1. Introduction

Recent studies have demonstrated a causal relationship between particulate matter concentrations in atmospheric air and increased human deaths [1]. It has also been proved that amongst different species of atmospheric particulate, the most abundant and dangerous to human health consists of carbonaceous particles, mostly produced in combustion processes. Besides natural sources (forest fires) of particulate matter, a large amount of polluting material is produced by electrical combustion power stations and by vehicle engines, the latter particularly affecting urban areas. Important studies on combustion processes occurring in vehicle engines, for instance, and a detailed characterization of atmospheric particulate led to the design and realization of filtering devices that effectively reduce pollutant emissions. The effectiveness of such devices is particularly high for micrometric scale particles, but drops to inadequate values in the case of very small (nanometric) particles. On the other hand, ultra-thin particulate matter with characteristic size of few nanometers is extremely dangerous to human health due to its ability to penetrate biological defenses and to its sometimes high diffusivity in water and consequent assimilation by the human body.

Moreover, because of their size and optical properties, carbonaceous nanometric particles may also have significant effects on climate [2], being suspected to significantly contribute to the greenhouse effect.

The whole process of soot production by chemical reaction from the unburned fuel/oxygen mixture up to large soot particles formation can be roughly divided into three phases: i) The molecular phase, where all processes can be described in terms of chemical reactions and the largest molecules, basically consisting of polycyclic aromatic hydrocarbons (PAH) still have sizes in the sub-nanometer range; ii) The nanometric phase, where PAH molecules react with small radicals forming larger molecular clusters of few nanometer size; iii) The micrometric phase, consisting of large soot particles grown from 10–20 nm size primary soot particles. Most work has been done on the first and third phases, but detailed characterization of the nanometric phase is still lacking, likely because the detection of nanometric species produced in combustion processes is a critical challenge [3].

Difficulties arise in developing diagnostics and evaluating their performance for particles in the transition range between the molecular and the large particle phase, and appropriate standards for calibrating these systems are not readily available. Consequently, there is an increasing demand for diagnostics capable to determine physical and chemical characteristics of particles in the 1–10 nm size range.

Optical techniques are extensively used for particulate matter measurement, mostly because they are non-intrusive and provide essentially real-time data. Recently, various optical techniques for measuring particulate size and concentration, as well as light scattering, absorption and extinction coefficients have significantly advanced. Amongst these, it is worth mentioning the laser-induced incandescence technique, enabling the detection of soot particles in the 10–20 nm range [4]

In recent years, the availability of ultra-fast laser sources and detectors allowed the development of the time resolved fluorescence polarization anisotropy (TRFPA) technique, which resulted to be a very powerful tool for providing simultaneous information on both the chemical composition of the particles and their size. This technique has been extensively used to study biological systems, providing very interesting information on rotational property and shape of proteins and macromolecules [5,6]. It is based on the fact that linearly polarized light preferentially excites those molecules or particles whose transition dipole moment forms a given angle with respect to the light field. Early fluorescence light also results to be partially polarized. Subsequently, due to rotational diffusion of the excited particles, their average dipole moment looses any partial orientation, and the same occurs to the polarization direction of the fluorescence light. The depolarization rate is obviously related to the mass and size of the fluorescing particle and, thus, by measuring the characteristic time for such depolarization process, it is possible to determine the particles’ size.

In the present article, we apply the TRFPA technique to two classes of samples, namely, nanometric carbonaceous particles collected from laboratory laminar flames and from the exhaust of gasoline and diesel vehicle engines. The aim is to partly fill a lack of information in the soot formation process by applying a well established technique to a new class of samples, namely carbonaceous particles produced in combustion processes. It is worth mentioning that all the samples studied in the present work did not need any previous treatment with label dyes, because they showed sufficient absorbance and fluorescence at wavelengths within the range of our light sources and detectors, respectively. Moreover, since the fluorescence light is emitted by the same particle under study, it is possible to obtain some information on the chromophores’ composition by correlating the fluorescence band with the particle size.

