Photo-acoustic sensor for detection of oil contamination in compressed air systems
Mikael Lassen, David Baslev Harder, Anders Brusch, Ole Stender Nielsen, Dita Heikens, Stefan Persijn, Jan C. Petersen
PPhoto-acoustic sensor for detection of oil contamination in compressed air systems
Mikael Lassen, ∗ David Baslev Harder, Anders Brusch, Ole StenderNielsen, Dita Heikens, Stefan Persijn, and Jan C. Petersen Danish Fundamental Metrology, Matematiktorvet 307, DK-2800 Kgs. Lyngby, Denmark VSL - The Dutch Metrology Institute, Thijsseweg 11, 2629 JA Delft, The Netherlands
We demonstrate an online (in-situ) sensor for continuous detection of oil contamination in com-pressed air systems complying with the ISO-8573 standard. The sensor is based on the photo-acoustic (PA) effect. The online and real-time PA sensor system has the potential to benefit a widerange of users that require high purity compressed air. Among these are hospitals, pharmaceuticalindustries, electronics manufacturers, and clean room facilities. The sensor was tested for sensitivity,repeatability, robustness to molecular cross-interference, and stability of calibration. Explicit mea-surements of hexane (C H ) and decane (C H ) vapors via excitation of molecular C-H vibrationsat approx. 2950 cm − (3.38 µ m) were conducted with a custom made interband cascade laser (ICL).For the decane measurements a (1 σ ) standard deviation (STD) of 0.3 ppb was demonstrated, whichcorresponds to a normalized noise equivalent absorption (NNEA) coefficient for the prototype PAsensor of 2.8 × − W cm − Hz / . I. INTRODUCTION
Compressed air is an essential asset for many industriestoday. It is safe and relatively inexpensive to operate andvery reliable [1, 2]. However, compressed air is suscepti-ble to oil contaminants, which even in minuscule quanti-ties can be disastrous for a number of industrial applica-tions and be health threatening for humans. In 1991 theInternational Standards Organization (ISO) establishedthe ISO-8573-1 standard on purity of compressed air [3].This was done in order to govern compressed air systemcomponent selection, design and measurement. Accord-ing to this standard all hydrocarbons with 6 or morecarbon atoms per molecule are considered as ”oil”. Thefollowing organic compound classes (hydrocarbons) aretherefore all ”oils”: solvents (e.g. toluene, hexane, de-cane,... ), VOCs (Volatile Organic Compounds), adhe-sives, thread and surface sealants, fragrances (air fresh-eners, perfumes, etc.), detergents/cleaning agents. Whenoil is present in pipelines of the compressed air system,it has always one of the three forms: Aerosols, wall flowand/or vapors (oil mist). Aerosols are partially removedby coalescing filters and appear as condensate. Wall floweither appears in condensate or travels along the walls ofthe pipeline to the end user, again partially removed bycoalescing filters. Vapors and oil mist are not removedby coalescing filters and can therefore be a huge problem.The lack of any reliable, highly sensitive, online sensorsystem has forced critical industries to rely on manualsampling and subsequent laboratory analysis, which islabour intensive, inefficient, and cannot guarantee con-formance of compressed air systems to the mandatory orindustry adopted regulatory norms. The main character-istics of the sensor should therefore be its capability todo online measurements of ”oil” contaminant concentra-tion at 8 ppb or lower in order to meet the sensitivity ∗ Corresponding author: [email protected] requirements of the ISO-8573 standards for compressedair purity levels of Class 1 or better. Two kinds of oilare used for air compressors today: synthetic and min-eral oils. According to a 2007 publication by Colyer [4],classical hydrocarbons had a market share of about 80%,however polyglycol lubricants are claimed to have nu-merous benefits for use in compressors and are heavilypromoted by compressed air companies. a) b) PA cell ICLFlow control
FIG. 1. a) 3D design drawing of the prototype sensor. Thesensor platform consists of three units: the sampling system(not shown), the optical spectroscopy unit, and the electron-ics/software. The front box is the PA sensor head, while therear box is the system control unit. The system control unitconsist of a combined field-programmable gate array (FPGA)and microprocessors. Through software the system controlunit provides control and monitoring of the vital parts in thePA sensor. b) Picture of the inside of the PA sensor head.
