Continuous crystallisation with in-line filtration: enabling multi-step synthesis and accelerating process optimisation

As continuous chemistry systems are becoming more widely adopted, there is a push towards more complex, multi-stage systems. This allows for more complex products, with the potential for improved safety, efficiency and sustainability. This chaining of unit operations comes with physical and chemical challenges that must be addressed. Often, this is done using in-line purification methods, with these typically utilising liquid-liquid multi-phasic techniques. In this thesis, a solid-liquid multi-phase platform is developed utilising continuous crystallisation, in-line filtration and controlled dissolution to perform the necessary treatments of the process stream. Digital and artificially intelligent methods are incorporated to accelerate process development and improve performance.

The design of the platform is discussed in detail, with evidence and justification for the equipment, materials and methods used. The platform consists of a miniature CSTR cascade for continuous antisolvent crystallisation, followed by an switching valve array. This allows for the diversion of slurry flow onto one of two custom in-line filters. Whilst one filter collects crystallised material, another has a clean solvent stream passed over in order to dissolve the material into a separate stream. This parallel configuration allows for continuous operation, with the desired species isolated from the process stream.

It is demonstrated how the operating parameters can be controlled to attain the desired conditions. The solution concentration, antisolvent content and residence time are varied to control the crystallisation outcome. Purification is achieved in this way, by the selective crystallisation of a multicomponent solution. The filtration time and solvent flowrate are varied to control the dissolution time of the crystals, such that parallel operation can be performed without material accumulation. By dissolving crystals in a different solvent at suitable conditions, it is shown how continuous solvent switching can be performed alongside selection of the outlet stream concentration and flowrate.

Automated pump control and switching valve automation are used, along with in-line UV-vis spectroscopy and online HPLC analysis. Combined, this allows for automated testing of multiple experimental conditions and allows for adoption of self-optimisation technologies. Bayesian optimisation is introduced to optimise the discovery of relevant process conditions in both simulated and physical experiments.

With blockages identified as a bottleneck for process development, new techniques are developed with the aim of simultaneously optimising important process measures, such as yield and purity, and avoiding conditions that cause blockages. Digitally driven blockage-avoidant approaches are tested, including single- and multi-objective Bayesian optimisation, Transfer Learning and dynamic optimisation. Statistical techniques such as logistic regression are used to perform classification tasks and build further understanding of the system with regard to blockages and the range of operable conditions that can be employed.

This research shows that the challenges identified in the literature can be overcome in a single platform if crystallisation is possible. With appropriate solid-handling, including the ability to clear blockages automatically using a flow reversal back-flush arrangement, the inclusion of solids can be welcomed, rather than avoided. By doing this, complex multi-stage systems are enabled, and the applicability of continuous chemistry is increased.