Digital Twin for the Digital Win

Learn more about the Executable Digital Twin (xDT).

The transcript below has been edited for clarity.

Jim Anderton: Hello everyone. And welcome to Designing the Future. Today we’re going to talk about how the digital twin will change the way the engineering design and development process operates in multiple industries. Joining me is Tommaso Tamarozzi, Senior Research Engineering Manager in the Simulation 3D RTD group inside the simulation and test solutions business of Siemens digital industry software. His team of application experts, software developers, and researchers work on new products and technologies, including advanced numerical strategies and executable digital twin solutions for advanced smart sensing and reduced order modeling. Tommaso, thanks for joining me on the program.

Tommaso Tamarozzi: Hi, good evening.

Jim Anderton: Tommaso, just to begin, for those viewers who are watching who are perhaps not as familiar with digital twin, can we talk about what digital twin is? For centuries engineering was about just imagination, transferred to a rendering, a bill of materials, and then documents, piles of documents. In the ’70s and ’80s we went to computer aided design, so we digitized the process, but it was still much the same conceptually. Digital twin is different. Tell me how?

Tommaso Tamarozzi: That’s actually a very interesting point of view, the one that even if there was a digital transformation partially already in the ’70s and the ’80s. Well, the thing is that the process was very similar. That’s entirely right. And a lot of the old habits, some of which are effective, some of which are actually not, have still continued. And if you look at the digital twin, this changes the concept of having piles off of documents, even if digital, to instead a concept that is more, I would say holistic, because a digital twin is a precise visual representation of a physical product or even of a physical process as a whole.

And you can see it as a collection of a lot of different information, everything that forms the digital footprint of a product, of a physical product, or process. So, there is the bill of material, there is CAD models, but there are also numerical models, which can be very simple up to very complex. And the most important thing is that it is a leading concept. It’s something that adapts itself during the life cycle of the product. So, it lives together with the product.

I think it’s actually really like if you would have a digital replica of yourself, if you want, if it would be a digital twin of a human, you could see it as a collection of information about how you look like, what’s your behavior, how your heart interacts with your brain. And of course, what’s your status, what’s your health.

And this of course will evolve. And the digital twin, the information that this digital twin carries, needs to evolve as well with the physical part.

Jim Anderton: In the traditional document-based paper-based world that I’m referring to, the rendering of a part or an assembly or a process, for example, the applied mathematics is buried behind. It’s not immediately apparent in the object or the rendering that you’re looking at. And to find it, to backstep your way through a process to find how we arrived at a solution, is often difficult or impossible or involved having to go to two different set of documents or different set of experts even. Are we talking about something that sort of fuses these different aspects?

Tommaso Tamarozzi: Yes. It’s a sort of glue that allows also people who are not experts in these fields like numerical modeling, for example, to have information and gather information that is useful for their work, despite not being expert in that field. It allows people who have no background in, I don’t know, computational fluid dynamics, to still be able to use the results of a computational fluid dynamic tool for the usage that they need. It would allow sometimes users even to create these models themself because the creation is automated in a way that even non-experts can use it. Of course, up to a certain level where some experience can still be needed but the overall goal of the digital twin is really also to democratize the usage of complex models to a broader audience and for multiple usages. So yes, absolutely. You can see it as a glue, I would say.

Jim Anderton: Tommaso, I’ve noticed in my experience that engineers are trained to think in Cartesian ways is that we think of a resolution of forces. And as a result, we think in those very rigid Cartesian coordinates, and that is sometimes reflected in design in some cases where it’s easy to conceptualize things like forces, moments, even in electrical world, things like currents, voltages. And when we look at highly successful designs in nature, for example, that have evolved over millions of years, they’re much more organic and fluid. And still individuals like myself of a certain age think of those complex forms as nothing more than those Cartesian axis shrunk to a very, very small mesh, a very, very large mesh. Is that mindset, is that old mindset still a factor in the industry, do you think?

Tommaso Tamarozzi: In the industry in general, or do you mean more related to digital twins?

Jim Anderton: Tell me both.

Tommaso Tamarozzi: Yeah. Well, for example, if you look at, from a digital twin point of view, a lot of the information that the nature provides naturally is integrated in the digital twin idea and concept in a form or in another. And a very interesting concept of digital twin is when you are able to connect the digital twin to real sensor data, which represent potentially some of your physical assets, as you said, in a more fluid way, in a way in which it’s more naturally evolving and able to actualize your representation, which would be maybe your Cartesian coordinate representation, as you mentioned, by using the sensory information and make this representation look more real and give you information that are closer to reality. I don’t know if that’s what you were looking or what you are hinting for, but there is this parallel between the real physical and more fluid world and our simplifications of how reality would be and should be.

