Quantification and prediction of water uptake by soot deposited on ventilation filters during fire events.

Soot samples from different fuels were produced in small and pilot combustion test benches at various O2 concentrations, and were then characterized in terms of primary particle diameter, specific surface area and oxygen content/speciation. Water sorption measurements were then carried out for soot compacted into pellet form and in powder form, using both a gravimetric microbalance and a manometric analyser. Water adsorption isotherms are all found to be Type V, and reveal the central role of the specific surface area and the oxygen content of soot. A single parametrization of the second Dubinin-Serpinsky model gives a proper fit for all isotherms. To the best of our knowledge, this is the first study to provide physico-chemical parameters and water sorption results for fire soot. This enables a better description of the soot cake formed on filters during a fire, in particular in industrial confined facilities as simulated in this study. Humidity can be then explicitly considered in the same way as other parameters influencing the aeraulic resistance of soot cakes. These results can be used to improve predictions of the consequences of fires on the containment of toxic materials within industrial facilities.

Introduction

The need for efficient filtration devices is of major importance in relation to the containment of hazardous materials, such as nuclear materials, nanoparticles or pathogens at basic nuclear installations, nanomaterial manufacturing plants and, in light of the SARS-CoV-2 pandemic of 2020, at biological/virological research institutes, respectively. In addition to their initial efficiency, which is generally of a high standard, characterized as High Efficiency Particulate Air filters (HEPA), their performance must be maintained even in hazardous situations which could occur within a contained or ventilated industrial or research facility. Furthermore, dramatic fires such as of fires at the Notre-Dame Cathedral (15–16 April 2019 in Paris, France), Lubrizol (26 September 2019 in Rouen, France) and in contaminated surroundings at the Chernobyl site (04−05 April 2020, Chernobyl exclusion zone, Ukraine) highlight the need to develop tools capable of predicting the consequences of such fires in terms of the dispersion of hazardous materials in the atmosphere (Ouf et al., 2020; Howes, 2019; Evangeliou et al., 2016). The exposure of firefighters and populations to such hazardous airborne particles dispersed during wildland or industrial fires is also of societal importance and raises questions regarding the performance of personal protective equipment in such accident situations and, more specifically, regarding filtration efficiency and clogging on filtering facepieces.

Soot particles, defined by Petzold et al. (2013) as agglomerates of monomers consisting solely of carbon with small amounts of hydrogen and oxygen, are inevitably produced during combustion processes encountered during wildland or industrial fires. In the case of facilities where hazardous materials are manufactured or handled, soot emitted during a fire can be confined in these facilities thanks to HEPA filters, but massive emission can clog those filters due to the formation of “cakes” (Mocho and Ouf, 2011), increasing the aeraulic resistance of the air flow passing through the filters. This increases the mechanical strain applied to the structure of the filters, leading in extreme cases to their rupture. Similar phenomena could also occur in the case of personal protective equipment worn by firefighters or the general public, which could increase the pressure drop of the filtering part, exacerbating leaks and, as a consequence, decreasing the protective function of these devices. For each of these applications and especially with regard to the safety analysis of nuclear facilities, it is crucial to be able to describe this clogging phenomenon in order to predict the consequences of fire on the containment of radioactive materials. A key parameter influencing filter clogging is the porosity of the cake, formed by interstitial spaces between soot aggregates. Typical porosity of soot cake has been reported above 95 % (Thomas et al., 2014). The aeraulic resistance of soot cakes also depends on humidity (Gupta et al., 1993; Joubert et al., 2011), which can reach high values when water spray systems are used to extinguish fire. The presence of organic vapours also influences aeraulic resistance (Mocho and Ouf, 2011). Thus, refined clogging models must take into account the dynamic regime of the clogging process, including high humidity variations and structural changes to which the filter and the cake may be subjected (Lazghab et al., 2005; Yu et al., 2003). Furthermore, the cake can be restructured in the presence of liquid water due to capillary condensation (Schnitzler et al., 2017; Adachi et al., 2010), which also affects aeraulic resistance (Ribeyre et al., 2014). However, not all these phenomena have been specifically modelled for soot emitted during a realistic fire. Indeed, due to their morphology and their potential hydrophilic character, fire soot particles can strongly adsorb water molecules, leading to such capillary condensation. To develop a relevant clogging model, it is necessary to determine the water uptake by fire soot as a function of humidity and, in this context, it is particularly interesting to determine the transition between adsorption and capillary condensation. It is also relevant to understand the effect of the fire conditions (O2 concentration, air flux) on the water uptake, as these conditions can influence the physico-chemical properties of soot (Ouf et al., 2015). Water uptake measurements have been performed for chemically- or thermally-treated synthetic carbons, such as activated carbon (Pastor-Villegas et al., 2010; Choma et al., 1999), mesoporous carbon (Horikawa et al., 2015) and commercial carbon blacks (Kiselev and Kovaleva, 1959; Carrott, 1992), but have never been reported for fire soot. Carbon blacks could be considered, from a size and morphological point of view, as relevant surrogates of soot (Ferraro et al., 2016), but not from a chemical point of view since they are mostly composed of elemental (graphitic) carbon, while combustion soot usually indicates significant oxygen (Ferraro et al., 2016) and organic content (Ouf et al., 2015). Water uptake has been also measured on reference soot, emitted by laboratory burners or engines running on various fuels, including diesel and kerosene (Popovicheva et al., 2008; Chughtai et al., 1999). In all cases, the main parameters influencing the water uptake are the sample porosity and the chemical composition of the particles, and the main adsorption mechanism is the formation of water clusters at hydrophilic adsorption sites (Liu et al., 2017). Numerous models for water adsorption on carbon have been proposed, especially for very porous activated carbon (Furmaniak et al., 2008). Among these, the Dubinin-Serpinsky (DS) adsorption model (Dubinin et al., 1955) is valid for both porous and non-porous carbonaceous material. In the case of cakes of fire soot, the DS model seems relevant, as water adsorption can take place at the surface of the non-porous soot aggregates forming the cake.

