Enabling Automation Podcast: S4 E1

We’re excited to bring you the fourth season of our podcast series, Enabling Automation. This monthly podcast series brings together industry leaders from across ATS Corporation to discuss the latest industry trends, new innovations and more!

In the first episode of season 4, we welcome host Ben Hope who is joined by Stan Kleinikkink. to discuss the future of AI in automation.

What we discuss:

  • Why innovation is important to ATS
  • What is AI in the context of Industrial Automation?
  • How does AI impact traditional technologies?
  • What are the biggest opportunities for AI in the industrial automation industry over the next 3 to 5 years?

Host: Ben Hope, ATS Corporation (ATS Products Group)

Ben has 25 years of experience in the automation industry, spanning both technical and commercial roles. He’s seen firsthand how technology can transform every phase of the automation lifecycle, from concept to engineering to assembly,  integration, operation and service.

Guest: Stan Kleinikkink, ATS Corporation 

Stan has been with ATS for 30 years and is currently the Director of New Innovation, with a focus on the ATS Innovation Center.

——Full Transcript of Enabling Automation: S4, E1—–

BH: Welcome to the Enabling Automation podcast. The show where we bring together thought leaders and innovators from across the ATS group of companies to explore trends, technologies and conversations shaping the future of industrial automation. I’m your host, Ben Hope. I have nearly 25 years of experience in the automation industry, spanning both technical and commercial roles. I’ve seen firsthand how technology can transform every phase of the automation lifecycle, from concept to engineering to assembly,  integration, operation and service. It’s a journey that continues to evolve, and today’s topic is one that’s accelerating that evolution at lightning speed. Today we’re diving into the world of artificial intelligence, or AI. Is it just another buzzword? Or is it a transformative force that’s here to stay? More importantly, what role will AI play in shaping the future of automation?

BH: Joining us today is Stan Kleinikkink. Hi, Stan. Great to have you here. Thank you and glad to be here. Can you start by telling us a bit about yourself and your role within the ATS group?

SK: Thanks, Ben. I’m an engineer here at ATS. I’ve been here also kind of a bit longer than yourself, I hate to say. It’s been 29 to 30 years that I’ve been, living my life here. And my role is quite exciting. I am the director of New Innovation. So it’s been a long time to get to this role. And it’s an exciting role that we’ve really developed over the years to, to focus on new innovations. So a lot of times we’ve been very good at executing, and we have a lot of areas of interest that we developed very advanced solutions. And as Director of New Innovation, I’m able to monitor those areas and move them forward where it’s appropriate, but also look into some white space areas where ATS hasn’t played and do an evaluation. If those technologies or those customer problems, we could add value, we could add solutions, and we could add benefit to automation into our customers by implementing new solutions. So my focus I’m within the Innovation Center, and I’m very focused on looking at new technologies to implement at ATS.

BH: Okay. Thank you for joining us today. So talking about innovation and how ATS approaches innovation, why do we do it and why is it important to the company? So we’ve always been an innovative company. We are there to innovate for customers. We’re there to find new solutions. We’ve seen the evolution of automation over the last 30 years, and it’s just kind of part of our DNA to work with the latest hardware and work with the latest solutions. So we approach it two ways true innovation, let’s say true new novel ideas. We don’t want to be tied to a contract which has a set deliverable and a set timeline. So we pull that out of our delivery, and we put that into the innovation center, that we can then develop the best solution without the risk of disappointing our customers if we’re not ready on a date. That that has been able to make, enable us to make step changes in innovation.

BH: And it allows you to fail, which I think is important in the learning process.

SK: Failure, I hate to say, is part of innovation. And then bringing in failure we fail fast. So we are set up to try the risky parts of the innovation, implement solutions and ideas quickly in a rapid prototype phase. Find what’s capable and not capable in the quick way, and then continue the investment on once we’ve eliminated a lot of risk. So and then why is it so important to the company strategy? The second part of the question, it’s critical to ATS that we innovate. We see the entire industry changing. We see that we can’t deliver the same solution to our customers that we did five years ago, because the industry moves forward. And we’ve always had the capability and want to continue to be on the leading edge and not be a follower of others, but set the path to find new solutions for our customers and driven those new innovative solutions into the industry, as opposed to following other leaders.

