Turbulence characteristics of tumble flow in a four-valve engine

Turbulence characteristics of tumble flow in a four-valve engine Kern Y. Kang a,*, Je H. Baek b a Head of Low Emission Engine Laboratory, Korea Institute of Machinery and Materials (KIMM), 171, Jang-dong, Yuseong, Daejon 305-600, South Korea b Associate Professor, Department of Mechanical Engineering, Pohang University of Science and Technology (POSTECH), P.O. Box 125, Pohang, Kyungbuk, South Korea Received 30 March 1997; received in revised form 9 February 1998; accepted 27 April 1998 Abstract Tumble flow has been adopted to increase the precombustion turbulence level in four-valve engines, since it can be e€ectively generated with negligible adverse e€ect on the flow coecient. In this study, turbulence characteristics of the tumble flow in a four- valve engine were investigated by laser Doppler velocimetry (LDV) and analyzed by means of turbulence intensity, integral time and length scales, and energy spectrum. The data rate was suciently high to allow the bulk velocity to be characterized in individual cycles at 500 and 1000 rpm for two di€erent intake ports. The integral time scales obtained by three kinds of di€erent definitions and their characteristics were compared. The results show that tumble causes turbulence intensity to increase considerably during the compression stroke and its distribution to be homogeneous. The e€ect of tumble on integral time scale is negligible, while the in- tegral length scale increases by tumble. Ó 1998 Elsevier Science Inc. All rights reserved. Keywords: Internal combustion engine; Tumble flow; Intake port; LDV measurement; Cycle-resolved analysis; Turbulence intensity; Turbulence scales 1. Introduction It is well known that the turbulent flow in the cylin- der of an engine plays an important role in determining the combustion characteristics and thermal eciency of the engine. Automotive engineers have endeavored to utilize the turbulence by changing the shape of the combustion chamber and the inlet system geometry, in order to reduce exhaust emissions, improve fuel econo- my and extend the lean operating limit of an engine. A number of attempts have been made to increase the precombustion turbulence levels through either the generation of turbulence due to shear strain rate during induction process or the breakdown of large scale flow structures into small scale turbulence during the com- pression process. However, induction-generated turbu- lence was proven dicult to be maintained up to the end of compression stroke for practical valve lifts [1,2]. Turbulence at the end of the compression stroke was found to be enhanced by the breakdown of large scale structures retained through the induction and com- pression strokes, or by the squish motion occurring near top dead center (TDC) of compression in certain com- bustion chamber geometries. However, in four-valve engines squish is very weak and its e€ect as a turbulence promoter is limited as well as localized around the squish lip [3]. Tumble has been considered to o€er more advantages in four-valve engines equipped with pent- roof combustion chambers, since it can be e€ectively generated with a straight and symmetrical dual intake port with negligible adverse e€ect on the flow coecient. The formation of tumble was revealed from visuali- zation studies in transparent engine cylinders [4], and first quantified by experimental and computational study [5]. The strength of the tumble was later quantified in water analogue rigs using particle tracking ve- locimetry [6–8], and its more detail structure was ana- lyzed by multi-dimensional numerical experimentation [9]. The concept of inclined tumble, which is a combi- nation of swirl and tumble flows, aiming to extend the lean limit in four-valve engines, was investigated in [10] for a range of intake port designs. The turbulence en- hancement mechanism of tumbling motion was studied and was found to be composed of the spinning up of the tumbling vortex, and its breakdown with the associated release of its kinematic energy during compression [11]. Experimental Thermal and Fluid Science 18 (1998) 231–243 * Corresponding author. Tel.: +82 42 868 7380; fax: +82 42 868 7305; e-mail: kykang@mailgw.kimm.re.kr. 0894-1777/98/$19.00 Ó 1998 Elsevier Science Inc. All rights reserved. PII: S 0 8 9 4 - 1 7 7 7 ( 9 8 ) 1 0 0 2 3 - 7 The tumble formation and decay mechanisms in four- valve engines were analyzed by means of tumble mo- ment and two non-dimensional parameters by the au- thors [12]; that is, the tumble center location relative to cylinder center and its velocity profile were proven to strongly a€ect the tumble decay mechanism. This paper describes detail characteristics of the tur- bulence enhancement by the tumbling motion in a four- valve engine. It’s aim is to provide better understanding of the e€ect of the tumble on turbulence during the in- take and compression strokes. Two di€erent intake-port configurations were selected to vary the characteristics of the tumble flow in the cylinder of a four-valve engine, and laser Doppler velocimetry (LDV) was used to measure the development of tumble structures and tur- bulence under motored engine operation. The turbu- lence is quantified by a cycle-resolved analysis and then characterized in terms of its intensity, and integral time and length scales during the compression stroke. 