Kirill Eremenko: 00:00:00
This is episode number 421 with simulations expert, Theunis Barnard.
Kirill Eremenko: 00:00:12
Welcome to the SuperDataScience Podcast. My name is Kirill Eremenko, Data Science Coach and Lifestyle Entrepreneur. And each week we bring you inspiring people and ideas to help you build your successful career in data science. Thanks for being here today and now, let’s make the complex, simple.
Kirill Eremenko: 00:00:44
Hello, and welcome back to the SuperDataScience podcast everybody’s super pumped to have you back here on the show. Very exciting episode, just finished recording with Theunis about 40 minutes ago. Very, very interesting if you are curious about the topic of digital twins. You probably have heard about it, it’s a new trendy thing, and very powerful thing as well in the space of data and believe it or not, it falls, at least in my classification, falls into the space of Data Science because it uses a lot of data, so Theunis is an expert. He’s a senior engineer and BI manager at Simulation Engineering Technologies in Johannesburg, South Africa. And he is an expert in simulations and digital twins.
Kirill Eremenko: 00:01:36
In today’s episode, you’ll find out quite a lot of things about this topic and understand what it is, so you can have very informed discussions with colleagues, friends or just people around, and at the same time, there will be some advice, some ideas on how you could maybe include this in your career as well. So, this podcast will be useful to anybody who is interested in digital twins and simulations and how it all works, and how it all happens.
Kirill Eremenko: 00:02:08
So, some of the topics we touched on or we discussed today are: digital twins, industry 4.0 or the fourth industrial revolution. You will hear some mining industry stories, they will put into context why digital twins actually have a place and what happens when they don’t exist. We will talk about digital twins, of course you will hear about creating systems and created systems. Very interesting insight there. You will also hear about data science in digital twins, and how data science in digital twins is different to data science and machine learning. Then Theunis will also help us understand process digital twins versus system digital twins and what the difference between them is. And at the end there will be some career advice in case you’re considering exploring this space for yourself.
Kirill Eremenko: 00:03:02
So, there we go, very exciting podcast coming up, lots of cool insights about digital twins. And without further ado, I bring to you simulations expert, Theunis Barnard.
Kirill Eremenko: 00:03:13
Welcome back to SuperDataScience podcast everybody. Super excited to have you back on the show. Today we have a very special guest calling in from South Africa, Theunis Barnard. Theunis, welcome, my friend, super pumped to have you here.
Theunis Barnard: 00:03:31
Thank you, Kirill, very, very excited to be here.
Kirill Eremenko: 00:03:35
Yeah. Is this such a random thing, like when we met at DataScienceGO, it was so cool, because I love South Africa. I lived in Zimbabwe for a while, so I was like, this is awesome, somebody from South Africa called in. Tell us a bit about that, how did that feel for you, just randomly meeting people from all over the world?
Theunis Barnard: 00:03:56
No, it was amazing. Let me maybe start of, when I met you, I couldn’t believe… I always had this aspiration to be on the podcast show, and it was a bonus to hear that you from Zimbabwe originally. To have that in common was great, and on the topic of conferences where I think I told you, we had a conference in South Africa, a virtual one, a few weeks ago. So, it was amazing to see what you guys did with your platform, and the networking aspect is amazing. Just to meet people from all over the world, and we’re lucky to have gotten that opportunity.
Kirill Eremenko: 00:04:37
Yeah, absolutely. Can you give us some examples, of like, who did you meet? From which countries or what professions?
Theunis Barnard: 00:04:45
Sure, so I met a guy who’s in Canada in the same industry as what we’re in, that would be more in the civils environment doing very similar things to what we doing, at the same time, not really a competitor, so it was just great to exchange details. Met a student from China, busy with his PhD, and I think he really enjoyed it, because he is obviously at home for seven or eight months now with COVID studying, so it was probably nice for him to also meet people. And just to see what kind of talent is out there is amazing.
Kirill Eremenko: 00:05:22
Okay, that’s awesome. You’ve listened to, as I understand quite a few podcast episodes on this show? How does it feel to be on this show yourself now?
Theunis Barnard: 00:05:34
No, it’s… I still have to pinch myself to… I told my wife about it, I think at first she didn’t believe me, but I’m here and it’s amazing. I’m really looking forward to connecting to people, and thanks for the opportunity.
Kirill Eremenko: 00:05:50
My pleasure. Topic for today, so everybody knows, is digital twins. I’m super excited to dive in, when you said at DataScienceGO, you’re in that space of digital twins and that’s what you do. I was like, “I got to get him on the podcast,” because we’ve only had one guest before speaking about this and it’s a really burning topic. But before we dive into it, a bit about South Africa, as we briefly mentioned before the episode, I think it’s a great country, when we lived in Zimbabwe, we did trips to South Africa, like weekend or maybe a five day trip. Fantastic, my favorite place is Sun City, it’s a wonderful city, of course Cape Town, Johannesburg. How’s the situation in South Africa, more importantly, how is data science in South Africa, these days?
Theunis Barnard: 00:06:44
Yeah, so for those who don’t know much about South Africa, we’ve got a very checkered history, lots of political shifts over the past decade or two. And that’s made it a interesting mix of cultures in South Africa, what that’s meant is that there’s been a lot of progress on the human aspects in terms of our, we’ve got, I think rated one of the best constitutions in the world. And freedom for people is extremely important, at the same time, there’s been quite, we call it a bit of the brain drain, lots of people leaving overseas. Which is not really such a big issue if you take into consideration we can do a lot of work remotely nowadays, but I think we’re definitely seeing a gap between wage incomes and obviously you’ve got the challenge of trying to address social, political, economic issues, and I think data science is a really powerful tool for that in our country. So, a unique place to be applying it.
Kirill Eremenko: 00:07:53
Absolutely. The world’s moving forward and we’re seeing quite a lot about ethnics and equality in the space of data science, a lot of discussions in that space. So, this is a great time for all countries to implement these things, and definitely I think South Africa could benefit from a lot of the developments in this space.
