Artificial Intelligence in Structural Engineering 3

[justhumm]

My random thought of the morning...Artificial Intelligence (AI).

If it can't be called so already, is looking like it's going to be a major player in the future of computing across numerous industries.

And I suspect that a large part of the technical aspect of that is going to be created by software companies.

Right now, I'm wondering what structural engineers, themselves, are using AI for in your day to day work...?

generating technical report drafts?
finite element analysis (FEA) input files?
generating code (LISP, VBA, C#, etc.) for repetitive tasks?
creating pictures of your pets playing poker?

Cheers!

Deep Dream Generator

 

[Lomarandil]

"I searched for God Structural AI and found only myself engineers"

 

[drawoh]

My random thought of the morning...Artificial Intelligence (AI).

If it can't be called so already, is looking like it's going to be a major player in the future of computing across numerous industries.

And I suspect that a large part of the technical aspect of that is going to be created by software companies.

Right now, I'm wondering what structural engineers, themselves, are using AI for in your day to day work...?

generating technical report drafts?
finite element analysis (FEA) input files?
generating code (LISP, VBA, C#, etc.) for repetitive tasks?
creating pictures of your pets playing poker?

Cheers!

Deep Dream Generator
View attachment 3196

Cool. How are you going to build that thing?

 

[JoshPlumSE]

I don't think AI is very good at thinking for itself. However, it is really good at looking up information from specified sources. Or, copying existing things. My thoughts about how it could be used in structural engineering:
a) In the software industry, we spend a good amount of time putting together documentation that answers all the normal questions we receive from our engineer users. We put together example problems. I have a feeling that we'll see some on-line AI Bots that are smart enough to serve as 1st tier technical support. Meaning, they can analyze a user's question and then point them to certain answers that already exist in the documentation, examples, verification, or wiki-pages. Questions like how to I create a non-linear spring support? What does this warning message mean?

I don't think AI will be good enough for 2nd tier support..... Where user's are asking more complex questions or questions that are not frequently asked. Anything like, "here is a calculation that I've done that (or an example from a text book I have) and the the program results don't match. Am I doing something wrong or is this a bug?"

b) I think there is the potential for engineering adjacent industries to use AI in the same way as software companies might use it in (a). Let's say Hilti, Weyerhaeuser, Simpson, APA, AISC, ACI, et cetera. These companies put together such good documents and presentations and such. The AI bots can absorb all this information and then quickly point customers in the mostly correct direction.

c) For day to day engineering? I wonder if drawing creation can be more automated than it is now. Like instead of using AutoCAD manually, you could have an AI enabled assistant that automatically updates stuff for you and looks for conflicts. I especially see this being useful for concrete detailing, steel shop drawings.

d) I could see a lot of AI helping with field visits. Like we scan this site with our phone. Then have AI use that data to dimension out everything so that field visits are less cumbersome than they are currently.

e) I think it could easily help with creating first draft of technical reports. Especially if it's the type of report where you have multiple examples that it can study. I think it could also do a good job editing reports for grammar and consistency. I can't tell you how often I see inconsistencies in engineering reports. Let's say you have bullet points. Is the first word capitalized in all the bullets. Is there a period after some bullets, but not others. When you're reference a certain greek letter, is that letter always in italics or sometimes not. Not horrible stuff. Simple stuff that is sometimes tough to catch without a really careful review.

 

[Eng16080]

If by AI, you mean general AI, like chatGPT, I don't use AI for anything. I doubt I would ever use it for anything technical, unless it's for something that could easily be checked. AI is notoriously good at creating stuff that seems correct, but when you look closer, you realize it's not. Maybe it's good for some creative content (writing, artwork, etc.), but to be honest, I can't think of anything scarier as a structural engineer.

 

[PMR06]

Just recently I saw a new grad engineer using an AI tool to generate some pretty complex VB scripts to run in the background of a new design tool he was developing.

(I've got almost 25yrs experience)
I joked that he was ChatGPT'ing us out of jobs.
He quipped that it took him an hour what it might take me 2-3 days to code and debug. Who's getting laid off first?
I shot back I'm the one who knows whether your tool is correct or just more garbage in/garbage out, faster. (all in good humor)

Until these "AI" tools are truly intelligent, I see them utilized as others have described. They do not appear to have true problem solving, adaptive, creative capabilities yet.

