Abstract
Fire simulations are becoming more and more widely used in fire protection practices. In order to achieve more accurate results, it is inevitable that simulations are always developed. This study investigates the fire behavior of various complex wooden geometries. This research aims to enhance the understanding of fire propagation of different geometries made of wood. The simulations are performed using fire dynamics simulator, which incorporates heat transfer, combustion, and fluid dynamics principles. Key parameters like temperature and heat release rate are analyzed for each of tree geometries. The research contributes to the development of more accurate fire models. It also provides the basis for further development of simulations including more complex geometries.
1 Introduction
Fire modeling and simulation play a key role in fire protection [1], enabling a deeper understanding of fire dynamics and effects. In this study, the fire simulation of wooden objects with different complex geometries using Fire Dynamics Simulator (FDS) is investigated [2]. The aim of the study is to prepare the simulation environment for the simulation of more complex practical tasks like an automobile fire simulation with the help of the presented research results. FDS is a Computational Fluid Dynamics (CFD) model of fire-driven fluid flow [3, 4]. It is specifically developed for simulating fires capable of modeling fire propagation, heat transfer, and the movement of smoke and gases in detail.
Nowadays different approaches to carry out the fire simulation of complex geometries were made. In [5] FDS was used to predict the flow patterns around a person in a ventilated room. Human bodies of different complexity models were simulated. All of the models used box-like obstacles. The simulations were compared to measurements found in the available literature. It was observed that the most accurate simulation took the combined effects of obstacles and heat sources into account. In [6] the CFD simulation of automobile fires using FDS is presented. Three simple cases of typical automobile fires were investigated with simulation and fire experiments: automobile engine compartment fire, automobile passenger and compartment fire, and fire spread from a burning car to another car. The car geometry was created from blocks of obstacles. In [7] the fire spread on different geometry facades of buildings with wooden claddings was investigated. It was found that the facade geometry can greatly influence the behavior of fire and its propagation. The horizontal projection could be used for deflecting the flame. Window size had also a great effect on the fire spread on the facade. The combination of horizontal projections and sloped surfaces could minimize the fire risk. For modeling PyroSim, for numerical solution FDS, and for visualizing the results Smokeview code was used. In [8] the fire simulation of different curved geometries like a curved wall, a cylindrical column, a ramp, and a hemisphere are presented using PyroSim code, FDS, and Revit Architecture. Five methods of meshing were tested. The different methods were compared with temperature, Heat Release Rate (HRR), computational storage, computational time, ease of meshing, prone to numerical instability and curved surface meshing quality. A single mesh was found the most suitable in regards of numerical stability and the achieved results.
It is visible that only some of the scientific literature deals with the simulation of the complex geometry wooden objects. Most of the research deals with the numerical simulation of wooden pallets expanded with fire experiments [9–12].
In this research, different wooden models with varying geometric complexity were simulated. Wood was selected as material as it is widely used in fire experiments, therefore the simulation can easily be validated [13, 14] later, and all of the properties were known for the simulation. The goal is to understand how the geometry of wooden objects affects fire propagation and related heat transfer processes.
The presented results can contribute to the further development of fire protection strategies and risk reduction measures and provide a strong base for further simulation developments.
2 Materials and methods
Complex geometry can be constructed using Blender code, which has an extension to FDS [15]. Blender geometry can be imported to FDS in 2 ways: using a GEOMetry (GEOM), which follows the complex geometry accurately or using an OBSTruction (OBST), in which case the geometry is constructed from small cubes according to the mesh [16]. In this paper, two studies are carried out. In the first case, different simple objects were simulated. The aim of this study was to investigate the differences between certain objects with the same volume and to explore the possibilities of Blender FDS. In this case, all of the objects were modeled as GEOM. In the second study there was a more complex geometry, which was based on the bodywork of a car. In this study, the differences between the two models (GEOM and OBST) were examined regarding simulation results and simulation time. This study is the basis for more complex simulation tasks, like the simulation of a complete car. The material was wood (pine) in all cases with the following properties: density: 600 kg m−3; specific heat: 1.76 kJ kg−1·K−1; conductivity: 0.12 Wm−1·K−1; reference temperature: 260 °C, heat of reaction: 6,500 kJ kg−1; heat of combustion: 16,750 kJ kg−1 [17, 18]. It was assumed that all of the objects were created from the same block of wood. The objects were ignited with 1,000 °C temperature sparks in all cases. In this research therefore the material-based approach was selected, in which the properties of the fire depend fully on the material [2]. For visualizing the results Smokeview was selected.
