Iam working on engine flow field modelling using a commercially available CFD code. The simulation involves moving valve and moving piston with bowl-in configuration. I am imposing a time varying pressure boundary condition at the intake valve, which has been experimentally recorded.

I have used k-e and also RNG k-e turbulent model for approximating the reynold stress and flux quantities. I am getting result for turbulent intensity (i.e turbulent kinetic energy) which is much higher than the experimetnal results normally quoted in literature for similar configuration.

I need information of the following nature. Does the results of turbulent kinetic energy or intensity from the computation include only the turbulent fluctuation component or does it include any cycle to cycle variation component? The mean quantities are based on esemble/favre averaging. I want to know the right basis for comparing computation and expt results. Can somebody clarify.

Thanking in advance.

Sridhar

(1). I don't think you are trying to simulate the turbulence. (2). So, the turbulence kinetic energy is a function of (x,y,z,t). And you are free to run your engine in the way you like. (3). It is likely that your turbulence model is not validated for your engine applications. So, perhaps it is time to do some turbulence modeling on your own for your engine. (4). One thing you have to remember is that: there is no such thing as the universal turbulence model or constants. And you should not expect to get perfect match between the simulation and the test data.

What kind of experimental results do you have and you want to compare it with CFD?

There are experimental results available in the literature for similar configuration engine cylinder-piston arrangement. The results are bascially availble as ensemble averaged mean velocity, cycled resolved turbulence intensities and rms mean velocity fluctuations.

The expt results is obtained more or les as follows. The velocity -time series data obtained using LDV over a number of cycles are ensmeble averaged to get the mean velocity. Further the ensmeble averaged mean velocity is subtracted from the indiviual cycle to get the turbulent fluctuations. The turbulent fluctuations in turn are filtered based on certain cut-off frequency to classify into turbulent intensity and mean velocity variations.

The results obtained using computation (k-e modelling) i.e turbulent intensity more or less matches the experimental results, which is rms sum of turbulent intensity and mean velocity fluctuation(cycle -cycle variation).

The precise reasons for the cycle-to-cycle variations in engines are unknown. People have attributed these variations to inflow turbulence characteristics, acoustic feed back from flow downstream of the engine, spark locations, fuel injection among other things. The chaos in the engine (that leads to the variations) is sensitive to one or more of these characteristics.

Most simulations are deterministic (unless you use random numbers somewhere). The k-e type models are designed to capture flow behavior averaged over an ensemble. The ensemble has several realizations of the flow field. Each realization is different despite the same average behavior because the system is chaotic and the chaos can be triggered by changes/disturbances that may not be quantified/represented using mean quantities. So the question now is what type of disturbances. Is it just the fluid dynamic disturbances or also those that can cause cycle to cycle variations. My guess is the former. (Most) turbulence models are designed to capture averages over different realizations resulting from fluid dynamic disturbances.

The problem with turbulence modeling is that you have to prescribe all inflow/boundary conditions for all quantities being modeled. This can be a big problem not only in RANS approaches like k-e, second order models but also non-algebraic LES models (like the one-equation models). So, the question to you is how did you presribe k and e in the code. You can measure k but e is often an estimate. Are your results sensitive any of the inflow conditions (including k and e).

The initial conditions used for simulation are very small values for k & e through out the flow field. I did some variations in initial conditions i.e in the velocities and pressure, but not the k & e. The results were almost identical.

If you are predicting higher turbulence intensities than in experiments, perhaps, your k-e model is not calibrated well. The question of cycle-to-cycle variations in a simulation is difficult one to answer. At this point, you should turn to some one who has experience with actual engine simulations (most of us often speculate).

Here are two people I can suggest. Profs Celic at Univ. of W. Virginia and Haworth of Penn. state. Both these people have worked on LES also in addition to k-e models (within KIVA). Recently, I heard Prof. Celic claim that the cycle-to-cycle variations have been captured in a LES. These (predicted) variations have very low frequencies than I expected. He suggests that the residual effects of turbulence from the previous cycle influences the evolution of the flow in the present cycle in a rather chaotic manner thereby leading to cycle variations. You may have better luck with your question with these people.

(1). All I can say is, IC engine transient simulation is as difficult as testing.

I was not aware that such kinds of experiments were possible in IC engines. It's good you've got such good info to compare to. My only comment on your question is that you may want to consider using a stress tensor turbulence model (I rarely suggest this but your case seems like one where good results with an isotropic model would be luck)

Let us say you conduct an experiment on a IC engine and measure the flow field over just one cycle. There is no question of cycle-to-cycle variations. Then, if you were to simulate the flow over just one cycle, what do you expect to see ? Since both your experiment and simulate involve just for one cycle, you are not aware of the so called cycle-to-cycle variations.

The real question is whether the flows in different cycles are "realizations" of the same mean field or completely different fields. So you have to answer the question if the differences in cycles arise from turbulence or from some other mechanism (like the spark characteristcs, variations in the injected fuel-air, acoustic feedback etc.). It could also be that the differences are due to turbulence in the flow outside the cylinder. In all cases except where the differences are due to in-cylinder turbulence (in which case each cycle is a different realization of the same mean field), the k in k-eps model does not include contributions from cycle-to-cycle variations.

[machineengine]

How I can numerically calculate to SMD, spray penetration in Fluent Program. Must I use emprical equations for SMD or Cone angle in Fluent?