Numerical simulation of the fire behaviour of facade equipped with aluminium composite material‐based claddings‐Model validation at large scale

The recent fire events in buildings involving combustible cladding systems have raised concerns regarding the risk that these systems can pose. Understanding such facade fires is complex as they involve a combination of various products and system. Facade fire propagation tests at ISO 13785‐1 intermediate scale were performed on different combinations of aluminium composite material (ACM) claddings and insulants. Simulations were addressed to reproduce these tests and were validated in terms of thermal conditions in the system. This allowed additional investigation and understanding of fire propagation on the facade and more accurate determination of the fire behaviour of the overall system. In this paper, the scaling influence on the fire behaviour of ACM clad systems is investigated with simulations performed to reproduce fire tests at the BS8414‐1 larger scale on three different combinations of ACM and insulants. The contributions of the cladding and insulant were numerically investigated. The fire behaviour of each component and of the overall system is validated by comparison with experiments. Simulations and tests show that the ACM cladding is the most important element driving the global fire behaviour of the systems. In particular, ACM‐PE‐based cladding systems, whatever the insulant, show extensive fire propagation while its degradation affects the integrity of the cavity.

Following the Grenfell disaster, fire behaviour of facades has become a topic of great interest. [3][4][5] Nevertheless, understanding such facade fires is very complex as they involve a combination of various products and system parameters. The fire behaviour of an external facade insulation system is dependent on the overall system" 's performance, rather In the context of fire safety, it should be noted that for systems such as external thermal insulation composite systems (ETICS) or ventilated facades, the materials used (cladding or insulation) may be combustible. In addition, in ventilated facades, the air gap may be a vector of fire propagation via the chimney effect. 5,[7][8][9]10 Thus, both the materials taken independently and the system as a whole (combination of materials and assembly) can potentially contribute to fire propagation.
At the present time, the use of fire barriers or compartmentation systems, as requested by national regulations, can hinder these problems, but they, too, constitute additional variables in the system. Assessment of a specific facade system's fire performance can be undertaken using large-scale testing, in accordance with local regulations. 5 However, these large-scale tests are pass/fail oriented; and they give very little quantitative information for further interpretation of the fire behaviour of the tested system; if the flames pass over the top of the frame, the test will be stopped and this will not allow a thorough performance investigation. 11 Following the fire at Grenfell Tower in London on 14 June 2017, the UK government established an Independent Expert Advisory Panel to advice on immediate measures that should be put in place to help make high rise residential buildings safe. On 6 July, the Independent Expert Advisory Panel recommended that a series of large-scale BS 8414-1 tests 12 be carried out in order to help building owners make decisions about any further measures that may need to be put in place.
This series of tests included seven combinations of cladding systems.
Previous studies [13][14][15] report on the three facade fire tests that were performed,accordingtotheBS8414-1standardonvariouscombinations of two grades of aluminium composite material (ACM) and two different insulants, similar to those subsequently tested in Guillaume et al.# 3 Several studies have shown the feasibility and usefulness of numerical simulation for facade fires using different test facilities or methods and using different simulation codes. [16][17][18][19][20][21][22][23][24] Published results have shown the feasibility of modelling such test methods using large eddy simulation (LES), especially when incombustible claddings were considered. However, great attention must be paid to the numerical model sensitivity, in particular to correctly representing the behaviour of the flames near the facade system, and thus the thermal stresses received by the facades.
A series of intermediate facade fire propagation tests according to the ISO 13785-1 standard, 25 with additional heat release rate (HRR), and gas analysis using FTIR was addressed in Guillaume et al. 3 The test series comprised nine different combinations of three grades of ACM and three different insulants.
Based on the ISO 13785-1 facade fire tests detailed in Guillaume et al, [3] a preliminary numerical study 26  The objective of the present study is to accurately numerically reproduce the thermal load to which the tested system is exposed, the thermal behaviour of the system and the fire propagation via the facade. The simulations are carried out with FDS software, based on the numerical hypothesis and input dataset previously validated. 26 Iterative calculations are performed to verify the model's consistency with, and relevance to, the experimental results obtained during the reference fire tests referenced 13-15 on various combinations of two grades of ACM and two different insulants, similar to those subsequently tested at intermediate scale. 3 In the first step, the BS8414-1 test is numerically modelled, to reproduce the fire test behaviour of a facade system comprising [ACM-PE] cladding and [PIR] insulant. In order to accurately reproduce the fire propagation on the system, the simulations are performed with a fine numerical grid similar to that used in the previous step, detailed in Dréan et al. 26,28 Comparisons are made between numerical and experimental results for temperatures based on a previous study. 13 The second step of the study consists in using coarser numerical cells, that are more commonly encountered in large-scale simulations and engineering studies. Using such a coarse grid is necessary because of the difficulties and time taken in modelling larger scales, such as a full-scale facade on a high rise building. However, a numerical hypothesis must be fixed in order to apply the model developed using an accurate fine grid, to a model developed with a coarser grid. The main objective is to reproduce the thermal gradients in gas and solid phases achieved with the initial model. For example in the coarser grid model, the air cavity now has the same thickness as the cell size. The exchanges between the materials and the gas phase are also evaluated in larger cell.
In the third step, the numerical model is modified to verify the combustion behaviour of each part of the system. These additional DRÉAN ET AL. 982 6 includes not only the cladding and the insulant's characteristics but also those of cavities, cavity barriers, mounting and fixing, substrate, than the performance of each individual component. The set-up of the experimental facility was developed according to BS8414-1 specifications. 12 This is a large-scale facade build-up as detailed in previous studies. [13][14][15] It consists of two frames with dimensions 2600 × 8000 mm for the back wall (main face) and 500 × 8000 mm for the side wall (wing) with calcium silicate (CalSil) boards as a support for the tested system and on which the system is installed ( Figure 1A).
A wood crib with a HRR close to 3.0 ± 0.5 MW is installed in the combustion chamber of the facility, and its dimension are close to 1500 × 1000 × 1000 mm (l × w × h). The complete system is then placed under a large hood to collect the effluents. 11,12 The instrumentation used during the test is fully detailed in previous studies [13][14][15] and indicated in Figure

