The All-Axis Podcast

How AI is Revolutionizing Quality Inspection in Manufacturing

• Michael Thiessen, Keven Wang

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 31:09

Quality control in modern precision manufacturing is rapidly shifting from subjective, manual visual inspection to automated, high-speed computer vision powered by artificial intelligence.

In this episode, host Michael Thiessen sits down with Keven Wang, founder of UnitX, to explore how AI-driven vision systems are transforming shop-floor quality inspection. Keven shares how advanced machine learning algorithms analyze surface defects, geometric tolerances, and complex material anomalies in real time, eliminating human error and reducing costly scrap rates. Together, they discuss the technical realities of deploying AI visual inspection, integrating edge computing hardware onto active production lines, and how automated quality data creates a continuous feedback loop for machining process optimization.

Whether you manage a precision machine shop, quality department, or high-volume production line, this conversation offers a practical look at automating visual inspection to maximize yield and efficiency.

In this episode: 
00:00 - Introduction: The Challenges of Manual Quality Inspection 
04:15 - Computer Vision & Deep Learning: Beyond Traditional Machine Vision
09:40 - Real-Time Defect Detection on High-Speed Production Lines 
15:20 - Overcoming Implementation Obstacles & Edge Computing Integration
21:10 - Using Quality Inspection Data to Optimize Machining Processes 
26:45 - The Future of Automated Quality Control in Precision Manufacturing

 To learn more about UnitX, visit their website. 

Tell us what you think - send us a text message!

Thanks for listening!

Speaker - Voiceover

Welcome to the All Axis podcast from the Experts and Tebis. Each episode we'll talk about technology topics, trends, or solutions that manufacturers need to know in today's rapidly evolving times. Now, sit back, enjoy the discussion and let us know what you think.

Speaker - Michael Thiessen

Hello, my friends. Welcome to the All Axis podcast. I am your host, Michael Thiessen. And this is the site where we bring you insightful topics that can help you out in your daily work, in your four walls, in your mold, your die shop, your aerospace shops. And hopefully, we can bring some interesting topics to your daily lives. At least that's what we're trying to do. Today we're talking about intelligence. Not my intelligence, because this would be a very short broadcast to you, but we're talking about artificial intelligence. And obviously, as we have in every single episode, we have experts in the industry that can give us a little bit more insight on what the artificial intelligence can do for you in the manufacturing area, but also on what value these kind of technologies can bring to you. Some people say that AI is the next industrial revolution that we see in our times. And for that, of course, we're going to be looking into this a little bit closer. We're not talking about ChatGPT or GROC or any of the other AI platforms. We're looking particularly at AI being utilized in the manufacturing sector, specifically on the automotive manufacturing or e-vee uh manufacturing, uh, any of those things. That that's the area that we want to concentrate our AI capabilities to. And for that uh conversation, we've invited a very unique individual in the industry. This individual has started in 2018 a company that is focusing on the visual inspection side of things in order to make perfect parts with their visual inspection, finding defects, but then incorporating AI into the situation. But this is not where this gentleman ended uh with the technology. He's also utilizing robotics in order to automate all of these things in there with the robotics and the visual inspection combined. The company that I'm talking about here is called UnitX. You can look them up. And uh, we are very pleased to have Keven Wang on this particular show, on this episode, talking about AI and how this particular technology, first of all, has elevated his company into the manufacturing side, but also asking him a little bit about what AI does for the manufacturers and how potentially you could be utilizing those types of tools in your daily environment. You probably have dabbled maybe with some AI uh information already on maybe your CNC machines or maybe your software products, but you probably have never looked at uh in the visual inspection side of things. So for that, I'm pleased to introduce Keven Wang. Thank you for the all on thank you for coming on the show.

Speaker - Keven Wang

Thank you, Mike, for the opportunity.

Speaker - Michael Thiessen

Oh, you're welcome because uh, you know what, everywhere where you're looking nowadays in every magazine news article, I don't think we can get out of this entire news cycle without hearing about AI. I mean, AI is everywhere nowadays. We have it on our phones, we have it on our computers with Copilot or Grok or Chat GPT, or if if you don't sound intelligent enough yet, you put it in Chat GPT and something smart comes out of it, and that's it. So tell me, Keven, how did you get involved in this entire AI?

