Numerical Model Development of the Air Temperature Variation in a Room Set on Fire for Different Ventilation Scenarios

Abstract

Statistics show that most fires occur in civil residential buildings. Most casualties are due to the inhalation of hot air loaded with smoke, leading to intoxication with substances harmful to the human body. This research aimed to develop a CFD model that relates the operation of the sprinkler system to the operation of the ventilation system through the air temperature in a specific point close to the sprinkler position. A real-scale experiment was carried out, and a CDF model was developed. Several parameters of the CFD model (thermal conductivity of the experimental test room walls, numerical grid elements’ dimensions, burner heat release rate variation) were imposed to the model, so that the resulting entire time variation of the temperature next to the sprinkler location corresponds to the real measured variation. Two other experiments were used to validate the numerical model. Besides the air temperature, at this point, other essential parameters were determined in the entire experimental space: indoor air temperature, visibility, oxygen concentration, and carbon dioxide concentration. We found that if the ventilation rate increases, the indoor temperatures in that specific point decrease, and the sprinkler is activated later or, in some cases, it might never be activated. However, this conclusion is not valid for the entire analyzed space, as the ventilation system alongside the natural air movement imposes specific air speed and specific temperature distribution inside the analyzed space.

1. Introduction

In Romania, in the first six months of 2013, 6227 fires occurred in a population of about 20 million inhabitants. Most fires occurred in civil residential buildings and outbuildings. About 280 victims were reported, including 23 children; among these victims, 189 were injured and 91 died [1]. Therefore, fire protection is one of the most important aspects during the early stage of a building design. To make a building safe, in the event of a fire, the main requirement is the capability to save the occupants [2]: maintaining optimal conditions for evacuating people is a priority. This can be achieved by ensuring low indoor temperatures, a good visibility, high oxygen concentrations, and low carbon dioxide concentrations, resulting from combustion, for a specified period. Smoke and hot gas poisoning are currently the main cause of injury or death to the users whenever a fire starts in a building. A good example of when these conditions were not met was on the night of 30 October 2015, when a strong fire occurred inside a private club in Bucharest, Romania. The fire was caused by the fireworks used during a concert, which caused the ignition of combustible materials. A total of 64 people died because of this accident, as well as the high temperatures, smoke poisoning, toxic combustion gases, high levels of carbon monoxide, and carbon dioxide. More than 100 people have fallen victim but survived with very serious injuries [3]. In this case, the amount of pollutants increases considerably, even up to 50 times, when the space that the fire develops is not ventilated (smoldering occurs). We presented these statistics to underline that fire protection is one of the most important aspects during the early stage of a building design.

The smoke-controlled ventilation and sprinklers represent two of the main active systems for fire protection that should be carefully analyzed during the building design stage. The main aim of smoke-control ventilation is to extract hot and toxic gas out of the building that caught fire, in order to allow occupants to evacuate the enclosure before untenable conditions are reached. The purpose of a sprinkler system is to control, contain, and extinguish a fire, as well as to prevent the smoke from being generated. Hot smoke and toxic gases, however, are responsible for more deaths in fires than direct exposure to flames. In order to ensure human fire protection, the correct design of these systems is of the upmost importance, and it should fire modeling take into account as an important part of building design.

There are currently three main categories of methods for modelling the reaction to fire of buildings [4]:

• simplified mathematical models for the description of the evolution of certain parameters describing the fires, such as the temperature-time logarithmic curve in EN 1991-1-2 [4];
• zonal models who divide the burnt space into two zones, each zone being characterized by an average temperature: the warmer upper zone and the colder lower zone [5,6]; and
• computational fluid dynamics (CFD) models that allow the simulation of the fire closer to the reality. CFD models are based on the finite volumes numerical method, involving the division of the model space into a fine grid of three-dimensional cells. For each cell, the physical and chemical parameters are known at a given time, by solving a particular form of the Navier—Stokes equations [7,8,9].

The concern for the fire safety of buildings, as an engineering science, is relatively new, and it is developed in two main directions: experimental and numerical studies (or a combination of the two). The purpose is to understand fire evolution and find solutions to increase the safety of buildings under the probable action of fire. In the early phases of the research, experimental studies were predominantly used in the absence of high-performance computers, which would have made it possible to use CFD methods. For example, Guillaume et al. [10] conducted several experiments at the real scale of a building by setting an apartment on fire. During these tests, data have been recorded with thermocouples, heat flux meters, opacimeters, and gas analyzers [10]. Tong [11] investigated two hybrid ventilations: natural ventilation through a roof opening combined with mechanical suction and mechanical jet flow. An experimental model was reproduced as 1/10 of the prototype [11]. Among the few examples of real-scale experimental investigations, we could cite the study of Jiang et al. [12], who conducted several real-scale experiments aimed at burning a car on the street or inside a parking lot.

