Candle flame soot sizing by planar time-resolved laser-induced incandescence

Soot emissions from flaming combustion are relevant as a significant source of atmospheric pollution and as a source of nanomaterials. Candles are interesting targets for soot characterization studies since they burn complex fuels with a large number of carbon atoms, and yield stable and repeatable flames. We characterized the soot particle size distributions in a candle flame using the planar two-color time-resolved laser induced incandescence (2D-2C TiRe-LII) technique, which has been successfully applied to different combustion applications, but never before on a candle flame. Soot particles are heated with a planar laser sheet to temperatures above the normal flame temperatures. The incandescent soot particles emit thermal radiation, which decays over time when the particles cool down to the flame temperature. By analyzing the temporal decay of the incandescence signal, soot particle size distributions within the flame are obtained. Our results are consistent with previous works, and show that the outer edges of the flame are characterized by larger particles (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx 60\,\hbox {nm}$$\end{document}≈60nm), whereas smaller particles (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx 25\,\hbox {nm}$$\end{document}≈25nm) are found in the central regions. We also show that our effective temperature estimates have a maximum error of 100 K at early times, which decreases as the particles cool.


Results
Primary soot particle temperature. As a first analysis step, the energy balance equation of a single soot particle (Eq. 1) in the Methods section below) is solved for spherical particle diameters in the range of values between 1 and 105 nm. The assumed initial particle temperature (2,000 K) employed in the iterative processes, is a typical diffusion flame temperature, and the applied fluence during 8 ns is 0.148 J/cm 2 . The resulting theoretical temperature histories may be observed in Fig. 1, where the very short initial heating period that occurs during the laser pulse heats the soot particles to temperatures greater than 3,000 K. This figure shows that the temperature decay time of the smallest soot particles is on the order of a few nanoseconds, whereas the largest particles show www.nature.com/scientificreports/ temperatures exceeding 2,500 K even after 2 µs . It is then clear that the temporal temperature decay rate may be used for the purposes of discriminating between primary soot particles with diameters spanning several orders of magnitude, as previously underscored by several works 27,28,32,34 . Note that, in the framework of the TiRe-LII technique, the calculated temperature is used to determine the theoretical effective temperature, T e (d pg , σ g ) , through Eq. (10), where d pg and σ g are the geometric mean particle diameter and the geometric standard deviation of an assumed lognormal particle size distribution 35 .
Temporal signal decay. The temporal decay behavior of the raw LII signals is the fundamental quantity used in the TiRe-LII technique. Figure 2 presents this decay along the flame for the studied candle. Note that since axial symmetry is assumed, only one half of the candle flame directly exposed to the laser signal is presented here, and that the origin of the vertical coordinate, HAB (Height Above the Base), lies at the candle wax pool surface.
Each of the LII intensity fields given in this figure is the outcome of averaging 200 instantaneous images, each of which corresponds to the the ICCD camera gated at 20 ns. During the experiments the camera opening delay with respect to the laser pulse was varied from 20 ns to 1,000 ns. In Fig. 2 representative images of the decay process are depicted, taken at the delays of 20, 40, 300 and 670 ns. These delay values have been chosen so as to represent the prompt, short time, intermediate time and long term LII signal behaviour. Figure 2a, b also give the temporal decay at the two detection wavelengths used in this study, i.e., = 450 and 650 nm, respectively. These detection wavelengths have been chosen to avoid fluorescence from gas-phase flame species (PAH and others) 36,37 , to account for the camera spectral sensitivity -thus yielding comparable LII signal intensities-and to provide an adequate spectral separation. Indeed, the emission intensity of the particles is higher for 2 = 650 nm than for 1 = 450 nm 38 , compensating for the signal ratio loss due to the spectral quantum efficiency variation of the ICCD camera.
The LII signal is larger for the longer wavelength, with local signal maxima at each HAB, possibly highlighting the position of the higher soot volume fraction (f v ). The LII intensities are highest at 20 ns, with a peak region located around HAB ≈ 20 mm . Therefore, this particular high intensity region is used here for analysis purposes, since it may be assumed that a fixed proportionality factor between the S LII intensity and f v 17,21 exists. Figure 3 presents the decay history of the LII signal for the two measurement wavelengths, i.e., S LII,450 and S LII,650 , at two different locations along the flame, as well as the corresponding effective soot temperature. These regions correspond to the position along the flame centerline where the maximum LII signal is observed and to the location of maximum soot volume fraction (f v,max ). In both regions, S LII,450 < S LII,650 (cf. Fig. 3a). Furthermore, at each wavelength, as expected, the signal at the maximum soot volume fraction region is larger than that at the chosen centerline position. Figure 3b also depicts an exponential fit of the experimental effective temperature T e,exp , which is needed for comparison purposes with that numerically calculated T e (d pg , σ g ) , and to obtain the lognormal diameter distribution parameters. The experimental effective temperature [cf. Eq. (11) below] is found to decay monotonically from 3,200 to 2,600 K, as may be verified in Fig. 3b. These temperature values, lower than the carbon sublimation point, are consistent with the desired application of the TiRe-LII technique.