Theunis Barnard: 00:08:27
Yep. No, definitely, and I think we are definitely able with technology nowadays to get people educated and up to speed within the country, so the main thing is just enabling people through learning new skills. There’s so much out there, they could really… They just need to know which platforms to reach out to.
Kirill Eremenko: 00:08:48
For sure. Which city are you in, by the way?
Theunis Barnard: 00:08:52
I’m in Pretoria, midway between Johannesburg and Pretoria, the capital, in the heart of the busy business center of South Africa.
Kirill Eremenko: 00:09:04
Okay. Got you, cool. All right well, South Africa, and I like what you mentioned, that a lot of things can actually be done virtually. This is an interesting time, especially with this coronavirus that more and more people are going to recognize that and maybe people in different countries, are not going to be like, if you’re in the US you’re helping the US economy all the time, or in UK you always have the opportunities. Like as you say, there is a big landscape for data science in South Africa, because I guess it’s not as competitive as Silicon Valley, and maybe people will be able to help or contribute to some projects.
Theunis Barnard: 00:09:55
Yeah, no, definitely. That’s, I think, one of our aims is to reach out and get as much work done for overseas companies to keep us competitive, and at the same time use the skills we learn to train new younger generation within the country.
Kirill Eremenko: 00:10:14
Nice. That’s very cool.
Kirill Eremenko: 00:10:18
Hope you’re enjoying this amazing episode. We’ll get back to it after this quick, short announcement.
Kirill Eremenko: 00:10:22
And the quick short announcement is that, we have DataScienceGO Connect. You’ve probably already heard of DataScienceGO, which is the conference run in California, you’ve probably also heard of DataScienceGO Virtual, the virtual conference we run several times per year, and in order to help our community stay connected throughout the year, we’ve started running these virtual events which happen every single month. So, you can find them at DataScienceGO.com/connect. They’re absolutely free, you can sign up at any time, and then once a month we run an event, where you will get to hear from a speaker, or a panel discussion. Maybe an industry expert Q&A, but very importantly there’s also Speed Networking sessions, where you can meet like minded data scientists from around the world. This is a great way to stay up to date with industry trends, hear from amazing speakers, and also meet peers and exchange details and stay in touch with the community.
Kirill Eremenko: 00:11:19
So, once again, these events run monthly. You can sign up at DataScienceGO.com/connect. Would love to see you there.
Kirill Eremenko: 00:11:29
Well on that note, let’s move on to your work. So, you work in a very interesting space of digital twins, or simulations. The company is called Simulation Engineering Technologies, or SET, and it’s part of a group called 4Sight. So, quick run down, tell us a bit about the companies, the structure or the main mission, objectives, just a bit to know what it’s all about.
Theunis Barnard: 00:11:57
Yeah, sure. So, the company currently consists of about five sub-groups, subsidiaries. These joined together roughly about five years ago, but each of the companies themselves have a good 20 plus years experience in their respective fields. And, I think it was about four years ago they listed on the Johannesburg Stock Exchange, a big moment for us.
Kirill Eremenko: 00:12:22
Yeah. Well done.
Theunis Barnard: 00:12:23
And, I’ll be honest, thanks, the past few years have been quite exciting, hasn’t come without its own challenges. Board changes, and new management, internal management coming in, and ever since then we’ve seen a lot more steady growth.
Theunis Barnard: 00:12:44
And I think what makes it so exciting is that it’s such a new field, not just digital twins, but I’m going to throw a buzzword in here, but the Industry 4.0. So, this group of companies, obviously I’m with the simulation group, but we all have various different domains of expertise. So, we’ve got software development, we’ve got telecoms, we’ve got advanced process control and then the simulation modeling fits in that multi-strategic level within businesses. So, we cover from low level, we call it devices and hardware to very high level strategic work, which makes it very exciting.
Kirill Eremenko: 00:13:28
Wow, fantastic. So that’s your division, covering the low level insights and the strategic?
Theunis Barnard: 00:13:36
So, we actually do more the high level, so we’ll get into the digital twins and basically what that means, is that our simulations models get hooked on to the actual live company or organization systems to make them digital twins, and then you get access to all the low level data from various devices.
Kirill Eremenko: 00:13:58
Cool, okay. So, you mentioned the buzzword Industry 4.0, what does that mean?
Theunis Barnard: 00:14:06
Right. So, as far as I understand it, these things always get a few tales put on the stories. The German government decided a number of years ago, to come up with this term, Industry 4.0 to incentivize the integration of automation and digital systems. So, the first industrial revolution, was your steam train, steam engines. Second, your electrical systems, and then the third would have been the creation of computers, the internet, telecoms. So, the fourth industrial revolution looks at integrating all of that, and just sharing data between any platform and optimizing your industrial systems. So very much your future view of where technology is going.
Kirill Eremenko: 00:15:02
Okay, very cool. So, how far are we in to the fourth industrial revolution?
Theunis Barnard: 00:15:11
So, I think, we’re very new in it. Obviously certain countries are way ahead of others, and I think South Africa is probably lagging quite a bit, and then at the same time you’ve got industries that are ahead and others that are lagging. So, mining and manufacturing, the two industries we largely focus on definitely lag its, in that sense. Because you still have a big component of the human element in there. Whereas let’s say, entertainment, telecoms are probably further ahead because, there is so much more you can automate and put online these days. So, I think the future’s, we’re going to see that curve of maturity is going to keep on accelerating for the next five years. So, I think fairly early.
Kirill Eremenko: 00:16:01
It’s interesting, with this mining industry, I got a really cool example of what you said that many cases they’re lagging behind. In some cases it’s surprising how far ahead, there’s in Western Australia there’s this Rio Tinto mine of the future, and they started I think in 2008 or something, ages ago. And they have these automated trucks, a fleet of 200 trucks that rock up, and then the excavator puts the soil and stuff in the truck, the truck goes, and it’s all like 200 trucks are run by a team of 15 data scientists, logistics people sitting somewhere remotely. So, some are ahead, but then some mines, it’s like every single mine has its own little story, as opposed to telecoms which are like competing fiercely for attention.