Recently, numerical studies tended to dominate the literature. Kadlic and Magdolenova [13] have analyzed the impact of input parameters in the computer modeling of fires, using them as an example a tunnel. Zhisheng [14] performed a numerical study to see the effect of different ventilation rates inside a tunnel. Cheng [15] have also conducted a numerical study to analyze the effect of fire on the structural elements of a building. There are also studies on the influence of the ventilation systems on indoor parameters, such as temperature, oxygen concentration, or visibility during a fire [16]. Lu et al. [17] performed a numerical study on the effect of the ventilation system during a fire in an underground parking lot, and they concluded that an impulse ventilation system can prohibit smoke spreading and maintain an acceptable visibility. Merci and Shipp [18] conducted an experimental study on the control of hot air and smoke inside a closed parking lot, and they established that the exact position of the extraction fans is not essential when they are not close to the fire source. However, there are very little experimental studies evolution [19] carried out upon indoor environment during fire evolution; consequently, there is a lack of CFD modelling of indoor spaces under fire conditions.

However, several other research were carried out on the effect of room ventilation during fire. Huang et al. [20] conducted a numerical study using FDS to track the distribution of smoke in a 4 × 7 × 10 (m³) burning room in different possible scenarios. The main parameters that were followed were the visibility and speed of the smoke distribution, in relation to the fire source. Lai et al. [21] studied, using FDS, the distribution of smoke in a burning room that had natural ventilation. The influence of natural ventilation on the smoke layer was investigated by measuring the distribution of the indoor temperature in time and space. Refay et al. [22] studied the effect of the geometry of vents air lets, upon the hot air exhaust and cold air compensation, using the CFD method with a k-ε model for describing the turbulence. Enright [16] performed a numerical study on the impact of the ventilation system on the triggering of sprinklers. Li et al. [23] found that the ventilation rate has only small influence upon the temperature variation during fire scenarios. Other research also found that the use of mechanical devices for ventilating spaces during a fire has both advantages (e.g., better visibility and low amounts of pollutants) and disadvantages (e.g., high oxygen concentrations maintaining combustion, as well as lower temperatures delaying the start-up of sprinklers) [24]. We note here that there is little research carried out on the way these two installations (ventilation and sprinkler installations) influence each other in fire scenarios.

The purpose of this study is to experimentally develop a CFD model that can be used to simulate the evolution of the fire indoors, as well as to analyze the relationship between the operation of the sprinkler system and ventilation system through the air temperature in a specific point close to the sprinkler position. The CFD model will also be used to understand the evolution of the indoor environment field parameters during a survival time of 15 min, based on rules laid down by specific technical standards [25]. The aim is to experimentally develop such a CFD model for a real indoor geometry. Further investigation targeted the adaptability of the model for different ventilation air flows indoors. This study has great potential, because such a CFD model might become a very important tool for an appropriate design of the HVAC systems and their control strategies during fire evolution.

The paper presents: the experiments, CFD modelling, CFD model validation for different ventilation airflows, and results.

2. Experiments

2.1. Introduction

The purpose of the experimental part is to provide suitable data for the numerical modelling of both the development and testing. The purpose of the experiments was to calibrate a numerical model developed for the solver Fire Dynamics Simulator (FDS) and then validate it.

The experiment took place under laboratory-controlled conditions, in order to match the indoor conditions of a real building indoor space. The analyzed space is a real-scale enclosure, as recommended in fire safety research literature [22]. The area of the considered enclosure is A = 18.00 m². The experimental was carried out in the courtyard of the Building Services Engineering Faculty at the Technical University of Civil Engineering of Bucharest. A real-scale experiment is usually preferred, being characterized by better precision, compared to a reduced scale model. Our experimental study involved a real application, with complex parameters, which evolution could not be represented at reduced scale. In our case, for a different scale, it would have been difficult to find the appropriate similitude criteria for the simultaneous involved phenomena: air flows, fire reactions, and sprinkler activation.

The experimental test chamber was equipped with burners, air inlet and outlet, ventilation system, liquified petroleum gas (LPG) ducts, thermocouples and data acquisition system, video survey cameras, a sensor for the unburnt gas leaks detection, and CO alarm systems.

Three experiments were performed, the only parameter that differed being the air flow, extracted by means of the ventilation system. Further in the paper, the following notations will be used for the three experiments: CALIB1 (used for numerical model calibration), CALIB2, and CALIB3 (both used for numerical model validation and testing).

2.2. Experimental Set-Up

A small room is reproduced, by means of a precast container (dimensions 3.00 × 6.00 × 2.70 m³) (Figure 1a). The walls of the container are made of assembled sandwich panels, containing 10 cm of non-combustible mineral wool [23]. The room features on the long side a fire-resistant door EI90-C (dimensions of 0.90 × 2.10 m²) and short side a fire-resistant window E30 (dimensions 0.60 × 1.20 m²). Two holes were executed in the container walls: (1) one square hole above the window (dimensions 0.25 × 0.25 m²) (Figure 1d) for the air outlet for the smoke extraction fan (Figure 1c) and (2) another rectangular hole on the opposite short side of the container (dimensions 0.4 × 0.1 m²) for the air inlet.

