CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   Main CFD Forum (https://www.cfd-online.com/Forums/main/)
-   -   Cost of a simulation (https://www.cfd-online.com/Forums/main/101907-cost-simulation.html)

 Edu-R May 16, 2012 19:39

Cost of a simulation

Dear all!

Yesterday I was thinking about the CFD companies, and the projects and simulations that have to do to compensate the amount of money they paid for the software.

Could anyone tell me how much cost a simulation? I know that it is a very generic question and that it can vary a lot, so I'm going to put a simple example:

Imagine that you have a car (or an airplane) modelled in CAD, and you want to know the pressure distribution over the model, velocity contours, and lift and drag coefficients, but you donīt have a CFD software, so you must go to above companies. Assuming that the CAD is properly done (no surface repairing is required), that the resultant mesh is not required to have more than 8 million elements (because k-e or k-w will be used with standard wall functions), and that is a steady simulation, how much would you pay for this information? And if you need any advice to improve your model (efficiency or whatever)?

I have no idea about how much it can cost, and I donīt need an exact values or prices. It's only to have an idea

Thanks in advance! ;)

Regards,

-Edu-

 arjun May 16, 2012 20:03

Quote:
 Originally Posted by Edu-R (Post 361534) Dear all! Yesterday I was thinking about the CFD companies, and the projects and simulations that have to do to compensate the amount of money they paid for the software. Could anyone tell me how much cost a simulation? I know that it is a very generic question and that it can vary a lot, so I'm going to put a simple example: Imagine that you have a car (or an airplane) modelled in CAD, and you want to know the pressure distribution over the model, velocity contours, and lift and drag coefficients, but you donīt have a CFD software, so you must go to above companies. Assuming that the CAD is properly done (no surface repairing is required), that the resultant mesh is not required to have more than 8 million elements (because k-e or k-w will be used with standard wall functions), and that is a steady simulation, how much would you pay for this information? And if you need any advice to improve your model (efficiency or whatever)? I have no idea about how much it can cost, and I donīt need an exact values or prices. It's only to have an idea Thanks in advance! ;) Regards, -Edu-
For 8 million cells calculation and k-w model , I will only pay 150000 yen (in japan) and buy myself a 64 bit desktop. I will use iNavier and get the results.

I in fact I have done same thing in past using k-w and SA model for someone using my desktop.

 SergeAS May 17, 2012 00:52

Quote:
 Originally Posted by arjun (Post 361536) For 8 million cells calculation and k-w model , I will only pay 150000 yen (in japan) and buy myself a 64 bit desktop. I will use iNavier and get the results. I in fact I have done same thing in past using k-w and SA model for someone using my desktop.
Good point. Looks like cost of simulation it is cost of your hardware :-)
I also confirm this rule for our contracts for simulation tasks.

 Far May 17, 2012 03:11

What about the software cost and man-hours ???? :confused:(70-80% of total cost)

 arjun May 17, 2012 03:50

Quote:
 Originally Posted by Far (Post 361580) What about the software cost and man-hours ???? :confused:(70-80% of total cost)

It depends on the user experience. But if grid is not very very bad, then solver usually at least converges. For the external aerodynamics , the new version of iNavier is yet to diverge for the tests i have done.

Accuracy is another thing though and solution can depend on grid and non orthogonality. I am working on this aspect and would try to create better solver.

 SergeAS May 17, 2012 03:59

Quote:
 Originally Posted by Far (Post 361580) What about the software cost and man-hours ???? :confused:(70-80% of total cost)
Do you add cost for in-house software in to the contract ? :)

 Far May 17, 2012 04:02

Ya, should be in some form. This is how business is done ;)

But the question was related to companies (you are outsourcing) and they shall solve problem with /without commercial solvers they have .

 SergeAS May 17, 2012 04:19

There is a problem, unless the contract provided that you are the developed of software for customer, then the customer can claim the for this code. If the customer receives only the simulation results that this question does not arise, but there is a "cost of modifying in-house software" for the problem of the customer, but it is not 70-80% of total cost

 arjun May 17, 2012 04:42

I thought by software code and man hours he means the cost of learning the solver and then cost of meshing etc.

