Multi-objective multi-disciplinary design optimisation: from benchmark development to real-world applications in public health

Large and complex systems with multiple objectives and many constraints, components, and design variables requiring optimization are more prevalent now than ever before. Optimization of such complex systems is often approached as a single, large problem. However, the natural partitions of these systems, often associated with decision-making units, should be considered in research, industry, and organisations alike. Partitioning these systems into disciplinary sub-systems is a potential approach for the handling of the overarching optimization problem. One way that these problems are formulated is using a method called multidisciplinary design optimization (MDO). While a number of recent efforts have focused on MDO solutions, it is difficult to compare the effectiveness of these methods in the absence of a benchmark problem. In this thesis, first a scalable standardised MDO test problem is developed, based on a popular multi-objective optimization problem - the ZDT test suite. A multi-objective MDO framework is used to solve this problem, and then the results are compared with the original non-MDO test set. The problem is then expanded to a full test set, and the system topologies are varied. This benchmarking test set is then evaluated using a different MDO architecture. Following this, the multi-objective MDO framework is used to solve a case study from the public health domain. It is found in general that increasing the number of design variables and disciplines leads to slower convergence. In the scalable benchmark problems, when compared to the non-MDO problems, the solutions have a smaller hypervolume, meaning they are further away from the Pareto front. The main limitations in this research stem from the use of the ZDT test suite for the development of the multi-objective MDO problems, due to features such as separability and lack of scalability in objectives.