We present two implementations of the TRFPA technique, one employing low energy (few nanojoules), high repetition rate (82 MHz) laser pulses and a streak camera as a light detector, whereas the other makes use of 1 mJ laser pulses at 1 kHz repetition rate and a simple detection system consisting of a fast photomultiplier tube and an oscilloscope. The former achieves short acquisition times and high time resolution, resulting in a rapidly converging data processing, and, thus, in smaller uncertainties. On the contrary, the data analysis of the latter method is more critical, and, consequently, a lower resolution is achieved, although its experimental realization is much simpler and cheaper. It is worth remarking that, however, the results obtained with the two methods are essentially comparable, although the higher sensitivity of the first one allows to resolve finer details of the samples under study (internal motion, wobbling, energy migration, energy transfer) that are not observable with the second technique. Finally, both methods described in the present paper require overall acquisition times that are considerably shorter compared to other techniques, as, for instance, time correlated single photon counting.

The following Section 2 is dedicated to a description of the operating principle of the TRFPA technique. In Section 3 we describe the two methods for analyzing the fluorescence light emitted by the excited samples, the data processing techniques for determining the fluorescence anisotropy and the procedures for samples’ collection. In Section 4 we apply both detection methods to molecular standards in order to test their reliability and to establish some calibration procedures. Section 5 is dedicated to the analysis of the results obtained on particles collected from laminar flames and from gasoline and diesel engine exhausts. Concluding remarks and future perspectives are reported in Section 6.

2. Time resolved fluorescence polarization anisotropy

The theory of fluorescence anisotropy decay was first developed by Perrin in 1934 [7]. The theory is mainly based on the work of Einstein [8] and Debye [9] on the rotational diffusion of molecules due to Brownian motion, and was presented in its complete and final version in the late 70’s [10,11]. The basic idea is that when randomly oriented ground state molecules are excited with short-pulse linearly polarized light, a partially oriented ensemble of excited molecules is created. Early fluorescence light will result to be partially polarized according to the average orientation of the excited molecules. At later times, as a consequence of rotational diffusion due to Brownian motion or energy transfer of the excitation energy to another molecular state or to a different molecule, the average orientation of the emitting dipoles changes, giving rise to a different (lower) degree of polarization. Eventually, at very long times after the exciting pulse, the fluorescence light will be completely unpolarized. When energy transfer between different molecular states or molecules can be neglected, the fluorescence polarization anisotropy measurement reveals the average angular displacement (rotation) of the fluorophore that occurs in the time lag between the absorption and the emission of a photon. This angular displacement depends on the rate and extent of rotational diffusion during the lifetime of the excited state. For a vertically polarized exciting field, the fluorescence anisotropy is detected by subsequent measurements of the vertically (I _‖(t)) and horizontally (I _⊥(t)) polarized components of the fluorescence light intensity emitted by the particles. It can be shown that these components are given by [12]:

where P ₂ is the second order Legendre polynomial, ê(0) and ê(t) are the directions of the transition dipole moment u at time 0 and t, respectively, and K(t) describes the temporal behavior of the fluorescence decay. The notation 〈P〉 refers to the ensemble average.

The phenomenon of fluorescence anisotropy is characterized by the parameter r(t), known as the fluorescence depolarization anisotropy or, simply, anisotropy, and defined by the following relation:

It can be shown that expected values of r(t) for t=0 and t→∞ are, respectively, 0.4 and 0 [13]. The factor of 2 in the second term of the denominator accounts for the existence of two distinct directions orthogonal to the exciting field, so that the whole denominator has the meaning of the total amount of emitted light. A calculation of I _‖(t) and I _⊥(t) in their time evolution leads, in the general case of an anisotropic diffusers, to an expression containing at most five decaying exponential functions [12].

In the frequent case when the directions of the absorbing and the emitting dipoles coincide with the same principal axis, the anisotropy decays with a double exponential. This may further reduce to the case of a spherical rotator (that is a good approximation in our case), when the emission anisotropy decays single-exponentially, namely:

where r ₀ is the initial value of r (expected to be equal to 0.4), τ_rot=1/(6D_r), and D_r is the isotropic rotational diffusion coefficient given by the Stoke-Einstein equation:

Here, k_B is the Boltzman constant, T and η are, respectively, the temperature and the viscosity of the medium, and V is the average volume of the diffusing particles. Thus, by determining r(t), one can evaluate the characteristic decay time τ_rot and consequently D_r, and, knowing the other experimental parameters (temperature and viscosity), it is possible to estimate the volume of the particles under study.

