Lyceum 2021 | Together Towards Tomorrow
Open pit and underground mining operations dedicate significant resources to collecting costly and highly detailed datasets, often with a narrow scope of use.
As a geoscientist, gaining access to such information can provide an opportunity to re-asses the understanding of mine scale structural geology and mineral system, with significant implications for assessing geotechnical risk. This session showcases some novel avenues of investigation relevant to the often-neglected overlap between structural geologists and geotechnical engineers.
Senior Structural Project Geologist, Mining One
(bright electronic music)
I am Anthony Reed, a Structural Geologist
working with Mining One,
and today I’m presenting analysis of high resolution
and consistent data for structural geology
and geotechnical studies.
I have to put your standard disclaimer in there
not to make any critical business decisions
based on this presentation alone.
But do reach out to me if you’d like
a similar treatment of your data.
So a little bit about Mining One,
we’re one of the largest mining consultancies in the world
at the moment, and we service, primarily,
mining companies of course,
but actually provide services to a number of other
sectors that are related to mining as well.
We have offices all over the world
and we operate in all those orange countries
that you can see on the screen there.
Got most of the big players in the mining industry,
as clients already, but always happy to add some more.
And we cover everything in the mining space,
from conceptual stages through to the implementation
of underground and open cut pits
all the way through to mine closure.
And you can see from that list,
generally everything that you need to run a mine,
we’re involved with.
Honing in on the geology space,
we do cover all stages of geology from early exploration,
through geoscientific modeling,
mineral resource modeling and reserve estimation,
independent valuations for internal or external valuations.
We can design drill programs,
and we have competent people qualified
under JORC and the NI-43-101 codes to sign off.
We conduct a QA/QC and have independent experts
qualified under VALMIN.
In the geotechnical space,
we cover pretty much everything to do with mine stability,
so very large geotechnical team.
And while there’s a large list
that I’m not going to go through,
I’d like to focus on down the bottom,
our partnership with a software package called Cavroc,
which is essentially a front end plugin
for the FLAC3D geotechnical modeling software,
which is mostly command line based.
It consists of an Octree Mesher
which brings geotechnical data into the block model space,
much as you would a resource modeling technique.
It’s got our IUCM,
which is a physics engine that uses that data
to model rock performance.
And then we’ve got two additional plug-ins,
StopeX and SlopeX, so one’s open pit
and one’s for underground,
which are GUI’s specifically designed for FLAC3D
to make the entire process far more accessible
to people on sites,
and it turns into a very agile and iterative process.
So that’s the sort of thing
that we would roll out for mines.
Now, why do I care about Cavroc?
So I’m a structural geologist and I’ve been working very
heavily with geotechnical engineers
for over a year at Mining One.
And what a surprise,
we have significant overlap in our spaces.
In the rest of the industry that I’ve been involved with,
structural geologists and geotechnical engineers
are usually kept at arm’s length,
but working together we’ve both had our eyes opened
into each other’s fields.
So the sort of data that’s particularly useful
for geotechnical modeling, that I also care about,
are discrete faults,
diffuse faults and zones of weakness, alteration zones
and halos, lithological volumes and boundaries,
which all geologists are familiar with,
and any anisotropy that’s in the rock mass, i.e., foliation.
Back to the title screen,
what do I mean by high resolution, consistent data sets?
Well high resolution sort of self explains.
Consistent, I mean a data set that is collected
with a well understood method
that’s very well leveled and very repeatable.
So things that I would normally consider that fit that bill
in the geology space are assays from Diamond Drill
and RC Core for exploration of resource,
hopefully that data’s in good quality
if you’ve got an operation.
More in the geotechnical and mining space,
we have huge data sets such as blast hole assays,
which geologists don’t usually get to see,
geotechnical parameters get logged, such as RQD,
machine learning derived data, such as color or clustering
on core or rock chips.
You can have observations directly from modeled geometry
of other objects,
so things like implicitly modeled orebodies
or other geology surfaces, and of course, photogrammetry,
which I’m sure everyone is into now,
and laser scanning for things such as stope pickups.
They’re all good examples of very high quality datasets.
I’ll start with the geotechnical data
and what that looks like.
So RQD is not something that I’ve played with
a hell of a lot in the past, as a geologist,
doing geological demanding,
but it’s a very consistent data set that remains consistent
from operation to operation.
