International journal of automotive technology, tập 10, số 1, 2009

International Journal of Automotive Technology, Vol 10, No 1, pp 1−7 (2009) DOI 10.1007/s12239−009−0001−9 Copyright © 2009 KSAE 1229−9138/2009/044−01 EFFECTS OF MIXTURE STRATIFICATION ON HCCI COMBUSTION OF DME IN A RAPID COMPRESSION AND EXPANSION MACHINE G S JUNG1), Y H SUNG1), B C CHOI2) and M T LIM2)* 1) Graduate School of Mechanical Engineering, Chonnam National University, Gwangju 500-757, Korea School of Mechanical Systems Engineering, Chonnam National University, Gwangju 500-757, Korea 2) (Received April 2008; Revised 10 June 2008) ABSTRACT−Compression ignition of homogeneous charges in internal combustion (IC) engines is expected to offer high efficiency of DI diesel engines without high levels of NOx and particulate emissions This study is intended to find ways of extending the rich limit of HCCI operation, one of the problems yet to be overcome Exhaust emissions characteristics are also explored through analyses of the combustion products DME fuel, either mixed with air before induction or directly injected into the combustion chamber of a rapid compression and expansion machine, is compressed to ignite under various conditions of compression ratio, equivalence ratio, and injection timing The characteristics of the resulting combustion and exhaust emissions are discussed in terms of the rate of heat release computed from the measured pressure, and the concentrations of THC, CO, and NOx are measured by FT-IR and CLD The experimental data to date show that operation without knock is possible with mixtures of higher equivalence ratio when DME is directly injected rather than when it is inducted in the form of a perfectly homogeneous fuel-air mixture Although fuel injected early in the compression stroke promotes homogeneity of the DME-air mixture in the cylinder, it causes the mixture to ignite too early to secure good thermal efficiency and knockfree operation at high loads Low temperature reactions occur at about 660K regardless of the fueling methods, fuel injection timing and equivalence ratio The main components of hydrocarbon emissions turned out to be unburned fuel (DME), formaldehyde and methane KEY WORDS : DME, HCCI, RCEM, Exhaust emissions, Perfectly homogeneous fuel-air mixture, Direct fuel injection INTRODUCTION 2005) have been suggested to resolve this difficulty by relaxing almost simultaneous heat release of HCCI combustion In actual engines, various causes for non-homogeneity exist including imperfect mixing of fuel, air, and residual or EGR gas, differences in gas flow and boiling points of fuel blends, and varying levels of heat transfer along the cylinder walls The objective of this study is to investigate the effects of stratification caused by uneven fuel-air mixing on the HCCI combustion and consequent emissions Different ways of adding fuel to air are tried including premixing and direct injection at various timing A rapid compression and expansion machine (RCEM) is used in this study in order to focus on the effects of fuel-air mixing caused stratification with minimized contribution of other factors like residual gas and gas motion DME (di-methyl ether) is the selected test fuel, since its combustion-related properties are close to those of diesel fuel, and it burns cleanly without generating soot DME also exhibits two-stage heat release, which is one of the distinctive characteristics of HCCI combustion (Ogawa et al., 2003; Teng et al., 2004; Yao et al., 2003) New technologies that increase thermal efficiency and clean exhaust gas of automobile engines are desperately being sought to mitigate the problems of energy shortage & environmental pollution Homogeneous charge compression ignition (HCCI) is one of those representative technologies being developed in the field of engine combustion HCCI engines are expected to have higher efficiency than SI engines due to their high compression ratio and dispensability of throttle valves Compared to diesel engines, they emit less particulate matters and less NOx because only lean premixed combustion without local fuel-rich zones is present (Gray and Ryan, 1997; Thring, 1989; Chung et al., 2008) Despite these advantages, commercial mass production of HCCI engines have not yet been realized One of the obstacles to the commercialization of HCCI engines is objectionable knock occurring under heavy load conditions Knock results from an excessive rate of pressure rise in the combustion chamber (Gray and Ryan, 1997) Methods of utilizing mixture stratification (Inagaki et al., 2006; Kumano and Iida, 2004; Sjöberg and Dec, 2006) and thermal stratification (Dec et al., 2006; Lim et al., 2006; Sjöberg et al., EXPERIMENTAL SYSTEM AND TEST SCOPE As shown in Figure 1, the experimental system consists of *Corresponding author e-mail: mtlim@chonnam.ac.kr G S JUNG, Y H SUNG, B C CHOI and M T LIM Table Specifications of RCEM Figure Schematic diagram of the experimental system Item Specification Combustion chamber Bore/Stroke Displacement volume Connecting rod length Crank Radius Compression ratio Top clearance height Disc type 100/450 mm 3.534 dm3 900 mm 225 mm 8~23 17.5~61.8 mm Table Specifications of fuel injector Item Specification Valve Lift 45~50 μm Nozzle Open pressure Needle lift Seat diameter Nozzle holes 5~6 MPa 0.35 mm Ø 2.25 × Ø 0.22 Table Experimental conditions Premixed Direct injection Figure Definitions of LTR and HTR an RCEM and subsystems for fuel supply and injection, for heating cylinder charge, and for acquisition of pressure and temperature signals Basic structure of the RCEM (specifications shown in Table 1) is similar to that of a reciprocating piston engine The RCEM is a single-shot engine, whose crankshaft is driven by the kinetic energy stored in a large flywheel Before the combustion test, residual gas is removed from the combustion chamber by a vacuum pump, and fresh air is charged Design and specifications for this RCEM are explained elsewhere in detail (Cho et al., 2004) The initial temperature of the air or fuel-air mixture in the combustion chamber of the RCEM is controlled by operating the electric heater wrapped around the cylinder head, and also by circulating heated water through the jacket around the cylinder The DME fuel is supplied to the combustion chamber either premixed with air or by direct injection Fuel and air are mixed and stabilized for perfectly homogeneous charge tests approximately a day before use Direct injection of liquid DME is accomplished at 50 MPa by controlling the timing and duration of the injector open time Table provides the major specifications of the fuel injector in detail An in-house designed computer program performs the injection control based on the crank angle signal from a rotary encoder (Autonics, E40S-360-3-5) Since viscosity of DME is insufficient for self-lubrication of the nozzle (0.33 cSt at 293 K and 50 MPa), 750 ppm of viscosity improver (ETHYL, hitec-4140, 17.2 cSt) is added to DME (Longbao et al., 1999) Fuel Revolution speed Compression ratio Initial temperature Fuel supply pressure Equivalence ratio, ø Injection timing DME 100 rpm 12 303 K − 0.1, 0.2, 0.25, 0.3 − 50 MPa 0.1, 0.2, 0.25, 0.3, 0.35, 0.4 BDC, 150°, 120°, 90°, 60°, 30° BTDC Combustion tests were run with fuel introduced either premixed or directly injected into the combustion chamber under various conditions of equivalence ratio and injection timing as described in Table Combustion characteristics are analyzed using the pressure in the combustion chamber taken by a piezoelectric sensor (Kistler, 6016B) and a charge amplifier (Kistler, 5011A) Exhaust gas or combustion product is displaced out of the cylinder and fed to the analyzers after each run of the combustion tests Engine exhaust gas analyzers are inapplicable to RCEM tests because RCEM conducts only one cycle at a time, producing a fixed volume of the exhaust gas (about 3.5 dm3 in the present case) which is too small to keep a continuous sample flow to the analyzers Concentrations of unburned hydrocarbon (HC) and carbon monoxide (CO) are therefore measured by an FT-IR (MIDAC, I2004) that requires about 0.3 dm3 of sample gas, and oxides of nitrogen (NO and NOx) are measured by a CLD (Thermo Environmental Instruments, 42C-HL) that requires about 25 cc/min of sample gas EFFECTS OF MIXTURE STRATIFICATION ON HCCI COMBUSTION OF DME IN A RAPID COMPRESSION 3 EXPERIMENTAL RESULTS AND DISCUSSIONS 3.1 Characteristics of HCCI Combustion LTR (low temperature reactions) and HTR (high temperature reactions) in HCCI combustion are identified in this study as the first and the second phases of the oxidation process on ROHR (rate of heat release) curves as depicted in Figure The particular curve in this example is the one actually obtained from the pressure record of a combustion test using a perfectly premixed DME-air mixture with equivalence ratio of 0.2 The LTR and HTR are found to be associated with formation of formaldehyde and formation of CO and CO2 respectively (Hamada et al., 2005) Figure 3(a) shows a group of curves representing respectively the motored cylinder pressure, and fired cylinder pressure along with the associated ROHR and the gas temperature during a test run at an equivalence ratio of 0.1 This figure gives an overall idea about what happens during the entire period of compression and expansion strokes in the combustion chamber during the test Figure 3(b) shows the similar sets of data for the combustion tests run at equivalence ratios of 0.1 and 0.3, but the range of data is restricted only to the late phase of compression and early phase of expansion The blue curves in the latter figure denote the results from the tests run with perfectly homogeneous fuel-air mixture supplied, while the red lines denote those run with fuel directly injected at the start of compression (i.e at bottom dead center) The combustion pressure in figure (a) gradually rises at first during the compression, just tracing the motored cylinder pressure before the start of reactions at about 19°BTDC, and then it rapidly increases thereafter The in-cylinder gas completes combustion, and its temperature reaches 930 K at TDC The two peaks of LTR and HTR appear after ignition on ROHR curve Although auto-ignitions are observed to occur at about 20o~15o