Mining - Underground Mining - The Mathematics of Mine Sampling

The problem of estimating the precision of systematic samples from a mineral deposit is attacked by interpolating the quality, or other attribute measured, by using Fourier approximation. Such approximation using trigonometric functions yields a particularly simple expression for the variance of the mean value of the attribute sampled. The increasing volume of publication in which the science of statistics has been applied to the problem of predicting the quality of an ore deposit from sets of samples taken from it might lead one to believe that engineering judgment can be replaced by statistics. There is, in fact, an implicit application of the principle of indifference or of "no advance knowledge" which has been discredited in modern statistical theory and practice. These pages will show that it is not always advantageous, efficient, or safe to apply this principle. This problem will be dealt with by using series of orthonormal functions to describe the distribution of the quality of ore in space as contrasted with the more commonly used distribution of quality as a frequency function. Mining engineers are rarely well acquainted with the use of such series or of the functions comprising them. These uses are commonplace in the fields of communication and information theory and it is not surprising that applications taken from these fields should be found appropriate for the analysis of the information contained in a set of samples from an ore deposit. It is hoped that presentation of the following basic, but incomplete, results will generate wider interest in this method and a fuller development of it as well as of the necessary computational techniques. FUNDAMENTAL PREMISES Random variables in the strictest sense are not very common in mine sampling. The sets of samples are almost always taken systematically and are, hence, not subject to all the laws governing random events. Therefore, unless there is indeed random experimental error in the determinations of quality, the treatment of systematic samples of n members as though they were random samples of n members often leads to calculation of confidence limits for the mean quality which are larger than they should be with the result that unnecessary additional sampling is done. The reason for this will emerge in the following discussion leading to Eq. 24 and its generalization. In order to inquire into such questions on a rigorous basis and at the same time avoid arithmetic complications that would obscure the concepts involved, a one dimensional distribution of quality will be assumed. The quality, Q, in suitable units is expressed as a function of the distance, x, along some straight traverse of the deposit. Such a function