We demonstrate an oil contamination sensor for on-line monitoring of compressed air. The sensor is basedon the photo-acoustic (PA) effect [5–9]. Photo-acousticspectroscopy (PAS) is an established technique for en-vironmental, industrial, and biological monitoring andimaging, due to its ease of use, compactness and its capa-bility of allowing trace gas measurements at the sub-partsper billion (ppb) level [10–17]. The design and layout ofthe prototype PA sensor is depicted in Fig. 1. [18].The PA system consists of a sample collector piece thatis installed into the compressed air system, using the so- a r X i v : . [ phy s i c s . i n s - d e t ] D ec called isokinetic sampling method, not shown in the fig-ure, and two units; a PA sensor module and a drivermodule. The PA technique is based on the detection ofsound waves that are generated due to molecular absorp-tion of modulated optical radiation. The generated PAsignal is proportional to the density of molecules, whichmakes the PA technique able to measure the absorptiondirectly, rather than relying on having to calculate it fromthe transmission of the radiation. The PAS technique ishowever not an absolute technique and calibration is re-quired against a known sample (known concentrations ofthe gas/oil).Explicit measurements of hexane (C H ) and decane(C H ) vapors via excitation of molecular C-H vibra-tions at approx. 2950 cm − (3.38 µ m) were conductedwith an interband cascade laser (ICL) [19]. The PAsensor has an oil concentration sensitivity lower than 1ppb with traceability to certified reference gases (veri-fied by VSL, The Dutch Metrology Institute), applica-ble for Class 1 detection (the most stringent classifica-tion according to ISO 8573). A linear dependency be-tween the spectroscopic signal and the hexane and decaneamount of substance fractions was observed, confirmingthe proper functioning of the PA sensor for all classes ofoil contamination. The interface between the compressedair system and the online PA sensor is critical as it war-rants that representative samples of the compressed airare taken. The sample system should be able to functionup to a pressure of 16 bar and be able to stand pressurepulses up to 32 bar (¡1 second). The design is there-fore robust and heavy designed in order to withstandhigh pressures. Continuously air samples are taken usingthe so-called isokinetic sampling method. The sensor isflowed continuously with 1.7 L/min for fast sampling ofthe compressed air. The sampling probe is placed at thecenter of the pipes of the compressed air system. Due tothe particular choice of tube lengths and diameters, thespeed of air in the compressed air system is the same asthe speed of air in the probe. More details on the isoki-netic sampling can be found in the ISO 8573 standard[3].The sensor was tested for the following parameters:Sensitivity (ppb level), accuracy and standard deviation(STD) of the measurements, long term stability, responsetime, reproducibility and bias/background signals, self-cleaning methods and flow noise immunity. In this contri-bution we present data for the sensitivity, response timeand reproducibility of the PA sensor and the stabilityover time. The ability to detect oil contaminants online,will significantly enhance the capabilities of manufactur-ers and users to guarantee the quality of their productsand eliminate a number of risks and civil liabilities thatare associated with non-conformance. II. THE PHOTO-ACOUSTIC EFFECT
The sensor is based on the photo-acoustic effect [7, 8].The PA technique detects sound waves that are gener-ated due to absorption of modulated optical radiationin various molecular species. In a gas the sound wavesare generated due to local heating via molecular colli-sions and de-excitation. Absorption of laser radiationexcites the molecules in the PA cell. The added energyis via collisional processes converted to local heating andde-excitation of the molecules in the PA cell thus gen-erating sound waves. A pressure sensitive device (e.g.microphone, tuning fork or cantilever) is used to moni-tor these modulated sound waves. The magnitude of themeasured PA signal in volts is given by: S P A = S m P F α, (1)where P is the power of the incident radiation, α is theabsorption coefficient, which depends on the total num-ber of molecules per cm and the absorption cross sec-tion, S m is the sensitivity of the microphone and F is thecell-specific constant, which depends on the geometry ofacoustic cell and the quality factor Q of the acoustic res-onance [7]. Ideally a highly sensitive PA sensor shouldonly amplify the sound waves and reject acoustic andelectrical noise as well as in-phase background absorp-tion signals from other materials in the cell (walls andwindows). III. EXPERIMENTAL SETUP