Speaker - Keven Wang

So my mom uh was a mechanical engineer, and she worked in a factory uh designing air conditioners, and she's a very wise mom. And when I was little, she told me, don't study mechanical engineering. Go go do something uh better, right? Of course. And as a as a as a good kid, I listened to my mom and uh I went into software. But uh, you know, my my mom being a mechanical engineer, she she distilled in uh in me a sense of uh appreciating the the physical beauty. My mom would tell me, look at how beautiful this this this this desk is designed, it's got rounded corners. And I love building things, making things with my hands, airplane models when I was a kid. But I I listened to my mom, and being a good kid, I studied the software and studied uh AI uh when I went to grad school. And and this was the moment when uh AI is really on a cusp of the breakthrough. You're hearing AI is detecting images better than the human experts, distinguishing this breed of cats from that breed of cats, you know, not as useful tasks, but still becoming very good, very accurate. And uh this was in 2017, and me coming from uh mom's uh inspiration, my dream is to combine these two things, like how to make AI useful and apply to the physical world of manufacturing where things are made. Uh so combining AI and manufacturing is like a dream come true for me.

Speaker - Michael Thiessen

Now, AI, I mean, obviously has been just a recent concept, or maybe I should say 2017, 2018, or maybe even going back to 2015. It's not that we've had it around for 20, 30 years already. This is fairly recent that AI has become a very powerful solution and it can be applied for so many different things. And you said you went to grad school, I believe uh, if I remember correctly, it was Stanford that you went to. Did you know anything uh during the university days about AI? Did you learn about it? And you go, huh, I need some more intelligence, I'm gonna look into this AI. Or what got you to that point where you're saying, you know what, this AI could really be a big deal?

Speaker - Keven Wang

Um, I remember there was this uh uh newscard article I read uh when I was in grad school, uh, that the AI actually beat the human expert in uh classifying objects. Basically, there are 1,000 types of objects. This is the test that's given. And the task is to tell, given this picture, uh, which one out of 1,000 types of objects it belongs to. And the AI's accuracy rate for the longest time before 2017 is is much worse than the humans. But in 2017, using the the neural network is a new form of machine learning algorithm that that mimics how the human brain processes images, right? The accuracy actually beat the human experts. So that that was like a pivotal moment. And uh I thought, whoa, it's it's actually good enough.

Speaker - Michael Thiessen

So then you took this idea that you saw that you studied and did your grad uh project on at Stanford, and you say, you know what, I want to utilize this. We're gonna utilize these pictures or this this technology in manufacturing, since you said you love building stuff with your hands, and you said, Yeah, you worked on in a manufacturing floor uh on CNC machines and milling and and all of that, but software was really on your heart. So you must something must have struck you when you were on the shop floor that says, you know what? I need to apply this AI to inspection routines in order to make a great part and utilize that technology to identify people probably broken areas or bad quality or anything like that. Is that how it went?

Speaker - Keven Wang

Yeah. Um I'm fortunate to have uh some uh uh family uh relatives uh in manufacturing. And uh after I tell them, you know, I'm working on this uh AI and I want to apply it in manufacturing, and one of them immediately told me, okay, we have we have this problem, uh, which is visual inspection. And specifically, you know, we have all these cameras, we've installed them in the last 10 years, and they're very good at uh detecting what you tell it to detect, like you know, measuring this straight line to next straight line, you know, finding these straight lines. But what is not very good at is finding random defects, such as a scratch. So that problem is not solved by the legacy rule-based vision, uh, traditional computer vision. And can AI solve it? So that's the the question that that they threw at me. And then uh we started uh uh with this project, and uh then we realized yes, AI is actually good enough to solve these random defects.

Speaker - Michael Thiessen

Interesting. So you did you go out of Stanford and say, you know what, I'm gonna start my own company because uh 2018 was just around the corner, and that's when you started and co-founded UnitX. Was this right away your idea once you've graduated from Stanford? Uh, that you say, you know what, I'm gonna utilize this technology right away to start UnitX?