Jim Anderton: Well, you mentioned CFD. And of course, it’s one of the things that most of us learn in the process, the education, is if you’re thinking about things like moving fluids, through surfaces, across surfaces, laminar flow is desirable, turbulent flow is not desirable. However, more sophisticated analyses of course have shown us that in some cases, a little bit of turbulent flow actually is highly beneficial to moving fluids in some circumstances, hence, that the skin of a shark is not perfectly smooth. And similarly, the airplane wing, the Formula One car aerodynamics – you have a Formula One car background in Ferrari, I understand. Is digital twin a way in which you can get individuals to think beyond those rigid thoughts about rules that say, “we must go in this direction,” when the mathematics perhaps suggests something else?

Tommaso Tamarozzi: Yeah, it helps to break these schemes. Especially it helps breaking these schemes because it allows you to perform and test different scenario into a virtual world that allow even the more skeptical engineer to be convinced about the fact that some of these counter-intuitive, designs that, for example, nature would provide and show are actually extremely effective. Even if you look in the aero industry, for a long time wings of planes have had their wings slightly curved up uphill, and that comes from CFD, but mostly it comes from looking at eagles. So, from the fact that that eagles would navigate by lifting their feather.

And I think the combination of looking at nature and combining this intuition that the engineer still needs to have with extremely effective computational methods gives more confidence to the engineers that it’s not just your gut feeling that tell you that you need to go there, but there is a solid mathematical reason to go in that direction.

Jim Anderton: Tommaso, we know that FEA, CFD, combined with today’s computers which are very powerful, allow us to sort of approximate our way to very, very good solutions, solutions that engineers can have great confidence in. You mentioned the aerospace industry. That’s an industry where it’s still common to build a model inserted into a wind tunnel, and then blow air across it, and measure pressures.

Is digital twin, in that example, a way in which you can do away with much of that actual physical experimentation?

Tommaso Tamarozzi: Yes. On the other hand, I think that the combination of the two is extremely important. I come from a world in which test engineers believe that their sensor data is ground truth, and simulation engineer believe that their accurate simulation is ground truth, and it’s difficult to have the two convince each other that none of them is right or both are.

So, in fact, if you look at it, you can have the most complex model you want of a certain object, but if you look at it in today’s complexity…

First of all, there is so much change in each product and customization that is extremely difficult to have a good complex model for each individual piece that you do in a reasonable time, unless you do it properly.

Secondly, even in a production, not every car is the same, not every airplane is the same. So, your results of your wind tunnel would only test one of these assets. And the combination of the simulation data that you would get from the operational life of an aircraft or of a car, linked to this advanced numerical method to make sure that these digital twins are actualized, are brought to a level of fidelity that you are really looking at a digital replica of that specific car and not just of a car that is representative. That really, really makes the difference.

If you look at, from reducing testing, that’s also true, but I think that you should look at it from the perspective of, with the same amount of testing I was able doing before, now I can get 10 times more information.

I can go much deeper and much broader into what I can look to. I am not forced anymore to look at point sensors, where I measure an acceleration, or where I measure a strain, or where I measure a localized force, but I can know all about that specific asset that I’m looking to. So I think that’s extremely powerful.

Jim Anderton: In the production automotive world that I come from, time is always a major constraint. It would be lovely to have an infinite amount of money to test, but also more importantly, it’d be nice to have more time to test and sometimes developing a new design or a new product or a new concept was limited, not by creativity or even not by technology, but by the lack of available time to completely prove it, to place it into production for a specific deadline.

Is digital twin, is that one way to defeat this time problem?

Tommaso Tamarozzi: Yes. That is actually for really one of the key goals that we have, especially in the design and engineering world.

There is several ways in which you can achieve it.

If you look at complex testing, you mentioned different fields, but it can be in automotive, but it can be also in the aerospace industry, in any industry, a big part of testing, actualizing the instrumentation, in the calibration, into making sure that everything works fine.

Then, you do your test, and you have to analyze the data.

Now, imagine that time to build that test setup could be cut in half or a third, because instead of instrumenting 500 sensor, you can instrument 10 of them. And all the rest is given by the connection between these physical sensors and the model that you have of that assets. That really splits, reduce the time tremendously.

Also, another aspect that you can look into, if you look at reducing time, is that many times, if you have at least – I give you the example in the wind industry, for example. That is an example I’m familiar with, but I’m convinced that it’s probably the same in the automotive.