Kirill Eremenko: 00:16:52
Mines have their own, as I understand it, they have their own deposits of minerals and whatever else, and they need to work on them. As long as they can get them out at a cost that’s lower than the market price. It’s a commodity right, you can sell as much of it as you want. They’re making profit, so they are not really bothered in many cases about these things. And so the story I have is, I know this from friends, and I won’t disclose the name of the friend or the mining company, but basically at one running company, they get this ore and they need to crush it down into little things, before they put it through a chemical process to extract gold or whatever it is, silver or platinum.
Kirill Eremenko: 00:17:39
So, the way they crush it, is they put it into a big spinning thing, like a big washing machine, and so it’s huge, three stories high, 30 meters or 20 meters or so. And it’s got big metal balls in there, as it goes these metal balls fall down and they hit the ore and they crush it, and so it’s spinning and the ore is falling and crushing it. And so you can imagine, a washing machine, 20 meters high with metal balls inside, and you chuck in the ore, they fall down, crush it. The question is how fast do you spin it? So, do you spin it too fast and the metal balls will go up too high and they won’t be falling at the optimal angle, and you will be wasting electricity, which is a ton of electricity for inefficiency.
Kirill Eremenko: 00:18:32
If you spin it too slow the metal balls won’t have enough velocity when they hit the ore and they won’t be crushing it as effectively. So, there’s a optimal speed. So my friend tells me this story, the way they find the optimal speed, you probably going to laugh me out of the room here. The way they find the optimal speed and this is a mine offshore, some island somewhere, some country where you get expats. People with a lot of experience in this specific thing, they come and they walk around, they are very important indeed, but they don’t want to really share their knowledge because that’s their secret sauce.
Kirill Eremenko: 00:19:07
So what they do, this guy has been doing this for 20 years, he comes up to this machine and they all have to wear earphones to protect them and so on. He comes up to this machine, to the wall of it, not the spinning wall but the outer wall. And he takes his ear thing off, puts his ear to it and listens to how many hits he hears, how many metal balls he hears per minute and then from that he’s like, “We need to speed it up, or we need to slow it down.” I’m like “Boom.” How ridiculous is that?
Theunis Barnard: 00:19:38
Yeah. That’s sounds very familiar and I don’t know, this is probably, correct me if I’m wrong, but probably quite a relatively advanced mine.
Kirill Eremenko: 00:19:47
I don’t know much about it. But, yeah.
Theunis Barnard: 00:19:52
No, we hear all those stories of rudimentary ways of people doing things all the time. So, I think the mining, when you’re onsite it’s an engineer’s playground, because it sounds simple, the processes, there’s one commodity flowing on the conveyor, goes through a crusher, the plant and then out the other side. But there’s so many cool little inventions along the way of the way people have done things. So, I like the mining industry, they’re got a shoot, where you’ve got the ore and they’ve got a big, I think it’s like 20 story building. So the ore on the final step goes all the way up to this top of this building, and the way they get the diamonds out of it, is this ore falls all the way through the building and they’ve got a little laser, so there’s the tech part, that shines a light and when it picks up a reflection on this ore falling, then it knows, okay there’s a diamond. And then it actually shoots a little bit of jet stream of air and pushes that ore into another bucket and that’s how they get the diamonds out at the end.
Kirill Eremenko: 00:21:00
That’s crazy.
Theunis Barnard: 00:21:03
Yeah. Amazing.
Kirill Eremenko: 00:21:04
That’s cool. So, some things are like, that is cool automation, then there’s, I forget the company that does it, but there’s some companies that do that for mining companies. But somethings are well automated and well designed, but somethings are so pre-historic and I can see what you mean. They’re lagging, you need the whole system to be updated and to be using AI to its full potential, if you have a few links in the chain that are not automated to that extent, there’s room for improvement.
Theunis Barnard: 00:21:36
Yeah. No sure, and what we try and do is, we try and integrate. The first step is digital transformation, get all the systems digitized automated as far as possible. Doesn’t mean remove the human element, it just means automate it, a part of it. Once you digitized, you can analyze, we call it digitize, analyze, optimize. So, understand what’s going on in bigger system, and then when you get to the optimize step, that’s when the simulation comes in. So, what’s great about the simulation is we take all the data, typically sit in a office off-site, build a model of the entire system. It could be a couple of mines actually put together, and then we understand where’s the biggest constraint within the system. And that’s where we focus on. So, it’s nice to combine the high-level, low-level view.
Kirill Eremenko: 00:22:31
Awesome, okay.
Theunis Barnard: 00:22:31
And we also take into consideration the human element.
Kirill Eremenko: 00:22:36
Very cool. So, let’s talk about digital twins then, what’s the definition of digital twins? And as I told you I did some homework, I watched some videos, so I also have a definition prepared, one by IBM actually. So, we can compare and let’s talk about the topic. What are digital twins?
Theunis Barnard: 00:22:52
Yes, so I’ve actually decided to when people ask me that question, I stay away from pure definitions because… I’m of the opinion that these things are to a degree constantly evolving. The same with the term, data science to a degree. You probably know way more than me on the topic, but as technology’s evolving, these things are just growing and to even understand all of this in your mind becomes quite complex. Our definition of a digital twin is, when you have a model that represents the real life system, you have now extracted that system on to a model. And the model is never exactly perfect, it can never be exactly the same as the real thing.
Theunis Barnard: 00:23:48
That’s a whole philosophy on its own, but once you now integrate that model with your life system, in other words you are getting data directly fed from the system in real time, and real time is also a very relative term. You can have something running with real time, one second or every 10 seconds or every 10 hours or whatever. But for the industry and environment you’re in, once the data is fed real time then, now in our terms you end up with a digital twin representative of that. And it’s across the life cycle system, so you actually start building your model before the system is implemented, so in design phase. And that model lives on all the way through to decommissioned phase at the end.