Using small and pilot combustion test benches, we have produced fire soot under conditions representative of fire events (Ouf et al., 2015; Alibert et al., 2017). Different O2 concentrations of the oxidizing gas have been used to mimic the real case of a poorly ventilated and confined fire, representative of an industrial facility handling hazardous materials (biological or radioactive nano/microparticles). For applications in the field of nuclear safety, we have considered fuels commonly found at nuclear facilities or that have been extensively studied in the past (Ouf et al., 2015). Soot was collected on polycarbonate membranes in order to perform physico-chemical analysis and sorption measurements. Transmission Electron Microscopy (TEM) image analysis (Bourrous et al., 2018), nitrogen sorption measurements (Sing, 1985), elementary and X-ray Photoemission Spectroscopy (XPS) analysis (Parent et al., 2016) have been used to determine the diameters of the soot primary particles, the Brunauer-Emmett-Teller (BET) specific surface area and the bulk and surface oxygen contents, respectively. Water sorption measurements were carried out for soot compacted into pellets and in powder form, using a gravimetric microbalance and a manometric analyser, respectively. Regarding temperature, the influence of this parameter on water uptake is still up for debate in the literature. Liu et al. (2017) mentioned a complex effect associated with a threshold temperature (defined as the temperature above which water adsorption becomes weaker as the temperature increases) of 298 K. In our study, and as a first step for safety applications, we have investigated conditions associated with the last filtration barrier placed at the outlet of industrial ventilation networks, generally mounted several meters from the fire. In such conditions, temperature could be close to ambient due to dilution and cooling of gases within the ventilation ducts. Given the lack of knowledge on the effect of temperature, and with a view to investigate the most favourable conditions for water uptake on soot cake, a value of 25 °C (298 K) has been considered for this experimental study.

Experimental procedure

Soot production

Soot samples were produced in two cone calorimeters, one at laboratory or “lab” scale and one at “pilot” scale, differing in their dimensions, their oxidizing gas flowrates and the residence times of the emitted particles inside their combustion chambers (Fig. 1 ). The laboratory scale calorimeter consisted of a combustion chamber of 0.03 m3 topped by a column in which the combustion aerosol is transported to the sampling point. We used 40 mL of heptane (Sigma-Aldrich) or DTE Medium (Exxon Mobil), a hydraulic oil used in the French nuclear industry, as fuel, placed in a cylindrical container with internal diameter of 5.7 cm. We also tested polymethyl methacrylate (PMMA), the major compound of glovebox walls usually used for the containment of hazardous materials (along with polycarbonate). The total flowrate of the oxidizing air was set at 10.8 Nm3 h−1, and the O2 concentration was varied by changing the air to nitrogen ratio with two mass flow controllers (model 5853S, Brooks). We chose three O2 concentrations of 15 % (highly depleted air), 18 % (depleted air) and 21 % (ambient air) in the oxidizing gas for fires with heptane, Hydraulic oil and PMMA. For each fuel, the soot samples are named according to these oxidation conditions (i.e. Heptane 15 %, Heptane 18 % and Heptane 21 %). Soot was collected on a cellulose acetate membrane (type 11106, Sartorius Stedium Biotech) placed in a high-volume air sampler (TE-2000PX, TISCH Environmental Inc, Ohio) at a flowrate of 8 Nm3 h−1. In order to carry out different ex situ analyses, soot was kept in a dry hermetically-sealed desiccator container, and away from light. The calorimeter at pilot scale has a significantly larger size and longer residence times compared to the laboratory scale calorimeter (Alibert et al., 2017). PMMA sheets and heptane pools were used as fuels, and their respective dimensions are reported in Table 1 .

