Explore how simulation, digital twins, and AI are shaping smarter manufacturing with Andy Siprelle in this insightful PcVue Podcast episode.

Watch the Episode: Unlocking the Power of Data Simulation

In this engaging episode of the PCVue Podcast, Jim Huysentruyt sits down with Andy Siprelle, CEO of Chia Ha and Simulation Dynamics, Inc., to explore the evolving role of simulation technology in manufacturing. They dive into the challenges of data collection and readiness, the power of discrete rate simulation, and how AI-driven digital twins are transforming decision-making and operational efficiency.

Andy shares insights on leveraging data architecture for process optimization, how IoT and affordable sensors are revolutionizing supply chain management, and why making simulation tools more accessible is key to driving manufacturing innovation. The discussion also covers how companies can overcome barriers to digital transformation and implement predictive analytics for smarter operations.

Key topics covered in this episode:

• The impact of simulation technology on manufacturing efficiency
• How to improve data collection and overcome data readiness challenges
• The benefits of discrete rate simulation in optimizing production processes
• AI and digital twins for real-time decision-making
• The role of IoT, sensors, and line event data systems in manufacturing
• How predictive analytics is shaping Industry 4.0
• The importance of a data-driven culture for operational success

If you’re interested in industrial engineering, process optimization, or the future of AI in manufacturing, this episode is packed with valuable insights.

Other podcasts

1. Podcast – P1 | Mastering MES, SCADA & Change Management in Industry with Catherine Bouchard
2. Podcast – P2 | Smart Manufacturing, AI and the Future of Industrial Simulation with Andrew Siprelle

Transcript

Jim Huysentruyt (00:03.214) I’m Jim Heisertreit and welcome to our podcast series sponsored by PCV SCADA Solutions, Let’s Engineer. Today, I’d like to welcome CEO of Chia Ha and Simulation Dynamics, Inc. Andy Siprelle. Welcome, Andy. How are you doing today?

Andy Siprelle (00:20.577) I’m doing great. How are you doing, Jim?

Jim Huysentruyt (00:23.158) I’m doing real well. I have some questions here. We’ll get started. But really, the first thing I’d like to do is go ahead and tell us a little bit about Keeaha and simulations dynamics and kind of what you do there and a little bit about your journey.

Andy Siprelle (00:42.83) Sure, sure, Yeah, so I am CEO, founder of both Simulation Dynamics, which I founded some 30 plus years ago, 32 years ago. am Virginia born and raised, well, not born, I’m actually born in Tennessee here where I still live today. But in Virginia raised dangerously close to the swamp and I

Andy Siprelle (01:12.503) to two Virginia schools, two Virginia universities. University of Virginia first, and then Virginia Tech. it was Virginia Tech where I fell in love with simulation. I got a degree in industrial engineering and operations research, or IEOR. But yeah, Virginia.

Andy Siprelle (01:41.07) You know, dangerously close to the swamp, big, big family, big Catholic family, fifth of five kids, 30 cousins. And we all share a property out in Montana, you know, we’re, we’re, I’m near relation to the brothers featured in that show, that, that Fox business made during COVID called, American gold, the legend of Bear Gulch. So, so yeah.

Andy Siprelle (02:11.937) So actually transferred to Virginia Tech originally to be a mining engineer and I had family.

Jim Huysentruyt (02:12.142) Okay.

Andy Siprelle (02:20.045) Then I got down there and my Uncle Dick said, know, there’s really a bad choice. You you got to go somewhere foreign to make money.

Andy Siprelle (02:36.75) kidnapped. And so I discovered industrial engineering and fell in love with it. And you might say that I’m a hardcore IEOR junkie, it were. And I’m a bit of a coding junkie as well. sort of got into that early. so simulation dynamics, if you will, is 30 plus years later, you could, to sort of paraphrase Tom Waits, you know,

Andy Siprelle (03:07.025) I’m living proof you can make a career out of a college course.

Jim Huysentruyt (03:12.632) There

Jim Huysentruyt (03:12.882) you go. really when you think of operations research and maybe we could give a little bit of background, especially back then what operations research, because there’s a lot of, I guess, kind of taking your linear algebra and moving it, right? And putting constraints inside the quantitative analysis, right? That’s, you know, as far as that goes. you know, really how did you apply that, you know, to the real world?

Andy Siprelle (03:27.415) Sure.

Andy Siprelle (03:37.89) Well, that’s that was the thing while I was always, you know, growing up in Northern Virginia, but then being exposed to the real world in Montana and doing mining with my great uncles out there, you know, in junior high or in high school. I remember I spent a summer with them and they had me on a ledge, you know, with a rock drill, you know, and I was covered in rock dust, you know, and

Andy Siprelle (04:06.559) And they had retired and they sort of bought up this canyon and they were prospecting for gold. you know, growing up in Northern Virginia was very sort of unreal. My dad worked for the Bureau of Engraving and Printing. He was a mechanical engineer for them printing money, you know. And things didn’t really seem quite real, you know, in the Northern Virginia swamp.

Andy Siprelle (04:34.894) And so I yearned for real. I got offered a job working for Arthur Anderson. And I said, no way, no way. You can throw me into that. And I’d love it, doing all that stuff, computer type stuff. But I really wanted to see manufacturing. And so I took a job with Alcoa, all because they were talking about simulation and modeling.

Andy Siprelle (05:03.543) how they used it for capacity analysis. Alcoa always did a good job, you know, going way back to, you know, the thirties, right? Well, even before that, you know, they were, you know, if you remember your American history, they were part of like the early antitrust cases because they were too good at predicting when to add capacity, you know? And so I did.

