Open-path cavity ring-down methane sensor for mobile monitoring of natural gas emissions.

We present the design, development, and testing results of a novel laser-based cavity ring-down spectroscopy (CRDS) sensor for methane detection. The sensor is specifically oriented for mobile (i.e. vehicle deployed) monitoring of natural gas emissions from oil and infrastructure. In contrast to most commercial CRDS sensors, we employ an open-path design which allows higher temporal response and a lower power and mass package more suited to vehicle integration. The system operates in the near-infrared (NIR) at 1651 nm with primarily telecom components and includes cellular communication for wireless data transfer. Along with basic sensor design and lab testing, we present results of field measurements showing performance over a range of ambient conditions and examples of methane plume detection. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement

where a vacuum pump pulls the sample through an inlet and flow cell located between HR mirrors [31][32][33]. An immediate advantage of open-path for mobile sampling is the improved time response (essentially instantaneous relative to flow time-scales) meaning that the spatial shapes of observed plumes are faithfully recorded, while other benefits are associated with ease of vehicle integration, i.e. reduced power draw and sensor mass as discussed in Section 3. When closed-path flow systems are used, there is a time lag associated with passage of the sampled gas through the inlet to the cavity as well as potential broadening of impulsive signals. For example, past work by Von Fischer and colleagues with closed-path CRDS has found varying lags of ~3-9 s (corresponding to spatial shifts of ~50-150 m for typical vehicle speeds) which complicate analysis [34,35]. More generally, several studies have recognized the deleterious effects of insufficient time response to mobile sampling applications [25,26].
The closed-path flow cell architecture is primarily used to lower the sample pressure, typically to ~0.1 atm, to narrow the spectral lines as well as fix pressure and temperature which are necessary to determine gas concentration. The inlet also allows use of an inline particle filter for the dual purpose of 1) maintaining HR mirror reflectivity while also 2) reducing signal fluctuations due to additional (varying) optical loss from Mie scattering by particles in the beam path [36]. Our past work has shown that with appropriate software filters (against the Mie scatter noise) one can still maintain very adequate sensitivity [30]. Here, we focus on the design and field testing of an open-path methane CRDS sensor with emphasis on field performance in real outdoor ambient conditions.

