Sampling and Online Analysis of Alkalis in Thermal Process Gases

Aug 18, 2011 - When using biomass or waste as a feedstock for power production, alkalis are a major concern because of their negative effects on equip...
0 downloads 4 Views 5MB Size
ARTICLE pubs.acs.org/EF

Sampling and Online Analysis of Alkalis in Thermal Process Gases with a Novel Surface Ionization Detector Marco Wellinger,*,†,‡ Serge Biollaz,† J€org Wochele,† and Christian Ludwig*,†,‡ † ‡

General Energy Research Department, Paul Scherrer Institute (PSI), CH-5232 Villigen PSI, Switzerland Ecole Polytechnique Federale de Lausanne (EPFL), ENAC-IIE, CH-1015 Lausanne, Switzerland ABSTRACT: When using biomass or waste as a feedstock for power production, alkalis are a major concern because of their negative effects on equipment. To investigate the release behavior of alkalis and the effectiveness of countermeasures, it is indispensable to quantify alkali emissions with a high time resolution (less than 1 min). This paper presents a newly developed alkali detector based on the principle of surface ionization. The detector includes a number of improvements compared to previous designs that enable a reproducible measurement of alkalis in heavily tar- and particle-laden product gases. Our redesigned alkali detector has demonstrated high sensitivity and improved characteristics for making measurements of tar- and particle-laden process gases. Using an ultrasonic nebulizer as a source for alkali aerosols, we could conduct a calibration of the alkali detector over 4 orders of magnitude. In combination with a dilution setup and a sampling lance, we could take online measurements of gasifier producer gas with a high degree of reproducibility and a high time resolution of 1 s. The measurements of product gas have proven the feasibility of using the detector for field measurements.

’ INTRODUCTION Role of Biomass and Waste as an Energy Feedstock. The sustainable potential of biomass as a feedstock for heat, power, and fuel generation is limited for economic, ecological, and technical reasons.1 An ecologically efficient and economically acceptable bioenergy supply is therefore indispensable.2,3 In this context, waste materials are most interesting as low-cost feedstock and thermochemical processes promise high conversion efficiencies.4 However, wastes contain various heteroatoms, which can be ecotoxic or harmful for the facilities. Alkalis are one of the most critical gaseous emissions generated from the use of biomass and wastes as a feedstock. Hightemperature agglomeration, caking, fouling, and corrosion can shorten the service life of boilers and entire facilities or lead to severe damage of high-cost downstream components,510 such as catalysts, fuel cells, or turbines. For example, alkalis have been reported to cause performance losses and cell degradation in solid oxide fuel cells because of corrosion of the anode and poisoning of the nickel catalyst.11,12 Besides toxic heavy metals, which are of mainly ecological relevance, alkali metals occur at elevated trace concentrations in biomass and waste materials, especially in herbaceous biomass, such as straw or grass. The distribution of alkalis between different compounds and phases shows a strong variation between different feestocks.1315 Sulfur and chlorine species can further interact with alkalis. The release and volatilization of alkalis have been linked to fuel parameters, such as the contents of chlorine, sulfur, and clay minerals, and process parameters, such as temperature, pressure, and air/fuel ratio.1622 The thermochemical behavior of the alkalis during gasification and combustion decides their fate in the facility and their potential to damage downstream processes. Therefore, it is most important to monitor and finally control the concentrations of alkalis in the r 2011 American Chemical Society

process gases. Today, no flexible and sophisticated process controls for alkalis are available to guarantee a stable operation. When alkali signals are analyzed in real time, emergency measures can be taken (e.g., shutdown) to prevent damage to the equipment if critical limits are exceeded. It should be noted that the maximum allowed alkali concentrations specified for the use of product gas from biomass conversion can vary over several orders of magnitude even for a single type of application. For example, with combined cycles where the product gas is fed to a gas turbine, the specified alkali limits vary from values as low as 6 parts per billion by weight (ppbw)23 to values up to 1001000 ppbw.24 An often quoted alkali limit for feeding of synthetic natural gas to gas turbines is 24 ppbw.25 To capture the dynamics of thermal conversion processes that pass off in a matter of seconds or minutes, a high time resolution of the analytical instrument is necessary. The online measurement of the process gases would allow for real-time control of the process. Quantification of Alkalis in Process Gases. Among the conventional alkali measurement techniques are gas-washing methods, where the sampled gas is led through a series of impingers or scrubbers to capture the alkali species in a liquid solvent.26 The solution is subsequently analyzed by a standard analytical method, such as atomic emission spectrometry or liquid chromatography. These techniques are only applicable as a batch method and lead to average values over a period of more than an hour.22 Moreover, there have been reports of a high transmission especially of small alkali particles27 through such systems, which raises doubt about the validity of such methods in general. However, such methods are clearly not suitable to accurately represent any variation of the process in the order of seconds to minutes. Received: June 1, 2011 Revised: July 26, 2011 Published: August 18, 2011 4163

