Imaging of flame propagation and temperature distribution in an all-metal gasoline engine with endoscopic access via anisole fluorescence

Optical diagnostics are fast and non-intrusive tools used to investigate in-cylinder processes in internal combustion (IC) engines in high temporal and spatial resolution. These are typically applied on single-cylinder “optical” engines with transparent liners and pistons. These optical engines have an advantage of large optical access, but the thermodynamic properties and speed/load range are different from those of all-metal engines. To apply optical diagnostics in more realistic environment, minimally invasive endoscopes are used in all-metal engines. These “endoscopic” engines have the advantage of having nearly unmodified thermodynamic properties and speed/load range as the production engines. Different endoscopes have been used in the past to visualize various in-cylinder phenomena. Most of the endoscopes used are only transparent in the visible range and have poor light collection efficiency. Richter et al. compared a standard Karl Storz endoscope with widely available UV-Nikkor f/4.5 lens and found the collection efficiency of Karl Storz endoscope to be 30 times worse (Richter et al. 1998). Therefore, the application of such endoscopes is limited to relatively bright events in the visible range only. Few examples are visualization of broadband combustion in gasoline (Drake et al. 2003; Fach et al. 2022) and in diesel engines (Eismark et al. 2009; Miers et al. 2005), temporally resolved soot pyrometry (Bakenhus and R. D. Reitz 1999; Shiozaki et al. 1996), visualization of the spray via Mie scattering (Sczomak et al. 2010; Sementa et al. 2012; Fach et al. 2021), and particle image velocimetry (PIV) (Fach et al. 2021, 2022; Dierksheide et al. 2002). Soot formation has also been investigated using endoscopic imaging in direct-injection spark-ignition (DISI) engines by Burkardt et al. (2021) and Etikyala et al. (2022). To apply a wider range of optical diagnostics in engines with endoscopic access, in our previous work, a large-aperture hybrid UV-transparent endoscope system was developed (Reichle et al. 2012; Zimmermann 2006). It is now commercially available from LaVision. Other versions of UV-transparent endoscopes are used to investigate the combustion in hydrogen-fueled engines by imaging OH radical in Eicheldinger et al. (2021) and Karmann et al. (2022).

Tracer-based laser-induced fluorescence (LIF) is a widely used optical diagnostic tool for quantitative measurement of fuel concentration, temperature, and fuel/air ratio in combustion applications (Schulz and Sick 2005; Deguchi et al. 2002; Kuwahara and Ando 2000). LIF has also been used to detect the flame front via the “negative-LIF” technique (Bladh et al. 2005; Attar et al. 2015) (i.e., the disappearance of the tracer where the fuel is burned), to investigate the spray in a diesel marine engine (Hult and Mayer 2013) and formation and evaporation of piston fuel films (Alger et al. 2001; Geiler et al. 2017; Hochgreb 2001). Previously, toluene (methylbenzene, CH₃C₆H₅, boiling point 111 °C) has been well investigated as a tracer for LIF applications (Düwel et al. 2003; Faust 2013; Koban et al. 2005) and used for measurements in combustion engines, e.g., to image equivalence ratio and fuel distribution (Frieden and Sick 2003; Smith and Sick 2005; James and Smith 2005), to investigate fuel films in DISI engines (Geiler et al. 2017; Schulz and Beyrau 2018; Schulz et al. 2016), and to image the gas-phase temperature distribution via the ratio in single (Fuyuto et al. 2006) and two spectral bands (Gessenhardt et al. 2015; Luong et al. 2008; Peterson et al. 2013). Since toluene is a component of typical commercial gasoline, pure toluene has also been used as a surrogate fuel for LIF applications in combustion engines (Warey et al. 2002). More recently, anisole (methoxybenzene, CH₃OC₆H₅, boiling point 154 °C) was characterized spectroscopically for engine-relevant pressures and temperatures (Tran et al. 2013; Benzler 2019; Faust et al. 2013; Zabeti et al. 2017) and emerged as a potentially advantageous alternative to toluene (Faust 2013; Benzler 2019; Faust et al. 2014). The effect of tracer quantity added into the non-fluorescent base fuel can generate a substantial systematic error due to their differences in physical properties, and hence prevent the accurate quantification of LIF signals (Schulz and Sick 2005). The absorption cross section of anisole, when excited at 300 K with 266 nm laser radiation, is about 50 times larger and the fluorescence quantum yield is 63% larger than that of toluene (Faust et al. 2013). Therefore, anisole produces a stronger LIF signal per volume than toluene, which may enable better measurements, with less effect on the combustion process due to the lower tracer concentration required.