Kirill Eremenko: 00:24:34
Okay. Very cool [crosstalk 00:24:36]. Yeah, no that’s very cool. So, as I understand, that’s a very good important clarification made there that, it has to be updated life with data, so you can’t just build a digital twin of an airplane and then just, okay, now there’s this airplane and then there’s this digital twin and that’s it, they’re separate. You have to whenever you can, update the data.
Theunis Barnard: 00:24:58
Yes. And I think that’s also, we use the term loosely because obviously it’s a growing field and people like to hear that you’re doing digital twins, but I think what we often try… We try and explain that to our clients because we’re not, I think in any specific industry, we have that perfect digital twin, you always going to have some aspect of it that might be left out, or might be a strategic level.
Kirill Eremenko: 00:25:23
Gotha. So, yeah the IBM definition is pretty much the same, so we’re not going to go through that. But another thing you pointed out, they also, in this video I watched on YouTube, they also point out that a digital twin helps with three phases. It helps with the design phase, the build phase of whatever it is you are building and the operation phase. So, I used to think, that it’s only just for the operations part, but actually it’s for all three. So, you even create the digital twin even before… How does that work? Do you create the digital twin before the design phase or do you create the digital twin to inform the design phase? Is that right?
Theunis Barnard: 00:26:11
Yep. So, I think probably… I always like to break problems down to something like a toy problem, something small. So a easy way to understand this is when you building a house, you might start a sketch on a piece of paper, so that’s already getting the creative process going. But then you start putting that into a CAD program on a computer, so then you already got a digitized model going. And then obviously as you start building this house, the model might even be refined and to a degree the model is living alongside. So, that’s I think how you can see it informing the design.
Theunis Barnard: 00:26:49
And I’m actually working on an interesting project at the moment where there’s an existing plant, it’s for a manufacturing industry and the plant exists, so we’ve got a good idea of what kind of model to build. But they want to do a new plant, Greenfields, so now we’re using our model to inform the design, once the design’s done they obviously going to building it and operating it, and our model lives alongside that. But then they want to do expansion, so then once again we’re informing on a design level. So this is the iterative process between design and execution and operation.
Kirill Eremenko: 00:27:31
And digital twins can be used at all sorts of scales, right? You can be at the scale of a plant, you can be at the scale of one machine. It can be at the scale of even a human. You could have a digital twin of an organ, of your heart or something like that. And feed data to it from your, you might have a heart implant or from your Apple Watch or something like that. And just observe how that changes and get informed. Tell us a bit about that, what are some of the common or most exciting use cases of digital twins out there?
Theunis Barnard: 00:28:08
I think you hit it spot on, if you look at social networking, so I’m obviously talking about platforms like LinkedIn, Facebook. In essence those are starting to create digital twins of people, and obviously it’s up to the company how ethically they use it. Won’t get into that debate. But-[crosstalk 00:28:28]
Kirill Eremenko: 00:28:28
It’s too late for that. The Social Dilemma. Have you seen it?
Theunis Barnard: 00:28:32
Yes, that is exactly the one I’m thinking of. [crosstalk 00:28:37] I think they bring it out in the way they portray this person. Making decisions for him. It’s very interesting.
Kirill Eremenko: 00:28:49
Good movie to watch and then delete your social media accounts.
Theunis Barnard: 00:28:57
I was working for an aerospace company and I think that’s probably the most I learned from that industry on technology and digital twins actually to a degree, because the aerospace defense environment, I think there’s a lot of history obviously going back to World War II. They actually worked out a lot of things on pen and paper, which we are using in our systems today. And I think, for someone that hasn’t been in that environment, that history just gets lost. But this one guy I was working for, he said, you get a creating system, and you get a making system. So there is a system that, what he referred it to as, the made system and the creating system. For example in a mine you would have a plant and the mine itself would be the creating system and the ore would be the, it sounds weird to say, but a created system.
Theunis Barnard: 00:30:00
The same goes for, let’s say, building airplanes. You’ve got your creating system, your manufacturing plant and the created system is your airplane. So it goes a lot more beyond just a certain product and that opens your mind to thinking about any industry in a new way. The interesting question is with social platforms, who’s the creating system and who’s the created system.
Kirill Eremenko: 00:30:30
Got you. That is very interesting, so you have the creating system and you have the created system. So as you said in planes, you have the factory is your creating system, the plane is your created system. But did you say that in mining it’s the other way around? The ore is your creating system and the mining plant is actually the created system.
Theunis Barnard: 00:30:56
No, it’s the same in mining.
Kirill Eremenko: 00:30:58
It’s the same.
Theunis Barnard: 00:31:00
Yes. The difference is that your created system is a very, I don’t want to say less complex system, but it’s not an interactive system. You’ve got ore that comes out at the end. But-
Kirill Eremenko: 00:31:13
It’s not an airplane.
Theunis Barnard: 00:31:14
Yup, exactly.
Kirill Eremenko: 00:31:15
Okay, so the plant is the creating system, the ore coming out then is the created system. And so the question in The Social Dilemma, right, what’s the creating and what’s the created, right?
Theunis Barnard: 00:31:30
Yes. I suppose at the end of the day, Social Dilemma talks about us as being the product, and I don’t have a particular view about it, I think it doesn’t help to think too negatively about these kind of things. But if you’ve got a digital twin, the closer that comes to representing a human being, once that platform starts providing insights to the human being on the decisions it should make, instead of the other way around, then you’ve got an interesting situation. But just to get back to the mining versus the manufacturing example, I think what opened my eyes there as well, is that you’ve got these industries that look to a degree vastly different. But we doing the manufacturing plant as a creating system and taking what one learns there in terms of data science. You could take that and apply that to an industry such as mining, if you understand that now all of a sudden your created system is not so complex, it might be higher volumes but you could still use the same algorithms and knowledge there, just in a different way.
