [+15s] Burner 6a Out 17:09: Open Shutter 8 17:09: [+12s] Open Shutter 7 17:10: [+12s] Open Shutter 6 17:11: [+10s] Open Shutter 5 17:12: [+7s] Open Shutter 4 17:13: [+10s] Open Shutter 3 17:14: [+10s] Open Shutter 2 17:15: [+10s] Open Shutter 1 17:16: [+10s] Burner 6b Out 17:17: Cut Flow to Burners 17:23:

66 7.9 EXPERIMENT Fire Mode As noted in section a Mode 1 fire was simulated in Experiment 8 corresponding to the fully-developed fire with the instantaneous spread rate. Experimental conditions were repeated from all previous experiments that simulated a Mode 1 fire with slightly different target HRR s. Due to complications with the gas burner system initial target HRR s were not achieved. The actual HRR achieved for the duration of Experiment 8 is illustrated in Figure 31. Figure 31: Experiment 8 HRR and target HRR Ventilation Mode Experiment 8 was also conducted with variable ventilation conditions (i.e. the Mode 3 ventilation mode). Again, as defined in Table 6 the ventilation conditions start with Openings uncovered. Throughout the experiment a shutter was removed at shutter rate Mode 2 (i.e. 5 per minute as defined in Table 7) until all shutters were removed. As with Experiment 7, the ventilation conditions started as under-ventilated and throughout the duration of the experiment transitioned to over-ventilated conditions. Again, it is important to note that shutter removal times were increased at the beginning of the experiment because of more shutters to be moved. Additionally, warping of the guide rails from previous experiments also increased to removal times. 46

67 7.9.3 Summary of Events Table 15 represents the noted events of Experiment 8. Table 15: Noted events during Experiment 8. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter. Event Local Time Relative to Time Zero (s) Start Lighting Pilots 10:50: All Pilots Lit 11:04: Start Lighting Burners 11:08:00-27 All Burners Lit 11:08:27 0 Burner 2a Out 11:10: Burner 1b Out 11:11: Open Shutter 12 11:13: [+40s] Burner 3a Out 11:14: Burner 4a Out 11:14: Burner 6a Out 11:14: Burner 6b Out 11:15: HRR Drops to 1.25MW 11:15: [+22s] Open Shutter 11 11:18: [+23s] Open Shutter 10 11:23: [+37s] Open Shutter 9 11:28: [+60s] Open Shutter 8 11:33: [+93s] Open Shutter 7 11:38: [+28s] Open Shutter 6 11:43: [+13s] Open Shutter 5 11:48: [+13s] Open Shutter 4 11:53: [+13s] Open Shutter 3 11:58: [+10s] Open Shutter 2 12:03: [+16s] Open Shutter 1 12:08: [+8s] Propane Supply Cut 12:14: EXPERIMENT Fire Mode In contrast to Experiments 7 and 8 a Mode 3 fire was simulated for Experiments 9 and 10, i.e. a travelling fire. Experimental conditions were attempted to be repeated from all previous experiments that simulated a Mode 3 fire. However, issues with a gas 47

68 burner extinguishing and propane flows moving from buoyant to jet-like for the remaining burner risked flame impingement on the ceiling and ultimately damage to the material composition of the ceiling. If this scenario occurred the HRR of the experiment was reduced to half of when a full pair of burners was ignited (i.e. 0.5MW) to avoid damage to the ceiling. The HRR and target HRR are both shown in Figure 32. Figure 32: Experiment 9 HRR and target HRR Ventilation Mode Experiment 9 was again conducted with Mode 3 ventilation conditions. This ventilation mode is defined in Table 6. Throughout the experiment a shutter was removed at shutter rate Mode 1 (i.e. 1 per minute as defined in Table 7). Complications with the shutters guide rail system did not allow for the planned removal of shutters (refer to Table 16 for actual removal rates). Ventilation conditions started as under-ventilated and throughout the experiment the level of ventilation was increased. 48

69 Summary of Events Table 16 indicates the noted events for the duration of Experiment 9. Table 16: Noted events during Experiment 9. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter. Event Local Time Relative to Time Zero (s) Start Lighting Pilots 11:12: All Pilots Lit 11:23: Burners 1a and 1b Lit 11:27:45 0 Burner 1b Out 11:28:45 60 Burners 2a and 2b Lit, Burner 1a Out 11:40: Shutter 12 Half Opened 11:41: [+15s] Burner 2a Out 11:42: Burners 3a and 3b Lit, Burner 2b Out 11:43: Shutter 12 Fully Opened 11:44: [+10s] Burner 3a Off 11:46: Open Shutter 11 11:46: [+32s] Open Shutter 10 11:46: [+47s] Open Shutter 9 11:47: [+22s] Burners 4a and 4b Lit, Burner 3b Out 11:48: Open Shutter 8 11:52: [+30s] Burner 4b Off 11:54: Burners 5a and 5b Lit, Burner 4a Out 12:00: Open Shutter 7 12:08: [+30s] Burner 5b Out 12:10: Burner 6a On, Burner 5a Out 12:13: Burner 6a Off 12:14:

70 7.11 EXPERIMENT Fire Mode As previously mentioned in section Experiment 10 was conducted using a travelling fire simulation (i.e. Mode 3 fire). Experimental conditions were again attempted to be repeated from all previous experiments that simulated a Mode 3 fire. However, the same fire simulation issues from Experiment 9 were again present in Experiment 10 in terms of potential damage to the ceiling (refer to section ). See Figure 33 for the actual HRR recorded and the target HRR for Experiment 10. Figure 33: Experiment 10 HRR and target HRR Ventilation Mode Experiment 10 was the final experiment in the variable ventilation series. The Mode 3 ventilation conditions that were simulated is characterised in Table 6. Throughout Experiment 10 a shutter was removed at shutter rate Mode 2 (i.e. 5 per minute as defined in Table 7). Due to the short duration of the experiment, a limited number of shutters were removed from the openings. Ventilation conditions started as underventilated and throughout the experiment the level of ventilation was increased. 50

71 Summary of Events Table 17 illustrates the noted events for Experiment 10. Table 17: Noted events during Experiment 10. Note*: The numbers shown in brackets in the last column was the time taken for the removal of the shutter. Event Local Time Relative to Time Zero (s) Start Lighting Pilots 09:41: All Pilots Lit 09:49: Burners 1a and 1b Lit 09:52:28 0 Burner 1b Out [HRR to 0.5MW] 09:54: Burners 2a and 2b Lit, 1a Out [HRR to 1.0MW] 09:55: Open Shutter 12 09:57: [+20s] Burners 3a and 3b Lit, 2a and 2b Out 09:57: Burners 4a and 4b Lit, 3a and 3b Out 10:00: Open Shutter 11 10:02: [+20s] Burners 5a and 5b Lit, 4a and 4b Out 10:02: Burner 5a Out [HRR to 0.5MW] 10:03: Burners 6a and 6b Lit, 5b Out [HRR to 1.0MW] 10:04: Burner 6a Out [HRR to 0.5MW] 10:05: Propane Supply Cut 10:07:

72 7.12 EXPERIMENT SUMMARY Table 18: Summary of experiments Experiment No. Ventilation Mode Shutter Mode Inverse Opening Factor (Ф) Fire Mode Spread Rate 1 Mode Mode 1-2 Mode Mode 2 2 burners added every 2.5 minutes 3 Mode Mode 3 2 burners moving every 2.5 minutes 4 Mode Mode 1-5 Mode Mode 2 2 burners added every 2.5 minutes 6 Mode Mode 3 2 burners moving every 2.5 minutes 7 Mode 3 Mode Mode 1-8 Mode 3 Mode Mode 1-9 Mode 3 Mode Mode 3 2 burners moving every 2.5 minutes 10 Mode 3 Mode Mode 3 2 burners moving every 2.5 minutes 52

73 8 EXPERIMENTAL DATA ANALYSIS 8.1 MATLAB MATLAB was the computational program that was utilised for the majority of the analysis. It is a user friendly platform that has been optimized for solving engineering and scientific problems (MathWorks, 2016). The language used by MATLAB is matrix-based that allowed for an easy interaction between Excel which was used to store the raw data. This program had the capability to handle large amounts of data and had good visualisation tools which was ideal for the analysis of the experiments. 8.2 RAW DATA TREATMENT The data collected from the experiments was collated into Excel documents. In order to treat any gross errors in the recordings Excel s inbuilt max and min functions were used. These functions indicate either the maximum or the minimum values in a selected range of data. Once the maximum and minimum values of the raw data were calculated, they were assessed for gross errors. If the value calculated appeared to be an outlier, the data recorded by the specific instrument was assessed. Any instrument that was recording faulty data was fixed by using a weighted average from the next two closest non-faulty test instruments from the same instrument type. This process was repeated until all of the data was within an acceptable range. Once the data was treated, it was transferred to MATLAB. MATLAB and Excel are quite compatible and allowed for an easy transition from the Excel raw data to MATLAB raw data. Once the data was stored in MATLAB the raw data was saved such that it could be readily accessed for analysis Velocity Calculation The final velocity equation used to convert the measurements recorded by the velocity probes is shown below. v = γ 2 P ρ air (2) where v is velocity (m.s -1 ), γ is the McCaffrey probe constant (0.94), P is the differential pressure (Pa) and ρ air is the density of air (kg.m -3 ). 53

