The tangent that sizes a plant

The closed column is a complete kinematic-wave problem: a flux curve that bends both ways, a falling interface set by a chord, a rising sediment shock set by a tangent, and a graded layer between them that never quite finishes settling. It ends by naming the calculation it did not do, which is what happens when the cylinder stops being closed.

A continuous thickener is a tank a hundred feet across, fed with dilute suspension somewhere near the middle of its depth, drawing clarified liquid off the rim and thickened slurry from a cone in the floor. Every mineral concentrator and every water-treatment works has several. They have been sized from a cylinder test since the 1910s, and the sizing is a construction on the same curve.

Two ways for a particle to go down #

In a closed cylinder a particle descends for one reason: it settles through the liquid, at a speed the local concentration decides. In a thickener there is a second, and it is usually the larger.

Slurry is being withdrawn from the floor, so the whole mixture — liquid and solids together — is being drawn downwards at a bulk velocity $u$, which is the underflow volumetric rate divided by the tank’s area. Below the feed, therefore, the solids flux is

$$F(\phi) = f(\phi) + u\,\phi,$$

the batch flux plus the bulk transport, and it is the sum rather than either term that has to carry the feed. Both terms are functions of concentration alone, which is what makes the whole problem a kinematic wave problem rather than a dynamical one.

That sum has a minimum in the concentration, and the minimum is the whole problem. The batch flux $f$ rises, peaks and falls back to zero at packing; adding a straight line $u\phi$ to it leaves a curve that still dips. Every level in the thickening zone has to pass the feed’s worth of solids, so the feed cannot exceed the smallest value $F$ takes anywhere between the feed concentration and the underflow. Ask for more and there is no steady state at all.

The second term is usually the larger one and it is worth seeing by how much. At the operating point computed below, the bulk draw contributes $u\phi^* = 0.078$ of flux against the settling’s $0.028$ — nearly three times as much. A thickener is mostly a pump with a settling correction, and the settling is what decides where the correction bites rather than how much of the work it does. That is the opposite of the impression a closed cylinder gives, where settling is the only mechanism there is.

It also explains a feature of thickener behaviour that is otherwise puzzling. Closing the underflow valve does not thicken the underflow indefinitely; it reduces $u$, which reduces the second term everywhere, which lowers the limiting flux — and the tank begins to fill. The control that looks like it should make the product thicker is the one that makes the plant fail, and the construction says so before any of it is built.

The line that must touch rather than cut #

Writing the condition out gives the construction. At the minimum $f'(\phi^*) = -u$, and the value there is

$$F_L = f(\phi^*) - f'(\phi^*)\,\phi^*,$$

which is the intercept on the flux axis of the tangent to $f$ at $\phi^*$. The underflow concentration follows as $\phi_u = F_L/u$, which is where that same tangent meets the concentration axis.

The construction is the flux curve’s tangent rather than its chord, and the difference between the two is exactly what a chord cannot settle.

So the whole design is one straight line: take the tangent to the batch flux curve; where it cuts the flux axis is the plant’s capacity, and where it cuts the concentration axis is the underflow it delivers. Nothing else is needed, and nothing about the tank’s depth, its rake, or the height of the feed point enters.

There is one restriction and it is geometric. A line can only touch a curve from below where the curve is convex, so the tangent point must lie past the flux curve’s inflection — at $2/(n+1)$, which for Richardson and Zaki’s exponent of 4.65 is a concentration of 0.354. On the concave branch below that, a line through any point on the axis cuts the curve instead of touching it, and there is no construction to make.

It is worth pausing on what the tangent point is, because it is not a place in the tank in the obvious sense. It is the concentration at which the thickener is working hardest — the level where the total downward flux is least, and therefore the level that decides how much the whole column can pass. Every other level has capacity to spare. A thickener is limited at one concentration, as a road is limited at one junction, and raising the flux anywhere else in the tank changes nothing.

Operators call it the critical layer, and in a working thickener it is visible: a level in the tank at which the concentration gradient is steep and above and below which it is not. The construction predicts which concentration that is from a cylinder test, and the prediction is testable in a plant with a sampling lance, which is one of the few things in thickener practice that can be checked directly.

And the construction has no time in it at all. Nothing above refers to how long a particle spends in the tank, how deep the bed is, or how fast the rakes turn. That is the kinematic assumption doing its work: a flux that depends only on the local concentration makes the steady state a question about one curve, and the answer is a straight line laid against it.

