Lyceum 2021 | Together Towards Tomorrow
Susan’s presentation focuses on how to break silos between data and disciplines, and use geospatial visualisation to promote transparency of material properties in real-time, both on-site and remotely.
Discover how to take a whole-system approach to the mine value chain and bring decision-making into focus with proactive and dynamic influence characteristics. Find out how the IntelliSense.io Stockpile Management application can deliver value to your operations and increase production.
Global Field Engineering Lead, IntelliSense.io
<v ->Hello, my name is Susan Kennedy.</v>
I’m here from IntelliSense.io.
We are an industrial AI company
focusing on the mining sector with a mission
to make mining operations, efficient, sustainable, and safe,
through trusted and transparent AI solutions.
We optimize every step of the mining value chain.
Today, I’m here to talk about stabilizing mine production
with end-to-end material tracking.
Within your mine,
high resolution visibility of your material
means predictable and stable material input
into your plant.
In your plant, the foresight into the incoming material
allows optimization of the plant controls and decisions.
End to end material tracking breaks down silos
between different disciplines and data onsite
and aids the realization of value of mining operations.
On your screen, you can see
our intelligent mine experience center,
which we have developed with our partners BSF.
Which is as you can probably make out a digital mine.
Now we’re here to talk about end to end material tracking.
So the obvious place to start this story
is in the resource model, underneath the pit.
Within your resource model,
you have information on your material properties
and where these are spatially located.
Once you blast, you may have a model
of how this material is expected to have spread.
And from there you will go to either your plant
or a stockpile.
So in this case, I’ve got open cut mine,
and I’ve got trucks that are hauling the material up
to a jaw crusher here and onto a conveyor belt
into the remainder of the plant.
But it’s actually the stockpiles
that I want to talk more carefully about.
So if we’re not going straight to the plant,
then we’re probably going to a stockpile.
And this is where information,
a lot of information is traditionally lost
or heavily diluted.
So I’m going to talk to you about the hurdles
of material tracking in traditional stockpile management
and how we have overcome this with a spatially aware model.
I’ll take you through a tour of the model
and how it integrates with other software.
And then I’ll take you into IntelliSense’s platform
Brains app, where I’ll talk further about the value.
After we talk about stockpiles, we’ll go into the plant.
To talk about the value of more stable
and predictable material information within the plant.
I’m going to talk specifically about heap leaching
as an example, but the concepts
that I’m going to talk about apply to other areas
of the plant as well.
So to talk about stockpiles, on the left of my screen
is a typical model of a stockpile,
a Weighted Average Model.
In a model such as this,
all material information is averaged.
And during planning and reclaiming
material variation cannot be distinguished.
Are the models currently in use include first in first out
or last in first out,
but all of these models are quite simple,
with the result being,
not having an understanding of material distribution
and potentially unstable and unpredictable feed
into the plant.
A spatially aware model is far more representative
of reality and that’s what we have developed
and what you can see on the right of my screen.
Using a model like this,
you can manage both dumps and reclaims better.
This model can be used to either plan and track blending
within your stockpile,
or strategically plan reclaims to suit your mine plan
and achieve more stable results
as the blend goes into your plant.
To show you a real life example of results achieved,
on the screen is a mine plan versus plant actuals
at a copper mine.
The red and blue bars are percent difference
and standard deviation of versus actuals
over a four month period prior to the introduction
of our Spatial Model,
when a Weighted Average Model was used.
And then the green and yellow lines
are after our Spatial Model was introduced,
which we call GRIDS, also over a four month period.
All sample days or data points were taken
from when there was over 60% plant feed from stockpiles.
So these days we’re heavily reliant
on stockpile information accuracy.
And as you can see using the Spatial Model,
they have been able to get much, much closer to their plan.
So I’ve just jumped into Seequent Leapfrog.
Now, as I said earlier,
I was going to talk about integration.
Our application is agnostic to systems
and can use data from various sources.
Typically FMS if it’s available, mine block models,
topographical scans and SCADA systems.
We’re also using machine learning and AI
to improve anomalous data.
Outputs are also integratable with other applications.
And our 3D block model can be opened up into any geological
modeling or visualization tool.
And as I mentioned,
I’ve actually got one of our models open now
within Seequent Leapfrog.
So this is a copper stockpile I’m displaying
and at the moment I’ve got total copper on display.
So you can see the spatial distribution of the copper.
Now I can change this to any other material property
that I’m tracking and I can track whatever
material properties that I have information on.
So for example, I’ve just changed to arsenic
and I can also see the spatial distribution
of arsenic within the model.
Another thing I can see within the model
is how long the material has been sitting on the stockpile.
So as you can see, this stockpile was created
from the Southeast and then went Northwest.
And this shows me the age in weeks of the material.
And this can be particularly important
for any tufts of material where oxidation is important,
for example, in a coal mine.
Now the model behaves realistically,
it’s not simply discrete blocks.
Within the model, the material blends and moves.
It flows downhill just as it would in real life.
If there are dozers on the stockpile,
either trying to blend or simply manage the stockpile,
the material movement and blending is also taken to account.
And ultimately information is retained.
You are no longer diluting information you had
from your Institute model.
And as I said, from here, you can start making decisions
on the information that you have.
So now I’ve jumped into the Brains app,
which is our platform where I’m displaying
a little more information about the stock piles.
So I’ve got open, a material reclaimed screen
about the Stockpile Calipso.