It’s a fairly simple calculation.
Even when it’s logged manually,
you do still have to consider that there’s a human factor,
and occasionally you’ll get some leveling issues
with different people logging during different shifts.
But mainly the only problems with the data
is things like the first five meters of holes
sometimes have to be excluded
because you are hitting a rock with a large machine
and it doesn’t always reflect actual rock quality.
Things like drilling artifacts can be stuck in the database,
like Navidril or Wedges that need to be taken out.
Cover sequences can swamp any deeper weaker signals
that you might want to consider.
And occasionally you might want to throw out entire
campaigns, like the exploration call from the surface.
It could be something to do with drilling technique
rather than logging, but you know,
sometimes you want to make sure
you have a really consistent data set,
so that has to be done sometimes.
And what are we looking for?
Well the image on the left-hand side
is a really nice planar fault.
You’ve got a lovely, bright red signal there,
and that’s something that can be digitized and wireframed,
and you’ve got a brilliant fault
that can head through to your geotechnical modeling.
And what we would do with this data set is zoom around
and rotate it until we can find all of those
very clear planar features and model them all up
in a fault network.
Once you’ve done that,
that data can then be excluded and you can re-level
the remaining data set to look for weaker signals.
Now on the right, you can see from the upper left
to the lower right of that image with all the drill holes,
some fairly strong striations in that data set.
Remember this is RQD data,
so that is showing a rock weakness
that is lighting up in a plane.
It turns out that they are, or one of them at least,
is in the same position as a logged slate horizon
in this deposit.
However, you can see there’s a number of striations
and we really only have a single slate horizon logged,
so something else is going on.
So in this case, we did some further investigation
and found just enough evidence in the rest of the logging,
looking at a whole myriad of codes
that could be related to shearing and slates.
But yes, we do indeed have a lot more slate horizons
that are heavily sheared in this deposit.
And so we’ve used the geotechnical data
and interpreted that as a lithology log
to build a whole stack of new features.
In addition to that,
we found these ramping structures that run between them,
and after a lot of discussions with the site geologists,
as well as looking at core photos,
the entire set of slates has kind of been reclassified
as a set of a shear zone network,
where pretty much all of the things that are modeled
as slates and these ramping structures between them
are some sort of fault,
so they’re very, they’re geotechnically important.
They’re very, very thin,
but we finally have a way to visualize them and model them.
Now from a geotechnical modeling’s perspective,
they’re so thin that in the block model space
they don’t take up enough volume to even assign blocks
with their own sort of rock parameters.
So instead, in this case, we would have treated them
as discrete faults and fed that through
into the Cavroc plugin for numerical modeling.
Blast hole assays are a absolutely brilliant data set,
and as a geologist, I’ve only recently started asking
to look at these.
But while they usually lack in a breadth of assay types,
they make up for it in the sheer amount of data
that you get.
So what we’re looking at here is iron assays
up high on the pit wall of a deposit.
Well and truly outside of your normal resource drilling.
So this is essentially a blank slate up here
until we get the blast hall assay data.
In this case, it’s been deemed of critical geotechnical
value to understand where the dikes are in this deposit.
And a lot of them haven’t been picked out from the previous
drilling or mapping efforts because they’re difficult to see
on the ground, and we don’t fully understand the geometry
from these sparse resource drill outs.
But as you can see from the blast holes,
they just shine.
Blast holes at any deposit really give you the best insight
into the shape of your deposit.
And a lot of how their mineral style is actually working
for you there as well.
With some aid of photogrammetry,
’cause I had it in this case,
but mainly tracing those features
we were able to build that into a dike network,
that again, gets fed into your geotechnical model.
Now in the previous images we had iron highs,
but this time we’ve got an economic mineral
and you can see the dikes are taking that material out,
and so you’d be tracing the lows in this case.
And frequently the same feature can track from a signal
of being anomalously high to anomalously low,
as it tracks through an orebody.
Now from sparse resource drilling,
from one drill hole to another,
the rock can look completely different and you cannot link
those features, but in blast holes,
the drill spacing is so close and the data
is such high quality that you can see
that it’s actually the same feature tracking
all the way through.
So using this sort of data to reconcile your geology model
and your understanding of your orebody
is really quite an amazing opportunity.
CMS stope scans, so once we have a stope,
we often send in the CMS to laser scan that stope.