3. Experiment

The anisotropy coefficient, r(t), typically decays in a rather short time, which in our experimental conditions ranges between tens of picoseconds and few nanoseconds. According to Eq. (2), the determination of r(t) requires a direct measurement of both I _‖(t) and I _⊥ (t). Therefore, in order to obtain significant results, it is necessary that the excitation pulse duration is much shorter than the fluorescence decay time of the samples. Therefore, we employed 100 fs laser pulses in two complementary configurations of pulse energy and repetition rate, in order to match the requirements of two different light detection methods. The first one was based on a streak camera, whose versatile acquisition settings allow efficient detection of extremely dim light signals by both increasing the micro-channel plate (MCP) gain and averaging the light signal over a large number of events. In this case, low energy (10 nJ) pulses generated by a laser oscillator at high repetition rate (82 MHz) were chosen, and demonstrated to be a suitable choice for sample excitation. The alternative detection system employing a conventional light detector (fast photodiode or photomultiplier tube (PMT)) and an oscilloscope required larger pulse energies capable to produce intense fluorescence light, and, thus, allowing single shot detection of the time profiles of I _‖(t) and I _⊥ (t). In this case, use was made of a laser amplifier delivering 1 mJ pulses at 1 kHz repetition rate. The two systems will be described in the following subsections, and, throughout this paper, will be referred to as first method and second method, respectively.

3.1 Using a streak-camera

The excitation light pulse was provided by a Titanium:Sapphire (Ti:Sa) laser oscillator, delivering linearly polarized, 10 nJ, 100 fs pulses at about 800 nm wavelength and 82 MHz repetition rate. A schematic view of the optical layout is shown in Fig. 1. The laser beam was focused (f-number=500) onto a type-I beta-barium borate (BBO) crystal for second harmonic generation (λ_SHG=400 nm), with approximately 15% conversion efficiency. Doubling the laser fundamental frequency is necessary because all samples studied in this work are almost transparent at 800 nm wavelength, and present a broad absorption band in the ultraviolet region (typically around 200 nm) with a tail extending up to about 400 nm.

 

Fig. 1. Optical layout employed with a streak camera as a time resolved fluorescence light detector. BBO: second harmonic generator BBO crystal; HS: harmonic separator; CGF: color glass filter; HWP: half-wave plate; GT: Glan-Taylor polarizing cubes; L: lenses; S: sample cell; BPF: band-pass filter; SC: streak camera; PD: trigger photodiode.

Download Full Size | PDF

A harmonic separator (highly reflecting at 400 nm and transmitting at 800 nm) and a low-pass color-glass filter (with a cut-off wavelength of 480 nm) removed the residual infrared laser radiation. The combination of a half-wave plate and a Glan-Taylor polarizing cube achieved both a continuous variation of the pulse energy and a high contrast vertical polarization of the exciting light pulse. The 400 nm radiation was then focused by a 35 mm focal length lens onto the sample, which was suspended in a proper solvent contained in a 10×10 mm2 quartz cell. Optical densities of all samples were kept below 0.2 OD to avoid complications from self-absorption of fluorescence radiation.

 

Fig. 2. (a) Streak image of a 100 fs light pulse. The vertical width corresponds to 2.11 ns. (b) Intensity profile of the streak image with a FWHM of 12 ps.

Download Full Size | PDF

The fluorescence light was analyzed in a direction perpendicular to both the exciting radiation field and the laser propagation direction. A pair of collecting lenses (focal lengths of 35 mm and 100 mm) imaged the fluorescing region onto the entrance slit of a streak camera (Hamamatsu, mod. H6780). A Glan-Thompson polarizing cube was placed in front of the streak camera and alternatively allowed the analysis of the two polarization components of the fluorescence light, parallel (vertical) and orthogonal (horizontal) to the exciting field. An additional band-pass optical filter (Δλ=200 nm) centered on 550 nm got rid of the residual scattered 400 nm and 800 nm radiation. The streak tube was time-synchronized with the 82 MHz pulse train by means of a fast photodiode delivering the proper trigger signal. The time gate for the MCP gain was set at 56 ms (corresponding to an average over 4.6×10⁶ laser pulses) and we further averaged the streak signal over 1,000 gate events. The streak camera was controlled by a personal computer (PC) that also performed data acquisition and subsequent analysis.

The time resolution of the streak camera was estimated as the width of the acquired intensity profile corresponding to a very short light event. When using scattered light from the 400 nm beam, lasting as short as 100 fs, with the acquisition time window set to 2.11 ns, we measured a full width at half maximum of about 12 ps, which is assumed as the time resolution at typical operating conditions. Figure 2(a) shows a streak image of scattered 100 fs light pulses, and the corresponding, nearly Gaussian intensity profile (Fig. 2(b)) is assumed as the instrument response function f_r(t).