BH: Yeah, that’s really exciting. I think it’s inspiring for employees to walk through the Innovation center, see what we’re working on, how it can really drive things forward. So I think that’s very cool. So looking at our topic today, artificial intelligence, AI, let’s start by saying what exactly is AI in the context of industrial automation?

SK: So there’s two words that come around, One is artificial intelligence and one is machine learning. In my mind, artificial intelligence is a bit larger than machine learning because it doesn’t necessarily have to be adaptive or predictive. Machine learning is deep neural nets and a lot less predictive in its outcome. Both ways I’d say when we talk about AI, one element of it in manufacturing can be the generative AI, where you’re working with text and document histories and generating new answers. One element of it can be deep neural nets on image processing that you see in autonomous robots, autonomous cars. You can also see it in the manufacturing on quality inspection of of parts that you can do a deep neural net on an image. Third element of it is more of a data driven data feature where you’re doing a multivariate. You’re looking at multiple influencers on a part or on a quality or on a process. You can feed those into a model. You can train the model to be predictive and have the outputs. So to summarize, see three areas that AI is influencing our automation today; vision, generative AI, and just call it numeric modeling of multiple variable problems.

BH: And how do you think that’s going to impact the industry in general from traditional automation technologies?

SK: AI is one element that’s driving change in the industry faster than we’ve seen in other areas. The ability to create a model that can adapt to new situations, that doesn’t require a hard coding, predetermined set of conditions to meet those scenarios is the real driver that AI you can train off of data. You can it can learn to create a better solution, and it can apply that better solution to areas unseen. Imaging is one of the easiest ways to maybe describe that in an easier form. With imaging, we could take a set of good images. You can train that the good images are acceptable. And if any image is not within, range or within the level of variability seen within those good images, it can create a reject. So you’re not having to create thresholds and deterministic tools to process the image. You are training the model based off of a set of data. And then it can go forward and make the decisions for you within AI. That’s the imaging example. It’s not limited to imaging, but that’s one of the biggest power that we see in AI today.

BH: It’s a great example to help understand the power of AI. If we get a little more. So we talking about imaging and the impact on imaging. What other areas of the automation lifecycle are most ripe for disruption or enhancement through AI?

SK: There’s a inherent amount of knowledge required to be a proficient operator, process engineer, quality person, and to really drive these high volume, high precision processes to their potential to eliminate waste. So the lifecycle is really looking at where is there waste in the system. Is there a way to generate or have data to quantify that waste. And then is there a way to eliminate it with AI so that maybe very general and high level and that’s kind of the opportunity level that that we’re looking at, if you dig down a lot of it is, is a prediction on what could be happening tomorrow in quality or in breakdowns. Can we drive the data to see predictive elements in the manufacturing? Implement in AI fashion And then rather than doing a time based maintenance or rather than waiting for failure, we can have an insight or a drive to say go and now’s the time to make a change. Understood if the full solution is potentially having an automated response to it and you don’t even need the person, I think. Predictive maintenance kind of. Yeah, yeah. But the real tangible difference is instead of having to collect data or instead of having to go and analyze data as a subject matter expert, can you take that and just provide direct direction to people to go and make the improvements required to improve the assembly process?

BH: That’s really cool because I find in my experience with an operator, sometimes a fault will come up on the machine and it’s some cryptic fault that nobody understands. But can we get to the point where the HMI can tell the operator, this is what happened, this is why it happened, and this is what you have to do to get out of it.

SK: Exactly. And when we get back to the case that the younger generation and the way labor is changing is there aren’t people there that are running the equipment for 10, 15 years. That is a very much a goal is can someone come in? Can we give the instructions to the person that within their first shift to their first couple of shifts, can they utilize the tools available to them to run the equipment at a very high level of efficiency? Be productive right away? Yes.