Kirill Eremenko: 00:32:49
Okay, so it’s transferable knowledge.
Theunis Barnard: 00:32:52
Yes.
Kirill Eremenko: 00:32:54
That’s very cool. How is data science in digital twins different to data science in machine learning?
Theunis Barnard: 00:33:06
So I think for us, where we do a lot of simulation work, before they go into digital twins, there is a abstraction of the real world data. We might get info from spreadsheets and the models we built, are built on a lot of assumptions. And we calibrate these models, to make sure they represent the real life system. But given these assumptions, you do these longer periods of studies on your own. Whereas with machine learning, AI applied to a real world system, directly working on the live data. It’s an interesting dynamic, you get feedback a lot quicker, and you actually see the results of your efforts a lot quicker. If I change this lever on a conveyor belt or specify a certain algorithm for processing the ore in a certain way, a week later you will already see the results of that. Whereas with strategic modeling you typically would give a report and a lot of that is based on assumptions and there would be interpretations. So, there’s quite a bit of a difference. And we use more heuristic algorithms whereas AI is not so relevant in the simulation modeling.
Kirill Eremenko: 00:34:25
Okay. Heuristic, you mean like as opposed to if else algorithms, as opposed to brute force just process all the data algorithms?
Theunis Barnard: 00:34:33
Yes. Correct.
Kirill Eremenko: 00:34:38
Why is that?
Theunis Barnard: 00:34:39
Why is-?
Kirill Eremenko: 00:34:40
Why do you use heuristics? Heuristics is like for instance if we look at, what’s it called, the game of Go versus chess, Deep Blue back in the 1998 or whatever beat Kasparov based on heuristics. Like if else, lots of scenarios, picking the right one and so on. Whereas AlphaGo or now AlphaGo Zero is just brute forcing its way through with deep learning, algorithms and not going through all the possible combinations because that’s impossible, but it’s actually just learning on the go. So, why in the digital twins space do you still use heuristics rather than allowing AI, using something like deep learning and allowing AI to learn on its own?
Theunis Barnard: 00:35:27
Yeah, that’s a interesting question and it’s actually got quite a lot of context to it. So, I think the first thing I left out up front is that our models are large UI digital twins, you get different types. Ours is a process digital twins, so you would have digital twins of a physical system so the airplane again is a good example. You’re doing fluid dynamics, CFD’s and all sorts of other calculations around the engine of a airplane. So that’s not what we do, we do process digital twins, we look… It’s actually something you can’t always see, so we model the process and then we come up with answers how to optimize that system. So that’s the first distinction which means that, the simulations models we build are not purely, they’re data driven, but they’re not…
Theunis Barnard: 00:36:23
We actually build a physical layout of a site and the heuristics come in where you’ve got, let’s say, a truck has to drive from one side of the plant to another, they’ve got three different routes they can take, which one should they take? So you don’t really, obviously AI would help you there, but for the sake of the types of projects we work on, coming up with the heuristics to say, choose a path with the least traffic on, those kind of rules makes it a lot easier and it’s more transferable to the actual operation, because you might not have trucks working with that AI. But then we do have the simulation per definitions runs, brute force, lots of scenarios to test, which would be the best outcome.
Kirill Eremenko: 00:37:13
Gotcha. So, the way I understand with the heuristics is that you have certain constraints that exist in the real world and you just have to adhere to them so you got to code them in as heuristics. [crosstalk 00:37:24]
Theunis Barnard: 00:37:23
That’s right.
Kirill Eremenko: 00:37:26
Gotcha. You said simulations. What kind of simulation algorithms do you use?
Theunis Barnard: 00:37:32
Right, so we rely on software, proprietary software that we use to build our models. The one package we use is Simio, so we agnostic to different tools, but we found this package to work really well. And it allows us to built almost any kind of physical environment where you have to hook your process model onto some physical layout so the trucks is a good example there. You’d struggle to do that maybe in a MATLAB, where you can’t physically see the trucks moving and understand what the dynamics are. And then we hook that model onto a interface like, let’s say Power BI or Tableau and that gives us that capability for client to obviously publish the dashboard to environment organization or just as a presentation. And it will also makes the whole process of doing a study a lot quicker. And then if we need to, we start using something like Python or R if we need to do really difficult calculations. Typically upfront before we do the model, and then also that can help us do the visualization aspects as well. So, it’s a narrow but nice toolset to work with.
Kirill Eremenko: 00:38:53
Okay. Gotcha. What does your day to day look like?
Theunis Barnard: 00:38:59
Right. So, I typically… The environment I work in is a very flexible environment, we… What I like about the work we do is, we work independently, together in groups on projects but very independently, so your deliverables are directly attributed to your work. Which means I could work remotely anywhere if I wanted to, or go in to the office and we’ve got a very collaborative environment. Each day has its own way of panning up, but obviously there’s a lot of planning that goes into it upfront.
Kirill Eremenko: 00:39:43
So, do you, for instance if you’re on a project, do you do the whole end-to-end or do you do a specific part of the digital twin project?
Theunis Barnard: 00:39:51
Yes, so because these, let’s take mining for example, because these mines have such a long life cycle, some mines you easily get up to 50 year life of mine, so you might come in, this mine might be 30 years old and you jump in and build a model that has to model the operations. So, no design influence there. On other projects you might be in conceptual phase, which has a very different dynamic on its own. And then you’ve got the manufacturing projects which are typically shorter term, so the example I gave where we starting off, right at inception and if all goes well we’ll continue with that for the next couple of years.
Kirill Eremenko: 00:40:37
Okay. But in your specific day… Like how some companies they pigeon hole you into one specific, okay, you’re always going to be writing Python code for this specific thing. Do you get the opportunity to work across different tools and different challenges within an individual project or do you specialize in something?
