What next for CFD?

For an industry that revolves around solving some very old equations, the continued innovation in CFD might come as a surprise to some. I asked a dozen industry experts what they thought was the aspect CFD that is currently most in need of innovation. What is holding the industry back? But, perhaps more importantly, what they would like to see done about it?

I got some great responses, with plenty of themes that weren’t on my radar. In general, the responses reflect the professional interests of the respondents. But then again, they really should do — after all:

“If you’re not working on your best idea right now, you’re doing it wrong.”

David Heinemeier Hansson — Creator, Ruby on Rails

And these guys are all working to improve what they see as the biggest problems that we have in CFD.

I’ve loosely grouped the responses into themes. Some respondents just could’t keep it to one problem, so they show up in a few of the themes. But I guess this just shows that there’s still plenty of room for improvement in the world of CFD.

All about the mesh

As the responses started to roll in I thought this was going to turn into the meshing roundup. Over half of the respondents pointed the finger directly at some aspect of the meshing process as being most need of innovation.

The problem is so clear to some that Derrek Cooper (Autodesk) took just 10 words to sum make his assement of the innovation gap:

“CAD 2 Mesh is the biggest problem plaguing the CFD market.”

Derrek Cooper — Director Simulation Products, Autodesk

Claus Abt (Friendship Systems) was even more succint, proposing a solution to that problem in just 4 words:

“simulation ready variable geometry”

Claus Abt — Co-Founder of Friendship Systems AG

Expanding a little on Derrek’s sentiment, John Chawner, one of the originators of the Pointwise meshing software, identified the following:

“When it comes to CFD workflow the obvious place for improvement is preprocessing: moving from a geometry definition to a CFD-ready mesh. Most estimates rank preprocessing as consuming 60-80% of an engineer’s time when it comes to applied CFD.”

John Chawner — President & Co-Founder, Pointwise

John went on to identify almost every stage of preprocessing as still in need of innovation:

“Advances are needed in geometry interoperability, better NURBs handling, better faceted geometry handling, feature detection, defeaturing, cleanup/repair, simplification. Meshing algorithms need to be more robust, target specific analysis needs, faster, etc.”

Henrik Diamant, Head of CFD at current Formula One world champions, Mercedes AMG Petronas F1 Team, added his vote to the meshing contingent:

“As we know, to produce an accurate CFD solution you need a high quality mesh featuring a cell distribution tailored to the problem you are analysing. Whilst an experienced Engineer has a good grasp of what mesh to use in different regions of the computational domain to accurately capture the fluid flow, it would be better if the meshing software was ‘intelligent’ and able to make this choice automatically, having first established the physics of the system.”

Henrik Diamant — Head of CFD, Mercedes AMG Petronas F1 Team

This would have plenty of upside, as Henrik explained:

“The advantages of such an approach are significant and speak for themselves; ‘right first time’, less room for inconsistencies, less time spent fine tuning meshes and improved accuracy.”

Which leads nicely into our next theme — time.

A Waste of Time

This was also a popular theme in one sense or another. In particular, the time spent by experienced engineers either administrating CFD runs or waiting for them to complete. David M. Holman, creator of the XFLOW CFD package, explains how he sees this problem:

“We see so many CFD engineers spending most of their time preparing cases, cleaning up the input geometries or adapting their workflow to the peculiarities and limitations of their CFD code… We want CFD engineers to be able to focus on analysis, on getting new insights. Analysts should be freed from traditional software constraints and be allowed to explore more challenging applications, backed up by data they can trust from high-fidelity predicting tools.”

David M. Holman — VP at Dassault Systèmes, SIMULIA R&D

A sentiment echoed by David Heiny, co-founder of the simulation platform SimScale:

“The traditional workflows an engineer uses to get an accurate flow simulation of his design or application are still to some extent tedious. So from my perspective, the most needed innovations are the ones that reduce the time to (an accurate) result in CFD.”

David Heiny — Co-Founder of SimScale

David goes on to highlight one of the more overlooked time sinks in the CFD process:

“As soon as a company starts using CFD more heavily, the need for computing power increases. But deploying an HPC system and maintaining it is typically nothing a mechanical engineering company wants to deal with. Removing this burden from engineers would allow them to spend their time more doing actual engineering work.”

Karim Fahssis, innovator in the emerging wind-power CFD sector, comes at this from a slightly different angle. He suggests that it’s not just the time spent getting to a result, but the time spent waiting for results that needs to be improved.

“In order to get the most out of their expertise, CFD engineers should not waste time waiting for results. Instead, they should be able to get (almost) instant results so that they can spend more and more time analysing more and more results in order to improve one final result. This is now possible thanks to cloud computing with unlimited scalability and on-demand unlimited computational power.”

Karim Fahssis — Co-Founder of AZIUGO

Appification

I expected “the cloud” to be a theme in the responses. In hindsight I think we’ve moved on a little. The industry has realised that “the cloud” itself doesn’t mean anything that new. Much more interesting are the innovations that can occur when you can easily offload your number crunching. One of the themes that was mentioned by several respondents was that of industry (or vertical) specific apps. Jeff Waters (ESI) succinctly expresses this theme as:

“Hyper-customization for 1-off, industry/vertical apps…”

Jeff Waters — Sr Account Executive, Siemens PLM

A proposal seconded by Paul Bemis, who has taken this approach with his CoolSim software which is niched to data center thermal modeling.