Two recipes, thirty years apart #

Thickeners were being sized before Yoshioka drew that line, by a rule that looks nothing like it.

Coe and Clevenger, in 1916, argued as follows. To thicken from $\phi$ to $\phi_u$, each unit of solids has to shed a volume of liquid $1/\phi - 1/\phi_u$, and that liquid has to travel upwards through the suspension at the settling velocity $v(\phi)$. The area needed is therefore $(1/\phi - 1/\phi_u)/v(\phi)$, evaluated at whichever concentration demands the most — and the concentrations to try are read off a series of batch settling tests.

That is a recipe about liquid moving up, and Yoshioka’s is a line drawn on a curve of solids moving down. They are the same calculation. Maximising Coe and Clevenger’s expression gives $f'(\phi) = f(\phi)/(\phi - \phi_u)$, which is the tangency condition rearranged, and the value at the maximum is the reciprocal of the limiting flux.

There is one difference and it matters. Coe and Clevenger’s expression diverges as the concentration goes to zero, so its global maximum over a dilute feed is at the feed itself rather than at the interior stationary point. That is not a failure of the rule; it is a second requirement — the area needed for the slowest particles in a dilute feed to reach the bed at all, which is clarification rather than thickening. A thickener has to satisfy both, and which one binds depends on how thin the feed is.

What is being economised, and by how much #

A thickener is a large and simple object, and it is worth knowing what the construction is buying.

A mineral concentrator handling ten thousand tonnes a day of ore at a few per cent solids has to recover the water, and the area the tangent construction calls for is the area of the tank. Get the limiting flux wrong by twenty per cent and the tank is twenty per cent too small or too large — the same leverage a bed’s own weight has on a sedimentation calculation; at a hundred feet across, that is a substantial fraction of the cost of a plant, and it is decided by a line drawn on a curve inferred from a cylinder of slurry left to stand.

The alternative to the construction is not a better theory but a bigger test. Before Coe and Clevenger, thickener areas were chosen by scaling up from a pilot unit, which means building a small thickener and running it — an experiment costing weeks and a slurry pipeline, giving one point. What the construction replaced was not ignorance but expense, and the reason it survived a century of being partly wrong is that even a partly wrong sizing from a cylinder test is a better bargain than a pilot plant.

That is the general shape of what a similarity or a construction is worth, and it recurs in quite different subjects: a relation that turns one cheap measurement into a family of answers is valuable in proportion to how expensive the alternative measurement is, rather than in proportion to how elegant it is.

What a thicker underflow costs #

Once the construction is drawn, the design trade is visible as a geometry rather than as a table.

The mechanism is worth stating in words because it is the reason a thickener cannot be improved by running it differently. A thicker underflow means drawing slurry off more slowly, which is a smaller $u$, which is a shallower operating line. A shallower line touches the flux curve further up, where the curve is lower, and its intercept on the flux axis is smaller. Every unit of extra thickness is paid for in capacity, and the exchange rate is the shape of the curve.

There is no setting at which both are good, and there is no arrangement of the tank that changes that, because nothing about the tank is in the argument.

Past the limit there is no steady state #

The failure mode is not a degradation and it is worth knowing which.

Feed a thickener ten per cent more than its limit and it does not thicken ten per cent less well, which is the same kind of hard ceiling as a throat that stops listening downstream. The solids that cannot pass accumulate, the bed rises, and it rises at a constant speed — because the surplus flux is constant and the bed’s concentration is fixed. Ten per cent over gives a bed rising at 0.017 of the particles’ own settling speed; sixty per cent over gives 0.105.

The linearity is the useful part. Because the rate is exactly the surplus divided by the underflow concentration, a measurement of how fast the bed level is climbing is a direct measurement of how far past the limit the plant is being run — with no model in the conversion beyond a mass balance. An operator watching the bed rise ten centimetres an hour on a tank whose underflow is thirty-five per cent solids knows the overload to the accuracy of the bed-level instrument.

The bed reaches the feed point after a time that is simply the distance divided by that speed, and then the thickener stops working entirely: the feed enters a bed rather than a clear zone, and the overflow that should be clarified liquid carries solids over the rim. A thickener does not warn before it fails; it fills, and the operator’s first evidence is the overflow going cloudy hours after the overload began.

The band the theory allows, and the one plants use #

The construction is exact, the two classical rules agree with it, and its central prediction is wrong by a factor of two.

