Identification and characterization of design fires and particle emissions to be used in performance‐based fire design of nuclear facilities

CERN operates one of the most complex particle accelerator facilities in the world. Several different hazards, including fires, are present and need to be investigated and reduced to a tolerable level. Toward this goal, CERN aims at developing a catalog containing detailed fire dynamics descriptions of combustible items present in its facilities. This paper contributes to this catalog in two ways. First, through the development of a design fire calculator for electrical cabinets that allows the determination of potential design fire curves for any number of electrical cabinets/racks. The second contribution was to experimentally characterize the smoke production rates and smoke particle properties of the most common cables and insulating oils used at CERN by coupling a fast particle mobility analyzer to a cone calorimeter. The two particle size modes (accumulation and nucleation mode) could be linked to the fire properties and heat release rate. Accumulation mode particles (~200 nm) were associated with high heat release rates and high soot emissions from the flame. This study identifies a necessity to consider ultrafine particle emissions with low mass emissions but high number emissions in relation to risk assessments pertaining to nuclear facilities and dispersion of radioactive aerosols to the surrounding environment.

recommendable. Therefore, a performance-based fire design is suggested.
In comparison to specification-based prescriptive design, the suggested performance-based design has three main advantages: 1 It allows the designer to address the unique features and uses of a building. 1 2 It allows a better understanding of how a building would perform in the event of a fire. 1 3 It ensures applying the most effective cost-benefit compensatory measures.
The long-term objective of CERN is to develop a catalog that contains detailed descriptions of combustible items present in its facilities, including their individual and combined heat release curves, as well as the propagation modes and other relevant characteristics (CO and CO2 yields, smoke particle size distributions etc.). This paper presents methodologies for assessing heat release rates in electronic cabinets and for assessing particulate and gaseous emissions. These results can be directly implemented in the CERN catalog.

| Objective
The first objective of this study was to assess the fire hazard imposed by electrical cabinets and racks used at CERN, and to address them with appropriate design fires. The second objective was to characterize the smoke particles that will be released by fires in diverse CERN facilities. The smoke characterization was performed by conducting fire experiments using the most common cables and insulating oils found at CERN.
To reach these objectives: 1 Representative cabinet types used at CERN were categorized and the most common cabinets were compared to the literature/reference values of electrical cabinets fires 2 A simple methodology was developed for calculating the expected design fires for any number and distribution of electric cabinets and implemented in a Microsoft Excel© spreadsheet. 3 An experimental set-up was designed to characterize smoke emissions from laboratory scale experiments using oils and cables by connecting a fast particle analyzer to a cone calorimeter.

| Motivation
Particle accelerator and nuclear facilities are usually complex and unique, with highly extensive and complex equipment installed in their premises. In many cases, standard, off-the-shelf equipment does not fulfill the specific needs of these special facilities. In many of them, most of the equipment used is custom made for unique purposes, which adds extra complexity to assessing all potential hazards, including fire hazards. Cabinets and racks used at CERN vary in geometry, contents, configuration (a single cabinet, 10 cabinets and racks combined in one row, 20 closed cabinets divided in two rows etc. -see Figures 1 and 2) and location (in control rooms above ground or in tunnels 100 m below ground). In case of CERN, as for many other facilities, inspection of most of cabinets is not possible due to operation complexity constraints. Ideally, a design fire gives the heat release rate over time for a certain scenario. When the combustible materials and their heat release rates, and the geometry of the room are known, it is possible to construct a design fire scenario. From these, a conservative "safe" estimate of an envelope case or a worst-case scenario can be evaluated.
The second part of this paper is dedicated to smoke particulate matter analysis of the three most common types of cables used at CERN, as well as two insulating oils used in transformers and klystrons. The term aerosol refers to a suspension of liquid or solid particles in a gaseous medium. Many commonly known phenomena such as dust, suspended particulate matter, fume, smoke, mist, fog, haze, clouds, or smog can be described as aerosols. 2 Smoke from combustion is a mixture of gasses, vapors, and particulates. The particles in smoke include: (a) Nanometer sized nucleation mode particles often in droplet forms, formed from condensed organic vapors, possibly with F I G U R E 1 Open/closed cabinets and racks in a row -picture taken at one of CERN facilities an inorganic core and (b) Carbonaceous accumulation mode soot particles formed in the flame with agglomerated structures and consisting of a large number of partly fused spherical primary particles. 3 In fire models, the accurate prediction of aerosol and soot formation (number, mass and size distributions), as well as aerosol and soot deposition thicknesses on surfaces is important for a wide range of applications, including human egress calculations, heat transfer in compartment fires, and forensic reconstructions of fires. 4 In case of a fire in one of CERN tunnels during experiments, it is expected that smoke particles, produced by burning activated materials, will be radioactive too and will further carry and eventually deposit the radiation on surfaces inside the CERN tunnels or lead to radioactive emissions to ambient air. Particle size controls the lifetime of the particles in the air, for example by controlling deposition velocities. This is a serious threat that likely has to be approached from a worst case scenario. For this reason, CERN has a need to obtain detailed knowledge about smoke particle yields and size distributions, which can be used in flame dynamic simulations and aerosol dispersion modeling. The sizes of inlet and outlet openings determined the amount of oxygen available for combustion, which had the greatest effect on peak HRR. Ventilation conditions had much greater impact to peak HRR in comparison to other parameters of the cabinets that were investigated -namely ignition location, amount of fuel, special arrangement in the cabinet and the cabinet filling. Each of the experimental campaigns resulted in models for peak HRRs where the only variables are inlet and outlet sizes and vertical distance between them.