Theunis Barnard: 00:41:05
Yes, so I think I had to make the call a couple of years ago whether I was going to go highly specialized or diversify and maybe even management and I just found that I liked both. So I moved from a corporate company to the company where I am at now, is small within the larger group, which gives us a lot of autonomy. And I’ve had the opportunity in the past two years to work on, just to give you an example, I’ve done these simulations models, which is actually a highly specialized work, and I mean we’ve got many guys going doing Masters PhD, and at the same time I’ve been able to be a project manager of a digital transformation project for a mine. Looking at every single department and coordinating a team of 30 or 40 people. So, it’s really been great to get that diversity but I think at some point one has to make that call which, I wouldn’t say which specialization you going to go into, but for long term you want to keep building on the same thing.
Kirill Eremenko: 00:42:17 Okay. I have an interesting thing here we can do is, if maybe a few more… You’ve already given a few examples of digital twin projects, maybe you can do a few more, like I have one. I’ll start first, I have one that I found online where digital twins, just to give our audience a feel for, what else can be digital twins be used for, so what I found really cool is, if you have a big manufacturing plant and you have a robot, basically a machine that needs to cut in a straight line or cut certain, I don’t know, let’s say a piece of a car, a door. It needs to cut the metal in a certain shape. That’s really cool, and they’re very accurate but because it’s such a big manufacturing plant there might be other things going on like vibrations coming from some big device hitting something or doing some testing or something.
Kirill Eremenko: 00:43:23
So unless you have the luxury of having a huge space and everything isolated, great, but what if your building is in a smaller confined space, if there’s some certain vibrations, they can impact this machine. So, in this case you would build a digital twin, this is the part that was really impressive to me, they built the digital twin not of just that robot that’s cutting, they built a digital twin of the whole system, of the whole manufacturing plant and then you know exactly when these vibrations are happening, which way they are going and so on, what’s producing them. So, then you can feed that information to the robot real time and it can adjust by millimeters its cutting trajectory to take those vibrations into account. So you don’t have to stop it and wait for the vibrations to go away, you can actually use that while it’s working. How cool is that?
Theunis Barnard: 00:44:15
That’s amazing. So, my mind’s already going to that systems view, where now you’ve got a… Your created system might be your part, your car part that, that plant’s producing, your creating system is your plant. But now you’ve got another third system around that interacting with both these two systems, so it just blows your mind when you think of all the different scenarios. So, yeah, that’s a brilliant example. And what people underestimate is, you look at a manufacturing plant, and you think, oh well, there’s complexity, there’s the robots, but from a process perspective it can be incredibly complex to understand and optimize that, as well as the part that is being produced. And so you’ve got digital twins of all these things popping up everywhere.
Kirill Eremenko: 00:45:05
Gotcha, do you have a cool mind blowing example?
Theunis Barnard: 00:45:10
Of a digital twin in manufacturing?
Kirill Eremenko: 00:45:13
In any thing, really.
Theunis Barnard: 00:45:15
Well, I think-
Kirill Eremenko: 00:45:16
What’s your favorite?
Theunis Barnard: 00:45:19
I really envy anyone that builds digital twins in the aerospace field, I think that’s on another level. The platforms that are being created, first of all on the manufacturing side, to model the flow of how these pods get put together, having the right components ready at the right time. Then that engine is built and now your designer who designed this engine has his model and that thing gets put in a plane, and he couldn’t have seen upfront exactly on that engine part how that would… What forces would eventually end up interacting on that engine once it’s up in the air. Obviously he models it, and he’s got all the experience and insight that goes into that, but only once that thing is up in the air and let’s say something goes wrong, then he might get the opportunity to get that engine back, download the data obviously stored on that platform and then see what happened between the model and the engine. What exactly happened there.
Theunis Barnard: 00:46:31
So, I’ve got another example from the defense industry we worked in where we built parts that actually detect if you were being fired at from a long range, and this part had to pick up the direction, the [inaudible 00:46:50] and the heading in which, let’s call it now missile, is coming in at. So, just by the way, all the work we did was defense not attack. So, now you’ve got this model of how this part should be working to pick up all these things firing at it, but now you put it on a helicopter, and now you’ve got a whole new dynamic because this helicopter is moving, the weather conditions. So, how do you build a thing like that without being able to test it? You can’t just go quickly outside, put it on a couple of million dollars worth of aerospace equipment or aviation equipment and see what could happen. You have to wait for that part to be put on a platform a year later, it actually gets into the field, so the amount of skill you have to have, to get something like that to work properly is truly amazing.
Kirill Eremenko: 00:47:53
And then you wait until it needs to get shot at-
Theunis Barnard: 00:47:56
Exactly.
Kirill Eremenko: 00:47:56
And then only you can test it out. That’s crazy. Cool. What’s the future of digital twins and where are we heading to? Is everything going to have a digital twin? Is my chair going to have a digital twin? How’s it look like?
Theunis Barnard: 00:48:17
Yes, I think hopefully the human race will be, I am sure the human race will be clever enough not to apply it to everything. I think there are just certain things that it becomes overkill. I’ve actually decided to go with a relatively cheap Casio watch because I just want to know the time, I don’t have to have notifications come up on my phone the whole time. But I think it is inevitable that we are going to have, we are going to see integration of or digitization on almost every single platform you can think of. So, I think, I just hope that we stay in control of the technology and it doesn’t become a master of us. And I think if we can do that the future outlook looks great.
Kirill Eremenko: 00:49:17
How soon will we have our day to day lives impacted? It’s really cool to hear about stories from manufacturing and mining and other industries, but how soon will we notice an impact on our day to day lives from digital twins?
Theunis Barnard: 00:49:38
Yes, I think the people in control of the digital twins, I would definitely say can already in industry see the power of it and it’s making their lives a lot easier, unfortunately if you at the receiving end of that in terms of losing the control of the digital twins, where it might be replacing your job then obviously you’re not even going to really know about it. On a personal level I think we’re definitely seeing it with social media, the impact of digital twins in our lives-
Kirill Eremenko: 00:50:18
Which is negative. Is there any positive examples?