3 Results and discussion
3.1 Simulating simple geometries
The task was to simulate the fire spread of objects with different geometry, like a cube, a cylinder, a cone, and a sphere. The volume of each object was 1 m2. The size of the room was 2 × 2 × 2 m and the mesh size was 0.05 × 0.05 × 0.05 m. The room was totally closed. The ignitor consisted of 64 particles at 1,000 °C, which were placed 0 and 0.1 m high. The object was placed in the bottom of the room reaching coordinates (0, 0, 0) with their top. The objects were created using Blender and were then exported into FDS. The objects are shown in Fig. 1.
Simple geometries in FDS (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
For the comparison of the fire cases, the temperature was measured at 4 points, which are located in the middle (0, 0, 0), at the top in the middle (0, 0, 0.8), at the top in the corner (0.8, 0.8, 0.8) and at the bottom in the corner (0.8, 0.8, −0.8). The Heat Release Rate (HRR) value was also measured. The simulation time was set to 600 s. The results are shown in Figs 2–5. In the case of sensors 2 and 3 the signal was noisy, so the moving average was also shown.
Temperature versus time diagram in the case of the first sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of a) the second sensor and b) the moving average (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of a) the third sensor and b) the moving average (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of the fourth sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
It can be stated that the temperature roses fast at the top of the geometry in all cases. The temperature was the highest in the case of the cube and the first cylinder and the curve was also similar in this case. The reason behind this is that a large area was ignited. The temperature first increased, then decreased then became constant. In the case of the sphere the temperature first increased, then decreased then increased again. The explanation for it can be that it has a curved surface. A similar phenomenon could be observed in the case of the second cylinder. However, this simulation did not run till the end because of numerical instability. In the case of the cone, the temperature increased, then became constant, then increased slowly. The explanation for it is that there was a small surface at the top of the object, which caught fire. Except for the cone, the temperature increased above 1,000 °C in all cases. The highest temperature could be observed in the case of the first cylinder.
It can be seen that the temperature increased above 700–800 °C at the top sensor in all cases except the cone, where it was above 400 °C. The average temperature was the highest in the case of the cube. In this case, the temperature first increased and then decreased to 600 °C. In the case of the first cylinder, the temperature increased above 800 °C then decreased, then oscillated. The average temperature was smaller than in the case of the cylinder, but the maximum temperature was higher. Similarly, the temperature increased above 700 °C in the case of the second cylinder, then started to decrease. In the case of the cone the temperature increased above 400 °C, then oscillated around a constant value. In case of the sphere the temperature rose above 700 °C. It first increased then decreased then started to oscillate.
In the case of the top corner sensor, the temperature rose around 250–300 °C in the case of every objects. The temperature curves were also similar. They first increased then started to oscillate around a constant value. The highest temperature was reached in the case of the sphere and the highest average temperature could be observed in this case too.
In the case of the fourth sensor the temperature did not rise above 50 °C for a long time, then at around 250 s a fast increase in temperature could be observed in the case of the cube. The maximum temperature was almost 200 °C in this case. After a fast temperature increase, it decreased below 50 °C. The explanation for this can be that hot gases reached the sensor at that time. A similar phenomenon could be observed in the case of the first cylinder and the sphere. In these cases, the temperature rise was not so high, but lasted longer. In the case of the cylinder, the temperature rise started only at 320 s and the temperature decrease was also smaller. In the case of the cone, the temperature rise was continuous. At the end of the simulation, the temperature was the highest in the case of the sphere. The highest average temperature could be observed in the case of the sphere (almost 60 °C).
The HRR versus time diagram is shown in Fig. 6.
HRR values versus time diagrams (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
It can be seen that the HRR value increased fast above 30 kW in all cases except the cone. The highest HRR could be observed in the case of the cube. In all cases except the cone, the HRR first increased, then decreased, then increased again, then finally decreased, when the fire was extinct because of the lack of O2. The time of the different phases was different. In the case of the cylinders, the phases happened faster, which was followed by the cube and then finally the sphere.
It can be concluded that the highest temperature and HRR could be observed in the case of the cube as it had the largest flat area. In the case of the first cylinder, the results were similar to the cube as it also had a large flat area. The next largest results could be reached in the case of the sphere. It also has a large volume, but it is curved, therefore fire spreads differently. It also had a time delay compared to the cube and the cylinder because of this property. The HRR and the temperature in the case of the cone was the lowest. The explanation for this is that there was only a slight area near the ignition source. In the case of the second cylinder numerical instability occurred. The cause of it was that the area at the ignition source was too curved. It can however be improved using more vertexes to model the cylinder or reducing the mesh size. The largest values could be measured in the case of the top of the objects, where there was the ignition source, which was followed by the sensor at the top in the middle. The next largest values could be measured with the sensor at the top in the corner. The smallest values were measured with the bottom sensor. This observation is according to reality as the hot gases fill the room from the top to the bottom. It can be concluded that Blender and FDS can be effectively used to model the fire spread on complex geometries.
