Reproducible control-system experiments from nonlinear modelling to portable C99 runtimes
This repository is a control-engineering portfolio covering nonlinear systems, state-space control, LQR, observers, actuator saturation, anti-windup, numerical validation, and deployment-oriented C implementations.
It is designed for engineers and students who want more than isolated scripts: each experiment documents its model, assumptions, controller design, measurable results, automated evidence, and limitations.
The five documented studies, independent Python reference models, MATLAB tests and two portable C99 runtimes are implemented and maintained. Current work is focused on clearer traceability and portability rather than adding unverified hardware claims.
The published numbers are reproducible simulation results. No experiment in this repository has been validated on physical laboratory hardware.
- Compare multiple controller architectures under documented conditions.
- Reproduce numerical results from executable MATLAB and Python models.
- Inspect portable C99 implementations with fixed-size state and no dynamic allocation.
- Trace engineering requirements to automated MATLAB, Python, and C tests.
- Study where simulation evidence ends and hardware validation must begin.
| Project | Main methods | Published result |
|---|---|---|
| Nonlinear control loops | Jacobian linearisation, equilibrium stability, controllability, local state feedback | Linearisation error increases away from the operating point |
| Elastically mounted rotary arm | Fifth-order trajectory, feedback, feedforward, 2-DOF control, load disturbance | 2-DOF tracking RMSE is about 70% lower |
| Quarter-car active suspension | State-space modelling, LQR, road disturbance, actuator saturation | RMS body acceleration is reduced by about 35% |
| Magnetic levitation | Nonlinear plant, pole placement, Luenberger observer, sensor noise, convergence study, portable C runtime | Position-estimation RMSE is below 0.001 mm |
| Two-tank process | Nonlinear hydraulics, PI control, saturation, back-calculation, portable C runtime | Recovery improves from about 429 s to 312 s |
Detailed numerical results are generated from executable models and published in the results summary.
| Active suspension | Magnetic levitation | Two-tank anti-windup |
|---|---|---|
 |
 |
 |
run_allRun the MATLAB test suite:
results = runtests('matlab/tests', 'IncludeSubfolders', true);
assertSuccess(results);cmake -S c -B build/c -DCMAKE_BUILD_TYPE=Release
cmake --build build/c --parallel
ctest --test-dir build/c --output-on-failure
./build/c/maglev_observer_demo
./build/c/two_tank_demopython -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
ruff check .
ruff format --check .
pytest -q
python -m validation.generate_reference_figures --check-only
python -m validation.report --check- Nonlinear differential-equation modelling
- Operating-point and Jacobian linearisation
- State-space modelling, controllability, and observability
- Pole placement and eigenvalue analysis
- Luenberger observers and output-feedback control
- LQR control and disturbance rejection
- P and PI control with back-calculation anti-windup
- Actuator saturation and control-effort analysis
- Fourth-order Runge-Kutta integration and convergence studies
- Deterministic sensor-noise scenarios
- Portable C99 observer and PI-control runtimes
- Fixed-size state storage without dynamic allocation
- CMake, CTest, GCC, Clang, and strict compiler warnings
- Direct MATLAB and multi-version Python CI
| Purpose | Technology |
|---|---|
| Control experiments | MATLAB R2024b scripts and matlab.unittest |
| Independent references | Python 3.10–3.12, NumPy and SciPy |
| Portable runtimes | C99 with fixed-size state and no dynamic allocation |
| Build and runtime tests | CMake, CTest, GCC and Clang |
| Automation | GitHub Actions, Ruff and pytest |
| Layer | Evidence |
|---|---|
| MATLAB | Direct matlab.unittest execution using MATLAB R2024b |
| Python | Numerical regression, behaviour, and report-freshness checks on Python 3.10–3.12 |
| C | GCC and Clang builds with warnings treated as errors |
| Runtime | CTest checks for configuration handling, tracking, saturation, determinism, anti-windup, and convergence |
| Traceability | Requirements mapped to automated evidence in the verification matrix |
projects/ MATLAB experiments and project-specific functions
matlab/+control_lab/ Shared MATLAB numerical and metrics utilities
matlab/tests/ Direct matlab.unittest verification
validation/ Independent numerical reference models and reports
tests/ Python numerical and behaviour tests
c/ Portable C99 control runtimes
c/tests/ CTest-based runtime verification
docs/ Architecture, results, and requirements traceability
.github/workflows/ MATLAB, Python, and C CI workflows
See the architecture document for the system structure and extension points.
Focused contributions that improve numerical validation, portability, documentation, or test coverage are welcome. Read CONTRIBUTING.md before opening a pull request.
Useful starting points include:
- independent numerical checks;
- additional deterministic disturbance scenarios;
- compiler and platform portability improvements;
- documentation corrections;
- requirements-to-test traceability improvements.
These are simulation and software-validation projects, not production controllers. They do not establish:
- hardware-in-the-loop performance;
- real-time scheduling guarantees;
- fixed-point numerical behaviour;
- MISRA-C compliance;
- robust stability outside the tested parameter ranges;
- functional-safety certification.
Each project README documents its specific assumptions and validation boundaries.
- Add more independent reference cases around saturation and disturbance rejection.
- Extend compiler and operating-system portability checks for the C runtimes.
- Document fixed-point and timing requirements before attempting embedded deployment.
- Add hardware-in-the-loop work only when a real interface and measurable acceptance criteria exist.
The original laboratory exercises were completed in a four-person academic team. This repository contains independently structured portfolio implementations and documentation. See NOTICE.md.