| PROPOSED DESIGN FIRE FOR MULTIPLE ELECTRICAL CABINETS AND RACKS
At CERN facilities, the electrical cabinets and racks can be distributed from a single closed cabinet to a set of two rows consisting of any number of columns of combined racks and open and/or closed cabinets ( Figures 1 and 2). Therefore, in order to cover the common and possible fire scenarios a simple methodology for calculating the expected design fires for any number and distribution of electric cabinets was developed and implemented in a Microsoft Excel© spreadsheet. In theory, obtaining the accurate heat release rate for a single cabinet is possible only if exact contents and specifications are known, as well as the precise geometry limits of the cabinet. Nevertheless, in F I G U R E 2 20 closed cabinets in two rows -picture taken at one of CERN facilities practice we know from tests 9 that unexpected events in the combustion process can occur (internal combustible failing, moving parts...), and affect the outcome HRR. Therefore, this further reinforces the necessity of making conservative conclusions when estimating electrical cabinet design fires. As seen in Figures 1 and 2, cabinets and racks with their contents are custom made, and the contents, specifications, and geometry vary from case to case. Therefore, estimation of the heat release rates were based on values found in the literature. Following guidelines given in these documents would certainly result in using more realistic HRR values, and consequently having more realistic risk estimations. Nevertheless, it is impossible for CERN to inspect every single cabinet, and also equipment used in CERN is often custom made, as it was explained in motivation Section 1.2. Therefore, following the precautionary principle adopted by CERN, the extreme values proposed in these documents were used in order to simulate scenarios to be used in a conservative risk assessment.
Several models have previously been proposed for obtaining the peak HRR of closed-door cabinets. 6,8,13 First, in Reference 6, a simple cabinet flow model assuming small vents and thus unidirectional flow is analyzed and a dimensional equation for peak HRR _ ðQÞ in the following form is developed: where H (m) is the vertical distance between the vents of the cabinet, while A i (m 2 ) and A e (m 2 ) are areas of inlet and outlet openings respectively.
After a new series of similar experiments, 10 years later, in 2004.
Mangs 13 proposed a similar but improved version of Equation (1), including the combustion efficiency χ that takes into account incomplete combustion: Finally, in 2011 IRSN 8 have considered cabinets with vent areas that are non-negligible in comparison to the cabinet size, and also taken into account acceleration of the fluid due to its heating.
Bernoulli equation yielded the following equation for the steady state mass flow rate (q*): The peak HRR in the cabinet is determined from the oxygen available, that is, from q*: where ΔH c, air = 3.144 MJkg −1 is a constant valid over a large range of fuels at standard conditions of pressure and oxygen concentration. 8 In all three proposed equations, values of HRR strongly depend on the cabinet height. Largest cabinets at CERN are 3 m high, which can be seen as an envelope case when determining HRR according to these equations. Peak heat release rates for a 3 m high cabinet are calculated according to all three equations proposed, for variety of vent size ratios and the results are shown in Table 1.
[Correction added on 5 August 2020, after first online publication: Table 1 citation has been added.] The peak HRR value for majority of inlet/outlet ratios is obtained using Equation (2) -referred to as Mangs 2004. When using Equation (2), combustion efficiency is taken to be χ = 0.7, which is seen as a conservative assumption as fires in closed cabinets are virtually certain to be under-ventilated. As it can be seen in summary of cabinet experiments in Reference 8, the most extreme case of ventilation sizes was A in = 0.1 m 2 and A out = 0.1 m 2 . In general vent sizes were much smaller.
Thus, peak HRRs were analyzed for the vent areas ratio of up to that size. It can be seen that even for the most extreme case when both inlet and outlet areas are equal to 0.1 m 2 , the peak HRR value is 494 kW, which is still below the 500 kW, which was the value taken as PEROVI C ET AL. a peak HRR value for closed cabinets proposed by Reference 10 and used in the Excel calculator. This validates the assumption of peak HRR for closed cabinets and further validates the excel calculator.
The equations supporting Table 1 presume that the fire size is limited to the combustion air entering the target. However, external burning could occur at the top vent allowing for larger HRR if sufficient pyrolysis can occur inside the cabinet. This limitation should be kept in mind. Nevertheless, it seems that test data, which shows HRRs <500 kW suggests that ventilation does not support significant excess pyrolysis.