BTDC, in all the cases shown in Figure 3, the early burning rate in the cases of the premixed combustion is much larger, resulting in a shorter combustion duration and higher peak pressure On comparison of the cases in Figure 3(b) where the equivalence ratio is 0.3, the cylinder pressure during combustion rises fast enough to cause obvious knock in the premix tests as characterized by the sharp peak and the fluctuations in the pressure curve, but knock is not observed in the case of direct injection test Figure shows the maximum rate of combustion pressure measured in tests with various equivalence ratio and fueling strategies In the cases of premixed mixture the maximum rate of pressure rise exceeds 4.3 MPa/ms (Lim et al., 2006), which is taken as knock borderline in this study, at equivalence ratios over 0.25 However, knock is not seen to take place at equivalence ratios of up to 0.4 when fuel is directly injected at various injection timing This implies that the rich operation limit of HCCI combustion can be extended by direct injection of fuel The effect may be attributed to the stratification of the mixture and latent heat Figure Cylinder pressure, temperature and rate of heat release as a function of crank angle for DME-air mixtures of different equivalence ratios and mixing processes of vaporization The maximum rate of pressure rise seems to gradually increase as either the equivalence ratio increases, or the fuel injection is retarded at equivalence ratio over 0.35 Figure represents the timing and duration of LTR and HTR in the tests with various fuel injection timing and equivalence ratios The left, blue portions of the bars indicate time for LTR while the right, red portions indicate this for HTR Although the ignition delay is significantly shortened, the start of LTR and HTR is delayed less than 5° CA as the fuel injection is retarded from BDC to 30° BTDC When compared with the premixed fuel-air mixtures, the ignition starts later, and combustion lasts longer when mixtures are formed through direct injection The ignition timing, taken as start of HTR, is slightly delayed when fuel G S JUNG, Y H SUNG, B C CHOI and M T LIM Figure Maximum rate of pressure rise and knock as a function of crank angle for DME-air mixtures of different equivalence ratios, formed by premixing or direct injection Figure Released heat and combustion efficiency as a function of equivalence ratio and injection timing Figure Combustion temperature as a function of equivalence ratio and injection timing Figure Combustion timing as function of fuel injection timing and equivalence ratio injection is substantially delayed The delay is bigger as the equivalence ratio increases, possibly due to the cooling effect of the fuel vaporization and also due to the higher degree of mixture stratification Mixtures of greater equivalence ratio also have shorter combustion durations (especially for HTR) Figure shows net heat release and combustion efficiency in tests of mixtures with various equivalence ratios, where fuel is either premixed or directly injected at different crank angles Hatched columns in the figure indicate total amount of released heat, while green curves represent the corresponding combustion efficiency The combustion efficiency is rather low, between 40 and 70 percent, with higher values associated with greater equivalence ratio Further observation reveals that the combustion efficiency does not depend on the fuel injection timing at equivalence ratio of 0.2 or 0.3, but it gradually decreases at equivalence ratio of 0.1 as the injection is retarded from BDC toward 30° BTDC Figure shows bulk combustion temperature of the incylinder mixtures during the combustion tests under various fueling strategies The blue bottom portion, the red upper portion, and the highest point of the bars, respectively, indicate the temperatures during LTR, during HTR, and at the final stage of combustion The in-cylinder gas temperature before the ignition is calculated from the equation (1) assuming an adiabatic compression process because the fuel-air mixture in the central part of the combustion chamber, where the ignition is most likely to occur first, gets least influenced by the heat loss to the walls (Iida et al., 2004; Sato et al., 2003) P ( θi ) ⎞ T ( θ i )=T ( θ i – ) ⋅ ⎛⎝ P ( θi – ) ⎠ γ–1 γ (1) Since the adiabatic assumption included in the isentropic process is not valid once ignition has occurred, the gas temperature is calculated using the equation (2), the ideal gas equation of state EFFECTS OF MIXTURE STRATIFICATION ON HCCI COMBUSTION OF DME IN A RAPID COMPRESSION Figure Imep as a function of fuel injection timing and equivalence ratio P ( θi ) ⋅ V ( θ i ) ⋅ n ( θi – ) ⋅ T ( θi – ) T ( θ i )= P ( θ i – ) ⋅ V ( θi – ) ⋅ n ( θi ) (2) T [K], P [Pa], V [m3], n [mol], γ, and θ [deg.] in the above equations represent the temperature, pressure, volume, specific heat ratio of the gas in the cylinder, and the engine crank angle, respectively A specific heat ratio of 1.3 was obtained from the logarithmic plot of the measured pressure and volume pertaining to the engine operating conditions It is clear from the figure that LTR begins at about the same temperature of 660 K ± 10 K regardless of either the equivalence ratio or the fuel injection timing Figure shows indicated mean effective pressure (imep) obtained under various fueling conditions As the fuel injection is retarded from BDC to 30° BTDC, imep steadily increases a little (up to about 0.05 MPa) except at the leanest condition The general trend of increasing imep with delayed ignition, caused by retarded fuel injection, is attributed to the decreasing compression work due to the delay in pressure rise during the compression stroke Figure Concentration of unburned hydrocarbons in combustion products of DME-air mixtures formed by premixing or direct injection 3.2 Pollutants Formed in HCCI Combustion HCCI engines produce significantly more hydrocarbons than conventional diesel engines, contrary to the early expectations Figure shows concentrations of total hydrocarbons and the three major species measured in combustion products of the tests under various fueling strategies As can be seen in the figure, hydrocarbons consist mostly of the unburned DME fuel, formaldehyde and methane Among the three fueling scenarios for each of the same equivalence ratio, the case of premixed fuel has the highest concentration of THC in Figure followed by the case of earlier fuel injection The trend can be easily explained since fuel injected in the center of the combustion chamber at an earlier timing will have more chance of getting trapped in the top ring crevice, which is considered to significantly contribute to DME emission in HCCI operation Among the cases where fuel is supplied in the same manner, THC seems to decrease as the equivalence ratio is raised from 0.1 to 0.2, and then slightly increases (premix fueling), or stays about at the same level (fuel injected at BDC) for further increases of equivalence ratio from 0.2 to 0.3 The major decrease of THC in the first interval of equivalence ratio results from a similar drop in DME Since equivalence ratio 0.1 is close to the lean limit of auto ignition, which will tend to help more DME near the walls or in the crevices avoid combustion, the highest concentration of DME and the low combustion efficiency is measured at equivalence 0.1 When the equivalence ratio is increased to 0.2 or above, more DME will burn to reduce THC in the exhaust gas, while more DME becomes trapped in the crevices to increase THC The latter effect will play a more important role when the cylinder charge is more homogeneous, as explained above, and also become more pronounced at conditions of greater equivalence ratio This may be the factor to explain the significant, in the case of premixed fuel, or marginal, in the case of fuel injected at BDC, increase of DME and also THC at the equivalence ratio of 0.3 Summarizing the above discussions, one may cautiously model HC formation in these cases so that the prevailing mode in the very lean range of the equivalence ratio below Figure 10 NOx and CO concentrations in combustion products of DME-air mixtures formed by premixing or direct injection G S JUNG, Y H SUNG, B C CHOI and M T LIM 0.1 is bulk quenching of the extra-lean mixture due to incomplete ignition or cold walls in the neighborhood, while the crevice volume effect is the dominant one in the richer range Figure 10 shows NOx and CO concentrations in the combustion products of mixtures formed through either premixing or direct injection Smaller amounts of CO are generated at higher equivalence ratios regardless of the fueling strategy, most likely due to higher combustion temperature associated with the greater equivalence ratios Premixed mixtures look to generate a little more CO at every equivalence ratio than mixtures prepared by direct injection This is considered quite natural because more fuel will exist in the close vicinity of the cold walls if fuel is premixed before induction than if it is injected later near the center of the cylinder NOx concentration appears to be roughly proportional to equivalence ratio, which is logical because of the expected variation of the combustion temperature Nevertheless the absolute levels of NOx concentration are fairly low in comparison with those in the conventional diesel combustion CONCLUSIONS Characteristics of DME HCCI combustion and composition of the combustion product were experimentally investigated, while various quantities of DME fuel were introduced by different methods in the combustion chamber of an RCEM Major findings of the study are as follows (1) The knock-limited rich limit of HCCI operation is expanded by directly injecting fuel (2) A homogeneous DME-air mixture is associated with shorter combustion duration and higher peak pressure than a mixture formed by direct fuel injection (3) The timing of LTR is slightly delayed when fuel injection timing is retarded (4) Imep increases with delayed timing of fuel injection (5) The main components of THC are unburned fuel (DME), formaldehyde and methane, and more fuel remains unburned when fuel is supplied premixed (6) More THC and CO are formed for any equivalence ratio tested when fuel is introduced