1. INTRODUCTION

Understanding internal combustion engine (ICE) exhaust gas properties is crucial for improving engine performance and reducing emissions of common pollutants. Commercially available instruments are regularly used to measure thermodynamic properties of ICE exhaust such as temperature and gas composition. However, commercial instruments exhibit some significant drawbacks that include: low measurement rates (typically between 1 and 10 Hz), probes that stick into and perturb the flow field, and extractive gas sampling.

Laser absorption spectroscopy (LAS) offers a solution to the aforementioned drawbacks by providing high-rate (kHz to MHz) non-intrusive in situ measurements of gas properties in harsh environments [1]. In the mid-infrared (MIR), free-space emitting lasers are typically used for LAS measurements due to the lack of commercially available fiber-coupled lasers, but for certain applications, fiber coupling the lasers is required to achieve robust optical access [1–4]. LAS has been used by many researchers to acquire in situ measurements of temperature, mole fraction, pressure, and bulk flow velocity in harsh environments. Some examples of prior LAS measurements include: scramjet engines [5], post-detonation fireballs [6], flames from solid rocket propellants [7], rotating detonation rocket engines [4,8,9], coal-fired power plants [10,11], and shock tubes [12]. There have also been numerous studies in which LAS was used to study various aspects of ICEs using near-infrared (NIR) lasers [13–20], MIR lasers [21–26], and fiber-coupled MIR lasers [27,28].

The research utilizing NIR lasers to study ICE systems includes temperature and species measurements in exhaust, in-cylinder, intake runners, and diesel engine exhaust gas recirculation (EGR). Jatana et al. [13] utilized NIR and visible LAS to acquire measurements of water, oxygen, temperature, and pressure in ICE exhaust at a rate of 5 kHz. Notably, a multi-pass cell was employed to enhance the measurement path length of oxygen due to its low absorption near the 760 nm transition. Stritzke et al. [14] used a NIR diode laser near 2.2 µm for ${{\rm NH}_3}$ concentration measurements in diesel engine exhaust. The laser beam was split into eight coplanar beams to obtain 1 Hz measurements of 2-D ${{\rm NH}_3}$ distribution using tomography. Cassady et al. [15] used a NIR tunable diode laser near 1.8 µm to acquire crank-angle resolved LAS measurements of temperature, water vapor, and pressure. The measurements were acquired in the intake of a single-cylinder gasoline research engine. An LAS sensor for in-cylinder measurements [16,17] provided crank-angle resolved thermometry and measurements of water or gasoline mole fraction at rates up to 7.5 kHz. The in-cylinder measurements were acquired near 1.3 µm using single-cylinder optical test engines designed for enhanced optical access. The work done to study diesel engine EGR [18–20] focused on making measurements of temperature and either ${{\rm H}_2}{\rm O}$ or ${{\rm CO}_2}$ using tunable diode lasers (TDLs). The measurement rates varied between 100 Hz and 5 kHz. The LAS thermometry and speciation measurements from these studies demonstrated superior time resolution compared to the measurements obtained from thermocouples and conventional gas analyzers.

The studies on MIR lasers for ICE systems mainly focused on exhaust gas thermometry and speciation; however, they were also used to provide measurements in-cylinder and in diesel engine EGR. Louvet et al. [21] performed exhaust gas measurements at a rate of 1 kHz to determine temperature and CO mole fraction in a single-cylinder test engine using a laser source near 5 µm. Additionally, measurements of NO mole fraction were obtained near 5.2 µm. A decade later, Kasyutich et al. [22] employed LAS to measure NO in the exhaust of a test engine at a rate of 953 Hz, utilizing a laser source operating near 5.2 µm. Other MIR LAS ICE exhaust measurements [23,24] were conducted using on-board sensors and multi-pass cells to measure ${{\rm NO}_x}$ emissions. Although the measurement rates were relatively slow (approximately 1 Hz), these portable sensors were battery-powered and capable of in-vehicle operation while the vehicles were in motion, demonstrating exceptional sensitivity with detection limits as low as 1 ppb. In-cylinder measurements of temperature and gasoline concentration at 10 kHz were made in a 4-cylinder production engine by Jeffries et al. [25] who used a MIR laser near 3.36 µm. Green [26] made ${{\rm CO}_2}$ measurements near 4.4 µm to monitor cylinder to cylinder EGR distribution in the intake of a diesel engine.

LAS has been also used with fiber-coupled MIR lasers for ICE measurements when the combustion environment being studied was too harsh or too spatially constrained for the laser to be in close proximity to the measurement location. Recently, Schwarm et al. [27] made 10 kHz measurements of temperature, CO, and NO in the exhaust of a production grade polyfuel engine. This was accomplished by fiber coupling a QCL near 4.85 µm and an ICL near 5.18 µm into separate fibers with separate beam paths and detectors. Around the same time, our group fiber coupled two MIR QCLs near 4.85 and 4.96 µm into one single-mode ${{\rm InF}_3}$ patch cable for 15 kHz measurements of temperature and CO mole fraction in 8-cylinder ICE exhaust on a vehicle [28]. The QCLs were time-multiplexed and the laser beams from each laser were monitored by the same detector.

This work builds on our prior work in two primary ways. (1) To our knowledge, this paper describes the first MIR LAS sensor to acquire simultaneous measurements of temperature and three species (CO, NO, and ${{\rm CO}_2}$) in engine-out exhaust, especially from a multi-cylinder ICE integrated into a vehicle. (2) This was enabled by a robust fiber-coupled sensing package employing only two single-mode ${{\rm InF}_3}$ fibers and two colinear beampaths. The measurements provided by this sensor enabled detailed quantification of transients in critical exhaust gas properties induced by engine start up and advanced engine control modes.

