A metallurgical investigation establishes why a plant's performance has changed: recovery has fallen, reagent costs have risen, or the mill cannot hold its throughput. The cause must be identified and demonstrated before money is spent on a remedy.
Three factors make this difficult. The ore is never the same from one day to the next, so plant figures move even when nothing is wrong. The figures themselves come from samplers, instruments and laboratories that can be in error. And several things usually change at once, so the obvious explanation is often not the correct one.
The method below draws on the scientific method, on problem-solving tools from manufacturing quality, and on the statistics of plant trials. Its contribution is discipline: a fixed sequence of steps that prevents the team from settling on a preferred answer and ceasing to look.
Most plants have a record of a remedy that was installed and acclaimed and later found to have had no effect, because the ore changed at the same time. A structured investigation avoids that outcome. This article sets out a practical method for recovery losses, reagent consumption, throughput problems and metal accounting variances, with a worked example from a gold CIL plant.
The investigation cycle
The steps form a loop. If the remedy does not hold, the investigation returns to the evidence with what has been learned.
Problem definition
"Recovery is down" is not a problem statement. A useful statement specifies what changed, where, when it started and by how much. One of the oldest tools for this purpose is the is / is not table from Kepner and Tregoe's problem analysis method, which requires the investigator to record where the problem appears and where it could appear but does not. The contrast often points directly to the cause.
| Is | Is not | |
|---|---|---|
| What | Gold recovery down from 92% to 89% | Throughput, grind size, residue on carbon fines |
| Where | Solid residue grade up from 0.15 to 0.22 g/t | Tailings solution gold, which is unchanged |
| When | Started three days after the new ore blend began | Before the blend change |
| How much | About 3 percentage points, steady since it began | Not getting worse day by day |
This illustrative table is already informative. The gold is being lost as undissolved gold in the solids, not as dissolved gold in solution, so the problem lies in the leach rather than the carbon circuit, and the timing points to the ore blend.
Data verification
Before a number is explained, it must be confirmed as real. The tailings and feed samplers should be checked for changes, wear, blockage or relocation. The assay standards and duplicates for the period should be reviewed, together with any change of method, analyst or laboratory. Calibration dates for the weightometer and flowmeters should be confirmed, and probes checked for readings that do not vary through the day.
The plant mass balance should still close; a balance that has stopped closing suggests a measurement problem rather than a process problem. Re-assaying a few retained samples, or sending duplicates to a second laboratory, costs little compared with remedying a problem that does not exist.
Locating the loss
The next step narrows down where in the process the change appears. For a recovery loss, a survey down the circuit, sampling every tank or stage on the same shift, shows where the profile changed. Size-by-size assays of the tailings show whether the lost gold is in the coarse or fine fractions: coarse losses suggest liberation or grind, and fine losses suggest slimes or solution chemistry.
Diagnostic leaching of the residue dissolves the gold in stages to show how much is free, how much is locked in sulphides and how much is locked in silicates. Mineralogy by automated scanning electron microscope shows the gold and its associated minerals directly.
Hypotheses
The geologist Thomas Chamberlin argued in 1890 for the "method of multiple working hypotheses": several explanations are kept under consideration at once, so that the investigator does not become attached to the first. The fishbone diagram developed by Kaoru Ishikawa is a practical means of doing this, with each bone representing a family of causes:
The diagram should be completed as a team, with operators, maintenance, the laboratory and geology represented. Those who work closest to the equipment often know of changes that were never recorded.
Weighing the evidence
For each hypothesis, the evidence for it and the evidence against it are recorded. A hypothesis contradicted by the evidence can be set aside, and those that remain are ranked by likelihood and by the cost and speed of testing them.
| Hypothesis | Evidence for | Evidence against | Next test |
|---|---|---|---|
| New blend contains more sulphides that consume oxygen | Timing matches; dissolved oxygen in tank 1 lower since the change | None yet | Bottle-roll leach of the blend at normal and high oxygen |
| Coarser grind | None | Cyclone overflow sizing unchanged | None needed |
| Carbon activity dropped | None | Tailings solution gold unchanged | None needed |
| Tailings sampler bias | Residue grade rose sharply | Duplicate cuts and check assays agree | None needed |
The "five whys" technique from the Toyota production system traces each surviving hypothesis back to a root cause. Recovery is down because undissolved gold is up; undissolved gold is up because the leach is slower; the leach is slower because dissolved oxygen in the first tanks is low; dissolved oxygen is low because the new ore consumes more oxygen; and the new ore consumes more oxygen because it contains more reactive sulphide. The final answer is one on which action can be taken.
