Using the Simulation

Run these commands from the repository root of the v2.2 branch after installation. The campaign launcher exports validated benchmark recordings. The custom launcher produces a raw OpenModelica result.

Run all ten benchmark datasets

Inspect the work list, then run up to four simulations at once:

python code/campaign.py --plan
python code/campaign.py --jobs 4

The complete campaign has 350 recordings: 160 healthy and 190 single-fault recordings. Each of the ten topologies has 16 healthy recordings (10 training, 3 validation, 2 calibration, 1 test). Each selected fault proxy has two recordings: a persistent window and a temporary window. The launcher runs the seed-matched healthy test control before its fault recordings.

The default release root is output/v2.2/; build artifacts and worker logs are under build/v2.2/campaign/. Set --jobs to the number of concurrent OpenModelica runs your machine can sustain. Every worker receives a separate build directory.

To place generated files elsewhere, use separate output and build roots:

python code/campaign.py --jobs 8 \
  --output /path/to/results \
  --build-root /path/to/build

After an interruption, rerun the same command with unchanged source and configuration. The launcher verifies and reuses complete recordings. Choose fresh output and build roots after changing the source or configuration. It moves an incomplete recording into the build logs’ incomplete/ directory before retrying it. See Campaign Output and Data Directory for the complete layout and validation files.

Run one benchmark dataset

--dataset accepts ds1 through ds10 and produces a complete shard for that topology:

python code/campaign.py --dataset ds1 --plan
python code/campaign.py --dataset ds1 --jobs 4

For ds1, this produces 16 healthy and 12 fault recordings. The release summary is output/v2.2/ds1_release_manifest.json. A full ten-dataset run instead writes output/v2.2/release_manifest.json.

Simulate your own topology or module

Copy examples/custom.json from the v2.2 branch. It connects a source, the example PassThrough.mo module, and a sink. Set model.modules and model.edges for your plant; every declared port must be connected exactly once. Model paths in files are relative to the JSON configuration.

A new module declares type: custom, modelica_class, input_ports, output_ports, files, and an empty faults object. Its Modelica class must define a replaceable Medium package. You can also combine the built-in modules without changing their models.

Run the included example:

python code/simulate.py --config examples/custom.json --dataset custom1

The raw result is output/custom/custom1/processPlant_res.csv. The same run directory contains the generated plant, OpenModelica script, and compiler logs. This command exports all Modelica variables as raw CSV and does not create the benchmark’s Parquet release or campaign metadata. To run again, pass a fresh --output location; an existing run directory is protected from overwrite.