Adress: Office 1404, Al Thuraya Tower, Dubai Media City, Dubai, UAE
Email: artur@oesv.ae
LinkedIn: oesventures
Case study · Coal · Sakhalin
Balancing a 1,000-unit fleet at Solntsevsky coal mine
An open-pit coal operation on Sakhalin with roughly 1,000 units connected around the clock. Two years of work on one asset, with the effect of balancing separated from everything else that changed at the same time.
- Operation
- Solntsevsky coal mine, Sakhalin
- Scale
- ~1,000 units connected 24/7
- Mining method
- Open-pit, truck and shovel
- Period
- 2025 compared with 2021–2022
- Modules
- AI balancing · Fleet management
- Status
- In production
The problem
At this scale a dispatcher is not short of information — they are short of time. A pit with around 1,000 connected units generates thousands of deviations per shift: a truck waiting at a face, a shovel with nothing to load, a stop nobody coded, a machine drifting below its norm. A human can act on a fraction of them, and the fraction they act on is chosen by whatever is loudest, not by what costs most.
The specific loss we were asked to attack was shovel idle time waiting for trucks on the 22 m³ class — the machines whose stoppage is most expensive per minute.
What we did
We deployed AI-based control of the haulage fleet: assignments of trucks to shovels are recalculated continuously to minimise queues and idle time, equipment is positioned for the shift change so the next shift starts in rhythm, and refuelling is timed so that no shovel is left without haulage while a truck is at the fuel bay.
The same platform classified every stop against the norm for its equipment type, which is what made the loss measurable rather than anecdotal.
How we measured it
Shovel idle time waiting for trucks on 22 m³ shovels fell from 14.6% to 8.9% of calendar time — 5.7 percentage points, comparing 2025 with the 2021–2022 baseline.
Loading cycle time fell from 4.0 to 3.5 minutes on 20+ m³ shovels and from 4.4 to 3.9 minutes on 12 m³ shovels. Three factors worked together: machinist coaching, continuous monitoring of cycle parameters by the multi-agent system, and maintenance that kept machines at speed.
Diesel intensity fell from 93.2 to 88.9 g per t·km and CO₂e intensity from 3.2 to 2.8 kg per m³ of rock mass over the same period.
What we do not claim. Of those 5.7 percentage points, 2.5 points — 44% of the improvement — are attributable to balancing. The remaining 3.2 points came from higher truck availability, a maintenance result rather than a dispatch one.
Over the same period the average haul distance shortened from 3.9 to 3.6 km. That change improves speed and cycle time on its own, and any honest reading of the speed figure has to account for it.
Run the same measurement on your pit
The assessment produces your baseline first. Only then does it make sense to talk about what a system would change.