Artificial Intelligence in Structural Engineering 5

[tolchijb]

We human engineers will always have something that AI won't and can't...Professional Liability Insurance. Until AI can be sued and held liable, we will have a job. That said, AI will be an invaluable tool once the marketplace matures. Right now AI is just a buzzword to most without proven, tangible results to back it up. Our industry tends to be slow to adopt, often rightfully so. Confidence should be earned not given.

 

[Tomfh]

We human engineers will always have something that AI won't and can't...Professional Liability Insurance. Until AI can be sued and held liable, we will have a job.

Phew! There’s still hope—we can scrape by charging a middleman fee to transfer liability from the AI designers to the liability insurers.

 

[Flotsam7018]

Before this thread dies.

I’m sure most of us are anticipating AI integrating into our workflow at some point, but I was surprised to come across this software available now which is what I was expecting to see in the coming years. https://www.genia.design/

The price point likely isn’t there for many, but I have no doubt this will be the future - whether we like it or not. Our job very well will become reviewing AI designs in the very near future.

 

[Greenalleycat]

That was a depressing way to start the day. I can't wait for someone to print off some plans for $500, tell me they want my signature, then get pissed off when I try to charge them $3000 to redesign and draw the first round of crap it spits out.

Then in 6 months they fix all the errors and I'm unemployed

Woo

 

[Ron247]

The Genia was interesting to peruse, but they won't see $4k a month nor $400 a month from me without some really serious test cases or trial runs for free.

I have used ChatGPT a lot to write VBA code to run inside Excel even though I know VBA, Fortan and some Python myself. The results have been really good, but coding is more structured as far as rules, methods and structure. Engineering design is not as structured.

I recently had ChatGPT create a picture of a typical metal building under construction. Here is the result. Note the left wall panels on the inside, x-brace on the bottom girt only on the right wall, finished concrete and plants outside but the building is not complete yet, the front rigid frame to "nowhere" and rafters in the air. And for the architects, ChatGPT did get the x-bracing correct when it put the bracing in front of a window, we engineers did teach it that.

 

[SWComposites]

and the car parked over the curb ...
the rafters remind me of the Pink Floyd album cover with flying pigs

 

[Flotsam7018]

This isn’t ChatGPT. It was created by an ex Amazon and Arup engineer, who created their own AI model for the software. The solutions claim to be compliant with the IBC.

 

[GregLocock]

Josh:" I wonder if drawing creation can be more automated than it is now. "

There are more than one startups with a text to CAD proposal. It isn't very useful as of now, but in 5 years time, and access to all the world's online CAD, who knows?

Zoo: AI CAD Model Generator | Create CAD Files With Text

Text-to-CAD is an open-source prompt interface for generating CAD files through text prompts. Generate models that you can import into the CAD program of your choice.

zoo.dev

The Best Text to Cad Drawing AI Generator (for Free)

Generate precise technical drawings with our AI Cad Drawing Generator! Perfect for engineering and design.

openart.ai

 

[phamENG]

I, for one, am excited about this. This is precisely the tool I've been wanting to see. I'm not terribly interested in the analysis piece, though if they can get that right I won't complain. But anything that can streamline my workflow and improve my efficiency - particularly in the modeling/drafting department - is at least worth a look.

Will this lead to a reduction in demand for engineers and drafters? Yes - it will. If this thing can actually do what it says, I could ostensibly replace up to 10 employees with that $4k per month. Just like the advent of CAD reduced the demand for drafters - a drawing that had to be painstakingly hand drawn over the course of weeks can be done in days by a proficient operator. But the engineer in charge will remain, and there will still be a demand for personnel to use these programs.

The danger is in the misuse of these programs by firms. People who take the output, stamp it, and move on without reviewing. Shops that train their junior engineers to run the program rather than design the building. But these problems already exist with analysis software that we've been using for decades. Lots of junior engineers only know how to draw beams in RISA, and if it doesn't show up red they think it's fine. Who cares that they failed to properly define bracing or that the axis of the beam is out? The program says it's fine. So the problems are nothing new, they'll just be exacerbated.