And in many cases, especially where we have overbreak,
everyone runs around
hoping that they can get more information
about why a stope overbroke.
So in this case, you can see from upper left
to lower right, there are a series of structures
that look like they’re involved in some overbreak
in this image.
You’ve got large blocks breaking back
to a very planar feature.
Now those were joints or faults that were not picked up
in the logging, not picked up in the mapping,
but clearly quite important in this area.
And so from analysis of this series of CMS scans,
we’ve identified in the yellow ovals
areas that we deemed risky,
even though they’re adjacent to a fallen stope anyway,
so they would have been thought of as risky,
but we’ve decided there’s a mechanism there.
Lo and behold, two months later,
we’ve got the rest of the data,
and yes, those zones were of risk,
they overbroke in the same way,
and there’s a whole lot more data to suggest that those
faults were involved in that area as well.
So, you know, the structures that are important
to the stope scale, are best visualized in the stope itself.
So it’s worth your geologists getting in there,
looking at the scans of the stopes
after they’ve actually been mined out.
Often, not an accessible dataset to a geology department.
So mapping on high resolution photogrammetry,
everyone’s on the photogrammetry bandwagon,
but I’m going to look at it again anyway,
just to show you what we do with it.
So what we’ve got is a couple of really obvious structures
sitting in this image.
You know, we’ve got one on the left-hand side
that’s colored by our dip direction on the right-hand side,
it’s just the photographic image.
And there’s a number of ways to model such a feature
for use in a geotechnical model.
So for instance,
we can see features where blocks
have fallen out of a pit wall
and literally just trace it up multiple benches.
And we can do that with a series of structural disks,
if we can really pick up the orientation well.
And if we can’t pick up the orientation,
then we can use striations,
so things like polylines or points to tie it together,
and often there’s a combination of the two.
Part of the problem that I have
with some of the more automated data sets that you can get
from Surrovision or ADAM tech is that the photogrammetry
meshes themselves, do have some inaccuracies.
And so if you’re trying to perform structural geology
on this data, you can’t always rely on an automated pickup
with flat faces.
You do need to have your geological brain on.
Does this make sense?
Is it a planar feature?
Does it bend while you’re doing the interpretation?
And from all that interp, we can make a structural network,
as you can see on the right-hand side there.
So just a little bit of a case study of what to do
with some really complex structural environments.
So this is, we had some failures in a pit,
and it was deemed very
important to get a structural geologist out there,
to do some mapping and find out what we could.
Only 12 out of the 40 days he was on site
was he able to even access the pit.
There was logistical issues with even getting in there.
The maximum approximate density that you can collect
when you’re walking around the pit is about one measurement
every 20 meters, realistically,
just in terms of time constraints.
There are heaps of inaccessible areas that we really needed,
of course, the ones close to various failures
are the ones that we didn’t get data for.
The positional accuracy of the GPS wasn’t particularly good,
and the relevance of individual measurements is questionable
because you really can’t tell a global context
when you’re walking out.
So in order to augment this study,
we decided to do a drone survey
and we have covered the entire pit in this,
but this is one little example of one little corner.
Some ground control points, just using a consumer drone,
a simple inclined grid planned with a back and forth
And we processed fit for purpose, high detail products
within a day using Agisoft Metashape.
And this is what they look like.
So one of the joys of doing it all in house
is that we can tweak the final products
to be exactly what we need to pick up the features
that we want.
And if it’s not quite right,
we can dial up that detail and reprocess it again.
Now, if you’re dealing with a open and shut
request to provide photogrammetry data
from someone that’s not involved in the structural or
geotechnical model, often the product that is returned
is not always appropriate to do that level of mapping.
The outputs, we configure them to be compatible
with all the major stakeholders on the site as well,
so everyone was able to have a look at this data.
In this particular case that we’re seeing on the screen
is this 30-meter-high slip along bedding.
There’s no physical access to this location.
There’s rubble, there’s overhangs,
and even if you were able to walk up to it on foot,
you’re unlikely to get a representative measurement.
As you can see, that slip surface is bending
quite considerably, even just this plane of view.
So instead took the photogrammetry output into a package
called Cloud Compare and in a combination of picking faces
in Cloud Compare and manual digitizing in Leapfrog,
we’ve got a decent dataset together of bedding and joints,
all digitized on the computer screen.
