Can AI Replace My CAM Programmers?

JOSH READER: Welcome, everyone, and welcome to our talk exploring AI in CAM. We're going to be clearing the air of will AI replace your CAM programmers? We can move on.

So before we start this talk, I need to mention that during this talk, we will be talking about products pre-release-- sorry. We'll be talking about parts of the product that are pre-release, and please do not make any purchasing decisions based on these products. I'll give you 30 seconds or so, or you can pause the video to read through our safe harbor statement, as well as this. We'll be talking about our own opinions on the current situation in AI in CAM.

So who are we? My name is Josh Reader. I'm from Birmingham in the UK, and I'm a manufacturing specialist at our Birmingham Technology Center. I specialize in CNC milling and CAM, and I spend my days in the workshop verifying that our software is posting out the right code to make your machines run correctly. Over to you, Rob.

ROB BOWERMAN: Thanks, Josh. So hi, folks. My name's Rob Bowerman. I'm product manager at Autodesk, also based in the UK like Josh. I work in a slightly different part of the organization, so I focus on AI technology and bringing some AI technology into our product.

Over the past year, 18 months, we've had quite a heavy focus on where we could use AI technology within the manufacturing space. So I'm here today to share with you some of how we frame AI within manufacturing and some of the exciting explorations that we've been doing more recently.

JOSH READER: So here's a quick reminder of what the objectives are of this talk. We're going to be identifying the applications that we currently have in AI and CAM. And that's internal Autodesk and our partners. And we're going to be analyzing the benefits of the implementation that we do. And then this is really going to give you an idea of what's going on with AI in manufacturing at the moment to give it more trust in the future.

ROB BOWERMAN: So I'm going to kick off the meat of this talk. And we thought before we got into showing you some of the things that we've been working on, that we would frame automation a little bit. So when we talk about AI for CAM, we think of it as an automation tool, really. So automating CAM is obviously a large drive within the industry, and it's something that is obviously quite important for us to be able to do in our products to ensure that we remove some of the bottlenecks and make our users as efficient as possible when they are using our CAM software.

We know that manufacturing is really difficult. It's really complicated. There are lots of decisions that the manufacturer or the CAM user has to make every time they make a part. It's things around what tools do I use to cut this component, what types of strategies do I use, what parameters do I use within those strategies, and how do I know if I'm doing it right. Am I using optimized selections for the thing that I'm doing?

And often, you have to make these decisions time and time again for every single part that you want to make. And you're often doing it under time pressure as well. So any circumstance where we are making these decisions that are maybe slightly different but similar in some ways is an area that's ripe for automation.

And we also know that automation isn't something that's binary. You don't either do automation or don't do automation. We know that automation is a journey, and you transfer the responsibility of automation from the human over to the machine over a period of time.

Now, we can look across to other industries to try and draw parallels in this. Now, the automotive industry has a very good parallel to this, where they're trying to hand off the responsibility of the driver to the car to create fully autonomous vehicles. And what we're seeing is over the past 5 to 10 years, and also into the future, this transfer of responsibility as technology is developed and deployed.

So originally, obviously, we have the human fully driving the car, and then we have things like driver assistance being added into vehicles, which enabled you to take your feet off the pedals. We had things like lane assist or tracking of the car in front, which enabled you to take your hands slightly off the wheels. And we will at some point in time get to a future where the passenger in the vehicle-- where you would be a passenger now. The driver can completely disengage with the driving process. So there is a removal of inputs from the user to achieve the same result.

And it's really important that we have this transfer of responsibility because what that does is that also builds trust with the user. I can imagine that if we had self-driving cars and the laws to be able to use them from day one, that very few people would actually trust them or go ahead and use them. So we have to go through this period of time to have this transfer of responsibility so that people trust these systems.

And we can see similar parallels within the manufacturing industry. So there is a series of automation steps that you can take to get yourself to a place where you are fully automating your workflows. And there are tools that are being developed at each of these stages to enable you to get there. So things like-- products like Fusion today enable you to do some level of standardization. So you can create tool libraries, material libraries. You can create presets as to how you cut those materials.

