1. Introduction

Biomass is a renewable feedstock that can be used as an alternative to fossil fuels in thermochemical conversion processes for production of heat, power and chemicals [1,2]. The chemical composition of biomass is significantly different from that of fossil fuels and varies considerably between different types of biomass. Common to most types of biomass is that they contain high amounts of potassium (K), which is to a large extent released to the gas phase as potassium hydroxide (KOH), potassium chloride (KCl) and atomic potassium, denoted K(g), during conversion [3,4]. The K species are subsequently involved in ash, slag and aerosol formation, which, in turn, may lead to severe depositions, fouling, reactor material corrosion and emission of harmful fine particles, even at gaseous K species concentrations in the parts per million (ppm) range [5,6].

Numerical simulations of single solid particles [7,8], the gas-phase reactions [9] and pulverized biomass combustion [10] are commonly employed to understand and optimize the involved process chemistry and flow dynamics. Recently, two-dimensional (2D) axisymmetric computational fluid dynamics (CFD) frameworks have become instrumental to aid characterization of processes, where a one-dimensional (1D) approximation of the flow is not applicable [11]. To validate the numerical models and mitigate the operational issues, fast, quantitative in situ detection of the K species close to the fuel particles during conversion is needed. This can be achieved using optical measurement techniques [12].

Optical methods for in situ detection of alkali hydroxides and chlorides at high temperatures include broadband ultraviolet (UV) absorption spectroscopy [13,14] and photofragmentation (PF) techniques with fluorescence- [15,16] and absorption-based [17,18] fragment detection. K(g) can be measured with a dynamic range of up to 10⁶ employing tunable diode laser absorption spectroscopy (TDLAS) in the visible or near-infrared [19]. UV absorption spectroscopy allows for compact setups, but lacks sensitivity (low ppm levels require a path length of ∼1 m) and may suffer from spectral interference, which hampers selective quantification of KOH and KCl in complex matrices. A higher sensitivity (low ppm levels at 0.01 m path length) has been achieved with photofragmentation fluorescence, where the K(g) fluorescence is detected after fragmentation of K species [16]. The precursor molecules can be distinguished, if the fragmentation of KOH and KCl results in K fragments in different excited states. While fluorescence techniques enable spatially resolved detection, they have higher demands on optical access compared to line-of-sight methods, require frequent calibration, and may suffer from self-absorption. By detecting the fragments with absorption methods, the sensitivity of PF techniques can be further increased, and calibration can be circumvented.

One of the most sensitive techniques for detection of the K compounds is collinear photofragmentation and atomic absorption spectroscopy (CPFAAS) [17,18]. This method uses two pulsed UV lasers to fragment KOH and KCl, and a collinearly aligned probe laser, whose frequency is locked to an atomic K transition, to detect the background atomic K(g) concentration, as well as the additional K(g) present just after the UV pulse. The momentary increase in K(g) absorption after the UV pulse can then be used to calculate the precursor molecule concentration. This technique has the advantage of being self-calibrating, if the spectroscopic parameters of precursor molecule and fragment transitions are known [17]. However, the method is less suitable for optically thick conditions, i.e. when the product of K(g) line strength, concentration and path length becomes large and the transmitted light intensity at the absorption line center is below the detector threshold [19]. This is likely to happen in gasification processes (high atomic K), when converting K-rich fuels and/or when measuring in large-scale facilities [14,20,21].

In order to reliably quantify the K species under optically thick conditions, a weaker K(g) transition can be used [14], or the fixed probe laser wavelength can be positioned in the wings of the K(g) line shape. A recently introduced third solution is to combine photofragmentation with TDLAS (PF-TDLAS), such that the probe laser is scanned across the K(g) line shape and the UV pulses are triggered to occur at a certain position in the line shape profile [22]. This facilitates flexible detuning of the UV pulse position within the K(g) line shape and accurate K(g) quantification by fitting the unsaturated wings of the absorption profile [14,19]. The PF-TDLAS method has so far been demonstrated for detection of K(g) and KOH in premixed methane/air flat flames.

In this work, the PF-TDLAS technique is extended to include detection of KCl, and the sensitivity of the setup is further improved by employing a high power probe laser. A theoretical description of the PF-TDLAS signal and the optimal UV pulse position in the K(g) line shape is presented and experimentally verified. The system is validated for KCl detection in a tube furnace, followed by axial measurements of all three K species in a fuel-lean and a fuel-rich flat flame seeded with KCl salt. The obtained line-of-sight K species concentrations are compared to 2D axisymmetric reacting flow simulations of the flames, including the K species reactions. Finally, the method is employed to quantify the release of K species from biomass particles during conversion in the flat flames.

