Use of Computer Simulation

[xxChrisxx]

It has NOTHING to do with the people nor their abilities as engineers. Not being able to get an accurate answer first time it is an inherent flaw in SIMULATION. I find it highly annoying that you are bashing these people for being crap, when you simply don't know what is involved with simulation.

I will give you a very brief intro to CFD simulation. The stuff below doesn't een come close to accurately describing the complexity of the modelling.

Refining a mathematical model (this is NOT the shape of the car but the equations modelling the physics of the flow) is an iterative process. A good guess to he right flow physics, say choosing the right flow model will get you 90% to a decent solution. To get more accurate required very detailed and very boring tedious work, analysing model outputs and comparing them to computer code.

We have several models for flow already, they all basically do the same thing, which is solve Navier-Stokes equation numerically (as a non-linear pde it can't be solved y conventional techniques:


The fact that its a numerical solution used, mean that it's impossible to get a correct answer, you can just get answers of increasing accuracy. This is done for the amount of cells you have in your CFD and it can track the changes to pressure/velocity of the fluid.

HOWEVER! The solution doesnt invovle plugging ang chuggin this, becuase it would try to solve the equation for, basically every molecule of air. Which is computationally almost impossilbe on current systesm.

As such very smalls flows are 'guesstimated at' by turbulence models, taken from the Reynolds averaged version of the N-S equation (RANS). There are lots of tubulence models. On the package i've used.

S-A
http://www.cfd-online.com/Wiki/Spalart-Allmaras_model
K-e
http://www.cfd-online.com/Wiki/K-epsilon_models
K-o
http://www.cfd-online.com/Wiki/K-omega_models

http://www.cfd-online.com/Wiki/RANS-bas ... nce_models


These basically involve putting in constants, to describe the turbulence. The engineer does not know these constants becuase they change for every flow. So all the data in the world will not give you the correct constants, they have to be found by trial and error.

This is done by comparing model output to something that has more fidelity to reality. In this case a windtunnel. Windtunnels are incredibly accurate if set up and the flow measured correctly. This can be then fine tune the model to gie correct results.


Now do you see, why removing the thing to compare the model against doesn't work. You can't fine tune it to get the correct answer. This is NOT the fault of the engineer/aero guy. Even if you didn't understand a word of this, please stop insulting the engineers who do this for a living.

[SZ]

Thanks, Chris, for going into the maths I couldn't be stuffed posting... well in.

And for the uninitiated, his post is the tip of the iceberg.

[xpensive]

Thank's Chris, well xplained. But before comparing the CFD results to the windtunnel's, the tunnel itself should be
calibrated to reality by running the car up and down an airfield or something, correct?

[SZ]

Thank's Chris, well xplained. But before comparing the CFD results to the windtunnel's, the tunnel itself should be
calibrated to reality by running the car up and down an airfield or something, correct?

Think about what you'd look at when calibrating/validating...

[xpensive]

Downforce and drag comes to mind, anything I forgot?

[autogyro]

Thanks, Chris, for going into the maths I couldn't be stuffed posting... well in.

And for the uninitiated, his post is the tip of the iceberg.

Hmmmm.
I have watched the development of wind tunnels and CAD for nearly four decades and been honored to have discussed aero design with some of the top aircraft and car designers.
I am pleased in the statement made, that accepts the discipline as partly an art and not pure science and math.
It is of course, highly complex and attracts extremely clever people. I do not deny this.
However, I still cannot understand why sufficient experience and data has not been collected over the many decades of wind tunnel use, to allow the calculations needed today, to be done almost completely on computer.
Repeating your true statement, that it is a highly complex and difficult discipline is not sufficient to answer my question.

[xxChrisxx]

However, I still cannot understand why sufficient experience and data has not been collected over the many decades of wind tunnel use, to allow the calculations needed today, to be done almost completely on computer.
Repeating your true statement, that it is a highly complex and difficult discipline is not sufficient to answer my question.

These basically involve putting in constants, to describe the turbulence. The engineer does not know these constants becuase they change for every flow. So all the data in the world will not give you the correct constants, they have to be found by trial and error.

Jesus wept... read the damn posts would you. Data can give you an indication, but for every change of scenario, you need to find new constants that best fit reality.

That is why. That IS the answer. It's perfectly sufficient.

Also CFD IS NOT CAD.


Now to my main point. Stop putting designers down for things that they have no control over. You are arguing and judging them when you are clearly ignorant as to how CFD simulation works. If you did know the first thing, you'd realise what you were expecting them to do is impossible at this time. Desigers could fluke a high fidelity simulation, but to accurately work towards one requires practical testing.

[autogyro]

Blimey mate, I expect Jesus did weep and perhaps I am closer to reasoning why.
So fluid dynamics is sooo complex and difficult, it is impossible to work out constants.
It has been in use since before WW2 and yet it is still a black art.
Wow.
Surely this is 'witchcraft' and not science?

[xxChrisxx]

Blimey mate, I expect Jesus did weep and perhaps I am closer to reasoning why.
So fluid dynamics is sooo complex and difficult, it is impossible to work out constants.
It has been in use since before WW2 and yet it is still a black art.
Wow.
Surely this is 'witchcraft' and not science?

Thats correct, its impossible to accurately work out the constants of a turbulence model. You can use previous data and scenarios to get you close to a good fit but not a fit to the level of accuracy needed in F1.

Pre-70's they got around this problem by not acutally solving the problem. It was totally impractical to solve the N-S equations, so they used empirical methods (like windtunnels, smoke flow, tape, wool tufts) and simplifications. For example, Bernoullis equation is a highly simplified form of the Navier-Stokes equations. Hoever it's not the solution of these equations that poses the problem.