Fig. 1

Experimental setup of the controlled atmosphere cone calorimeters at lab scale (left) and at pilot scale (right).

Table 1

Summary of operational conditions for fire tests in the laboratory and pilot scale cone calorimeters.

			Laboratory scale	Pilot scale
Combustion chamber volume (m3)			0.03	22
Oxidizing gas flowrate (Nm3 h−1)			10.8	1500
Residence time (seconds)			10	53
Flowrate of high-volume air collector (Nm3. h−1)			8	60
Fuels and dimension (cm)	PMMA plate		5 × 5 × 1	40 × 40 × 3
	Liquid fuel in a cylindrical container	Heptane	∅: 5.7	∅: 21
		Hydraulic oil		–

Physical and chemical analysis of the samples

The true density of the primary particles of carbonaceous aggregates was determined using a measurement technique based on the displacement of a liquid induced by the immersion of a known mass of sample. This method, described in the ISO 787-23 standard and based on the Archimedes buoyancy principle (using ethanol as the displacement liquid), has recently been demonstrated to be relevant regardless of the type of soot sample used (Ouf et al., 2019). The diameters of the soot primary particles were determined by transmission electron microscopy (TEM). For TEM sampling, soot particles were first diluted in ethanol and mixed in an ultrasonic bath for several minutes to give a homogeneous suspension. A microlitre of this solution was deposited on TEM grids (holey carbon film on 300 mesh Cu (X25), S-147-3H grids from Agar Scientific®) and left to dry. Soot micrographs were recorded with a Jeol 100CXII microscope equipped with a CCD camera (Gatan® Erlangshein Dualvision 300 W, 780 model). A hundred of TEM images were analysed for each sample, both manually and automatically using ImageJ software and a semi-automatic software (Bourrous et al., 2018), respectively. Measurements of nitrogen sorption at 77 K using a manometric analyser (ASAP 2020, Micromeritics) provided the specific surface area of each sample from the conventional BET analysis (Gregg and Sing, 1982) and for relative pressure of nitrogen (p/p0) ranging from 0 to 0.30 (0 to nearly 300 mbars). Prior to the measurements, the powdered samples were pumped into a primary vacuum (0.01−0.10 mbars) during at least 12 h at a temperature of 25 °C. The elemental compositions of soot were determined with an organic elemental analyzer (FlashEA 1112, Thermo Scientific). Carbon, hydrogen, nitrogen and sulfur (CHNS) contents are inferred from the gas analysis emitted during a flash combustion at 920 °C under oxygen. According to previous studies on carbon black samples (Pastor-Villegas et al., 2010; Ferraro et al., 2016; Miura, 2016), the soot oxygen content [O]diff (in weight %) can be determined from this analysis. In the present study, soot particles were produced from fuels mostly composed of carbon and hydrogen, and we thus expect to detect no other elements than C and H, except nitrogen and oxygen resulting from the reactions with the oxidizing gas (Ferraro et al., 2016; Boehm, 1994). The surface oxygen concentration [O]XPS of several soot produced at laboratory scale was determined by X-ray photoelectron spectroscopy (XPS). Prior to the XPS analysis, the samples were compacted into 7 mm diameter pellets using a hand press (Pike Technologies®). The experiments were then performed under ultra-high vacuum using a Resolve 120 hemispherical electron analyser (PSP Vacuum) and a TX400 (PSP vacuum) unmonochromatized X-ray source (Mg Kα at 1253.6 eV) operated at 100 W. The surface oxygen concentration was obtained from the quantification of the main C1s and O1s lines of the survey spectra ([O]XPS= [O]/([O]+[C]), after correction from the relative sensitivity factors provided in the CasaXPS program used for the XPS analysis (Parent et al., 2016). The XPS C1s and O1s lines were further deconvoluted using Gaussian/Lorentzian profiles and after Shirley-type background subtraction. A common set of graphitic (Csp2 and shakeup) and carbon oxides (C-O, C=O and HO-C=O) contributions was used for the C1s line. The O1s line was fitted with a set of conjugated C=O (quinone), C=O (carbonyl), C-O (ether), OH-C=O (carboxyl) and water contributions (Niessner, 2014) (see Supplementary Information I). These analytical methods have been used in preference to thermo-desorption analysis of organic carbon to elemental carbon ratio (OC/EC) since this method is not specific to surface composition and not only includes oxygen-containing species but also alkyl or aliphatic groups (Ess et al., 2016) as examples.