2. Experimental apparatus and method 2.1. Transparent engine The experimental facilities consist of a transparent engine, an LDV system and a particle seeding device as shown in Fig. 1. A 125 cc air-cooled motorcycle engine was modified into an optically accessible transparent engine. The combustion chamber of the engine is formed within the head and has a pent-roof geometry with twin intake and exhaust valves, as shown in Fig. 2. The general specifications of the engine are as follows: 56.6 mm bore, 49.5 mm stroke, and a com- pression ratio of 8.5. The opening and closing time of the intake valves are 10° before top dead center (BTDC) and 30° after bottom dead center (ABDC), respectively. Optical accesses are provided by two ar- rangements: in the first, two quartz windows are lo- cated on opposite sides of the combustion chamber for the forward scattering optical configuration, and in the second, a flat quartz window is installed in the piston head, with a mirror positioned at 45° inside the elon- gated piston, for the backward scattering optical ar- rangement. The original piston rings were replaced by two Teflon-impregnated carbon-rings and a rubber O- ring, enabling the engine to be motored without any lubrication and preventing air leakage through the crevice region between the piston and cylinder wall. The engine was motored by a 10 kW dc motor, whose rotational speed could be controlled from 300 to 2000 rpm within ‹2 rpm. The experimental engine was equipped with a shaft encoder having a 0.5° crank angle resolution. Fig. 1. Schematics of the experimental setup. 232 K.Y. Kang, J.H. Baek / Experimental Thermal and Fluid Science 18 (1998) 231–243 2.2. LDV system The LDV system consisted of a 5-W argon-ion laser (Spectra-Physics), a fiber optic transmitter, and a signal processor. The single-component laser was operated at the green wavelength of 514 nm. The laser power during the experiment was adjusted between 0.6–2.0 W. Two types of LDV optical arrangements for transmitting and receiving optics were used for the measurement of air motion inside the cylinder as shown in Fig. 3. First, a backward scattering arrangement was installed for the bulk flow measurements in the entire cylinder, in order to take full advantage of the transparent piston. The laser beams were transmitted into the combustion chamber after being reflected on the mirror centered in the elongated piston. Secondly, a forward-scattering optical arrangement was used for the turbulence mea- surements in the top region of the combustion chamber. The beams entered the engine through one of the win- dows located in the upper part of the liner, and then the scattered light was collected through the opposite win- dow. A Bragg cell of 40 MHz was used to eliminate the directional ambiguity. The dimensions of probe volume were 0.63 mm by 0.08 mm, with the fringe spacing of 2.12 lm. The optical probes could be moved by a 3-di- mensional traversing system with 0.1 mm resolution. The LDV signal was processed by an FFT type signal processor (BSA, Dantec) and the Doppler signals from the photomultiplier were transmitted to the BSA to- gether with the crank angle information provided by the shaft encoder. 2.3. Particle seeding In this experiment alumina powder of nominal di- ameter 1 lm was supplied as seeding particles into the intake air stream through a fluidized bed. Before en- tering the seeder, the intake air was dried to its dew point of 2°C and then heated to the ambient tempera- ture by an electric heater. This procedure ensures that the seeding material is always available in a non-ag- glomerated form and that the engine may be run with- out condensation of water inside the cylinder, resulting in improved LDV signals. The dried particles were e- ciently mixed in a surge tank with bypassed intake air, and then introduced into the engine. 