2. LAS FUNDAMENTALS

In LAS experiments monochromatic laser light with incident intensity (${I_0}$) and frequency ($\nu$) traverses a gas in which photon energy is absorbed by specific molecules. The transmitted intensity (${I_t}$) of the laser beam is measured by a photodetector. The amount of absorption that occurs is dependent on the thermodynamic properties of the test gas. These properties include temperature, pressure, species concentration, and path length. Beer’s law relates the amount of absorbance to the incident (i.e., baseline) and transmitted light intensities. A spectroscopic model is then used to relate the absorbance to thermodynamic properties of the gas as shown in Eq. (1):

Here, ${\alpha _\nu}$ is the absorbance, $({I_t}/{I_0})$ is the fractional transmission, $j$ denotes a given absorption transition within the range of optical frequencies across which the laser is tuned, ${S_j}(T)$ (${{\rm cm}^{- 2}}\;{{\rm atm}^{- 1}}$) is the linestrength of transition $j$, ${\Phi _{\nu ,j}}(\nu ,T,P,\chi)$ (cm) is the lineshape function of transition $j$, $P$ (atm) is the gas pressure, $\chi$ is the mole fraction of the absorbing species, and $L$ (cm) is the path length of the laser beam through the absorbing gas. The linestrength and lineshape are both important spectroscopic quantities that determine the absorbance at a given frequency.

The linestrength at a given temperature can be calculated using Eq. (2):

Here, ${T_o}$ is the reference temperature of 296 K, ${S_j}({T_o})$ (${{\rm cm}^{- 2}}\;{{\rm atm}^{- 1}}$) is the transition linestrength at the reference temperature, $Q$ is the absorbing species’ partition function, $h$ is Planck’s constant (${\rm J} \cdot {\rm s}$), $c$ is the speed of light (cm/s), $E_j^{\prime \prime}$ (${{\rm cm}^{- 1}}$) is the lower-state energy of the absorption transition, ${k_B}$ is Boltzmann’s constant (J/K), and ${\nu _{o,j}}$ (${{\rm cm}^{- 1}}$) is the linecenter frequency of the absorption transition.

The lineshape function represents a probability distribution function that describes how the integrated absorbance (i.e., the area under an absorption curve) is distributed in frequency space. The lineshape function is dependent on a variety of line broadening and, to a lesser degree, line narrowing processes. Two of the most important broadening processes are collisional (i.e., pressure) broadening and Doppler broadening. These are both accounted for, assuming they are not coupled, in the Voigt profile. Doppler broadening occurs when atoms or molecules have a velocity component in the direction of the laser beam, which leads to the molecules absorbing light at slightly upshifted or downshifted frequencies. Since the absorbing molecules in the gas are moving with a wide range of velocities, the absorption spectrum is broadened in frequency space. In contrast, collisional broadening occurs when molecules experience inelastic collisions, thus reducing their lifetime in an absorbing or emitting state. Heisenberg’s uncertainty principle states that when this occurs, the energy of the molecule becomes less certain, which leads to the absorption of photons with a wider range of energies (i.e., frequencies), thereby broadening the lineshape. Collisional broadening can also be caused by elastic collisions that dephase the molecular motion or by elastic collisions that change the molecular angular momentum vector. The Doppler full-width at half-maximum (FWHM), $\Delta {\nu _{D,j}}$, and collisional FWHM, $\Delta {\nu _{c,j}}$, of a given transition can be calculated using Eqs. (3) and (4), respectively:

 

Fig. 1. Stick plot of absorption transition linestrengths for CO, NO, ${{\rm CO}_2}$, and ${{\rm H}_2}{\rm O}$ for the wavelengths between 4.1 and 5.4 µm. Yellow dotted lines with labels show the location of each of the absorption transitions that were used in this work. The stated gas conditions are representative of ICE exhaust.

Download Full Size | PDF

In Eq. (3), $M$ (g/mol) is the molecular weight of the absorbing species. In Eq. (4), ${\gamma _{i,o,j}}$ is the collisional-broadening coefficient of the $i$th collision partner at the reference temperature ${T_o}$ for transition $j$, and ${n_{i,j}}$ is the exponent for transition $j$ for the power law temperature scaling model. Here, the algorithm developed by McLean et al. [29] was used to model the Voigt profile.

LAS thermometry can be performed by measuring the integrated absorbance (${A_j}$), given by Eq. (5), of two separate absorption transitions of the same molecular species. A two-color ratio ($R$) of these integrated absorbances (i.e., areas) can be taken so that the pressure, mole fraction, and path length cancel out leaving only integrated areas and temperature-dependent linestrengths in the equation. As a result, the two-color ratio is a function of temperature and spectroscopic constants only as shown in Eq. (6). The gas temperature can be determined by comparing the measured ratio of integrated areas to the calculated ratio of linestrengths. Once the temperature is found, the mole fraction of the absorbing species can be determined using Eq. (5):

When performing LAS thermometry, it is imperative to use absorption transitions that yield a two-color ratio that is a strong function of temperature. The temperature sensitivity, $\sigma$, is a metric that can be used to help quantify how sensitive (i.e., precise and accurate) the temperature measurements from two given absorption transitions can be, and it is defined as the unit change in $R$ per unit change in temperature as shown in Eq. (7). Well-selected transition pairs will have a strong temperature dependence, which reduces the error in LAS temperature measurements induced by signal noise and error in the reference linestrengths:

The temperature sensitivity scales linearly with the difference in lower-state energy between the two transitions as can be seen in Eq. (7). This means that a large difference in $E^{\prime \prime}$ is necessary for accurate and precise LAS thermometry.

