Edinburgh Research Explorer Early flame propagation in a spark-ignition engine measured with quasi 4D-diagnostics

This paper presents the first results towards experimentally resolving the local three-dimensional (3D) flame propagation and turbulence-chemistry-interaction in a spark-ignition engine using temporally resolved multi-planar laser diagnostics. The experimental method utilizes simultaneous dual-plane laser induced fluorescence (LIF) of OH and stereoscopic PIV (SPIV) to locally resolve 3D flame displacement speed during the early flame development when less than 5% of the mass has been consumed. OH-LIF is used to track the reaction-zone position and flame normal direction in 3D space, while SPIV measures the convection of the identified flame contours. Based on the vectorial difference of the 3D convection and absolute propagation of the reaction-zone, the 3D displacement speed (s T ) is calculated. An instantaneous flame realization shows a large dynamic range of local s T and local flow transport, while also revealing the importance to resolve these quantities in 3D. Several flame-flow configurations are shown along the flame surface and each uniquely defined the local flame transport along the individual flame realization. A detailed uncertainty and sensitivity analysis is performed, confirming the validity of the s T distribution resolved for the methodology and operating conditions. A discussion on the different mechanisms leading to the large distribution of s T for the given operations is included and testifies to complex nature of the in-cylinder flame development in this early stage. The limitations of the presented methodology are discussed particularly in the need for improved spatial resolution and additional volumetric information. The merits and limitations of the presented work provides an improved understanding of what is further needed to better resolve local 3D flame transport in engines for both experimental and numerical methodologies.


Introduction
A detailed understanding of local flame propagation in spark-ignition (SI) engines is required to further improve combustion performance, realize next-generation combustion strategies, and provide valuable data to build predictive models in turbulent combustion.Flame-front progression within premixed systems is primarily governed from two components: (1) unburned-gas advection and (2) flame-normal displacement speed [1].The latter has been recognized as one of the most important flame properties [2] particularly regarding numerical models using flame surface density (FSD) [3] and level-set [4] approaches, where such local effects must be included into sub-grid scale models [2,5].In SI engines, like most turbulent combustion environments, flow-direction and flame-geometry are highly threedimensional (3D).Determination of the flame-normal displacement speed and advection velocity within a 3D domain is necessary to appropriately resolve local displacement speeds in turbulent combustion.
Laser-based diagnostic techniques provide non-intrusive measurements of leading scalar and vector properties that quantify fundamental processes in combustion systems.With regards to flame propagation, traditional methods such as Schlieren, chemiluminescence, Mie scattering, and laser induced fluorescence (LIF) have measured integral flame growth rates for spherical-based flame geometries, which have provided the basis for most combustion models in practical systems [6][7][8][9][10][11][12][13].A more comprehensive understanding of flame propagation however, has come from multi-parameter diagnostics that utilize multi-planar measurements to resolve coupling between turbulent flow and reaction chemistry.Measurements employing such techniques have provided detailed statistics of the 3D flame-normal, flame stretch, advection velocity, and local 3D displacement speeds in turbulent combustion [5,[14][15][16][17].Sophisticated, well-designed diagnostic methods are considered high-priority for an improved understanding in turbulent combustion.
Within SI engines, experimental efforts have primarily resolved integral flame growth rates from singleplane imaging, line-of-sight, and optical-fiber probes [9,11,12,18].Local 2D flame displacement speeds have been resolved utilizing high-speed laser tomography and particle image velocimetry (PIV) in a boosted optical SI engine [19].Using short laser-pulse separation, Mie scattering images resolved planar flame displacement, while local advection velocities were obtained from PIV.Although these measurements provided valuable classification of turbulent premixed flames within engines, the assumption of 2D flame propagation biases displacement speed calculations [15] and resolving the 3D flame-normal and flow velocity must be considered.
This work utilizes the dual-plane OH-LIF, stereoscopic PIV (SPIV) methodology from [14] to resolve local 3D displacements speeds in an optical SI engine.The engine is operated at 800RPM with stoichiometric, premixed isooctane-air mixtures.Images were acquired during the early-flame development when less than 5% of the mixture is consumed, typically regarded as the crucial stage of flame development defining subsequent combustion [20].Measurements reveal individual contributions of local flow velocity and flame displacement speed on the overall transport of the flame surface.A detailed uncertainty and parameter sensitivity analysis is presented to discuss the validity of the results and limitations of the measurements.

