At an iron ore operation, an oversized rock or foreign object can cause downtime and expose people to manual intervention. BHP describes a computer vision system that looks for those materials during loading and along conveyors before they reach critical equipment.
Detect before the material reaches the crusher
A concrete industrial problem
The BCA case says oversized rocks and foreign material had contributed to around 1,100 hours of crusher downtime each year and increased reliance on higher risk manual intervention. The system uses real time computer vision and machine learning to identify problematic material before it reaches crushers or port systems.
BHP's published targets include a 20 percent reduction in crusher downtime, a 60 percent reduction in port downtime and approximately 0.8 million tonnes per annum additional throughput at scale. The case describes these as targets and value estimates, not verified realised outcomes.
The safety system surrounds the model
A detection must trigger a controlled, safe action. Teams need to know whether the model pauses a conveyor, alerts an operator or diverts material, and what happens if the camera is dirty, the lighting changes or the alert is wrong.
That means designing sensors, procedures, control logic, maintenance and human authority together. A high detection score in a test set is not enough if an operator cannot respond safely in the real plant.
How to scope a workplace safety pilot
Pick one measurable hazard and map the existing controls. Run the vision system in observation mode first, compare alerts with incident and inspection records, then review with workers who know the operating conditions. Only change a safety control after formal engineering and safety review.
- Track missed hazards and false alerts.
- Involve operators and maintainers in testing.
- Keep existing safety controls until the new one is validated and approved.
Want to explore how this could apply to your organisation?
Explore Private AI ↗