In order to spectrally resolve the fluorescence light, a spectrograph should be coupled to the streak tube that, unfortunately, was not available during our work, and all spectroscopic results were obtained with the experimental setup described in the following subsection.

3.2 Photo-detector – oscilloscope system

Single shot detection of the fluorescence intensity profiles required larger excitation pulse energies. To this end, the oscillator laser pulses were amplified by a Ti:Sa regenerative amplifier based on the chirped-pulse amplification (CPA) method. The amplifier output consisted of 1 mJ, 100 fs pulses at 800 nm wavelength with a relatively low (1 kHz) repetition rate. The laser beam presented a nearly Gaussian transverse profile with a diameter of about 7 mm. An iris diaphragm was used to select the more homogeneous central part of the laser beam with a diameter of about 2 mm. The second harmonic of the laser fundamental frequency was obtained in the same BBO crystal described in the previous subsection without beam focusing, with conversion efficiency up to 20%. In order to benefit of the large absorption band centered at 250 nm presented by many carbonaceous samples, we could also mix the fundamental laser radiation (λ=800 nm) and its second harmonic (λ_SHG=400 nm) in a second BBO crystal for generating third harmonic UV light pulses (λ_THG=266 nm).

 

Fig. 3. Optical layout for detection of the fluorescence time profiles with a monochromator (M), a photomultiplier tube (PMT) and an oscilloscope (O).

Download Full Size | PDF

The interaction of the frequency up-converted light pulses with the sample and the collection of the fluorescence light was realized with an optical layout similar to that described above. Two minor differences were: i) A 250 mm focal length lens was used for focusing the laser beam onto the sample; ii) A single collecting lens (100 mm focal length) was employed for imaging the fluorescing region onto the entrance slit of the detection system (see Fig. 3). The fluorescence light was analyzed by a grating monochromator tuned within the sample fluorescence band, and detected by a photomultiplier tube (Hamamatsu, mod. H8670) placed at the monochromator exit port. The voltage signal produced by the PMT was, then, fed into a 2 GHz bandwidth digital oscilloscope.

As in the previous case, we determined the instrument response function of the PMT-oscilloscope system by recording the intensity profile produced by scattered 400 nm, 100 fs laser light. The response function resulted to be nearly Gaussian with a full width at half maximum τ_rf=0.84 ns, thus comparable to the decay times to be measured (up to few nanoseconds). Figure 4 represents the detected signal corresponding to a 100 fs light pulse. As in the previous case, it is assumed as the instrument response function f_r(t).

 

Fig. 4. Instrument response function of the PMT-oscilloscope system.

Download Full Size | PDF

3.3 Sample collection

In the present subsection, we briefly describe the methods employed for collecting carbonaceous particles produced in laminar laboratory flames and in vehicle engines.

Ethylene/air flames were generated by a bronze-plate, water-cooled McKenna burner, stabilized with a stainless steel plate. The cold flow velocity was 9.0 cm/s, and ethylene/air (combustible/oxidant, C/O) ratios of 0.77 and 0.92 were investigated. The products of these flames were studied with extra situ sampling techniques, using a stainless steel, water cooled probe. Combustion products were bubbled in bi-distilled water, where hydrophobic products were removed by a vacuum pump and only hydrophilic particles were retained. The latter fraction of combustion products is of great importance in the environmental pollution, because it tends to stick to atmospheric water and represents a serious risk to human health. Samples collected from these slightly sooting flames were probed at a height of 3.5 mm above the burner surface. Since the soot-formation region started at about 5 mm, the samples were substantially soot-free. For more details on the sampling procedure see Ref. [14]. Finally, a quartz cell of 10 mm side was filled with water samples (approximately 20 ppm).

We have also analyzed carbonaceous particles produced during the combustion process occurring in two vehicle engines, namely, a direct-injection, spark-ignited engine, fuelled with unleaded gasoline, and a high-pressure injection (common rail) diesel engine, fuelled with commercial, low-sulfur light oil. The vehicles were run at the chassis dynamometer on the standard New European Driving Cycles (NEDC) [15]. Our water diluted samples were collected directly at the exhausts of the two engines, and particular care was made in order to obtain a high concentration of nanometric organic compounds (NOC) without interfering soot particles. The concentration of NOC in water was deduced from their UV-visible extinction spectra, and resulted to be about 550 ppm and 880 ppm for the gasoline and diesel sample, respectively [16].

