Flotation Process in
Mining Technology Explained

A bubble rising through pulp collides with a mineral particle. Between them sits a thin film of water. For attachment, the film thins, ruptures, and a contact line forms where solid, liquid, and gas meet. The whole sequence takes somewhere between 1 and 100 milliseconds.

The film does not simply pop on its own. Electrostatic repulsion between the bubble surface (negatively charged in most flotation conditions) and the mineral surface resists thinning. The film drains hydrodynamically under the pressing force of the bubble-particle collision, and rupture occurs only if the film thins below a critical thickness before the bubble and particle separate again. The collision contact time is set by the local turbulence; the critical film thickness is set by the surface chemistry. When people talk about "making a mineral floatable," what they mean in physical terms is shortening the critical film thickness and the drainage time so that rupture happens within the collision window.

This is where electrochemistry enters. On a sulfide mineral surface, the electrical double layer is not just a function of pH and dissolved ions. It is a function of the oxidation state of the mineral itself, the adsorption products of whatever reagents have reached the surface, and the galvanic environment the particle experienced during grinding. Two chalcopyrite particles ground in different mills can carry different surface charges, different oxide layer thicknesses, and different collector adsorption densities, and respond to the same bubble in completely different ways.

The first-order kinetic model used in circuit design, R = 1 − e^(−kt), assumes a single rate constant for the whole feed. This holds for narrow size fractions of clean mineral in a lab cell. For a plant feed containing particles ranging from 5 to 150 microns, variably liberated, with three or four sulfide minerals present at different galvanic potentials, the single k is a fiction. The rate constant is a distribution. The fast-floating particles inflate the apparent k in a lab test; the slow-floating fraction, which is exactly what the scavenger circuit has to deal with, barely registers in the fitted value. This is why scavenger circuits routinely underperform their design recovery: they were sized using a k that describes the easy material, not the material they actually treat.

Xanthate on a sulfide mineral surface is not a case of a molecule sticking to a solid. The mineral surface is an electrode. At anodic patches, xanthate ions give up an electron and dimerize into dixanthogen, an oily hydrophobic film. At cathodic patches, dissolved oxygen grabs that electron and is reduced to hydroxyl ions. The mineral floats because of the dixanthogen layer, and the dixanthogen layer forms only if the surface potential is in the right range. Too low (too reducing), and xanthate stays ionic and washes away. Too high (too oxidizing), and the mineral surface itself oxidizes to a metal hydroxide that buries the dixanthogen under a hydrophilic crust.

Grinding environment controls the starting Eh of every particle that enters the flotation circuit. Steel balls are anodic relative to most sulfide minerals. During grinding, the steel corrodes, dumping ferrous and ferric ions into solution and releasing iron hydroxide colloids. These colloids deposit on every surface they encounter. Meanwhile, the galvanic coupling between the steel and the sulfide minerals shifts the mineral surface potentials. In a mixed sulfide ore, pyrite (the most electrochemically noble common sulfide) gets cathodically protected by the steel and by less noble sulfides like chalcopyrite and sphalerite. The pyrite surface stays metallic and clean. It picks up collector. It floats. The metallurgist sees high pyrite in the copper concentrate, adds more depressant, and fights a losing battle because the problem was created in the mill, not in the flotation cell.

Iron hydroxide slime coatings from steel media corrosion deserve specific discussion because they cause so much misdiagnosed trouble. The colloids are sub-micron. They deposit as a continuous film on mineral surfaces, gangue and valuable alike. The film is hydrophilic. Collector molecules arriving at the surface land on top of the iron hydroxide layer and cannot access the mineral beneath. The resulting surface has collector on it and is still hydrophilic, which baffles anyone reasoning from the assumption that more collector equals more hydrophobicity.

Replacing steel grinding media with inert alternatives (ceramic, high-chrome) eliminates the problem at source. The cost difference per tonne of media is significant. On a life-of-mine net present value basis, the cost difference shrinks or reverses if the recovery gain is even one or two percentage points, because the revenue from recovered metal dwarfs the media cost. The calculation is straightforward. The obstacle is that media purchase decisions sit with the maintenance or procurement department and flotation recovery sits with the metallurgy department, and in many organizations those two departments do not share an optimization framework.

