Sie können Operatoren mit Ihrer Suchanfrage kombinieren, um diese noch präziser einzugrenzen. Klicken Sie auf den Suchoperator, um eine Erklärung seiner Funktionsweise anzuzeigen.
Findet Dokumente, in denen beide Begriffe in beliebiger Reihenfolge innerhalb von maximal n Worten zueinander stehen. Empfehlung: Wählen Sie zwischen 15 und 30 als maximale Wortanzahl (z.B. NEAR(hybrid, antrieb, 20)).
Findet Dokumente, in denen der Begriff in Wortvarianten vorkommt, wobei diese VOR, HINTER oder VOR und HINTER dem Suchbegriff anschließen können (z.B., leichtbau*, *leichtbau, *leichtbau*).
Der Artikel geht auf das kritische Zusammenspiel zwischen Heizungs-, Lüftungs- und Klimaanlagen (HLK) und Branddynamik ein, wobei ein besonderer Schwerpunkt auf bedarfsgeregelten Lüftungssystemen (DCV) liegt. Es unterstreicht die Bedeutung des Verständnisses, wie HLK-Systeme die Brandentwicklung beeinflussen und umgekehrt, insbesondere in Gebäuden, in denen die Druckkontrolle von entscheidender Bedeutung ist, wie etwa in Nuklearanlagen und Gesundheitseinrichtungen. Die Studie stellt eine umfassende numerische Analyse mit dem Fire Dynamics Simulator (FDS) dar und vergleicht Simulationen mit Daten einer realen experimentellen Kampagne in einem simulierten Schulgebäude. Zu den zentralen Themen gehören die Modellierung von klappenoptimierten DCV-Systemen, die Auswirkungen von brandinduzierten Drücken auf Lüftungsströme und die Herausforderungen der Synchronisierung experimenteller Daten mit Simulationsinputs. Der Artikel diskutiert auch die Beschränkungen und Unsicherheiten bei der Vorhersage von Brandverhalten und HLK-Wechselwirkungen und liefert wertvolle Erkenntnisse zur Verbesserung des Brandschutzes und der HLK-Konstruktion in verschiedenen Gebäudetypen. Die detaillierte Untersuchung von Druckdynamik, Lüftungsraten und Gaskonzentrationen bietet ein differenziertes Verständnis, wie Brände die Leistung von HLK beeinflussen und wie diese Systeme für eine bessere Brandkontrolle und -sicherheit optimiert werden können.
KI-Generiert
Diese Zusammenfassung des Fachinhalts wurde mit Hilfe von KI generiert.
Abstract
Modern heating, ventilation, and air conditioning (HVAC) systems have evolved from simple on-off, fan-driven systems to highly complex, energy-optimized systems involving sensors monitoring the building whose outputs result in dynamic changes to the HVAC system operation. In some buildings, the HVAC system is intended to aid in smoke and pressure control during the event of a fire. In such a case, the smoke, heat, and pressure from fire growth and spread interact with the HVAC system, while the control logic may react to the fire alarm and increase ventilation rates. A series of tests investigating the performance of modern damper-optimized demand control ventilation (DCV) systems during a fire and its effect on smoke and pressure control was recently performed. This paper examines the ability of Fire Dynamics Simulator (FDS) to model a DCV HVAC system undergoing a dynamic response change due to the presence of fire. Results show that the FDS HVAC model is capable of such simulations. However, there were challenges in the modelling process due to the limitations on the experimental data obtained from the real-world building management system software. A path forward for more complete simulations is identified.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
1 Introduction
Heating, ventilation, and air conditioning (HVAC) systems have become integral to many buildings, e.g., residential, commercial and industrial buildings. An HVAC system consists of several elements, such as ducts, filters, fans and dampers. The primary objectives of HVAC in residential buildings are thermal comfort (by regulating the room temperature and humidity) and air purification [1, 2]. However, HVAC systems are also used to control the pressure balance in buildings. In nuclear facilities, the mechanical ventilation system is designed, in normal operating conditions, to create a pressure cascade to prevent the release of radioactive materials to the surroundings [3]. In healthcare facilities, negative pressure rooms are used to isolate rooms to prevent the spreading of airborne infectious diseases. The effect of different ventilation strategies, i.e., negative and positive pressure rooms, has been studied, for example, during the recent COVID-19 pandemic [4]. In most buildings, however, a balanced pressure distribution is desired so that the effort for opening and closing doors is not affected by any unintended pressure buildup due to an unbalanced HVAC system.
In the event of a fire, the ventilation strategy in a building plays a major role in smoke and pressure control. Some buildings employ a compartmentalization strategy in the case of a fire, which means that the HVAC system is turned off and fire dampers close in response to heat at a pre-defined temperature and seal the ventilation ducts which enforces a physical barrier for smoke movement between compartments. In other buildings, the ventilation system is used for smoke ventilation. In this case, there is a strong coupling between the fire and the ventilation flow rates delivered by the HVAC system, especially through pressure effects in air-tight configurations, as is the case in passive houses or nuclear facilities. In passive houses, fire-induced pressures might hinder evacuation in the early fire stages and can even lead to structural damage at a more developed stage [5]. In nuclear facilities and healthcare institutions, the pressure cascade may be compromised. Furthermore, reverse flows might occur, leading to the spread of combustion products through the fresh air inlet ducts [3]. The fire has thus a substantial effect on the HVAC. On the other hand, the HVAC system may influence the fire development leading to different regimes such as the specific fire oscillatory behavior regime observed in [6].
Anzeige
The strong coupling illustrated above renders very important the analysis of the interface between HVAC and fire [7], as already addressed, for example, in the early 1980s by Rye [8], with the intent to design efficient mitigation measures for smoke control. Subsequently, fire models, i.e. zone models or Computational Fluid Dynamics (CFD) models, are coupled with HVAC to serve the purpose of such analysis, e.g. [5, 9, 10]. For example, an HVAC module is incorporated in the CFD code Fire Dynamics Simulator (FDS) [11]. The HVAC model contains all the essential features mentioned above such as ducts, filters, fans and dampers.
In most of the previous modelling validation studies on the interface fire - HVAC, the behavior of the HVAC system is dominated by the fire conditions. For example, in Wahlqvist and Van Hees [12], the flow rates delivered by the fans positioned in the inlet and exhaust ducts are a function of the fire-induced pressure and the fan characteristics (e.g. stall pressure and volume flow rates at ambient conditions). However, demand-controlled ventilation (DCV) systems in modern buildings can be adjusted dynamically by being controlled by an automation system and are therefore not (solely) controlled by the fire. The control mechanism of such systems has evolved to highly complex, energy-optimized systems due to the ever-more stringent energy regulations for buildings. Reducing ventilation rates reduces the fan power and heating or cooling demand, constituting a significant energy saving [13]. The automation system consists of several elements (i) sensing, (ii) data gathering, (iii) central control system, and (iv) transmission system. Some key parameters for the automation system are the pressure, the flow and the damper and valve positions [8]. The most energy-efficient DCV system is the so-called damper-optimized system, which ensures a minimum pressure rise over the fan. The required fan speed in a damper-optimized DCV is calculated by an optimizer, a dedicated control unit, which takes the required airflow rate based on room sensors, the supplied airflow rate and the damper angle of all DCV dampers into account. Each DCV damper in such a system measures the airflow rate and adjusts its damper angle accordingly [2]. Damper-optimized systems can also be integrated with fire alarm systems such that the ventilation rates increase on detection in order to aid smoke evacuation. This is, for example, a common practice in Norway [14].
The effects of an automation system can be modeled in FDS. Changes in the flow rate of fans can be accounted for by use of a fan model or via the use FDS control system logic. Similarly, FDS has inputs to specify variable flow losses due to a damper changing position. This paper limits its focus to modelling of damper position and fan effects are specified a priori in the model inputs using system measured flowrates. To the authors’ knowledge this is the first published effort using the FDS variable loss feature.
