high-speed laser-induced fluorescence and spark plug absorption sensor diagnostics for mixing and...

High-speed laser-induced fluorescence and spark plugabsorption sensor diagnostics for mixing and

combustion studies in engines

Michael Cundy,1,* Torsten Schucht,2 Olaf Thiele,2 and Volker Sick1

1Department of Mechanical Engineering, University of Michigan, W. E. Lay Automotive Laboratory,1231 Beal Avenue, Ann Arbor, Michigan 48109-2133, USA

2LaVision GmbH, Anna-Vandenhoeck-Ring 19, D-37081 Göttingen, Germany

*Corresponding author: [email protected]

Received 31 July 2008; revised 14 November 2008; accepted 19 November 2008;posted 26 November 2008 (Doc. ID 99673); published 22 December 2008

Simultaneous high-speed in-cylinder measurements of laser-induced fluorescence of biacetyl as a fueltracer and mid-infrared broadband absorption of fuel and combustion products (water and carbon diox-ide) using a spark plug probe are compared in an optical engine. The study addresses uncertainties andthe applicability of absorption measurements at a location slightly offset to the spark plug when infor-mation about mixing at the spark plug is desired. Absorbance profiles reflect important engine operationevents, such as valve opening and closing, mixing, combustion, and outgassing from crevices. © 2008Optical Society of America

OCIS codes: 060.2370, 120.1740, 110.6915, 300.1030, 300.2530.

1. Introduction

Spark ignition direct injection (SIDI) stratifiedcharge engines offer significant fuel economy im-provements when compared to conventional port fuelinjected engines [1]. At full load, these engines areoperated with a homogeneous charge like port fuelinjected engines, but since liquid fuel is injected di-rectly into the cylinder its evaporation leads tocharge cooling. This allows higher loads and com-pression ratios to be used, which results in thermalefficiency gains. The brake-specific fuel consumptionof these engines can approach that of diesel engines,without the associated noise or extra weight. Load iscontrolled by varying the amount of fuel injected, asopposed to varying the amount of fuel and throttlingthe intake air to create a homogeneous, flammablemixture. This requires that the mixture is suitablystratified to ensure ignition and complete combus-tion for low to mid load engine operation. A locally

rich region around the spark plug is desirable atthe time of ignition to facilitate ignition and to pro-mote early flame propagation because of its higherlaminar flame speed. This is critical when operatingan engine with overall high excess air or exhaust gasrecirculation levels, which are approaches to mini-mize emissions. One particular engine concept thatutilizes an inhomogeneous (stratified) mixture thathas attracted considerable interest recently is thespray-guided (SG) SIDI stratified charge gasolineengine.

In general, proper mixture preparation in internalcombustion engines is essential to simultaneouslyachieve maximum engine performance and mini-mized emissions. Cycle-to-cycle variations in com-bustion efficiency including partial burns andmisfires are affected by ignition and early flame de-velopment near the spark plug, which are in turn lar-gely affected by the local fuel/air/exhaust gasrecirculation mixture at the spark plasma and inthe regions surrounding the flame kernel. The pro-cesses involved vary spatially and temporally fromone engine cycle to another. Studies of ignition and

0003-6935/09/040B94-11$15.00/0© 2009 Optical Society of America

B94 APPLIED OPTICS / Vol. 48, No. 4 / 1 February 2009

combustion under such conditions will require diag-nostic tools that can resolve relevant scales. In par-ticular, the rate at which spatially resolvedmeasurements of fuel could be performed was lim-ited until recently. This has motivated the develop-ment of high-speed laser-induced fluorescence(LIF) fuel imaging diagnostics [2] and absorption-based spark plug optical probes [3,4]. LIF measure-ments deliver images at adequately high rates butrequire the use of an optical engine. The imageseither exclude the third dimension or spatially aver-age over it in the case of volume illumination. Asidefrom observing the fuel plume development and eva-poration, statistical information can be extractedfrom quantitative LIF images using data analysiswindows, and these cannot always be selected inideal locations. Absorption-based spark plug sensorscan be used in any engine with minimal or no mod-ification and offer even higher measurement rates,but they lack spatial resolution. Especially underhighly stratified fuel/air mixing situations, such asin a SG SIDI engine, the fuel concentration at thespark gap and the absorption path (typically a fewmillimeters away from the gap) may be quite differ-ent. The present paper evaluates the use of an opticalspark plug probe in a SG SIDI engine. The probemeasures infrared absorption of hydrocarbons, car-bon dioxide, and water to derive fuel and productgas concentrations. Simultaneous measurementswith the spark plug sensor and high-speed LIF wereconducted in an optical SG SIDI engine at crank an-gle degree (CAD) resolution (3600Hz).Comparison of the results of the two techniques

allowed an evaluation of their precision when the en-gine was operated with a homogeneous fuel/airmixture. The higher signal-to-noise ratio of the ab-sorption measurements allowed an assessment ofthe precision limitation of the LIF measurements,which are noise limited due to limited laser pulse en-ergies at high repetition rates. LIF and absorbancedata that were acquired under highly stratified con-ditions were normalized with data that were takenunder homogeneous mixing conditions. This elimi-nated pressure, temperature, and number density ef-fects on the signals, thereby allowing spatial andtemporal analysis of the signals, which are measuredmillimeters from the spark gap in different locations.