2.3. Burner Conditions

In this experiment, we used the LPG gas and a tank of 60 (l) tank as fuel. This fuel type presents the following advantages: (1) known fuel consumption during the experiment (measurable tank mass by means of a scale) and (2) relative constant gas insertion over time (due to the pressure regulator). The use of liquid or solid fuels would have made it almost impossible to replicate the experiment under similar conditions for model testing.

Two LPG gas burners were used to simulate the fire (Figure 1b). The fuel consumption was determined by measuring the weight of the LPG tank at the beginning and end of the experiment period. We opted for this method, instead of a continuous measurement of the LPG mass, because the mass difference between two consecutive measurements is smaller than the measurements error. Any possible error can be further corrected during the model development.

2.4. Exhaust System and Conditions

The extraction of the indoor polluted air is performed using the mechanical ventilation system (Figure 1c). A centrifuge fan, PR-Q 456T AT, was used for extracting smoke and hot air (placed above the experimental stand). The fan can be used for temperatures up to 200 °C, to ensure extracted air flows between 0 m³/h and 1500 m³/h. The air intake is ensured through a permanently open gap in the facade, i.e., naturally organized ventilation (similar to building façade windows joints). Therefore, we expect the intake of air flow is equal to the exhaust air flow, and their value is known and imposed by the fan voltage regulator system.

During the first experiment (CALIB1), the fan was off and the air extraction outlet was blocked. During the other two experiments (CALIB2 and CALIB3), the fan was on, operating at different ventilation airflows. The corresponding values of the air speed in the outlet hole, v_a (m/s), and the airflow, Qv (m³/s), are presented in

Table 1. Average air exhaust speeds and extraction rates.

Case	va	Qv	Qv/(100 m2)
	(m/s)	(m3/s)	(m3/s)/(100 m2)
CALIB1	0.0	0.0	0.0
CALIB2	1.0	0.060	0.3
CALIB3	4.5	0.270	1.6

. The ventilation airflows were chosen to analyze the applicability of the model for several types of buildings. The design norms in fire safety in Romania [26] require a hot air extraction air flow higher than 1.0 m³/s per 100 m² of fan-protected space. Therefore, the model developed during experiment, CALIB1, will be tested for both small airflows (CALIB2—corresponding to general apartment buildings without ventilation systems) and large airflows (CALIB3—corresponding to buildings with ventilation systems for the smoke and hot gases exhaust).

2.5. Sensors

The main parameter recorded during the experiments was the indoor air temperature. Several K thermocouples were used; this thermocouple type is the most used sensors in experimental fire studies today [27]. The measurement error of the temperature sensors, 0.1 °C, is acceptable, given the range of working temperatures in our field of research. The calibration of the sensors was carried out using a hot water calibration bath with known temperatures. The data were recording using an ALMEMO710 data logger. The time step for the measurements was established at 5 s. This time step was considered relevant for our study (the duration of the fire evolution was monitored during about 760 s). We also observed, during preliminary experimental tests, that the temperature variation below this time step was generally less than 1 °C.

Two of the temperature sensors were arranged outside the experimental stand to record the outdoor temperature (the experiments were conducted at an outdoor temperature of 20 °C), while the rest of the thermocouples were arranged inside the experimental booth in two orthogonal planes: one horizontal (at the top of the booth) and one vertical (arranged longitudinally through the middle of the booth) (Figure 2).

3. Modelling, Parameter Adjustment, and Results

3.1. Introduction

The purpose of this paragraph is to present the numerical model useful for the fire simulation propagated in indoor environment, in order to understand the temperature evolution in time. The purpose of our study is to understand the relationship between two active systems for fire protection, the ventilation and sprinkler systems, and how the operation of the ventilation system could influence the efficiency of the sprinkler system. These two systems influence one another, by means of the air temperature around the sprinkler position. This is the why we chose temperature as the calibration parameter at a point located close to the sprinkler position (point S24). Besides this main parameter, other criteria are also used for the modelling of this phenomenon. The model is determined on the measured data CALIB1 and tested on the experimental data from CALIB2 and CALIB3.

The Fire Dynamics Simulator (FDS) solver was used for modelling of fire dynamics and the temperature evolution indoors. FDS is a practical software, commonly used in fire modeling [28,29]. This software used an approximate form of Navier–Stokes equations, appropriate for low Mach. The large eddy simulation (LES) technique was adopted, in order to handle the sub-grid scale convective motion. FDS requires the geometry of the room, calculation time, choice of the grid density, characteristics of the material for the walls, amount of heat introduced over time, and combustion characteristics, such as the amount of smoke produced, extracted air flow, and so on.

During the experiment, the locations of the thermocouples (Section 2.5 and Figure 2) are represented by means of the two planes inside the experimental booth. During the numerical simulation, we determined the temperature field all over the simulated domain. The calibration of the numerical model was carried out according to the temperature value of point S24 (Figure 2). This point represents the position of the 24th experimental thermocouple, and it is located at the intersection of the two planes (horizontal and longitudinal). Moreover, this temperature sensor presents the advantage of being in an approximately median area of the container, in the immediate vicinity of a sprinkler. Sensors in the flame area, hot air outlet, cold air inlet, and immediate vicinity of the envelope have been avoided.