For a certain vehicle, or in the wind turbine example, for a certain turbine, you need to do multiple type of tests. You need to test static, you need to test dynamics, you need to test durability, you need to test remaining useful life, and things like this. And all these tests require different setup. Imagine you could combine them with just a single setup that contains the key information for the digital twin of your asset to then expand it to all the rest you need. This would cut time tremendously. And it does.

Jim Anderton: Time is an important factor.

Also, you mentioned things like durability, service life. Historically, engineers are quite conservative. Often safety factors are very generous in designs and intuitively, even though we protect ourselves with a fat safety factor, the reality is we’re leaving cost and profitability behind when we do that. We’re leaving something on the table.

Is digital twin away to perhaps operate with narrower safety margins, but still have assurance that the product itself will perform?

Tommaso Tamarozzi: Yes, yes. You’re absolutely right.

It still happens in a lot of times, even if you have this complex simulation and algorithms behind that. 

You reach a result.

You say, “This should work.”

And then you say, “Yeah, let’s just multiply everything by two, so you’re safe.”

And that “multiply everything by two” is a big problem, because it leaves a lot of money, a lot of money not made out of a certain product that would be much more valuable. And not only that, now in terms of green energy, just multiplying everything by two is not even possible anymore, because rules are more restricting.

Heavier cars, for example, means more emissions, or heavier car means that the already slim range of your electric vehicle is going to be cut in half. Just because you needed to multiply everything by two to be safe and that you can’t do anymore.

So yes, digital twins allow you to thin that safety factor source so that if an engineer goes to sleep, instead of saying, “Let’s multiply everything by two,” you will end up saying, “I will multiply everything by 1.2, still going to be fine.”

Then I think that’s already a great achievement. And digital twins help you do that.

Jim Anderton: Tommaso, we’ve talked about exotic things. In some ways, aerospace, Formula 1 Motorsports, some automotive applications, wind turbines, energy; all these areas are areas where I hear a lot now about carbon fiber composites, lightweight materials, new materials, but these materials don’t operate the way the metals that I’m used to operating with function.

These operate by statistical processes. I recall once telling a passenger on a Boeing Dreamliner, looking out at the wing flexing, and I said that the likelihood that this wing will fall off spontaneously in flight is that is a non-zero probability. It is a very small probability, but it’s non-zero because most flexing carbon fiber structures operate by probabilistic mathematics, not by the conventional Young’s Modulus stress-strain that we would think of the materials.

Is that a place where digital twin can live, in composite materials?

Tommaso Tamarozzi: Yes. The ability to play what if scenarios with these characteristics, it’s key to this.

You mentioned statistical analysis, which is really true for composite, not only for composite. If you look at 3D printing too, a lot of the properties of those materials are extremely difficult to get, and they are dependent on the process. Digital twin can help you a lot, because instead of only having a digital twin of the actual carbon fiber object or 3D printed object, you have a digital twin of the process that makes that component together in the same digital twin of the object itself.

So, what you’re able to do now is to perform multiple analysis in a relatively short amount of time that give you a much more precise idea of what type of statistical variability you will get inside of your product in a near production.

And that zero probability of failure is maybe not going to be exactly zero, but it’s going to be much closer to that than what we already use now, actually, if you want.

So yes, the digital twin there of a process I think is extremely important to help in that respect.

Jim Anderton: You mentioned three 3D printing, and 3D printing of course has opened up many possibilities that were never possible before.

And one future projection for 3D printing that I’m hearing in the industry is the possibility in the future we will see a composition gradient in a part. From the center to the outside with possibly using different materials, which are dynamically blended and injected into the part.

So, it, when you not only have then a form factor, and of course the standard sort of Young’s Modulus calculation, with a composition gradient, which might be quite complex inside the part, the mathematics would be fabulously complex to trya and determine, things like yield strength or pressures is. Do we have the computing power? Can we apply digital twin to complex problems like this?

Tommaso Tamarozzi: That’s a very good question. So, I think you can see it from two perspective. The first is, do we have the computing power to simulate everything we want? The reality is no, we don’t have that yet. I’ve seen simulations going at conferences of molecular simulation that takes thousands and thousands of parallel course to do just a small millisecond of simulation of a molecule moving; on the other hand, we have enough computing power to solve a lot of engineering problems. And there are technologies that are helping us to make sure that complex problems like the one that you described can be solved in a much, much shorter amount of time. And that they occupy less resources like memory, for example, in this space. So that the amount of data that they produce is also still informative but reduced. This field is an extremely important field of research that we are very active with in Siemens Digital Industry, which is called model order reduction.