74 The raw voltage data collected by the velocity probes was treated and converted into velocity. Initially the voltage was treated with MATLAB s inbuilt smoothing function, this included the use of a moving average filter. The baseline voltage was determined for each velocity probe using an average of the first 50 data recordings (before the experiment started). The calculated baseline was removed from all of the velocity probes so the voltage difference from the baseline could be used to calculate a change in pressure. The change in pressure calculation was as follows: P = Voltage Difference 10 (3) where the constant, -10, was the pressure transducer conversion factor (Pa.V -1 ). To convert the pressure difference to a velocity the density of air surrounding the probe was required. ρ air = M airp atm R(T + T K ) where M air is the molar mass of air (kg.mol -1 ), P atm is atmospheric pressure (Pa), R is the Ideal gas constant (J.mol -1.K -1 ), T is the temperature measured in degrees Celsius ( C) and T K is a conversion factor from degrees to kelvin (K). (4) The following is a list of assumed values used in the calculation of ρ air. M air = 29 kg.mol -1 P atm = 101,325 Pa R = J.mol -1.K -1 T K = K To calculate the density of air, ρ air, a temperature recording was required at the velocity probes. The thermocouple trees at the opening were placed such that there was a thermocouple within 50mm of the velocity probe. The recordings from these thermocouples were used in the calculation of the density of air. Finally, the calculated change in pressure and density of air at each probe was substituted into Eq. (2) to calculate the velocities of the gasses leaving and entering the compartment. 54

75 It was found that some of the velocity probes were recording faulty data. These velocity probes were: Opening 3 lower probe Opening 4 upper probe Opening 5 upper and lower probe Opening 9 upper probe. Note*: as the neutral plane was always recorded between the location of the velocity probes the upper probe typically recorded the flow out of the compartment whilst the lower probe typically recorded the flow into the compartment. To treat the faulty velocity sensors an average was taken between the next two closest sensors. For example, the Opening 9 velocity probe was averaged with the Opening 8 and 10 velocity probes. A special case was when there was an under-ventilated fire scenario simulated. A faulty probe affected Opening 3 and there were no relatively close sensors to use for an average. To account for the error the temperature contours were assessed (see Figure 34-Figure 36). Figure 34: Opening 3 temperature contour 55

76 Figure 35: Opening 8 temperature contour Figure 36: Opening 13 temperature contour It can be seen that the temperature contour for Opening 3 and 8 were similar. Therefore, it was assumed that the velocity recordings would also be similar. As a result, the faulty probe was replaced with recordings from the non-faulty probe that most closely matched the temperature contours shown at Opening 3. Another unique scenario occurred when the experiments that simulated window breakage were conducted. When the shutter was removed from the opening with a faulty sensor, the recordings were replaced with next closest sensor that was not blocked by a shutter. Once the shutters continued to open, a weighted average was 56

77 used when the next closest non-faulty sensor became available. As a result, the method of treatment for the faulty sensors became the same method used in the overventilated scenarios Incident Heat Flux Calculation The incident heat flux was calculated for all surfaces and it was based off recordings produced by the TSC s. The fundamental equation used to calculate the incident heat flux is shown below: q inc = 1 α disc (1 C) [m disc c S P ( T disc ) + ε disc t disc σt 4 disc + h c_disc (T disc T g )] (5) where q inc is the incident heat flux (W.m -2 ), α disc is the absorptivity of the TSC disc (-), C is a TSC conduction correction factor (-), m disc is the mass of the TSC disc (kg), S disc is the surface area of the TSC disc (m -2 ), c P is the specific heat capacity of the disc (J.kg -1.K -1 ), T disc is the temperature of the TSC disc (K), t is the differential change in time (s), ε disc is the emissivity of the TSC disc (-), σ is Stefan-Boltzmann s constant (W.m -2.K -4 ), h c_disc is the convective heat transfer coefficient of the TSC disc (W.m -2 ), T g is the temperature of the gasses surrounding the TSC (K). The following is a list of assumed values used in the calculation of q inc : α disc = 1.0 m disc = kg S disc = m 2 c P = T disc T disc T disc 3 J. kg 1. K 1 t = 1 s ε disc = 0.4 σ = 5.67 x 10-8 W.m -2.K -4 h c_disc 16 W.m -2 The convective heat transfer coefficient, h c_disc, was calculated using a correlation between Nusselt s and Reynolds number (Incropera et al., 2011). The following equations are representative of the method used to calculate the convective heat transfer coefficient: 57

78 h c = N uk L where N u is the Nusselt Number (-), k is the thermal conductivity of the gases (W.m - 1.K -1 ) and L is the characteristic length (m). (6) N u = 0.453R e 1 2P r 1 3 (7) R e = vl ν where R e is Reynolds Number (-) and P r is Prandtl s Number (-). (8) R e = vl ν where v is the velocity of the gases (m.s -1 ) and ν kinematic viscosity of the gases (m 2.s -1 ). (9) This correlation was used as the maximum velocity within the compartment which was assumed as 3m/s. This was based on results from the experiments which resulted in a R e that corresponded to a laminar flow. A limiting condition was that P r 0.6 which was always the case for the duration of the experiment. P r was predetermined based off the recorded data from the thermocouples as was R e. The characteristic length of the TSC was assumed to be 80mm which was the diameter of the TSC component. The TSC conduction correction factor, C, was calculated using an empirical formula. C was defined as a function of the TSC disc temperature and illustrated in Figure 37 (Hidalgo et al., 2015). 58

79 C / Correction factor C = T R² = Temperature / C Figure 37: Conduction correction factor, C, as a function of temperature As seen in the figure above the correction factor was calculated as the following equation: C = (( )T disc ) (10) where T disc was measured using the recordings from the TSC disc (K) and the temperature of the surrounding gasses, T g, was calculated as the recording from the closest thermocouple (K). 8.3 ENERGY CONSERVATION IN A CONTROL VOLUME To quantify the energy distribution within the compartment conservation of energy within a control volume was considered. For each experiment the control volume was considered as the compartment. To assess the energy conservation, the following formula was adopted: Q in,propane Q loss,surfaces Q loss,opening = Q gas (11) This equation is described in terms of a rate of energy transfer (i.e. heat). The first term, Q in,propane, is the heat supplied to the compartment through the propane burners. The second term, Q loss,surfaces, is the heat losses into the surfaces. The third term, Q loss,opening, is the heat losses out of the opening. While the last term, Q gas, is the heat stored within the gas phase. The quantification of these terms are detailed in the following analysis. 59

80 8.4 HEAT SUPPLY TO THE COMPARTMENT To calculate the amount of heat that was supplied to the compartment during each experiment the amount of heat produced by the propane gas burners was quantified. This was calculated using the energy release rate equation (Karlsson and Quintiere, 1999). Q = m H c (12) where m is the mass flow rate of the fuel (kg.s -1 ) and H c is the heat of combustion of the fuel (J.kg -1 ). The mass flow rate of the fuel was quantified based on a recorded portion of the maximum flow rate of propane (3.2kg/min) through the burners. The combustion process of the propane was assumed to be 100% efficient. The following equation demonstrates the calculation of the mass flow rate of propane: m = % recorded (13) Finally, to calculate amount of energy entering the compartment the heat of combustion for propane was required. The heat of combustion is the amount of energy produced as heat if the combustion process is 100% complete. A heat of combustion of 46.45x10 6 J.kg -1 was assumed for the purpose of the analysis (Drysdale, 2011). The following equation represents the amount of heat produced by the propane burners: Q in,propane = m H c,propane (14) As a result of the above calculations, the amount of heat supplied to the compartment for the duration of the experiments was described. It should be noted that there was an additional source of heat into the compartment through the inflow of the surrounding ambient environment. However, this source was considered negligible because the inflow was at a low temperature. 60

81 8.5 NET HEAT FLUX CALCULATION The net heat flux at the exposed surface of the solid was derived from first principles. The net heat flux was taken as a heat transfer between the gasses within the compartment and the exposed solid. The following equation is representative of the fundamental heat transfer equation for a solid: 4 q net = α solid q inc h c_solid_exposed (T g T s ) ε solid σt s (15) where q net is the net heat flux by the surface (W.m -2 ), α solid is the absorptivity of the exposed solid (-), q inc is the incident heat flux, h c_solid_exposed is the convective heat transfer coefficient of the exposed solid (W.m -2 ), T g is the temperature of the surrounding gasses (K), T s is the surface temperature (K), ε solid is the emissivity of the exposed solid (-), σ is the Stefan Boltzmann constant (W.m -2.K -4 ). The heat transfer equation is a combination of conduction, convection and radiation components. The first term of the net heat flux equation, α solid q inc, represents the absorbed radiant heat flux. The second term, h c_solid_exposed (T g T s ), is representative of the convection component of the heat transfer. Whilst the third term, ε solid σt s 4, is the radiation emission component back into the compartment. The following is a list of assumed values used in the calculation of q net: α solid = 1.0 h c_solid_exposed 6 13 W.m -2 ε solid = 0.9 σ = 5.67 x 10-8 W.m -2.K -4 The same method of calculation as presented in section for h c_solid_exposed was used (see Eq. 6-9) except with a variance in the calculation of the characteristic length. The characteristic length was calculated as the tributary area divided by the tributary perimeter (Incropera et al., 2011). In order to solve the above equation using the data in its present state a numerical method was adopted. 61

82 For time-step i: For time-step i + 1: q net i = α solid q inc i h i c_solid_exposed (T g T i i 1 ) ε solid σt 4 1 (16) q net i+1 = α solid q inc i+1 h i+1 c_solid_exposed (T g T i+1 i+1 1 ) ε solid σt 4 1 (17) The value q inc was calculated using Eq. (5) as highlighted in section while T g was the gas temperature recorded for the area subject to the net heat flux for the appropriate time-step. For the first time-step, iteration T 1 i was assumed ambient temperature (15 C). However, as the heat transferred through the solid the temperature was affected by the properties of the materials that made up the surface. This was important to analyse as the temperature profile within the surface had a significant effect on the net heat flux of the surfaces. As a result, to calculate the temperature of the next time-step a full understanding of the entire temperature profile was required. The following series of equations demonstrates how the temperature profile was calculated using the finite difference method. The following differential equation (Fourier s law) describes the net heat flux in terms of the energy conducted. q net = k dt dx (18) x=0 + To solve the above differential equation a numerical method, the finite difference method, was established and solved in MATLAB. The following equation illustrates the portion of the numerical equation that solved the net heat flux of the first element that is exposed to the fire: = ( k 1 i + k 2 i q net 2 ) ( T 1 i i T 2 x ) + (ρc i P) x 1 2 (T 1 i+1 i T 1 ) (19) t where k 1 i is the thermal conductivity of element 1 at time i (W.m -1.K -1 ), k 2 i is the thermal conductivity of element 2 at time i (W.m -1.K -1 ), T 1 i is the temperature of element 1 at time I (K), T 2 i is the temperature of element 2 at time i (K), x is the element increment i length (m), ρ 1 is the density of element 1 at time i (kg.m -3 i ), c p,1 is the specific heat of element 1 at time i (J.kg -1.K -1 ), x is the element increment length (m), T 1 i+1 is the temperature of element 1 at time i+1 (K) and t is the time-step (s). 62