2. Background

A. Planar Laser-Induced Fluorescence

Fuel imaging using LIF is accomplished by adding afluorescence tracer to a nonfluorescing base fuel. Avariety of tracer/fuel combinations and excitation-detection strategies are described in the literature[5]. For high-repetition-rate imaging only one tra-cer/fuel system (biacetyl/iso-octane) has been used[6]. Biacetyl has been selected to take advantage ofthe higher pulse energies that can be achieved at355nm compared to 266nm using diode-pumpedNd:YAG lasers. For the 100ns pulse duration and

a sheet with a cross section of 20mm × 0:5mm thatwas used in the present study the saturation pulseenergy would be ∼4:3 J. In this experiment, the max-imum pulse energies approach 1mJ so the LIF signalvaries linearly with laser energy according to Eq. (1):

SLIF ¼ cILaserðλÞV�Xtracerp

kT

�σðλ;p;T; yÞφðλ;p;T; yÞ;

ð1Þ

where c is the overall efficiency of the collection op-tics; ILaser is the laser energy density; V is the detec-tion volume; the quantity ðX tracerP

kT Þ represents thenumber density of the fluorescing species, deter-mined by the mole fraction of the species of interest,X tracer, the pressure, P, the Boltzmann constant k,and the temperature T; σ is the absorption cross sec-tion, f is the fluorescence quantum yield; and y is themixture composition.

The photophysics of biacetyl for engine-relevantpressure and temperature conditions have been stu-died by Wermuth and Sick [7], Guibert et al. [8], andSmith and Sick [9]. Early engine biacetyl LIF experi-ments reported signal decay that was suggested to becaused by chemical decomposition at ambient condi-tions, but a dedicated species stability study byZhang et al. [10] related that to sequential evapora-tion. Adequate chemical stability of biacetyl as afluorescence tracer for fuel under spark-ignition con-ditions was found in a numerical study by Sick andWestbrook [11].

Some of the images presented in this work wererecorded under late injected stratified charge condi-tions. In order to produce quantitative fuel concen-tration data the images have to be processed usingbackground subtraction, light sheet division, and fil-tering to reduce noise. Recall that the available pulseenergy was rather low and the fluorescence imageswill exhibit significant shot noise.

Background light, both 355nm scatter and fluores-cence from, e.g., surface deposits, is scattered bypassing through windows and by reflecting off of fea-tures in the combustion chamber. Scattered 355nmlight is effectively absorbed by a GG420 filter. Toeliminate remaining background signal from theimages of interest, background images were takenat each CAD when the engine was motored (no fuelinjection) for a statistically significant number of cy-cles (∼100). Images taken at a given CAD were aver-aged and then subtracted from the light sheet andlate injected images at the same CAD.

In order to generate quantitative data for strati-fied fuel images, reference LIF images can be usedthat were taken with a known fuel concentration, setby the fuel injection duration. This is accomplishedby capturing LIF images in the compression strokeafter a known amount of fuel is injected early inthe intake stroke, producing quasi-homogeneousmixtures that correspond to a defined fuel concentra-tion. These images then automatically include theeffects of temperature, pressure, number density,

1 February 2009 / Vol. 48, No. 4 / APPLIED OPTICS B95

and incongruities in the light sheet and detectionpath for each crank angle [9]. Light sheet imagesare taken in between late injected runs, so it is rea-sonable to assume that the laser sheet alignment andcondition of the windows are sufficiently stable. It isassumed that temperature and pressure are homoge-neous within the illuminated area. The division ofthe late injection images by the light sheet imagesresulted in quantitative fuel distribution maps.The final processing step is spatial filtering by

either a 3 × 3 averaging filter or an adjustable non-linear anisotropic diffusion filter [12]. The latter fil-ter is used to smooth out shot noise within flowstructures while preserving large gradients.The cumulative one standard deviation error of

17% for the LIF diagnostics is described in detailby Smith and Sick [9]. The use of higher laser powerin the present study and spatial averaging allowedan improved uncertainty.

B. Absorption Measurements

In contrast to tracer-based measurements via LIF,infrared absorption allows the direct measurementof fuel as well as an aggregate of combustion productgases; in the case of this study carbon dioxideand water.Molecular absorption is governed by the Beer–

Lambert law. The absorbance is defined in Eq. (2):

− ln�II0

�¼ σ · c · L; ð2Þ

where σ is the absorption coefficient, L is the absorp-tion path length, and c is the density of the absorbingspecies. For the work presented here the absorptionpath length is constant. The absorption coefficient isassumed independent of temperature and pressureas a first approximation, but fully quantitative con-firmation of this assumption is still pending, whilethe work presented here addresses this assumptionto some extent. Additionally, the absorption condi-tions have to be assumed constant along the pathlength L. Then, the absorbing molecule density canbe determined using Eq. (2).Spectrally highly resolved measurements, e.g.,

using diode lasers, usually require information onwavelength-dependent absorption cross sections thatare functions of temperature and pressure. The ap-proach taken here is based on broadband absorptionwhere the rotation structure of the absorption spec-trum is not resolved. A broadband light source isused, and detection covers a large portion of the en-tire fundamental vibrational band. Integration overthe entire rotational manifold is expected to mini-mize temperature sensitivity. However, vibrationalpopulation and line broadening effects due to pres-sure and temperature may still affect the absorptionsignal and eventually will need to be quantified. Thebroadband absorption strategy makes the measure-ment sensitive to cross talk with absorption by other

molecules, although in the present case only to asmall degree.

Absorption bands in the mid-IR are much strongerthan the overtone bands that are often used in tun-able diode laser absorption spectroscopy. A single-pass absorption setup with a path length of a fewmillimeters will then be sufficient to achieve a signif-icant absorption that can be measured with a highsignal-to-noise ratio. Broadband absorption mea-surements can be conducted with simple incandes-cent lamps such as tungsten halide mid-IR lamps.

Figure 1 shows the absorption spectrum for CH4(representing hydrocarbon fuels), CO2, and H2O cal-culated from the HITRAN database [13]. Note thatthe absorption bands of hydrocarbon fuels and wateroverlap slightly. Potential cross talk of water absorp-tion with fuel measurements may occur in caseswhere large amounts of water are present, althoughthe highest water absorption coefficient is approxi-mately two orders of magnitude lower in that spec-tral region. A HITRAN simulation for 600K (themaximum temperature during our experiments)shows a cross talk of no more than 4%.