ICL Gas InletGas Outlet
Flow CellDetector HR MirrorPAS cell
SW SW SW
PAS CellICL Amplifiers Operating SystemICL Controller 16 bit DAQLock-InAmplifiers PAS cell ControllerOptical Detector
MIR Light
PAS Sensor Head a) b)
FIG. 2. a) Block diagram of the main parts of the setup. ThePAS cell controller regulates the temperature, the flow andthe pressure inside the cell. The ICL controller controls thetemperature and the modulation of the ICL current. b) Showsthe optical parts of the sensor head and inner PAS cell. A3.38 µ m laser beam from the ICL (Interband Cascade Laser)is aligned to make a double pass through the cell in orderto enhance the interaction with the oil molecules. Mount forthe ICL; PAS cell; back coupling mirror; SW: Silicon Window.The center of the PAS cell shows a COMSOL simulation of theacoustic resonance mode at 6.5 kHz. Two 30 mm in diameterplates constitute the acoustic resonator, and are the maincomponents of the PAS cell. The block diagram of the main parts of the setup andthe schematics of the sensor head are shown in Fig. 2.A typical setup for PAS involves an amplitude modu-lated light source and a resonant absorption cell withmicrophones, where the PA signal is enhanced by theacoustic resonances. The highest conversion efficiency oflaser power to PA signal is obtained when the laser spec-trum is narrower than the absorption feature, since laserpower at frequencies outside the absorption feature willnot contribute to the PA signal. The C-H stretch bandof the oils to be monitored are located at approximately3.38 µ m and is shown for hexane in Fig. 3(b). This bandis very broad and spectrally dense, thus high resolutionspectroscopy probing individual ro-vibrational absorp-tion lines in the oil molecules is not required. Therefore abroadband custom made ICL is used. The ICL spectrumis shown in Fig. 3(b) together with the hexane absorp-tion spectrum. Additional requirements to consider arethe power level required for the PAS measurements, com-pactness, form factor and price. The laser is temperaturecontrolled with a Peltier element and kept at 20 ◦ C (+/-0.1 ◦ C). If the temperature of the ICL is not kept con-stant the laser spectrum shifts in frequency. This willchange the overlap between the laser spectrum and theoil absorption spectrum resulting in a changed PA signaland thus in a change in the evaluated oil concentrationlevel. The modulation of the light intensity is controlledby modulating the current, where a square-wave is fedinto the laser controller. The duty cycle used is always50/50. The laser controller can be operated with fre-quencies from 100 Hz to 20 kHz. The output beam fromthe ICL is collimated with a molded IR aspheric lens.The lens is AR coated in the wavelength region 3 - 5 µ m. After the collimating lens a beamsplitter is insertedto tap approximately 0.5% of the laser light onto a MIRphoto detector (PbSe, 1.5-4.8 µ m, AC-Coupled Ampli-fier) for optical power measurement and normalizationof the PA signal. The optical transmission through thecell is approximately 97% at 3.38 µ m and the absorp-tion coefficient is approximately 10 − cm − for each cellwindow. The 3.38 µ m ICL can deliver over 60 mW of op-tical power. The PA signal can be enhanced by opticalmulti-pass techniques resulting in an increase of the sen-sitivity of the PA spectrometer due to the increased lightabsorption path length from multiple reflection. Variousmulti-pass and single-pass configurations have so far beenexploited for PAS configurations, such as ring cells, cavitybased cells and transverse square cells [20–25]. Here weuse a double pass configuration with a highly