The paper presents a numerical study of an automated HVAC system interacting with a fire and compares the predictions with data collected in a real-scale experimental campaign. This is in the context of the BRAVENT project which is targeted at examining smoke control systems in public schools in Norway[15‐17]. It is important to mention that, in the numerical study presented in this paper, we do not attempt to fully simulate the automated system, but rather present the capability to adjust dynamically the HVAC flow rates during a fire, by varying a posteriori the damper positions based on experimental data. The BMS software and optimizer algorithm is therefore not modeled. The dynamically changing HVAC losses and the ability of FDS to predict the resulting flow rates under fire conditions is the novel aspect of this paper.
Anzeige
2 Experimental Setup
The numerical model developed in the present study is based on and compared to experimental data obtained during a campaign of 14 different large-scale fire experiments in a mock-up school building [15‐17] equipped with an HVAC system controlled by a building management system (BMS). This section provides a brief overview of the major relevant details for the modelling in this work. A more detailed discussion can be found in the aforementioned references. The model development in the present work is based on Test 12.
2.1 Test Building
The fire experiments were performed in a test building with a ventilation system (discussed in Sect. 2.3) representing a school with three rooms representing a classroom, an office, and a corridor (see Fig. 1). However, the results obtained in the present study should also apply to other types of buildings using such systems such as offices and hospitals. Since the test building was built on a slightly sloped concrete surface, the room heights varied from 3.01 m in the corridor to 3.18 m in the classroom (bottom to top of Fig. 1). Building surfaces are described below with material properties given in Table 1. The layers for the exterior surfaces are described from inside to outside.
Fig. 1
The test building and ventilation system. Exhaust air is shown with yellow ducts, and supply air is shown with blue ducts
Ceiling - Three layers: SeaROX (0.03 m), OSB (0.0098 m), and Rockwool (0.095 m in the classrooms and 0.12 m in the corridor).
Floor - Concrete (0.254 m)
Interior Wall - Five layers: Gypsum (0.0125 m), OSB (0.012 m), Air (0.094 m), OSB (0.012 m), Gypsum (0.0125 m).
Exterior Wall - Two layers: Gypsum (0.0125 m) and OSB (0.012 m).
Above the Rockwool insulation, the roof was covered with corrugated steel sheets. All doors were EI30 rated fire doors. Door construction details were not known and the doors were modeled as two (0.004 m layers of OSB with a (0.038 m air gap. As the doors are a very small fraction (\(\approx\)3 % in the office and corridor) of the surface areas of a room, they are not a significant contributor to wall heat transfer, and this assumption was not a significant source of predictive error.
Table 1
Material properties
Material
Density
Specific heat
Conductivity
\(\hbox {kg}\,\hbox {m}^{-3}\)
\(\hbox {kJ}\,\hbox {kg}^{-1}\,\hbox {k}^{-1}\)
\(\hbox {W}\,\hbox {m}^{-1}\,\hbox {K}^{-1}\)
Air
1.0
1.0
0.03
Concrete
2400
0.75
1.6
Gypsum
930
1.09
0.17
OSB
650
1.55
0.13
Rockwool
20
0.70
0.03
SeaROX
150
0.70
0.95
2.2 Air Leakage Measurements
A series of blower door tests according to ISO 9972 [18] was conducted for all rooms before the start of the experimental campaign. Overpressure (blower flow into the compartment) and underpressure (blower flow out of compartment) tests were performed for each of the three compartments. For the classroom and office tests, the corridor door at the office end of corridor was open for makeup air. For the corridor tests the doors to the office and classroom were closed. Two sets of corridor tests were performed where one set sealed the gaps around the door edges with tape. A blower door test measures the volume flow rate, \(V_L\) in \(\hbox {m}^{3}\,\hbox {s}^{-1}\), of leakage for a pressure difference, \(\Delta P\) in Pa, across the blower door. The equation used is shown below.
$$\begin{aligned} \dot{V}_L = C_d A_{L,ref} \left( \frac{\left| \Delta P \right| }{\Delta P_{ref}} \right) ^{n_P-0.5} \text {sign}\left( \Delta P \right) \sqrt{2\frac{\left| \Delta P \right| }{\rho }} \end{aligned}$$
(1)
where \(C_d\) is a discharge coefficient (typically assumed to be 1 for a leakage test), \(\rho\) is the ambient air density in \(\hbox {kg}\,\hbox {m}^{-3}\), \(A_{L,ref}\) is reference leakage area in \(\hbox {m}^{2}\), \(\Delta P_{ref}\) is the reference pressure (typically 50 Pa), and \(n_p\) is the pressure exponent (the leakage area typically increases with \(\Delta P\)). Each blower door configuration was performed for multiple values of \(\Delta P\), and a least squares fit was performed to determine \(A_{L,ref}\) and \(n_p\) for that configuration. A summary of results from all blower door tests is shown in Table 2 where the uncertainty represents the uncertainty of the least-square fit.
Table 2
Blower door test conditions, leak area and pressure exponent measurements
Room
Test conditions (50 Pa)
Leak area (\(\hbox {m}^{2}\))
Pressure exponent
Corridor
Overpressure
0.0096 (±3.2%)
0.762 (±0.045)
Corridor
Underpressure
0.0079 (±0.6%)
0.717 (±0.008)
Corridor Doors Sealed
Overpressure
0.0084 (±9.0%)
0.755 (±0.123)
Corridor Doors Sealed
Underpressure
0.0075 (±2.5%)
0.700 (±0.032)
Office
Overpressure
0.0031 (±7.1%)
0.842 (±0.104)
Office
Underpressure
0.0024 (±7.1%)
0.724 (±0.016)
Classroom
Overpressure
0.0056 (±9.7%)
0.738 (±0.124)
Classroom
Underpressure
0.0052 (±9.7%)
0.765 (±0.022)
The leak areas and pressure exponents that were obtained in the blower door tests represent the net effect of all the available flow paths for each test. These must be apportioned over the different leakage paths in the FDS model. For example, the corridor leak area includes leakage from the corridor to the ambient through the corridor walls and ceiling, leakage from the corridor to the ambient through the second door in the corridor, leakage into the classroom and office via the walls to those rooms, and leakage into the classroom and office via the doors to those rooms. For the FDS model, these individual flow path areas and leak pressure exponents must be defined. To do this, all the leak test data were entered into a Microsoft Excel spreadsheet. Leakage equations were defined for each possible flow path (each unique pair of rooms via a wall or via a door). The data from Table 2 were used to define initial guesses for \(A_{L,ref}\) and \(n_P\) for each leak path. The door \(A_{L,ref}\) guess was estimated using the difference in \(A_{L,ref}\) for the sealed vs. unsealed corridor tests. For walls the initial guess apportioned the total \(A_{L,ref}\) based on the flow path area (i.e., if the office wall to the corridor was 10 % of the total corridor area it was given 10 % of the value of \(A_{L,ref}\)).The Microsoft Excel genetic solver was used to optimize the the values for \(A_{L,ref}\) and \(n_P\) for each leak path by using the root mean square error between the predicted leak flows and the measured leak flows during testing. The resulting optimized values, shown in Table 3, were used as the FDS inputs for describing leakage.
Table 3
Adjusted Leak Areas and Pressure Exponents
Leak path
Area (\(\hbox {m}^{2}\))
Pressure exponent
Corridor-Ambient
0.0055
0.714
Corridor-Office
0.0010
0.714
Corridor-Classroom
0.0011
0.714
Classroom-Ambient
0.0043
0.749
Office-Ambient
0.0017
0.771
Doorway
0.00075
0.780
2.3 Ventilation System
The test building was equipped with a damper-optimized, balanced demand-controlled ventilation (DCV) system designed according to the room sizes and intended usage. The minimum, \(\dot{V}_{min}\), and maximum,\(\dot{V}_{max}\), design airflow rates in \(\hbox {m}^{3}\,\hbox {s}^{-1}\) are shown in Table 4, and a schematic of the ventilation system that was included in the simulations is shown in Fig. 1. In Test 12, the air flows were set to \(\dot{V}_{min}\) at test start. The ductwork pressure drop during V\(_{min}\) and V\(_{max}\) operation was calculated using the software package MagiCAD for Revit 2022 UR-2 and used to determine the loss coefficient inputs for the FDS simulations (\(K_j\) in Eq .2).
