2983: RoviSys - How AI is Transforming Manufacturing

[00:00:01] Have you ever wondered how the manufacturing industry is transforming with the advent of digital innovation? Well today I want to learn more about the future of manufacturing with a visionary leader who is at the forefront

[00:00:15] of this revolution. His name is Brian Deboy and he is the Director of Industrial AI at RoviSys. They're a company renowned for integrating AI solutions into the world of manufacturing. So in our conversation today I want to explore how the market is evolving, what steps industry leaders

[00:00:36] should take to adapt to digital transformation and yes AI, and from starting a digital transformation journey to unlocking new possibilities with AI, I'm going to ask Brian to share his insights for any medium to large manufacturing company. Reaching listeners in a hundred and sixty-five countries

[00:00:54] every day is testament to the unwavering support of you my listeners and our sponsors without whom this podcast just simply wouldn't be possible. And it also gives me a chance to talk about the fact that legacy DRM failed to securely

[00:01:07] enable external collaboration, especially on sensitive files, and how every organisation faces this risk trust contradiction where they can share content with untrusted third parties yet expected to protect that data. So it's time for something more modern, a DRM solution that solves that dilemma

[00:01:25] without compromising security or productivity. And you can do all that with a company called Kiteworks that will enable you to say goodbye to deployment headaches, file transfer risks, collaboration barriers and productivity constraints so you can experience a more modern way to collaborate on sensitive

[00:01:43] content without sacrificing control or security. Please visit kiteworks.com to get started today. That's kiteworks.com to get started today. Now is the moment you've really been waiting for. It's time to get today's guest on. So

[00:01:58] buckle up and hold on tight as I beam your ears all the way to Wyoming where Brian is waiting to join us today and we're going to talk about all this and uncover the exciting advancements and opportunities in manufacturing so if

[00:02:12] you're ready to embrace the future of technology and manufacturing you're gonna love this one. So a massive warm welcome to the show. Can you tell everyone listening a little about who you are and what you do? Yeah great to be here Neil.

[00:02:28] So I am Brian Dubois. I am the director of industrial AI for a company called Rovisys. And Rovisys, what we do is we are a system integrator and we can talk a little bit more about that in a minute and kind of what that means but

[00:02:43] we really work primarily with manufacturing and industrial customers and we've been around since 1989. And when you look at the world of manufacturing there is typically a lot of very legacy equipment, very old equipment. This equipment often time these assets are a couple

[00:03:04] decades old at least and so in that world you have these companies called system integrators and so they're going to take all of that equipment, they're going to take all of the different disparate technology

[00:03:19] from a lot of different vendors and as a system integrator we're going to put that together into a solution that solves a problem for that manufacturer. So that's kind of what Rovisys does. I've been with Rovisys for 24 years now and five

[00:03:33] years ago I was fortunate enough to be named director of this new division industrial AI and really what we were doing with that is taking what we've always done which is bring solutions to manufacturers but now extend it into

[00:03:47] that artificial intelligence space. And I will be talking I'm sure a lot about that today in this episode but kind of at high level it involves both AI ML as you would imagine but then there's a lot that we do in my division around data

[00:04:04] infrastructure as well because as you know you know these AI algorithms are hungry for data so we have to do a lot of kind of data infrastructure type of work to get to that AI future that all of our customers are trying to get to.

[00:04:19] That's one of the reasons I invited you on the podcast I always try and get people thinking differently about technologies like AI the kind of areas that it's transforming at the moment because it's very often areas you don't

[00:04:29] associate with technology and before you we started recording today you were telling me in Wyoming giving a digital transformation talk there but that's right largely the mining industry so I'm curious in what ways have you seen AI redefine digital transformation specifically within that manufacturing

[00:04:47] industry that we're talking about here? Yeah you know it's it's been really interesting and I'm not going to tell you that it's it's widespread adoption by any means yet right so but I will say that every company that I talked to is

[00:05:02] interested is either dipping their toe in today or is you know is considering it you know for the for the near future. The impact that it's having though because and we'll talk a little bit more about some of the specific types of AI

