The Field data models menu option creates a field meta model of optimal prognosis (field-MOP) based on previously imported field response and input parameter samples. Only active samples are considered. The field meta model is stored internally.
Note: You can view the current list of created field meta models in the Field data models window. For more information, see Field Data Models Window.
The Create Field Metamodel of Optimal Prognosis (Field-MOP) window has two tabs: Select training data and Settings.
On the Select training data tab, you select the inputs and outputs for the field-MOP from the list of field and scalar quantities. oSP3D does not allow you to drag and drop a quantity to a location where it is not supported. Inputs that you drop into the Input (mandatory) location are included in the field-MOP, even if their sensitivity indicates a negligible influence on the field-MOP.
On the Settings tab, you specify the settings to use. Descriptions of all possible settings follow. However, as you specify a setting, the visibility and available of other settings can change.
Settings
- Maximum number of shapes
The upper bound for the number of scatter shapes to be computed in the internally used random field model. The default is 10. This option is used to limit hardware resource usage.
- Use a single cross-correlated model, ident
Clear this check box to create an individual field meta model for each selected quantity. Select this check box to collect all selected quantities into a single field meta model. In the text box on to the right, specify the identifier for the new data object to create. The number of active designs must be equal for all selected quantities.
- Use interpolated missing items
Clear this check box to use the mean value for missing items or eroded elements. Select this check box to use the interpolated (filled) value for missing items. If the missing item originated from an incompatible mesh projection, selecting the check box is recommended.
- Compute sensitivities
Available only when you are not using the MOP backend, select this check box to calculate the sensitivity of the field-MOP’s output with respect to each input parameter. A
F-CoP[input parameter]
field data object is created for every relevant or mandatory input parameter.- Compute R*sigma
Select this check box when not using the MOP-backend to generate data objects that contain the F-CoP values in absolute numbers, which is the square root of the F-CoP scaled by the individual standard deviation for each spatial point.
- Default meta model settings
Use default settings, balancing accuracy and performance. oSP3D has built-in checks to improve performance for a large number of designs:
Number of RBF localization points is limited to 200.
Number of reshuffles is limited to 200.
Anisotropic function kernels are disabled if the number of designs is larger than 500.
- Fast mode for meta model creation
Settings are optimized for performance at the expense of accuracy. Check the command log for the settings that are used.
- Use custom settings of internal meta model
Selecting this option displays the Custom meta model settings area with additional settings. For more information, see Custom Meta Model Settings.
- Use custom settings for MOP backend
Selecting this option displays the MOP settings area with the advanced MOP settings that are used in optiSLang. For more information, see MOP Settings
Custom Meta Model Settings
- Use anisotropic function kernels
Select this check box if you want the RBFs scaled with a distance factor individual for each input parameter. This tends to improve the F-CoP at the expense of performance.
- Use custom localization points, number
Number of support points to use for RBF localization. This number must not be greater than the number of designs. You use this option to reduce computation time of the field-MOP.
- Filter input parameters (input-output polynomial correlations)
Select this check box to eliminate input parameters that do not contribute to a polynomial regression model.
- Do not filter inputs with a linear correlation significance level
When filtering, do not exclude input parameters above the specified linear correlation significance level. The default is 0.99.
- Use custom number of reshuffles for sensitivities
Number of reshuffles of test sets being used to estimate the sensitivity indices. A greater number leads to more accuracy.
- Filter input-input correlations, limiting eigenvalue of correlation
Smallest eigenvalue of the input-input correlation matrix being used to identify dependent inputs. For example, use this in robustness sampling or with measured input data.
- Iteration tolerance
Termination tolerance for internal iterative optimization loop (for gradient condition and function decrease).
MOP Settings
From the check boxes on the left, select the models to use to build the MOP:
Use polynomial models
Use Moving Least Squares models
Use Kriging as model
If Kriging: Use anisotropic function kernels
Use Box-Cox transformation
From the check boxes on the right, apply filters:
Use input-output correlation filter
Use input-input correlation filter
Use significance filter
Use CoI filter
Use CoD filter
Output Objects for Each Quantity
For each quantity, these output objects are generated:
Field-MOP model
Mean value vector and standard deviation (
mean[FMOP]
andsigma[FMOP]
) that are usedF-CoP for each input parameter and the accuracy of the whole field meta model (
F-CoP[Total]
).F-CoP values in absolute numbers (
R*sigma
).Log messages:
oSP3D determines the fraction of explainable variation that is obtained for individual numbers of scatter shapes and prints this table in the log messages.
oSP3D prints further intermediate results to the log messages.