Jim Huysentruyt (05:28.302) Mm-hmm. Yeah, so…

Andy Siprelle (05:31.053) I did capacity

Andy Siprelle (05:32.335) analysis for them and they even studied their customers. So we modeled them too.

Jim Huysentruyt (05:36.27) How are you collecting the data

Jim Huysentruyt (05:39.864) back then? Were you collecting it manually?

Andy Siprelle (05:42.626) yeah,

Andy Siprelle (05:43.306) yeah, yeah, back, way back before big data, you know, and so, so when I first started joining, when I first joined Alpola, that was the beginning of the personal personal computer revolution. They didn’t even have PCs and plants. And I was using the very first ones there, you know, and they were doing hardcore.

Andy Siprelle (06:12.919) computer, they were doing hardcore simulation models of their plants. And we had gigantic, you know, spreadsheet models that we did as well back in the day. Excel, Excel on the Mac, actually, the very first version of Excel was way more capable on the Mac. Yeah. So yeah.

Jim Huysentruyt (06:18.605) Okay.

Jim Huysentruyt (06:26.902) Okay, I’ll use the load it’s or something back then, right?

Jim Huysentruyt (06:33.036) Okay, there you go. Okay.

Jim Huysentruyt (06:37.966) I remember that.

Jim Huysentruyt (06:39.546) So now you’re getting this data as far as the technology. Maybe you could talk a little bit on how that technology evolved on just collecting the data and doing the computational analysis.

Andy Siprelle (06:51.086) Yeah, it’s very interesting. As I said, I was a computer. I got hooked on computers in school, taking a Pascal course at University of Virginia first. And then as we talked, I got into Turbo Pascal and later into Delphi and whatnot. But really, what

Andy Siprelle (07:20.141) What got me hooked was this general purpose computer simulation, general purpose discrete event simulation package that I discovered by reading an article of Byte Magazine sometime in 89 or something. getting into that, you know, it was just that became the basis of a whole tree.

Andy Siprelle (07:47.436) We trained people in computer simulation and building how to build models of their lines and multi-stage plants and then later entire supply chains. And grappling with data was always a problem.

Andy Siprelle (08:09.069) companies that were were pragmatic about this, you know, it was it was almost hard to believe in the early days, how hard it was to do this, you know, because as far as the data goes, oftentimes, people would would go through a massive static data collection, build this enormous model and then throw it away, you know, after a one time study. You know, you know,

Jim Huysentruyt (08:37.134) Oh, geez. And so it’s gone

Jim Huysentruyt (08:38.734) because they’re just doing it in spreadsheets. So it’s like, I don’t need this anymore. Okay.

Andy Siprelle (08:39.298) Gone. Yeah. Yeah, right.

Andy Siprelle (08:43.801) So we saw a lot of that. we worked with this general purpose package trying to doing some very audacious models. We built libraries on top of this general purpose package in a subset of C called Mod-L.

Andy Siprelle (09:07.167) And so we built our own libraries and use this mainly in the consumer products space, you know, food and beverages, some mining, but also for the DOE and the military and all kinds of folks. A turning point for us really in our whole journey was

Andy Siprelle (09:28.749) We all get so so we are we were dealing with lots and lots of data all the time, you know, and like I said, trying to turn this into a company asset was hard because for numerous challenges to this, you know, people would show up at our at our training having having received a model, you know, been been beneficiaries of a model, but they they were showing up at the training just because

Andy Siprelle (09:58.402) They couldn’t understand how to drive the car without knowing how to beat car mechanics, you know. Yeah.

Jim Huysentruyt (10:03.384) Well, yeah.

Jim Huysentruyt (10:04.975) And I think that was part of it, back in the dawn of when simulations to me and actually use. and where is all this data? Is it in spreadsheets? Is it in databases? So you got to know how to extract it.

Andy Siprelle (10:09.1) Yeah.

Andy Siprelle (10:14.849) Yeah, yeah.

Andy Siprelle (10:20.085) Yeah, and the data was getting ever bigger and bigger. we, sometime about 20 years ago, we just were doing projects that were just too big for this package. And so we burst out of the seams of it. And we got into, we ported some of our supply chain simulation libraries into .NET. so the first big job we did was, and this was really the turning point because this was the beginning of what I

Andy Siprelle (10:49.016) called model-based apps. And so this was a job for a joint venture of Chrysler and Union Pacific Railroad. And it was actually a transportation logistics job. And they had this massive system called Vinvision. So they managed the distribution of brand new vehicles, Chrysler’s from plants, from fab plants, out to dealers via rail.

Andy Siprelle (11:18.977) barge, know, Mad Max, know, retired couple truck trailer out of ramps. And so we built an entire, you know, model. a key to this was it was really important. was such a, such a, you know, it was a good use of the technology. because this VIN vision system they had already built was this

Andy Siprelle (11:48.522) real-time visibility system. You could type in

Jim Huysentruyt (11:51.31) and

Andy Siprelle (11:51.385) a VIN, you know, a vehicle identification number, and it would tell you where it was. And so what they did, they asked us to do, was to build this model in such a way that they could, that we could suck in the state of where it was at time equals zero, you know, suck in all the status of these vehicles.