Soot temperature and diameter. In order to further characterize the LII signal at early times, Fig. 4 gives the fields of the signal ratio at the two detection wavelengths ( S LII, 1 /S LII, 2 ), the maximum effective temperature computed at 20 ns and the fitted temperature decay rate between 20 and 100 ns. The signal ratio varies between 0.3 and 0.5, with the smaller values obtained at lower portion of the flame, closer to the wick, and the larger being characteristic of the outer, oxidizing, region. The effective temperature at this early time is as high as 3,200 K at the outermost regions of the flame, and descends to ≈3,000 K towards the flame axis. This rather uniform effective temperature distribution indicates that, regardless of their diameter or concentration, the soot particles are being heated to similar temperatures by the laser pulse. The corresponding temperature decay rate exhibits a higher absolute value at the innermost, central flame regions, and smaller at the outer regions. Furthermore, The effective temperature decay rate distribution indicates that the larger soot particles should be located in the vicinity of the oxidizing region, and that a rather uniform diameter distribution should arise along the flame centerline. This may indeed be verified in Fig. 5a, b, where the obtained soot particle diameters are given. The Sauter mean diameter ( d 32 ), is a classical particle size average, defined as the ratio of the third to second moments of the particle diameters distribution 39 , and represents the diameter of a particle whose surface-tovolume ratio is equal to the entire soot particles ensemble at the region of interest 40 . Accordingly, the geometric and Sauter mean diameter values reach 50 and 60 nm at the outermost regions of the sooting region, respectively, whereas smaller corresponding values-35 and 25 nm-arise near r = 0 . In order to gain further insight on the soot particle diameter distribution within the candle flame, Fig. 5c shows the computed values of the primary particle diameter, d s , obtained by Eq. (8). Attention is drawn first to Fig. 5d, where the probability distributions of primary particle diameter are plotted at two positions within the flame. The first of these positions is characteristic of the flame centerline behavior, and the second of the maximum soot volume fraction location. This figure indicates that a narrower primary particle diameter distribution is observed at the centerline position, i.e., 20 < d s < 35 nm, whereas a wider one ( 20 < d s < 90 nm) characterizes the maximum soot volume fraction region. The corresponding field of primary particle diameter d s may be seen in Fig. 5c. These diameters are found to range from 20 nm, near r = 0 , to 50 nm, at the vicinity of the soot oxidation region, where the soot volume fraction is maximum. As expected, the overall diameter spatial distribution is similar, regardless of which -d 32 , d s or d pg -is considered (cf. Fig. 5). The primary particles diameter, d s , obtained here are comparable with previous results obtained by TEM images (20-50 nm) 18 and point TiRe-LII measures ( ≈ 55 nm) 20 .