And as the word says, it reduces the complexity of model in semi-automatic ways. It can use AI methods, artificial intelligence methods, like typical neural networks or deep learnings and things like this, or it can use your physics background in order to make assumptions in an automatic way that allow you to have maybe a slightly less accuracy for a speed up factor of maybe two or three orders of magnitude. And then if you look at it that way, then you can go two or three level of complexity further, and then you’re back to the limit of your computational power.

So the answer is a lot of technologies are being made now to make sure that we can with the technology we have now, not only by brute force and computing power solve complex problem, but also by using mathematics.

Jim Anderton: You briefly mentioned sensors. I know you’ve done work in this field in particular. That’s an interesting subject in engineering in general. The traditional world of sensors, these are devices that generate analog signals and those analog signals are pumped to an analog to digital converter. Those raw digital signals are then sent to a central processor, and then some filtering happens, something happens to the signal, and then we hope that the digital signal is a true representation of what is happening at the sensor.

Today’s generation Sensors are smart. The computation is happening in the sensor. So there’s no real way to visualize the difference between the analog signal being generated by that piece electric crystal or that Thermo couple and the digital output. But they’re also sort of self-diagnostic, in some cases, they’re quite intelligent; but there’s also I sense, a trust gap that can be generated by very smart sensors as well. A sense that “am I sure that this thing is giving me valid data?” But they’re also proliferating in numbers because they’re cheap.

Tommaso Tamarozzi: Yeah, so there are cheaper sensors. There is MEMS technology that have evolved tremendously. There is still computational power, it’s limited, but still computation power really, really close to sensors. And that’s an interesting feel because it still allows you to use much more sensors that are cheaper. On the other hand, the gap in trust it’s real. It’s true. And so, we have ways to handle it and digital twins can help also there.

In fact, what we call a smart virtual sensor is a sensor that takes the information from a standard sensor, which maybe doesn’t have that computing power that you were saying, which is just your raw signal. And it blends it in with your model to give you a reliable, cheaper sensor where you did not instrument anything.

We have recently created the concept of stress camera. So we just instrument a couple of points, a couple of strain gauges on metallic structures. And we use a digital twin to create an augmented reality deformed representation of the object with a full strain field. It gives you almost infinite information about your strain field given only a couple of those sensors.

But then the question is, why should you trust those? They go through a digital twin. How much do you trust your digital twin? And that is where software technology can help a lot because the difference between a good model and a bad model sometimes is subtle. And so, we develop a lot of tools to make sure that before you deploy these smart virtual tool sensors, you can actually have a confidence that everything is being done to assure your certain level of confidence in the results and accuracy in the results.

So, you will be able to virtually test them, maybe test even what would happen if the input to your smart virtual sensor would be wrong data, maybe a failed sensor, would this lead to a catastrophic failure of your algorithm? And we have a lot of technologies that we provide to our users to be able to themselves build in an easy way that confidence and deploy these cheaper sensors.

We actually call these smart tool sensors, but the concept of embedding models very close to the actual object is something that we recently have called “executable digital twin”, which is a concept that is basically extrapolating part of the digital twin and brings it close to the physical object as close as possible to the physical.

Jim Anderton: Now that’s interesting. We hear a lot about 5G will enable the internet of things. So the manufacturer of the car tire or the athletic shoe will be able to track in real time because of low cost sensors, the ability to connect them, things like how fast the shoe sole wears out or the rubber wears out in the tire, and feed vast amounts of data back to the, the engineering team in real time. So, you now have a river of information that’s pouring back, which has to be filtered presumably, screened presumably by something that’s pseudo-AI or least something sophisticated. But it also, that would then suggest running changes in the design and engineering process perhaps in real time also.

So, what is the single source of truth in this future world at this point? Is there such thing anymore as a design or is there simply a model that is fluid in changing as time moves forward?

Tommaso Tamarozzi: Yeah. A lot of the machines will be able to self-customize depending on the conditioning. You made a funny example about, about rubbers and shoes, but imagine if your shoes could have through a model, the capability of changing its own friction coefficient, if the pavement is too slippery. Would you call that a single design of a shoe? I don’t know if I would call it a single design it’s a continuous self-design of the shoe. The shoe would self-change itself based on the information that the surrounding world provides to it and would adapt; and okay, that maybe the shoes is not something that will have happen tomorrow in this form. I can’t imagine that happening tomorrow, but in a lot of Industrial world application, this can happen. Imagine adapting the forces of machines during manufacturing based on environmental condition or is on the customization of small batches of products that you want to make.

