Carbon inventory extrapolation calculator
Survey the tank just before a carbon pull and again just after it. The drop in the survey against the mass removed shows how far your survey method under- or over-reads. Extending the line to zero gives the carbon actually in the tank, and the correction factor fixes your carbon matrix equation.
The example values are filled in so you can see how it works. Results update as you type.
Estimating carbon inventory from one pull and two surveys
Limitations of carbon surveys
Gold on carbon is often the largest single component of gold in circuit in a CIL or CIP plant. It is calculated tank by tank as carbon mass multiplied by gold loading. The loading is determined by assay and is generally reliable. The carbon mass is determined by survey, and that is where most of the uncertainty lies.
A typical survey takes a known volume of slurry from the tank, screens out the carbon, measures it and scales the result up to the slurry volume in the tank. Some methods weigh the carbon; others measure its settled volume and convert it to mass with a bulk density. Errors can enter at three points. The sample may not represent the tank, because carbon is not evenly distributed through the slurry and the sampling points, depths or timing favour some zones. The bulk density or moisture basis used in the calculation may no longer match the carbon in the tank, which changes as it wears, breaks down and fouls. The slurry volume may be incorrect, because the tank has sanded, the level measurement is inaccurate, or the slurry is aerated.
Repeat surveys do not reveal these errors. They show how well a survey repeats, not whether it is accurate. Testing accuracy requires an independently known change in the carbon in the tank, and a carbon pull provides one.
Principle of the pull test
Suppose the survey reports a fixed fraction B of the carbon actually in the tank: 0.8 if it misses a fifth of the carbon, 1 if it reads correctly. If a weighed mass R is removed, the actual carbon falls by R and the survey reading should fall by B × R. Dividing the fall in the survey reading by the mass removed gives B, and dividing the survey reading by B gives the carbon actually in the tank.
When the survey reading is plotted against carbon removed, the two surveys lie on a straight line that reaches zero at that same mass, which is the origin of the term extrapolation.
Assumptions and limitations
The calculation is exact, but the result depends on three assumptions, which two surveys can only partly verify.
First, the survey must report a fixed fraction of the carbon. A proportional error, such as an incorrect bulk density or a sampler that under-collects carbon, is exactly what the test detects. A fixed error is not. If carbon in a sanded or dead zone is never sampled and never pulled, the test gives an incorrect result with no indication in the data.
Second, the pull must be the only change in the tank's carbon between the two surveys. Carbon advanced into or out of the tank, or fresh carbon added, is treated by the calculation as part of the pull.
Third, the pull must be measured independently of the survey, which in practice means weighing it. If the pull is calculated from a volume using the same bulk density as the survey, any error in that bulk density cancels and cannot be detected.
In addition, a single pull gives limited precision. A pull of 10% of the tank's carbon multiplies survey scatter by about 9. The test is therefore a reliable indicator of whether the carbon matrix is substantially in error, and of the approximate size of the error, but on its own it is a weak basis for revising the matrix. It should be used together with the checks described under Diagnosing an inaccurate carbon matrix. The graph adds no information beyond the calculation: with two surveys it is the same result presented visually.
Planning the pull
Because only one pull is available, its size and the number of repeat surveys should be decided in advance. The result depends on the fall between the two surveys, which is small relative to the survey readings, so survey scatter is magnified. The degree of magnification depends only on the fraction of the tank's carbon removed (see Precision of the result).
First establish the repeatability of the survey. On a day with no pull, survey the same tank two or more times in succession using the normal method, and calculate the relative standard deviation. Then decide the precision required, for example ±10% at 95% on the carbon in the tank, and enter both values in the calculator's precision section. The calculator reports the fraction of carbon the pull must remove for that precision with the planned number of surveys, and the number of surveys required on each side with the expected pull. If neither is practical, a wider range must be accepted, and the test treated as a check for a large bias rather than as a source of a precise factor.
Test procedure
- Stop all carbon movement into and out of the tank, other than the pull, from the first survey until the second. Do not advance carbon or add fresh or regenerated carbon. Record the mass of any carbon that does move, because the calculation assumes the pull is the only change.
- Survey the tank immediately before the pull, repeating the survey as many times as planned, using the normal method and sampling points. Record every result and the slurry level.
- Pull the carbon and weigh it on the same basis as the survey. If the survey reports dry carbon and the pull is weighed wet, determine the moisture on a dried subsample. If the pull can only be measured by volume, refer first to Pull measured by volume.
- Allow the tank to redistribute the remaining carbon, then repeat the survey the same number of times and in the same way: the same sampling points, the same sample volumes and, where possible, the same personnel. Record the slurry level again.
