Burning material behaviour in hypoxic environments: An experimental study examining a representative storage arrangement of acrylonitrile butadiene styrene, polyethylene bubble wrap, and cardboard layers as a composite system

Cone calorimeter and controlled atmosphere cone calorimeter experiments were conducted on various samples. The intent of the tests was to examine the behavior of uniform and composite samples in a range of thicknesses, irradiances, and oxygen concentrations. Single, uniform layers of acrylonitrile butadiene styrene (ABS) were compared to a composite mix, comprising of ABS with a surface layer of cardboard and a secondary layer of polyethylene bubble wrap (intended to represent a potential storage arrangement). The horizontal samples have been tested at irradiances of 25 and 50 kW/m2 and oxygen concentrations of 20.95%, 17%, and 15% to examine a range of significant variables. Results for the uniform arrangement indicated various correlations, previously observed in the works of others, such as the relationships typically described between applied heat flux, ignitability, heat release rate and the effect of the introduction of hypoxic conditions. However, results were shown to change significantly when samples were arranged to feature composite layers. A hypothesized cause of the behavioral change, namely the soot and char residual introduced from the incomplete combustion of the cardboard layer, highlights further important variables that require consideration in material testing under hypoxic conditions. Such variables, namely specific material behaviors and sample orientation, must be sufficiently captured in the design methodologies of systems reliant upon the introduction of hypoxic conditions. It is concluded that sufficiently capturing a wider range of variables in burning materials under hypoxic conditions will introduce further design resilience and help optimize fire protection/prevention methods.


| INTRODUCTION
There is a limited body of knowledge within the fire research community concerning the study of materials formed of composite layers, particularly when exposed to hypoxic conditions. 1 While there is more data available observing the behavior of materials of a uniform composition under fire conditions, it is again limited where exposure to hypoxic conditions was the focus. Available research is restricted further still when concerning the polymer acrylonitrile butadiene styrene (ABS). While ABS is often used in the formation of high end electronic equipment there appears to be limited research coverage compared to other common polymer types within the field of Fire Safety Engineering. Due to its common use and predictable burning behavior, there is a large collection of data related to polymethylmethacrylate (PMMA) in reduced oxygen environments. [2][3][4] Similarly, polyethylene (PE), polypropylene (PP), and polystyrene (PS) have also received both theoretical and experimental discussion in the works of researchers such as Tewarson et al. 5,6 and Kashiwagi. 7 ABS has been covered in previous works by others 8 but some of this work focuses on different types of polymer degradation and does not directly cover combustion reactions of ABS exposed to hypoxic conditions. 9 Other works have indeed covered the combustion of ABS in hypoxic conditions. However, they have discussed the material in a much broader context alongside many other materials and have not considered its behavioral changes when stored in a composite arrangement. 10,11 The testing methodologies proposed in EN16750:2017 12 and VdS3527:2018 13 are intended to identify an ignition threshold for oxygen reduction systems (ORS). The ignition threshold is a target oxygen concentration that applies an approximate limiting oxygen concentration (LOC) of the materials contained within the protection zone and an additional safety margin (typically 0.75%-1.0% O 2 concentration). The application of the ignition threshold is in order to ensure the lowest concentration of oxygen necessary to prevent the ignition of stored materials. The test used to define the ignition threshold has been noted as being very material dependant and primarily for the testing of generic materials. Such limitations are of concern because of the likelihood that composite materials are contained within the protection zone. 1 As a relatively new method of fire protection, the installation of ORSs is limited to 700 installations outside North America (as recorded in 2014). 14 As such, there is a limited number of case studies available in which to learn and improve system design. The limited number of installed systems is likely to slow the iterative evolution of system design.
Due to the value, in both financial and societal terms, of the contents protected by ORSs, there is significant motivation to explore composite material behavior in hypoxic conditions. This motivation is further supported by the limited consideration of composite behavior within the experimental methodology of ORS design guidance as reported. 1 It is not the intention of this study to replicate the testing methodologies proposed for ORSs, but rather, to supplement the knowledge of material behavior in hypoxic conditions in order to improve the theoretical discussions used to guide design methodologies.

