Hi all, I am a bit curious about the Lattice-Boltzman code Powerflow. Is there anyone out there who is actually using it in vehicle production work? You always seems to hear that "Oh yes, company XXX is using Powerflow" but I would like to hear from someone directly. And especially from someone that also has a thorough background in aero cfd, using traditional codes as Star-CD, Fluent or Cfx, and hear what he thinks of Powerflow. How do you think it performs in terms of leadtime from "dirty" cad and accuracy etc?

Regards Anders

Anders,

I was a Fluent/star user before and had similar curiosity about powerflow and had actually use it for a while before I decided that enough is enough and said goodbye to EXA/Powerflow (& I feel a lot better now that I don't see a single sale man from EXA and don't have to point out the lies/false claims they were making.

If you're prepared to take a lot of pain to find out yourself, that's upto you, but be warned. You maybe very disappointed & the polished claim is one thing & the product is another. Oh another thing you might need to be aware, when I complain about the version of the powerflow I used, EXA always said the problems will be solved in the NEXT VERSION. To give you an example, In powerflow 3.1, It predicts falsed separation which doesn't happen in a real car, then they said it'll be sorted in Powerflow 3.2, so from Powerlfow 3.2p1, p2, p3, .... to p5, the falsed separation disappeared, but the problem is that now the code become steady state where there should be separtion, it doesn't predict. So EXA said it'll be sorted in Powerflow 3.3. Guess what happened, in powerflow 3.3, it over predicted separation again (similar to powerflow 3.1). Then EXA said it'll be sorted in powerflow 3.4, then powerflow 3.4 is very similar to powerflow 3.3 and so on...... until I had enough and throw it away completely. It may well be sorted in powerflow version 10, but by then I probably be retired - LIFE IS TOO SHORT TO BE MESSED ABOUT ......so please judge for yourself, but be careful & don't take what EXA tell you as the truth......

Heh, this story is the same for most commercial code vendors I think.... the next version will always solve your problems and in reality it rarely makes any difference at all.

Hi Jonas and ex-powerflow user,

Jonas, it is true that most vendors fix problems in the next release. Still it was a very interesting answer to hear from someone that had actually used powerflow and had previous Star/Fluent experience. I do work in the automotive industry and as I indicated in my first mail I have got so many questions from non-experience CFD persons why we are not using the super duper code Powerflow. The latest instant was a symposium that a hardware vendor had and they also asked me the question about Powerflow. As I understood, they were selling quite large systems dedicated for Powerflow simulations. The questions also came from a automotive company why we did not use powerflow. So, since I do not know anyone that uses it, I wanted to know more how experienced CFD persons that has used powerflow, experienced it. To me, it seems to exist an huge confidence out there for Powerflow, and especially from people with no or limited experience from CFD.

Personally, there are no benchmarks known to me or other material that has convinced me to try Powerflow. But on the other hand, one should always try to have an open mind.

Regards Anders

Dear Anders,

actually I do not really have experience in using Powerflow. About 2 years ago I had an evaluation licence from Exa, but I dropped it after a few weeks for several reasons: - There was no support, at least no useful one.

- Tremendous memory requirements.

- For some cases where we expected to get additional information using a LES (the basically LES code) Powerflow performs worse than any RANS code we have checked.

As our team is working for one of the largest supplier to the automotive industry, the lack of interest from Exa to sell Powerflow was suprising.

A discussion with a friend, who is actually using Powerflow (German Automotive Industry) results in the following: "Powerflow is very memory intensive (at least 12 GB for some reliable results for full car aerodynamics). We are statisfied with the results, but for the development we are using ...."

Best regards, F & S

Does anybody know what brand of CFD code used Boeing, Aerospatial, Snecma, Pratt, Rolls-Royce....

The fact is that most companies involve money and YEARS of benchmarking to get confident in several codes, then make theyr choice upon different criteria, predictivity is one of them. So they won't tell for free what they have choosen and why.