The real-time modeling solution was adopted, which, as small-scale fire modeling, was not sufficiently studied [30]. The calibration of the numerical model is the process by which the input parameters of the numerical model are determined, so that the numerical model results would be close to those obtained experimentally. The calibration of the model was performed knowing the temperature distribution during the CALIB1 experiment. The order of the model development phases were as follows: the (1) geometry was defined in FDS, (2) boundary conditions and the initial conditions were setup, (3) grid cell dimension of the modelling domain was analyzed, (4) ventilation exhaust setup was introduced in the model, (5) thermal conductivity of the experimental test chamber was calibrated, and, finally, (6) HRRPUA in the model was calibrated, in order to suit the real HRR variation that occurred due to the pressure dropping in the LPG tank. The HRRPUA parameter was left at the end of the calibration process; in order to avoid that, this correction would cover the significant correction of the thermal conductivity of the experimental test chamber.

3.2. Geometric Model

The envelope of the room was considered uniform in the model from the point of view of the surface geometry and material characteristics (Figure 2). Elements, such as the door, window, or other details, such as the joining method of the panels, doorknob, etc., were not taken into consideration in the numerical model. The interior dimensions of the room in the model were those determined by direct measurements on the experimental booth: 3.00 × 6.00 × 2.70 m³. The envelope has a similar thickness to that of the experimental stand: 0.10 m. The dimensions of the hot air exhaust outlet were determined by direct measurements: 0.25 × 0.25 m². The dimensions of the cold air input grids were also determined by direct measurements: 0.40 × 0.10 m².

The analysis domain was extended by 0.50 m, outside the outer limit of the experimental booth walls.

The thermocouples were mounted inside the container: 15 thermocouples in the vertical longitudinal plane, 15 thermocouples in the horizontal upper side plane (red crosses in Figure 2), and two thermocouples on the walls of the container.

3.3. Boundary Conditions and Initial Conditions

The initial and boundary conditions are described below, some being implicitly set by the FDS software, others being set by the user.

The following initial conditions, applied for CALIB1 case:

• the initial temperature at all points in the range was considered to be equal to +20 °C;
• the initial air speed at all points in the range was equal to 0 m/s;
• the initial oxygen concentration was equal to 0.208 mol_O2/mol_air;
• the initial carbon dioxide concentration was equal to 0.000387 mol_CO2/mol_air;
• the initial visibility, in all points, was equal to 30 m.

The following boundary conditions, applied during the CALIB1 case:

• the air temperature at the burner surface was imposed, depending on the amount of heat with a linear evolution of the HRPPUA;
• air speed in contact with the inner face of the booth walls is equal to 0 m/s;
• air speed at the air outlet interior surface was equal to 0 m/s;
• material characteristics of the envelope do not change during the 760 s period.

The definition of the numerical model reunites the following characteristics: the geometrical shape of the analyzed air and walls domains, burner modelling (by means of its main characteristics (dimensions, location, the SQ, and HRRPUA variation)), walls thermal conductivity, fan operation, mesh dimension, boundary conditions, and initial conditions.

3.4. Choosing the Grid Density

The choice of the numerical model grid density was made to balance the precision advantage with necessary computation time. We studied several mesh types (from coarser meshes (cell dimensions 0.25 m × 0.25 m × 0.25 m) to fine meshes (0.05 m × 0.05 m × 0.05 m), and we aimed to achieve good results with short modeling times. The temperature variation profiles over time were compared to the experimental temperature variation profile for each grid’s dimensions (Figure 3).

The temperature accuracy for the entire time interval (i = 0 ÷ 760 s) is indicated by a parameter regrouping the temperature prediction error (Equation (1)) at all-time steps during the entire monitoring interval. We used the standard deviation error (Equation (2)) to characterize the numerical model precision for the entire time interval. For each grid’s dimensions, the calculation time and the standard error were estimated.

Table 2. Characteristics of the model, according to grid dimension.

Grid Dimension (m × m × m)	Number of Cells	Calculation Time (h)	σErr °C
0.25 m × 0.25 m × 0.25 m	7888	0.5 h	21.0
0.20 m × 0.20 m × 0.20 m	15,120	1 h	19.8
0.15 m × 0,15 m × 0.15 m	34,944	2 h	12.8
0.10 m × 0.10 m × 0.10 m	117,936	7 h	2.1
0.05 m × 0.05 m × 0.05 m	943,488	120 h	1.8

reunites the main characteristics of each grid type. One can notices for the course meshes the calculation time is very short (about 30 min), but the error is too large considering the studied phenomenon (about 21 °C). On the contrary, the fine meshes (cell dimension 0.1 m and 0.05 m) present smaller errors but increased calculation time. We considered the grid with cell dimension 0.1 m is the best option, based on the two criteria.

$$\mathrm{Err}={\mathsf{\theta }}_{\mathrm{i}}^{\mathrm{model}}-{\mathsf{\theta }}_{\mathrm{i}}^{\mathrm{experiment}} °\mathrm{C}$$

(1)

$${\mathsf{\sigma }}_{\mathrm{Err}}=\sqrt{\frac{\sum {\left({\mathsf{\theta }}_{\mathrm{i}}^{\mathrm{model}}-{\mathsf{\theta }}_{\mathrm{i}}^{\mathrm{experiment}}\right)}^2}{\mathrm{N}-1}} °\mathrm{C}$$

(2)

3.5. Fan Modeling

The position and dimensions of the outlet air and inlet air were determined by direct measurements. Measurements of the velocity were made during the experiments using an anemometer. We determined the velocity in the points from a grid (25 × 25); then, we calculated the average air exhaust velocity and the mass flow rate through the outlet. We further imposed this mass flow rate as a boundary condition in the FDS outlet model.