Like most aromatic species, the fluorescence of both toluene and anisole shows a red-shift with increasing temperature. This can be exploited to measure temperature by using two-color LIF thermometry. In this technique, the LIF signals are measured in two spectrally distinct regions (namely red and blue channels for shorter and longer wavelength bands, respectively). Due to the red-shift of LIF spectra with the increase in temperature, accompanied by an increase in the spectral width, relative signals increase in the red channel and decrease in the blue. The signal intensities detected in each channel, resulting from the combination of the temperature-dependent LIF spectra with the respective wavelength-dependent detection quantum efficiency, yield a signal ratio that can be used to determine the temperature. Since the measurement is based on taking the ratio between two spectral bands, it is very sensitive to the noise that accompanies low fluorescence signals (Peterson et al. 2014). In this work, it was investigated if the previously reported endoscopic two-color LIF thermometry (Gessenhardt et al. 2015) could be improved to single-shot imaging by substituting anisole for toluene. In addition to the (desired) red-shift of the fluorescence spectra with increasing temperature, anisole also has an undesired, but lesser, red-shift with increasing oxygen partial pressure (p_O2) that needs to be considered. The effect of p_O2 is more pronounced at pressures lower than what corresponds to atmospheric pressure, i.e., 210 mbar p_O2 (Faust et al. 2013), which is relevant to the in-cylinder conditions in the intake stroke. Even though toluene has no p_O2-dependent red-shift when excited with 266 nm laser (Koban et al. 2005) which makes it more convenient to use as a tracer in two-color LIF thermometry technique, the stronger LIF signal of anisole fluorescence spectra make the latter more advantageous for single-shot temperature measurements. The temperature and p_O2 dependence of the anisole LIF spectra was considered in a spectroscopic model for determining the gas-phase temperature from the pixel-wise red/blue signal ratio and p_O2, the latter determined from the in-cylinder pressure data. This theoretical spectroscopic model was then calibrated in situ. A simple calibration equipment was designed and used to perform in situ calibrations over a broad range of temperatures with air and nitrogen as bath gases. Different mathematical interpolation models were then used to extend the range of the spectroscopic models. These models were then used to determine temperatures along the compression stroke, and the results were compared to the adiabatic-core temperature. Finally, a selected model was used to visualize the temperature distribution during the compression stroke, and temperature stratification and mixing of fresh charge and internally recirculated exhaust gas during the gas exchange process.

Measurements were performed in a production line, four-cylinder, inline, naturally aspirated, spark-ignited engine (BMW N46 B20). The engine operated with stoichiometric fuel/air ratio and was equipped with port fuel injection (PFI) and a mechanically actuated variable intake valve train (Valvetronic). The engine was coupled to a speed-controlled dynamometer, with the torque controlled by an analog input to the engine control unit (ECU). Actuators, valve lift and timing, phasing and injection were managed by the production ECU accordingly, to which there was read-out access only. The engine geometry and operating parameters are given in Table 1.

The engine is modified in the fourth cylinder (the one nearest to the flywheel) to have two mutually perpendicular small endoscopic ports of 13 mm diameter each, one for illumination and other for detection. More details about the engine and the modifications for endoscopic accesses can be found in Gessenhardt (2013) and Witzel (2007). The hybrid UV endoscope system consists of two parts: a wide-angle front endoscope and a refractive-diffractive (“hybrid”) relay lens for chromatic correction in a pre-designed range (Reichle et al. 2012; Zimmermann 2006). The front endoscope is mounted on the engine and produces an intermediate image on the field lens at its back. This intermediate image is passed onto the hybrid relay lens without a rigid physical connection and projected onto a camera. The detection hardware (filters, relay lenses, and cameras) is mounted on a separate rigid frame. The absence of a rigid connection between the engine and the detection hardware (see Fig. 1) makes the latter isolated from engine vibrations. Moreover, filters and other optical elements can be placed between the front endoscope and the hybrid relay lens. The field of view (FOV) is approx. 30 × 30 mm² area at a distance of 35 mm (Reichle et al. 2012).

