targeting certain feed blends, while maintaining high mill power draw set points, as shown in Figure 52b. An unexpected outcome was the effect of tramp metal in the mill load, which was removed by the operations team approximately once per week. This took place right before the commencement of the final block of runs, such that the third block corresponded to a ‘clean’ mill, devoid of tramp metal in its load. This effect is shown in Figure 53, where it can be seen that the removal of tramp metal from the mill led to a strong reduction in fine particle generation. It was inferred from this that the tramp metal was behaving similarly to grinding media in a ball mill, with more severe grinding impacts exacerbating fines generation, and with its effect becoming more severe over time as it accumulated inside the mill load. Figure 53 - Surface plots of AG product -75 µm (%) at a mill power draw of 120 kW during the (a) first time-block; (b) final time-block (reproduced from Vizcarra et al. (2018)). The results of the experimental design were significant for Boyne who were able to devise strategies to better control the product size distribution of the grinding circuit, without the need for capital expenditure. In addition to planning more targeted feed blends and higher mill power draws, the mill is now cleaned of tramp metal twice per week instead of once per week. SIMPLEX SELF-DIRECTING EVOLUTIONARY OPTIMISATION There can sometimes be a reluctance to undertake CCRD experimentation as its design will often prescribe runs that depart from typical process conditions. This entails some risk of a deterioration in circuit performance during the course of the experiment (though it must be understood that: a) this will only be temporary; and b) mapping regions of performance decline is a key step in delineating response surfaces such as those shown in Figure 53). There exists a class of experiments where response optima are identified through smaller perturbations of the process. The response of the system after a given run is then used to determine the process variable set-points in the following run. In this way, the experimental design is not determined a priori, but evolves as the experiment progresses via an approach called evolutionary optimisation. Several types of evolutionary optimisation experiments are outlined in Napier-Munn (b) (a)
RkJQdWJsaXNoZXIy MTM0Mzk2