Thank you very much, Viral, for welcoming me here. I'm absolutely excited to speak here today. And the topic is on digital twins, which is a very, very interesting, exciting topic, very interesting from an academic perspective, interesting from a business perspective. And originally when I signed up, I was going to give a talk as a CTO of Ansys, but now three days ago I joined the executive leadership team of Synopsys. And so I'm going to speak, so I had to change the format of my slides to Synopsys last night.
So the agenda is I'm going to first talk about the Synopsys and Ansys introduction because it's a really interesting thing that is going on in the industry, and I want you to know why these two large companies got together, right? And then dive into the technical topic of digital twins, how it is relevant to various industries like automotive, aerospace, energy, and so on, and then go into the core technology. Right, there are two broad approaches: data-driven approaches and physics-driven approaches, and sort of the hybrid approach which we do. And then I'm going to talk about some of the key innovations that are going on in this general area of digital twins.
And then I will end my talk about the future of digital engineering, which is, I hope I can get to these things because I'm painting a picture of what's out there, right? And again, I'm honored to have people like Alan and Bob in the audience because this is the picture that I'm painting on behalf of the Ansys-Synopsys documents. We are all actually going in that direction, so I want you all to see where the world is headed. And I will tell you up front, JuliaHub is going to be a very, very important part of that journey. So that's basically the thing: the future is digital twins, digital engineering, and what this community is building, Julia, is going to be a most important part of that journey. So with that, let me begin with what just happened last week.
So 18 months ago, a year and a half ago, Synopsys, which is the leading electronic design automation company in the world with about six billion in revenue, decided to acquire Ansys, the leading simulation company with about three billion. The question is why in the world of electronic design automation, right? When you're designing electronic chips like Nvidia's Blackwell GPU chip with 300 billion transistors, that requires electronic design automation. I mean, when I was a professor at the University of Illinois in 1985, I used to teach students how to do custom IC design, where on an IC you have masks. You have a red rectangle which is polysilicon, and when you draw a red rectangle polysilicon over a green rectangle which is diffusion, it forms a transistor. Those are the two mask layers that if you do it now with that technology, you could design maybe 100,000 rectangles, so 10,000 transistors.
So early part of VLSI design was computer-aided design. You had to design the transistors, the layouts, and the chip. Then the world of automation came in where you essentially took standard cell designs like a two-input NAND gate, four-input NAND gate, and you entered your design not in drawing rectangles, but you enter it at the gate level, and then you use an automation tool like a placement tool, a routing tool to connect it up, and that's how you generate the rectangles. Then the technology went to RTL synthesis using technology like Design Compiler and say, okay, now enter it at RTL, VHDL, Verilog languages like Julia, right, and push of a button gets the netlist of gates, push of a button gets the rectangles. And then there was high-level synthesis, which is what I did when I was at Axell Chip, right, taking high-level languages like C or MATLAB and generate the RTL, VHDL, generate the netlist, and so on.
So this is the world that Synopsys has been leading, right? All these semiconductor companies, the chips when you design them from Intel, from AMD, from Nvidia, from Broadcom, Qualcomm, so the top 200 companies that are designing chips, they use Synopsys or Cadence. Those are the two big companies that do it. So that's the world of electronic design automation.
In the world of systems, when you are, I'm sure you came driving a car or flying an airplane or powered by the energy of a wind farm or a solar cell, or you are using some kind of a pacemaker, I hope not, but if you, so in all the industries like automotive, aerospace, energy, industrials, high-tech, and healthcare, that world uses, you need to design and you need to simulate. Ansys is the leading simulation company.
And so the combination is, it turns out that on a Boeing airplane, right, that this now it is no longer only the airfoil and whatever, right? There's a lot of software that is driving the airplane stuff, right? On a Tesla car, there's 100 million lines of code on a Tesla car, right? So it's a hardware-software combination ecosystem. So the world is moving towards hardware-software systems design, right? And while Synopsys was focused on only the design of a chip, Ansys was working on the system. Bringing it together, we are now going to help customers build silicon to system. So that is literally why Synopsys paid a lot of money, $35 billion, to acquire Ansys last Thursday.