premixed than when it is directly injected at BDC (7) As fuel injection is retarded, ignition occurs later with less THC and more CO formed (8) As equivalence ratio is raised, THC emission decreases in the low range of equivalence ratio (0.1 to 0.2), but it increases (fuel premixed) or levels off (fuel injected) in the higher range of equivalence ratio (0.2 to 0.3) More nitrogen oxides are generated as the equivalence ratio increases, but their absolute levels are quite low ACKNOWLEDGEMENT−This research was conducted as a part of “Development of Basic and Practical Technology for HCCI Engine” under the financial sponsorship of the Ministry of Knowledge Economy REFERENCES Cho, S H., Kim, K S and Lim, M T (2004) Development of a rapid compression expansion machine and compression ignition combustion of homogeneous premixtures Korean Society of Automotive Engineers 12, 2, 83−90 Chung, J W., Kang, J H., Kim, N H., Kang, W and Kim, B S (2008) Effects of the fuel injection ratio on the emission and compression performances of the partially premixed charge compression ignition combustion engine applied with the split injection method Int J Automotive Technology 9, 1, 1−8 Dec, J E., Hwang, W and Sjoberg, M (2006) An investigation of thermal stratification in HCCI engines using chemiluminescence imaging SAE Paper No 2006-011518 Gray, A W and Ryan, T W (1997) Homogeneous charge compression ignition (HCCI) of diesel fuel SAE Paper No 971676 Hamada, K., Niijima, S., Yoshida, K., Shoji, H., Shimada, K and Shibano, K (2005) The effects of the compression ratio, equivalence ratio, and intake air temperature on ignition timing in an HCCI engine using DME fuel SAE Paper No 2005-32-0002 Iida, N., Yamasaki, Y., Sato, S., Kumano, K., Kojima, Y (2004) Study on auto-ignition and combustion mechanism of HCCI engine SAE Paper No 2004-32-0095 Inagaki, K., Fuyuto, T., Nishikawa, K., Nakakita, K and Sakata, I (2006) Dual-fuel PCI combustion controlled by in-cylinder stratification of ignitability SAE Paper No 2006-01-0028 Kumano, K and Iida, N (2004) Analysis of the effect of charge inhomogeneity on HCCI combustion by chemiluminescence measurement SAE Paper No 2004-011902 Lim, O T., Nakano, H and Iida, N (2006) The research about the effects of thermal stratification on n-heptane/ iso-octane-air mixture HCCI combustion using a rapid compression machine SAE Paper No 2006-01-3319 Longbao, Z., Hewu, W., Deming, J and Zuohua, H (1999) Study of performance and combustion characteristics of a DME-fueled light-duty direct-injection diesel engine SAE Paper No 1999-02-3669 Ogawa, H., Miyamoto, N and Yagi, M (2003) Chemicalkinetic analysis on PAH formation mechanisms of oxygenated fuels SAE Paper No 2003-01-3190 Sato, S and Iida, N (2003) Analysis of DME homogeneous charge compression ignition combustion JSAE Paper No 20030236 Sjöberg, M and Dec, J E (2006) Smoothing HCCI heatrelease rates using partial fuel stratification with twostage ignition fuels SAE Paper No 2006-01-0629 Sjöberg, M., Dec, J E and Cernansky, N P (2005) Potential of thermal stratification and combustion retard for reducing pressure-rise rates in HCCI engines, Based on multi-zone modeling and experiments SAE Paper No EFFECTS OF MIXTURE STRATIFICATION ON HCCI COMBUSTION OF DME IN A RAPID COMPRESSION 2005-01-0113 Teng, H., McCandless, J C and Schneyer, J B (2004) Thermodynamic properties of dimethyl ether - An alternative fuel for compression-ignition engines SAE Paper No 2004-01-0093 Thring, R H (1989) Homogeneous-charge compressionignition engine SAE Paper No 892068 Yao, M., Zheng, Z., Xu, S and Fu, M (2003) Experimental study on the combustion process of dimethyle ether (DME) SAE Paper No 2003-01-3194 International Journal of Automotive Technology, Vol 10, No 1, pp 9−16 (2009) DOI 10.1007/s12239−009−0002−8 Copyright © 2009 KSAE 1229−9138/2009/044−02 EXPERIMENTAL INVESTIGATION OF CHARACTERISTICS OF PRESSURE MODULATION IN A FUEL INJECTION SYSTEM D HUANG1,2)* and M.-C LAI1) 1) Mechanical Engineering Department, Wayne State University, Detroit, Michigan 48202, USA 2) Mechanical Engineering Department, Shanghai University, Shanghai 200072, China (Received 25 July 2007; Revised 24 April 2008) ABSTRACT−A piezoelectric atomization device achieves fuel pressure modulation through vibration of a piezoelectric pressure modulator As a consequence, the fast alternating and slow moving streams collide with each other and further break up the fuel drop In this paper, an experimental investigation was carried out to study the fluid dynamic characteristics of the spray atomization process of automotive port fuel injectors with a piezoelectric pressure modulator The investigation mainly focuses on: (a) the coupling characteristics between the piezoelectric stack and the hydraulic as well as the transfer characteristics of pressure modulation from the piezoelectric modulator to the point above the orifice; (b) the time history of the pressure dynamic response at the point above the orifice under a typical modulation frequency, which reflects the variation of pressure modulation while the fuel injector is working; and (c) the time-variation characteristics related to mechanical structure and fluid dynamics The experimental results expose some important dynamic characteristics of pressure modulation, which will be very significant and lead us to greatly improve the fuel injection system, optimize the control parameters and implement spray atomization with a high quality performance in the near future KEYWORDS : Pressure modulation, Fuel injection system, Dynamic characteristic, Spray atomization INTRODUCTION 330 Hz) produced by a pressure modulation machine Although different models (Ren and Mally, 1996) have been developed to describe the fuel injection system, few can accurately predict the transient response of a fuel injection system, especially its internal pressure modulation In order to fill the defect of the numerical computation, the present paper is experimentally investigating the fluid dynamic characteristics in a fuel injection system, which contains a piezoelectric modulator This study was motivated by the success of earlier work on fuel spray atomization in the automotive port injector After carefully observing the coupling phenomenon of the piezoelectric-hydraulic system and the dynamic response of the fuel injection system with pressure modulation, the characteristics of pressure modulation in a fuel injection system are summarized Along with this summary, this manuscript also describes the experimental setup, procedures and signal processing employed in the experiments Experimental results and future work are then discussed Spray atomization is very important to reduce the fuel/air mixing time and minimize the attachment of liquid fuel to port surfaces in automotive and turbo fan engines Wellatomized gasoline spray has a high potential to reduce hydrocarbon emissions and improve the engine cold starting properties Dressler developed a pressure modulator based on the piezoelectric principle (Dressler, 1993) The pressure modulator is installed inside the fuel line and generates the pressure modulation It efficiently enhances the atomization characteristics of gasoline spray injectors (Zhao et al., 1996; 1995; 2002; Sipperley et al., 1998; Schiller et al., 2006; Kim at al., 2004) To optimize the performance of the pressure modulation fuel system in spray atomization (Hu and Wu, 2001a; 2001b), presented mathematical modeling of an individual injector and an entire fuel injector system by considering one-dimensional, unsteady Bernoulli’s equation and loss factors of kinetic energy Kf and Ko based on a discrete segment of the injector (loss factors Kf and Ko are used to account for the losses of kinetic energy as fluid enters the injector through the filter at the top and discharges through the orifice at the bottom, respectively) (Miller, 1990) They predicted the dynamic response of an automotive fuel system under low frequency pressure fluctuation (around EXPERIMENTAL SETUP Figure shows the scheme of an experimental pressure modulated port fuel injector system The pressure modulator, as shown in Figure 1(a), is composed of a piezoelectric driver mounted inside a circular housing and bolted to one end The other end of the circular housing is closed by a rigid endplate upon which the fuel injectors are *Corresponding author e-mail: hdishan@shu.edu.cn 10 D HUANG and M.-C LAI Figure Schematic of experimental setup mounted by means of an adapter The piezoelectric driver consists of a basemount, a pair of piezoelectric disks, and a piston A hollow bolt clamps these items together into a cylindrical assembly The liquid from the fuel tank, pressurized by high pressure nitrogen gas, is introduced into the piezoelectric driver through the basemount It flows through the hollow bolt and a passage in the piston and then enters the fluid manifold surrounding the piston The fluid flows through a hole in the end plate and the injector adapter, and then enters the gasoline port injector The piezoelectric elements receive the electrical signal and convert it into a longitudinal motion Such a process produces a transient variation in the pressure of the fluid as it passes through this gap, allowing for pressure modulation of the fuel system Therefore, the device here is referred to as a pressure modulator It should be denoted that even though there is significant pressure perturbation inside the fuel line, the liquid flow rate remains nearly constant While the device can operate at many frequencies, the optimized operating point can be determined in order to minimize energy loss and maximize pressure perturbation The device has a small size and can be installed easily between the port fuel injector and the fuel line There is no extra control difficulty associated with this system since the fuel injection rate and injection timing are controlled by the conventional fuel injector metering valve It should be noted that while ultrasonic atomizers are only advantageous for atomization at low flow, the pressure modulator technique can provide good atomization over a wide range of fuel flow The working fluid used in the experiment is surrogate fuel (Viscor 16B), which has the same viscosity and density as gasoline, but is more safe to