Is anyone else using AI to support engineering design work?

[Arbutus]

So we have all heard about AI systems - they exist in various forms, from authoring and writing tools to fantasy art. There is no question that the tools are useful although their veracity sometimes borders on hallucinatory.

I've been using chatGPT 4 (32K tokens) for engineering support for over a year now. Generally to verify my assumptions, often to research a particular design detail, and very often for advanced C++ code development and refactoring.

The main lesson I have learned, is to spend the time to understand the limitations of these LLMs (Large Language Models) and understand the boundaries of their interpretive memory (the token bank), otherwise the responses received will likely be junk, misleading or simply fantasy. It would be very easy for someone with a limited technical background to believe the answers as fact, but that is not always the case.

I think the AI systems available to us peons are years behind the much more sophisticated systems in use by large corporations and the military. In fact i would say the public AI is just about to leave kindergarten for high school. Soon, the depth of the infiltration into our lives will be deeply impactful and this change will happen very quickly.

Has anyone else explored AI as part of their engineering work?



Here's an example of a simple conversation:

Me: Explain how the function of a ground lathe turning tool is affected by the rake, face and clearance angles


The geometry of a ground lathe turning tool is critically influenced by the rake, face, and clearance angles. Each of these angles plays a significant role in the tool's cutting performance, chip formation, tool life, and the quality of the machined surface. Here's a detailed explanation of how each angle affects the geometry and performance of the turning tool:

1. Rake Angle​

The rake angle is the angle between the tool face and the perpendicular to the workpiece surface. It can be positive, negative, or zero.

• Positive Rake Angle: This angle makes the tool sharper and facilitates easier cutting by reducing cutting forces and improving chip flow. It is beneficial for machining softer materials and producing finer finishes. However, a too-large positive rake angle can weaken the tool's edge, making it more prone to chipping.
• Negative Rake Angle: This angle increases the tool's strength and is suitable for machining harder materials. It can handle higher cutting forces and temperatures but may result in higher cutting forces and energy consumption. It generally leads to less smooth finishes compared to a positive rake angle.
• Zero Rake Angle: This is a compromise between the positive and negative rake angles. It balances cutting ease and tool strength, making it versatile for various materials.

2. Face Angle​

The face angle, also known as the angle of inclination of the tool's face, influences how the chip flows off the cutting edge.

• Steeper Face Angle: A steeper face angle promotes better chip evacuation, reducing the chances of chip clogging and helping in maintaining a clean cutting area. It can improve surface finish but may also increase the tool's cutting edge wear rate.
• Shallow Face Angle: This can improve tool life by reducing wear on the cutting edge, but it may cause poor chip control, especially in ductile materials. This angle is more suitable for hard and brittle materials where chip control is less of a concern.

3. Clearance Angle​

The clearance angle is the angle between the flank of the tool and the workpiece surface. It prevents the tool from rubbing against the workpiece.

• Larger Clearance Angle: This angle reduces friction between the tool and the workpiece, which minimizes heat generation and wear on the tool's flank. However, an excessively large clearance angle can weaken the tool and reduce its strength.
• Smaller Clearance Angle: This angle increases the contact area between the tool and the workpiece, enhancing the tool's strength but also increasing the risk of friction, heat generation, and potential for tool wear. It can lead to poor surface finishes and higher cutting temperatures.

Combined Effects on Tool Geometry​

• Tool Sharpness: Positive rake and larger clearance angles generally make the tool sharper and more efficient at cutting, but at the cost of tool strength.
• Tool Strength: Negative rake and smaller clearance angles enhance the tool's strength and durability but may result in higher cutting forces and potential for poorer surface finishes.
• Chip Formation: The rake and face angles critically determine how chips are formed and evacuated. Properly optimizing these angles can ensure smoother cutting and efficient chip removal.
• Heat Dissipation: Clearance angles influence how well the tool dissipates heat generated during cutting. Optimal clearance angles reduce heat buildup and prevent thermal damage to the tool and workpiece.