3. SENSOR DESIGN

A. Line Selection

Figure 1 illustrates transition linestrengths at 800 K for the fundamental vibration bands of CO, ${{\rm CO}_2}$, NO, and ${{\rm H}_2}{\rm O}$ as well as the location of the transitions utilized to measure temperature, CO, ${{\rm CO}_2}$, and NO in engine exhaust. Figure 2 shows simulated absorbance spectra for each line at gas conditions representative of the engine-out ICE exhaust. For CO, two QCLs were used to access the $P(0,30)$ and $P(0,20)$ lines near $2013\;{{\rm cm}^{- 1}}$ (4.96 µm) and $2060\;{{\rm cm}^{- 1}}$ (4.85 µm), respectively. For NO a single QCL was used to access the $R(0,15.5)$ transition near $1927\;{{\rm cm}^{- 1}}$ (5.19 µm). For ${{\rm CO}_2}$ a single ICL was used to access several absorption lines near $2395\;{{\rm cm}^{- 1}}$ (4.18 µm), specifically, the ${\nu _3}$ (${00^0}0$) $R(92)$, ${\nu _3}$ (${00^0}0$) $R(94)$, ${\nu _3}$ (${00^0}0$) $R(96)$, and ${\nu _3}$ (${00^0}0$) $R(98)$ transitions, where ${\nu _3}$ is used to indicate the fundamental asymmetric stretch bands where $\Delta {{ v}_3} = 1$. At higher temperatures, some higher-$J$ ${{\rm CO}_2}$ transitions are also visible in the measured absorbance spectra, although they are much weaker than the aforementioned lines. Table 1 shows the key spectroscopic parameters for each of the prominent absorption transitions.

 

Fig. 2. Simulated absorbance spectra for each species at conditions representative of ICE exhaust: (a) CO near $2013\;{{\rm cm}^{- 1}}$, (b) CO near $2060\;{{\rm cm}^{- 1}}$, (c) NO near $1927\;{{\rm cm}^{- 1}}$, and (d) ${{\rm CO}_2}$ near $2395\;{{\rm cm}^{- 1}}$. The temperature is 800 K, the pressure is 1 atm, and the path length is 6 cm for all plots, and the mole fraction of the relevant species is listed within each subplot.

Download Full Size | PDF

Table 1. Spectroscopic Parameters of the Absorption Transitions for CO, NO, and ${{\rm CO}_2}$  Used in This Work^a

View Table

These absorption transitions were chosen for three main reasons. (1) The linestrength of each transition is large (${\gt}0.1\;{{\rm cm}^{- 2}}\,{{\rm atm}^{- 1}}$) at temperatures representative of the engine exhaust as shown in Fig. 3. Large linestrengths are important because they directly translate to larger absorbance, which results in increased accuracy and reduced sensitivity to measurement noise. (2) All of the transitions selected for this work are isolated from other MIR-absorbing species such as water. This is most critical for the NO transition as discussed by Chao et al. [31]. (3) The $P(0,30)$ and $P(0,20)$ absorption transitions were used for thermometry because (i) their difference in lower-state energy is sufficiently large ($\approx 1000\;{{\rm cm}^{- 1}}$) for high-temperature sensitivity (${\gt}1$) at the temperatures of interest (see Fig. 4), and (ii) their linestrengths are large at all temperatures relevant to this work. This approach was also taken in our recent work [28]. It is important to note that in this work the gas temperature is often too low (${\lt}\approx 650\;{\rm K}$) (e.g., during engine crank and start up) to enable high-accuracy and -precision single-QCL thermometry via adjacent CO transitions in the ground and first-excited vibrational states, for example, using such transitions near $2060\;{{\rm cm}^{- 1}}$ as done by several researchers in other applications [27,32].

 

Fig. 3. Linestrengths as a function of temperature for each of the dominant CO, NO, and ${{\rm CO}_2}$ transitions used in this work.

Download Full Size | PDF

 

Fig. 4. (a) Ratio of linestrengths for the P(0,20) and P(0,30) transitions for CO. (b) Corresponding temperature sensitivity for the two CO absorption transitions at temperatures relevant to this work.

Download Full Size | PDF

Unfortunately, the ${{\rm CO}_2}$ transitions are not ideal for ${{\rm CO}_2}$ mole fraction measurements. This is because their high lower-state energies (near $3500\;{{\rm cm}^{- 1}}$) cause their linestrengths to be very sensitive to temperature. As a result, moderate errors in measured temperature correspond to relatively large errors in ${{\rm CO}_2}$ mole fraction. That being said, these transitions were chosen primarily to provide desirable absorbance levels ($0.1 \lt \alpha \lt 3$) in the ICE exhaust while still utilizing mid-infrared wavelengths. The latter facilitated combination with the QCLs used for CO and NO sensing. Utilizing ${{\rm CO}_2}$ transitions in the fundamental ${\nu _3}$ absorption band with more desirable lower-state energy would correspond to prohibitively large linestrengths (see Fig. 1) and absorbance signals. In future work, the authors would consider utilizing a diode laser near 2.8 µm to measure ${{\rm CO}_2}$ via weaker combination band transitions; however, this was not an option for this project. As a result, the ICL was scanned across four strong ${{\rm CO}_2}$ transitions to (i) help reduce measurement error and (ii) enable ${{\rm CO}_2}$-based thermometry to mitigate potential species-specific line-of-sight biases induced by, for example, imperfect thermal and chemical mixing in the exhaust flow. At some engine operating conditions, this proved to be important, and ${{\rm CO}_2}$ mole fraction measurements had to be extracted using a ${{\rm CO}_2}$-specific temperature. While the span in lower-state energy of the four dominant ${{\rm CO}_2}$ transitions is not sufficient to yield high-temperature sensitivity in the context of a two-color thermometer, it in combination with small absorbance contributions from underlying higher-${J}^{\prime \prime}$ transitions enabled accurate and precise ${{\rm CO}_2}$-based temperature measurements as discussed in Section 5.

