Sampling and Mass Balancing in Mineral Processing: The Complete Guide

Sampling and mass balancing are the analytical foundations on which the entire metallurgical accounting and process control framework of a mineral processing plant rests. When a typical large copper concentrator processes 100,000 tonnes of ore per day, the composite daily sample sent to the assay laboratory weighs approximately one gram. That single gram must represent one part in 100 billion of the day’s feed to within better than 5 percent accuracy. This is not a trivial problem. The science of achieving representative samples from heterogeneous ore streams, quantifying and minimising the errors involved, and then using the resulting data to construct consistent mass balances across the plant circuit is both intellectually demanding and practically consequential. Errors in sampling protocols propagate directly into incorrect metallurgical accounts, poor process control decisions, and ultimately into misallocated capital and operational expenditure. This guide provides a comprehensive treatment of sampling theory, sampling equipment, mass balancing methods, data reconciliation, and process control instrumentation for mineral processing engineers, metallurgists, and students. The primary reference is Wills’ Mineral Processing Technology, 8th Edition (2016), supplemented by current industry practice and peer-reviewed literature.

Why Sampling Matters in Mineral Processing

The purpose of sampling in mineral processing is to obtain a small portion of material that accurately represents the chemical composition, particle size distribution, and physical characteristics of a much larger quantity of material. The data obtained from these samples form the basis for metallurgical accounting, process control, plant testing, and production reporting. Every major decision made in the operation and management of a concentrator relies, directly or indirectly, on sample data.

Despite its importance, sampling is frequently undervalued in plant design and operation. Processing plants are often designed and built with inadequate provision for proper sampling points, insufficient infrastructure for representative sampling devices, and sampling protocols that introduce systematic bias into the measured data. The consequences of poor sampling practice are insidious: biased assay data lead to incorrect calculation of metal recovery, concentrate grade, and tailings losses, which in turn lead to incorrect process control responses and inaccurate production accounts.

The challenge of representative sampling arises fundamentally from the heterogeneous nature of ore particles. Individual particles differ in mineral content, density, size, and shape. The composition of a sampling stream varies over time and space as ore grade fluctuates, grinding circuit conditions change, and reagent additions vary. A sample can only represent the larger lot from which it was taken if every particle or fluid element in that lot has an equal probability of being included in the sample. This probabilistic principle is the cornerstone of all rigorous sampling theory and practice.

International standards organisations, including ISO, ASTM, Standards Australia, and JIS, have published detailed standards for the sampling of bulk materials including coal, iron ore, precious metal concentrates, and metalliferous concentrates. Compliance with these standards is a regulatory and contractual requirement in many commercial transactions involving bulk mineral commodities. Regulatory frameworks for mineral deposit reporting, such as the JORC Code in Australia and NI 43-101 in Canada, also impose specific requirements on sampling and assaying procedures to protect investors from fraudulent or misleading resource estimates.

The economic value of correct sampling is best appreciated by considering the consequences of systematic bias. A consistent 2 percent overestimation of concentrate grade in a large copper operation translates directly into incorrect smelter payment calculations, inaccurate production targets, and potentially significant over- or under-declaration of mineral inventory. Getting sampling right is not merely a technical nicety; it is a commercial and regulatory obligation.

Gy’s Theory of Sampling: The Fundamental Sampling Error

The theoretical foundation of modern sampling practice in mineral processing was established by the French engineer and statistician Pierre Gy (1924–2015), who spent over 25 years developing the Theory of Sampling from the 1950s onward. Gy’s work identified the sources of sampling error, quantified their magnitudes, and provided practical tools for designing sampling systems that meet specified precision requirements. His contribution to the field is widely regarded as one of the most significant methodological advances in applied mineral processing.

At the heart of Gy’s theory is the concept of the Fundamental Sampling Error (FSE), which is the irreducible minimum error associated with taking a finite-sized sample from a heterogeneous population of particles. The FSE exists because individual particles differ in their composition, and a sample of finite mass cannot perfectly represent the average composition of the entire lot. Crucially, the FSE cannot be reduced to zero, but it can be minimised by careful design of the sampling protocol.