| Tested systems in the experimental reference
In the reference test reports, [13][14][15] facade fire tests have been performed on three combinations of two different compositions of ACM and two different insulants (Table 1). Information regarding these insulants, such as density or thermal conductivity, is available on product datasheets from their respective manufacturers. The However, attention must be paid to the test repeatability and to the variability of the heat released by the fire source since a wood crib was used.
Concerning mounting and fixing, the cladding systems were assembled on calcium silicate boards. The cladding was made of 3 × 4 panels for the back wall (1B to 3E in Figure 1) with two additional panels around the combustion chamber (0B and 0C in Figure 1) and four panels (0A to 3A in Figure 1) for the side wall (or test wing). Gaps between cladding panels were 20 mm wide.
A set of four horizontal intumescent cavity barriers, with a 25 mm thickness of intumescent, were installed. Vertical cavity barriers were placed: two for the back wall and one for the side wall (see Figure 2).   So the available experimental test data only covers the period from ignition until the tests were extinguished.

| NUMERICAL SET-UP
Preliminary numerical development was carried out by means of iterative calculations, which were performed to verify the model's consistency and relevance with the experimental results obtained during the BS8414-1 fire tests. The objective of this study is to accurately numerically reproduce the thermal load to which the tested system is exposed, the thermal behaviour of the system and the fire propaga-

| Numerical tools
The numerical simulations are performed with the CFD code FDS ver- Mesh size is taken at 20 × 20 × 20 mm for the facility, so that the grid is refined to capture accurately the combustion and turbulence phenomena of the system. A total of 11.25 million cells are used.
In the FDS reference guide 27  This technical choice is made to conserve reasonable calculation costs and regarding the later upscaling application of this numerical model. Compromises were needed to develop a robust numerical model able to be used for this application. The selected cell size is enough to capture the main features of local effects, not in details, but sufficiently to reproduce the fire behaviour in the present application. Furthermore, quickly after the beginning of the test, the fire propagation from the burner to the system leads to its combustion.
Thus, the cladding panels, and thereby the gaps between them disappear in the first minutes of the test.
In the numerical model, virtual instrumentation consisting of thermocouples are placed at the same locations as during the real test ( Figure 1).

