Flame Speed Measurement of a Racing Engine by IR Method and Chemiluminescence Method

Transcription

1 Lisbon, Portugal, 7-1 July, 28 Flame Speed Measurement of a Racing Engine by IR Method and Chemiluminescence Method Yuji Ikeda 1, Atsushi Nishiyama 2, Thierry Baritaud 3 1: Imagineering, Inc., Welv Rokko 2nd Bldg. 351, Fukada-cho, Nada, Kobe, Japan, yuji@imagineering.jp 2: Imagineering, Inc., Welv Rokko 2nd Bldg. 351, Fukada-cho, Nada, Kobe, Japan, atsushi@imagineering.jp 3: Ferrari spa, Via Ascari 55/ Maranello, Italy, tbaritaud@ferrari.it Abstract In this study, the optical diagnostics techniques of infrared (IR) absorption and chemiluminescence were applied to a racing engine. A small M5 IR sensor and Cassegrain optics sensors were developed to measure in-cylinder fuel concentrations and local chemiluminescence intensities. The small IR sensor was able to withstand racing engine operating conditions up to 16, rpm and exhibited little vibration noise. The IR sensor was mounted in the engine in place of a pressure transducer. It had an optical path length of mm, which proved to be adequate for practical applications. The results contained little vibration noise and provided an adequate signal for analyzing the flame propagation speed. In separate tests, two micro Cassegrain (MC) sensors were installed in the cylinder head near the wall to evaluate the cylinder-to-cylinder fluctuations and the flame propagation characteristics. The local air excess ratio (λ) was estimated from the ratios of the emission intensities using the chemiluminescence and the micro Cassegrain system (MCS) after deriving a calibration function. The averaged values of λ in cylinders 2 and 3 were similar, but cylinder 2 had higher cyclic variations. The cyclic fluctuation of the flame propagation speed in cylinder 2 was also larger, indicating that combustion was more stable in cylinder 3. The measured flame propagation speed increased with the engine speed, but the propagation speeds normalized by the piston speed remained at a constant level. The flame propagation speed increased with and was strongly influenced by the engine speed. Under the high engine-speed conditions found in a racing engine, the flame propagation speed was influenced by the fluid dynamics of the in-cylinder flow. The IR and MC sensors were successfully used to analyze the flame propagation characteristics in a racing engine at high engine speeds and high loads. The MC system detected the radical signals without any problems, despite transient engine operating conditions, and could be used to measure the flame characteristics. 1. Introduction Pressure data are the most useful tool for diagnosing engine performance when developing racing engines. However, the data depend on the robustness and resolution of the sensor [1, 2]. Pressure values can vary from cycle to cycle and from cylinder to cylinder. Since the intake air flow can affect the inertia, the air flow may change during engine operation. Cycle-to-cycle fluctuations during combustion in a spark-ignition (SI) engine seriously affect the engine performance and exhaust gas characteristics. For multi-cylinder engines, the cylinder-tocylinder combustion fluctuations also affect the engine response, output power, and vibration. These cycle and cylinder fluctuations are difficult to control in real time. The fuel concentration around the spark plug and the fluid motion strongly influence the duration of the combustion initiation, causing cycle-to-cycle variations that can become large in lean-burn engines. A diagnostic tool that can determine the gasoline distribution is useful to better understand how to achieve an appropriate local mixture and control large-scale stratification. Instantaneous data for fuel concentration and flame characteristics in firing engines would greatly aid in improving and optimizing engine designs. Ionization probes have been used to investigate combustion in high-speed racing engines. Shimasaki et al. [3] used onboard diagnostics to measure the ionic current at the spark plug to detect misfires, hesitation, and detonation. Pfeffer et al. [4] used twelve ionization sensors to investigate the direction of flame propagation under two different tumble-ratio conditions. A fast flame - 1 -