Engine
Experiments were performed in a single-cylinder optical SI engine [21].The engine was operated at 800RPM with port-fuel injection of isooctane (27.4mg/cycle) to provide a stoichiometric, homogeneous mixture.The operating conditions shown in Table 1 are chosen to mimic low-load, idle engine operation.
Low-load operation is technically relevant as it can be prone to combustion instabilities [20].Moreover, the operating conditions provided low turbulence levels ( 1m/s (Reynolds decomposition) not shown) and repeatable thermodynamic conditions for reliable measurements of local flame propagation.
The fuel-air mixture was ignited by a spark plug (NGK Spark Plug Co., Ltd.) with dwell of 3.5ms.Spark timing, ST=19 crank-angle degrees (CA o ) before top-dead-center (bTDC), was chosen to optimize indicated mean effective pressure (IMEP) and combustion stability (COV of IMEP <2%).The engine was fired 200 cycles before acquiring measurements and provided stable engine boundary conditions.

Diagnostics
The optical setup is similar to that used in [14] and shown in Fig. 1.A frequency-doubled Nd:YAG dualcavity laser (Edgewave, INNOSLAB IS4 II-DE, 532nm) operating at 4.8kHz was used for SPIV measurements.Laser light passed through focusing optics to provide a laser sheet (0.5mm thickness), which was centered vertically within the tumble plane and bisected the spark plug center electrode.
Two CMOS cameras (Phantom V.711, double-frame exposure) in Scheimpflug arrangement placed on each side of the engine imaged Mie scattering off chemically inert boron nitride (BN) particles (0.5µm diameter), which were seeded into the intake flow.The cameras were mounted =14° off-normal to the LIF cameras and imaged a 15x25mm 2 region centered around the spark plug.
The combustion radical hydroxyl (OH) was imaged simultaneously in two parallel, vertical planes by two independent double-pulsed UV laser systems.Each laser system used a frequency-doubled dye laser (Sirah, Precision Scan, Rhodamine 6G) pumped by a double-pulsed Nd:YAG laser (Spectra Physics, PIV400, 532nm) each providing two UV laser pulses (24mJ/pulse) temporally separated by ∆t LIF .The dye lasers were tuned to 282.9nm to excite the Q 1 (6) line of the A-X(1-0) transition of OH.
The laser beams from each UV laser system were sent through independent focusing optics and spatially separated to provide two parallel UV light sheets (0.2mm thickness) offset by ∆z=±0.5mm on each side of the PIV light sheet (Fig. 1).Experiments were also conducted with ∆z=±0.25 and ±0.75mm.The sheet separation and thickness was adjusted by UV-sensitive beam-monitor (DataRay).Fluorescence emission from each laser sheet passed through high-transmission, band-pass filter (Laser Components, UV-B) and was imaged onto two separate image intensifiers (LaVision, High-speed IRO) coupled to 14-bit CCD cameras (LaVision, ImagerProX, double-frame exposure) arranged on each side of the engine.
Spontaneous OH* chemiluminescence and flame luminosity were suppressed by gating each intensifier to 300ns.The projected pixel resolution of both LIF detection systems was 20 m, while the spatial resolution was 80µm determined by a Siemens-stern (contrast transfer function).
Integral flame growth was imaged by chemiluminescence.An unintensified CMOS camera (LaVision, HSS5, 10bit) operated at 5kHz and imaged onto a 40x55mm 2 area through the piston via the crank-case mirror.Two identical long-pass 550nm filters (CVI) were used to suppress 532nm laser light.
An optical crank-angle encoder (AVL) was used to synchronize all lasers and cameras to the engine.The LIF systems were synchronized by a programmable timing unit (LaVision, PTU) to provide images at a fixed CA o .Each double-pulsed LIF system operated at 10Hz providing two temporally resolved LIF images (t 0 , t 0 + t LIF ) in each plane at 14°bTDC (i.e. 5 CA o after spark), when the flame has consumed less than 5% of the mixture.The UV laser pulses between each laser system were offset by 400ns to avoid cross talk between the LIF images in each plane.PIV lasers and cameras operated at 4.8kHz to measure the threecomponent (3C, u,v,w) velocity from 20°-2° bTDC.The first PIV laser pulse was triggered 12.5µs after the first UV pulse and the pulse separation for the PIV (Δt PIV ) was 25µs, while the pulse separation for each LIF laser system (Δt LIF ) was 50µs.LIF images were recorded every other cycle for a 200 cycle sequence, while PIV images were recorded for 200 consecutive cycles.Chemiluminescence images started 1 CA o before spark and recorded 20 images/cycle at 5kHz for 200 consecutive cycles.