Gy expressed the variance of the fundamental sampling error as:

σ²_f = C · d³ / m_s

where σ²_f is the variance of the fundamental sampling error, d is the nominal particle top size (typically the 95% passing size, in centimetres), m_s is the mass of the sample taken (in grams), and C is a mineralogical constant that characterises the specific ore assemblage being sampled. This simplified form applies when the mass of the lot being sampled is much greater than the sample mass, which is almost always the case in plant practice.

The mineralogical constant C is the product of four dimensionless factors: f (a shape factor, typically 0.5 for most ores and 0.2 for gold ores), g (a particle size distribution factor ranging from 0.25 for wide distributions to 1.0 for uniform size fractions), l (a liberation factor that accounts for the degree of mineral liberation), and m (a mineralogical composition factor that depends on the density and grade of the valuable mineral relative to the gangue).

The liberation factor l is of particular significance. For an ore where the valuable mineral particles are poorly liberated from the gangue at the sampling size, l is large and C is large, requiring a much greater sample mass to achieve a given precision. As the sample is ground to finer sizes and liberation improves, l decreases substantially. This is why sample preparation protocols for assaying invariably include a size reduction stage: grinding the sample finer before splitting and assaying dramatically reduces the required sample mass and the associated sampling error.

A practical worked example illustrates the power of Gy’s formula. For a lead ore containing 5 percent Pb, sampled on crusher product with a top size of 2.5 cm, requiring 95 percent confidence limits of plus or minus 0.1 percent Pb, the calculation yields a required sample mass of approximately 175 to 350 kg. This represents a striking practical challenge: 350 kilograms of crusher product must be collected, sub-sampled with further size reduction, and assayed to meet the stated precision requirement. The calculation immediately highlights the need for multi-stage sampling systems and appropriate size reduction between sampling stages.

Gy’s formula has been the subject of continuing debate and refinement in the literature. Some researchers have argued that the exponent in the liberation factor relationship should not be universally 0.5 but should vary between 0 and 3 depending on the mineralogical texture of the ore. Gold ores, with their extreme nugget effect, are particularly challenging for standard Gy theory. Nevertheless, the framework remains an indispensable practical tool for sampling system design and continues to evolve through the work of researchers such as François-Bongarçon, Pitard, and Minnitt.

The Seven Sampling Errors: Sources and Minimization

While the Fundamental Sampling Error represents the irreducible minimum error associated with the heterogeneity of individual particles, a real sampling system generates additional errors from a variety of other sources. Pierre Gy identified up to eight distinct sampling errors that collectively constitute the total sampling error. For practical mineral processing applications, these can be grouped into four main categories that capture all significant sources of variance.

The first and most fundamental is the Fundamental Sampling Error (FSE), also called the composition error or composition heterogeneity error. This arises from the variation in composition between individual particles, and is the irreducible component quantified by Gy’s formula. It is determined by particle top size, mineralogy, degree of liberation, and sample mass. It cannot be eliminated, but it can be minimised by reducing particle size before sampling and by taking larger sample increments.

The second major error source is the Grouping and Segregation Error (GSE), also called the distribution error or distribution heterogeneity error. This arises from the fact that material within the sampling unit is not uniformly mixed but is heterogeneous on a larger scale. A stockpile, for example, is segregated by particle size due to gravitational settling during filling, with coarse material concentrating at the outer edges and fine material accumulating near the stacking point. A mill feed stream varies in composition over time as ore grade from different parts of the mine changes. The grouping and segregation error can be reduced by increasing the number of sampling increments taken from different locations or at different times within the sampling unit, but it cannot be reduced by increasing the mass of individual increments.

The total variance arising from FSE and GSE can be expressed following Visman’s formulation as the sum of composition variance and distribution variance components. The composition variance is reduced by increasing both the mass and number of increments; the distribution variance is reduced only by increasing the number of increments, not their mass. This distinction has important practical implications: where distribution heterogeneity is the dominant error source, the emphasis should be on collecting more increments, not heavier ones.

The third error category is the Preparation Error, which includes all errors introduced during the handling, size reduction, splitting, and sub-sampling stages between the primary sampling device and the final assay laboratory. Sources include contamination between samples, loss of fines during transfer, selective breakage during grinding, and non-probabilistic splitting using riffles or coning techniques. Preparation errors are reduced by careful laboratory procedure, use of properly calibrated equipment, and the application of probabilistic splitting devices such as rotary table riffles that give every particle an equal chance of entering either half of the split.