2 Lisbon, Portugal, 7-1 July, 28 ionization detector (FID) is often used to measure fuel concentrations in an engine cylinder; however, there is some delay in obtaining the measurements [5]. Recently, in situ fuel concentration measurements have been performed by using infrared absorption [6 17]. In particular, a µm He Ne laser has been used to obtain fuel concentrations for combustion diagnostics [6, 7]. Tomita et al. [12, 13] used an optical sensor with a pair of sapphire rods, the tips of which were cut at an angle of π/4 radians, to pass laser light through the combustion chamber of a practical engine, and they discussed some of the factors that affected the measurement accuracy. This sensor has also been applied to practical SI engines and direct-injection gasoline engines [14]. A new infrared (IR) optical spark plug sensor with a double-pass measurement length has also been developed [15, 16]. The measurement accuracy has been confirmed by measurements of the concentration of a homogeneous methane air mixture in a compression expansion engine. The spark plug sensor has been applied to a practical SI engine using isooctane as fuel; the study confirmed that fuel concentrations measured using the sensor agreed with the preset concentrations under firing conditions. An optical method of chemiluminescence measurement has been used for a practical spark ignition engine with a micro Cassegrain system (MCS) [18 21]. An MCS is a useful tool for measuring the local air excess ratio (λ) in the cylinder of a practical SI engine. This method has been used to measure the transient uniformity of the mixture, optimize fuel injection, and examine mixture quality in a racing engine. We are interested in inter-cylinder and cyclic variations associated with the flame front structure in racing engines, as well as the local and time-series values of λ. In this study, an IR absorption measurement system was developed to investigate its potential to support the development of racing engines. We used a small infrared (IR) sensor to measure the amount of laser light absorption by the fuel and the flame arrival timing at the measurement point. Separately, a micro Cassegrain (MC) sensor was used to measure the chemiluminescence inside the cylinder after the flame arrived at the measurement point. Figure 1 compares the IR absorption and chemiluminescence measurement techniques. Both methods can be used to determine flame arrival timing and thus the flame propagation speed. The results obtained from the two methods were compared. IR absorption Chemiluminescence Spark ignition Before combustion Time sequence Line of sight Fuel dependence Pressure dependence Air flow dependence Residual gas dependence Flame propagation Time evolution Local point Excited radical concentration Flame stoichiometry Heat release Temperature Figure 1 Comparison between IR absorption and chemiluminescence measurements - 2 -

3 Lisbon, Portugal, 7-1 July, Methodology IR absorption measurement Hydrocarbons have a strong absorption band close to a 3.4-µm wavelength, due to the stretching vibration of the C-H bonds in the molecular structure. Therefore, when hydrocarbons exist in the measurement region, the light near 3.4 µm is strongly absorbed. When the incident intensity of the light I decays to the value of I through a gas, the transmissivity I/I can be expressed using Lambert-Beer s law as follows: log( I I ) = εcl (1) where ε, C, and L denote the molar absorption coefficient, the molar concentration of the gas absorbing the light, and the measurement length, respectively. When the measurement length L is constant, the molar concentration can be determined by measuring the transmissivity, provided that the molar absorption coefficient of the absorber ε is known. The measurement system is shown in Fig. 2. A He-Ne laser with a µm wavelength (laser power: 1 mw) was used as the light source. The light from the laser was guided into a fluoride fiber (core diameter: 15 µm, length: 3 m) with a manipulator. The other end of the fiber was connected to a micro IR sensor installed on an engine with a FC connector. The laser light passed through the in-cylinder gas at the tip of the sensor and was guided to an InSb IR detector through a band pass filter with a fluoride fiber (core diameter: 4 µm, length: 3 m). The transmitted IR intensity was recorded by a personal computer through an A/D system together with the in-cylinder pressure signals, an encoder signal, and the top dead center (TDC) signal at every crank angle degree. The IR sensor had the same dimensions as an M5 pressure transducer and was installed instead of the pressure transducer. The sensor had two fibers. The IR light from a laser was led to the tip of the sensor through one of the fibers (core diameter: 25 µm). The reflected light was transmitted into the optical component, and then returned via a different path to enter the receiving fiber (core diameter: 425 µm). The fiber diameter and the optical component were selected and designed to Manipulator µm (1 mw) He-Ne laser Optical fiber1 - Fiber length: 3 m - Core diameter: 24µm FC connector Engine (V8) A/D PC Detector -Pressure -TDC - Encoder Manipulator with BPF Optical fiber2 - Fiber length: 3 m - Core diameter: 4µm IR sensor - Fiber length: 1. m M5 Optical fibers Spark plug IR sensor Adapter Cylinder head surface 1.3 mm Measurement region Thermocouple Figure 2 Schematic diagram of experimental setup for IR absorption measurement - 3 -