Discussion
In this work, the planar two-color time-resolved laser induced incandescence (2D-2C TiRe-LII) technique was used to characterize a candle flame operating below the smoke point. To the best of the authors' knowledge, at present there are no works that report soot particle sizes for a candle flame using this technique for two dimensions. Nevertheless, TiRe-LII has been applied to Candle point measurements 20 , n-heptane 34 , ethylene [31][32][33]41 and ethane 24 flames, yielding similar results both for the S LII and particle diameter distributions. The effective soot temperatures decay from 3,200 to 2,600 K, which is consistent with TiRE-LII applications at the laser fluences used in this study. The larger soot particles, with d 32 ≈ 60 nm, tend to be located at the outer edges of the sooting region, whereas the flame centerline is characterized by the presence of smaller particles ( d 32 ≈ 25 nm). Figure 6 depicts the vertical evolution of a set of properties characterizing the sooting flame. Following the line of maximum soot volume fraction, this figure presents the geometric mean diameter, which exhibits a non monotonic behavior. Soot particles are first detected with d pg ≈ 32 nm , at HAB ≈ 9 mm , and the diameter is found to increase up to 64 nm at HAB ≈ 17 mm . A diameter decrease is then observed until HAB ≈ 35 mm , where the detection limit is again reached. Following previous works 6 , this figure also shows the radially integrated soot volume fraction evolution with HAB, normalized with respect to the peak value β = 2π ∞ 0 f v (r) rdr. . Note that from the relationship between f v and S LII , the normalized β values are related to the incandescence signal, which is taken at the prompt detection time, where the signal reaches its maximum value. For this purpose, either 1 or 2 may be used, since f v is a physical property of the flame. The results of the evolution with height of the integrated soot volume fraction are compared in Fig. 6 with previously measured values using Modulated Absorption Emission (MAE) for the same candle flame 6 . The previous measurements detected soot in significant amounts ( β = 0.3 ) at HAB ≈ 7 mm , whereas in this work these amounts are observed further downstream ( HAB ≈ 11 mm ). This discrepancy is related to the different methods used to determine β , since in the present work S LII nearly vanishes below HAB ≈ 11 mm , whereas the MAE signal is strong in the lower parts of the flame 6 . The maximum value of β is found to occur nearly at the same position ( HAB ≈ 20 mm ) for the two measurements, and the progressive integral soot volume fraction decrease rate with HAB is nearly identical. A   Fig. 7 presents both the effective temperature difference ( ǫ ), given by Eq. (14), and the normalized likelihood estimator ( χ 2 ), as defined by Eq. (15). The first figure-of-merit, ǫ , is examined at two different times, 20 ns and 600 ns in Fig. 7a, b, respectively. These figures show that both the magnitude and the distribution of this error change with time. Indeed, at earlier times the maximum error, which is on the order of 100 K, is found to occur at the innermost parts of the measurement region, whereas at later times the maximum is displaced towards the outer edges of this region. Furthermore, at 600 ns, ǫ is smaller than 50 K at the central part of the flame but, at 20 ns it remains larger than 60 K at the upper parts of the flame centerline. These differences between the experimental and theoretical effective temperatures are mainly associated to the S LII measurements. By comparing the signal ratio S LII, 1 /S LII, 2 , given in Fig. 4a, and ǫ at 20 ns (see Fig. 7a), it is possible to verify that the largest temperature difference occurs at the area of the smallest signal ratio value, and vice-versa. This underscores the close relationship that exists between the signals ratio and the effective temperature difference. Furthermore, considering the values of ǫ at 600 ns (see Fig. 7b), it can be deduced from Fig. 3a that the S LII ratio decreases with time, so that the difference between the effective temperature is smaller at the selected points. As a consequence, at later times the temperature difference ǫ is larger at the outer flame zones, where the temperature decays faster, than at the inner regions. The χ 2 likelihood estimator field depicted in Fig. 7c is evaluated at a radial location of 2 mm from the flame axis and a height of 22.5 mm as a function of the diameter distribution parameters, d pg and σ g . This position has been chosen because it corresponds to the maximum soot volume fraction region. This figure shows the minimum value of χ 2 that characterizes the distribution shown in Fig. 5d. The χ 2 determination has been performed for each pair of parameters, for time intervals ranging from 20 ns to 1 µ m. The estimate range for d pg spans from 0 to 50 nm with a 0.05 nm step, and the corresponding one for σ g is from 1 to 2 with a 0.01 step. Figure 7c indicates that a χ 2 mathematical minimum exists for the chosen parameter variation of the lognormal distribution. In particular, the region of d pg from 30 to 50 nm and σ g from 1.1 to 1.3 is where an absolute minimum seems to lie, i.e., where the difference between the values of effective temperatures is the smallest. The parameters of the lognormal distribution at this region (seeFig. 5d) are those that minimize the χ 2 -value. However, since a minimum value may not be sharply distinguished, the values obtained above ( d pg and σ g given in Fig. 5d) are assumed to be representative of those that minimize χ 2 -value.

