Application at Ok Tedi Mining of a Neural Network Model within the Expert System for Sag Mill Control

An expert system applied to the control of mineral process unit operations is, by its nature, the site best operating practice, having been designed by the local experts in conjunction with consulting metallurgists. The “common sense” layer in a MET (MinnovEX Expert Technology) solution determines an appropriate action for the current operating conditions through a developed heuristic model of the process (expert system). The magnitude of the setpoint change that is made is proportional to the confidence that the system has in its interpretation of the current conditions (fuzzy logic). The solution is always striving to take the process to a constraint and thereby optimise the operation. An accepted limitation of the solution is that the only way the system can determine if a constraint has been reached is to exceed it and then withdraw – essentially “experimenting” with the process. By leveraging a model, which can accurately “simulate” the process, this restriction can be circumvented by pretesting the set point change through a “what if” analysis and then modifying the solution as appropriate. At Ok Tedi, the introduction in 1999 of a MET expert system for SAG milling resulted in significant improvement in circuit throughput. Yet, with the additional inclusion of an accurate model of the process there was still more that could be achieved. Hence, in 2002 a MET Neural Network model was developed and deployed to enhance the existing expert system solution. The principal objective of this model was predicting the effect that changes in SAG tonnage and density would have on the power draw of the mill essentially a power predictor. The results have seen a decrease in process variance with an associated increase in average circuit performance. This paper describes the methodology by which the Neural Network model was implemented, the considerations and limitations when applying this technology and ultimately the results that were achieved.. SGS MINERALS SERVICES TECHNICAL PAPER 2003-18