Theunis Barnard: 00:50:22
Yeah, I think there is definitely positive examples. The past few months with COVID where we haven’t been able to be in contact with family and friends. The technology’s been amazing to help connect people and then we’ve also, I’ve got many friends overseas and it’s great to keep in contact. You don’t always have the time, especially with a family, you don’t have the time to call someone or arrange a call, so just to see that they’re doing okay, they’re loving life. I think that’s great.
Kirill Eremenko: 00:50:58
You mean social media?
Theunis Barnard: 00:51:00
Social media, yeah. Instagram, Facebook whichever it is.
Kirill Eremenko: 00:51:04
But what about digital twins? Do you have an example of digital twins impacting or maybe soon impacting our day to day lives in a positive way?
Theunis Barnard: 00:51:15
Well, I think I’m really excited about the resurrection of the space race. So, I think digital twins are going to be invaluable in exploring space, with us going to new planets or sending equipment up to do exploration. So, I think that’s brilliant. I can’t think of a more noble cause for digital twins than that.
Kirill Eremenko: 00:51:43
Okay. All right. I’ve got an example, might be more down to earth example. For instance, what’s it called, digital twins, IBM uses digital twins to analyze the comfort levels in their offices and how people navigate through, what paths they take to walk and then they restructure the layout of the office, using that information. That’s another way that maybe people working in offices will see. Maybe we’ll have that for homes, maybe things like, what’s it called, Nest, that controls your temperature in your home, maybe once that is integrated with a digital twin of your home, it can even better do its job.
Theunis Barnard: 00:52:35
Yes, no definitely and you just triggered another thought of mine, is that we also do digital twins in the transportation industry, so a lot for rail networks and I’ve actually had… It was a couple of years ago, visiting Australia, the Australia, what you call it, public transport is really excellent. I was amazed by what you’ve got going there. The way that they’re using data to optimize transport traffic, so apparently, I don’t know if this is, this must be implemented by now, but to regulate the traffic lights based on the congestion. That’s brilliant.
Kirill Eremenko: 00:53:22
That’s really cool. That would be a digital twin as well, right?
Theunis Barnard: 00:53:25
Yes. [crosstalk 00:53:26]
Kirill Eremenko: 00:53:26
You can’t get away, sorry, you can’t get away with just like, okay, this one road we will put a sensor, or even machine learning algorithm and so, you need the whole system. You need to know, okay if we put a green light here, what will happen three roads down, what will happen on this block. No, there’s a firetruck coming here. We need the whole thing working together.
Theunis Barnard: 00:53:46
Yep. And obviously you need some kind of AI ready, because just to come up with basic rules or even do mass simulations to understand that flow, you not going to get it in time so you need a system that can take a vast amount of inputs, lots of mass of data and make those decisions very quickly. And learn from it as it goes on.
Kirill Eremenko: 00:54:08
You raise a good point there, digital twins are a big thing with… It’s funny I thought of it for like three seconds the best word I could come up with, was big. So funny, big, big. Anyways, they’re a big deal for [crosstalk 00:54:30] cities, I heard San Diego for instance in the US, they’re working with some, and somebody was on the podcast, I just don’t remember the episode number or exact who the guest was, but somebody was talking about how they’re working with the City of San Diego to build a digital twin of a lot of things that are going on, from infrastructure to pipes and electricity and so on to roads, networks and so on to optimize to allow the city to operate efficiently because we’re growing, population is growing, like we always have.
Kirill Eremenko: 00:55:02
LA for instance is stuck in traffic a lot of the time. And a lot of cities are facing these problems and yet there are times when there’s no cars on the roads, like in the middle of the night. Why not do road works then, or whatever else, but cities have so much data, so much going on, they’re drowning. A lot of them are drowning in all this, and humans are not capable of even teams of humans are not capable of just sitting down, okay let’s make these decisions. And then test it out, roll it out, let’s see what happens in the next month in our city. With a digital twin they could benefit, you can simulate stuff and you can roll out these different scenarios and see what happens before you actually implement them in the real world and suffer the consequences.
Theunis Barnard: 00:55:44
Yes, that’s actually a field that I’m so interested in, I think we’ve just in South Africa, we’re not mature yet for that, but in terms of infrastructure, but I come from a family of architects and so buildings and design of buildings really interests me. And the big movement in data is now, the call it BIM, I think there’s a BIM 360 Building Information Modeling. That basically means that exactly the same digital twin concept you applying to buildings. So, you’ve got your model, construct your building, but then you actually end up with sensors in your building, so you getting temperature information, vibration information, electrical information, all of that getting fed into a model and you could optimize whatever you wanted there.
Theunis Barnard: 00:56:32
And obviously the possibilities are very big, but now apply that to a city and I think you’ve got some really good, powerful capability to take a city and just make it better in whichever aspect you look. Whether it’s getting better use of electricity, optimizing traffic flow, even social aspects of where people meet and move around.
Kirill Eremenko: 00:57:01
And you got a lot of sensors right, you’ve got a lot of… You could have a sensor on every corner and every video or on every window. That’s so much information coming in.
Theunis Barnard: 00:57:15
Yep.
Kirill Eremenko: 00:57:17
Okay, what’s your advice for somebody who’s starting in to the space of data science, or maybe transitioning from some other background into data science? What’s your advice for them to, if they’re interested in exploring this space of digital twins and simulation more and they want to maybe direct their career path in that direction? What recommendation would you give?