If the solver is easy to learn cost of learning is very low.
Then meshing is the one that takes time. If mesh is too bad user will keep on getting divergence and it is time consuming.

If grid is good or solver is good enough to handle bad mesh than solution at least converges. (this is the start)

Then you are looking at the accuracy, it means the way to test if results are even good or usable.

An experienced user who understand CFD would be able to create good meshes that solver converges easily and provide accurate results. (accurate enough to be usable).

The new versions of iNavier are very stable, I have been testing them on very very bad meshes. So this part of time would be saved.

Interesting that while people spend time trying to converge, I spend time trying to create meshes that should make solver diverge. :-D

 Far May 17, 2012 04:57

Quote:
 Interesting that while people spend time trying to converge, I spend time trying to create meshes that should make solver diverge. :-D
In starting of my career, I was expert at creating the poor meshes, which never converge:D.

Even solver (Fluent) by looking at the mesh (mesh check), I can feel it is saying Aaaargh :p

 Far May 17, 2012 05:01

Quote:
 I thought by software code and man hours he means the cost of learning the solver and then cost of meshing etc.
By software cost I mean you pay for it on yearly basis+ initial cost. And by man hours, I mean cleaning geometry, learning about problem, grid generation, solver setting and solution, validation, postprocesing and report writing.

 Edu-R May 17, 2012 07:02

Hi all!

First of all thank you for your answers!

I know that for the example I've put the best solution (economically) for the company will be to have a free CFD solver and learn how to use it.

Reading your answers I forgot to mention some things that could be important, so I'll modify it:

Imagine you need an accurate simulation of the model, close to y+=1, and it means that you will need a mesh of about 100 million elements. You need to know the parameters explained in the first post, but you donīt know anything about CFD, so you won't buy any software (even if it is free, because you don't know how to use). The other possibility is to go to an external company, which makes a lot of projects with CFD, and which has a computer capable to do this simulation. The problem is that they use a commercial solver (such as Fluent or Star CCM+), and they have to pay an anual license for it. It must be taken into account that the people of this external company knows perfectly how to use this software, because they have a good experience with it, and because of that, the CFD engineer defines the mesh in 2 hours, and the computer builds this mesh in 8 hours and solves the problem in 24 hours (for example). Is also assumed that the computer is already amortized, and the energy consumption cost is added directly to the final price, so we can neglect it for this example.

How much would you pay?
How much could be the /h of the engineer (defining mesh, post-processing, etc.) and the /h of the computation (meshing and solving)?

If there is something I've forgotten don't hesitate to comment ;)

Regards,

-Edu-

 SergeAS May 17, 2012 07:39

I'll try to present the situation from customer's point view.

Cost of simulation is closely related to the cost of full-scale experiment because the simulation can reduce the cost (to cut the program) full-scale tests. It is natural for the customer the expected economic effect of reducing the full-scale testing program exceed the costs of the numerical simulation. Ratio of simulation cost / full-scale experiment cost is different for different areas (aerospace, automotive, etc)

 sail May 17, 2012 08:05

I'll just give my 0.02...

First of all bear in mind that my situation is not the norm for cfd consultancy: we are a very small group of cfd consultant, we don't own very larg clusters but prefer to rent the computer time, and use mostly open-source software. this allows us to keep a very low overhead and to pay for resources only on a as-you-go base.

That being said, our major cost is man-hour so the time to clean the geometry, mesh, set up the simulation, postprocess the results, write the report, wash and rinse. If we are dealing with a problem we've not encountered before there is to add a fraction of the time to prepare some testcases and validation aganist known benchmarks.

A company that uses commercial software have to pay the licences and the computer, but if it is big enough the cost will be distribuited across many customers, and I won't expect it to be a big part of the total sum.

