Lifecycle Insights CEO and Chief Analyst Chad Jackson and Industry Analyst Arvind Krishnan discussed the latest computer aided engineering trends. Here’s the transcript from their discussion: Chad Jackson: Hey everybody. Thanks for joining us today. We’re going to be talking about digital twins and what’s been going on in different events. This will be an …

Computer Aided Engineering Trends: Digital Twin Read More »

Computer Aided Engineering Trends: Digital Twin

Lifecycle Insights CEO and Chief Analyst Chad Jackson and Industry Analyst Arvind Krishnan discussed the latest computer aided engineering trends.

Here’s the transcript from their discussion:

Chad Jackson:

Hey everybody. Thanks for joining us today. We’re going to be talking about digital twins and what’s been going on in different events. This will be an evolving topic, but Arvind Krishnan and I will be diving in. I hope you enjoy it.

All right Arvind, I’m excited to have you on and for us to have this discussion. This will be really good, talking about the digital twin and how simulation plays into it. So, you recently attended the online event for ESI Group. It’s called ESI Live, right?

Arvind Krishnan:

Yes. ESI Live 2020.

Chad Jackson:

Okay. All right, good. And you’ve also written up a post and published that recently about actually some of this topic. So on there, you listened to a presentation by what was their name again?

Arvind Krishnan:

Yeah, the gentleman’s name is Francisco Chinesta.

Chad Jackson:

So in that, he talked about a few different types of digital twins, right? So the first one he talked about was the virtual twin. So tell me a little bit about their view on what that is and what it’s used for.

Arvind Krishnan:

Yeah. I mean in short, according to Francisco and the ESI Group, the virtual twin really refers to a simulation model. So it could be any of their simulations, they have a bunch of them. So anything they can use as a model to represent the real life behavior of the physical asset can be considered as a virtual twin.

Chad Jackson:

Okay. All right. Good. Okay. So that’s not a crazy concept that a lot of our followers are unfamiliar with, so that’s good, so it just needs a little translation over. But then they talked about the digital twin. So let’s put that into context. What did he describe as that?

Arvind Krishnan:

Yeah. In the context of ESI Group, the digital twin really is a set of data that comes from the physical asset. So it could be sensor data for all you know, most likely the sensor data. It could be thermal sensors, which are transmitting temperatures. It could be pressure sensors, which could be giving you pressure loads, for example, on a wind turbine blade.

So it’s a bunch of sensor data that can be collected by this digital twin. So really the digital twin in the context of ESI Group’s definition, is really a bunch of sensor data that can be used.

Chad Jackson:

Gotcha. Okay. That makes sense. Well, let me ask you a clarifying question here. So is that really just from say service, or is it also from test as well?

Arvind Krishnan:

That’s a very good question, Chad. It could be both, technically. Service is something where the physical asset has already been deployed and then it has some sensors and there’s probably some kind of an internet of things service that actually gets these values.

But on the other hand, during the product development phase, you could as well build a prototype and test this prototype and get the test data as well. So it really does not differentiate whether the source comes from a prototype, or the source comes from a live physical asset, which is operating in the real world.

Chad Jackson:

Okay. All right. That makes sense. But in both cases, regardless of where you get the data from, we’re really not looking at a simulation in this context, right? We’re really just talking about a set of data that describes the actual physical behavior of a physical thing, right?

Arvind Krishnan:

Yes. The digital twin refers to that, correct.

Chad Jackson:

Okay. All right. So there’s this third concept that they described called the hybrid twin. Tell us about that.

Arvind Krishnan:

Yeah. The objective of the virtual twin or the simulation, is really to mimic the reality, what really happens to the physical asset. The reality as you know Chad, is really complex. It cannot be defined by an abstract model. I mean, that’s the reality, right?

So what happens is the simulations to some extent, capture the behavior of the physical asset, but it may not be perfect. And in most scenarios, it’s not perfect. There is a little bit of, you can call it error, but it’s really the inability of the simulation to capture the reality.

Now, the digital twin is actually helping you, I mean, helping the virtual twin in bridging this error gap. How? That’s the concept of hybrid twin. So it actually gets the real data from the sensors. It compares it with these simulation results and it actually measures the error. And now it bridges the gap.

It learns with more and more data, not enough, but sufficient to make sure that error is minimized. And technically, you can always get it to almost a zero error. And now you have a model which is really what they conceive as a hybrid twin, which is kind of perfect. It mimics the reality.

Chad Jackson:

Interesting. Okay. So there is this concept of capturing sensor data and using it as an input to a simulation, then you have independent output. But I guess what you’re saying is, you’re comparing the output of the simulation and the output of the digital twin, the sensor data, to see where the error is. And you’re using that to converge?

