An Introduction To Intelligent Digital Twins Webinar

During this informative session, XMPro CEO, Pieter Van Schalkwyk will delve into the concept of Intelligent Digital Twins and discuss how they differ from traditional Digital Twins. This webinar...


During this informative session, XMPro CEO, Pieter Van Schalkwyk will delve into the concept of Intelligent Digital Twins and discuss how they differ from traditional Digital Twins. This webinar... hi everybody I'm Peter from Skull click

um the CEO of Ericsson Pro and the topic

for today is intelligent digital twins

um I wish I I could take the credit for

coming up with the concept of

intelligent digital twins but we're

really standing on the shoulders of

giants um Dr Michael Greaves who

is also known as the father of digital

twins actually started the concept or in

his work looking at where digital twins

are going

came up with it

concept of industrial of intelligent

digital Twins and you can also find

um a great paper that he where he

explains the whole concept

um at the digital and we'll also make the the um

the link available after the webinar and

I I was fortunate I had the opportunity

to meet with Dr Griggs at one of the

digital twin Consulting member meetings

and we discussed where things are going

with digital Twins and this is a diagram

that he's got in in the piper

um and he explains kind of the evolution

and right now we're seeing quite a lot

of ad hoc so sometimes people refer to

this as digital Shadows where there's

one-way communication so we've got the


items and that updates a a digital

version so I'm creating a static

repository of data and we've seen more

and more digital twin platforms emerging

where there's some replication of data

going around

and and the future is moving towards uh

intelligent digital twins with things

like front running simulations and

everything so

really expensive paper that explains it

all my summary of the paper

um quickly ran through that is we're

moving from a focus on data to focus on

data flow so the difference between a

traditional digital twin where it's just

a passive repository and we kind of

taking data from my physical and just

representing it in in a virtual way to

something which is active and always on

so it continuously monitors the

environment and the the assets itself

where the traditional digital twin is

more of an offline and it writes for the

physical to actuate it so even if you

think of something like airbags there's

a a


it's as soon as an event happens on the

physical it updates the digital on the

online one this this continues running

on the side and it monitors scans the

environment and based on that also

creates actuation of of potential

actions that come out of it

with traditional digital twins we give

it a a goal and then we create kpis and

performance measures to see how well

we're doing based on the goals that we

are with that we've set for that

with intelligent digital twins we can

now use more intelligent approaches to

make it more goal-seeking to try and

optimize where we augment what humans

are doing with AI

and smarter digital twins to to support


and lastly we have predictive

capabilities so we can predict but it's

not really optimizing so it's not

looking for certain optimal set points

operating points


maintenance intervals those kind of

things where going forward doing things

like front running

simulation we are able to speed up and

that's what we mean by manipulate time

we can take real-time data we can take

contextual data we can take historical

data and speed that all up and then

based on that anticipate what is going

to happen and see what are the better

responses for those and great example of

front running simulation is what happens

in Formula One car Rising

and example is a partner with Dell it's

this is not built on XM Pro but this is

work that Dell has done with McLaren

around things like front running

simulation so this is one of the fastest

Edge devices out there

um it goes uh 200 miles an hour or 360

km 360 uh

kilometers per hour

and at that it generates about a hundred

thousand data points per second so a lot

of information that you can put together

and based on that determine fuel

strategies and a whole bunch of other

things during race time so ability to

speed up so take the data run it faster

than real time to speed it up and then

decide and

um on on on certain actions so this is

the McLaren example and this also

great work done by Amazon AWS and others

on on on on similar things so things

like again tracking information on the

vehicles they can track information

combining all of that you can create

multiple use cases and I think that's

one of the the other elements of digital

twins that's that's quite uh that's not

often understood it's not about one

application or use case that I'm trying

to do I can now take that data and I can

actually facilitate a multiple number of

different things that I can do so

multiple use cases in this instance of

battle Focus pitch strategy Striker

Performance Tire performance exit speeds

a whole bunch of different front running

simulations that I can do in a specific

interesting one is undercut thread so

um great video on the AWS website as


um but with undercut thread it's

actually deciding when to bring your car

in and also looking at what are your

competitors potentially doing so are

they looking at undercutting you by

bringing in a vehicle by bringing in a

race car at a certain window of

opportunity now in order to do that you

need to meet you need to measure the

real-time speed of all your other

competitors look at their way by using

