AI isn’t yet impacting real-world CAE. In fact, it’s even causing issues for engineers – but there’s still huge potential for it to totally transform workflows that haven’t changed in decades, writes Laurence Marks
Given AI’s limited use in computer-aided engineering (CAE) workflows that I’ve seen to date, it’s safe to say that the technology has not yet hit its peak in engineering circles.
In fact, AI is in part to blame for some of the big problems that simulation faces today. Some hardware costs have dropped – you can now buy a cluster for the equivalent cost of my old Sun Workstation – but increasing demand for AI data centre capacity has impacted availability and cost of hardware like RAM (which currently costs around five times what it did a year ago). And if you want serious SSD storage, then you’ll have to search hard and be prepared to pay up when (and if) you find some available.
Hopefully, these issues will ease as the novelty and general business panic around AI wanes, enabling supply chains to realign and bringing this period of craziness to an end.
Little change so far
Part of the wider problem is that while simulation and analysis tools and software have advanced considerably, workflows really haven’t changed massively for some 30 years.
Back in the late 1980s, I was building small finite element analysis (FEA) models by drawing them on a board and typing the node and element definitions into a teletype terminal. (The last time I saw one of those was on a trip to a museum.)
While I detest overuse of the term ‘disruptive’, it’s high time something new came along to truly shake the tree and I can’t help thinking AI will be a lot of things that it isn’t just yet
While I detest overuse of the term ‘disruptive’, it’s high time something new came along to truly shake the tree and I can’t help thinking AI will be a lot of things that it isn’t just yet
Five years later, I was working in pretty much the same way that I work today. Over those three decades, toolsets have become a lot neater, faster and more affordable. Software costs have kept pace with inflation, and there are even credible, usable FEA packages available for free download.
What hasn’t changed in CAE is the ‘what’ and ‘how’ of getting work done. While I detest overuse of the term ‘disruptive’, it’s high time something new came along to truly shake the tree.
I can’t help thinking AI is eventually going to be a lot of things that it isn’t just yet. It’s not a fast solve for CAE, that’s for sure – even if software companies present it that way.
Behind every piece of marketing that claims that huge computational flow dynamics (CFD) models are being solved in a matter of seconds also comes with a footnote that reveals how long it took to train the model to perform that trick in the first place.
However, AI is a powerful tool to search the complex responses that simulation models generate if you vary their input parameters. Because of this power, calculating what happens at the intermediate points is practically instantaneous. However, as a way of solving the equation set for a single one-off model, it’s not a technique that works.
The traditional brick mesh is still superior in many ways when it comes to getting reliable results rapidly, but its main disadvantage is the extensive manual intervention required, unlike tetrahedral meshes that dominate modern software thanks to their willingness to be automated.
Thirty years ago, a big part of my job was manually meshing 3D geometries that mostly hadn’t even been modelled in 3D. In other words, I was dividing geometries into small brick-meshable blocks and doing it all by eye. The results worked, but headaches and eye strain were the uncomfortable consequences of that effort.
So surely this is an area where using AI might deliver a real step forward, although I’ve yet to see anyone tackle this issue. And why can’t we use AI to augment the manual checking of models? This is a slow, demanding and unreliable process. I’m not suggesting we go crazy and let computers sign off on our work, but anything that can help identify potentially catastrophic mistakes in models has to be a good thing.
Bigger picture
Overall, when we want to achieve something with simulation, it’s not individual models that we concentrate on but simulation workflows.
Elements like the process of model creation and how interpretation fits into the overall design and operation lifecycle – PLM if you want to call it that – is where I’m sure big benefits lie in wait.
In the marketplace, companies like Physics-X, Neural Concept and Luminary Cloud are all announcing amazing efforts in this space, but figuring out what the actual picture looks like today, let alone what it might look like in the future is mindbogglingly difficult.While I detest overuse of the term ‘disruptive’, it’s high time something new came along to truly shake the tree and I can’t help thinking AI will be a lot of things that it isn’t just yetWhile I detest overuse of the term ‘disruptive’, it’s high time something new came along to truly shake the tree and I can’t help thinking AI will be a lot of things that it isn’t just yet
Whatever the end game, the process of shaking the tree is one that can only be good for simulation circles – even if all it does is help us improve on jobs we already perform today.
There’ll be much more to come, I guarantee. But after 30 years of only slight shifts in the game, I’m excited for what low-hanging fruits AI could target in order to transform workflows
This article first appeared in DEVELOP3D Magazine
DEVELOP3D is a publication dedicated to product design + development, from concept to manufacture and the technologies behind it all.
To receive the physical publication or digital issue free, as well as exclusive news and offers, subscribe to DEVELOP3D Magazine here