83 Figure 38 illustrates how the material was broken up into smaller increments to solve the numerical equation. Figure 38: Element increments for temperature profile calculation It is important to note that for the first time-step the temperature of all elements of the material structure was assumed ambient temperature. Also, the first increment was x 2 to increase the accuracy of the heat transfer calculation of the first element. The first term of Eq. 19 represents the conduction whilst the second term introduces storage of energy for the increment of material because the material has a thickness due to the use of the numerical method. The combination of x and t was carefully selected such that the Fourier Number, F o, was less than 0.5. For a one-dimensional analysis, F o must be restricted to less than 0.5 such that the solution converges to steady-state conditions (Incropera et al., 2011). If the conditions of Fourier s Number are not satisfied the adopted numerical solution may introduce undesirable results as it is not unconditionally stable. Oscillations may be introduced that can become unstable and result in a divergence from steady-state conditions. The Fourier Number was represented by the following equation: F o = τ t ( x) 2 (20) 63

84 where τ is the thermal diffusivity (m 2.s), t is the time-step (s) and x (m) is the incremental lengths. Note*: the thermal diffusivity term from herein will be used as τ because the typical nomenclature for this term has already been used. Table 19 highlights the Fourier Number calculated for all materials used in the experiment (see Table 20 for a full list of the material properties). Table 19: Fourier Number results for appropriate materials Material τ (m 2.s) x10 7 t (s) x (m) F o (-) Aircrete Rockwool Flexi (50mm) Rockwool Flexi (140mm) Rockwool Flexi (220mm) Rockwool Beam Clad Plywood i+1 i+1 In order to solve the numerical method T 1 needed to be calculated. Once T 1 is i calculated this becomes T 1 for the next time-step. As a result, Eq. 19 was rearranged i+1 and solved for T 1. First element: ( k 1 i + k 2 i q net 2 ) ( T 1 i i T 2 x ) = (ρc i P) x 1 2 (T 1 i+1 i T 1 ) (19) t T 1 i+1 = T 1 i + 2 t (ρc P ) 1 i x [q net ( k 1 i + k 2 i 2 ) ( T 1 i i T 2 )] (21) x 64

85 Eq. 21 shows that T 1 i+1 is dependent of the temperature of the next element, in this case T 2 i. In fact, each element is affected by the next. As a result, in order to calculate i+1 T 1 which is essential for the calculation of q net the temperature profile of the entire material structure needed to be calculated. Eq. 22 and 23 are appropriate for the interior elements. Interior elements: ( k i j k j i ) ( T i j 1 x T j i ) ( k j i i + k j+1 2 ) ( Tj i T i j+1 x ) = (ρc P ) i j x ( T j i+1 i T j ) (22) t i i T i+1 j = T i t j + (ρc P ) i j x [(k j 1 + k j i ) (T 2 2 j 1 T i j ) ( k j i i + k j+1 2 ) (T i i j T j+1 )] (23) The fundamental difference between the interior elements and the exposed element is there is only a heat transfer between the two elements (via conduction) and no additional energy introduced into the system. As a result, there is a conduction component between the two interior elements and an energy accumulation within the material. A slightly different variation of Eq. 22 and 23 was used when the element increment included two different materials. The following formula was adopted: Interior element with different materials: ( k i i N 1 + k N 2 ) ( T i N 1 x 1 T N i ) ( k i N i + k N+1 2 ) ( T N i T N+1 ) x 2 = ( x 1 2 ρ N 1c P,N 1 + x 2 2 ρ N+1c P,N+1 ) ( T N i+1 i T N ) t i (24) i i T i+1 N = T i 2 t N + ( x 1 ρ N 1 c P,N 1 + x 2 ρ N+1 c P,N+1 ) [(k N 1 + k N 2 ( k N i i + k N+1 2 ) ( T N i i T N+1 )] x 2 ) ( T i N 1 i T N ) x 1 (25) 65

86 The small difference between Eq. 24 and 25 against Eq. 22 and 23 is the halving of the element increment length. This was done so the transition period between one material to the other was captured. Figure 39 illustrates the unique situation as described in the Eq. 24 and 25. Figure 39: Element increments between two different materials Finally, to describe the entire temperature profile within the material structure the numerical method was generated for the last element that was exposed to the ambient environment. This was termed the unexposed section as it was not exposed to the experimental conditions. Last element: ( k i i N 1 + k N 2 ) ( T i N 1 i T N ) h x c_solid_unexposed (T N T amb ) 4 ε solid σt N x = (ρc P ) i N 2 (T N i+1 i T N ) t (26) T i+1 N = T i N + 2 t i i (ρc P ) i N x [(k N 1 + k N 2 ) ( T i N 1 i T N ) x h c_solid_unexposed (T N T amb ) ε solid σt N 4 ] (27) The unexposed element portion of the numerical solution was similar to the exposed element. Again, Eq. 26 and 27 incorporated a conduction term for the solid materials and an accumulated energy term for the energy stored within the solid. The equations 66

87 also included terms to account for convection and radiation heat losses. With the above numerical solution adopted the entire temperature profile of the material structure was described. As a result, there was enough information available to calculate the absorbed heat flux, q net. The calculation of q net was consistent for all of the surfaces except for the overhang and shutters when they were used. The overhang was calculated using the results from the closest sensors in the ceiling because they were not equipped with their own TSCs. The material composition of the overhang was similar to that of the ceiling so this assumption was upheld. Whilst, the energy absorbed by the shutters was neglected because results from the similar wall sections demonstrated negligible amounts of net heat fluxes. Even further, the emissivity and conductivity of the shutters were lower than the walls so this theoretically resulted in even lower q net results. A key parameter required to calculate q net was the properties of all the materials that were subject to the heat transfer. Table 20 lists all of the materials and their properties that made up the surfaces. Table 20: Material properties Material ρ (kg.m -3 ) Cp (J.kg -1.K -1 ) α (-) ε (-) k (W.m -1.K -1 ) Aircrete 500 [1] 1000 [1] 0.63 [1] 0.9 [1] 0.15 [1] Rockwool Flexi (50mm) Rockwool Flexi (140mm) Rockwool Flexi (220mm) Rockwool Beam Clad 35 [3] 840 [2] [1] [3] 45 [3] 840 [2] [1] [3] 45 [3] 840 [2] [1] [3] 180 [3] 840 [2] [1] [3] Plywood 540 [4] 1210 [5] [4] Plasterboard 800 [1] 840 [1] [1] [1] (Derrick, 2006), [2] (The Engineering ToolBox, n. d.), [3] (ROCKWOOL LTD, 2014), [4] (Austral Plywood PTY LTD, 2016), [5] (Engieering.com, 2006). 67

88 All of the surface types (i.e. walls, ceiling and floor) had unique material composition. While the overhang had a similar material composition as the ceiling. Table 21 illustrates the composition of the surfaces. A more detailed description of the surface layouts can be found in Chapter 4. Table 21: Material structure composition Surface Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 Ceiling Rockwool Plasterboard Rockwool Rockwool Plasterboard Beam (15mm) Flexi Flexi (15mm) Clad (50mm) (220mm) (25mm) Walls Aircrete Rockwool Rockwool Plasterboard - (50mm) Flexi Flexi (15mm) (50mm) (140mm) Floor Rockwool Plywood Flexi (15mm) (140mm) Note*: Black Steel sheet was ignored from the heat transfer calculation because it was considered highly conductive and to have minimal effect on the heat transfer through the material structures. Finally, to fully quantify the amount of heat that was transferred to the surfaces the heat flux per unit area, q net, for each TSC was calculated for the tributary area of the TSC. The tributary area of the TSC was calculated as the midpoint distances between the TSC s and the boundary of the surfaces. The following equation represents the total heat losses accounted for within the compartment surfaces: Q loss,surfaces = q net A i where A i is the tributary are for each TSC (m 2 ). i (28) 68

89 8.6 GAS PHASE HEAT STORAGE CALCULATION To calculate the amount of heat stored in the gas phase the specific heat formula was used. This equation describes the amount of energy required to raise the temperature of a substance (Liberman, 2008). The following equation is the specific heat formula: Q gas = mc p dt dt where m is the mass of the substance that is being heated (kg), c p is the specific heat capacity of the substance (J.kg -1.K -1 ), dt is the differential change in temperature (K) and dt is the differential change in time (s). (29) A discrete method of calculation for the amount of energy stored in the gas phase was adopted for the purpose of the analysis. To calculate the mass, m, of the substance (i.e. air), a control volume with constant temperature was assumed around each thermocouple. The boundaries of each control volume were set at halfway to the next closest thermocouple. If the thermocouple was next to a surface or the opening the constraining dimensions of the experiment acted as the boundary for the control volume. To calculate a mass for each control volume the following equation was used: m = Vρ air (30) where V is the volume around each thermocouple (m 3 ) and ρ air is the density of the air around each thermocouple (kg.m -3 ). The density of air, ρ air, was calculated using the recordings from each thermocouple. Based on the temperatures recorded a density was calculated using a linear interpolation of predetermined densities and their corresponding temperatures. The same method of calculation for the density of air was used for the calculation of the specific heat capacity of air, c p. To calculate the temperature rise, T, the recordings of the thermocouples were used. At the start of the experiment an ambient temperature of 15 C was assumed. As a result, the rise in temperature was calculated as the following: 69