2. Methods

2.1 Photofragmentation atomic absorption spectroscopy

Photofragmentation (also photolysis or photodissociation) is the process of dissociating molecules into fragments using light. The target molecule absorbs a high-energy photon, typically supplied by a pulsed UV laser, which excites the molecule to a repulsive state that causes the molecular bond to break. The molecule is momentarily dissociated into fragments, which then recombine again to restore chemical equilibrium. Reaction kinetics studies of K species revealed recombination rates on the order of ns to µs (depending on the process conditions) with exponentially decaying fragment absorbance [23]. Properties of the precursor molecules can then be indirectly determined via the fragments.

For absorption-based fragment detection with CPFAAS, an expression for the PF-signal intensity, I_PF, has been given by Sorvajärvi et al. [17]. Assuming that the photofragmentation event occurs at time zero (t=0), the probe laser intensity is given by

2.2 Tunable diode laser absorption spectroscopy

In TDLAS, the laser frequency, ν, is tuned across one or several absorption lines of a target atom or molecule, typically at rates of tens of Hz to kHz. This allows detection of the absorption profile, whose shape and width depend on the surrounding environment (temperature, pressure, gas matrix) and are usually unknown prior to the measurement. For laser light passing through an absorbing gas sample, the relation between incident intensity, I_0,T, and transmitted intensity, I_T, is given by Beer’s law according to

2.3 Photofragmentation tunable diode laser absorption spectroscopy

By combining PF and TDLAS, the K(g) absorption line shape can be measured simultaneously with the PF-induced fragment absorption, which allows accurate determination of the background K(g) concentration. Moreover, the UV pulse position in the K(g) line shape, i.e. the probe laser wavelength in CPFAAS, can be freely chosen, which extends the dynamic range of KOH and KCl detection. However, the analytical signal obtained with PF-TDLAS (the PFT-signal) will now include the (slow) variations due to the TDLAS-scanned fragment line shape. The PFT baseline intensity is then given by

Following the frequency dependence of the absorption cross section, the magnitude of the PF-absorbance will decrease with detuning from the center of the K(g) line shape. On the other hand, also the noise in the PF-absorbance will decrease with detuning from line center, due to the higher transmitted light intensity. There is thus an optimum position of the UV pulse in the K(g) line shape, where the sensitivity is maximized. This position can be found by analyzing the signal-to-noise ratio (SNR) of the PF-absorbance as a function of probe laser frequency. The noise in the PF-absorbance, Δα_PFmax, can be estimated from the SNR of the PFT baseline intensity as

2.4 Experimental PF-TDLAS setup

The experimental setup of the PF-TDLAS system is schematically shown in Fig. 1(a). An external cavity diode laser (ECDL, New Focus TLB 6700-LN) with a beam diameter of 2.3 mm was used as probe laser to detect the background K atoms by TDLAS and to probe the atomic K(g) fragments induced by photofragmentation. The ECDL was scanned over 0.158 nm (2.7 cm^-1) across the potassium D1 line at 769.9 nm at a rate of 40 Hz. Pulsed diode pumped solid state (DPSS) lasers at 266 nm (DPSS1, CryLas FQSS 266-200) and 355 nm (DPSS2, CryLas FTSS 355-300) operating at a repetition rate of 40 Hz and synchronized with the TDLAS scan were employed to fragment the KCl and KOH molecules, respectively. Synchronization of the UV laser pulses with the TDLAS scan resulted in one KOH and one KCl PFT-signal per K(g) line shape. The 266 nm and 355 nm UV pulses were separated in time by 155 µs to avoid overlapping PFT-signals.

 

Fig. 1. (a) Schematic diagram of the experimental PF-TDLAS setup. ECDL - probe laser, DPSS - fragmentation laser, C - collimating lens, DM - dichroic mirror, BS - beam splitter, UVD - UV detector, F – optical filter, L - lens, HBWD - high bandwidth detector, LBWD - low bandwidth detector, Dump - UV beam dump, B - burner. (b) Burner top view showing the laser beam (red arrow) above the platinum sample plate. (c) Burner side view without flame (left) with the axisymmetric measurement and simulation domain (box) highlighted, and with flame and plume during KCl sample conversion (right).

Download Full Size | PDF

The power of the probe laser light incident on the sample was 8 mW (still significantly below the limit for optical saturation), and the pulse energies of the 266 nm and 355 nm lasers were 97 ± 0.6 uJ and 240 ± 1.5 uJ, respectively, as measured prior to experiments with an energy meter (Thorlabs ES111C). In addition, beam samplers (BS1, Thorlabs BSF10-UV) reflecting 1% of the UV light were inserted before and after the sample volume to continuously monitor the incident and transmitted pulse energies using photodetectors (UVD, Thorlabs DET10A2) equipped with attenuating filters (F1, Thorlabs NDUV20A and NDUV10A) and focusing lenses (L1, Thorlabs LA4917-UV). Stable pulse energies were observed throughout the experiments, including biomass conversion, and there was no measurable attenuation after the sample volume. This implies that UV light extinction and scattering losses due to soot or polycyclic aromatic hydrocarbons were negligible, and thus motivates the use of Eq. (2) for calculation of the KOH and KCl concentrations.

