Chapter 1 · Material, Measurements, and Balances

Particles and process questions

A concentrator transforms a population of particles. At the boundary of a unit operation, the objects of interest are the amounts of material in that population with each relevant property. An ore described only as 180 tonnes per hour and 1.1% copper can represent a readily separable feed or a difficult one. If most copper lies in liberated intermediate particles, a flotation rate inferred from a brief test may transfer reasonably to a similar feed. If much of it lies in locked coarse particles and ultrafines, the same bulk grade says little about its recovery after a change in grind. The process model succeeds when its representation preserves the differences that control the engineering decision.

A model starts with a question stated as a change and an outcome. Suppose a plant intends to raise fresh feed from 180 t/h to 205 t/h without increasing grinding power. The relevant outputs may be final size, circulating solids, cyclone water return and flotation feed composition. The model needs a response of breakage to energy or residence time, a classifier response to feed conditions and a balance on the water passing through the circuit. A measured mill product curve alone answers what passed the screen on the survey day. It gives no law for how that curve should change when the solids rate changes. Conversely, a hand-calculated solids balance may be sufficient for estimating a required recycle belt capacity if the split is known from a representative test.

A useful description of a process model has three elements. It specifies the quantities carried by a stream, a mathematical transform acting on those quantities, and the evidence that sets the parameters of the transform. Conservation follows from the first two elements. Predictive merit depends on the third. A mass-conserving model can represent the wrong mechanism, just as an accurate fit can mask a wrong balance if the observations allow compensating errors.

The stream as a population

Let M[c,i] be the dry mass flow rate of component c in size class i. Choose indices consistently for the entire chapter or calculation. Every component is defined on the same dry solids basis, and the complete component set spans the solids. For a plant with measured elemental copper and iron assays, those elements may be useful conserved tracers. Their masses belong to minerals and other phases. They cannot generally be added to a set of full mineral mass fractions as extra exclusive components. One may instead use an exclusive set such as chalcopyrite, pyrite and residual gangue, then calculate Cu and Fe assays from mineral compositions where those compositions are known. A tracer model with copper and balance-of-solids can be useful without claiming mineral speciation.

S = Σᵢ Σ꜀ M[c,i], m[i] = Σ꜀ M[c,i], H[c] = Σᵢ M[c,i].

The size marginal m[i] describes size regardless of composition. The component marginal H[c] describes the bulk component rate regardless of size. The within-size grade is g[c|i] = M[c,i] / m[i] where m[i] is positive. Knowing both marginals leaves the joint distribution underdetermined. A 2 × 2 matrix with fixed row and column sums still has one free entry. With C components and I sizes, the corresponding number of degrees of freedom is (C − 1)(I − 1), before considering additional physical or assay constraints. This dimensional fact sets a limit on what can be inferred from simple plant samples.

Mass units make the balance intuitive. For numerical storage, a program can hold the dry solids rate, the vector of retained fractions and a matrix of component fractions within each sieve class. Multiplication recovers the constituent mass rates. The safest form for mixing and audits is the mass matrix itself. An empty class has zero mass for every constituent, so its conditional composition is undefined for reporting and irrelevant to its immediate mass balance. Retaining a default grade on such a class must never create mass when the class becomes active in a later mixing operation.

Water ordinarily needs its own flow variable. A hydrocyclone can route fine solids with water while concentrating coarse solids into underflow, so treating a single slurry fraction as the split for both phases loses the selective behavior. Additional liquid species may require a separate dissolved component balance in applications where their concentrations matter. Soluble losses should then be represented in the physical phase where they occur.

Three linear operations and a nonlinear plant

Mixing adds corresponding stream masses. A size splitter multiplies each size class by a fraction for the chosen destination. A size reduction operation gives each parent class a distribution over its descendants. These calculations are linear in an incoming mass vector when the relevant mixing rule, partition or breakage operator is fixed. Let f denote a column vector indexed by size. A separator partition t and a breakage matrix B have the forms below.

y[i] = t[i] f[i], r[i] = (1 − t[i]) f[i], p[i] = Σⱼ B[i,j] f[j].

For pure size reduction with no loss, every column of B sums to one and represents descendants of one unit of parent mass. A diagonal can retain some material in the original class. With rows ordered coarse to fine and a pure fragmentation mechanism, the entries in rows coarser than their column vanish. An agglomeration or classification-plus-recombination model may call for a different structure. The fact that a transform preserves the total does not identify its physical mechanism.