computer vision and a whole bunch of

things die away

um trying to figure out what their fuel

consumption is take all of that and run

it in your strategy so these are very

sophisticated and advanced examples of

front running simulation but we're

starting to see that move into the

industrial space and other areas where

we are trying to do this so

from a digital twin perspective how we

make decisions is changing as you saw in

those very sophisticated examples we can

bring a whole lot of information from

multiple different places

and that's really the essence of

decision intelligence

where we've got external information

we've got internal information and

traditionally what we've done with kind

of the more static digital Twins or or

conventional digital twins is really

decision support

so there's a business process there's a

human in that Loop and we are now

um trying to give them some decision

support and traditionally it's been

dashboards and business intelligence

condition monitoring basic predictions

as well

um so again the mind of of decision

support we're seeing more and more

decision augmentation where we can bring

in smarts from Ai and other tools

so um we still have the human in the

loop that make the decision but that

human is now the the decision process is

augmented with

um information that can be processed at

a speed that humans can't do it the

volume of information that that can be

processed again is done at a speed that

we can't do as as humans but

at the end the decision Still Remains

with us so this is where front running

simulations prescriptive recommendations


is is coming in and this provides us the

opportunity to create a closed loop a

feedback loop and learn from that and

kind of improve our decision making as

well as improve the models that we have

and this is kind of where a lot of um

focus is at the moment in terms of

moving towards more intelligent digital


but there is a future where we also look

at decision automation so once we have a

high level of confidence on in some of

these decisions or or some of these

models we can actually let the machine

make the decision through business

process Automation and create this

distributed intelligence system

where we can maintain the rules and the

models and everything centrally and it

it it it provides us the opportunity to

get to algorithmic business where again

a certain number of these of of these


business processes that can be done in

an automated advice so if you think

about fly by wire which you can do with


you might be able to operate by wire in

a certain envelope those decisions can

all be made by machines so that's

the future that we see in terms of where

this is all heading distributed

intelligence systems and now if you

combine that with the control

environments if you look at a

distributed control systems and you add

distributed intelligence systems my

personal views that's probably the

future of what operations will look like

now what does this mean for intelligent

digital Twins and I refer back to the

digital twin Consortium


definition of what a digital twin is

it's a virtual representation of

entities uh of real world entities and

processes to synchronized at a certain

frequency frequency and fidelity

interested in being on it's synchronized

and it helps us with decision making and

taking effective action again it's about

decision making it uses all sorts of

data and it is motivated by outcomes and

it's focused around use cases again that

example that I showed with the front

running simulation a whole bunch of


use cases

um that are being facilitated for that

and we implement it in it requires the

main knowledge and implemented through

this so

at a traditional digital twin has got

um that synchronization Sometimes some

of the information may go back

but it's really passive it's offline

um in the sense that it's waiting for

the thing on the on the left hand side

the physical twin something to happen

and that will then update that's not um

and it's called Givens I've got kpis and

things that I'm trying to to measure and

yeah we can predict but it's not focused

on optimization

going forward with intelligent digital

twins we see that decision intelligence

structure that I showed a little bit

earlier make its way into kind of

operating on the side so

we've got expert knowledge business

rules all the math and physics models

that we have and quite a few of those

are already being used in the uh more

decision support type digital twins that

we have we're now starting to see convey

what is regarded as conventional AI so

regression models and all of those but

then also the new generative AI that

we've seen lightly and large language

models that are making its way in

and some of the more sophisticated

techniques like deep learning neural

networks and those so that in

intelligence making is it's why

into digital Twins and then providing

the opportunity to Market goal seeking

and learning and doing this front

running simulations now the question

that we get is how do how do we make

this happen how do we do this because in

order in order to do this we need this

thing that intelligence that run on the

side almost and we go from where in the

previous one we had information at the

bottom there we now have prescriptions

and that is synchronized at a different

rate so it's not just the information

we've seen in back we actually send a

prescription in terms of what to do

um back now this could be augmented or

potentially automated as well but we

have to have this mechanism on the side

that continuously runs so now going from

data this requires a data flow where it