A bandpass filter that coincides with the funda-mental vibrational–rotational absorption band of hy-drocarbons centered at 3:4 μm (0:297 μm width)isolates light that is absorbed by the fuel. Biacetylalso has C─H bonds and will absorb within thisband, although the band may be slightly shifted be-cause of the C═O bonds. The net effect of adding bia-cetyl to iso-octane is not characterized. In general,even if it may be a good assumption to take the ab-sorption coefficient as constant, which is checked aspart of this study, the absolute magnitude of the ab-sorption coefficient will depend on the molecularstructure and composition of the fuel. A calibrationmeasurement with a known fuel concentration isthus necessary. A bandpass filter centered at 2:7 μm(0:231 μmwidth) is used to measure the combined ab-sorption by CO2 and H2O. Note that at present, thetemperature and pressure dependence of the absorp-tion in this bandpass are not as well quantified asthat of the fuel.

The high intensity of the IR lamp combined withsensitive detection allows measurements with high

Fig. 1. Absorption spectrum of fuel, water, and carbon dioxide at296K The spectral simulations used the HITRAN database.


signal-to-noise ratios with potentially significantlylower noise than single-shot LIF imaging. Absorp-tion measurements can therefore be used here asa baseline to assess the precision of the high-repetition-rate LIF measurements.

3. Experimental Setup

As the goal of this work is an in-depth comparison ofhigh-speed LIF and absorption-based in-cylindermeasurements it is imperative that both measure-ments are conducted simultaneously. The presentedmeasurements are a combination of high-speed CADresolved planar LIF measurements recorded with ahigh-speed intensified camera and absorption mea-surements using an IR light spark plug sensor. Mea-surements were conducted in an optical single-cylinder engine with a four-valve head, 8-hole fuel in-jector, a transparent piston bowl, full quartz liner,and windows in the cylinder head. For the resultspresented below, the engine was run at 600 rpm withthe air intake temperature fixed at 45 °C. Moisture inthe air intake is removed using filters. The air intakepressure was set at either 35 or 95kPa. 10mg of fuelwere injected with an end of injection of 250° beforetop dead center (BTDC; CAD ¼ 0 at TDC) for thehomogeneous experiments, and 40°BTDC for thestratified charge experiment. For fired engine opera-tion, the spark timing was 30° BTDC (the otherswere motored). More details on the engine can befound in Table 1 and [14]. A schematic of the LIF di-agnostics setup can be found in Fig. 2.The signal-to-noise ratio of the LIF images is lim-

ited by the low pulse energy of high-repetition-ratelasers. For this experiment the combined beams oftwo frequency-tripled Nd:YAG lasers (QuantronixHawk II) emitting at 355nm were used to excitethe fuel tracer biacetyl (10% by vol.), which wasadded to the iso-octane base fuel. The beams weremade to overlap using a waveplate and a polarizingbeam splitter cube and were fired simultaneously.High transmission at 532nm=high reflectivity at355nmmirrors were used to both redirect the beamsand remove remnants of frequency-doubled light. Acylindrical lens (þ1000mm) was used to assist inthinning the light sheet. A second cylindrical lens(−100mm) was used to expand the beams in the ver-

tical direction. A sheet-forming telescope (Roden-stock) was then used to focus the sheet. The sheetentered the engine between the crankcase and com-bustion chamber, where it reflected off of a 45° mirrortowards the cylinder head. It passed through the pis-ton, which has a large axial window as well as a bowlin crown with side windows for viewing. This bowl iscommonplace in SG SIDI engines along with an ex-tended length spark plug to help confine the strati-fied fuel/air mixture around the spark plug and toinitiate combustion near the center of the combus-tion chamber.

After traveling through the piston, the light sheetimpinged on the cylinder head, extending from thefuel injector to slightly past the spark plug installa-tion hole, bisecting the spark plug electrodes and ab-sorption optical path. A schematic of the spark plugprobe can be found in Fig. 3, and a schematic of thecylinder head showing the light sheet position andfeatures within the combustion chamber can be seenin Fig. 4. Note that one of the eight fuel spray plumesis directed towards the spark plug. Also, note that thespark plug portion of the probe lies∼2mm in front of

Table 1. Optical Engine Parameters

Parameter Value

Bore 8:6 cmStroke 8:6 cmConnecting rod 15:92 cmHead volume 62:24 cm3

Displacement 499:92 cm3

Compression ratio 9∶1Intake valve open 362° BTDCIntake valve close 126° BTDCExhaust valve open 139° ATDCExhaust valve close 373° ATDCValve lift 10:3mm

Fig. 2. Schematic of the experimental setup. The cylinder outlinegeometry is based on a computational grid (courtesy of Brian Pe-terson). The labeled optical elements are A&F, dielectric mirrorHT532/HR355; B, half-wave plate; C, beam splitter cube; D, cylind-rical lens; E, cylindrical lens combination; G, high-speed image-in-tensified relay optics; H, GG420 color glass filter.