use in tests of fluid dynamics and atomization As shown in Figure 1(b), the system consists of a center port injection (CPI) injector developed by (Zizelman et al., 1992) The CPI injector has a single spray hole with a diameter of 500 μm, and is connected to the pressure modulator through an adaptor The baseline case is the spray with static pressure of 276 kPa (40 psi) regulated by the pressure of the nitrogen gas The pressure modulator receives a sinusoidal signal from a wave generator operating between and 30 kHz, and the signal is amplified by an amplifier before it is used as a modulation voltage driving The final signal delivered to the crystal stacks is a 25~250 V sine wave When the CPI injector is controlled by a pulse wave, it alternatively opens and closes and the fuel passes through the orifice as instructed The pulse width is set at 16 ms for the pressure modulation test To measure the fluid dynamic pressure in the piezoelectric pressure modulator and injector, three KulteXT123C-100 pressure sensors are installed The first one is located at the point above the injector orifice and its responding pressure is P1, which directly affects spray atomization The second one is near the entrance of the injector chamber and its responding pressure is P2 The third one is located in the middle of the pressure modulator and its responding pressure is P3, which directly reflects the source of pressure modulation All sensors are connected to computer data acquisition A computer collects data from the measuring points and processes the data for different statuses With this setup, one can simulate various working conditions with different modulation frequencies, extract the dynamic characteristics of pressure modulation in the fuel injection system and observe the fuel spray atomization COUPLING CHARACTERISTICS The measurement of the coupling characteristic between the piezoelectric stack and the hydraulic system is carried out with the port injector being closed The single frequency sine wave is applied to the piezoelectric stack, and as a result, the pressure modulation is generated in the fuel injection system By recording the pressure P1 at a point above the orifice of the injector and its modulation frequency, and using the signal filtering process to extract the pressure modulation component, the responding amplitude is given for different driving voltages Repeating this procedure for a series of modulation frequencies, the measurement result for the coupling characteristics is obtained, as DRIVING ENVIRONMENT ASSESSMENT USING FUSION OF IN- AND OUT-OF-VEHICLE VISION SYSTEMS gazes the defined inattention regions with a large TDGR, visual inattention warning is generated as shown in Figure 18(c) CONCLUSIONS Compared to the previous systems which used limited states and gave many false warnings, the proposed driving situation assessment system can generate proper warnings and reduce the false warnings by using the following metrics: (1) Improved EVD detection methods − Driver’s gazing region estimation using 2D+3D AAM fitting and projection of the driver’s gaze onto the frontal plane − Vehicle detection using SVM and on-line learning and the probability of non-road region (2) Long-term state prediction: TDNN fuses EVD states and predicts long-term EV states from various real driving training DB (3) Multiple driving situation assessment: Fuzzy inference generates lane departure, vehicle collision, and visual inattention warning signals by using heuristic EVD MFs and fuzzy if-then rules This research focused on the EVD states from the forward area and assumed that the height of driver’s face and the distance between the driver and frontal plan are constant Future work should cluster general driver’s gaze motion and analyze EVD states from all around the vehicle to extend this research to more general driver assistance systems ACKNOWLEDGEMENT−This work was supported by the Basic Research Program of MOST in Daegu Gyeongbuk Institute of Science and Technology (DGIST), Korea REFERENCES Apostoloff, N and Zelinsky, A (2004) Vision in and out of vehicles: Integrated driver and road scene monitoring Int J Robot Res 23, 4–5, 513–538 Cheng, H., Zheng, N., Zhang, X., Qin, J and Wetering, H V E (2007) Interactive road situation analysis for driver assistance and safety warning systems: Framework and algorithms IEEE Trans Intell Transp 8, 1, 157–167 Choi, H C and Oh, S Y (2006) Real-time recognition of facial expression using active appearance model with second order minimization and neural network Proc IEEE Conf Systems, Man, and Cybernetics, 1559–1564 Chung, T., Yi, S and Yi, K (2007) Estimation of vehicle 113 state and road bank angle for driver assistance systems Int J Automotive Technology 8, 1, 111–117 Davies, B and Lienhart, R (2006) Using CART to segment road images Proc SPIE Multimedia Content Analysis, Management, and Retrieval, 60730U, 1–12 Fletcher, L., Loy, G., Barnes, N and Zelinsky, A (2005) A correlating driver gaze with the road scene for driver assistance systems Robot Auton Syst., 52, 71–84 He, Y., Wang, H and Zhang, B (2004) Color-based road detection in urban traffic scenes IEEE Trans Intell Trans 5, 4, 309–318 Kim, J H., Kim, Y W and Sim, K Y (2007) Quantitative study on the fearfulness of human driver using vector quantization Int J Automotive Technology 8, 4, 505– 512 Kim, S Y., Kang, J K., Oh, S Y., Ryu, Y W., Kim, K S., Park, S C and Kim, J W (2008) An intelligent and integrated driver assistance system for increased safety and convenience based on all around sensing J Intell Robot Syst., 51, 261–287 Lowe, D G (1999) Object recognition from local scaleinvariant features Proc Int Conf Comput Vision 1150– 1157 Mamdani, E H and Assilian, S (1975) An experiment in linguistic synthesis with a fuzzy logic controller Int J Man-Machine Studies 7, 1, 1–13 Mandic, D P and Chambers, J A (2001) Recurrent Neural Networks for Prediction John Wiley Chichester New York Matthews, I and Baker, S (2004) Active appearance models revisited Int J Comput Vision 60, 2, 135–164 McCall, J C., Wipf, D P., Trivedi, M M and Rao, B D (2007) Lane change intent analysis using robust operators and sparse Bayesian IEEE Trans Intell Transp 8, 3, 431–440 Stiller, C., Färber, G and Kammel, S (2007) Cooperative cognitive automobiles Proc IEEE Intell Veh Symp., 215–220 Viola, P and Jones, M J (2004) Robust real-time face detection Int J Comput Vision 57, 2, 137–154 Wu, Y.-J., Lian, F.-L, Huang, C.-P and Chang, T.-H (2007) Image processing techniques for lane-related information extraction and multi-vehicle detection in intelligent highway vehicles Int J Automotive Technology 8, 4, 513–520 Xiao, J., Baker, S., Matthews, I and Kanade, T (2004) Real-time combined 2D+3D active appearance models Proc IEEE Conf Comp Vis and Pattern Recog 535– 542 International Journal of Automotive Technology, Vol 10, No 1, pp 115−121 (2009) DOI 10.1007/s12239−009−0014−4 Copyright © 2009 KSAE 1229−9138/2009/044−14 PERFORMANCE ANALYSIS OF A CVT CLUTCH SYSTEM FOR A HYBRID ELECTRIC VEHICLE W RYU1), N CHO2), I YOO3), H SONG4) and H KIM5)* 1) Transmission Test Team, Hyundai Powertech, 447-26 Onseok-dong, Hwaseong-si, Gyeonggi 445-110, Korea 2) New Technology Development, NGV, 314 dong Seoul National University, 599 Gwanak-ro, Gwanak-gu, Seoul 151-742, Korea 3) HEV System Engineering Team, Hyundaimotor Co Ltd., 772-1 Jangdeog-dong, Hwaseong-si, Gyeonggi 445-706, Korea 4) Department of Mechanical Design, ANSAN College of Technology, Gyeonggi 425-792, Korea 5) Sungkunkwan University, School of Mechanical Engineering, Gyeonggi 440-746, Korea (Received 30 November 2007; Revised 10 July 2008) ABSTRACT−In this study, the performance of a CVT clutch system for a hybrid electric vehicle was investigated To analyzed the vehicle performance at restart, the restart delay and driveshaft torque was investigated by simulations and experiments It was found from the simulation results that the vehicle restart response depends on the clutch pressure buildup time to the point where the clutch torque begins to overcome the vehicle road load, and driving comfort at restart is directly related to the rate change of the clutch pressure KEY WORDS : CVT, Hybrid electric vehicle, Clutch, Restart response INTRODUCTION Engagement characteristics of the clutch are of major importance for increasing the smoothness at the start of acceleration However, an improvement in smoothness requires a long engagement time, which leads to high thermal load and increased deterioration (Holgerson, 1997) Since the acceleration performance depends on the vehicle powertrain system dynamics, including the clutch control system, there are many complex phenomena that must be analyzed in the entire powertrain, making dynamic modeling and analysis a requirement A torsional vibration model coupled with a vehicle body longitudinal model was developed to investigate the system response for steady-state and transient running conditions during and after clutch engagement The gradient of friction was proposed to be a major cause of an unstable response (Crolla and Rabieh, 2006) In regards to the hybrid electric vehicle response with continuously variable transmission (CVT), up- and downshift response of the electromechanical controlled CVT ratio feedback system was investigated using the hardware in a loop simulation (Yeo et al., 2004) In this study, restart performance of a soft type HEV is investigated In order to evaluate the restart performance, dynamic models of the CVT hydraulic control system are obtained, including the clutch and CVT variator A hybrid CVT vehicle performance simulator is developed using AMESim software Performance of the clutch control system with the CVT variator is investigated and compared with experimental results Using the simulator, performance Generally, hybrid electric vehicles (HEVs) are classified into 1) soft types and 2) hard types In a soft type HEV, an electric vehicle mode where the vehicle is propelled only by the electric motor is not possible because the motor capacity is relatively small Meanwhile, the EV mode can be implemented in a hard type HEV due to its relatively large motor capacity In a hard type HEV, the engine clutch plays an important role, changing the operation mode from the EV mode to the hybrid