In real, growing and ventilated fires, the initial fire development is nearly always accelerating, 15 in comparison to for example smoldering fires, that can smolder for a very long time, prior to the initiation of a potential accelerating growth and expansion. A simple way to describe the accelerating growth is to assume that the energy release rate increases as the square of the time. By multiplying time squared by a factor α, various growth rates can be simulated, and the energy release rate as a function of time could be expressed as: where α is a fire growth rate (often given in kilowatts per second squared [kW/s 2 ]) and t is the time from established ignition, in seconds. The fire development in all the cabinet fire experiments was slow according to the NFPA classification. 6 Therefore α = 0.003 kW/ s 2 was chosen.
The decay phase HRR curve can be characterized according to the following exponential function 16 : where _ Q td ð Þ is the heat release rate at the time of the start of the decay phase, t d is the time at the start of the decay phase, and τ is the decay time constant. In Reference 13, the decay time constants used to fit the exponential functions that describe the decay were in the range between 13 and 23 minutes. As no additional data is available, it was decided to take the mid value, that is, τ = 18 minutes. Using a decay time constant τ = 18 minutes gives a decay time of approx. 50 minutes. Decay time is defined as the time from when the fire starts decreasing from the peak value, until the fire dies out. Comparing the adopted value of 50 minutes to decay times between 18 and 40 minutes found in Reference 5, proves that our assumption is conservative.
We now know the HRR peak values for both closed and open cabinets and the way to describe both the fire growth and decay fire stages. It now remains to determine the duration of the HRR peak. In the case where the amount of fuel is not known, the duration of burning at the peak HRR (steady stage burning time -t s ) needs to be determined. Kassawara 10 recommends t s = 8 minutes to be taken.
Following the precautionary principle adopted by CERN a 50% higher t s (t s = 12 minutes) is used. In the case of closed cabinets, as their peak HRR is 50% smaller than peak HRR of open cabinets, it is assumed that duration of steady burning (t s ) will be 2 times longer than t s of open cabinets. Thus, t s for closed cabinets is taken to be t s = 24 minutes. This is decided according to the engineering estimation that the same amount of fuel will take twice as long to burn if the peak HRR is 2 times smaller. The final HRR curves for closed and open cabinets are presented in Figure 3A,B.
Once that the burning behavior of a single cabinet is completely addressed (growth stage, peak HRR, duration and decay stage), we only need to describe the fire spread between the adjacent cabinets in order to obtain a complete methodology to assess the HRR of a fire of multiple electrical cabinets. Fire spread to the adjacent cabinet occurs in 11-16 minutes according to the experiments done in References 6 and 13. To be on the safe side, it is assumed that the fire will spread to the adjacent cabinet after 10 minutes. The mode of fire spread are conduction and radiation. The walls of the fire cabinet heat up, then conduct the heat to the adjacent cabinet wall, which finally heats up it's contents which then catch fire. In the experiments, the bundle of cables with outer sheaths coated in PVC The worst-case scenario is when a fire starts in a middle cabinet.
In a simple one-row case, fire would spread to two adjacent cabinets every 10 minutes. In a more complicated case with 2 rows, after ignition of the first adjacent cabinets, further cabinets in the row opposite from the "fire cabinet" row would be heated by 2 cabinets at a time, thus it would take less than 10 minutes for each of them to ignite. An approximation of the time until the ignition of each of the cabinets is given in Figure 4. It is observed that cabinets sometimes have glass facing (Figure 1).
In case the cabinet has plastic or glass facing, it is advised to consider it as an open cabinet due to the conservative assumption that the facing would inevitably fail leading to unlimited oxygen supply inside of the cabinet.
Also, the proposed calculation methodology is valid only for adjacent cabinets, that is, cabinets that have the walls touching each other. This is the most common case in CERN, as it was shown in Figures 1 and 2. Spread to opposite cabinets is not covered in this work.
A schematic of the Excel calculator for design fires in electrical cabinets in CERN facilities is presented in Figure 5. • The mode of heat transfer for spread to adjacent cabinets is assumed conduction.
• The cabinets are assumed to be adjacent (ie, next to each other without air gaps).
• If the amount of fuel is known, combustible fraction will determine the duration of the steady burning.