4. Data analysis

In most instances, the experimental determination of the anisotropy coefficient, defined in Eq. (2), requires spectral filtering of the fluorescence radiation. As a consequence, the really measured quantities are not the horizontal and vertical components of the fluorescence light intensity emitted by the sample, but those actually passing through the detection system. Thus, it is necessary to take into account the different transmission efficiencies of the two polarization components, which is done by inserting in Eq. (2) a correction factor, G [13]:

Here, the quantities I _‖ and I _⊥ refer to the measured intensities. The G factor is defined as the ratio I _‖/I _⊥. It was simply estimated by illuminating our detection systems with an unpolarized light emitted by a Tungsten lamp and measuring the transmitted intensity for both positions (horizontal and vertical) of the polarizer in front of the monochromator. Its value was 1.02±0.04 when the wavelength selection was obtained by means of band-pass filters, and 1.42±0.02 when we used the monochromator. Moreover, the G value resulted nearly constant in the investigated wavelength range (400–600 nm).

The measured intensity profiles of the quantities I _‖ and I _⊥ corrected by the proper G factor would represent the true emitted profiles only if the time width of the apparatus response function resulted much shorter than the typical decay times. In our experiment, this was the case only when we used the streak camera. On the contrary, due to the relatively small bandwidth of the PMT-oscilloscope system, in this case the instrument response function (f_r(t)) had a time width only slightly smaller than the characteristic decay times.

Generally speaking, any detected light signal, A_obs, is the time convolution of the true profile, A_true, and the response function, f_r:

By knowing the latter, it is possible to retrieve the true profile through a common algorithm, based on the method of Least Squares Iterative Reconvolution [17]: Starting from the assumption that A_true is a combination of exponential decay functions, the method consists in assigning suitable initial values to its parameters and convolving this test function A_true with the measured instrument response function. The parameters of A_true are then varied until the least square differences with the measured profile, A_obs, are minimized. Obviously, the method is the more sensitive and the results are the more accurate, the shorter is the response function time width. In fact, we tested the algorithm for signals detected with the streak camera and observed a very quick convergence to a retrieved true function that resulted substantially undistinguishable from the detected signal A_obs, that would occur (see Eq. (6)) if f_r(t-t′)≈δ(t-t′), where δ is the Dirac delta function.

As an example, in Fig. 5(a) we show the time profiles of I _‖ (open red diamonds) and I _⊥ (blue squares) for a gasoline sample at a concentration of 550 ppm. Due to the high time resolution, the two curves result consistently different at short times, thus, allowing the determination of the anisotropy ratio. In fact, Fig. 5(b) reports the anisotropy ratio (green squares) as defined by Eq. (5), and the best fitting double exponential decay (orange line). It can be seen that r(t) is sufficiently clean as to allow a direct fitting with a double exponential function.

 

Fig. 5. (a) Time profiles of I _‖ (red open diamonds) and I _⊥ (blue squares) for a gasoline sample. (b) Anisotropy ratio r(t) (green squares) and double exponential fitting function.

Download Full Size | PDF

On the contrary, the reconvolution procedure resulted less rapidly converging for decay profiles detected by the PMT-oscilloscope system, due to its slow response function. In such cases one has to proceed in a different way: It is necessary to separately apply the above described method to the exponentially decaying quantities U(t) and S(t) (with decay times τd and τ_f, respectively) [13], defined by the following equations:

so that:

According to Eq. (7-b), τ_f is the overall fluorescence lifetime, whereas τ_d is given by [12]:

Thus, from determining τ_f and τ_d, it is possible to evaluate the rotational time τ_r=τ_dτ_f/(τ_f-τ_d), and, knowing the temperature and the viscosity of the liquid, the hydrodynamic radius rhy of the spherical fluorescing particle (see Section 2) can be evaluated:

As an example of the processing procedure, we report in Figs. 6(a) and 6(b), respectively, the time profiles of U(t) and S(t) (green solid lines) as defined in Eqs. (7), for a sample collected from a laminar flame with a combustible/oxidant (C/O) ratio of 0.92. Because of the low time resolution available with the second method, the time profiles of I _‖ and I _⊥ are very similar even at short times, which implies very small values and extremely noisy behavior of the anisotropy ratio. In such a case, we separately perform a reconvolution least square minimization for the quantities U(t) and S(t) (see Eq. (6)). The best fitting convolutions are shown with orange lines in Figs. 6(a) and 6(b). Both test functions are single-exponential decays, because the low time resolution does not allow the observation of a fast (<100 ps) decay. From their decay times τ_d and τ_f, corresponding to U(t) and S(t), respectively, one can determine the rotational decay time, τ_r, according to Eq. (9).