dx.doi.org/10.1021/ef200811q | Energy Fuels 2011, 25, 4163–4171

Energy & Fuels There are several techniques that allow for a time-resolved online quantification of alkalis. One of them is the excimer laserinduced fluorescence (ELIF). The detection principle bases on the fragmentation and excitation of alkali compounds by a laser beam and subsequent in situ detection of alkali compounds in process gases. Dependent upon the employed power of the laser, it is possible to discriminate between the gas- and condensedphase alkali species. At lower laser intensities, only gaseous alkali compounds are detected, whereas higher laser intensities also permit the detection of condensed alkali compounds. The ELIF method has a lower detection limit of 0.1 ppb and allows for a sampling interval of 10 s.17 Another online detection method is plasma-excited alkali resonance line spectroscopy (PEARLS), where the gas-phase as well as particle-bound alkalis are atomized and, afterward, either the excited photons or absorbing properties are used for quantification.28 A third online detection method for alkalis is molecular beam mass spectrometry (MBMS). In MBMS, the detection of the gas-phase as well as particle-bound alkalis can be measured. The sample gas is led into a vacuum chamber where a molecular beam is formed that is analyzed by mass spectrometry.2931 For the PEARLS as well as the MBMS method, the equipment reach a size that makes transport of the sampling system to a field site difficult. Monkhouse published a review of the methods discussed above.22,32 A technique that also allows for a time- and additionally size-resolved measurement of alkalis is the rotating drum impactor.3335 The very low detection limit of this method (1015 Ω cm at 20 °C and 1013 Ω cm at 200 °C) can be kept outside of the sample gas stream and, therefore, prevent them from contamination, which can lead to signal artifacts. This design change makes it possible to conduct long-term measurements (several hours to days) in heavily tar- and particle-laden gases. Inside the ceramic tube, there is an alumina ring in which a platinum filament (0.3 mm diameter, 99.997% purity, Johnson Matthey and Brandenberger AG) is spanned in parallel lines with an exposed total length of 464 mm (see Figure 2B). The filament geometry produced this way is highly reproducible. Having a second alumina ring with a premounted and preconditioned platinum filament available allows for swapping filaments within less than 20 min. The collector is made from a perforated metal plate, which is mounted on a metal rod that connects to the electrical feedthrough. To change the distance between the filament and collector, the connecting rod was built to variable lengths. Additionally, an auxiliary heater was integrated into the detector. It is located on the outside of the ceramic tube and helps minimize tar deposition inside the detector. The detector can be operated at excess pressure of up to 6 bar. Because platinum changes its electrical resistivity with the temperature, the temperature control of the filament is implemented by keeping the platinum filament at a constant total resistance. The temperature

ARTICLE

calibration was conducted with an optical pyrometer. To correct for a change in emissivity within the calibration range, a correction equation

Figure 2. Alkali signals from the nebulization of aqueous solutions of (A) sodium and potassium in air and (B) potassium in nitrogen. (1) Stop of aerosol feed for 1 min. Concentration of the solutions in chronological order: (A) 0.7, 7, and 70 μg of K/mN3 and 70 and 700 μg of Na/mN3 and (B) 0.7 (run time of 57 min), 7, 70, and 700 μg of K/mN3.

Scheme 1. (A) Longitudinal Section of the Detector Including the Measurement and Control Setup and (B) Alumina Filament Ring with the Mounted Platinum Filament

4165

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels for variable emissivity was calculated on the basis of the correction curves (given for fixed emissivity values) supplied by the manufacturer. Additionally, the temperature calibration was verified by a comparison to a calculated temperature change based on the change in resistivity of platinum with the temperature.38 The neutralizing current that is induced by ions that migrate from the filament to the collector is measured with a Keithley 487 picoammeter with an integrated voltage source. To improve the yield of ions, a bias voltage of +500 V is applied to the filament relative to the collector. The bias voltage causes a Schottky barrier lowering effect, resulting in a reduction of the energy needed to create an alkali ion on the platinum surface. Using a voltage of 500 V and a minimum distance between the filament and collector of 40 mm, the Schottky barrier lowering effect is 0.004 eV. Given that tabulated work functions for positive ion emission from platinum show differences of several tenths of electron volts and have error margins in the range of one tenth of an electronvolt,45 the influence on the overall ionization probability can be neglected. Sampling Setup. Water, tars, and particles present in product gases from gasification tend to condense, agglomerate, and finally block sampling lines.46 Therefore, special attention has to be paid to ensure a representative and reliable sampling of alkali-containing gas streams. In our case, two strategies are applied to increase transport efficiency and overall representativeness of the sample. The first approach consists of heated sampling lines to prevent condensation. Through heating, the condensation of water can be avoided easily, because in practice, a temperature of 120 °C ensures that all water is kept in the gaseous form. The tars present in gasification product gas would require heating to more than 250 °C because the polycyclic hydrocarbons, such as naphthalene or phenanthrene, have boiling points of 218 and 340 °C, respectively. This poses considerable problems in the use of material for the sampling lines as well as the power density of the heating equipment. The sampling setup developed to take the measurements presented in this paper is depicted in Scheme 2. Its main features are an online dilution system and a sampling lance that is adapted per sampling port and reactor type. The dilution setup consists of three mass flow controllers (MFCs) that are either outside the sample gas stream (MFC 1 and MFC 2) or shielded by filters from particles and tars (MFC 3). All MFCs were bought from V€ogtlin and have a guaranteed accuracy of (0.5%. When a backpressure regulator (Porter, model 9000 regulator) is inserted downstream of MFC 1 and MFC 2, their operation can be decoupled from varying backpressure in the case of a change in the flow regime. First, the MFCs on the input side (MFC 1 and MFC 2) are adjusted to match the total output through MFC 3. The measurements shown in this paper have all been conducted at a total flow through the detector of 105 mN3/s. The absolute error between input (MFC 1 and MFC 2) and output (MFC 3) can be measured experimentally by quantifying the gas flow at the sample gas inlet. Next, the sample gas intake flow is determined by adjusting MFC 2, because it has a lower measurement range (1.67  106 mN3/s) than the other two MFCs (3.33  105 mN3/s) and, therefore, a smaller margin of error. Therefore, the precision in the sample gas flow is only dependent upon the long-term stability of all three MFCs plus the precision of the small MFC. During the measurements made at a BFB reactor, a lance was used to sample the product gas (see Scheme 2).