4 Lisbon, Portugal, 7-1 July, 28 achieve high transmitting efficiency while maintaining their robustness and vibration resistance. The output power from the sensor was almost 1 V, as shown in Fig. 3, which also displays an example of the time series IR signal at 7,5 rpm with a wide open throttle (WOT). These results indicate the high efficiency of the sensor (almost 1 V output with no absorption) and its resistance to vibration. Adequate absorption was obtained around TDC. A thermocouple was installed to measure the temperature 5 mm from the tip of the sensor. The temperature was 2ºC at 1, rpm and 245ºC at 18, rpm. I/I Pressure, MPa rpm TDC TDC TDC TDC TDC Crank angle, deg. #1 #3 1 5 Voltage, V Figure 3 Example of the pressure and IR signal history (75rpm, WOT) Chemiluminescence measurement (micro Cassegrain system) We used a micro Cassegrain system (MCS) to measure the in-cylinder λ and flame propagation speed. The system consisted of small chemiluminescence micro Cassegrain (MC) sensors, a spectrometer, and related software. Photographs of the MC sensors and the spectrometer are shown in Fig. 4. There are several types of these sensors, including M5, spark-plug, and glow-plug sensors. We used M5 MC sensors. The measurement volume was less than.1 mm and the focal length was 3 mm. A metallicized optical fiber capable of withstanding temperatures up to 25 C was used. The spectroscopic unit (SPB1) detected the flame emissions, and was connected to the optical fiber from the MC sensor. The flame emissions were separated into five wavelengths by a dichroic mirror and a band-pass filter (OH*, CN*, CH*, C 2 *(,), and C 2 *(1,)); these were then detected by photomultipliers. The intensities of each radical were acquired by a personal computer (PC) connected to the SPB1 by a USB cable. The local λ and flame propagation speed were calculated using data-analysis software. Figure 5 shows the locations of the micro Cassegrain sensors and pressure transducers. Two micro Cassegrain sensors were installed to measure the flame chemiluminescence of cylinders 2 and 3. Additionally, pressure transducers were installed to allow simultaneous measurement of the pressure data in cylinders 1 and 4. The cylinder pressure was measured using a piezoelectric transducer for 35 cycles at.5-degree crank angle resolution. The measured data were recorded by a combustion analyzer, which also computed the cyclic variation using the coefficient of variation (COV) of the indicated mean effective pressure (IMEP). Our sensors were installed near the engine cylinder wall in place of pressure transducers. The measured optical data were transferred to the signal-processing unit. We conducted engine tests that involved changing the engine speed, preset λ, load, and injection mode. The engine was operated at steady-state conditions from 11, to 18,5 rpm with different values of λ, injection timings, and ignition timings. The flame spectrum and the cylinder pressure were measured under all operating conditions simultaneously

5 Lisbon, Portugal, 7-1 July, 28 Calibration experiments were performed with the preset λ ranging from 12.2 to 14. based on the airflow rate and injection quantity. The mean and root mean square of the fluctuation of numerous samples of the OH*/CH* ratio were calculated from instantaneous values. The measured chemiluminescence intensity ratio was dependent on the λ ratio. The dependency of the OH*/CH ratio on the λ ratio was approximated by a second-order curve fit [18 21]. M5 type Glow plug type Spark plug type Figure 4 Photograph of the micro Cassegrain sensor and SpectraBox (SPB1) Spark plug #1 cylinder #2 #3 #4 Air 39mm Measurement point of MC sensors Location of the pressure sensors #8 #7 #6 #5 MC sensor Adapter Cylinder head surface Spark plug Spark plug 3 mm 39 mm Figure 5 Installation locations of MC sensors in a V8 racing engine 3. Results and Discussions The measurements were carried out on an eight-cylinder racing engine with a four-valve pent-roof cylinder head. Table 1 lists the experimental conditions. The IR absorption and MCS tests were performed separately. The IR sensor was installed in cylinder 2. The MC sensors were installed in cylinders 2 and 3 simultaneously to measure both the cyclic and cylinder-to-cylinder variation. The MCS was also used in a transient test to detect the flame characteristics during transient operations