electric vehicle mode and vice versa In the soft type HEV, even if the operation mode is not changed by the clutch, the engine clutch plays an essential role in vehicle restart performance such as restart delay and driving comfort when the vehicle starts from engine idle stop When the HEV stops, the internal combustion engine becomes off-state If the driver lifts their foot from the brake pedal to restart the vehicle, the motor propels the engine and the engine begins firing when the engine speed reaches a pre-determined speed At this moment, the oil pump, which is connected to the engine, begins to supply hydraulic pressure to operate the clutch In order to achieve faster start performance, faster build-up of the clutch pressure is required However, a faster increase in the clutch pressure may degrade the start feel, thus requiring a compromise between the expectation of a smooth start feeling and faster start performance *Corresponding author e-mail: hskim@me.skku.ac.kr 115 116 W RYU et al of the restart delay and driving comfort are investigated and a clutch pressure control scheme is proposed to improve restart performance and driving comfort DYNAMIC MODELING OF THE CVT CLUTCH CONTROL HYDRAULIC SYTEM AND CVT DRIVE TRAIN 2.1 Hydraulic Control Valve Figure shows a schematic diagram of the soft type HEV investigated in this study along with the vehicle specifications The 12 kW motor is directly connected to the engine, and a metal belt CVT with wet-type multi-disc clutch is used as a transmission The oil pump is driven by the engine and supplies hydraulic oil to the clutch through the valve body In Figure 2, the CVT hydraulic control system of the HEV investigated in this study is shown The line pressure of the entire hydraulic system is controlled at the secondary valve At the primary valve, the primary pressure is controlled by reducing the line pressure to generate the primary belt clamping force, which controls the speed ratio The clutch engagement is controlled at the clutch valve by a variable force type solenoid valve (VFS) In Figure 3, a schematic diagram and bondgraph model of the clutch valve are shown If the clutch solenoid pressure Pclutch is supplied through port #1, the spool moves to the right side, which opens port #4, which allows the oil Figure CVT hydraulic system flow to port #3 to the clutch actuator Since there is an area difference between land A and B, the spool moves to the left side, which closes port #4 and opens port #2, the exhaust port This leads to oil flow from the clutch actuator to the exhaust port, which results in a decrease of the clutch pressure, and the spool moves to the right side again Therefore, the desired clutch actuator pressure is obtained at the position where the forces acting on the spool due to the solenoid pressure, clutch pressure, and the spring reaction are balanced From the bondgraph model in Figure 3, the dynamic equations of the clutch valve hydraulic system can be obtained as, M clutch x··clutch =P clutch ⋅ P clutch −B clutch ⋅ x··clutch −k clutch ⋅ ( x + x clutch ) – A clutch ⋅ P clutch V clutch0 + A clutch1 ⋅ x clutch- · -P clutch1 =Q clutch1 – A clutch1 ⋅ x· clutch (2) V clutch -3 · P clutch =Q clutch4 – Q clutch2 (3) Qclutch1=Cd ⋅ Aclutch1 ⋅ sign( Pclutch_sol,Pclutch1 ) ρ- Pclutch_sol – Pclutch1 (4) β β Figure Soft type HEV (1) where M is the valve spool mass, x is the clutch piston displacement, P is the pressure, K is the spring stiffness, A is the area, V is the volume, Q is the flow rate, Cd is the discharge coefficient, ρ is the oil density, β is the bulk modulus, and the subscript indicates the clutch Figure shows a schematic diagram of the primary valve When the control pressure Psol of the ratio control solenoid valve (RCSV) increases, the spool moves to the right side This causes the supply port A4o to open, which results in a pressure increase in the primary actuator The primary pressure force is also applied to the spool in the left side due to the land area difference The primary pressure is determined at the spool position where the control pressure force is balanced with the solenoid pressure and PERFORMANCE ANALYSIS OF A CVT CLUTCH SYSTEM FOR A HYBRID ELECTRIC VEHICLE 117 Figure Primary valve Figure Relationship between the axial displacement of the primary movable sheave and the speed ratio Figure Clutch valve and bondgraph model the spring reaction force (Ryu et al., 2005) From the force equilibrium and the continuity equation, dynamic equations of the spool motion and the primary pressure can be derived as mps ⋅ X·· PS =A1 ⋅ Psol – A3 ⋅ P3 – KPS ( XPS + XPS0 )−BPS X· PS −Fr (5) V P + AP XP · P P =Q P – A P X· P (6) β In Equation (7), dXP/di can be obtained from the relationship between the axial displacement of the primary movable sheave and the speed ratio, as shown in Figure As for the CVT shift dynamics di/dt, the empirical equation (Ide, 1995) is used as di - = α ( i ) ⋅ ω P ⋅ ( P p – P *p ) = α ( i ) ⋅ ω P ⋅ Δ P dt where α (i) is a coefficient that is a function of the speed ratio i, ωP is the primary speed, Pp is the primary actuator p ressure, and P *p is the primary actuator pressure at a steady where mPS is the spool mass, A1, A3 are the spool land areas, P3 is the primary pressure, XPS is the spool displacement, KPS is the spring constant, BPS is the damping coefficient, Fr is the flow force, QP is the flow rate to the primary actuator, VP is the initial volume of primary actuator, AP is the actuator area, and XP is the axial displacement of the movable sheave The time derivative of the primary movable sheave displacement X· P is related to the shift speed di/dt as follows, dX P- di dX P- = -dt di dt (7) (8) Figure CVT hybrid vehicle powertrain 118 W RYU et al state The hybrid electric vehicle powertrain investigated in this study is shown in Figure The motor is directly connected to the internal combustion engine (ICE) Power from the engine and motor is transmitted to the CVT through the wet clutch pack Two spring-damper models are introduced at the clutch shaft and output driveshaft Dynamic models of the HEV powertrain are implemented using AMESim software In modeling the HEV powertrain, the clutch torque Tclutch is represented as 3 ( Ro – Ri ) T clutch = μ NA c P clutch - ⋅ ( R o – R 2i ) (9) where μ is the friction coefficient, N is the number of the clutch friction plate, Ac is the actuator area, and R0 and Ri are the clutch outer and inner radius The friction coefficient μ is represented as a function of the relative sliding Figure AMESim CVT hybrid vehicle simulator speed, μ = ( μ s – μ k )e ω-⎞ ⎛ –Δ ⎝ ω ⎠ s + μk (10) where μs and μk are the static and kinetic friction coefficient, respectively, Δω is the speed difference between the input and output shaft, and ωs is the clutch output shaft speed VEHICLE PERFORMANCE SIMULATION AND VALIDATION Based on the dynamic models of the CVT hydraulic control system, variator, and hybrid vehicle powertrain, the vehicle performance simulator is developed using AMESim Figure shows the AMESim simulator for the CVT hybrid vehicle The simulator consists of three modules: (1) hydraulic system, (2) variator, and (3) vehicle dynamics In PERFORMANCE ANALYSIS OF A CVT CLUTCH SYSTEM FOR A HYBRID ELECTRIC VEHICLE 119 Figure CVT system test rig Figure 10 Comparison of primary pressure characteristics Figure Comparison of steady-state characteristics for clutch valve the hydraulic system, hydraulic valves such as the manual valve, clutch valve, pressure reduction valve, primary valve, secondary valve, and lubrication valve are included For the CVT variator, dynamic models developed in this study are implemented instead of using the AMESim module In order to validate performance of the simulator developed in this study, simulation results of the clutch valve, primary and secondary valve, and vehicle restart response are compared with the experimental results Figure shows a CVT test rig developed for the experiment A 30 kW AC motor drives the CVT variator The CVT variator, which consists of the primary, secondary pulley, and the metal belt, is fabricated inside the box The motor power transmitted through the CVT variator is balanced at the hydraulic load simulator that consists of the gear pump and proportional relief valve The CVT control pressure is supplied through the valve body from a hydraulic power unit In Figure and Figure 10, simulation results of the hydraulic control valves are compared with the experiments The simulation results of the pressure response characteristics of the clutch control valve, and primary and secondary valves are in good accordance with the experimental results, which demonstrates the validity of the simulator It Figure 11 Idle stop and restart performance of CVT HEV can be noted from Figure that the clutch pressure does not build up for the clutch solenoid valve duty, 0~13%, which is the ineffective duty range Figure 11 shows a comparison of the vehicle test results In the vehicle test, the restart performance was evaluated when the vehicle restarts after an idle stop As shown in the 120 W RYU et al test results in Figure 11(a), when the driver lifts a foot up from the brake pedal at t = 0.1 s, the ICE speed ωe begins to increase The clutch pressure Pclutch begins to build up at t = 0.5 s via the oil pump connected to the ICE The vehicle is observed to restart after 0.51 seconds from the moment when the ICE speed reaches ω3 = 600 rpm In this study, the restart delay is defined as the delay time from the brake pedal off to the vehicle restart moment, and it is found from Figure 11 that the restart delay is t = 0.68 s The simulation results of the ICE speed, clutch pressure, CVT primary pulley speed, and vehicle velocity are observed to be in good accordance with the experimental results, which demonstrates the validity of the performance simulator developed in this study PERFORMANCE SIMULATION AND DISCUSSION The restart response time of the CVT hybrid vehicle investigated in this study is found to be relatively slower as compared to the response time of the conventional automatic transmission vehicle, since the torque amplification effect by the torque converter does not exist Since the typical restart delay of the automatic transmission vehicle should be designed to be less than t = 