| Methodology
The goal of the experimental campaign was to assess relationships between HRR, smoke production rates, and smoke particle size distributions for a set of specimens commonly used at CERN (three type of cables and two types of oils). In order to capture the transient com- ticle size is used as input to an inversion matrix that provides the number based particle size distribution. The data reported here are fitted using two lognormal size modes, a smaller nucleation mode and a larger accumulation mode. Fire smoke particles may have complex shapes and the definition of particle size is not trivial. As the particle sizing in the DMS is based on electrical mobility, the size measure is the equivalent mobility diameter. The mass concentration and mass weighted particle size distribution were estimated using assumptions of the particles mass-mobility relationship as discussed below.  Table 2:

| Samples and preparation
The tested oils were the most common ones used as electric insu-   (Figures 8B and 9B).

| Results
The nucleation and accumulation modes apparent from Figure 8 were determined by lognormal fits to the data. The time evolution of HRR, nucleation mode and accumulation mode particle number concentrations are shown in Figure 9A,  The 3D plot in Figure 10 shows the evolution of particle size distributions over time for C01 cables (blue). In the cable example, the nucleation and accumulation modes are apparent for different burning modes of the cable. 3D plots for the remaining fuel types -C02 (black) cables and C04 (brown) cables are shown in Appendix A. Figure 11 a-c shows time series of the heat release rate (HRR) and the nucleation and accumulation mode particle number concentrations from a single test with each of the three tested specimens.
Looking at C01 (blue) cable results in Figure 11A, upon ignition, a short period of moderate HRR and bimodal size distribution is visible.
After this follows a longer period (200-400 seconds) with reduced HRR, where nucleation mode particles are dominant. We interpret this as the outer sheath burning. In the time period 400 to 900 seconds the HRR is increased and accumulation mode particles became dominant. We interpret this as burning when fire caught the inner sheath. Thus, nucleation mode particles observed in the first part of burning represent burning of the outer sheath of cable.
Accumulation mode particles present in second part of burning represent burning of inner wires' sheath. Accordingly, the main (outer) cable sheath and inner wire sheath are made of different materials. The inner sheath produces larger particles, which is most likely connected to the significantly higher burn ratio (higher HRR) and formation of soot in the flame. This also suggests that while the outer sheath shows some positive fire rating, the inner sheath contribution worsens the global performance. Therefore, peaks in aerosol emissions and smoke production rates in cable fires can be expected when the fire reaches the inner wires' sheath and the soot formation rate increases with the HRR.
When comparing these results with the other two cable types the main features of the HRR and particle concentration trends cables. This may be related to different thicknesses of outer sheath material, with high releases of non-combusted pyrolysis gasses being a possible origin of these short-lived peaks of nucleation mode particles.
Particle number and mass concentration graphs, obtained using an estimation of the particle effective density vs size, from a single test on C01 cables (blue) are shown on Figure 12. Figure 12A shows the number weighted particle concentration (average and all the time points) demonstrating bimodal particle distribution.
The smaller particles are more likely to be spherical while the larger ones might be highly irregular as they often consist of an agglomeration of smaller primary particles. 19 In Reference 20, an aerosol particle mass analyzer was used to measure the mass of diesel exhaust particles as function of mobility diameter. They determined the effective density and found that it decreases as the particle size increases. By observing TEM (transmission electron microscopy) images, they realized that this phenomenon occurs because the particles become more highly agglomerated as size increases. Later data has shown that this phenomenon is similar for a range of soot particle sources. 21 By using effective densities, ρ eff [Equation (7)], of diesel exhaust particles, the number concentration, N i , in each size channel was converted to the mass concentration, m, in the size channel for our tests and the result is shown in Figure 12B. It is important to have in mind that the mass concentration results are uncertain, as mass of smoke particles and their effective density for cables and oils used were not known, thus the values of diesel exhaust were used as an F I G U R E 1 1 Single experiments heat release rate (HRR), nucleation mode particle number, and accumulation mode particle number -A, C01 (blue) cables; B, C02 (black) cables; C, C04 (brown) cables approximation. Also, when comparing Figure 12A,B, it is interesting to note how nucleation mode particles (smaller than 100 nm), even though high in number concentration ( Figure 12A), have negligible contribution to the size distribution based on total mass concentration ( Figure 12B). This is because of the low mass of the smaller nucleation mode particles. Mass is calculated according to the Equation (7). 20