3.6. Modelling the Material Characteristics of the Envelope

The characteristics of the experimental room envelope are important for a correct consideration of the temperature evolution over time. Hence, their correct assessment is essential for good and reliable model results. The main characteristics of the materials, which are relevant to the present study, are the: walls material density, the walls specific heat, and thermal conductivity. The specific density and heat used the values from the data sheet of the facade panels: ρ = 120 kg/m³ and c_p = 0.840 kJ/kg/K.

The distribution of temperatures inside the booth walls and heat loss through the walls depends on their thermal conductivity [31], and its influence is further transferred through convection to the indoor air temperature. It was necessary to determine an equivalent thermal conductivity λ_eqv W/m/K of the entire booth envelope to match the temperature profile of the numerical model to the experimentally obtained profile. Generally, suppliers provide determined thermal conductivity, under standard conditions, for panels (field thermal conductivity). However, it is necessary to correct this value, taken from technical data sheets, due to the existence of doors and windows on the real booth that are not present in the model. The equivalent thermal conductivity must also consider the existence of thermal bridges, such as the joints of the panels, intersection with the door, the windows, corners of the container, in between metal structure of the panels, and so on.

Five different values of the equivalent thermal conductivity were analyzed in the FDS model: 0.024 W/m/K, 0.044 W/m/K, 0.061 W/m/K, 0.078 W/m/K, and 0.098 W/m/K. The temperature evolution profiles in point S24 corresponding to the five different equivalent thermal conductivities. Figure 4 presents similar values for the beginning of the monitoring period (0–200 s) and differ considerably beyond this time interval.

It can be observed that for λ_eqv = 0.061 W/m/K, the overlap of the numerical prediction of the temperature evolution (violet curve in Figure 4) over the experimental profile (black curve in Figure 4) is the closest. The model accuracy is indicated by a parameter regrouping the temperature prediction error (Equation (1)) at all-time steps during the entire monitoring interval. We used the error standard deviation (Equation (2)) to characterize the numerical model precision for the entire time interval (i = 0 ÷ 760 s).

The variation of the errors, corresponding to the five equivalent thermal conductivities (Figure 5a), are characterized by different standard deviations (Figure 5b). The minimum value of the error standard deviation is σ_Err = 1.5 (°C), corresponding to λ_eqv = 0.061 W/m/K.

3.7. Burner Modeling

For the modeling of the burner, three aspects have to be considered:

• Geometrical aspects: burner dimensions and its location inside the analyzed domain;
• the total amount of heat introduced into the booth by burning the fuel;
• the heat release rate per unit area, HRRPUA (kW/m²) evolution over time [30,32,33].

Burner dimensions and location was set in the experiments with the dimensions 0.25 × 0.60 m², and were located at 0.3 m from the floor and 0.3 m from the outside wall on the longitudinal plane.

The FDS software allows for two ways of modelling the heat source: either by introducing the mass of fuel consumed over time, Mi (kg), or introducing the heat emitted by the heat source over time by the burner, S_q (MJ). In order to avoid the error that might be introduce the type of LPG used in the model, we preferred to use the heat source, instead of the fuel mass. The total amount of heat introduced into the analyzed domain is equal to the real experiment heat release of the burner, S_q (MJ), and was determined as a function of the mass of fuel consumed during the experiment, M_i (kg), multiplied by the lower calorific value of the LPG gas, Q_i (MJ/kg), (Equation (3)) [30]. Thus, the average value of the HRRPUA was evaluated using Equation (3), depending on the measured mass of the gas.

$$\mathrm{HRRPUA}=\frac{Sq\left(\mathrm{MJ}\right)}{S_{burner}\left({\mathrm{m}}^2\right) · T\left(\mathrm{s}\right)}=\frac{Mi\left(\mathrm{kg}\right) · Qi\left(\frac{\mathrm{MJ}}{\mathrm{kg}}\right)}{S_{burner}\left({\mathrm{m}}^2\right) · T\left(\mathrm{s}\right)}=\frac{0.315\left(\mathrm{kg}\right) · 38\left(\frac{\mathrm{MWs}}{\mathrm{kg}}\right)}{0.15\left({\mathrm{m}}^2\right) · 760\left(\mathrm{s}\right)}=105\left(\frac{\mathrm{kW}}{{\mathrm{m}}^2}\right)$$

(3)

The profile of the temperature evolution over time is largely influenced by HRRPUA variation over time. In order to establish the correct method of introducing HRRPUA in the FDS, we want to match the profile variation of the temperature over time, obtained from the numerical model to the experimentally obtained profile. Several cases of HRRPUA variation were analyzed in FDS (for each case the total amount of energy released is the same, Sq = 0.315 kg · 38 MJ/kg = 11.97 MJ = 3.325 kWh): (1) one case with constant HRRPUA value, (2) one case with linear increase of HRRPUA, and (3) three cases with linear decrease of HRRPUA (different slopes of variation).