 

Fig. 6. Time profiles of U(t) (a) and S(t) (b) (green lines) observed for a sample collected from a laminar flame (C/O ratio of 0.92). The orange lines represent the best fitting convolutions of single exponential decay functions with the instrument response function.

Download Full Size | PDF

Most of our samples have been analyzed with both experimental methods described above. For the sake of brevity, we do not report the corresponding comparison, but it is worth remarking that the results obtained with the second experimental method and processed with the above described reconvolution procedure proved to be in good agreement with data collected with the first one, thus confirming the validity of the second method. The main difference consisted in a larger uncertainty on the estimated decay times.

5. Results

5.1 Measurements on molecular standards

The apparatus was first tested on some standard molecules of known dimensions, with the aim to better characterize the experimental set-up and determine the optimal conditions for reliable measurements. The samples used as reference standards were commercial Rhodamine 6G dye in water and solar dye in N-methyl-2-pyrrolidinone (NMP). Their structure, shape and dimensions are well known and have been previously characterized using the TRFPA technique [18,19,20]. The products were supplied by Sigma-Aldrich and used without further purification. All the measurements were realized at room temperature (T=300 K) and no significant increase in the sample temperature was observed during laser irradiation.

For Rhodamine 6G, the mean diameters determined with the streak camera and with the PMT-oscilloscope detection methods were, respectively, (1.20±0.04) nm and (1.3±0.3) nm. The two estimates of the diameters are consistent with each other and in excellent agreement with other authors’ results, reporting values ranging from (1.20±0.08) nm [18] to (1.12±0.06) nm [19], and obtained in experimental conditions similar to ours.

The mean diameter of solar dye molecules resulted to be (1.75±0.05) nm and (1.6±0.3) nm, respectively, with the first and second method. These results are comparable with the value (1.82±0.03) nm reported in Ref. [20], where toluene was employed as solvent. The advantage of using NMP instead of toluene resides in the different viscosities (16.7 mP and 5.6 mP, respectively) of the two solvents. In fact, according to Eqs. (3) and (4), the larger viscosity of NMP implies longer rotational times compared to toluene, which increases the sensitivity and reliability of our method.

An interesting additional test was performed on C₇₀ fullerene dispersed in toluene. These was the first carbonaceous material investigated with our two methods. As a matter of fact, the characteristic rotational times of C₇₀ molecules are too short for detecting the rapidly decaying anisotropy with the low time resolution of the second method. On the contrary, the first method proved to be capable of determining the time profile of the anisotropy function, as was also reported in Refs. [21,22] with the time correlated single photon counting technique.

Table 1. Measured diameters of different test molecules.

View Table | View all tables in this article

The fluorescence anisotropy of C₇₀ allowed an estimate of the characteristic molecular diameter that resulted to be (0.73±0.07) nm, in agreement with previous measurements [23].

All the results on molecular standards are summarized in Table 1 and compared with existing data.

5.2 Laminar flame samples

The combustion of hydrocarbons in fuel-rich conditions gives rise to the formation of solid soot particles as well as many other hydrocarbons and carbonaceous materials, in particular, polycyclic aromatic hydrocarbons (PAH), tar-like materials [24,25], and nanometric organic compounds (NOC), with size of about 2 nm and transparent to the visible light [26,27]. The soot growth mechanism is an extremely complicated process, and is still under debate. It basically consists of growth of hydrocarbon fuel molecules, containing few carbon atoms, up to the birth of soot precursors that finally originate the first soot particles containing as many as 10⁵ carbon atoms. In particular, the characteristics of soot precursors, which can have very different physical-chemical properties with respect to those of the larger soot particles, are rather obscure [28,29].

 

Fig. 7. Absorption spectra of NOC samples collected from laminar flames at different C/O ratios.

Download Full Size | PDF

Laminar flames allow to investigate the evolution of various molecular and particle species occurring during the entire combustion process: The height from the burner surface is an increasing linear function of the time a given species has spent within the flame[3].