Fig. 3. Schematic of the spark plug probe. The left image is shownwith one channel for illustrative purposes, and the right imageshows the position of the absorption path relative to the spark plugelectrodes.


the light sheet and the absorption sensor portion liesbehind the sheet by about the same distance. Thespark plug probe package includes spacers of variousthicknesses but cannot in its current form extend farenough into this engine for stable stratified chargeoperation with the standard injector. The sensor plugis a few millimeters shorter than the standard sparkplug for this engine, and thus the fuel plume missesthe spark plug. This alleviated any concern that li-quid fuel accumulated in the absorption optical pathduring late injected runs.Images were captured with a high-speed compli-

mentary metal oxide semiconductor camera (VisionResearch, Phantom v7.3). Signals were digitized at14 bits and recorded until the 16GB of on-boardmemory are filled. The weak LIF signals are ampli-fied with a lens-coupled image intensifier (LaVision,IRO-HS). Synchronization of lasers, camera, and in-tensifier timing with the engine crank angle positionwas accomplished with a software-controlled high-speed timing unit [LaVision, High Speed Controller(HSC) and DaVis 7.2 software]. This also allowed ad-justment of camera parameters and the setting of agate on the camera and intensifier triggers to captureimages only during the portions of the engine cycle ofinterest, enabling recording of hundreds of consecu-tive cycles.The spark plug sensor system uses an incandes-

cent (tungsten) lamp that is modulated at 30kHzusing a chopper wheel. The modulation is requiredto record a background signal to account for back-ground signal light that originates from the thermalradiation of the heated gases and straylight. Themodulated light travels through a sapphire fiber tothe spark plug, enters the cylinder through a sap-phire window, is reflected by the mirror, and is trans-mitted back to the detection unit via a second fiber.The light is spectrally filtered and converted into anelectrical signal using a two-stage TE-cooled IRphotovoltaic detector (one for each channel). Theseanalog signals were acquired with a 14 bit analog–digital converter (National Instruments) at a rateof 2:5MHz. The chopper modulation reference signaland the crank angle encoder signals from the enginewere recorded at the same rate. All these data wereprocessed online in the sensor electronics to produceCAD resolved data and to split (demodulate) the

traces into “light on” and “light off” signals. The datawere transferred to a control computer via Ethernet,where three additional processing steps were per-formed to get the desired relative density traces.First, the background (light off) signal was sub-tracted from the light on signal, resulting in an ab-sorption trace. Then the light intensity I0 [see Eq. (1)]was determined dynamically for each individual cy-cle as the maximum recorded signal intensity in agiven range where the fuel and product gas concen-trations are at their lowest levels. The measurementof the reference light intensity through the entire ab-sorption path allows accounting for any fouling of themirrors or coupling lenses during engine operation.It also compensates for any potential intensity fluc-tuations of the incandescent lamp. In principle, I0should bemeasured in the absence of fuel (or exhaustfor that channel). Splitting off a fraction of the incom-ing light and sending it through the same spectralfilter bandpass as for the absorption detector wouldallow accounting for intensity fluctuations due to thelamp but not for window and mirror fouling. Thiswould also require additional detectors. The impacton the accuracy of the absorptionmeasurement whendetermining I0 from the highest measured transmis-sion could be assessed quantitatively. Approximately11% of the gas volume will remain in the cylinderafter the exhaust stroke. This amount is given bythe compression ratio of the engine (9∶1) and the ex-haust backpressure. Thus, at any time during the cy-cle fuel absorption will be measured unless fuelinjection is stopped for several cycles to purge the cy-linder. Back-to-back measurements with and with-out fuel injection were conducted and I0 wasdetermined for both cases. The difference in mea-sured absorbance, which is directly proportional toconcentration, is less than 2%. The absorbance mea-surement with I0 determined dynamically is lower bythis fraction.

Conventional engine parameters were recordedvia a high-speed data acquisition system (NationalInstruments Compact DAQ controlled by OptimumPower software, PTrac) that allows recording of 28channels at CAD resolution, digitized at 16 bits.The engine crank shaft position signals were firstsent to the engine controller, where they were distrib-uted to the spark plug sensor control and acquisitionelectronics and the data acquisition system men-tioned above. The spark plug sensor electronics thenrelayed the signals to the HSC, which controlled thetiming of the laser pulses, camera, and intensifier.All three systems were synchronized to record thesame engine cycles.

4. Results

Results are presented on the precision, accuracy, andresponse to large spatial gradients for both measure-ment techniques. Simultaneously acquired absorp-tion and LIF data are compared for this purposefor different mixing conditions. A discussion ofabsorption-based combustion product measurements

Fig. 4. Horizontal and vertical cross sections shown to illustratethe locations of the LIF and spark plug probe measurements rela-tive to the spark plug electrodes. The LIFmeasurements are takenin the hatched area in the right image.


that were carried out simultaneously with fuel mea-surements examines the use of broadband mid-IRabsorption for aggregate water and carbon dioxidemeasurements.

A. Homogeneous Charge Motored Measurements

The signal-to-noise ratio of the absorption measure-ments [4] is expected to be higher than that of theLIF measurements [9], allowing the absorption datato be used as a benchmark for the precision of the LIFdata. For this purpose, measurements in the motoredengine were conducted with early fuel injection(end of injection at 250°BTDC) to produce quasi-homogeneous mixtures around the typical time ofignition. Figure 5 shows the results of this experi-ment, which also serve as light sheet images to nor-malize late injection images. An 11 × 12 pixel dataanalysis window corresponding to approximately1:2 × 1:3mmwas selected in each LIF image, the spa-tial average within the window was calculated, andthe resulting values were averaged at their respec-tive CAD over 100 cycles. The location of the windowis indicated in the inset picture on the plot. The en-semble standard deviation was also calculated forboth the LIF and absorbance data and is indicatedon the plot as variation bars. The images do not in-clude any corrections (temperature, pressure, num-ber density, etc.), and only a spatial 3 × 3 slidingfilter has been applied. The spark plug probe mea-surements are presented in terms of absorbance[− lnðI=I0Þ], corrected for background light. TheLIF data axis was scaled arbitrarily to match the dis-play of the absorbance data at 39°BTDC.In the absence of any temperature and pressure

effects on the LIF and absorption signals the signalswould vary inversely proportional to the in-cylinder

volume once the intake valves are closed (126°BTDC). Thus, a comparison is presented in Fig. 5to a scaled ideal number density value that is peggedto the absorbance data at intake valve closing. It isnoticeable that the absorbance data closely follow theexpected number density, but starting at approxi-mately 20°BTDC the absorbance data fall belowthe expected values. The reason could either be a re-duction of the absorption coefficient as pressure andtemperature increase or mass loss by blowby acrossthe nonlubricated piston rings. The observed magni-tude of the signal reduction is commensurate withknown significant blowby mass loss (>10%) fromthe cylinder in nonlubricated optical engines [15].The substantial mass loss is also the reason whypeak signals for LIF and absorption are observed be-fore the piston reaches TDC.

