Add a Design of Experiments node to evaluate outputs with the Latin hypercube or Morris sampling methods. The evaluation is done by solving the COMSOL model for certain parameter values. The values for these parameters are generated with these sampling methods. The outputs are any scalar quantity that can be evaluated for the underlying study. Click the
Run button (
) to run the design of experiments.
When a Design of Experiment node is used by UQ, you will only see a disabled setting,
Defined by Uncertainty Quantification job, which shows which node uses this node and controls all its settings. When this is the case, a
Disconnect from UQ Job and Edit button is available. By clicking this button, you can take over the control of the node.
The main setting at the top of the section is the Sample method list. Choose between
Latin hypercube (the default),
Morris, and
Specified. For the
Latin hypercube method, you can use the
Number of model evaluations to set how many model evaluations you want to do in total. A global optimization method is used to find a Latin hypercube of good quality. Use
Maximum number of iterations for LHS to specify how many steps to allow for the global optimizer for each initial configuration. Use
Number of restart points for LHS to prescribe how many initial configurations to optimize. The best one of these will be used for the sampling. See the section
Data Sampling — Latin Hypercube Sampling for more information.
From the Random seed type list, choose
Automatic (the default) to generate a random seed automatically (the random seed used is displayed), choose
Manual to enter a seed in the
Random seed field, or choose
Current computer time to use that time as the random seed. The
Automatic method adds 1 to the seed for each run. This method is useful if you want reproducible results, but not identical sampling for repeated runs. The
Manual method gives the same sampling each time you run with the same random seed. The
Current computer time method typically never gives the same sampling when you run repeatedly.
Choose one of the Output data methods:
Recompute or
Compute and append. The
Recompute method clears the
Design table in the
Output table group and adds the sample parameter values and the corresponding output variable evaluations to this table. The
Compute and append method appends new sample parameter values and output variable evaluations to the existing data.
See the Error section for Parametric Sweep (Job Configuration).
See the Cluster section for Parametric Sweep (Job Configuration).
See the Log section for Parametric Sweep (Job Configuration).
) and select Solver and Job Configurations in the Show More Options dialog.