90 T = T TC T Amb (31) T = T TC 15 where T TC is the thermocouple temperature recording ( C) and T Amb is the ambient temperature ( C). With all the components required to solve Eq. 29 acquired the heat in the gas phase for each time-step was calculated. To convert this into a rate (i.e. per time-step), the change in heat was calculated between each time-step for the duration of the experiment. The following formula represents the change in heat stored in the gas phase: Q gas = Q gas,n Q gas,n 1 (32) where Q gas,n is the gas phase energy at time n (J) and Q gas,n 1 is the gas phase energy at time n-1 (J). n was bound by the following conditions: 2 < n < tspan where tspan is the time span of the experiment. Once the above equation was derived and run for the duration of the experiments the amount of heat stored in the gas phase was quantified. 8.7 HEAT LOSS OUT OF THE OPENING The last component of heat loss within the compartment was the hot gases leaving through the opening. To quantify the amount of heat leaving through the opening the specific heat formula was used again. The following formula illustrates the specific heat formula used: Q loss,opening = m c p T (33) where m is the mass flow rate across the opening (kg.s -1 ), c p is the specific heat capacity of the hot gasses leaving the compartment (J.kg -1.K -1 ) and T is the temperature difference between the hot gasses leaving the compartment and cooler gases entering the compartment (K). To calculate the mass flow rate there were velocity probes placed at the opening so a velocity profile could be quantified. Figure 40 illustrates the locations of all sensors used at the opening. 70

91 Figure 40: Probe Locations at the Openings In order to quantify the velocity profile a linear profile was assumed due to a lack of velocity probes available at the opening. After research was conducted in this area it was also evident that assuming a linear velocity profile for the two sensors was appropriate. Results from experiments such as Steckler s experiments showed that the velocity profile recorded at a simulated door recorded almost resembled a linear profile (Yuen et al., 2006). Kumar also conducted compartment fire experiments and simulated results from CFD modellings represent the velocity profile at the opening as a linear profile (Hasib et al., 2006). As seen in Figure 40 there were two velocity probes used to record data. This probe layout was repeated for all fifteen openings. Using the recorded data, the linear velocity profile was created. As the hot gasses were leaving the compartment these gases were getting replaced with ambient air. The upper velocity probes measured the velocity of these hot gases leaving the compartment at the set probe height. While, the lower velocity probes recorded the velocity of the ambient air that was entering the compartment. Figure 41 illustrates typical velocity profiles that occurred throughout the duration of the experiments. 71

92 Figure 41: (a) Typical Case Calculation Method of Energy Leaving the Compartment. (b) Special Case Calculation Method of Energy Leaving the Compartment Once the linear profiles were quantified for each opening and the duration of the experiments the volume flow rate was easily calculated. An approach termed the trapezoidal method was adopted for this calculation. The velocity profile was broken up into sections where a thermocouple was the bounding condition for all the sections which resembles trapezoidal sections as seen in Figure 41. It is important to note that the first section above the neutral plane (i.e. zero velocity) was triangular in shape. The area of the trapezoidal and triangular shapes was calculated based off the velocity recordings within each section as seen in Figure 41 (a) and (b) and the physical heights of each Tributary section. To fully quantify the volume flow rate, the areas of the velocity profiles were applied to the entire width of each opening (i.e. 1.1m). The velocity value required for the trapezoidal shapes were the midpoint values (i.e. between thermocouples) and for the triangular shape were the velocities at the first thermocouple above the neutral plane. The following equations illustrate the calculation of the volume flow rate: 72

93 Trapezoidal sections: V = hv mid w (34) where h is the vertical height of each trapezoidal section (m), v mid is the velocity at the midpoint of the trapezoidal section (m.s -1 ) and w is the width of the opening (m). Triangular sections: V = hv TCw 2 where v TC is the velocity at the first thermocouple above the neutral plane (m.s -1 ). (35) Finally, to convert the volume flow rate into a mass flow rate a constant density was applied to the trapezoidal and triangular sections. The density applied to the sections were calculated from the recordings of the thermocouples. Similar to section 8.6 for a given temperature recording the density was calculated based off a linear interpolation between predetermined densities and temperatures. As illustrated in Figure 41 (a) the recordings from each thermocouple were applied to the sections above the thermocouple. This was done so there was no overestimation incorporated into the amount of heat leaving the compartment through the opening. A special case is shown in Figure 41 (b) where the triangular section was calculated with the upper thermocouple. This would result in an overestimation of mass flow for that particular section. However, the significantly small magnitude of the overestimation with comparison to the rest of the flow profile was deemed to have insignificant effects on the results. To fully quantify the amount of energy leaving the compartment the specific heat capacity, c p, and temperature differences, T, for each section needed to be calculated. The specific heat capacity was calculated in the same method as the density (i.e. linear interpolation). Whereas the temperature difference was calculated as the difference between the thermocouple recordings and the minimum temperature of the thermocouples at the opening. This process was adopted to try and account for a change in ambient conditions due to a temperature rise within the burn hall caused by the experiment. 73

94 Once all of the above components were derived the amount of energy leaving the compartment through an opening was quantified as Eq. 33 could be solved. This method was repeated for all openings without shutters. 8.8 SENSITIVITY ANALYSIS Throughout the analysis extensive sensitivity tests were conducted to validate results with respect to the net heat flux calculations. The following is a list of the sensitivity analysis that were performed: Time-step and material thickness; Material composition; and Smoothing of q inc results Time-step and Material Thickness Sensitivity Analysis The first sensitivity analysis that was conducted was an analysis of the effects of changing the time-step and material thickness. Both the time-step and material thickness were assessed simultaneously such that the Fourier number could remain the same which ensured the solution would converge. The initial analysis was conducted by assuming a time-step of 1 second and a material thickness of 5mm per iteration. To assess the accuracy of these parameters they were altered to two different combinations of time-step and material thickness whilst keeping the Fourier number the same. The parameters used for the sensitivity analysis are listed below: Sensitivity Case 1: Time-step = 0.25s and material thickness = 2.5mm. Sensitivity Case 2: Time-step = s and material thickness = 1.25mm. The sensitivity analysis was conducted for a single TSC located on the back wall. Typical results of the sensitivity analysis are shown in Figure

95 Figure 42: Sensitivity analysis of the change in time-step and material thickness to 0.25s and 2.5mm respectively Figure 42 illustrates that with a change in time-step and material thickness to 0.25s and 2.5mm respectively there was little to no change in the calculated q net. As a result, the time-step and material thickness of 1s and 5mm respectively were accepted and used for the analysis as the computational times were reduced. Results from the sensitivity analysis for the other time-step and material thickness, s and 1.25mm, is illustrated in 12.2 Appendix B Material Composition Sensitivity Analysis Material Composition Sensitivity Analysis Due to issues with large computational times the material structure composition shown in Table 21 was modified in the model. The surfaces were modelled with adiabatic boundary conditions at the location of an internal layer of Rockwool Flexi due to this material having a low thermal conductivity. An adiabatic boundary implies that there are no heat losses at this location of the material. Table 22 shows the modelled material composition of the surfaces. 75

96 Table 22: Modified material composition for modelling Surface Layer 1 Layer 2 Layer 3 Ceiling Rockwool Beam Clad Plasterboard (15mm) Adiabatic Boundary (25mm) Walls Aircrete (50mm) Adiabatic Boundary - Floor Rockwool Flexi (140mm) Plywood (15mm) - In order to demonstrate that this assumption did not affect results a sensitivity analysis was performed to show that there was no variation in the calculated q net. To perform this sensitivity analysis a section of each surface was analysed using both the original material composition and the modified material composition and the results were compared. Figure 43 highlights the typical results of the sensitivity analysis. Figure 43: Sensitivity analysis of the change in material composition for the back wall From the above figure it can be seen that the modification of the material composition highlighted in Table 22 for modelling purposes resulted in essentially no variation of the calculated q net from the non-modified material composition highlighted in Table 21. As a result, modification of the material composition was accepted for modelling purposes. Results from the sensitivity analysis for the other surfaces are illustrated in 12.2 Appendix B Material Composition Sensitivity Analysis. 76

97 8.8.3 Smoothing of q inc Sensitivity Analysis Lastly, the final sensitivity analysis that was performed was an analysis of the effect of smoothing the calculated q inc. This assessment was required only to validate this as a method of eliminating noise from the overall results. To do this the effects of smoothing the q inc results using an inbuilt MATLAB function were assessed by comparing the effect on the temperature profile throughout the material composition. Further analysis was conducted by assessing the effect on the results of q net. Figure 44 - Figure 46 illustrate the results of the sensitivity analysis. Figure 44: Result of smoothing the calculated q inc 77

98 Figure 45: Temperature profile before and after the smoothing of q inc Figure 46: Change in calculated q net after smoothing q inc 78

99 The above figures show that the smoothing of q inc has essentially no effect on the overall result of the calculated q net. Most importantly the difference in temperature profile of the material has only a minute affect which is deemed negligible. As a result, the smoothing of q inc was accepted as it was shown to only eliminate the noise in the calculated q net and not affect the overall results of q net. 79

100 9 RESULTS AND DISCUSSION 9.1 FULLY-DEVELOPED FIRE EXPERIMENTS 1, 4, 7 AND 8 In order to discuss the results of the experiments they were categorised into groups. Each group consisted of a constant fire mode (e.g. Experiments 1, 4, 7 and 8 were all conducted with a fully-developed fire). The first group of experiments to be discussed were all conducted under fully-developed fire conditions. However, before the full analysis could be achieved each experiment within this group was assessed individually Experiment 1 As Experiment 1 is the first in the series of experiments to be discussed, a slightly more in depth analysis will be conducted for this experiment to avoid unnecessary repetition of content throughout the analysis As mentioned in Section 5.4 the velocity was recorded at the opening to capture the flow into and out of the compartment. Figure 47 and Figure 48 are representative of the recorded outflow and inflow velocities respectively. Both figures illustrate that the magnitude of each flow type increases throughout the duration of the experiment. This result was expected because the heat supply into the compartment through the gas burners was systematically increased throughout the experiment which corresponded to the same time the increases in flow occurred (see 12.3 Appendix C Experimental Flow Results for additional inflow and outflow data). 80