Andy Siprelle (12:16.233) upwards of 150,000 vehicles to represent now. And then from now, now into the future, pump out continuous vehicles from fab plants based on schedules. And so we could give, you know, loading ramps and unloading ramps, you know, heads ups on what labor they would need. We could help them during, they used it, you know, to equilibrate, figure out how long it took to equilibrate.

Jim Huysentruyt (12:19.544) Mm-hmm. Mm-hmm.

Jim Huysentruyt (12:28.302) Okay. Okay.

Andy Siprelle (12:45.07) equilibrate the network during hurricanes. The CFO even used it to do the budget annually. So this was a turning point because we could leverage the big data very importantly, very effectively with .NET. But most importantly, we could provide a UI that allowed a business analyst to steer it.

Jim Huysentruyt (12:55.19) Right, So.

Andy Siprelle (13:14.369) And that was big. That was a totally different thing. That was not an IE. And so they used that model for a decade and it led to enormous savings.

Jim Huysentruyt (13:20.556) Right, right.

Jim Huysentruyt (13:25.006) Well, I can imagine, I can imagine

Jim Huysentruyt (13:26.766) taking it and, you know, anything from food processing, chemical processing, you could take that same model. so, you know, and having a UI where a business guy can do it. And, you know, I’m not saying, you know, I mean, I was just saying, you don’t have to be a computer guy anymore. Right. that, that’s right. And that’s, yeah. So, so with this model, cause I know you’ve worked in

Andy Siprelle (13:45.675) Right, you don’t have to be an IE and that’s who they had running it before.

Jim Huysentruyt (13:54.862) with chemical and other type of plants process and applying that to process. Cause now you’re collecting data in different ways, right? You’re getting data, you know, it’s, it’s a, where you’re getting, you know, you’re getting your throughputs, you’re getting real data, but whether you’re doing on packaging or you’re doing it on, you know, or maybe on the, from the batch on down, I don’t know, or from what you need to put into the batch, right?

Andy Siprelle (14:05.089) Yes.

Jim Huysentruyt (14:23.662) And, you know, as far as the data now, because we still, now we’re talking about this data and, know, how do you, when you go in to a system, you’re saying, okay, they’re saying, oh, we love your simulation model. We want to do all this stuff. Obviously the benefits are huge, right? I mean, if you could just, if you could just get the throughput up by 5 % by that much.

Andy Siprelle (14:23.969) Right, right.

Andy Siprelle (14:46.679) Yeah.

Jim Huysentruyt (14:52.238) You know, that’s millions and millions of dollars,

Andy Siprelle (14:55.083) Yes.

Jim Huysentruyt (14:55.2) but you’re sitting in a situation where they have old legacy systems and then they have newer systems and then they have whatever their ERP system is. How do you put it in a format that you could read?

Andy Siprelle (15:08.34) It’s

Andy Siprelle (15:11.297) Well, everybody’s different, you know, and everybody starts at a different point on this, right? It just so happens that who we’ve worked with, you know, has been what we would say suitable conditions for life to work in this space. You know, not everybody can, not everyone has the information architecture to pull up valid data from.

Andy Siprelle (15:40.064) every manufacturing line all over the world, right? But some do. And I can speak to the issue of data readiness. It’s been all over the map who we’ve worked with. When we went back in the old days, when we went into plants, they didn’t have line event data systems. And what this is is

Andy Siprelle (16:08.491) And they still don’t. Yeah, yeah, yeah, yeah. Yeah, what is a line event data system? Well, so first of all, let me answer your question by answering old school, this is the consulting practice. Say we’re building a model of a breakfast cereal.

Jim Huysentruyt (16:10.638) Okay, actually, I want to take a step back and maybe you can explain what is a line event data system.

Andy Siprelle (16:36.162) building, you know, we’re trying to understand the throughput and the availability, the capacity of the entire line. We would actually, you know, step through every piece of the production process and ask the question, what slows it down? You know, what constrains it? And in addition to that, things like, you know, what makes it stop, most importantly,

Andy Siprelle (17:04.717) Because as an old boss of mine said, know, Andy, you know, from Alcoa, you know, it’s not the little fine tuning of things that is really, you know, causing manufacturing facilities to go out of whack, you know, like fine, you know, control. It’s really the big crash pang booms. It’s like the things that we do on a casual basis, you know, shut down the 049 slitter in the North plant.

Jim Huysentruyt (17:23.436) Mm-hmm.

Andy Siprelle (17:34.647) you know, and move it to the West plant or vice versa. You know, so it’s the crash bang booms, right?

Jim Huysentruyt (17:35.15) Mm-hmm.

Jim Huysentruyt (17:37.762) Right. you also have, I

Jim Huysentruyt (17:42.975) mean, because when you have a line stop event, you have really changeovers. You have little things like if you’re replacing film or what have you. And there’s a lot of them. And getting those is,

Andy Siprelle (17:50.369) Yes. Yes. Yes.

Andy Siprelle (17:55.447) That’s right, yes.

Andy Siprelle (17:58.638) Yep. Yes, exactly.

Jim Huysentruyt (18:00.276) that’s an art.

Andy Siprelle (18:02.239) And so this is a critical source of us. I’ll call it raw material for prediction, if done right, a line event data system. So with a line event data system, we want to record through sensor or through a combination of sensor and operator.

Andy Siprelle (18:26.291) every reason that the line goes down. We want to know what it’s for, no matter if it’s a failure mode or it’s a planned stop. And if we have that data, we can make a lot of really, really good prediction. In fact, we can make predictions. can… Simulation is like the analytics of the future. You know, we can…

Andy Siprelle (18:55.918) If we have data on stops and it’s configured well, we’re tracking the right things, then we can not only validate systems of the baseline that exist today and figure out what improvements to make on that system, but we can also do analytics.

