Numerical simulation and
Steady-state flow rigs are extensively used in the design of diesel engine intake port designs. Because of the relative simplicity of the intake port and dummy cylinder assemblies, these experiments are well suited for numerical simulation by a state-of-the-art Computational Fluid Dynamics (CFD) package. In this paper the validation of a CFD code in conjunction with the employed simulation models is presented. A total of 4 intake port configurations were simulated and tested. The basic comparisons are made by using Laser-Doppler Anemometry (LDA) measurements of the flow in the dummy cylinders and the results of the numerical simulations. Traditional steady-state flow rig results are used to compare the flow and swirl coefficients.
In the engine design process, two dimensionless numbers, among others, play a dominant role in the initial layout of the engine and its thermodynamic performance (including emissions); the flow, or discharge, coefficient and the swirl number. The first is a measure of the pressure loss introduced by the port/valve combination and the second relates to the rotation of the fluid in the cylinder. These numbers are normally gathered in tests on a steady-state flow rig using either cylinder heads or concept intake port models, normally known as flow boxes. The steady state flow rig measurements are suited for numerical analysis because:
Thus, steady state flow rigs are popular benchmarks for CFD codes because of the importance of flow rigs in the engine design and their suitability to numerical analysis. A number of papers have been published on the subject,ranging from 2D axisymmetric intake ports, as in /1/, to more complex configurations like in /2/ and /3/.
It is widely expected that CFD will play a major role in future
engine designs. As there are a number of commercial and in-house CFD
codes available today, each of them employing one or more numerical methods and
techniques,the validation and proved accuracy of a model of one code cannot be transferred
to another. Thus, whenever a CFD code goes into service somewhere,
the modeling and analysis techniques need to be validated and known.Especially the latter
is very important otherwise the quality of predictions cannot be reliably estimated.
In today's diesel engines two basic intake port designs are widely used:
The ports basically differ in the shape around the valve stem. The helical port concept relies on the assumption that the flow upstream the valve curtain is brought in rotation and therefore exits the valve with an angular velocity component around the valve stem axis. This in turn should result in a higher fluid rotation in the (dummy) cylinder.The swirl of a direct port relies on the interaction of flow and cylinder wall only. More information on the subject of port design can be found in papers like /4/, /5/, and /6/.
As both types of ports are used in today's diesel engine designs, the choice was to include at least one of both types in the project. The helical port is further referred to as Head 1, and the direct port as Head 2.
CFD codes need a geometrical description of the space the flow is in or around. The effort needed to generate complicated meshes is very high. Thus it is useful to use a mesh, or parts of it, more than once. This has been done in this project with the direct port (Head 2). By rotating the port around the valve axis three distinctively different flow situations have been created (see fig. 1):
Fig. 1: Definition of Head 2 port orientations (view from top)
In order to reduce modeling time only one valve lift of each cylinder head has been used.
Traditional flow rigs measure two integrated values, which are normally made dimensionless: the pressure drop and angular momentum. However, they cannot be used to reveal any flow details and hence to verify numerical velocity profiles. Therefore other measurement techniques were needed for this purpose. In this project LDA has been used to obtain tangential and axial velocity profiles in the dummy cylinders at several planes and some diameters.
Fig. 2: Definition of Head 1 measurement planes and diameters
The basic measurement planes in this project are located 1.75 D, 1.0 D, and 0.5 D downstream of the head, using the inner cylinder diameter D as reference length (referred to as Plane 1 to 3 respectively, see fig. 2). Especially plane 1 is important because the impulse torque meter used on the flow rig begins at this distance from the head. It is assumed that the meter destroys all the angular momentum entering it without influencing the flow upstream of the device. However, this ideal assumption does not occur in reality and therefore the angular momentum at 1.75 D which might be calculated from the LDA experiments, or from the numerical results, will not match the results from the impulse torque meter exactly.
In order to gain some insight in the quality of the numerical results near the valve, two additional planes were defined near the valve of Head 1. The planes are located 2 h and h downstream of the head (referred to as Plane 4 and 5),where h is the valve lift.
The velocity measurements were taken along three diameters in each plane. Diameter B is perpendicular to the diameter joining the valve and cylinder axes. Diameters A and C were chosen +/-30° to Diameter B to yield more information spatially about the flow induced by the two heads without overlapping the valve when measuring velocities at Plane 5 (see fig. 2). This configuration is also a compromise between a circumferential equidistant diameter specification and local flow resolution.
A total of 84 velocities profiles has been measured. As the data of Diameters A and C follow the trends of diameter B closely, the results presented in this paper are restricted to the results of Diameter B only.