A partially liberated particle has a composite surface. The contact angle on the exposed mineral patch is high; the contact angle on the exposed gangue patch is near zero. The effective contact angle is a weighted average (Cassie equation), and flotation response drops steeply below about 70% mineral exposure. Grind finer, liberate more, float better. Grind too fine, produce slimes that coat other particles, float worse. The optimum is somewhere in between and it shifts as the ore changes.

In most circuits the rougher concentrate goes to a regrind mill (usually a stirred mill) and then to cleaners. The stirred mill does two things. It breaks composite particles into smaller, better-liberated fragments. It also abrades the particle surfaces, stripping off whatever was on them: oxidation products, iron hydroxide coatings, and collector. The surface that exits the regrind is not the same surface that entered. In cases where the rougher particles carried heavy coatings, the stripping is beneficial and cleaner performance improves. In cases where the rougher had produced a good collector layer on clean mineral surfaces, the stripping removes collector that the cleaner was depending on, and cleaner recovery tanks.

A size analysis of the regrind product tells you whether the mill did its comminution job. It tells you nothing about what happened to the surface. A plant that installs a regrind mill and monitors only P80 will never diagnose collector stripping. Adding 5-10 g/t of collector to the regrind discharge fixes the problem when it exists. Finding out whether it exists requires either a surface-sensitive measurement (EDTA extraction, ToF-SIMS, contact angle on the regrind product) or a structured plant trial where collector addition after regrind is tested against no addition and the cleaner performance is compared.

This section is long because the froth phase is where most of the unexploited optimization in operating plants sits.

The froth is a foam. Bubbles packed together, separated by thin liquid films (lamellae), draining under gravity. Water flows down through the lamellae and the Plateau borders where three lamellae meet. Entrained gangue particles, which were not attached to any bubble and were simply suspended in the water that got swept into the froth, drain downward with the water. Hydrophobic particles attached to bubble surfaces stay. As the froth ages (meaning as it sits in the cell between formation and overflow), it gets drier, its bubbles get bigger through coalescence, and some attached particles detach as the bubble films they were riding thin and merge.

The froth recovery factor is the fraction of particles that entered the froth attached to bubbles that make it to the launder. It can be anywhere from 15% to 95%. In a deep, slow-moving, unstable froth, most of what enters at the bottom never reaches the top. In a shallow, fast-moving, stable froth, almost everything does. The cleaner circuit froth is deliberately deep and slow because the goal is drainage: let the entrained gangue fall back, let the weakly attached gangue detach and fall back, and overflow only the well-attached valuable mineral. The rougher froth is shallower and faster because the goal is recovery.

Frother chemistry controls bubble size in the pulp and froth persistence above it. MIBC produces small bubbles and a relatively brittle froth. Long-chain polyglycol ethers produce small bubbles and a much more persistent, wetter froth. The CPC (critical coalescence concentration) of a frother describes the minimum concentration needed to prevent bubble coalescence; below CPC, bubbles merge and get big, reducing collection efficiency. Above CPC, bubble size is roughly constant but froth properties keep changing with increasing frother dose. Most plants operate above CPC. The action of additional frother is mostly on the froth, not on bubble size.

There is evidence from UBC and McGill that frother molecules co-adsorb with collector on mineral surfaces and affect the contact angle. This means frother dose and collector dose are coupled variables, not independent. The standard approach in reagent optimization is to hold one constant and vary the other, which misses the interaction. The interaction is most pronounced with the longer-chain polyglycol frothers. In practice, plants using DF-250 or similar polyglycols have different collector demand curves than plants using MIBC, even on the same ore, and the difference is not just about froth stability. The contact angle on the mineral surface itself is different.

Now, froth management at the cell level.

An experienced operator watching a rougher cell sees all of this simultaneously: the color of the froth tells them mineral loading (dark froth is loaded, pale froth is barren or carrying gangue). The texture tells them drainage state (smooth, glassy froth is wet and poorly drained; rough, knobbly froth is drier). Bubble size at the surface tells them frother concentration relative to what the cell needs. Froth mobility (how fast it moves from the center to the launder) tells them air rate versus froth removal rate. A periodic surge of dark loaded froth followed by a retreat of pale barren froth signals that the cell is oscillating, probably a level control issue. Froth that piles up in the center and does not move laterally signals insufficient lip length or a dead zone in the launder. Froth that cascades over the lip in waves signals too much air or a froth that is collapsing and reforming rather than flowing steadily.