Speaker - Keven Wang

Yeah, essentially we got that first project with that uh for the first customer, if you will. They're a factory that makes a motor per uh commutator, part in the in the motor uh that rotates. And you know, they they really care about the scratches on that and that part, and they they have uh not been able to inspect them very well. And they receive about 200 customer complaints every year, a lot. And they they want some way to better inspect those. So we started with that one uh project with the first customer. And initially we didn't have a company, and you know, when when we started putting a lot of work and the customer say, Okay, well, you're you're doing good work, you know, we we should really enter into a contract, pay you some down payment, and then I give them my personal bank account and say, You can't buy the money here, and they're like, No, we need a business account. That's when we started the business. We realized, okay, this may have some legs.

Speaker - Michael Thiessen

Okay, so you're now what seven, eight years in the business. How has this transformed? I mean, you probably every project you got smarter. The AI learned from all of this, and you're applying this. Uh, you I know that you have multiple different products uh on the inspection side that you're applying this technology to. How has this now changed? How has it changed at all from the beginning when you started to where you are now?

Speaker - Keven Wang

It has changed a lot. Even the AI, uh, you know, AI is such a buzzword, right? But really, AI has gone through uh, I would say, three iterations uh since when our company got started. Yeah. Specifically, um, you know, when when AI uh was first introduced, it takes a lot of images, a lot of samples to teach. To teach a scratch, you may need a a thousand, one thousand scratch pictures to tell it. This is a scratch. And then after you know, a hundred, it it may not be very good. It will only detect very obvious ones. And after 500, it starts to get good enough, but it's not, it will still miss some very subtle scratches. After a thousand, it may be good. You know, that was seven years ago. But you know, it really has improved a lot. And and right now, our system uh we take up to 20 uh samples to to achieve uh acceptance. So that that's kind of the state of the art. That it takes about 20 uh images only, in some cases, even fewer. We see as few as five to to uh uh achieve the site acceptance testing, basically reach the sign-off of the customer. So AI has really improved a lot.

Speaker - Michael Thiessen

So where would you find your particular systems? What industry? Is it any? Is it any manufacturing industry, or is it geared more towards the automotive or the aerospace? Or where can you find this AI-powered solution?

Speaker - Keven Wang

The technology is is uh versatile and it's literally anywhere there is a mission critical inspection needs, anywhere where quality is uh critical. So lots of applications of uh AI inspection today are in automotive. And uh our company, uh we our biggest business is in automotive. Yeah, we have some in uh EV batteries, uh, some in medical devices, consumables. Uh we're also looking into uh consumer packaged goods, and uh, anywhere uh there is a need to inspect defects where the quality is important, where the uh defect escape may cause uh safety consequences or functional consequences, sometimes cosmetic.

Speaker - Michael Thiessen

Okay.

Speaker - Keven Wang

This is like a class A surface uh on a car door panel.

Speaker - Michael Thiessen

So can you get an inspection? I mean, I'm very familiar with blue light scanners and white light scanners, which were before that. Is this looking at the the same type of kind of architecture where you have a large capture frame or is it relatively small capture frame and you need to move the inspection? You just mentioned a door frame or or door uh to look at this, or do you have to move your camera or your device constantly on the part to capture the entire image?

Speaker - Keven Wang

It could be either, Mike. Um, like you said, um you can have large parts, you can have small parts, and uh you can have different uh sizes of the defects. Um so depending on uh the application, let's say sometimes we have a big uh washer panel and we want to see a very tiny dent, and then uh you you would want a uh very high resolution camera, but you know you don't want to have too many pixels that the camera gets overly expensive, then it gets cheap, it's actually cheaper to move the camera to scan the big panel. In some cases, you can get away with uh uh less pixels because the defects are bigger, more obvious. Then you can have a one fixed stationary camera to take a picture of a big part, one shot. Yeah, so so it really depends. And but yes, there's uh two components in the system, in the inspection system. There's uh the eyes or the eyes that takes a picture. Right, the lenses, brain, yeah, the brain that processes the picture and decides is it good or bad. Both are important.

Speaker - Michael Thiessen

So let me take that analogy then that you just painted. Can you get inspection systems that have less brain and more vision, or uh more vision and less and more brain?

Speaker - Keven Wang

Totally you can. Yeah, you can have you can you can kind of build Lego blocks uh together. So you could have in the uh one brain uh with uh uh four eyes, and each eye is looking at a different area of the parts or from a different perspective. Uh yeah, so you can definitely make sense.

