Approximation Model Types
You create an approximation using the Approximation Trainer adapter or Results Analytics. Results Analytics allows you to export the approximation to a coefficients file, a Functional Mock-up Unit
(FMU), or a 3DEXPERIENCE
approximation object. See Approximations in the Results Analytics User Guide for more information.
The Approximation adapter supports each of the formats created by Results Analytics, as described below:
- Coefficients file
-
A coefficients file is a text file containing the data points used to build the
approximation and the coefficients of the approximation equation. Results Analytics can use either of the following techniques to create an approximation that can be
exported as a coefficients file:
- Radial Basis Function (RBF), a neural network-based approximation technique
that includes radial and elliptical basis functions.
- Response Surface Model (RSM), a polynomial-based approximation technique.
When you export the approximation to a coefficients file, Results Analytics downloads the file to a local text file. You must use the Approximation Editor to
upload the file to a 3DEXPERIENCE document to make it available to the Approximation adapter.
See Approximation Techniques in the Results Analytics User Guide for more information.
- Functional Mock-up Unit (FMU) file
-
An FMU file is a single file in compressed format containing the interface
definition of the approximation model together with the compiled implementation. For
more information, see Functional Mock-up Interface (FMI) in the Dymola Behavior Modeling User's Guide.
When you export the approximation to an FMU file, Results Analytics downloads the file to a local .fmu file. You must use the
Approximation Editor to upload the file to a 3DEXPERIENCE document to make it available to the Approximation adapter.
All the input and output FMU parameters are set using the activity input
parameters. All of the output parameters are exported to the output activity
parameters. For an FMU to participate in this type of process, the FMU must use the
model-exchange format. Using an FMU as part of a
co-simulation analysis with other processes is not supported.
- 3DEXPERIENCE Approximation object
-
An approximation object is a mechanism that implements combinations of the RBF,
RSM, and FMU techniques. For example, the approximation can select between the
techniques depending on the value of an input parameter.
The Approximation Trainer adapter creates a 3DEXPERIENCE Approximation object.
When you export the approximation to an approximation object, Results Analytics saves it as a 3DEXPERIENCE object from which it is available to the Approximation adapter.
Approximations and Parameters
An approximation object can be thought of as a black box that uses a set of input
parameters to calculate a set of output parameters.
The approximation object's input and output parameters are called model parameters; Optimization
Process Composer parameters are called activity parameters. You use the Approximation adapter to map a
model parameter to an activity parameter. Mapping the parameters allows you to provide input
to the approximation model and to access the output from the approximation model.
You can map model parameters to only activity parameters defined as real scalars. You
cannot map a model parameter defined as an input to an activity parameter defined as an
output. Similarly, you cannot map a model parameter defined as output to an activity
parameter defined as input.