101 Figure 47: Outflow velocities for each opening during Experiment 1 Figure 48: Inflow velocities for each opening during Experiment 1 To understand the fire dynamics that were occurring within the compartment during the experiments temperature contours were created. Figure 49 illustrates the crosssections taken for the temperature contours. While Figure 50 illustrates the series of images that show the temperature contours through the centreline cross-section of each opening at 875 seconds of the experiment time. 81

102 The right-hand side of the images is representative of the opening whilst the left-hand side of the image corresponds to the back of the compartment. The large spikes of temperature in some of the images were generated by the thermocouple trees that were located close to the gas burners. It can be seen that the hot layer that had formed in the compartment was approximately at the height of the last sensor at the opening. It can also be seen that there was a component of the flow at the opening that is vertical which is characterised by the slightly non-horizontal contour at the opening. The colder ambient air that was entering the compartment can be seen as the darker portion of the temperature profile at the bottom right of the images. Figure 49: Cross-sections take for analysing the temperature contours within the compartment 82

103 83

104 Figure 50: Temperature contours of cross sections 1-1 to at t = 875s for Experiment 1 The amount of heat that was transferred into the surfaces was quantified by the net heat flux, q net, which in this case is in terms of per unit area. Figure 51 illustrates the distribution of the neat flux across the surfaces of the compartment (at time 875 seconds). As expected due the characteristic rise of heat which results in the formation of an upper hot layer within the compartment (see Figure 50), the quantity of net heat flux across the ceiling was larger across the ceiling than all other surfaces. This phenomenon was further demonstrated as the net heat fluxes across the walls were larger in the upper portion of the walls. It can be seen that there was a concentration of net heat flux in the back right hand portion of the compartment. This was caused by the propane distribution system not distributing the propane evenly across all of the gas burners. It was also shown that the net heat fluxes were overall stronger towards the back of the compartment. This was explained by the vertical lean of the flames to the back of the compartment caused by the inflow. Lastly, it could be seen that the floor was subject to comparatively negligible net heat fluxes. This was due to the floor being in the cold layer. It is important to note that the overhang was left out of Figure 51 because there were no TSCs in this portion of the compartment, for the purposes of the analysis the overhang was assumed to receive the same incident heat flux as the closest row of TSCs within the ceiling (see 12.4 Appendix D Experimental Heat Flux Results additional heat flux results from each experiment). 84

105 Figure 51: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 1 Experiment 1 as highlighted in Table 18 consisted of a fully-developed fire with overventilated conditions. Figure 52 illustrates the results of the energy distribution within the compartment. The energy distribution for Experiment 1 and the rest of the fullydeveloped fire series was discussed in terms of a rate of energy transfer (i.e. heat). Immediately it can be seen that the trend of the total heat out of the compartment follows the trend of the heat supplied to the compartment. It was also immediately obvious that the amount of heat leaving the compartment through the opening was considerably larger than any other source of heat loss. Consequently, compared with the amount of heat supplied to the compartment the amount of heat distributed to the surfaces was considerably lower. The surface subject to the largest portion of heat was the ceiling whilst the rest of the surfaces were subject to almost negligible levels of heat. 85

106 It can be seen that there were disturbances in the level of heat subject to each source of energy out. This was most noticeable in the variation of heat stored in the gas phase. This disturbance was caused by a change in experimental conditions which affected the equilibrium conditions within the compartment which are naturally assumed (i.e. a change in HRR, ventilation, the burners activated, etc.). With this change in experimental conditions began a transient period where the conditions within the compartment transitioned from an unbalanced state back to a state of equilibrium. The experimental conditions that changed during Experiment 1 were the increases and decreases in HRR which is clearly shown by the changes in heat supply to the compartment in Figure 52. It can also be seen that there is a difference in the amount of heat being supplied to the compartment and the amount of heat leaving the compartment (i.e. conservation of energy was not fully satisfied in the analysis). There were several possible reasons for this mismatch between the heat in and out. Firstly, the highest sensor at the opening was 300mm below the top of the opening. This potentially resulted in the hottest temperatures that left the compartment not being quantified. This negative affect on the results would be amplified if the transition between the hot and cold layer (i.e. the neutral axis) was located between the top of the opening and the highest thermocouple. This scenario becomes more likely with experimental conditions that are subject to lower levels of energy into the compartment (which can be seen in later experiments). This is due to an increased chance of the neutral axis being located in a higher portion of the compartment because there is not enough heat within the compartment for the hot layer to descend past this point. Another potential source of error in the calculation came with the velocity measurements. The velocity probes were only capable of recording the horizontal portion of the flow which does not account for the entire portion of the velocity due to the buoyant nature of the hot gases produced by a fire. Again, this magnitude of error would increase with lower heat experiments because of a tendency of the hot gases to leak out of the compartments openings rather than flow out. If the gases leaked out of the compartment the velocity probes were unlikely to record this flow accurately because the probe measuring the outflow was also located 300mm below the top of 86

107 the opening. Furthermore, a linear velocity profile has been assumed in the analysis and is potentially not accurate for all fire scenarios. Finally, the amount of energy leaving the compartment did not include a radiation component which is potentially a significant source of heat loss. Figure 52: Energy distribution recorded for Experiment 1 Once the distribution of energy within the compartment was quantified the energy distribution was normalised with respect to the amount of heat into the compartment. Figure 53 illustrates the normalised plot. This plot shows that 80-90% of the heat was accounted for with the method of analysis described in Chapter 8 for Experiment 1. Approximately 70% of this heat was accounted for by heat leaving the compartment through the openings. A further 10-15% of the heat was distributed to the ceiling. Of the remaining surfaces the distribution of heat was less than 4% and on average was approximately 1% each. The portion of distributed heat was only considered when approximate equilibrium conditions were achieved such that the results were not misleading. 87

108 Figure 53: Normalised energy distribution with respect to the energy supply for Experiment Experiment 4 Experiment 4 which is summarised in Table 18 was representative of a fully-developed fire and under-ventilated conditions. The results of the energy distribution for the duration of this experiment are illustrated in Figure 54. It can be seen that there was a large portion of the heat loss from the compartment not accounted for. However, the portion of heat leaving the compartment through the opening was still the largest portion of heat loss. The next largest components were the amount of heat distributed to the ceiling and the back wall respectively. The other surfaces were distributed a small portion of the heat. Even though there was a large portion of heat loss from within the compartment not accounted for the total heat loss followed the trend of the heat supply. The potential sources of error with all experiments were highlighted in the analysis for Experiment 1. These sources of potential error were essentially all with respect to the portion of heat leaving the opening. A valid point to make was that the recorded data by the TSCs which was essential for calculating the amount of heat distributed to the 88

109 surfaces was much less affected by the different fire dynamics of each experiment. This was in comparison with the recorded data for quantifying the amount of heat leaving the compartment because the velocity probes were susceptible to error while the hottest gasses leaving the compartment were potentially not being fully quantified. It is also worth noting that this experiment was conducted with the use of shutters which were not accounted for as a source of heat loss. However, it was deemed negligible because of the low absorptivity of the material and the low portion of heat that was distributed to the opposite wall (i.e. the back wall). Figure 54: Energy distribution recorded for Experiment 4 Again, as with Experiment 1 the distribution of energy within the compartment was normalised with respect to the energy into the compartment. Figure 55 illustrates the normalised plot. This plot shows that approximately only 40-50% of the total heat loss was accounted for. The largest portion of this heat loss was heat leaving through the opening which ranged from approximately 25-30%. The next largest portions were the 89

110 heat losses to the ceiling and back wall which was approximately 10% and 3% respectively. The rest of the surfaces accounted for approximately 1% each. Again, careful attention was payed to any misleading results which can occur in the transient portions of the experiment. Figure 55: Normalised energy distribution with respect to the energy supply for Experiment Experiment 7 Experiment 7 (summarised in Table 18) simulated a growing fire, variable ventilation and a shutter removal rate of 1 per 2.5 minutes. Figure 56 illustrates the energy distribution calculated for Experiment 7. Firstly, the total heat leaving the compartment was accounted for through the duration of the experiment and for a significant portion of the experiment even more than the heat supplied to the compartment was accounted for. The total heat leaving compartment demonstrated spikes in the heat loss which was explained by the analysis method for the heat leaving through the opening. The spikes in the amount of heat leaving the compartment can be seen to coincide with the removal of a shutter and simulate window breakage. This result is expected as ventilation conditions change dramatically and at this point the analysis triggered an additional opening which accounted for heat leaving the newly available space. 90

111 The amount of heat leaving the compartment through the opening was a very significant portion of the total energy out. Again, consistent with the results from the earlier discussed experiments the next largest distribution of heat was distributed to the ceiling and back wall. The other surfaces were again distributed much less significant amounts of heat. The resulted total heat loss plot appears to follow the trend of the heat supplied to the compartment. There were significant portions of the experiments that were subject to experimental conditions that were not in equilibrium (i.e. a transient period). This was due to constant changes in conditions because of the shutter removal rate used to simulate variable ventilation. The issue with the results was that for a significant portion of time the experiment a small over-estimation of the amount of heat leaving the compartment was quantified. Another potential source for error which was not previously mentioned was in the measurement of the propane gas flow. This occurred more prevalently in the later conducted experiments with increased complication that potentially affected the accuracy of the measurement of the actual propane flow into the compartment. As a result, this was a potential source for the over-estimation seen. However, on a whole the over-estimation was not too significant in magnitude given the level of assumptions made. 91