A custom fuel supply allowed “clean” operation on neat surrogate fuels without unknown fluorescing compounds in the endoscopically accessed cylinder, while a commercial fuel pump supplied high-octane commercial fuel to the other cylinders. All measurements were performed in continuously fired operation at 2000 min⁻¹ and 4 bar indicated mean effective pressure (imep). Throttling to this part load was achieved by an intake valve lift of 2.2 mm. Ignition occurred at -35 °CA. (0 °CA is compression top-dead center, i.e., crank angles during intake and compression are negative). On the endoscopically accessed cylinder, the crank angle resolved pressure was measured and recorded. The temperatures of the intake and exhaust gases, engine oil, and cooling water during the measurements are given in Table 2. The cameras and laser were controlled and synchronized with the engine by DaVis (version 8): imaging software from LaVision.

A schematic of the optical geometry of the modified cylinder, camera, and laser guiding arm is shown in Fig. 1. The dashed boxes represent two independent and rigid frames. The beam of an Nd:YAG laser at 266 nm with a maximum repetition rate of 10 Hz was introduced into the light sheet endoscope via a laser guiding arm. The laser energy was 2 mJ/pulse as measured before the endoscope that generated a 0.5 mm thick, vertically diverging laser light sheet through a set of cylindrical optics. A long-pass filter (LP 266) mounted behind the detection endoscope blocked reflected laser light. A beam splitter (LP 310) transmitted light above 310 nm onto a 12-bit ICCD (max counts: 4096) yielding an image from the “red” part of the LIF spectrum, while the reflected light below 310 nm was detected by an identical camera in the “blue” channel. The linear bias in measured signal ratio due to the angular dependence of the transmission spectrum of beam splitter, as considered in Peterson et al. (2014), was not considered in this case because the light coming from the field lens was near-parallel (Shahbaz et al. 2023) (The change in incidence angle is only 0.44% from the parallel beam). Band-pass filters further constrained the spectral range of each “color” channel (BP 320/40 for the red channel and BP 280/20 for the blue channel). Both cameras were gated for 200 ns bracketing the laser pulse.

Figure 2 gives an overview of the most important spectral features in emission and detection. As the fluorescence spectra red-shift with increasing temperature, as shown in Fig. 2, the ratio of the signals detected on each camera changes. This ratio can be related to local temperature by a suitable calibration.

Based on our previous work on two-color tracer-LIF (Gessenhardt et al. 2015), 20% toluene by volume was added in isooctane base fuel. For experiments with anisole as the tracer, only 2.5% anisole was added to the isooctane base fuel to keep the absorption of the laser beam across the field of view similar.

Figure 10 shows the image post-processing steps for extracting the signal ratios for each image at a crank angle. The images were first coarsely registered (mapped onto each other) in DaVis with a UV-illuminated target mounted on the detection endoscope in place of the field lens, and then more precisely based on actual LIF images in MATLAB R2017a (The MathWorks Inc., Natick, USA). After registration, the regions outside the FOV and, where applicable, those blocked by the piston are masked out. The dynamic background (BG) was then subtracted from each image by averaging a 30 × 100 pixel area from the shadowed region by the spark plug’s ground electrode (shown as red rectangle in Fig. 10 (1)).

Next, as shown in Fig. 10 (3–5), the ‘burnt area’ where the signal is only background was masked out at later crank angles. To achieve this, a flat field was obtained by averaging 100 images obtained at each crank angle from the same experiment under fired conditions. The flat-field images were homogeneous during most of the compression stroke, i.e., before ignition at − 35 °CA bTDC. However, after ignition the fluctuating location of the flame starts affecting the homogeneity of flat-field image. For the images where the flame kernel was small (i.e., − 20 °CA bTDC), the effect was limited to the regions close to the spark plug. However, for the later crank angles where the flame front was larger and fluctuating, the effect was not negligible. However, since it was not possible to obtain a flat-field image with unfired operation on the test bench, the images obtained under the fired operation were used to obtain flat field. The flat field obtained at later crank angles under fired conditions still yielded sufficient correction for successful binarizing by thresholding. The single-shot images were divided by this flat-field image. If this flat-field correction is not performed, binarizing by simple thresholding would also eliminate some ‘useful’ regions on the right side of the image that have lower signal due to laser absorption. The flat-field corrected images were then smoothened by a 5 × 5 mean filter, normalized by the maximum count value and binarized by thresholding at 18% of that maximum count value. This binarizing process eliminated all of the burnt gas and unilluminated regions.