So because of it now, essentially you have the number one EDA company, which is Synopsys, number two is Cadence, number three is Siemens EDA, formerly Mentor, combined with silicon IP. When you design a Blackwell chip, right, you don't design all 300 billion transistors from scratch. You say, hey, I need an HBM design, I need a PCI Express, I need an Ethernet controller, Bob, all the technology that you build, right, these are predefined IPs. I need a GPU core, I need a CPU core, I need an ARM, all of those things you put together, right? So Synopsys is actually the number two silicon IP provider. Number one is ARM, number two is Synopsys, and number three is Cadence.
So, and when you're designing these chips, you need to have the electronic design automation and you also need the IP. Now you combine it with systems. This is a very, very interesting combination. Ansys is today a $9 billion revenue company. We had our executive leadership meeting yesterday and we are literally going to change the world to do silicon to systems. This is our vision. So in that vision, you will see that Julia will play a big role and hopefully I'll get to that point.
So now the technical part of the talk. So this is about digital twins. What is a digital twin? You have a physical asset like an airplane, right? This is the thing that is there, right? Or the engine on an airplane or the car or the engine in the car, whatever is a physical asset, that's a physical asset. And then you have a digital model of that thing, right? In the digital world.
And so sometimes people say, oh, a digital twin is just a simulation model. No, it is a simulation model, but that model is connected to the physical asset through, I put some sensors on that asset, right? And I collect data through an IoT platform, etc. And there's two-way information flow that happens. So you say, here is an aircraft engine, it is getting overheated. Oh my god, it's getting overheated. What if I could cool it down? I put cold water here, I do oil there. So I do all these what-if scenarios. So I do those what-if scenarios on the digital world and say, oh, I should put a mixture of oil and vinegar or some random thing, right? And based on it, you make the decision, you do it, and you do it in the other part, right? That's what a digital twin is good for.
And those of you who have, who I know have watched the movie Apollo 13, that disastrous thing that happened, right? When people were stuck up there, right? NASA created a digital twin model of those things and say, oh shoot, how am I going to bring those people back? I put oil, vinegar, this, that, etc. Did all those things on the digital model. If I did this, we'll bring the astronauts back. So the concept of digital twin actually came out of NASA, right? So that is the origin, but this is kind of what digital twins are.
So digital twins allow you to track the past, deeper insights on the present, and predict the future. If we can say this airplane is going to crash if you don't do certain stuff, this thing is, so if you could predict that with high accuracy, you can literally solve the world. So it turns out that digital twins, even though it got started with NASA stuff, right, is now used in aerospace, that aircraft engine like GE basically they do predictive analytics of that airplane engine, right? These are normal things.
You all own cars, right? You are supposed to do tire rotation every 5,000 miles or kilometers, right? Why do you do that? Because statistically speaking, if you drive your car normally, right, you are doing this, that, where the wear and tear of the tire will be such that if you don't change it, you'll have a tire burst, right? You have to do an oil change every 5,000 miles. Statistically speaking, if you don't do it, mess up. But practically speaking, if you look at Priit, I drive like a madman, right? My tire, if I don't rotate it every 4,000 miles, if I don't change my oil every 2,000 miles, my car is going to explode.
This digital twin on Priit's car will monitor the oil pressure, my tire things, etc., and say, hey, it's time for an oil change, time for whatever. And if I can do this accurately, instead of that $400 million airplane crashing, I will prevent that crash. So that is why having accuracy of prediction of a digital twin in aerospace, automotive, manufacturing, buildings, oil and gas is so important.
Now digital twins has been used in different phases, right? So in the product design phase, in the manufacturing phase, in operation phase. Let me give you an example. GE engines, aircraft engines, you all fly planes, Boeing planes, etc. You have those little engines there, right? The blades going around and so on. So when GE engineers design that engine, right, there are 52 blades on the engine. They have to have exactly that curve. So it's a perfect ideal CAD model of that blade.