The solution isn't to stop the new tools, it's to figure out how to use any tool correctly.

 

[Flotsam7018]

@phamENG I appreciate your outlook. I am excited about certain aspects of this - if presenting multiple structural solutions for a given layout in a matter of hours, each highlighting different project impacts such as material cost, schedule, etc. become a possibility, this very well could change the entire way we operate and be a net positive for the built environment.

My concern isn't so much with misuse, but with the potential demise of the profession - at least as we know it. If you're already licensed and have had the traditional experience, yes - this will likely be a great tool and reduce some of the more mundane tasks. But the path to getting to that position and understanding what aspects of the software input/output you need to home in on for any particular project will become more difficult than it already is by multiple orders of magnitude.

 

[ThomasH]

I occasionally use ChatGPT, mostly for fun but sometimes work related. I basically consider it an advanced web browser.

I think it can be useful for translations. It can help me regarding formulations it I want things more or less formal. It is not perfect but significantly better than Google Translate.

For technical stuff I am more skeptical. I have tried asking questions where I know the answer and the result is sometimes good but to often bad. Therefore, I often use it on a "Trust but verify" approach. I try to verify the answer, and if I can't, I don't trust it.

I have a thread in the forum right now regarding "Torsional vibrations". I started with ChatGPT and it took a while to get the correct unit for the vibration, Hz. But I have so far been unable to make ChatGPT give me a result that is close to the FEM-analysis. And that means that I don't trust the result.

 

[canwesteng]

"AI", or more specifically generative algorithms with transformers, will surely have a place in engineering. Think of turning laser scans into models, reading existing drawings and turning them into models or first pass prints, possibly turning sketches into draft cad files, turning physical models into analytical models and back and forth.

A lot of the stuff genia is promising is not something AI is good at and not stuff that is hard or eats up a ton of time to begin with, and lots of this is already pretty automated (turning 3d layouts into 2d drawings for example). I think there's going to be a lot of AI bubble stuff like this that comes before the wheat is separated from the chaff

 

[phamENG]

My concern isn't so much with misuse, but with the potential demise of the profession - at least as we know it. If you're already licensed and have had the traditional experience, yes - this will likely be a great tool and reduce some of the more mundane tasks. But the path to getting to that position and understanding what aspects of the software input/output you need to home in on for any particular project will become more difficult than it already is by multiple orders of magnitude.

I think that is misuse. Just like telling a junior engineer to just "go figure it out" and then following up with "what did the model say?" with no meaningful interaction in between is misusing that technology. It's deluding yourself into thinking that learning to use a program is a replacement for dedicated training. The generational knowledge loss we're experiencing now is not unique to engineering, but the onus is on us to make sure that it is unique to our generation of engineers. We have to make sure we're doing better and finding ways to teach the fundamentals that allow you to not only understand the output of these tools but, more importantly, to call them on a garbage output.

For instance, another member of this forum and I were chatting earlier and he, looking closer than I was at the youtube video above, noticed some negative utilization values for shear walls. That's a pretty big red flag. I'm interested to see just what it's doing and how. But without training, somebody might just slap that on some sheets and ship it. Sadly, some people will do that.

 

[ViaParadise]

For me it's about a 90/10 mix of glorified search / bouncing ideas off of it. "It" here being any "AI" or at least tool that can come up with some means of doing what I ask of it. Say I want to write a module in VBA or create a new notebook to organize ideas in Notion or ask what would be needed in grasshopper to accomplish something then it's pretty good.

The challenge for me is that beyond compiling tailor-made documentation to programming problems I've struggled to make AI all that useful for me. In fact, everything I ask of it now were things that I simply searched forums and documentation for just a few years ago.

Which leads me to my concern that a lot of people are and will take for granted the ability of AI to present itself as a perfect encyclopedia that can adapt to its users asks. Improvements in technology like search powered by AI is great except when over reliance on a searching mechanism causes people to no longer try and understand what they're actually doing, software/engineering/cooking/etc., and instead take absolutely what AI tries to tell them it believes (was fed and tuned to) is the case when reality is much different or can be made better by applying understanding and creativity.