112 Figure 56: Energy distribution recorded for Experiment 7 Again to conduct further discussion, the results for Experiment 7 were normalised. Figure 57 represents the results of normalising the distribution of energy with respect to the amount of heat into the compartment. The amount of heat loss out of the opening ranged from 80% to almost 100% of the total heat supplied. The amount of heat portioned to the ceiling was approximately 15-18% while the back wall received an increased portion of heat which ranged from 4-8%. The rest of the surfaces were subject to a smaller portion of heat, approximately 2% each. As previously stated, transient periods during this experiment occurred quite often. As a result, careful attention was required such that spikes caused by non-equilibrium conditions did not affect the analysis of the results. 92

113 Figure 57: Normalised energy distribution with respect to the energy supply for Experiment Experiment 8 The final experiment in the series of fully-developed fires was Experiment 8. The summary of this experiment can be found in Table 18. Experiment 8 was a simulation of a fully-developed fire, variable ventilation and a shutter removal rate of 1 shutter every 5 minutes. The results of the distribution of energy for Experiment 8 are shown in Figure 58. It can be seen that variation between the amount of heat loss versus the amount of heat supplied to the compartment is almost negligible. While the most significant portion of heat loss was again heat losses across the opening. The next most significant losses of heat were also consistent with previously discussed experiments as heat losses into the ceiling and back wall respectively. While the other surfaces accounted for minimal amounts of heat loss. The transient states of the experiment can again be seen through the variation of energy stored within the gas phase. While these variations of gas phase energy again coincide with a change in experimental conditions. 93

114 Even though the results show a well correlated trend there was still the same potential for error as stated with the previous experiments. These potential errors may account for the minor differences between the heat supplied and the heat loss but their affects were much less in magnitude. Figure 58: Energy distribution recorded for Experiment 8 For the purpose of a more accurate analysis Figure 59 illustrates the normalised energy distribution with respect to the energy supplied to the compartment. The total amount of heat loss accounted for was approximately % of the heat supplied. Of this total heat loss approximately 70-80% was accounted for through losses across the opening. The next most significant heat loss source was into the ceiling at approximately 20%. All other sources of heat loss were approximately less than 5%. 94

115 Figure 59: Normalised energy distribution with respect to the energy supply for Experiment Comparison of Fully-developed Fires To analyse the fully-developed fire experiments across all of the simulated ventilation conditions Figure 60 and Figure 61 were created. Figure 60 shows the results for heat losses accounted for across the opening and into the ceiling. While Figure 61 illustrates the heat losses into all of the surfaces. The plot represents the average amount of normalised heat losses with respect to the heat supply for the periods of non-transient behaviour. The error bars are representative of the maximum and minimum normalised heat losses. The two figures were separated such that the results for the surfaces that contributed to minimal amounts of heat loss were not skewed by including the heat losses through the opening. Figure 60 highlights that the majority of heat losses were accounted for by the losses out of the opening for all fully-developed fire experiments. It is also highlighted how variable the results are in terms of the opening losses, this is a direct result of the potential sources of errors that have been highlighted in the individual experimental analysis. Figure 61 illustrates that the heat losses for each surface are marginally larger for over-ventilated conditions when compared with underventilated conditions. It can also be seen that the amount of heat loss to a surface 95

116 increases significantly for both variable ventilation scenarios. The maximum heat loss into a surface for any fully-developed fire scenario was approximately 20% of the heat supply. This occurred with the variable ventilation scenario and shutter rate 2 (i.e. 1 shutter removal per 5 minutes, see Table 7). Figure 60: Comparison of fully-developed fire results across different ventilation conditions (ceiling and opening heat losses) 96

117 Figure 61: Comparison of fully-developed fire results across different ventilation conditions (surface heat losses) 9.2 GROWING FIRE EXPERIMENTS 2 AND 5 The next series of fire scenarios was growing fires. This series consisted of two experiments with constant ventilation conditions. Again the energy distribution was spoken about in terms of a rate of energy transfer (i.e. heat) Experiment 2 Experiment 2 simulated an over-ventilated fire scenario (see Table 18). The results of the experiment are illustrated in Figure 62 below. It can be seen that for the majority of the experiment the total heat losses followed the trend of the total heat supply. It can also be seen that of this total heat loss the most significant portion was accounted for in losses across the opening. The next most significant contributors to heat losses was the ceiling and back wall respectively. This was consistent with the results seen in the fully-developed fire scenarios. A noticeable difference between Experiment 2 and the fully-developed fire was the magnitude and length of the transient period that can be seen in the gas phase portion of the energy distribution graph. This larger variation 97

118 within the gasses was most likely caused by the introduction of a gas burner that provided a larger heat release rate (HRR) rather than an increase in HRR of already operational burners. For the duration of the earlier stages of this experiment the results show a good correlation. However, towards the end of the experiment the total amount of heat losses that were accounted for exceeded the recorded heat supply. Similar to Experiment 7 there was a potential error introduced in this experiment through malfunctioning pressure gauge recordings. As the target HRR was actually a 2MW fire for this portion of the experiment as highlighted in Figure 25 this may well have been the actual HRR achieved. However, this potential source of error is not conclusive. Figure 62: Energy distribution recorded for Experiment 2 Figure 63 shows the normalised energy distribution with respect to the energy supplied to the compartment through the gas burners. The normalised energy distribution shows a large range of variability in terms of the heat losses. This was due to the lengthy transient periods as highlighted earlier. As a result, the careful analysis of the results 98

119 was made sure not to include these transient periods. From this it can be seen that % of the total heat losses were accounted for. Of this total heat loss 70-90% was accounted for as losses across the opening. The two next most significant portions of heat losses were losses to the ceiling and back wall surfaces which contributed to approximately 20% and 5% respectively. All of the other sources of heat loss contributed to less than 5%. Figure 63: Normalised energy distribution with respect to the energy supply for Experiment Experiment 5 Experiment 5 simulated a growing fire and an under-ventilated scenario (see Table 18). The results of this experiment are illustrated in Figure 64 seen below. Firstly, it can be seen that there was a missing portion of the heat released into the compartment through the gas burners. This was due the data from the propane flow gauges not being set to record the flow from the start of the experiment. However, for the data that was recorded the total heat loss within the compartment followed the shape of the heat supply curve but was offset by a significant portion (i.e. the total heat supply was not accounted for in the analysis). The major contributing components to the loss of heat 99

120 within the compartment were the losses across the opening and into the ceiling and back wall surfaces respectively. The rest of the surfaces made up a much less significant portion of the heat losses. The graph also showed the transient periods during the experiments through the variation in heat within the gasses as represented by the gas phase results. In this experiment there was a clear error seen in the poor correlation between the heat supplied to the compartment and the total heat loss. This was explained thoroughly in the discussion of Experiment 1 and 4 and will be reiterated briefly here. Essentially there was a significant potential source of error with respect to the heat loss across the openings. This was due to the location of the velocity probes and the density of velocity probes used. As a result, the hottest gasses leaving the compartment were not fully quantified. This error was then potentially amplified if the neutral plane that separated the hot and cold layers within the compartment was located between the top of the opening and the sensor recording outflows. Also, the velocity probes used to measure the flows in and out of the compartment only had the capacity to measure the velocities in the horizontal plane which ultimately affected the quantification of mass flow in terms of hot gases leaving the compartment. Figure 64: Energy distribution recorded for Experiment 5 100

121 To further quantify the energy distribution with the compartment the above results were normalised with respect to the energy into the compartment. This is illustrated in Figure 65. To quantify the results accurately only periods of non-transient behaviour were assessed. It can be seen that only 50-60% of the total heat loss was accounted for in Experiment 5. Of this total heat loss, the largest contribution came from heat losses through hot gases leaving the opening accounting for approximately 35%. The two next most significant losses were losses into the ceiling and back wall surfaces which accounted for approximately 13% and 5% respectively. While the rest of the surfaces contributed to approximately 2% each. Figure 65: Normalised energy distribution with respect to the energy supply for Experiment Comparison of Growing Fires To compare the growing fire experiments with different ventilation conditions Figure 66 and Figure 67 were used as illustration. Figure 66 is representative of the results for the normalised heat losses across the opening and into the ceiling surface. While Figure 67 is illustrative of the normalised heat losses into all of the surfaces. These 101

122 plots are representative of the normalised heat losses with respect to the heat supply for periods of non-transient behaviour only. The error bars used in Figure 66 and Figure 67 show the maximum and minimum range of the normalised heat losses. As with the fully-developed fires it can be seen how variable the results of the heat losses out of the openings can be. Experiment 2 showed an average contribution to the total heat losses across the opening as approximately 90%. Whereas, Experiment 5 showed and average of approximately 35%. Again illustrating the potential sources for error in the analysis was most likely within the quantification of the heat losses across the opening as previously highlighted. Similar to the comparison shown for the fully-developed fires Figure 67 illustrates clearly that the magnitude of distribution of energy to the surfaces was marginally larger for all surfaces subject to the overventilation conditions. Overall, for the growing fire experiments the maximum heat loss to a surface was 13-17% into the ceiling surface and for over-ventilation conditions. Figure 66: Comparison of growing fire results across different ventilation conditions (ceiling and opening heat losses) 102

123 Figure 67: Comparison of growing fire results across different ventilation conditions (surface heat losses) TRAVELLING FIRE EXPERIMENTS 3, 6, 9 AND 10 The last series of fire scenarios were the travelling fires. This series consisted of four experiments with the same ventilation conditions as shown in the fully-developed fire series. These experiment results are presented and discussed in the same order in terms of ventilation as the fully-developed fire series results. The energy distributions shown for the travelling fire series of experiments was again spoken about in terms of a rate of energy transfer (i.e. heat) Experiment 3 Experiment 3 consisted of a travelling fire scenario with over-ventilated conditions (see Table 18). Figure 68 is illustrative of the energy distribution results within the compartment. The first distinct characteristic of the results shown below was that the total heat loss did not match the heat supply. However, the general shape of the total heat loss curve was somewhat similar to the heat supply curve but offset significantly. 103