The two UV laser beams were collimated to beam diameters of 2.1 ± 0.1 mm (266 nm) and 2.2 ± 0.1 mm (355 nm) and collinearly aligned with the probe laser beam using dichroic mirrors (DM1, Semrock FF376-Di01 and DM2, Semrock FF310-Di01-25 × 36). After passing the sample volume, the UV beams were again separated from the probe beam employing a third dichroic mirror (DM3, Semrock FF365-Di01-25 × 36). Using a beam splitter (BS2, Thorlabs BSX11), the major part (90%) of the probe laser light was focused onto a high bandwidth detector (HBWD, Thorlabs DET10A), while the remaining light was directed to a low bandwidth detector (LBWD, Thorlabs PDA36A-EC). Both detectors were equipped with spectral band-pass filters centered at 770 nm (F2, Thorlabs FB770-10) and focusing lenses (L2, Thorlabs LA1131-B), which were slightly tilted to suppress etalon effects and avoid back-reflections. The electrical HBWD signal was amplified (SRS SR445A) prior to data acquisition.

A low-bandwidth digitizer (National Instruments PXIe-6356) was used to sample the LBWD signal at 1 MHz and to generate the 40 Hz probe laser scan (triangular wave) and UV laser trigger signals (square waves). The HBWD and UVD signals were sampled at 1 GHz using a high-bandwidth digitizer (GaGe RMX-161-G40), whose acquisition was triggered by the UVD installed prior to the sample volume. The position of the PFT-signal in the K(g) line shape was chosen by controlling the relative phase between the probe laser scan and UV laser trigger signal.

2.5 Flat flame burner and KCl sample preparation

A water-cooled flat flame burner with a diameter of 38 mm (Fig. 1(b) and (c)) based on a design by Hartung et al. [26] was used for sample conversion. The burner was operated on a CH₄/air mixture controlled by mass flow controllers (MKS GM50A) and included a nitrogen shroud (flow rate 5 L/min). To enable radial and axial displacements of up to 25 mm, the burner was mounted on a set of translation stages. Samples (KCl salt or biomass) were placed on a round platinum plate with a radius of 3 mm, which was suspended at a height above burner (HAB) of 2 mm using platinum/rhodium wires, which also served as type S thermocouple. The KCl samples were prepared by placing droplets of KCl solution on the platinum plate, which was then dried on a hot plate and weighted using a laboratory scale with 10 µg resolution (Mettler Toledo XS205). The weight of the KCl samples was approximately 3 mg.

A fuel-lean flame at fuel-air equivalence ratio ϕ = 0.8 and a fuel rich flame at ϕ = 1.3 were used in this work. The CH₄ and air flow rates for the two mixtures, the temperatures of the platinum plate and the flame temperatures at a height above the plate (HAP) of 15 mm are presented in Table 1. The flame temperatures were determined using a separate wavelength modulation spectroscopy system operating at 1400 nm [27]. During sample conversion in the flames, K species concentrations were measured either as function of conversion time at a fixed HAP (20 averages, 0.5 s time resolution) or as a function of HAP (100 averages, 2.5 s time resolution) by shifting the burner vertically. The optical path length was determined by shifting the burner horizontally, perpendicular to the laser beam direction, and measuring the K species absorption. The resulting K species distributions resembled Gaussian functions, from which the path length could be estimated to 21 ± 1 mm, similar to what was reported in [22] and [28].

Table 1. CH₄ and air flow rates, plate and flame temperatures at HAP 15 mm for the two flame conditions.

View Table | View all tables in this article

2.6 PF-TDLAS signals and species quantification

Typical TDLAS-measured K(g) absorption spectra and PF-absorbance signals measured in the fuel-lean and fuel-rich flames are presented in Figs. 2(a) and 2(b), respectively. Figure 2(a) also shows the trigger signals of the two UV pulses (blue dashed lines) in the wings of the respective line shape, separated by 155 µs. The UV pulse detuning from line center was about 0.2 cm^-1 for the fuel-lean case and 1.1 cm^-1 for the fuel-rich case.