Hi, Anders,

Before I used Powerflow, during one of the international conference, I discussed Powerflow with one of the CFD vendors who told me this 'Stone Soup' Story (some of you may have heard it before, cann't remember the exact words, something like below):-

Two people came to a small town and told the local people who suffer from starvation that they can make soup from stones. People don't believe them, so they started to put stones in a pan and put some water in it and started cooking and tasting the coup while they're cooking. They give the expression that the soup was so delicious and people started to wonder, then the two said to each other, if we could have just a bit of tomato, the soup will be perfect, someone out of curiosity brought a tomota for them to put in the soup, then they said if we could have just a bit of ...... it'll be...... evenly, the soup was made - all using the material given by local people......

The vendor said that people selling Powerflow is like cooking soup using stone - ask you for more and more and eventually make something out of stuff given by the users themselves.....

At that time I didn't believe this CFD vendor, but later experience proved that this vendor was right, some examples are as follow:-

When EXA sell powerflow, they claim that they have aeroacoutics capability, when we started using powerflow for aeroacoutics, it turn out to be rubbish (predicting many false peaks in the SPL/frequency figures), then EXA said if you give us 200,000 dollars, we'll form a consortium (hire some Univ experts & new engineers) to develop the aeroacoutics capability for you...... with serval car makers giving 200,000 dollar each, they hired some people to work on aeroacoutics now. EXA also claiming thermal capability, then if one is serious about thermal prediction, EXA said they will form a thermal consortium with member contribution 200,000 each to develop thermal capability in Powerflow..... Doesn;t this sound like the 'stone soup' saga ??????!!!!!!

IF THAT CFD VENDOR IS READING THIS NOTE, I'D LIKE TO THANK HIM FOR HIS WARNING......

It looks like Exa's marketing stratgey is working (in the short term):

"1. Sell to people who has non or limited experience with CFD"

2. Take advantage of people's curiosity"

For the longer term, we shall see how successful this marketing strategy (trick) will be:-

1. some people with non or limited experience will become more experienced and find out the liars themselves

2. Once the curiosity is gone, there is nothing to keep the potential users interested. For example, they made a big claim on Lattice methods, when the curiosity is gone, one may ask: (1)if Powerflow's lattice is so good, why exa still use turbulence models ?

(2) Since powerflow uses turbulence models, all the short-coming associated with turbulence models will surely be associated with Powerflow as well......

So please judge for yourselves - look at some of the internet companies who promised a heaven 3 years ago end up today......

Hi Anders,

I have not used Powerflow but I have been involved in two benchmarks including Powerflow as you probably know. I would like to give EXA credit for the ability to mimic known experiments very well. Unfortunately, they seem to perform much worse when the solution is not known beforehand. I have recently heard from a mutual friend that recently used Powerflow, that some things never change....

Regards

Sveko

[littlebroca]

Hello everyone,
I had to use PowerFLOW for three years, wanting to study aeroacoustics and aeraulics on fans and train motors.

I had so many computing problems due to its requirements in RAM (i had to do internal flows in a very complex industrial configuration), and in capacities (the files' size are up to 200Go). It needs a cluster that's fully optimized for this software. Ewa people always say thay manage to launch any type of calculation, but they don't specify that their cluster is exactly configured FOR powerFLOW

Then, i found that their support is really unefficient. I hated that every part of this code is closed, we can't change anything. The output files can only be read by their post processing tools. If you want a special tool, they take forever to give it to you. I needed a Ffowcs-Williams Hawkings routine, that took into account the rotation effects, and it took forever to get it.

Then, they promised a sliding mesh for fan noise studies. They said it'll be delivered by december 2007. By the end of 2010, nothing ever came out.

Aside of that, it's still a very powerful tool, quite easy to use (especially for the meshing part), and the Lattice Boltzmann method is very good to avoid convergence issues.

The major issue with it, is that to ensure a convergence, they have a very very very dissipative turbulence model, that doesn't allow to do very good aeroacoustics. And you can't change this model because everything is closed.
Then, the boundary layers are not very well modelized. Associate that and the dissipative turbulent model, and you get 40% error for frequencies for very simple cases!

To sum up, it's a very expensive tool (may be the most expensive one). The licenses are expensive (a little bit more than fluent), and you will need an installation that will allow you to run any type of calculation you want. Therefore, it's again an expense that is not negligeable at all.

I don't really recommend it. Only bad memories and headaches over this thing.