B. Experimental Setup

The hardware used by the LAS sensor consisted of three main components: (1) the lasers and their driving electronics, (2) the optics needed to fiber couple the lasers and combine four colors onto a single beam path, and (3) the equipment used to provide optical access in the vehicle’s exhaust manifold.

Three QCL’s (Alpes Lasers) and one ICL (Nanoplus) were used. For simplicity, the lasers will be referred to as QCL 1, QCL 2, QCL 3, and the ICL. The absorption transitions they targeted are as follows: QCL 1 scanned across the $P(0,30)$ transition of CO near $2013.35\;{{\rm cm}^{- 1}}$, QCL 2 scanned across the $P(0,20)$ transition of  CO near $2059.91\;{{\rm cm}^{- 1}}$, QCL 3 scanned across the $R(0,15.5)$ transition of NO near $1927.27\;{{\rm cm}^{- 1}}$, and the ICL scanned across the ${{\rm CO}_2}$ transitions near $2395\;{{\rm cm}^{- 1}}$. The QCLs emit beams of highly collimated light with beam diameters of $\approx 1.5\;{\rm mm}$ and output powers of 50 to 100 mW. The ICL produced a larger beam with a diameter of $\approx 3\;{\rm mm}$ and a lower output power of approximately 10 mW.

All four lasers were operated in a time-multiplexing mode, which is a method where multiple laser signals are transmitted into one data stream by separating the signals in time. This was accomplished by scanning all four lasers in sequential order (i.e., one laser is on while the others are off) and repeating the pattern in time. Figure 5 shows an example measurement of raw detector signal during an engine test that illustrates how the lasers were time-multiplexed. The four-color scan rate (i.e., the four-color measurement rate) in this work was 5 kHz, meaning that all four lasers were fired in a time period of 1/5000 s (200 µs) for a single scan. It should be noted that when scanning at 5 kHz in a four-laser time-multiplexed setup, the effective scan rate of each laser is four times faster (i.e., 20 kHz). This is important because it has implications for the required frequency tuning, detector bandwidth, and sample rate. This 5 kHz scan rate was fast relative to exhaust gas transients produced by the engine (e.g., the time between crank shaft revolutions at 2000 RPM is 30 ms compared to the time between LAS scans of 200 µs) so the gas conditions were considered to be frozen during a given scan of all four lasers.

 

Fig. 5. Example measurement of raw detector signal illustrating one scan for each laser in a time-multiplexed configuration. The waveforms from left to right are for QCL 1, QCL 2, QCL 3, and the ICL.

Download Full Size | PDF

The four lasers were each sent a sawtooth waveform via two dual-channel arbitrary function generators (AFG) (Tektronix AFG-3252). The first AFG controlled the first two signals and the second AFG controlled the third and fourth signals, which are described as follows: (1) a 5 kHz, 100–800 mVppk waveform was sent to the controller (Wavelength Electronics) of QCL 1, which had temperature and current set points of ${-}{0.416^ \circ}{\rm C}$ and 88 mA. (2) A 5 kHz, 100–790 mVppk waveform was sent to the controller (Arroyo 6310) of QCL 2 that had temperature and current set points of 11.65°C and 100 mA. (3) A 5 kHz, 100–900 mVppk waveform was sent to the controller (Arroyo 6310) of QCL 3 that had temperature and current set points of 4.31°C and 112 mA. (4) A 5 kHz, 100–1200 mVppk waveform was sent to the controller (Arroyo 6305) of the ICL, which had temperature and current set points of 8.3°C and 19 mA. The transfer function of the QCL controllers was 100 mA/V, and the transfer function of the ICL was 50 mA/V. The wavelength scanning of each laser induced by these waveforms was characterized via a $3^{\prime \prime}$ long solid germanium etalon with a free spectral range of $0.0163\;{{\rm cm}^{- 1}}$.

Figure 6 shows schematics of the fiber coupling setup along with all the equipment used to operate the lasers and achieve optical access in the engine exhaust pipe. All three QCLs were placed on an $18^{\prime \prime}\times 18^{\prime \prime}$ aluminum breadboard (Thorlabs MB18) and fiber coupled into a 2 m long single-mode indium trifluoride (${{\rm InF}_3}$) patch cable (Thorlabs P3-32 F-FC-2) that was connected to a five-axis fiber port with a lens coated for a wavelength range of 2–5 µm (Thorlabs PAF2-4 E). This was accomplished by aligning all three beams to be colinear using gold mirrors (Thorlabs PF10-03-M01) that were placed in kinematic mounts (Thorlabs KM100) and two 50:50 calcium fluoride (${{\rm CaF}_2}$) beamsplitters with 2–8 µm coatings (Thorlabs BSW510). The ICL was placed on a separate $12^{\prime \prime}\times 12^{\prime \prime}$ breadboard (Thorlabs MB12) and fiber coupled using another set of the same components (fiber, fiber port, mirrors, and mounts). This was done to avoid using additional beamsplitters in the ICL beam path due to its lower output power compared to the QCLs and to avoid the complications associated with coupling four laser beams into one single-mode fiber.