| Numerical model for thermal analysis
The thermal characteristics of the system components are integrated in the numerical model in terms of density, thermal conductivity, heat capacity, emissivity, heat of combustion, ignition temperature, mass loss rate, and species release rates, for every material involved. All thermal and combustion properties considered, for the material making up the systems, are taken from the numerical model validated at intermediate scale. 26,28 A justification for the numerical model used for thermal degradation analysis of the materials and overall combustion is deeply discussed in Dréan et al. 26,28 The burning rates of the materials are indeed imposed. The burning rate of the facade is however more complex as flame spread occurs.
The most challenging point of this approach was to find the suitable parameters, with a physical meaning. This parameters included thermal parameters (so the heat transfer is correctly modelled) and the right combustion properties (ignition temperature and mass loss rate).
These complete sets of parameters have to be found for several materials with strong interaction from one to another. Thus, the numerical results achieved will not fit correctly the experimental ones because mass loss rate is prescribed but because the thermal properties and the fire properties are suitable. Indeed, the materials properties, ignition temperature, and burning rates of the materials are imposed. The thermal properties of the calcium silicate (CalSil) supports are implemented to ensure a correct thermal transfer in terms of loss from the facade system backing. Exposed boundary conditions are thus considered.

| Numerical model for the fire source
The model of the fire source evolution of the basic test rig was previously validated by comparison with several calibration tests (plasterboard facade) as presented in Appendix C (see Supporting   Information). The numerical model assumed the HRR indicated in Figure 3 for the wood crib combustion. According to the previous study, 12 this heat source releases a nominal total heat output of 4500 MJ over 30 minutes with a peak rate of (3 ± 0.5) MW. The HRR is comparable with that indicated in Anderson and Jansson. 31

| Comparison between experimental and numerical values for the [ACM-PE + PIR] configuration
The comparisons between numerical and experimental results for temperatures are analysed. Experimental data and numerical results are smoothed using a rolling average over 30-second periods. During the test, the crib was extinguished at 395 seconds because the "flame height" fail criterion of BR135 was exceeded. However, the numerical simulation was performed up to 10 minutes.
The overall uncertainty of a numerical prediction is the combination of the uncertainties of both the numerical model and of the input parameters. 32 The numerical uncertainties are evaluated following McGrattan and Toman 33 and are indicated in Table 2. Numerical parameter uncertainty is an important consideration to assess the reliability of the results and the impact of the input parameters of a model. In this study, the input parameters are taken from the literature and used to fit the experimental results. Thus, no input uncertainties are evaluated. However, a sensibility analysis was performed for the ignition temperatures of the cladding and the insulant.

| Fire behaviour observations
A comparison of the experimental fire behaviour of the system with numerical observations is presented in Figure 5 at different elapsed times. The visualization begins at 2 minutes and is shown for every minute thereafter and illustrates the fire development and the system behaviour during the simulated test. Unfortunately, no experimental observations were available. Thus, the modelled fire behaviour helps to understand the flame propagation over the tested system.
It can be seen that the fire propagation from the wood crib to the external cladding starts early, at around 2 minutes. Then the fire propagates to the whole system, and flames are visible through the gap between the first and second row of panels above the combustion chamber, at around 3 minutes. A quick propagation of flames on the cladding surface appears between 4 and 7 minutes, and the side wall starts to contribute after 5 minutes. Some counter-current fire propagation at side wall is visible after 6 minutes. Then, the fire intensity tends to decrease, corresponding to the lack of combustible materials.
Slowing propagation is observed on the side wall between 6 and 10 minutes. Furthermore, the real test was ended after 395 seconds.

