Plant Sampling: Where Metallurgical Accounting Fails

A representative increment is a portion of a process stream collected in such a way that every particle in the stream has the same probability of being selected. That sounds simple, but it's rarely achieved by hand. A cross-stream cutter is the standard tool: a narrow, rectangular opening that traverses the entire flowing stream at a constant speed, cutting the stream at right angles. The cutter must be wide enough to avoid bridging—typically at least three times the top particle size—and it must travel completely across the stream without splashing or overflowing. Cutter design matters because a poorly designed cutter selectively rejects coarse or dense particles, introducing systematic bias.

Sample mass is tied to particle size. The larger the top particle size, the larger the minimum sample mass needed to keep the fundamental sampling error low. The relationship is non-linear: as top size increases, required sample mass grows much faster. You can't just take a fixed scoop and expect it to represent a coarse stream. If the sample mass is too small for the particle size distribution, the sample cannot contain a statistically meaningful number of the largest particles, and the grade will swing wildly from increment to increment. This is why a few grams from a fine flotation concentrate can be adequate, while the same few grams from a coarse crusher product is almost meaningless.

A grab sample is a single instantaneous portion taken by hand—a scoop from a launder, a handful from a belt, a bucket from a sump. It is easy and fast, and it is almost always wrong. The stream varies minute to minute in grade, particle size, and moisture. One grab cannot represent hours of production. A composite sample is a collection of increments taken over a defined period and combined proportionally to flow. The difference is not just quantity; it's representativeness. A composite built from regular, correctly designed increments averages out short-term fluctuations and approaches the true mean of the stream.

To build a proper composite, first decide the sampling period—a shift, a day, or a complete batch. Second, set the increment frequency based on stream variability and sample mass requirements. Third, collect each increment with a correctly designed cutter. Fourth, store increments in a sealed container, preventing moisture loss or contamination. Fifth, after the period ends, mix the increments thoroughly and split them down to the analytical sample. Skipping any step reintroduces bias. If you need a quick check, a grab sample is better than nothing, but you should never use it for metallurgical accounting. It simply doesn't carry the statistical weight.

Defensible sampling is an engineering discipline. It starts with sampling theory, moves through equipment selection, and ends with audited procedures. The Canadian Institute of Mining, Metallurgy and Petroleum (CIM) publishes guidance on sampling and metallurgical accounting that sets a defensible baseline. In practice, you need trained personnel, documented sampling protocols, and periodic bias tests. A bias test compares the sampler against a reference method—usually a full-stream cut taken by stopping the entire flow for a few seconds. If the routine sampler and the reference disagree beyond an acceptable tolerance, fix the sampler before trusting any numbers.

Metallurgical testwork programs—like those described in metallurgical testwork—generate the design data, but plant sampling sustains it. A good testwork program will specify expected grades and variability, which in turn define the sampling system requirements. When the plant is running, that system becomes the front line of metallurgical accounting. Treat it as a critical instrument, not an afterthought. Calibrate it, maintain it, and audit it regularly. Your recovery numbers are only as honest as the samples behind them.