Abstract
Global sensitivity analysis of mathematical model outputs is recommended for the systematic and holistic quantification of the effects of uncertain model inputs on model outputs. In the field of global sensitivity analysis, a variance-based method called constrained global sensitivity analysis (cGSA) has been introduced for wider classes of models with inequality constraints. The approach of cGSA is highly applicable to chemical and process engineering, where constraints play a pivotal role in the simulation of physicochemical processes as well as the design, control and optimisation of systems subject to key quality, safety, sustainability or economic indicators. However, the computational requirements of cGSA are significantly high and for this reason the wide adoption and development of cGSA applications may be hindered.
The current thesis elaborates on the computational framework and implementation of cGSA and contributes to the development of computational approaches to sampling constrained subspaces, such as process design spaces, and metamodelling techniques that reduce the computational requirements of cGSA. The application of cGSA is demonstrated in the context of Quality-by-Design in biopharmaceuticals manufacturing and the benefits as well as the computational challenges of cGSA are identified and compared with conventional global sensitivity analysis without constraints. The results of the thesis aid the advancement of methods towards computationally efficient cGSA by introducing a novel adaptive sampling method, design space modelling techniques, and metamodelling approaches based on an algorithm of adaptive Gaussian random fields.