Predictive Maintenance With Xmpro Idts
Predictive Maintenance with XMPro iDTS
Dive into an in-depth exploration of Predictive Maintenance powered by XMPro iDTS. This video aims to provide engineers, subject matter experts, CTOs,...
Transcript
Predictive Maintenance with XMPro iDTS
Dive into an in-depth exploration of Predictive Maintenance powered by XMPro iDTS. This video aims to provide engineers, subject matter experts, CTOs,... in our overview of a video we describe
the process that we followed to get from
our assets and real-time data from the
assets through to some actions and
recommendations that we want to drive
and the outcomes that we're looking for
now the typical process is that we bring
in data through our data stream designer
you could then run some analytics
through AI or do some of it inside the
data stream designer see the front end
in the app designer and then lastly have
recommendations and manage
recommendations inside the
recommendation manager from a
demonstration point of view I'm going to
start at the app designer and then we'll
go into that also includes
recommendations we'll go into the data
stream designer and show you how we do
it in the uh behind the scenes how do we
get the data in and also touch on how we
apply AI to this so let's jump in and
get started
this is the app designer the green icons
at the top
this is the data stream designer and
this is AI so we'll go through all three
of those but I'm going to start kind of
at the end I'm going to show you the
result of what what you see when you use
XM Pro and I'll start with uh just a a
simple example around multiple
facilities or assets that we have in
this instance I'm just I've set the
filter just to show me
um
some of my
Wind forms and I can the colors and
things will change based on the severity
the health of the overall Farm not
necessarily the individual ones I can
set that up to so that I can actually
drill down and get into the actual form
itself and drill down right down you'll
see this is a viewport that is using
esri I'll touch a little bit later on
how we can edit this and what all the
different types of viewports that we
support but this is an example of
something like that here I can see some
of the assets that need maintenance
there's also some recommendations on
some of this so at a holistic view as a
maintenance manager facility manager I
can very quickly see the health and
status of my different facilities and
where I need to go now this again is a
specific type you can have a much more
sophisticated example of this as well
I'll um I'll go into and just show you
again as a maintenance manager
maintenance planner looking at all the
facilities that we have
um I can see kind of the alerts to work
requests work requests to work orders
work orders to open work orders to to
closed and how long it takes to actually
resolve it and
um how what is the efficiency how many
open work orders do I have and what are
the recommendations around all the
different machines again this is
everyone's you'll see there's a whole
bunch
listed through here so and they range in
severity in this instance they rank by
severity
I can look at
the different installations or
facilities that I have and in this
facility this is a wind turbine facility
this is a solar array so
I'll go into the wind turbine one so
I'll have a quick look at what's
happening at the wind turbine and this
gives you an idea and I'll share a more
advanced example of this but I can see
some safety and health information I can
see the overall time usage profile
current metrics that it's running some
of the and again
maintenance records maintenance
information some of the corrosion damage
tables and depending on the type of
modeling that you use I can get into it
you can see this updates in real time if
it's a Unity or Omniverse or a a more
interactive model or a card based model
I can actually get into it and I'll show
you an example of one of those but this
is
with this I can now actually get to a
recommendation so I can see there's low
gearbox oil on this reported and there's
the instance of that that triggered that
so there's a rule that runs and when it
goes below a certain threshold this is
very simple rule around a certain
threshold on the on the on the gearbox
oil low warning and there's potentially
some triage instructions this one
doesn't have too much I can also see how
many times this has occurred and if I
look at kind of a longer duration what
are all the other things that are
happening to this specific asset that
has happened over time I could create a
form or I can fill in a form right here
and this could be a work order request
this could be a root cause analysis a
failure mode analysis
um
you can associate
a large number of different act action
forms do this in terms of the kind of
action that you want to take I can also
Mark this as a false positive so that
later on we can start analyzing how many
times are we getting this kind of false
positives out of this so this is one
example of a
facility predictive maintenance I'll
show you a slightly more advanced
example in in this example this is a
processing plant in this instance plant
but it could be a Water Filtration plant
it could be
any processed plant and in that
again analyzing the data looking at
real-time anomalies and looking at
defects where certain type of defects
that are that are happening so this is
on the quality side so I could bring in
some of the process information
there's fill rates efficiency on on the
energy consumption but likewise with
that esri map where I had a
um uh Green Dot or radar or whatever
represent that there's an issue you can
actually see on this one that
there's a there's an issue here so these
will appear or disappear depending on
the recommendations and the current
state so it's very quick to see as a
planner
or maintenance supervisor maintenance
manager plant manager you know what is
happening and I can have a very simple
2D model of a pump with some real-time
data coming through for the pump from
sensors that we have I can see the
nameplate data for the pump
what is writing at or what it's writed
at so just very quickly I can see how
far is it off in terms of its actual
performance I might be running a
predictive model which again will touch
on a little bit later a predictive model
that's pretty predicting the the
remaining is for life
so that I can determine how much time I
have before there I may have an issue
