poor decision-making can be effectively quantified and managed. This becomes critical when investment is needed to facilitate the change being investigated, for example, if higher operating expenditures or even significant capital expenditures will be incurred. Three case studies are presented in the following sections to demonstrate the power of this approach in facilitating continuous improvement at a production-scale. QUANTIFYING METAL RECOVERY IMPROVEMENTS WITH A RANDOMISED PAIRED TRIAL The randomised paired trial is perhaps one of the most powerful experimental designs in minerals processing, due to its ability to detect small differences that exist between comparative datasets otherwise subject to large degrees of uncontrolled variability. The experimental design is established so that ultimately, comparisons are undertaken within data pairs that are most ‘similar’ (i.e. as similar as is possible to achieve in a mineral processing context), systematically differing only in the condition that is being tested. In production data, ‘similar’ data pairs are those that are adjacent in time, as these have the highest chance of comprising shifts where ore types, throughputs, operating strategies, etc. are ‘similar’. Therefore, comparisons would only take place between (for example) Monday night-shift and Tuesday day-shift. Comparisons would not take place between Monday night shift and (for example) Friday day shift, as the large time difference would likely result in different circuit, feed and operating conditions that would otherwise exacerbate data variability and further obscure the real effect of the trial. There will, of course, be variability within each data pair, as no two-shifts are ever identical. However, this is overcome in the experimental design by ensuring that the overall length of the trial is sufficient to allow the effect of the tested variable (if any) to be isolated from the inherent variability in the data. Furthermore, the application of the experimental vs. baseline conditions is randomised within each pair, to guard against any underlying time-based trends that may exist in the production data that would bias the outcome of the trial. For instance, in JKTech’s experience, it is common for recoveries during night-shift to systematically outperform those achieved during day-shift. To eliminate this source of bias, the trial condition is tested during both night and dayshifts, with the application changing randomly within a given pair (Figure 2). Figure 49 – Randomised paired trial design where a single ‘time-block’ corresponds to one day. In a recent JKTech study, the effect of a flash flotation cell treating cyclone underflow material in a grinding circuit was tested. The production data in Figure 50 is identical to that shown previously in Figure 48, except with baseline (flash flotation OFF) and trial (flash flotation ON) periods also indicated. Baseline Trial Trial Baseline Trial Baseline Baseline Trial etc. Day 1 Day 2 Day 3 Day 4 Time
RkJQdWJsaXNoZXIy MTM0Mzk2