| Comparison with experimentally measured values
The simulated temperatures at the back and side walls at external L1 level ( Figure 8  In the corresponding test report, level 2 external thermocouples reached the external fire spread criterion at 360 seconds. Thus, this criterion seems to be reached numerically 30 seconds before the experiment at some location. The main differences observed can be attributed to the fact that thermocouples can slightly move in the cavity due to local turbulence during the test or to the thermocouple inertia when the crib is extinguished. Furthermore, only one test was performed, so attention must be paid to its repeatability. Furthermore, the numerical model does not take into account mechanical changes of the system (local distortions were observed after the test) or displacement of thermocouples inside the cavity. Regarding the 5% and 8% uncertainties on gas temperature evaluated experimentally and numerically, respectively, the simulated temperature are very comparable with the experimental data or comprised in the confidence range. However, the heat source from wood crib is 3 ± 0.5 MW, thus 3.5 or 2.5 MW as a range, it is around 36% and it is acceptable in the standard.
However, the numerical simulation illustrates the full development of the fire, after the time at which the "flame height" fail criterion of BR135 was exceeded and the crib was extinguished. This

| Numerical evaluation of the HRR during the test for the [ACM-PE + PIR] configuration
The total HRR during the BS8414-1 test was evaluated numerically.
The HRR evolution indicated in Figure 12A corresponds to the heat released by the tested system and by the fire source (wood crib).   Figure 12B. A maximum value around (4.5 ± 0.8) MW at 7 minutes 30 seconds of the test is observed.
The energy released by the tested system without the contribution of the wood crib is also indicated.

| Variance of the numerical model for the [ACM-PE + PIR] system
The method 35 Table 3. The minimum cosine value evaluated, close to 0.87, is associated with the maximum relative difference value close to 41% and concerns the external system temperature at the side wall in position L2. This low cosine value and high relative difference value can be attributed to the fact that thermocouples can slightly move in front of the system due to local turbulence during the test or to the thermocouple inertia when the crib is extinguished. Furthermore, only one test was performed, so attention must be paid to its repeatability. However, this relative difference is in the range of the numerical and experimental uncertainties.
For all the other temperatures at the back and side walls, as well as the air cavity temperatures, cosines are in all cases higher than 0.86, and relative differences lower than 34%. The values for the relative differences are also in the range of both numerical and experimental uncertainties. For the external gas temperature at L1 level at back wall, the relative difference is very low (9.5%), associated to a cosine close to 0.94. Thus, the numerical predictions at this location are accurate enough to properly represent the flame load from the fire source and the ignition of the tested system in this area.
As a preliminary conclusion, the numerical model applied to the [ACM-PE + PIR] system is valid and can be used for further investigations.

| NUMERICAL EVALUATION OF THE MODEL FOR THE [ACM-PE + PIR] SYSTEM, USING A COARSER GRID
Once the small 20-mm grid numerical model validated, a model is created with coarser numerical cells, which are more commonly encoun-

| Numerical set-up
The  total HRR achieved numerically for the wood crib and facade system (Q = 7 MW), the mesh size should be close to 130 mm. Thus, the con-

| Comparison between experimental and numerical values for the [ACM-PE + PIR] system using a coarse grid
The simulated temperatures at the back and side walls at external L1 level ( Figure 13) and L2 level ( Figure 14) using the coarse grid are comparable in magnitude and evolution with the experimental tempera-

| Numerical evaluation of the HRR for the [ACM-PE + PIR] configuration on coarse grid
The total HRR during the BS8414-1 test is evaluated numerically using the coarse grid ( Figure 17). Its evolution corresponds to the heat The THR evaluated with the fine and the coarse grids are also compared. Good overall agreement is found, indicating that fuel stoichiometry and the fuel mass released are correctly taken into account in the simulation with the coarser grid.