Now you try to manufacture the blade using injection molding or whatever, that blade, which is such a beautiful picture with a circle and spheres and so on. When you manufacture it, it's a slightly different blade, right? So now you have a digital twin that designed ideal world, now you manufactured it with all the imperfection. So every engine of GE in the actual manufactured world is slightly different, right? So literally if you could track every one of the engines, and that's what they'll really do with some sensors. Here is how I designed it, here's how I manufactured it. I can track how that engine will work.
And now this plane is an Air India plane flying from Mumbai to Ahmedabad to whatever, and all the smog and whatever flying into China doing this. So that same aircraft engine on, you know, airlines only flying between San Francisco, New York, clean places versus sort of Mumbai and Beijing, right? That airplane engine is quite different after 20 years. And then a bird strikes that engine, right? So now that engine is slightly different.
So in the operations phase, even we started with a perfect ideal world of a CAD model, in the operational world it got all messed up, right? Because there's a little bend there. So I need to predict through the digital twin the model of that aircraft engine in the design phase, in the manufacturing phase, and so on. And you can do it beautifully through the world of simulation powered by JuliaHub, which is where I'm getting to.
So now there's sort of two broad approaches to digital twins. One is you say, okay, here is an engine, right? And that I would put some sensors on the engine. I'll just measure the pressure, temperature, whatever, etc. on that engine, on the blade. If I were to put sensors on everything, right? And collect a lot of data.
And so if you collect a lot of data, it's just that time series data, pressure, temperature, etc., etc., and you build a model, all the AI stuff that you see today, right? It's basically doing that data-driven modeling. That approach is totally general. It doesn't care about the physics. You do structural physics, you do electromagnetics physics, you do fluids physics, everything can be captured by data. The trouble is, so it is very general. The trouble is it requires an incredible amount of data to train that model, right? It's like millions and millions of times of data.
If you capture it, then you will predict exactly how that aircraft engine will work. The other trouble is it will only train it based on what it has seen. So if it has seen its aircraft going from San Francisco, New York, San Francisco, New York, going from this to this, etc., it will never be trained to go into the streets of Mumbai with all the smog and so on, or whatever in Beijing, right? Because it has not seen that data. If it has not seen a bird strike, it will never be able to model the bird strike.
But the bottom part is if you do simulation-based modeling with Ansys Mechanical, here is this model of an airplane wing, here is a bird that struck it, right? With LS-DYNA, I can model exactly how much it'll bend and whether that blade got cracked or not. It's physics-based simulation. So that is, I mean, it's like the ultimate thing, right? Because physics doesn't lie, right? The world around us is governed by the laws of physics.
If I say this plane is going to fly, it will fly. If Fluent says this thing is going to fly, it is going to stall, it will stall. I mean, I kid you not because it is so accurate because the world is following second-order partial differential equations, Navier-Stokes equations or Euler equations or Maxwell's equations. That's what we do in the physics-based. The trouble is that takes a long time. You try to model a bird strike, right? This thing, bird is striking this thing, it's like 10,000 hours, right? Oh shoot, I needed the result every second, I cannot.
So then what you need is you take that simulation model, you have to build what is called reduced order models to not do the simulation at the level of the 1,000 million finite element method, but you reduce those orders from 10,000 million to a thousand so that you can run it really fast. So there are these two broad approaches, and where we have landed is you need to do hybrid. Take the best of both worlds. Do data analytics to do the sort of easy cases and then correct it through simulation, right?
So the intuitive explanation is the following. Suppose you had a bridge, right? And there's a six-inch crack on the bridge. Statistically speaking, of all the bridges in the world over the last 30 years, right? When you had a six-inch crack in the bridge, within 18 months that bridge collapsed. Statistically speaking, that's what our insurance companies, State Farm, says.
Hey, based on all the people who are driving cars, right? If you are male, you have this risk. If you are female, you are this risk. If you're a male who is crazy like Priit, you have this risk. So they're putting it in a bucket of a risk, right? That's statistics.
So that bridge with that six-inch crack would statistically speaking crack after 18 months. Now if you could put a sensor on that crack and you monitor the crack and now you use a simulation to do crack propagation. If I have a six-inch crack on this bridge and I have a thousand cars going on the bridge with all the stress strain that is going on, I will predict that crack will be 18 inches in six months. Pure simulation.
