Data-driven geological modelling might be on the way to being as “ubiquitous as mobile phones and the internet”, but machine learning models are only just starting to unlock significant value from deeper in the mine value chain, this week’s AusIMM International Mining Geology Conference heard.
PETRA Data Science founder and CEO Dr Penny Stewart told the conference enhanced “orebody learning” from the fusion of mining operational data – including tens-to-hundreds of millions of tonnes of ore mined and processed at sites each year – and geological and other data, could materially enhance drill and blast, mobile fleet and processing outcomes.
Machine learning models are also helping unlock new operational optimisation opportunities and provide fresh insights into mine orebodies. The changes could position mine geologists “at the centre of the modern mine”.
“Orebody learning is really expanding the role of mining geology to help other mining professionals across the whole value chain,” Stewart said.
“Mining geologists now are, potentially, creating new insights.
“The limitation at the moment really is peoples’ interest in looking further and deeper into the models.
“Some of the insights may be things we already know but some of them may actually uncover new aspects of the orebody systems that previously people hadn’t understood. I think this [geological science] area is actually really interesting and at the moment I have to tell you that this is completely untapped.
“People using orebody learning in operations are really not having the time to report these insights to other geologists within the mining companies. Some are starting to do that through internal communication and presentations, but there is a massive opportunity to look at these charts and the predictors and identify how that may inform geological science more broadly.”
Stewart said while mine geologists were not traditionally responsible for increasing orebody knowledge – typically the domain of resources geologists – “modern mines” were building greater orebody knowledge from operational experience, with the help of technology, and “mine geologists with their understanding of operations and geology are ideally placed to take a central role in a modern mine that continuously learns how to achieve its best performance”.
She said as well as the potential for discovery of new geology insights, AI-backed experiential learning opened many operational improvement doors, from “getting the best blast patterns for specific geology, and plant set point optimisation … to ore blending optimisation”.
“Mining geos can provide inputs for resource modelling through this process: recoveries, specific energy, throughput, and product quality.
“The example of specific energy [is] quite topical at the moment. The variability in the kilowatt hours per tonne required for every block in a block model [is] important for mine schedulers. The true cost of grinding that ore is quite different depending on the hardness of the ore. It may be that certain hard ores are no longer going to be economic and conversely the softer ores will become economic because their specific energy is lower and therefore their costs of processing decrease and when the mining engineers come to do their mine planning those blocks then become economic and are included in the schedule.”
Maptek’s Steve Sullivan said at the conference as the benefits of wider use of data-driven geological modelling became widespread, “we will look back and wonder why it took so long to embed it into our daily processes”.
“The new approach will become ubiquitous in our [resource geology] world, just as mobile phones and the internet have changed the way we work.”