3.2 Simulating complex geometry
The complex geometry was based on a car's bodywork, but on a reduced scale to fit into the 2 × 2 × 2 m room. The mesh size was 0.025×0.025 × 0.025 in order to capture the main dimensions of the bodywork. The material was wood in this case as well. The geometry was created in Blender, which was exported to FDS later (see Fig. 7). For exporting the geometry two methods were compared: in the first case, the geometry was exported as GEOM, which follows the geometry accurately. In this case, burning on is not allowed. In the second case, the geometry was exported as OBST, which creates the geometry from small cubes. In this case, burning on is allowed.
Complex geometry in Blender (left) and in FDS GEOM (middle), FDS OBST (right), (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
The location of the sensors was the same as in the previous simulation. The ignition source consisted of several particles at 1,000 °C temperature and was placed in the back of the bodywork. The simulations run on a HPC cluster using Message Passing Interface (MPI). The results are shown in Figs 8–12.
Temperature versus time diagrams in the case of the first sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of the second sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of the third sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
Temperature versus time diagrams in the case of the first sensor (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
HRR versus time diagrams (Source: Authors compilation)
Citation: Pollack Periodica 2025; 10.1556/606.2024.01225
It can be seen that the temperature curve is similar in the case of sensor 1. The temperature reached around 350 °C and then oscillated around this value. In the case of sensor 3, the curve is similar (see Fig. 10), but the highest temperature was reached faster. The cause of it is that the hot smoke fills the room from the top to the bottom and this sensor was below the bodywork.
It can be seen that temperature increased rapidly in the case of sensor 2, especially using the GEOM model. The temperature reached almost 1,200 °C in this case. After a fast increase, it oscillated around 600–700 °C. In the case of the OBST model the temperature reached 400 °C fast then it became constant then increased to around 600 °C at 230 s. Then the temperature oscillated between 400 and 600 °C. The maximum temperature was 832 °C this case. The cause of the temperature difference can be explained by 2 things: in the OBST model burning on was allowed and there were missing parts compared to the GEOM model.
In the case of the third sensor, the temperature reached 300 °C rapidly and then oscillated around this value in both models. There was a difference in the temperature at the end of the simulation, which was decreasing in the case of the OBST model. The reason for that is that burning on was allowed and there remained less material for burning.
In the case of sensor 4, the temperature did not reach 90 °C till 200 s in the case of the GEOM model. After that, it increased to 200 °C then started to decrease. In the case of the OBST model the temperature increased similarly at the beginning of the simulation after that it increased at a slower pace to 125 °C. The maximum temperature in this case was 176 °C. The difference between the curves can be explained by that in the case of the GEOM model more material remained and then hot gases filled in the bottom of the room.
In the case of the GEOM model, the HRR increased to 60 kW. After that, it started to decrease for a short time and then increased rapidly. After that, it started to decrease as the O2 was consumed and the intensity of fire therefore decreased. The average value of the HRR was 32.75 kW, which is a high-intensity fire for a relatively small wooden part. In the case of the OBST model the HRR increased fast and reached its maximum, which was 49 kW at 160 s, then started to decrease, when the O2 and the burnable material was consumed. The average HRR was 30.67, which is similar to the GEOM model.
The calculation time was also compared. The simulations were run on an HPC cluster using MPI with 8 processes. The calculation time in the case of the GEOM model was 22:42:28 and in the case of the OBST model 16:28:38. The OBST model is faster even with burning on allowed.
To summarize, the OBST model seems to be more effective. Not only the simulation is faster, but also the burning on of the object can also be included. The temperature curves are similar, only in the case of the sensor at the top has large differences. The explanation is that in the case of the OBST model, some of the geometry was lost and the burning on did not allow the temperature to increase higher. For further simulation, the OBST model will therefore be used. With a high-resolution mesh the OBST model can be nearly as accurate as the GEOM model. However, this can increase the simulation time to a great extent. The main question for further research is to find the equilibrium between the simulation time and the accuracy of the model with an adequate mesh.
4 Conclusions
In this paper, the numerical simulation of different geometry tree objects is presented. It could be concluded that to achieve a more precise and accurate fire simulation, the first step is to select an appropriate simulation model. The comparison of each model is only possible after running their simulation and analyzing their results. Simulations approximately present the fire spread and temperature changes in closed spaces on different objects. It was concluded that the hardware demand of fire spread simulations is great; it requested almost a day even on supercomputers. However, simulation results are very important from the point of view of real fire case validation or the evaluation of 1:1 ratio examination. Practical experiences provide a strong basis for the setting of simulation parameters, mesh, and result evaluation. The research contributes to the development of more precise fire models and provides a solid base for future studies to explore even more complex geometries. These advancements in fire modeling could enhance fire safety strategies and improve our ability to predict and manage real-world fire scenarios involving complex structures.