Once you have those standardized libraries, we also enable you to gain things like insights. So can we use things like feature recognition to find simple geometries on your models and offer those to you as a level of insight to help you make better decisions?

Going a step further than that, we can give you things like templates, which means that you can define recipes that you use time and time again. A step beyond that is using something like the API to be able to hard code rules for automation. So if you're familiar with Fusion, we have quite an extensive manufacturing API that enables users to begin to do this today. But all of this is what we would call a kind of traditional or procedural automation, where you're hard coding those rules.

Now going forward to maybe get to full automation, we might have to look to other technologies to get to that final 20% of the journey. And this is where AI can be used.

So I mentioned before procedural automation, which is fantastic. It can get you a long way on this automation journey. But there are some troubles with using this type of technique.

So firstly, you need to be able to code the rules, which means that you need to know what the rules are. So if you don't know what the rules are, you're not going to be able to code them.

And then if you do know what the rules are, coding can take a long time. And what we often find is with any automation process where the rules are coded, there are always edge cases and caveats where the automation then falls over. So what do you do in that case?

Now, we find that to be most useful, the actual automation workflow needs to be tailored to the individual use case. And you often have a couple of options to do that. You can pay somebody to come and tailor it for you, or you can do it yourself.

And to do it yourself, that obviously means you have to be highly skilled, have a high level of expertise, to do that. And that's something you've either got to invest in, or you have to outsource. And then finally to round this off, the automation only improves when you update the code. So that means that you need to be capturing data and analyzing that data and drawing conclusions on that data throughout the manufacturing process to know where you can improve.

So you need to change-- you need to turn those conclusions into some actionable changes, and you go back round in the loop and code some new rules for automation. So this is a great approach, but there are lots of areas where it can fall down. And it can be fairly resource intensive to get to something that's working. And then when you have to change that thing that's working, you have to go back round this loop.

So the question that we ask ourselves is, Can we accelerate manufacturing automation with the use of AI? So there might be people in this talk, in this audience, who are experts in machine learning or AI. There might be people here who have no awareness or understanding or never even heard of ChatGPT. You've been living under a rock for the last couple of years. [LAUGHS] So we just wanted to level set a little bit on some of the things that AI is good at before we get into the next section.

AI is not a silver bullet for everything. There are certain tasks that it can do really well today, and there are certain things that it does terribly at today. So we've put together a bunch of use cases here where we think the technology is most suitable. And just highlighting a few of those, AI is very good when we want to classify or cluster data into different classifications.

AI is very good when we want to predict or forecast what might happen next when we're using something like time series data. An example of this might be if you have a factory running and you're monitoring all your machines, can you monitor that data and maybe predict things like when is my machine going to fail; when is it going to need maintenance, this type of thing. And that's a very common application of predictive AI today.

AI is very good at making recommendations based on your peer group or people who are maybe in a similar circumstance to you. And we also-- we finally see a new wave of AI that's coming through, which is generative AI. And this is AI that can generate new data. So you will have seen this in the form of ChatGPT, which can generate text data or things like Midjourney and DALL-E, which can generate image data for you. This is the real new wave of AI, which is grabbing all of the headlines.

OK, great. So now we wanted to touch a little bit on what we've been exploring in terms of some of these AI applications for manufacturing. And we've been exploring three proofs of concept. We're going to show you three proofs of concept today that we've been exploring. And really, they fall into these three categories, which you might see spoken about in some other talks at AU. And those categories are analyze, augment, and automate.

So one of these projects helps you gain insights into your data by analyzing it. One of these helps you automate your workflow. And one of these helps augment how you interact with the product to hopefully make you more efficient.

So when we were thinking about exploring AI for manufacturing, we really asked ourselves three questions. And those questions are, how might we guide a user through the manufacturing process, how might we enable a user to program a part quickly, and how might we reduce the need for expert knowledge when you are programming components. And so the three projects that we've been exploring address these questions exactly.

So to address the question of how might we enable a user or guide a user through the manufacturing process, we've been working on something called the manufacturing assistant. And this is a tool that's providing users with an on-demand Fusion manufacturing expert.

