A mass balance rests on a simple rule: what enters a process must leave it or remain within it. If 300,000 tonnes of ore containing 540 kg of gold entered a plant during a month, then 540 kg of gold must be found in the gold produced, the gold remaining in the plant and the gold lost to tailings.
In operating plants the balance never closes exactly, because every figure is measured and every measurement carries some error. The belt scale may read half a percent high, the feed sample may not fully represent the ore, and the assay has its own scatter. Combining several slightly erroneous figures leaves the balance out by a small amount even when nothing has gone wrong.
Data reconciliation is the standard, defensible method of dealing with this. It adjusts each measurement by the minimum necessary to close the balance, adjusting the least reliable measurements most, and it indicates whether the imbalance that had to be closed was of normal size or large enough to indicate a genuine problem. Without it, the difference must be explained, written off or carried in an "unaccounted" line at month-end.
Sources of imbalance
A monthly balance combines figures that were each measured with some error. Belt weightometers drift between calibrations; Gaylard and co-authors cite a target of ±0.5% accuracy for metal accounting weighers. A cross-stream cutter that is too narrow, too fast or worn delivers a sample that does not represent the stream, and sampling error is often the largest term in the error budget. Fire assay and solution analyses have their own precision, visible in duplicate and standard results.
Stocks and timing also contribute. Gold held on carbon, in solution and in the mill circuit changes from month to month, and material fed late in the month may not be recovered until the next. Even when every error is small and random, the errors do not cancel to zero, so some imbalance is the expected outcome rather than a failure.
Weighted least squares reconciliation
Data reconciliation finds a set of adjusted values that satisfy every balance exactly while remaining as close as possible to the measurements. Closeness is weighted by the standard deviation, σ, of each measurement. The method minimises:
where
- measured value of quantity j (tonnes, grade or gold mass)
- reconciled (adjusted) value of quantity j
- standard deviation of measurement j, in the same units
- balance equation k, written as metal in minus metal out
A reliable value, with a small σ, is barely moved, and a less reliable value is adjusted more. This is the approach used by commercial mass balancing packages, which report balanced flows, assays and recoveries together with confidence intervals.
Reconciliation requires redundancy: more measurements than the minimum needed to calculate the unknowns. Where a stream is not measured at all, the balance can estimate it, but nothing can then check the estimate.
Worked example: one month in a CIL plant
The table shows an illustrative month, with uncertainties expressed as one standard deviation. Gold recovered is gold poured plus the change in gold held in the circuit.
| Measurement | Measured | σ | Reconciled | Change |
|---|---|---|---|---|
| Dry tonnes milled | 300,000 t | 1.5% | 298,493 t | −0.50% |
| Head grade | 1.80 g/t | 3% | 1.760 g/t | −2.24% |
| Residue grade | 0.150 g/t | 5% | 0.151 g/t | +0.52% |
| Gold recovered | 480,000 g | 0.5% | 480,266 g | +0.06% |
The month has one balance equation for gold:
where
- dry tonnes milled (t)
- ,
- head grade and residue grade (g/t)
- gold recovered: gold poured plus the change in gold held in the circuit (g)
As measured, gold in feed less gold in residue is 495,000 g, while 480,000 g was recovered, leaving 15,000 g, or 2.8% of the gold in feed, unaccounted for. After reconciliation the balance closes exactly. Most of the adjustment falls on the head grade, the least certain figure relative to its influence, while the poured gold, the most precise measurement, moves by 0.06%.
The three bases for stating recovery show the significance of this:
| Basis | Recovery |
|---|---|
| (head − residue) / head, as measured | 91.7% |
| Gold recovered / gold in feed, as measured | 88.9% |
| Reconciled | 91.4% |
Reporting 88.9% treats the whole imbalance as lost gold, and reporting 91.7% ignores the gold that cannot be located. The reconciled figure uses all of the information and is consistent with every measurement.
The chi-square test
The sum of the squared adjustments, each divided by its σ, indicates whether the imbalance is consistent with the stated uncertainties. If the measurements carry only random error of the stated size, this sum follows a chi-square distribution with as many degrees of freedom as there are independent balance equations. For one constraint, the 95% critical value is 3.84.
where
- sum of the squared normalised adjustments
- degrees of freedom: the number of independent balance equations
- chi-square critical value at 95% confidence; 3.84 for one equation
In the example the sum is 0.69, well below 3.84. The 2.8% imbalance is consistent with the stated uncertainties, so no gross error is indicated. Had the sum exceeded the critical value, the next step would be to examine the measurement with the largest normalised adjustment, or to question whether the σ values are realistic.
Good practice
Standard deviations should be based on evidence from sampling studies, duplicate assays and weightometer calibration records, and should never be tuned until the balance passes. The direction of the adjustments should also be monitored. Reconciliation assumes random error, so if the head grade is adjusted downward every month, the head sampler or assay is probably biased and the correction belongs in the plant rather than in the spreadsheet.
Gold on carbon, in solution and in the mill circuit should be confirmed by physical stocktakes at set intervals. An audit trail should record the measured values, the σ used, the adjustments and their approval; the AMIRA P754 Code expects metal accounting to be transparent and auditable.
Saldora's month-end close reconciles the month totals by weighted least squares and tests the closure against the chi-square 95% point. For circuit surveys, Ajax reconciles survey data before it fits model parameters.
Glossary
- Mass balance
- An account showing that what enters a process equals what leaves it plus any change in what is held inside.
- Reconciliation
- Adjusting measured values so the balance closes, in proportion to how uncertain each one is.
- Standard deviation (σ)
- A measure of how much a measurement typically scatters around the true value.
- Weighted least squares
- A method that finds the smallest set of adjustments, weighted by each measurement's uncertainty, that makes the balance close.
- Redundancy
- Having more measurements than the minimum needed, so that they can check each other.
- Bias
- A systematic error that always pushes a measurement the same way, such as a scale that always reads high.
- Chi-square test
- A statistical check of whether the total adjustment is consistent with the stated measurement uncertainties.
- Weightometer
- A belt scale that weighs ore on a moving conveyor.
Frequently asked questions
Does reconciliation hide problems?
Not when it is applied correctly. It removes random error and reports the size of the adjustments. A failed chi-square test, or adjustments in the same direction every month, indicate a genuine problem.
Where do the standard deviations come from?
From evidence: sampling studies, duplicate and standard assays, and weightometer calibration records. They should never be tuned to make the balance pass.
How large an imbalance is acceptable?
That depends on the precision of the measurements. The chi-square test answers the question for the plant's own data, which is preferable to a fixed percentage.
Can a balance be reconciled with only one measurement per stream?
Yes, provided the balance has some redundancy. Where a stream is not measured at all, the balance can estimate it but nothing can check it.
Discuss your plant with a metallurgist
Tiegel is a Perth metallurgy company. We help with testwork, plant optimisation, metal accounting and investigations, and we build the Tiegel Suite software.
Talk to a metallurgistReferences and further reading
- Gaylard, P.G., Randolph, N., Wortley, C.M.G. and Ralston, I. (2006). Designing for metal accounting. Journal of the Southern African Institute of Mining and Metallurgy, 106. Link
- Metso. Mass balancing of concentrator data (2017, updated 2022). Link
- Applying the AMIRA P754 Code of Practice for Metal Accounting. AusIMM MetPlant 2008. Link
- Wills, B.A. and Finch, J.A. (2016). Wills' Mineral Processing Technology, 8th edition. Butterworth-Heinemann.
Worked examples and charts marked illustrative use invented figures, not data from any operation.