| THEORETICAL BASIS FOR THE USE OF OXYGEN CALORIMETRY
Oxygen consumption calorimetry is a long-established method of determining the heat release rate (HRR) of a material. 15 The HRR is determined by the principle of oxygen calorimetry, conceptually recognized by Thorton 16 and empirically confirmed by Parker 17

and
Huggett, 18 where organic liquids and gases were shown to consume approximately the same mass of oxygen per net amount of heat that is released (13.1(±5%) kJ per 1 g of O 2 ). The method is widely used and standardization 19 has allowed for the production of replicable data assuring its wide-ranging use within the research community. 20 Evidently, there are some critical design limitations to the standard cone calorimeter (CC) that limit its applicability to evaluate fire in hypoxic conditions, not least is its open design. It is the intent of the controlled atmosphere cone calorimeter (CACC) to maintain the operating capabilities of a typical cone calorimeter (CC) while being capable of operating at oxygen conditions which are different to those encountered in ambient environments. As noted by Leonard,21 one of the primary differences between a typical CC and the controlled atmosphere alternative includes the introduction of a flow of N 2 /air to lower the volume fraction of O 2 . A CACC, therefore, introduces hypoxic conditions through vitiation of the environment by displacing oxygen with nitrogen rather than 'under-ventilation' where the flow of air is limited into the enclosure. This is important to note as ORSs also create hypoxic conditions through vitiation and key differences in material behavior have been observed between vitiated and under-ventilated conditions. 22 Where a standard CC is used, ISO5660-1:2019 calculations can be used to determine the HRR. The study conducted made use of the alternative methods outlined by Werrel 23 where the HRR is calculated with a time-dependent mole fraction of oxygen in incoming air, X A 0 O2 , rather than the constant value used in the typical CC method.
This calculation method, first proposed by Werrel, 23 describes how the use of time-dependent mole fractions in HRR calculations is due to the changing ratio of the mass flow rate into the exhaust duct during experiments using a CACC. This is due to the change in temperature, and therefore density, of the exhaust gases from the CACC chamber.
The fan controlled extract system located in the CACC exhaust duct, typically running at a flow rate of 24 L/s (0.024 m 3 /s), has a constant volumetric flow rate while the fluctuating density within the CACC chamber will change the mass flow into the exhaust ductwork. This change in mass flow into the exhaust duct results in the changing rate of entrained air from the laboratory, mixing via the open chimney connection between the chamber and exhaust duct. Subsequently, there is a change in the ratio of the mass flow rate of the exhaust gases out of the chimney to the mass flow rate of the entrained ambient air. This process leads to dynamic baseline mole fractions rather than constant values found in the standard CC apparatus. This invalidates the assumption, made in standard cone calorimetry experiments, that mass flow rate into the exhaust flow stays constant and therefore baseline readings of effective mole fractions for constituent gases (used to calculate mass of oxygen consumed) stay the same during the experiment.
T e = absolute temperature of gas at the orifice meter (K).
X A CO = measured mole fraction of carbon monoxide in exhaust gases (-). X A S O2 = measured mole fraction of oxygen in the surrounding air (-). ϕ = oxygen depletion factor (-).   With the oxygen depletion factor in this case defined as: The methods described for determining heat release rate are one measure of burning behavior that is typically used in Fire Safety Engineering to quantify, evaluate, and compare material performance.
These methods are not used in the testing regimes prescribed for ORSs, with instead, samples of 200 mm by 25 mm by 5 to 25 mm in reduced sized protected zones used to replicate the proposed system.
The ignition threshold is subsequently determined through the use of observable performance to assign pass/fail criteria. However, oxygen consumption calorimetry been considered appropriate in the current study in order to create reproducible and comparable data for academic study beyond simple 'no-go/go' criteria used in practice. It should be noted that the experimental work was planned and conducted in early 2020, prior to the release of ISO 5660-5:2020. 26 However, results can be considered broadly comparable as the methodology followed and apparatus design described in this work is and ABS were used for testing. These materials as shown in Figure 2.