 

Fig. 6. Schematic of the experimental setup with the electronic equipment used on the left, the fiber coupling setup in the middle, and the integration of the optics into the vehicle exhaust pipe on the right.

Download Full Size | PDF

The beam paths from each fiber were made colinear via another beamsplitter and then directed through the engine exhaust using additional hardware as shown in Fig. 6. Due to space constraints on the underside of the vehicle, cage-cube-mounted right-angled prism mirrors (Thorlabs CCM5-M01) were used to turn the beam 90° to enable the pitch and catch hardware of the sensor to protrude downwards, rather than horizontally into the vehicle frame. Each fiber optic cable was connected to a fiber collimator (Thorlabs F028APC-4950) that was placed in a kinematic mount (Thorlabs KC05-T). The two beams were made colinear via a beamsplitter. The light then reflected off the first prism mirror and passed through a custom window holder that held a ${1/2}^{\prime \prime}$ diameter wedged sapphire window (Thorlabs WW30530) positioned between two graphite gaskets that provided a gas seal while a custom retaining ring held the windows in place. Following the first right-angled mirror, the light traversed the 6 cm ID exhaust pipe and exited through a second window and reflected off another prism mirror. The beam then passed through an iris (Thorlabs SM1D12D) and focused through a ${1}^{\prime \prime}$ diameter ${{\rm CaF}_2}$ plano–convex lens with a 40 mm focal length (Thorlabs LA5370-E) that was mounted in an ${X} \text{-} { Y}$ translating lens mount (Thorlabs CXY1). The light was then passed through a longwave-pass filter (Spectrogon LP-3750) and finally into a mercury–cadmium telluride (MCT) photovoltaic detector (Vigo PVI-4TE-10.6) with a 10 MHz bandwidth.

C. Optical Challenges

Several challenges had to be overcome in the optical design. First, relatively strong reflections off the ${{\rm InF}_3}$ fiber tip introduced pronounced optical noise and wavelength jitter. While optical isolators can be used to mitigate this problem, they are expensive for MIR wavelengths and only recently became commercially available. Instead, thin LEXAN sheets were placed at various angles in the beam path to attenuate and redirect the back-propagating beam away from the laser cavity. This reduced the intensity noise and wavelength jitter to manageable levels, but they remained one of the most significant components of signal noise. Figure 7 illustrates an example of the optical noise for the ICL with varying amounts of LEXAN sheets in the beam path with a scan rate of 500 Hz and a scan amplitude of $0.6\;{{\rm cm}^{- 1}}$. The results indicate how increasing the number of LEXAN sheets reduces the optical noise.

 

Fig. 7. Example of the optical noise caused by back reflections off the fiber optic tip entering the cavity of the ICL. Each plot corresponds to differing amounts of thin LEXAN sheets that were placed in the beam path.

Download Full Size | PDF

Second, it was crucial to strategically balance transmission losses induced by fiber coupling and beam splitting. This consideration was driven by the need to ensure substantial and relatively uniform optical power for all four lasers reaching the detector, as reduced transmission could potentially lead to a lower signal-to-noise ratio (SNR). Achieving this balance among various factors necessitated careful placement of lasers within the optical setup. The three QCLs were positioned on a single breadboard featuring multiple beamsplitters. The ICL was located on a separate breadboard for fiber coupling, without any beamsplitters, due to its lower power output. Specifically, the ICL produced a maximum power output of approximately 10 mW with an optical fiber transmission rate of roughly 97.5% per meter. Among the QCLs, QCL 3 was configured to pass through only one beamsplitter, while QCLs 1 and 2 each passed through two beamsplitters. This configuration was chosen due to QCL 3’s observed weakest transmission through the optical fiber. QCLs 1 and 2 near 4.9 µm and QCL 3 near 5.2 µm exhibited varying power outputs, between 50 and 100 mW, and lower transmission rates of approximately 87.5% per meter and 75% per meter, respectively. These transmission rates align with the data sheets for ${{\rm InF}_3}$ fiber optic cables available on the Thorlabs website.

As an illustrative example of the laser intensities reaching the detector, let us assume that each QCL outputs exactly 50 mW of power, the ICL outputs exactly 10 mW, a fiber-coupling efficiency of 40% is achieved for each laser, each beamsplitter is a true 50/50 splitter, and the transmission of light from each laser through the fiber optic cable is 87.5% per meter for QCLs 1 and 2, 75% per meter for QCL 3, and 97.5% per meter for the ICL. Under these conditions and without accounting for other losses (such as transmission through sapphire windows), the laser power reaching the detector would be 1.9 mW for QCLs 1 and 2, 2.8 mW for QCL 3, and 1.9 mW for the ICL.

Third, during a previous test campaign [28], significant transmission losses were observed due to water condensation on the windows during engine cold start tests. To address this issue, heating tape was applied to the window holders to warm them to approximately 110°C before each test, effectively preventing water condensation on the windows. No significant transmission losses occurred after implementing the heating tape.

Fourth, identifying an appropriate optical filter for the detector was somewhat challenging. Ideally, a bandpass filter spanning only the laser wavelengths (4.2 to 5.2 µm) would be used; however, it was challenging to find a bandpass filter that transmitted this wavelength range while effectively blocking wavelengths outside this range. This was problematic since background emission from hot exhaust pipes as well as ${{\rm H}_2}{\rm O}$ and ${{\rm CO}_2}$ in the exhaust gas could saturate the detector. To mitigate this, an iris, along with a longwave-pass filter cutting on at 3.75 µm, was employed. Fortunately, the background emission in engine tests proved to be relatively minor, thereby enabling successful measurements with this filter.