Spectral simulations
In CRDS, the analyte species concentration is found by recording the ring-down times, τ(ν), versus laser frequency, which can be readily converted to absorption coefficients k(ν). Figure  1 shows a simulated absorption spectrum (given as vacuum wavelength) for the conditions of interest including 4 methane absorption lines (6057.08, 6057.13, 6057.09, and 6057.10 cm −1 ; which blend into a single feature at ambient conditions) as well as proximate water and carbon dioxide absorption lines. The final spectrum is computed as the sum of the contributing features: where b is the cavity baseline loss (units of cm −1 ), i is an index to sum over the needed lines, φ L are frequency-normalized Lorentzian lineshapes (units of cm) with center frequency ν 0i and width Δν i (units of cm −1 ), P is the total pressure (units of atm), S i the line strength (expressed in units of cm −2 atm −1 ), and n i the species concentration (in units of mole fraction). The use of Lorentzian lineshapes is justified by simulations showing that the effect of Doppler (thermal) broadening is negligible at atmospheric pressure. The spectroscopic constants ν 0i (given above), Δν i and S i are from HITRAN 2012, where the width is found from the pressure broadening coefficient [37]. Equation (1) is used to simulate spectra as well as to fit CRDS measurements to retrieve concentrations. In the latter case, we employ 3 free fit-parameters: the baseline loss (from mirrors, diffraction, Rayleigh scattering and any possible broad spectrum absorption interferences), laser frequency offset (as the experiment uses relative frequency given slight drift in absolute laser frequency), and CH 4 concentration. (Note that while Fig. 1 includes the H 2 O and CO 2 for purposes of exposition, we have also shown through simulation, that for the selected laser scan range, these non-methane species can be neglected from the spectral fits without compromising the retrieval accuracy of methane.) Accurate fitting requires knowledge of temperature and pressure which is achieved using real-time measurements of these parameters. Note that for plume detection, as opposed to long term ambient monitoring, one does not need to correct for dry air molar fractions since the water content (humidity) is relatively constant over the plumes (~10-100 m and ~10 2. Most eliability, istributed ng in the ler (TEC) recorded er current vent back r which is ing-down after the the AOM C-8X-C). matched to the dualparated by sed silica substrates and have reflectivity of R~0.99994 at 1651 nm. Light exiting the cavity is coupled with an off-axis parabolic mirror to a high-gain low-noise InGaAs photodetector amplifier module. The basic operational sequence is to scan the laser across the range of interest ( Fig.  1) and, at each passage across a cavity resonance (where power is injected to the cavity and recorded with the exit detector), use a trigger circuit to turn off the AOM voltage (deflecting the beam away from the cavity) allowing a ring-down event to occur. A National Instruments sbRIO-9651 based on the Xilinx Zynq-7020 system on chip (SoC) is coupled with a custom carrier board to form the electronics platform core. Acquisition, triggering and laser scanning are implemented on the FPGA side of the SoC and ring-down fitting and analysis reside on the dual-core ARM Cortex-A9 CPU running a real-time operating system. The overall scan and trigger scheme follows past work from Lehmann's group [39] with more specific detail in our past publication [30]. Figure 3 shows a solid model of the full sensor package. The sensor is comprised of two main assemblies: an optical head which contains the optical cavity (and detector) and an electronics enclosure which contains the microprocessor, controllers, circuitry and wireless gateway. The entire system was designed to handle a range of ambient conditions (temperature, vibration, possible precipitation and splash) while also optimizing size, weight and power (SWaP) for seamless vehicle integration. (An exception, as discussed below, is the enclosure used for the electronics box which, in this design iteration, was not yet optimized for size and weight.) The sensor is powered via direct connection (including inline fuse) to the vehicle battery (~12-18 V). Given the open-path design (lack of main flow pump), the power draw is quite low -ranging between <30 W with heaters off to maximum of 150 W with heaters on -and does not affect the battery and vehicle operation. The sensor mounts to the vehicle via a standard roof-rack with a custom aluminum mount using wire-rope isolators to reduce the effects of vibration. The electronics are housed in an IP-68 rated NEMA enclosure that has been designed for thermal performance and all-weather operation. In addition to the microprocessor and electronics, the enclosure also contains the laser source (and inline isolator and AOM), laser current and temperature controllers, PID temperature controllers for heating of the sensor head, pump line for mirror purge [30] and LTE cellular wireless gateway. The gateway is used to relay aggregated sensor data to a computational cloud on a lossless store-and-forward basis. Large aperture cellular antennas are mounted outside the enclosure to ensure reliable connectivity in remote areas where low-band operation is prevalent. In addition, 2x2 MIMO is employed to further enhance link reliability and multi-carrier support is accommodated via dual radio architecture. The enclosure also includes a GPS sensor (Linx, RXM-GPS-FM-T) for timestamping and spatial localization, as well as sensors for external temperature (Omega, RTD-806) and pressure (Honeywell, PX2AM1XX001BAAAX) for calculation of concentration. The top of the enclosure includes a solar shield to prevent excessive conduction heating due to sunlight exposure. The electronics enclosure is designed to operate over −30 to + 50 °C ambient and is limited primarily by the size of the laser Peltier element and thermal coefficient of expansion (CoE) of the mechanical structure. This proof-ofconcept minimizes power consumption but, to minimize design complexity in this iteration, we have used a stock NEMA enclosure not yet optimized for size and weight. Future work will significantly reduce size and weight primarily through adoption of miniaturized electronics and a high level of integration.

Sensor packaging and vehicle integration
The sensor head houses the optical cavity, along with the final steering mirrors and photodetector. Light is coupled into the head via a fiber running from the enclosure box. A critical design aspect is to maintain cavity alignment and mirror reflectivity for field use where driving can induce significant vibration, temperature can change, precipitation can be present, and contaminants (dust, moisture, splash etc.) can be present in the air. The main structure of the cavity uses carbon-fiber to achieve low mass with relatively high rigidity and low thermal expansion. A single 2" diameter tube connects the two optical ends of the sensor to define the structure and provide a conduit for purge air and wire harness routing. Rounded, removable plastic caps enclose the head on either side. The head is thermally insulated and includes heaters to maintain uniform temperature for the optical cavity with <120 W of heating power. The heating circuit comprises two ceramic cartridge heaters that operate at <20 W each in each of the 2 caps along with a single tube heater at <40 W which surrounds the 2" tube; these are controlled by 3 PID temperature controllers (one for each cap, one for tube heater). Finally, to maintain cleanliness of the two HR mirrors, small purge flows are used to prevent ambient particles or splash from reaching the HR surfaces [30]. A single compact rotary-vane pump with inline particle and moisture filters pulls in air from the atmosphere (~3 slpm) and then passes that (filtered) air to small positive pressure regions near either HR mirror (see also Fig. 2). For (on-axis) CRDS the beam size within the cavity is only ~3 mm meaning that relatively low purge flows can be used (cf. multi-pass cells techniques with larger mirror diameters [22,40]). In some cases, we also shield the entire beam line with a perforated tube to further protect the mirrors without impacting the overall air flow through the open-cavity.