Water uptake measurement

For water uptake measurements, soot samples were either in their natural powdered form, or compacted into a cylindrical pellet using a laboratory-made press functioning with a torque wrench (torque set at 0.5 N.m). Knowing the true density of soot, the global porosity (Table 2 ) of non-compacted εpowder or compacted samples εpellet can be deduced either from the volume of the powder in the cylindrical glass container, or from the diameter and the height of the pellet, respectively. Water sorption measurements were performed using gravimetric and manometric methods (Emmett and Anderson, 1945; Wolf et al., 1984). These agreed whether conducted under static or dynamic sorption conditions (Arlabosse et al., 2003; Belmabkhout et al., 2004). Gravimetric measurements were performed only on pellets using a “dynamic vapor sorption” (DVS) Vacuum microbalance (Surface Measurement Systems, SMS). Pellets were first pumped into high vacuum (10−6 mbar) for 6 h at 25 °C to remove water and adsorbed impurities at the sample surface. For water sorption measurements, this degassing is recommended (Snoeck et al., 2014) over the conventional thermal pre-treatment, which can alter surface properties, for instance by removing of hydrophilic adsorption sites (such as carboxyl groups reported on our samples by XPS, see SI-1, and decomposing at a temperature close to 473 K (Figueiredo et al., 1999)). By monitoring high vacuum pressure and sample mass, the initial mass of the dried sample was determined after 6 h of outgassing, which was enough to reach mass equilibrium (with a mass change of less than 0.05 wt.% for a time period of 10 s). Subsequently, humidity steps were gradually applied at a constant water vapour flowrate and at a constant temperature of 25 °C. Water was swept over the sample with a limited residence time in the microbalance system cell, enabling continuous renewal of the vapour phase in contact with the soot surface. Relative humidity RH (as a %) in the microbalance is defined as the ratio between partial pressure of water PH2O,vap within the cell divided by saturation pressure PH2O,sat (Eq. (1)) at the measurement temperature T:

Table 2

Summary of the physico-chemical properties of soot and carbon black samples studied.

1From Ref. Ouf et al. (2019).

2From manufacturer.

Under the conditions studied (25 °C), a RH range of 0–95 % was considered, corresponding to partial pressure of water in the measurement chamber ranging from 0 to 22.6 torr (0–30 mbar). Water uptake a(RH) (Eq. (2)) is then defined as the ratio of the mass of water adsorbed (mH2O, adsorbed), determined according to sample mass at each relative humidity msample(RH), and the reference mass mreference, which is measured at RH = 0%. It is worth noting that, in the rest of this study, water uptake will be presented in percent (by multiplying a(RH) by 100).

The transition between two humidity steps depends on the time needed to reach the thermodynamic equilibrium and stabilization of the sample mass with an accuracy of 0.1 μg, according to SMS. Given this value and the uncertainty propagation principle, the water uptake uncertainty is less than 10−4 %. After a preliminary optimization phase to identify the most relevant parameters to take into consideration as an equilibrium criterion, time duration of humidity steps was set at 6 h for 0 %≤RH<40 % and 10 h for 40 % ≤ RH (since equilibrium takes longer time to reach for higher RH values). For soot and carbon black samples, whose expected affinity with water is low, these time criteria were relevant in order to reach, at the end of each relative humidity step, a mass change of less than 0.003 wt.% for a time period of 60 s. This gives the evolution of mass over time and the maximal water uptake for a defined humidity (Fig. 2 ) which ultimately, provide the sorption data needed to plot the sorption isotherms to a very high level of accuracy. Manometric measurements were performed on non-compacted powered soot only, using a 3FLEX analyser (MICROMERITICS). Soot samples were first degassed in a cell at 0.1 mbar and 25 °C using the VacPrep 061 low vacuum pump. The cell was then placed in the 3FLEX analyser, and water was pumped through the cell. Pressure measurements were then performed only when a stabilization criterion of 0.01 mbar min-1 was reached.

Fig. 2

Example of gravimetric measurement using the DVS microbalance (with evolution of sample mass in red and evolution of relative humidity in blue).

Qualification of water uptake measurement protocol

Prior to taking the measurements on soot, the experimental protocols and apparatus were validated using microcrystalline cellulose (MCC) as a reference. For MCC, gravimetric, static and discontinuous sorption data are available for two kinds of sample: MCC Avicel Ph-101 provided by FMC (Wolf et al. (1984)) and MCC RM 302 (Jowitt and Wagstaffe, 1989), which are used as references for the COST90 European standard procedure. Due to the significantly higher water affinity of MCC, the equilibrium criterion for this sample was fixed at 0.05 wt% for a time period of 10 s. Fig. 3 shows the water uptakes of these two MCCs following the COST90 procedure at different RH obtained with salt solutions (Wolf et al. (1984)), and those obtained with the DVS microbalance at the same RH steps (Levoguer and Booth, 2014). Our results are in agreement with the COST90 data, except for slight discrepancies at high water uptakes for the MCC Avicel Ph-101. This can be explained by differences in the samples, such as size distribution or specific surface area (Doelker, 1993). Fig. 4 compares water uptakes obtained on the MCC Avicel Ph101 in powdered form using the 3FLEX analyser (circles) and those obtained using the gravimetric (DVS) methods (squares). As can be seen, there is good agreement between the methods, as reported by previous authors (Emmett and Anderson, 1945).