BH: And I think looking at other areas like supply chain; optimization is an interesting area of AI too. And you look at inventory management or even forecasting, which sounds like very kind of simple things to do, but can sometimes be very, very complex. There’s lots of different variables. And getting AI to help kind of in that area.

SK: And one of the exciting things I think, is I’ve always driven for data driven decisions. AI also drives people to say, well, what’s the data? So another process you can implement is if there is a problem, well what is the data? Can you, can you get the data? Can you then analyze the data? And can you manage a solution? So very much I think a very positive fix is it drives people to a data driven decision in manufacturing. They get the data. We’ve had opportunity where the solution ends up being on AI, but it’s a data driven solution and it’s a simpler solution than AI. And it’s implemented and drives change. So both AI itself is easier to deploy. It’s quite scalable. It eliminates a lot of complexity. But the process of implementing AI has also led to both AI and non AI solutions improving the quality and the performance of the equipment.

BH: So talking about data and data is the foundation of any AI implementation. And I think everybody’s probably played around with ChatGPT or Copilot or these kinds of tools. And what you put in is very, very important to what you get out. Kind of like the analogy to good cooking, if you use crappy ingredients. You’re going to get a crappy meal. So how do we ensure data quality? As kind of a first step.

BK: There’s such a wide range of data. Maybe here we do use some Copilot, which goes and looks at all of SharePoint and can drive some answers. And then you are looking at a large swath of data that’s maybe not curated and not checked. This is specifically generative AI. Can you then focus your chat bots and focus your data libraries to be on the curated/review data sets that the answers presented to the user are off of higher quality data sources than just any data available to the user? So that’s for generative AI we find that important and a higher level of accuracy, very much from the incoming data. The other element of actual machine data. So I’m going to go back to the predictive maintenance elements of motor currents motor torques. You come back to performance of individual devices of the pneumatic devices and how that timing works. You come back to vision results and characteristics of your equipment. The first source is do you actually have data? So a lot of times this data doesn’t exist. And if it doesn’t exist on your equipment, there could be a generic model that you could deploy based off of others data. And it’s but then it’s not tied into your equipment and your problem. It is the first step is getting data, here at ATS. We’ve been collecting data for 20 years. So we do have a set of data going back a long ways which does enable AI. The other thing that we’ve seen practically with our testing is if you test AI and you do some root cause on why it’s right or why it’s wrong, or giving the desired answer versus the non desired answer, just to maybe be redundant, it is The answers reflect your incoming data. So if you’re incoming data is wrong, your outgoing predictions is wrong. If you have 1 or 2% of your data that are outliers, incorrect, it becomes very difficult to have confidence in the output of your model because you don’t have confidence of what goes into your model. So how do we ensure data quality? There’s the actual how getting into the engineering of it. It is engineering data system that can collect data, and you need to check and verify and even validate if it’s critical for quality, that the data you think you’re gathering is equal to the data, that is, that the data is correct. It is a process to ensure your data is correct.

BH: So it’s you still need expertise or subject matter experts to understand the quality of the data. It’s never or not yet anyway, to the point where you can start implementing AI to do things that you might not necessarily understand how they’re done.

SK: Correct. And AI can be another tool to help assist checking data and look for abnormalities in data and some of the of the outliers. Then the first process is as well as this an outlier because there’s an outlier in my process? Or is this an outlier because it’s a data problem? And that process, even using AI can also increase your data verification, but it’s still a process that can’t be skipped. And it does need a human or a source of truth in the loop to validate that your data is true. So it’s really when we started working with AI, there’s two ways that we would create models and two big hurdles that we would see. One is to actually validate in a controlled lab environment that AI would perform as expected, and that the model was capable of delivering results that could impact the business. The next level is actually taking that lab pilot proof of concept and deploying it, that it can impact people in the organization. It can be used in a production environment, whether that be in, an office environment or on the shop floor. So with any technology, any innovation, the movement from that working concept into impacting people’s lives is one of the riskiest areas of innovation and one of the riskiest areas of AI. So to move it into there, we’ve implemented a lot of user testing in the pilot phase. So we take the best adopters and the people most open to change. We’ve expose them to the model. We expose them to the functionality of the intended system. We receive their feedback directly from using it in the pilot phase. And we’ve really wanted to transform those adopters into users familiar with the technology before it’s into production. That has really helped us take something that works in the lab, have advocates in the organization that are ready to use it, and smooth the transition from the lab into actual use. So I think a lot of people don’t do that. Or if you are doing that, you just it’s much more difficult to get a user acceptance of the model and the system is meets the needs. And it’s harder to get the system to meet the needs because you haven’t heard direct from the users. So engaging your users in your testing, getting their feedback early on in the pilot phase has the two fold effect that your product gets better and you immediately have people ready to adopt it.