2.4. Experimental conditions Measurements were performed for the two intake port configurations shown in Fig. 4. The conventional port, which is the original intake port of the engine, is aligned with two intake valves inclined 25° to the axis of the cylinder. The tumble port, on the other hand, has Fig. 3. LDV optical arrangements for accessing the combustion chamber. Fig. 2. Schematics of the transparent engine. K.Y. Kang, J.H. Baek / Experimental Thermal and Fluid Science 18 (1998) 231–243 233 straight intake ports with 30° inclination with respect to the cylinder head plane and the same intake valves as the conventional port. The engine was motored at 500 and 1000 rpm. Measurements were made throughout the whole 360° period, from the start of the intake stroke to the end of the compression stroke. Velocity measure- ments for the tumbling motion in the cylinder were obtained at 20 points with a 3 mm interval along the cylinder axis and at 9 points along a radius on two cross- sections of the Section 1 (1 mm below from TDC) and the Section 2 (17 mm below from TDC) planes. In this study, only the tumble velocity component, which is normal to the plane of cross-section A–A in Fig. 2, were measured in the cylinder by the single-component laser. Typically, 16 000 samples were collected at each mea- surement point, and the data rate for the back-scatter was normally 0.8–1.5 kHz (valid rate 60%) at 1000 rpm. The measurements for the cycle-resolved analysis were obtained at the cylinder center (rˆ 0) and at 9 mm away from the center (rˆ 9 mm) on the plane of 1 mm above TDC, and typically 400 000 samples were collected at each point, and data rates were normally 40–60 kHz (valid rate 80%) at 1000 rpm. 2.5. Experimental uncertainties The overall error of a Doppler burst measurement is estimated to be approximately 0.7% with downmixer [13]. Since part of the error may be due to be noise, operation of the system was checked by confirming that in the absence of particle seeding the data rate was zero. It was also found to be useful to block one beam and to make sure that the data rate was again zero. Approxi- mately 90 valid measurements for each crank angle window were considered in the ensemble averaging process, resulting in conservatively estimated uncer- tainties of 5% in the mean velocity. The statistical un- certainty in autocorrelation coecient estimates depends on the reciprocal square root of the number of data records (80) and is expected to be less than 16% in the results presented here. In the cycle-resolved analysis, the statistical uncertainty in the turbulence intensity is essentially the same as that of autocorrelation coe- cient. However, it is considered that the analysis of the results and the related conclusions are not influenced by the overall experimental uncertainty. 3. Data analysis 3.1. Mean velocity The flow inside the engine cylinder are quasi-periodic because of the cyclic nature of the engine operation. Moreover, the flow in one cycle is unsteady with short period so that the averaging techniques used for steady turbulent flows are not applicable. Basically two ap- proaches to defining the flow characteristics in an engine have been used; ensemble-averaging analysis and cycle- resolved analysis. The mean velocities of tumble flow were determined by the ensemble-averaged analysis as follows: U…h† ˆ 1 NM…h† XNc tˆ1 U…h;D; i†; …1† Fig. 4. Intake port configurations. 234 K.Y. Kang, J.H. Baek / Experimental Thermal and Fluid Science 18 (1998) 231–243 where h, i, Nc, and NM(h) are crank angle, the ith cycle, the number of cycles, and the total number of mea- surements at crank angle h over Nc cycles respectively. In the above ensemble averaging, Dh is the crank angle window for decreasing the statistical uncertainty [14] and the data are supposed to be same within each crank angle window. In this study 2° crank angle windows were used and, on average, 90 samples per each crank angle window were processed. 3.2. Cycle-resolved turbulence Generally in the ensemble-average analysis any cycle- to-cycle variation are interpreted as turbulence. In this study the velocities measured in the top region of the cylinder were processed by cycle-resolved analysis to obtain the exact information about the turbulence characteristics. However, the diculty of cycle-resolved analysis for in-cylinder LDV data is in defining what constitutes the mean velocity (U ) and what part of the velocity variations constitutes the turbulence (u) in each cycle where the velocity components are expressed as follows: U…h; i† ˆ U…h; i† ‡ u…h; i†: …2† In this study the approach to the above problem is basically to follow that of Liou and Santavicca [15,16], in which digital low-pass filtering was used to determine the bulk velocity in each cycle, even though there still remains the uncertainty for selecting a cut-o€ frequency. Here, the cut-o€ frequency was determined by the spectrum analysis of the ensemble-average of time-av- erage filtering velocities, as suggested by Catania and Mittica [17,18]. The data reduction procedure of the cycle-resolved analysis is summarized as a sequence of four steps, shown in Fig. 5. Before the main process of data re- duction, any LDV measured data which are not within three times of standard deviations from the mean, after the data are grouped into the two crank angle degrees, are excluded. The excluded data were found to be less than 1% of all the data, which provides confidence in the accuracy of the velocity data. The first step is the time-average filtering. Fig. 5(a) shows the time average of raw data and its curve fitting in a single