The LIF signal strength does not follow the in-crease in number density, pointing to a pronouncedeffect of temperature and pressure on the signalstrength. The peak temperature during motored en-gine operation does not exceed ∼600K [15] at a peakpressure of 15 bars. There is no evidence that biacetylwould be chemically decomposed under the experi-mental conditions of this study [11]. LIF signalstrength for biacetyl after excitation at 355nm wasshown to drop with increasing pressure and tem-perature [9], and for the conditions shown in Fig. 5the observed signal evolution is in good agreementwith expectations. A full correction of the presentedLIF signals for temperature and pressure effectswould require accurate temperature data. For the gi-ven experiments, the engine exhibited unusuallyhigh blowby due to a slightly malfunctioning valve;this will cause a higher uncertainty in temperatureanalysis based on measured in-cylinder pressuretraces. Therefore, only noncorrected LIF data forthe homogeneous engine runs are presented.

Fig. 5. Comparison of the spark plug probe absorbance measure-ments with the LIF signals in the homogeneous charge motoredexperiment. The LIF data are spatially averaged within the rec-tangle shown (132 pixels) and then averaged at each CAD over100 cycles. The data contain one standard deviation variation bars.The data reflect the change number density as the chamber vo-lume changes when the piston moves during the engine cycle. Thischange in the absence of engine blowby losses is shown as a solidcurve.

Fig. 6. Individual data from simultaneous LIF and absorptionmeasurements for 100 cycles measured during the homogeneouscharge motored experiment. The average and standard deviationof the same data are shown in Fig. 5 as a function of crank angleand illustrate the change in slope observed here as a pressure andtemperature influence primarily on the LIF signal strength.


The correlation of LIF and absorbance data areshown in Fig. 6 where individual data points from100 cycles from 39°BTDC to 70°ATDC are plotted.The nonlinearity of this correlation reflects a differ-ence in the temperature and pressure dependence ofthe signals. Recall that the LIF data have not beencalibrated and have not been processed for tempera-ture and pressure. A LIF signal decrease withincreasing temperature and pressure is expected[9] and is observed in Fig. 6, as the highest LIFand absorption data were measured around TDC,where temperature and pressure are highest. This,along with the better correlation of the absorbancedata with the in-cylinder volume before 20°BTDC,supports the assumption of reduced temperatureand pressure dependence of the IR fuel absorptioncoefficient.An evaluation of the measurement precision is

based on the standard deviation of the data from100 consecutive engine cycles. To eliminate any influ-ence of pressure and temperature on this analysis,the ensemble standard deviation values were nor-malized by the ensemble average at each CAD. Forthe LIF data a spatially averaged signal across132 pixels was used from within a specified data ana-lysis window as shown in Fig. 5. The results areshown in Fig. 7. The signal fluctuations for thetwo measurement techniques are initially higher,and then at TDC for the LIF signals and afterTDC for absorbance data the variation decreases.The measured normalized standard deviation showsthat homogeneity is increasing after TDC until fluc-tuations begin stabilizing at around 5%. The noiselevel for absorbance was determined to be at or below1% from transmissivity data (I=I0) from experimentswithout fuel injection. This marks the lower noise

limit for the entire setup at the engine. The data con-firm that local fluctuations in mixture homogeneityremain indeed at a level of 5% for this engine oper-ating condition.

B. Stratified Charge Motored Measurements

The ability to use the absorption measurement to de-duce information about fuel concentration and itsvariation at the spark plug needs to be evaluated.This is important since the spark plug electrodesand absorption path are several millimeters apartfrom each other and spatial fuel gradients in SG SIDIengines can be very large. Correlations of simulta-neously acquired LIF and absorbance data can beused for this evaluation. It is important to restate,though, that the LIF measurement volume is be-tween the electrode gap and absorption optical path.Eliminating differences in LIF and absorption datathat stem from temperature and pressure effects isimportant for this evaluation. Biacetyl LIF imagesfrom late injection engine runs were normalized withaveraged LIF images from an early injection run. Si-milarly, absorbance data from late injection runswere normalized with averaged absorbance datafrom the same early injection run. The relative fuelconcentrations that are reported from both measure-ments could be scaled to equivalence ratios or molefractions after determining the amount of air presentin the cylinder. Given the high blowby losses for thepresent experiment, we did not perform this scalingto avoid the introduction of a potential bias error thatwould depend on crank angle position. This wouldnot prevent the evaluation and discussion of the fluc-tuations and the correlations between LIFand absor-bance that was observed but it would also not addany relevant information to the present study, i.e.,an assessment of local fluctuations and how differentthey are at the two measurement locations. There-fore, no absolute scaling to equivalence ratio or molefraction was performed.

Processed LIF images (which have been correctedfor temperature, pressure, number density, lightsheet inhomogeneities, etc., as described above) froma single cycle during a late injected motored experi-ment are presented in Fig. 8. This particular imagesequence was chosen to illustrate two points.

First, it is evident that the fuel misses the sparkplug probe on its first pass and is recirculated fromthe piston bowl back to the cylinder head. Fromthere, the amount of fuel present in the images ap-pears to decrease, suggesting significant fuel motionin the third dimension or outside of the light sheetarea. A fuel cloud reaches the data analysis window,shown in the image at 39°BTDC just to the left of thespark gap, at approximately 19° BTDC.