The system was operated as a once-through system (no recirculating flow). A regenerative heat exchanger (RHE) coupled the supply and exhaust flows. The RHE is an energy-saving device that uses the exhaust air being discharged to outside to precondition the supply air being taken from the outside. During testing, the fire heats the exhaust flow causing the RHE to preheat the supply air.
Table 4
Design minimum \(\dot{V}_{min}\) and maximum \(\dot{V}_{max}\) air flow rates for the test building
As seen in Table 4, the ventilation system was dimensioned to provide air to additional rooms, which were not built, with a total area of 300 \(\hbox {m}^{2}\). This was done to maintain a realistic ratio between the smoke extracted from the fire room and fresh, colder air extracted from all other rooms and to have a realistic ratio between the pressure build-up in the fire room and the total capacity of the air handling unit (AHU). The ducts to and from this extra 300 \(\hbox {m}^{2}\), along with the AHU and RHE, were located outside the test building. All ventilation ducts were uninsulated and suspended from the ceiling using standard ventilation brackets.
2.4 Demand Control Ventilation (DCV) Dampers
Two different types of DCV dampers were used during the experimental campaign. The supply and exhaust extraction dampers to the classroom and the corridor were the brand name type LEO, which is a complete Variable Area Volume (VAV) measuring and control unit. For Test 12, the dampers on the fire room’s exhaust and supply ducts were positioned in the corridor outside the classroom. The airflow rate for the LEO damper is measured with a thermal-mass flow sensor. The measurement uncertainty for a thermal-mass flow sensor is expected to increase once the air temperature inside the dampers exceeds the upper operational range of 50 °C of the sensor. In Test 12, the dampers for the office were of the brand name type TVE which has different construction than the LEO type. It uses the same thermal-mass flow measuring principle as LEO. All dampers and ductwork for the office and the corridor were 0.16 m in diameter. The ductwork and dampers for the classroom had a larger diameter of 0.315 m. As non-fire rated dampers, plastic was used in the damper and flow sensor construction.
The DCV damper flow losses depend on the damper position, which varied over time during the test. The pressure drop for the \(\varnothing\) 0.315 m and \(\varnothing\) 0.16 m LEO dampers was measured for nine damper positions using a test duct. The test results, shown in Table 5, were used as a look-up table for the flow losses in the FDS model as a function of the damper positions over time as discussed in Sect. 3.1.1. The flow loss was interpolated from the values in Table 5 using the BMS reported damper postion. An exponential interpolation was used to reflect the general exponential nature of the position vs. loss curve as seen later in Fig. 8a
Table 5
Measured loss coefficient \(K_v\) based on damper position (i.e., opening degree) for LEO with \(\varnothing\)0.315 m and \(\varnothing\)0.16 m
\(\varnothing\)0.315 m
Position %
24
28
38
42
47
57
64
71
100
\(K_v\)
278.9
130.1
59.98
31.86
18.4
6.61
3.39
1.69
0.67
\(\varnothing\)0.16 m
Position %
26
29
38
41
52
62
72
80
100
\(K_v\)
3492.6
936.3
177.2
64.54
16.24
5.3
2.17
1.08
0.6
Damper position was continuously controlled by a programmable logic controller (PLC) using the BMS measured pressures, flows rates, temperatures, and CO\(_2\) concentrations. The details of the logic for determining damper position during normal operation was proprietary and not known to the test team. Upon fire detection via multicriteria smoke detectors (AutroGuard®V-430) mounted in each room, the BMS system responds to provide \(\dot{V}_{max}\) to all spaces. This is accomplished by increasing the AHU power to obtain 100 % of the designed maximal airflow rate for the building. The optimizer function of the PLC adjusts the DCV damper positions to supply \(\dot{V}_{max}\) to the compartments. As a once-through system, providing 100 % supply means also having 100 % exhaust which means the RHE will provide maximum heat exchange between the exhaust and supply air streams.
During Test 12, approximately 950 s after ignition, plastic components in the classroom exhaust damper melted, leading to faulty airflow rate measurements. This was not a planned event. The present study simulates the first 900 s of the experiment to avoid making comparisons with questionable flow rate data due to temperature effects on the BMS sensors located in the HVAC system. Temperatures in the other BMS flow measurement locations remained within the range of normal operation.
2.5 Fire Scenario
For Test 12, two separate fires were started in the classroom. The first fire was a polyurethane (PU) mattress that primarily served as a source of soot, and the second fire was a propane burner that primarily served as a known source of heat. The mattress was placed inside a box of Siporex blocks (autoclaved, aerated concrete) in order to reduce the access to oxygen and to increase the smoke production during the fire. The top side of the box was open. The propane fire used a 70 cm \(\times\) 30 cm \(\times\) 29 cm size burner filled up to 2/3 of the height with leca balls (light expanded clay aggregate). Ignition of the matress used a rock wool wick with 120 mL placed on the floor next to the mattress. The test started by first manually igniting the mattress, then manually igniting the propane, and lastly closing all doors in the test building as the person who lit the fire exited the building. There was a 26 s delay between the start of the mattress fire and the start of the propane fire resulting from the time to change location in the fire room followed by the time for propane to flow into the burner and reach a flammable concentration. Following ignition, the propane gas flow was step-wise increased, reaching a maximum flow rate of 5.83 \(\hbox {g}\,\hbox {s}^{-1}\), held constant for around 360 s before being reduced to a lower plateau.
Video footage was used to approximate the size of the PU fire. The video shows the fire took approximately 60 s post-ignition to grow to a quasi-steady size estimated to be 20 kW based on the flame length and diameter estimated from the video. This fire maintained that size for approximately 440 s and then decayed to 0 kW (no visible flame or glow in the video) over 740 s. The growth and decay were defined as linear ramps. The measured PU mass loss for the test was 669 g. Using the assumed fire profile, this would correspond to a 22 kW quasi-steady fire size, indicating that the assumed 20 kW fire size is reasonable.
The propane flow rate was controlled by a mass flow controller. However, at test start the burner and the piping between the flow controller and the burner was not filled with propane. Both the burner and the piping were assumed to be air filled prior to ignition. Using the burner construction and the length of piping, the burner volume and the volume of piping between the flow controller and the burner was estimated. A control volume calculation of the piping and burner was done to estimate the volume fraction of propane leaving the burner as fraction of the volume flow based on the mass controller measurement. For the calculation, flow in the pipe was assumed to be plug flow and the gas in the burner was assumed to be perfectly mixed. This was used as the initial guess for the propane fire. The heat release rate (HRR) calculated based on the propane flow rate (Flow Meter) and based on the control volume calculation (Hand Calc) is shown in Fig. 2. The control volume curve shows a delay due to the time to purge the piping of air and reach a flammable mixture in the burner followed by an initially lower HRR due to the mixing time needed to push the remaining initial air volume out of the burner. The FDS Input curve is the final input used for FDS reflecting additional adjustments to the Hand Calc curve. The development of these inputs are discussed in Sect. 3.2.3.