Data Processing and flame speed calculation
The absolute velocity of the reaction-zone ( ) is defined as sum of the local unburned convection velocity ( ) and the flame displacement speed (s T ) relative to the flow in the flame-normal direction ( ).
The image processing procedure presented in [14] was used for the reconstruction of the flame surface onto the SPIV plane and to extract s T from the acquired images.The reaction-zone was identified in the LIF images by applying a non-linear diffusion filter [22] based on anisotropic operating splitting in combination with a Canny edge-detection algorithm.The identified reaction-zone is superimposed onto a raw LIF image in Fig. 3a.The red dashed line highlights a region of high in-plane curvature towards the products, which is not resolved due to weak LIF gradients.Although such areas represent <5% of the overall detected contour, systematic errors in s T will occur in such regions.
A NURBS spline interpolation [23] was used to construct the 3D surface between the flame contours identified in each OH-plane and a patch diffusion algorithm [24] removed numerical noise from the constructed surface.The 3D flame surface is projected through the SPIV-plane providing the local flamenormal at z=0mm.The convection velocity ( ) was extracted 0.4mm in front of the flame surface (flame-normal direction) on the SPIV-plane at t 0 .1D laminar flamelet simulations [25] demonstrated this distance is sufficient to avoid effects of pre-heating and expansion.SPIV images were processed with a commercial software (LaVision DaVis 8.1).Spatial calibration and dewarping of the PIV images were accomplished with a 3D target (LaVision, Type7).Self-calibration was accomplished from 200 Mie scattering images and provided a remaining average pixel disparity <0.01 pixels.Image cross-correlation and vector calculation were performed with a decreasing window size multi-pass algorithm.Final interrogation window size was 16x16 with 75% overlap providing 0.1mm vector spacing.
The integral flame growth rate (from chemiluminescence) was calculated assuming spherical flame growth [29].Burned and unburned gas densities to determine the integral growth rate were resolved from 1D flamelet calculations.

3D flame surface and local displacement speed realization
This section presents a single realization describing the flame propagation with respect to local 3D flame-orientation, convection velocity, and 3D displacement speed.A statistical distribution of s T is shown and further discussed in sections 4 and 5.
The reconstructed 3D flame surface projected onto the 2D3C flow-field is shown in Fig. 3b   The distribution of s T along the flame surface for 80 cycles is shown as bars in Fig. 6.It reveals that s T covers a broad range of values ranging from -5 to 15m/s, with mean value =2.6m/s.Discussion of this range is presented within the following sections.

Uncertainty and Sensitivity Analysis
Calculation of s T is based on the precision of each input parameter in eq. 1 as well as the quality of the linear approximation of the 3D flame surface.Discussion of measurement uncertainty (accuracy, precision) and parameter sensitivity is considered within this section.
Accuracy of the calculated s T is assessed in comparison to integral flame growth rate obtained from chemiluminescence imaging [29].Chemiluminescence images depicting the integral growth rate (not shown) provide a spherical global flame growth of 2.1±1.2m/s at the same CA o (statistic from 80 cycles).
This is in good agreement with .

