Key Takeaways
- Geometallurgical reconciliation compares predicted and actual mill performance to recalibrate your models. It has a documented real-world AI implementation at the Tropicana Gold Mine.
- The Tropicana approach recalibrates Work Index and geomet block models in near real time, based on the gap between predicted and actual mill throughput and recovery.
- Cancha productizes a similar reconciliation workflow for teams that don’t want to build custom ML pipelines in-house.
- This is one of the more mature AI applications in the geomet space. It is “Yes” rather than “Emerging” because there is a published, working case study, not just a vendor pitch.
TL;DR
AI geometallurgical reconciliation feeds actual mill performance back into your prediction model, so the geomet block model recalibrates itself as ore is processed. Tropicana Gold Mine’s published approach does this in near real time, and Cancha offers a productized version. Most sites start with flagged updates that a geologist reviews.
What Is Geometallurgical Reconciliation?
Every geomet model is a prediction made from limited sample data. It says how hard each block is, how fast the mill will process it and how much metal it will recover. Every mill run then generates ground truth that either confirms or contradicts that prediction.
Reconciliation is the discipline of closing that loop. You compare what the model predicted for the ore that was milled with what the plant actually achieved. Then you use the difference to correct the model. However, doing this manually, in quarterly spreadsheet reviews, means running on stale assumptions for months at a time.
How Do I Streamline Geometallurgical Reconciliation With AI?
The published approach at Tropicana Gold Mine automates this loop. In practice, it breaks into three stages.
Capturing mill data at block scale
Start with data granularity. Mill performance data (throughput, recovery, comminution behaviour) must map back to specific ore blocks or geomet domains. That usually means tracking ore from the pit through stockpiles to the mill feed. Without that link, the model cannot learn which predictions were wrong.
Automating geometallurgical reconciliation
As mill data comes in, the system compares it against what the geomet block model predicted for that ore. It then uses the discrepancy to recalibrate Work Index and other geomet parameters going forward. This differs from a one-off audit. It is a continuous feedback loop, closer to how a recommendation system updates on new data than an annual resource-model review. If you don’t want to build that pipeline yourself, Cancha offers similar functionality within its broader geometallurgical reports platform.
Choosing how much to automate
Next, define which comparison you track: recovery, throughput or Work Index. Then decide whether recalibration should be fully automatic or flagged for geologist review. Most operations start with review, and only later trust a fully closed loop.
What to Expect From Geometallurgical Reconciliation
Reconciliation depends on clean plant data. Mill throughput is easy to log, but recovery needs reliable assays of feed, concentrate and tailings. Stockpiles blur the link between a block and a mill shift, so expect some averaging.
The payoff reaches beyond the model. Better Work Index predictions help the grinding circuit plan throughput, and they improve mine schedules. As a result, reconciled models are often the foundation for dynamic block models that update continuously.
FAQs
What is geometallurgical reconciliation?
It is the process of comparing predicted plant performance from the geomet model with actual mill results. The gap is used to recalibrate model parameters such as Work Index and recovery. It keeps the block model aligned with how the ore really behaves.
How does AI improve geometallurgical reconciliation?
AI turns a periodic manual review into a continuous loop. Models compare predictions with incoming mill data and update geomet parameters automatically or flag them for review. As a result, the model stays current instead of drifting for months.
What data do I need to start?
You need mill throughput, recovery and comminution data at a resolution that maps back to ore blocks or domains. You also need the original geomet predictions for that ore. Ore tracking from pit to mill is usually the hardest part.
Should recalibration be fully automatic?
Not at first. Most operations begin by flagging proposed changes for a geologist to approve. Once the team trusts the model’s behaviour, some parameters can update automatically.
Try It With Geocluster
Building or evaluating a reconciliation pipeline like Tropicana’s means comparing several possible ML approaches for the recalibration step. The Geocluster Research Harness is designed to help you research and validate exactly that kind of AI workflow choice for your own operation. Read how the Geocluster Research Harness works to see where it fits.