Speaker - Michael Thiessen

And now I know why I don't see very well. I have only two eyes and one brain. So I understand that now. But let me let me ask this, uh Keven, though. From your experience now, and you've now been out doing this for eight years, of course, there's a lot of inspection systems out there. I don't need to name all of them because there's there's numerous inspection systems. If I'm looking at probing, if I'm looking at laser, if I'm looking at white light, if I'm doing uh if I look at your systems, so all of them have a very particular application. Yours is more geared towards the quality loop or the quality inspection systems uh that you're there. So why would a company that maybe already has an inspection solution, why would they look at, okay, I need to have a much better solution with AI? Do you find additional issues with the parts that the regular inspection system just did not find? Or why would a company's go that go that route and maybe employ you and your your company?

Speaker - Keven Wang

The application is really uh random defects. Okay. Where it's a uh quality inspection. Uh, you know, we're looking at uh random uh scratch of a random shape, random size, random locations. This is where the the traditional rule-based vision cameras have a hard time. The traditional rule-based camera is very good at finding what you tell it to look for, like a straight edge or a corner or a circle. Very good. Excellent repeatability and precision. Uh but when it comes to random defects that can happen anywhere, any shape, any size, this is where the AI, using neural network, we call it neural network architecture, that's where it shines.

Speaker - Voiceover

Yeah.

Speaker - Keven Wang

So it's able to learn from data. So instead of uh in the you know traditional vision where you have an engineer to program, here's what to look for. Here are the pixel values to define a straight line. In the AI world, you show examples to AI, and the AI learns those rules itself. So it does take more data to learn, but it's much more robust and versatile when it comes to these random defects.

Speaker - Michael Thiessen

So obviously, I don't need to convince you that uh, or I think you are convinced that AI is the future. But let me ask this this random question. Do you believe AI is gonna be an item that is not going away from the manufacturing sector? That it's always from now on will be implied applied in more and more areas because you've dabbled with AI already. So you are the the best expert that I can talk to about AI because you've dabbled with it. What do you see?

Speaker - Keven Wang

I think AI is definitely here to stay. And I also think it's uh it's important to view AI just as uh uh another technology, that it's not a silver bullet, uh, that it's not you know this magic thing that's gonna solve all the problems, right? It's just another technology in the in the toolkit. Uh and uh there's so much buzz around AI, but when it comes to creating value, right, how does it help uh manufacturers? I think we we we do need to be very specific and look at the ROI. And specifically uh for machine vision, AI is ready for prime time. The technology is mature enough, the accuracy we're seeing is much better than any rule-based uh legacy vision system. But specifically, you know, you asked uh Mike, you asked how does it benefit uh manufacturer, why do they use it uh instead of rule-based vision? Well, I could give you a story. Um, one of our customers they came to us uh because they were using a legacy camera using the traditional rule-based vision. And when they installed it, it was working decently well. Um, so they accepted it. But three months after, there's a new defect that happened. The process can drift over time. You know, for example, your your upstream machining tools, uh tool blade may wear out and sometimes not replaced in time, or your uh upstream suppliers' material changed the color slightly from this shade of gray to that shade of gray.

Speaker - Michael Thiessen

Right.

Speaker - Keven Wang

It's still a perfectly okay part. And that threw off the vision system. Now the vision system rejects 20% of parts every day. It could have been good parts. So you're scrapping a lot of good parts. Um that that's causing waste.

Speaker - Michael Thiessen

Correct.

Speaker - Keven Wang

That customer went to came to us, and uh basically within a week using AI, it's able to learn these uh defects and accommodate it and reduce the uh scrap rate down to about 1%. In addition, the the software is simple because that with AI, you you you would not need an experienced vision engineer to program the rules. You just show it data, you just show it pictures of examples, and then the AI will learn itself how to differentiate between good part and bad part. So then the customer is able to um self-serve, they're able to take over this machine on their own. So next time when you know a new uh defect happens, the the customer is able to do it on their own. And if they need help, you know, they they can obviously reach out.