| Variance of the numerical model for the [ACM-PE + PIR] system
The relative difference (hybrid method) and the cosine associated with each quantity to be validated are presented in Table 4. The minimum cosine value evaluated, close to 0.93, and the maximal relative difference value, close to 28%, concerns the external system temperature at the back wall in position L2 and the external system temperature at the side wall in position L1, respectively.
For all the other temperatures at the back and side walls, as well as the air cavity temperatures, cosines are in all cases higher than 0.95, and relative differences lower than 27%. The values for the relative differences are also in the range of both numerical and experimental uncertainties. For the external gas temperature at L1 level at back wall, the relative difference is very low (16.7%), associated to a cosine close to 0.96. Thus, the numerical predictions at this location are accurate enough to properly represent the flame load from the fire source and the ignition of the tested system in this area, even if a coarse grid is used.
As a preliminary conclusion, the coarse grid numerical model of the [ACM-PE + PIR] system is valid and can be used for further investigations.

| Investigation into the influence of the wood crib
According to the testing conditions given in BS8414-1, the HRR of the fire source (wood crib) can commonly range between 2.5 and 3.5 MW   The influence of the wood crib HRR on the numerical model of BS8414-1 with the coarser grid was investigated, using three different values of the wood crib HRR: 2.5, 3, and 3.5 MW. In the numerical simulations described above, the assumed HRR for the wood crib was close to the maximum HRR allowed by BS8414-1 (eg, 3.5-MW plateau).

| Influence on the HRR of the overall system
The overall HRR during the BS8414-1 test of the [ACM-PE + PIR] system was evaluated numerically using the coarse grid for three values of the peak HRR of the initial wood crib. The HRR evolution indicated in Figure 18 corresponds to the heat released by (a) the tested system and the fire source and (b) the tested system only.
The HRR achieved with the 3.5-MW wood crib shows that the combustion of the system starts more quickly due to more important flame above the fire room. The overall kinetics is similar whatever the peak HRR of the crib. The maximum total HRR values are close to 8 MW (± 0.5 MW) for all three cases investigated. The contribution of the evaluated system appears to be only time shifted when this range of wood crib HRRs is considered.

Examples of comparisons between numerical and experimental results
for external temperatures at level L1 on the back and side walls are presented in Figure 19 for the three different wood crib peak HRRs. properties. Thus, in this configuration, the ACM cladding will not have any contribution to the heat released. Furthermore, no fire retardants were considered in the ACM cladding. The simulations were performed using the coarse grid described previously. The experimental results for these systems can be found elsewhere. 14,15 The

| Heat release rate
The HRR curves from the numerical analysis of the tests of the three configurations are presented in Figure 27.  were numerically modelled following previous BS8414-1 experiments. [13][14][15] In a preliminary study, 3  In this paper, the behaviour of the different facade systems was predicted at a large scale through the modelling of the abovementioned BS8414-1 tests. Gas temperatures and the general behaviour of system were predicted with good agreement between the model outputs and the experimental data. In all the simulated configurations, the strong combustion of polyethylene cored ACM cladding leads to the quick consumption of the material soon after its ignition. The ACM cladding burns in wellventilated conditions because of its external location, and it disappears at early stage of the fire and thus reduces the cavity performance. The insulant is exposed to the fire contribution of the cladding and to the flames in the cavity, even if the cladding has disappeared. During the fire test, the insulant can then burn in well-ventilated conditions because it is quickly exposed to the external environment once the cladding has disappeared. Furthermore, the ACM-PE represents more than 90% of the value of the peak of HRR and of the total energy released.
An additional sensitivity analysis was performed on the [ACM-PE + PIR] system, to evaluate the influence of the wood crib on HRR. It can be concluded that once the combustion of the system above the fire room starts, the allowed variance in the heat output of the wood crib has only a minor influence on the fire propagation of the system. The