To address the question of how might we enable users to program more quickly, we've been exploring some functionality we're calling CAM Similar Search. And this enables a user to gain insights on previous parts that they manufactured which are similar to the current part that they're manufacturing so that they can get a head start on things like tool selection and toolpath programming.

And to address the question of how might we reduce the need for expert knowledge, we've been focusing specifically on an area of recommending cutting feeds and speeds to a user. Now, I imagine everybody in the audience is probably familiar with cutting feeds and speeds. It's a bit of a dark art when it comes to manufacturing. And we wanted to explore a tool that would give the user a starting point as to those parameters.

So we're going to dive a little bit more into each of these POCs. Now, as I-- sorry, as Josh mentioned previously, this is all unreleased functionality. And the reason that we're showing this to you is we would love to get feedback, and we would like to understand that we are going in the right direction with this type of technology and understand your concerns and really get you excited about it because, at some point, we would love to be able to get this in the hands of people to use it to get more interactive feedback.

So the first thing that we will dive into in a little bit more detail is the manufacturing assistant. So this is currently a preview in Fusion. If you are part of Fusion's Insider Program, you will be able to have access to this. You can enable the functionality in the Preview panel. And this is essentially like a chatbot style experience which uses third-party technology. So it's not an Autodesk-developed AI model. But the AI model that we're using has seen Autodesk help documentation, so it can answer specific questions about Fusion and Fusion [INAUDIBLE].

Now, you can see once the tool opens, we give you some sort of example questions. Here, we're clicking on one of those example questions, which is, Where is my toolpath trimming command? And then you'll see very quickly the AI will come back with a response telling you where that trimming command is. And then it will also give you a help documentation reference that you can follow if you want to read more details.

Now, as you can imagine, you might want to ask this any question So we've put in some guiderails to try and remove the various questions that people might ask that we don't want to give information on. So for instance, if you ask something out of context, like how is the weather today, we will say, sorry, we're unable to provide you information on that.

Going beyond this sort of Q&A-style experience, we're also exploring-- and I don't have a demo of it here-- but we're also exploring how this tool can action events on your behalf. So you might ask, how do I create a new manufacturing setup, and it'll give you advice on how to do that. But going beyond that, can we have the tool actually create that setup for you or create that toolpath for you or edit that toolpath for you? So this is something that we are also starting to explore.

So I'm going to quickly pass back over to Josh, who's going to talk through some examples of questions that people have asked previously and the types of responses that we're getting.

JOSH READER: Cool. So you may be thinking, if you're not a Fusion user right now, how does a Fusion user come across getting an answer to a question they have. And that's where our community gets involved in our forums.

Our forums are a great place to ask questions, but they're not the quickest. So you could have a simple question that it takes a couple of hours to get answered. And sometimes, it doesn't get answered at all. And that relies completely on human interaction and our community.

So here's a few examples of forum posts that took a while to get answered, and we can get an answer in seconds using our manufacturing advisor. So a quick example on this one-- the user was confused at the difference between the chamfer width and chamfer tip offset. And as an experienced user, I know there are ways of finding that out using the tooltips. But using the Manufacturing Advisor, We can get the exact response we're looking for in this. It's looking for the definition between the chamfer width and the chamfer tip offset.

I've copied and pasted some part of the forum post into the Manufacturing Advisor, left in some of the spelling mistakes, left in some of the grammar issues, and it's given me the answer I exactly want. So it's telling me the difference between the chamfer width and the chamfer tip offset with answering his question at the bottom saying, "A large chamfer tip offset moves the tool lower but does not change the chamfer width." So it's giving exactly what he wants in that circumstance.

We go to the next slide. And same again here-- so we've got a user where their 2D contour is cutting on the wrong side of their selected sketch. So as a user, many of you may know that when you're using 2D contour, it's not model aware. So therefore, it will go either side of that line with the tool. And sometimes, it's not the correct side of the line. So we have moments where, in this circumstance, the tool is now gouging the component.

And it can be answered on the forum. It took a few hours to get a response back. And with the Manufacturing Advisor, we've just asked it, "My contour is cutting on the wrong side of the model. Why is that?" And it gives me the simple answer of, "The red arrow needs to be ticked to the other side." And that's just a simple mistake of it going on the wrong side of the line and can be fixed very easily.