again I can get into the recommendation
in a minute but here's a bunch of
real-time data and around temperatures
vibrations
and all the maintenance records so I can
see maintenance schedules maintenance
history
and I can see what is currently planned
I can also see what has been what has
been recently completed in terms of work
order history really just helps with
um and again I could drill down on this
next level before we do that what I'd
like to do though is actually get a
better view of this type of pump and
if you if if for certain type of
equipment you may want to
expand the capabilities and being able
to see you know what is really happening
so if I look at the discharge exception
or the remaining useful life on the
bearings
I can move this around and actually have
a view on that now
if I click on this view more you'll see
I get to that same recommendation that I
complete but before I go there this was
a discharge exception and I may want to
go to something like cha GPT or
or
generative AI large language models and
actually ask it and I've typed this in
before so that's why it comes up but I
can actually ask it top five root causes
for centrifugal pump where there's a
loss in discharge pressure
and what it does it gives me just some
direction and what I could potentially
do with this
is just copy that
so that when I do go to the the the
recommendation itself
I can kind of create a bit of a starting
point in terms of you know where we
should potentially be looking for
um
for certain issues or and again it's the
same and this again this is a work order
request but it could be root cause
analysis could be different types of
forms that you want to associate with
associate with it this could be have
more advanced triage instructions so you
know these are the typical things that
you could look for Block suction by
blocked impellers
um
it's not necessarily the information
that you send right through to the work
order level it might just be something
for you to help triage what is the
potential issue based on the combination
of this as well as the drills
instructions that we have here and again
the analytics across this to say well
you know I want to know what are all the
issues that we've seen on this pump
lately is it a kind of recurring pattern
or what is happening so that's a key
application for us around
predictive maintenance in facilities
to be able to and condition monitoring
is to be able to get a view of the
um the overall acid and then being able
to get down right into a recommendation
I'll show you how we set up the
recommendation data in a minute but
before we do that I would like to show
you how we do the back end data that
goes into this
now
for that pump we'll just go to the smart
asset that bump that I just showed you
there are two different data streams
here we refer to these as data streams
so they're streaming real-time data this
is a very simple condition monitoring
and then I will also show you a little
bit more then and bonds predictive
maintenance example this is using Azure
digital twin as an example it doesn't
have to be but it uses Azure digital
twin as the digital twin repository or
the the
asset master or the asset model for for
this it could be any existing eam or
other system that you that you have
what we have here is we're getting
real-time pump Telemetry using mqtt
which is a protocol so I'm getting
real-time data in I clean up that data
because industrial data is never clean
and in this instance I just I take this
and I send it to go and update my Azure
data Explorer which will give me a
really nice time series visualization of
all the information that I can drill
down on that and do thumb series slice
and dice
uh information but I can also
do some calculations around the pump
efficiencies and you know all the pump
metrics that we are trying to calculate
and update that state to Azure digital
twin so that we have the lightest on
that I also take that same data and I
will
I run it through to the recommendations
but before I do that I will
contextualize it using pump make model
all of that again from an eam system or
a maintenance system or the digital twin
system so that I have Rich data that
sits over here now how we put that
together so these are all listeners
you'll see the blue ones that's how we
get real-time data in and this is just a
subset but you'll see there's a huge
bunch of different protocols and
applications and services
and streaming platforms and and that we
pre that we support in in putting this
together the context might model all
this this is all fast moving data all
the slow moving data
um I can get through from all the
contextual data mic model
um
yeah again
weather patterns anything that I require
so if I'm doing flood predictions or
that kind of thing you know I can get it
for Weather Services or against
different
contextual data sources Transformations
is when we change the shape of the data
so cleaning it doing calculations I'm
changing it and you'll see there's a
there's a number of ones that are around
cleaning the data missing value
substitution
aggregation calculations uh
normalizing
setting up thresholds and um
and a number of these where we actually
transform the data this one doesn't have
machine learning in I'll show you an ex
machine learning example in a minute but
this is where we can bring in I can just
drag on anomaly detection and it's not
probably the right place to do it here
but just to illustrate the concept and
then I configure that each of these are
configurable so you can see this
pantelemetry one there's the
configuration for that so in order and
it will interrogate the underlying
service and come back with what are the
fields that this thing actually has and
I can now use that in my data stream or
if I go to something more sophisticated
like as a digital twin example it will
use some of the
xero trust capability credential
management which is all in our
subscription and in our in our
subscription manager installed in the in
the in the variable side but this is
where I would create for example I can
create a whole new instance in Azure
right from the application in here so
these are the kind of applic the the
configuration in a no code way of
setting up these data streams and each
of these have their own unique set of
properties
that we activate for each of those
I just discarded
um and then
that's the machine learning part so I