This is this continuous amount of information that comes from the boundary condition and the environment that is fed in your digital twin, and then provide more and more information to your machines or products to be optimally self-designed for the usage they need to be used for. I think the ground truth will be not anymore single, but it will be fluid too.

Jim Anderton: Well, I’m glad you brought that example up, because we’ve talked a lot about the sort of the end user product, if you will. The output of the engineering process, but for production things like shoes and airplanes and automobiles, there’s also a production process itself that must be engineered. There are factories or assembly lines, highly complex, very dynamic systems. I’ve seen them simulated literally with physical models, rooms, tables in which machines and objects are moved like chess pieces on a chess board in attempt for the production engineer to visualize things like material flows on the inside. But I’ve also seen architects using virtual reality goggles and digital twin technology so they can walk through a process and attempt to visualize it that way. Is the digital twin something that you think will be more valuable for its ability to optimize processes or its ability to let engineers actually understand processes and visualize them?

Tommaso Tamarozzi: I think both. If you look at from an understanding point of view, there is a part of each of digital twins that will allow you to explain more that you could explain with just the raw data you had before or just with your engineering intuition if you want or your experience. The example you gave about augmented reality visualization and being able to… We have an example which we have seen recently in which one of our engineer literally was flying inside of a turbine blade of a jet engine to look at how the flow was changing. It was basically a funny way to show that you can go and access with your eyes things that before you were not able to access and therefore understand much, much better what you need to do as an engineer to improve it.

But it’s also true that digital twins there are problems that are so, so complex even for engineers to understand despite having the means to do that, that sometimes you need to trust that your understanding stops at the tools you create to analyze the problem. If you have good tools to analyze a very complex problem then your intuition will follow. You will be able with a digital twin to create so many what if scenarios of complex situation that are difficult to understand and give you an optimal solution for your process, for example, which will optimize your process and you’ll go and try it and it’ll optimize the process. You will trust the fact that your tools, the tools you use to do that are properly engineered.

Jim Anderton: Tammaso, you are Dr. Tamarozzi. You have a PhD in numerical methods for mechanical engineering. Most of us in the engineering world are not so blessed with such a high level of understanding of numerical methods. Do these tools require a higher level understanding of mathematics of computational methods or will these things work for the men in the street who’s attempting to make those running shoe soles.

Tommaso Tamarozzi: Yes, that’s actually one of the key challenges that we have. Actually, one of my personal target is to democratize this tool for people who do not need to be and are not experts because not everyone who is and not everyone needs to be because there is other problems to concentrate than trying to tune your model to be accurate and if someone can do it for you, then the better. So, the answer is we are trying to put in all our tools enough engineering knowledge and numerical methodologies know-how to make the complex choices for our users such that they can only be left with a limited amount of choices to make, which let them still have the freedom to choose between different levels of complexity of the model without understanding deeply what’s happening behind.

This though is also a weapon with two blades: it should be used carefully. So for example, giving the opportunity with a very well-tuned method to a user to define, I don’t know, a vehicle model as simple, medium or complex, just to make a very simple example, could be brilliant if that simple, medium and complex really means what the user wants that method to do. With a lot of methods, for example, like AI and things like this, which are fully data driven, this level of customization requires a lot of numerical and mathematical skills such that the results of those simplifications can still be trusted. So it’s a lot of work for us which we really like and that’s a lot of focus we are putting into making sure that a lot of people can use numerical methods without having a deep understanding of the math behind it.

Jim Anderton: Tommaso, a final question; it’s been said that an engineer is 80% mathematician, 20% artist. You’re developing tools here which could potentially turn this upside down. Will the engineer of 50 or 100 years from now be 100% artist with no need to understand the mathematics at all?

Tommaso Tamarozzi: I think that it can be that only a few engineers will have to then be 150% mathematicians for the other one or to have the freedom to be 100% artists. I like the way in which you put it. I never thought about it from this point of view but there is a level of freedom that you gain by not having to think too deeply about the complexity of numerical models and digital twins. Yes, I think you’re right, this probably will lead engineers to concentrate more and more into their artistic side and this will probably lead to good innovation. That’s a good way to put it, interesting.

Jim Anderton: Dr. Tommaso Tamarozzi, Siemens Digital Industries Software. thanks for joining me on the show.

Tommaso Tamarozzi: Thank you very much for having me.

Jim Anderton: And thanks for watching and Designing the Future. See you again, next time.

Learn more about the Executable Digital Twin (xDT).