 

Fig. 2. (a) Typical K(g) line shapes measured with TDLAS (markers) at HAP 5 mm in the fuel-rich flame (circular markers) and HAP 20 mm in the fuel-lean (square markers) flame, together with Voigt fits (red solid lines) and UV pulse trigger signals (blue dashed lines), which indicate the PFT-signal position (0.2 cm^-1 and 1.1 cm^-1 detuning). (b) Typical PF-absorbance signals following the 355 nm pulse (black lines, square markers) and the 266 nm pulse (blue lines, circular markers) together with curve fits (red solid lines). For clarity, a reduced number of experimental data points is shown.

Download Full Size | PDF

The concentrations of KOH and KCl were calculated using Eq. (2) based on the maximum PF-absorbance in the decay curves in Fig. 2(b) and absorption cross sections for KOH and KCl reported in [14]. The absorption cross section for atomic K(g) was calculated from the line strength reported in [19]. While KCl only absorbs at 266 nm, KOH absorbs at both 266 nm and 355 nm. Therefore, the KOH concentration is first evaluated at 355 nm, and then the influence of the KOH absorption on the 266 nm PFT-signal is taken into account, when determining the KCl concentration.

2.7 Furnace setup for validation of KCl detection

The PF-TDLAS system was previously validated for KOH detection [22]. Here, validation for KCl was achieved using a laboratory furnace [19] to produce stable, saturated KCl vapor in the temperature range 676-1023 K, and by comparison of the measured path-averaged KCl concentrations to those predicted by thermochemical equilibrium calculations (TEC). KCl salt was distributed along an open-ended ceramic tube of length 265 mm and diameter 15 mm placed inside the furnace, with the laser beams passing above the salt. Prior to each measurement, the furnace set temperature was kept constant for 10 minutes to ensure stable gas concentration. For each set point, the temperature profile along the tube was measured using a type-N thermocouple. The average temperature was then used to evaluate the experimental PFT data. The equilibrium KCl concentrations were obtained by averaging over the concentration profile obtained based on the temperature profile and TEC. It should be noted that at average furnace temperatures >850 K, the UV pulse was attenuated while propagating through the sample. Thus, Eq. (2) was no longer valid, and a more general expression for the target molecule concentration was used [17].

2.8 Thermochemical equilibrium calculations

To estimate the final state of the K species concentrations in the different processes, TEC were performed using FactSage 8.0 [29], where the Gibb’s free energy of a closed system of compounds is minimized. The TEC system considered for validation of KCl detection in the furnace consisted of air (O₂ and N₂) and KCl salt at atmospheric pressure. The amount of solid KCl in the calculation was increased until saturated conditions for KCl vapor were reached, i.e. when further addition of KCl resulted in increased liquid, but not gaseous, KCl.

For the flat flames, the TEC system consisted of CH₄, air and KCl at atmospheric pressure and temperatures of 1830 K and 1890 K for the fuel-lean and fuel-rich flames, respectively (Table 1). The initial, seeded KCl concentration was based on the total amount of K measured at HAP 15 mm. It was assumed that the conditions at this HAP were close to equilibrium, and that the KCl salt will mainly be converted to gaseous K, KOH and KCl [14].

2.9 2D axisymmetric computational fluid dynamics simulations including K species

The introduction of the platinum plate 2 mm downstream of the burner surface induces non-uniformities in the flow. While conventional 1D simulations might be applicable at higher HAP, 2D models are necessary to describe the situation in the vicinity of the plate, where the release of K compounds to the stream takes place.

In order to describe the reactions, convection and diffusion of species in the plume above the platinum plate in sufficient detail, a reacting flow was simulated in a 2D axisymmetric geometry by solving the complete system of Navier-Stokes equations, including transport processes and Arrhenius chemical kinetics. A transient, pressure-based finite-volume solver based on the standard OpenFOAM solver reactingFoam [30] was employed to obtain steady reacting flow solutions. Calculation of the transport properties of the gas mixture, such as viscosity, thermal conductivity and mixture-averaged mass diffusivities were based on the mixture-averaged approach [31]. Spatial discretization was second order accurate for the advection and diffusion terms. Time integration was realized using a first order implicit Euler scheme. The calculation of the chemical source terms was based on OpenFOAM’s native implementation of operator-splitting method and a stiff ODE solver. A 30-species skeletal mechanism for methane oxidation based on GRI 3.0 [32] coupled with gas-phase K/H/O and K/H/O/Cl subsystems [33] was employed.

A sketch of the numerical setup is shown in Fig. 3(a). The gas stream at the flame portion of the bottom inlet boundary contained the products and intermediates of CH₄/air combustion at HAB 2 mm. The composition, temperature and velocity of this stream was obtained through separate 1D Chemkin PREMIX [34] calculations of a burner-stabilized premixed CH₄/air flame. The detailed chemical kinetics, thermodynamic and transport properties were computed with CHEMKIN [35] and TRANSPORT [36] libraries. Gaseous KCl was introduced to the stream at the plate portion of the bottom boundary, where the flow velocity was assumed to be zero. The supplied KCl concentration was set such that the total amount of K in the stream at HAP 15 mm agreed with the experimentally measured total concentration at this HAP. The radius of the plate was 3 mm and its temperature was set according to Table 1. The computational mesh was rectangular, with the interfaces between the cells being parallel to the radial and axial directions. The size of the calculation domain was 240 by 240 cells, corresponding to physical dimensions of 19 mm (radial) by 30 mm (axial).