In summary, the geometry of a ground lathe turning tool is a careful balance of rake, face, and clearance angles. Adjusting these angles appropriately based on the material being machined and the desired cutting conditions is essential for optimizing cutting performance, tool life, and the quality of the machined product.


OK pretty simple, but not detailed enough, so I try another probe:

Me: What are typical angle values for a ground carbide tool for use with mild steel?

For machining mild steel with a ground carbide turning tool, typical angle values are chosen to optimize cutting performance, tool life, and surface finish. Here are the typical angles:

1. Rake Angle​

• Positive Rake Angle: Generally, a positive rake angle of about 5° to 10° is used for mild steel. This helps in reducing cutting forces and improving chip flow.

2. Clearance Angle​

• Side Clearance Angle: A side clearance angle of 6° to 8° is common. This prevents the tool from rubbing against the workpiece, reducing friction and heat generation.
• End Clearance Angle: An end clearance angle of 6° to 8° is typically used. This ensures adequate clearance to avoid interference with the workpiece surface.

3. Tool Nose Angle​

• Tool Nose Radius: A nose radius of 0.4 to 0.8 mm is often used. A larger nose radius can improve surface finish and tool strength but may increase cutting forces.

4. Back Rake Angle​

• Back Rake Angle: For mild steel, a back rake angle of 5° to 10° is typical. This angle helps to control chip flow and reduce cutting resistance.

5. Side Rake Angle​

• Side Rake Angle: Similar to the back rake angle, a side rake angle of 5° to 10° is common. This angle aids in chip removal and minimizes cutting forces.

These angles can vary slightly depending on the specific cutting conditions, tool geometry, and machine setup. However, the values provided are a good starting point for general machining of mild steel with carbide tools. Adjustments may be necessary based on specific requirements or performance observations during the machining process.

Example Tool Geometry for Mild Steel​

• Side Rake Angle: 7°
• Back Rake Angle: 6°
• End Clearance Angle: 7°
• Side Clearance Angle: 7°
• Nose Radius: 0.4 mm

These typical values strike a balance between effective cutting performance and tool longevity when machining mild steel. Adjustments might be needed based on specific machine capabilities and workpiece characteristics.

 

[Susquatch]

When you posted this on your profile I thought you were asking if anyone had used AI to generate a CAD model. I was very sceptical of its ability to do that. I still have my doubts about that working at all - let alone being practical. But never say never.

However, your much larger question posted here seems much more likely to produce results. If one of the goals of AI is to find and accumulate existing knowledge in order to synthesize new or different applications, then as your examples show, it can provide lots of information much faster than we can do it on our own.

Whether that information is correct or even applicable is another matter. Unlike you or I, AI cannot reliably distinguish between the ramblings of a self professed idiot and a real expert. Especially not for machining where the idiots outnumber the experts 1000 to 1. I think the stats alone would add credibility to what is otherwise garbage.

For your question, I didn't think the answers were all that bad. But I think that's because most idiots don't pontificate on such matters, but cutting tool manufacturers often do. And their information is usually pretty good.

When I do a search for information like that, I generally ignore sources like YouTube preferring to read technical papers or blogs by tooling manufacturers. But that can take days or weeks to find what I need.

So I think AI could be quite useful in such a quest. The fundamental problem is that you are still stuck wondering if it's right unless you already know - in which case why ask at all?

Very interesting @Arbutus. I can't wait to see what others think.

 

[DavidR8]

I'm using ChatGPT to design load bearing trusses for my new shop.


















Not.