0.5 s, the restart response of the CVT clutch system needs to be improved In order to improve the restart response the clutch pressure should be supplied as soon as possible, but this may cause a torque shock that deteriorates the smoothness in the start feel; in other words, it affects the driving comfort The driving comfort at the moment of vehicle restart mainly depends on the clutch slip and lock-up In this study, the restart response and driving comfort are investigated for various clutch pressure profiles that are derived based on the simulation results In Figure 12, clutch torque and driveshaft torque (b), clutch actuator pressure and vehicle velocity (c) are compared for three clutch solenoid valve input duty profiles (a) The solenoid valve duty profiles (a) are selected by considering the ineffective duty characteristics of the clutch solenoid valve It is noted that the clutch pressure Pclutch and Pclutch (Figure 12c) in Profile and Profile (Figure 12a) build up much faster than Pclutch in Profile that is created by neglecting the ineffective duty range For Profile and Profile 3, the initial duty is maintained at 15%, which is slightly higher than the effective duty start point (Figure 9) When the input duty increases from 15% at t = 0.4 s, the clutch pressure Pclutch and Pclutch begin to rise after 36 milliseconds, which is caused by the response delay of the solenoid valve Pclutch in Profile begins to increase only after the solenoid duty exceeds 13% (Figure 12a, c), showing a time delay due to the solenoid valve dynamic characteristic The clutch torque Pclutch is generated at first by the clutch pressure Pclutch The clutch torque begins to build up only after the clutch clamping force from the clutch pre- Figure 12 Vehicle starting performance for various clutch pressure profiles ssure overcomes the clutch return spring force It is noted that the clutch pressure when clutch torque generation occurs is around Pclutch = 0.9 bar The clutch torque Tclutch shows a time delay even if the clutch pressure profile Pclutch has a similar value as Pclutch This is because Pclutch remains around 0.9 bar from t = 0.6 s to t = 0.8 s after the initial increase Tclutch shows the slowest response since Pclutch increases slowly to the clutch torque generation point The driveshaft torque shows a similar response as the clutch torque as shown in Figure 12(b) In Figure 12(c), the vehicle velocity responses are compared The restart response of the vehicle is determined by the driveshaft torque The velocity of the driveshaft torque Td3 shows the fastest start response, while the velocity from the Td2 shows a hesitation at the starting moment The torque overshoot in Td2 is due to the clutch pressure profile (Pclutch 2) which results from the duty profile It is found from the simulation results in Figure 12 that the vehicle is able to start when the clutch torque, or the driveshaft torque, begins to overcome the road load Therefore, the faster the clutch actuator pressure rises, the faster the start response can be achieved However, faster pressure build-up in the clutch actuator may PERFORMANCE ANALYSIS OF A CVT CLUTCH SYSTEM FOR A HYBRID ELECTRIC VEHICLE 121 Figure 13 can be used as an initial design guideline in the tuning process CONCLUSION Figure 13 Vehicle starting performance for various clutch pressure profiles cause a driveshaft torque shock that deteriorates driving comfort In Figure 13, the rate changes of the driveshaft torque are compared for various clutch pressure profiles For the three clutch pressure profiles, each pressure profile is designed to rise to P = bar at t = 0.8 s in order for the vehicle to start at the same time, even if each pressure profile has different shape It is seen from Figure 13 that the rate change of the driveshaft torque depends on the rate change of the clutch pressure From the simulation results in Figure 12 and Figure 13, it is found that the vehicle restart response depends on the clutch pressure build-up time to the point where the clutch torque begins to overcome the vehicle road load For driving comfort, which is evaluated by the rate change of the driveshaft torque, it is found that the driving comfort at restart is directly related to the rate change of the clutch pressure Therefore, in order to improve the restart performance and driving comfort, first, it is necessary to control the initial duty of the clutch control valve in order to supply the clutch torque that overcomes the vehicle road load at a standstill state and also to select the appropriate duty ratio that minimizes the rate change of the driveshaft torque Even if driving comfort is normally evaluated by a subjective manner in the field test, the simulation results in A clutch actuator pressure control scheme is investigated for a soft-type hybrid electric vehicle, which adopts a metal belt CVT, to improve the restart response while maintaining driving comfort To evaluate the vehicle performance at start, dynamic models of the CVT hydraulic control system, wet clutch pack, CVT variator, and drive train are obtained An AMESim vehicle performance simulator is developed based on the dynamic models and is validated by comparison with the vehicle test results Using the simulator, performance of the restart delay and driveshaft torque are investigated It is found from the simulation results that the vehicle restart response depends on the clutch pressure build-up time to the point where the clutch torque begins to overcome the vehicle road load For driving comfort, which is evaluated by the rate change of the driveshaft torque, it is found that driving comfort at the restart is directly related to the rate change of the clutch pressure Therefore, in order to improve the restart performance and driving comfort, a compromise between the fast clutch pressure build up and a smooth rate change of the clutch pressure is required REFERENCES Crolla, D and Rabieh, E (2006) Coupling of driveline and body vibration in truck SAE Paper No 962206 Holgerson, M (1997) Apparatus for measurement of engagement characteristics of a wet clutch Wear, 213, 140−147 Ide, T., Udagawa, A and Kataoka, R (1995) Simulation approach to the effect of the ratio changing speed of a metal V-belt CVT on the vehicle response Int J Vehicle System Dynamics, 24, 377−388 Ryu, W., Nam, J., Lee, Y and Kim, H (2005) Model based control for a pressure control type CVT Int J Vehicle Design, 39, 175−188 Yeo, H., Song, C H., Kim, C S and Kim, H S (2004) Hardware in the loop simulation of hybrid vehicle for optimal engine operation by CVT ratio control Int J Automotive Technology 5, 3, 201−208 International Journal of Automotive Technology, Vol 10, No 1, pp 123−129 (2009) DOI 10.1007/s12239−009−0015−3 Copyright © 2009 KSAE 1229−9138/2009/044−15 DEVELOPMENT OF A KNOWLEDGE-BASED HYBRID FAILURE DIAGNOSIS SYSTEM FOR URBAN TRANSIT H J KIM1), C H BAE1), S H KIM1), H Y LEE2), K J PARK2) and M W SUH3)* 1) Graduate School of Mechanical Engineering, Sungkyunkwan University, Gyeonggi 440-746, Korea 2) Korea Railroad Research Institute, 360-1 Woulam-dong, Uliwang-si, Gyeonggi 437-050, Korea 3) School of Mechanical Engineering, Sungkyunkwan University, Gyeonggi 440-746, Korea (Received 27 June 2007; Revised October 2008) ABSTRACT−Urban transit is a complex system that contains both electrical and mechanical entities; therefore, it is necessary to construct a maintenance system for ensuring safety during high-speed driving Expert systems are computer programs that use numerical or non-numerical domain-specific knowledge to solve problems This research aims to develop an expert system that diagnoses the causes of failures quickly and displays measures to correct them For the development of this expert system, the standardization of a failure code classification and the creation of a Bill of Materials (BOM) were first performed Through the analysis of both failure history and maintenance manuals, a knowledge base has been constructed Also, for retrieving the procedure of failure diagnosis and repair linking with the knowledge base, we have built a Rule-Based Reasoning (RRB) engine with a pattern matching technique and a Case-Based Reasoning (CBR) engine with a similar search method Finally, this system has been developed as web based in order to maximize accessibility KEY WORDS : Expert system, Knowledge base, Rule-based reasoning, Case-based reasoning, FMEA (Failure Mode and Effect Analysis), FTA (Failure Tree Analysis), DT (Decision Table) INTRODUCTION quickly and accurately Therefore, many studies have been conducted on failure diagnosis systems Dendral (Patterson, 1995) developed a failure diagnosis system for solving material and chemical equations, and Mycin (Patterson, 1995) studied a medical diagnosis system A failure diagnosis system requires the use of the following research: systematizing heuristic, diagnosis and maintenance procedures, failure mode and effect analysis (FMEA), failure tree analysis (FTA), and decision tables (DTs) (Suh and Kang, 1999) investigated a rule-based expert system using standardization of the failure mode, but the application model was limited to a singleness system such as machine tools (Lee and Kim, 1998) constructed a hybrid expert system by mixing rule and case-based reasoning for a copying machine However, compared with a web-based system, the system had limited knowledge accumulation because it was a Client/Server version (Zhang et al., 1998) This research aims to develop a hybrid failure diagnosis system by mixing rule and case-based reasoning to accomplish the efficient maintenance of a complex system such as urban transit through the following: Constructing a Bill of Material (BOM), failure code classification, and a knowledge-base A knowledge-base can be constructed by analyzing historical failure data and manual information For searching the procedure of a failure diagnosis and repair linking in the knowledge-base this research establishes a Rule-based reasoning (RBR) engine with a pattern matching technique and a Case-based reasoning (CBR) engine Urban transit has become the principal form of public transportation and has been designed with consideration to improved safety Safety expresses freedom from unexpected risk of harm and can be defined as: ‘The ability