Andy Siprelle (19:24.041) on systems that haven’t even been built yet, but still using that equipment. You know what I mean? Just in different arrangements, serial, parallel, et cetera. Yeah.

Jim Huysentruyt (19:29.496) Right.

Jim Huysentruyt (19:35.31) So, yeah. So, or, or you could try to predict the change because you’re going to, you’re going to introduce a change, right? But you have all the stuff that’s, you know, all the basically data of, you know, of incoming and outgoing, right? As far as how fast you can go.

Andy Siprelle (19:44.876) Yes.

Andy Siprelle (19:53.166) The highest,

Andy Siprelle (19:55.247) that’s right, the highest fidelity behavior of the internal stops on a line is critical. Because it’s the external stops, the blocking and starving and the possibility of that that we actually simulate. we can take the data and create very, very high fidelity simulations.

Jim Huysentruyt (20:11.534) Thanks

Jim Huysentruyt (20:14.925) Mm-hmm.

Andy Siprelle (20:22.893) Now, some folks don’t have the data and certainly data readiness is one of the big, big challenges. And I can speak to that some more. We could talk about that all day long. But, you know, I think that a lot of customers in this space that we address know that 90 % OEE is possible, operational efficiency and effectiveness. They just don’t know how to get there.

Jim Huysentruyt (20:48.199) Right.

Andy Siprelle (20:51.549) And really it’s quite amazing that…

Andy Siprelle (20:56.288) In modeling, there’s been a lot of emphasis on the physical models, computational fluid dynamics and finite element design for packaging, all of those things that they get a lot of attention and the manufacturers court that kind of modeling because they know they need it. But what I see is that I talk to people even at big companies where

Jim Huysentruyt (21:15.982) you

Andy Siprelle (21:25.323) You know, I talked to a guy who works in a big pharmaceutical plant and he said, you know, our plant still operates like it does in the eighties, you know, so they’re, generally in this space operating in the sixties or lower. They know 90 is possible. They just don’t know how to get there.

Jim Huysentruyt (21:40.654) Okay. Right.

Jim Huysentruyt (21:44.097) Right. Well, I from the OE side is, I find going in and you’re talking OE and really they have a baseline where they’re at. really it’s, there’s so many in the thirties and the forties, I mean, just in that range. which is, so working with OE, it’s kind of nice to be in the first coming in because we can get some victims fast and we can get that.

Andy Siprelle (22:11.435) Yep, yep.

Jim Huysentruyt (22:12.462) That

Jim Huysentruyt (22:12.742) first 10, we can usually get pretty fast. But it’s when you’re flying to when you’re getting to those next to the next level where you can get to like, you can get to 70 7580. That’s, that’s world class what I see in most in most cases. So yeah, yeah.

Andy Siprelle (22:15.373) Yep.

Andy Siprelle (22:27.467) Indeed. Yep. But

Andy Siprelle (22:29.499) what we see is that, you know, when it comes to answering the strategic questions for supply chain throughput and reliability, you know, that’s not really happening much at all. You know, that’s a big challenge for manufacturers. You know, the companies selling equipment, they often use simulation models just as part of a selling exercise.

Jim Huysentruyt (22:54.562) Yeah, yeah, yeah. a lot of times I’ll put OE rates on them that aren’t achievable, but that’s another story. Because it’s a perfect situation, right?

Andy Siprelle (23:00.791) That’s right. Well, that’s part of the story. Yeah. A lot of times

Andy Siprelle (23:04.925) they, you know, their motivation is to sell equipment. And so, you know, they’ll quote you repair times, you know, that are just based on their technicians repair time, you know, like they’ll

Jim Huysentruyt (23:08.718) you

Jim Huysentruyt (23:16.566) Right, right. Yeah, right.

Jim Huysentruyt (23:18.739) they’re, they’re doing it all the time. That’s the thing, right? So that’s, yeah. Yeah. Right.

Andy Siprelle (23:20.757) Right, they’ll quote you 20 seconds when in reality in the plan it takes two or three minutes because your operator is not standing there with a toolbox all day like their technician. So that’ll

Andy Siprelle (23:33.795) make a huge difference.

Jim Huysentruyt (23:36.526) Okay. So tell us a little more about some of tricks that you’ve come up with. Because I know you have some things, especially with with GAAHA that you’re doing some things with discrete rate simulation and you know, you’re basically have a model of competing risk. Maybe you could tell us more about some of that.

Andy Siprelle (23:50.573) discrete rate.

Andy Siprelle (23:56.078) Yeah, it struck me about a year ago after 30 years of simulation dynamics being like an incubator for ideas. We ended up stewarding a matrix of innovation projects, I guess you’d say, and stewarding them from, I’ll say, zany idea.

Andy Siprelle (24:25.549) or some grad student’s project in Python or whatever, all the way up through potentially enterprise apps scaled on a worldwide scale. And then edge apps are something we plan as well. So I wanted to, a year ago, because of what we just discussed as the challenges and also the opportunities, you see all of this IoT and you

Andy Siprelle (24:55.566) IOT and you see the possibility of cheap, you know, sensors being used and all of the opportunities these days that manufacturers have and I thought, you know,

Andy Siprelle (25:10.829) that this little trick I invented, as you call it, some 30 years ago called discrete rate, really could play and help people in this field. So about 15 years ago, the trick I’m you’re alluding to is a first of all, a term of art of my coinage called discrete rate simulation. And it’s a variant of

Andy Siprelle (25:39.818) of discrete event simulation. That’s something that we all industrial engineers learned in school. some 15 years ago, 20 years ago, I developed a library on top of a general purpose package. And then I marketed it commercially as well in a joint venture with the vendor.

