Multi-objective model-based design of experiments of pharmaceutical tableting process

The development of high-fidelity and predictive models is one of the cornerstones of process engineering. Often mathematical models involve several unknown parameters to be identified from experiments. Designing the minimum set of information rich experiments for precise estimate is a critical to reduce the examination costs and help improve model prediction capabilities. In this work, we present a novel Muti-Objective Model-Based Design of Experiments (MOMBDoE). The proposed approach is based on the simultaneous maximizing the D-optimal design of experiments criterion and the estimability potential of the model parameters (Estimability). Global sensitivity analysis is used to build the Fisher Information Matrix (FIM), which allows the maximization or minimization of MOMBDoE criteria. The Pareto optimal solutions which represent the best experimental compromises were ranked using a multicriteria decision aiding method to help identify the best alternatives for experimental validation. To validate the proposed MOMBDoE framework, a tablet lubrication process is used as a case study. Kushner and Moore's model is used to predict the tensile strength and hardness of the tablets as one of the main critical quality attributes in tablet manufacturing.