In the next step, the binary image was then multiplied with the background-corrected images in both channels and 3 × 3 mean filtering was applied to reduce the noise (Fig. 10 (6)). The image from the red channel was divided pixel-wise by the image from the blue channel to yield the signal ratio image. The average signal ratio was determined by averaging just the non-zero pixels in a 100 × 150 pixel ROI below the spark plug, as shown in Fig. 10 (7). If the nonzero pixels filled less than 25% of the ROI, as could be the case at later crank angles with larger burnt regions, the results were discarded. This resulted in 100 values for the signal ratio at crank angles before and just after ignition, becoming less and less as the flame propagated. For example, at TDC 9 out of 100 images could be used to obtain a value for signal ratio.

To calculate the average temperature at each crank angle, the average signal ratios for each image were obtained at each crank angle with the method explained above, and then ensemble-averaged. p_O2 for each image at each crank angle was obtained by crank angle resolved pressure trace considering 19.1% internally recirculated exhaust gas (Gessenhardt et al. 2015). The ensemble-averaged signal ratios were then used in combination with the ensemble-averaged p_O2 at that crank angle to calculate the average temperature. The signal ratio images were also converted into temperature images by calculating the temperature for each pixel assuming a constant p_O2 throughout the field of view. In addition to signal ratios for each image at all crank angles, the mean and standard deviation of pixel values in the ROI were also determined for BG-subtracted red and blue images, and for the signal ratio images.

Anisole, due to its photophysical properties, is a strong candidate for use as a tracer in non-intrusive LIF diagnostics in combustion engines and other applications. Here, it was used as a tracer with the base fuel isooctane in an all-metal engine with endoscopic access. Imaging in two spectral portions of anisole fluorescence was performed with two ICCD cameras through a UV endoscope system. Anisole LIF signal levels were first compared to those of toluene. Toluene is one of the best-characterized tracers, but anisole yielded higher signal levels when both tracers’ concentration was limited by laser attenuation, which here was the case for 20% toluene vs. 2.5% anisole. The anisole signal was ~ 5 times higher in the spectral range of 320 ± 20 nm with 8 times less concentration at − 60 °CA (before TDC during compression stroke). Small structures of the turbulent flame burning into the anisole/isooctane mixture were clearly visible after ignition.

Since this test was promising, anisole was used for ratiometric two-color LIF thermometry, a technique we had implemented endoscopically before with toluene, but mostly without achieving reasonable quality single-shot images (Gessenhardt et al. 2015). From spectroscopic data in the literature, different empirical signal models considering the effects of temperature and oxygen partial pressure (p_O2) were assembled. The models can be categorized mainly as two-step fit function, fitting separately for p_O2 and temperature, and a single 3D surface fit function incorporating the effect of both parameters on the signal ratio. Each category had further variations in terms of the function governing the effect of temperature on the signal ratio. An in situ calibration was performed in nitrogen and air at temperatures ranging from room temperature to 573 K. The calibrated models were used to convert the red/blue LIF signal ratio to temperature in the compression stroke, which was compared to the adiabatic pressure-based temperature. It was found that the 3D-surface fit functions greatly overestimated the temperatures in the late compression stroke. The two-step fit functions with a two-parameter exponential temperature-dependent part yielded temperatures closer to the adiabatic temperature until ignition, but with some deviation when the p_O2 in the cylinder increased beyond that in available spectroscopic data and the effect of increasing p_O2 on the signal ratio was extrapolated.

The two-step fit function with a four-parameter exponential temperature-dependent part was then used to visualize the spatial temperature distribution at different crank angles during the compression stroke. As expected for port fuel-injected operation, the temperature was found to be mostly uniform across the FOV, with later crank angles showing more noise because of lower signal due to the fluorescence lower quantum yield at higher temperatures and pressure. The spatial temperature distribution around the gas exchange TDC were also visualized. The residual gas was assumed to be unburnt anisole containing fuel. The residual gas mixture was observed moving along the in-cylinder tumble at crank angles shortly before TDC with measured temperatures close to the temperature measured in the exhaust downstream of the valve. Clear temperature segregation was observed when fresh charge started entering the cylinder with temperature decreasing with progressing intake stroke, consistent with what was observed in our previous work with toluene LIF (Gessenhardt et al. 2015).

The results showed that for anisole LIF two-color thermometry considering the effect of oxygen partial pressure is also important. In this respect, the lack of spectroscopic data is an issue, since the available data only cover about 7% of the range encountered in p_O2. If further spectroscopic data were available, the accuracy of the temperature determined by two-color thermometry with anisole may be improved. The correlation of signal ratio and temperature also needs to be more accurately established with the help of more spectroscopic data, especially at high temperatures. The available data covered more than 50% of the required temperature range, but inaccurate extrapolation nevertheless may result in inaccuracies at high temperatures.