Fig. 3

Comparison of water sorption data on different microcrystalline celluloses (MCC Avicel Ph-101 and RM302) obtained with the COST90 procedure and DVS microbalance.

Fig. 4

Water adsorption isotherms of MCC Avicel Ph-101 obtained with the DVS microbalance (squares) and with the 3FLEX analyser (circles).

Since the quantities of soot were limited, experimental isotherms were generally determined once. Prior to this determination, the experimental repeatability of the DVS microbalance was checked using the MCC and two commercially available carbon black samples as size/morphological and size/morphological/chemical surrogates for these preliminary tests, namely Printex 90 and FW200 (characterized with oxygen content close to those reported for soot particles) from Orion®, respectively. Fig. 5 shows the coefficient of variation (Eq. (3)) for MCC Avicel Ph-101, Printex 90 and FW200 for respectively seven, five and three repeated measurements. The coefficient of variation is less than 10 % at water uptake higher than 1%, indicating excellent repeatability of the measurement.

Fig. 5

Repeated water sorption measurements for different samples (MCC Avicel Ph-101, PRINTEX 90 and FW200).

Results

Sample properties

Measured soot densities ρtrue range from 1492 kg.m−3 to 1780 kg.m−3 (Table 2). Upper values are in agreement with the literature for soot having a low oxygen content (Dobbins et al., 1994; Newman and Steciak, 1987). Lower density values measured for some of our samples can be explained by the presence of an organic carbon phase, which is less dense than the elemental carbon phase, reducing the overall density (Ouf et al., 2019). The diameter of the primary particles dpp ranges from 22.3 nm to 43.3 nm (Table 2), in agreement with values reported for soot emitted by gaseous flames (Prado et al., 1981; Megaridis and Dobbins, 1990), and for more complex liquid and solid fuels (Ouf et al., 2015). For all samples, particles indicate fractal morphology typical of soot particles and examples of TEM images are available in Supplementary Information II (Tables SI-1 and SI-2). Close agreement, in terms of size and morphology could then be reported between carbon black and soot samples. The specific surface area SBET ranges between 52 m² g−1 and 100 m² g−1 (Table 2). This is also typical of non-porous adsorbents and agrees with values determined for soot emitted in various combustion processes (Ferraro et al., 2016; Popovicheva et al., 2008; Chughtai et al., 1999; Levitt et al., 2007). This surface area is mostly due to the surface developed by the primary particles composing soot aggregates (Bourrous et al., 2018). The smaller the primary particle diameter, the higher the specific area. This explains the high surface area of PRINTEX 90 (341 m² g−1), whose particles are small (25 nm) compared to FLAMMRUSS 101, whose surface area is 24.4 m².g−1 because of large particles of 136 nm. We note that the SBET of soot from the hydraulic oil fire is slightly lower (53.3–54.2 m² g−1) than the soot produced using PMMA and heptane (75.5–97.9 m² g−1), due to the larger diameters of their primary particles. The elemental oxygen content [O]diff ranges between 6.4 wt.% and 11.8 wt.%, in agreement with values commonly reported in the literature (Ferraro et al., 2016; Liang et al., 2014). The oxygen content is higher (up to 10 wt.%) in soot than in the carbon blacks, which are known to be mostly composed of elemental carbon with oxygen concentrations of less than 2 wt.%. The oxygen concentration at the surface of the particles [O]XPS (Table 2) is in good agreement with the oxygen content found in the bulk [O]diff, indicating a homogeneous distribution of oxygen within the particles. Deconvolution of C1s and O1s spectra highlights significant amounts of carbon and oxygen functionalities at the surface of soot particles (see Supplementary Information SI-III) with major abundance of carbonyls C=O and C-O in ethers and alcohols, regardless of the fuel or [O2] concentration. Lower additional contributions of quinone C=O and carboxyl OH-C=O groups were also reported. This limited contribution of carboxyl groups (nearly 10 % of all oxygen groups reported) partially explains the poor water affinity of the studied samples, since beyond all the oxygen functions reported on soot and carbonaceous surfaces, carboxyl OH-C=O is known to exhibit water affinity at a higher order of magnitude (Liu et al., 2017) than any other oxygen group (carbonyl, hydroxyl, etc.). Furthermore, O1s analysis shows a water contribution of less than 10 % of all oxygen functions, confirming a limited amount of water (given an oxygen content of soot below 10 %, see Table 2) prior to outgassing and initially adsorbed on samples.

Global porosity ε (Eq. (4)) was been estimated using true density ρtrue of soot particles, the sample mass m, the radius r and height H of the cylindrical soot pellet or of the glass container for the uncompacted samples. For soot pellets, the global porosity εpellet ranges between 29 % and 66 %. This wide range of values probably results from different adhesion properties of each sample during the compaction process. The global porosity of the powdered soot samples εpowder is around 96 % for all samples, close to typical soot cake porosities reported on HEPA filters (Thomas et al., 2014).