6 Lisbon, Portugal, 7-1 July, 28 Table 1 Experimental conditions Engine speed, rpm 1 ~ 16 1 ~ 185, Transient Load WOT WOT Measurement point 39mm from the spark plug 39mm from the spark plug Micro IR sensor Cylinder number 2 MC sensor Cylinder number 2 and 3 Figure 6 shows the averaged P-V diagrams and cyclic variations observed in the pressure traces for cylinders 1 and 4 at high-speed full-load conditions. The values were normalized with the mean pressure of cylinder 1. The average P-V patterns differed slightly, but the variation of the peak firing pressures always fluctuated within 2% of the mean pressure. The variation in the peak combustion pressures were attributed to the non-homogenous air-fuel supplied to the engine cylinder. Normally, the combustion phases in each cycle differ for various reasons, such as the mixture formation and in-cylinder flow; these are related directly to the flame characteristics. Hence, the flame characteristics are important indicators of engine performance and can be used to reduce fluctuations in the combustion phase. Figure 6 Cylinder-to-cylinder variations of in-cylinder pressure (15rpm, WOT) Figure 7 illustrates the transmissivity time history obtained from the IR-absorption and CH*- intensity measurement techniques. These are the average results of 2 cycles, and the vibration noise observed in previous reports was absent. There was a very clear and strong IR signal for each cycle. We used fixed reflection optics in the sensor and along the optical paths, and determined that noise due to engine vibration was not present at all tested engine speeds. During the intake stroke, the transmissivity decreased when the rich mixture passed through the measurement region, as indicated by the ellipsoidal dashed line. During the compression stroke, the transmissivity decreased as the in cylinder volume decreased and the molar concentration of the mixture increased. Around TDC, the transmissivity suddenly increased as the flame propagated through the measured region and removed the hydrocarbons. During the exhaust stroke, the transmissivity remained constant at unity because almost no hydrocarbons were present near the measurement site. The IR sensor had very robust characteristics at high engine speeds. There was a peak in the radical intensity when the flame arrived at the measurement point of the MC sensors. The intensity of the CH* peak differed with the engine speed. The ratios of the radical intensities were converted into λ values using the calibration function prepared in the wide λ setup test. The minimum transmissivity and maximum CH* intensity indicated the flame arrival timing. Therefore, the flame propagation speed could be obtained from both the IR and chemiluminescence measurements. These values are compared in Fig

7 Lisbon, Portugal, 7-1 July, 28 IR signal (#2 cylinder) CH* (#2 cylinder) CH* (#3 cylinder) 1. 1rpm Flame arrival timing rpm Transmissivity, a.u rpm 1..5 Intensity, a.u. Transmissivity, a.u rpm 1..5 Intensity, a.u rpm rpm Crank angle, deg. ATDC Crank angle, deg. ATDC Figure 7 Transmissivity measured by IR absortion method and the CH* intensity measured by MCS Figure 8 shows the transmissivity under five conditions ranging from 1, to 16, rpm. The flame propagation speed (arrival time) was calculated from the arrow indicated in the figure using the distance between the spark plug and the measurement point divided by the time from spark ignition to CA (I/I)min, where CA (I/I) min indicates the flame arrival timing at the measurement location. The time of the flame arrival from ignition was denoted as the flame propagation speed, not the local value. Figure 9 shows the flame propagation speed at six engine speeds. The left-hand side of the graph indicates the relationship between the engine speed and flame propagation speed; the right-hand side shows flame propagation speed divided by mean piston velocity (V mean ). As the engine speed increased, the flame propagation speed also increased due to the higher turbulence intensity. However, the flame propagation speed divided by V mean remained almost the same at these six engine speeds. Figure 1 shows the cyclic variation of the flame propagation speed for different engine speeds. The flame propagation speed increased with the engine speed. The fluctuations of the flame propagation speed were larger at lower (1, rpm) and higher (16, rpm) engine speeds; however, these fluctuations were small. Transmissivity, I/I a.u Flame propagation 16,rpm 15,rpm 14,rpm 11,rpm 1,rpm.2 Ignition timing Flame arrival timing at the measurement point CA, deg.atdc Figure 8 Transmissivity measured by IR mothod - 7 -

8 Lisbon, Portugal, 7-1 July, FPS,m/s 25 FPS/Vmean Engine speed, rpm Engine speed, rpm Figure 9 Standard and normalize flame propagation speed obtained from IR measurements rpm 11rpm 12rpm FPS, m/s rpm rpm Cycles 16rpm PDF, % 5 Figure 1 Cyclic variation of the flame propagation speed for each engine speed condition obtained from IR measurements Cylinder-to-cylinder fluctuations during combustion are one of the most significant issues that must be overcome to achieve smooth acceleration and high output power during transient operations and while cornering. The MCS was applied to measure the local λ cylinder difference and the flame propagation speed. Figure 11 shows the measured air-to-fuel ratio for cylinders 2 and 3. Both cylinders had similar averaged values, but cylinder 2 had higher cyclic variation than cylinder 3. Therefore, the combustion was more stable in cylinder 3. In a typical racing engine, the combustion process has to occur at high engine speeds. These inter-cylinder variations have a detrimental effect on engine performance. Figure 12 compares the flame propagation speed calculated from the radical signal intensities under the same conditions as used for Fig. 11. The variation in cylinder 3 was smaller than in cylinder 2, and the average speed was higher. This reinforces the finding that the combustion was more stable in cylinder

