
Real-Time Interface – Blocksets and Toolboxes Support 5
Relevance:
MATLAB Releases: R14 to R2010b
dSPACE Platforms: all
dSPACE Releases: 4.1 to 7.0
RTI single-processor and RTI-MP
Problem Description:
For most Fuzzy Inference Systems (FIS), the Fuzzy Logic controller block uses the Fuzzy
Wizard to automatically generate a hierarchical block diagram representation. As only standard
Simulink blocks are used, compact and efficient code generation with the Real-Time Workshop
is possible.
Sometimes the Fuzzy Wizard cannot create a hierarchical block diagram from the FIS (e.g., if
custom membership functions are used). In these cases the Fuzzy Logic Controller Block uses
an S-function to simulate the FIS and the block performs a lot of time-consuming computations.
For this reason the block rapidly imposes a fixed-step size that is very high. For more details,
refer to 'The Fuzzy Logic Controller Block' in the Fuzzy Logic Toolbox User's Guide from The
MathWorks.
3.4 Neural Network Toolbox
3.4.1 General Limitations
Relevance:
MATLAB Releases: MATLAB 5.2.x to R2010b
dSPACE Platforms: all
dSPACE Releases: dSPACE CD 1.3 to Release 7.0
RTI and RTI-MP
Problem Description:
Some of the blocks provided by the Neural Network toolbox are not supported by Real-Time
Workshop. Hence they cannot be used with RTI.
Known unsupported blocks are:
compet block (uses MATLAB Fcn block, up to R2010a)
NN Predictive Controller block (uses M-file S-function)
3.4.2 Run-Time Issues
Weights for neural networks
Relevance:
MATLAB Releases: R11.x to R2010b
dSPACE Platforms: all
dSPACE Releases: dSPACE CD 2.1 to Release 7.0
RTI and RTI-MP
Problem Description:
The weights for neural networks that were modeled in Simulink must be determined prior to the
build process. It is not possible to directly train neural networks in real time. If training of a
network is required in conjunction with a real-time program, a possible strategy could be: upload
Commentaires sur ces manuels