489 Figure 7—Starting centroid and simplex in SSDEVOP. The coordinates of this ‘worst’ run are then reflected across the simplex, to generate the setpoints of the fourth run (Figure). Figure 8 —Generation of second simplex in SSDEVOP. This workflow repeats, with the ‘worst’ run of the second simplex identified. These coordinates are then reflected across the second simplex to generate the set points for the 5th run. These results are then assessed against the remaining vertices of this new, third simplex (not shown). The approach is therefore to continuously select conditions that are, in a sense, ‘the opposite of bad’ which, over time, further encourage conditions that drive the process response towards a desired optimum. The data of Chauhan (2014) was used to illustrate how an SSDEVOP experiment would unfold in practice. The P80 of a lab mill discharge was measured after different combinations of grind time and mill speed. As the experiment was originally undertaken using a CCRD structure, generating an associated predictive model of P80, this was used to simulate hypothetical SSDEVOP runs (which in a real scenario, Variable 2 → Variable 1 → ¬'Worst' run Variable 2 → Variable 1 → ¬Set points for 4th run
RkJQdWJsaXNoZXIy MTM0Mzk2