214
the core are often junior members of staff and their lack of experience can mean important
features are overlooked or misclassified. Inconsistent logging can have a flow-on effect to the
quality of the geological models produced.
Mineralisation is controlled by geological features. Errors in the modelling of the geology
will cause errors in the understanding of the mineralisation distribution. This is particularly
true when computers are increasingly relied on to build the geology model. An artificial intelligence technique called implicit modelling uses algorithms to generate the geological shapes.
Although inconsistent logging will impact the quality of a human created model, a human has
some capacity to generate models that account for the inconsistencies. However, a computergenerated implicit model based on the same data will simply follow the input data and produce
a model that is inherently wrong. As companies become more reliant on computer generated
models, the possibility of lower quality models caused by low quality input increases.
The concept of geological consistency extends to stratigraphic sequences. Often there is an
expected lithological order down a hole and the geologist assumes to see rocks within a drill
hole in this order. If this sequence is not observed, it could indicate the geological sequence
has been disrupted by folding or faulting, or there is an error in the logging. Artificial intelligence can be used to identify drill holes that fail to conform to the expected pattern. Such
holes could be flagged for review by senior staff.
Understanding the engineering strength of a rock effects the stability of the mine. Rocks
with lower strengths require lower angles for pit walls and greater support underground.
Major factors in determining the rock’s strength are understanding the joint patterns, density,
geometry and joint infill. The geotechnical properties of the rock not only effect the safety of
the mine but also its economic viability. It is therefore highly advantageous to train a computer
system to identify and measure joints and fractures in the rock from core imagery.
As the mining industry becomes more automated, the need for equipment to monitor itself
and correct its actions based on monitoring results increases. An example of this is an automated blast rig drilling on a coal deposit. The blast holes are drilled so explosive charges can
be used to fracture the rock overlying the coal. This makes it easier for equipment to remove
this overburden. The blasting should not extend into the coal seam. The holes drilled for
blasting should stop before the coal is reached. Having drill rigs that can detect the approaching top of a coal seam and so stop drilling is helpful.
Often when a hole is drilled, geophysical instruments are lowered down the hole to take measurements of the surrounding country rock. These measurements provide a valuable source of
objective data. Certain strategic horizons have very pronounced geophysical signatures. This is
particularly true for coal seams. These signatures could be used to help identify known coal seams.
The geophysical traces are useful in correcting the depth of samples and observations. As
the hole is drilled, the sample depth is estimated. Unfortunately, drill cores fracture or sections of the core can be destroyed in the drilling process. This makes the depth estimations
difficult. By comparing the geophysical traces of the surrounding country rock with the rock
in the core, it is possible to adjust the depth of observations and samples to a value closer to
the real collection depth. This process is currently undertaken manually.
4 CASE STUDY
Not only can machine learning assist in the areas of process efficiencies, data quality and
prediction, but it can be used to extract information from historic photographic datasets.
Core tray imagery contains a wealth of textural, mineralogical and geotechnical information.
However, this data source is poorly utilised because of the arduous manual process of clipping the regions of core from each core tray image.
MICROMINE Pty Ltd (MICROMINE) undertook a research project as part of the
Newcrest Crowd “Get 2 the Core” online competition. Machine learning was applied to
extract the core outline from core tray images. The spatial extent of the tray in each image
was determined and within that region, the rows containing the core were isolated. Masking
core tray images to isolate just the core is difficult because the tray boxes can be made from a
the core are often junior members of staff and their lack of experience can mean important
features are overlooked or misclassified. Inconsistent logging can have a flow-on effect to the
quality of the geological models produced.
Mineralisation is controlled by geological features. Errors in the modelling of the geology
will cause errors in the understanding of the mineralisation distribution. This is particularly
true when computers are increasingly relied on to build the geology model. An artificial intelligence technique called implicit modelling uses algorithms to generate the geological shapes.
Although inconsistent logging will impact the quality of a human created model, a human has
some capacity to generate models that account for the inconsistencies. However, a computergenerated implicit model based on the same data will simply follow the input data and produce
a model that is inherently wrong. As companies become more reliant on computer generated
models, the possibility of lower quality models caused by low quality input increases.
The concept of geological consistency extends to stratigraphic sequences. Often there is an
expected lithological order down a hole and the geologist assumes to see rocks within a drill
hole in this order. If this sequence is not observed, it could indicate the geological sequence
has been disrupted by folding or faulting, or there is an error in the logging. Artificial intelligence can be used to identify drill holes that fail to conform to the expected pattern. Such
holes could be flagged for review by senior staff.
Understanding the engineering strength of a rock effects the stability of the mine. Rocks
with lower strengths require lower angles for pit walls and greater support underground.
Major factors in determining the rock’s strength are understanding the joint patterns, density,
geometry and joint infill. The geotechnical properties of the rock not only effect the safety of
the mine but also its economic viability. It is therefore highly advantageous to train a computer
system to identify and measure joints and fractures in the rock from core imagery.
As the mining industry becomes more automated, the need for equipment to monitor itself
and correct its actions based on monitoring results increases. An example of this is an automated blast rig drilling on a coal deposit. The blast holes are drilled so explosive charges can
be used to fracture the rock overlying the coal. This makes it easier for equipment to remove
this overburden. The blasting should not extend into the coal seam. The holes drilled for
blasting should stop before the coal is reached. Having drill rigs that can detect the approaching top of a coal seam and so stop drilling is helpful.
Often when a hole is drilled, geophysical instruments are lowered down the hole to take measurements of the surrounding country rock. These measurements provide a valuable source of
objective data. Certain strategic horizons have very pronounced geophysical signatures. This is
particularly true for coal seams. These signatures could be used to help identify known coal seams.
The geophysical traces are useful in correcting the depth of samples and observations. As
the hole is drilled, the sample depth is estimated. Unfortunately, drill cores fracture or sections of the core can be destroyed in the drilling process. This makes the depth estimations
difficult. By comparing the geophysical traces of the surrounding country rock with the rock
in the core, it is possible to adjust the depth of observations and samples to a value closer to
the real collection depth. This process is currently undertaken manually.
4 CASE STUDY
Not only can machine learning assist in the areas of process efficiencies, data quality and
prediction, but it can be used to extract information from historic photographic datasets.
Core tray imagery contains a wealth of textural, mineralogical and geotechnical information.
However, this data source is poorly utilised because of the arduous manual process of clipping the regions of core from each core tray image.
MICROMINE Pty Ltd (MICROMINE) undertook a research project as part of the
Newcrest Crowd “Get 2 the Core” online competition. Machine learning was applied to
extract the core outline from core tray images. The spatial extent of the tray in each image
was determined and within that region, the rows containing the core were isolated. Masking
core tray images to isolate just the core is difficult because the tray boxes can be made from a