[adrpo]

Fixes #1674.

Adds OMSimulator.calibration. It fits the parameters of a Calibration activity of the OpenSCALING UQ layered standard to the reference data in a CSV file, on an SSP. The fit is bounded nonlinear least squares (scipy.optimize.least_squares). The UQ file is the one the SSD references by ssc:MetaData of type application/x-uncertainty-quantification.

OMSimulator --calibrate cal_01 --calibrationOutput calibrated.ssp model.ssp

from OMSimulator.calibration import calibrateSSP
result = calibrateSSP("model.ssp", "cal_01", output="calibrated.ssp")

How it works

• The SSP is instantiated once. Each evaluation runs reset, sets the fixed ParameterSet values and the targets, runs initialize, then stepUntil each reference time point and getValue on the observed variables. No result file is written.
• The model starts at its own start time. A calibration window only selects which samples are compared.
• Results go back into the UQ file. The ResultSet gets an ErrorMetric type="RMSE" per observed variable and refers to <activity>_calibrated.ssv, an SSP parameter set with the calibrated values written next to the UQ file. Both validate against their schemas (UQ specification-v1, SSP 2.0).
• The standard does not yet define the column mapping, weights, time window or optimizer settings. They are read from <uq:KeyValues type="org.openmodelica.omsimulator.calibration"> in the SimulationSetting, which the schema allows: dataName:<var>, weight:<var>, startTime, stopTime, scaling, relativeTolerance, maxEvaluations, diffStep. By default an observed variable maps to the CSV column with its name, optionally followed by |unit.
• numpy and scipy are imported on first use: pip install OMSimulator[calibration]. The omsimulator build-deps images already have scipy.

Tests

calibration/HelloWorldCalibration.py fits a and x_start of HelloWorld (der(x) = a*x) to exp(-t). cal_01 uses all samples; cal_02 uses only t ≥ 0.5 while the model still starts at 0. Both recover a = -1, x_start = 1. Checked that cal_02 fails (x_start = 0.6065) if the simulation starts at the first compared sample. Run with runtest.py against an installed OMSimulator (pin ffc28b1) plus this branch's Python sources.

Validation on the OpenSCALING DC-motor use case

This is the OpenSCALING DC-motor calibration (d2.1-uq-layered-standard, use_cases/dc_motor, activity cal_01): 6 parameters, 502 noisy samples of speed and current at 13 V / 0.25 N·m. The SSP was rebuilt from three FMUs exported by omc 1.28 as model exchange (one CVODE integrates the whole system), and it matches the omc simulation of DCMotor_L3 to six digits. The run takes 485 evaluations and 15 s:

parameter	nominal	this PR	reference notebook
R	0.1	0.099968	0.099973
L	1e-4	1.0246e-4	1.0348e-4
c_m	0.01	0.010013	0.0100116
c_g	0.01	0.010215	0.010696
J	5e-5	5.132e-5	5.426e-5
d	1e-4	1.067e-4	1.24e-4
cost	104360	104041	104740

Using the use case needed three changes, which were made in the test SSP, not in this PR:

• ParameterSet names constantVoltage.V / constantTorque.tau_constant are calculated parameters in an FMU; Vbattideal / Tmechload were used instead.
• The observed level_3_dc_motor.I is the negative of the CSV column dc_motor_electric1.I, so dc_motor.dc_motor_electric1.I is observed.
• The uq:Settings element is no longer in the schema and was dropped.

With co-simulation (the default for me_cs FMUs) the fit is off (c_g at its upper bound), because the electrical time constant of 1 ms is no larger than the communication step.

Not in this PR

• Every evaluation prints the native info: lines (step size, final statistics), since info messages cannot be silenced from Python.
• No unit conversion between observed variables and CSV columns.
• HTML report.

Error reporting

An invalid calibration problem (unknown activity id, no UQ file, unknown observed variable, ...) is reported as a single error: line on stderr with exit status 1, not a Python traceback.

Coverage workflow

The calibration test needs numpy and scipy, which the bare GitHub runner of the coverage workflow did not have. The job now runs in docker.openmodelica.org/build-deps:ubuntu-24.04-omsimulator, the image the linux64-noble Jenkins stage uses, so coverage and Jenkins test in the same environment. OpenModelica/build-deps#101 also adds numpy and scipy to every apt image, not only the omsimulator add-on.

OMEdit

The OMEdit front end is OpenModelica/OpenModelica#17360 (issue OpenModelica/OpenModelica#17359): a Calibrate dialog for SSP models that runs OMSimulator --calibrate. It needs an OMSimulator submodule bump after this PR is merged.

---

Generated by Claude Code

🤖 Generated with Claude Code

[arun3688]

looks good

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 28.35%. Comparing base (8a9d93d) to head (cba3cee).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1675      +/-   ##
==========================================
+ Coverage   27.69%   28.35%   +0.65%     
==========================================
  Files          72       72              
  Lines       13976    13976              
  Branches     9020     9020              
==========================================
+ Hits         3871     3963      +92     
+ Misses       8876     8775     -101     
- Partials     1229     1238       +9     

Flag	Coverage Δ
linux	28.35% <ø> (+0.65%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.