Speaker - Michael Thiessen

Uh right. So what do you say to these individuals? And there's lots of them out there that are car, of course, are very scared about AI. That of course the the notion is out there AI will take our jobs away, uh, AI will do our jobs. Of course, there's numerous movies that we've been made, either Terminator or or uh there was an is another Arnold Schwarzenegger movie where all of us in the future have become big and fat and machines are only doing our work. Is that what you see AI do? Because you taught you said that AI can do things better, faster than visual, than the human uh factor can do. Can you see that AI potentially will be a threat to uh what I just mentioned?

Speaker - Keven Wang

I think uh AI is uh like any technology, it's uh it's neutral, it's a it's a tool, and it's up to us uh human beings how we use the tool.

Speaker - Michael Thiessen

Correct.

Speaker - Keven Wang

And analogy is like a nuclear power. The nuclear power, if if used very, very badly, you know, it can it can become a nuclear weapon, it can cause a lot of damage. But if we use it peacefully, we can we can perhaps generate uh lots of energy uh very very safely. So AI is just like that. And I think it's up to us how we use AI. And I do believe AI has a lot of potential if we use it right. And I I believe the best manufacturers that we have seen today, ones that that puts AI to work for them.

Speaker - Michael Thiessen

Right.

Speaker - Keven Wang

They adopt AI and they let AI work for the factory.

Speaker - Michael Thiessen

Right.

Speaker - Keven Wang

And then you're seeing this uh uh improvement in quality, improvement in uh yield that otherwise would would take a lot longer and a lot a lot more expensive to achieve.

Speaker - Michael Thiessen

What would you say to manufacturers that are looking into AI-powered manufacturing tools or AI-aided uh solutions such as yours? Is should they be saying, yeah, I'm gonna look at any product that has AI in it, or are there specific items that you say, you know what, maybe for your application AI AI may not be the best, or for you, you need to be looking at XYZ, or do we say no, any any solution powered with AI is perfect?

Speaker - Keven Wang

Yeah, definitely depends. Um, you're right. The the the AI, when it comes to a quality inspection perspective, it is the best for, well, I want to say like high volume, low mix uh scenarios. Not all the applications are are suitable for AI because the AI does need uh samples to teach, to train. Although it's getting better and better uh every half a year, but it still takes uh samples to teach AI. So when you have a uh high mix situation, high mix low volume, let's say uh every day there's multiple changeovers, different part types, and the part sizes are very different. Yeah, AI is probably not the right solution.

Speaker - Michael Thiessen

Okay. So you kind of guide your clients that says, okay, you know what, AI may not be the best. Now now let me ask you this. Um just as we're talking here, my mind goes wild because AI is such an interesting topic. Besides inspection now, I mean you've now have this inspection solution, this robotics. Is that the end for you? Because it seems like that you're very interested in applying EI to the manufacturing solutions. Have you come to the end of the line with this inspection saying, Yep, I've done everything that I could uh with this particular product that I have at UnitX? I need to now apply my brain or my technology or this the solutions that I have to other applications. What does your future look like?

Speaker - Keven Wang

I think there's so much opportunity in applying AI in manufacturing. And and UnitX as a company, uh, we are focused on manufacturing as our uh uh customer market to serve. Yeah, we we love manufacturing, we have a deeper respect for manufacturing and people who are working in manufacturing. We think this is where a lot of real values are created. I'd love for us to have more manufacturing. So, yeah, there's so much more opportunity. Uh, I think installing uh machine vision, installing cameras is just a starting point. The the ultimate goal is to help manufacturers improve the OEE or the overall equipment effectiveness. That is the yield, uh, the uptime and the throughput of production. Where today, you know, there might be a tribal knowledge that, okay, when my yield drops like 5% every Tuesday morning, oh, it turns out it's this uh this uh washer fluid upstream that needs a replacement.

Speaker - Michael Thiessen

Right.

Speaker - Keven Wang

With with all this uh data that's collected by AI, we hope that uh this data can automatically uh find these insights and uh and send these uh notifications to the manufacturing engineers to help them realize, oh, actually, there's a pattern here, so I can more proactively, more quickly fix the process upstream, this drift that occurred to uh reduce the scrap and improve the EU?