Quantification Experiments with Aerosols Generated from an Ultrasonic Nebulizer (USN). To deliver a uniform flow of liquid to the USN (Cetac, U5000AT+), a peristaltic pump (Gilson, Minipuls 3) was used at a frequency of 5 rpm to deliver a sample flow rate of 0.43 mL/min. The efficiency of nebulizers can be expressed as a ratio of the mass flow output/mass flow input of the analyte. In contrast to conventional pneumatic nebulizers that show efficiencies around 24%, an USN can reach an efficiency of more than 30%,47 although the manufacturer of the instrument used in this study states an efficiency of 1015%. In our case, the nebulizing efficiency was determined 2-fold by

ARTICLE

Scheme 2. (A) Flow Sheet of the Sampling Setup Used with the Alkali Detector and (B) Schematic Cross-section of the Sampling Lance

measurements taken with an inductively coupled plasmaoptical emission spectrometer (ICPOES). First, the volume and concentration of the analyte in both the feed solution and the drain solution were measured, and subsequently, the actual output concentration was calculated. Second, the signal intensities of nebulized aqueous solutions were compared to those using gaseous standards that were fed directly to the ICP-Torch. Both methods showed a nebulizing efficiency of 10% for the used setup. For the experiments, solutions of sodium nitrate and potassium nitrate have been produced. For both compounds, concentrations of 1, 10, 100, and 1000 ng/mL have been used. A MFC was used to supply a constant flow of 105 mN3/s of either nitrogen or air to the USN. The nebulizer has an integrated evaporation and condensation loop to reduce the water load in the sample gas. The evaporation loop was set to 120 °C, and the condensation loop was set to 2 °C. The background signal as recorded before and after each measurement of a standard solution was below 1 nA. For a new filament, the background is below 0.1 nA, and after a cleaning interval of 2 h, the background signal after the calibration measurements dropped below 0.5 nA again. BFB Reactor. The BFB reactor has been described in detail elsewhere.15 The reactor was fed with commercially available wood pellets (Buerli Trocknungsanlage, Alberswil, Switzerland) at a rate of 0.58 kg/h, and air was used as a gasification agent at a gas flow rate of 2.85  104 mN3/s. Alumina (Al2O3) was used as material in both experiments. The air/fuel ratio was set to 0.44, and the reactor bed temperature was around 755 °C. Gas Chromatograph. In parallel to the measurements at the BFB reactor taken with the alkali detector, the permanent gas compositions 4166

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels

Figure 3. Calibration curves for the surface ionization detector. The black and gray lines represent the power correlation curves in nitrogen and air, respectively. (4) Sodium in air, (2) sodium in nitrogen, (O) potassium in air, and (b) potassium in nitrogen. In brackets, are the values that account for the whole peak area (see Figure 2).

ARTICLE

Figure 4. Calibration of the dilution setup. The error bars represent the 3-fold standard deviation of the measured gas flow converted to the gas sample fraction.

were analyzed using a gas chromatograph [Varian CP 4900; detector, thermal conductivity detector (TCD); columns, MSA5]. A quench and gas-washing system (further referred to as gas washer number 1) similar to the one described by Kowalski et al.27 was used to separate water, soot, and tars from the permanent gases before being fed to the gas chromatograph. A second type of gas washer (further referred to as gas washer number 2) was used in this study that can tolerate high particle loads but is also characterized by a high dead volume, leading to a longer lag time and a stronger tendency to even out fluctuations in gas concentrations. The ability to tolerate high particle loads stems mainly from different types of pumps used (peristaltic pumps instead of piston pumps) and the fact that the used tubes have a larger diameter. To calculate the total volume flow through the reactor, helium was used as an internal standard. To this end, a known volume flow (9  107 mN3/s) of helium is continuously injected into the gasification reactor along with the pressurized air. Through a comparison of the absolute volume flow of helium to the percentage that is measured at the gas chromatograph, the total volume flow coming from the reactor can be calculated.

correlation to have an intercept at about 0.7 nA. On the other hand, a linear correlation forced through zero led to an overestimation of the concentrations below 7 μg/mN3. The signals at 700 μg/mN3 showed much lower values than expected from the correlations. To test whether the signal tailing (see Figure 2) could close this gap between measured and expected values, the cumulated charge over the whole peak was divided by the effective feed time. The corresponding values are shown in brackets in Figure 3. Power correlations were used to establish a mathematical relationship, which more accurately represents the measured signals. The correlations including both sodium and potassium nitrate with the concentrations of 0.7, 7, and 70 μg/mN3 gave the following calibration equations:

’ RESULTS

where I denotes the background-corrected detector signal in nA (109 A) and c is the mass concentration of alkalis in μg/mN3. Verification of the Dilution Setup. Figure 4 shows the results of this experiment, where the sample gas fraction is plotted against the setting of MFC 2 (see Scheme 2) at a fixed setting of MFC 1 and MFC 3 (the total flow through the detector was kept constant at 105 mN3/s). The sample gas fraction was calculated by dividing the measured gas flow (rate of intake) at the sampling lance by the total gas flow measured at the outlet of the sampling lance. The calibration of the dilution setup showed a linear correlation with a R2 of 0.9997 and a maximum relative standard deviation of less than 1.5% for the tested dilutions between 1.6 and 10.9% sample gas fraction. Because the standard deviation itself was too small to be clearly visible on the graph, the y error bars represent the 3-fold standard deviation of the replicate measurements. Measurements of Gasifier Product Gases. In a series of measurements at a pilot-scale (∼5 kWth) BFB reactor, we could test the potential of the instrument for online quantification of real process gas. The gasifier was run with wood pellets as feedstock. Figure 5 shows the results from a calibration experiment using the sampling system for an online dilution of gasifier product gas. Figure 5 shows the signal of the detector with time resolutions of 1 s (Figure 5A) and 10 s (Figure 5B). In Figure 5C,

Analyte Sensitivity of the Detector. In a series of laboratory experiments, the SID II was fed with aerosols generated by an USN using aqueous solutions of sodium nitrate and potassium nitrate. Figure 2 shows the alkali signals with a time resolution of 1 s using air (Figure 2A) and nitrogen (Figure 2B) as the carrier gas. In Figure 2A, the signals from five different solutions are shown. Figure 2B shows the signals from four different solutions of potassium nitrate using nitrogen as the carrier gas. The solutions with the highest alkali concentration (700 μg/mN3) show considerable tailing, meaning that the alkali signal takes up to 40 min to return to the background level (stepdown response). In Figure 2B, the reaction of the detector to a shortterm stop of aerosol feed was tested. The feed of aerosol was stopped for 1 min (indicated with a “1” in Figure 2B), during which the detector signal drops from 200 to 132 nA. After the aerosol feed is turned back on, the signal returns to its previous level within 5 s (stepup response). Figure 3 shows the calibration curves for aqueous solutions of sodium nitrate and potassium nitrate. The relative standard deviations for all measurements were below 6%. The curves show a linear response from 0.7 to 70 μg/mN3. Below 0.7 μg/ mN3, the signal sensitivity decreases, which causes the linear

for measurement in nitrogen : I ¼ 1:741x0:952 ðR 2 ¼ 0:9995Þ

ð1Þ for measurement in air : I ¼ 1:407x0:956 ðR 2 ¼ 0:9985Þ

4167

ð2Þ

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels

Figure 5. Alkali signal with (A) 1 s and (B) 10 s time resolution from diluted gasifier product gas. (C) Calibration of the detector signal against the sample gas fraction. The equations in the chart show the free linear correlation and the linear correlation forced through zero. (b) Anchor point of the estimation and (O) estimated value.

the measured values are plotted against the estimated values according to the calibrations of the dilution setup and the analyte sensitivity of the detector. The expected alkali signals were calculated using the initially set dilution of 11 wt % sample gas as an anchor point. Applying the analyte sensitivity calibration of the detector to the initial alkali signal resulted in an alkali concentration of 5.6 μg/mN3. Using the alkali concentration measured by the detector and the initial dilution level, the alkali concentration of the sample gas was calculated. The alkali concentration of the sample gas was then used to convert the two other dilution levels (7 and 5%) to actual alkali concentrations, which in turn were entered into the analyte sensitivity of the detector to yield an estimated alkali signal (see Figure 5C). The values in the chart represent the mean of 55 values, each averaged over a 10 s interval (to a total of 9 min and 10 s). We also tested the dependence of the heating power (needed to keep the filament at a constant temperature) upon the sample gas fraction because Kowalski21 reported linear dependencies of the heating power from the methane content because of the higher heat capacities compared to nitrogen. The linear correlation according to the least-squares fit method gave the calibration equation P = 0.3455r + 66.702, with a R2 of 0.99958, where P is the heating power in Watts and r is the sample gas fraction in percent. The intercept of this equation corresponds to the required heating power in pure nitrogen. After the testing of the dilution setup, several measurements with gasifier product gas have been conducted. Figure 6 shows the results from a measurement conducted over a period of almost 5 h. During the experiment, the fuel was fed for 3.5 h, and afterward, the gasifier was run for an additional 1 h to burn out yet unconverted fuel in the reactor. Figure 6A shows an online measurement of the raw product gas diluted to a fraction of 11 wt % in nitrogen. The alkali concentration during the duration of the wood feed was 38 μg/mN3 on average. Data points were recorded in intervals of 1 s, although for reasons of clarity, only average

ARTICLE

values of 10 s intervals are shown in Figure 6. The alkali detector showed a strong change in the signal within less than 1 min to both the start of the wood feed at 10:07 (with an increase) and the feed stop at 13:48 (with a decrease). Although from experiments with the USN, the reaction time of the detector itself was determined to lie below 5 s. The observation of sharp peaks each coinciding with a mechanical shock of the reactor through vibration (incurred by the rotary feeder) at 11:03 and 11:19 has prompted us to test whether or not the two observations are linked in a causal manner. To this end, we induced shocks by knocking on the sampling lance at 11:24 and 11:52. Especially at 11:52, the detector showed a very high signal. The signal averaged over a 10 s interval corresponded to a concentration of 510 μg/mN3, whereas the full time resolution showed that, for a single second, the concentration even rose to more than 1500 μg/mN3 (although this value bears a considerable uncertainty; see Figure 3). The bed temperature of the reactor showed a flat profile (756 °C on average) with minor fluctuations (7 °C standard deviation). In contrast, the temperature of the freeboard showed a steady rise during the run from 340 °C (at 10:07) to 445 °C (at 13:48). The data of the gas chromatograph (Figure 6B) was gathered using a gas washer number 2. This leads to a longer lag time and slower increase of the gas concentrations than the response observed with the alkali detector (e.g., compare Figure 6A to Figure 6B just after the start of the experiment). The total gas flow was reduced from 2.36  102 to 1.81  102 mN3/s after the fuel feed was stopped because of a smaller product gas production. Therefore, to accurately compare the gas production before and after the feed stop, one has to compare the absolute volume flow of the different gases (see Figure 6C). After the fuel feed has stopped, the volume flows of hydrogen and methane start declining after a lag time of 8 min from the feed stop. During the first 20 min after the feed stop, the absolute volume flow of carbon monoxide is reduced approximately by half, whereas the produced carbon dioxide is reduced by roughly 30%. From 14:25 to 14:35, the carbon dioxide production rises again because there is barely any more residual fuel and, therefore, the relative amount of available oxygen rises. Figure 7 shows another run of the BFB reactor using wood as feedstock. The alkali detector was again fed with sample gas diluted to a fraction of 11 wt % in nitrogen (see Figure 7A). In contrast to the run shown in Figure 6, where fresh bed material was used, this time, the bed material was reused from a previous run. Another difference to the previous run is that the wood feedstock had a slightly lower water content, with 8.5% compared to 8.9% in the run shown in Figure 6. At 10:05, the flow of air was switched on but without feeding any wood. This resulted in a steep and short peak of alkali emissions, as measured by the alkali detector. At 10:06, the feeding of wood was started. With a lag time of less than 1 min, the detector recorded a high concentration of alkalis. The bed temperature of the reactor was not yet stabilized at the beginning of the run and showed an increase until 10:55. Thereafter, the temperature was stable with an average of 751 °C and a standard deviation of 5 °C. After the temperature stabilized, the temperature of the freeboard showed a constant rise from 331 °C (at 10:55) to 397 °C (at 12:08). Data on gas composition were recorded from 10:25 on, with the help of a gas washer number 1. At 10:45, the fuel feed was stopped for 3 min. The change in the process condition was reflected with a decrease in the alkali signal, with a lag time of less than 1 min. Similarly, the measured gas concentrations drop to 4168

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels

ARTICLE

Figure 6. (A) Alkali concentration in the raw gas of a wood fed gasifier: (1) start of the wood feed, (2) alkali peaks after vibration transmitted from the gasifier feed system, (3) self-induced alkali peaks with a knock to the sampling lance, and (4) stop of the wood feed. (B and C) Gas concentrations measured by a gas chromatograph expressed as (left) percentages and (right) absolute gas flows.

Figure 7. (A) Alkali concentration in the raw gas of a wood fed gasifier: (1) start of the air flow, (2) start of the wood feed, (3) short-term interruption of the wood feed, and (4) stop of the wood feed. (B) Gas concentrations as measured with a microgas chromatograph.

almost zero with a lag time of 9 min (see Figure 7B). At 12:08, the wood feed was stopped for good, while continuing the flow of pressurized air for another 1 h.

’ DISCUSSION In the experiments with the USN, we could demonstrate a strong correlation of the detector signal to the mass concentrations of alkalis fed. Although for 70 μg/mN3, there was a considerably lower signal from sodium compared to that of potassium. A possible reason for this difference is that the surface work function of platinum was actually lower under the used conditions than the suggested literature value of 5.61 eV.45 This would impact sodium by a much higher degree than potassium because the ionization potential of sodium is 5.14 eV, whereas that of potassium is 4.34 eV. Another possible explanation for the observed signal of potassium is its tendency to be in a solid rather than in a liquid or gaseous state. This is due to its higher melting point and probability to form a gaseous compound at the filament temperature of 1200 °C. Thermodynamic equilibrium calculations using HSC software (version 7.0, Outokumpu) with water and alkali nitrate indicate that potassium is only present in the gaseous form at 1200 °C, whereas the same simulation for sodium indicates that only 40% is in the gaseous form. In the range below 0.7 μg/mN3, the curves diverge because of a lower signal-to-noise ratio.42,44 The main reason for the increase in noise is electrostatic interference from the environment. At a concentration of 700 μg/mN3, the momentary signal intensity is considerably lower than predicted according to the sensitivity correlation. On the other hand, when the signal tailing was included, the measured signal intensity and that predicted by the sensitivity correlation were in agreement. This means that the alkali signal turns from an online measurement into an offline measurement above a signal intensity of 100 nA. Preliminary tests have shown that the signal intensity can be increased by a higher filament temperature, suggesting a lack of thermal energy 4169

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels to melt the high load of impinging particles at concentrations of 700 μg/mN3. A further explanation for the lower signal sensitivity could be a limitation in patchy faces on the platinum surface with sufficient high local surface work function.45 During the measurements at the BFB reactor, single high peaks appeared as documented in Figure 6A. At full time resolution (1 s intervals), we could observe that all of these high peaks have as a minimum three data points that lie 10-fold outside the otherwise observed fluctuations. This supports the notion that the observed signals are indeed detected alkalis, instead of just being electronic artifacts. The last very high peak consists of more than 30 single measurements. The conducted tests where the sampling lance was knocked on, suggest that mechanical shocks induced the detachment from alkali-containing particles inside the sampling lance. This was confirmed by char particle deposits, which were found inside the sampling lance after decommissioning, which further supports the notion that the observed peaks are caused by detachment of char particles from the sampling lance, leading to an unusually high alkali load being transported to the detector. Wet chemical analysis has confirmed that the char particles indeed show a high alkali concentration of more than 10 wt %. In both runs of the BFB reactor, the alkali signal followed a slow decrease over dozens of minutes after the fuel had been switched off. Analogously, the alkali signal does not drop to zero during the short feed stop in Figure 7 but rather stays at an intermediate level of alkali emission. In the run shown in Figure 6, this burnout phase lasted 50 min, whereas the same phase lasted only 30 min in the second run (Figure 7). This duration seemed roughly proportional to the duration of the wood fed to the gasifier, which was 4 h for the first run and 2 h for the second run. Furthermore, the comparison of alkali and gas concentration data revealed that the carbon monoxide percentage and alkali concentration showed a similar progression of curves. It is known from previous experiments that, at the λ values used in these experiments, a floating layer of char is being built-up over time on top of the fluidized bed.15 The initial surge of high values observed in both runs of the BFB reactor could be caused by a filtering effect of char that gradually builds up on the top of the fluidized bed. In those cases where the run of the gasifier was not finished with a burnout phase (where no fuel is supplied), a layer of char was found on the top of the bed material when the reactor was cleared out. As a direct consequence of this phenomenon, the trend of a slightly rising alkali concentration observed in both runs could well be linked to the increasing temperature of the freeboard. From the dilution experiments with the raw gas from the BFB reactor, an important implication can be deduced. In the tested range of 115%, the gas matrix of the gasifier product gas seems to have a negligible effect on the analyte sensitivity of the detector. This can be deduced because the measured values for 5 and 7% are in agreement with the predicted values using the correlation from the experiments with the USN and dilution setup. Kowalski reported for the SID I a relative increase in the heating power from pure nitrogen to pure product gas of 31%. In the measurements presented in this paper, the SID II showed an increase in heating power from pure nitrogen to pure product gas (with a similar composition to that used by Kowalski) of 55%. Considering that the SID I has a completely different flow geometry and markedly less filament (about 60% less) than the SID II, this increase was expected.

ARTICLE

’ CONCLUSION We presented the development and application of a novel alkali detector with improved characteristics. Namely, the filament geometry of the detector enables an increased surface area of platinum and thereby an increased sensitivity. At the same time, the filament geometry can be accurately reproduced when a filament change is necessary, which is crucial in keeping the sensitivity of the device constant. Furthermore, the parallel alignment of the gas flow with the flight of the ionized alkali species decreases the sensitivity of changes in total flow through the detector. The analyte sensitivity calibration conducted with an USN enables an accurate measurement of alkalis in process gases. Furthermore, we could demonstrate that the gas matrix evidently does not influence the signal sensitivity in the range between 5 and 11% sample gas fraction. Beside the measurements of alkali emissions from energy generation processes (gasification or combustion) using wood, with the help of the dilution setup, the device is also possible to measure alkali emissions from feedstock containing high amounts of alkalis, such as straw or common pasture grass. Furthermore, the alkali detector could be used in thermal processes in the waste and cement industry, as well as metallurgical processes. ’ AUTHOR INFORMATION Corresponding Author

*Telephone: +41563105862. Fax: +4156310. E-mail: marco.wellinger@ psi.ch. (M.W.). E-mail: [email protected] (C.L.).

’ ACKNOWLEDGMENT We thank Jan Hovind, Albert Schuler, Johannes Judex, and Emmanuele Japichino for their contribution in designing the SID II. Further, we thank J€org Schneebeli for the efforts in the development of the dilution system and sampling lance. Financial support was obtained from Swisselectric research (Project TREPGAS) and the Swiss Federal Office of Energy (Project 102093). Both of the aforementioned projects were imbedded in the framework of the Competence Center Energy and Mobility (CCEM) Project “Woodgas-SOFC”. ’ NOMENCLATURE ELIF = excimer laser-induced fluorescence PEARLS = plasma-excited alkali resonance line spectroscopy MBMS = molecular beam mass spectrometry SID = surface ionization detection SID I = previous version of the surface ionization detector SID II = newly developed surface ionization detector MFC = mass flow controller USN = ultrasonic nebulizer ICPOES = inductively coupled plasmaoptical emission spectrometer BFB = bubbling fluidized bed ’ REFERENCES (1) Steubing, B.; Zah, R.; Waeger, P.; Ludwig, C. Renewable Sustainable Energy Rev. 2010, 14, 2256–2265. (2) Steubing, B.; Zah, R.; Ludwig, C. Biomass Bioenergy 2011, 35 2950–2960. (3) Sipilae, K.; Johansson, A.; Saviharju, K. Bioresour. Technol. 1993, 43, 7–12. 4170