 

Fig. 3. (a) Sketch of the axisymmetric 2D simulation setup. HAB – Height above burner surface. (b) Axial component of the flow velocity for the fuel-rich flame (ϕ = 1.3).

Download Full Size | PDF

The axial component of the flow velocity in the fuel-rich flame is shown in Fig. 3(b), illustrating a region of low velocity formed in the vicinity of the platinum plate. The flow around the plate determines the convection and diffusion of KCl evaporated from the plate, therefore, the presence of the low-velocity region is expected to have a limiting effect on the overall release of K compounds to the stream of the burnt gas. Flow recirculation at the edge of the platinum plate enhances the entrainment of methane combustion products and intermediates to the vicinity of plate surface, which, in turn, affects the rate of formation and release of gas-phase KOH in the stream.

3. Results and discussion

3.1 Validation of quantitative KCl detection

In order to accurately measure KCl concentrations, first the width of the K(g) fragment line shape needs to be determined. This was achieved by tuning the position of the PFT-signal, i.e. the UV pulse, across the K(g) absorption line shape. Figure 4(a) shows the K(g) absorption cross section obtained from the maximum PF absorbance (markers) as a function of detuning from the line center at a furnace temperature of 848 K. A curve fit using a Voigt-profile (solid line), with temperature-inferred, fixed Gaussian width (0.043 cm^-1) and the Lorentzian width as open fitting parameter, reveals a FWHM of 0.30 cm^-1. The average width for the entire furnace set temperature range was 0.28 cm^-1 (no clear temperature dependence observed), which is comparable to the value of 15.7 pm (0.267 cm^-1) measured by Sorvajärvi et al. [37].

 

Fig. 4. (a) K(g) absorption cross section based on PFT signal scan (FWHM = 0.30 cm^-1) as a function of detuning from resonance, fitted with a Voigt profile. (b) Measured KCl concentrations (markers) compared with equilibrium concentrations (line).

Download Full Size | PDF

A comparison between the path-averaged measured and TEC-predicted KCl concentrations as a function of path-averaged temperature between 676 K and 1023 K in the tube furnace is shown in Fig. 4(b). A good agreement was achieved in a wide dynamic range, except at the lowest temperature point (676 K), where the expected KCl concentration, 1.3 parts per billion (ppb), was close to the detection limit (0.8 ppb). The predicted background K(g) concentrations in the furnace were in the low parts per trillion (ppt) range, which was below the detection limit for K(g) in this work.

3.2 PF-absorbance signal-to-noise ratio validation

To validate the expression for the PF-absorbance SNR, Eq. (8), and to find the optimum UV-pulse position in the K(g) line shape, PF-TDLAS measurements were performed in three fuel-lean CH₄/air flames (Table 2), i.e. at K(g) concentrations low enough to fully resolve the PF-absorbance over the entire line shape.

Table 2. K species concentrations measured in the flames used to validate Eq. (8) for the PF-absorbance SNR.

View Table | View all tables in this article

Since temperature and gas composition in the flames are different from the furnace conditions, it was necessary to again determine the width of the K(g) fragment line shape. The line shape was obtained by recording the PF-absorbance during a scan of the 266 nm UV laser pulse position across the K(g) line shape, and subtracting the contribution from KOH to obtain the KCl absorbance. Figure 5(a) shows the line shape of the K(g) fragments (markers) together with a Voigt-profile curve fit (solid line) for the ϕ = 0.7 flame at HAP 15 mm. The corresponding TDLAS-measured line shape of background K(g) (dashed line) is shown for comparison. In the experiment, the scans were conducted with both UV lasers simultaneously, both resulting in identical line shapes of the K(g) fragments. From Fig. 5(a), the FWHMs of the absorption cross sections of background and fragment K(g) were determined to be 0.19 cm⁻¹ and 0.26 cm^-1, respectively, where the fixed Gaussian width (0.064 cm^-1) was inferred from the temperature and the Lorentzian width was an open fitting parameter. The larger fragment width is probably related to the higher velocity of the fragments compared to the background K atoms, leading to an increased collision rate and line broadening. The obtained K(g) fragment width was assumed to be valid also under fuel-rich conditions.