Now I actually monitor what happened to the crack in six months. If the crack is 24 inches, I say, oh shoot, I have under-simulated. So I need to push it, right? Push that simulation. Or if the crack was only 12 inches, I need to over-simulate. So with simulation I predict that and then with actual measurement I bring it down. That is intuitively what is going on with hybrid. There's a lot of math, which the famous professors here will know, but I'm just giving you the intuitive explanation of why hybrid digital twins work.
So we have sort of put together at Ansys, now Synopsys, this overall view of digital twin. So you have to start with the assets. The assets are the wind turbine engine, the car, the aircraft engine. You name the asset and you put some IoT sensors on the asset. You collect a lot of data. You put it on the cloud on AWS or Microsoft, whatever, right? They need to do data-level analytics, which is what, as I said, most people when they start doing digital twins, that's what.
When I was CTO at ABB, CTO at Schneider, this is what I did to those companies, right? They have a lot of assets, robots, switches, transformers, and so on. I just created this Ability platform or EcoStruxure platform, build the digital twins so ABB could say this robot is about to fail and let me send a repair person to fix it. And if there's an accuracy, then I came to Ansys and say, hey, let's do simulation-based or hybrid and we'll do the accuracy.
So what good is this digital twin? What it is, is it allows you to create virtual sensors. So if you could put a sensor on every part of the blade on a GE engine, right? There's 52 blades and on that blade you got to put it here, here. You could analyze the heck out of that engine. What is going to happen? The trouble is if you put all those sensors, that engine will not work as an engine, right? Because you're doing an invasive change in that thing, right?
Or you have a large oil well, right? 5,000 kilometers of oil wells and so on, right? And you're putting a sensor at every meter, every inch, every whatever, right? You could do all kinds of predictions. Unfortunately, that is not real. So in the world, what you can do is you can do virtual sensors. Meaning just imagine a pipe. The pipe has oil coming in from this side and the pressure, temperature, velocity of that thing is here.
Now, if I had a sensor after one meter, it would measure this because through fluid simulation, I can predict because hey, this is a pipe, this is a fluid flow, etc., with all those turbulence and so on. It has this pressure temperature here. If I had a sensor, this is what it would measure.
But I cannot have a sensor. So I'm creating a virtual sensor at the first meter, second meter, third meter, fourth meter, fifth meter. At the 100th meter, I actually put a sensor to say, hey, did I do it right or did I screw up? No, you screwed up. You need to correct it. So that's how you do virtual sensors.
So to measure missing data, to perform what-ifs, right? So here is this Apollo 13 sort of up there screwed up, right? How am I going to bring it down? I have to do, if I did this oil and vinegar plus this, then this is going to work. So all those what-if scenarios are due in the digital twin, digital world, and then based on that I make a decision, apply, do very accurate predictions.
Hey, this bridge is going to crash. Today's Wednesday, July whatever, 23rd. It will crash on July 28th. If you could predict that so well, my God, we could make a lot of money because you could avoid a lot of disasters and explore the, hey, this thing, when this Air India plane crashed, right? And you have heard these things, right? Why did it crash? All these things happened. You could literally do physics-based stuff to do all kinds of analysis because the black box recorders have measured, then you combine it with physics, you could do amazing things. So that's where the power of digital twins.
So now let me talk of the technology. So Ansys has the core physics. We have the leading mechanical solver called Ansys Mechanical, LS-DYNA. We have the leading fluid solver called Fluent. We have the leading electromagnetic solver called HFSS and Maxwell. So we take each of these physics and we can simulate it with the highest fidelity. I mean, if I say this is going to crack, it is going to crack. I tell you not, right? But it requires 100 million finite element points, right? It will take 10,000 hours. The trouble is that is too long.
So what we have a technology is we have this reduced order models. So think of if a one million finite element problem, right, basically it gets resolved into sparse linear solvers. It is a one million by one million matrix with sort of some sparse elements, and now you have to solve Ax=b. That Ax=b is an order n cubed computation, right, which is taking you forever. Now looking at that matrix, you can, you going back to your math days of numerical linear algebra, right, there are these things called eigenvalues and eigenvectors and principal vectors.
