What bubble-size distributions have taught us about recovery.

Flotation is a game played by bubbles. A particle leaves the pulp attached to a bubble, rides the froth phase, and reports to concentrate — or it drains back and is lost. For decades the industry proxied this with recovery and grade assays, which arrive late and tell you only what already happened. The research question has been quieter and more fundamental: how much of the recovery outcome is already visible in the bubble-size distribution at the froth surface?

The literature's answer has shifted. The link is tighter than operators used to assume.

What the research has converged on

Across pilot studies and plant-scale work in the last decade, a few findings keep repeating:

• Bubble-size distribution (BSD) is a leading indicator of recovery, not a lagging one. BSD shifts respond to collector dosage, frother dosage, air rate, and pulp chemistry on timescales shorter than the assay cycle.
• The modal size and the width of the distribution both matter. A tight, small-mean distribution typically supports higher recovery on fine particles; a shift toward larger bubbles often signals coalescence from a chemistry disturbance before grade reports it.
• Surface velocity is a complementary channel. BSD describes the froth's structure; surface velocity describes its mobility. The two together explain more of the recovery variance than either alone.
• The signal is non-stationary. Froth imaging that averages over too long a window washes out exactly the transients that matter for control. Short, overlapping windows beat long, clean ones for closed-loop use.

Where the open literature stops

The academic work has mostly been about demonstrating the correlation: given a dataset, how well does froth imagery predict recovery. The next step — running the measurement continuously, at plant cadence, as a control input — is a different engineering problem. It needs robust segmentation under variable lighting, stable BSD estimates under heavy loading, and a latency budget tight enough for a controller to act.

Where our camera picks up the baton

Our Froth Analyzer is built to run the measurement live. It reads BSD and surface velocity on short overlapping windows, hands the controller a signal with a known lag, and treats the research literature's correlation findings as the starting assumption rather than the destination.

The froth has been trying to tell operators what the assay is about to say. Reading it in real time — not as a research exercise but as a control input — is what turns that signal into recovery that stays in the concentrate instead of walking out in the tails.