Andy Siprelle (26:09.479) And at the heart of it, it’s about representing production processes using rates as alternatives to items. And it’s very applicable to CPG and food and beverage processes. We even went to the one year we went, two years we went to the Powder and Bulk Solids Show conference. Now that’s an exciting title for you.

Jim Huysentruyt (26:19.171) Okay.

Jim Huysentruyt (26:36.845) Hahaha

Andy Siprelle (26:39.374) But the thing about discrete rate that was new is that it’s ridiculously easy to understand as a worldview. Lots of folks think in terms of rates anyhow, know, electrical engineers, chemical engineers. It’s only us, we weirdo, know, industrial engineers who talk in terms of items. So it’s ridiculously easy to model and it has

Andy Siprelle (27:08.437) It’s easy to teach folks. We’ve tried to teach folks discrete event simulations systems. I taught hundreds of classes back in the day. And our success rate for this stuff was very, very small. So we know. But another property about discrete rate is that like discrete event, and we’re talking about models that run in compressed time.

Andy Siprelle (27:38.444) not real time, okay? Because we need to run them in compressed time to make strategic decisions, right? So one of the properties of discrete event is that it compresses time so that the models run faster, hopefully, than real time, right? And discrete rate does this even more drastically. So it results in, and really, models that run so fast that you can do what I call

Jim Huysentruyt (27:53.134) Okay. Okay.

Andy Siprelle (28:08.151) practical true stochastic problem solving and even develop an intuition for how to improve the system through running simulations.

Jim Huysentruyt (28:13.674) Okay.

Jim Huysentruyt (28:20.79) and so just a little bit at what the data needs to be. mean, because you’re probably talking about a lot of data that you’re reading fast. Do you have a, do you take that data and put into a different database or?

Andy Siprelle (28:30.453) rate.

Andy Siprelle (28:33.718) Sometimes databases and table structures are important. And most importantly, yes. A big part of what this innovation is is that it’s what I call a composable simulator, not a simulation, a general purpose language.

Andy Siprelle (29:03.103) And so we can develop some very effective models that are what I call model-like data heavy. We’re used to working with people and teaching them general purpose simulation only to see them building these Picasso style things that, you know, they’re obviously trying to get to the moon, you know, by way of climbing one branch at a time. And so

Andy Siprelle (29:32.15) So part of this innovation I’ve come up with is nodes that you use to model and that have a very rich construct underneath machines or anything that’s likely to stop. And this construct is using a model of competing risk that was developed for this industry from back at 20 years ago in the National Lab.

Andy Siprelle (30:01.869) And so what we’re looking for, you know, to model, first of all, is rates, right? We want to know what the max rate of machines are. That’s important. And then we want to know when they stop. And that involves interrupts, which we discussed, right? So when you’re driving your car and you get a flat tire, these are

Jim Huysentruyt (30:15.118) Okay. Okay.

Andy Siprelle (30:30.443) discernible and actionable tuples of time to failure, time to repair. And so we need to somehow take the data on all the stops from something, hopefully, like a line event data system.

Jim Huysentruyt (30:36.11) Mm-hmm.

Jim Huysentruyt (30:51.47) Mm-hmm.

Andy Siprelle (30:53.334) And we need to coalesce all of those stops into interrupts. these interrupts essentially have a time to failure, a time to repair, a clock type. They run according to certain clocks, either the wall clock or runtime. And then they have a group cause name.

Andy Siprelle (31:21.677) That’s how they coalesce. So if you get a flat tire and you fix it with one of those fix-a-flat things versus you fix it with having to change the tire or say you got no spare, those are three different interrupts. And they are distinctly different interrupts because of their repair mode.

Jim Huysentruyt (31:38.478) Yeah, okay. Right, right, right.

Andy Siprelle (31:50.326) Now, on the actual event, if in their recording there’s some operator or manual entry, they can say anything like, know, change tire or used spare, right? And so our job is to clean map and clean and map those into coherent actionable interrupts.

Jim Huysentruyt (32:07.746) Right.

Jim Huysentruyt (32:12.989) Normalize the data to where, yeah, normalize the data. Right. Okay. Right.

Jim Huysentruyt (32:19.39) Okay. Okay. And, and really with everything you’re talking about now and yeah, well, sorry, using AI. Are you using AI yet? mean, with everything that’s

Andy Siprelle (32:31.693) AI is

Jim Huysentruyt (32:32.31) going on in that world.

Andy Siprelle (32:34.573) really interesting because, of course, digital twin plays heavily in AI. people are, I think,

Jim Huysentruyt (32:44.098) I’m going

Jim Huysentruyt (32:44.849) to take a step back. You mentioned Digital Twin. I want you to define Digital Twin for our audience because I think there’s different views of what Digital Twin really is.

Andy Siprelle (32:50.414) Digital twin.

Andy Siprelle (32:53.837) Yeah, yeah. Yeah, there are a lot of folks using that term these days. We were working with a fellow who just developed a real-time dashboard on Power BI, and he said, look at my digital twin. And of course, I think it’s a great term, OK? I’ll say that.