The fourth category is the Analysis Error, introduced during the final chemical or physical measurement of the sub-sample. This includes instrumental calibration errors, matrix effects in X-ray fluorescence analysis, and random analytical precision limits. Analysis errors are typically the smallest component of total sampling variance, often constituting only 1 to 4 percent of the total. They can be estimated by performing replicate analyses on the same sub-sample and calculating the variance of the results.

In addition to these statistical error categories, Gy’s framework also identifies several systematic errors related to sample extraction methodology. The Delimitation Error occurs when the spatial or temporal boundary of the sample increment is not correctly defined, for example, when a sampler cuts only part of the stream width instead of the full cross-section. The Extraction Error arises when particles within the defined boundary are not all collected with equal probability, as occurs when coarse lumps are deflected by the cutter edge or fine particles are left behind on the belt surface. These systematic errors are the most insidious because they introduce bias rather than random variance, and bias cannot be detected or corrected by statistical analysis alone; it requires comparison with reference measurements obtained by independently verified probabilistic methods.

Practical rules of thumb derived from Gy’s work and widely applied in industry include: the sample mass should be at least 1000 times the mass of the largest particle; the sampler opening slot should be at right angles to the material flow and at least three times the width of the largest particle; the cutter traversal speed should not exceed 0.6 metres per second for material with particle sizes above 1 mm; and cutter blade edges should be knife-edged to prevent deflection of impacting particles toward either side.

Manual vs. Automatic Sampling Systems

Mineral processing plants require two broad types of sampling systems: manual systems for plant surveys, bias testing, and reference sampling, and automatic systems for continuous metallurgical accounting, on-stream analysis, and process control. Each type has distinct strengths and limitations, and the choice between them depends on the purpose of the sampling, the required precision, the sampling frequency, and the practical constraints of the installation.

Manual sampling, when correctly performed, can provide the most reliable reference measurements in the plant. A stop-belt manual reference sample, in which the conveyor belt is stopped and the entire cross-section of material on a defined length of belt is collected, is the closest achievable approximation to probabilistic sampling on a conveyor system. Similarly, manual cross-stream cuts made with a hand-held sampler during shutdown for calibration purposes can provide a reliable reference for assessing the bias of installed automatic samplers. The limitations of manual sampling are its labour intensity, the practical impossibility of achieving the sampling frequency required for continuous metallurgical accounting, and the safety risks associated with working around moving conveyor equipment.

Traditional manual methods such as grab sampling from a conveyor belt, riffle splitting, and coning and quartering are inherently non-probabilistic and should not be used for metallurgical accounting purposes. Grab sampling introduces severe bias because the sampler preferentially picks up accessible surface material rather than a true cross-section of the stream. Riffle splitting can introduce bias if the outside slots feed to the same container, if the splitter is not exactly as wide as the combined slot width, or if feeding is slow and uneven. These methods may be acceptable for low-stakes quality monitoring but should never be used for commercial accounting or metallurgical balance calculations.

Automatic sampling systems provide the high sampling frequency and consistency needed for metallurgical accounting and process control. Well-designed automatic systems, based on probabilistic sampling principles, can collect representative samples at intervals of minutes or hours, accumulate them into composite samples representative of production periods of any desired length, and deliver sub-samples to the assay laboratory or on-stream analyser with minimal human involvement.

A key operational requirement for automatic sampling systems is regular bias testing against manual reference samples. No mechanical system is immune to wear, blockage, or miscalibration, and a sampler that was correctly installed and bias-free at commissioning may develop systematic errors over time due to wear on the cutter blades, changes in the stream characteristics, or gradual changes in the geometry of the system. Periodic comparison of the automatic sampler output against carefully collected manual reference samples is an essential component of sampling system quality assurance.

The design of the overall sampling installation, from primary sampler through secondary and tertiary sub-sampling stages to the final assay preparation laboratory, must consider the required sample mass at each stage, the splitting ratio between stages, and the material handling logistics. A modern large concentrator may have 20 to 30 automatic sampling systems installed across the circuit, each delivering samples to a central sample preparation facility where drying, crushing, splitting, and dispatch are carried out to ISO laboratory standards.