ROB BOWERMAN: OK, awesome. Thanks, Josh. So just some really nice examples of how this tool can quickly answer questions and get a mover moving when they might otherwise be stuck waiting for some advice. So I think that's a really nice example of how we can use that type of technology to make you more efficient as a new or even experienced user.

Touching on the two other projects, these are not quite as developed, so don't really have too many nice demos here. But it's still worth us just talking through them. The second one was called the CAM Similar Search project. So this is whereby we want to look through all of your previous manufactured components to show you ones that are similar to the part that you're manufacturing at the moment.

So how would this work? So it's kind of like a search functionality. You're in Fusion. You have a new component. It's just CAD at the moment. You have no toolsets. You have no toolpaths programmed, no setups or anything like that, or no machine selected.

We will take that geometry, and we will compare that geometry in a very quick fashion to all of the previous geometries that you manufactured. And this is the kind of USP for using AI in this use case is the speed of doing this. So you could search through thousands, tens of thousands of parts in a matter of seconds to find the ones that are most similar.

This is a really common problem in a lot of companies where they make lots of parts that are really similar, and they end up having to reprogram them time and time again. This could save a lot of time in quickly giving you that starting point by showing you a similar component. From that point, you could then import the assets from that previous project into your new project-- so your tool libraries, your machines, maybe even your templates and your recipes that you used-- and apply them to your new part to get you going.

And then finally, the last project that I wanted to touch on is the manufacturing feeds and speeds recommendation. So as I said before, feeds and speeds are a bit of a dark art in manufacturing. Everybody has an opinion on them. And how people get their feeds and speeds varies from workshop to workshop.

Often, you'll have a new material, and you have a tool, and maybe you'll go and ask your mate in the workshop what would you use. And they'll give you some advice, and you'll use that as a starting point. And then you'll do some trial and error from there.

We want to try and remove that barrier of having to have that expert knowledge or find that person that knows that or go on the internet and search that thing by giving you a reasonable starting point within the product. And we want to do that by building it directly into the CAM tool. So what you're seeing here is an example of how we've added a Suggest Feeds & Speeds button directly into the toolpath dialog to give you a reasonable starting point.

Now, this takes into consideration various things. It obviously takes into consideration your tool, the type of tool you're using, the material of your tool, the material that you're cutting. But at the moment, there are lots of things it doesn't take into consideration. So there are areas-- many areas where we could improve this by taking into consideration things like your depth of cut, your width of cut, the machine that you're using to understand the torque and spindle speed ranges that you can deal with.

And what you're seeing is once we press that suggested Feeds & Speeds button, we make a request to our AI model, which then returns some values. And we're not just returning one set of values. We're actually returning three sets of values. And that enables you as a user to look through those three sets of values and choose something that you think is most suitable.

The difference between those three sets is supposed to be something that's perceived as more cautious-- so maybe slower cutting; something that's perceived as being a bit more aggressive if you want to cut the part faster; and then something in the middle. And like I said, at the moment, the intention is to not make something perfect or optimized, but it's to make something that can just get you going, get you started.

And with all AI models, it's something that can improve over time. And it may even be able to start to learn and understand the preferences of that individual user. But that's something.

So that rounds off the three explorations that we've been looking at recently and hopefully, that's interesting for folks in the audience. And yeah, we would love to hear your feedback on those at some point in time.

OK, cool. So I'm now going to pass over to Josh for the next section, whereby he is going to talk about partnerships and Fusion add-ins that can also help you automate your manufacturing workflows.

JOSH READER: So yeah, as Rob said, we're going to be talking about what our partners are doing at the moment and what's actually out there. So the things I'm going to be talking about are available to the user now-- so yeah, if you just skip into the first one.

So CloudNC, we'll start off with them. They were the first company to implement AI into Fusion programming. But their main aim of CAM assist is to reduce programming time and to eliminate CAM bottlenecks, which this application does very well. Its original product focused on 3-axis milling but has now branched onto 3 plus 2, cutting parameters, fixture design, and quoting.