4. DATA PROCESSING

Measurements of temperature, CO, NO, and ${{\rm CO}_2}$ mole fraction were obtained from best-fit parameters provided by a least-squares spectral-fitting routine employing MATLAB’s lsqcurvefit and absorption spectroscopy models developed in our lab. Two fitting methods were used and are discussed in the next paragraph: (1) a conventional independent-line model for CO and NO measurements and (2) a multi-line database-constrained model for ${{\rm CO}_2}$, hereafter referred to as the HITEMP model.

The independent-line model involved floating a linecenter frequency (${\nu _o}$), an integrated absorbance ($A$), and a collisional full-width at half-maximum ($\Delta {\nu _c}$) for each individual absorption transition, and it used these parameters to calculate absorbance profiles for each line. The CO absorbance spectra from QCL 1 and QCL 2 were fit simultaneously in the same routine for each scan, which was done to enable their collisional widths to be linked. Individual linecenters and integrated absorbances were fit for each line but only the collisional width of the $P(0,20)$ line ($\Delta {\nu _{c,2}}$) was floated. The collisional width of the $P(0,30)$ line was constrained to be a scalar multiple of $\Delta {\nu _{c,2}}$. This scalar multiple was found by employing collisional-broadening parameters from HITRAN 2020 [30] for the two investigated CO transitions. This included calculating the ratio $\Delta {\nu _{c,1}}/\Delta {\nu _{c,2}}$ within a temperature range representative of engine exhaust conditions, where this ratio was found to remain approximately constant. The NO and ${{\rm CO}_2}$ spectra were also fit using the independent-line model. It should be noted that for simplicity and convenience, only the two strongest ${{\rm CO}_2}$ transitions were included in the routine for the gas-cell data since fitting the extra smaller lines did not yield any benefit to the ${{\rm CO}_2}$ measurements. Baseline errors were corrected for by multiplying the estimated baseline by a polynomial (linear or cubic), where the polynomial coefficients were free parameters in the fitting routine. This approach is analogous to superimposing a baseline correction on the simulated absorbance spectra. Once the fitting routine for a given scan was completed, temperature was calculated via Eq. (6). The mole fractions of CO, NO, and ${{\rm CO}_2}$ were then calculated via Eq. (5), and the gas pressure was measured by a pressure transducer.

 

Fig. 8. Example measurements of absorbance spectra and corresponding best-fit spectra and peak-normalized residuals for each laser. The measurements were acquired in the heated gas cell at 745 K and 1 atm with a path length of 9.4 cm, a CO mole fraction of 1.8%, a ${{\rm CO}_2}$ mole fraction of 2.0%, and a NO mole fraction of 4.3%. The best-fit spectra were produced by the independent-line model.

Download Full Size | PDF

The HITEMP model works by performing a line-by-line simulation of ${{\rm CO}_2}$ absorbance spectra using parameters from the HITEMP 2010 database [33] while floating temperature, mole fraction, linecenter shift parameters for each prominent transition, a collisional width scaling factor, and polynomial coefficients for baseline correction in order to obtain a best-fit spectrum. This approach is typically used for LAS with broadband light sources such as what was done by Tancin et al. [34]. This is because broadband LAS measurements have many absorption transitions in the measured absorbance spectra and there would be too many free parameters for a least-squares fitting routine to model each line independently. The HITEMP model used here calculates the simulated absorbance spectra similar to the description provided by Goldenstein et al. [35]. As a result, this model also accounts for weaker transitions in the absorbance spectrum that are too weak to be easily seen (e.g., high-J transitions that have wrapped around the nearby bandhead of the $\Delta {{v}_3} = 1$ fundamental band of ${{\rm CO}_2}$) or modeled well using the independent-line model. As a result, the HITEMP model can provide better quality fits and improved temperature measurements from the measured ${{\rm CO}_2}$ absorbance spectra. The HITEMP model was used to process absorbance data acquired in ICE exhaust as described in Sections 3.A and 5.B.

5. EXPERIMENTAL RESULTS

A. Gas-Cell Experiments

The accuracy of the fiber-coupled LAS sensor was validated with measurements of temperature, CO, NO, and ${{\rm CO}_2}$ mole fractions in a heated static-gas cell at various temperatures and pressures. The gas cell is described in detail by Schwarm et al. [36], and as a result, only the most critical details are repeated here. The gas cell can be heated to a maximum temperature of 1200 K and can be pressurized up to 200 atm at temperatures below 1000 K. ${{\rm CaF}_2}$ rods inside the cell are used to bypass the thermally non-uniform zone and provide a 9.4 cm long thermally uniform zone for LAS measurements. Three thermocouples were placed in distinct locations in the gas cell to determine the temperature. In this work, two different gas mixtures were used. The first mixture consisted of 1.833% CO and 2.007% ${{\rm CO}_2}$ with a balance of nitrogen. This mixture was used for validation of CO-based thermometry and measurements of CO and ${{\rm CO}_2}$ mole fraction. The second mixture consisted of 4.33% NO with a balance of argon, and it was used to evaluate the accuracy of the NO mole fraction diagnostic. Both mixtures had a manufacturer stated uncertainty of ${\pm}2\%$ for each constituent.