On that matrix, if I were to take the most important 100 values, the principal components, that's 100 of those, those things are the most, I'll take those one million by one million vector matrix into a 100 by 100. Those 100 by 100 are the most important things. That's fundamentally what is going on with reduced order models, right? Now instead of taking order n cubed time, order one million cubed, I solve order 1,000 cubed, which is much easier to solve. That's basically what is going on with digital models. So we take all the core solvers, build the reduced order models, and do all these funky stuff with hybrid digital twins.
Now we have a flow. Now there's, within ROMs, there's all kinds of ROM technology. Again, this is just to impress you. Some of you are in the audience. Hey, Priit is just not doing nonsense. There is technologies on linear ROMs, nonlinear ROMs, static ROMs, dynamic ROMs. Okay, I have impressed you.
So now we have built this thing, the Twinbuilder. So this just incredible amount of stuff that is going on today, and that's where Julia comes in, right? So we have to take this model of an electric vehicle or an autonomous airplane or this wind farm. So this really complicated systems having hardware and software in it, right? And this combined Ansys-Synopsys stuff, right? And you have to do this model building for either brownfield application, you have a stuff that is there, or for greenfield, right? How do you build the model? Well, I will come to that.
It is actually JuliaHub is a fantastic thing that we are looking forward to partnering with, right? Allowing reuse of existing simulation models with AI/ML. I mean, AI is actually changing the whole world with ChatGPT and others, but it is really changing the world in our area of AI/ML applied to simulation, enabling scaled runtimes, and driving a lot of ecosystem partnership with the likes of AWS, PTC, etc. So there's stuff that is going on.
So let me just focus on the system-level simulation part. So let's say you're trying to design an electric vehicle, right? In an electric vehicle, you will have a motor, electric motor. You can actually model it in Julia, which is a high-level system-level motor. Now that electric motor is powered by DC power. I have AC power. What do I do? I take the AC into DC. So I need some converters, or I take a AC into DC inverter. So I do all this. I'm sorry, DC to AC need inverters. So I need a battery, I need a motor and inverters.
Then you need to have something that once the motor goes, that motor will connect with a gear and then it will connect to the actual things and that's how the car moves. So you could actually go in and model the motor, all of them in Julia and build a whole electric vehicle simulation model at a high level.
But then you say, oh my God, this electric motor has these rotors and stators and all that kind of stuff. Will it really, if I apply this much voltage, my model says that torque will be this much. Is it really that much? Well, we go into Maxwell, which is Ansys, right? And we look at the actual CAD model of the stator and the rotor and all those fantastic Maxwell's equations with all the partial differentials and said the torque will be 3.212.
In Julia, you said 3.1. No, it's 3.212. But it required now 1,000 hours of compute to get to that accuracy, right? So what in this world, what you need is you need a system-level model to do this kind of stuff, but you need to go and validate with the detailed models, the 3D models. That is, and if you can bring those things right together, that is sort of the interesting thing.
Now in this world, for the last 100 years, 100 years, 30, 20, 30, 40 years, the leading system-level language has been MATLAB and Simulink, right? But as you guys all know, yes, they are literally the preferred things. You go into any automotive company, they will use MATLAB Simulink as the model-based design thing, right? Then they will, after they pick that motor, then they'll go into an Ansys or a Dassault or Siemens to actually with the 3D.
But with all the advantages that you guys are working on with Julia, right, with the compile stuff and so on. Thanks, Alan, for all the great work that you have done at MIT and the work that you guys have done in JuliaHub. This is where we at Ansys are looking for the partnership, right? So the Ansys Twinbuilder, this system. Samir, are you here in the audience? There you go. This is the man. He is the GM for our Twinbuilder.
He is working closely with the JuliaHub team to say, hey, this thing, can I integrate with Ansys and build this system-level model, right? And which would be better than MATLAB, better than Modelica, better, all that is sort of what we are working on, then building all these things with ROMs, scale.