The reason is one Kynch’s own theory records and does not follow up. A tangent construction needs the convex branch of the flux curve, the convex branch begins at 0.354, and every tangent drawn from beyond it meets the concentration axis above 0.583. There is no operating line in Kynch’s theory that delivers a dilute underflow.

What is missing is that a settled bed has strength. Below some concentration the particles touch, the bed carries part of its own weight through particle contacts rather than through the liquid, and the liquid is expressed slowly by consolidation over hours rather than by settling over minutes. That process depends on the history of the bed and on the weight above a given layer, and neither of those is a function of the local concentration — which is precisely what a kinematic theory assumes.

So the honest position is that Yoshioka’s construction sizes the clarification and settling part of a thickener correctly and says nothing usable about the underflow concentration, which is set by a compression zone the theory cannot see.

That is a sharper statement than “the model is approximate”, and it has a practical form. The construction’s capacity prediction is about the settling zone and stands; its underflow prediction is about the compression zone and does not. So a designer uses the tangent construction with the underflow concentration taken from a separate consolidation test rather than from the curve — which is exactly what thickener practice does, and which reads as an arbitrary hybrid until the reason is written down.

The same division explains an old disagreement in the literature. Coe and Clevenger’s rule and Talmage and Fitch’s graphical method give different areas for the same slurry, sometimes by fifty per cent, and both have their defenders. The difference is in how each treats concentrations past the inflection — the region where the theory is not describing settling at all — so the argument is about which way of misapplying a kinematic theory to a compressing bed happens to be closer.

What the picture cannot show #

Nothing here has a depth in it. The construction fixes an area and says nothing about how deep the tank should be, which is exactly the dimension the compression zone needs. A real thickener’s depth is chosen by residence time in the bed, from a consolidation test rather than from this.

The feed enters as a plane source and mixes with nothing. A real feedwell is a jet, and the dilution it causes — feed suspension entrained into the clarified liquid above — changes the concentration the thickening zone actually sees.

The particles are one size, so the mixture has one settling velocity rather than a viscosity made of particles. A distribution of sizes does not have a single flux curve; the fine tail settles far more slowly and is what controls the clarification requirement, which is why the overflow clarity and the underflow density are set by different ends of the same distribution.

And the flux curve is Richardson and Zaki’s. The exponent 4.65 is for small particles at a Reynolds number small enough for Stokes’ law, the cut-off at 0.64 is random close packing, and neither is a measurement of a real slurry. The construction is what it is; the curve it is drawn on should be measured.

Who found it, and when #

Coe and Clevenger published their unit-area rule in 1916, from batch tests and a liquid-displacement argument. Kynch’s kinematic theory is 1952. Yoshioka’s tangent construction is 1957, and Hassett’s and Talmage and Fitch’s closely related constructions are of the same decade. The identity between Yoshioka’s line and Coe and Clevenger’s maximum was noticed rather later, which is an ordinary fate for two statements written in different vocabularies forty years apart.

The surprising connection is with a machine that has no settling in it whatever. A thickener’s capacity is the minimum of a flux curve, and so is a nozzle’s: a converging duct passes the most mass at the section where the mass flux per unit area is greatest, and no amount of extra pressure moves it, because the limit is a property of the curve rather than of the driving. Both are a choked flow. A thickener fed past its limit does not pass more solids for the same reason a nozzle fed at a higher pressure does not pass more gas, and in both cases what happens instead is that the surplus backs up — a bed that climbs, or a pressure that rises upstream.

Still open: whether one test can separate the two zones #

A thickener is sized from a cylinder, and the cylinder measures two things at once.

The construction above uses the whole flux curve, which a single batch test gives only through its interface trajectory — and the interface only ever visits the concentrations on the concave branch, because that is where the chord rule puts it. The convex branch, which is the only part the tangent construction can touch, is inferred from the trajectory rather than observed on it. That inference is Kynch’s and it is exactly the step the compression zone invalidates: at the concentrations where the tangent lives, the bed is consolidating rather than settling, and its rate is not a function of concentration alone.

What would separate them is a batch test instrumented for concentration rather than for the interface — a column with a gamma-ray or conductivity traverse, reading the profile through the bed as it settles rather than the position of its top. The flux curve inferred from the interface and the one measured inside the bed should agree on the concave branch and part company on the convex one, and where they part company is the concentration at which the particles start touching. That single number would tell a designer which of the two theories to use at which level of the tank, and it does not appear to have been measured against a tangent construction directly.