continuously run on the site and

interact with that

so in order to do this we came up with a


as organizations are considering how do

we move to intelligent digital twins

well first of all it needs to be

integrated and it needs to be based on

standards models and have that

bi-directional integration that we that

I mentioned it needs to have

intelligence and we'll go into each of

these in a little bit more detail but

needs to be executable so it needs to be

able to run in real time we need some

way to make it Innovative and explore

experimentation and doing those

simulations but we also need to provide

an environment where we can bring in

the the the help from from the digital

Twin Side and augment so that we can

learn from that

and lastly we need to migrate

interactive so

um this is about the visualization so

how do we how do we provide

recommendations how do we make it in a

generative multi-experience user


because all of this is becoming the

foundation for the industrial metaverse

whatever the metaverse looks like when

it comes out in order to do this at

scale we also need to do this on the

composable kind of


framework where we can reuse components

almost like the Lego blocks that kids


um toys with you can actually reuse

components and and and have a plugable

composable uh price for this

so if we briefly look at the integrated

side of things

um standards-based apis for these

capabilities that we package together a

model driven approach and bi-directional

in order to do this again this is about

data flow so example that we're showing

here this is our data stream designer is

being able to create standard


to the apis of different systems then

being able to create a model now the

benefit of this is I can apply this to

um uh 100 wind turbines or a thousand

when turbines is exactly the same model

so in terms of the data model that

supports it and potentially the digital

twin model also model driven but in

terms of of creating The Logical data

flow structure making that model driven

and at the bottom right

it's not just about in uh have receiving

information one way but also sending

information back and potentially

changing set points um based on

recommendations that could come from Ai

and some of the other elements

so that's the the from integration

perspective making it more intelligent

where we now adding this capability

first of all in terms of those three

elements we need to make it executable

we need to provide an environment to to

to to for um to create these AI elements

and then we need to augment the user

experience with that and again different

audiences which this applies to so if I

look at that executable how do I bring

it into the data flow traditionally we

would take real-time data we would have

a model and

um so we've got the streaming agent that

with that with um

bring in real-time data we've got a

configuration of what the fly looks like

and that just gives us a result

with executable AI

um we can create a training model and

I'll briefly Show an example we'll

create a training model I'm using the in

this instance our XM Pro AI notebook

deploy that model into into an ml Ops

environment because again we need to now

look at how do we do this at scale if

I've got tens of thousands of models how

do I do the model management and as part

of the digital twin management as well

and then bring that in through bring

that model in through again an agent

that's got the intelligence bringing

live data on that can now run on that

training model and again in our in our

now code in environment you can

configure all of this and that will give

you the prediction and simulation

so what does it look like when you

actually do it

um what you can see here is a very basic

Way Reading uh tanks type and you'll see

where the yellow perform machine

learning analysis that actually calls a

beer quality model that sits in uh

uh the through um ml flow where where

which is the data repository for the

model so that's how we make the digital


executable in the data streams that we

have the next part is


to be able to bring in an environment

where we can make it more Innovative so

we've embedded jupyter notebooks but

we've added some functionality to that

so you can wire it up into our data

streams but you can also integrate it

with things like giant GPT and others to

help you in the process

so again aimed at

on the one hand in analyzing the data

but also building models

um for things like front running

simulations and some of the others in

this instance I've got chat GPT

and um I I can ask it to tags to create

a code for me which is the next part

over here to create the guide for me to

represent this data

in this in a certain way so having the


um the request


having a request here through GPT

how do I visualize this data it then

writes the python code for me and it I

hope if I run that and it gives me the

visualization here's another example of

of embedding


or augmenting

um the the user experience in this

instance there's a copilot so we ran

this and based on the recommendations at

the top we can see that there is

potentially a impeller problem but I can

also see my discharge pressure is not

what it should be

so I can ask you know what are the top

five root causes for centrifugal pump if

there's a loss in discharge pressure and

you can see what it came up with if we

have time on briefly jump into and

showing you what that

is um towards the international fact let

me briefly let me quickly do that




this is the Jupiter notebook with the




I can run through this it will generate

data for me



this is

machine generated data and now I ask

Chad GPT