I actually need to speed up because I really need to cover some of the other stuff. Okay, so there's other stuff. Sorry, Samir, I'm skipping through some of that stuff because I need to, they'll catch you after for the details.
So then the thing is we are working on an ecosystem where these digital twin models, you can actually visualize it. So those of you who have seen Jensen's keynote at GTC, he talks about the Omniverse. This is this visualization platform. You have a manufacturing thing at BMW factory, whatever, can see all those things. So that's that powered by GPUs and so on.
So what all these simulation models that we have from Ansys powered by Samir's twin is now you can visualize on the Omniverse. And so you can actually have the physical asset, which is this whatever thing it is, right? Is this the electromechanical stuff? Here's a physical stuff, and you can actually see it on the Omniverse, and you can predict, hey, if this has happened, I put oil and vinegar, this is going to happen, it'll show up on the picture.
So imagine if NASA had this Twinbuilder and Omniverse in 19, whenever Apollo 13 came down, it would have done amazing things. But we are here today, and I hope there's no more Apollo disasters.
Then there's this technology that Samir has been cooking on this thing called Ansys Cosim, right? Basically you have a multiphysics simulation stuff, right? You have multiple things and you have to simulate a distributed system. Some of them are with sort of continuous time, some with discrete time. It is this really complicated stuff, and so we have a technology of Ansys Cosim which can do this massive distributed simulation together.
And I know I'm rushing up because I want to cover the future of digital engineering. So all these things happen, customers love it. Sorry, Samir, we have used it in Tata Steel, we have used it in glass manufacturer. So this thing actually does work. What I really want to talk about is the future.
Because this is the academic audience, right? I just want to give you a peek of where are things headed. Where it's headed is in the industrial landscape. We, you all know, we have gone from Industry 1.0, which is mechanization, to Industry 2.0, which is electrification, Industry 3.0, which is the automation and the electronics. Industry 4.0, which is cyber-physical, IoT, sensors, etc. We are in Industry 5.0, which is the AI and human and all that kind of stuff.
The technologies that are pushing us in that area is AI. All the ChatGPTs and Claude and Anthropic and GenAI, the blockchain, the digital twin, the metaverse, Omniverse is Nvidia's name for metaverse, quantum computing, big deal, right? And 6G, right? With all the IoT connectivity. So that's kind of where things are headed from the Industry 5.0, the actual end users.
So in that world, the world is moving towards, if you look at how people changed from building enterprise resource planning, ERP systems, right? That was a big change that happened in the 80s, to supply chain management systems, to PLM systems, to model-based systems engineering. The world that we live in today is digital engineering ecosystem, right? So in the world of digital, I'm designing an airplane, right, with 100 million lines of C code, etc. And you will predict this autonomous electric airplane will fly, right?
Boeing is trying to design the 9, sort of 787 Dreamliner set, the future versions, right? The entire thing, the specifications will come in in languages like CML, and the output will be a beautiful thing that will, the airplane, right, fully validated and will run. So in that world, this is kind of the picture of digital engineering. And if there is one picture you want to take, it is this. Yes, I'm, oh, I take it in with me there, right? Priit talking about future of digital. Yes, and so in that world.
Remember I talked about the world of synthesis, right? Synopsys transformed the world of EDA to do electronic design synthesis. I mean, instead of drawing rectangles on the screen, you enter it in RTL, VHDL, Verilog, out comes gates, place and route, etc., etc., and do the rectangles. That is electronic synthesis.
Can we do system-level synthesis? Once you do that, can you virtualize? Can you validate? Can you do this with the physical assets, the IT integration? Can you do the digital operations and have the embedded digital twin? So this is the whole journey, and I have one slide on each of them just to give you an example, right? So imagine...
