Hydrocyclones and Neural Networks
  Biological computing aids mining industry with real time estimate of particle size distribution
The physics are not clear, but the computer does not care.



Mineral processing is largely about taking big pieces of blasted rock and crushing and grinding them into fine powder. Only after the host rock is pulverized can the desired mineral particles be separated. Grinding mills are expensive to operate and maintian, so it is undesirable to grind the ore finer than required for efficient  mineral extraction. A common method to continously and automatically separate the material that is fine enough from that requiring more grinding is to pump the rock and water slurry through one or many hydrocyclones. A hydrocyclone is a particle size classifier which has no internal moving parts yet using centrifigal fields and vortices is able to split the feed entering the unit into an overflow stream of small light particles and an underflow stream of larger, heavy particles. The overflow stream leaves the grinding circuit for further processing and the grinding mill operator wants a continuous measurement of the finished particle size - in the range of  150-170 microns. Seven thousand metric tons per hour of rock may pass through a bank of parallel operating cyclones, yet the mining industry lacks a reliable online method of measuring the size of particles exiting the overflow.


                                                                                                              hydrocyclone


Hydrocyclones are used throughout the mining industry to classify
particle sizes





                                                                    Simple schematic of a closed loop continuous milling                                                                          circuit. The ore is mixed with water as                                                                                               it enters the grinding mill. The ground rock slurry is                                                                             pumped through the hydrocyclone where the finer                                                                                 particles are removed and the coarser particles                                                                                        returned to the mill.

                                                                                                                                            Drawing by Keith Palchikoff
circuit schematic


                                                            
photos from Fort Knox Gold Mine




      Trying to Sample Mud is Not Easy



analyzer

This is an onstream particle size
analyzer which uses an onboard
computer to sequence slurry samples
through automated calipers. The slurry passes between the calipers which open and close hundreds of times per minute. After two minutes, the average of the measurements are reported as a % Passing 100 mess. This is the percentage of particles that would pass through a screen mesh with 100 holes per inch - this is a common US mining particle size measurement.

The machine has been out of service for two years because of ongoing high maintenance and constant repairs - trying to sample mud is not easy . It is estimated to cost $100,000 per year to maintain a particle analyzer.







Enter the Soft Sensor solution.            A neural network based model of hydrocyclone performance.

                                          biological neuron
A Soft Sensor is a software based virtual analyzer which uses secondary physical measurements of key state variables as inputs to a runtime process model and produces an inferred measurement of the desired variable. No more expensive hardware based analyzer to unplug. All that is
required is a reliable and practical model of cyclone performance.  Unfortunately, the physics of
particle behavior in a cyclone are not easily explained and is an area of active scientific research.

The internal forces in a cyclone may be difficult or impractical to accurately model based on first principles and solutions to partial differential equations. However, a numerical modelling technique
based on the way the human brain learns complicated relationships, has been successfully used to model and predict the particle size produced by an operating cyclone.






                                                                                                                                                                Drawing from Neural Network Design

Cyclone  Physics             Neural  Networks         Measurements         The Model     Appendix