Slurry Sampling: Cross-Stream and In-Line Samplers

From the grinding stage onward, most mineral processing operations handle material as slurry, and the majority of critical process streams that require sampling for metallurgical accounting are slurry streams. Slurry sampling presents unique challenges compared to dry solid sampling because of the need to collect a representative cross-section of a flowing mixture of particles and water, in which particle size segregation and density differences between species create additional sources of heterogeneity.

The gold standard for slurry sampling is the linear cross-stream sampler, in which a cutter traverses across the full width of the stream at a discharge point, collecting a slice of the entire flow over the duration of each cut. This design satisfies the probabilistic sampling requirement that every particle or fluid element has an equal chance of being collected. Linear samplers are installed at the discharge points of vertically falling slurry pipes, at conveyor head pulley transfer points where slurry is present, and at the outlets of launders and chutes. They are mechanically straightforward but require adequate clearance and structural support for the cutter mechanism and drive.

For secondary and tertiary sampling of smaller, already-reduced sample streams, the Vezin-style rotary sampler is widely used. The Vezin sampler has a cutter that sweeps in a circular arc around a rotating shaft, with the cutter cross-sectional area forming a sector of a circle typically occupying 1.5 to 5 percent of the total circle. This design ensures that the cutter always intercepts the same fraction of the falling stream, regardless of the rotational position, maintaining probabilistic sampling at a reduced scale. Multiple cutter heads can be mounted on a single shaft, increasing the sampling frequency without requiring additional drive mechanisms.

Gravity slot samplers and pressure pipe samplers are widely used in practice for process control and on-stream analyser feeding, but they represent compromises from strict probabilistic principles. A gravity slot sampler placed in a horizontal launder extracts a portion of the flowing slurry through an internal cutter, promoting turbulent mixing upstream to reduce segregation effects. While this design does not cut the full stream cross-section, the argument is made that under well-turbulated flow conditions the extracted sample is sufficiently representative for process control purposes.

Pressure pipe samplers extract a sample from either a straight vertical section or a Y-section of a pressurised pipeline. Internal rods or baffles are placed upstream of the extraction point to promote turbulent mixing and reduce segregation. These samplers are extensively used for feeding on-stream elemental analysers in flotation circuits, where they provide continuous sample delivery without interrupting the process flow. Their limitation is a tendency toward bias with coarser size fractions, which are more prone to segregation in the pipeline.

An important practical consideration in slurry sampler design and selection is the susceptibility to blockage by coarse particles, fibrous material, or high-density mineral species that settle preferentially. Sampler cutter openings must be sufficiently wide to pass the largest particles present in the stream without bridging, and the sample transport lines carrying the extracted sample to the sub-sampling or analysis stage must maintain sufficient velocity to avoid sedimentation. Regular flushing and maintenance protocols are essential for reliable long-term operation of slurry sampling systems.

The validation of slurry samplers against reference measurements is particularly important because the consequences of undetected bias are severe. Systematic underestimation of concentrate grade by even 0.5 percent absolute in a high-throughput copper operation represents a substantial annual revenue misstatement. Validation should be conducted at commissioning and repeated at defined intervals, using stop-belt or manual reference cuts collected under controlled conditions.

Solid Sampling: ROM Ore, Concentrate, and Product

Solid sampling encompasses the collection of representative samples from dry or moist ore on conveyor belts, in stockpiles, in railcars, in trucks, and in bags or other discrete containers. The challenges differ from slurry sampling in that there is no liquid phase to carry fine particles in suspension, making segregation by particle size and density an even more significant concern.

For continuous solid streams on conveyor belts, the preferred sampling approach is the cross-belt sampler, which mimics the stop-belt manual reference technique by sweeping a cutter across the full width of the belt at the head pulley, where material is discharged from the belt and the stream is momentarily in free fall. When correctly designed and maintained, the cross-belt sampler cuts the complete falling stream cross-section, collecting a probabilistic sample. However, the practical challenges of maintaining correct cutter geometry, especially when the ore contains very coarse lumps that may be deflected by the cutter edges, mean that regular bias testing is particularly important for cross-belt samplers on coarse ore streams.

