Plenum modelling in WAVE

[Mikey Antonakakis]

http://i7.photobucket.com/albums/y294/hhspunter/intake1-1.png

http://i7.photobucket.com/albums/y294/hhspunter/intake2.png

I'm making about ten hp less with the first design. I wish I could figure it out myself, but I don't know why it makes a difference (or more like why it is one way and not the other). I know the runners are totally different lengths, but I'm talking about when they are the same. I eliminated all of the factors I could think of and it came down to that. Any help would be greatly appreciated. I was VERY happy with the numbers WAVE was giving me for the "more powerful" version. I've checked out last year's intake in WAVE to see how close it is to what our dyno chart looked like. Apparently there is an 18% drivetrain loss, and a characteristic drop in torque at 5000 rpm that doesn't show up in the WAVE model, but other than that it's spot-on. That's really the only verification I've been able to do, and it was a log-type plenum which is pretty standard for cutting up. I also didn't have the right muffler in the model.

Sorry if this was a very rambling post, here's some cliffs:
anyone have suggestions for how to model that plenum to be as accurate as possible in my simulation? I am open to using CFD software if CFDesign can interface with WAVE.

EDIT
I should add that last year's numbers came from a chassis dyno with our car strapped down very tight, with racing slicks at about 20psi.

[Mikey Antonakakis]

http://i7.photobucket.com/albums/y294/hhspunter/intake1-1.png

http://i7.photobucket.com/albums/y294/hhspunter/intake2.png

I'm making about ten hp less with the first design. I wish I could figure it out myself, but I don't know why it makes a difference (or more like why it is one way and not the other). I know the runners are totally different lengths, but I'm talking about when they are the same. I eliminated all of the factors I could think of and it came down to that. Any help would be greatly appreciated. I was VERY happy with the numbers WAVE was giving me for the "more powerful" version. I've checked out last year's intake in WAVE to see how close it is to what our dyno chart looked like. Apparently there is an 18% drivetrain loss, and a characteristic drop in torque at 5000 rpm that doesn't show up in the WAVE model, but other than that it's spot-on. That's really the only verification I've been able to do, and it was a log-type plenum which is pretty standard for cutting up. I also didn't have the right muffler in the model.

Sorry if this was a very rambling post, here's some cliffs:
anyone have suggestions for how to model that plenum to be as accurate as possible in my simulation? I am open to using CFD software if CFDesign can interface with WAVE.

EDIT
I should add that last year's numbers came from a chassis dyno with our car strapped down very tight, with racing slicks at about 20psi.

[EPMAl]

When you're using wavemesh, the only thing it does is approximate it to a geometry wave can take (such as complex y-junction, ducts, orifices, etc). In your first design, your plenum is approximate as a cylindrical y-junction with an expanding length and surface corresponding to your half-sphere. In your second design, since you split it in multiple volumes mostly modeled as ducts, the divergence is approximate more accurately. In my case, I also have a 10% difference between the dyno curve and the wave curve. Wave can only be as precise as the input you put in your model. Get as much data about your engine as you can (valve discharge coefficient, fuel properties, etc), you'll see your model's accuracy improve (we were at a 15% difference before I changed our discharge coefficient and throttle body modelisation).

[Mikey Antonakakis]

I don't have a throttle body, but as far as I know the rest of the engine model is as accurate as I can make it. If it is indeed logical and accurate to model it as a small y-junction, with the rest of the plenum respresented by ducts, I am going to be a very happy man. We'll be making about 25% more power this year according to the models.

[Charlie]

Honestly, WAVE is just not the proper tool for this type of thing. Folks go crazy with Wavemesher, but it only converts the result into accurate volumes. It's good for the acoustic model, but not flow. It's 1D, it can't.

WAVE is all about accoustics. the flow losses come from coefficients. They are reasonably good approximations for simple models, but not for stuff like this. You need to measure flow losses and incorporate the results in WAVE for power level accuracy.

Use CFD to help you with complex intake geometery if you want to compare power level... not WAVE. CFD coupled with WAVE is ideal.

[Mikey Antonakakis]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Charlie:
Honestly, WAVE is just not the proper tool for this type of thing. Folks go crazy with Wavemesher, but it only converts the result into accurate volumes. It's good for the acoustic model, but not flow. It's 1D, it can't.