9 Lisbon, Portugal, 7-1 July, 28 #2 cylinder #3 cylinder λ λ λ λ #3-#2 cylinder (Ave, rms)=(.91,.3) (Ave, rms)=(.9,.2) 5 (Ave, rms)=(-.1,.4) Cycle PDF, % Figure 11 Cylinder-to-cylinder variations of the air fuel ratio measured by MCS (18,rpm) #2 cylinder #3 cylinder FPS, m/s FPS, m/s #3-#2 cylinder FPS. FPS, m/s (Ave, rms)=(41.5, 2.) (Ave, rms)=(43.6, 1.) 5 (Ave, rms)=(2.1, 2.2) Cycle PDF, % Figure 12 Cylinder-to-cylinder variations of the flame propagation speed measured by MCS (18,rpm) Figure 13 shows the preset λ effects on the flame propagation speed. The flame propagation speed was significantly affected by the engine rotation speed. Similar flame propagation speeds were obtained for a given engine rotation speed. Figure 14 compares the effect of engine speed on the flame propagation speed measured by the IR and MC sensors. The flame propagation speed was significantly affected by and increased with the engine rotation speed. The speed obtained using the MC sensors was 5 1% higher than that obtained using the IR absorption method. The increasing flame propagation speed and IMEP had similar tendencies, and were reduced near 15, and 18, rpm due to the dynamic compression at high engine rotation speeds. As a result of the valve timing, the intake resistance and all other factors that affected the cylinder filling created supercharging conditions due to the ramming effect. The intake gas temperature and density were also affected by the dynamic compression. The air pressure waves in the intake pipe at 15, and 18, rpm decreased the volumetric charging efficiency, causing acoustic resonance. This decreased the flame propagation speed. Therefore, the flame propagation speed was influenced by - 9 -

10 Lisbon, Portugal, 7-1 July, 28 the volumetric efficiency factors. Absorbance measured by IR absorption measurement also has the same tendency with the flame propagation speed and IMEP. The absorbance has the local minimum value around the 15 rpm and this was related with the low volumetric efficiency. In a normal engine, the flame propagation speed is related to the mixture distribution, mean λ, and in-cylinder flow characteristics. The increased turbulence intensity with higher inlet flow velocities was the predominant cause of the increased flame propagation speed. But in a racing engine, the flame propagation speed is not dominated by turbulence intensity; it also depends on the fluid dynamics, such as the intake air temperature, density, turbulence, flame stretch, heat release, and the presence of non-uniformities. This is the reason for the weak effect of the λ on the flame propagation speed in Fig.13. Figure 15 shows transient mode test results for 1, to 18,rpm. The micro-cassegrain system could detect radical signal without problems, and could measure flame characteristics. The maximum pressure and flame propagation speed had similar tendencies, as shown in Fig. 14, and the peak cylinder pressure increased with the flame propagation speed. Figure 13 Effect of preset λ on the flame propagation speed 5 FPS, m/s MCS, #3 cylinder MCS, #2 cylinder IR, #2 cylinder IMEP, a.u..8.6 Absorbance, a.u #1 cylinder 3 deg.btdc 74 deg.btdc #2 cylinder 15 2 Engine speed, rpm Figure 14 Comparison of the flame propagation speed between IR and MCS measurements with IMEP and absorbance - 1 -