Water isotherms of soot particles produced under different fire conditions

Fig. 6 presents the gravimetric water adsorption isotherms for soot compacted into pellets. The water uptake values were obtained considering a reference mass m as the mass of sample at RH = 0 %. Overall, all the samples present the same slopes and the water uptake ranges from 2.4 % and 3.6 % (at RH = 90 %). This is due to the fact that all forms of soot have similar physico-chemical properties in terms of structure and composition. From RH = 0 % to 80 %, the isotherms could be reasonably assumed as Type V isotherms according to the IUPAC classification (Rouquerol et al., 2014). They present a rather slight convex curvature at RH < 30 %. On the isotherms, this convexity combined with a positive slope is characteristic of low interaction with water at the surface, with locally high interaction with some hydrophilic adsorption sites of the soot surface (Velasco et al., 2016). A slight inflection point is observed at RH = 80 % for the Heptane 15 % soot obtained at pilot scale (Fig. 6 left, red triangles), revealing the beginning of capillary condensation. Hysteresis loop was found for most of the samples in pellet form (see Supplementary Information IV), confirming type V isotherms. Note a restricted area of hysteresis loop for all samples, mainly demonstrating the presence of pores with a width, as reported by previous authors (Liu et al., 2017; Nakamura et al., 2010), close to 0.5–2 nm.

Fig. 6

Water adsorption isotherms for soot produced with liquid fuels (left) and PMMA (right).

Water uptakes measured on the hydraulic oil soot samples is slightly higher than for the heptane and PMMA soot samples. Fig. 7 compares the water uptake obtained on fire soot samples produced at laboratory scale and at pilot scale. Except for the heptane soot produced at 21 % O2 concentration, water uptake is similar regardless of the scale, with a maximum difference of ±15 %. For the fuels studied, fire scale had no major influence on the water adsorbing properties

Fig. 7

Comparison between water uptake by soot samples produced at pilot and laboratory scale.

Fig. 8 compares the water uptake by soot produced at 21 % of O2 concentration (ambient air value) with that produced at depleted O2 concentrations (15 %, 17 % and 18 %), which adsorb up to 50 % more water than soot produced at ambient concentration (see Fig. SI-4 in Supplementary Information V). This highlights the significant modification in size (decrease of primary particle size and increase of specific surface area) and composition (increase of oxygen content) of soot particles at decreasing oxygen concentration reported in Table 2 and in agreement with previous findings (Ouf et al., 2015; Léonard et al., 1994).

Fig. 8

Comparison between water uptake obtained for soot samples produced at depleted oxygen concentrations (15–18 %) and ambient oxygen concentration (21 %).

Influence of specific surface area and oxygen surface content of water uptake

Fig. 9 presents the water uptake obtained at the maximum RH of 90 % as a function of the specific surface area, for all fire soot samples and for PRINTEX 90 and FLAMMRUSS 101. These carbon black samples are mostly composed of elemental carbon. Their water adsorption isotherms are Type III (see Supplementary Information VI, Fig. SI-5), which indicates low carbon black-water interactions and a water uptake related to the specific surface area only. Additional data on carbon black samples available in the literature - whose compositions are similar to our samples - are also plotted in Fig. 10 (identified by an asterisk “*” (Kiselev and Kovaleva, 1959; Carrott, 1992; Popovicheva et al., 2008; Charrière and Behra, 2010)). Over the whole specific surface area range, soot particles present higher water uptake than carbon black and do not follow the linear correlation proposed in Fig. 10. These discrepancies between soot and carbon black can be explained, in addition to the geometric surface associated with the cake structure, by their specific chemical composition.

Fig. 9

Effect of specific surface area on water uptake at 90 % relative humidity.

Fig. 10

Evolution of the number of ML with the surface concentration of oxygen, at RH = 30 % (left) and RH = 90 % (right).

To assess the influence of sample composition on water sorption capacity, water uptake can be expressed as the number of equivalent water monolayers (ML) needed to cover the entire surface of sample (Eq. (5)). This unravels the effect of the surface area from the chemical composition:where σ is the surface occupied by a water molecule (1.05.10−19 m²), a the amount of adsorbed water per mass of sample, N the Avogadro number (6.022.1023 mol−1) and S the specific surface area (m². g−1).

Fig. 10 shows the evolution of ML with the surface concentration of oxygen [O]surface, calculated as the mass of oxygen per surface area (Eq. (6)):

Oxygen content is known to significantly influence water sorption (Liu et al., 2017). The oxygenated chemical functions (mainly carboxyl and carbonyl as reported by (Liu et al., 2017)) located at the surface strongly interact with water molecules, especially in low humidity range, where they represent the energetically most favourable adsorption sites (Velasco et al., 2016). To highlight this effect, we plotted ML at RH = 30 % and RH = 90 % (left and right, Fig. 10 respectively) as a function of [O]surface for all our samples and those available in the literature. ML linearly increases with [O]surface with a slope of 5.33 (RH = 30 %) and 13.57 (RH = 90 %). This clearly shows that the oxygen concentration at the surface significantly increases the adsorption process, with a similar physico-chemical mechanism within the 30–90 % relative humidity range (as the same linear dependence is observed at RH = 30 % and 90 %).