WAVE is all about accoustics. the flow losses come from coefficients. They are reasonably good approximations for simple models, but not for stuff like this. You need to measure flow losses and incorporate the results in WAVE for power level accuracy.

Use CFD to help you with complex intake geometery if you want to compare power level... not WAVE. CFD coupled with WAVE is ideal. </div></BLOCKQUOTE>
Okay, so time to learn CFD! http://fsae.com/groupee_common/emoticons/icon_biggrin.gif

But in all seriousness, does anyone know off the top of their head if CFDesign will work with WAVE? I'm talking like stuff from the advanced tutorials here, like what VECTIS is for (if I remember correctly). If not, I'm guessing I would be inputting data from my CFD software manually... I think it's time I start using my free support http://fsae.com/groupee_common/emoticons/icon_confused.gif

[Professor Gas Can]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Charlie:
It's good for the acoustic model, but not flow. It's 1D, it can't.
</div></BLOCKQUOTE>

Just to clarify, 3-D flow yes. 1-D no.
1D is entirely capable of predicting mass flow rate into the cylinder (which predicts the power, which is what you need). You just need a correlated manifold with, like Charlie said, flow losses from coefficients. Some of the ways to do this is with a flowbench, or seperate CFD calcs. But yea, you would use CFD to design a manifold/port combination in terms of minimizing a pressure loss, then capture that loss in a coefficient and slapping it on the 1D for power predictions.
I think I basically just said what Charlie did.

[Mikey Antonakakis]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Professor Gas Can:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Charlie:
It's good for the acoustic model, but not flow. It's 1D, it can't.
</div></BLOCKQUOTE>

Just to clarify, 3-D flow yes. 1-D no.
1D is entirely capable of predicting mass flow rate into the cylinder (which predicts the power, which is what you need). You just need a correlated manifold with, like Charlie said, flow losses from coefficients. Some of the ways to do this is with a flowbench, or seperate CFD calcs. </div></BLOCKQUOTE>

So I'll probably be going with the CFD option, hopefully that software will be up and running tomorrow (working on getting a license at the moment). If I may ask, how would I go about getting these flow losses once I have my model meshed? I know from my VERY limited CFD experience that I have to choose my openings, walls, etc. Past that I haven't done any CFD, so I'm not even sure what kinds of things you can tell it to do. Do I, for instance, set up pressures at the openings, and then it will do calculations? Feel free to tell me to go look it up myself, I'm just kinda lost at this stage, and it's going to be a very busy semester to say the least. But I can't wait!

[Mikey Antonakakis]

Good to know we're all on the same page lol. I'll just bust open the CFDesign manual and start reading, you guys don't have to answer if it's something that is easy to find. I'll do my own research and come back if I have more questions. I appreciate the help guys.

[samphlett]

There is no currently supported link between WAVE and CFDesign. Supported links are VECTIS/STAR-CD/Fluent/FIRE.

There are two options here:

1) Write your own CFDesign link to WAVE. How to do this is is described in the WAVE installation (RSimlink).

2) Consider VECTIS. Also available for FSAE people.

[Mikey Antonakakis]

I do have access to Fluent, just not on the same computer I have WAVE installed on. I'd have to use the very slow university computers.

I did not know we could get VECTIS too, that's very good news! I'll get on that right away, thanks for the info!

[Charlie]

Even though WAVE is meant to be used with an IC engine, it doesn't have to be.

Therefore you can set up an individual component in WAVE and set the boundary conditions, just as you would in CFD or on a flowbench.

[Mikey Antonakakis]

I just have to say I love Ricardo. I asked for a license for VECTIS less than two hours ago and already have the license file, no questions asked.

[Bazanaius]

When I asked for a VECTIS license file (when we got WAVE originally) I was told they didn't give it to Students for free... :-(

I'll be going back to them!

[Mikey Antonakakis]

Okay so I THINK I got it working... I want to view the results in Phase6, but I can't find a POST file to open.... there were no errors during the run and as far as I know I set everything up correctly, I checked probably 20 times because I've been running into so many issues. Any ideas?