can bring in machine learning I'll show
you an example in a minute I could this
calculation might be quite Advanced and
I may actually want to do that in like
python so I want to maybe use my my
um bump calculations and efficiency
there might be a library that I'm
already using and again we support all
of those libraries out of the box
um in in this
the functions so that's more statistical
mathematical things like fast four years
geofencing and a whole bunch of goal
seeking similar to what you find in
Excel recommendations area that I'll
touch on in a minute but that's a whole
area on its own and then lastly action
agents so this is where we create
actions so it could be something like
send an email
or create ethereum smart contract
completely to ends of the scale but
that's really or create work orders in
Maxima update other databases
send it back to systems a very
comprehensive set now these are all this
is an extensible framework so if we
don't have a connector or don't have the
function that you that you require here
it's very easy to put that together we
can do it our partners do it our
customers do it and there's a framework
in order to create these connectors
yourself so this is at a really high
level what we're doing here so this is a
very basic one and there's actually a
live view of this so this is live data
coming through this is not the user
interface that I use every day to look
at my data but it helps me just to
understand am I getting the right data
at the right
point of this data flow and the way that
I for example
um
map and configure the endpoints or
getting the data in so you'll see in
this data flow it's interrogating adx
service and so what can you accept and
then we can bring in
from the from the
um
from the
next effect let me see if the Azure
digital twin one is configured to
show that up
so yeah as you'll see this one is
already configured to send the data into
Azure digital twin it will interrogate
the twin service and see what it can
accept this is what I have in my
Pipeline and we can Auto map that and it
will get that
into the data flow very simple example
but really effective in terms of getting
condition monitoring on assets going
as you get more advanced and more
sophisticated you may want to start
adding some predictive capability and in
this instance we're reading some some of
the data from a historian like osisoft
we're getting some other data from a
sensor based solution which has got IPC
UI running we've got some
Azure iot hubs and some of the other
capabilities and these are all the data
wrangling data cleaning and everything
that I need to do
and adding context from sap on mic model
and geolocation a whole bunch of other
information before I can actually and
converting failure tags and doing all of
this this is really hard if you do it in
code
what if you do it yeah it's really easy
to understand the logic but also to
troubleshoot and it's much easier to to
understand the logic of what you've done
in order to get this to this point what
I can now do is do my pump calculations
as you saw exactly the same calculation
as in the previous example but I'm going
to update my Azure digital twin in this
instance so I've got three actions
coming out of this block go and update
it with that data run an anomaly
detection on the pump performance
simple anomaly detection or have a more
sophisticated binary classification to
say to say is it likely to fail yes or
no and if it is you know what's a
remaining useful life model which we may
have for the pump
and um
and I'll show you an example of how that
can be put together merge all those data
and again send it to the recommend to
the what we call the recommendation
engine and run that recommendation you
may notice some little red legs on here
this is where you can configure a error
handling flow so if there's a problem
with this what do you want it to do who
must it alert what what must it let you
know
and or whatever system do you want to
activate when you have an error or
whether this data from the historian is
not coming through so this is the data
streams very very powerful this is where
80 of the work happens if you remember
the tip of the iceberg this is a nice
visualization all the heavy lifting
happens at the bottom this is what
happens over here
before we go back to the app designer
and recommendations and how we configure
them so as I said this remaining useful
life regression model
so where do you get them well inside XM
Pro we've now built in
jupyter Notebook so you can create these
models here and deploy them here here's
an example of reminding useful life
using a random Forest
um
to do the to do the um
the the model and I can see what what
senses and what influence and what is
the Affinity of and the correlation of
the data on this and then based on that
so this
um regression model that's built on
random forests
[Music]
um as an as an example we'll then output
the right model for me we also big
supporters of things like Auto ml so if
you don't know what model to use it can
suggest and um and the output of this
model then goes into something like ml
flow so just another one you saw the
example of the beer the beer quality
what we're doing in this one is we can
even generate synthetic data if we don't
have the right data we can kind of
generate the data that the more that we
want the model to be trained on
and we can even use chat GPT to ask it
to
help us with the
um the visualization of this information
so it will write the code so you can see
here we're asking GPT we've got a magic
command for charge GPT built into this
so how can I visualize the data as a
correlation Matrix and it gives you the
code we can I can then run that code and
this gives me the correlation Matrix for
that
this is where I get into creating the
actual model and deploying that to
something like ml5 there is a webinar on
our website that goes into a lot more
detail I'm not going to spend more time
on this right now but just you can
automate building models right through
so that in the data stream if even if I
retrain my model I don't have to go and
update it here it will automatically be
updated
the last thing I want to touch on is the
um
recommendations and as you see we've got
a recommendation here now how did we set
up these recommendations there's another
one over here so for the pump discharge
so how did we set up this recommendation
and there's a whole area where I can