As we know, AI is not a one-size-fits-all functionality. And currently, AI applications like this can complete about 80% of the process, significantly reducing the number of clicks a CAM programmer needs to make. This means that we're saving the program time on tedious tasks like simple part programming. Go on to the next slide.

So as I mentioned, they've branched into other sides with their business. So one of these is their Cutting Parameters tool. So as a machinist and a programmer, we all know that determining feeds and speeds can be a lengthy process, and dealing with tooling reps isn't always the fastest process. And we don't always have the perfect packaging to know exactly what that tool does. It might just be a tool that's in a drawer.

So with this tool, it means that we are able to get our physics-based feeds and speeds quickly and very close to our optimal settings in seconds. It really does remove a lot of the complicated mathematical equations for determining parameters like surface finish and the real gray area of preserving tool life versus increasing productivity. Personally, I've used this tool a few times, and it consistently provides great results and matching closely to what I've determined through experience.

I want to go to the next slide. So here's a quick video describing what the product looks like. So what we're going to look at is the tool library selection. It chooses your tools, the material that they're going to be using, and what machine it's on, and whether you want to do 3-axis or 3 plus 2. Then you select the current setups that you want to use.

And then on the next tab, we have our tooling area. So this is where we give our specific tooling parameters. And then the final tab, this is where our machines and programming knowledge is still required. We need to give it the right questions and the right answers so it can program the component correctly.

So this component that's in the video is a component that we did with Haas. And it's built on a curriculum, and you learn to program this over time. But we're just using this as an example for our CAM Assist.

Personally, I gave it a go with this component. It took me about eight minutes to program this side with a machining time of about eight minutes. So as we look at the product, the CAM Assist is giving me a load of toolpaths on our component.

So this is a simple simulation. We can see the green areas that are machined fully. And then we have our purple area, which is our roughed area. So as it goes through the simulation, we can see what areas have been machined and what haven't.

And, for me here, I can see on the outside of the component on one of the steep walls, it's not quite finished the component. So it's just missed one slight area. And we can now go back and take a look and change those programs.

So it's got me 80% of the way there. I can now jump into the wall section, find the toolpath that was the issue, and I can go in and edit that really quickly. So I've got very close towards my finishing line. I just have to make a few tweaks to that toolpath to make me happy. Again, if this was a production run, you would go in and make sure all of those settings are as fast as possible, and it would be a lot less time spent on programming the component and more time making that toolpath more efficient.

So if you want to skip along. So the next one, now let's take a look at another partner application, which is CAM Accelerator by Toolpath Labs. So their approach to AI in CAM is slightly different. Rather than focusing on the CAM, they focus on the quoting process to ensure the shop takes on a job that is profitable and good for their capabilities. This includes identifying potential issues, such as excessive number of setups or unavailable tooling before the job is taken on. This will reduce the risk of costly estimation mistakes and machinability errors.

And this really does take away the factor of machine downtime. These factors can take whether a machining business can run or not. With machine downtime, it's very costly and time consuming because those spindles need to be moving as much as possible. Additionally, Toolpath built a cost estimator to streamline your quoting process and one-click programming in their beta.

So now let's take a closer look at their application. So Toolpath at the moment is located on the web and looks like this. You can send the part through Fusion, or you can load a step file onto the web. And this is it. Straight away is where the calculations begin.

Initially, this is the second tab, so we're on the tools rather than the summary. This is where it's going to show you the part that's fully green, in this case, where everything is machinable. And that's with the tooling I've specified in the setup, as well as pointing out any sort of unmachinable corners or sharp corners or areas that are unreachable-- for example, an undercut.

So just to note, there's the three colors at the bottom. So there's the machinable. There's no tool area, so the tool can't reach down there. And there's the unreachable. So it will give you color-coded, nice and simple, can this part be machined or not.

And on the right-hand side, we have the tools that are required. And they're located on the right-hand side. And selecting the tool allows you to see where that tool is being used on the component.