| EXPERIMENTAL PROCEDURE
The ABS samples were sourced 27 and confirmed to contain the following distribution of base polymers; Styrene-58% to 63%, Acrylonitrile-20% to 25%, Butadiene-16% to 21%, Additives-1% to 4%. The bubble wrap sample was sourced 28 and was confirmed to be composed of polyethylene. Cardboard used for testing was sourced from spare cardboard packaging available within the laboratory. It is noted that, due to the wide-ranging potential of ORS, many different fuel arrangements could have been explored. ABS represents a popular thermoplastic for a wide range of uses because it is superior toughness and rigidity. These characteristics have been described as unusual for thermoplastics and explain the use of ABS for thousands of products in the housing, electronics, and automotive industry. 29 The cardboard and bubble wrap layers were chosen because cardboard was identified as the primary fuel load within ORS protected zones in a number of case studies. 30,31 One of these case studies featured the protection of a storage space containing high precision scales, which, from the product literature studied, often used ABS for the scale due to its robustness. 32 It should be noted that the chosen composite layers are somewhat arbitrary. They do not represent a particular arrangement but instead a hypothetical arrangement based on examined case studies.
While the CC has been said to adequately represent composite arrangements of 50 mm or less, 33 other researchers have reported complications in performance due to edge effects. Richardson and Brooks 34 reported that composites with a surface layer of low burning propensity but an underlayer of high burning propensity performed in an unrepresentative fashion at the edges due to the early involvement of the composite underlayer. This effect was not observed during the experiments conducted as the surface layer was reported to ignite and burn much faster than the underlayer in this instance due to the material properties of the layers involved.

| RESULTS AND DISCUSSION
Experimental results are as shown in Tables 1 to 3  Varying the sample thickness in uniform samples in this instance, between 5 and 20 mm, does not appear to alter the material thermal gradient to such an extent that heat losses from the material surface are significant enough to prevent or delay igniting the sample due to the influence of the applied irradiance. This is in alignment of the findings of Berg and Lindgren, 35 and can be observed in Figure 3, where there is no clear trend toward increased t ig values in thicker uniform samples.
To further investigate this, an approximation of whether the samples demonstrate semi-infinite behavior has been calculated using Equations (3) to (5)  An approximation of semi-infinite behavior in a onedimensional slab can be calculated using Equations (3) and (4), where L is the thickness of the sample (m), t is the time of exposure to a uniform heat flux (s), and α is the thermal diffusivity (k=ρc ) (m 2 =s ). Where Equation (4) equals approximately 2 then semi-infinite behavior can be assumed provided that Equation (3) is true.
It is also considered reasonable to assume semi-infinite behavior where Equation (5) is true 36 : Composite sample densities have been calculated based on the total sample weight divided by the total thickness of the system and not on the basis of each independent layer. b The error range of peak heat release rate measurements has been given by the range of values 25 seconds before and after the measured PHRR. Results from Table 4 indicate that, at the time of ignition, the 5-mm sample is not semi-infinite, whereas the 10-mm sample is semi-infinite. It would be expected that a sample with a heated unexposed face would ignite faster than a comparable semi-infinite sample. However, it is noted that the value used to determine whether the 10-mm sample in Table 4 is semi-infinite is close to the boundary used by Equation (5)  Results from Table 5 indicate that, at the time of ignition, markedly increased due to the increased irradiance, both samples are potentially semi-infinite. This would explain the similar t ig values recorded in Figure 3 as the ignition criteria were primarily caused by irradiation onto the sample surface rather than the progressive heating of the entire body of the sample.
Analysis using engineering calculations suggests that differences in thermal inertia between samples at thicknesses ranging from 5 to 20 mm was not significant enough to effect time to ignition. This appears to be because the thicknesses examined did not affect whether t ig was caused by a uniform rise in temperature of the entire sample body (as appears to be the case under 25 kW/m 2 ) or the application of sufficiently high irradiance to the surface (as was the case for the 50 kW/m 2 sample).
The changes to ignition time in samples of different thicknesses are more significant in the composite samples, however, this is thought to be due to the effect of the soot and char residual layer rather than the thicknesses of the ABS, cardboard, and PE bubble wrap layers as discussed later.