It can be noticed that, for the increasing HRRPUA, we obtained a temperature profile with values at the beginning of the time monitoring interval that was smaller than the experimentally measured ones and higher towards the end of the monitoring interval (Figure 6). On the contrary, for the decreasing HRRPUA, we obtained a temperature profile with values at the beginning of the time monitoring interval that were higher than the experimentally measured ones and smaller towards the end of the monitoring interval. Our numerical findings are in good accord with the phenomenological aspects.

It can be observed (Figure 6) that the closest temperature variation, obtained from the model, to the experimentally measured one is obtained for the HRRPUA introduction with a decreasing linear function, with HRRPUA = 105 (kW/m²) and the FDS HRRPUA variation factors F_{t = 0 s} = 1.15 (−) (corresponding to HRRPUA = 1.15·105 = 120.8 kW/m²) and F_{t = 760 s} = 0.85 (−) (corresponding to HRRPUA = 0.85·105 = 89.3 kW/m²). This observation is correct, from a phenomenological viewpoint; as fuel is consumed, the pressure in the LPG tank decreases and, despite the existence of the pressure regulator, the amount of fuel introduced to the burner slightly decreases over time.

All these characteristics were established for the CALIB1 ventilation operation type and further tested for CALIB2 and CALIB3. From the phenomenological viewpoint this numerical model, established for CALIB1, might be appropriate for other ventilation strategies, characterized by different airflows, because the numerical model has the same geometry, type of walls, thermal bridges of the walls, boundary, and initial conditions.

In the next paragraph, we shall analyze, in more detail, whether this model can be used for other ventilation operation case, thus, if the model is characterized by a larger applicability.

4. The Test of the Model for Different Ventilation Airflows

The aim of this paragraph is to understand the applicability of the numerical model, i.e., whether it is limited to the ventilation type used in CALIB1 or can be further employed for other ventilation types.

From the experimental viewpoint, two other experiments were carried out: (1) CALIB2 corresponding to small airflow rates (apartment buildings) and (2) CALIB3 corresponding to high ventilation rates (smoke and gases evacuation by means of ventilation system). The air speed was experimentally determined for all cases in the section of the air outlet in over 600 points, by means of an anemometer. The air speed profiles (Figure 7) were used to determine the average air speed, v_a, in the air outlet section and airflow, Qv. Higher ventilation rates correspond to higher airflow rates.

From the numerical simulation viewpoint, the same numerical model was used. The only parameter changed between the calibrated model (CALIB1) and current test cases is the ventilation rate. The experimentally measured values were introduced in the numerical model, as in the case of the inlet, by using a mass flowrate boundary condition at the inlet of the computational domain.

Both the experiments and the numerical studies lead to the variation of the temperature in control point S24 over time. The temperature variation profiles, predicted by the numerical model, were compared to the experimentally determined profiles. The numerical model is appropriate if the error between the numerically predicted and measured temperature is acceptable for our field of research.

We observe that the temperature variation profiles present have a logarithmic increase and tend to stabilize towards the end of the analyzed time period. A similar pattern is observed for all ventilation cases (Figure 8).

For CALIB1 vs. MODEL1 (green curves in Figure 8), the graph shows that temperatures tend to stabilize at about 100 °C. That point is the value at which the amount of heat introduced by burning the fuel becomes equal to that lost by the system. The two temperature profiles are almost superposed, denoting an excellent match of the numerical prediction over the experimentally measured values.

For CALIB2 vs. MODEL2 (yellow curves in Figure 8), the temperature profile, obtained by means of the numerical simulation, is almost superposed over the experimentally measured temperature profile. The standard deviation of the temperature prediction error is σ_Err = 2.6 °C. This value that is considered appropriate, being within the acceptable range for our phenomenon. This temperature profile tends to stabilize at about 90 °C.

For CALIB3 vs. MODEL3 (blue curves in Figure 8), we can see that the two-temperature profile almost overlapped. The standard deviation of the temperature prediction error is σ_Err = 1.7 °C, the smallest value of all three cases. This value is also considered appropriate, being within the acceptable range for our phenomenon. We can observe the temperature curve is stabilizing at about 50 °C.

From a phenomenological point of view, the higher the ventilation rate, the faster the indoor hot air is evacuated and replaced with cooler outdoor air, thus maintaining a cooler air indoor. Moreover, one can also expect that a higher airflow might influence the indoor air dynamics. Thus, the hot air might be drawn towards the center of the room (in the direct line between the air inlet and outlet) and less towards the upper side, leading to the same conclusion. In this study, both the experiments and the numerical models present the same conclusion: for null airflow, the temperature raises up the 100 °C, while for the higher airflow, the temperature raises to only 50 °C. This conclusion confirms the phenomenological conjecture and further presents a size order for the temperature in the control point.

