In Between the Tech: A Conversation with Peter Denzinger of Vista Solutions

In Between The Tech

Raymond Yin
Welcome to In Between the Tech. Today, we're wrapping up Mouser's look at Industrial Machine Vision. Mouser sits down with Peter Denzinger, Vice President of Engineering at Vista Solutions to get his thoughts on the momentum behind this technology and the opportunities for engineers in this space. Listen as he shares his unique perspective in this innovative field of machine vision.

Question
Thanks for joining us today, Peter. As the Vice President of Engineering at Vista Solutions, please tell us a bit about the company and the role they play in the Machine Vision space?

Peter Denzinger
Vista Solutions is a machine vision solution provider. What that means is that we just do machine vision, and we use commercially available technology like cameras, cables, PCs, lenses, lights, smart cameras, as well as commercially available components. We don't manufacture components on our own. We also are a machine mission integrator, meaning that we don't do PLC programming. We don't build panels; we don't build machines. All of our employees just do vision.

We use basically every tool in the vision toolbox because we like to stay in a niche. Meaning that we do the 2D technology, 3D technology, thermal, we're doing measurement type applications all the way up to artificial intelligence type applications. We'll use something as simple as a smart camera to solve a problem, all the way up to a custom developed PC-based solution, running complex artificial intelligence. And that allows us to play in a lot of different industry spaces, basically everything from agriculture to construction to medical. Sometimes it's a bakery, sometimes it's somebody making turbine blades.

Question
Artificial Intelligence is popping up on newsfeeds every day. As someone who has been in this space for several years, what is your perspective on where AI, Machine Vision, and these other complementary technologies are heading?

Peter Denzinger
I would say that at the moment, we're probably on the tail of some inflated expectations, and we're likely bound for a bit of a correction. Perhaps a little bit of disenchantment will hit us with what we're expecting AI to do immediately. Coupled with that is the fact that industry tends to be a little bit less agile than the rest of the world.

Industry likes to move a little bit slower. They like to be a little bit more predictable. So, where do I see this heading? Industry right now is in a phase, that's been coined Industry 4.0 and almost everything about it in industry has been pretty focused on data capture. Getting data out of something. And right now, what we're doing with that data is a little bit shallow. We're not getting very deep into that data. And so, I think the next step is some actual use of this data that we're able to capture these days. In addition to that, we're seeing industry finally starting to adopt some of the things that are necessary for AI to really take off.

Industry's been slow to allow things to connect to the internet. Seems mundane to us, everything's connected to the internet, but remember, industry takes a long time to warm up to that sort of thing. They're also warming up a little bit more to allowing things to save a lot of data and actually use all that data. So right now, you might have a sensor that can capture all kinds of data, provide all kinds of information, but nobody's really thought about storing it anywhere. Nobody's really thought about managing the data and putting it into a container that's useful, and I think we're on the cusp of actually doing something with that. It might sound a little mundane that that's probably the next immediate thing to happen.

If we're to look a little bit further and say, okay, what might be coming for machine vision? Like, what's the next thing machine vision's going to do in industry that it's not quite doing yet? You can look at all the traditional things that have been advancing slowly over the past two decades or more. Things have been getting slightly higher in resolution, slightly faster. I'm not sure I'm looking for big breakthroughs there, but I am looking for a little bit more of a breakthrough in the realm of process monitoring. So right now, machine vision is heavily used in industry to solve specific problems to guide industrial processes, to look for flaws in parts, to measure things, et cetera, guide one process to another so that they're coupled.

But something that is very reliant on the human, currently, is the process monitoring. So, did something go wrong? What went wrong? How do we know that something didn't fall outside of the scope of the system? And I think machine vision can be used in the same way that personnel on the plant floor are watching processes for things to go wrong. And right now, the AI tools, they're not quite at the point where they can grab that data and make sense of it in the way that we're able to use it. So, I think that that may be the next step in the specific application of AI in the industrial space.

Question
You touched on some possible applications and making better use of the data. Machine vision is a natural fit for industrial settings. Can you talk us through some of the other applications, beyond industrial, that are using or could utilize machine vision?

Peter Denzinger
Machine vision really thrives in industry for a lot of the reasons that technology itself thrives in industry. You've got that repeatable process, predictable steps, and operations. And so, machine vision, in its infancy, was able to be quite simple and solve a problem. You've also noticed that cameras are everywhere, but industrial processes are everywhere too. When I was thinking about this topic, one of the things that I thought of immediately, everything's kind of been industrialized. Not everything, but a lot of things have, things that you might not be thinking of.

For example, mining and agriculture and forestry, all of those processes are becoming industrialized, because we want to use technology on them. You might not think of shipping and receiving as an industrial process. I mean, perhaps you do now in the Amazon age, but traditionally that wasn't something somebody thought of as an industrial process. A bakery you might not consider to be an industrialized process, but the bakeries that feed all of us - very industrial environments as well. Machine vision has found its way into all of these industries, through the foothold that is industrialization.

But beyond that, perhaps some places that you might not realize machine vision is active, are these places where cameras have ended up. And now because we have advanced processing techniques to deal with the non-conforming or not repeatable nature of the world outside of industry, we can now actually do machine vision, not just observe, not just capture images, but do machine vision in these environments you might not really expect that autonomous processing is happening in.

One example that many are likely familiar with on a toll road machine vision is reading your license plate automatically. You've got cameras doing the same algorithms used to count the number of fish in a river. Passing through some point, detecting invasive species, is this type of fish that is not supposed to be in this river detected, if it is, sound, some alarms, get the naturalist to take a look at these images and verify. AI's enabling us to do all these interesting things.