| Influence of irradiance
Ignition times are shown to decrease under the exposure to higher levels of irradiation as has been shown in the work of Urbas, 38 where all but one fire rated timber material was found to ignite sooner where a higher level of irradiance within the CC was applied. This is clearly the case for uniform samples shown in Figure 4A,B. In  Figure 4D, the secondary peak behavior of the 5 mm ABS underlayer samples now demonstrate steeper rates of rise in HRR, which were exactly the same as the 20 mm ABS underlayer in the uniform samples shown in Figure 4B. These will be of further discussion where the impact of the composite layer is examined.

| Influence of oxygen concentration
Ignition, as illustrated in Figures 5 and 6, was measured through observation and then validated through reviewing video recordings of the experiments. The error bars in Figure 5 show the minimum and maximum t ig values, collected over three repeat tests, for the corresponding test group. There is a conflict within reviewed literature regarding the relationship between the t ig and oxygen concentration. There are studies indicating a strong correlation 39  nonfire rated wood. 40 There is some risk that the use of the inlet flow rate into the CACC may influence the t ig of samples and that this may be misinterpreted as being due to the oxygen concentration imposed rather than the effects of apparatus dependency. 2 Observations illustrated in Figure 6 appear to indicate a slight increase in t ig between ambient and hypoxic conditions for uniform samples of both 5-and 10-mm thick. However, the increase in t ig between 17% and 15% is slightly less conclusive as there is no pattern emerging to allow for a definitive conclusion from the data. While the data indicate that the t ig increased slightly when oxygen concentrations were reduced, further testing is necessary to determine the true extent of this effect and the dependency on apparatus design. The most notable cases where oxygen concentration influences t ig are all cases where there is a composite arrangement as shown in Figure 5. This is, in part, due to the changes in burning behavior due to the combination of composite layering and hypoxic conditions as discussed in a later section. Time to ignition has been taken as the mean value recorded over three repeat tests in uniform samples at the relevant thickness. b Where the preliminary criteria required in Equation (3) are not met, Equation (4) has not been recorded.
F I G U R E 4 Average of heat release rate per unit area (kW/m 2 ) values for uniform and composite samples at 20.95% O 2 over three repeat tests (A, Uniform samples exposed to 25 kW/m 2 at 5 and 10 mm; B, Uniform samples exposed to 50 kW/m 2 at 5 and 20 mm; C, Composite samples exposed to 25 kW/m 2 at 5 and 10 mm; D, Uniform samples exposed to 50 kW/m 2 at 5 and 20 mm)

| Influence of composite form
The composite form was shown to have a significant impact on ignition times in multiple ways. First, as shown in Figure 4,

| Influence of irradiance
Irradiance was shown to be the largest contributing factor to heat release rates. Higher levels of irradiance caused greater peaks in heat release rate consistently. This was particularly true for the uniform samples shown in Figure 4A,B but also the composite samples as shown in Figure 4C,D.