Fig. 2
Measured (flow meter), calculated, and adjusted propane heat release rate
The ventilation system was equipped with CO\(_2\) and temperature sensors in each room, pressure and temperature sensors in the supply and exhaust ducts and in the AHU, as well as airflow sensors in each damper. These were recorded by the BMS and used to control the AHU and damper position. Several 1.5 mm type-K thermocouple arrays were installed in each room and logged by a logging system independent of the BMS. The independent logging system was also used to log the wall temperature in each room, the temperature outside and inside of each damper, the flow in each damper based on bidirectional probes and a Setra267 differential pressure gauge (0-500 Pa), as well as the CO, CO\(_2\) and O\(_2\) concentrations in the supply and exhaust ducts connecting to the classroom. The wall temperature was measured with 1.5 mm encapsulated type-K thermocouples, which were stapled directly to the surface of the gypsum wall to ensure flush contact. The bidirectional probe in the classroom exhaust duct had an issue with the zero setting of the differential pressure measurement, resulting in uncertain differential pressure magnitude post-fire. Figure 3 illustrates the locations and type of instrumentation in the test building.
The uncertainties of the measurements reported by the BMS have not been quantified. However, due to this system’s intended use and very low logging frequency, it is assumed that its uncertainty is higher than the uncertainty of the dedicated logging system installed solely for conducting the fire tests. Uncertainties for the dedicated system are reported as follows based on in-house calibration and manufacturer specifications. The uncertainty associated with the temperature measurements of this system is ±3 °C within the relevant measuring range from 15 to 375 °C. The uncertainty of the room pressure measurements is ±0.1 %. The flow rate of the propane burner was around 0.291 kg/min for most of Test 12, at which point the measurement uncertainty is ±7%. However, in the initial stage of the test, the uncertainty is higher, up to ±15% during the first four minutes where the flow rate was only 0.129 kg/min corresponding to an HRR of approximately 100 kW.
Fig. 3
Illustration of instrumentation inside of the mock-up building (not-to-scale). The supply and exhaust ducts for the classroom and all supply and exhaust valves are illustrated with blue (supply) and yellow (exhaust) dashed lines. Grey dashed lines illustrate the fire source location
Video of the test was collected using four cameras. Their position is also indicated in Fig. 3. One camera was located in the office looking towards the corner opposite the door, one camera was located in the corridor at the side of the classroom looking down the length of the corridor towards the door at the office, and two cameras were located in the classroom looking towards each of the two corridor facing corners of the mattress enclosure. The classroom video was used to determine the relative ignition time of the two fires, assess the fire growth, and determine the time for closing the door to the classroom. The corridor camera was used to determine the time the corridor door to the outside was closed. Synchronization of the independent logging system files to the video used two thermocouples above the propane burner which provided a clear time mark at burner ignition. The BMS files were challenging to synchronize with the independent system given the low data rate (one to two points per minute). BMS data for the classroom fire alarm, classroom return duct temperature, return duct pressures, and return duct flow were used to estimate the synchronization of the BMS data with the video and logging system. However, given the sparsity of time points the uncertainty for the BMS synchronization is estimated as 10 s.
3 Modelling
3.1 Fire Dynamics Simulator
Simulations were performed with Fire Dynamics Simulator (FDS) version 6.9.1. FDS is a CFD program developed for modelling fires. Full details of FDS can be found in its documentation [19, 20] available online. A brief summary of FDS features key to this work (HVAC and leakage) is provided in the following two subsections.
3.1.1 HVAC
FDS has an HVAC submodel for solving flows through HVAC systems. The solver is based on the MELCOR thermal hydraulic solver [21]. The solver creates a map of an HVAC system consisting of ducts and nodes. A node is any location where the HVAC system connects to the FDS domain, to the ambient, or where multiple ducts connect (e.g., a tee). A duct connects two nodes. The solver consists of explicit equations for mass and energy conservation at nodes and an implicit equation for the momentum in ducts (Eq. 2):
where i and k are nodes, j is a duct, L is the duct length in m, K is the duct loss coefficient, p is a node pressure in Pa, \(\Delta p_j\) is a fan source term in Pa, and \(\left( \rho \textit{g} \Delta z \right) _j\) is a buoyancy term. When predicting, rather than specifying flow rates, K is a critical parameter for obtaining the correct steady flow rates.
The HVAC submodel has features for modelling fans, dampers, aircoils, and filters. The airocoil and damper features were used in this work. An aircoil heating or cooling of a duct and is done by adding an energy source term at the downstream node to add or remove the required energy from the duct mass flow. Other features, not used in this work, include the ability to specify filters or fan curves. Details of the aircoil and damper inputs are provided in Sect. 3.2.2.
3.1.2 Leakage
FDS has two approaches to defining leakage which are computed using the FDS HVAC model. Each leakage path is treated as a duct connecting the two sides of the leakage path. The duct area is assigned as \(A_{L,ref}\) from Eq 1, and the duct loss is assigned as \(1/C_d^2\). The pressure exponent, \(n_P\), is also specified as part of the leakage inputs.
The first approach is called the pressure zone leakage approach. This approach is used for the bulk leakage that occurs through the bounding surfaces of a compartment where there isn’t a clearly defined location for the leak. For this approach, \(A_{L,ref}\), \(C_d\), and \(n_P\) are defined for each pair of pressure zones that are connected by leakage paths. Individual surfaces in the domain are then assigned to the leakage paths. Leakage flow is computed using the background pressure in each pressure zone, and the volume flow is assigned as a uniform velocity boundary condition to the surfaces in each pressure zone assigned to that leakage path. This approach does not preserve the enthalpy of the leaking gas. It is assumed that the leakage consists of many small leakage paths where the surface area of material the gas passes through is large enough to return gas temperatures to near ambient conditions as it passes through walls, ceilings, and floors.
The second approach is called the localized leakage approach. This approach is used when there is a known location for the leak to occur, for example, the undercut beneath a closed door. For this approach, a pair of boundary conditions are defined in FDS to denote the two sides of the leakage path, and the pair are applied to one or more grid cells on each side. The two boundary conditions are then connected by a special HVAC leakage component where \(A_{L,ref}\), \(C_d\), and \(n_P\) are defined. Since this approach has a defined location, it uses the local pressure on either side of the leak (background pressure plus the perturbation pressure). This approach can also be applied in an enthalpy conserving manner for leak paths where the flow area is large enough for heat losses to be small. Leak enthalpy conservation was enabled for the simulations.
3.2 Model Implementation
3.2.1 Geometry
A geometry model was created consisting of the classroom, the office, and the corridor. Obstructions representing the HVAC ducts were included in the model to approximate the free air volume occupied by the HVAC system. Figure 4 shows two views of the FDS model. The surfaces are colored based on the leakage path through that surface. In the classroom, the red topped obstruction on the floor is the propane fire, and the orange topped is the PU mattress fire. Details of the leakage are given in Sect. 2.2. The domain was meshed with a 0.1 m mesh using a total of 310,770 grid cells over 12 meshes. A grid study doubled and halved the mesh size to 0.2 m and 0.05 m respectively. Doors were defined as two layers 0.004 m OSB surrounding an 0.038 m air gap. Ducts in the model were treated as adiabatic surfaces.
During the test, the outside air temperature was lower than the initial inside air temperature by \(\approx\)2 °C. To account for this in the wall heat transfer, the backside gas temperature for external facing walls and the ceiling was set to 18 °C. A non-fire simulation was run with a small volumetric heat source present in the domain (\(\approx\)2 \(\hbox {W}\,\hbox {m}^{-2}\)). This source was sufficient to keep the inside air temperature close to the pre-test temperatures. The resulting steady temperature profiles for the exterior walls and ceiling of the three rooms were set as the initial profiles for the fire simulations. Wall heat losses at test start were approximately 4 \(\hbox {W}\,\hbox {m}^{-2}\) and the ceiling losses were approximately 0.5 \(\hbox {W}\,\hbox {m}^{-2}\).
Since there was no recirculation flow and since total exhaust and supply flows for building and the rooms were measured by the BMS, the geometry model focused on the HVAC ducts inside of the three rooms. That is, the AHU was not modeled, and the measured flows were imposed as boundary conditions to the HVAC network. The included portion of the HVAC network was the main supply duct from where it entered the corridor from the AHU to all the endpoints in the corridor and rooms and the exhaust ducts from each room and the corridor until a point before the building total exhaust duct joined with the remainder of building exhaust duct. The FDS supply and exhaust networks are shown in Fig. 5. The building and duct obstructions have been hidden. Duct areas and loss coefficients were taken from the MagicCAD calculation discussed in Sect. 2.3.