reflectionmirror (on the order of 99% and a radius of curvatureof 500 mm) to reflect the beam back into the PA cellas illustrated in Fig. 2(b). The double pass configura-tion is used in order to enhance the interaction with theoil molecules, thereby enhancing the PA signal and thesensitivity.The acoustic resonator of the PAS cell has an open cellconfiguration and is made by two circular plates with adiameter of 30 mm and separated by a distance of 10 mm.The acoustic response of the PA cell is shown in Fig. 3(a).In order to have a continuous flow of air through the Hexane Absorp. ICL 3.38 um N o r m a li z ed I n t en s i t y / ab s o r p t i on Wave Number [cm -1 ] N o r m a li z ed PA s i gna l [ a . u .] Frequency [Hz]
Q ~35
The Acoutics mode a) b)
FIG. 3. a) Shows the acoustic resonance at approximately 6.5kHz and the associated drum skin mode. The visualization ofthe drum skin mode was simulated using COMSOL. b) Thenormalized spectrum of the ICL (red curve) centered at 3.38 µ m. Plotted together with a Hexane absorption spectrum(black curve). PAS cell (up to 2 l/min), input nozzles and output noz-zles have been attached on opposite sides of the flow cell.The dimensions of the nozzles are 10 mm in diameter andthe flow is kept stable using massflow controllers. Theacoustic resonator is covered by an outer cell with an 80mm diameter. It acts as the buffer zone for the acousticflow noise and the acoustic signal generated by windowabsorption. The design of the buffer flow zone has beenboth theoretically and experimentally investigated. Thetheoretical investigation included the use of the Open-FOAM software, however the simulations are not shownhere. The aim of the analysis was to investigate and op-timize the influence of the cell geometry and flow rate onthe gas distribution within the cell and how flow noise isaffecting the acoustics resonator. The PA cell is heatedto 65 ◦ C ( ± ◦ C). The temperature control of the PAis very important since the exact resonance frequency isa function of temperature and shifts in temperature willadd an uncertainty to the PA signal.The microphone is positioned in the middle of theplates, thus at the maximum acoustic amplitude. Thesignal from the microphone were amplified with a home-build low noise amplifier with variable gain and filteredwith a 7 kHz bandpass filter (3-10 kHz) before furthersignal processing. All data was processed using a lock-inamplifier. The ICL modulation is controlled by a signalgenerator, which also acts as the local oscillator for thelock-in amplifier. The data from the lock-in amplifier iscollected with a 250 kS/s data acquisition (DAQ) cardwith 16 bit resolution.
IV. RESULTS USING CERTIFIED OILSAMPLES PREPARED AT VSL
The PAS technique is not an absolute technique andrequires calibration using a certified gas reference sample.The experiments were performed by excitation of molecu-lar C-H ro-vibrational modes of hexane and decane. Thesamples were prepared using the dynamic generation sys-tem at VSL, the Dutch Metrology Institute. Part of thissystem is shown in Fig. 4. The dynamic system is a two-
FIG. 4. The two stage dilution system set up used during testsat VSL. The VSL measurements are conducted by measuringthe mass loss of the liquid mixture over time with a sensitivebalance. Information about the mixing ratios for the variousdilution steps determines the concentration. step dilution system. A small amount of substance (inliquid phase) is forced via a small capillary into an oven,which is heated and flowed with known amounts of nitro-gen and pure air. The next dilution step further decreasesthe concentration by dilution with a known amount ofpure air. The mass loss is measured every 60 seconds.Knowledge of the mass loss of the substance togetherwith the various flow rates allows the hexane and decaneconcentration to be calculated with high precision. Inthe following data all concentrations and associated un-certainties for the hexane and decane are determined bythe mass loss of the oil substance. Note that all mea-surements with the PA sensor have been tested experi-mentally with 1.7 l/min of flow and the PA sensor cellpressure was maintained at approximately 1 bar for allexperiments.
