PyDelica: Serverless OpenModelica with Python

About

PyDelica is an API providing a quick and easy to use interface to compile, customise and run OpenModelica models with Python. Unlike OMPython it does not require the launch of a server session to use OMShell but rather dynamically adjusts files produced after model compilation in order to update options and parameters. The lack of server requirement means models can be run in tandem without calling multiple OMC sessions which can be very taxing on CPU. Furthermore PyDelica is able to detect errors during model runs by reading the stderr and stdout from OpenModelica and throw appropriate exceptions to terminate the program execution.

Installation

To use PyDelica you will require an installation of OpenModelica on your system, the API has been confirmed to work on both Windows and Linux, but if OM is installable on macOS it should also be possible to still use it.

To install run:

pip install pydelica

Getting Started

For the purposes of demonstration the included test model tests/models/SineCurrent.mo will be used.

PyDelica Session

All uses of PyDelica require an instance of the Session class:

from pydelica import Session

with Session() as session:
    ...

It is strongly recommended that this class be used via the context manager to ensure cleanup of temporary directories.

Logging

The Session class has one optional argument which is used to set the log level output within OpenModelica itself. The options are based on the -lv flag within OM. By default the level is set to Normal which means no log level output.

As an example if you wanted to run with statistics logging -lv=LOG_STATS you would setup with the following:

from pydelica import Session
from pydelica.logger import OMLogLevel

with Session(log_level=OMLogLevel.STATS) as session:
    ...

See the source for more options here.

Building/Compiling Models

Before you can run a model you must first compile it. This is done using the build_model member function which takes the path to the Modelica source file.

model_path = os.path.join('tests', 'models', 'SineCurrent.mo')
session.build_model(model_path)

If the required model is not top level, that is to say it exists within a module or , we can optionally specify the address within Modelica. This is also required if the required model is not the default. For example say model A existed within module M:

model_path = 'FictionalModelFile.mo'
session.build_model(model_path, 'M.A')

The build_model function also allows you to specify additional flags/options to hand to the OMC compiler, these are given in the form of a dictionary where the value can be None if the flag does not take any input. You can also directly set the profiling level for profiling the Modelica code. When set, the profile dictionary is also stored in the session after the simulation and is accessible via the code_profile and code_info attributes:

session.build_model(
    model_path,
    model_addr='M.A'
    profiling="all",
    omc_build_flags={"-g": "MetaModelica"}
)
session.simulate()
print(session.code_profile)
print(session.code_info)

Specifying Additional Model-Based Dependencies

If additional model files are required to execute the main model these can be specified with the extra_models argument:

session.build_model(
    model_path,
    extra_models=["extra_model.mo"]
)

Using Alternative Inputs Location

If your model inputs are stored in an alternative directory, this can be specified with the update_input_paths_to argument:

session.build_model(
    model_path,
    update_input_paths_to="/path/to/inputs"
)

Examining Parameters and Options

We can examine all parameters for a given model using the get_parameters method which will return a Python dictionary:

session.get_parameters('SineCurrentModel')

if the parameter is unique to a single model then the model name argument can be dropped. Returning the value for a single parameter is as simple as:

session.get_parameter(<parameter-name>)

For simulation options the analogous methods are get_simulation_options and get_simulation_option respectively for general case, for more specific see below.

Setting Parameters and Options

Set a parameter to a different value using the set_parameter function:

session.set_parameter(<parameter-name>, <new-value>)

Further Configuration

The output file type can be specified:

from pydelica.options import OutputFormat
session.set_output_format(OutputFormat.CSV) # Other options are MAT and PLT

Set the solver:

from pydelica.options import Solver
session.set_solver(Solver.DASSL)    # Other options are EULER and RUNGE_KUTTA

Set the time range:

# Each argument is optional
session.set_time_range(start_time=0, stop_time=10, model_name='SineCurrentModel')

Set tolerance:

# Model name is optional
session.set_tolerance(tolerance=1E-9, model_name='SineCurrentModel')

Set variable filter for outputs:

# Model name is optional
session.set_variable_filter(filter_str='output*', model_name='SineCurrentModel')

Failing Simulation on Lower Assertion Level

By default PyDelica will look for the expression assert | error as an indication of a Modelica assertion failure and then terminate when this is violated. You can override this behaviour using the fail_on_assert_level method of the Session class:

from pydelica import Session

with Session() as pd_session:
    pd_session.fail_on_assert_level('warning')

Possible values ranked by order (highest at the top):

Running the Simulation

To run the simulation use the simulate method. If a model name is specified then that model is run, else this is the first model in the model list. At the simulation step parameter values are written to the XML file read by the binary before the model binary is executed.

# Model name is optional, verbosity is also optional and overwrites that of the session
session.simulate(model_name='SineCurrentModel', verbosity=OMLogLevel.DEBUG)

Retrieving Results

To view the results use the get_solutions method which will return a python dictionary containing the solutions for all models after a model run:

solutions = session.get_solutions()

The variables for each model are stored as a Pandas dataframe.

Using Alternative Libraries

NOTE: Currently only works in WSL on Windows machines.

You can use an alternative library version via the use_library method:

session.use_library("Modelica", "3.2.3")

you can also optionally specify the location of this library:

session.use_library("Modelica", "3.2.3", library_directory="/home/user/my_om_libraries")

Including Extra C Resources

When building a model extra C file resources can be specified using the c_source_dir argument to build_model:

session.build_model(
    model_path,
    c_source_dir=os.path.join(model_dir, "Resources", "Include")
)

Docker

A Docker image is available for OpenModelica with Pydelica:

$ docker pull artemisbeta/pydelica

You can try out Pydelica within a Jupyter notebook by running:

$ docker run -ti artemisbeta/pydelica jupyter notebook --ip 0.0.0.0 --no-browser

and opening the resulting URL within your browser.

Troubleshooting

Simulation fails with no error thrown

Try setting the assertion level to a lower level, for some reason OM ranks missing input file errors as type debug, see here.

stdout | info | ... loading "data" from "Default/myInput.mat"
assert | debug | Not possible to open file "Default/myInput.mat": No such file or directory
assert | info | simulation terminated by an assertion at initialization

PyDelica cannot find OMC

PyDelica relies on either locating OMC on UNIX using the which command, or in the case of Windows using the OPENMODELICAHOME environment variable. Ensure at least one of these is available after installating OpenModelica.

Use Cases

Pydelica is currently being used in the following projects, if you would like to be included in this list please open an issue:

• Power Balance Models, United Kingdom Atomic Energy Authority: A tokamak power balance model with Python API and CLI