Fig. 11 compares, for each relative humidity step (+/− 1%), the water uptake obtained for the pellets using the DVS microbalance and for the powders using the 3FLEX analyser (the corresponding water adsorption isotherms are available for 3FLEX in Supplementary Information, Fig. SI-3). This figure shows that water uptake for powder and compacted samples is generally equivalent within a +/− 15 % interval. However, exceptions can be observed for Heptane 15 %, Heptane 21 % and PMMA 21 % [pilot scale], where adsorption is higher for powder than for pellets, especially at high water uptake (related to higher relative humidity). This may be explained by different mesostructures of these soot samples when they are characterized in pellet or in powder forms, changing the accessibility and quantity of surface available for water adsorption. This hypothesis is supported by hysteresis loops, presented in Fig. SI-3, which indicate a higher hysteresis area for powder samples. As reported by previous authors (Liu et al., 2017; Nakamura et al., 2010), any increase in the hysteresis loop area is generally associated with an increase in pore size. Such an evolution confirms, in the present case, that compression applied when preparing the pellet samples induced a decrease of both pore size and accessibility and, as a result, a less pronounced hysteresis loop.

Fig. 11

Parity diagram comparing water uptake between pellets and powders at different humidity steps.

Discussion

Until now, water sorption on carbon systems has mostly been modelled for activated carbon or carbon/zeolite (Do, 1998), whose isotherms are systematically Type V. Several common water sorption models have been established for such isotherms, considering a primary adsorption on specific hydrophilic surface sites followed by adsorption on already adsorbed water molecules. This second adsorption process is driven by water-water interactions, which are, overall, more favourable than those between water and a generally hydrophobic carbonaceous surface. Being an associating fluid, water can fill in the micro- and mesoporous pore volume. Unlike activated carbon, soot particles are made of non-porous hydrophobic carbon, porosity only being due to interstitial spaces between the primary particles in the soot cakes, which also varies with the sample form (powder or pellet and as supported by analysis of hysteresis loops). The surface of soot is mostly hydrophobic, with some hydrophilic adsorption sites related to the presence of surface oxygen. As mentioned above, one of the most suitable sorption models for such a porous solid is the Dubinin-Serpinsky (DS) model (Dubinin, 1980; Dubinin and Serpinsky, 1981). This model describes a mechanism of water cluster formation on the adsorbent surface sites, which can be followed by filling the pore volume (Barton et al., 1991; Dubinin et al., 1982). In the DS model, the adsorption process is considered as an equilibrium state of a chemical reaction between water molecules in the gaseous phase H2Ogas and the adsorption sites (mainly carbonyl and carboxyl groups identified by XPS analysis, see Figs. SI-1 and SI-2) located on the adsorbent surface Sadsorbent (Fig. 12 ). These sites can be primary or secondary, corresponding respectively to the initial number of hydrophilic sites a (%g. g−1 adsorbent) and the already adsorbed water molecules a per gram of sample (expressed here in terms of %g. g−1 adsorbent). This equilibrium is formalized by the equilibrium constant c (defined as the ratio between kinetic constants associated with adsorption kads and desorption kdes), used to express a, the total amount of adsorbed water (Eq. (7)), including the water vapour relative pressure h.

Fig. 12

Schematic diagram of the proposed mechanism of water adsorption and the equation used to express the Dubinin-Serpinsky model.

Among all versions of the DS model (Furmaniak et al., 2008), the second version, commonly called DS2, has the simplest analytical form and takes into account the limitation of adsorption with water uptake due to the steric hindrance. For this purpose, a dimensionless and strictly positive factor (1-ka), decreasing with the adsorbed water amount, has been added to the original DS equation (Eq. (8)). The constant k therefore has a value that ranges from 0 to strictly below 1/a. This constant relates to the proportion of water molecules which no longer act as a secondary adsorption site (steric hindrance). It leads to the analytical form of the second version of the DS model (Eq. (8)):where k is the coefficient related to the steric hindrance due to the formation of water clusters (%gadsorbent. g−1).The adsorption isotherms obtained in this study all successfully fit with the DS2 model in the relative humidity range of 0 %–90 % (Fig. SI-6). Fitted parameters k, a0 and c for all samples are also available in Table SI-3, with regression coefficients R² all above 0.99. This indicates a unique adsorption process consisting in the formation of water clusters on few hydrophilic sites, which we assume to be oxygen groups present at the surface of soot particles.

