Fit Surface Based Optimisation

The workflow for setting up an optimisation that uses a fit surface in HyperStudy is described.

The workflow is the following:
  1. Setup
    Specify the .cfx file, Import variables, define the lower and upper bounds of the design variables and define the output responses.
  2. Design of experiments (DOE)
    Use MELS and set the Number of Runs1.
  3. Testing DOE (optional)
    Add an additional DOE, use Latin HyperCube or Hammersley and set the Number of Runs to between 30 and 50.
  4. Fit surface
    Add a fit response, set the Input matrix to the DOE. If a test DOE was used, add this as a second matrix with the type set to Testing. Use the FAST method, Evaluate Tasks and check the accuracy of the fit surface.
  5. Optimisation
    Define the design goal Objectives and Constraints, change the Evaluate From field to the fit surface for all design responses. Select the optimisation method and Evaluate Tasks. The optimum can be found highlighted in green in the Iteration History table.
  6. Implement the optimum model
    Select the row of design variable values of the optimum in the Iteration History, copy them, add a new DOE, set the Number of Runs to 1 and press Apply. Edit the run matrix, select the design variables row and paste the optimum values here. Run the DOE. The Feko model with the optimum design variable values is created and run.
1 The required number of runs depends on number of design variables, but also how quickly the responses change in the design space. HyperStudy suggests a minimum number of runs, but in most cases this will need to be increased.