Theunis Barnard: 00:57:42
Okay, so I think my suggestion would be is, before you decide… The world of data science, AI, machine learning is so big, before you decide where in that field you want to go, is first to decide which domain do you want to apply this in. So, I heard a comment once where someone said, you have to define the problem before you choose the tool. Don’t let the tool determine the problem. Not to go out and just do data science for the sake of data science but actually figure out where do you want to apply this. Because that’s in my opinion, that’s going to spur on your passion for whatever you doing. Because, I think it’s important to have understanding of the real world implications of what you busy with. Especially with this world becoming so interconnected, you don’t always have the view of the greater impact of what you doing.
Theunis Barnard: 00:58:42
And then I would say, in terms of simulation modeling, there’s some really good courses on simulation modeling so you could have a look at Simio is the one platform, AnyLogic, there’s quite a number of them that’s more on the process level. And look at a couple of courses and then identify which kind of company you want to work with. And I actually listened to the previous podcast on career success, and I think that advice is brilliant to say, luck is when hard work, preparation meets opportunity. So then just stick at looking for those opportunities and put in the hard work while you can in preparation for that.
Kirill Eremenko: 00:59:30
Awesome, and I think this will be a cool time to revisit or maybe or cement it in, because this is an important choice, right? You said, process digital twins versus system digital twins. Like in a nutshell, you already talked a bit about this, but in a nutshell, what’s the difference and how does one pick which one they want to do?
Theunis Barnard: 00:59:55
I think on the process side, I’ve got a bit more of an interest on the business level, how things impact on a business level. So that’s where processes are very important. If you are looking at being a lot more, let’s say you’re a hardcore engineer working with materials, I would say look more at your physical digital twins. More the scientific…
Kirill Eremenko: 01:00:28
I understand, process is business results orientation, that side of things. And systems is how’s this thing got to be designed or how’s this thing working, you’re more focused on the asset or the object itself. That’s really cool. Thank you. That clarifies it a lot.
Kirill Eremenko: 01:00:47
Awesome, fantastic. Well, Theunis. What else, we’re coming to the end, so I wanted to know, very interesting work you’re doing. If there’s companies out there that want to work with your company and want to get a consultation or some maybe digital twin project or maybe there’s somebody who’s very interested in this and wants to get a job in this space. How can they find your company, SET or 4Sight?
Theunis Barnard: 01:01:18
Sure, so they could do a Google of 4Sight, so that’s four, the letter four, and then Sight, S I G H T. The site company I work for is Simulation Engineering Technologies, if you Google that you’ll find our site directly. And then please feel free to add me on LinkedIn. And you could even send me a email address to, my email is, theunis@setec.co.za
Kirill Eremenko: 01:01:46
S E T E C.co.za
Theunis Barnard: 01:01:50
Yes.
Kirill Eremenko: 01:01:51
Okay, gotcha. Cool all right and what’s… One final thing… Let’s do the book first, what’s a book you want to recommend to our listeners?
Theunis Barnard: 01:02:08
Yeah, so I love reading books. I try to read at least three or four at a time, slowly and the one that I am currently reading at the moment is Deep Work by Cal Newport.
Kirill Eremenko: 01:02:22
Great book.
Theunis Barnard: 01:02:23
You’ve read it?
Kirill Eremenko: 01:02:24
Yeah, I read more about half of it, really cool insights, love it. Totally cool.
Theunis Barnard: 01:02:31
Yeah, brilliant. I am actually reading the book, listening to the audiobook at the same time. So, just a whole philosophy, I think it’s very applicable to data science. If you going to want to be a good data scientist, you going to have to put in solid blocks of time to figure out your routine. So, excellent book to read.
Kirill Eremenko: 01:02:53
Absolutely. And what’s your, one final piece of advice, or I don’t know, wish for our listeners on the podcast in order to help them in their careers?
Theunis Barnard: 01:03:10
Yeah, so another book, I’m going to take it from this book because it’s meant a lot to me the last while. It’s a book called, The Slight Edge, and basically it’s… You’ll find these principles probably in most career books, but it’s just the idea of taking things, little bits at a time, over a long period of time to realize that right now, your dream is to be at a certain point three or four years from now. But focus on the here and now. I heard a quote, “Be where your feet are.” And do the little bits, focus on the process, work towards the outcome and when you look back over a year or two you will be amazed by the results you see.
Kirill Eremenko: 01:03:53
That’s cool advice. Still a lot of time we get caught up trying to shortcut to what we want. That’s not always, in most cases it doesn’t work. Got to put in.
Kirill Eremenko: 01:04:08
Theunis it’s been a huge pleasure, thank you for joining me today. And I learned a lot about digital twins. Thank you.
Theunis Barnard: 01:04:14
Thank you to you Kirill. Really appreciate it.
Kirill Eremenko: 01:04:22
So, there you have it everybody. Hope you enjoyed this podcast as much as I did, and got some valuable take aways from it. My favorite part was how Theunis explained process digital twins versus system digital twins. And especially what he said at the end about career guidance. If you’re interested in the object and the asset or the design, the materials behind something, then you’d probably would be more interested in system digital twins, or if you’re interested in the business results and applications you might be more interested in the process digital twins. And of course there were plenty of other interesting take aways.
Kirill Eremenko: 01:05:04
As always you can find the show notes at SuperDataScience.com/421. That’s SuperDataScience.com/421. There you will also find the transcript for this episode, any materials that we mentioned on the podcast and links to the company where Theunis works, in case you want to talk to them about working together in your organization or maybe you’re looking for a job. Theunis said they are always on the lookout for talented and passionate people.
Kirill Eremenko: 01:05:31
And of course you’ll find the URL for Theunis’ LinkedIn, make sure to connect with him too. And yeah, so that’s us for today, if you enjoyed this episode and you know somebody who’s interested in this space of simulations and digital twins, or in general would like to broaden their knowledge of technology in this fourth industrial revolution, feel free or we would really appreciate it, if you send them this episode and help spread the word. It’s very easy to share, just send them the link, SuperDataScience.com/421.
Kirill Eremenko: 01:06:02
On that note we appreciate your time, and you being here today and look forward to seeing you back here next time. Until then, happy analyzing.