Title: Predicting shear strengths of mine waste rock dumps and rock fill dams using artificial neural networks
Authors: Rennie Kaunda
Addresses: Mining Engineering Department, Colorado School of Mines, 1500 Illinois Street, Golden, CO 80401, USA
Abstract: Three new back-propagation artificial neural networks (ANNs) for predicting the shear strength envelopes of mine waste rock dumps and rock fill dams are presented. Fourteen key material properties for rock fill characterisation are mustered for model development and evaluation. Using principal component analysis, the initial fourteen parameters are ultimately reduced to six through data compression. The neural net with the fewest parameters (ANNOPTC) is consistent over the widest range of confining stresses for shear strength prediction. The three artificial neural networks also perform better than multiple regression conducted on the same global database. Sensitivity analysis ranks input parameters as normal stress, minimum particle strength, dry unit weight, 10% passing sieve size, the coefficient of curvature of the particle size distribution, and the coefficient of uniformity of the particle size distribution in decreasing order of significance. Excellent agreement is observed between predicted and measured shear strengths for new cases.
Keywords: ANNs; artificial neural networks; mine waste rock; rock dumps; rock fill dams; shear strength prediction; large triaxial tests; direct shear testing; modelling; principal component analysis; PCA; normal stress; minimum particle strength; dry unit weight; passing sieve size; particle size distribution.
DOI: 10.1504/IJMME.2015.070378
International Journal of Mining and Mineral Engineering, 2015 Vol.6 No.2, pp.139 - 171
Received: 07 Mar 2014
Accepted: 21 Dec 2014
Published online: 03 Jul 2015 *