Jim Huysentruyt (33:09.304) the

Andy Siprelle (33:19.731) And I think that the term has had a good effect because it’s really rejuvenated and put a spotlight on my field simulation. But in addition, the term emphasizes a focus on the decisions instead of the method. And a customer of ours developed a saying, I’m not sure where he got it, but we quote it sort of

Andy Siprelle (33:49.214) mantra like, know, it’s, it’s the decisions. It’s a digital twin. The definition is this. It’s about decisions, past, present and future automated.

Jim Huysentruyt (34:04.334) There you go.

Andy Siprelle (34:07.437) But in my world, I see a lot of folks equating this with 3D animated movie looking simulations. That that’s what it’s about. It’s an overemphasis on the geographical or geometric or kinematic or physical details.

Andy Siprelle (34:36.627) of the process.

Andy Siprelle (34:43.128) To really leverage, I think, a digital twin for your problem, it’s got to not only look like it, your process, it can’t be skin deep, right? It’s got to behave like your process. I think that the connotation, the idea connotes or brings up in our mind, like, if we had a twin, what would we do with it, right? And so…

Andy Siprelle (35:08.833) In my mind, it’s really got to behave like your system and on a level that’s useful in a kind of a George Box way, in honor of him. All models are wrong. They’re only useful.

Andy Siprelle (35:28.811) And so I think I spent 30 plus years of my career really trying to not only make models more right, but most of all more useful. And so what people really struggle with still to this day is these questions that we answer. We answer strategic questions about your production line. How many stretch wrappers do you need? What speeds and…

Andy Siprelle (35:57.132) and reliability of the components do you need? How much? What are the relative rates of these different machines? How many buffers do you need and how big should they be? Should you run some parts of the line faster than others? And it’s all in the context of Gold Rat and theory of constraints and lean production and Toyota.

Andy Siprelle (36:26.221) It’s a dumb idea to have have idle assets, right? Is what what they say. But it’s the context of it, you know, and how it’s used in the law.

Andy Siprelle (36:39.725) So we have our own software that does that and helps people to do that to figure out those questions, to predict availability. Sorry.

Jim Huysentruyt (36:41.784) So.

Jim Huysentruyt (36:46.158) Right. And, and, and at the end of the day, right. And then be

Jim Huysentruyt (36:51.438) able to, sorry, and I jumped on you a little bit, but, at the end of the day, it really is, I mean, your success is, you know, increasing through foot, right?

Andy Siprelle (37:03.189) Yes. Yes.

Jim Huysentruyt (37:04.018) And you know, that is, you know, at end of the day is if you can increase that rate, that rate, that’s right. And you can increase that rate on a consistent basis.

Andy Siprelle (37:10.112) Yes.

Jim Huysentruyt (37:13.742) Right. And then our, to put it in terms of OE, you can get that OE rate up to where it is down, pop it up there. You know, the guys in the suit, they’re happy. Right. Cause that’s really. Cause you know, there’s, there’s the one and they’re the ones we have to make happy at the end of the day. And from like, from the skater world, our job is to get data in a format. mean, amongst other things, but one of our big jobs now is getting data, bringing it in.

Andy Siprelle (37:23.447) That’s right. That’s all they care about.

Jim Huysentruyt (37:42.19) doing whatever massaging we could do, pre-massaging we could do and putting it into a format to where you could use it. And of course we’re not it, we’re not it. Because there’s a lot of different places that you’re getting your data and of all these different technologies now. And really one of the questions I have for you now, because I’m looking at this evolution of what you’ve done from back.

Andy Siprelle (37:48.577) Yep.

Andy Siprelle (37:51.159) That’s right.

Jim Huysentruyt (38:11.566) back in the day, right? Back when you’re in school and then just getting out of school. so, to now, right? And you have, you know, it seems like the softwares are getting better. It’s right. But also with the technology, it’s being able to present it to, you know, the, I’m just going to call it the more average businessman, right? The guy on the, you know, the guy doing the financials, the guy

Andy Siprelle (38:17.857) Yep.

Andy Siprelle (38:20.735) An avalanche of data. An avalanche of data, yeah.

Jim Huysentruyt (38:41.994) Ordering, right? The guy doing the ordering could get data from that. as far as it goes, it goes to a bunch of different areas. And really, as far as it goes, look at it. How would you predict the field in five, 10 years with this technology? Where do you see it?

Andy Siprelle (38:46.475) Right.

Andy Siprelle (39:01.014) I think, gosh, well, see, I’ve been very, like I mentioned before, I’ve been very lucky to work in a space where there’s been a real emphasis placed on architecture for information and data readiness. Almost all of what we do, as I mentioned,

Jim Huysentruyt (39:24.462) Thanks.

Andy Siprelle (39:30.409) involves eventually building a mature model-based app. And so the company that the ecosystem in which I’ve worked primarily has built fantastic data systems for collecting this data right off the line. And

Andy Siprelle (39:54.048) Well, we’ve worked in, and you mentioned SCADA, and we’ve worked in all three layers except maybe control, I’d say. But we’ve been involved in wiring up lines and stuff like that. Interestingly, those models are real time versus compressed time, which is where we started.

Jim Huysentruyt (40:03.576) Okay.

Andy Siprelle (40:22.237) Even the same, even in these huge organizations that have built out this information architecture to do this stuff, it’s been incredibly difficult to build a pipeline of innovation, which is what we really do. In our ecosystem, we serve as an incubating service for zany ideas. And we also serve as an enterprise and edge level DevOps organization for mature smart apps.