And on this screen,
you can see a lot of information about the material
that has been reclaimed from the stockpile.
If I show you this graph, you can see how the model
has spatial understanding.
So instead of being a flat line of your reclaims
from a stockpile, you can now see
that the material information that I’ve got displayed,
and I’ve got total copper recovery
and concentrate displaying on my screen,
how they vary depending on exactly where
the material was reclaimed and the material properties
at that location within the stockpile.
Traditionally in a Weighted Average Model,
it would look more like this red line.
It might be just completely flat or maybe slightly sloping.
And the difference between the red line and the blue line,
blue line representing what’s actually happening
within the stockpile,
is the difference between what you think you’re getting,
with a Weighted Average Model
and what you’re actually getting.
So in this example, we’ve got 2.05% copper,
is what you would think you’d be getting.
Whereas you might actually be getting 2.2% or 1.85%.
And so with this,
you’re not only not reaching your feed plan,
but if you’ve got levers to pull in your plant
for recovery or throughput like additive
in your concentrator, you will not only,
you’re also not optimizing your output
and there’s opportunity cost there.
So it’s not only about reclaiming from your stockpile,
but there’s also the opportunity to manage
the build of your stockpiles and actively manage them
in real time.
So I want to show you a couple of interesting cases
that we’ve seen.
In the first one, there was over a hundred thousand tons
of waste, unknowingly dumped onto a stockpile
when some road widening activity was being carried out.
This was clearly seen in the 3D block model,
as represented by the blue area in the model
and the alerts in the application as shown
on the bottom in the, as shown as out of range dumps.
So this means it was out of range
for what was expected for that stockpile.
If this has entered the plant,
the opportunity costs of not processing ore instead of waste
was over 5 million US dollars.
So instead the material was isolated from the stockpile.
In the next example, 20,000 tons of ore
was used to create a platform in a pit, again unknowingly.
The material was labeled to go to a stockpile,
which you can see within the red boundary on the screen,
but was actually dumped in a pit.
And they found out was used to build a platform.
The data quality screen that you can see
on the bottom right, was able to highlight the anomaly.
And a plan was put in place to reclaim the material
once the platform was no longer required.
And this ore was valued over US 1 million.
So to continue talking about stockpiles.
Now, once the material comes from a stockpile
and goes to the plant, what you can see on my screen
at the moment,
is the combination of various different sources.
So in the green here,
I’ve got 52% of material is coming from my pit,
going into my plant.
And the remainder, 48%,
is coming from three different stockpiles.
And you can see in this graph,
the combination of all those different material sources
and the ultimate material properties entering the plant
I’ve got the average over that time,
down the bottom in these text boxes.
I have 11 days filtered, but you might want to filter
for a few hours, for six hours, for a shift, for a day,
so you can see exactly what is coming
into the plant in near real time.
So it will hit your crusher and then that will be,
then the people in the plant will have an understanding
about what is coming in very soon into their plant.
So this gives them more information
about what is entering the plant
then what was just in the mine plan.
So this will allow decisions to be made
to optimize recovery or plant output.
So generally the various solution panels available,
such as the two that I’ve shown you,
are useful by geologists and grade controllers
my mine plan is to check compliance to plan
and the plant team to understand
what material is entering the plant.
So if we take this and move to the plants now,
as I mentioned, I’m going to talk about heap leaching,
but the concepts of what I’m talking about
can be applied to other areas of the plant.
I’m going to talk about two challenges within heap leaching
and how material tracking can improve the outcomes.
The first challenge is the presence of highfine material
causing poor leaching kinetics,
and excessive consumption of asset or cyanide.
Secondly, the presence of moisture in the tailing
causes the material to behave with greater fluidity,
which can prevent the construction of a stable heap,
reduce the rate a rotary shovel can reclaim
and result in increased storage requirements
and thus required environmental permits.
We use machine learning to model the influence
of the different material properties
on process performance metrics,
giving the marginal contribution of each factor
to every prediction made.
And this is what we call the Material Influence model
in the screen have now open.
So by knowing the planned and actual material input
to input to your plants,
you can the fine fraction and residual moisture.
So on my screen here,
these limits here in red are what we want to stay under
and this is our prediction of fine fraction.
So you can see, we want to keep the green and blue lines
underneath this red line.
And the same can be said
for this residual moisture prediction.
So we want to keep our residual moisture
underneath this red line.
Now, how do we do that?
So we can look at the material influence.
So once we know when the fine fraction
or residuals might likely to exceed the limit,
this is where we go.
For the fines, some of the influencing factors
are linked to the material properties
and some linked to the crushing circuit.
And it is a similar story for residual moisture.
Adjustments can then be made when high fines
or high moisture is predicted.
This can be adjustments around the crusher blending ore,
or adjustments to your agglomeration variables,
such as adding more binder.
For moisture content, blending is also a possibility
or adding salt.
And down here, I can see the opportunity cost of action
is not taken.
So for this particular timeframe that I have filtered,
which is a month, which I can see up here,
if I don’t take actions,
when the predictions are above this red line,
then there’s a potential to, there’s opportunity costs
of over a hundred thousand US dollars.
So this is just one example in the plant
of how effective material tracking
can not only stabilize your product,
but also increase the value of production.
And as you can probably imagine, from what you’ve seen,
end to end material tracking helps break down silos
between disciplines and data and facilitate constructive
conversations between teams.
And that is all from me today.
Again, my name is Susan Kennedy from IntelliSense.io.
I hope you found that interesting and useful.