The SMArtInt Library is designed to integrate various artificial intelligence (AI) models seamlessly into Modelica-based simulation tools. SMArtInt, short for Simple Modelica Artificial Intelligence Interface, provides a user-friendly interface that bridges advanced AI capabilities with the power of Modelica simulations, enhancing both modeling efficiency and simulation accuracy.
Currently, it supports the following tools
- Dymola
- OpenModelica
with
- TensorFlow models exported as TFLite
- ONNX models.
The repository contains a compiled version of the interface for usage in windows. As a starting point open the Modelica Library. It contains some ready to run examples (SMartInt.Tester) which demonstrate the usage. The corresponding python files which create the TF-Lite and ONNX models are located in Resources\ExampleNeuralNets.
Hints for usage in Dymola: Currently, only a 64-bit version is available. If the variable Advanced.CompileWith64 is set on its default value 0, Dymola will automatically compile a 64-bit Dymosim.exe after giving a remark in the translate log file. In case Advanced.CompileWith64=2 a 64-bit dymosim.exe is created anyway and in case of Advanced.CompileWith64=1 compilation will fail.
SMArtInt uses other software - the source code is included as submodule and/or as compiled version for direct usage_
- Tensorflow (https://github.com/tensorflow/tensorflow)
- Bazel.exe (https://github.com/bazelbuild/bazel)
- ClaRa Delay (https://github.com/xrg-simulation/ClaRaDelay)
- ONNX Runtime (https://github.com/microsoft/onnxruntime)
This work was carried out within the framework of the research project DIZPROVI, supported by the Federal Ministry of Education and Research (number 03WIR0105E).