- Enter the results in the calculator, with the repeat results for each survey in one field, separated by spaces. Results may be entered as carbon mass, or as concentration in g/L with the slurry volume at each survey.
Calculation
If the survey reading Y is proportional to the actual carbon M, then Y = B × M. Before the pull, Y1 = B × M0. After removal of R, Y2 = B × (M0 − R). Subtracting the second equation from the first gives Y1 − Y2 = B × R, from which B and then M0 follow. Where surveys are repeated, the averages are used.
where
- ,
- carbon your survey method shows, and the carbon actually in the tank (kg)
- survey response: the fraction of the actual carbon the survey shows (1 if it reads true)
- ,
- average survey results just before and just after the pull, by the same method; from concentration, c (g/L) × slurry volume V (m³) gives kg
- carbon removed in the pull, on the same basis as the surveys
- ,
- carbon actually in the tank before and after the pull
- correction factor: multiply survey results from this method, in this tank, by K
- ,
- error magnifier (a 1% error in either survey average moves M₀ by E%), and the share of the tank’s carbon removed
- , ,
- relative standard deviation of one survey, number of repeat surveys averaged on each side, and relative standard deviation of the pull mass
- ,
- precision wanted (half-width of the 95% range, as a fraction), and the smallest share of the carbon the pull must remove to reach it
Against carbon removed, the survey reading follows Y = B × (M0 − X), a line of slope −B that reaches zero at X = M0. The carbon remaining in the tank follows M0 − X, which falls one for one and reaches zero at the same point. The difference between the two lines is the carbon that the survey does not detect.
Concentrations are converted to mass before the calculation. Since 1 g/L equals 1 kg/m³, concentration in g/L multiplied by slurry volume in m³ gives kilograms. The volume at each survey is used, so a change in tank level between the surveys is taken into account.
The range is calculated by first-order propagation of the uncertainty in the two survey averages and in the pull mass. Where surveys are repeated, their own scatter is used, pooled as a relative standard deviation, with the Student t value for the number of repeats. Otherwise the repeatability entered is used. In both cases the errors are assumed to be random and independent.
Worked example
The following figures are illustrative. A tank surveys at 8,000 kg of carbon. A pull of 1,000 kg is removed and weighed. The tank then surveys at 7,200 kg.
| Step | Working | Result |
|---|---|---|
| Fall in the survey reading | 8,000 − 7,200 | 800 kg |
| Survey response B | 800 ÷ 1,000 | 0.800 |
| Carbon in the tank before the pull, M0 | 8,000 ÷ 0.800 | 10,000 kg |
| Carbon in the tank after the pull | 10,000 − 1,000, or 7,200 ÷ 0.800 | 9,000 kg |
| Correction factor K | 1 ÷ 0.800 | 1.250 |
| Fraction of carbon pulled, f | 1,000 ÷ 10,000 | 10% |
| Error magnifier E | 7,200 ÷ 800 | 9 |
The survey under-reads by 20%, reporting 8,000 kg where 10,000 kg is present. The line through (0, 8,000) and (1,000, 7,200) falls by 0.8 kg for each kilogram removed and reaches zero at 10,000 kg.
Expressed as concentrations, with 1,000 m³ of slurry in the tank at both surveys, 8.0 g/L and 7.2 g/L correspond to 8,000 kg and 7,200 kg. The actual carbon concentration was 10.0 g/L before the pull and 9.0 g/L after it.
Correcting the carbon matrix
The carbon matrix converts each tank's survey result to carbon mass. For the tank tested, the correction is to multiply the survey result by K. In the example, a later survey of 6,000 kg corresponds to 1.25 × 6,000 = 7,500 kg. In terms of concentration and volume, carbon (kg) = K × c (g/L) × V (m³). Where the equation includes a single multiplying calibration constant, such as the bulk density used to convert a measured carbon volume to mass, the same correction can be made by multiplying that constant by K.
Because gold on carbon is carbon mass multiplied by loading, the same factor applies to the gold on carbon in that tank, and from there to gold in circuit and reconciled recovery. The test, its date, the factor and its range should be recorded with the month's working.
The range is as important as the factor. If the 95% range of K includes 1, the test has not demonstrated a bias in the survey, and changing the equation would only add noise. If the range is narrow and excludes 1, the correction is supported.
A factor determined in one tank applies to that tank and that survey method. Carbon varies along the train: the lead tanks hold the most heavily loaded carbon, which has also been in the circuit longest since regeneration, and mixing and sampling conditions differ between tanks. Applying one tank's factor to the others is an assumption, should be recorded as such, and can be tested without a further pull using the checks below.