A plant can be nonlinear even when each individual accounting step has this appearance. A hydrocyclone partition depends on the incoming volumetric flow, slurry concentration and geometry. The slurry concentration depends on cyclone underflow return and dilution water. A mill rate can depend on the size composition of its holdup or on specific power, which depends in turn on solids throughput. The fitted upstream and downstream transforms must then be recomputed on the circulating load. Fixed conditional operators embedded in a loop give a useful analytical check before adding these dependencies.

Worked example of an invisible grade response

Consider two feed composites, each 100 t/h of dry solids, with equal bulk valuable mineral rate of 12 t/h and equal total size fractions. The coarse class and fine class both carry 50 t/h. In Feed A the valuable mineral favours the coarse class. In Feed B the association is reversed.

PopulationCoarse valuable t/hFine valuable t/hCoarse gangue t/hFine gangue t/h
Feed A934147
Feed B394741

The rough separation sends 80% of coarse material and 20% of fine material to one product. It is a size separation, with no distinction between minerals within each class. Both feeds give 50 t/h to the selected product, because 0.80 × 50 + 0.20 × 50 = 50. For A, the valuable rate is 0.80 × 9 + 0.20 × 3 = 7.8 t/h, or 15.6% grade and 65% valuable recovery. For B the rate is 4.2 t/h, or 8.4% grade and 35% recovery. The two assays differ by 7.2 percentage points despite identical feed grade, feed size curve and product yield. Replacing each within-size grade by the bulk 12% would predict 6 t/h valuable mineral in the 50 t/h product for both feeds. It would remove every discernible effect of the size separation on grade.

This error has a diagnostic signature. Size partition and the sum of the output component rates both close. The model misses an observed redistribution of components between products. A plant survey can reveal the missing state by assaying at least a few size fractions, leaving a sensible composite for the fine classes if sample mass is limited. If the coarse and fine assays drift in changing ore domains, the calibrated separator could have been sound even when its predicted grade drifts.

State resolution and mineral liberation

A modal mineral assay quantifies the mineral composition of the size fraction. A liberation measurement describes how the mineral is distributed among particles with different exposed or volumetric fractions. Two size fractions with equal chalcopyrite content may have very different surface exposures. If flotation recovery or density separation depends on those exposures, a single compositional grade and one rate per constituent lumps together distinct populations. A practical expansion is to use classes such as largely liberated valuable mineral, middlings and low-grade composites with suitable physical attributes, with each class contributing a known amount of valuable mineral to the balance.

Every new grouping expands the observations that would identify the model. A 24-class size vector for two material groups needs fewer measurements than 24 classes crossed with ten mineral types and seven liberation intervals. Conventional plant samples rarely constrain all the latter groupings independently. Choose enough classes to predict the targeted response, then assess whether aggregation across an important screen cut or flotation threshold alters that response. Keeping a coarse model together with a more detailed one is informative when both agree on reconciled balances but disagree about a proposed change.

Exercises with solutions

Exercise 1. Suppose size and grade marginals from the worked example remain known but size-resolved grades were never sampled. Bound the valuable mineral flow to the selected product using only the marginals and physical nonnegativity.

Let x be the coarse valuable flow. Fine valuable flow is 12 − x. The two capacities are 50 t/h, so x lies between zero and 12. Product valuable flow is 0.8x + 0.2(12 − x) = 2.4 + 0.6x, giving 2.4 to 9.6 t/h. Product grade ranges from 4.8% to 19.2%. These limits isolate structural uncertainty. Increasing assay precision for the two known marginals will not narrow the range.

Exercise 2. A 100 t/h stream has 70 t/h water. If a device sends 35% of the solids and 20% of the water to underflow, compute its solids mass fraction. What happens if the slurry is instead sent with a uniform 35% split?

The correct calculation is 35 / (35 + 14) = 71.43% solids by mass. Applying 35% to both phases yields 35 / (35 + 24.5) = 58.82%, equal to the feed solids fraction. A uniform split preserves the mass fraction and cannot model thickening.

Exercise 3. A composite made of 30 kg calcite and 70 kg silicate contains approximately 12 kg Ca attributable to the calcite. Test two complete solid descriptions.

An exclusive mineral description has 30 kg calcite plus 70 kg silicate. A simplified elemental tracer description could carry 12 kg Ca plus 88 kg balance by mass where total calcium is confirmed and no additional traced elements are counted. A representation that carries 30 kg calcite, 70 kg silicate and 12 kg Ca as three additive component masses creates 12 kg solids. The safer layer is a mapping from mineral mass flows to predicted assays.