| Repeatability of the tests
Overall, the repeatability of the tests had proven to be high. Figure 14A shows HRR plot for 3 repeats on Midel oil, while Figure 14B shows 3 repeats for total particulate concentration (nucleation + accumulation) for the same oil. Figure 15 shows HRR and total concentration repeatability for However, the charge level for a given particle size in the DMS depends on particle shape, especially for larger particle sizes. 18

| Summary and conclusions
In the first part of this paper, an Excel calculator that provided design fires for electrical cabinets and racks present in CERN was developed.
Findings from experiments conducted in the USA, Finland and France were used to obtain an envelope case covering the worst possible conditions. The user can specify any combination between the most basic one -a single cabinet, to the most severe one -2 rows containing 10 cabinets each.
In the second part, an experimental campaign was conducted on three types of the most common cables used at CERN, as well as on two of the most common insulating oils used at CERN. Our results demonstrate that in these cable fires, nucleation mode particles were emitted during low heat release rates primarily during burning of the outer sheath of the cable or at the final stage when nearly all material had been combusted. Accumulation mode particles were dominating the particle emissions (ie, smoke) at high heat release rates in later stages of burning when the sheath of inner wires was burning. Accumulation mode particles were always dominating the particle mass emissions, and their concentration correlated well with the smoke production rates derived from obscuration measurements. Nucleation mode particles were often found to dominate the particle number emissions. Particle emissions, both number and mass, from the oils were dominated by accumulation mode particles. However, a burst of nucleation mode particles was observed toward the end of combustion for the high flashpoint oil sample, suggesting that nucleation mode particles may also be important emissions from fires in the klystron and transformers.
It is instructive to compare our results to recently published data on rich biomass combustion emission and fire properties investigated with a combination of a modified cone calorimeter and the DMS500. In addition, that study showed a bimodal size distribution, with the fraction of accumulation mode particles increasing with increasing HRR. 24

| Future work
The literature on fires in electrical cabinets is existing, but limited. To The goal of the experiments was to obtain detailed smoke characterization data of key specimens, which is required to initialize and further develop CFD modeling software that can resolve aerosol agglomeration and deposition to better model particle dispersion. Particle accelerators and nuclear facilities might be interested in these results; smoke particles are expected to carry radiation further away from the seat of a fire of activated material, which is a potential hazard both for the facility and for the environment. Our results show that the nucleation mode particles observed with the DMS500 did not contribute significantly to the total aerosol particle mass nor to the aerosol light extinction. However, our measurements reveal that these ultrafine particles (<100 nm) can dominate the particle number emissions from cable fires. With respect to fires in environments susceptible to ionizing radiation, our measurements show that consideration of these ultrafine particle emissions may be crucial for accurate fire-propagation and aerosol dispersion models. In particular, chemical and elemental analysis of these particles can reveal whether they have high probabilities of containing radioactive isotopes during or after exposure to ionizing radiation. Smoke emissions from various sources to ambient air are also of relevance for adverse health impacts and effects on global climate. 25 The adverse effects can be related to the particle properties such as size and chemical composition.