 

[mbond]

RE this specific output, the conclusions on negative rake angle seem wrong

You mention C++. Extending that to programming in general, I think that the 1000 idiots to 1 expert ratio is generous, but this is my area of particular expertise. Unfortunately, many poor programming patterns are expounded very loudly including in articles published by major players - Microsoft, Intel, IBM, Redhat all have abject nonsense written by employees on their various websites. And there is way more garbage written on the many independent web sites

 

[ChazzC]

AI includes a lot of applications in addition to search engines and research tools. I use several digital photography tools that incorporate AI to enhance exposure, color, contrast, shadows, etc. based on the software’s interpretation of the subject matter. AI is also used to “separate” subjects from the background automatically so that alternate backgrounds can be used - while this feature works well for people, it doesn’t do a good job with machines (yet) which is why I still place a contrasting color behind my mill & lathe when taking photos for publication.

 

[Susquatch]

Another field where AI is growing leaps and bounds for those big outfits who can afford it is in on-line sales and support. For example, many major insurance agencies are using it to evaluate risk profiles for potential customers and answer questions. The customer doesn't even know they are talking to AI. But AI can research a client's background while they are talking and know more about that potential clients insurance risks than any real human ever could.

We have all seen the major shift to offshore customer support with agents that can barely speak your language. The next wave will be to AI agents who speak your language perfectly and answer all your questions on the spot. Perhaps even escalating the call to another agent or supervisor or manager algorithm.

Scary or comforting?

 

[Susquatch]

RE this specific output, the conclusions on negative rake angle seem wrong

I think that your view of this depends on your own definition for positive and negative rake. I see many different definitions. To the point where I don't care anymore what anybody calls it. Instead I look at the job and the tool and ignore the words and names. And even then we get all caught up in what is a face and what is an edge. It drives me crazy. But a rose by any other name smells the same.

This is the definition I prefer to embrace. It doesn't require words that confuse us. 1 is negative 2 is neutral, and 3 is positive. This gets complicated further when you add chip breakers. Unless I misunderstood, I think the definition in the photo fits with the AI description. But that's an opinion..... LOL!

 

[mbond]

But that's not actual negative rake. That's holding a positive rake tool (or insert) at a negative angle. Actual negative rake is useful not for hard materials but for plastic ones

 

[PeterT]

I started dabbling with ChatGPT but have kind of migrated to Gemini. Nothing serious yet, just messing around. Some things it does very well, others...not so much. I think this stuff is still in its infancy & AI is already a maligned marketing term. But specific to these apps I'm generally impressed with some aspects. Like text based tools - summarize this, or distill that, or rewrite this from a particular perspective. Some of my programming friends use it a lot to debug, improve code or look things up, just quicker. I know enough VBA to be dangerous so I pasted some code with planted errors. It not only spotted them, but provided me background information & better recommendations. Google search would have directed me to Bust Buy to buy a new PC. So despite early days, it's really quite exciting, especially for something that is free. But it also doesn't take much to confuse it.

I'm sure there are lots of AI books out there, but FWIW I just finished this on Audible & it was pretty good: A Brief History of Artificial Intelligence What It Is, Where We Are, and Where We Are Going By: Michael Wooldridge.

 

[Susquatch]

But that's not actual negative rake. That's holding a positive rake tool (or insert) at a negative angle. Actual negative rake is useful not for hard materials but for plastic ones

That's true. But it's only to graphically show what they think rake is. Lots of other aspects are missing too. I only chose that graphic because it makes it very clear what they think rake is and it happens to line up with my own definition.

 

[mbond]

I think I need a drink

 

[mbond]

It will help much better than bad fish

 

[PaulL]

I am probably the person on this forum most qualified to speak to the ability of AI. If you ask I'll list my qualifications.

GPT knows *nothing*. Every question X you might ask you should reframe as "what would an answer to X look like". It then uses a model of all text that was ever written to form an answer that "looks like" the answer to the question you asked.
It doesn't know anything about the domain.
It doesn't check if it's true, it doesn't check if it's lying to you.

It just bullshits at you.

You'll have to judge if this reply was chatGPT BS or if someone with some knowledge wrote this.

Paul

 

[Arbutus]

If one of the goals of AI is to find and accumulate existing knowledge in order to synthesize new or different applications, then as your examples show, it can provide lots of information much faster than we can do it on our own.

