ENBIS-11 in Coimbra

4 – 8 September 2011 Abstract submission: 1 January – 25 June 2011

Using Automated Adaptive Experimentation to Achieve Constrained Optimisation

6 September 2011, 11:45 – 12:15


Submitted by
Chris Marley
Chris Marley, Dave Woods, Sue Lewis
University of Southampton, UK
Modern chemical processes are often highly multivariate, in terms of both input and output variables, and there may be complex relationships between the responses and predictors. There is therefore a need to identify robust and optimum operating conditions in terms of both desirable outputs and constraints on, for example, by-products.

Recent advances in technology mean that it is sometimes possible to automate a series of chemical reactions to be performed one at a time, without the need for any reprogramming of machinery or further intervention of a chemist. This is known as “continuous flow mode”, and enables these systems to be left running continuously (for instance overnight), hence reducing costs.

To enable such systems to be fully utilised for process optimisation, the reactions to be performed need to chosen adaptively and automatically. We propose a flexible approach for choosing these reactions based on the Expected Improvement criterion, often used for computer experiments. This general framework allows us to adaptively select design points with the goal of optimising an objective function subject to constraints. We present a simple example incorporating the choice of initial design, the adaptive selection of design points, model fitting and sensitivity analysis.

Return to programme