Overall then, what you are paying for is the time, brain and experience of the cfd analyst. It is not too hard to get the results from a cfd simulations. It is to get the right results, the right postprocessing and the right feedback to improve your design.

I hope my post haven't turned out too much as a commercial. :D

 Edu-R May 17, 2012 08:05

Quote:
 Originally Posted by SergeAS (Post 361646) I'll try to present the situation from customer's point view. Cost of simulation is closely related to the cost of full-scale experiment because the simulation can reduce the cost (to cut the program) full-scale tests. It is natural for the customer the expected economic effect of reducing the full-scale testing program exceed the costs of the numerical simulation. Ratio of simulation cost / full-scale experiment cost is different for different areas (aerospace, automotive, etc)

Dear SergeAS,

Thanks for your answer!

I know that the cost depends on multiple factors, and that's why I've put an example as complete as I can.

The aim of the question was to know an order of magnitude of the cost. I mean, to know if the simulation could be in the level of 1k, 10k or 100k € (or maybe 100€ :D). I donīt need to know (and we can't) that it would cost exactly 2565€ (for example), because we can't calculate it due to the lack of few parameters: we (or I) don't know the cost of the software (which vary a lot), the power consumption cost, etc.

If you tell me "Ok, if you are going to ask to a company to do this simulation they could charge you around 50k€, because it is a normal price" (for example), I will be happy, even if the simulation cost 25k or 60k, because I will have an order of magnitude of the cost.

However, all the answers and information are welcome, so thank you again for your answer! ;)

Regards

 Edu-R May 17, 2012 08:20

Quote:
 Originally Posted by sail (Post 361651) I'll just give my 0.02... First of all bear in mind that my situation is not the norm for cfd consultancy: we are a very small group of cfd consultant, we don't own very larg clusters but prefer to rent the computer time, and use mostly open-source software. this allows us to keep a very low overhead and to pay for resources only on a as-you-go base. That being said, our major cost is man-hour so the time to clean the geometry, mesh, set up the simulation, postprocess the results, write the report, wash and rinse. If we are dealing with a problem we've not encountered before there is to add a fraction of the time to prepare some testcases and validation aganist known benchmarks. A company that uses commercial software have to pay the licences and the computer, but if it is big enough the cost will be distribuited across many customers, and I won't expect it to be a big part of the total sum. Overall then, what you are paying for is the time, brain and experience of the cfd analyst. It is not too hard to get the results from a cfd simulations. It is to get the right results, the right postprocessing and the right feedback to improve your design. I hope my post haven't turned out too much as a commercial. :D
Dear sail,

Thanks for your answer!

I agree with you at all, especially with the text marked ;)

ŋCould you tell me (if you can or want) how much costs for a customer an hour of work of a cfd analyst (not how much he/she earns per hour, which normally is different)?

Regards

 Martin Hegedus May 17, 2012 23:43

Are you looking for an aero/fluid dynamicist or a CFDer? Are you looking for design support or an analysis? In general an aero/fluid dynamicist is more expensive than someone who runs a CFD code. The aerodynamicist will oversee what the CFDer does, if they do not do the work themselves. The cost of an aero person is about 2x to 3x their salary. So take what you think the yearly salary of a professional of the caliber you are looking for should be and multiply it by 2.5.

One of the first things you and your aerodynamicist should do, before you even touch a keyboard, is to determine what physics are involved with your problem. The more non linear and/or sensitive your problem is, the higher the cost and uncertainty will be. In general, outlining the physics should not take much time and should be inexpensive. A good aero/fluid dynamicist should basically be able to quickly sketch out the problem for you. That sketch(es) will then be the foundation of the work plan.

If your problem has a lot of uncertainty then a big part of the cost will be how much you want to drive down the uncertainty or determine what the uncertainties (error bars) are. Depending on what your project is, some control systems need accurate models of the plant.

