Process and Product Characterisation
Driven by industrial needs, Newcastle is developing new approaches to integrate fundamental process knowledge into empirical approaches. Exemplar projects have included:
- In the area of genetic programming, dynamic models using multi-objective optimisation have been developed - this technology has been adopted by Dow (USA) replacing their use of neural networks for inferential estimation.
- Hybrid models based on the integration of a neural network and mechanistic model have been developed - this fundamental research has been rolled-out on a polyethylene process.
Newcastle is also undertaking fundamental theoretical research under this theme, with the focus being on bringing new concepts from other areas, such as statistics, to realise process and product characterisation. Techniques have included:
- particle filters for on-line robust state and parameter estimation;
- Gaussian processes for automatic variable selection and confidence limits realising enhanced process modelling;
- the development of a dependency measure to select the operating regime which yields desirable performance at a smaller scale but which provides comparable behaviour on process scale-up.