[00:05:18] that Rovasis is working with but the impact that we're having is we're now able to move some of these metrics that you know that that have really not moved for a long time so some of these processes that we're talking about if

[00:05:34] you look at the glass industry if you look at the steel industry the aluminum industry some of these processes were developed a hundred years ago and have not really you know changed that much right so these are well understood well

[00:05:47] optimized processes and so the only way that that these folks are going to get you know single digit and even double digit percent improvements in these processes is to adopt AI. To me that's really in the next step and when we look

[00:06:05] at I'm sure you've probably heard a lot that we are in the fourth industrial revolution right and so when you look at that when we look at all of the other industrial revolutions you don't know until you look backwards what was that

[00:06:19] revolution all about? I really firmly believe that at the end of this fourth industrial revolution we're going to look back and realize it was about getting to AI. So like I mentioned I've been at this for 24 years so if I go

[00:06:32] back to the early years of my career we were doing a lot to just collect data so we were implementing time series databases which in our world are called historians we were implementing these historians on the planet floor to

[00:06:45] collect all that data and a lot of people were saying okay we're gonna collect all this data but what are we gonna do with it? Well now we know what we why we were collecting all this data for all that time we were collecting

[00:06:56] it to be able to feed these industrial AI algorithms. And you mentioned there that we're not a mainstream adoption levels yet in the world of manufacturing so what would you say the current state of AI adoption in the industry is and

[00:07:12] how is the sector experiencing this gradual shift towards AI? Well and every every company is experiencing a little differently right so you've got those companies that are very forward-thinking now you know those tend to be the companies that have always kind of focused on continuous improvement and

[00:07:29] have always had continuous improvement dollars available and so those companies tend to be the ones who are the most forward-thinking and they may already have you know I'll go in there and they may already have a data science team or

[00:07:39] something like that that will work alongside of. And then you've got companies I've got a steel customer right now where the CEO has said nope no AI we're not doing it it's it's not that's not gonna happen here right so okay I

[00:07:54] mean that's fine I guess but what we're doing then is this we're doing a lot of data readiness with those type of customers and what's interesting is is that you know that particular company that steel company has thousands of

[00:08:08] employees around the world most of those employees work at a site and they are interested in getting to an AI future and so in some ways they kind of know that it's just a matter of waiting out the CEO and so we're doing projects so

[00:08:23] that when you know maybe new leadership is in there that has that's a little more forward-thinking they are ready to start to move on that. And so it's the whole spectrum of all of those different types of things I think that a big

[00:08:37] challenge though and so you mentioned yeah I'm out in Wyoming this week doing a digital transformation workshop for some mining customers and I think one of the things you know that I'll spend a lot of time on today with them is the

[00:08:49] education aspect of it because most of these customers don't even really know what the state of the art is what is industrial AI capable of today they and it's funny because they oftentimes either way underestimate what we can do

[00:09:03] with AI or way overestimate what we can do with AI that they they either thing you think it's it's well it's not much different than some of the other technologies that we've already had and to some extent you know but it does go

[00:09:16] further than that or they think we're at the Skynet level and and I'm like okay well no we're definitely not there yet so part of it is just the education aspect of it to get them to understand and then once I can get them

[00:09:30] understanding what the state of the art is and what it's capable of now I can start to get them thinking about how we can apply that to the specific problems you know in their industry. And they do say we often overestimate what we can

[00:09:47] achieve in that short term but often underestimate what we can achieve in the the long term there. Yeah right. What would you say the various types of AI being utilized in manufacturing sector because you said there's so many

[00:10:01] different areas and there's no such things are one size fits all so what are you saying that what different types of AI are being used in the sector? Yeah so generally when I talk about this there's three primary categories of AI

[00:10:15] or machine learning really because machine learning is what underlies all this. There's three primary categories of machine learning that we can put all of industrial AI that is in use today in manufacturing we can put into those

[00:10:28] three buckets then there's a fourth but that we'll talk about here in a second the three are the first category is what's called unsupervised learning and so that is in our world in the industrial world we typically describe

[00:10:40] that as anomaly detection and so what that is is you can hook up a model to the process. It does not require any previous training it doesn't require any a priori knowledge you just hook it up it starts to monitor the