The very small errors between the numerically predicted and experimentally determined temperature profiles represent the main conclusion of this study. We developed a numerical model that can be used to predict the temperature variation indoors, during a fire scenario. The large applicability of the model is given firstly by its geometrical shape (dimensions like common indoor rooms), similar to both rooms, as well as the evacuation hallway and escape lane, in case of fire scenarios. The applicability of the numerical model is further amplified, due to its precision for different ventilation strategies, adequate for different building destinations. It can be concluded that the established model is valid for other ventilation cases and can be successfully used, in order to study the effect of different ventilation control scenarios over the fire dynamics and indoor environment conditions.

It can be assumed that the model is highly suitable for values of the average velocity of the exhaust air, between 0 m/s and 4.5 m/s, with all other parameters remaining unchanged.

5. Numerical Results

The FDS numerical model was developed, adjusted, and tested for the other two ventilation scenarios. In the following, we will present a few of the results of the modelling. The FDS numerical model allows us to track other parameters that could not be determined experimentally: oxygen concentration in the air, carbon dioxide concentration in the air, and visibility. These parameters, together with the internal temperatures, are indispensable for ensuring survival of the people during fires [31]. In the following, we shall present the variation of these parameters for the three ventilation scenarios.

5.1. Indoor Temperatures

In all simulations, the indoor air temperature starts at 20 (°C) and increases logarithmically (Figure 9). It is observed that the indoor temperatures depend on the ventilation flow, with lower temperatures obtained at higher flow rates. Although the amount of fuel is not high, temperatures rise rapidly when ventilation does not work (CALIB1), approaching 100 °C. In CALIB 2, temperatures exceed 80 °C during the analyzed period. In CALIB1 and CALIB2, in less than three minutes, the temperatures exceed 60 °C, and the interior conditions are no longer suitable for survival and evacuation of the people.

We can observe that ventilation rates that are too low do not significantly improve the indoor conditions (CALIB2). At high ventilation rates (CALIB3), the air temperature does not exceed 60 °C, maintaining good conditions for people evacuation. We conclude that mechanical ventilation affects indoor temperatures. Moreover, the mechanical ventilation extracts polluted air and brings fresh air with higher oxygen concentration, maintaining good conditions for people evacuation. However, the negative effect of the high oxygen concentration is that it sustains the fire.

The current regulations [26] establish a correlation between the hot exhaust air flow and surface of the fan-protected space. The indoor air temperature represents the result of the heat gains (heat release from the fire) and heat loss (through air exhaust and through walls). Therefore, another useful conclusion might be to improve the current norms, so that the exhaust air flow should also depend on the fire heat release or the combustible materials in the room.

5.2. Visibility

The visibility is another parameter resulted from the FDS numerical model simulation. The visibility is an important parameter for the people evacuation in case of fire scenarios. A visibility higher than 10 m is perceived as very good visibility, and visibility higher than 6 m is suitable for evacuation [34]. The visibility in the FDS depends not only on the amount of fuel burned but also on the type of the burned material (amount of smoke resulting from the complete combustion depends on material type). In all three cases, we considered in FDS that SOOT_YIELD = 0.05, meaning that 5% of the amount of combustible material is transformed into smoke particles. We used this percentage as an average value, corresponding to a fire in a civil building.

The time variation of the visibility for the three ventilation scenarios (Figure 10) shows that the higher the exhaust air flow, the better the visibility, because the hot air filled with smoke particles is partially exhausted.

For CALIB3, visibility is good during the entire study time interval (0 ÷ 760 s), stabilizing at about 6 m. On the contrary, the low ventilation rates (CALIB2) or no ventilation rates (CALIB1) lead to improper evacuation conditions in less than one minutes.

The FDS numerical simulations present values for all parameters inside the entire analyzed domain. These parameters are used to observe the variation of the temperature and visibility along the entire longitudinal plane of the analyzed space (Figure 11). We note that the increase of the ventilation rate simultaneously leads to a general smaller temperature and general, slightly better visibility inside the analyzed space.

5.3. Oxygen Concentration

Oxygen is an indispensable element for human survival during fire scenarios. Its concentration represents another result of the FDS simulation. As reference oxygen concentration levels, we evoke that for concentrations below 0.140 molO2/molair, the combustion ceases completely; for concentrations below 0.180 molO2/molair, the people evacuation conditions are believed to be affected [35,36]. We also recall the double effect of the oxygen: high concentration is very good for people’s health and their evacuation, but it also sustains the combustion. The higher the oxygen concentration, the more intense the combustion and resulting heat release inside the analyzed space, therefore leading to higher indoor temperatures.

We observe that high ventilation rates lead to higher oxygen concentrations, C_O2 (Figure 12a). It can be observed that, as time passes, the oxygen concentration decreases in the cases of CALIB1 and CALIB2 (the fresh air introduced does not represent a significant oxygen source and, consequently, the combustion consumes the oxygen inside the room). In the case of CALIB3, the oxygen concentration decrease is smaller, and it reaches its even out value inside the 760 s time interval (this situation corresponds to the case when the oxygen introduced is equal to that consumed during combustion).