BH: You see value. It’s critical to really implementation. Then moving on to the implementation and looking at kind of common challenges that companies may face when implementing AI solutions in industrial environments. So how can they overcome the idea of, let’s say, implementing something that doesn’t really return any value? How do you avoid that? And maybe it’s the data, the quality of the data, but understanding what you want to do.

SK:  Yeah. And when we have often for innovations 4 phases, we, we start with an idea. So if someone comes up with an AI idea or non-AI idea very much at that point it’s a good idea. We listen to the people and we say and we record that idea we do then say okay, compared to everything that we’re doing, do we actually have time to assign resources and think about it because we can’t just jump around. So once we have time, we do a deeper thought analysis, try to get down to the root cause of what’s the problem? The person’s coming up with, how can we solve the root cause problem? That’s the second phase of doing that analysis. Then it is the pilot phase. Could be one pilot, could be multiple pilots, and each pilot is designed to eliminate risk. To learn more about the performance of the system and learn the capabilities. So the very first pilot of an AI project could just be a data validation. Just doing data analysis, looking at trends in the data could be just looking at images. It could just be looking literally at documentation and doing searching of documentation for accuracy. Second pilot could be some very rudimentary models to see. Is there a performance or predict element? And you could run multiple pilots to optimize or improve your models. And from that, then you start to build up the infrastructure around deliver a solution to the customer. So that’s the third element where you’re more into a prototype phase than a pilot. Finally, it always ends up with the handoff where you’re trying to smooth that transition into production. One of the areas that gets you into pilot purgatory is very much I want to apply AI, and I am looking for a problem to apply AI. I don’t think a company needs an AI PhD person, but a company does need the ability to evaluate. Is this a real problem and is AI an appropriate solution for the problem? And maybe the third element I just need to add to it is this attempting to boil the ocean, or is this a solvable problem that I can go drive it? So ATS’s path We did research AI for two years and without even attempting an implementation, but it was a rapid prototype fail fast, learn more, learn to see if it can move forward or not. And it was more like an intentional. This is a pilot to not drive to production. This is a pilot to learn and to understand the limits of it. From there learning, you understand the ability and the timelines to drive a pilot how long a pilot could take. So once we got to the point that we said, hey, we think AI is a real tool, we think we can predict a duration of a pilot accurately, and we see a pathway to deployment. So those once those elements were in place, the pathway to deployment became more predictable. And it enabled us to go forward. Another element, it is a large group that needs to get engaged for a deployment into a production environment. You need quality buy-in, you need these groups that you’re not going to have people that have a true stake in the company and a true stake in the process to not understand and not be able to accept the technology. So that communication upfront, when you start a project indicating the goals of the project, indicating the process that’s going to happen, to move through the pilot phase and into the production phase is a very important communication phase. So once you’re finished your pilot, you don’t surprise people. They understand the scope, they understand the risks. And you can then make an intelligent decision at the completion of your pilot to move forward or not, rather than that educational decision, that educational transfer to say, hey, here’s what we intend to do. It’s more of a risk based analysis at that point for evaluating the pilot results.