Class 2 PA Lo ck - i n S i gna l [ V o l t] Hexane Concentration [ppb]
Class 1 a) b) PA s i gna l m ea s u r edb ys en s o r [ V o l t s ] VSL ppm levels for Hexane and DecaneDecane Hexane
FIG. 5. The sensors linear dependency between the lock-inamplifier signal and the hexane and decane amount of sub-stances (concentrations). a) ISO class 3 and ISO class 4 mea-surements of Decane and Hexane. b) ISO class 1 (below 80ppb) and ISO class 2 (below 800 ppb) measurements of decaneand hexane. Note that ppm: µ mol/mol and ppb: nmol/mol. Figure 5(a) demonstrates the linear dependency be- tween the lock-in amplifier signal and the hexane anddecane concentrations with an uncertainty of less than3% for each measurement. The fitted line for the de-cane measurements has the best linear response. This isprobably due to a small drift in the alignement and sta-bilization of the gas sample in the PA sensor. Since thedecane concentrations were measured within a few hourswhile the hexane measurements were measured during 2days without realignment of the optical system. Fig. 5(b)shows measurements of hexane measured within a fewhours and it is observed that the linearity of the PA sen-sor is better. These results also stress the importance forcalibrating and recalibrating of the PA sensor in orderto achieve trusted values for the concentrations. FromFig. 5(a) it can be seen that the decane signal is approx-imately 1.5 times larger than the hexane response whichis in good agreement with the normalized absorption as afunction of carbon number [26]. The data shown in Fig. 5is processed with a lock-in amplifier with an integrationtime of 3 seconds. D e c ane C on c en t r a t i on [ ppb ] Time [Hours]
Class 0
Mean = 11.5 ppb STD = 0.3 ppb
Background Signal
Mean ~ 0.98 ppbSTD ~ 0.03 ppb D e c ane C on c en t r a t i on [ ppb ] Time [Hours]
VSL mean ~ 1 ppb a) b)
FIG. 6. a) Concentration measurements of decane (C H )as a function of time using the PA sensor. The mean con-centration used was 11.5 ± . Figure 6(a) shows the result of measurements per-formed on an 11.5 ± . σ ) STD of 0.3 ppb. Decane has amaximum absorption of 5.5 × − cm − for 1 µ mol/molat 3.38 µ m, thus the maximum normalized noise equiv-alent absorption coefficient is 2.8 × − W cm − Hz / .This makes the PA sensor comparable with state of theart PA sensors [13, 27, 28]. The data shown in Fig. 5and Fig. 6 demonstrates the potential of the PA sensorfor measuring all five ISO classes of oil contamination incompressed air.Figure 6(b) shows the measurement of the backgroundsignal and a class 0 decane concentration. The dilutionstage measured a mean concentration of 1 ± .
53 ppb overthe 2 hours of measurements. This value is used to nor-malize the data seen in Fig. 6(b). The background signaldue to unwanted absorptions in the PA cell (windows,acoustic plates, ...) is approximately 0.93 ppb. This valuedetermines the absolute sensitivity of the PA cell. Thegraph shows measurements of a 1 ppb decane concen-tration with a S/N (Signal to background) of 1.1. Thismeans that the S/N for a class 1 detection would be bet-ter than 11 and this demonstrates that the prototype PAsensor has the needed sensitivity to detected class 1 orbetter concentrations. All data shown in Fig. 5 are pro-cessed with a lock-in amplifier with an integration time of10 seconds, except the gray shaded area which were pro-cessed with a lock-in amplifier with a integration time of1 second. Note that using heavier oils e.g. going fromhexane to decane as demonstrated, improves the sensi-tivity of the PA sensor we would therefore expect fortypically oils used for compressors (with up to C ) thatthe sensitivity would be one order of magnitudes higher[26]. a) b) Time [Hours]
PASS i gna l ( m H e x ane ) [ V o l t] Differences between the peaks are 0.5% La s e r I n t en s i t y [ V o l t] H e x ane C on c en t r a t i on s [ ppb ] Tme [Hours]
FIG. 7. Test of reproduciblility and response time of the sen-sor. a) The sensor is flowed with 10 ppm of hexane followedby a flow with pure air. The blue curve shows the laser inten-sity over the measurement time. b) The sensor is flowed with222 ppb of hexane followed by a flow with pure air. Thenflowed with 65 ppb and 24 ppb of hexane followed by a flowwith pure air.
