CITIC Pacific Mining: An A.I hub in disguise.

CITIC Pacific Mining: An A.I hub in disguise.

It's fair to say that most mine sites would not seem, at first glance, to be A.I or Machine Learning hubs.

When I conjure up an image of a West Australian mine site, I think of a huge, hot dusty hole in the ground filled with burly blokes and big machinery. About as far from the image of savvy young tech AI entrepreneurs that the socials espouse.

CITIC Pacific Mining explodes that vision. Hiding its AI credentials in plain sight.

The Force Behind CPM's AI Initiatives

Mark O'Brien is the pragmatic innovator helping to guide some of CITIC Pacific Mining's AI success. He is the General Manager for Digital Technology & Innovation (the CIO) at CPM, Australia's largest magnetite mining and processing operation.

In a recent interview with SAUG, Mark talked openly and candidly about his experiences and shared insights into his pragmatic approach to solving real problems with AI.

Who is CITIC Pacific Mining?

CITIC Pacific Mining operates a vast magnetite mine at Cape Preston in WA. Mark explains that magnetite is not the usual "rusty dirt" type iron ore. Magnetite, instead, comes out of the ground at a much lower quality that hematite, but after extensive processing it becomes a sought-after product used in special steels and prized as a "green steel" product because of the lower emissions during smelting.

Because it requires much more processing, it involves many large, fixed assets, and presents far more interesting challenges and large-data opportunities.

Mark describes the operation as a Mega Project and it's easy to see why. The site is truly massive. So massive that it boasts its own power station and desalination plant.

It contains lots of things performing an array of complex tasks and generating truly HUGE amounts of data.

Data is the new oil.

The mathematician Clive Humby declared that "The world's most valuable resource is no longer oil, but data." In the realm of AI, that is exponentially so.

With huge amounts of data available, the opportunities for CITIC Pacific Mining to leverage AI lay just beneath the surface.

Unearthing Opportunities for AI.

One such opportunity was unearthed a few years back in their Mine-to-Mill operations.

The Mine-to-Mill process is complex and involves all the planning and production elements required to proceed from a plan built from a geological model, drilling, blasting, crushing and processing, all the way to the refined product exiting the mills.

When seeking to optimise and gain efficiencies in these processes one faces a Sudoku problem.

You cannot solve the puzzle by looking at each component square separately. You must see the whole and address the challenge at the system level.

The data for this whole process at CITIC Pacific Mining was held in several key data sets. Data from different sources, in varying formats and in separate locations, all connected by time. Lots of data!

Mark and his team held the belief that meaningful insights lay in this data. Insights that had the potential to significantly impact production.

But how to go about it? Clearly spreadsheet wrangling was not up to the task.

The construction and training of large models was the answer. The data was stitched together and an AI model specially trained to look for correlations and offer insights.

With so many variables and interdependencies it had been impossible to answer such a seemingly simple question as:

What makes a good day?

Understanding where a good day starts, what are the events of most significance and where to focus attention has been largely down to what Mark calls, "Machine Whisperers". Those highly experienced operators who instinctively know what's happening but who can never fully explain how.

Now, thanks to the developed AI models, CITIC Pacific Mining was able to identify what matters in the process, when an event means something in the context of the overall process and where to intervene.

These early experiments showed the way, and further work has continued to take these insights and build them into the management of the processing plant.

Continuous Innovation

Mark and his team never seem to stand still and are right now continuing to work on the models to further improve site performance, improve ramp-up times, have an even deeper understanding of what impacts production, and correlating factors.

Mark's Rules & Lessons Learned

  1. Become an experimenter. Be prepared to test, fail, discover, and share.
  2. There is no set-and-forget. To be effective, models need constant revision and rebuilding.
  3. AI at scale is hard. Going from POC to production is a huge step.
  4. Data is crucial. There is so much data and so much more that can be captured with cheap devices. Gather as much as you can. Then some more.
  5. Not all data is equal. The reality is that only when you begin to use the data do you realise that not everything you've been collecting is immediately useful. Another good encouragement to get using the data early so you can improve your collection.

This one example demonstrates just one of the ways in which mining companies are using AI in their businesses, in everything from exploration to safety, optimisation and automation. It would be fair to say that almost all the technology platforms existing in modern mining are embedded with AI tools which exploit the rich data sources available from plant and equipment in a variety of ways.

About Mark O'Brien

With an involvement in mining technology that stretches over 30 years, Mark is also Vice Chair of the Board of Directors for Global Mining Guidelines (GMG), the Inaugural Chair of the GMG AI in Mining Working Group, Chair of the SAP Australia User Group (SAUG) Working Group on AI, as well as a regular contributor to national and international collaborations around mining technology. Alongside his board work, he also provides advisory services on leadership, technology, AI and cybersecurity.

Displaying item 1 of 149