Sighting down measurements for Leapfrog digitization,
which is often a little bit better when you’ve got,
you know, rough edges, but you can still see a good feature.
And the context is decided during the digitizing process.
So I knew I wanted to make a structural model,
so I digitized with that context in mind,
rather than just ending up
with a cloud of unlabeled measurements.
It took only about two hours to do this entire area.
So a similar number of measurements
to our structural geologist friend
that was there for, that was in the pit for 12 days.
It’s not automatic or replacing a skilled geologist,
it requires a skilled geologists to do this work.
We haven’t quite managed to automate that out of it yet.
We picked up all the ethanol and the joints of any visible
outcrop, and it’s easy to refine targets
for walk-up in the pit, if we’re not sure about something.
More to the story, three years prior,
there were some pseudo bedding models created.
And it certainly looks like from the old measurements
that there’s a scoop geometry that aligns pretty well
with where the failure was in the pit.
So the thought was, well we think we understand
that there is a failure risk along bedding.
Can we highlight that risk in great detail
on the photogrammetry?
Often in a pit study, I’ll be asked,
can I please have a dip and dip direction
of the features that we’re interested in?
And as you can see from that image there,
the feature that we’re interested in,
doesn’t have a single dip and dip direction,
it meanders all over the place,
so the answer would be, no,
we need to find some way a little bit cleverer to do that.
And hopefully we did.
We interpolated the unit vectors from the raw bedding data
that was taken from the photogrammetry and the previous
studies and mapped them directly
onto the photogrammetry data set.
So we’ve found the angular difference calculated
between the rock face from a photogrammetry
and the expected bedding,
and there’s a couple of little issues with needing,
we need trig functions in Leapfrog
to make this a bit more accurate,
and if we can get access to the,
the form interpolant algorithm
to sort of directly map values onto points
that would make this a lot more accurate as well.
But if you’re close to data,
this method works pretty well.
And this is the output.
So what we’re looking at here is that same survey area,
where we have very bright yellow,
is very close to aligned to bedding.
So faces in the rock that exists during
a photogrammetry survey that match the bedding orientation.
And down to the red pixels are five degrees away from that.
So the brighter, the color,
the closer to the bedding we were.
And you can see from that image,
that in the zone of the failure, towards the left-hand side,
there’s a huge area of yellow, and over to the right,
where the bedding veers away from the exposed,
from the pit plan, from the bench angles,
you get little planes of failed rock,
but none of them are really huge.
So from that data, we turned a couple of hundred
measurements manually into possibly millions
and picked up everywhere in the rock face
that looks to be relevant to bedding failure.
And from that we can evaluate, you know,
risk at any given location
based on what we can see in the pit walls.
But I think it’s a good case for having
some augmented reality as well,
where you could walk around the pit and see
what’s in front of you.
What is bedding?
What is a joint?
Is it close to your expected angle,
or is it something completely different?
Okay, onwards to our implicit vein
model structural analysis.
So we’re looking at the analysis
of an implicitly modeled object.
In this case, it’s a high grade ore lens.
It’s a fairly planar feature,
but there are a number of structures that interact with it.
We know from looking at the core and underground,
that there are structures in a number of orientations
that are important, but their persistence,
the significance of them and how they interact,
and yeah, it’s a little bit difficult to work out.
I call this a consistent data set
because we’ve got fairly good drill hole coverage
for this object, about 12 1/2 meter centers
across the entire thing, along a strike of 2 1/2 k’s,
and down dip of about one kilometer.
And it’s modeled from assays.
So even though there’s some geology interp,
it’s still a fairly consistent shape,
and every contact is very similar to the next contact along
in terms of how it’s defined.
So is there a way to get a bit more value
out of seeing those perturbations in that shape
than just looking at it?
And the answer is yes.
Again, we’ve taken the vertices from that object
into Cloud Compare
and looked at the different dip direction
and the constituent unit vectors that are orthogonal
to that mesh.
And in the top left-hand side,
you can see our dZ unit factor,
which is actually fairly similar to dX and dip
in this instance,
because it’s a north south striking orebody.
And we can see those horizontal features that were visible
are now far more continuous.
And we can also see some features that are ramping up
towards the left as well, that are fairly clear.