We're also well aware of driver assist technologies. You know, many of us who have driven a current vehicle maybe are lucky enough to have one that not only has driver assist technology - just to show you something or to give you a light - but it's culminating in autonomous vehicle operation. These are machine vision or algorithms operating on the cameras and the LIDARs installed on modern vehicles.

We've been peripherally aware that cameras are popping up in every everything. These cameras that once were just maybe looking for any kind of motion or any kind of brightness change, they are able to do much more intelligent things. Your doorbell went from detecting motion to detecting humans to detecting the faces of recognized individuals. And this is something that became normal or appeared and then became normal within a five-year span. I don't want to sing the praises of AI too loudly, because it might seem like I'm trying to plug something, but it gives us that ability to do something and elevate that camera to a vision system. And this elevation is happening basically everywhere cameras have found themselves.

Question
Is there a particular solution that you found to be very meaningful or an incredible learning experience?

Peter Denzinger
Absolutely. I'd say something that probably breaks outside of the box the most significantly that we've been working on for a while and, have seen some interesting success with, probably a little bit more success than many would've expected is a road repair robot.

This is a robot that rides in the back of a truck and it 3D scans the road, and it will fill cracks in the road with the sealant, filling cracks in parking lots. You'll also see it a lot in the middle of a road, where the crests meet, that's a common place for cracks.

We have a partner company that we've worked with that does robot work. We provide the vision part, in this case. And so, our vision part, combines two exciting things, and one is 3D vision. So, 3D data capture, getting real world measurements as we drive down the highway. And then we're processing these 3D images with AI techniques to extract where the cracks are and to create pathways for the robot to travel on while it dispenses the goop. It's very technically impressive that we were able to get that working. There's also a big safety boon in using something like this, because obviously road work is very hazardous. It's a very dangerous job, especially in the cell phone age. It's only getting worse for the folks who are risking their lives to keep our roadways safe and maintained. This could be really something that has a big impact.

Question
Speaking of robots, industrial machine vision balances a working harmony between humans and machines on a factory floor. What advancements on the component level are needed to expand this relationship to other environments such as those in conditions unfit for humans – such as extreme temperatures, pressure, water, etc.

Peter Denzinger
I would say as a preface that the environment that we don't want people to go in, the scope of what that is increases every year, as it should. As we want to keep humans safer and we're able to keep humans safer, and we don't subject them to the same hazards that they've had to in the past and perhaps are still subject to today. Cameras have been in these environments that are unsuitable for humans for quite some time. And as a result, there is a lot of great technology that exists to ruggedized cameras, to the point where they can withstand blasts in a mining environment. They can self-clean. Obviously, there's waterproof camera enclosures and air conditioned ones. Now, of course, almost every camera has a problem with heat. It would be great if components could be ruggedized to the temperatures without cooling because implementing cooling, greatly increases the bulk of a camera enclosure. Now, when we're talking about the entire vision system, of course, the camera isn't the only component. When we're looking at AI systems, they're not tiny. They do require a processing unit, either a circuit board running some kind of GPU, all the way up to a full pc.

And of course, we can put these in large boxes and air condition them and waterproof them and do all kinds of nice things. But the bulk becomes a big problem. What do we do with this bulk? Two different fronts, I think, would be great.

One is that 50 degrees Celsius is that temperature where things start to need to be actively cooled. Cameras don't like being very hot because they will sense heat as light, and they'll increase the amount of dark noise in the sensor, and your image quality will decrease, as the temperature increases above 50 degrees Celsius. Also, electrical components, their lifespan will go down the longer they spend at that 50, especially imaging sensors. So, if there is a materials advancement that just pushes that boundary a little bit and increase to 60 would actually do a lot.

Perhaps a more realistic place to look is in miniaturization and the miniaturization of AI processing hardware so that we can get it into the same enclosure would go a long way as well. We don't always want an entire refrigerator on the mining site or in the foundry. It blocks access to the forge. We need to get that as small as possible. Another place where this could go is in data transfer. So, one great way to keep things safe is to not have it there at all and just send the image away from the camera.

And, of course, in the industrial environment, people don't love wireless. Perhaps that's going to change with advances in wireless technology, maybe more reliable wireless data transfer, perhaps further distant wireless data transfer. Perhaps more use of 5G. So, 5G is used right now in a lot of applications where the camera is in a place that is just inaccessible even to wires, very far away from where you can get a WiFi network. Perhaps we'll see more of that technology on board with cameras.

Question
We’ve talked about the opportunities, and you touched on some of the challenges, but what do you see as the key challenges or roadblocks facing machine vision?

Peter Denzinger
It's an interesting question because some of the roadblocks you can look at as opportunities to solve a problem in some way.

I mentioned this earlier. Everything is capturing data or has the opportunity to capture data. It was the big push of Industry 4.0. It was the first piece that was necessary for us to do any Industry 4.0 things, and everyone focused on that. But not many people actually embraced it enough to want to say, oh, I actually want to store this data, I actually want to manage this data, I want to turn this data into something that will work for me and I can do something with. And, I believe that one of those roadblocks right now is that there just isn't a good way to capture the data in a way that we actually want to act on and use.

For people trying to get off the ground with vision and with AI, where we might have a lot of focus on collecting data. AI needs data in order to actually function. In order for data to become useful, you'd need an AI to train. You need both of those things, but often what I see is that people only focus on one piece of the puzzle, and they try to solve the whole problem.

I think we have to expect that data needs to be created before the AI can do anything with it. And then once the AI has some data, perhaps new systems can be created using that data.

Raymond Yin
We hope you've enjoyed this episode of In Between the Tech. This podcast is only one part of Mouser's in-depth look into the rapid adoption of Industrial Machine Vision. Explore the entire Empowering Innovation Together offering on this subject, with articles, videos and more at mouser.com/empowering-innovation.