References
- [1]↑
Á. Zs. Mohai, B. Elek, and M. Kovács, “Environmental aspects of using fire protection systems with a sustainable development approach,” J. Integrated Disaster Risk Manage., vol. 13, no. 2, pp. 22–39, 2024.
- [2]↑
K. McGrattan, R. McDermott, S. Hostikka, J. Floyd, C. Weinschenk, and K. Overholt, Fire Dynamic Simulator User’s Guide. NIST Special Publication, 2019.
- [3]↑
T. Pusztai and Z. Simenfalvi, “CFD analysis on a direct spring-loaded safety valve to determine flow forces,” Pollack Period, vol. 16, no. 1, pp. 109–113, 2021.
- [4]↑
A. Saliby and B. Kovács, “CFD modeling for phase change materials integrated to building envelope,” Pollack Period, vol. 19, no. 1, pp. 67–72, 2024.
- [5]↑
G. Villi and M. De Carli, “Detailing the effects of geometry approximation and grid simplification on the capability of a CFD model to address the benchmark test case for flow around a computer simulated person,” Build. Simul., vol. 7, pp. 35–55, 2014.
- [6]↑
L. Halada, P. Weisenpacher, and J. Glasa, “Computer modeling of automobile fires,” in Advances in Modeling of Fluid Dynamics, C. Liu, Ed. Ch. 9, 2012.
- [7]↑
M. P. Giraldo, J. Avellaneda, A. M. Lacasta, and V. Rodríguez, “Computer-simulation research on building-façade geometry for fire spread control in buildings with wood claddings,” in World Conference on Timber Engineering, Auckland, New Zealand, July 16–19, 2012, pp. 311–314.
- [8]↑
C. H. Zhong, “Fire Dynamics Simulation (FDS) study of fire in structures with curved geometry,” BSc Thesis, Universiti Tunku Abdul Rahman, 2013.
- [9]↑
R. Hansen and H. Ingason, “Heat release rates of multiple objects at varying distances,” Fire Saf. J., vol. 52, pp. 1–10, 2012.
- [10]
J. Degler, A. Eliasson, J. Anderson, D. Lange, and D. Rush, “A-priori modeling of the Tisova fire test as input to the experimental work,” in The First International Conference on Structural Safety under Fire & Blast, Glasgow, Scotland, United Kingdom, September 2–4, 2015, pp. 1–10.
- [11]
X. Dai, A. Gamba, C. Liu, J. Anderson, M. Charlier, D. Rush, and S. Welch, “An engineering CFD model for fire spread on wood cribs for travelling fires,” Adv. Eng. Softw., vol. 173, 2022, Art no. 103213.
- [12]
M. K. Cheong, M. J. Spearpoint, and C. M. Fleischmann, “Calibrating an FDS simulation of goods vehicle fire growth in a tunnel using the Runehamar fire experiment,” J. Fire Prot. Eng., vol. 19, no. 3, pp. 177–196, 2009.
- [13]↑
D. Gross, “Experiments on the burning of cross piles of wood,” J. Res. Natl. Bur. Stand. C. Eng. Instrument., vol. 66 C, no. 2, pp. 99–105, 1962.
- [14]↑
R. Kuti, G. Zólyomi, G. László, Cs. Hajdu, L. Környei, and F. Hajdu, “Examination of effects of indoor fires on building structures and people,” Heliyon, vol. 9, no. 1, 2023, Art no. e12720.
- [15]↑
E. Gissi, F. Notaro, G. Longobardo, and F. Valpreda, Development and testing of BlenderFDS, the open, community-based, user interface for NIST FDS, 2024. [Online]. Available: https://www.academia.edu/10074830/Development_and_testing_of_BlenderFDS_the_open_community_based_user_interface_for_NIST_FDS. Accessed: Aug. 21, 2024.
- [16]↑
M. Vanella, R. McDermott, G. Forney, and K. McGrattan, Fire dynamics simulator: Advances in simulation capability for complex geometry, 2016. [Online]. Available: https://files.thunderheadeng.com/femtc/2016_d1-07-vanella-paper.pdf. Accessed: Aug. 21, 2024.
- [17]↑
R. K. Janardhan and S. Hostikka, “Predictive computational fluid dynamics simulation of fire spread on wood cribs,” Fire Technol., vol. 55, pp. 2245–2268, 2019.
- [18]↑
ChemPages Netorials: Thermodynamics: Heat and Enthalpy, 2024. [Online]. Available: https://www2.chem.wisc.edu/deptfiles/genchem/netorial/modules/thermodynamics/enthalpy/enthalpy3.htm. Accessed: Aug. 16, 2024.