Beneath that we’re coloring with dY,
which is analogous to aspect in this case,
a different value, but a similar pattern.
And so some features that are running up and down
the length of this are seen quite prominently
in the purple and the green bands next to the purple.
The top right-hand side, we can see thickness,
which is calculated from the Leapfrog vein algorithm.
And those horizontal features can be seen causing a extreme
thinning of the orebody in the same location.
So not as much with the ones that ramp up
towards the left-hand side,
but definitely those horizontal ones are important.
The vertical ones in that image
appear to be thickened rather than thinned as well,
so starting to understand what the structures are doing
and what their angles are,
how they interact with the orebody
and why they’re important.
In the lower right-hand side, they’ve all just been traced.
So running around with the digitizing tool
and picking out everything we can see
from any of these datasets.
And of course there was a large amount of mucking around
with color and contrast in order to really understand
where everything was.
And from that, we generated a structural network.
Now we didn’t necessarily have the exact angles
for this study.
They can be tied into underground observations
when there’s time,
but what we did get out of this is an understanding
of where families of structures interact with the orebody,
where they interact with each other,
and what would be considered a structurally complex zone
in this orebody versus one that’s not.
And that is quite an important outcome.
So we went from,
we went from an understanding
that there are structures everywhere,
and we have a risk everywhere in this orebody
to really nailing down some locations of importance.
Finally, so all of this is innate of,
fill in a geotechnical numeric model
so we get the best evaluation of rock quality,
rock mechanics, and risk we can in open pit or underground.
Now Leapfrog does have a part to play
in populating this block model as well,
because, especially with the edge module,
we can populate a number of things into a block model space
that is useful in Cavroc.
So things like lithological instructional domains,
structures themselves, structural trends,
and they can be individual per domain,
or they can be global.
And we can feed, yeah,
we can feed the trend data into Cavroc as well.
For the geotechnical logging, RQD can be interpolated
fairly nicely in Leapfrog.
You do have to flip it around to do 100 – RQD
as the RBF algorithm likes to contour around high values.
And so I find that visualizing RQD as it’s being
interpolated is an extremely powerful thing
that a lot of people don’t do.
And you can ensure that your RQD interpolation
is geologically reasonable before it gets sent onto
a geotechnical numeric model.
Orientation interpolation, there is actually a workflow
using edge to get downhole measurements
through into the block model space
that involves creating form interpolant surfaces,
and then using VAT to populate a variable
interpolation field in edge.
But you can see from the lower right image,
the final result of that is an evaluation of angle
from the input data at every centroid in a block,
which is a particularly useful thing.
The final geotech block model,
in this case from Leapfrog only,
I was able to populate the domain, foliation orientation,
foliation intensity, the RQD, distance functions to objects,
and distance to reliable data in the model.
And so while it’s not my place to evaluate risk,
I was able to do little calculations
like looking inside the orebody for poor quality rock
that’s highly foliated
that’s sitting next to some valid data and pointing out,
areas of particular interests, geotechnically,
that are worth focusing on.
And that block model can just be delivered directly
to our numeric models.
So some final thoughts,
there’s a lot of opportunities for geology
in the geotechnical space.
Geotechnical engineers often have access to a lot more,
really high quality data than a geology team
would normally see.
And so it’s worth the two teams working together,
the data of critical mass to do this level of interpretation
in the structural space,
and just in the orebody knowledge space
is usually when a mine is more mature
than when you’ve got your exploration department going
great guns on your deposit.
Interpretation and reconciliation of the geology model
using these massive data sets at a mine
is not commonly done from what I can tell.
Most of the data is restricted to very narrow use cases,
such as the blast hole databases being for mining block
models, rather than being seen as a opportunity
to really understand your orebody again.
And better geology always feeds into better
geotechnical modeling, the understanding and the products.
So it’s really worth having this partnership
between geologists and geotechnical engineers.
Sadly, geotechnical numeric modeling is often the only good
justification for getting this work done.
They tend to be the ones that have the budgets and the need,
and the link directly to production,
rather than a structural geologist, such as myself,
who would often only be asked to do this
as part of a more academic study.
So hopefully what I’ve shown is that
it actually does provide a level of rigor to geotechnical
numeric modeling to get some really high quality structural
geology in there,
and especially from these datasets that geotechnical
engineers get to use every day.
So thank you very much for your time.
(bright electronic music)