In the manufacture of high-value chemical products, such as pharmaceuticals, separation and purification from complex, multicomponent mixtures are crucial steps. Crystallisation, often accompanied by other separations steps (extraction, distillation) is nearly always deployed, due to its separation effectiveness and extreme flexibility. These possibilities provide a multitude of choices for selecting the optimal design solution, in terms of product quality and process efficiency, as well as waste production, material and energy consumption. However, the optimal process design requires accurate and reliable thermodynamic data for multicomponent mixtures in the system over a wide range conditions (e.g., solvents, impurity concentrations, temperatures, pressures) which are often missing, especially for new compounds. These data can be obtained by performing a multitude of measurements across a broad range of conditions; however, this is time consuming, expensive and requires significant quantities of material which may not be available (e.g., in the early stages of drug discovery). Thermodynamic models can be used to estimate these data; however, they have limited accuracy and reliability, especially for compounds and process conditions not previously encountered. In principle, some of these models can be successively improved by incorporating additional experimental measurements. Each measurement has an information value, but it also has an associated cost related to the energy, effort, time, and materials required to perform the measurement. Certain measurements are likely to be more valuable, resulting in more reliable thermodynamic predictions. However, it is not clear what sequence of which experiments would yield most valuable thermodynamic information. The ability to balance this trade-off and follow an optimal sequence of experiments would be transformative for our ability to rationally design efficient industrial separation processes. The aim of the proposed work is to develop an intelligent decision system, implemented as a software application tool, that will specify the precise sequence of experiments that will most efficiently provide the information to produce an accurate thermodynamic model of pharmaceutical compounds in complex, multicomponent mixtures. The intelligent decision system will be adaptable, so users can define what is meant by optimal: minimizing economic cost, time, complexity/availability of experimental technique(s) or required amount of materials.
The proposed work will use machine learning to develop an automated methodology to efficiently construct accurate mathematical representations of the thermodynamics of complex multicomponent mixtures. This will provide engineers and scientists a novel intelligent decision system to determine the optimal experimental measurements that will most effectively improve the thermodynamic model and yield most accurate predictions. It will radically improve our ability to rapidly and efficiently construct accurate thermodynamic models, which is key to the development of any chemical manufacturing process. Although the proposed project is focused on pharmaceutical separation processes, it will have direct relevance to other industries, such as fine chemicals production, biorefining and bioprocessing. It has a strong potential to be transformative in enabling rational design of efficient separation processes across these industries and we will make it widely accessible to the user community through a user friendly app based interface.
In addition to undertaking cutting edge research, students are also registered for the Postgraduate Certificate in Researcher Development (PGCert), which is a supplementary qualification that develops a student’s skills, networks and career prospects.
Information about the host department can be found by visiting: