In recent years, the negative effect of so-called greenhouse gases
like carbon dioxide (CO
), methane (CH
) and nitrous oxide (N
O) on global climate has become evident. The impact of different gases can be directly compared by means of their global warming potentials (GWP); a numerical value describing the propensity to trap heat expressed as a multiple of the potential of carbon dioxide [1
]. As depicted in Table 1
, the perfluorocarbons (PFCs) tetrafluoromethane (CF
) and hexafluoroethane (C
), have very high GWP values indicating significant contributions to the greenhouse effect despite low atmospheric concentrations.
The aluminum smelting process, also known as the Hall-Héroult
process, is one of the largest anthropogenic contributors to atmospheric PFCs [2
]. When the dissolved alumina content in the cryolitic bath falls below ~1.5%, a rapid rise in the operating voltage occurs and the carbon anodes begin to react with the fluoride-based electrolyte (Na
) forming PFCs. These transient events are referred to as anode effects (AEs) [3
]. Recent reports indicate that in addition to AEs, PFC generation may also occur during normal potroom activities that result in either (1) a localized increase in anode current density or (2) a localized deficiency in alumina concentration; PFCs emitted during these periods not associated with AEs have been referred to as low-voltage PFCs (LV-PFCs) [4
Fourier Transform Infrared (FTIR) spectrometers are commonly employed by the industry to measure PFCs in real time, providing an accurate snapshot of the emission profile during typical potroom operation. The technique relies on extractive sampling from the exhaust stream and subsequent conditioning of the sample gas prior to the measurement. The exhaust stream at aluminum smelters is rather hostile to FTIR spectrometers: the sample gases contain a high dust loading, chemically aggressive species (e.g., hydrogen fluoride, HF), and spectrally interfering compounds (e.g., water vapor, H
]. As such, deployment of an FTIR spectrometer for continuous monitoring of PFCs would require significant upkeep and daily maintenance. Instead, the industry relies on short-term in-plant campaigns to spectroscopically determine levels of PFCs emitted during a representative sampling of AEs spanning typical production cycles. Combining PFC emission values with production and AE metrics yields plant-specific slope terms used to estimate the long-term relationship between PFC emissions and smelter process parameters [8
In-plant measurements are also performed in order to ensure regulatory compliance, at a frequency prescribed by the ruling government (e.g., triennial for Quebecois smelters [9
]) or upon significant changes to the process control algorithm impacting AE metrics.
Tunable laser absorption spectroscopy (TLAS) is nowadays considered to be a mature technology for in-situ measurements allowing to perform year-round continuous emission measurements (CEMS) with a higher accuracy and faster response time than extractive methods while requiring significantly lower maintenance.
The aluminum industry is using TLAS sensors for many years in potrooms and exhaust stacks to perform in-situ
measurements of HF [10
]. This success has led to the aluminum industry’s interest in evaluating a similar technology for in-situ
offering the capability to continuously monitor PFC emissions.
2. In-Situ Tetrafluoromethane Sensor
According to the HITRAN (high-resolution transmission molecular absorption database) [11
] and PNNL (vapor phase infrared spectral library) [12
] spectroscopic databases, CF
does not absorb in the near-infrared but displays a strong absorption feature in the mid-infrared range located around 1283 cm
. A simulated transmission spectrum of 20 ppb CF
in ambient air using an optical path length (OPL) of 1 m and ambient pressure is shown in Figure 1
. The simulated spectra were assembled using data available from the HITRAN and HITEMP (high-temperature molecular spectroscopic database) [13
] databases. A gas temperature of 125
C was chosen since it is a typical value for emission monitoring applications. Under the described conditions, the absorption feature for CF
is nearly 1 cm
broad, containing more than 8000 absorption lines. Interference from H
, and N
O is significant, whereas CO
does not absorb in this wavelength range.
First reports of mid-infrared TLAS sensors for in-situ
measurements of PFCs employed lead-salt lasers [14
]. These sensors achieved detection limits in the low parts-per-billion (ppb) range with a few seconds acquisition time, the systems were employing multi-pass cells and extractive sampling. However, lead-salt lasers require cryogenic cooling and are in many cases unreliable; therefore they are not generally considered suitable light sources for sensors in permanent industrial installations.
Recent developments in quantum cascade laser (QCL) technology [17
] have made compact and reliable mid-infrared laser sources available that are able to operate in continuous-wave (cw) mode at room temperature and even above. QCLs have proven to be feasible in several field applications; e.g., the measurement of nitric oxide (NO) in household waste incinerators [18
] or atmospheric sensing of nitrous oxide and carbon monoxide (CO) [19
sensor presented in this work is based on the commercial LaserGas Q™ platform [20
] and is an on-stack
sensor consisting of two units: a transmitter and receiver. In the transmitter, a distributed feedback (DFB) QCL from Corning Incorporated (now Thorlabs Quantum Electronics, Inc., Dearborn Heights, MI, USA) in an optimized HHL-package [21
] is used. The laser has an optical output power of about 35 mW at +40
C (driving conditions: 320 mA, ~9.5 V), a current tuning rate of 0.015 cm
/mA and a temperature tuning rate of 0.09 cm
/K. The beam is collimated and focused by antireflection (AR)-coated CaF
-lenses. A mercury cadmium telluride (MCT) detector from Vigo System S.A. (PVI-2TE-8) is used in the receiver unit.
Optical detection of the sensor is based on the Wavelength Modulation Spectroscopy (WMS) technique previously described by, for example, Linnerud et al.
] and Kluczynski et al.
The emission optical frequency of the quantum cascade laser was adjusted to about 1283.9 cm
by setting the operation temperature to +37.5
C while applying a direct current of approximately 290 mA. A current ramp of 58.5 mA is subsequently used to tune the optical frequency within a narrow range of about 0.9 cm
. The tuning range is indicated by the green arrow in Figure 1
. The current ramp, and thus the emission wavelength of the laser, is modulated to enable WMS-based detection. The laser beam is passing through the sample and a detector is measuring the transmitted optical signal on the receiver side. A hardware mixer extracts the second harmonic component within the preamplified detector signal. The second harmonic signal is sampled by an analogue-digital converter; 64 points of the scan are recorded. Adjustable low-pass filters embedded in the signal-processing software are used to further improve the signal-to-noise ratio.
To help reduce the strong interference from H
O and other gases (Figure 1
), only a small portion of the CF
absorption feature is scanned. The wavelength region of the strongest curvature change at about 1283.3 cm
is used to perform measurements with a modulation amplitude that is much smaller than the width of the absorption feature [23
]. The measured trace is therefore closer to the true second derivative of the direct absorption signal. This is different to conventional WMS, where modulation amplitudes comparable to the width of the absorption line are applied [10
]. Although the sensitivity of this unique approach is reduced (as will be shown in the noise analysis in Section 3
), a much better selectivity can been achieved. This is especially useful to detect broad absorption features with sharp edges typical for the Q-branch of ro-vibrational spectra.
3. Laboratory Evaluation
In order to calibrate and characterize the sensor’s analytical performance, it was mounted onto a heated single-pass cell with an OPL of 700 mm and capable of heating gas mixtures up to 400
C. A gas cylinder with 4.65 ppm CF
in nitrogen (certified accuracy ±5%) and a gas mixer (HovaGas Digital G6) were used to perform a calibration of the sensor. Different concentrations ranging from 0 ppb to 1310 ppb CF
in nitrogen were generated and filled into the absorption cell. Figure 2
a shows second harmonic spectra collected at a cell temperature of 125
C. In Figure 2
b, the set concentrations of the gas mixer are plotted versus the measured concentration of the developed sensor. As can be seen, the response is linear within the accuracy of the gas mixer (±1%) and the noise level of the CF
Following analysis with spectra provided by the HITRAN and HITEMP databases, the response of the CF
sensor was evaluated with respect to potentially interfering gases by sequentially filling the heated single-pass cell with 20 ppm CH
, 10 ppm N
O, and 5% H
O (generated with a HovaCAL digital 112-MF vapor generator). The recorded second harmonic spectra are plotted in Figure 3
. A comparison of the measured spectra and the HITRAN simulation confirms good agreement between line position and absorption strength of the interferents evaluated.
quantifies contributions from the different gases to the CF
measurement. Despite the close proximity of interfering absorption lines, the influence on the CF
measurement is limited for typical emission gas mixtures in the aluminum industry.