[Mikey Antonakakis]

Turns out I didn't have VECTIS and WAVE coupled correctly. Everything is working now, I just have to strike a balance between run time and timestep length (too few timesteps per cycle, and it crashes). One solution I am exploring is the Computer Science department, haha.

[Mikey Antonakakis]

Turns out I didn't get it working, hahahaha

My units were wrong.

So for those of you looking to do this, when you save a CAD file to import to VECTIS, make sure it's in mm!

Simulation is running now, and should have an elapsed time of two hours with timesteps that correspond to a crank angle of one degree at 11000 rpm. Turns out things work a lot better when your intake model isn't 25.4 times too small. No more mach 4 air in the restrictor, no more crashes. And so much faster too! It would have taken 80 hours before, if I would have ever gotten it to work.

One question though, I'm sure it's been discussed already, but it's a simple yes or no... is it logical for the velocity to be slightly over mach one just after the restrictor throat?

[roepke44]

Yes it is logical. A choked nozzle will go supersonic after the throat. Then if the downstream pressure is too great (and it is in an engine), it will cause a normal shock in the diverging section to bring the flow subsonic. Normal shocks are a huge source of total pressure loss, but they are unavoidable when the flow is choked.

[samphlett]

Mikey,

When you've mastered the beast that is phase6, be sure to post some pretty pictures.

Well done on getting it all to run.

- Steve (Not that one)

[Mikey Antonakakis]

Phase6 is actually what I found easiest, haha (that is, once I got past not having any POST files to work with... turns out phase6 is pretty tough where there are no files to open haha).

As of right now, I'm trying to run a several-case sim... like 5000rpm, 7k, 9k, 13k.... just to fill in a very rough torque curve to see which 1-d setup is more accurate. I'm having trouble getting that to run though, I feel like I'm doing something that wasn't really designed into the software... or it's just user error, as usual http://fsae.com/groupee_common/emoticons/icon_rolleyes.gif But in all seriousness, I'm running case 1 and 2 just like you would for a regular coupled sim (the 3-step process, then trying to use case 3 and 4 for a different RPM, 5 and 6 for another rpm, and 7 and 8 for a final rpm. Is this the correct way to do that? I haven't found anything in any of the documentation about it. Any suggestions would be greatly appreciated.

Anyway, here's a GIF
http://i7.photobucket.com/albums/y294/hhspunter/intake7-new.gif

[samphlett]

"As of right now, I'm trying to run a several-case sim..."

Have you tried the "postponed" option? It allows the 3-stage coupling to run in a single case. The next case kicks it all off again with a different VECTIS post file. At least I think that's how it works.

[Mikey Antonakakis]

Okay so from data points I have at 7k, 9k, 11k, and 13k, it seems the more generous design is more accurate, which is good news that makes me happy.

Also, some more good news is that last year's log-type plenum has been greatly improved upon as far as air distribution to cylinders is concerned. For those same data points, last year's ranged from 6.2% to 9.0% difference in mass flow to the cylinders. This year the values range from 0.49% to 1.0%. I'm pretty happy about that, too. Sure will make my life easier when it comes time to tune.

[Mark100]

Hello. I don't normally post here, but a friend of mine sent me a link to this discussion. I've been doing engine modeling simulation (professionally) for a bit over four years, on both four and two strokes, designing intakes, mufflers, cams,etc. Hopefully this gives me some qualification to make some statements on what you can acheive with engine modeling. My aim here is to perhaps help in your use of engine modeling/simulation.

Here are a few points.....

1. Wavemesher will allow you to cut a geometry into a mesh, but I prefer using an automatic mesher tool such as the complex geometry tool (aka WB3D). Wavemesher is extremely tedious to get a reasonable mesh. I would suggest you construct your geometry using primitive bodies in WB3D and mesh it there. Wavemesher is also not going to be consistent as using the automatic mesher.

2. With regards to modeling intakes for power prediction, WAVE or GT-Power are excellent tools. You can get extremely close to the real VE curve even with complex intake geometries. If you're simply looking for power prediction, it doesn't take a very fine mesh, to get reasonable VE curves. The Y-junctions in WAVE do capture flow losses. Losses are captured both by abrupt expansions of ducts into volumes and by expansion losses throughout the volume mesh itself. Wall friction is also captured but is really of a much smaller concern anyway. These losses will be captured with no other input from you, the user, other than merely putting in the correct geometry. (The exception here being the restrictor (see bottom)).

