🐈
» Groups » Upwork Engineering » Forum » Re: AI risks for Engineers?
Page options
25005175
Community Member

AI risks for Engineers?

So, with all of the hubbub surrounding the effect of ChatGPT and similar text- and image-generation AI, I wonder if anyone here has given thought to any existing or upcoming AI that affect engineers and CAD modelers.

 

As a 3D CAD designer/engineer, I am unfamiliar with any 2D AI tools, so I won't address those. But I am interested if anybody has knowledge/experience with any.

 

Regarding 3D tools, there are already the generative design and topology optimization tools like those found in Fusion360 and Ansys. They are not free, only available at the part level, usually don't optimize for stress-relief, and they don't optimize for any manufacturing technique (not even additive). They exist and can be useful, but are poor tools for most applications, like drawing a simple bracket or creating an enclosure. But there is at least one tool that will be good for that (albeit overkill for basic jobs): Hyperganic.

 

Early last year, Titans of CNC produced a complex part that was designed using Hyperganic. Currently, the software is only for commercial/research entities. I'm not even sure if it is free or purchased, but there is definitely an application process, according to the website. But I can say that it is game-changing for high-level design. If that level of AI becomes readily-accessible and simple to use, then there will be a massive shift in the levels of work that require skilled engineering 3D CAD modelers. Certainly, design/engineering productivity for companies would skyrocket. I don't know that something as complex as Hyperganic will ever be "simple" to use with precision, but I don't see any reason that a chat-based system would not eventually be able to do things like design and iterate on simpler models like brackets, enclosures, and power-transmission assemblies.

 

9 REPLIES 9
k_froudarakis
Community Member

Hey Jonathan,

 

You raise a very good point. But I think we are way off from these tools posing a threat to human engineers in terms of employment. Most of these tools are iterative algorithmic tools (not AI) and use FEA analysis to give a "score" to each solution. They test out many options for each step and create a population with solutions and scores and then select the ones to move on to the next generation. They are more heavily based on Genetic Optimization algorithms, which have been around for years, than pure AI.

However, I am sure that you noticed that the results are fairly elaborate. And that most of them are only manufacturable with various 3D printing technologies.

So, I think that humans will be better at the following problems for a fair bit more:

  • Designing simple parts to solve simple problems. Not everything needs to resemble the pelvis of a cat...
  • Design for certain manufacturing technologies with specific requirements (minimum bend radii in sheet metal, avoiding sharp internal corners for CNC manufacturing, etc.)
  • Selecting manufacturing technology and adjusting the design accordingly to reduce cost.
  • And others.
williamtcooper
Community Member

Jonathan,

 

During the next 2 to 3 years, most knowledge worker jobs will radically change due to AI tools. Many of the entry and intermediate jobs will be gone or significantly changed. The winners will be Experts that use AI and that are great salespeople. Upwork will be totally different within a few years. The changes are already occuring this year; just look at all the posts in this community complaining of having no job offers.

 

I attended the annual stockholders meeting for Upwork yesterday and you can read my thoughts and information from the CEO of Upwork in the Coffee community where I posted.

celgins
Community Member

Jonathan L and Konstantinos F,

 

Very interesting!

 

I have not visited the Upwork Engineering Group because almost all of my engineering skills have diminished over the past 10 years and I don't want to embarrass myself!


But I have an observation/question.


I have worked in the military/defense sector for many years. One of my clients is a defense contractor and I sometimes communicate with 3D CAD Designers/Engineers who generate engineering diagrams and drawings for military satellite systems (e.g., rack assembly schematics, isometric drawings, wiring diagrams, exploded view drawings of equipment, etc.) They use Autodesk AutoCAD, Solidworks, Solidedge (and some others) to develop brackets and other pieces. (As you can imagine, it's a lot of COTS hardware/software cobbled together with military hardware/software/systems to create a specific, tactical environment).

 

Weeks ago, I was having a discussion with one of those engineers regarding the AI risks for engineers. Also, I watched the Titans of CNC video where they used Hyperganic to print the injector head. This is the type of thing I thoguht about while having the discussion with the engineer.

 

My question is: Are you saying that you believe algorithms / tools like Hyperganic are close to a point where the human 3D CAD Modeler can be fully replaced to develop complex parts, but it is way too expensive and complext to operate right now?


Then Konstantinos mentioned iterative algorithmic tools. Are those more akin to how robotics in automobile manufacturing plants are used to paint, weld, assemble, etc. based on set data/models? Meaning, those robots operate on existing software that was written with specific, iterative steps?


Clark S wrote:

Then Konstantinos mentioned iterative algorithmic tools. Are those more akin to how robotics in automobile manufacturing plants are used to paint, weld, assemble, etc. based on set data/models? Meaning, those robots operate on existing software that was written with specific, iterative steps?


No, not quite. The way industrial robots operate, they execute a program line by line. It is literally called line-by-line execution. It's the same way that a CNC machine works. Now, if you incorporate variables or parameters into that program you have something that can be called an algorithm. But the way these optimization software work is different.

 

They are based, with additional proprietary tools and methods I imagine, on the mathematic algorithms of Genetic Optimization Algorithms.

https://www.mathworks.com/help/gads/what-is-the-genetic-algorithm.html

 

The difference here is the intermediate step of evaluation. This step might contain an entire FEA simulation or even multiple. They also incorporate a degree of uncertainty, ingeniously named "mutations", so as not to have the algorithm get trapped on local optimal (minima and maxima) instead of the actual global optimal.

 

I am sure that Hyperganic and Autodesk's Generative Design and every other tool are not simply executions of Genetic algorithms. They will have a bit more in there, I hope. But the core and the idea of it is what is described in the article above. So, now we have a method to optimize a problem (design a rocket nozzle or tune a PID controller). But what's the problem?

 

A human is always, or for very much longer, going to be better at the abstract task of parametrizing the problem. It is something that many designers do instinctively. Some call it DFMA, some call it experience, and some call it "they don't teach you the important stuff anymore". But generally speaking, it is the process of drawing on experience and constructively and innovatively defining a problem so that it has a solution. It's just that sometimes, computers are better at the next step of solving that problem.

Konstantinos,

 

Well said. Thank you for that explanation

 

I'm going to take a look at that Genetic Algorithm piece at MathWorks.

25005175
Community Member

You can look at a simple implementation of genetic algorithms in the software Deepnest.io which is an open-source 2D nesting software. It is nowhere near as powerful as commercial software like Nestfabb, but its free and works for me!

Slightly off-topic, but it speaks to the strength of the Engineering Group here on Upwork. This high-level AI conversation that went through neural network based AIs and Genetic Algorithm based design software has led me to discover Deepnest!

Thanks Jonathan, I actually needed that type of software. So cool.

Quick tip 1: Deepnest plays much nicer with SVG than DXF.

Quick tip 2: Deepnest may forget to append the file extension when exporting the nest, so you will have to change the filename to append the extension yourself.

celgins
Community Member

I am also hearing of Deepnest for the first time. I know some folks who migth be interested in this--especially the fact that it is free!

 

Thanks Jonathan.