Finally the mass flow rates were initially specified using the standard flow rig pressure drop of 10 kPa. However, during the LDA measurements of Head 1 it was found that there were particle seeding problems resulting in low data rates. In order to accelerate the measurements of Head 2 the pressure drop was reduced by 30% resulting in lower mass flow rates.
A finite volume code solves the ensemble averaged conservation equations for mass, momentum, and energy, using the standard k-e turbulence model to link the Reynolds stresses with the mean ensemble averaged properties. During the numerical simulations no attempt has been made to tailor the various physical model constants.
Especially the k-e turbulence model constants were left to their default settings (see table 1).
The numerical models were made using some CAD data. Of both heads CAD surface model were available and splines were defined on the outer surfaces. The spline definitions were then transferred to the preprocessor and formed the basis of the port shapes. Drawings of valves and valve seats were used to complete the models. The capability of the code to use unstructured non-orthogonal meshes was heavily relied upon. Quasi-polar meshes were employed around the valve stem and the outer field of the cylinder whereas quasi-rectangular meshes can be found in the inlet and under the valve in the dummy cylinder. The resulting numerical models are shown in fig. 3 and port details in fig. 4. Table 2 shows a summary of model data.
Fig. 3: Meshes of Head 1 and 2 (MS Position)
Fig. 4: Port meshes of Head 1 and 2
It can be remarked that the original CAD data of the port lay-out of Head 2 only consisted of some cross-section definitions. CAD surfaces were made by using this wire frame model and there is no guarantee that the real port was exactly reproduced. Especially the vital area near the transition of port intake and valve seat was rather poorly defined and had to be guessed.
The alternative geometry definitions of Head 2 were easily derived from the initial MS position model by rotating the cells of the port around the valve axis. The valve curtain was divided circumferentially equidistant in 48 cells, thus each row of cells spans 7.5 degrees. The 67.5 and 45 degrees port rotations were set to match the initial cylinder mesh exactly.Therefore the effort needed to create the alternative Head 2 models was minimal.
Because of the high flow rates or pressure drops the flows cannot be considered incompressible (maximum Mach numbers range between Ma 0.42 and 0.5 near the valve seats). Therefore all the simulations were done with the density as a function of local pressure and temperature (ideal gas law). Although the temperature has been included as a variable, gravitational buoyancy effects were not taken into account because it was assumed that the flows are completely dominated by forced convection. The model walls were assumed to be adiabatic.
At the inlet cross-sections the velocity, turbulent kinetic energy and dissipation profiles were prescribed to be uniform. Turbulent kinetic energy intensity was set to be 8% and the characteristic turbulent length scale was assumed to be 10% of the inlet height. As the computations were done fully compressible, mass flow rates were specified at the inlet instead of fixed velocities.
The Linear Upwind Differencing scheme (LUD) was used for all the variables except for the enthalpy which employed the standard Upwind Differencing scheme (UD). The latter was necessary to reduce some, possibly hazardous,instabilities in the residuals of the enthalpy equation.
Finally the well-known SIMPLE velocity pressure coupling algorithm was used throughout all the calculations.
4.1 LDA Measurements
The LDA measurements were performed by LSTM at Erlangen University, Germany. The detailed description of equipment and instrumentation are beyond the scope of this paper and are therefore not presented. Only the major features will be listed below.
First of all, LSTM uses a steady state flow rig with the blower positioned downstream of the head. The air is drawn in form the environment surrounding the flow rig. The flow boxes were extended by conical inlet channels and a straight pipe containing flow straighteners. This flow rig is further referred to as the LDA rig.
Numerical studies revealed that the shape of the velocity profile at the start of the flow box can have a significant impact on especially the swirl number. To determine the uniformity of the inlet flow, an adapter which provided optical access was mounted between the inlet channel and Head 1. It was found that the velocity profile at the start of the flow box was indeed uniform. It is assumed that the Head 2 velocity profiles are uniform too.
The length of the cast Plexiglas cylinders was equal to 3 times the inner diameter. Preliminary tests of the system showed no differences in the desired velocity data between cylinders 2D in length and 3D in length. The Plexiglass was chosen so as to provide sufficient optical access to the entire flow in the cylinder for LDA measurements of the axial and tangential velocities.
Fig. 5: LDA steady-state flow rig
The overall LDA rig is shown schematically in fig. 5. It consists of a base unit with settling chamber upon which the flow boxes and cylinders are mounted. The mass flow rate was determined using a turbine flow rate meter and the air temperature and pressure in the flow meter. LSTM assumed that the temperature of the inlet air did not change enough as it passed through the flow rig to significantly affect the mass flow measurement. The numerical simulations however, indicate a slight temperature decrease of about 7 K. Consequently the mass flow during the measurements was roughly 2% higher than indicated.
