Grinding media charge prediction of wet ball mills for optimal power draw using artificial neural networks

Abstract

Undercharging of grinding media leads to grinding inefficiencies. However, overcharging on the other hand increases the overall mill weight thereby increasing the torque of the mill motor accordingly. Counterbalancing the increase in load torque by mill motor is achieved by drawing more current to produce enough power capable of overcoming the load. More power used means increase in cost. Studies reported in the literature have established that both undercharging and overcharging are undesirable since they contribute negatively to the overall economics of the grinding process. In this research, a non-linear ANN-based predictive model of optimal grinding media charge of the ball mill with consideration of minimum power draw was developed. Operational data of the ball mill involving nine variables were collected over a period of time from a gold mine in the Western Region of Ghana and employed in developing four predictive models. Further, the best performing model was optimised using Grey Wolf Optimisation (GWO) algorithm and Optimal Charging Practices (OCPs) datasets for the purpose of predicting 60 mm grinding media balls. Finally, investigations on how sensitive power draw and grinding media charge were to changes in selected input variables namely, throughput, ore hardness and grinding media wear rate were conducted. Both single and multiple variables analyses were performed in an attempt to find out the possibility of grinding with minimal power draw and minimal grinding media charge while maximising throughput. The analysed scenarios and cases revealed that, it is desirable to grind at 80.0% passing 106 m and that minimisation of 60 mm grinding media charge is achievable at the expense of the mill power draw.