Theoretical background: the TiRe-LII model. The laser-induced incandescence technique (LII) is
widely used to study the formation of soot particles within flames 42 . This technique is implemented by applying a nanosecond laser pulse that increases the soot particles temperature to levels where detectable incandescence signals arise ( S LII ) 43,44 . After the laser irradiation, the S LII decays as the soot particle temperature returns to the surrounding flame condition. Since smaller soot particles cool down faster than larger ones 22,45 , due to the larger surface area to volume ratio and smaller thermal inertia, the soot particle size distribution may be determined from the temporal decay study of S LII . This may be performed by comparing the measured soot temperature decay with the computed temperature results obtained by assuming a soot primary particle size probability distribution 21,44 . Please refer to Fig. 8 for a visual summary of the technique. The temperature time-history, T(t) , of a single soot primary particle, with diameter d s , density ρ s (1,900 kg/m 3 ) and specific heat c s with a temperature dependence according to Liu's model 29 , may be modeled by the energy balance equation 28,29 : (1) www.nature.com/scientificreports/ which states that the internal energy rate of change is equal to the sum of laser energy absorption, C a F 0 q(t) , and the energy loss by thermal radiation, q r , conduction, q c , and sublimation, q s . In the laser irradiation term, C a is the absorption cross section, F 0 is the laser fluence and q(t) is the power density history of the laser pulse 46 . Note that other physical and chemical processes, such as photodesorption, annealing, and soot oxidation, may arise during particle cooling, but can be considered negligible when low or moderate laser fluences are used 21,47 , which is the case of the present work. The different processes influencing the soot particles temperature decay rates given by Eq. (1) should be adequately modeled in order to allow particle diameters to be determined from the measured temperature history. In the Rayleigh limit, the absorption cross section is given by: where E(m) is the soot absorption function 48,49 , and represents the laser irradiation wavelength. The energy lost by thermal radiation is given by 47 : where h, k and c are the Planck and the Boltzmann constants and the speed of light, respectively. N p is the aggregate size 46,50 and the integration yields a constant value of 24.886 51 . Under ambient pressure conditions, the energy lost by thermal radiation can be considered negligible 22,51 . Using the hypothesis of free-molecular regime for the soot particle cooling by conduction 50 and the Fuchs approach 52 , the conduction cooling rate of soot particles may estimated as: where α , p g and T g are, respectively, the soot thermal accommodation coefficient assumed as 0.37 46 , the ambient gas pressure and the temperature 6 . The mass of the surrounding gas molecule is m g and γ * represents the average value of the surrounding gas specific heats ratio 50 . The energy lost due to particle sublimation may be modeled as 22,29 : where M v is the soot molecular weight and H v represents the enthalpy of formation of carbon clusters. The particle mass loss rate by sublimation, dM/dt , may be written as 22,29 : (2) C a = π 2 d 3 s E(m) , Fig. 8. Schematics of the TiRe-LII methodology: (a) Heating process and relevant properties: HAB is the height above the base, i.e., measured from the wax pool surface; soot particle and aggregates characteristic dimensions; laser energy temporal evolution; experimental and theoretical effective temperature decays; and (b) Flowchart for geometric mean soot particle diameter; Step 1: theoretical temperature decay; Step 2: experimental effective temperature decay and Sauter mean diameter; Step 3: error minimization, lognormal distribution properties.
Scientific RepoRtS | (2020) 10:11364 | https://doi.org/10.1038/s41598-020-68256-z www.nature.com/scientificreports/ where α M is the mass accommodation coefficient, p v is the average partial pressure, R p and R m are the universal gas constant expressed in different units and W v is the average mass of the sublimed clusters 29 . Following 53,54 , the soot primary particles diameter probability distribution within the probe volume can be assumed as exhibiting a lognormal size distribution. i.e., where d pg and σ g , represent the geometric mean particle diameter and the geometric standard deviation, respectively. The primary particle diameter, d s , has been obtained from the first moment of the lognormal distribution for given d pg and σ g values, following If the laser probe volume is small enough to allow for the assumption of an optically thin path, and the soot particles are uniformly distributed, the modeled total thermal emission intensity of this particle distribution at wavelength i may be expressed as 28 : where T(d s ) is the solution of Eq. (1). Then, the theoretical effective temperature time-history may be obtained from the ratio of total thermal emission at two different wavelengths ( 1 > 2 ), by using the two-color pyrometry equation 38 : where C 2 = hc/k is the second Planck constant and the Wien approximation, exp(hc/k T)≫ 1, has been employed. On the other hand, after the laser pulse, the effective temperature time history of the soot particle distribution can be determined from the measured TiRe-LII signal as: where the two-color pyrometry relation has also been used. Furthermore, the Sauter mean diameter ( d 32 ) may be related to the initial temperature decay rate as follows 28 : which is evaluated at time ( t = τ max ) where the effective temperature peak ( T