Precision of local s T values arise from the precision of each input parameter, namely reaction-zone
identification and unburned gas velocity.Each LIF-detection system had a spatial resolution of 80 m.This could provide a maximum possible offset of 160 m between FS1 and FS2.Assuming the LIFdetection within this offset is Gaussian distributed, a reasonable offset between flame surfaces is 53 m (i.e. 1 ), providing an uncertainty ±1.1m/s, while the maximum offset provides ±3.1m/s.SPIV measurements have an estimated precision of 10% from [30] corresponding to a maximum uncertainty of 1m/s.A root mean square estimation for s T provides a 1 uncertainty of ±1.5m/s, or maximum (3 ) of ±4.5m/s.The calculation of s T is also based on several limitations of the measurement systems.It is anticipated that flow and flame detection is under-resolved in regards to fine-scale structures that exist in turbulent flows [4].Furthermore, the linear interpolation of the 3D flame surface and laser sheet separation ( z) provides limited out-of-plane resolution.Therefore it is necessary to address the sensitivity of system parameters on the calculation of s T .As depicted in Fig. 3a not all regions of the flame are resolved.
However, this consists of less than 5% of the flame contour for all cycles reported.The low occurrence of these areas will not significantly affect the s T distribution and is therefore not included in this analysis.
The following sensitivity analysis discussion is separate from the s T uncertainty values already reported.
SPIV was processed with a 16x16 pixel window size and data processing was repeated for a 32x32 pixel window to assess the sensitivity of SPIV resolution for s T .s T was compared at each location for both processed datasets and s T is the local difference.This resulted in an average s T =0.7m/s and did not significantly alter the s T distribution (red line, Fig. 6a).The extraction location of the unburned convection velocity was also varied (0.1-0.6mm, 0.1mm increments) for the 16x16 pixel window size dataset to assess the sensitivity of flow location in the flame-normal.The average s T is 0.8m/s and reveals a similar s T distribution (not shown for clarity).
Experiments were repeated with z=0 mm to assess differences of the detected flame contour imposed from the processing methods and different LIF-detection systems.An average deviation between detected flame contours was 120 m, which will produce an artificial normal flame-angle of ±12 o .
Displacement speed calculations were repeated with ±12 o and revealed a maximum s T of 0.4m/s.
Additional experiments were repeated with z=0.25 and 0.75mm to assess the sensitivity of z.This revealed similar s T distributions as z=0.5mm (Fig. 6a black and blue lines vs. bars).In-plane flame curvature revealed a statistic of =-0.2±0.8mm - from the 80-cycle distribution.Assuming that out-ofplane curvature is similar to in-plane curvature, the linear reconstruction method for z =0.25 and 0.5mm is argued valid to capture the 3D flame structure and assumed that sub-wrinkling not captured by the linear reconstruction is minimal.
Out-of-plane convection and flame orientation can still bias the multi-planar measurements.Therefore, results were conditionally sampled to only include measurements that exhibited an out-of-plane convection angle | |<45 o and out-of-plane convection velocity |w|<2m/s.Although 40% of the measurement data was removed, the s T distribution remained nearly the same (Fig. 6b, red and black lines vs. bars).Results were further conditionally sampled to include data that exhibited absolute displacement speeds >2m/s (i.e.>0.1mm total flame displacement) to exclude data that are susceptible to uncertainties of the flame front detection due to limited LIF spatial resolution.
Conditional sampling on all aforementioned conditions (blue line, Fig. 6b) results in the same distribution, providing trust that the resolved s T distribution is valid for the methodology and operating conditions.