It certainly does that. It can also help with the mundane when fully instructed. Take C++ coding for example. When correctly prompted, the code that is generated is really quite reliable and always syntactically correct. Errors occur, but in my experience those are the result of loose specifications and instructions from me. The benefit is the reduction of both time and debugging frustration.

I find the image generators to be really fascinating. There are many available (MidJourney, Dall-E, etc) each of which have a very unique way of interpreting the instructions and sometimes provide quite astonishing responses. For example, I was exploring some options for a sustainable, off-grid, net-zero home. I fed the system several sets of queries. Many of the instructions resulted in garbage images which completely missed the mark. By training the system (thumbs up or down, or text feedback) the AI learns to reach towards what it is you are looking for.

This is the query: "produce a realistic image of an off-grid, net-zero home in the Canadian Rockies with the emphasis on sustainability and aesthetics"

And this was the response after some training:


And with that established I could use the AI and it's knowledge of buildings to drill down on specifics such as square footage, layouts, heating & cooling, water storage etc.

My point is that these systems have deep and broad knowledge which can be used to advantage, if the user understands the limitations as well as the broader benefits. As an ancient engineer, without the collegial support I once had, I find the AI to a valuable companion these days as well!

 

[PaulL]

Except that they have *zero* knowledge. All they can do is rewrite text that they scraped. Sometimes that's enough to *look* like knowledge. But most of the time it's just bullshit.

 

[Arbutus]

Except that they have *zero* knowledge.

I disagree. Part of that conversation was a discussion about roof and wall R values and how that could be achieved. I was impressed by the response which was much deeper and more inclusive and insightful than a simple Google search would have produced. It included the aluminum sheathing and the airspace requirement as well as the calculated final R value without prompting and therefore included extracted and inferred data from the Building Code manual. Now whether or not that is "knowledge" or "machine inference" or whatever you call it, the response was valuable.

 

[slow-poke]

I'm not using AI, however my daughter works for a large pharmaceutical company as a data scientist. The masters program she took in preparation for the job was AI focused. My daughter tells me that she "obviously' couldn't do her job without the AI tools she uses daily. She definitely uses it for code generation however that's just one aspect and wonders incredulously how I operate without.

Solving tricky math problems, I was pretty impressed with mathcad 20 years ago, but as stated you need to understand enough to know when the answer produced is actually valid.

 

[Susquatch]

Wow! When I first saw @Arbutus's question, I knew that this would be a fascinating discussion.

Except that they have *zero* knowledge.

I think that very much depends on your definition of knowledge.

If knowledge implies understanding, then I must agree.

If knowledge implies internal data storage, I also agree.

But if knowledge implies ready access to documentation and information (knowledge) stored elsewhere on the Internet, then I disagree.

In academic circles, the search for new knowledge is "what they do". (if you ask a group of taxpayers, the majority will say their job is to educate, but if you ask a group of academics, the majority will say their job is to do research and create new knowledge. An debate about that could consume another whole thread.) As a direct result, the number of papers they produce is a foundational measure of performance. These papers are published and available on the Internet. If those papers are knowledge, then clearly AI has access to knowledge. Better faster access than any of us do.

A much better debate might be to ask if Artificial Intelligence has intelligence!

In no way shape or form am I challenging your qualifications even though I might choose different definitions.

 

[combustable herbage]

Right from the horses mouth.

 

[Bandit]

It seems everyone has forgotten the "old saw", definition of AI-- blond dies her hair different color.
Don't get me wrong, there some dam smart blonds out there, -not to sure about the one that married me, the marrying part anyway, or maybe it the staying with me part!
It appears that AI is not intelligent in anyway, it only gathers information and uses the common? 50%+ answer. For RIGHT or WRONG! So, 2+2 could equal 5 if the majority of answers were 5.
Keep in mind, I likely don't know what I am talking about, many people tell me that!