Algorithms for Self-Optimising Chemical Platforms

The appreciable interest in machine learning has stimulated the development of self-optimising chemical platforms. The power of harnessing computer aided design, coupled with the desire for improved process sustainability and economics, has led to self-optimising systems being applied to the optimisation of reaction screening and chemical synthesis. The algorithms used in these systems have largely been limited to a select few, with little focus paid to the development of optimisation algorithms specifically for chemical systems. The expanding digitisation of the process development pipeline necessitates the further development of algorithms to tackle the diverse array of chemistries and systems .Improvements and expansion to the available algorithmic portfolio will enable the wider adoption of automated optimisation systems, with novel algorithms required to match the previously unmet domain specific demands and improve upon classical designed experiment procedures which may offer a reduction in optimisation efficiency. The work in this thesis looks to develop novel approaches, targeting areas currently lacking or under developed in automated chemical system optimisations. This includes development and application of hybrid approaches looking at improving the robustness of optimisation and increasing the users understanding of the optimum region, as well as expanding multi-objective algorithms to the mixed variable domain, enabling the wider application of efficient optimisation and data acquisition methodologies.