ITEA 4 page header azure circular

Neural Network library for eFMI export

Project
22013 OpenSCALING
Type
New library
Description

The open source eFMI Modelica library maintained by the Modelica Association Project eFMI (MAP eFMI) provides support models like Modelica Standard Library (MSL) table adapters or GALEC built-in function implementations to ease the application of Modelica in eFMU development contexts. As part of its version 1.0.2 release support for neural networks as pure Modelica equations has been added in the new eFMI.NeuralNetworks package (https://github.com/modelica/efmi-testcases/tree/main/eFMI/NeuralNetworks).

Contact
Christoff Bürger, Dassault Systèmes AB (SWE)
Email
Christoff.BUeRGER@3ds.com
Technical features

The new eFMI.NeuralNetworks can be used to implement trained Neural Networks as pure Modelica code. The application area of eFMI.NeuralNetworks are neural networks intended for embedded use; hence, the library is not suited to represent large models since such anyway are not suited for final embedded deployment.

Integration constraints

eFMI.NeuralNetworks enables eFMI capable Modelica tooling to generate safety-critical, hard real-time domain suited code generation for the neural networks of PeN-ODEs.

Conditions for reuse

See https://github.com/modelica/efmi-testcases/blob/main/LICENSE

Confidentiality
Public
Publication date
04-09-2025
Involved partners
Deutsches Zentrum für Luft- und Raumfahrt (DLR) (DEU)
Dassault Systèmes AB (SWE)