Machine vision systems capture some of this. They segment bubble size distributions, estimate color indices, and measure velocity fields. What they do not yet handle well is temporal patterns: the way froth behavior changes over 30-second or 60-second cycles. A human observer watching a cell for two minutes sees the cell "breathe" and identifies the periodicity. A froth camera taking one image per second has the data to detect this but the image-processing pipelines in commercial systems are mostly not set up for time-series pattern recognition.

Pulp level control. The pulp-froth interface position is set by the tailings discharge valve (dart valve, pinch valve, whatever the plant uses) and a level sensor (float, pressure, ultrasonic). The control loop adjusts the valve to maintain a setpoint. When the loop is well tuned, the interface is stable and the froth above it drains predictably. When the loop is poorly tuned, which is common because level control tuning is low on the maintenance priority list and because the process dynamics change with feed rate and air rate, the interface oscillates. Each upward surge pushes a slug of undrained pulp into the froth. That slug overflows before the gangue can drain out. The effect on concentrate grade is periodic and sharp, visible on an on-stream analyzer as a grade oscillation that correlates with level oscillation. On grab samples taken every four hours, it averages out and is invisible.

Roughers for recovery, cleaners for grade, scavengers for the last bits of mineral. Cleaner tails and scavenger concentrate recycle to create circulating load. The architecture is standard. The implementation varies.

Circulating load is the diagnostic number most metallurgists do not track closely enough. When it climbs, the circuit is failing to make a clean split somewhere. Typically the cleaner is rejecting too much material back to the rougher feed, which means either the cleaner froth is too aggressive (dumping attached mineral back) or the rougher is sending too much gangue to the cleaner in the first place. Tracing circulating load back to its root cause is one of the most effective troubleshooting approaches in flotation, and it requires mass-balancing the internal streams, not just the circuit feed and products.

The original flowsheet for most plants was designed on a feasibility-study composite sample, five to eight years before commissioning. The ore has changed. The flowsheet should change with it. Rebalancing cell volume allocation between roughing and cleaning, moving reagent addition points, redirecting a recycle stream from the cleaner tailing to the scavenger feed instead of the rougher feed: these are zero-capital or low-capital changes that can recover performance lost to ore drift. Plants that treat the original flowsheet as fixed lose roughly a point of recovery per decade to this drift. On a large copper concentrator, a single point of recovery is worth $20-30 million per year at current metal prices.

Mechanical cells (impeller-driven) and column cells (sparger-driven with wash water) are the two workhorses. The Jameson Cell, the Reflux Flotation Cell, and several other designs decouple the functions of bubble generation, particle collection, and froth separation. The Reflux Flotation Cell is the most interesting of the recent entries because its inclined channels exploit gravity within the froth zone, which extends the upper particle size limit for flotation and could allow coarser grinding. Coarser grinding means less energy. On a 100,000 tpd operation, grinding is 50-70% of total energy consumption. Even a modest increase in the flotation-effective top size saves megawatts.

Air rate profiling across a rougher bank: the lead cells see fast-floating material and should run at moderate air rate to produce a drainable, high-grade froth. The tail cells see depleted feed with slow floaters and need more air to maximize collision chances. Running all cells at the same air rate is suboptimal. Valve adjustment to create a rising air rate profile across the bank costs nothing and is done at a minority of plants.

Lip length and launder design determine how far froth must travel laterally before it overflows. Froth mobility decreases as the froth ages and drains. A cell with all its launder on the periphery forces center-of-cell froth to travel the maximum distance, and that froth arrives old, dry, and partially collapsed. Internal launders reduce the maximum travel distance. The choice of launder geometry is a choice about froth residence time, and it has direct consequences for the local grade-recovery tradeoff. In retrofit situations, adding internal launders to existing cells has recovered performance that was attributed to ore deterioration.

Depressant addition point matters more than depressant identity in many circuits. Depressant at the mill discharge encounters bare mineral surfaces ahead of collector and can occupy adsorption sites that the collector needs. The same depressant at the cleaner feed encounters surfaces already carrying collector and competes with an established adsorption layer. Depending on which mineral you want to depress and whether it has already picked up collector, the earlier or later addition point can be correct. There is no universal rule. The competition between collector and depressant for surface sites is a kinetic process, path-dependent, and the order of addition changes the outcome.