Another item to be aware of, and I'll be straight forward here, my experience has been, in the case of design projects rather than just analysis, that when a client, without some fluid dynamics background or support, comes to me with a design already established, there is a chance that things do not go according to the clients wishes. Frequently, they find the performance of the design not meeting their expectations. And sometimes, for various reasons, they are hesitant to change the design in a meaningful way. Trying to fit a square peg in a round hole, so to speak, becomes VERY expensive. And, frankly, it is a cash cow to an aerodynamicist.

 Martin Hegedus May 18, 2012 00:20

@Edu-R

Just for fun, you mention "Imagine that you have a car (or an airplane) modelled in CAD, and you want to know the pressure distribution over the model, velocity contours, and lift and drag coefficients"

So, as an aerodynamicist, I'm going to ask you, for what do you need the pressure distribution, velocity contours and CL+CD for? And why use CFD? For example, in regards to CL+CD for a car, what will the added accuracy by using CFD rather than historical knowledge give you?

 Far May 18, 2012 01:17

Quote:
 And, frankly, it is a cash cow to an aerodynamicist.
Great. This explains the whole story.

 Edu-R May 18, 2012 16:05

Dear Martin,

Thanks for your answer!

Quote:
 Originally Posted by Martin Hegedus (Post 361771) Are you looking for an aero/fluid dynamicist or a CFDer? Are you looking for design support or an analysis? In general an aero/fluid dynamicist is more expensive than someone who runs a CFD code. The aerodynamicist will oversee what the CFDer does, if they do not do the work themselves.
The first aim was to know the cost of a simulation done by a CFDer, but there is no problem to know also the cost of aerodynamicist :D

Quote:
 Originally Posted by Martin Hegedus (Post 361771) The cost of an aero person is about 2x to 3x their salary. So take what you think the yearly salary of a professional of the caliber you are looking for should be and multiply it by 2.5.
This is the datum I was looking for, even without exact values ;) Thanks!

Quote:
 Originally Posted by Martin Hegedus (Post 361771) One of the first things you and your aerodynamicist should do, before you even touch a keyboard, is to determine what physics are involved with your problem. The more non linear and/or sensitive your problem is, the higher the cost and uncertainty will be. In general, outlining the physics should not take much time and should be inexpensive. A good aero/fluid dynamicist should basically be able to quickly sketch out the problem for you. That sketch(es) will then be the foundation of the work plan. If your problem has a lot of uncertainty then a big part of the cost will be how much you want to drive down the uncertainty or determine what the uncertainties (error bars) are. Depending on what your project is, some control systems need accurate models of the plant. Another item to be aware of, and I'll be straight forward here, my experience has been, in the case of design projects rather than just analysis, that when a client, without some fluid dynamics background or support, comes to me with a design already established, there is a chance that things do not go according to the clients wishes. Frequently, they find the performance of the design not meeting their expectations. And sometimes, for various reasons, they are hesitant to change the design in a meaningful way. Trying to fit a square peg in a round hole, so to speak, becomes VERY expensive. And, frankly, it is a cash cow to an aerodynamicist.
Totally agree, especially with the last paragraph :D

Quote:
 Originally Posted by Martin Hegedus (Post 361772) @Edu-RJust for fun, you mention "Imagine that you have a car (or an airplane) modelled in CAD, and you want to know the pressure distribution over the model, velocity contours, and lift and drag coefficients" So, as an aerodynamicist, I'm going to ask you, for what do you need the pressure distribution, velocity contours and CL+CD for? And why use CFD? For example, in regards to CL+CD for a car, what will the added accuracy by using CFD rather than historical knowledge give you?
Another example:

A small company, that does a little number of cars (those companies that make the cars "by hand"), has one designed, and the owner doesn't want to modify this design because it is visually nice (for example). Maybe he wants to know the pressure over the bodywork to have an idea of the force applied over it, and thus, the deformation that it would suffer, or maybe he wants to know this datum to calculate the thickness of material they have to put in order to obtain a desired deformation or to achieve the less weight as possible.

It is strange, and maybe a bit irreal, but it can be possible :p

Thanks again!

Regards,

-Edu-

All times are GMT -4. The time now is 07:39.