Figure 8 displays the measured absorbance profiles, best-fit spectra, and corresponding residuals for each laser, showcasing the high SNR and small residuals. Figure 9 shows a brief time history of LAS measurements using the CO-CO₂-N₂ mixture as well as the NO-Ar mixture at 745 K and 1 atm. The LAS temperature measurement exhibited a time-averaged value of 743 K (0.27% error) with a standard deviation of 5 K. The average CO mole fraction was 1.84% (0.38% error) with a standard deviation of 0.02%. The ${{\rm CO}_2}$ mole fraction had a mean of 2.09% (4.14% error) with a standard deviation of 0.08%. The time-averaged mole fraction of NO was 4.35% (0.46% error) with a standard deviation of 0.08%.

 

Fig. 9. Measured time histories of temperature and mole fractions of CO, ${{\rm CO}_2}$, and NO acquired in the heated gas cell. A comparison to known values is also shown. The pressure was 1 atm and the temperature was 745 K.

Download Full Size | PDF

 

Fig. 10. Summary of LAS measurement accuracy for gas-cell experiments conducted at 1 or 2 atm and temperatures from approximately 300 to 1000 K. The LAS measurements of temperature and mole fractions of CO, ${{\rm CO}_2}$, and NO are accurate to within 3.1%, 4.8%, 5.1%, and 4.6%, respectively. The error bars represent the uncertainty in the LAS measurements and were calculated from the 95% confidence intervals corresponding to the best-fit spectra and uncertainties in linestrengths (taken from the HITRAN database or the literature), pressure, and path length. The individual sources of uncertainty were added in quadrature to determine the reported measurement uncertainty.

Download Full Size | PDF

Validation tests, lasting 50 ms each, were conducted in the heated gas cell at temperatures from 298 to 1000 K. Tests were conducted at pressures of 1 and 2 atm, which are representative of the pressure range encountered in the vehicle exhaust. Figure 10 shows a summary of the LAS sensor’s accuracy at various pressures and temperatures. Across all the tests, the CO, ${{\rm CO}_2}$ and NO mole fraction measurements agreed within 4.8%, 5.1%, and 4.6% of known gas bottle values and the temperature measurements were within 3.1% of the average thermocouple measurement. The LAS sensor exhibited detection limits (SNR of one) of 250 ppm for CO, 1000 ppm for ${{\rm CO}_2}$, and 200 ppm for NO for a 5 kHz measurement rate and path length of 9.4 cm at gas-cell conditions. The SNR was determined by comparing the measured peak absorbance at line center with background noise. The background noise is primarily electronic bit noise from the DAQ for QCL 1, QCL 2, and the ICL while there was also electromagnetic interference (EMI) noise present in QCL 3’s waveform. This background noise was quantified through the calculation of the standard deviation across frequencies for a given scan within non-absorbing regions of the absorbance profile. We expect that significantly lower detection limits could be achieved upon reducing fiber-coupling induced noise in future work.

B. Demonstration in Engine Exhaust

The LAS sensor was demonstrated by acquiring measurements in the engine-out exhaust of a light-duty pickup truck with an 8-cylinder gasoline ICE. Figure 11 shows examples of measured absorbance profiles and their corresponding best-fit spectra for a single scan of each laser acquired in the vehicle exhaust. Each absorbance spectrum has a sufficiently high SNR and peak-normalized residuals below 5% with the exception of QCL 3 for NO between 1927.3 and $1927.4\;{{\rm cm}^{- 1}}$. This is due to EMI noise present in the scan of QCL 3, which appears as a wave-like distortion in the raw data and absorbance profile. This noise appeared in the engine test bay and could not be removed due to time constraints.

 

Fig. 11. Example measurements of absorbance spectra and corresponding best-fit spectra and peak-normalized residuals for each laser. The measurements were acquired in ICE exhaust at the following conditions: CO temperature of 1247 K, pressure of 1.1 atm, CO mole fraction of 0.78% (7800 ppm), NO mole fraction of 0.15% (1500 ppm), ${{\rm CO}_2}$ temperature of 1266 K, and a ${{\rm CO}_2}$ mole fraction of 13.1%.

Download Full Size | PDF

Figure 12 shows temperature and mole fraction time histories for the first few seconds of a cold-start test (i.e., all vehicle components and fluids are below the engine operating temperature, usually at ambient temperatures) with the engine idling under 5% rich conditions. All four quantities move towards a quasi-steady value within the first second with the CO mole fraction displaying the most transient behavior at the beginning of the test. Figure 12 clearly illustrates the transients in temperature and species concentration that are induced by individual engine cylinder firing and exhaust events. Similar dynamics were shown in our prior work reporting temperature and CO measurements in ICE exhaust [28]. This is most evident in the temperature time history where the gas temperature repeatedly spikes and decays (see zoom view in Fig. 12). The larger amplitude spikes in temperature come from the cylinders that exhaust closer to the measurement location and the smaller spikes come from cylinders located further from the measurement location. This could be due to enthalpy losses along the exhaust manifold or non-uniform mixing. With each large spike in temperature, there is a corresponding spike and decay in NO mole fraction. This suggests that higher levels of NO exist in hotter portions of exhaust gas, which may result from the fact that NO formation occurs more rapidly at higher temperatures due to the well-known Zeldovich mechanism. The maximum amount of NO measured was approximately 0.5% (5000 ppm). In general, the CO mole fraction decreases and the ${{\rm CO}_2}$ mole fraction increases when there is a large increase in temperature. This is also understandable since a higher ${{\rm CO}_2}$/CO ratio corresponds to greater combustion efficiency and higher temperatures with all else equal.