124 The majority of the heat loss was accounted for by losses across the openings. However, marginally less heat loss was distributed to the ceiling surface. The next most significant portion of heat loss was accounted for by the losses into the back wall. While the rest of the surfaces received a much smaller portion of the heat introduced into the compartment. The transient periods within the experiment can be seen by the variation in heat of the gasses shown in the gas phase plot. The change in the experimental conditions which caused the disturbances of equilibrium conditions was accounted for by the moving of the burner pairs activated as part of the travelling fire simulation. It was seen that Experiment 3 showed a poor correlation between the heat supplied and the total heat losses. As previously discussed in other experiments that show a similar poor correlation the error is most potentially within the calculation of the heat loss through the opening of the compartment. Most notably, for Experiment 3 the HRR of the propane burners was at the lowest rate used for any experiment (1MW). As a result, the potential for the hot layer to be within the highest thermocouple at the opening and top of the opening itself was more significant. Also, as the ventilation conditions were representative of an over-ventilated scenario and coupled with the lower HRR there was also more potential for the hot gasses to leak out of the opening being dominated by the buoyancy affect. As a result, the mass flow of the hot gasses leaving the compartment would yield a more significant error. 104

125 Figure 68: Energy distribution recorded for Experiment 3 Again following the systematic method of analysis adopted, Figure 69 illustrates the normalised energy distribution with respect to the energy into the compartment. It can be seen that only approximately 50-60% of the heat within the compartment was accounted for in terms of the total losses. Of this total heat loss, the most significant portion was accounted for by losses across the opening at approximately 30% of the heat supplied. The next most significant portions of heat distribution were to the ceiling and back wall surfaces at approximately 18% and 5% respectively. Finally, the heat losses to the other surfaces were no greater than 2%. It is important to note that this portion of the discussion only considered the non-transient periods of the experiments. 105

126 Figure 69: Normalised energy distribution with respect to the energy supply for Experiment Experiment 6 Experiment 6 simulated a travelling fire scenario and under-ventilated conditions (see Table 18). The results of the energy distribution for this experiment are shown in Figure 70 below. As with Experiment 3 it can be seen that the total heat losses were underestimated with comparison to the heat supply to the compartment. However, the general shape of the total heat loss curve followed the shape of the heat supply curve with a clear offset. In contrast to all previous experiments discussed the most significant source of heat loss was with respect to losses into the ceiling. This was closely followed by the heat loss out of the openings and much less significantly by the back wall and other surfaces respectively. The transient periods of the experiment are clearly defined by the variation in energy of the gasses shown by the gas phase results. This variation of energy follows the trend seen in Experiment 3 because the same source of change in experimental condition occurred in both experiments (i.e. the activation and deactivation of burner pairs to simulate the travelling fire scenario). 106

127 As previously stated there is an obvious error within the total heat loss analysis. The highest potential source has been highlighted several times. Again for this experiment this source of error is most likely within the quantification of the hot gasses leaving through the opening (refer to Experiment 3). Figure 70: Energy distribution recorded for Experiment 6 The normalised energy distribution with respect to the energy supplied to the compartment is show in Figure 71. Figure 71 was analysed for the portions of the experiment that displayed non-transient behaviour. It can be seen that approximately 50% of the total heat supplied the compartment was accounted for in terms of the total heat losses. The most significant portion of heat loss was contributed by losses into the ceiling at approximately 20% with respect to the heat supply. This was closely followed by the heat losses out of the opening which contributed to 18% percent of the heat loss. The back wall accounted for the next most significant portion of heat loss at approximately 5%. While the other surfaces accounted for approximately 1-2% each. 107

128 Figure 71: Normalised energy distribution with respect to the energy supply for Experiment Experiment 9 Experiment 9 was a simulation of a travelling fire scenario, variable ventilation conditions and shutter removal rate 1 (i.e. 1 shutter removal per 2.5 minutes) Table 18 for a summary of the experimental conditions. Due to complications with the guide rail system used to remove the shutters during this experiment the shutter removal rate was not as specified pre-experiment (see Table 16 for a history of the major events during Experiment 9). Additionally, as the moving rate of the travelling fire was a constant rate (see Table 18) not all of the shutters were removed. This ultimately did not allow for the full analysis of the entire ventilation range possible. During Experiment 9 there were also large spikes shown in the heat supply to the compartment. This was a result of a gas burner self-extinguishing. At this point of the experiment the HRR was dropped to 0.5MW as an almost jet flame was be induced in the remaining operable gas burner (when the HRR was 1MW) which had the potential to cause significant damage to the materials in the ceiling. 108

129 The results of the Experiment 9 in terms of the energy distribution in the compartment are shown in Figure 72. Immediately it can be seen that the correlation between the total heat loss and heat supply curves were strong. In portions of the experiment the heat loss through the opening and into the ceiling were quite similar. However, in general the largest contribution in terms of heat loss was accounted for by losses across the opening. The next most significant source of heat loss was into the ceiling which was followed by the losses into the back wall. Clear variations in the energy of the gasses were seen in the gas phase curve denoting the transient periods of the experiment. These transient periods were induced by the change in the burner conditions and the removal of shutters. Experiment 9 was again subject to the same potential sources of error as all of the experiments conducted. However, this experiment showed a good correlation between the total heat loss and heat supply suggesting that the potential errors highlighted did not have a significant effect on the results. Figure 72: Energy distribution recorded for Experiment 9 109

130 The normalised energy distribution with respect to the energy supply is illustrated below in Figure 73. In order to avoid misleading results the transient periods of the experiment highlighted in the previous graph were ignored. It can be seen that for the majority of the experiment 100% of the total heat loss within the compartment was accounted for. The most significant portion of this heat loss was accounted for by the losses across the opening. However, this was highly variable due to the quite significant and random changes of experimental conditions with an account of approximately 40-70% of the heat losses. The next most significant portion of the heat loss was accounted for by the losses into the ceiling at approximately 30-40%. Finally, the next most significant source of heat loss was accounted for by losses into the back wall ranging from 5-10%. While the other surfaces accounted for a much less overall significant portion ranging from 1-4%. Figure 73: Normalised energy distribution with respect to the energy supply for Experiment 9 110

131 9.2.7 Experiment 10 Experiment 10 was the final experiment in the travelling fire series. In addition to the travelling fire simulation, variable ventilation conditions were simulated with shutter removal rate 2 (i.e. 1 shutter removal per 5 minutes). Table 18 provides a summary of the experimental conditions. Similar to Experiment 9 the moving rate of the fire across the room was constant with all travelling fire scenario (see Table 18) and coupled with the slower shutter removal rate a limited range of ventilation conditions were assessed. As with Experiment 9 there were dramatic variations of HRR during the experiment. This was again explained a propane burner self-extinguishing and creating a jet like flow from the other operable burner within the pair when the HRR was 1MW. As a result, the HRR was reduced to 0.5MW until the fire front systematically moved to the next burner pair to avoid damage to the ceiling materials. The energy distribution within the compartment is shown in Figure 74. Figure 74 shows a good correlation between the heat supply to the compartment through the gas burners and the total heat losses. The most significant portion of heat loss was accounted for through the flow of hot gasses out of the opening. The next most significant portion of heat distribution in terms of losses was attributed to the ceiling. This was followed by the heat losses into the back wall while the other surfaces were less significant. The transient periods which are shown by the variation of energy within the gasses are clearly seen. While these periods were again induced by the change in burner conditions and shutter removal. The inherent error that has been well described within the other experimental discussions was again shown to have much less significant effect on the results for Experiment 10. This is highlighted by the strong correlation shown between the heat supply from the propane burners and the total heat losses. 111

132 Figure 74: Energy distribution recorded for Experiment 10 To account for a more accurate analysis of the energy distribution within the compartment for the duration of Experiment 10 a normalised energy distribution plot with respect to the amount of energy supplied to the compartment was created. Figure 75 is illustrative of the normalised energy distribution for Experiment 10. Consistent with the rest of the analysis only transient periods were considered when quantifying the heat losses. It can be seen that the total heat loss within the compartment accounted for was approximately 100% due to the strong correlation previously stated. The heat loss through the opening made up the largest portion of the total heat loss which ranged from 50-60%. The next most significant portion was approximately 30% of heat loss accounted for by losses into the ceiling. While the back wall received approximately 10% of the heat supplied. The other surfaces received approximately 2-4% of the heat supplied. 112

133 Figure 75: Normalised energy distribution with respect to the energy supply for Experiment Comparison of Traveling Fires Finally, to compare all of the travelling fire experiments across the different ventilation conditions Figure 76 and Figure 77 were generated. Figure 76 represents the results for the normalised heat losses across the opening and into the ceiling. While Figure 77 is illustrative of the normalised heat losses distributed to all of the surfaces. These plots are only representative of the normalised heat losses for periods of non-transient behaviour only. The error bars seen in Figure 76 and Figure 77 are representative of the maximum and minimum range of normalised heat losses. Similar to the other two fire scenario comparisons the variability in results of the heat loss across the opening between experiments was obvious highlighting the potential error with this portion of the analysis. In contrast to all other experiments the underventilated scenario demonstrated less heat loss across the opening than into the surface. However, there was a large portion of the total heat loss not accounted for in this experiment. In terms of the heat losses into surfaces, Figure 77 illustrates marginally larger heat losses for the under-ventilated scenario compared to the over- 113

134 ventilated scenario. As with the results shown in the fully-developed fire comparisons there was a significant increase in heat distribution to the surfaces for both variable ventilation scenarios. The overall maximum heat loss into a surface for the travelling fire simulations was an average of 32% and peaked at almost 40%. These peak heat distributions occurred in the variable ventilation scenario conducted with shutter rate 1 (i.e. 1 shutter per 2.5 minutes, see Table 18). Figure 76: Comparison of travelling fire results across different ventilation conditions (ceiling and opening heat losses) 114

