Froth Vision Analytics

Application

The appearance of the froth on a flotation cell carries a great deal of information about how the cell is performing, yet on most operations this information is read intermittently and subjectively by operators, or inferred only after delayed downstream metallurgical assays become available. By the time a laboratory result confirms that recovery or grade has drifted, the froth condition that caused it may have changed several times over.

Minealytics Froth Vision Analytics places cameras above flotation cells and continuously analyses the froth surface in real time, extracting the key froth properties that experienced metallurgists watch for:

• Bubble size distribution (p80): an indicator of particle attachment and overall froth health, reflecting how well mineralised bubbles are forming and reporting to the concentrate.
• Froth velocity (speed): the rate at which froth travels across the surface toward the overflow lip — a strong proxy for the recovery, or pull, rate of the cell.
• Bubble burst and collapse rate: the frequency with which bubbles rupture at the surface, providing an early-warning indicator of froth instability before it becomes visible in downstream results.

By measuring these properties continuously rather than sampling them occasionally, the system gives the control strategy a direct view of froth condition as it evolves.

Solution

The Minealytics vision layer converts the live camera feed into continuous, actionable process signals that represent the “quality” of flotation at each cell. Instead of relying solely on delayed and infrequent metallurgical assays, the control system can act on froth health directly and immediately.

These signals are made available to the plant process control system (PCS) over standard industrial protocols such as Modbus, so they can be trended, alarmed and consumed by control logic alongside conventional instrumentation. In this way the froth itself becomes a measured variable rather than a qualitative observation.

This sensing capability is the foundation for closed-loop flotation control: every higher-level optimisation strategy depends on first having a reliable, real-time measurement of what the froth is actually doing.

Technology

Froth Vision Analytics uses object-detection and segmentation neural networks from the YOLO family to detect and characterise individual bubbles within each frame. Running on an industrial AI edge controller located at the plant, the models process the camera stream locally and in real time, avoiding any dependence on remote infrastructure for time-critical sensing.

From the detected and segmented bubbles the system derives the froth metrics — bubble size distribution, surface velocity and burst rate — and publishes them as continuous signals. Because the analysis runs on rugged edge hardware integrated with the PCS, the vision layer can be deployed cell by cell and scaled across a bank or circuit as more cameras are added.