Grade & Recovery Soft Sensor

Application

The true measures of flotation performance — concentrate grade and recovery — are confirmed by laboratory assay, but those results are delayed and infrequent. Control strategies therefore lean on proxies such as froth velocity, which correlate with recovery but are not the metallurgical outcome itself. The gap between what can be measured continuously and what actually matters is a persistent limitation in flotation control.

Minealytics closes that gap with a real-time soft sensor for grade and recovery. By learning the relationship between observable conditions and measured metallurgical results, the soft sensor predicts the outcomes that would otherwise have to wait for the laboratory.

Solution

The soft sensor is a neural network trained on historical fused data — froth-vision metrics together with process-control signals — matched against the corresponding laboratory assay results. From these paired records it learns to map the live, continuously available signals onto the metallurgical outcomes they produce, and then predicts concentrate grade and recovery directly in real time.

Once the soft sensor has been validated against ongoing assays, its predictions replace the simple froth-velocity proxy in the supervisor’s objective function. The optimiser then targets true metallurgical performance — grade and recovery — rather than a stand-in for it. This effectively closes the loop between froth appearance and metallurgical result: the control system optimises for the outcome that the operation is actually paid for.

Benefits

• Targets real performance: the supervisor optimises grade and recovery directly instead of a proxy, aligning control with metallurgical and commercial objectives.
• Reduced reliance on lab assays: continuous predictions lessen dependence on delayed, infrequent laboratory results for moment-to-moment control decisions.
• Faster response: because predictions are available in real time, the control system can respond to grade and recovery drift as it develops rather than after it has been confirmed downstream.
• Builds on existing sensing: the soft sensor reuses the froth-vision and process data already gathered for control, requiring no additional plant instrumentation.