Mining

The bottleneck in mine planning is compute

Across strategic, long-term and medium-term planning, the same limit keeps showing up: a plan can only carry so much detail before it stops solving in the time available. Hyperbridge builds the compute to lift that limit.

What planners tell us

Model detail is rationed by solve time: planners make careful calls about how much operational detail a model can carry and still solve, still be trusted, still be defended. Expertise outruns the tooling: deep orebody and operational knowledge often cannot reach a viable recommendation while the decision still matters. The plan and the operation drift apart: actuals move away from the plan as faces, trucks and priorities change, and the re-plan cadence cannot catch up before the gap compounds. Where the bottleneck is compute, Hyperbridge works to lift it.

One bottleneck, several horizons

Strategic and long-term planning sets the mine's shape over years: method, production rate, scale, sequencing, and the capital decisions behind them. Medium-term planning covers monthly and quarterly scheduling, twelve to twenty-four months out, carrying financial targets alongside production. Operational and shift planning keeps the live operation and the plan together as faces, fleet and priorities move through the day, and is not a horizon we sell into today. At every horizon, how much detail a model can hold is set by how fast it solves.

What we have measured in mining

A ten-year open-pit strategic plan, run twice: once through Google OR-Tools and once through Cognomine, on the same orebody, the same variable count and the same constraint set. Cognomine returned a schedule in about 25 seconds. OR-Tools was stopped unconverged at twelve hours, which puts the gap at more than 1,000x and makes that figure a lower bound rather than a ceiling.

The plans themselves differed as well. The Cognomine plan carried

73M more discounted NPV, about 16% higher, on
49.5M less equipment capital. It reached that with a peak fleet of two diggers and seventeen trucks, against five and forty-one.

The baseline is open source because open source is reproducible. Full method, hardware detail and footnotes are at https://cognomine.com/benchmark.

A schedule that survives contact with the pit

Fast schedulers usually get their speed by dropping the constraints that make a plan executable, and the planner spends the following week putting them back by hand. Our engine holds those constraints inside the solve. Drill-and-blast lag is modelled, digger separation is enforced, slope constraints are solved in rather than applied afterwards, and within each month diggers are given territories and walked through nearby blocks in a sensible working order. Every plan comes out at monthly cadence, life-of-mine included, so a ten-year schedule is 120 months of sequencing. Pit selection, block scheduling and haulage resolve in a single pass rather than as three sequential steps in three tools.

What we expect at the operational horizon

Shift-level re-planning is the same class of problem as the strategic solve, one horizon down and one cadence up. We expect improvement of a similar order there. We have not measured it, so we are not going to publish a number for it. What we want is an operator or a technology partner willing to put a real re-planning problem in front of the engine and let the result settle the argument.

Cognomine: our mining product

Cognomine covers strategic, long-term and medium-term mine planning. It returns a scheduled plan in minutes, so a study can test the assumptions that actually move the answer rather than the two or three there was time for. Product detail, the full benchmark, pricing and the security and data model live at https://cognomine.com.