In order to test the reproducibility and the responsetime of the sensor the sensor was flowed with two differ-ent concentrations of hexane, 10 ppm and 222 ppb, re-spectively, followed by a flow with pure air. Figure 7(a)shows that the reproducibility of the sensor system sig-nal when using a very high concentration of oil (10 ppm).The reproducibility is within 0.5% peak to peak. The re-sponse time of the sensor is fairly quick and that levelstabilizes within 30 seconds. Figure 7(b) shows that thereproducibility when the system is flowed with 222 ppbof hexane of the PA sensor signal is within 1% from peakto peak and with a class 1 sensitivity (STD ¡ 8 ppb). Theresponse time is around 15 seconds for reaching the fulllevel of 222 ppb again. However in both for the high con-centration and the low concentration the response time depends on the lock-in amplifiers integration time (inthese measurements the integration time of the lock-inamplifier was 3 seconds). From this we conclude that thetime response of the sensor will not be the limiting factorfor notifying the customer or end user of an oil breach inthe air compressor filters.
V. CONCLUSION
In conclusion the prototype PA sensor provides con-tinuous online (in-situ) measurements of oil contamina-tions in compressed air systems. The PA sensor provide aunique sensitivity that allows detection of less than 1 ppbcontaminations levels of all lubricates and combinationsof oils. The prototype PA sensor with flow buffer zone hasbeen tested experimentally with 1.7 l/min flow and meandecane concentration measurements down to 1 ppb weredemonstrated. It was found that after 8 hours of flush-ing the PA sensor with a decane concentration of 11.5ppb in air, a (1 σ ) STD was of 0.3 ppb was obtained, andthus a normalized noise equivalent absorption coefficientof 3.1 × − W cm − Hz / was demonstrated. Fur-ther the measurements demonstrate that the PA sensorfulfills many of its requirements, namely the sensitivity,reproducibility, flow immunity, response time, linearityof the PA signal to oil any oil concentration within theISO-8573 classes and repeatability of the background biassignal (no contamination). To summarize the PA sensorsspecifications are: sensitivity: 0.3 ppb (class 0: ¡8 ppb),reproducibility within 1%, time response less than 15 sec-onds, flow immunity up 2 l/min flow without reducing thesensitivity.A recently started Eurostars project titled PASOCA(Photo-Acoustic Sensor for Oil detection in CompressedAir) will take the developed prototype PA sensor with-out impairing it to a certified commercial product. Wetherefore believe that with further miniaturization andproper mechanical design the PA sensor technology de-veloped here will be ready for mass production. We fore-see that the developed oil contamination PA sensor willbenefit a large category of industries that require highpurity compressed air including hospitals, pharmaceuti-cal industries, electronics, clean rooms and many otherindustries which are in need of pure compressed air. FUNDING AND ACKNOWLEDGMENTS
We acknowledge the financial support from EUREKA(Eurostars program: E10132 - PASOCA) and the Dan-ish Agency for Science, Technology and Innovation. TheEuropean Unions Seventh Framework Programme (FP7)managed by REA 8211 under Grant Agreement N.286106. We would like to thank Poul Jessen and SørenLaungaard from PAJ Group ([email protected]) for fruitful dis-cussions and collaboration.
