The Model and Physics
A recipe for the NN model approach
does not offer explicit new theoretical substance but ...


                                                                                                              
The softsensor described was built and tested at a  large copper concentrator in Chile. A similar project is underway at Fort Knox. The NN approach did create a numerical model that converged to a solution. The computer algorithm derived a mathematical function to describe the relationship between cyclone feed flow, density, new feed and cyclone particle size. However, the model results do not shed any theoretical light on the subject of vortex behavior. The NN "learns physics" but only as a numbers game.

To design a successful network requires a good understanding of the underlying theory of what is being modeled. This understanding will allow for correct selection of the required secondary measurements and proper evaluation of the required variance in the training data.


 



Cyclone  Physics             Neural  Networks         Measurements         Main Page         Appendix