Copper sulfate activation of sphalerite depends on pH (copper hydroxide precipitates above about pH 9 and is no longer available as Cu²⁺), on sphalerite iron content in the crystal lattice (higher iron makes activation harder and shifts the copper sulfate demand upward), on competing ions in solution (particularly lead and iron), and on conditioning time. Conditioning time is rarely optimized at plant scale because the circuit layout fixes the available conditioning volume and therefore the time at a given throughput. Retrofitting additional conditioning capacity is expensive. Most plants live with whatever conditioning time the original design provides.

Process water in a recirculating plant contains calcium, magnesium, sulfate, thiosulfate, residual reagents, colloidal metal hydroxides, and organic matter. Each species has a specific interaction with mineral surfaces and with reagents.

Calcium activates quartz. Ca²⁺ adsorbs on the quartz surface through electrostatic interaction with the negatively charged silanol groups, partially neutralizing the surface charge and creating patches where collector or frother can co-adsorb. The quartz becomes partially hydrophobic and reports to the concentrate. The metallurgist sees high silica in concentrate, increases CMC dose, and may or may not get relief depending on whether the CMC can compete with the calcium for the quartz surface at the prevailing pH and ionic strength.

The fix is straightforward: measure water chemistry regularly (Ca, Mg, total dissolved solids, residual frother by TOC or surface tension measurement) and re-optimize reagent conditions whenever the water changes significantly. A water assay costs almost nothing relative to the metal value flowing through the circuit. It is not done regularly at most plants. There is no particularly good reason for this except that it has not traditionally been part of the daily routine, and daily routines have enormous inertia.

Temperature swings the whole system. Cold water holds more dissolved oxygen, shifting Eh. Viscosity increases, slowing bubble rise and drainage. Collector kinetics slow. Frother performance shifts because micelle formation and surface tension are temperature-dependent. In arid climates with 10°C diurnal swings in feed water temperature, froth stability cycles daily and gets blamed on ore variability. A thermocouple on the feed line and a frother dosage trim tied to the reading manages it. Simple instrumentation, simple control loop, rarely installed.

The metallurgical balance at most plants is calculated from grab samples of the feed, concentrate, and tailing streams taken every two to four hours. Grab sampling of slurry streams is biased toward fine particles because coarse particles segregate in the pipe or launder. The bias is systematic, not random. The calculated recovery can differ from the circuit's performance by several percentage points, and the direction of the error depends on the grade-by-size distribution, which means it is not even consistently optimistic or pessimistic.

Cross-stream cutters that traverse the full stream cross-section eliminate the bias. They exist at some plants, usually on the final concentrate and tailing but often not on internal streams. The cost of installing them everywhere they are needed is small relative to the value of the information. The reason they are absent is partly capital competition (every small project competes with every other small project) and partly that the people who understand sampling bias are rarely the people who control the capital budget. Gy's sampling theory and Pitard's extensions of it provide the full mathematical framework for sizing and designing samplers. The framework is taught in specialized short courses and is not part of standard metallurgical engineering curricula, which is a curriculum problem.

Ores are getting finer-grained. Below about 10 microns, bubble-particle collision efficiency drops steeply because the particle follows the fluid streamlines around the bubble instead of impacting it. Conventional cells are ineffective below 5 microns. Microbubble generation, nanobubble nucleation on hydrophobic surfaces, oil agglomeration of fines before flotation, and carrier flotation (attaching fines to coarser particles) are all under investigation. Lab results exist for all of these. Scaling any of them to 100,000 tpd is a different problem.

Tailings disposal constraints will reshape circuit design. Dry-stacked or paste-thickened tailings require specific particle size distributions and specific residual reagent levels. The flotation circuit will eventually be designed to produce a tailing with the right physical properties for disposal, not just to maximize recovery. This adds a constraint that current design practice does not incorporate.

Sensor technology is advancing faster than the modeling that would use the data. On-line Eh, pH, particle size, froth imaging, XRF on slurry streams: the data is there or becoming available. The bottleneck is integrating it into control actions. Hybrid models that combine flotation kinetics equations with machine-learning correction terms are the most promising approach because they degrade less catastrophically when the ore moves outside the training data range than pure data-driven models do.

Surface characterization (ToF-SIMS, XPS) applied to plant samples rather than laboratory specimens would transform troubleshooting from inference to direct observation. The technology exists at the right sensitivity. The turnaround time and cost per sample are still too high for routine use. Both are falling.