Model-Based Design of Experiments for Processes with Uncertain Inputs

Abstract

A new methodology is proposed for sequential model-based design of experiments (MBDoE) for chemical processes with uncertain inputs. New A-optimal experiments are designed sequentially for a pharmaceutical batch reactor case study involving the addition of an uncertain amount of trimethylamine (TMA) reactant. The batch reactor model has four parameters, which are estimated along with the uncertain concentrations of TMA in each run. The MBDoE problem consists of four, eight and twelve decision variables when one, two or three experiments are sequentially designed, respectively. Monte Carlo simulations show that the proposed method selects experimental conditions that lead to better parameter estimates, on average, than a traditional MBDoE method that ignores input uncertainty. The typical uncertainty of the input TMA concentration was increased by two and five times its usual level to determine the effectiveness of the proposed methodology at much higher uncertainty levels. Results confirm that the proposed methodology becomes even more important at high input uncertainties.