Confirmation lithium mining heavyweight Pilbara Minerals has joined a growing list of metals producers using Plotlogic’s ore characterisation technology continues the rapid emergence of one of the most impressive newcomers in the global mining technology space.
The A$11.5 billion ASX-listed miner has given Plotlogic 12 months to demonstrate the value of its OreSense rock-mass mapping and AI-driven analytical platform at Pilbara Minerals’ flagship Pilgangoora hard-rock operation in Western Australia’s Pilbara region.
Plotlogic’s aim is to use the technology to improve ore and waste delineation in-pit and in stockpiles, with faster and more accurate inputting of crucial planning and modelling data giving Pilbara Minerals the opportunity to optimise operational processes and, potentially, risk management.
It’s a similar value case to the one Plotlogic founder and CEO Andrew Job has successfully pitched to majors BHP and Vale, as well as diversified miner South32 and an expanding list of other large and smaller miners. BHP, through its venture capital unit, is also a Plotlogic investor.
The company raised US$18 million of equity funding last year that brought in former Google CEO Eric Schmidt’s Innovation Endeavors and other US venture capital funds.
Job left a high-paying mine management job a few years ago to go back to university, to add to his mining engineering and finance degrees. He says he had to address a “technical deficit” if he was going to pursue his mission to try to change an industry he believed was central to a new world industrial revolution, but one that needs to address its own efficiency and environmental issues.
So he got a mechatronics, robotics and automation engineering doctorate, with his PhD focus being precision resource extraction using advanced sensors and artificial intelligence.
While still at the University of Queensland the launch timing for Plotlogic became compelling and Job joined a rare group of CEOs who can brag about putting lines of code into the core of their company’s tech platform.
Not that he’s doing too much of that in an enterprise that now has nearly 100 staff including some of the best and brightest software engineers in Brisbane.
“I hadn’t done any computer programming since I was doing university the first time around and apparently Fortran was no longer a thing,” Job said of starting his PhD.
“Fifteen years is an extraordinarily long time, [as] anybody familiar with the software and AI space would know.
“But for me just taking this snapshot of software engineering and computer programming from one point in time to then literally 15 years later and seeing how much it had moved and how much more sophistication was coming to the industry … That was a massively steep learning curve for me.
“I’m glad now I have a team of software engineers who routinely remind me to not write any more lines of code, but I’m proud of the fact that at least the foundational piece of work I was actually able to cobble together myself.”
The anecdote itself belies the sophistication of Plotlogic’s product offering and the speed at which it has gained traction in the market.
The company’s integration of LiDAR positioning and hyperspectral rock-property scanners, site edge algorithms, and then cloud-based AI analytics coupled with web data reporting and presentation interfaces, is unique.
“We scan with our sensors, we run our AI – not every mine is connected to high-speed internet connection, so we run a whole bunch of edge algorithms, build our AI in the cloud and then deploy it on the edge so we can run these algorithms – essentially in real time, and provide that information to the mining client,” Job says on a recent Forward podcast.
“One of the reasons this tech stack we’ve built is so important is because it is quite ubiquitous. It can work across any commodity type and can be deployed with any mining method. We built the tech stack from the ground up with that in mind.
“We’ve been building our [AI] models out over time and getting to a point of higher commercial utility.
“We work on the basis of building generalisable models that can work, essentially, straight out of the box on any mine site they go to. We started in iron ore. We’ve done additional models into the copper space and manganese. And we’re just building out our nickel model as we speak.
“But essentially we had this idea of building a generalisable model that works straight away.
“So it doesn’t matter which mine site you go to anywhere in the world, we can give you results immediately. We then do some localisation … where we might tune the model to specific conditions.
“Having the additional training data come in is valuable for that local site, but also it feeds back into our generalisable model as well.
“As we bring in more sites the models become more and more robust.
“We’re seeing that play out into every commodity that we go into.”
Job says operational improvements have been apparent at a number of sites at which Plotlogic has worked so far.
“We’re still building the instrumentation that actually measures those uplifts,” he says.
“The measurement of uplift in a complex system is, as you can imagine, non-trivial, because there are multiple inputs and outputs.
“But essentially we target a 20% uplift in mining performance, which is quite substantial for those mining operations.
“And the way that we do that is by reducing the amount of waste material that goes through the mill and also making sure that the blend of actual ore material that goes into the mill is the perfect blend to be able to give the best performance.
“They’re a couple of the big levers.
“There’s also a reduction in the unit cost from having to haul less material than you would otherwise do and making sure you’re not double-handling material, and all those things.
“So that’s at the macro level.
“We also get benefits at the unit level where essentially we take a traditional process which is quite expensive and quite time consuming and we automate that so there are fewer people physically in the field, which helps drive higher degrees of remote operation and automation, which results in less labour on site, less fuel consumption, etc.”
Job says conventional orebody sampling is typically slow and expensive – with millions of dollars a month spent on assaying that can take 7-10 days to turn around – which impacts mobile fleet deployment and certainly overall planning and execution efficiency.
Sampling methodology was also typically based on sparse sampling of a large rock mass.
“0.05% of all the rocks on a mine site are actually sampled in traditional approach,” Job says.
“We end up sampling 26-27% via various technologies. So about one in four rocks.
“And we can do it non-invasively in real time, so it gives you this unprecedented accurate knowledge of the actual rock mass that you’re extracting.
“Scanning in real time gives the geo an opportunity to run more scenarios to optimise their extraction sequencing, and it means they’ve always got a plan so the machines [operators] know exactly where they’re digging next so there’s no idle time.”
Job says the veracity of data fed into the information loop guiding physical operational activity is obviously critical.
“It’s very important for us to have commercial grade, very reliable, very repeatable and auditable AI results,” he says.
“One of the things that we do is take a small percentage of samples – about 10% – and do our own laboratory assessment on those as part of our audits, checks and balances, and we use that for two things. One to provide the client with an assurance that there’s a second independent method – checking the checking of some of these results – that they maintain their reliability and usability. And then the second thing is it enables us to increase our … training data, which of course then enhances our models even further.
“It’s a function of ordinance and governance processes that we’ve built to give our system that commercial grade of reliability.”