Andy Siprelle (40:51.886) So we manage a matrix of these ideas along a product life cycle. And they always start with the consideration of models. for models, know, models have to work. And assessing the data readiness, you know, with often the goal to steward them into a production, edge or enterprise app involves a ton of time validating them. And we had

Andy Siprelle (41:20.397) You know, when this big data thing came on, we actually had a model ready for all this that was, you know, that big data was going to feed. we had an analytics person validating the data for four to five years.

Jim Huysentruyt (41:41.198) Wow. OK.

Andy Siprelle (41:42.358) Yeah.

Andy Siprelle (41:43.318) And so I predict a lot of churn for the next five. See, I think that the ecosystem I’ve been working in is about 30 years ahead of the industry of the rest of the CPG and foods and beverage firms. so, know, chat GPT, chat can certainly help, you know, because analysts spend about 80 % of their time in this space.

Andy Siprelle (42:11.053) cleaning and collecting and all that. And so, yeah, that’s some of what I had to say about that. mean, there are certainly a ton of things that manufacturers can leverage around cloud computing and high performance computing. There’s lots of techniques from practical AI that we can…

Andy Siprelle (42:40.046) use and are using to do the data cleaning thing around what I was just talking about.

Andy Siprelle (42:48.738) But you have very useful things like Decision Trees and Monte Carlo Tree Search and Neural Nets. I think there will be a recognition more and more of what higher level things can be answered through more robust and more advanced models. And I think it’ll become more and more evident to the person on the shop floor. think…

Andy Siprelle (43:12.885) I think in some ways, you know, chat offers the possibility to radicalize the user interface for these models. Given that, and so I think, I think too, that a cultural shift is coming and has to come because of what’s required for innovation. I run a seven ring circus in my business and my teams are

Andy Siprelle (43:42.731) work on totally different planes. And the things that we do require some real risk taking. So that’s going to have to become accepted and nurtured within organizations.

Jim Huysentruyt (43:45.678) You

Jim Huysentruyt (43:59.276) But

Jim Huysentruyt (43:59.546) I mean, you do essentially mitigate the risk in a sense, right? I mean, can you kind of front load the risk a little bit so you don’t, because obviously four or five years is a long time, right? It’s a long time. yeah. Yeah. Yeah. And that’s just one of those, I mean, to get a CEO and get the, you know, get the, you know, you know.

Andy Siprelle (44:05.599) Sure.

Andy Siprelle (44:13.431) That’s a long time. That’s a lot of churn, five years of churn.

Jim Huysentruyt (44:29.55) you know, that C group to, to buy into that, you know, because that’s not an easy thing to do because it’s, it’s money. Right. And then, you know, and a lot of times if there’s looking two, three years out, that’s, that’s a long time for, I’m going say, especially the medium, because the bigger companies, they have, you know, you know, they, you know, they have departments are thinking that right. And they go, go that while they

Andy Siprelle (44:34.754) Right.

Andy Siprelle (44:40.267) Right.

Jim Huysentruyt (44:58.296) continue to do their business. But how do we take what you’re doing, right? And get it to the medium sized company or the small companies, you know, as far as,

Andy Siprelle (45:10.391) That’s what we’re trying to do. We’re trying to show them what they can do with just a little bit of data. What’s ridiculous is that even the simplest of questions that are possible to address using simulation are not accessible to those sized companies. And I just think that’s ridiculous, and that needs to change.

Jim Huysentruyt (45:19.308) Right.

Andy Siprelle (45:39.574) I mean, even if you, I mean, we found in one instance, you know, you don’t need to start with a ton of data. mean, for instance, you do need to know simple things like, you know, if you’re making French toast sticks, you know, how many sticks go in a box? And, you know, for this view, you know, how many boxes go in a case? And, you know, how many cases go into a pallet? That kind of data,

Jim Huysentruyt (46:06.542) Mm-hmm.

Andy Siprelle (46:08.749) on physics and is very important to understanding. Those are production physics we need to know and are accessible, right? It gets harder when they try to quote their reliability. know, back in the day, we simulationists would go into a plant and we’d say stuff like, you know, well, how often does it stop? And what’s your percentage?

Andy Siprelle (46:37.385) uptime, you know, and also when you repair it, you know, give us a triangular distribution for it, you know, a minimum, a mode, and a max. know, nowadays people have the possibility of a lot better information.

Jim Huysentruyt (46:46.69) Yeah, right. Right,

Jim Huysentruyt (46:57.11) Yeah. And there are systems out there like for OEE, you know, there are systems out there where you can put something in there and you can get that data. and because that’s part of the culture, think a big part of it is, is like you talked about the whole culture on the, for the whole big piece, but really just on the flat, on the plant floor, getting them to collect data, know your OEE. And when you know your OEE, not just your stoppage, know your rates. Why is it running slow? Right. Versus what, you know, cause you’ll watch

Andy Siprelle (47:00.054) Yes.

Andy Siprelle (47:07.629) That’s right.

Jim Huysentruyt (47:26.88) OE, sometimes the rates will do this throughout the day. it’s not a stoppage, And knowing why it’s giving that, right? Getting that information.

Andy Siprelle (47:29.419) Right. Yeah. Yeah.

Andy Siprelle (47:35.96) That’s right.