Discussion
Thus far it has been shown that s T is locally dependent along the flame surface and a large distribution exists for the operating conditions reported.The uncertainty and parameter sensitivity analyses performed provided confidence that the distribution is credible.This section presents a discussion describing the range of s T values reported.
The aforementioned effects can vary locally along the flame surface, which can also lead to local situations where s T >7.2*s L .The large dynamic range of flame development and limited out-of-plane resolution can also result in large s T values for the current methodology.As depicted in Figs. 4 and 5, the flame surface is not equally transported.For the fixed t LIF = 50 s, local flame segments drastically advance and change shape.If opposing surfaces of the flame unite (e.g.closing of concave wrinkle) the locally resolved flame progression will appear substantial and increase s T [26].To better resolve these effects, a large temporal dynamic range and an improved out-of-plane resolution (e.g.tri-plane, or volumetric tomography) are required.
Twenty-percent of the distribution reported in Fig. 6 contains negative displacement speeds.This occurs when local regions of the convected flame contour are transported past FS2 and the resulting flamenormal is towards the products rather than reactants.The notion of negative s T does not pertain to rates of fuel consumption or heat release, but instead pertains to the flame displacement relative to the flow [27].Although uncertainties due to limited spatial resolution or measurement precision can yield negative s T , these occurrences still exist beyond the maximum uncertainty.Thus, negative s T values are statistically significant (i.e.6.5% occurrence considering 3 uncertainty), which support findings of [2,14,15,26,27,28].Direct numerical simulation (DNS) studies have identified three primary mechanisms attributed to negative s T : (1) regions of high positive flame curvature exhibiting a larger diffusive flux than convective flux [27], (2) regions of high compressive and tangential strain [26], and (3) sensitivities of the prescribed isolevel (e.g.OH or fuel mass fraction) within the reaction front layer that is used to calculate local displacement speed [5,28].In this work like other experimental investigations [14,15] no correlation is found between negative s T and local values of flame curvature or strain.Such findings are based on the available detection system resolutions (temporal, spatial) and given operating conditions.
Note that our operating conditions are quite different than the aforementioned DNS and experimental efforts, which exhibit atmospheric conditions within simpler flame-facilities.At our imaging timing, the engine operates with 12 bar, 550K unburned gas-temperature, complex flame-flow configurations, and solid boundaries -all which can lead to different mechanisms for flame transport or negative s T .
The measurements presented are the first of its kind that quantify the distribution of local 3D flame propagation in an SI engine.They reveal the complex 3D flame transport due to advection and diffusion.
Similar data processing methods are transferrable to numerical models for meaningful statistical comparisons and model development.However, the presented results demonstrate that it is important that numerical models account for negative displacement speeds such as in [27] for meaningful comparisons and further understanding of the mechanisms attributing negative s T values.The measurements also provide an understanding of the required numerical mesh sizes needed to resolve the flame displacement speed beyond the merits of the presented work.
While these measurements provide a unique opportunity to understand flame transport in engines, they also have several shortcomings, which should be well-addressed.With the conventional lenses and cameras used, it is recognized that our detection systems under-resolve the spatial-and temporal-scales of the flame transport, which results in a filtered s T value.To properly resolve local flame transport, it is recommended to resolve several measurement points within the reaction zone thickness ( L 45µm at the imaging timing (1D flamelet calculations)).Improved spatial resolution in the z-direction is also required.The current experimental setup, however, vastly approached the physical limits of the conventional detection systems used and the physical space around the optical engine.Measurements and simulations with improved temporal and spatial resolution, with additional volumetric information and detection of several reaction species will provide further insight of flame transport beyond the capabilities of the presented work.Verification of such complex methodology undertaking is first intended for simpler flame-facility geometries (e.g.laminar flames or freely propagating turbulent flames as presented in [14]).

Conclusions
This paper describes an experimental investigation to resolve the early 3D flame propagation in a singlecylinder SI optical engine.The engine is operated at 800 RPM with premixed stoichiometric isooctane-air with mechanisms found to cause negative s T from previous DNS studies.For presented findings, there appears to be several physical mechanisms responsible for the large s T distribution, which attributes to the complex nature of early flame development in SI engines.This experimental effort was the first of its kind to resolve local 3D displacement speeds in an SI engine and will be further used to study flame transport behavior for higher engine speeds and different fuel-blends.Further improvements in spatial resolution and additional volumetric information within experimental and numerical investigations will provide further insight of flame transport beyond the capabilities of the presented work.

