Intake CFD

[BeaverGuy]

I was wondering how everyone set up their CFD analyis of their plenums. Whether you do a steady state analysis with all runners pulling air at the same time or set up a transient analysis based on the engine timing and pressure at the runner base? For those doing a transient analysis how big a time step in terms of crank angle are you using and how many complete intake cycles are you simulating?

I ran a transient analysis using a time step equivalent to one degree of crank rotation over two revolutions of the engine. However the induction strokes that overlapped the start and end of the simulation were greatly affected. To alleviate this I am thinking of running the analysis over six revolutions that way all cylinders would have at least two uninteruppted induction cycles. Then on the time side I think a 5 degree crank angle step should give me good results while reducing the processing time even with the longer analysis.

[BeaverGuy]

I was wondering how everyone set up their CFD analyis of their plenums. Whether you do a steady state analysis with all runners pulling air at the same time or set up a transient analysis based on the engine timing and pressure at the runner base? For those doing a transient analysis how big a time step in terms of crank angle are you using and how many complete intake cycles are you simulating?

I ran a transient analysis using a time step equivalent to one degree of crank rotation over two revolutions of the engine. However the induction strokes that overlapped the start and end of the simulation were greatly affected. To alleviate this I am thinking of running the analysis over six revolutions that way all cylinders would have at least two uninteruppted induction cycles. Then on the time side I think a 5 degree crank angle step should give me good results while reducing the processing time even with the longer analysis.

[woollymoof]

I use steady state with all runners pulling at the same time. The reason I choose this way was primarily due of lack of computer power. People at uni that know about CFD say that you should really do transient analysis lasting for AT LEAST one second. But, they were really talking about transient analysis that tries to alleviate diversion.

[Ben Inkster]

I am using fully transient CFD analysis coupled to 1D boundary conditions. This usually takes about 5 engine cycles to become reliable and i have experimented with different crank steps ranging from 0.1 degrees to 0.4 degrees. I was able to get away with 0.4 degrees for simulations under 9000 rpm.

There are many big advantages to tranisent simulations, we have done plenty of steady state CFD and found mass flow distribution predictions through the runners of 10% difference, whereas in reality it is only 2%. Transient can completely remove these errors and clean up any problems associated with 1D analysis as well. The problem (of course) is the computational time, i have been running some simulations for over a week! but most have been 2 days.

I'm not sure what CFD package you are using BeaverGuy but i am using Ricardo Wave and Vectis.

Hope this is some help.

Cheers

-Ben Inkster

UWA Motorsport

[Mark100]

Ben, what kind of computer are you using for your simulation?

[Marc Jaxa-Rozen]

What kind of mesh sizes are you guys using for reasonable accuracy? Two revolutions with 3 degree crank steps in Fluent takes me about 30-45 minutes on a simple log manifold, which I guess means my mesh is really too simple.

Marc Jaxa-Rozen
École Polytechnique de Montréal

[BeaverGuy]

I used a level 3 auto mesh and auto refine turned on in FloWorks and it took about 40 minutes with 8 refinements on a 1.2Ghz Athlon w/768MB of ram. As that was a preliminary run through I figure I might want to reduce the mesh size but the plenum is a fairly simple eliptical cylinder.

[Ben Inkster]

I did have access to a HP 3ghz quad parallel processor for a month in which time i tried many different mesh sizes up to 300 000 nodes using quads, but i found this to be way too extreme.

At the moment i am using a standard 2.5ghz pentium with 512 ram that seems to work fine as long as i keep the mesh density low. The only problem i have encountered is memory consumption during post processing.

The meshes i have constructed are usually really coarse in the plenum volume and runners, but I add refinement around the runner/plenum interfaces, the restrictor and diffuser, this is where i find the most complex flow.

cheers

-Ben
UWA Motorsport