In Fig. 7, we show the absorption spectra of the flame samples collected according to the procedure described in Sec. 3.3. Besides the strong peak around 200 nm, we can still observe a residual absorption at 266 nm and 400 nm. Figure 8 and its inset show the fluorescence spectra for excitation wavelengths of 266 nm and 400 nm, respectively. The spectra for the two samples mainly differ in the absolute fluorescence intensity, but have very similar shapes, with a peak value around 310 nm and a second band in the visible, and with the maximum at 480 nm for excitation wavelength of 400 nm (the spectra in the inset).

 

Fig. 8. Fluorescence spectra of the two laminar flame samples excited at 266 nm. The inset shows the fluorescence spectra for 400 nm excitation wavelength.

Download Full Size | PDF

The hydrodynamic diameters of the sample particles were measured with the TRFPA technique with excitation at both 266 nm and 400 nm and selecting different fluorescence wavelengths within the fluorescence band. Wavelength selection was achieved by the grating monochromator and the intensity time profile detection was performed with the PMT-oscilloscope system. The experimental results are summarized in Fig. 9, where the measured particle sizes are plotted as a function of the emission wavelength. For the sake of clarity, it is worth to remind that the experimental points at 330 nm were obtained following excitation at 266 nm, whereas for the remaining diameters, the exciting radiation wavelength was 400 nm. Similar trends are observed for the two C/O ratios, an indication that the composition of the samples was substantially the same. In particular, we can infer that the samples mainly consist of two species of particles, one fluorescing in the 300–470 nm spectral range and with an average size of about 1.7 nm, and the other emitting between 490 nm and 570 nm and with a mean diameter of 2.3 nm.

 

Fig. 9. Spectral distribution of particle diameters collected from laminar flames with C/O ratios of 0.77 (blue squares) and 0.92 (red bullets).

Download Full Size | PDF

The two species of particles characterized by different size and fluorescence wavelength can be associated with the number of aromatic rings constituting them. In particular, 1.7 nm particles with fluorescence wavelength below 480 nm basically consist of large molecules with two or three aromatic rings, whereas larger (2.3 nm size) particles fluorescing at longer wavelengths consist of molecules with at least four aromatic rings. This interpretation is in agreement with the kinetic model introduced by D’Anna and Violi [30] who predicted that primary particles produced during a combustion process consist of a structure with two aromatic rings, whereas the growth process proceeds with subsequent aromatization, thus increasing the number of rings within the particle structure.

5.3 Gasoline and diesel engine exhaust samples

The size distribution of particles produced in premixed laminar laboratory flames is unambiguously determined by the combustion conditions, which can be easily controlled and monitored during the combustion process [31,32]. On the contrary, for combustion occurring inside engines, it is very difficult to monitor and control the combustion conditions, and, obviously, to predict the generated particle distribution.

Recently the experimental characterization of carbonaceous particles collected from the exhausts of modern combustion sources (vehicle engines and industrial burners) has shown a bimodal distribution [33,34,35]. Particle distribution basically consists of soot particles with a rather large average diameter (20–60 nm), and nano-sized organic compounds (NOC), whose size is about an order of magnitude smaller (2–5 nm). Very often, the smallest particles’ size is well below the detection threshold of commercially available particle sizing equipments. As a consequence, there is a large demand of particle sizing techniques being sensitive to the full size range of the bimodal primary particles formed in combustion reactions and found both in combustion exhausts and in the atmosphere.

Both samples from diesel and gasoline engines present a non-vanishing absorption at 400 nm. Moreover, the overall shape of the absorption profile shows a striking similarity with the absorption spectra of samples collected from laminar flames in sooting and non-sooting conditions [16,33].

 

Fig. 10. Comparison of fluorescence spectra of water samples collected from diesel (dark green triangles), gasoline (light green diamonds) engine exhausts and from laminar flames with C/O ratios of 0.77 (blue squares) and 0.92 (red bullets). The excitation wavelength was 400 nm.

Download Full Size | PDF

Figure 10 reports the fluorescence spectra of the two engines’ samples excited with a radiation of 400 nm wavelength. Both spectra present a broad band in the range from 450 to 650 nm, which was used for TRFPA measurement of the average particle size. In Fig. 10 the two spectra are also compared with those corresponding to samples collected from laminar flames in non-sooting conditions (C/O ratios of 0.77 and 0.92). Besides the absolute values, all the spectra have very similar shapes.