Separation is not truly predicted as it's not a full 3D (ie- CFD) solution. However, this is a touchy subject, as depending on your geometry and mesh, you can capture some "separation" or high velocity gradient effects. Again though, most of the losses from this geometry would be captured at the abrupt expansion.

Also, you have to be careful with mesh layout. Tilting the mesh with respect to the axis of the highest local flow rates can artificially increase flow losses in the intake. This isn't a huge effect, but something to watch for.

In general if you want to see how your model is performing, at least with only pure flow losses, you can compare your WAVE flow prediction against a flow bench. While a flow bench is not an engine, it does allow you to separate flow from pulse/wave effects. This way you can see how well the flow prediction part of your model is doing. To give you an idea, it is very reasonable to expect a WAVE muffler model to flow within 2-4% of prototype mufflers. The flow prediction is before a proto is even made. No "tweaking" or model manipulation is done to get within the 2-4%, and this is a muffler with lots of perf sections (ie - lots of highly separated flow). This is not hard if you use an acoustic grade mesh.


3. Mesh density. What are trying to acheive? With the plenum showed in the first post, five junctions would be sufficient to capture power tuning effects, and VE imbalance to a somewhat reasonable degree. If you want to capture noise output
accurately, five junctions would be woefully inadequete. Here the rule of thumb given is 6 elements per wavelength. I feel this is too low, and 10-12 per wavelength is more realistic. Choose your highest frequency of interest. For an Yamaha R6, say you're interested in sound output up to 15,000rpm, going up to 8th Order, that's going to be 2000Hz. At c=340m/s, and 10
elements/wavelength, you'd be at about 17mm elements. Probabaly don't want to run that fine, but now you see why wavemesher is an utter and total waste of time if you are really interested in acoustics.

4. Specifically in regard to the plenum displayed in the first post. This is really not an accurate representation of the intake as the engine will truly see it. There is only Y-junction with a very bad aspect ratio at the bottom. All four runners are drawing from the same exact volume. That intake model is basically saying all runners come into the EXACT SAME LOCATION.
Do you see why? There is no spacial distance between the runners when you connect them to one volume. Also having that very thin Y-junction might drive your time step down making your model take much longer to run (remember that it's an explicit solution).

Where you cut the restrictor is fine. Leave that as a duct. Cut the plenum in half using the same plane orientation you have already used. Then cut the bottom volume in four quarters so you have a separate volume over each runner. You should now have 5 y-junctions. This is the minimum I would go with. Do not use ducts to model large volumes.

5. CFD is great, but it is computationally very expensive. If you can get a decent model, you allow yourself to spend more time to burn through iterations. With that said, the downfall of the WAVE model will not be the intake plenum volume itself. The downfall and the major reason there will be a disconnect between a GOOD MODEL and a real VE curve (notice I didn't say torque curve), will be due to the separation losses in the restrictor. WAVE is assuming 1D flow in the restrictor, which means you have basically made an assumption of no separation, which is invalid. This is where the CFD will shine. The 1D engine model will even pick up the supersonic flow, however, again, the separation losses in the restrictor (which it doesn't know about) will dominate the losses.


6. You can run multiple rpms with a coupled solution, but be mindful of the number of cycles your running. Higher intake volumes, high overlap, higher rpm, and separation losses will tend to increase the number of cycles required for convergence. How many cycles are you running for each rpm? 5, 10, 15, 20?

Sorry for the long-winded posting. I hope this is somewhat helpful.

Cheers,
Mark

[Mikey Antonakakis]

Mark, that was an EXCELLENT post. Thank you so much.

I'd first like to say that we are on the exact same page for your #4. I got that suggestion from Ricardo Support, and the results seem to be the most accurate so far. I'm actually running some coupled simulations as I write this (well, only one at the moment). But when I go to sleep here in a few minutes, I am going to set up a few to run at the same time. I'm just doing each separate RPM as a different WAVE file, rather than different cases within a single WAVE file. I couldn't figure out why it wasn't working as multiple cases, so I just gave up. It has no real effect on computing time, just makes it a little annoying to make sure I am only changing one variable (RPM) between several different models. To go back to your WAVEMesher suggestion, here is the latest mesh I have been working with:

http://i7.photobucket.com/albums/y294/hhspunter/FinalIntake.png

I'm pretty sure that is what you were suggesting.