| Influence of oxygen concentration
HRRPUA were generally shown to be lower when hypoxic conditions were introduced. This was true for all uniform samples. Reducing the oxygen volume fraction was also shown to alter the relationship between material thickness and peak HRRPUA as is shown in Figure 9.
Thicker, uniform samples demonstrate a greater reduction in peak HRRPUA when oxygen levels are reduced with a more gradual incline in heat release rate growth over time. As the HRRPUA lowers the material thickness of the sample has a greater significance on the overall transference of heat internally within the sample. For uniform samples at ambient conditions, the comparable HRRPUA peaks for different sample thicknesses suggest that the thermal inertia of these samples has a negligible effect on peak heat release. Following from this, it is also evident that thermal inertia, and therefore sample thickness, becomes more critical to peak HRRPUA as the oxygen volume fraction is lowered.
The combined effect of hypoxic conditions and increased material thickness in the uniform samples are not demonstrated in composite samples. This is as shown in Figure 10 where As can be observed in Figure 9, the relationship between material thickness and the exposure to lower volume fractions of oxygen for the uniform sample set displays a trend where, as the volume fraction of oxygen is reduced, the HRRPUA peak for samples is reduced also. These reductions of HRRPUA were observed to be greater for thicker samples.
The uniform samples observed correlation between heat release rates, material thicknesses, and oxygen volume fractions. The relationship between HRRPUA and oxygen concentration in uniform samples, in this case from Tests 6, 15, and 16, has been shown in Figure 11.
Such trends are not observed in Figure 12 where the influence of material thickness can no longer be discerned by the shape of the can see that five out of six of the hypoxic tests had higher peak heat release rates than any of the three ambient tests.
Broadly, there appears to be two primary contradictions between the composite and uniform tests: • Typical HRRPUA indications of thicker/thinner behavior, based on the thickness of the sample, are no longer clearly observed.
• Typical indications of the effect of hypoxic conditions on HRRPUA are no longer clearly observed.
It is hypothesized that there are two primary causes for the alteration of typical behavior in composite samples exposed to hypoxic conditions in the tests conducted. These are: • The initial burning of the cardboard layer, its combustion efficiency during this stage and the coverage of the subsequent soot and char residual layer created once the cardboard layer has ignited.  • The likelihood of the uniform samples to fail to ignite under hypoxic conditions did not increase. However, thicker uniform samples did demonstrate a greater reduction in peak HRRPUA as the volume fraction of oxygen was lowered than was observed for the thinner sample groups. It is likely that this was due to the change in ratio of 'heat released in flaming combustion' and 'heat losses into the material' when hypoxic conditions are introduced. Greater heat losses into a thicker material become a more relevant factor where the heat production rate from flaming combustion is reduced.
• Composite tests demonstrated that behavioral interactions between different layers of the sample could lead to significantly conflicting results. In numerous tests, the typical thicker/thinner behavior patterns shown in uniform samples were no longer apparent due to the effect of the protective soot and char layer over the ABS as well as the subsequent increase in preheating through exposure to an external irradiance. Test data even suggests that, in certain conditions, the HRRPUA under hypoxic conditions could be higher than otherwise expected in ambient conditions.
• The added complexity introduced by the impact of composite layers also suggests that the required minimum number of repeat tests should be scrutinized by the laboratory team where composite arrangements are of interest.
• Comparisons in t ig appear to show that hypoxic conditions did increase the time to ignition when compared to ambient test data.
However, the change from 17% and 15% O 2 had a negligible effect on t ig for uniform samples. Differences in ignition times were much greater in composite samples. Composite samples saw t ig increase by hundreds of seconds, compared to tens of seconds for uniform samples, where hypoxic conditions were introduced. There was also significant variability introduced to t ig results where composite layers were present.

| CONCLUSIONS
The study conducted has focused on the comparison of ambient and hypoxic atmospheres in bench scale tested samples. The tests carried out aimed to provisionally identify characteristic behaviors that differentiate material performance of uniform/generic materials and those stored in composite arrangements when exposed to an ignition source. The primary aim of the work undertaken was to supplement academic knowledge of material behavior in hypoxic conditions in order to improve the theoretical discussions used to guide ORS design methodologies.
One to provide results that are simultaneously able to appreciate complex interactions between experimental variables while maintaining cost efficiency. Comparisons between small-and large-scale tests will produce data that can potentially prove useful in offering more robust testing methodologies than the 'pass/fail' tests currently used in hypoxic system design but are unlikely to allow the prediction of real world fire events.