I come up with a system-level requirement that I need to design a Tesla beater. I am designing an electric car that will go from 0 to 60 in 1 second and a range of 1,000 miles. I mean, if I could build that car, Tesla will be out of business, and the cost is like less than $20,000. So that's my requirement. If I could specify a requirement like that, can I synthesize a design? So if you were to work in an automotive company, right, you say, oh, I need to go a range of 1,000 miles, right? My last car, electric car that I had, could go only 200 miles, and I had a battery of this size. So if I have to go three times as much, I needed three times larger battery. That's how a human thinks. I need to go from zero to 60 in one second, right? My previous thing that I had went from 0 to 60 in say six seconds. So I need to go six times faster. That one was with the 200 horsepower battery. Now I need a 1,000 horsepower battery. That's how a human thinks.
But today the human is making every one of those steps, saying, oh, for this specification, I think you need a 1,000 horsepower motor with a whatever size battery with that gear. So the human is choosing all those steps. What if automatically I could synthesize all the stuff that is in the human's head, right? All the knowledge is captured, and this is kind of what we are working on. So you do all these things, you say this is the EV car, and I have actually shown Viral a demo of this, right? There are six different designs that automatically came in. Then with JuliaHub, you are actually simulating all of those. Say this is one case of this combination of the electric car which has the motor with in-wheel drive, with all the motors in all these wheels in parallel, that is going to be the architecture, right? Or it's architecture with only front-wheel drive, rear-wheel. So all the choices that a human designer does, automatically, and that's system-level synthesis.
The second one is virtual validation. You are designing this car, right? It is going on the streets of, I mean, driving from Los Angeles to Las Vegas, right, in the heat and so on. It's just super hard. And then the thing is going over these things, right? So how do you actually test, virtually test before you build that car that it will actually work, right? So the normal way that you do is you do the unit testing, test the motor, test the battery, test the gears and so on, and then you do physical testing and do the combination of physical testing and so on. The future is all virtual because if I am saying if the world is governed by the laws of physics, if Navier-Stokes equation doesn't lie, if Maxwell's equation doesn't lie, you don't have to build a physical prototype, go to zero physical prototyping, right? And we have actually done things with digital twins. The picture is, if you could show those videos, it actually shows here is a car going right from Las Vegas to whatever, really hot things over this highway, and is there overheating happening or not, right? So you could actually validate.
Then you have this IT integration, right? I'm just going through this, right? You have all these different things, and you put sort of sensors, you put PLCs, etc., etc. So there's the OT part is having all these Modbus, OPC UA, all those standards that you have, and then you connect, use all the latest Ethernet technologies developed by Bob here, right? And this is the world, the world is MQTT broker, etc., etc. That's how in a real system you are actually building these things with IT integration. Then you did the whole digital operations. I'm flying this airplane from New York to Kathmandu or New York to Mumbai, right? And I will do all this thing, right? This is going, this plane, etc. I have all the sensors, so doing this whole thing with digital missions engineering, digital twin, that's sort of digital missions engineering. The last thing is digital twin. This is kind of where somebody is headed with this stuff.
Right. So we did a little interesting project in this Covisa joint work, right? With companies like Bosch and all. So imagine you are driving an electric car, right? And you say it's like only 30% charged. Oh my god, if I don't find a charging station by the time this thing goes, right, I am toast because I am on the highway. I need a charging station. So this is called range anxiety. So let's imagine that I am a driver and I have connected to all the power and electronics and AC and all that kind of stuff, right? All the things says I have to go only 30% things, and with that I could drive only 100 miles, but my charging station is like 200 miles away. What the heck do I do? If you're like me, say, oh sure, come on, turn off the lights, turn the AC. So you do all those things, right? This thing would say, 'Prit, just relax. Don't turn on the lights. Don't turn the AC. You just reduce the AC from 75 to 42 and you'll be fine.' And you're completely crazy. You just go in with the last amount of juice, go into a charging station, you're done. If you could predict it, well, that would be amazing. So it's an example of an embedded digital twin. I thought this was an entertaining morning.
So what I talked to you about is the concept of digital twin, which can be done with pure data analytics or with simulation, and I think the world is headed towards hybrid. I am biased. I think we are well positioned. Samir and I, we agree on that. But the future is about this digital engineering area, and this will literally be powered by JuliaHub. I kid you not. So we are excited to work with JuliaHub and all of you in the audience. So thank you very much for your attention. Thank you.