[00:10:58] process and it can tell you when things start to go awry. Now it can't importantly it cannot tell you why things are going weird all it can say is look I've been watching this process for the last week I've got a pretty

[00:11:08] good feel for what normal looks like and I can tell you right now things are starting to go abnormal. Okay that's unsupervised learning. The second category is what's called supervised learning and that encompasses most of

[00:11:19] what we think of in terms of machine learning today and that's where you take large volumes of very clean very correlated data and you you send it into the ML model and you are trying to train it to be able to predict a single value

[00:11:36] and that's all it can ever do is predict a single value. So anytime you hear anything described as predictive predictive maintenance predictive quality that's supervised learning we're going to take large volumes of data

[00:11:46] we're gonna train it to predict a value that value could be in the case of predictive maintenance how many days until this piece of equipment is gonna go down. In the case of predictive quality what's the final quality of this batch

[00:11:57] gonna be before I even finish the batch predict for me what the final quality of this batch is going to be. Now there's an important assumption built into supervised learning and that's that somebody knows what to do with that

[00:12:10] prediction right so somebody can take action based on that prediction to fix things. So in the case of predictive maintenance if I can predict for you how many days until this piece of equipment is gonna go down does somebody know what

[00:12:24] to do to fix that with that information. If I can predict for you what the final quality of this batch is gonna be does somebody an operator a supervisor do they know how to fix that batch then to get it back on spec let's say I predict

[00:12:37] that it's it's gonna go out of spec that that batch we're gonna have to derate it or it's not gonna be a good batch does somebody know how to fix that problem. So that's kind of an assumption built into supervised learning. The third type and

[00:12:49] this is the type that is unique to Rovisys there's only a handful of system integrators in the world who can do this is called autonomous AI and this is where we actually train a neural network and it does what frankly a lot of people

[00:13:08] think AI can do and that's that it takes an action. So it actually knows what's the next best action to take and it can tell you so we can look at the state of the system and it can say based on everything I'm seeing here's your next

[00:13:22] best move and that's very unique in this space it's relatively new but not that new we've got customers running it in production today and typically that that autonomous AI is deployed in what's called decision support mode first so

[00:13:39] that's where it's making a recommendation but a human operator is taking that action based on its recommendation now you can take it from there and put it in to what's called direct control that's where it's actually sending that set point down to the control system itself without a

[00:13:53] human in the loop but we always start with decision support because we want to make sure that all the humans involved trust that it's making the right decisions those are the three big categories of industrial AI today. The

[00:14:07] fourth category and what's coming is generative AI and so we can talk a little bit more about that if you'd like to but that's the stuff that's coming but is not quite there yet for the manufacturing world. And when we've got

[00:14:21] all these different categories how do you distinguish between AI for production processes and I offer other parts of the business? Yeah so typically the way I look at it is you know a manufacturing industrial customer they have kind of

[00:14:41] when we look at the problems you're trying to solve there's two kind of big categories of problems what I would call operational problems and knowledge problems. So on the plant floor we were primarily interested in in operational

[00:14:55] problems and solving those types of operational problems and we can do that with any of the three types of AI that I talked about we can solve those operational problems. Where so generative AI and when I say that for your

[00:15:08] listeners that's the chat GPT those are the large language models the LLMs that's that's that whole category of AI is called generative AI. That is really what I am looking for in the future to solve the knowledge types of problems

[00:15:22] that you find at these companies. Those are really how I kind of split those two and in the manufacturing world like I said I don't feel like we're ready to start to adopt the generative AI yet. And for any business leaders listening in

[00:15:40] the world of manufacturing they're hearing about artificial intelligence in just about everything they know they need to be doing something but where should manufacturing companies be looking for for those ideal places to begin that journey and start adopting AI. They don't want to just shoot for the

[00:15:56] moon and stray away. Where should they begin? Right yeah it's a good question I think that where we typically particularly around autonomous AI because of its ability to make human-like decisions and actually build long-term strategy and then follow that strategy over time. We are really