 

Fig. 5. (a) Background (FWHM 0.19 cm^-1) and fragment (FWHM 0.26 cm^-1) K(g) absorption cross sections as a function of UV pulse position measured at ϕ = 0.7 and HAP 15 mm. (b) Experimental (markers) and simulated (lines) PF-absorbance SNR from KCl absorption as a function of detuning from K(g) line center. For clarity, the ϕ = 0.7 and ϕ = 0.75 curves are vertically shifted by +100 and +600, respectively.

Download Full Size | PDF

Figure 5(b) presents the measured PF-absorbance SNRs for KCl (markers) in the three different flames, together with the corresponding PF-absorbance SNRs (solid lines) calculated using Eq. (8) and the experimentally determined K species concentrations given in Table 2. For clarity, the curves for ϕ = 0.7 and ϕ = 0.75 are vertically shifted by +100 and +600, respectively. The behavior of the experimental PF-absorbance SNRs is in good agreement with the predictions. For low K(g) concentrations, the optimum PFT-signal position is at the line center (zero detuning), since the PF-absorbance noise does not change significantly across the K(g) absorption profile. However, as the K(g) concentration increases, the PF-absorbance SNR decreases at line center, and the optimum PFT-signal position shifts to the line shape wings.

It is important to note that although the scans at ϕ = 0.7 and ϕ = 0.75 shown in Fig. 5(b) were measured under optically thin conditions, they nevertheless indicate an optimal UV pulse position slightly detuned from line center. For a detuning of ±0.2 cm^-1, the SNR can increase by 50% (Fig. 5(b)). Under optically thick conditions, the optimum detuning is usually significantly larger (>1 cm^-1, as in Fig. 2), since the line shape is saturated around line center. The sensitivity of the SNR to changes in UV pulse position decreases with detuning from line center, due to the decreasing slope of absorption cross section.

In practice, it is difficult to in advance know the (range of) K(g) concentrations that will prevail in the experiment and to adjust the UV pulse position accordingly. In the flame experiments presented in this work, the detuning was first chosen based on an estimate obtained using TEC. In the beginning of an experimental run, the PFT-signal amplitude was then maximized by slight adjustments of the detuning. Since, in PF-TDLAS, the actual K(g) line shape, the PFT-signals and UV pulse positions are recorded simultaneously in real-time, the necessary data for a correct evaluation is always available.

3.3 K species time series during KCl sample conversion

In order to be able to conduct reliable quantitative measurements as a function of HAP, time series of K species release were first recorded at a fixed HAP to investigate the temporal release behavior of KCl samples converted in the flat flames. Figure 6 shows typical time series of the three K species concentrations (0.5 s time resolution) measured simultaneously at HAP 15 mm in the fuel-lean (Fig. 6(a)) and fuel-rich (Fig. 6(b)) flames defined in Table 1. The overall conversion time was longer in the fuel-rich case. KOH and KCl show similar concentrations in both flames and are the dominating species under fuel-lean conditions, whereas K(g) dominates in the fuel-rich environment.

 

Fig. 6. Time series of K(g), KOH and KCl concentrations at 15 mm HAP. (a) Fuel-lean condition, ϕ = 0.8. KOH and KCl are the most abundant K species; (b) Fuel-rich condition, ϕ = 1.3. K(g) is the major K species in the gas. The concentrations are relatively stable during the 250-350 seconds plateau regions during which the HAP-scans are performed.

Download Full Size | PDF

During most of the conversion process, the KCl samples vaporizes at a rather constant rate, resulting in plateau regions (shaded areas) of length 250-350 s, during which HAP scans were conducted by vertically shifting the burner. The variations in concentration during conversion may be caused by inhomogeneous sample distribution on the platinum plate and/or the plate orientation in the flame. If the plate is not exactly parallel to the burner surface, the sample vaporization rate may vary spatially, due to the steep temperature gradient at HAP 1-3 mm [27]. Contrary to previous experiments with conversion of KOH salt [22], it was not possible to close the mass balance for the gaseous K species using the time series measured in this work. A possible reason is instantaneous sample loss at ignition, which could also explain the sharp concentration spikes in the beginning of the conversion (Fig. 6(a)).

3.4 2D simulation results

Figure 7 shows the concentration fields of K(g), KOH and KCl obtained in the 2D axisymmetric simulations for the fuel-lean flame (Figs. 7(a)-7(c)) and the fuel-rich flame (Figs. 7(d)-7(f)). As expected, the KCl concentration peaks at the plate due to KCl release from the sample, and K(g) is immediately formed. Both KCl and K(g) concentrations decay quickly with increasing HAP. The KOH concentration field is lifted above the plate, especially in the fuel-rich case, and relatively stable throughout the domain up to HAP 25 mm. As was observed in the time series and previous work [14,22], the KOH and KCl concentrations are similar in the two flames, whereas K(g) is much higher at ϕ = 1.3. In order to compare the 2D simulations with the experimental data, simulated column densities of K(g), KCl and KOH were obtained by multiplying the radial cell size with the concentrations of the individual species in each cell at a given HAP, and by summing up these products along the radial direction of the computational domain. The column densities were then divided by the experimentally observed absorption path length (21 mm) to obtain the species concentrations. The simulated concentrations were also averaged vertically over 2.3 mm to account for the finite laser beam diameter.