It's the turbulence models that pose the problem. There are literally hundreds of purpose written turbulence models, some are valid for only a very small range of flow conditions. So you could apply 1 model that worked for a previous car, but a new diffuser would create different flow conditions, making the model brake down in this area. Getting the correct model, with the correct constants first time, requies a huge amount of luck, or painstaking testing.

And you are perfectly right, CFD is a black art. It's a scientific approach, but does require some luck, judgement annd voodoo in attempting to get a high fidelity model.

Eventually computers will become fast enough to allow a direct solution to the equations, no turbulence model needed. On that day you wouldnt need to test agaisnt a wind tunnel as everything coud be done on computer reliably. This will require orders of magnutude more computing power.

Edited out my analogy, I don't want to distract from the main topic.

[noname]

Blimey mate, I expect Jesus did weep and perhaps I am closer to reasoning why.
So fluid dynamics is sooo complex and difficult, it is impossible to work out constants.
It has been in use since before WW2 and yet it is still a black art.
Wow.
Surely this is 'witchcraft' and not science?

what do you mean by "constants" and how would you like to use them as a help for designers ?

[Jersey Tom]

Thank's Chris, well xplained. But before comparing the CFD results to the windtunnel's, the tunnel itself should be
calibrated to reality by running the car up and down an airfield or something, correct?

Think about the kind of detailed outputs you'd look for in CFD... sensitivity to yaw, pitch, front and rear ride height... not going to get yaw sensitivity just driving in a straight line unless you start doing NASCAR-esque stuff with rear toe.

Plus you'd want to be able to separate out aerodynamic sideforce and yaw moment from tire contribution.. good luck with that, even with WFT's.

Not to mention it's not exactly a controlled test environment. What if there's wind? What if the wind is different on one side of the field than the other? What if it's a hot day and the air at the ground level is hotter and lower density than say the air getting to the rear wing?

However, I still cannot understand why sufficient experience and data has not been collected over the many decades of wind tunnel use, to allow the calculations needed today, to be done almost completely on computer.
Repeating your true statement, that it is a highly complex and difficult discipline is not sufficient to answer my question.

Aero engineers have developed all sorts of models and technology over decades of testing, but there are still different ways of tackling the problem. There are quite a few turbulence models for example. Personally I have very little CFD background, but the fact that you're trying to approximate really small scale extremely dynamic flow, with a 'macro' level simulation, is inherently going to be inexact. So of say 10 turbulence models you can use, all equally valuable in different applications, which do you choose? When?

Also, in theory you could go wild with mesh refinement but then you run into computational problems. Large scale CFD is computationally demanding even if you have massive distributed computing to run it on. Let's say you have 10 different front wing configurations, 10 rear, 10 diffuser, et cetera. Depending on how you set up your DOE you're talking about dozens or hundreds of combination.

If you have 100 different total design options you can evaluate in CFD.. do you want to run each one to an extreme level of total detail which will take a week to solve, and you won't be done in 2 years time? Or do you go with a simpler, less exact solution that takes a day to solve so you can actually get some meaningful trends to impact your design?

In my experience, engineering (particularly with race) is often not about finding the 100% exact solution.. it is about finding the solution that's accurate enough, quickly enough.

Once you run through all your CFD to get a 'pretty good idea' of what to expect, you've got to validate it somehow at each major iterative stage in design. It's not a one time process.

[Fil]

xxChrisxx thanks for a fantastic explanation & peak at the tip of the CFD iceberg. Especially for those mere mortals who didn't know how far CFD has come, yet how far it still has to go. CFD in layman's terms - didn't think it was possible!

Once you run through all your CFD to get a 'pretty good idea' of what to expect, you've got to validate it somehow at each major iterative stage in design. It's not a one time process.

Yet Manor are skipping the validating step?

Obviously their major point of validation is the first rollout of the car.. but that's scarily late compared to the others competing in the race.

They bravely (naively?) have huge faith in their witchcraft.

[Ciro Pabón]

It's not witchcraft, it's engineering. Allow me to tell another of my lame jokes about physicists and engineers:

A mathematician, a physicist and an engineer were asked about the volume of a red rubber ball.

The mathematician measured the diameter, took an integral of the volume with respect to radius and calculated the volume.

The physicist submerged the rubber ball in water, measured the displaced volume of water and calculated the volume of the ball.

The engineer looked the serial number written on the bottom of the ball and looked for the volume in his "Red Rubber Ball" catalog.

[xpensive]

An MD, an Artist and The Engineer were asked about their female preference; A wife or a mistress?

- Definately a wife, says the MD.

- Absolutely the mistress, goes the Artist.

- Both, reasons The Engineer.
When I'm not with my wife, she thinks I'm with the mistress and when I'm not with her, she thinks I'm with the wife.
In the mean time, I can be at the office getting some work done.

[Jersey Tom]

xxChrisxx thanks for a fantastic explanation & peak at the tip of the CFD iceberg. Especially for those mere mortals who didn't know how far CFD has come, yet how far it still has to go. CFD in layman's terms - didn't think it was possible!

Once you run through all your CFD to get a 'pretty good idea' of what to expect, you've got to validate it somehow at each major iterative stage in design. It's not a one time process.

Yet Manor are skipping the validating step?

Yup, exactly. IMO, not a wise move. It may just be out of necessity. Wind tunnel time costs real money, CFD costs computational time. As a startup F1 team, having to develop everything, they may just be strapped for cash.