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, would be measured experimentally). The starting simplex is shown in Figure 56 with corresponding P80s of the mill discharge at each vertex. As the objective was to minimise P80, the vertex with the coarsest corresponding P80 (162 µm in this instance) was identified as the ‘worst’ and reflected across the simplex to generate a fourth vertex. Figure 56—First two SSDEVOP simplexes generated using the data of Chauhan (2014). The milling conditions of this new vertex corresponded to a P80 = 143 µm, and so in the second simplex, the vertex with the coarsest P80 (154 µm) was reflected to generate the conditions of the fifth run (not shown). This workflow was repeated until a point where the simplexes started to ‘spin’ (Figure 57) – an indication that a process has approached an underlying optimum (or minimum in this instance) that is otherwise invisible to the experimenter. 285 290 295 300 305 310 315 320 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Mill speed (rpm) Grind time (s) 150 µm 162 µm (worst result,starting simplex) 154 µm New set-points for 4th vertex
RkJQdWJsaXNoZXIy MTM0Mzk2