Run-to-Run Optimization

Run-to-run optimization takes advantage of the repetitive nature of discontinuous operations. Measurements from previous runs are used to improve the current run, the objective being to get to the optimum over a small number of runs. NCO tracking provides a systematic way of performing these adjustments, which are often performed in a heuristic manner in practice. The input profiles are parameterized and the input parameters specified before the transient operation (run), while the NCO elements are measured at run end. Hence, run-to-run optimization can be reformulated as a static optimization problem, for which the dynamics get lumped in a nonlinear static map between the input parameters and the NCO elements. From a methodological viewpoint, this feature facilitates the analysis of convergence, while from a practical viewpoint it leads to run-to-run control strategies that are relatively straightforward to implement and therefore well accepted by practitioners.
Run-to-run optimization via NCO tracking is capable of rejecting the effect of (i) structural plant-model mismatch and parametric uncertainty, provided that the uncertainty does not affect the model of the solution, and (ii) process disturbances that persist over several runs. We have applied run-to-run optimization to several simulated and experimental processes, including industrial ones, in the chemical and machine-tool industries. It is also particularly well suited for scaling-up reaction processes, for which the « optimal recipe » determined at the lab scale can be progressively modified to be optimal at the industrial scale.