max ) is observed, which typically occurs 20 ns after the laser pulse 28,32,33 . Also, T 0 represents the initial soot temperature, which may be determined by using a two color pyrometry technique, in the absence of laser irradiation. The parameter represents a cluster of all the surrounding gas and soot particles thermal properties 28 . Under the assumption of a lognormal soot particle diameter probability distribution, the Sauter mean diameter ( d 32 ) is related to the two distribution parameters ( d pg , σ g ) by: These soot particle distribution parameters ( d pg , σ g ) are estimated by minimizing the difference between the computed and measured values of the soot effective temperature, T e , according to the iterative process shown in the Fig. 8b. Indeed, the Sauter mean diameter ( d 32 ) obtained from the experimental effective temperature (step 2), Eq. (12), is used as an input parameter to compute a modeled effective soot temperature, Eq. (10), and its corresponding lognormal distribution parameters (step 3), d pg and σ g , which are related via Eq. (13). For this purpose, Eq. (1) is numerically solved within a range of spherical soot primary particle diameters (1 to 105 nm) 28 (step 1). The difference between the numerical and experimental effective temperature decay is then iteratively reduced until reaching a minimum error. The quality of the parameters estimation results is evaluated by considering the absolute error ( ǫ ), which compares the predicted and measured effective temperature values at each the flame location: In addition, the Chi-Squared test ( χ 2 ) also is used as a suitable test for the purpose of evaluating the effective temperatures agreement 53 :  Experimental apparatus. The experimental setup used for the TiRe-LII measurements is shown in Fig. 9a. In the adopted configuration, the second harmonic (532 nm) of a Nd:YAG Litron Aurora II laser (3), with a 9 mm beam diameter and 10 Hz repetition rate, has been used to excite the incandescence ( S LII ) of the soot particles formed in the candle flame (1). The candles are composed of Sasolwax 6203 paraffin, and have been manufactured in house according to procedures detailed elesewhere 6 . These candles are characterized by wick diameter and length values of D wick = 3 mm and L wick = 7 mm, respectively. These wick dimensions have been chosen since they correspond to the stablest experiments previously performed 6 , and lead to flames that operate below the smoke point. The height of the candle flame measured from the wax pool ( L f ) is consistent with previous measurements 4, 6 , which have been performed for D wick = 3 mm and L wick = 7 mm. As evidenced by Eq. (1), the accurate characterization of the laser pulse is paramount for the TiRe-LII technique. Accordingly, Fig. 9b shows the temporal profile of a typical laser pulse (8 ns span), which has been measured with a fast photodiode ET-2030 (4) coupled to a 1 GHz oscilloscope (LeCroy Wavesurfer 3104Z). An iris is employed to select a 6 mm diameter central portion of the laser beam that is expanded, with the help of a spherical convex lens ( f = 750 mm ) and a concave cylindrical lens ( f = −50 mm ) to form a thin laser sheet, which crosses the flame centerline. As observed in Fig. 9c, the sheet thickness is 120 µm , whereas its height is 65 mm. Note that the sheet thickness is roughly 1/10 th of the maximum flame diameter, 2 mm. To correct both for the laser intensity non-homogeneity and for the shot-to-shot laser fluctuations, the corresponding spatial and temporal energy distributions are respectively mapped with a beam profiler (Coherent LaserCam-HR II) (5) and by an energy sensor (Coherent J-50MB-YAG) (6), which is coupled to a Coherent Labmax TOP energy meter.
The laser incandescence signal emitted by the soot particles is captured with an intensified CCD camera (Andor Istar DH334T) with a 1024 × 1024 px 2 matrix (2), that is coupled with a Nikon AF Nikkor 50 mm lens (f/1.4). The LII signal first passes through narrow band filters (40 nm FWHM) centered either at 450 nm or at 650 nm, which have been selected in order to improve the signal to noise ratio. All the measurement devices are synchronized with an external (Quantum sapphire 9200) pulse generator (not show here). A motorized linear stage (7) has been used to progressively raise the candle as the wax is consumed, thus ensuring that the flame lies at the same measurement region during all tests. Since the TiRe-LII technique requires for the laser illuminated soot particles to be heated below the carbon sublimation temperature, care must be taken when selecting the www.nature.com/scientificreports/ laser fluence. Accordingly, the normalized fluence curve of the candle flame, shown in Fig. 9d, indicates that the plateau region should be achieved beyond 0.16 J/cm 2 . Then, in order to reduce the soot sublimation effect, a laser sheet fluence of 0.148 J/cm 2 has been used in this work to obtain the S LII . In order to improve the signalto-noise-ratio, 2 × 2 pixel binning has been used, thus resulting in an image resolution of 12.6 px/mm. For each measurement delay after the laser pulse, 200 images have been captured at each wavelength. The effective soot temperature uncertainty may be estimated following previous works 55 , and is attributed to the wavelength separation 38 , the standard deviation of incandescence images due to inherent noise of the ICCD cameras, and the value of absorption function, which is currently matter for debate 21,49 . It should also be stressed that, to the best of the authors knowledge, the value of E(m) in candle flames has not been previously reported. Therefore, a classical correlation 48 is assumed to hold in this study.