VI. REFERENCES
Photoacoustic Spectroscopy in Trace Gas Mon-itoring, in Ensyclopedia of Analytical Chemistry ed. byR. A. Meyers, (John Wiley & Sons Inc., 2000).[7] A. Rosencwaig,
Photoacoustics and Photoacoustic Spec-troscopy (John Wiley & Sons Inc., 1980).[8] A. G. Bell,”The production of sound by radiant energy,”Phil. Mag. , 510 (1881).[9] A. C. Tam, ”Applications of photoacoustic sensing tech-niques,” Rev. Mod. Phys. , 381–431 (1986).[10] S.-L. Chen et al., ”Efficient real-time detection of tera-hertz pulse radiation based on photoacoustic conversionby carbon nanotube nanocomposite,” Nat. Photon. ,537–542 (2014).[11] L. V. Wang and S. Hu, ” Photoacoustic tomography: invivo imaging from organelles to organs,” Science ,1458–1462 (2012).[12] A. Szab´o, A. Mohacsi, G. Gulyas, Z. Bozoki, and G. Sz-abo, ”In situ and wide range quantification of hydrogensulfide in industrial gases by means of photoacoustic spec-troscopy,” Meas. Sci. Technol. (6), 065501 (2013).[13] J. Peltola, T. Hieta, and M. Vainio, ”Parts-per-trillion-level detection of nitrogen dioxide by cantilever-enhancedphoto-acoustic spectroscopy,” Opt. Lett. , 2933–2936(2015).[14] V. Spagnolo, P. Patimisco, S. Borri, G. Scamarcio,B. Bernacki, and J. Kriesel, ”Part-per-trillion levelSF detection using a quartz enhanced photoacous-tic spectroscopy-based sensor with single-mode fiber-coupled quantum cascade laser excitation,” Opt. Lett. , 4461–4463 (2012). [15] M. Lassen, L. Lamard, Y. Feng, A. Peremans, and J. C.Petersen, ”Off-axis quartz-enhanced photoacoustic spec-troscopy using a pulsed nanosecond mid-infrared opticalparametric oscillator,” Opt. Lett. , 4118–4121 (2016).[16] C. K. N. Patel, ”Laser photoacoustic spectroscopy helpsfight terrorism: High sensitivity detection of chemicalwarfare agent and explosives,” Eur. Phys. J. Special Top-ics , 1, 1–18 (2008).[17] M. Lassen, A. Brusch, D. Balslev-Harder, and J. C.Petersen, ”Phase-sensitive noise suppression in a pho-toacoustic sensor based on acoustic circular membranemodes,” Appl. Opt. Chemical Properties Handbook (New York:McGraw-Hill, 1999)[20] A. Miklos, S.C. Pei, and A.H. Kung, ”Multipass acousti-cally open photoacoustic detector for trace gas measure-ments,” Appl. Opt. , 2529–2534 (2006).[21] J. Saarela, Johan Sand, T. Sorvajarvi, A. Manninen, andJ. Toivonen, ”Transversely excited multipass photoacous-tic cell using electromechanical film as microphone,” Sen-sors , 5294–5307 (2010).[22] A. Manninen, B. Tuzson, H. Looser, Y. Bonetti, and L.Emmenegger, ”Versatile multipass cell for laser spectro-scopic trace gas analysis,” Applied Physics B , 3, 461–466 (2012).[23] M. Lassen, D. Balslev-Clausen, A. Brusch, J. C. Petersen,”A versatile integrating sphere based photoacoustic sen-sor for trace gas monitoring,” Opt. Express , 11660–11669 (2014).[24] M. N¨agele and M. W. Sigrist, ”Mobile laser spectrome-ter with novel resonant multipass photoacoustic cell fortrace-gas detection,” Appl. Phys. B , 895–901 (2000).[25] J. Rey, D. Marinov, D. Vogler, and M. Sigrist, ”Investi-gation and optimisation of a multipass resonant photoa-coustic cell at high absorption levels,” Appl. Phys. B ,261–266 (2005).[26] The PNNL Quantitative Infrared Database for Gas-Phase Sensing, nwir.pnl.gov.[27] P. Patimisco, G Scamarcio, F. K. Tittel and V. Spag-nolo, ”Quartz-enhanced photoacoustic spectroscopy: Areview,” Sensors , 6165–6206 (2014).[28] A. Sampaolo, P. Patimisco, M. Giglio, L. Chieco, G. Sca-marcio, F. K. Tittel, and V. Spagnolo, ”Highly sensi-tive gas leak detector based on a quartz-enhanced pho-toacoustic SF sensor,” Opt. Express24