Development of artificially intelligent reactors for precision polymer synthesis

The digitisation of chemistry has been a rapidly evolving field of research both in academia and
industry in the last 20 years. Miniaturisation of equipment and increased computational power
have aided in the evolution of the way chemists conduct and optimise their reactions. Pharmaceutical drugs and small-molecule industries have taken precedence when it comes to artificially
intelligent reactor platforms. Flow chemistry has enabled the development of user-independent
and modular platforms with interchangeable components. In polymer science, there has been
a delay due to challenges associated with the accuracy of liquid handlers, the increasing viscosity, and the use of specialised characterisation equipment. In precision polymer synthesis,
low molar mass dispersity and high conversions are targeted; however, this is not trivial in
reversible deactivation radical polymerisation (RDRP); a trade-off in these objectives complicates optimisation. Prior to the work in this thesis, multi-objective optimisation algorithms
had been used in a droplet flow reactor and a batch reactor; however, a completely operator independent platform had not been reported. A mechanistic understanding of polymerisation
techniques can assist polymer chemists in predicting the reaction space of interest. Current
models require the use of software or programming abilities which are not always accessible.
Efforts towards explicit quantitative equations for conversion exist for most RDRPs, including
reversible addition-fragmentation chain transfer (RAFT). Full predictive equations for dispersity, that also account for termination, exist for atom transfer radical polymerisation (ATRP),
nitroxide-mediated polymerisation (NMP), and cationic polymerisation, but this does not exist
yet for RAFT. The aim of this thesis is to bridge the gap between polymer synthesis, kinetics, and
autonomous self-optimisation. Bayesian optimisation will be used to facilitate high-throughput
experimentation and will be applied to RAFT polymerisation. Furthermore, extensive reactor
design will be applied that will lead to a platform capable of exploring a wider reaction space.
A conversion model will be coupled to a newly completed equation for dispersity as a function
of conversion, allowing for rapid in-silico kinetic modelling of RAFT. Inclusion of the model will
also be used in tandem with the self-optimisation platform to direct reaction space exploration.