“The tools need to be made application specific and highly automated. The learning curve for general purpose tools is too steep for occasional users in the design arena.”

Paul Bemis — President, Applied Math Modelling Inc.

Paul also highlights something many other industries noticed some time ago, that “The tools need to be mobile and able to run from a low cost laptop or tablet.”

Wolfgang Gentzsch adds his vote to this theme of appification, noting that:

“pre-canned ready-made apps software for solving problems in a specific area (e.g. combustion) will help a lot”

Wolfgang Gentzsch — Co-founder and President of UberCloud

While “the cloud” didn’t appear as a theme in and of itself. It is an enabler for new approaches to software delivery and customisation that were too gnarly a problem only five or so years ago. Wolfgang Gentzsch goes on to suggest that:

“What we need is seamless, on-demand, intuitive access and use of CAE Cloud resources — hardware plus CAE software fully bundled and accessible at your fingertips — with all current hurdles (security, licensing, data transfer,…) removed.

Using just a workstation is perfect for our daily work, but for our ever and more complex applications with finer meshes and more sophisticated physics workstations alone are limiting our innovation and the competitiveness of our products.”

Thankfully for us, he’s bullish on a timeline for this change in resource provision, noting that “I can see that this will happen within the next two years.”

How much is this going to cost me?

Not unlike all other consumers, CFD users (or potential users) are sensitive to cost. Joe Galliera of Solidworks Corp. notes that from his experience in the field:

“A very large barrier to wider acceptance (beyond technical limitations) is price of software and cost of ownership (annual maintenance and having someone qualified enough to know how to use the software tool).”

Joe Galliera — Senior Technical Sales Specialist for SOLIDWORKS

So if CFD is moving towards application-specific apps and on-demand software / hardware provision. Will CFD pricing still be fit for purpose?

Jeff Waters points out, with reference to hyper-customised apps, that we need "…radically different pay-as-you go cost and delivery models."

Paul Bemis also points to the need for change when he says that “tools also need to be lower cost and priced as a function of usage.”

But even in the world of downloadable software things aren’t all that rosy.

“It seems you can go online and find the price of almost anything these days. You want to know the price of a Mercedes, just visit their website and there it is. If you want to know the price of a Honda, same again — it’s easy. However, if you want to know the price of commercial CFD software then you are out of luck. I guess it's a case of ‘if you have to ask, you can't afford.’”

Dr Richard Smith — Principal & Founder of Symscape

Luckily Richard has an innovative & ground-breaking solution for this problem:

“Symscape is one of the few exceptions — if you happen to stop by our Caedium product pages you’ll see how much our CFD software costs — what an innovation, actually publishing CFD software prices!”

The emergence of new delivery models, such as software-as-a-service or infrastructure-as-software are challenging the age-old licensing models as David Heiny (SimScale) points out:

“Depending on the company and the software, licensing can be a challenge. High fixed costs, restriction regarding the hardware and usage as well as the varying need for simulation can make it complicated for a company to have the right amount of simulation capacity available when they need it. Innovations here could be more flexible licensing models that meet better the needs of each engineer or company.”

David Heiny — Co-Founder of SimScale

Almost themes

The following three quotes weren’t widely suggested so didn’t quite make it to a theme. But they do raise great points so I couldn’t just overlook them.

All GUIs are not created equal

“Given the adoption and success of modern graphical user interfaces (GUIs) across a broad range of software, you’d think that making the case for one would be redundant. However, for some reason CFD was a late adopter of GUIs, and for what is predominantly a visual set of disciplines — think geometry, mesh, solver feedback, 3D visualization, and 2D plots. This late adoption is still evident in CFD software today.”

Dr Richard Smith — Principal and Founder of Symscape

The “Know-How Barrier”

“Most of todays CFD tools and methods are very powerful and enable the user to run very sophisticated numerical analyses. However to generate accurate simulation results with them requires the engineer to have both physics as well as tool expertise, which in turn implies a lot of training and education. If this “know-how barrier” would be reduced, it would allow more engineers to make effective use of CFD and would simplify the start with new analysis types for experienced CFD engineers. Advances in this field could be new types of learning resources, more specialized and therefore easier-to-use simulation tools or a more collaborative simulation approach.”

David Heiny — Co-Founder of SimScale

Perhaps the online education initiative, www.LearnCAx.com and other similar offerings might have a role to play in this area?

Open-Source

“Lack of transparency is another major issue when running CFD simulations with proprietary software. Indeed, CFD results too often depend on choices made by the software developer and the user should be able to understand the code in order to understand the risks he is taking while using any particular software. For that, using open-source CFD is very important and transparent simulations with reproducible results are now possible thanks to OpenFOAM.

By combining Open Source with cloud computing, I believe these issues can be solved and this could potentially revolutionise the way engineering simulations are made nowadays by transforming technology, business and culture with industrial-level performance improvements.”

Karim Fahssis — Co-Founder of AZIUGO

What themes have we missed? If you’ve come into contact with CFD then I’m sure you’ll have an opinion on what needs changing. Tweet me @cfdengine and let me know.