The lower detection limit (LDL) of the sensor was determined by mounting the sensor on an absorption cell (OPL = 735 mm) filled with ambient air, putting it into a climate chamber and measuring the CF
concentration for several hours. To simulate real-world ambient conditions, the set temperature of the climate chamber was changed in several steps from ambient to 55
C and back to ambient again. For this test the averaging time of the sensor was set to 5 s. From Figure 4
, a short term noise level of less than 10 ppb·m (peak-to-peak value), corresponding to a sensitivity of 3.5 × 10
[rel. abs.], was deducted. The achieved sensitivity is in good agreement with other WMS QCL sensors, as recently reported by, for example, Ren et al.
] and Cao et al.
]. In previous work using a similar setup applying traditional WMS to measure sulphur dioxide at 7.5 µm a short term sensitivity of 2 × 10
[rel. abs.] was demonstrated [21
]. As expected, the new approach is less sensitive than traditional WMS while providing a higher selectivity for molecules with broad absorption features. Concentration values reported by the QCL sensor remained below 20 ppb·m across the testing period, so it can be assumed that this is the long-term LDL of the QCL CF
sensor. The Allan-Werle deviation [26
] and thus the theoretical LDL was not investigated since in this context only the LDL for the targeted application was of interest.
The sensitivity is limited by optical etalons which are seen as long-term variations in Figure 4
. Etalons arise from undesired multiple reflections of the laser beam within the transmitter and receiver unit.
4. Field-Test at an Alcoa Aluminum Smelter
Feasibility of the QCL CF
sensor under real-world conditions was tested by installing the unit at an Alcoa aluminum smelter. Transmitter and receiver unit were mounted on opposite sites of an exhaust duct servicing exhaust gas from 72 smelting pots (Figure 5
). The duct diameter is about 2.4 m, the pressure is around ambient and the gas temperature varies between 80
C and 100
C. Measurements were performed continuously between 25 September and 19 October 2015.
An extractive FTIR spectrometer was installed at the same location in order to perform comparative measurements. Spectral averaging time of the FTIR spectrometer was set to 20 s. To synchronize both sensors, the averaging time of the QCL sensor was configured to be the same. Figure 6
is showing a 24 h plot of the concentration readings from both instruments; the purple trace depicts the FTIR readings and the green trace the concentration obtained with the QCL CF
sensor. To make the baseline more visible, the y-axis is limited to a concentration of 200 ppb.
Baseline, LV-PFC emissions ranging 20 ppb to 100 ppb CF
and several AEs with concentrations of several parts-per-million (ppm) were recorded during the presented measurement interval. Good agreement between both measurement platforms was observed, with even small fluctuations in emitted values distinctly visible in both profiles. The offset between FTIR and QCL measurement can be attributed to interference from CH
) and appears to be nearly constant across the measurement time. Within the concentrations of interest to this study, CF
levels measured by the FTIR spectrometer are unaffected by CH
. In contrast to the QCL sensor, the prediction algorithm in the FTIR spectrometer utilizes a broader spectral range covering the entire spectral feature. As such, gas concentrations measured are less susceptible to the presence of known interferents. At the time no efforts were attempted to compensate for the CH
contribution in the QCL sensor. Variations in the offset might arise either from optical etalons slowly moving across the CF
absorption (e.g., the offset at around 5:00 is much smaller than during the rest of the measurement time) or from variations of the CH
concentration in the exhaust gas. The short term noise level of the QCL sensor is about 5 ppb and is comparable to the lab results taking the optical path length of 2.4 m into account. The value is of the same order as the FTIR noise level.
shows the concentration readings in full scale for the time period marked by the red bar in Figure 6
; multiple anode effects are occurring. As can be seen, the overall agreement is very good. However, the QCL CF
sensor displays a nonlinear behavior at concentrations above 4 ppm. The nonlinearity arises due to strong absorption of the analyte: According to simulations using HITRAN, 4 ppm CF
at ambient pressure, gas temperature of 80
C and OPL of 2.4 m results in 45% absorption. For such strong absorptions the linear approximation of WMS signals is no longer valid [10
]. To overcome this drawback, incorporating quasi-simultaneous direct absorption measurements alongside the WMS can be considered. This approach was recently demonstrated by Klein et al.
Additional background information and a detailed evaluation of the obtained measurement data can be found in Espinoza-Nava et al.