Designing chemical processes with rigorous simulators is accurate but expensive: every evaluation can take minutes. This project explores how cheaper, approximate models can guide the search so that the expensive simulator is only used where it matters most.
- Combine several models of the same process at different levels of accuracy and cost.
- Use cheap models to screen the design space and expensive simulations to confirm promising candidates.
- Let a Bayesian optimiser decide, step by step, which model to query next, balancing information gained against computational cost.
- Respect engineering and safety constraints throughout the search.
A methanol synthesis process modelled in an open-source process simulator.
Python · Gaussian-process surrogates · Bayesian optimisation · process simulation
Research in progress. A journal article is in preparation.
This repository is an overview only. The source code is kept private. If you're interested in the work or a collaboration, feel free to get in touch.