[00:16:16] recommending to customers that they look at where where they have places in the process right now that are too complicated for traditional control systems or traditional optimization techniques and so they've had to add

[00:16:30] humans into the mix. Those are the types of places there's a human in the loop in the decision-making loop those are great places to look to adopt autonomous AI so for instance if you've got an enemy it can be anything from hey this

[00:16:46] part of the process is so finicky and it's so difficult to control that we always have to have a human there that's a great place to look at training autonomous AI to be able to do that. Or even at things as varied as

[00:17:01] like scheduling problems on this part of the process you know we always have to have a human production scheduler because it's too complicated we can actually train autonomous AI to be able to take the place of that human

[00:17:12] scheduler or at least augment maybe a less experienced human scheduler in that in that process. The other place that we oftentimes will look at adopting AI is in what's called unknown starting and unknown system conditions so unknown

[00:17:31] starting conditions are things like so for instance Microsoft has a use case that they used autonomous AI and it's public so I can talk about it. They trained it for PepsiCo to make Cheetos and the moisture content of the corn

[00:17:48] when you're making Cheetos can vary significantly and so what they do is they just start popping Cheetos and then they've got experienced operators that will sit there and they'll make adjustments to the knobs until they're

[00:17:58] making good Cheetos but then as you know they dump the next batch of corn in and the moisture content can be different and so then they maybe start making out-of-spec Cheetos and so then the the operators have to make adjustments and

[00:18:10] get back to making good product. That's unknown starting conditions. We can train an autonomous AI we call it a brain a neural network trained this way as a brain we can train an autonomous AI brain to actually be able to get to

[00:18:22] making good product faster even in the face of unknown starting conditions. The other one is unknown system conditions so this happens in processes like polymer reactors where you have to have a highly trained typically seven to ten year

[00:18:36] expert operator because you can't measure exactly what's going on in the reaction directly so you have to use what are called proxy measures. So you have to have a really experienced operator that has built a mental model

[00:18:47] of what's going on in that reactor and can read the proxy measures and and be able to react to them to control that that reaction. We can train an autonomous AI brain to be able to read those proxy measures like a human can and be able to

[00:19:03] control it at least as well as an expert operator. So all of those are great places to look for adopting AI particularly if you're trying to make you know million dollar a year plus ROI types of improvements those are great

[00:19:20] places to look. And again for that same business leader in manufacturing you could be listening to our conversation today anywhere in the world. Are there any stories that you might be able to share about somebody that's been on this

[00:19:32] road already they've got that AI success under their belt you must hear a lot of things so is there any you are able to share today? Oh for sure I've got a couple. The one is that I'll lead off with is so this was a glass bottle

[00:19:47] manufacturer and they had developed a brand new way of creating glass bottles and again this is a process that has not changed a lot in a hundred years and they've created a brand new way of making it and it will be cheaper it will

[00:20:00] use a significantly less amount of energy. The footprint of these glass bottle facilities is going to be significantly smaller so in every regard this is an improvement however there was a problem and that was that the

[00:20:16] process was very finicky so you had to have an expert operator who could really get dialed in and then you'd be making good bottles but then over time it would start to drift and you'd be making out-of-spec bottles and if that you know

[00:20:28] that expert operator can't be there 24-7 standing at the controls you know making adjustments if they go to lunch for an hour and they come back you're making bad bottles so that was a big problem. Well so we actually built an autonomous

[00:20:41] AI brain that could control that process as well as their expert operators and it's in production today and once this company gets to the point where they start right now they're just making bottles for themselves but the long-term

[00:20:56] goal is to actually market this and sell this as basically a glass bottle in a you know in a facility that you could buy and a bottle manufacturing facility you could buy and every one of those will ship with this autonomous AI brain

[00:21:11] to be able to control this process better. Another example is in so this was a sausage manufacturer and one of the challenges they have is of course sausage is a natural product and so the moisture the amount of moisture in

[00:21:26] that meat coming in is gonna vary and they had real issues with controlling that moisture to the point where they would go and they would smoke those sausages and it could smoke for anywhere from six hours to 14 hours trying to get