manage all of them look at how and look
at Who's got what and where it is but
there's also rules where I set up the
rules side so if I go to the bump you'll
see these all the different categories
so the pump discharge pressure was the
one that we had a rule
so there were actually two rules
so it would First Look for out of
efficiency range and then it will look
out of optimal range so if this one is
not if it's not true then it will
continue and to go so it enforces a
execution order and it gets data from my
pump Telemetry data stream
that that sends the the data through and
that's how it interrogates that
um if I look at I'll just use this quick
example to show you this is the nice
description that you saw at the top with
a nice icon and everything but this is
the heart of it where I have the
um
flow rate
and less than a third between a certain
band and the discharge project is
listening as soon as that rule is true
it puts that little red dot on there for
me and puts the it puts the
recommendation on the on the list and I
can now get that information now where I
get this flow rate you'll see there's a
whole bunch of parameters that's
actually what comes out the bottom here
so that what comes at the end of the
pipe over there
is what I now have available
to build this for me and I can use
different calculations and this can be a
value or it could be another parameter
so I can actually build a very Dynamic
set of rules including predictions and
everything that can come down the pipe
now this is not a valid rule where the
flow rate is not equal to the motor
current well hopefully that's not the
case but
that is gives you an idea of how you can
construct this so it could be completely
different because I can bring in very
different information I can bring
weather information I can bring all
sorts of
um I can bring maintenance record so if
there's if there's no
um
if the
if a certain condition exists then it's
minus 40 degrees outside then don't
create the work order because no human
can work in those conditions we have
seen some of those applications as well
so if a safety factor and all of those
you can bring in if they're or if
there's hot equipment nearby then notify
that
so that's how I built the rule and as
you saw in this instance I've enabled
the form and the form was a work request
but there are some other form types and
you can build your own forms as well and
this is just some of the how many times
does it need to log it can it
automatically resolve it so if it
condition is not true anymore will it
automatically resolve
these all capabilities that are built
into our recommendation engine very
sophisticated capability that you can
track but also this is what I put in
triage instructions notifications so
do I want to know when there's a new
alert when there's a status change how
might you know if there's certain
thresholds being not met on time because
we're waiting too long for someone to
respond to it so those are all
capabilities of
the um and the last thing I want to do
is just very briefly touch on as I said
I'll explain a little bit you can see
now that I've played around with it
um
there's definitely a higher
probability of failure that has gone up
due to some of some of the things that
I'm doing at the back here
um that's not what I wanted to do
this is um so when I clicked on the
pencil I have access to be able to edit
this you can actually see both of these
the 2D and the 3D they just overlaid on
top of each other and there's the name
black so that's how I got that done this
is a page builder web page builder in
this instance this is a Unity model and
this is just a 2d graphic both of them
have the same data that it displays and
these are all wired up data sources
and when I'm what I mean by a wired up
data source if I click on this Unity
block well let me first explain the
concept of a block there's a whole bunch
of different types of blocks
there are recommendation blocks action
blocks
and visualization blocks so that's where
we had Autodesk Forge and esri and unity
and all of these different visualization
capabilities that we have and we can
also create widgets so if I like this
style of something that I've built I can
actually go and create a widget for that
and these are just some examples so we
just can be saved and they can be shared
so if I've built a really nice widget
based out of a grouping of things so for
example I just press the
save there and whatever built in here
and now is available as a widget next
time I just drag the implied on and I
have a a name plate available they might
even be on here already
but that's typically how I create
widgets the data that sits behind this
so again I'm clicking on the unity model
here you'll see it gets the yellow line
around it
this is just how I'm the layout style
how it will react when it goes on a
mobile device in terms of its Flex
layout or but the block properties this
is where the actual
models reside
and we support both newer and older
versions of
of unity as an example these are where
the files reside and if I look at the
data the that is that resides on this um
certainly just um
so the data sources is the pump readings
um but on this overall page these are
all the data sources and if I look at my
live data where it's coming from
that's actually from the data source so
you'll see the
um
this is this is not the expression don't
want to bring the expression so I just
want to find the data source for it so I
can build very sophisticated
data connectivity this one is using
um that that data stream
connector it could be using a digital
twins it could use maintenance schedules
SQL data there's many different
connectors that we have oops
don't want to delete that as you can see
there's some some built-in capabilities
for people like myself that are not
coders or so it will help with not
screwing it up completely so
this is and also starting a new
application you can start from templates
so you can say well this one with the um
this pump one I'm not sure what so
you'll see this one has actually got a
series of drill down pages and
everything that's already associated
with it so I can use this as a starting
template to get started quickly
so this is how we look at
um
providing digital twin capabilities in
terms of Maintenance predictive
maintenance condition monitoring for
facilities management
if you have any questions please reach
out and happy to address them thank you
Last updated