An example of fluorescence decay has already been shown in Fig. 5(a), for the case of a gasoline sample with a concentration of 550 ppm. Also reported in Fig. 5(b) is the time profile of the corresponding anisotropy ratio r(t). The best fitting function is a double exponential decay, with estimated time constants τ ₁=(27±15) ps and τ ₂=(200±20) ps. The second value is basically due to the rotational diffusion of the particle, corresponding to an average diameter of (1.21±0.09) nm. The shorter decay time is probably connected to faster decays of internal degrees of freedom of the whole particle, namely, internal rotation of the chromophore or energy transfer and energy migration among different components of the carbonaceous particle. The correct interpretation of the fast decay time and the determination of its origin would need further investigation, as, e.g., spectral resolution of the fluorescence radiation [13]. It is worth noticing that a double exponential decay is only observable when a high time resolution is achievable, which, in our case, occurs with the use of the streak camera.

Table 2. Decay times of fluorescence anisotropy from NOC particles collected at the exhaust of gasoline and diesel vehicle engines at two different concentrations, and corresponding average diameters.

View Table | View all tables in this article

In order to verify whether any concentration effect occurred on the measurement of the particle diameter or not, both samples were diluted at 10 times smaller concentration. The analysis was carried out for a gasoline sample at 55 ppm and for diesel samples at both 880 ppm and 88 ppm. The results are summarized in Table 2. From these data, it is clear that a larger effort is necessary in order to reduce the relative error on the “fast” decay times, so as to understand their origin. Moreover, the sample concentration does not seem to have a uniquely determined effect on τ2 and more detailed measurement at different concentrations are necessary in order to investigate the role of particle coagulation within the samples.

6. Conclusions

In this paper, the time resolved fluorescence polarization anisotropy technique was used to determine the average size of combustion-generated carbon nanoparticles. In particular, we focused our attention on NOC particles collected from ethylene/air laboratory laminar flames and from the exhausts of gasoline and diesel vehicle engines. A significant characteristic of the investigated samples is the non-vanishing value of their absorption coefficient at wavelengths ranging between 200 nm and 400 nm. This allowed the use of a Ti:Sa laser source with relatively simple frequency doubling and tripling. Moreover, they also present strong fluorescence bands in the visible range, so that the analysis and detection of fluorescence radiation did not require sophisticated equipments, nor complex labeling of particles with fluorophores.

Although the sample fluorescence is induced by a rather complex and expensive femtosecond laser system, it must be pointed out that, nowadays, similar laser sources are available in many research centers, where they are run as laser facilities. Therefore, when the laser source is already available as a beam line, amongst various experimental techniques for determining the time profile of the fluorescence light, the one described in this paper based on conventional light detectors and oscilloscopes, presents some advantages: It is very simple and cheap, does not require additional electronics (as for time correlated single photon counting), and results in very short acquisition times. In spite of its simplicity, it lends itself to measurements on a wide range of molecule and particle sizes, providing results that are in good agreement with those obtained with more expensive equipments. The drawback of the method described in the present paper is the smaller accuracy, compared to other techniques. On the other hand, we also report on the use of a streak camera for determining the time profile of the fluorescence radiation. This method achieves a much higher time resolution in determining the anisotropy ratio decay. When ellipsoidal particles are analyzed, the lack of spherical symmetry implies that the anisotropy ratio decays with a multi-exponential behavior with time constants of the same order of magnitude, which can only be observed with high time resolution measurements [13]. It must be pointed out that in the present work we always employed the spherical approximation, so that the measured diameter should be considered as that of a spherical particle having the same volume of the actual one. However, we are carrying out further studies aimed to the determination of the particles’ shape from high resolution TRFPA measurements.

We have also demonstrated that the high time resolution of the first method allows the observation of a double exponential decay of the anisotropy ratio, with two time constants differing about by one order of magnitude. The longer one (of the order of hundreds of picoseconds) is related to the particle rotational diffusion, and, thus, allows an estimate of the average particle diameter, whereas the shorter one (about 40 ps), not observable with low resolution methods, may be attributed to a fast decay of some internal degree of freedom.

Finally, when spectral resolution of the fluorescence light is available, one can also deduce some information on the sample composition by analyzing the spectral distribution of the measured particle diameters.

An interesting aspect of our investigation methods is the fact that their sensitivity is limited to the fluorescent fraction of particles within the sample. As a consequence, they can be adopted to determine the average size of the smallest fluorescing fraction of carbon particles sampled from combustion systems, without the interference of larger, non-fluorescing particles. On the other hand, when analyzing a completely unknown sample, nothing can be said on its composition, because non-fluorescing particles cannot be detected.