[00:21:40] down to the moisture content that they need. That's a big swing and that has significant impact on things like they don't run 24-7 so you know you could run into a situation where that shift has to leave and so you've got to kind of

[00:21:53] leave it for overnight I mean there's a lot of issues with with having that kind of variability we actually don't see that type of variability very often in our world you know four hours versus you know or six hours versus 14 hour you

[00:22:05] know those types of swings and so what we were able to do is we were able to build an autonomous AI brain that actually monitors I call it a moisture czar. It monitors everywhere in the line where we're adding moisture to that and

[00:22:17] dials in very specifically dials in at all those places to try to make sure that we're not adding more moisture than is absolutely necessary based on the amount of moisture that came in through the meat product. Now

[00:22:32] again that's something where a human could have done that but there's no human whose job it is to be the moisture czar. We built a brain to basically be able to do that and so I mean that's just too we've got lots and lots of success

[00:22:45] stories that we can talk about but in a lot of ways what I what I like to say is that we're really solving the unsolvable problems. We're solving problems that have never been attempted by any of these companies to try to solve before.

[00:22:59] And there are two big trends at the moment I think if you look at every business one is AI that we've talked about today and the other is sustainability and the responsibilities around that we're seeing things like ESG

[00:23:11] scores etc but of course AI does have a well-documented energy problem so how does sustainability and AI go together how do you see this relationship evolving or are there any synergies there? I think there are I think that when

[00:23:27] you look at sustainability you know most of these projects if not all of these projects that we talk about trying to solve with AI end up having an impact on sustainability. When you've got when you're you know that glass bottle

[00:23:44] manufacturer right so if I can get dialed in with with AI and they can make better on spec bottles and so because all the off spec bottles that they make all of that gets remelted all of that is energy lost all of that is energy wasted

[00:24:00] right so if I can make if I can get to making on spec bottles faster and sustain that on spec bottles now I'm not wasting all that energy on remelting that glass. We had another project where a customer makes flat glass and so you

[00:24:18] cut out different window shapes out of this flat glass and one of the challenges is that the glass may have minor defects it may have inclusions and you have to basically plan the cuts around those inclusions so that they don't end up in

[00:24:31] the customers window and they had built a very rudimentary optimization algorithm to lay out how to cut the glass. Well we built an AI optimized algorithm to so that we could vary as close as possible we could squeeze as much as many window

[00:24:49] cuts out of that sheet of glass as possible while still missing any of the inclusions and things like that in the glass. I mean they were saying that they were telling us anecdotally that we probably saved them hundreds of

[00:25:04] thousands of dollars a week in energy costs in terms of not having to remelt all that extra glass and that's great in terms of the bottom line but when you look at it from a sustainability standpoint that's a significant

[00:25:17] improvement in the amount of energy wasted in the past versus adopting AI. So I think that one of the things that that folks on the plant floor don't really we oftentimes don't think because I work with these folks you know one-on-one and

[00:25:32] and they don't think in terms of sustainability they think in terms of you know energy costs and they think in terms of reducing waste and they think in terms of increasing throughput but all of those things have a significant

[00:25:45] and material impact on the sustainability of that company and so I think sometimes it's just a matter of kind of connecting those two worlds and realizing you know that we're making major improvements to sustainability through this adoption of AI.

[00:25:58] And at the end of every podcast I always try to leave my guests some actionable insights and tips so for anybody listening today maybe we've set off a few of those light bulb moments what steps can manufacturers take to ready

[00:26:12] themselves for AI driven changes any tips or advice there? Yeah I mean so this is typically how I wrap up my presentations right so today when I talk to these mining customers I'm gonna talk to them about even if your company is

[00:26:26] not ready to jump you know into AI with both feet there are things that you can do today so look at your infrastructure we are doing more infrastructure upgrade projects now Rovisys than we've ever done in our history because

[00:26:40] folks are recognizing that to get to an AI future they're going to have to make those investments so look at your infrastructure are there places where we need to improve networking are there places that we need to improve

[00:26:51] data collection then look at your OT data strategy your operational technology data strategy and look at all the places where you have gaps in your data collection and what can we do to improve that then look at now don't look