Heck, I bet some of the construction mess-ups I've seen would crash these server farms if you asked them for a fix!

 

[RenHen]

I use OCR sparingly & with the knowledge that, while it is one of the more accurate applications of "AI", it's still often incorrect. Other than that, I don't directly use any AI that I'm aware of. Having read a lot of research regarding how these tools (particularly LLMs/generative AI) work, I don't anticipate that changing on my end; just a tool that is deeply unsuited to the work that I do. Even if it worked & didn't just spit out nonsense that I need to spend as much time or more double checking & correcting, it would be on tasks that I enjoy doing or processes that improve my skills/knowledge.

 

[JStructsteel]

I tried to get AI to go into a crawl space the other day and measure a beam, but it said "hell no, there are spiders down there" So I asked it to give me a fix for a beam with a notch in it. Gave me a W10x22.

So, right now, gonna skip using AI.

 

[EngDM]

By the time AI can do basically everything an engineer is required for, plenty of other fields will already be taken over. I wonder what the world will be like once AI takes over all thinking-based jobs. Are we all now manual labourers for our AI overlords?

 

[271828]

At this point, I've only used AI as a really good search engine.

For example, I've had a nagging question for years about whether one of my calculation methods is correct. I've verified it to death, but was still at about 98.7% sure. I've never been able to find any source that was close enough to help. I typed three or four sentences describing what I'm trying to do into the MS Copilot. In seconds, it came back with the answer and suggested Python code. It was exactly what I needed to put that question down for the count. This one was a little odd because it didn't cite anything like they usually do. I don't know how it figured that one out.

 

[canwesteng]

Someday I think it may be able to turn drawings into 3D models, and maybe even laser scans into 3D models. Not necessarily accurate models, but good enough models that a person comes back and cleans up later.

 

[GregLocock]

I use ChatGPT to write simple matlab scripts, but it needs a lot of badgering to get even close, and to be honest I always end up taking the well written framework it has written and debugging it myself. As an example it tried solving a propped beam with a non uniform distributed load by applying the usual canned formulae, but after 6 iterations was stuck in a loop of failing to meet bcs.

So I told it to use an FEA approach, and after I'd stripped out all the bodges got a solution that worked.

 

[Brad805]

There are lots of everyday things I use ChatGPT for. You need a paid account. They are better. It takes practice to get it to produce useful results. Some examples:

• Data manipulation. You can feed it large amounts of data, and it can sort thru the columns and find what you need. In one volunteer role I prepare canvassing sheets for our annual auction. I needed to find the unique people in a spreadsheet with 900 rows. I also wanted to know which club the individuals were from. I exported a .csv file, explained each column and what I wanted. A few minutes later I had a list how I wanted. It had a few mistakes, but that is partly because of our database errors.
• I had been using it to help script daily work problems in Python. It works well for that, but I got to the point where I decided I need to take some courses to become faster. Our IT guy programs in C Sharp, and he said he uses it frequently to convert scripts into different langauges he is less familiar with.
• One day I wanted to generate a spreadsheet of all the file names in a folder. That folder has 1400 files or so. I was not sure how best to do that, so I asked, and in a few minutes I created a script I could use in Python.

It works best for problems where it has vast amounts of data to search thru. I doubt it will ever get good at a lot of things we do for the simple fact you must pay for all our codes. I think it should remain that way if we want to have code bodies in the future.

 

[JoshPlumSE]

271828: At this point, I've only used AI as a really good search engine.

Exactly. That's what it's best at. Or, at least that's the easiest thing to teach AI to do. Variations on this will likely be the primary driver for the use of AI for the time being.

You have to remember, the only thing AI can do is what you teach it. If you teach it things that aren't accurate, then you will get bad results.... Like we saw with Googles first version where it was hyper skewed to be "diverse".... which was evident when people asked it to generate images of the founding fathers of the US and it came back with a bunch of black versions of Washington, Franklin and such. LOL.

For creative types, I've seen people use it to create some pretty good memes... Like creating a white trash / trailer park version of game of thrones. But, I guarantee that it took a lot of interacting with the AI to get it to do that so well.