Fig. 5
FDS HVAC network layout. Green squares are supply vents, magenta squares are nodes, and blue lines are ducts
Two HVAC flow conditions were used in the FDS simulations. The first specified the supply and exhaust flows directly at the supply and exhaust vents for each compartment (specified room flow). The second specified the total flows in the main supply and exhaust ducts which are the ducts extending just beyond the corridor wall in the lower left corner of each image in Fig. 5 (specified building flow). The two sets of flow rates were defined as time-dependent volume flows using the BMS sytem reported values shown in Fig. 6ba
Fig. 6
Specified HVAC supply and exhaust rates for the building (a) or individual rooms (b)
The RHE was modeled with two approaches. For specified room flow simulations, the measured supply duct temperatures for each room were specified as part of the supply vent boundary conditions. For specified building flow simulations, the FDS aircoil feature was used. Using the BMS measured flow rates, inlet air temperature from outside the building, and the post-RHE temperatures, the heat added to the supply air (treated as if it were the building initial ambient temperature) was determined. This was the specified as a fixed heating rate aircoil in FDS. The temperatures and heating rate are shown in Fig. 7. The FDS model used the inside temperature as the ambient temperature; therefore, at the test start there is a slight negative heating rate due to the outside temperature being cooler than the inside. It isn’t until the fire starts that the exhaust air is warm enough to raise the supply air above the FDS ambient temperature.
Fig. 7
Specified HVAC supply air temperatures for each room (a) or the total building supply air heating rate (b)
As discussed in Sect. 2.4, only the LEO type dampers had measured losses; however, Test 12 had TVE type dampers in the office. Prior to the fire, the BMS data show well balanced flows into and out of each compartment such that no significant pressure differences existed between the rooms or between the rooms and ambient. With leakage flows near zero, and correct loss values, FDS should be able to correctly predict the pre-fire room flows. If the LEO data were used for the TVE damper, FDS predictions of pre-fire room supply and exhaust flows had an RMS error of 30 %.
Another test in the test series, Test 8, did use LEO dampers in the office. The pre-fire BMS data for that test also showed a well balanced system. An FDS simulation for the Test 8 pre-fire flows had an RMS error of 8 % indicating that with good data for all the variable dampers that the fixed duct losses were reasonable.
To generate TVE type damper loss data, it was assumed that the LEO and TVE type dampers had similar shaped loss curves that have a shifted response to the damper position. The TVE curve was assumed to equal the LEO curve when the indicated damper position (0 to 100 % open) is adjusted as (100 − indicated position)/offset. For Test 12, an offset of 1.22 best balanced the room pressures and resulted in RMS errors of the pre-fire flows of 18 %. There was still a mismatch for the classroom and corridor. This is due to a combination of measurement uncertainties in the flow rates, the damper pressure drop measurements, and the damper position indication used to determine the damper loss. It was assumed that this error was solely related to the position indication and that the overall curve was correct given the true position. A position indication error of 2 % resulted an RMS error of 11 % for the pre-fire flows, similar to the errors for Test 8. The final damper position vs. loss coefficient curves are shown in Fig. 8a
Fig. 8
Damper loss coefficients as a function of damper position and time
The sparsity of the BMS data discussed in Sect. 2.6 can be seen in Fig. 6baand Fig. 8b
3.2.3 Fire
Two, fast (e.g., mixed-is-burned), FDS combustion reactions were defined for the simulations. One for the polyurethane fire and one for the propane fire. The polyurethane fire used a fuel molecule of CH\(_{1.12}\)N\(_{0.16}\)O\(_{0.33}\) with a 1.6 % CO yield (based on cone calorimeter testing of the material) and a 10 % soot yield [22] with a 26,600 \(\hbox {kJ}\,\hbox {kg}^{-1}\) heat of combustion (based on cone calorimeter testing of the material). The propane fire used a 0.5 % CO yield and a 2.4 % soot yield [22] with the heat of combustion calculated by FDS using the FDS predefined heats of formation for propane, oxygen, and the products.
A simulation was run using the growth rate estimated based on the burner volume calculation from Sect. 2.5. Early in the fire scenario, when the bulk gas in the room is still relatively cool, heat losses to the walls will be a small fraction of the convective heat release rate of the fire. From the FDS Validation Guide [23], it is known that FDS makes very accurate predictions of gas temperatures for known fires. This means any discrepancy in the early bulk temperature rise can be largely attributed to errors in the assumed initial growth rate. The FDS predicted vs. measured room temperatures of the upper two thermocouples were compared after the end of the initial rapid temperature transient in the compartment, approximately 70 s after ignition of the PU. The relative difference in the predicted vs. measured temperatures were used to estimate how much just using the control volume calculation over-estimated the heat release rate, i.e., a 10 % overprediction in temperature would imply a 10 % error in the total heat release.
Video of the test showed a fire that took a short period of time post-ignition to spread across the surface of the gas burner and for the flame height to reach a quasi-steady size. Using the video to estimate fire diameter and flame height suggested a fire growing in an approximate t\(^2\) manner in the moments immediately following ignition. A t\(^2\) growth was then fit starting at the ignition time in the video (which was delayed from the estimated time in the control volume calculation). The control volume calculation fire size at increasing times was used as the second point for the fit until a time was found that gave the estimated total heat release based on the predicted temperatures. A second simulation was performed with the adjusted curve. This still showed an excessive rise in temperature during the initial transient, and the early fire size peak was decreased by 15 % during the growth period which resulted in a better match to the early transient in temperature. The fire size as measured by the propane flow rate, estimated by the control volume calculation, and as finally input to FDS is shown in Fig. 2.
3.3 Simulations Performed
Three sets of simulations were performed:
1.
Classroom only: Only the interior of the classroom was modeled. The HVAC network was not modeled; however, duct obstructions were kept. Leakage through the classroom wall and door to the corridor was reassigned as leakage to the ambient. The measured supply and exhaust flows to the classroom were applied as time-dependent volume flows at the supply and exhaust vent locations. The measured supply flow for the entire classroom was evenly divided over the three supply vents. Three grid resolutions were run 20 cm, 10 cm, and 5 cm. This variant served as a grid study and as a means to evaluate the self-consistency of the inputs for flow, fire, and leakage without the added complexity of predicting flows in the HVAC network. Additionally, this simulation setup was used to examine potential sources of input errors. In plots, this simulation is labeled as "Cl".
2.
Whole building, specified room flow: The domain was as shown in Fig. 4 at a 10 cm grid resolution. The HVAC network was not modeled, and the supply and exhaust flows were directly specified at the supply and exhaust vent locations in each room. The duct obstructions were kept. By specifying flows for all three rooms, this simulation represents a modest increase in complexity from the single room simulation as the model must now account for the leakage flow between rooms. In plots, this simulation is labeled as "All".
3.
Whole building, specified building flow: The prior simulation with the HVAC network inside the building modeled. The total building supply and exhaust flow rates were specified where the main supply and exhaust ducts enter the building. This simulation used the adjusted and shifted damper curve. In plots, this simulation is labeled as "Bldg".
4 Results and Discussion
The following subsections discuss the modelling results for the classroom only, specified room flow, and specified building flow simulations. The first two subsections show classroom only results for the grid study and for a study on input sensitivity. The remaining subsections provide results for individual predicted quantities for all three simulation variants at a 10 cm resolution.
4.1 Grid Sensitivity
Fig. 9 shows the pressure, O\(_2\), and Tree 1 temperature predictions for the classroom only simulation for the three grid sizes. Results for the 10 and 5 cm grids are very similar. Notable differences are seen for the 20 cm grid for O\(_2\) and temperature. Based on this, the 10 cm grid was selected for the comparisons in the subsections that follow.