 

Fig. 7. 2D axisymmetric simulations of the K(g), KOH and KCl concentration fields for the fuel-lean (a-c) and the fuel-rich (d-f) flames.

Download Full Size | PDF

3.5 Axial K species profiles

The measured concentrations of K(g), KOH and KCl as a function of HAP (markers) are presented in Fig. 8, together with the beam diameter averaged line-of-sight concentrations extracted from the 2D axisymmetric CFD simulations (red solid lines) and the final concentrations predicted by TEC (blue, dash-dotted lines). The fuel-lean case is shown in Figs. 8(a)-8(c) and the fuel-rich case in Figs. 8(d)-8(f). The color and shape of the markers denote two different experiments (2 replicates).

 

Fig. 8. Measured concentrations of K(g), KOH and KCl (markers, two replicates) as a function of HAP in the ϕ = 0.8 flame (a-c) and the ϕ = 1.3 flame (d-f), together with the corresponding K species concentrations extracted from the 2D axisymmetric CFD simulations (red lines) and the predicted equilibrium concentrations (blue dashed lines).

Download Full Size | PDF

The detection limits for KOH/KCl (dashed lines) given an optical path length of 21 mm in the plume were 9/16 ppb under typical combustion conditions (low atomic K) and 310/630 ppb under gasification conditions (high atomic K). This shows an improvement of almost a factor of 10 compared to the previous implementation of PF-TDLAS [22], and thus constitutes the most sensitive in situ measurement of gaseous KOH and KCl to date. The detection limit for K(g) was around 3 ppb, limited by residual etalon effects originating from the dichroic mirrors.

The uncertainty in the TEC results due to the uncertainty in temperature is 5-10%, approximately the thickness of the dash-dotted line. The K(g) measurement uncertainty (1σ) is estimated to 5% based on the uncertainties in the absorption path length and the temperature-dependent line strength. The uncertainties in KOH and KCl concentration are estimated to 13% in both fuel-rich and fuel-lean conditions, calculated from the uncertainties in UV pulse energies, gas temperature, UV beam radii, PFT-signal noise and path length, where the last three parameters contributed the most.

Except for the region close to the platinum plate (HAP 0-5 mm), the measurements show a good qualitative agreement with the simulations. For KCl and K(g), the concentrations are high close to the plate, followed by a fast decay converging to the TEC-predicted levels. The measured KOH concentrations peak 2-3 mm above the plate and are then stable up to HAP 25 mm. Quantitative agreement is reasonably good for KCl and K(g), both with CFD and TEC. However, it is clear that the distribution between the K species is not correctly predicted. In particular, under fuel-rich conditions, KOH is over-predicted at the expense of K(g). Similar discrepancies between measured and predicted K species concentrations for KCl seeded flames have previously been reported by Weng et al. and could be explained by an incomplete reaction mechanism [38] and radical removal reactions caused by K species [39].

The discrepancies in shape and absolute concentration between measurements and simulations at lower HAP, which are more pronounced in the fuel-lean case, can have several reasons: The fact that the measured KCl concentrations are lower could indicate that the sample does not vaporize only as KCl, which is assumed for the plate portion of the bottom boundary in the CFD simulations, but as a mixture of K species. Also, the high concentrations of KCl and K(g) at low HAP are not only caused by the release of KCl and primary reactions forming K(g), but are also partly caused by an accumulation of K species just above the plate due to the low velocity region and flow re-circulation around the plate (Fig. 3(b)). The simulated flow profile may thus not exactly correspond to the actual flow conditions in the experiment. In addition, at HAP 0.5 mm and HAP 1 mm, there were strongly inhomogeneous conditions for K(g) along the line-of-sight, which lead to errors in the TDLAS spectra evaluation. At these HAPs, the absorption profiles could not be accurately curve-fitted using a Voigt profile.

3.6 Release of K species from biomass particles

The sensitivity of the PF-TDLAS system, and its ability to investigate K release from biomass in well-controlled laboratory settings with short absorption path lengths, is demonstrated by quantitative detection of the three K species close to biomass particles converted at high heating rates in the two flat flames (Table 1). A woody biomass, spruce softwood, and an agricultural waste product, wheat straw, with significantly different ash compositions and total K contents (Table 3) were used. The biomass particles weighted around 30 mg and were placed on the platinum plate at HAB 2 mm for conversion. The position of the laser beams was at HAP 5 mm. Both the devolatilization stage and the char burning phase were explored.