Speaker - Michael Thiessen

So would you? I mean, we only have a uh a couple more minutes left here, but let me ask you this AI, these AI solutions, do we need to implement those already in high school so that of course kids need to get used to it and apply it and even further it even further? Do we need to implement that already in high schools and colleges and all of that in order so that because if you're saying AI is not going away in manufacturing or in our lives, so to speak, do we need to implement that already early on in our child's education system in order to further it further in the manufacturing environment stuff?

Speaker - Keven Wang

Absolutely. I I think it's it's it's very it's gonna be extremely helpful. Is uh technology is here to stay, and it's up to us how we utilize technology, how we adapt to it. And education is uh plays a big role. If we can educate the future workforce from an earlier young age, that they can embrace this technology and make this technology serve them, then they're more better positioned to uh adapt.

Speaker - Michael Thiessen

Yeah. Have you ever run into a situation with your company, and this may be a gotcha question or whatever. Have you ever run into it that says uh that you said to a customer, sorry, I can't figure this out, my AI is not smart enough, and we can't do this?

Speaker - Keven Wang

Oh, we have, certainly. There are applications where you know it's uh extremely high mix. You know, the customer would switch over part types uh multiple times a day. And our typical rule of thumb is uh 200,000 parts made per year of the same part type. Then that's a threshold you you can use the AI. Otherwise, it's still gonna be cheaper and better to use uh conventional inspection.

Speaker - Michael Thiessen

So let me ask you this just as closing for you, because we have a lot of manufacturers listening to um our podcast. What would you advise the manufacturers out there when they look at AI or maybe even look at your system, AI inspection system, in order to determine if AI could help their manufacturing? What would you advise them to do?

Speaker - Keven Wang

I recommend start small, uh, start with one project and then experiment with it and realize it's going to take time. Uh the time is not necessarily the technology, it's more the human organization. It's getting your team, getting the manufacturers, engineers, the quality production team departments uh familiar and comfortable with the technology. That takes time, that takes alignment, but uh yeah, it always takes longer than uh initially expected. But once the organization is familiar and uh accepts this technology, then it's a beautiful future. Then you can really improve the uh OEE productivity and uh improve the quality.

Speaker - Michael Thiessen

Keven, remind our listener base if they want to reach out to UnitX or to you, Keven, where would they need to go to to find some information about you and your your company?

Speaker - Keven Wang

You can go to our website, it's uh unit xlabs.com, unit x l-a-b s dot com. Or you can send me an email. Uh my my first name uh is K-E-V-E-N. So Keven with two E's, K-E-V-E-N at unitxlabs.com.

Speaker - Michael Thiessen

And it's all because your mother told you do not do what I'm doing, do something different. And there you are doing manufacturing, applying your knowledge, your software in AI, and uh making the manufacturing world a better place. Awesome. Keven, thank you for being on this podcast and giving your knowledge and your intelligence in that particular field to our listening base. So thank you.

Speaker - Keven Wang

Thank you, Mike, for the opportunity. My pleasure.

Speaker - Michael Thiessen

Well, my friends, you've heard today all about AI and how beneficial this particular technology is in your manufacturing environments. I'm glad to hear that AI is not the perfect solution. It is a tool, it is a hammer, it's an it's a tool that you use on a daily basis if applied correctly. So, with that said, I urge you to look into AI power or AI solutions, dabble with it, do like what Keven said, start small, look at a certain project, start with one project maybe at a time and see how it could maybe improve your manufacturing processes. And uh hopefully this particular episode will steer you into that direction so you can utilize that particular tool in order to make you a more successful operator and a more successful manufacturer. We thank you all for tuning in to this episode of the All Axis podcast. We hope you found our discussion today insightful as well, as well as valuable. If you enjoyed today's episode, don't forget to subscribe on Spotify, Apple Music, or wherever you get your podcast. Feel free to share the episode with your network, leave us a review or hit the subscribe button. We love to hear back from you guys. For more updates, follow us on Instagram X and well as LinkedIn. And I wish you guys a wonderful manufacturing time. Until next time, keep shaping the future of manufacturing, and I'll see you next time on the All Axis Podcast.

Speaker - Voiceover

Thank you for joining this episode of All Axis. Please leave a comment or review with your feedback or what you'd like to hear in future episodes. To learn more about manufacturing technology solutions and Tebis's capabilities, visit our website at Tebis. That's T E B I S dot com.

Head Shot

Michael Thiessen

Host