Figure 9ashows the FDS predicted pressure in the classroom vs. the measured pressure for the classroom only simulation. FDS overpredicted the first pressure peak by 210 Pa, the second peak by 100 Pa, and the remaining pressure history by 50 to 100 Pa. The reported FDS error in pressure is 19 % [23]. A 210 Pa pressure difference could result from any one of the causes below or a proportional combination of more than one:
A mass excess of approximately 0.4 kg in the room at the time of the peak plus additional mass excess due to higher pressure meaning leakage flows would remove more mass. Sources for this error would be errors in the measured supply and exhaust flows plus errors in synchronizing the BMS system time.
A net heat release enthalpy error of 51 kJ during the rise to peak. This is 19 % of the energy release of just the polyurethane fire or 4 % of the total energy release for both fires plus additional error due to the higher leakage flows removing enthalpy. There is a significant uncertainty in the early fire size which drives the first pressure peak.
A 55 % error in the leakage area. It is unlikely that leakage errors are this high; however, leakage is likely a contributing factor.
Additional sources of error for the pressure rise are the wall convective heat losses and the radiant fraction of the fire. During the first pressure peak, bulk room temperatures have not risen significantly and temperatures where the plume impinges on the ceiling are also not very high. Convective heat loss is probably not a significant contributor. A higher radiant fraction would radiate more the walls during the early fire growth which would reduce the enthalpy of the gas. Most of the heat release rate is in the propane fire and the FDS assumed radiant fraction for propane of 29 % is not likely a significant error.
To investigate these potential error sources the simulation was rerun varying the various error sources one at a time. Results for the classroom pressure are shown in Fig. 10aand 10b
The supply flow was reduced by 10 % (vol flow). This dropped the first error by 28 % with no significant impact on the long term trend.
The timing of the supply and exhaust flows were shifted by 10 s earlier (time-) and later (time+). This respectively dropped the first peak error by 6.2 % and 9.5 % with no significant impact on the long term trend.
The propane heat release rate was reduced by 10 % for the entire fire duration (HRR). This is slightly less than the potential 15 % error during the start of the fire and slightly more than the 7 % error during the end of the fire as reported in the prior discussion on uncertainty in Sect. 2.6. This reduced the first peak error by 29 % with no significant impact on the long term trend.
The leakage area was increased by 50 % (leak). This reduced the first peak error by 53 % and resulted in a good match for the long term trend.
The radiant fraction for the propane fire was increased from the default value of 29 % to 40 % (\(\chi _r\)). This reduced the first peak error by 22 % with no significant impact on the long term trend.
The heat transfer coefficient for the room surfaces was set to 20 \(\hbox {W}\,\hbox {m}^{-1}\,\hbox {k}^{-1}\) (htc). This reduced the first peak error by 43 %. Improved agreement was seen following the first peak but the long term trend did not change significantly.
These are all relatively large variations in the parameters and none results in matching the first pressure peak. Only leakage showed a significant impact on the long term trends. It is likely some combination of error sources plus predictive error in FDS.
Fig. 10
Sensitivity study results for classroom pressure for the classroom only simulations
Pressure predictions for the three variants are shown in Fig. 11. In the plot, FDS pressures have been filtered using a 5 s moving average. The unfiltered peak pressures for the first peak are over predicted by 210 Pa for the classroom only and specified room flow and and 62 Pa for the specified building flow. Since the measured data are at 1 Hz with only a handful of measurement points over the peak, the actual test peak pressure may have been significantly higher but not recorded. After the first peak, the classroom only and specified room flow simulations show significant over predictions in pressure for the classroom of 100 and 50 Pa for the second and third peaks. The specified room flow flow simulation also overpredicts the office and corridor pressures with longer term errors of 30 and 20 Pa respectively. For the specified building flow simulation the pressure over prediction in the office and corridor is twice that for the specified room flow simulation. One or more of the errors identified in Sect. 4.2 likely contribute to this predictive error.
Fig. 11
Pressures for the building specified flow simulation
Flow rates for the rooms are only predicted for the specified building flow model. The predicted flow rates are shown in Fig. 12. Leakage flow rates are shown in Fig. 13 where a positive flow is from the first compartment listed in the legend to the second. In the plot, FDS flow rates have been filtered using a 5 s moving average. The supply flows in each compartment match closely to the data especially in the classroom and corridor. At the time the fire room door closes (39 s) there is a spike seen in the classroom and office flows that is not seen in the data; however, the spacing of BMS data are too large for that spike to have been captured. The classroom exhaust prediction is larger than the BMS measurements and the reverse is true for the office and corridor. This would account for the changes in pressure. If more flow is being exhausted from the classroom than being supplied, that must be made up by leakage flow into the classroom which would take higher corridor pressures. Since the corridor is coupled to the office via leakage, higher corridor pressures would act to increase the office room pressure as well.
Substantial leakage volume flows are seen out of the classroom. The duct supply and exhaust flows do not differ by as large a value; however, the exhaust flow is significantly hotter meaning that it is significantly less mass flow. This means achieving equilibrium in the classroom mass requires the mass excess to leave via leakage. Much of this mass leaks to the outside but some leaks into the corridor. Without a fire in either the corridor or office their pressure remains similar to one another and leakage flows between those compartments are near zero.
Figure 14 shows the O\(_2\), CO\(_2\), and CO concentrations for the classroom exhaust duct. For specified room flow simulations ("Cl" and "All"), the FDS results are measured as the average value over the plane of grid cells adjacent to the exhaust vent. For the specified building flow simulation ("Bldg") the FDS results are from inside the exhaust duct. The test data are measured inside the duct. Transport delay times to the gas analyzers were estimated by comparing the time the return duct bi-directional temperature probe first showed change to the times the CO\(_2\) concentration first showed a change (the CO concentration is not as strong a signal and the instrument used for O\(_2\) only started logging after a drop occurred which is why the initial drop is a step function).
Results are essentially identical for the three simulations. O\(_2\) and CO\(_2\) predictions closely follow the test data. After the initial transient, from 100 s to 200 s, the FDS CO\(_2\) predictions are 10 % low. After 500 s, the predictions are 5 % high. O\(_2\) shows similar but opposite trends. A combination of errors in the fire size combined with errors in the classroom ventilation rate adding up to approximately the prediction error would account this predictive error. With the classroom pressure over predicted in all models and the class room exhaust over predicted in the specified building flow model, this suggests some error in the fire size as too
The CO concentrations are under predicted by a factor of five. As this error is not seen in the CO\(_2\) predictions, it is entirely due to the assumed CO yield for the propane burner and the polyurethane fire. Minor product yields can vary with scale and the specific burning conditions. A small portion of the error may result from too low of a heat release for the polyurethane fire; however, not more than 10 % given the measured mass loss vs. the assumed PU fire. This error, therefore, is largely due to the effective combined yield of the two fuels being much higher than assumed in the inputs with the PU likely being the major contributor due to its higher CO yield.
Figure 15 shows the temperatures measured at the two thermocouple trees in the classroom. The early rapid temperature rise predicted by FDS matches well with the measured values. This is a period of time with little surface convective heat transfer due to low temperatures throughout most of the compartment and with the HVAC system not yet having a significant response to the fire. Matching temperatures well under those conditions indicates that the initial assumed growth rate of the fire reasonably matches that of the test. Had the FDS growth rate resulted in significantly too much or too little energy release, that would have been reflected as errors in the temperature rise during the early growth. At approximately 40 s, the HVAC system begins responding to the fire. This is also near the end of the initial growth of the propane fire.