 

Fig. 12. Measured time histories of temperature and species mole fractions in ICE exhaust for the first three seconds of a cold-start test with the vehicle running at a 5% fuel rich calibration. A zoom view from 1.1 to 1.4 s is shown to highlight the LAS measurement precision and transients in exhaust gas conditions.

Download Full Size | PDF

The exhaust gas temperature inferred from ${{\rm CO}_2}$ using the HITEMP model is, on average, approximately 160 K higher than the temperature inferred from CO via the independent-line model; however, the temporal dynamics are nearly identical for both temperature diagnostics as shown in Fig. 12. It is important to note that for the first 0.625 s of data, the ${{\rm CO}_2}$ temperature and mole fraction time histories were obtained from best-fit spectra fit to only the strongest three ${{\rm CO}_2}$ transitions. The fourth transition (closest to ${{\rm CO}_2}$ bandhead) was ignored due to its lower SNR, which was problematic early in the test. There are three potential explanations for the observed temperature difference between the CO and ${{\rm CO}_2}$ diagnostics. First, the large lower-state energies for the ${{\rm CO}_2}$ transitions will desensitize the ${{\rm CO}_2}$-based measurements to gas in the colder boundary layer (see [1] and references therein). This effect will be more pronounced for ${{\rm CO}_2}$ than CO (see Table 1), thereby causing the line-of-sight temperature of ${{\rm CO}_2}$ to be higher than that of CO. This effect is most certainly at least partially responsible for the observed results. Second, it is also possible that the exhaust gases are not uniformly mixed and that the concentration of ${{\rm CO}_2}$ is higher in hotter pockets of gas or that CO is biased towards colder pockets of gas. Either of these scenarios would also cause the path-integrated temperature of ${{\rm CO}_2}$ to be higher than that of CO. Lastly, a third explanation is that the beam paths for CO and ${{\rm CO}_2}$ measurements are not perfectly colinear and are passing through regions of gas with different temperatures; however, this seems less likely given (i) the magnitude of the temperature difference, (ii) expected degree of thermal and chemical mixing, and (iii) the care taken in aligning the laser beams.

The estimated uncertainties in the mole fractions of CO, ${{\rm CO}_2}$, and NO are 2.9%, 8.3%, and 10.6%, respectively. Similarly, the estimated uncertainties in the temperatures of CO and ${{\rm CO}_2}$ are 1.5% and 7.8%, respectively. The uncertainty in the LAS measurements was found using the Taylor series method for uncertainty propagation, where contributing uncertainties are added in quadrature. A more rigorous explanation of this general approach can be found in [8, 28].

The principal sources of uncertainty for ${{\rm CO}_2}$ temperature and mole fraction measurements are derived from the best-fit parameters and the uncertainties associated with the linestrength of each parameter provided by the HITEMP database. Among these factors, the uncertainty in the best-fit spectra contributes the largest amount of uncertainty. Since more than two ${{\rm CO}_2}$ absorption lines were measured, the impact of the linestrength uncertainty was found via a Monte Carlo simulation where each linestrength was randomly perturbed by a scaling factor that falls within the uncertainty bounds provided by HITEMP. The simulated absorbance spectrum was then reproduced, and the impact of the linestrength perturbation on the temperature and ${{\rm CO}_2}$ mole fraction was determined. The standard deviation in temperature and mole fraction were then determined for a total of 1000 random perturbations on each linestrength value, and these values were treated as the uncertainty induced by the uncertainty in transition linestrengths.

In the case of the CO diagnostic, temperature uncertainty arises from uncertainty in the best-fit spectra, specifically the corresponding best-fit values of integrated absorbance, which are utilized in temperature calculations. Additionally, linestrength uncertainties also contribute to temperature uncertainty; however, this was small compared to the uncertainty from the best-fit spectra. As for CO mole fraction, uncertainties originate from the best-fit spectra, linestrength, pressure, and path length, with the dominant contributor originating from the best-fit spectra. The effect of temperature uncertainty on the CO mole fraction calculation was accounted for in the linestrength uncertainty calculation as detailed in [28].

In the case of NO mole fraction, the same sources of uncertainty apply as for CO, with an additional contribution arising from the unknown species-specific temperature. To estimate the latter, the mole fraction of NO was calculated using the temperature from CO and the temperature from ${{\rm CO}_2}$. The difference in NO mole fraction between these two calculations was assumed to be the uncertainty induced by the unknown species-specific path-integrated temperature, which was the leading contributor to the overall uncertainty in NO mole fraction. It should be noted that the reference linestrength uncertainty for NO was taken from Chao et al. [31].

6. CONCLUSION

A fiber-coupled, mid-infrared four-color LAS sensor was developed for simultaneous measurements of temperature, CO, NO, and ${{\rm CO}_2}$ mole fractions at 5 kHz in harsh engine environments. The lasers were fiber-coupled into two single-mode fibers and then combined onto a single beam path to overcome spatial constraints that limited optical access around the vehicle’s exhaust manifold. The LAS sensor provided temperature and species measurements within 3% and 5% of known values in heated gas-cell experiments at temperatures from 298 to 1000 K and pressures from 1 to 2 atm. The sensor was then demonstrated in the field by acquiring measurements in the engine-out exhaust of a light duty truck with an 8-cylinder gasoline engine. The high-precision LAS measurements clearly revealed how the exhaust gas temperature and composition evolve due to individual cylinder firing events and on longer time scales. As a result, this work demonstrates the potential of multi-species MIR LAS measurements for studying a wide range of transient engine behavior in production vehicles and their impact on engine-out emissions.