Diagnosing an inaccurate carbon matrix
The matrix calculates each tank's carbon as slurry volume × carbon concentration in the sample × a conversion to mass, usually a bulk density. Where the matrix is inaccurate, the better approach is to identify which of these terms is in error and correct it, rather than apply a single factor to every tank. Most of the following checks do not require a pull.
Gold assays and solution gold
If the evidence of an error is that gold in circuit does not reconcile, the loaded carbon assays, the preparation of those samples and the gold in solution should be checked before the carbon mass is questioned. Gold on carbon is carbon mass multiplied by loading, and solution gold is added to it, so carbon mass is only one of the inputs.
Bulk density
For the next several surveys, measure each sample by the normal method, then wash, dry and weigh the same carbon. The dry mass divided by the mass given by the equation for that sample is a direct measurement of the bulk density correction, and it can be repeated as often as required, in every tank. The lead and tail tanks should be checked separately, because loaded, fouled and worn carbon differs from fresh carbon. A more robust option is to survey by dried weight and remove the bulk density term altogether.
Sampling
Sample one tank at several depths and positions at the same time. A consistent difference in carbon concentration between positions indicates that single-point sampling is biased. Samples should be composited from the full depth and from several positions, and larger samples should be used to reduce scatter.
Tank volume
Use the slurry level at the time of the survey rather than the nominal tank volume, and deduct the sanded or dead volume measured at the most recent shutdown inspection.
Shutdown counts and carbon balance
The most reliable calibration available to most plants is a full count when a tank is drained for maintenance. Survey the tank immediately before it is drained, then screen out all of its carbon and weigh it, determining the moisture on a dried subsample. This gives the true inventory of that tank, detects the fixed errors that the pull test cannot, and requires no additional pull.
A carbon book balance is a useful complement. Each month, take the opening inventory, add the weighed fresh carbon and deduct losses, such as fines to tailings and carbon lost through screen failures, then compare the running total with the matrix. Losses are difficult to measure, so the balance checks the trend rather than establishing the true inventory, but a matrix that diverges steadily from the book total indicates a problem.
Once the individual terms have been checked and corrected, the pull test provides a cross-check of the corrected matrix as a whole.
Implementing the correction
Correct the term that is in error, whether bulk density, volume or sampling, so that each tank receives its own correction. Introduce the change from a stated date, and report the resulting one-off step in gold in circuit as an adjustment in that month's reconciliation, so that it is not read as a change in recovery.
Interpreting the graphs
The first graph plots survey reading against carbon removed. The two surveys, or their averages with each repeat shown as a small point, are joined by a solid line that continues as a dashed line to zero. That point is the carbon in the tank before the pull. The dotted line is the carbon actually remaining after each amount removed, falling one for one. If the survey read correctly, the two lines would coincide. Where a range has been calculated, it is shown as a shaded bar on the axis.
The second graph is the calibration line for the survey method: survey reading against the carbon actually in the tank. The two surveys lie on a line through zero with slope B, and the dotted 1:1 line represents a survey that reads correctly. A later survey is corrected by reading across from the survey result to the solid line and down to the actual carbon, which is equivalent to multiplying by K.
Precision of the result
A 1% error in either survey average changes the result by E%, where E = Y2 ÷ (Y1 − Y2). This is equal to (1 − f) ÷ f, where f is the fraction of the tank's actual carbon removed by the pull, so it depends only on the size of the pull:
| Fraction of carbon pulled | Error magnifier | Uncertainty of the result, single surveys repeating to 1% |
|---|---|---|
| 5% | 19 | ±27% |
| 10% | 9 | ±12.7% |
| 20% | 4 | ±5.7% |
| 25% | 3 | ±4.2% |
| One third | 2 | ±2.8% |
| 50% | 1 | ±1.4% |
The last column is one standard deviation arising from the two surveys alone, √2 × E × 1%; the 95% range is approximately twice this. In the example, with single surveys that repeat to 1%, the result is 10,000 kg ± 1,273 kg, or approximately 7,500 to 12,500 kg at 95%. That is sufficient to show that the survey under-reads, but not to establish a precise factor. The uncertainty scales with survey repeatability, so surveys that repeat to 3% give three times these values.
With a single pull, precision can be improved in two ways. A larger pull is the more effective: increasing the pull from 10% to 25% of the tank's carbon reduces the magnifier from 9 to 3. Repeat surveys have a smaller effect, because the average of n surveys scatters √n times less than a single survey; four surveys on each side halve the uncertainty and nine reduce it to a third. Repeat surveys also allow the calculator to estimate scatter from the data rather than from an assumed repeatability.
Repeat surveys reduce scatter only. A consistent bias affecting both surveys equally is precisely what the test measures, and repetition does not change it. Where the range exceeds about 30% either side, the first-order range also understates the upper limit, and the factor should not be used on its own to revise the equation.