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171

Energy & Fuels (4) Fagernaes, L.; Johansson, A.; Wilen, C.; Sipilae, K.; Maekinen, T.; Helynen, S.; Daugherty, E.; den Uil, H.; Vehlow, J.; Kaberger, T.; Rogulska, M. Bioenergy in Europe: Opportunities and Barriers; VTT Technical Research Centre of Finland: Espoo, Finland, 2006; http:// www.vtt.fi/inf/pdf/tiedotteet/2006/T2352.pdf. (5) Penner, S. S.; Alpert, S. B.; Beer, J. M.; Bozzuto, C. R.; Glassman, I.; Knust, R. B.; Markert, W.; Oppenheim, A. K.; Smoot, L. D.; Sommerlad, R. E.; Wagoner, C. L.; Wender, I.; Wolowodiuk, W.; Yeager, K. E. Prog. Energy Combust. Sci. 1984, 10, 87–144. (6) Scandrett, L. A.; Clift, R. J. Inst. Energy 1984, 57, 391–397. (7) B€ohm, H. VGB Kraftwerkstech. 1993, 74, 173. (8) Osborn, G. A. Fuel 1992, 71, 131–142. (9) Hansen, L.; Michaelsen, H.; Dam-Johansen, K. Fluidised-Bed Combustion; American Society of Mechanical Engineers (ASME): New York, 1995. (10) Hald, P. Alkali metals at combustion and gasification equlibrium calculations and gas phase measurements. Ph.D. Thesis, Technical University of Denmark, Lyngby, Denmark, 1994. (11) Norheim, A.; Lindberg, D.; Hustad, J. E.; Backman, R. Energy Fuels 2009, 23, 920–925. (12) Nurk, G.; Holtappels, P.; Figi, R.; Wochele, J.; Wellinger, M.; Braun, A.; Graule, T. J. Power Sources 2011, 196, 3134–3140. (13) French, R. J.; Milne, T. A. Biomass Bioenergy 1994, 7, 315–325. (14) Khan, A. A.; de Jong, W.; Jansens, P. J.; Spliethoff, H. Fuel Process. Technol. 2009, 90, 21–50. (15) Judex, J. Grass for power generation. Extending the fuel flexibility for IGCC power plants. Ph.D. Thesis, ETH Z€urich, Z€urich, Switzerland, 2010; http://e-collection.library.ethz.ch/view/eth:1070. (16) Glazer, M. P.; Khan, N. A.; de Jong, W.; Spliethoff, H.; Schurmann, H.; Monkhouse, P. Energy Fuels 2005, 19, 1889–1897. (17) Gottwald, U.; Monkhouse, P.; Wulgaris, N.; Bonn, B. Fuel Process. Technol. 2002, 75, 215–226. (18) Laatikainen, J.; Nieminen, M.; Hippinen, I. Proceedings of the 12th Fluidized Bed Combustion Conference; San Diego, CA, May 913, 1993. (19) Lee, S. H. D.; Teats, F. G.; Swift, W. M.; Banerjee, D. D. Combust. Sci. Technol. 1992, 86, 327–336. (20) Korsgren, J.; Hald, P.; Davidsson, K.; Pettersson, J. Proceedings of the 15th Conference on FB Combustion; Savannah, GA; May 16, 1999. (21) Kowalski, T.; Ludwig, C.; Wokaun, A. Energy Fuels 2007, 21 3017–3022. (22) Monkhouse, P. Prog. Energy Combust. Sci. 2011, 37, 125–171. (23) General Electric Company (GE). Specifications for Fuel Gases for Combustion in Heavy-Duty Gas Turbines (GEI 41040j); GE: Fairfield, CT, 2007; http://www.gepower.com/prod_serv/serv/env_serv/en/ downloads/gei41040j.pdf. (24) Romey, I. F. W.; Garnish, J.; Bemtgen, J. Diagnostics of Alkali and Heavy Metal Release (Clean Technologies for Solid Fuels); European Commission: Brussels, Belgium, 1998. (25) Rubow, L.; Zaharchuk, R. Proceedings of the 2nd Annual Contractors Meetings on Contaminant Control in Hot Coal-Derived Gas Streams; Morgantown, WV, Feb 1719, 1982. (26) Mojtahedi, W.; Kurkela, E.; Nieminen, M. J. Inst. Energy 1990, 63, 95–100. (27) Kowalski, T.; Judex, J.; Schildhauer, T. J.; Ludwig, C. Chem. Eng. Technol. 2011, 34, 42–48. (28) H€ayrinen, V.; Hernberg, R.; Aho, M. Fuel 2004, 83 (7), 791–797. (29) Blaesing, M.; Melchior, T.; Mueller, M. Energy Fuels 2010, 24 4153–4160. (30) Dayton, D. C.; French, R. J.; Milne, T. A. Energy Fuels 1995, 9 855–865. (31) Dayton, D.; Jenkins, B.; Turn, S.; Bakker, R.; Williams, R.; BelleOudry, D.; Hill, L. Energy Fuels 1999, 13, 860–870. (32) Monkhouse, P. Prog. Energy Combust. Sci. 2002, 28, 331–381. (33) Cahill, T.; Goodart, C.; Nelson, J.; Aldred, R.; Nastrom, J.; Feeney, P. Design and evaluation of the drum impactor. Proceedings of International Symposium on Particulate and Multiphase Processes; Miami Beach, FL, April 1985.

ARTICLE

(34) Cliff, S. S.; Cahill, T. A.; Jimenez-Cruz, M.; Perry, K. D. Abstr. Pap. Am. Chem. Soc. 2003, 225–247. (35) Bukowiecki, N.; Hill, M.; Gehrig, R.; Zwicky, C. N.; Lienemann, P.; Hegedus, F.; Falkenberg, G.; Weingartner, E.; Baltensperger, U. Environ. Sci. Technol. 2005, 39, 5754–5762. (36) Langmuir, I.; Kingdon, K. H. Science 1923, 57, 58–60. (37) Zandberg, E.; Ionov, N. Sov. Phys. Usp. 1959, 67, 255–281. (38) Anderson, P.; Ross, R. G.; Backstrom, G. In Thermal Conductivity; Hust, J. G., Ed.; Plenum Press: New York, 1983; Vol. 17. (39) Holmlid, L.; Wall, S. Langmuir 1989, 5, 1170–1175. (40) J€aglid, U.; Olsson, J. G.; Pettersson, J. B. C. J. Aerosol Sci. 1996, 27, 967–977. (41) Olsson, J. G.; Jaglid, U.; Pettersson, J. B. C.; Hald, P. Energy Fuels 1997, 11, 779–784. (42) Davidsson, K.; Engvall, K.; Hagstrom, M.; Korsgren, J.; Lonn, B.; Pettersson, J. Energy Fuels 2002, 16, 1369–1377. (43) Svane, M.; Hagstrom, M.; Pettersson, J. B. C. Energy Fuels 2005, 19, 411–417. (44) Kowalski, T. Evaluating a surface ionisation detector for measuring alkalis in biomass gasification. Ph.D. Thesis, ETH Z€urich, Z€urich, Switzerland, 2007. (45) Kawano, H. Prog. Surf. Sci. 2008, 83, 1–165. (46) Hasler, P.; Nussbaumer, T. Biomass Bioenergy 2000, 18, 61–66. (47) Borkowska-Burnecka, J.; Lesniewicz, A.; Zymicki, W. Spectrochim. Acta, Part B 2006, 61, 579–587.

4171

dx.doi.org/10.1021/ef200811q |Energy Fuels 2011, 25, 4163–4171