cycle, as well as the ensemble-averaged value of the fitted curves over many cycles to get an ensemble- averaged smooth curve. The curve fitting was perform- ed, using cubic spline method, and the time-average interval of 12° suggested by Catania and Mittica was used. The second step is to analyze the frequency spec- trum of the ensemble-averaged smooth curve to obtain the cut-o€ frquency. Fig. 5(b) shows the comparison between the fast Fourier transform (FFT) of the en- semble-averaged smooth curve and the single-cycle raw data. The cut-o€ frequency was selected as the frequency at the end of the first sharp decay of the power spectrum of the ensemble-averaged velocity, when compared to that of the single-cycle raw data. The third step is to calculate the bulk velocity in a single cycle by low-pass filtering and the remaining component of the raw data, or the high-pass filtered parts, to attribute to turbulence. During this step, the time intervals for the digital low-pass filtering are 0.5° CA, which is equivalent to a frequency of 12 kHz at 1000 rpm. Therefore, the maximum frequency of the turbulence analysis is 6 kHz according to the Nyquist criteria. The last step in the cycle-resolved analysis is to calculate the ensemble-averaged turbulence intensity, u0EA, by ensemble averaging the turbulence fluctuation over many individual cycles as follows: u0EA…h† ˆ 1 NM…h† XNc tˆ1 u…h;Dh; i†2 " #1=2 ; …3† where 2° of crank angle window was also used here for decreasing the statistical uncertainty. 3.3. Turbulence scales In order to describe the frequency characteristics of turbulence in engines, autocorrelation coecients at various crank angles were first analyzed with turbulence components obtained from the cycle-resolved analysis. For a non-stationary flow, the Eulerian temporal auto- correlation coecient is defined as Rt…h;/† ˆ 1Nc XNc tˆ1 ui…h†ui…/† u0EA…h†u0EA…/† ; …4† where h is the crank angle, / the phase angle with re- spect to h. The integral time scale can be evaluated, in analogy to that defined in stationary turbulent flows, by taking the integral of Rt over / from 0 to /max, when Rt has a positive value and decays over a suciently long period, i.e. Lt…h† ˆ Z/max 0 Rt…h;/† d/: …5† However, if the autocorrelation coecients oscillate rapidly with their magnitude decreasing, then the inte- gral time scale should be defined in di€erent ways. Three di€erent definitions, as shown in Fig. 6, were used for determining the integral time scale in this study, since there has been no general consensus on that subject. The first definition is given by the value of / at which Rt has the local minimum, namely, the time to the dip point of Rt [19] as shown in Fig. 6(c). The other two definitions of the integral time scale can be obtained by expressing the temporal autocorrelation coecients as the follow- ing empirical formula [20]: Rt…h;/† ˆ exp …ÿ/=Lt…h††: …6† The second definition could be obtained as the 1/e decay time in the first decay region of the autocorrelation co- K.Y. Kang, J.H. Baek / Experimental Thermal and Fluid Science 18 (1998) 231–243 235 ecient [21], and the last definition could be considered as the 1/e decay time of the exponential envelope curve for oscillating peaks in the autocorrelation coecient [16]. The integral length scale of turbulence could be de- fined from the integration of the spatial autocorrelation coecient in the same way as the integral time scale is obtained from the temporal autocorrelation curve. The technique for measuring the integral length scale re- quires the use of two probes and the subsequent ma- nipulation of large amount of data to get an autocorrelation curve. Due to the diculty of this technique, most investigators who tried to estimate length scale in engines have employed an empirical correlation between length and the integral time scale using Taylor’s hypothesis: Lx…h† ˆ U…h†Lt…h†: …7† However, this hypothesis can be applied to determine integral length scale for the case of the flow with con- stant mean velocity, homogeneous turbulent intensity, and very smaller relative turbulent intensity u=U ; …u=U � 1† [22]. Since a strong mean flow may not exist in the cylinder near TDC of the compression stroke, the relationship given by Taylor’s hypothesis is not valid. Therefore, the alternative expression proposed by Tab- aczynski [23] was adopted in this study, since it is more acceptable for the flow condition in the late-stage of compression stroke in engines: Lx…h† ˆ Cu0…h†Lt…h†; …8† where C is a constant of order 1. Fig. 5. Turbulence determination by the cycle-resolved analysis. 236 K.Y. Kang, J.H. Baek / Experimental Thermal and Fluid Science 18 (1998) 231–243 4. Results and discussion 4.1. Tumble flow characteristics Fig. 7 shows the velocity distributions giving rise to a tumble motion in the longitudinal plane and on two cross-sectional planes of the cylinder for the case of the conventional port. The origin of tumbling motion dur- ing the in