135 Figure 77: Comparison of travelling fire results across different ventilation conditions (surface heat losses) 115

136 10 CONCLUSIONS AND RECOMMENDATIONS 10.1 KEY FINDINGS Through the series of experiments conducted there were several conclusions possible. Firstly, when comparing the traditional over and under-ventilated scenarios similar energy distributions were seen across all fire modes simulated. However, if there was any discrepancy between the two ventilation conditions two out of three fire modes showed a larger distribution of energy to the surfaces for over-ventilated conditions contradicting the original compartment fire framework assumptions. The highest distribution of energy was accounted for in the losses into the ceiling at approximately 20% of the energy supplied to the compartment through the burners. Overall, the travelling fire simulation resulted in the highest energy distributions to the surfaces. With respect to the variable ventilation experiments the distribution of energy to the surfaces was seen to increase compared to the constant ventilation experiments. The highest distribution of energy for the variable ventilation scenarios peaked at approximately 40% heat loss into the ceiling. Overall, the highest amount of energy distribution again occurred during the travelling fire simulations. On a whole, every experiment except one showed that the largest contribution of energy loss was accounted for across the opening. However, the largest amount of variability in results were also shown to be in the quantification of the losses across the opening. This was due to the potential error essentially created by the sensor layout at the opening. The highest thermocouple was 300mm below the top of the opening and resulted in the hottest gasses leaving the compartment not being quantified. This source of error was seen to increase in magnitude when the HRR of the burners were set to the lower end of their range (i.e MW). As the energy within the compartment was lower the hot layer was less likely to descend below the highest thermocouple and resulted in a gross under-estimation of the hot gasses leaving the compartment. Further potential error was seen in the use of the velocity probes for the mass flow calculations of these hot gasses leaving the opening. These probes only measured horizontal flows and as a result did not fully quantify the velocity leaving the 116

137 compartment. Coupled with the location of the velocity probe at 250mm below the top of the opening if the gasses leaked out of the opening the flow was not accurately quantified again. Also, with the lack of sensors at the opening an assumption of a linear velocity profile was made. While this may not be the case research showed that this was a reasonable assumption. A full quantification of the energy distribution within the compartment was required to satisfy the conservation of energy within the control volume. However, the quantification of the losses across the opening were not necessarily needed for understanding thermal loads rendering the inherent errors within this calculation not detrimental to the results. Overall, the losses into the surfaces was key to understanding the thermal load within the structure while the losses across the opening allowed for completeness in the conservation of energy within a control volume FURTHER RESEARCH There are several areas where further research can be conducted. Firstly, spatial analysis within the compartment could be conducted in order to identify any areas which are subject to large portions of the energy distribution quantified. Secondly, additional experiments could be conducted using slightly more appropriate sensor design in order to quantify sources of energy losses more accurately (i.e. more thermocouples and velocity probes at the openings). Lastly, a series of sensitivity analysis experiments could be conducted in order to establish the effect of varying the following: Ventilation arrangements; Surface materials used; Compartment size; Fire modes; or Fire sizes (in terms of heat release rate) The above list would further enhance the potential platform for fire safe designs of structures. 117

138 11 REFERENCES Austral Plywood PTY LTD. (2016). "Characteristics." Retrieved , from Austral Plywood PTY LTD. (2016). "Thermal Properties." Retrieved , 2016, from Bornakke, C. and Sonntag, R. E. (2013). Fundamentals of Thermodynamics. Michigan, Don Fowley. Buc, E. C. (2008). Fire Testing and Fire Reality: What Do Fire Tests Really Tell us about Materials? Derrick, B. (2006). Environmental design. CIBSE Guide A. Norwich. 7. Drysdale, D. D. (2011). An Introduction to Fire Dynamics. West Sussex, John Wiley & Sons. Engineer.com (2006). "Thermal Conductivity." Retrieved , from Harmathy, T. Z. (1972). "A New Look at Compartment Fires, Part 1." Fire Technology 8(3). Harmathy, T. Z. (1980). "The Possibility of Characterising the Severity of Fires by a Single Parameter." Fire and Materials 4(71). Harmathy, T. Z. (1981). "The Fire Resistance Test and its relation to Real-world Fires." Fire and Materials 5(112). Harmathy, T. Z. and Mehaffey, J. R. (1983). "Post-flashover Compartment Fires." Fire and Materials 7(2). Hasib, R., Kumar R., Shashi and Kumar, S. (2006). "Simulation of an experimental compartment fire by CFD." Building and Environment 42: Hidalgo, J. P. (2013). Compartment Design for the Research Project "Real Fires for the Safe Design of Tall Buildings". UK, University of Edinburgh. Hidalgo, J. P., Maluk, C., Cowlard, A., Abecassis-Empis, C., Krajcovic, M. and Torero, J. L. (2015). "A Thin Skin Calorimeter (TSC) for Quantifying Irradiation During Largescale Fire Testing." International Journal of Thermal Sciences. Incropera, F. P., Bergman (2011). Fundamentals of Heat and Mass Transfer. Jefferson City, John Wiley & Sons, Inc. Ingberg, S. H. (1928). "Tests of the Severity of Building Fires." Quarterly of the National Fire Protection Association 22(1). 118

139 Ingberg, S. H. (1942). Building Materials and Structures. Fire-Resistance Classifications of Building Constructions. Washington. Karlsson, B. and Quintiere, J. G. (1999). Enclosure Fire Dynamics. New York, CRC Press. Kawagoe, K. and Sekine, T. (1964). Estimation of Fire Temperature-Time Curve in Rooms. B. R. Institute. Japan. Majdalani, A. H., Cadena, J. E., Cowlard, A., Munoz, F. and Torero, J. L. (2015). "Experimental characterisation of two fully-developed enclosure fire regimes." Fire Safety Journal. MathWorks (2016). "MATLAB Product Description." Retrieved , from McCaffrey, B. and Heskestab, G. (1976). A Robust Bidirectional Low-Velocity Probe for Flame and Fire Applications. Combustion and Flame. O'Connor, D. J., Morris B. and Silcock G. W. H. (1997). Flux Measurement in Real and Standard Fires. Proceedings of the Second Cardington Conference: part 1, UK: Spon Press. ROCKWOOL LTD. (2014). "ROCKWOOL FLEXI." Retrieved , 2016, from ROCKWOOL LTD. (2014). "Thermal Performance of Fire Stopping/Protection Products." Retrieved , 2016, from Rodak, S. and Ingberg, S. H. (1967). Full-Scale Residential Occupancy Fire Tests of National Bureau of Standards Report #9527. US. Schumacher, P. (2011). The Autopsies of Architecture: A New Framework for Architecture. West Sussex, John Wiley & Sons. Stern-Gottfried, J. and Rein, G. (2012). "Travelling fires for structural design-part I: Literature review." Fire Safety Journal 54: Stern-Gottfried, J. and Rein, G. (2012). "Travelling fires for structural design-part II: Design methodology." Fire Safety Journal 54: The Engineering ToolBox. (n. d.). "Solids - Specific Heats." Retrieved , from Thomas, P. H. and Heselden, A. J. M. (1972). "Fully developed fires in single compartments." Fire Research Note No

140 Welch, S., Jowsey, A., Deeny, S., Morgan, R. and Torero, J. L. (2007). "BRE Large Compartment Fire Tests - Characterising Post-Flashover Fires for Model Validation." Fire Safety Journal 42: Yuen, R. K. K., Lee, E. W. M., Lo, S. M. and Yeoh, G. H. (2006). "Prediction of temperature and velocity profiles in a single compartment fire by an improved neural network analysis." Fire Safety Journal 41:

141 12 APPENDICES 12.1 APPENDIX A TIME-STEP AND MATERIAL THICKNESS SENSITIVITY ANALYSIS Figure 78: Sensitivity analysis of the change in time-step and material thickness to s and 1.25mm respectively 121

142 12.2 APPENDIX B MATERIAL COMPOSITION SENSITIVITY ANALYSIS Figure 79: Sensitivity analysis of the change in material composition for the ceiling Figure 80: Sensitivity analysis of the change in material composition for the floor 122

143 Figure 81: Sensitivity analysis of the change in material composition for the left wall Figure 82: Sensitivity analysis of the change in material composition for the right wall 123

144 12.3 APPENDIX C EXPERIMENTAL FLOW RESULTS Experiment 2 Figure 83: Outflow velocities for each opening during Experiment 2 Figure 84: Inflow velocities for each opening during Experiment 2 124

145 Experiment 3 Figure 85: Outflow velocities for each opening during Experiment 3 Figure 86: Inflow velocities for each opening during Experiment 3 125

146 Experiment 4 Figure 87: Outflow velocities for each opening during Experiment 4 Figure 88: Inflow velocities for each opening during Experiment 4 126

147 Experiment 5 Figure 89: Outflow velocities for each opening during Experiment 5 Figure 90: Inflow velocities for each opening during Experiment 5 127

148 Experiment 6 Figure 91: Outflow velocities for each opening during Experiment 6 Figure 92: Inflow velocities for each opening during Experiment 6 128

149 Experiment 7 Figure 93: Outflow velocities for each opening during Experiment 7 Figure 94: Inflow velocities for each opening during Experiment 7 129

150 Experiment 8 Figure 95: Outflow velocities for each opening during Experiment 8 Figure 96: Inflow velocities for each opening during Experiment 8 130

151 Experiment 9 Figure 97: Outflow velocities for each opening during Experiment 9 Figure 98: Inflow velocities for each opening during Experiment 9 131

152 Experiment 10 Figure 99: Outflow velocities for each opening during Experiment 10 Figure 100: Inflow velocities for each opening during Experiment

153 12.4 APPENDIX D EXPERIMENTAL HEAT FLUX RESULTS Experiment 2 Figure 101: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Experiment 3 Figure 102: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 3 133

154 Experiment 4 Figure 103: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Experiment 5 Figure 104: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 5 134

155 Experiment 6 Figure 105: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment Experiment 7 Figure 106: Net heat flux of all surfaces (excluding the overhang) at t = 875s for Experiment 7 135