Second, a fuel droplet can be seen moving from thefuel injector to the spark plug in the earlier images.The in-cylinder flow in this engine is dominated by astrong tumble flow that is directed from the fuel in-jector to the spark gap [16]. Using the known tempor-al separation between images and the spatial scaling

Fig. 7. Information on precision for LIF and absorption measure-ments from the homogeneous charge motored experiment is avail-able from normalized standard deviation data. Ensemble standarddeviation data from 100 consecutive engine cycles normalized bythe average at their respective CAD are used to eliminate the in-fluence of pressure and temperature dependence on the signals.Note that the noise limit of the absorption measurements was de-termined from runs without fuel injection to be 1% or less. LIFdata are not available after 70CAD.


it was determined that this droplet is moving at ap-proximately 2m=s. The separation of the LIF dataanalysis window and the absorption path is∼6mm. This suggests that the fuel will show up inthe LIF images approximately 9–10CADs before itregisters in the absorbance measurements. Figure 9shows a fuel concentration profile that was extractedfrom images in the same cycle along with normalizedabsorbance data for the cycle. LIF data prior to25°BTDC have been excluded from the plot sincean initial offset error in the LIF acquisition renderedimages taken prior to this CAD nonquantitative. A

temporal shift in the rise of the fuel concentrationat the absorbance measurement location is notice-able in the profiles, although to a lesser degree thanestimated. Variations on the delay are expected fromcycle to cycle since the in-cylinder flow is highly vari-able. This is especially true near obstructions such asthe spark electrodes and the absorption mirror ex-trusion. The systematic delay in arrival of the fuelcloud at the spark plug sensor is longer in the aver-age data shown in Fig. 10. This delay correspondsmore closely with the estimated velocity of the dro-plet. The significant discrepancies between the nor-malized LIF and absorbance probe signals is likelydue to the different probe measurement locations.One of the fuel plumes was directed towards the axisof the spark plug installation hole. The light sheetwas aligned between the spark electrodes, and theabsorption path and therefore this fuel plume almostfully lies within the path of the fuel spray. Conver-sely, the absorption path is about 2mm away fromthe light sheet, so the LIF data were expected toshow different dynamics.

An additional observation can be made in the ab-sorbance data set shown in Figs. 9 and 10. Before fuelfrom the current injection arrives at the measure-ment location, greater than zero concentrations weremeasured. There is an expected residual amount offuel in the cylinder from the previous cycle sincethe engine is just motored.

Figure 11 shows the results of this late injected ex-periment in scatter plot form. Data from 25° BTDC to70° ATDC are included in this plot. There are strik-ing differences noticed in comparison to the same as-sessment of the data from the homogeneous mixtureexperiment. The large fuel gradients, shown in theimages in Fig. 8, suggest that there are similar gra-dients between the location of the LIF measurementvolume and the absorption path. The result is that a

Fig. 8. Fully processed fuel concentration images from the lateinjected motored experiment. These images are used to visualizefuel plume development. Statistical information is also extractedfrom the data analysis window to the left of the spark electrodes(shown only in the 39°BTDC image for illustrative purposes) andis compared to the absorbance measurements. In images at33 − 23 °BTDC, circles are placed in the images to the left of thespark plug to identify a fuel droplet which moves from left to rightwith the flow field (see text for more details.)

Fig. 9. LIF and absorbance probe fuel concentrations from thecycle shown in Fig. 8 from the late injected motored experiment.The LIF measurement shows the presence of fuel approximately8CADs before fuel registers in the absorbance data. This delay isexpected from the spatial separation of LIF measurement windowand spark plug sensor based on the flow direction that was evidentfrom the droplet motion shown in Fig. 8.

Fig. 10. 100 cycle average of quantitative LIF and absorbancedata from the late injected motored experiment normalized bydata from the homogeneous charge motored experiment to cali-brate the signals (see text for details on calibration). The systema-tic delay in the arrival of fuel at the spark plug probe is consistentwith estimated droplet velocities that were extracted from LIF im-age sequences.


low correlation is expected, and this is clearly con-firmed. It is noteworthy to point out that the LIFdata are not expected to drop down to zero values;the images in Fig. 8 representatively show that atthis time we already observe fuel at the LIF detectionvolume, while it is plausible that the absorption pathdoes not yet have significant levels of fuel in it. Theinterpretation of either the LIF or absorption signalunder highly stratified conditions must take into con-sideration that neither measurement location maycoincide with the location of the spark plasma chan-nel. However, overall timing and fluctuation infor-mation can be obtained from both types of signals.

C. Homogeneous Charge Fired Experiments

While the LIF measurement delivers information onthe fuel concentration only, the spark plug sensormeasures fuel and combustion product gas concen-tration (aggregate of H2O and CO2) simultaneously.This capability was demonstrated for an early in-jected engine run with the intake air throttled downto ∼35kPa at 45 °C and a lean fuel/air mixture(ϕ∼ 0:8). The intake air was throttled to achievenear stoichiometric conditions since the quartz cylin-der is unable to withstand running at full load (atmo-spheric intake conditions and ϕ∼ 1). The spark wastriggered at 30°BTDC. This spark timing is some-what early for optimized combustion conditions ofthe engine when running at 600RPM and thus high-er than usual levels of unburned fuel were expected.The plasma luminosity is clearly visible from 29 −

22°BTDC in the LIF images shown in Fig. 12. Theflame kernel likely begins in between the light sheetand the camera (see Fig. 4 for an illustration of thelight sheet location) and eventually burns fuel withinthe light sheet beginning at 22°BTDC.Figure 13 shows the evolution of the LIF and ab-

sorbance signals that were recorded for the single cy-

cle shown in Fig. 12. The fuel absorbance rises from60 − 15° BTDC due to compression, as was observedin the motored measurements shown in Fig. 5. TheLIF data for this randomly selected cycle show con-siderable scatter at the measurement location to theleft of the spark plug. Fuel is not consumed at themeasurement locations until ∼17°BTDC, whereeach signal begins to drop and combustion productgas absorbance rises. The delay between the onsetof the spark and the appearance of the flame inthe measurement locations is likely due to the timerequired for energy transfer to the fuel and for thetime for the flame kernel to develop. The LIF signaldrops off more quickly than the fuel absorbance

Fig. 11. Individual, simultaneously acquired LIF and absorptiondata points that were used to generate the average values andstandard deviation bars in Fig. 10 are presented. The correlationis much weaker for stratified engine operation than for homoge-neous mixing conditions. See Fig. 5 for comparison. The large spa-tial and temporal fuel gradients in fuel concentrations aremanifested in this plot.