Laboratory testing
A series of tests were performed to confirm sensor accuracy and precision in the laboratory. For accuracy measurements, to deliver fixed and known concentrations, a closed-path setup was temporarily used (i.e. a pump and flow cell tube were installed). The gas delivery setup consisted of two gas cylinders each with a mass-flow controller. A mixture of 250 ppm methane with nitrogen balance (from one tank) was mixed with ultra-high purity grade nitrogen (from second tank) to set desired methane concentrations. Figure 4

Field test
We

Sensor performance under varying field conditions
We have analyzed the field data to examine sensor performance under varying field conditions. We are particularly interested in the sensitivity of baseline CH 4 (based on the 1second Allan Deviation) under different ambient conditions. For detection of elevated methane plumes one requires that they be discernible relative to the (~flat) baseline. To address this, we have extracted a series of data segments (30 s each) from the data. We only accept CH 4 data that pass certain Quality Control (QC) tests; in particular, we only maintain readings where the mirror reflectivity remains above R = 0.9999, as can be diagnosed from the baseline fit parameter, b, of Eq. (1). Other QC parameters include the number of ringdowns ingested per second, which we normally require to be >300 (but can very occasionally fall below the target, for example in cases where we drive immediately behind a vehicle emitting exhaust with large particle content), and the detector baseline level (voltage with no incident light) the variation of which is indicative of detector degradation. For looking at baseline stability, we reject segments containing elevated (plume) readings as these would bias the baseline statistics. We have examined slightly over 1 million raw (1 Hertz) CH 4 concentration readings from ~280 hours of drive data, from which we find ~25,000 usable baseline segments. The median (1-second) Allan deviation of these segments is 25 ppb (while the mean is 28 ppb). The first conclusion is there is slight degradation in sensitivity, by a factor of ~2, for mobile (driving) measurements as compared to laboratory measurements. This is explained by several factors: collecting some data with slightly poorer R, more particle loading in the air, and collecting data at colder temperatures where alignment is affected (see below). Maintaining a sensitivity of better than 30 ppb under these diverse field conditions is a very positive result for practical plume detection (see below). We have examined the data to consider the factors influencing the field sensitivity. We find that the vehicle speed (over ~0-75 km/hr) has very minimal influence on sensitivity given that the median deviation is 25 ppb for all stationary readings as compared to 26 ppb for all moving readings, indicating that effects of vibration have been successfully dampened. As expected from preparatory cold-chamber tests, the measurement sensitivity has some dependence on ambient temperature (owing to thermal expansion of the cavity). Figure 7 shows a plot of the segment Allan deviations versus temperature (randomly plotting only 1/30th of the overall segments for visual clarity). The green line shows the median of all segment Allan deviations. Based on fitting these data (red curve), we find that the sensitivity   Figure 9 provides two brief examples of detection of elevated methane plumes as the sensor is driven past real oil and gas infrastructure. In both cases, one can see elevated plumes downwind of the oil and gas equipment (with baseline on either side). The top panel is due to a single pass with peak CH 4 concentration of 4.1 ppm. The lower panel is due to multiple back-and-forth passes by a larger site with distributed equipment over about 12 minutes (during which time the wind will also have some variation). Again, it is beyond the scope of this paper, but these types of concentration readings along with wind readings (from either a vehicle mounted or fixed weather station) can be used to locate and quantify methane emission sources [18][19][20]. In the Pennsylvania study, our collaborators also recorded VOC emissions that were found to correlate with the methane data [43].

Conclusions
We have presented the design and testing of a purpose built open-path CRDS methane sensor for mobile sampling of natural gas emissions. The sensor is primarily designed for use on vehicles (e.g. trucks) but we have flown more compact version (without cellular and total mass including electronics and head of 7 kg) on a drone (Matrice 600 Pro) as will be published in the future. The open-path sensor yields sensitivity of 10-15 ppb in the lab and <~30 ppb for extended field use over a range of speeds, road conditions and temperatures (−10 to 35 °C; and these limits can be improved). The sensitivity is poorer than from commercial closed-path CRDS instruments (which are overly sensitive for this application), but is very adequate for practical emissions measurements. In addition to meeting the given sensitivity target, the sensor also meets the temporal response and accuracy needs given in Section I with favorable SWAP including low power draw allowing direct connection to automobile batteries without other modifications. Future sensor versions may operate in the mid-infrared (MIR) where cross-sections are stronger, but the NIR telecom components are relatively low cost and maturely engineered. Recent data [24] indicates that a sensitivity of ~100 ppb allows detection of mass flows as low as ~0.03 g/s within ~100 m of the source (and higher mass flows at larger distances), as we have shown with our instrument at the METEC facility. At the system level, detecting emissions of >~0.03 g/s corresponds to measuring well over 95% of total emissions from real oil and gas fields based on published inventories and given the long-tail distribution of site emissions and super-emitters (e.g [19,44,45].).
The simplicity and autonomous nature of the sensors makes them well suited for mobile application including research purposes as well as commercial use on oil and gas company work trucks for monitoring of basins or enterprise regions. Such measurements would have the spatial granularity to monitor individual well sites and enable data-driven asset management (e.g. prioritization of repairs). The spatially resolved measurements can also complement coarser resolution satellite measurements [8][9][10][11][12].

Funding
Advanced Industries program from the Colorado Office of Economic Development and International Trade.