Optimisation Workflows
Two optimisation workflows in HyperStudy are mentioned and the fit surface based optimisation is discussed in detail.
- Conventional optimisation
- Fit surface based optimisation
Conventional Optimisation
A simulation is run at each iteration of the optimisation.
Fit Surface Based Optimisation
A design of experiments (DOE) is computed and used to generate a fit surface response, which is used for running the optimisation.
- Multi-goal optimisation where goal weighting needs investigation.
- Multi-goal optimisation where trade-off analysis is needed.
- Whenever it is necessary to compare different optimisation algorithms.
- Where a stochastic analysis is also required (can be run using the same HyperStudy fit surface that is used for the optimisation).
The workflow is made up of three parts:
-
- DOE
-
The design of experiments is the main computational effort during this workflow. Feko is run for each permutation of the design parameters and the corresponding design responses are calculated. It is recommended to use the Modified Extensible Lattice Sequence (MELS) method for this analysis because it is well suited to space filling and can also be extended if more samples are required.
It is also possible to run a secondary DOE, which can be used as a testing matrix for the fit surface. In this case, it is recommended to chose a different method like Latin HyperCube or Hammersley for the validation DOE.
-
- Fit Surface
-
The fit surface response maps the relationship of the design variables to the output responses (computed in the DOE). The result is a continuous description of how the output responses vary with respect to changes in the design variables, and can be used to predict the optimum design.
HyperStudy offers a FAST method which should be used to compute the fit surface. It compares the different methods and parameters and selects the best configuration for each response.
-
- Optimisation Using the Fit Surface
-
Optimisation using the fit surface typically takes a few seconds to run, which is one of the main advantages of this workflow. The computational effort is shifted from the optimisation to the DOE, which makes it possible to run multiple optimisations at a negligible additional computational cost.
HyperStudy offers a variety of different optimisation methods to choose from. The Global Response Search Method (GRSM) or Adaptive Response Search Method (ARSM) is recommended.