Arvind Krishnan:

Yeah. That’s the understanding, yes.

Chad Jackson:

Okay. All right. Interesting. So how are they closing that gap? How are they converging? I mean, are they using AI machine learning? Are they using some other method? What’s the idea there?

Arvind Krishnan:

I think that here I would be speculating because it was not extremely clear, at least during the discourse, what exactly they are using. But I guess it’s a combination of these things. They probably have the physical prototype or the physical operating device, but you are limited by the number of sensors you can actually attach into those devices. So they probably get values at say a few different sensor points, but the simulation is a continuum.

You could technically place sensors anywhere you want. And I guess they probably use some training models to understand how they correlate the real life versus the simulation. And it’s possible that they could use some kind of machine learning algorithm to make sure that this model becomes better. But ultimately, the hybrid twin can become a reduced order model, which can in lightening speed, give you the results that you want. And they talk about that too.

Chad Jackson:

Okay. All right. Well, that’s really interesting too. So I think this is something we’ll have to follow up with them on to understand exactly how they are closing the gap there, because that would be some really interesting technology.

The other activity that we saw around digital twins is with Siemens Realize LIVE event. So their user event for Simcenter and all of their other solutions out there. So one of the concepts that they talked about there was the executable digital twin. It seems like we’re getting all sorts of different types of digital twins nowadays. So let’s talk about that. And what’s the idea here with the executable digital twin?

Arvind Krishnan:

As the name suggests, executable digital twin, consider that as a piece of software that can be installed on an edge, or it can be running on an IOT platform. All it is, is a reduced order model of the simulation, along with the ability to get sensor data into this executable software, which has the power to predict.

Chad Jackson:

Okay. Yeah, that makes sense. That makes sense. So the idea is, that it will mimic and predict the behavior of something, but you can almost put it anywhere. Is that the idea?

Arvind Krishnan:

Yes. That’s an interesting point. You can install it onto an edge device because it’s a packaged piece of executable. Think of it as an EXE file. It’s an executable, you can install it on an edge device. You can have it in the cloud, obviously. You can also have it in your on-premise server. And ultimately you can give it to your customers as well, so that they can install it and make use of this piece of technology, that’s lightweight actually. So it can be installed on a very rudimentary kind of electronic device and it can perform.

Chad Jackson:

Yeah. I think this is really interesting when you talk about electronics and embedded software that run on products. One idea here is you can put it on the edge, and the product itself can be on the edge.

But there’s an interesting side note here in that Mentor has a real-time operating system that it offers out that you can then run on the product. And it will be really interesting to see if there are some integrations there, to see if it performs better or something along those lines, more integrated, more out of the box. You don’t have to hand hold it to kind of prepare it as much. So that will be really interesting to see if they move in that direction as well. When I heard about this concept, it made me think of custom ICs and FPGAs.

So on today’s products, especially in safety applications, you see ICs fed sensor data. And because especially in safety applications, they need to be processed so fast. They don’t bother with software. Its pure gates on the IC because that’s the fastest way to execute it. And that way you can have it react very, very quickly to automatically brake or swerve to miss or something like that. I’m really interested to see eventually if they take the executable digital twin and one day it takes not the form of software, an executable, but if actually it takes the form of silicon, in terms of the custom IC or an FPGA.

I mean, that’s where you could go with this. You basically have a digital twin as a chip that you could then go install and it’s running very, very fast. And when you think about it, running these simulations in real time, that’s always been a problem. I mean, even in reduced order models, people have always been trying to get those to run faster. And I mean to your knowledge, have people really gotten to the point where you could run it continuously and feed it sensor data, and it’s always running in real time?

Arvind Krishnan:

I don’t know, but it’s really a very interesting future technology that I’m pretty sure we are going to see. Because I think the maximum benefits for these digital twins are going to come if they are always vigilant, if they are 100% ready to spot, let’s call it a black swan kind of an event, that is not really thought through during the design process. And that’s really the power of this. And you are absolutely right Chad, that’s an interesting piece of thought process in terms of future. Because what better, if the silicon is just meant to run that particular model. You can’t beat that.

Chad Jackson:

Yeah. That’s about as fast as you’re going to get, at least as far as we know today. This has been a really good discussion I think for everybody out there watching this. I think this is going to be a recurring discussion. Because a lot of these solution providers, a lot of manufacturers are continuing to adopt the digital twin in lots of different forms. Simulation has a big role to play in these digital twins. And I think you’re going to see more and more versions of digital twins come out over time. So hey Arvind, thanks for the discussion. Really good one. And want to do this again real soon.

Arvind Krishnan:

Thank you, Chad. Yes, it was my pleasure. And I look forward to such future discussions.

Share this post
LinkedInTwitterFacebookEmail