The fitted values of k are 0.24 ± 0.02 and 0.14 ± 0.03 for pellets and powders, respectively. The higher k value for pellets is a consequence of a limited secondary adsorption process, indicating higher steric hindrance to cluster formation compared to powders. This is most likely due to the higher compaction of the pellets, which facilitates the blocking of interstitial spaces between the soot particles. The decrease in the k values is consistent with the increase of mean concentrations of primary adsorption sites a (in %g g−1 adsorbent), from 1.0 % ± 0.4 % for pellets to 2.6 % ± 0.4 % for powders. The lower value of a for pellets indicates less accessibility to the adsorption sites due the filling of the pores where the primary adsorption sites are located, while more sites are available and more accessible in powders (with larger pore width and higher accessibility confirmed by hysteresis loops presented in Fig. SI-3). The mean values of equilibrium constant c, corresponding to the ratio of the kinetic constants between adsorption and desorption processes, are 2.6 ± 0.4 and 1.6 ± 0.4 for the pellets and powders respectively. Taking into account the related uncertainties, these values remain quite close and, at this stage, we cannot explain such a slight difference in the kinetic constant without further investigation.

As previously reported in the literature (Thomas et al., 2014), porosity of soot cake formed at the HEPA filter surface is generally between the values considered in this study for pellets (mostly 50–60 %) and powder (95 %). To provide useable values of k, a and c that could be implemented in clogging models for similar kinds of hydrophobic soot (Mocho and Ouf, 2011; Bourrous et al., 2016), we have averaged the fitting parameters k, a and c of the 22 samples studied, whether compacted into pellet form or not. These values are reported in Table 3 , along with their standard deviation. Fig. 13 presents the computed water uptake using these averaged values and deviations plotted against the experimental water uptake for all soot samples. We observe that 95 % of the water uptake (limited in the present case to water uptake values higher than 1%) can be satisfactorily represented by the DS2 model using these averaged parameters, within a confidence interval of ± 47 %.

Table 3

Calculated values from 22 water adsorption isotherms.

	k (%)	a0 (%g.gadsorbent-1)	c (−)
Mean value	0.20	1.54	2.23
Standard deviation	0.056	0.85	0.69
Expanded uncertainty for a 95 % confidence interval	0.012	0.18	0.15

Fig. 13

Comparison between water uptake in the DS2 model, computed according to mean constants, and experimental results.

Conclusion

This study aimed to measure the water uptake in soot cakes representative of those that form on HEPA filters during fire events occurring at nuclear plant. For this purpose, soot was produced at laboratory and pilot scales using different fuels - heptane, PMMA and hydraulic oil- representative of fuels encountered in a real nuclear plant. The physico-chemical properties of these samples have been determined ex situ using analytical techniques to provide the specific surface area, the primary particle diameter and the oxygen content/speciation, found in the ranges of 52 m² g−1–100 m² g−1, 22.3 nm–43.3 nm and 6.7 wt.% - 9.9 wt.%, respectively. We have shown that these physico-chemical properties are only slightly influenced by fire scale, fuel type, or O2 concentration of the oxidizing gas. Using gravimetric and manometric techniques, we measured the water uptake of these samples as well as two additional carbon black samples, left in powder form or compacted into pellets. Gravimetic and manometric approaches have proven to be equivalent, with a variation coefficient of less than 10 % across the whole relative humidity range (0 %–100 %). The adsorption isotherms of soot are Type V, present maximum water uptake between 2% and 4% at RH = 90 %, and indicate a hysteresis loop with a restricted area. Under the conditions investigated here, fire scale had no significant impact on the isotherm classification. Two physico-chemical properties of soot directly impact the sorption process: the specific surface area and the oxygen concentration per surface area. The good fit of the experimental isotherms with the second Dubinin-Serpinsky model (DS2) indicates that adsorption occurs by nucleation of water clusters on few oxidized and hydrophilic surface sites (namely carboxyl, carbonyl, quinone and ether groups). The DS2 parameters were obtained for 22 samples studied, in either compacted or powder form, and their average allows for the effective prediction of 95 % of the experimental data. This model can be used to improve simulation codes, therefore, to better predict HEPA filter clogging during a fire at a nuclear facility. In addition, the use of the second Dubinin-Serpinsky model appears relevant for representing the water sorption isotherms of fire soot.

CRediT authorship contribution statement

Laura Lintis: Investigation, Formal analysis, Validation, Methodology, Writing - original draft, Writing - review & editing. François-Xavier Ouf: Conceptualization, Supervision, Writing - review & editing. Philippe Parent: Investigation, Formal analysis, Writing - review & editing. Daniel Ferry: Investigation, Formal analysis, Writing - review & editing. Carine Laffon: Investigation, Formal analysis, Writing - review & editing. Cécile Vallières: Conceptualization, Supervision, Writing - review & editing.

Declaration of Competing Interest

The authors report no declarations of interest.