As for #5, what you stated is, along with my intake geometry, why I am using the coupled analysis. So far the differences have been small between my WAVE model and my coupled model. I'm actually using a different restrictor now than in the previous iterations (the one for sale from AT-Power), and I have that modeled up accurately in the WAVE model and the VECTIS model. I played around with my lengths in WAVE to get the shape I wanted, now I am using VECTIS to verify for a few RPMs (torque and power peak plus one or two others). As far as computing time, my WAVE model (without a muffler) was converging in around 800 timesteps when run as a standalone, and in five or less cycles (I think I said that right). When I tried to run all of the simulations in one run, using different cases for each RPM, I believe I set up the postponed process to run 7 coupled cycles.

Anyway, I hope I replied to all of your points. Again, thank you for the great response, it is very helpful. Our dyno has been orderd as far as I know, and my intake construction will begin this week, so I'll be sure to post again on the accuracy of my model.

EDIT
As far as computing time goes, without a muffler each RPM was taking about 2 hours to run, on a Pentium 4 desktop running XP. I got in contact with the computer science department but I have a feeling at this point it'd end up taking longer. I have to start building this thing if I want to meet our team's February 28th deadline.

[James_C]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">
Also, some more good news is that last year's log-type plenum has been greatly improved upon as far as air distribution to cylinders is concerned. For those same data points, last year's ranged from 6.2% to 9.0% difference in mass flow to the cylinders. This year the values range from 0.49% to 1.0%. </div></BLOCKQUOTE>

Interesting work...
Was your 2008 log plenum tapered? In the simulation did the cone plenum and log plenum have the same volume?

Intuitively it seems to me that the firing order of these inline 4's (1-2-4-3) might dominate the cylinder-cylinder filling disparity to a greater extent than plenum geometry.

[Mikey Antonakakis]

It was tapered. They have a similar volume, although I think the log plenum was a bit smaller.

[EPMAl]

Has anybody tried to do transient simulation in Wave? Rigth now, by enlarging my plenum volume, the power and torque curves go up (which is normal). I just don't want to hurt my transient response with a too large volume. So has anybody tried transient simulation in Wave and how good is it to caracterize response?

[Mikey Antonakakis]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by EPMAl:
Has anybody tried to do transient simulation in Wave? Rigth now, by enlarging my plenum volume, the power and torque curves go up (which is normal). I just don't want to hurt my transient response with a too large volume. So has anybody tried transient simulation in Wave and how good is it to caracterize response? </div></BLOCKQUOTE>
I did transient, but my variable was RPM, it seemed to work great...
As far as plenum volume is concerned, for us at least, it's not really worth going too into detail. Throttle response for most FSAE teams, I feel, is going to be almost solely limited by fuel and ignition mapping... if you're at the point where that is no longer your limiting factor, I'm impressed, and I hope someone can answer your question. WAVE support has been very helpful to me, and I think Mark100 from a few posts earlier would know how to answer your question as well.

[EPMAl]

Since we're running on a twin cylinder engine, I think the limit where the plenum volume comes into account is before when using a 4 in-line since each cylinder has twice the displacement volume as a 4 in-line cylinder. Therefore at the same rpm, it has to move more air and the inertial effects become predominant. I'm not sure what I'm saying isn't BS but just for safety, I'll keep my plenum volume not too big.

[samphlett]

EPMAI,

This is one of the first publications showing transient response validation data. It's quite old now, so some of the images look a bit different and it is a turbocharged diesel, but I think it demonstrates the principle.

http://www.ricardo.com/downloa...rrent_simulation.pdf (http://www.ricardo.com/download/pdf/wave_matalb_concurrent_simulation.pdf)

The thing to realise is that WAVE & other similar 1D simulations are transient by definition in that the flow is unsteady. When cycles start to repeat themselves, the model can be thought of as steady-state although there's nothing in the model that's in any way steady.

(The kind of transient Mikey mentions is really a quasi-steady transient. These are routinely used for intake/exhaust noise, because that's how they're typically measured on a dyno or test bed.)