Fig. 12. Single-cycle LIF image sequence from a homogeneouscharge part load combustion experiment. Note that the luminosityof the spark plasma is visible from 29 − 22 °BTDC. The biacetyl isconsumed in the flame showing the corrugated growth of the flamefront starting at 22 °BTDC.

Fig. 13. Single-cycle LIF, fuel, and combustion product absor-bance measurements from the homogeneous charge part load com-bustion experiment. The LIF data were extracted from the imagesshown in Fig. 12. Note that the location of the LIF data analysiswindow has been moved to avoid capturing spark luminosity.


signal because the LIF measurement location is clo-ser to the spark gap where the flame must havestarted. Note that the combustion product absor-bance data clearly indicate the presence of residualgas in the cylinder from the previous cycle. Althoughno correction has been applied to determine the mag-nitude of these data, it is above the noise floor of themeasurements and can be concluded to be residualgas. The absorbance signal of the (residual) productgasses also increases during compression along withthe fuel signal because of increasing number density.Once combustion starts the signal actually firstdrops due to expansion of the (hot) combustion gas,and then signals increase again due to continuedcompression and combustion.Figure 14 shows full cycle profiles of the results of

fuel and product gas absorbance measurements aver-aged over 200 cycles. Both traces start out at residualgas levels and then drop considerably as they are di-luted with fresh, dry intake air. Fuel was injected at250°BTDC and began to arrive at the spark plug sen-sor around 200°BTDC. As mentioned before, thespark was triggered at 30°BTDC. Similarly to whatwas observed in the single cycle shown in Fig. 13, onaverage the flame does not reach the measurementarea until several CADs after the spark is triggered.The product gas absorbance trace shows a smallhump near 15° BTDC. This reflects the expansionof the gases due to combustion as discussed for theindividual cycle above. The fuel trace also shows evi-dence of outgassing as fuel forced into crevices duringcompression is released during the expansion strokewithout immediately oxidizing completely. This fueleventually reached the probe around 30°ATDC. Afterthe exhaust valves open at 139°ATDC fuel and ex-haust signals began to rise and peak approximately28CADs later. The cylinder pressure when the ex-haust valves opened was∼65kPa, while the pressurein the exhaust manifold during the exhaust strokewas ∼96kPa, so this peak is attributed to residualfuel and exhaust gases from the previous cycle flow-

ing back into the cylinder. The background (light off)signal from the spark plug probe was flat in this re-gion, indicating actual gas dynamics and not somespurious input.

5. Conclusions

The use of optical diagnostics has assisted the designand development of modern internal combustion en-gines significantly. Now, more than ever, detailedtemporally and spatially resolved information ofmixing and combustion processes is needed to allowfurther improvements in the efficiency and the re-duction of pollutant formation. This work describessimultaneous high-speed in-cylinder measurementsof fuel concentration distributions using laser-induced fluorescence and fuel and exhaust gas con-centrations using a mid-infrared absorption probethat was integrated into a spark plug. LIF measure-ments were based on biacetyl fluorescence that wasinduced by two simultaneously fired diode-pumpedfrequency-tripled Nd:YAG lasers. The broadbandbut spectrally filtered mid-IR measurements usedthe fundamental C─H stretch absorption band of hy-drocarbons to detect fuel near 3:4 μm, and combinedabsorption by water and carbon dioxide near 2:7 μmwas used as a measure for combustion product gases.

The combination of both techniques allowed a mu-tual assessment of the precision and accuracy ofthe measurements and the applicability of a path-integrating absorption measurement at a locationslightly offset to the spark plug when informationabout mixing at the spark plug is desired underhighly stratified conditions.

The precision of the LIF measurements has beenimproved over previous work due to spatial aver-aging over 132 pixels and the use of higher laserpulse energies. Precision values as low as 4% areshown for a measurement with highest homogeneityduring the exhaust stroke. However, the absorptionmeasurements with a precision of better than 1%also show 4–6% fluctuations in the fuel signal, indi-cating that this is a true fluctuation for the given op-erating conditions.

The analysis of the correlation of fuel absorbanceand LIF signal intensity revealed that the assump-tion of a temperature and pressure independenceof the mid-IR fuel absorption coefficient was reason-able around TDC for the iso-octane/biacetyl mixturethat was used in the present study. A systematic timedelay between LIF data taken from a∼ 1mm2 areanear the spark plug and the fuel absorbance mea-surement for highly stratified engine operationwas linked quantitatively to the fuel motion acrossthe cylinder. No correlation between the LIF and ab-sorbance values was present in this experiment dueto this time delay and large spatial gradients in thefuel concentration, but the absorption probe can beused by itself to identify the arrival of a fuel cloudin the measurement path. Magnitudes of measurednormalized fuel concentrations are matched forLIF and absorption, thus the temporally resolved

Fig. 14. Fuel and product gas absorbance signals averaged over200 consecutive engine cycles. The traces show important engineoperation characteristics such as ignition, outgassing, gas dy-namics, and residual gas levels.