11 14th Int Symp on Applications of Laser Techniques to Fluid Mechanics Lisbon, Portugal, 7-1 July, 28 Figure 15 Transient test results measured by MCS 4. Conclusions Optical diagnostic IR absorption and chemiluminescence measurement techniques were applied to a racing engine. We developed a small IR sensor to measure local fuel concentrations and thereby determine and reduce the sources of pressure fluctuation. The small IR sensor was able to withstand racing engine operating conditions up to 16, rpm and exhibited little vibration noise. A small Cassegrain optics sensor was also developed to measure the local chemiluminescence intensities of OH*, CH* and C2* in a racing engine to evaluate the cylinder-to-cylinder fluctuation and the flame propagation characteristics; two of these sensors were installed near the wall in cylinders 2 and 3. The combustion pressure of cylinders 1 and 4 fluctuated from cycle to cycle. There was also a cylinder-to-cylinder variation. The flame characteristics are important indicators of the engine performance and can be used to reduce fluctuations in the combustion phase. The IR absorption measurements results indicated little vibration noise and provided an adequate signal for analyzing the flame propagation speed. The measured flame propagation speed increased with engine speed, but propagation speeds normalized by the piston speed remained constant. The local λ could be estimated from the ratios of the emission intensities using the chemiluminescence measurement technique with the MCS after deriving a calibration function. The averaged local λ values of cylinders 2 and 3 were similar, but cylinder 2 had higher cyclic variations than cylinder 3. The cyclic fluctuation of the flame propagation speed in cylinder 2 was also larger. Therefore, the combustion in cylinder 3 was more stable. The flame propagation speed was strongly influenced by and increased with the engine speed. At the high engine speeds found in a racing engine, the flame propagation speed is influenced by the fluid dynamics of the in-cylinder flow. The acoustic resonance in the intake pipe at 15, and 18, rpm decreased the flame propagation speed

12 Lisbon, Portugal, 7-1 July, 28 Both the IR and MC sensors could be used to analyze flame propagation characteristics in a racing engine at high engine speeds and high loads. The MCS system detected the radical signals without any problems, despite transient engine operating conditions, and was able to measure the flame characteristics. ACKNOWLEDGEMENT The authors are pleased to acknowledge the contribution of Dr. Ing. Vincent Pérémé, Dr. Ing. Gilles Simon and Ferrari technical staffs. References [1] G. H. Neo, and N, Collings: SAE Paper 9761, (1997). [2] H. Alten, and M. Illien: SAE Paper , (22). [3] Y. Shimasaki, H. Maki, J. Sakaguchi, N. Kondo and T. Yamada: SAE Paper , (24). [4] T. Pfeffer, P. Bühler, D. E. Meier, and Z. Hamdani: SAE Paper , (22). [5] H. Zhao, and N. Ladommatos, Engine Combustion Instrumentation and Diagnostics, Society of Automotive Engineers, Inc., (21). [6] R. Mongia, E. Tomita, F. Hsu, L. Talbot and R. Dibble: Proc. Combust. Inst., 26, p (1996) [7] J.G. Lee, K. Kim, and D.A. Santavicca : Proc. Combust. Inst., 28, p (2) [8] S. Yoshiyama, Y. Hamamoto, E. Tomita, and K. Minami: JSAE Review 17(4), p (1996) [9] M.J. Hall, and M. Koenig: Proc. Combust. Inst., 26, p (1996) [1] M. Koenig, and M.J. Hall: SAE paper , (1997). [11] K. Kawamura, T. Suzuoki, A. Saito, T. Tomoda, and M. Kanda: JSAE Review 19(4), p (1998) [12] E. Tomita, N. Kawahara, M. Shigenaga, A. Nishiyama, and R.W. Dibble: Meas. Sci. Technol. 14(8), p (23) [13] E. Tomita, N. Kawahara, A. Nishiyama, and M. Shigenaga: Meas. Sci. Technol. 14(8), p (23) [14] E. Tomita, N. Kawahara, S. Yoshiyama, A. Kakuho, T. Itoh, and Y. Hamamoto : Proc. Combust. Inst., 29, p (22) [15] A. Nishiyama, N. Kawahara, and E. Tomita, : SAE paper , (23). [16] A. Nishiyama, N. Kawahara, and E. Tomita, SAE paper , (24). [17] N. Kawahara, E. Tomita, A. Nishiyama, and K. Hayashi, SAE paper, , (26). [18] Ikeda, Y., Kawahara, N., and Tomita, E., Proc. of 6th COMODIA 24, Paper No.C3-3, in CD-rom (24). [19] Ikeda, Y., Nishiyama, A., Kawahara, N., Tomita, E., Arimoto, S., and Takeuchi, A., SAE Paper, No (25). [2] Kawahara, N., Tomita, E., Arimoto, S., Takeuchi, A., Ikeda, Y., and Nishiyama, A., SAE Paper, No (25). [21] Ikeda, Y., Nishiyama, A., Kim, S.M., Takeuchi, A., Winklhofer, E., and Baritaud, T., Proc. of 7th international symposium on internal combustion diagnostics, pp , (26)