In all three cases, the oxygen concentration remains high during the 760 s time interval and does not affect the combustion. In our case, the amount of fuel burned is low, therefore resulting in low oxygen consumption.

5.4. Concentration of Carbon Dioxide

Even though CO₂ is not perceived as a pollutant, it causes poor reactions in humans (high concentrations lead to dizziness and drowsiness). The concentration of carbon dioxide, C_CO2, is another result of the FDS numerical model simulation, and its indoor concentration is a result of the burning chemical reaction. The main source of CO₂ is indoors; therefore, one would expect its concentration to rise during the burning process and to be diminished as the ventilation intensifies.

It could be observed that the concentration of CO₂ is inversely proportional to the concentration of O₂ (Figure 12b); this result was expected, since CO₂ results from the combustion of O₂. For the high ventilation scenario (CALIB3), the CO₂ concentration evens out at about 0.2%. For intermediate ventilation rate (CALIB2), the CO₂ concentration evens out at about 0.8%, while for CALIB1 the CO₂ concentration exceeds 1.2%. The inverse correlation between the oxygen and carbon dioxide can be observed in the entire longitudinal plane (Figure 13). As the ventilation flow rises, we observe a decrease in oxygen concentration and increase in carbon dioxide concentration.

From the three considered cases, it could be observed that different rates of ventilation significantly affect the conditions of survival and evacuation of humans. It is found that better ventilation led to lower temperatures, increased visibility, increased oxygen concentration, and decreased carbon dioxide concentration.

6. Conclusions

The generality of the presented results always depends on the variation of different parameters. In this study, we focused only on the variation of the temperature in a specific point in space, close to the sprinkler position. We must clearly point out that our study was not intended to obtain a general conclusion but to allow for the validation of a verified protocol that can be used for the further construction of more complex models.

During this research, the influence of mechanical ventilation on the survival conditions of the users of a building was analyzed. This was achieved by observing four parameters: indoor temperature, visibility, oxygen concentration, and carbon dioxide concentration. The real phenomena, in such an environment, and corresponding modelling allow for the observation of these inside the considered space. However, the purpose of our study was to understand the relationship between two active systems for fire protection: the ventilation and sprinkler systems, i.e., how the operation of one system could influence the efficiency of the other system. These two systems influence one another by means of the air temperature around the sprinkler position. This is why we chose the calibration parameter to be the temperature in a point located close to the sprinkler position. The current paper presents the first step of a larger approach involving the evolution of the above-mentioned parameters, and this first step was the experimental and CFD modelling, focused on the two systems influence. We have also to point out that, in our quest dedicated to obtaining accurate models to be used in fire safety engineering, we did not find literature studies that evidenced the influence of the two types of active protection systems.

In the present study, three distinctive cases were analyzed: CALIB1- no air extraction, CALIB2- average exhaust speed at the outlet of 1.0 m/s, CALIB3- average exhaust speed at the outlet of 4.5 m/s.

By comparing the three situations, the following conclusions can be drawn:

• If the ventilation rate increases, then the indoor temperatures decreases. For the first two cases (CALIB1, CALIB2), up to t = 155 s temperatures go higher than 60 °C and no longer fall below this value. In the case of good ventilation (CALIB3), throughout the analysis period, the temperature does not exceed this value, the conditions being specific to survival (it has the disadvantage that it would not have triggered the sprinklers, if they had existed).
• If the ventilation rate increases, then the visibility increases. In case of poor ventilation (CALIB1, CALIB2), the visibility is maintained above 10 m, for a maximum of 35 s, after which it decreases below 2 m until the end of the study. In case of good ventilation (CALIB3), the visibility is maintained above 10 m to 125 s but remains above 5 m until the end of the study.
• If the ventilation rate increases, then the O₂ concentration increases. In all cases, the values remain within reasonable limits, but a decrease in oxygen concentration of 9.6% is identified for CALIB1, 6.2% for CALIB2, and only 1.9% for CALIB3.
• If the ventilation rate increases, then the CO₂ concentration decreases. In all three cases, the values remain within reasonable limits, but an increase in the concentration with carbon dioxide up to 1.1% is identified for CALIB1, 0.7% for CALIB2, and only 0.2% for CALIB3 (4.7-times higher than CALIB1).

Other studies, regarding the effect of the ventilation upon the temperature indoors, concluded that the ventilation has a small influence [23], while in our study we concluded (both experimentally and numerically) that the temperature varies considerably, due to the ventilation rate (see Figure 9 and Figure 11). Further, the ventilation has the potential to influence the activation of the sprinkler system. The differences are due to both the parameter used for the model adjustment and the value of the ventilation rate.

It is noted that the existence and flow of the ventilation system are very important for maintaining good conditions for users. However, the question arises: where should we stop with the ventilation rate, so that other fire-fighting systems, such as sprinklers, can be triggered?

For other studies, it would be interesting to establish a correlation between the fan hot air extraction flow and thermal load in space, i.e., the total amount of fuel.