APPENDIX


Information Sources

  1. M. Hagen, H. Demuth, M.Beale, Neural Network Design, Thompson Learning, 2002 edition, chapters 1 - 11, ISBN 981-240-376-0
  2. R.P. King, Modeling and Simulation of Mineral Processing Systems , Butterworth Heinemann,2001, chapters 2 and 4, ISBN 0 7506 4884 8
  3. D. Sbarbaro, J. Barriga, H. Valenzuela, G. Cortez L. Mujica, N. Perez, A Comparison of Neural Networks Architectures for Particle Size Distribution Estimation in Wet Grinding Circuits, ISA Mining and Metals Industries Division Newsletter, Summer 2002, pages 11-19
  4. Krebs Engineering, Design and Operating Factors Affecting or Influencing the Performance of Hydrocyclones, undated company report
  5. E. Hill, Grind Size vs. Gold Recovery, report from Fairbanks Gold Mining, Inc., Fort Knox Mine

  7. R.A. Arterburn, The Sizing of Hydrocyclones, Krebs Engineers, Oct. 1976, pp. 1-16
  8. Y.G. Du, R. del Villar, J. Thibault, Neural Net-Based Softsensor for Dynamic Particle Size Estimation in Grinding Circuits,Int. J. Miner Process.52 (1997) pp. 121-135
  9. F. D. Foresee, M.Hagen,Gauss-Newton Approximation to Bayesian Learning , Proceedings of the 19977 IJCNN, pp.1930-1935, 1997
  10. Fairbanks Goldmining,Inc., Fort Knox Gold Mine, site visit and mill photographs


biological neuron                working cyclone
                                      Neural Network Design                                                                                                                                    Fairbanks Goldmining, Inc.





analyzer         High maintenance and high cost onstream
particle size analyzer can be replaced by a runtime computer
generated model of hydrocyclone performance?  The two studies conclude a neural
network can successfully estimate particle size distribution but the question is how long will this model remain accurate?  The network is nothing more than a combination of matrix algebra statements which a computer will faithfully execute.  However, like any instrument, the virtual analyzer will need to be re-calibrated when the milling process changes - i.e. a new type or ore - and the hardware version will be required to supply the calibration data.
                                                                                                                      











Fairbanks Goldmining, Inc.

Cyclone Physics         Main Page         Neural Networks        The Model        Measurements