[00:27:06] at just one line or one site now look enterprise-wide and let's see what we can do to collect all that data correlate that data across all of those different sites and build an enterprise data warehouse that includes the plant

[00:27:19] floor data the OT data all of this I haven't even talked about AI yet right all of those are projects that you could tackle today even if your organization is not ready to take on AI and all of that will make you ready when it's finally time

[00:27:32] to start to dip your toe into that AI you know and start to take those first steps towards that AI journey what I typically say though in terms of a call to action is just don't wait right because here's the problem that steel

[00:27:46] manufacturer that I talked about while they're not taking on any AI projects right now their peers are and I know that because I'm working with them so while you may not be ready there's other people in your industry who are taking

[00:27:58] those first steps so don't wait take those first steps you don't have to swallow the elephant in one gulp you can walk before you run and the other thing that I always say and I know this sounds self-serving but I always say find an OT

[00:28:10] system integrator that you can trust who can help you navigate all the hype there's a lot of hype in this space there's a lot of vendors that are selling frankly kind of snake oil and so it helps to have an OT system

[00:28:21] integrator that you can trust who can help you with you know we're independent we don't have any skin in the game I don't care what vendor you end up choosing all I want to make sure is that you've got an AI solution that's going

[00:28:32] to solve the problem for you so fine even if it's not rove assist find an OT system integrator that you can trust I think it's that important when you're when you're taking these steps on this journey fantastic so many great insights

[00:28:43] and I cannot thank you enough for coming on and sharing those with everyone listening around the world and before I let you go I'm gonna ask you to leave one final gift for everyone and that is a book that meaning something to you or

[00:28:55] that you'd recommend it I can add to our Amazon wish list and the books can and the listeners can check out but what would you like to leave everyone listening with and why yeah I mean I would say right now probably my

[00:29:07] favorite book that I've read in at least the last 10 years was a book by Andy so if you have heard of the book called the Martian and then there was a movie the Martian his newest book is called project Hail Mary if you like it's the

[00:29:20] Martian I I loved this book even more and if you happen to have audible I don't want to give away too much but the audible production in particular based on some of the things that happened in the story the audible production is perfect

[00:29:33] for that I think that's the best way until they make a movie I would supposedly they're going to but I think it's the best way to consume that book project Hail Mary definitely recommend that and this podcast it just cost me one

[00:29:45] audible credit using that for anyone listening wanting to find out more information about Roe vs all the work that you're doing how people or how you can help anybody listening where would you like to point everyone yeah so if

[00:30:00] you just go to Roe vs comm so that's ROV as in Victor I s y s comm slash AI that'll get you to our industrial AI section of our website or you can always look me up

[00:30:11] on LinkedIn my name is Brian BRYN Dubois so if you just search for that on LinkedIn you'll find me there and I'm always happy to connect with anyone there well again a real pleasure to have you join me today and providing that

[00:30:25] vision for a future of manufacturing that is dominated by AI and how the market is evolving how industry leaders should adapt and also how to start a digital transformation journey for any medium to large size manufacturing company and ultimately unlock those new possibilities with digital

[00:30:45] innovation in manufacturing your pleasure for my side but thank you for sharing your story with me today thanks Neil appreciate it Wow what an enlightening discussion on the future of manufacturing and indeed the role of AI in driving that transformation and as Brian shed light

[00:31:01] on various aspects of industrial AI from anonymity detection to autonomous decision-making and highlighting the importance of investing in data infrastructure partnering with trusted integrators there is so much valuable advice in that so as we conclude I think it's clear that the path to digital

[00:31:19] innovation holds immense potential for enhancing sustainability and efficiency in every manufacturing process but hey I'd love to hear your thoughts on the ideas Brian shared today and how do you envision the future of manufacturing and what steps are you considering to adapt to these changes well let's continue

[00:31:40] this conversation share your insights and questions with me by emailing me techblogwriteroutlook.com Twitter LinkedIn Instagram just at Neil C Hughes but until next time keep innovating keep pushing the boundaries of what's possible in manufacturing and I'll be back again same time same place

[00:31:57] tomorrow morning speak to you all then