PythonFMU

A lightweight framework that enables the packaging of Python 3 code or CSV files as co-simulation FMUs (following FMI version 2.0).

How do I build an FMU from python code?

1. Install pythonfmu package:

pip install pythonfmu

2. Create a new class extending the Fmi2Slave class declared in the pythonfmu.fmi2slave module (see below for an example).
3. Run pythonfmu build to create the fmu.

usage: pythonfmu build [-h] -f SCRIPT_FILE [-d DEST] [--doc DOCUMENTATION_FOLDER] [--no-external-tool]
                       [--no-variable-step] [--interpolate-inputs] [--only-one-per-process] [--handle-state]
                       [--serialize-state] [--use-memory-management]
                       [Project files [Project files ...]]

Build an FMU from a Python script.

positional arguments:
  Project files         Additional project files required by the Python script.

optional arguments:
  -h, --help            show this help message and exit
  -f SCRIPT_FILE, --file SCRIPT_FILE
                        Path to the Python script.
  -d DEST, --dest DEST  Where to save the FMU.
  --doc DOCUMENTATION_FOLDER
                        Documentation folder to include in the FMU.
  --no-external-tool    If given, needsExecutionTool=false
  --no-variable-step    If given, canHandleVariableCommunicationStepSize=false
  --interpolate-inputs  If given, canInterpolateInputs=true
  --only-one-per-process
                        If given, canBeInstantiatedOnlyOncePerProcess=true
  --handle-state        If given, canGetAndSetFMUstate=true
  --serialize-state     If given, canSerializeFMUstate=true
  --use-memory-management
                        If given, canNotUseMemoryManagementFunctions=false

How do I build an FMU from python code with third-party dependencies?

Often, Python scripts depends on non-builtin libraries like numpy, scipy, etc. PythonFMU does not package a full environment within the FMU. However you can package a requirements.txt or environment.yml file within your FMU following these steps:

1. Install pythonfmu package: pip install pythonfmu
2. Create a new class extending the Fmi2Slave class declared in the pythonfmu.fmi2slave module (see below for an example).
3. Create a requirements.txt file (to use pip manager) and/or a environment.yml file (to use conda manager) that defines your dependencies.
4. Run pythonfmu build -f myscript.py requirements.txt to create the fmu including the dependencies file.

And using pythonfmu deploy, end users will be able to update their local Python environment. The steps to achieve that:

1. Install pythonfmu package: pip install pythonfmu
2. Be sure to be in the Python environment to be updated. Then execute pythonfmu deploy -f my.fmu

usage: pythonfmu deploy [-h] -f FMU [-e ENVIRONMENT] [{pip,conda}]

Deploy a Python FMU. The command will look in the `resources` folder for one of the following files:
`requirements.txt` or `environment.yml`. If you specify a environment file but no package manager, `conda` will be selected for `.yaml` and `.yml` otherwise `pip` will be used. The tool assume the Python environment in which the FMU should be executed is the current one.

positional arguments:
  {pip,conda}           Python packages manager

optional arguments:
  -h, --help            show this help message and exit
  -f FMU, --file FMU    Path to the Python FMU.
  -e ENVIRONMENT, --env ENVIRONMENT
                        Requirements or environment file.

Example:

Write the script

from pythonfmu import Fmi2Causality, Fmi2Slave, Boolean, Integer, Real, String


class PythonSlave(Fmi2Slave):

    author = "John Doe"
    description = "A simple description"

    def __init__(self, **kwargs):
        super().__init__(**kwargs)

        self.intOut = 1
        self.realOut = 3.0
        self.booleanVariable = True
        self.stringVariable = "Hello World!"
        self.register_variable(Integer("intOut", causality=Fmi2Causality.output))
        self.register_variable(Real("realOut", causality=Fmi2Causality.output))
        self.register_variable(Boolean("booleanVariable", causality=Fmi2Causality.local))
        self.register_variable(String("stringVariable", causality=Fmi2Causality.local))

        # Note:
        # it is also possible to explicitly define getters and setters as lambdas in case the variable is not backed by a Python field.
        # self.register_variable(Real("myReal", causality=Fmi2Causality.output, getter=lambda: self.realOut, setter=lambda v: self.realOut = v))

    def do_step(self, current_time, step_size):
        return True

Create the FMU

pythonfmu build -f pythonslave.py myproject

In this example a python class named PythonSlave that extends Fmi2Slave is declared in a file named pythonslave.py, where myproject is an optional folder containing additional project files required by the python script. Project folders such as this will be recursively copied into the FMU. Multiple project files/folders may be added.

Test it online

Note

PythonFMU does not bundle Python, which makes it a tool coupling solution. This means that you can not expect the generated FMU to work on a different system (The system would need a compatible Python version and libraries). But to ease its usage the wrapper uses the limited Python API, making the pre-built binaries for Linux and Windows 64-bits compatible with any Python 3 environment. If you need to compile the wrapper for a specific configuration, you will need CMake and a C++ compiler. The commands for building the wrapper on Linux and on Windows can be seen in the GitHub workflow.

PythonFMU does not automatically resolve 3rd party dependencies. If your code includes e.g. numpy, the target system also needs to have numpy installed.

Would you rather build FMUs in Java? Check out FMI4j!
Need to distribute your FMUs? FMU-proxy to the rescue!

Publications

Hatledal, Lars Ivar, Frédéric Collonval, and Houxiang Zhang. "Enabling python driven co-simulation models with PythonFMU." Proceedings of the 34th International ECMS-Conference on Modelling and Simulation-ECMS 2020. ECMS European Council for Modelling and Simulation, 2020.

Credits

This works has been possible thanks to the contributions of @markaren from NTNU-IHB and @fcollonval from Safran SA.