Testing the hypothesis
Laboratory tests are fast and controlled. A bottle-roll leach of the new blend with and without additional oxygen, for example, can confirm or eliminate the oxygen hypothesis within days.
Plant trials are more difficult, because daily ore variation obscures the effect being sought. Comparing one month before a change with one month after is the weakest form of evidence. Better designs, described in detail by Napier-Munn, include alternating on/off trials, in which the change is switched on and off in random blocks so that ore variation affects both conditions equally, and paired comparisons, in which two parallel lines fed the same ore operate with and without the change. In either design, the number of repeats needed to distinguish a real effect from noise should be calculated before the trial starts.
Implementation and verification
Once a cause is proven, the change is made and the result confirmed with the same measurements that defined the problem. In the example, the remedy was additional oxygen in the first leach tanks. Recovery should return towards 92%, residue grade towards 0.15 g/t, and dissolved oxygen in tank 1 to its target.
The investigation report
The operating procedure, set-points and alarm limits are updated, and the new ore type is added to the blend rules or the geometallurgical model so that future schedules allow for it.
A short report records the problem statement, the evidence, the hypotheses eliminated and the reasons, the remedy and the result. It is the most valuable product of the investigation, because it saves the next team from starting again.
Common pitfalls
Confirmation bias, the search only for evidence that supports a preferred theory, is the most common failure. Changing several things at once leaves no one able to say which change worked. Data can be trusted blindly or, conversely, blamed whenever a result is unwelcome. Short trials cannot separate the effect from day-to-day ore variation.
Regression to the mean is easily overlooked: a very poor week is usually followed by a better one, with or without a remedy, and the remedy should not be credited with the natural recovery. Finally, an investigation can stop at the symptom, for example by correcting the oxygen supply without informing the mine planners of the new ore type.
Theseus follows this method. The problem is described in plain English, and Theseus returns ranked hypotheses, the evidence for and against each, and the first test to run. Saldora holds the daily data and change history on which an investigation depends.
Glossary
- Hypothesis
- A possible explanation that can be tested and could turn out to be wrong.
- Root cause
- The underlying reason for a problem, which, if fixed, stops it recurring.
- Is / is not analysis
- A table comparing where a problem occurs with where it could occur but does not, from Kepner and Tregoe's method.
- Fishbone (Ishikawa) diagram
- A diagram grouping possible causes of a problem into families such as ore, equipment, process and measurement.
- Five whys
- Asking 'why' repeatedly until the answer is something that can be acted on.
- Diagnostic leach
- A sequence of laboratory leaches that shows how much gold is free, locked in sulphides or locked in other minerals.
- Plant trial
- A test of a change on the operating plant, designed so that ore variation does not hide or fake the result.
- Regression to the mean
- The tendency for an unusually bad or good result to be followed by a more normal one, whatever was done in between.
- Confirmation bias
- The tendency to notice evidence that supports an existing belief and to overlook evidence against it.
Frequently asked questions
How long should a plant investigation take?
As long as the evidence requires. A data error can be found in a day. A plant trial to prove a reagent change may need several weeks of alternating blocks to separate the effect from ore variation.
What is the most common cause of an apparent recovery drop?
There is no single answer, but measurement and sampling problems, ore changes and gold-in-circuit timing are all common. For this reason the data are verified before they are explained.
Why not compare the month before and the month after a change?
The ore and operating conditions also changed between the two months. Alternating or paired trials compare like with like.
Who should be involved?
Metallurgists, operators, maintenance, the laboratory and geology. Each sees part of the picture.
What should the final report contain?
The problem statement, the evidence, the hypotheses eliminated and the reasons, the remedy, the verification data and any changes to procedures.
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
- Chamberlin, T.C. (1890). The method of multiple working hypotheses. Science, 15, 92–96.
- Kepner, C.H. and Tregoe, B.B. (1965). The Rational Manager. McGraw-Hill.
- Ishikawa, K. (1986). Guide to Quality Control, 2nd revised edition. Asian Productivity Organization.
- Napier-Munn, T.J. (2014). Statistical Methods for Mineral Engineers: How to Design Experiments and Analyse Data. JKMRC Monograph Series, University of Queensland.
- Ohno, T. (1988). Toyota Production System: Beyond Large-Scale Production. Productivity Press.
Worked examples and charts marked illustrative use invented figures, not data from any operation.