The
3D displacement speed was determined by transporting the position of the reaction-zone at t 0 , z=0mm by the local 3C velocity in 3D space.The remaining distance between the transported contour and the nearest point on the reconstructed flame surface at time t 0 +∆t LIF in the flame-normal ( ) is representative of the local displacement speed s T .Local values were determined for individual points spaced 20 m (i.e.projected LIF pixel size) along the flame contour on the SPIV plane.The limited spatial resolution (80 m LIF, 400 m SPIV) is greater than the laminar (thermal) flame thickness ( L 45µm, from1D flamelet simulation).The detected flame front in the measurements is therefore recognized as a spatially filtered quantity.As a clear separation from previous work[2,14,15,26,27,28], which use various methodologies to assess the local laminar burning velocity, all results here are based on a filtered displacement speed (s T ), referred to as turbulent displacement speed.
and provides an overview of the flame development for an individual cycle.The red and blue contour lines represent the reaction-zones in the z=0.5mm and -0.5mm OH-LIF planes, respectively.The gray surface depicts the reconstructed 3D flame surface between the LIF contours and the white contour is the flame position on the SPIV plane.A side-view of the 3D flame surface (Fig. 3c) further reveals the flame geometry within the 3D domain.Velocity vectors (every 4 th vector) represent the local 3C velocity within the burned and unburned regions.The color-scale corresponds to the 3C velocity magnitude.The rectangles (white dashed lines 1 and 2) highlight flame segments discussed in Fig. 4. The flame surface shown in Fig. 3b surrounds the spark plug electrodes and is approximately 35mm in length.Both in-plane and out-of-plane wrinkling is evident for the reconstructed flame surface and is discussed further within section 4. The flow to the left of the flame exhibits high velocities (8-10m/s) directed towards reactants, resulting in strong transport of the flame surface.To the right of the flame, velocity magnitude is significantly lower (1-4 m/s) and primarily directed opposing the flame surface.

Figure 4
Figure4shows enlarged views of two selected flame segments (regions 1 and 2 in Fig.3b) to describe mixtures.Simultaneous dual-plane OH-LIF and stereoscopic PIV were used to measure the local 3D flame-normal and 3C convection velocity to resolve the local 3D flame displacement (s T ).Images were acquired 5 CA o after spark when less than 5% of the mixture was consumed.The processing methods used to create the reconstructed 3D flame surface and extract local s T is described.An instantaneous flame realization is shown as an example to describe the flame transport at the operating conditions.Both local s T and convection velocity exhibit a large dynamic range and many flame-flow configurations are shown, which individually define unique flame transport along the same flame front.Thorough uncertainty and parameter sensitivity analyses performed revealed that the s T distribution over the 80 engine cycles is credible.The distribution is centered around s T =7.2*s L with significant regions extending both to large and negative values.Local flame instabilities caused by high in-cylinder pressures, local thermodiffusive properties, and high turbulence levels including strong flame wrinkling are underlying physical reasons for s T >s L .Occurrence of negative s T support previous literature findings.However, under our operating conditions and filtered s T representation, negative s T values were not correlated

Figure 2 :
Figure 2: Vectorial schematic of local transport of flame surface.

Figure 5 :
Figure 5: (a) s T (blue) and convection velocity (black) along flame surface, (b) local velocity values along flame contour, (c) local out-of-plane flame-and convection-angle.Shaded regions correspond to flame segments shown in Fig. 5. Red line indicates cycle-average .

Figure 6 :
Figure 6: Sensitivity of s T -calculation on input parameters.(a) original s T PDF (bars) compared to PDFs using different PIV resolution (red) and experiments utilizing z=0.25, 0.75mm (black, blue), (b) original s T PDF (bars) compared to PDFs that are conditionally sampled accounting for out-of-plane convection angles -45< <45 (red), out-of-plane convection velocity |w|<2m/s (black) and absolute displacement >2m/s (blue).

Table 1 :
Engine details and operating conditions