absorption data can be used for parametric studies ofnear-spark plug events.The simultaneous recording of fuel LIF and fuel

and combustion products (aggregate of water andcarbon dioxide) absorption during homogeneouscombustion provided information about combustiononset and propagation as well as in-cylinder residualgases. Absorbance profiles reflect important engineoperation events, such as valve opening and closing,mixing, combustion, and outgassing from crevices.While the spark plug sensor measurements pre-

sented here indicate that the absorption coefficientfor fuel (iso-octane/biacetyl) is less sensitive to pres-sure and temperature than biacetyl LIF, an absolutecalibration strategy for absorption measurements re-quires quantitatively measured absorption coeffi-cients. An additional detector should be added as areference channel in a spectral region without ab-sorption by fuel and exhaust gases (around 3 μm).This channel allows for a true I0 measurement with-out assuming that at some time during the enginecycle absorption is negligibly small. This will espe-cially be required to measure absolute exhaust gasrecirculation rates. Exhaust gas measurementscould also be based on CO2 absorption (4:3 μm) to fa-cilitate absolute calibration independent of the ratioof CO2 and H2O.The present study shows that a mid-IR absorption

based spark plug probe simultaneously providescrank angle resolved data of relative fuel and ex-haust gas densities near the spark gap. The data cor-relate well with the characteristics of data from high-speed LIF imaging. In contrast to LIF, absorptiondoes not provide spatial concentration distributions,but it allows measurements in nonmodified engines.

This work was supported in part by GeneralMotors R&D within the GM-University of MichiganCollaborative Research Laboratory on Engine Sys-tems Research at The University of Michigan.

References1. F. Zhao, M.-C. Lai, and D. L. Harrington, “Automotive spark-

ignited direct-injection gasoline engines,” Prog. Energy Com-bust. Sci. 25, 437–562 (1999).

2. J. D. Smith and V. Sick, “Crank-angle resolved imaging of fueldistribution, ignition and combustion in a direct-injectionspark-ignition engine,” SAE Transactions Journal of Engines114, 1575–1585 (2005).

3. M. Koenig and M. J. Hall, “Cycle-resolved measurements ofprecombustion fuel concentration near the spark plug in a

gasoline SI engine,” SAE paper 981053 (Society for Automo-tive Engineering, 1998).

4. T. Berg, V. Beushausen, O. Thiele, and H. Voges, “Fibre opticsspark plug sensor for the optimisation of engine combustionprocesses,” Motortech. Z. 67, 2–6 (2006).

5. C. Schulz and V. Sick, “Tracer-LIF diagnostics: quantitativemeasurement of fuel concentration, temperature and fuel/air ratio in practical combustion systems,” Prog. Energy Com-bust. Sci. 31, 75–121 (2005).

6. J. D. Smith and V. Sick, “Crank-angle resolved imaging of bia-cetyl laser-induced fluorescence in an optical internal combus-tion engine,” Appl. Phys. B 81, 579–584 (2005).

7. N. Wermuth and V. Sick, “Absorption and fluorescence data ofacetone, 3-pentanone, biacetyl, and toluene at engine-specificcombinations of temperature and pressure,” SAE Trans. J.Fuels Lubricants 114, 804–814 (2005).

8. P. Guibert, V. Modica, and C. Morin, “Influence of pressure,temperature and gas phase composition on biacetyl laser-in-duced fluorescence,” Exp. Fluids 40, 245–256 (2005).

9. J. D. Smith and V. Sick, “Quantitative, dynamic fuel distribu-tion measurements in combustion-related devices using laser-induced fluorescence imaging of biacetyl in iso-octane,” Proc.Combust. Inst. 31, 747–755 (2007).

10. R. Zhang, S. V. Bohac, and V. Sick, “Stability of iso-octanemixtures with 3-pentanone or biacetyl as fluorescence tracersin combustion experiments,” Exp. Fluids 40, 161–163 (2006).

11. V. Sick and C. K. Westbrook, “Diagnostic implications of thereactivity of fluorescence tracers,” Proc. Combust. Inst. doi:10.1016/j.proci.2008.05.012.

12. P. Perona and J. Malik, “Scale-space and edge detection usinganisotropic diffusion,” IEEE Trans. Pattern Anal. Mach. In-tell. 12, 629–639 (1990).

13. L. S. Rothman, D. Jacquemart, A. Barbe, D. C. Benner,M. Birk, L. R. Brown, M. R. Carleer, C. C. Jr., K. Chance,L. H. Coudert, V. Dana, V. M. Devi, J.-M. Flaud,R. R. Gamache, A. Goldman, J.-M. Hartmann, K. W. Jucks,A. G. Maki, J.-Y. Mandin, S. T. Massie, J. Orphal, A. Perrin,C. P. Rinsland, M. A. H. Smith, J. Tennyson, R. N. Tolchenov,R. H. Toth, J. V. Auwera, P. Varanasi, and G. Wagner, “TheHITRAN 2004 molecular spectroscopic database,” J. Quant.Spectrosc. Radiat. Transf. 96, 139–204 (2005).

14. U. Fissenewert, V. Sick, and H. Pucher, “Characterization ofcombustion and NO formation in a spray-guided gasoline di-rect-injection engine using chemiluminescence imaging, NO-PLIF, and fast NO exhaust gas analysis,” SAE Trans. J. FuelsLubricants 114, 786–803 (2005).

15. W. Koban, J. D. Koch, V. Sick, N. Wermuth, R. K. Hanson,and C. Schulz, “Predicting LIF signal strength for tolueneand 3-pentanone under engine-related temperature andpressure conditions,” Proc. Combust. Inst. 30, 1545–1553(2005).

16. C. M. Fajardo and V. Sick, “Flow field assessment in a firedspray-guided spark-ignition direct-injection engine based onUV particle image velocimetry with sub crank angle resolu-tion,” Proc. Combust. Inst. 31, 3023–3031 (2007).


high-speed laser-induced fluorescence and spark plug absorption sensor diagnostics for mixing and...

Documents