to not cause injury to persons, significant material damage or other unacceptable consequences above a fixed level for a stated time interval when operating under given conditions’ It is essential to take measures to improve safety and extend the life cycle of the transit system for meeting the requirements of the citizens in the field of urban transit (Bae, 2005) The maintenance of a complex system, such as urban transit, generally occupies sixty percent of the total operational cost (Lee et al., 2003) Therefore, it is necessary to minimize costs by constructing an effective maintenance system Maintenance faults seriously influence social safety and the economy because they may result in big accidents due to devices failure In order to minimize the occurrence of failures through quick and accurate maintenance, it is necessary to analyze the cause of repeated failures/accidents by using a failure diagnosis system (Maeng, 1996) Experts have diagnosed device defects through experience and know-how obtained by trial and error If the experienced heuristic is systematized through IT technology, laymen could diagnose device defects of devices *Corresponding author e-mail: suhmw@skku.edu 123 124 H J KIM et al with similar search methods Finally, we develop a webbased system to maximize its accessibility STANDARDIZATION FOR FAILURE DIAGNOSIS SYSTEM Standardization is divided into a BOM and a failure code classification It makes the failure diagnosis system gather, accumulate, and process all historical failure data systematically 2.1 Construction of Urban Transit BOM The BOM, which is a hierarchical product tree, is necessary to list all the components and parts that make up a product in the inventory and consists of Line Replaceable Units (LRU) This research constructed a master BOM for urban transit and then designed a component-based BOM management system to extract the function BOM from the master BOM according to the rule sets shown in Figure The master BOM is the entity responsible for managing components and investigating their physical locations and their status Besides, constituent elements in the master BOM can interact with each other using a mediator called the rule-set The rule-set is a group of interface rules that constructs independent components and makes the BOM system suitable and functional The relationships between master BOM and function BOM are illustrated in Figure (Orfali et al., 1998, Mason and Towill, 1997) 2.2 Definition of Failure Code Classification As one of the standardizations for a case-based failure diagnosis system of urban transit, it is necessary to classify and standardize failure codes The failure code illustrates the conditions or reasons for a failed asset It is generally used in computerized maintenance management systems like failure code in the copier machine mentioned in the introduction However, in order to have a broader utility, such as being able to express various or complex failure modes, it needs be expanded An expanded failure code; which includes maintenance history data, classification of failure and the reasons for the failure mode, etc, makes possible data gathering and standardized maintenance (Kim et al., 2005) For this, we analyzed current failure data in the work space and determined the current failure classification Table shows the sample data for the current failure classification Finally, we made an expanded standard code system that includes the standardized classification rule of the failure and the code numbering system as may be seen in Figure The expanded failure code is an eight-position Table Sample data of current failure classification Failure Failure mode classification CM CM BRAKE BRAKE BRAKE BRAKE Actions taken CM inventor PCB rack exchange in failure CMSB box CM NG CM oil replenishment & air filter exchange CM failure Lead line exchange to PTR CM valve CMG failure, exchange discharge CM air dryer Wing valve fracture in leakage dryer, exchange CM inventor CM oil replenishment failure display Figure Generation of function BOM based on component from master BOM Figure Failure code indexing order Figure Construction of function BOM Figure Structure of failure diagnosis system DEVELOPMENT OF A KNOWLEDGE-BASED HYBRID FAILURE DIAGNOSIS SYSTEM FOR URBAN TRANSIT 125 pattern and consists of five items: failure classification, mid-class, failure modes, failure causes, and severity Failure classification and mid-class indicate the fourteen main systems of urban transit and the subsystems in the recently lower level, respectively Severity, an assessment on how serious a failure mode affects the safety of system, is a kind of weighting factor to give the priority of maintenance to the relevant failure HYBRID FAILURE DIAGNOSIS SYSTEM FOR URBAN TRANSIT A hybrid failure diagnosis system consists of a knowledgebase, inference engine, user interface, and DataBase Management System (DBMS) as shown in Figure The failure diagnosis system establishes inspection and repair procedures on the basis of accumulated knowledge history for maintaining urban transit The knowledge-based system is a computer program that contains the subject-specific knowledge of human experts An artificial algorithm, adapted to the failure diagnosis system, helps users decide an objective standard This maximizes the efficiency by minimizing the maintenance task There are generally two types of artificial algorithms in the failure diagnosis system, they are Rule-based and Casebased reasoning A rule-based failure diagnosis system (Lee and Kim, 1998), using technical knowledge such as a maintenance manual has several merits and demerits, which are described as follows: • Merits: Quick search time and high reliability of reasoning result • Demerit: Difficult knowledge (especially rules) acquisition A case-based failure diagnosis system, using experimental knowledge such as maintenance history, is described as follows: • Merit: Easy knowledge (especially cases) acquisition and good extension of application area • Demerit: Lower reliability when case is not sufficient This research presents a hybrid failure diagnosis system that mixes rule-based reasoning with case-based reasoning to take merits and compensate the demerits of each reasoning method and Rhyu, 2003) Case-based reasoning solves a problem through the use of a search method A search method using case-based reasoning is a document retrieval system designed to help find information stored on a computer system The architecture of case-based reasoning is described in Figure This search method allows one to ask for content meeting specific criteria and retrieves a list of items that match those criteria Figure shows the structure of each module that comprises the case-based reasoning system; these modules are defined as follows: • Retrieve Module: searching an existed case in the database for solving a new problem and evaluated the similarity between an input as a question and existing case • Reuse Module: solving a new problem through the searched solution of the case • Revise Module: In the case that the failure diagnosis system can not solve a new problem, Revise Module in the system researches the solution for a new problem • Retain Module: Retain Module acquired knowledge includ- Figure Architecture of case-based reasoning 3.1 Inference Engine An inference engine, which is the brain of a failure diagnosis system, derives proper answers from the knowledgebase using a computing technique This architecture relies on a data store, working memory, and a global database of symbols representing facts and assertions about the problem The inference engine type is decided by considering the size of the knowledge-base and the efficiency of the application 3.2 Case-Based Reasoning Case-based reasoning offers a solution using the used case and the experienced case for solving a new problem (Ha Figure Module of case-based reasoning 126 H J KIM et al ing the recording solution process of solving a new problem If the user searches relevant cases by using an inquiry keyword containing a space between words without a general information retrieval engine, the system would not be able to present appropriate cases However, if the general information retrieval engine for case-based searching is used, development and maintenance costs would increase because overall system architecture is complicated Therefore, it is necessary to apply a method which does not increase the complexity of overall system architecture and can also effectively search case-base The case-based failure diagnosis system in this paper adopts the combination of three search methods that could search inquiry keywords and case documents containing spacing words The first out of three search method is using a DataBase Management System (DBMS), and the second is a keyword searching method that distinguishes a compound noun using an analyzer morpheme of Korean This method offers convenience while searching by expanding inquiry keywords with a thesaurus for distinguished keywords The third is using the Bigram search method The combination of three search methods has two merits The first is that it does not increase the complexity of the overall system architecture, because it does not create an indexing database for each case The second is the efficient managing of the overall system caused by not indexing the database renewal periodically 3.3 Rule-Based Reasoning A rule-based failure diagnosis system reasons through solving the problem using experiential knowledge from expert-based rules (Han et al., 2003) An understanding of the inference rule concept is important in understanding failure diagnosis systems An inference rule is a statement that has two parts; an if-clause and a then-clause This rule is what gives failure diagnosis systems the ability to find solutions to diagnostic and prescriptive problems There are two main methods of reasoning when using inference rules; backward chaining and forward chaining In this study, rule-based reasoning is used for forward chaining Forward chaining starts with the data available and uses the inference rules to conclude more data until a desired goal is reached An inference engine using forward chaining searches the inference rules until it finds one in which the if-clause is