BH: Interesting. That kind of leads into the next question in what role does culture play in adopting AI across teams and organizations? And I think, are we seeing a mind shift in people? I think when we first started talking about AI, people were saying, oh AI is going to steal everybody’s jobs. And I think going through that learning phase, understanding the technology, the way that we approach innovation here at ATS, I think. How does culture and the culture of innovation impact kind of the implementation of AI in an organization?

SK: An organization that is that is change resistant AI is often a multiple change process, change in data, change in algorithms, change in work processes. So definitely all the normal change control required in a organization is required. One of the neat things about AI specific cultural changes that I’ve personally observed with the uptick of open AI and with the uptick of ChatGPT, and that usage of people using and implementing AI on their own time at home, it’s not 100% of the people. But suddenly you’re change leaders in your organizations and you’re you people looking for new technology have experienced AI on their own in a safe environment with non-critical quality data. And I’ve seen the results. The benefits of it. And the benefits, the risks, the failures of it. I know everyone I’ve talked to has asked ChatGPT a question that is more of a humorous answer than an actual answer and some validation of the answer required. So that’s a true risk, a hallucination. So that understanding that familiarity has really driven a different change control process within ATS. Because you’re more explaining the risks specific to the model and how it compares to maybe some known AI applications, as opposed to explaining the entire risk of AI to the organization.

BH: I think it’s exciting because for me personally, like AI using ChatGPT saves you a lot of time doing things it can help you kind of correlate your notes together and all of that. So I’m excited to see where it goes into the future and how it can really be applied across multiple different industries. But I’m curious, what do you think are the biggest opportunities for AI in industrial automation over the next 3 to 5 years?

SK: There is a lot of labor and a lot of costs associated with automation and in production, and a lot of the work is very repetitive and almost predictive in nature. So there’s the the new inventing can also benefit, but AI has the opportunity to come alongside workers. It has the opportunity to then alleviate a lot of the base work and starting from scratch on a lot of things, and providing insights and capabilities to that person that previously were unattainable. So I see it being an enabler for employees to accomplish more. An enabler for organizations and employees to leverage a lot of information that’s currently available in systems, but just not practical to go find and leverage on day to day tasks. And AI makes a leveraging of that existing information very possible. So I see a lot of productivity improvements throughout design, throughout procurement, throughout the tasks that really take a lot of labor and I see people leveraging that and becoming more proficient at their output. And a lot of that well, we came back and when we started, we talked about how operators online often take an extremely long period of time to become familiar, become proficient on the equipment AI drives the opportunity to leverage that previous history, provide more direct direction to the operators of the equipment, and enables equipment to run better. That those kind of maybe a series of wins of productivity gains will continue to accumulate over the next 3 to 5 years. I also see a lot of vision and these other applications that that’s just a way to gather data. But I slowly see then more data collections coming into systems and more ways than to drive unique models that we aren’t deploying today.

BH: That’s really cool. Thank you for sharing your insights, Stan. Is there any kind of closing statements or kind of message that messages that you want to.

SK: In closing, I’d like to say a few times. One of the things I’d like to say is I’m pleasantly surprised sometimes on the output of AI. So whether it be a generative answer that we’ve built up based off of a library of knowledge, or whether it be a prediction of temperature on a SuperTrak, sometimes you look at it and you’d say, wow, I, you know, just to be a bit humble on the models, I’m pleasantly surprised how that model prediction aligns with reality. So the AI always validate the answer. Always be sure to validate the data. I wouldn’t trust a model on the first run, and even after it’s been running for a while, I wouldn’t trust it without some testing and validation. But also the surprises on how it performed has been exciting and new and- impressive? Impressive.

BH: Okay, well, I think it’s clear that AI just isn’t a trend. It’s a powerful tool that, when used wisely, can unlock new levels of performance, efficiency and innovation in automation. And I think that’s exciting. So thank you to all our listeners for joining us on Enabling Automation. If today’s discussion sparked ideas or questions, we’d love to hear from you. Be sure to follow the podcast to catch future episodes where we explore more breakthroughs shaping the automation world. Until next time, keep innovating, stay curious, and thanks for tuning in.