According to the availability of LDA systems, velocity measurements were performed using two different, single component, backscatter LDA systems. Both anemometers used a small probe connected to the main optics by a10 m fiber optic cable. The main difference difference lay in the type of laser To measure the velocities resulting from Head1, a laser diode operating at an infrared wavelength was used. System details can be found in /7/. For Head 2, a similar system using an Argon-lon laser operating on the green fine was connected to the fiber optic probe. The entire systems, except for the probe, were located on an optical bench. The optical parameters for both systems are summarised in table 3 for comparison.
Particle seeding was provided by a medical atomiser which generates particles with a mean diameter of 2.25 mm at an operating pressure of 1.0 bar when water is used as the medium. For the current measurements, a mixture consisting of 85% water and 15% propylene-glycol was used. This mixture probably produced slightly smaller particles since the presence of the propylene-glycol serves to reduce the surface tension of the fluid mixture. These particles are assumed to be small enough to follow the flow accurately but large enough to provide signals from the Avalanche Photo Diode with sufficient amplitude and modulation. The particles were introduced upstream of the inlet channels so as not to disturb the inlet flow.
Because the measurements spread out over weeks the laboratory conditions changed over time. Thus the mass and volume measurements were converted to a standard set of conditions which were chosen as 1.013 bar and 288 K (Standard Atmosphere). The laboratory conditions for the velocity measurements are listed in table 4. Note that the standard deviations in the volume and mass flow rates are smaller than the uncertainties with which they can be measured. For this reason, these values can be considered stationary in time within the accuracy specified in table4.
Data rates ranged typically from 50 to 200 Hz with occasional values as low as 5 Hz (near the walls) or on the order of 1 kHz (near the valve). Approximately 2000 measurements were taken per velocity component at each point. The data were ensemble averaged, yielding the mean and fluctuation (rms) velocities of each component.
The blower of the steady-state flow rig (standard rig, fig. 6) is, in contrary to the LDA rig, positioned upstream the flow box. Thus the ambient pressure prevails in the outlet of the mounted impulse torque meter. In the discussion below it is assumed that the meter does not introduce an additional pressure loss.
Fig. 6: Standard steady-state flow rig
A separate numerical study into the effects of the atmospheric pressure and temperature on the resulting pressure drop showed that although the pressure drop differed in absolute values, the dimensionless flow coefficients did not. Thus numerically there is no difference in the LDA and standard rig results.
The influence of mass flow rate on the dimensionless flow coefficient can quickly be studied on a steady-state flow rig. This has been done and the result was that the coefficient was indeed locally independent within a range of 10% around the actual mass flow.
The impulse torque meter on the standard rig was used to measure the angular momentum in the dummy cylinders. The device starts at a distance of 1.75 D downstream of the head and directly records the angular momentum relative to the (dummy) cylinder axis. The effects of varying mass flow on the recorded torque have been studied too. The torque can be converted into a dimensionless swirl number which was found to be locally independent of the mass flow as well.
5.1 Velocity profiles
The numerical results have been gathered similarly on the equivalent diameters as the LDA measurements. In fig. 7 to fig. 14 these results are shown together with the LDA data. The marked lines are the primary velocity profiles and they are surrounded by confidence intervals.
The confidence intervals of the LDA data are directly derived from the rms values. They have been multiplied by two to yield the 95% confidence range. In order to be able to judge this range with the numerical results, an equivalent procedure has been followed using the calculated turbulent kinetic energy. First of all, in the k-e turbulence model all three fluctuating velocity components are assumed to be of equal magnitude. From the definition of k follows: u' = v'= w'. The resulting fluctuating components are also multiplied by two giving a numerical "confidence interval".
Fig. 7a: Head 1, Diameter B (icon, 28 kB)
The results of Head 1 are shown in fig. 7 and especially the profiles of Plane 1 to 3 are very satisfactory. As has been noted before the results on Plane 1 are important because the impulse torque meter starts here. Going upstream from Plane 1 to 5 the resemblance between measured data and the numerical results deteriorates. The largest discrepancies can be found on Plane 5 for the axial velocity profile in the neighbourhood of the valve. Detailed inspection of the numerical results showed that the flow out of the valve is deviated towards the head. This effect can be caused by the local spacing of the grid in the cylinder near the valve in this area. The sudden increase in grid spacing reduces the order of accuracy of the numerics. Moreover, the magnitude of velocity is locally very high and thus a small deviation in the angle results in a large axial velocity component variation. The results of the tangential components on Plane 5 are satisfactory. The agreement of the turbulent kinetic energy or confidence bandwidths is also satisfactory. In general it can be stated that the order of magnitude is correctly forecasted by the code. The largest deviations here are for the axial velocity component in Plane 1 and, again, in Plane 5. An impression of the resulting flow is shown in fig. 8. The rotation in the port is dearly visible together with the prominent band around the core vortex in the dummy cylinder.
