The Use of Artificial Neural Networks for the Delineation of Boundaries Between Ore Bodies Based on Geophysical Logging Data

Abstract

An artificial neural network model is applied to the delineation of boundaries between ore bodies based on borehole geophysical logging data, such as gamma-ray, density, neutron and resistivity in a Swedish underground mine, the Renstrom Mine. The input data set includes the four geophysical logging parameters and the output data to be predicted by the neural network consist of three rock classes, waste rock (rock class 1), semi-ore (rock class 2) and ore (rock class 3). Three boreholes, B33, B34 and B36, each of a length of approximately 40 m, were analyzed. Borehole B33 was divided into 20 sections based on the core log results. The four geophysical logging parameters, initially measured at 0.1 m intervals, were averaged in each section to form the training set for the neural network. The original data from boreholes B33, B34 and B36 were used as the test set. The optimum configuration of the neural network is a 4-layer neural network with 4 neurons in the input layer, 15 neurons in the first hidden layer, 5 neurons in the second hidden layer and 3 neurons in the output layer. The minimum error rate, 0.2160, in the test set, was obtained from training the network over 29,500 epochs. The ability of the neural network to delineate the boundaries between ore bodies from the geophysical logging data is encouraging, and this technique represents a great advantage compared to diamond core drilling and a qualitative and subjective judgement by a geologist in identifying the boundaries between ore bodies from geophysical logging data.