known to be true It then concludes the then-clause and adds this information to its data It will continue to this until a goal is reached This method is also called data driven because the data available determine which inference rules are used, However, rule-based failure diagnosis systems have two disadvantages First is the limit of expansive knowledge due to the creation of rules through expert’s experience The second disadvantage is that created rules not apply to every engineering field Therefore, we constructed a hybrid failure diagnosis system mixing a case-based failure Table Function of failure diagnosis system Case-based failure diagnosis system Index keyword and distinguish a compound noun using analyzer morpheme of Korean Rank indexed cases by using keyword Index Bigram using question and answer Search included Bigram cases using DBMS function Rank indexed cases using Bigram search method Rule-based failure diagnosis system Fast-Match Algorithm (RETE, TREAT Algorithm) Data Gathering, Indexing Figure Architecture of rule-based failure diagnosis system diagnosis system and a rule-based failure diagnosis system to compensate for the disadvantages of the rule-based failure diagnosis system in this study Table shows the functions of a case-based failure diagnosis system and a rule-based failure diagnosis system in the hybrid failure diagnosis system Figure shows the architecture of rule-based reasoning The rule-based reasoning system is composed of three modules Each module is illustrated as follows: • Knowledge-Base: This module creates experiential knowledge from expert-based rules and offers suitable rules for Figure Steps of rule-based reasoning DEVELOPMENT OF A KNOWLEDGE-BASED HYBRID FAILURE DIAGNOSIS SYSTEM FOR URBAN TRANSIT 127 Table Hardware spec CPU: 64Bit RISC 1.0GHz, CPU 2EA Main Memory: 8GB Internal Disk device: 15K rpm 200GB − Web Tier development environment: JSP, Java − Engine development languages: JDK 1.3.1_06, J2SDKEE 1.3.1 − DB: Oracle 9i − WAS: IBM WebSphere 4.0.3 solving the problem • Knowledge Acquisition: This module modifies unsuitable rules as occasions and expands the knowledge-base through modified rules • Inference Engine: This module solves the caused inference problem on reasoning using suitable knowledge and reasons using working memory of stack type Figure shows the steps of rule-based reasoning The first step is to systematize knowledge using books and the experience of an expert Finally, the user obtains a solution for a new problem through each step DEVELOPMENT OF A WEB-BASED SYSTEM Figure Flow chart of failure diagnosis system In the study the development languages of the hybrid failure diagnosis system are Java (Sun Microsystems, 2000) and JSP (Java Sever Page) (Sun Microsystems, 2000) which is known as the most effective web development language This system also used Oracle 9i (Raghu and Johannes, 2003) as the database operating tool for managing the knowledge-base Figure shows the logic of a hybrid failure diagnosis system and Table illustrates the hardware specification of the server for developing the web-based system 4.1 Database Design A database is organized so that it can be easily and quickly accessed, updated and extended It consists of thirteen tables as shown in Figure 10 The AP2RULE table has the results of rule-based reasoning The AP2RULE table consists of AP2RULECON-CLUSION, AP2RULECONDITION, AP2REPAIR, and AP2REAIRDETIL Each table contains the historical maintenance data of the parts These data are used for establishing a diagnosis and repair procedure through a reasoning process of interactive (Yes/No) type In the AP2RULE table, the BOM Figure 10 Failure diagnosis system ERD 128 H J KIM et al column contains the hierarchical tree structure of the system or equipment To store each part uniquely, we set up the car number, BOM code and position number as Primary Keys The AP3CASE table has the results of the case-based reasoning The AP3CASE table has the basic information according to car number, vehicle number and line number for urban transit The basic information contains the failure cause, failure mode, outbreak time, measure, notice, attached file, and so on In additional, AP1FAIUREBOARD manages atypical failure data AP1TBLMAINCD, AP1TBLSUBCD, and AP1TBL-DETAIL contain the failure code 4.2 Modules and GUI of Failure Diagnosis System A knowledge-based hybrid failure diagnosis system consists of rule and case-based reasoning modules 4.2.1 Rule-based reasoning module This module displays the BOM tree, part name, and rule conditions of the system, sub-system, equipment, and part by a line, a vehicle, and a car number for urban transit The Graphic User Interface (GUI) is illustrated in Figure 11 The results of this module are used to establish repair or inspection procedures An example of reasoning scenario is as follows: An example of a reasoning scenario is as follows: Rule-based Reasoning for Diagnosis Process • Device Keyword: Air Compressor • Diagnosis Reasoning: − Step 1: Is there continuously exhaust gas from safety valve? Y − Step 2: Is the V-three Valve covered with dust? Y • Result: Change the valve bracket Rule-based Reasoning for Repair Process • Repair Reasoning: − Step 1: Shut off the power supply − Step 2: Disassemble the air compressor − Step 3: Determine whether the valve bracket is damaged or not − Step 4: Change the valve bracket − Step 5: Clear out and assemble the air compressor − Step 6: Perform a test run 4.2.2 Case-based reasoning module This module represents the failure mode, failure cause, outbreak time, relevant part, and hereafter measure and notice by vehicles The GUI is illustrated in Figure 12 The results of this module presents the diagnosis and maintenance procedures through reasoning at the time of the failure’s occurrence An example of a reasoning scenario is as follows: The Occurrence of Failure Event or Maintenance Request • Date: 05/25/2006 pm • Related Vehicle: 4line – 414unit – 4414vehicle • Failure Mode: Sending out smoke from the truck (bogie) Figure 11 Rule-based reasoning Figure 12 Case-based reasoning Case-based Reasoning for Similar Case Search • Search Keyword: Sending out smoke from the truck (bogie) • Search Result: − Occurrence Date: 07/11/2004 pm Occurrence Place: Ji-Chuk Station − Related Vehicle: line-414 unit-4414 vehicle − Failure Mode: Sending out smoke from the truck (bogie) − Failure Cause: Damaged Valve seat of Air Compressor − Inspection Content: Test whether a valve bracket is damaged or not − Result: Inspect the air compressor CONCLUSIONS Our research developed hybrid failure diagnosis system for improving the reliability and efficiency of the maintenance process The system is composed of two modules that have a rule-and case-based reasoning engine The development of this system was accomplished through two tasks: construction of a BOM and failure code DEVELOPMENT OF A KNOWLEDGE-BASED HYBRID FAILURE DIAGNOSIS SYSTEM FOR URBAN TRANSIT 129 classification From these tasks we could obtain the standard maintenance data as input values for the failure diagnosis system Then, a rule-based reasoning engine was constructed by using the repair guide book and experts experience The case-based reasoning engine was built using historical maintenance data This research may summarized as follows: (1) The efficiency of maintenance data flow for failure diagnosis system was maximized by classifying the failure code (2) The standardization of the maintenance procedure was accomplished by extracting rules from the repair guide book and historical data (3) The developed system could guarantee a reasonable failure diagnosis and maintenance procedure by using experts experiences and the repair guide book (4) This research could be used to develop a web-based failure diagnosis system through IT Technology and also establish the basis for repetitious failure and maintenance procedures by developing the failure diagnosis system ACKNOWLEDGEMENT−This work was supported by grant No (R01-2004-000-10938-0) from the Basic Research Program of the Korea Science & Engineering Foundation and the fostering project of the Excellent Lab of the Ministry of Education and Human Resources Development (MOE), the Ministry of Commerce, Industry and Energy (MOCIE), and the Ministry of Labor (MOLAB) REFERENCES Bae, C H., Kim, S B., Lee, H Y and Suh, M W (2005) A study on development of the reliability evaluation system for VVVF urban transit Trans Korean Society Automotive of Engineers 13, 5, 7−18 Ha, C S and Rhyu, K S (2003) Design and implementation of intelligent web search agent using case based reasoning Korea Society of Computer and Information 8, 1, 20−29 Han, W Y., Sohn, J C., Ham, H S and Kang, J H (2003) A study on a system to develop rule-based applications Korea Information Science Society 30, 2, 259−262 Kim, H J., Bae, C H., Kim, S B., Lee, H Y., Kim, M H and Suh, M W (2005) A study on the development of web-based expert System for urban transit Trans Korean Society of Automotive Engineers 13, 5, 163−170 Lee, H Y., Park, K J., Ahn, T K., Kim, G D., Yoon, S K and Lee, S I (2003) A study on the RAMS for maintenance CALS system for urban transit Korean Society for Railway 6, 2, 108−113 Lee, J S and Kim, Y G (1998) A hybrid malfunction diagnosis System using customer-reported symptoms Korea Expert System Society 4, 1, 115−131 Maeng, J S., Choi, I K and Sung, K R (1996) Development of an expert system for design of high performance compressor valve system Trans Korean Society of Automotive Engineers 4, 6, 153−163 Orfali, R., Harkey, D and Edwards, J (1998) Instant CORBA John Wiley & Sons U.S.A 89−108 Mason, J R and Towill, D R (1997) Information enrichment: designing the supply chain for the supply chain for competitive advantage Supply Chain Management 2, 4, 137−148 Patterson, D W (1995) Introduction to Artificial Intelligence and Expert Systems Prentice Hall U.S.A 128− 172 Raghu, R and Johannes, G (2003) Oracle 9i User's Guide McGraw-Hill U.S.A Suh, D K and Kang, M J (1999) Development of an expert system for machine tool diagnosis Korean Society of Precision Engineering 16, 10, 217−224 Sun Microsystems, Inc (2000) J2EE(tm) Developer's Guide Sun Microsystems U.S.A Sun Microsystems, Inc (1999) JSSE for the Java SDK Standard Edition, V1.4 Sun Microsystems U.S.A