For ROM ore, sampling is particularly difficult because the extreme size range of run-of-mine material makes it practically impossible to collect a representative sample of sufficient mass by conventional means. The required sample mass, calculated from Gy’s formula for typical ROM particle sizes, may be hundreds of tonnes, well beyond what can be practically collected and processed for assay. In practice, ROM ore sampling is usually performed on the crusher product, where the top size has been reduced to a more manageable level, using multiple sampling increments collected over a representative production period.

Concentrate sampling for commercial accounting purposes, where the financial stakes are highest, must conform to relevant ISO standards for the specific commodity. ISO standards exist for the sampling of copper concentrates, lead and zinc concentrates, iron ore, coal, and many other bulk commodities. These standards specify minimum sample masses as a function of lot size and particle size, sampling equipment design criteria, sub-sampling protocols, and moisture determination procedures. Compliance with these standards is a contractual and regulatory requirement in most commercial concentrate sales agreements.

Sampling of concentrates in railcars or trucks presents a particular challenge because the material is a three-dimensional lot that cannot easily be sampled on a linear basis. The preferred approach is to sample the concentrate as it is being loaded into the container, using a cross-belt sampler at the loading conveyor head pulley. If this is not possible, car or truck sampling can be performed by collecting increments through the full depth of the material at defined grid points across the surface, but this approach represents a significant deviation from the probabilistic ideal and should be supplemented by loading-point sampling wherever possible.

Moisture content determination is an integral part of all solid sampling programs for metallurgical accounting. The dry mass of ore or concentrate is the relevant accounting quantity, and moisture content must be measured and applied as a correction to the wet weight. Samples for moisture are taken concurrently with assay samples, weighed immediately, dried at 105 degrees Celsius to constant mass, and reweighed. The drying temperature limit of 105 degrees Celsius is important for sulfide concentrates: higher temperatures cause sulfur dioxide loss from sulfide minerals, resulting in erroneously low dry weights and overstated moisture content.

Mass Balancing: Principles and the Two-Product Formula

Mass balancing is the process of applying conservation of mass constraints to measured data from a mineral processing circuit in order to calculate unknown flow rates, verify data consistency, and derive performance indicators such as metal recovery and concentrate ratio. It is a fundamental tool of metallurgical accounting, process optimisation, and plant performance assessment.

The simplest and historically most widely used mass balance technique is the two-product formula, which applies to any process unit with one feed and two product streams, such as a flotation cell producing a concentrate and a tailings stream. If the mass flow rate of the feed is W_F, and the assay values of a metal of interest in the feed, concentrate, and tailings are x_F, x_C, and x_T respectively, then conservation of mass gives:

W_F = W_C + W_T

and conservation of metal mass gives:

W_F x_F = W_C x_C + W_T x_T

Solving these two equations simultaneously yields the concentrate mass flow rate:

W_C = W_F (x_F − x_T) / (x_C − x_T)

This expression is the two-product formula. It is remarkably useful in practice because it allows the concentrate mass flow rate to be calculated from three assay measurements and one measured feed rate, without requiring direct measurement of the concentrate flow. The metal recovery R, the fraction of feed metal reporting to the concentrate, is then:

R = W_C x_C / (W_F x_F) = (x_F − x_T) / (x_C − x_T) × x_C / x_F

The simplicity of the two-product formula has made it a staple of metallurgical practice for well over a century. However, it has important limitations. It propagates measurement errors directly into the calculated quantities without providing any mechanism for detecting or correcting erroneous measurements. The variance of the calculated recovery is particularly sensitive to the separation efficiency of the process unit for the assayed metal: when x_C is close to x_T, indicating poor separation, small errors in either assay value produce large errors in the calculated recovery. The formula also applies only to a single process unit with one feed and two products; it cannot be directly extended to complex multi-unit circuits with recirculating streams without iterative application, which can accumulate and amplify errors.

The n-product formula extends the two-product formula to process units with n product streams, requiring n-1 assayed metals to solve for the n unknown product mass flow rates. While mathematically straightforward for up to four products, the formula rapidly becomes unwieldy for larger numbers of products and provides no mechanism for handling redundant data or detecting measurement errors. For these reasons, the n-product formula is no longer considered best practice for complex circuit mass balancing, having been superseded by statistical reconciliation methods.

Data Reconciliation: Weighted Least-Squares Mass Balance

Modern metallurgical practice requires mass balancing methods that can handle redundant data sets from complex multi-unit circuits, detect and correct measurement errors, and provide statistically optimal estimates of all stream flow rates and compositions consistent with the conservation of mass. This is the domain of data reconciliation, which uses weighted least-squares optimisation to adjust measured values to satisfy mass balance constraints while minimising the total weighted deviation from the original measurements.

The fundamental concept of data reconciliation is that when more measurements are available than are strictly necessary to satisfy the mass balance equations (a condition known as data redundancy), the excess data can be used to detect measurement errors and to improve the estimates of all measured and unmeasured quantities. The best estimates are those that minimise the sum of squared normalised residuals, where each residual is weighted by the inverse of the measurement variance. Measurements with smaller uncertainty (lower variance) are trusted more heavily and adjusted less than measurements with higher uncertainty.

Three main approaches to data reconciliation have found application in mineral processing mass balancing. The first is the node imbalance minimisation method, developed by Finch and Matwijenko (1977) and Lynch (1977), in which mass conservation equations are written with explicit imbalance terms representing the deviation from perfect balance due to measurement errors. The best estimates of unmeasured flow rates are those that minimise the sum of squared imbalances across all nodes in the circuit.

The second approach is the two-step least squares method, which separates the reconciliation problem into two stages: first, the unmeasured flow rates are estimated from the measured data, and then the measured values are adjusted to satisfy the conservation equations. This approach is computationally efficient and well-suited to implementation in spreadsheet or dedicated software environments.

The third and most statistically rigorous approach is the generalised least squares method, in which all measured and unmeasured quantities are simultaneously adjusted to satisfy mass balance constraints while minimising the weighted sum of squared adjustments. This approach provides the optimal statistical estimates under the assumption that measurement errors are normally distributed and uncorrelated. It requires knowledge of the measurement uncertainty (standard deviation) for each measured quantity, which must be estimated from calibration data, replicate measurements, or engineering judgment.

Commercial mass balance software packages such as BILMAT, JKSimMet, and Metsim implement these reconciliation algorithms and provide plant metallurgists with practical tools for routine mass balancing. These packages allow the metallurgist to define the plant circuit topology graphically, enter measured data from sampling surveys and continuous instruments, specify measurement uncertainties, and obtain a reconciled mass balance in which all stream flow rates and compositions are consistent with the conservation of mass.

An important concept in data reconciliation is the distinction between measured and unmeasured streams. In any plant circuit, some streams will be measured directly by flow meters, density gauges, or belt scales, while others are unknown and must be calculated from the mass balance. The reconciliation algorithm adjusts the measured values within their stated uncertainty bounds to achieve a consistent balance, and simultaneously calculates the best estimates of the unmeasured quantities. Streams that are not measured at all cannot be reconciled; they can only be calculated, with no uncertainty reduction from the reconciliation process.

Data redundancy is therefore a critical design consideration for any plant sampling program intended to support rigorous mass balancing. Having more measurements than the minimum necessary to satisfy the balance equations provides the statistical power needed to detect measurement errors and to improve the reliability of the reconciled results. The additional cost of extra sampling points and analyser installations is almost always justified by the improved quality of the resulting metallurgical data, particularly in large operations where small improvements in data quality can translate into significant process optimisation gains.

The sensitivity of the reconciled mass balance to measurement errors in specific streams can be assessed using error propagation analysis. Streams with assay values that are poorly separated across a process unit, or streams where the measurement uncertainty is large relative to the grade difference, contribute disproportionately to the overall uncertainty of the reconciled recovery. Identifying these sensitive streams allows the metallurgist to target measurement improvement efforts where they will have the greatest impact on mass balance quality.

Process Control and Instrumentation in Mineral Processing

The acquisition of representative, accurate sampling and mass balance data is not only important for periodic metallurgical accounting but also forms the real-time data foundation for plant process control. Modern mineral processing plants operate under multi-layer hierarchical control systems that use continuous instrumentation to maintain stable process conditions, optimise performance against economic objectives, and manage quality and environmental compliance.