to actually create the um

let me move down a little bit so I'm now

going to ask Jack GPT to create the code

for me

to visualize this data

and that's done that and now it's

created based on the code that it's

created for me in Python automatically

and also created the visualization

um and again this goes into more

actually putting this into into an email

slope on the the other example that I

briefly mentioned so


on the well Refinery operations

and I just go to that pump

we just got an issue

and you can see the normal information

that I have I've got all the contextual

information but I also want to ask it a

question now

um I'll just one of the top five root

causes and this thing interrogates using

um functionality of chat GPT to do that

so that's ways that we can bring

intelligence to it lastly on the

interactive side

to make it more accessible

in terms of providing recommendations

again uh

the intelligent digital twins have have

the opportunity to not just


tell you what is wrong but also give a

recommendation on on what you should be

doing triage instructions and some

additional context around what happened

in the past this gives you the

opportunity to also close the loop I can

create work orders and and things from

here and then be able to track that as

well to see how effective we are at

making certain decisions in terms of

creating a collaborative environment as

you can type as the previous example

which I briefly showed you

you can create very collaborative

environments and user experiences where

you now bring that intelligence

um into a front end that that users and

can use

and as I mentioned going forward you

know this will form the foundation of

the industrial nativist now the Brew the

Brewing example that you see here is is

actually an example that we built out

for the Dow validated

solution for the manufacturing Edge side

of things but this is what we see going

forward in terms of creating a meta

versus in a very interactive environment

where you know it doesn't matter what

your user experience is whether it's AR

VR desktops mobiles

um the intelligence is portable across

all of those

environments lastly and

the composable side


we came up with the the


positioning XM Pros that can composable

platform and


bringing in data from all the the the

underlying systems to be able to build

authorize various different use cases

and for that we have a whole bunch of


um modules that that support that so

in order to do that the basis for all of

that is is


capabilities and inside digital

Consortium we were instrumental in

creating the the

um capabilities periodic table these are

all capabilities that we see at a really

high level and that is used for digital

twins if you're wanting more information

on the digital Consortium website you

can download

the the whole

capabilities framework we also have a

webinar that we've done in the past

which you can have a look at where we go

into the capabilities and composability

and side for that so so here's our view

in terms of what the future looks like

oh sorry oh so the the the the

components that you need in order to

create intelligent digital twins which

we see as the future of where digital

twins are going we've created we've run

all that all into what we call

intelligent digital twin Suite example

idps consists of a number of modules and


happy to share more information on that

just lastly we also see that this way in

terms of where we are going and how

we're doing this at that line in future

may disappear and that intelligence all

being built into the

um digital twins of the future so with

that um there's a few minutes left any

questions that that everyone's got that

I can address

great presentation you do have a few

questions that came out there

um take it from the top uh do you one of

the questions that I that that we get is

you know do you supply the AI models


it is the reason why we created that

environment so inside

um the XM Pro AI notebook

um you can create your own models there

are libraries of existing models

um there are Auto ml there's a whole

um uh kind of Continuum of opportunity

to use pre-made libraries of of models

we don't specifically do do models but

um there's a large library of of

simulation models of

all the standard

um traditional IR models as well as as

you saw being able to bring in

generative Ai and others pretty easily

how do you get started with this

um we're happy to run you through kind

of the the kind of three-step process

um on on how to get started with these

digital twins but hopefully that's been

been helpful in giving you an

understanding of you know what we see as

the future and what digital twin what

what intelligent digital twins

look like going forward

any additional questions

I have one question that was sent to me


um so yeah again this is I think the the

the way that we


that we look at how to how to create

this look at it it needs to be

integrated that needs to be intelligent

needs to be interactive all on a

composable approach and then in terms of

um the question is really do are they

starters or examples we have a

blueprints that we are creating they are

they are started

examples available in GitHub so we're

really easy to export into your


um and then play with that we do have a

freemium option so if you go on a

website you can download premium option

um of us of our software and also some

of these libraries and examples that we


over that um thank you very much I

really appreciate you watching this and

we'll see you uh on the future webinar

we will send out the recording of this

um after the event thank you

Last updated