Up through 700 s, the FDS temperature rise values are within 7 % of the measured values for the upper two thermocouple locations (10 and 35 cm below the ceiling) with temperatures trending below the measured value. Both elevations are above the discharge elevation of the supply vents whose ducts turn down a short distance from the ceiling before terminating. After 700 s, predictions worsen slightly with the maximum errors reach 13 % of the measured values. FDS over predicts the temperature for the lower two thermocouple locations for the duration of the simulation with a larger magnitude of over prediction at the 100 cm below the ceiling location (up to 21 °C) than the 200 cm location (up to 13 °C). The percentage errors are 10 to 15 % at test end for all four lower thermocouple locations. Unlike the upper two locations, the bias for the lower two is established early on in the simulation and remains fairly constant. Since the lower two locations represent a much larger portion of the room volume, overall FDS is over predicting the energy content of the room. This is consistent with the earlier observations of over predicted pressure since pressure is a measure of the energy content of a volume. It is also consistent with under predicted O\(_2\) and over predicted CO\(_2\) suggesting to much retention of combustion products (potentially due to a positive bias in the input fire size).
Fig. 15
Change in classroom thermocouple tree temperatures, number in legend indicates distance below ceiling in cm
Temperatures in the office, Fig. 16, reasonably match the data. The 10 cm location prediction is within 1 °C of the measurement for the specified room flow and 2 °C for the specified building flow. The 35 cm location has larger errors of 3 °C and 4 °C respectively. However, the measured data for the 35 cm location show a temperature drop not seen in the other data indicating a potential issue with the thermocouple. The 100 cm location is over predicted by 1 to 2 °C, and the 200 cm location is under predicted by 1 °C. Some of the error is likely due to the FDS model not resolving the full details of the diffuser disk at the bottom of the supply vents. The overall performance of the temperatures combined with the good agreement on office supply and exhaust for the specified building flow simulation indicates the approach for handling the RHE was reasonable as that is the only source of heating for the office (there was near zero leakage exchange with the corridor).
Fig. 16
Change in office thermocouple tree temperatures, number in legend is distance below ceiling in cm
In the corridor, Fig. 17, FDS is under predicting the 10 cm thermocouple by 3 to 20 °C and the 200 cm thermocouple by 1 to 2 °C with the middle two being over predicted by 1 to 8 °C. The bare metal return duct from the classroom runs along the ceiling of the corridor. The temperatures inside the duct match the 10 cm to 35 cm temperatures in classroom. With over 10 m of 0.315 m diameter duct, the heat transfer from the duct to the corridor would be of similar magnitude to the heat addition from the RHE to the corridor. Some of the error for the 10 cm location is likely due to duct heating. Some error may be due to radiative heating of the thermocouples by the classroom exhaust; however, that is likely a small effect. Another potential error source is at 10 cm resolution the louvered end of the corridor duct is not resolved with the likely result of less of a downward component to the mixing in the FDS model.
The 10 cm thermocouple on corridor tree 3 shows a brief temperature spike when the classroom door was closed. FDS is predicting a brief flow reversal in the corridor exhaust duct, i.e, the classroom pressure increased enough to briefly push combustion products through the corridor exhaust. This does not appear to have happened in the experiment and was a consequence of the over prediction of the peak pressure in the classroom.
Fig. 17
Change in corridor thermocouple tree temperatures, number in legend is distance below ceiling in cm
The last set of simulations specified flows in the main ducts and the damper positions over time. This was still not a fully predictive simulation. A fully predictive simulation would model the BMS logic for adjusting the fan flow rate and the damper positions based on measurements made in the duct network. This was not possible in this work as the details of the proprietary controls for the system were not known. FDS, however, has the capability to perform this type of simulation if the details of the control system were known. There are two approaches that could be used for a fully predictive simulation.
The first approach would be to use the built-in FDS control system inputs. These inputs include basic math functions (add, subtract, multiply, sin, cos, etc.), basic logic functions (and, or, etc.), and a proportional-integral-derivative (PID) controller function. Complex behavior can be modeled by using multiple control functions together in series or parallel; the control function inputs can operate as a rudimentary programming language. The value of the fan flow rate and the damper loss could be tied to outputs of sequences of FDS control functions. The advantage of this approach would be a self-contained simulation. The disadvantage of this approach would be the likely substantial effort to convert the BMS control system logic to FDS control system inputs.
The second approach would be to utilize a recently added FDS capability for external control of the simulation. In this approach FDS would write outputs for all the system parameters monitored by the manufacturer’s control system. A second piece of software running the BMS control logic would monitor the FDS outputs and write a file containing the fan flow rates and damper losses. FDS would update its values for fan flow and damper losses using that file. An advantage of this approach is that it would likely be easier to either write develop the second piece of software using a standard programming language or toolkit (like Matlab) than to utilize the FDS control logic inputs. A disadvantage would be there would be a delay between FDS writing new system data and reading new system settings; however, FDS time step sizes are small (\(\sim\)0.01 s for these simulations) compared to the response time of a large blower or a motor controlled damper. A parallel calculation running slightly behind the FDS simulation would likely not have a significant impact on the results. The simulations in this paper had approximately 30,000 grid cells per FDS process, and took approximately one minute of wall clock time per second of simulation time. A parallel calculation for the control system should easily be able to closely keep up with the FDS calculation.
6 Conclusions
FDS was used to model a fire in a three room experimental facility representing a school and its HVAC system. While modelling of multiple room facilities plus HVAC have been the subject of prior work, this facility had the novel feature that its HVAC system was a intelligent system that dynamically adjusted flows to compartments to optimize energy usage, air quality, and comfort. These experiments provided a unique ability to test the ability using the HVAC capabilities of FDS to model a dynamic HVAC system.
6.1 Modelling Challenges
Modelling had to overcome two challenges with regards to the boundary conditions for the model. These were the fire size at test start and the time synchronization of the HVAC system data to video and test logger data.
The fire source was a mix of a PU mattress and a propane gas burner. The PU fire size was not measured and had to be derived via video and end-of-test mass loss measurements. While the propane burner had a flow meter, at test start the burner emits an time-dependent composition of propane and air before becoming pure propane. The time dependence of the fire size required estimation via analysis of the burner construction, video of the fire, and model iterations. This added significant uncertainty to the early fire size, which is also coincident with the time of the major response of the HVAC system to the developing fire conditions. For future experiments where a defined fire is desired, use of a sootier gas for the fuel such as ethylene or propylene would remove the solid fuel uncertainty. Additionally, characterizing the burner under a hood to better characterize the early fire growth rate and measure CO and soot yields for that specific burner would further reduce the input uncertainty for the fire size.
The BMS system recorded data at sparse time intervals, and it was not maintained synchronized in time with either the video or the logging system clocks. The long intervals, 30 s or more, between BMS system data points meant it was difficult to precisely synchronize the BMS system data to the ignition time. Additionally, the highly dynamic behavior predicted by FDS at the start of the fires only had a handful of BMS system outputs for generating the FDS inputs of valve position and flow rate. A recommendation for future experiments of this sort would be to ensure time synchronization and have the BMS system output data at 1 Hz rather than 0.03 Hz or less.
6.2 FDS Performance
Three simulation variants were run. The first was only the classroom with the supply flow rate, the supply temperature increase due to the RHE, and the exhaust flow rate specified based on the BMS measurements for flow and the logger measurements for temperature in the respective supply air DCVs. The second had all three rooms but still specified each room’s supply and exhaust flow and supply temperature. The third specified the total building flow in the main supply and exhaust ducts, the total supply air heating provided by the RHE, and the DCV time dependent flow losses based on valve position. This variant let the FDS HVAC model partition the total flow over the three compartments.
In all three variants, the temperature and major species concentrations in the classroom are well predicted by FDS, with temperature errors of 10 to 15 %. This is not a surprising result. The propane fire is two orders of magnitude larger than the PU fire, and its size is well known other than the initial transient. There was approximately one air change in the classroom over the simulation period, which limits the impact of flow rate predictions on gas concentrations and temperature. Under these conditions, existing FDS validation would suggest good performance by FDS. The exception was the CO concentration. The actual yield of the PU fire was not known and value from a well ventilated cone calorimeter test was used for the input. Similarly, the propane fire used a handbook value. The specific ventilation and burningconditions for the two fireswere a contributing factor to the uncertainty associated with the CO yield.