If you go to the next slide, Rob-- so the next slide is where we identify the number of setups. So really quickly, you can see whether this is going to be too many setups or not. So if I have a 3-axis machine, I don't want to spend too much time making a part that has four or five setups, for example, because that's where we're going to lose that time. The setup of the machine with an experienced machinist is going to spend a lot of time doing that. So we need to make sure that we want to bring on simple jobs that can be a two-setup job or a one-setup job if we possibly can to make sure we're using the machine's capabilities wisely.

So we can see on the right-hand side the setup's clearly defined. We've got setup 1 and setup 2. And all those features on the part are then recognized and categorized. So you can go into those folders to see where all of those features are.

So if we go to the next slide-- and this is where we start to import our toolpaths. So again, as I mentioned, this is part of their beta program. So there will be some serious development going on on this in the future. But on the left, we can see all the features that are categorized. And then they've put specific toolpaths on that inside the toolpath side. And then they match that to what's in Fusion and post it over to Fusion. So this is where that one-click programming comes into play, just like CAM Assist.

So go to the next slide. So finally, the cost estimator for toolpath.com. Here is where they have a cost breakdown of making the components. This is material dependent. So it takes into consideration the time taken to machine, the tooling cost, the shipping costs, and other factors.

So here, we can see an example of an aluminum part at $119. And then if we skip to the next slide, we can see there's a big difference by just changing to stainless steel, costing $424. So you can really see whether you want to be-- your shop's willing to take on the cost of making that component. And you can compare your profit margins from that as well.

So that's our two partners. We just spoke about what they do with their AI softwares. And we can also take questions from you to make sure that it gets to them, and we can hopefully answer them.

So if we go to the next slide-- so let's have a look at closing and what other research there is available. So first of all, I'm going to hand over to Rob, and he's going to just give a quick summary of what we've been talking about.

ROB BOWERMAN: Yeah, OK. Thanks, Josh. So yeah, in summary of the presentation, really, like we said at the top, manufacturing is complex, requires multiple experts at different levels to be successful. CAM as a process is extremely time-consuming. And it means that you have to make the same decisions time and time again on often things that are very similar.

We mentioned that automation gives you as a user the opportunity to be more productive, remove some of these bottlenecks that you are facing. And as we mentioned, automation comprises several technologies. It may be more traditional procedural or deterministic hard-coded rules, or it may be new techniques, things like AI and machine learning.

And as we mentioned, AI/ML, it's not a silver bullet for everything. And the combination of these technologies-- so the AI and the procedural technologies-- will probably at the moment result in the best types of outcomes.

However, I must caveat this. This is obviously a view as to where we are now. Who knows what might change or what might come in the future, particularly with the rate of development as we're seeing in this technology right now. So I'm just going to pass over to Josh to round off the presentation.

JOSH READER: So finally, the question-- will AI replace my CAM programmers? Our conclusion is no. But that doesn't mean never. But we're talking about in the near future and what it can-- I mean, I want to cover what it can help you do right now.

So the main aim of AI at the moment is to make us more efficient. So what can it help me with at the moment? So it's really going to reduce the amount of time spent on tedious jobs. As a CAM programmer myself, there's lots of different things that I could spend my time better. So we've got lots of things like quoting, basic part programming, which still needs a bit of experienced human interaction in, complicated feeds and speed generation, soft draw generation, fixture designing, tool recommendations, product support, and data insights.

There's lots of things that I can get AI to do to save me more time. I can spend more time working on the really high-complex work-- for example, some full 5-axis machining-- and spend less time on basic 3-axis jobs. And that's at the moment. Who knows what's going to happen in the future?

If we skip to the next slide-- so I just wanted to give a bit more information on what we have and what is available to you. So I've got three really quick QR codes that you can scan to get a bit more information. There's a great video, which is on the left-hand side, which is the programmer versus Cloud NC versus Toolpath. So this is where a YouTuber goes through and looks at all the different ways of programming and compares them together. So it's a great video to watch if you want to learn a bit more about that.

We have our AI for Manufacturing landing page on the Autodesk website, which is the middle QR code. And then there's a few blogs from CloudNC which are available, which the latest one is talking about ITAR compliance.

Thank you for listening, and we hope you enjoy the rest of your AU.