Table 3. Experimentally measured K mass in the gas phase and K, Cl and Si content in the fuels [8].

View Table | View all tables in this article

Figures 9 and 10 present time series of the concentrations of K(g), KOH and KCl above softwood (panels a and b) and wheat straw (panels c and d) in the fuel-lean and the fuel-rich flame, respectively. The upper panels (a) and (c) show solely the devolatilization phase (∼20 s), while panels (b) and (d) show the devolatilization and char burning phases (∼250 s). In general, the behavior is similar to what was observed for the conversion of solid KCl (Figs. 6–8), but the concentrations are much lower due to the small mass fraction of K in the biomass particles compared to in the pure KCl salt. For the fuel-lean case (Fig. 9), the behavior is also similar to predictions by a numerical particle model [8].

 

Fig. 9. Time series of K species released during conversion of softwood (a, b) and wheat straw (c, d) in a fuel-lean flame (ϕ = 0.8) measured at HAP 5 mm. Panels (a) and (c) show the devolatilization phase and panels (b) and (d) depict the devolatilization and char burning phases.

Download Full Size | PDF

 

Fig. 10. Time series of K species released during conversion of softwood (a, b) and wheat straw (c, d) in a fuel-rich flame (ϕ = 1.3) measured at HAP 5 mm. Panels (a) and (c) show the devolatilization phase and panels (b) and (d) depict the devolatilization and char burning phases.

Download Full Size | PDF

The devolatilization phase is shorter and shows steeper gradients in the fuel-lean than in the fuel-rich flame. In the former, all three K species are released, whereas mainly K(g) is present in the fuel-rich case, as also observed in the time series (Fig. 6) and the HAP scans (Fig. 8). Furthermore, a plateau-like release is found for softwood in combustion, while wheat straw shows a ramp shape, with higher concentrations towards the end of devolatilization. Most likely, the volatile Cl is quickly released in the beginning of the devolatilization, favoring KCl formation, similar to what was found in the studies by Sorvajärvi et al. and Qu et al. [8,18]. During char conversion, KOH dominates in the fuel-lean flame, while K(g) is the main species released in the fuel-rich flame.

Based on the time series in Figs. 9 and 10, the total mass of released K substance was calculated using the procedure introduced in [22], assuming a plume size of 21 mm and a gas flow velocity of 1 m/s. Table 3 shows a comparison between the total released K mass measured experimentally and the total K mass initially contained in the biomass particles, determined based on wet chemical fuel analysis [8]. The comparison shows that 30-50% of the initial K was released for softwood, but only 0.5-3% was observed in gas-phase above the wheat straw particles. Here, it has to be taken into account that a part of the conversion process (ash cooking stage) was omitted in the measurement. The total K release is lower in the fuel-rich flame than in the fuel-lean flame, which is more pronounced for wheat straw.

As can be seen in Table 3, there is a significant difference in K content between the two fuels, with wheat straw containing much higher K mass fractions. Thus, a significantly higher total K release could be expected from wheat straw, but this was not the case. Since also Cl is higher, more KCl should be present above wheat straw in the devolatilization stage, which was not the case either (Figs. 9 and 10). Both these effects could be connected to the high Si content in the agricultural biomass and the low plate temperatures (Table 1), which may promote the formation of silicates that bind K in the ash and hinder its release to the gas phase. Therefore, the Cl may be released in a form other than KCl, such as hydrogen chloride or in its atomic form.

4. Conclusions

Photofragmentation tunable diode laser absorption spectroscopy was employed for simultaneous detection of gaseous K(g), KOH and KCl in fuel-lean and fuel-rich methane-air flat flames with sub-second time resolution and unprecedented sensitivity. Mathematical expressions for the PF-TDLAS signal and the optimum detuning of the UV pulse from the K(g) absorption line center were presented and experimentally verified. The axial concentration profiles of the three K species measured during conversion of KCl salt in the flames agreed well with 2D axisymmetric reacting flow simulations and thermochemical equilibrium calculations. In fuel-lean flames, the K species concentrations deviated from the predicted behavior close to sample plate, which could indicate that the KCl salt is not only released as KCl, but also as K(g) and Cl. In fuel-rich flames further away from the sample plate, KOH was over-predicted at the expense of K(g), which points to an incomplete K reaction mechanism. A similar release behavior, with KOH dominating in fuel-lean and K(g) dominating in fuel-rich flames, was found when converting softwood and wheat straw particles. However, due to the presence of other ash-forming elements, only a small fraction of the total K in the biomass was released to the gas-phase, at least within the time frame of the measurement.