The other 10% is asking AI something along the lines of "I have a situation where torsion is transferred such and such and I considered such and such and its interaction as such and such, but I think that this can be avoided entirely if I do such and such, what do you think?". By asking the AI to tune its response it's as though I can talk to anyone from the prescriptive-only engineer to the minimize all work fabricator for their thoughts. Perhaps the answers ultimately don't really differ regardless of how I want to believe the AI's 'personality' matches my own beliefs of how it should be, but it does make for some fun discussions of limit states and how to avoid/solve for them without immediately diving into unnecessary FEA.

Personally, I would love to have an AI function along the lines of "I want to use rhino, grasshopper, AutoCAD, and SOFiSTiK to design a Vierendeel truss structure. I want to be able to modify inputs like section sizing and spacing using a combination of well-labeled excel and grasshopper inputs to produce well-labeled layers in rhino as well as TEDDY code. The analysis should be automated as well as reviewable. Analysis using groups to easily visualize WINGRAF plots that take advantage of color schemes, shapes and font sizing." If it could produce some files to do that while also interacting with those files as I do, then it would turn the process of ideation and manipulating shapes and sections for an easy to produce and understandable structure into an exercise of mentally and mathematically balancing inputs to get an output very quickly tailored to whomever on the project would benefit from that output.

 

[phamENG]

Personally, I would love to have an AI function along the lines of...

Agreed. One big opportunity I see with LLM based "AI" systems is the AI "Agent" that is able to take natural voice commands and interpret them into commands that the machine will understand. You see that in the programming world with platforms like Cursor and Lovable. Creating something similar that can interact with commonly used with structural and mechanical engineering design 'stacks' would be really neat.

In the meantime, I'm going to see what this thing is all about. I suspect it's a lot of hype and limited substance, like many tech startups are...until they aren't.

 

[EngDM]

Agreed. One big opportunity I see with LLM based "AI" systems is the AI "Agent" that is able to take natural voice commands and interpret them into commands that the machine will understand. You see that in the programming world with platforms like Cursor and Lovable. Creating something similar that can interact with commonly used with structural and mechanical engineering design 'stacks' would be really neat.

In the meantime, I'm going to see what this thing is all about. I suspect it's a lot of hype and limited substance, like many tech startups are...until they aren't.

One thing I am hopeful for, is the ability for AI to dumb down some of the legalese texts that I reference quite often. Sometimes I just information it dumbed down to really get an initial grasp.

 

[Flotsam7018]

For instance, another member of this forum and I were chatting earlier and he, looking closer than I was at the youtube video above, noticed some negative utilization values for shear walls. That's a pretty big red flag.

Ha! Hadn't caught that either. I have no doubt that it's extremely buggy in the current state, but also that in the next 5 years we will see software that is capable of reliably producing the results it claims - perhaps not by genia.

 

[Enhineyero]

AI is good for looking up information and probably engineering computations.

Hope it becomes so advanced someday that it can go on-site and have a cordial discussion with bob the builder and frank the architect on how to 'make the engineering work' on whats been built on site. I'd be more than glad to pay for a 'premium' subscription for this service.

 

[LOTE]

I am on the wait list for Work Beaver, which is an AI Agent of sorts and supposed to do tasks on your desktop using any application you tell it to. This is where I see the big time saver and benefit in AI tools. If we can eliminate all the non engineering stuff (creating folders, uploading files, entering all the data into project management/CRM tools, copying project info and design parameters from one program to another, etc.) from a typical engineer's workday and we can spend more time engineering and less time doing admin work that isn't easily outsourced, that is true value. Using existing tools but automating the data entry that you usually give to an intern.

As far as tools that claim to design it all for you, we are a long way from that I think.

 

[azcats]

I've successfully used it to quickly generate complex, multi-layered if formulae in excel. Just kept feeding it a word salad of what I wanted and we eventually got it tailored to work perfectly. Done in 5 minutes what would have taken me an hour to organize the logic in excel myself.