The office only saw heating due to the RHE heating of the supply air. This heating was specified as a boundary condition. This is effectively equivalent to FDS validation examples for a known fire in a compartment where good performance is expected. FDS predictions in the office had larger percentage errors in the temperature rise; however, in absolute terms, predictions were within 2 °C. In the corridor, FDS does not predict as large an overall temperature rise as seen in the experimental data; however, at this time the FDS HVAC model does not have an easy method for heat transfer between a duct and its surrounding environment. Based on the classroom upper layer temperatures and the length of duct in the corridor, heating of the upper corridor from the exhaust duct would have been similar to heating from the RHE. In practice, a typical office, school, or residential occupancy would often have ducts concealed from the occupied spaces (for example behind ceilings). However, industrial-type occupancies often see exposed ducts. This points to an area of development needed for FDS.
In the predicted simulation, flows were generally well predicted. The classroom exhaust was approximately 10 % high. The resulting shift in compartment pressures caused the corridor exhaust to be about one-third low (the classroom had a much higher total flow rate than the corridor). The office, which was more isolated from the direct effects of the fire room than the corridor, showed excellent agreement for the predicted flows. For future experiments, it would be useful to exercise the building HVAC system without a fire to provide a more extensive data set for validating model inputs. For example, running the building HVAC system at different total flow rates for different damper positions while measuring system flows and room pressures.
Pressure predictions were not as good as the flow, temperature, and species predictions. The early pressure peak was not well predicted by any simulation variant; however, as discussed, there is significant uncertainty in the early fire size and the exact timing and magnitude of the BMS response. Later in the fire test, when the BMS system had reached a quasi-steady state, pressure predictions were better than the initial peak; however, pressures in the office and corridor were two to three times those measured. For the final predicted flow simulation, the late simulation classroom pressures were within 10 % of the measured pressures. Since the pressure overprediction occurs in all variants, including the small room with specified flows, this is indicative of likely errors in either leakage or the BMS measurements. If both of those boundary conditions were precise, then FDS should have been able to capture the simple control volume problem for the office. As with the flow predictions, additional non-fire data for varying building flows and damper position would help to identify the likely source of the error.
As previously noted, these predictions were based on the damper positions over time and the flow rates being pre-defined. For future experiments, obtaining the details of the logic used to control the damper positions and fan flow rates would allow for a complete simulation.
The overall conclusion from this study is that FDS can be used to predict flows in HVAC systems with dynamic dampers and a time-varying fire. These predictions; however, rely upon detailed knowledge of the HVAC system design for obtaining fitting loss data and fan flow data and detailed knowledge of the effect of damper position on loss. These could be challenging to obtain early in a design. Specified flow simulations provided similar results to predicted flow simulation. This indicates FDS could be applied as a design tool to assess flow requirements in response to a fire to inform the HVAC system design and logic for the automated controls.
Acknowledgements
Funding for RISE Fire Research came from the project "BRAVENT - Efficient smoke ventilation of small fires", which is funded by the Research Council of Norway, grant no. 321099 and its project partners. Funding for the Fire Safety Research Institute came from internal funding.
Declarations
Conflict of interest
The authors have no financial or non-financial Conflict of interest in this work.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Absar Alam M, Kumar R, Yadav AS, Arya RK, Singh VP (2023) Recent developments trends in HVAC (heating, ventilation, and air-conditioning) systems: a comprehensive review. Mater Today Proceed. https://doi.org/10.1016/j.matpr.2023.01.357CrossRef
2.
Mads M, Peter GS, Axel C (2014) é: demand-controlled ventilation - requirements and commissioning. Technical Report ISBN 978-82-536-1414-4
Hostikka S, Janardhan RK, Riaz U, Sikanen T (2017) Fire-induced pressure and smoke spreading in mechanically ventilated buildings with air-tight envelopes. Fire Saf J 91:380–388. https://doi.org/10.1016/j.firesaf.2017.04.006CrossRef
6.
Pretrel H, Suard S, Audouin L (2016) Experimental and numerical study of low frequency oscillatory behaviour of a large-scale hydrocarbon pool fire in a mechanically ventilated compartment. Fire Saf J 83:38–53. https://doi.org/10.1016/j.firesaf.2016.04.001CrossRef
7.
Beji T, Hidalgo JP, Fateh T, Floyd J, Prétrel H, Hamins A (2024) Compartment fires: challenges for fire modeling as a tool for a safe design (IAFSS workshop, April 2021). Fire SafJ 144:104109. https://doi.org/10.1016/j.firesaf.2024.104109CrossRef
Chaudhari DM, Weinschenk C, Floyd JE (2023) Numerical simulations of gas burner experiments in a residential structure with HVAC system. Fire Technol. https://doi.org/10.1007/s10694-023-01390-yCrossRef
10.
Ghanekar S, Weinschenk C, Horn GP, Stakes K, Kesler RM, Lee T (2022) Effects of HVAC on combustion-gas transport in residential structures. Fire Saf J 128:103534. https://doi.org/10.1016/j.firesaf.2022.103534CrossRef
Wahlqvist J, Van Hees P (2013) Validation of FDS for large-scale well-confined mechanically ventilated fire scenarios with emphasis on predicting ventilation system behavior. Fire Saf J 62:102–114. https://doi.org/10.1016/j.firesaf.2013.07.007CrossRef
SINTEF (2023) SINTEF Byggforskserien 520.352 Brann- og røyksikring av ventilasjonsanlegg. Technical report
15.
Meraner C, Fjærestad JS, Haukø A-M (2024) On the performance of damper-optimised demand control ventilation systems during a fire. To be submitted
16.
Fjærestad JS, Yang A, Meraner C, Dovran F, Olsen J (2024) BRAVENT - Storskala branntester (del 1): Brannytelse for ikke-brannklassifiserte ventilasjonskomponenter. Technical Report RISE-rapport 2024:37, RISE Fire Research
17.
Meraner C, Fjærestad JS (2024) BRAVENT - Storskala branntester (del 2): Effekt av ventilasjonsstrategi på røykspredning og trykkontroll i en mock-up skolebygning. Technical Report RISE-rapport 2024:38, RISE Fire Research
18.
ISO 9972:2015 Thermal performance of buildings - Determination of air permeability of buildings - Fan pressurization method. International Organization for Standardization (ISO) (2015)
19.
McGrattan KB, Hostikka S, Floyd J, McDermott R, Vanella M, Mueller E (2024) Fire dynamics simulator technical reference guide volume 1: mathematical model. NIST special publication 1018-1 NIST SP 1018, National Institute of Standards and Technology, Gaithersburg, MD. https://doi.org/10.6028/NIST.SP.1018
20.
McGrattan KB, Forney GP, Hostikka S, Floyd J, McDermott R, Vanella M, Mueller E(2004) Fire dynamics simulator user’s guide. Technical Report NIST SP 1019, National Institute of Standards and Technology, Gaithersburg, MD. https://doi.org/10.6028/NIST.SP.1019
21.
Gaunt R (2000) MELCOR Computer code manuals: reference manuals version 1.8.5, Volume 2, Rev. 2. NUREG/CR-6119. Technical report, US Nuclear Regulatory Commission, Washington, DC
22.
Hurley MJ, Gottuk D, Hall JR, Harada K, Kuligowski E, Puchovsky M, Torero J, Watts JM, Wieczorek C (eds) (2016) SFPE handbook of fire protection engineering. Springer, New York
23.
McGrattan KB, Randall M, Vanella Marcos, Mueller Eric, Hostikka Simo, Floyd Jason (2024) Fire dynamics simulator technical reference guide volume 3: validation. Technical Report NIST special publication 1018-3, National Institute of Standards and Technology, Gaithersburg, MD. https://doi.org/10.6028/NIST.SP.1018