The next set of data is shown in fig. 9 for Head 2 in the low swirl (LS) position. Again the agreement between measured and calculated data is good, although the quality is slightly less than for Head 1. The absence of swirl in Plane 1 is dear. Note that the measured confidence bandwidths are all wider than the numerical bandwidths. The absence of swirl is illustrated in fig. 10.
The medium swirl position of Head 2 and the corresponding results are depicted in fig. 11. In this figure only the results of Plane 3 and the axial velocity component in Plane 1 might be considered acceptable. The calculated tangential velocity component in Plane 1, the swirl component, is roughly four times too large compared to the measured profile. The axial velocity component in Plane 2 is in fact completely wrong. Instead of the measured peak in the middle of the cylinder the code calculated a reversed flow there. It seems that in reality the flow in Plane 1mainly dives through the center of the cylinder and is surrounded by reversed flow regions. The onset of this effect can be seen in the axial velocity component at Plane 3. However, the agreement of the axial velocity profiles on Plane 1 is surprisingly good except for the large difference in turbulence/confidence bandwidths. As a whole it can be said that the code forecasts a completely different flow field for the MS configuration. Currently there is no solid explanation for this behaviour of the code.The graphical representation of the flow is in fig. 12.
Finally the results of Head 2 in the high swirl orientation is presented in fig. 13 and 14. The agreement of the velocity profiles in all planes is again very good, especially on Plane 1. However, the measured confidence bandwidths are,as for the LS configuration, all wider than the numerical bandwidths. Note that the streamline pattern in fig. 14 resembles the results of Head 2 MS. Summarising the results: The agreement of both the velocity and turbulence profiles ranges from very good for Head1 to good for Head 2 HS and Head 2 LS. In comparison with the other three configurations the striking failure of Head2 MS is remarkable.
Flow coefficients are related to the pressure drop of the flow as it passes the port/valve assembly. The definition of the flow coefficient is based on the Bernoulli equation. The recorded static pressure in the dummy cylinder is used to calculate a theoretical velocity (mass flow) based on the total pressure upstream of the flow box. The ratio of actual mass flow and theoretical mass flow is said to be the flow coefficient. Thus the flow coefficient is directly linked to the total pressure loss of the system.
The flow coefficient definition used here is:
The numerical models shown in fig. 3 and 4 start at the beginning of the flow boxes and not at the start of the mounted conical inlet channels. However, this is not an imperfection since the contracting flows in these inlet channels hardly introduce any total pressure loss. Therefore there was no need to include the inlet channels in the models as would have been the case if they would have been absent. As the velocities on both flow rigs are very low upstream of the conical inlet channels, the static pressure there equals the experimental total pressure. The calculated and experimental flow coefficients appear in table 5.
This table contains the following results:
These results do not allow any harsh conclusions towards the quality of the numerical predictions. It may be questioned whether the results of one rig are better than of the other. The high Cf value of Head 2 MS on the standard rig is at least peculiar. Certainly, if it is assumed that the majority of the pressure loss occurs in the gap between valve and valve seat, then the flow coefficient should be almost equal for Head 2 cylinder orientations between the LS and MS configurations. On the other hand, considering the 10% difference in Cf between the LDA and the standard rig, the result of Head 2 LS on the standard rig might be incorrect. However, the results of the standard rig are consistent since the flow(and swirl) coefficients were independent of the mass flow rate.5.3 Swirl
The swirl generated by the flow boxes can only be compared with the results of the impulse torque meter. Again a dimensionless coefficient is defined:
Table 6 shows the results:
Again some comments can be made:
The reason that the swirl of Head 2 MS is so large can be directly related to the comparison of the corresponding velocity profiles in paragraph 5.1. However, the same argument does not hold for the data of Head 2 HS. Here the axial and tangential velocity profiles of Plane 1 agree in all three diameters very well for the numerical and LDA results. The explanation is that the standard rig produces a considerably lower swirl number which can be attributed completely to the presence of the impulse torque meter.
In general it can be stated that the CFD code and the employed models predicted the flows well except for one case. The reason of this failure is currently unknown, although a variety of errors, numerical and experimental, might be held responsible.
Flow coefficients have been predicted with acceptable accuracy when compared with the LDA rig. The differences between the LDA and the standard rig remain unsolved and should be investigated. Swirl has been less accurate forecasted, although here too some interesting questions are left to be answered.
The numerical techniques and methods were adequate for this kind of flow. However, the importance of a smoothly varying grid should not be underestimated. The standard k-e turbulence model performed well for especially the high swirl cases. This is a remarkable conclusion since it is generally assumed that this turbulence model should fail for this kind of flow.
Please adress all enquiries
about this site to our webmaster.