Sampling Methods: Problems and Solutions

INTRODUCTION Although sampling techniques used by the mineral industry continue to improve, sampling is still an area which is often neglected. Frequently, sampling and sample preparation requirements are left to personnel who do not fully appreciate the significance and importance of sampling. Consequently, plant evaluations are often based on seriously biased samples, leading to mass balance problems and incorrect assessments of plant performance. The basic rule for correct sampling is that each particle of ore or concentrate must have an equal probability of being collected and becoming part of the final sample for analysis. If this is not the case, bias is easily introduced and the sample is not representative. For example, when ore is travelling on a conveyor belt, the lumps may come to the surface. Consequently, a grab sample taken from the top layers only will contain a greater proportion of lumps, i.e., the sample is biased. There seems little point in making major investments in precise analytical equipment, if the sample provided to the analytical laboratory is not representative in the first place. This paper describes the sources of plant sampling errors, reviews the methods of estimating their variance, presents some worked examples, and provides some practical advice for reducing sampling errors and their variance. SAMPLING BASICS It is generally accepted that the discharge point of an ore or concentrate stream is the most suitable sampling location. The stream can be intersected at regular intervals, and representative samples, referred to as increments, can be obtained by taking a complete cross-section of the ore stream. On the other hand, sampling devices that take part of the stream on an intermittent or continuous basis, e.g., a bleed from a pipe, may introduce serious bias, and should be avoided at all cost. Increments may be taken at completely random times or tonnages during the sampling operation, in which case it is called "random sampling". However, as shown below, random sampling results in the largest sampling variance, so it is preferable to divide the ore or concentrate stream up into strata of equal time or mass and take one increment from each stratum. This operation is called stratified sampling and may be either systematic sampling, in which increments are taken at the same point in each stratum, or stratified random sampling in which the increments are taken at random within each stratum. Systematic sampling results in the smallest sampling variance, provided periodic variations in quality or quantity are not present which may coincide with, or approximate to, any multiples of the proposed sampling interval. In such cases, it is strongly recommended that stratified random sampling within fixed time or mass intervals be carried out.