Model-Based Optimisation

The use of computational modelling in engineering has gathered momentum over the last three decades, driven mainly by the massive advance in computer power. Industry is making use of computational models to increase the speed of its technical development, which is an important factor in the overall competitiveness of a company. Environmental considerations, such as global warming and pollution reduction, demand constant development of new products to satisfy constraints imposed by the market and the government.

There are several important aspects of this process which require careful attention

  • Comprehensive access to experimental and model data
    • Advanced data models based upon XML and RDF, which can be employed to store experimental and model data in a standardised structure. Adopting these standards ensures reliable access to the data for ongoing model parameter optimisation against arbitrarily large databases.
  • Intelligent model parameter estimation
    • Due to limited computational resources or lack of understanding of the underlying physical processes, computational models are built upon assumptions, whilst not ideal this introduces model parameters. This aspect must be embraced as a critical step in the model development process as the methods employed for the optimisation of these parameters with respect to data (experimental or model) have serious implications for the robustness of the model when adopted later.
  • Inclusion of experimental and parametric uncertainty
    • We also have advanced algorithms that propagate the uncertainty in model parameters, through to the results. This enables engineers to have a measure of confidence in the robustness of their model.
  • Engineering process or design optimisation
    • Once the model is properly validated and engineers have enough confidence in its application, control inputs can be optimised to satisfy multiple engineering constraints whilst (e.g.) maximising efficiency or reducing emissions etc.

A user story outlines how this methodology was carried out for application for developing advanced soot and combustion models for IC Engines. A second example was applied to tackle similar challenges in modelling granulation processes.

How Can We Help?

  • Data standardisation – storage and communication of experiment and model data
  • Advanced optimisation routines – accounting for experimental and parametric uncertainty
  • Model response surfaces, reduced complexity models
  • Bayesian approaches
  • Advanced parameter estimation
  • Engineering process or design optimisation
  • Multi-objective, multi-response optimisation
  • Model discrimination – quantify usefulness of experiments for identifying model parameters