Pull measured by volume
If the pulled carbon is measured as a volume, for example the fill of an elution column, and converted to mass using the same bulk density as the survey, any error in that bulk density is present in both the surveys and the pull. It cancels from B and cannot be detected. The test still checks the other terms, sampling and tank volume, but the carbon it reports is expressed on the same bulk density basis as the survey. Selecting Measured as a volume in the calculator labels the results accordingly.
To check the bulk density as well, either weigh the pull or measure the bulk density of the tank's carbon directly: take carbon from the tank, measure it as the survey does, and weigh it on the same dry or wet basis. This check requires no pull and can be repeated as often as required, in every tank.
Sources of error
A fixed offset is the most significant. If the survey reads B × M + c, for example because carbon in a dead zone is never sampled and never pulled, the line reaches zero at M0 + c ÷ B rather than M0. Two surveys cannot distinguish an offset from the factor; only an independent full count, such as screening out and weighing all the carbon in a tank, can.
Carbon movement between the surveys has a similar effect. Carbon advanced from the next tank, fresh carbon added, carbon lost through a damaged screen, or fines generated by attrition and washed out all change the tank's carbon without being weighed.
Differences in survey method between the two surveys also introduce error. Changes in sampling points, depths, sample volumes, personnel or agitation alter B between the surveys, so the test measures the change in method rather than the carbon.
The basis of the pull must match the survey. A wet weighed mass compared with a dry survey gives an incorrect B, and a pull converted with the survey's own bulk density conceals any bulk density error.
Finally, the survey response may not be proportional. If the fraction of carbon collected by the sampler changes with the amount of carbon in the tank, the survey is not a fixed fraction of the carbon, and two surveys cannot detect this. Comparing the corrected inventory with the plant's carbon records over the following months, including fresh carbon added, carbon removed and fines lost, is the practical means of confirming that the factor holds.
Frequently asked questions
Is the pull test sufficient on its own to correct a carbon matrix?
No. It is a reliable check of whether the matrix is substantially in error, but with one pull its precision is limited, and it cannot detect a fixed error such as carbon in a sanded zone. First identify which term of the matrix is in error: dry and weigh the same samples to check the bulk density, sample at several depths to check the sampling, and use the actual slurry volume. Where possible, count all the carbon in a tank when it is drained at a shutdown, which is the most reliable single calibration.
Why not count all the carbon in the tank?
Screening out and weighing all the carbon in a tank is the most direct check, and the only one that can detect a fixed offset, but it takes that tank's carbon out of service. The pull test uses a single pull and requires only the surveys and a weighing.
How large should the pull be?
As large a fraction of the tank's carbon as is practical. Survey errors are multiplied by (1 − f) ÷ f, where f is the fraction pulled: 9 for a 10% pull, 3 for 25% and 1 for half. With the survey repeatability and the required precision entered, the calculator gives the pull size needed.
How many times should each survey be repeated?
As many times as the planning result indicates, and at least two or three so that the calculator can estimate the scatter from the data. The average of n surveys scatters √n times less than a single survey, so four surveys on each side halve the uncertainty.
Can one test correct every tank in the circuit?
Not without further checks. The factor applies to the tank and survey method tested. Carbon loading, wear and mixing differ between tanks, so applying the factor elsewhere is an assumption. Measuring the bulk density of each tank's carbon directly tests that assumption without a further pull.
What if the survey after the pull shows more carbon than the survey before it?
The surveys cannot then be compared. Either something other than the pull changed the tank's carbon, the volumes are incorrect, or survey scatter exceeds the fall in carbon. Check for carbon moved or added between the surveys, and plan a larger pull or more repeat surveys for the next test.
What does a factor above 1 indicate?
A correction factor above 1 means the survey under-reads, reporting less carbon than is present. A factor below 1 means it over-reads. A survey response B above 1, and therefore a factor below 1, can also indicate that carbon left the tank by some route other than the pull between the surveys.
Should the carbon be weighed wet or dry?
On the same basis as the survey. If the survey reports dry carbon, weigh the pulled carbon and correct it to dry mass using the moisture of a dried subsample. Comparing a wet mass with a dry survey introduces the moisture directly into the factor.
Want the background? Read Gold reconciliation from resource model to gold bar: F-factors, gold in circuit and scheduling effects.
Discuss your plant with a metallurgist
Tiegel is a Perth metallurgy company. We help with testwork, plant optimisation, metal accounting and investigations.
Talk to a metallurgistThese calculators use standard published formulas and are provided as a guide. Check results before relying on them for design or reporting.