At the instrumentation layer, the key continuous measurement devices in a mineral processing plant include belt scales (weightometers) for solids mass flow on conveyor belts, magnetic and ultrasonic flowmeters for slurry volumetric flow in pipelines, nucleonic density gauges for slurry density measurement, on-stream elemental analysers for continuous assay of process streams, and particle size analysers for grinding circuit control. Together, these instruments provide the continuous data stream required for real-time process monitoring and control.

Belt scales or weightometers are the most common method for measuring the tonnage of ore on conveyor belts. They consist of one or more weigh idlers mounted on a load-sensing bridge, combined with a belt speed sensor to give a continuous mass flow rate signal. Accuracy is typically 1 to 2 percent of full scale, sufficient for most process control applications but not always adequate for fiscal metallurgical accounting, where static weighing of discrete loads may be required for higher accuracy.

Magnetic flowmeters, based on Faraday’s law of electromagnetic induction, are the standard choice for volumetric flow measurement in slurry pipelines. The flowmeter generates an electromotive force proportional to the slurry velocity as it flows through a magnetic field, with the signal detected by electrodes flush with the pipe bore. Magnetic flowmeters have no moving parts in contact with the slurry, are unaffected by changes in density, viscosity, or temperature, and handle aggressive slurries without special wear provisions. Their main limitation is that they require the liquid phase to have adequate electrical conductivity, which is satisfied by most mineral processing slurries.

On-stream elemental analysis by X-ray fluorescence (XRF) is one of the most important analytical tools for continuous process monitoring in flotation circuits. Centralised XRF systems such as the Outotec Courier series sample multiple slurry streams sequentially, delivering each sample to a central analyser that measures the elemental composition in near-real time. The analytical cycle time of a few minutes per stream is fast enough to provide timely feedback for process control. In-stream probe systems, using isotope excitation sources encapsulated with a detector in a compact probe immersed directly in the slurry, offer an alternative for applications where representative sample transport to a central analyser is difficult.

On-belt elemental analysis using prompt gamma neutron activation analysis (PGNAA), exemplified by the Scantech GEOSCAN-M system, provides continuous non-contact elemental analysis of bulk ore on conveyor belts without the need for sample collection. A neutron source placed below the belt irradiates the material, and gamma ray detectors above the belt measure the characteristic emission spectra of all major elements simultaneously. When combined with a belt scale and moisture monitor, this provides a continuous tonnage-weighted elemental analysis of the ore feed at measurement intervals of 2 to 5 minutes.

The regulatory control layer of the plant control system implements PID (proportional-integral-derivative) feedback control loops that maintain individual process variables such as sump level, slurry density, and cyclone feed pressure at their set points. Advanced regulatory control strategies including cascade control, feedforward control, and Smith predictor control address the challenges of interacting control loops, significant measurement delays, and process transport delays that make simple PID control insufficient for demanding applications.

The advanced process control layer uses model predictive control (MPC) to manage multivariable interactions and constraints in a systematic, coordinated manner. MPC formulates the control problem as a dynamic optimisation over a finite future horizon, using a mathematical process model to predict future process behaviour and computing the sequence of control actions that minimises deviation from target values while respecting all operating constraints. Commercial MPC packages from vendors including ABB, Honeywell, Emerson, and Mintek are widely deployed in grinding circuit and flotation circuit control at large concentrators.

At the optimisation layer, real-time optimisation (RTO) systems use reconciled mass balance data and calibrated process models to determine the operating point that maximises economic performance, typically expressed as net smelter return (NSR) or some equivalent revenue-based objective function. RTO systems detect when the plant has reached a new steady state, reconcile the current measurement data, update the process model, and solve the optimisation problem to determine new set points for the advanced process control layer. A well-implemented RTO system can systematically push the plant toward its economic optimum in the face of changing ore characteristics, market prices, and operational constraints.

The integration of sampling, mass balancing, and process control into a unified digital plant information system is the hallmark of a modern, high-performance mineral processing operation. The data flow from automatic samplers through on-stream analysers, mass balancers, and optimisation engines to the control room display represents the information backbone on which operational excellence is built. Investment in this infrastructure, designed from first principles with attention to sampling theory and measurement uncertainty, consistently delivers returns that justify the capital and operating cost many times over.