How Coherent are Geological Logs with Geochemical Assays or Metallurgical Tests?

摘要

Geological logs, which record petrophysical attributes such as mineralization, alteration or rock type on core samples, are often qualitative and may be incoherent with chemical analyses and metallurgical tests. Some reasons for such incoherencies are the lack of chemical analyses during logging, the difficulty to estimate mineral percentages, the existence of overprinting processes that lead to complex rock textures, the lack of experience of loggers, or the lack of a unique logging criterion.In this context, different tools and approaches have been developed for validating geological logs, based on geological knowledge, statistics, geostatistics, neural networks or data mining. This work presents an alternative approach based on geostatistical modelling for identifying and reclassifying potentially mislogged samples when quantitative variables are available from geochemical assays or metallurgical tests. The principle is to: (i) define geological or geometallurgical domains for each quantitative variable by grouping log classes; (ii) transform the quantitative variables into normal scores, accounting for the previously defined domains; (iii) model the spatial correlation of the normal scores; (iv) cross validate the data and obtain predictions of the normal variables and the associated variance-covariance matrices of prediction errors for each sample and each class; (v) calculate a measure of likelihood of each class for each sample; and (vi) compare the likelihood of the actual logged class with that of other classes to identify potentially erroneous logs.The methodology is applied to an iron ore deposit, to validate the rock type logged by geologists against the information from measurements of grades and granulometry. The number of suspicious logs can be managed by the practitioner. A sample-by-sample analysis of these suspicious logs is then required to confirm whether or not a sample should be relogged.