Andy Siprelle (47:37.911) When we, I mean, it’s really an interesting thing, this whole thing about relative rates. I taught engineers simulation for many years and a big part of my teaching, I began to use this kind of sound effects of like slurping, you know, and I’d say, okay, you when you’re packing Folgers and big red can,

Andy Siprelle (48:03.915) you know, the packaging line has a higher slurp rate, you know, because it’s still pounding out at the same, you know, cycle. now, you know, that’s causing relative rates, you know, in the buffer to be sucked down faster. And so therefore, you know, we’re going to empty bins faster. It’s really important to show people these dynamics.

Jim Huysentruyt (48:08.43) It’s true.

Andy Siprelle (48:33.134) Throughout the 2000s, we also got involved in multi-stage packaging, multi-stage scheduling problems. And those are some of the most difficult problems to solve in the world. In a pet food plant, for instance, because of this relative rates thing and the scheduling and the amount of bins that are in place, it’s very much like a game of free sell. Have you ever played free sell?

Andy Siprelle (49:01.675) that card game where you’re where you can cheat. You’ve got four places you can kind of lay down cards and it’s like solitaire, but you can cheat. And you have these four like buffers in essence, and those buffers are more or less like analogous to interprocess buffer storage availability in a manufacturing line. And so this pet food plant

Jim Huysentruyt (49:09.475) Okay.

Jim Huysentruyt (49:23.182) All right.

Andy Siprelle (49:28.075) You know, everybody knows that, you know, in free sell to get better at it and when, you know, you’ve got to really be judicious about that space. You know, you can’t give it up easily, right? Because when you do, you give up all those free sells and you get stuck, you’re done. Right. And so the same thing happened in this pet food plant. And, you know, they get get all their bins loaded with all the wrong product in the day.

Andy Siprelle (49:55.276) And throughout the day, anytime throughout the day, they get it loaded up. And then it’s bin lock is where all of the stuff that’s sitting in bins can’t progress downstream because the schedule’s not ready to take it. And so this was a condition that, you know, they get three or four times a day, they call up the scheduler in the middle of the night or whatever. And this is a problem, I think that

Jim Huysentruyt (50:05.324) wrong stuff. Right.

Andy Siprelle (50:26.245) is largely unsolved still. And I just wonder how many in how many places this exists. And I think it’s everywhere.

Jim Huysentruyt (50:37.87) Yeah, it’s interesting

Jim Huysentruyt (50:41.131) because, and I could see it all over the place. Definitely. But I think some schedulers are better, but it depends. know, because I think companies that tend to go toward, companies that tend to run leaner and get the stuff out faster. But then you have dedicated lines and if you have a lot of change over, but we can go, we can keep going. I think we do need to kind of wrap this up here because we’re done.

Andy Siprelle (50:50.199) Sure. It’s an art.

Andy Siprelle (51:04.557) And there’s, you know,

Andy Siprelle (51:07.888) certain designs actually have an inherent maximum availability as well, you know. That’s something to think about as well. A lot of manufacturers design their systems for maximum flexibility, and then that requires them to run a maximum schedule, you know, and so.

Jim Huysentruyt (51:15.308) Right. Yep.

Jim Huysentruyt (51:30.734) Right.

Andy Siprelle (51:34.263) Yeah, challenges.

Jim Huysentruyt (51:34.414) Well, Andy, this has been really good. mean, we can go on forever. I could go on forever. But we need to wrap it up just because of time. But I do want to thank you so much for just coming on and just talking about simulations. it’s a whole different way of thinking.

Andy Siprelle (51:38.797) Yeah, sure.

Andy Siprelle (51:47.021) We probably do.

Jim Huysentruyt (51:59.206) It really is. just looking at lot of our customers and mutual customers, I the ones who are doing it, I think the ones that see the success and once they see the light, I think they’ll keep doing it. It’s the ones, I think there’s a lot that aren’t doing it right now and they’re, it’s just really what would be your biggest advice to somebody? mean, why should I start doing simulations?

Andy Siprelle (52:28.929) Well, think, yeah, I mean, that’s a great question because it might look like a daunting thing for a lot of folks to enter into. I guess that what I’d tell them is that they would do it to get operational certainty. know, when there’s operational uncertainty and the actual system is a battleground for

Jim Huysentruyt (52:48.312) reasons.

Andy Siprelle (52:54.483) set points and what the, you know, this shift runs these set points, that shift runs another, you know, that it’s to drive operational certainty and even I think it’s possible, I think it’s possible to see these systems and to feel the risks and, know, I think that

Andy Siprelle (53:24.587) There’s little certainty about what to work on next and what will really give me the biggest bang. There’s robotics all over the place. People can use robots to help their stretch wrappers out and so on. But one of the problems really in the industry is there’s very little analysts in the manufacturing plant itself. They’re not even there.

Andy Siprelle (53:54.542) You know, so that’s in a lot of the, a lot of the smarts, you know, a lot of the smart folks who’ve been there a while are retiring, you know, 30 years ago, 40 years ago, people operating an oven baking a cookie. There was a certain pride of craftsmanship there, you know, and I think that’s, that’s largely, largely gone. You know, I mean, I think, uh,

Andy Siprelle (54:21.505) how to transfer that knowledge. We do have the opportunity, I think, to transfer that knowledge through systems that can help operators give them guidance on the line in real time, et cetera. So I think it’s possible.

Jim Huysentruyt (54:40.654) OK,

Jim Huysentruyt (54:41.495) great. Well, thanks again. It’s been an absolute, absolute pleasure. So with that, want to thank our audience too. And we’ll see you next time.

Andy Siprelle (54:44.877) It’s been a pleasure.

Andy Siprelle (54:59.542) Okie dokie. Bye bye.