Case studies

Four pieces of work, each told the same way: the problem, the dynamics and constraints, what was built, how it was validated, and what it taught. Further material can be shared under NDA.

Discuss a similar problem

CERN · IAV · SkySails · own developments

Proof with context

References should show fit, not just impress

For a technical decision-maker the useful question is whether the background matches the current control, estimation or optimization problem.

Some proof is public — papers and certificates are linked below. Other material is more useful under NDA, once the real system, constraints and failure mode are known.

Rafał Noga in a technical design review with engineering parts and system diagrams on a table — References establish relevance before a serious technical conversation.

Cryogenics · NMPC · state estimation

Real-time NMPC for the LHC superfluid-helium circuit

Context: CERN, Geneva — researcher and doctoral student (2007–2013); PhD, University of Valladolid (2015).
Problem: Keep the superconducting magnets of the LHC at their operating temperature of 1.9 K. Above 2.1 K an interlock de-energizes the magnets; above 2.16 K the helium leaves the superfluid state and a magnet can quench.
Dynamics and constraints: Strongly nonlinear, distributed thermohydraulics with dead time, inverse response, actuator saturation and strong coupling — and only a few measurements along a long cryogenic circuit.
Existing approach: Conventional PI control, tuned conservatively to stay clear of the temperature limits.
Solution: A first-principles thermohydraulic model simplified for real-time use; nonlinear moving-horizon estimation of the unmeasured states (5 thermodynamic states from 3 pressure sensors and temperature, at 1 Hz); output-feedback economic NMPC with the temperature limits as hard constraints.
Implementation and validation: Semi-automatic code generation for the NMPC of a stiff, distributed-parameter system. About 1 s per estimation and 7–14 s per optimization. The controller was tested at CERN.
Result: After the same perturbation, the NMPC stabilized the bath temperature about ten times faster than the legacy PI control, without violating the 2.1 K limit.
Lessons: The estimator decides how good the controller can be. A simplified model that runs in real time beats a complete one that does not. Hard limits belong in the optimization problem, not in conservative tuning.
Evidence: Noga et al., IEEE CDC 2014 — moving-horizon estimation and control
Noga et al., IFAC NMPC 2015 — NMPC for superfluid helium cryogenics
Noga, PhD thesis, University of Valladolid, 2015
Noga & Ohtsuka, ICPC 2011 — code generation for NMPC of a stiff distributed-parameter system

CERN recommendation letter (Geneva, 10.03.2009): “I recommend Mr. Noga without any reservation.”

Wind energy · economic NMPC

Advanced control for utility-scale wind turbines

Context: IAV GmbH — control engineer, R&D in model-based and advanced control for multi-megawatt wind turbines.
Problem: Larger rotors make turbines more flexible and more strongly coupled. The controller must balance energy capture, structural loads, actuator use and stability across changing wind regimes.
Dynamics and constraints: Flexible tower and blades, pitch and torque actuators with limits, turbulent wind that is only measured indirectly, and operating regimes from partial to full load.
Existing approach: In conventional turbine control, pitch and torque loops are tuned separately and switched by supervisory logic between operating regions.
Solution: Control-oriented first-principles modelling and identification; nonlinear state estimation; multi-objective economic NMPC; controller architecture and requirements; diagnostics and parameter optimization; engineering tools for repeatable deployment of advanced-control methods.
Implementation and validation: Model-based design and simulation of the control architecture, and tooling to take advanced-control concepts toward industrial use. Details are confidential.
Result: A model-based control architecture and deployment tooling for multi-megawatt turbines; published work on super-short-term wind-speed prediction for turbine control (IECON 2018).
Lessons: On large flexible machines the problem often sits in the architecture between loops, not in one loop. The wind itself has to be estimated. Advanced control pays off only if it can be deployed repeatably.
Evidence: Luna et al., IECON 2018 — super-short-term wind-speed prediction for turbine control

Airborne wind energy · trajectory optimization

Flight automation and trajectory optimization for airborne wind energy

Context: SkySails Power GmbH, Hamburg — control engineer (until March 2024).
Problem: Make a full-scale prototype of a tethered, variable-trim kite fly reliably, and find the flight trajectories that produce the most power within the system limits.
Dynamics and constraints: Nonlinear flight of a flexible, tethered wing; strong wind disturbances; limited on-board sensing; aerodynamic, mechanical, electrical, geometric and operational constraints.
Existing approach: Inherited prototype controls that had to be tuned and improved on the real system.
Solution: Prototype tuning and flight-data analysis; identification and model-based feedback improvement; setpoint adaptation; UKF-based state estimation; nonlinear constrained optimization of 3-D flight trajectories for variable-trim kites.
Implementation and validation: Work on the full-scale prototype with flight data; optimization workflows for setpoints, performance studies, sensitivity analysis and component sizing; online trajectory optimization for autonomous power-generation flights.
Result: Continuous online replanning for a highly dynamic airborne system, and a trajectory-optimization method published as a paper (arXiv 2024).
Lessons: On a new system, make the prototype usable first. A trajectory optimizer is only as good as its model and the constraints it respects. Flight data closes the gap between model and reality.
Evidence: Noga et al., arXiv 2024 — 3-D flight-trajectory optimization of variable-trim kites

SkySails Power GmbH work certificate (Hamburg, 31.03.2024), translated from German: “…we were always extremely satisfied with Mr. Noga.”

Soft sensors · sensor fusion

Estimating what cannot be measured directly: a sensor-fusion variometer

Context: Own development (SSDV12) — and the same estimation pattern as in the cases above.
Problem: Paraglider pilots need a fast and accurate climb rate. Barometric variometers trade noise against delay, and GPS alone is too slow.
Dynamics and constraints: Pressure, inertial and GPS measurements with different noise, delay and bias; a pilot–wing system that moves in all axes.
Existing approach: Barometric variometers whose filtering trades noise for delay.
Solution: First-principles modelling and a sensor-fusion estimator combining pressure, inertial and GPS measurements.
Implementation and validation: Estimator developed and tested on experimental flights; a hardware prototype is planned.
Result: A working estimation approach — and the same pattern (model + available sensors + nonlinear estimation) used at CERN, IAV and SkySails.
Lessons: A soft sensor is a model plus an estimator. The noise, delay and bias of each sensor have to be modelled explicitly.
Evidence: SSDV12 sensor-fusion variometer — development page

How proof is shared

Public where possible

Papers, conference work and certificates establish technical depth without exposing confidential material.

NDA where necessary

For many industrial mandates the most useful evidence is shared after a first fit discussion and mutual confidentiality.

Decision-oriented

The point of proof is not volume; it is to show whether the background fits the problem at hand.

Have a similar control, estimation or optimization problem?

Bring the system, the main symptoms and the hard constraints. The technical fit is usually clear within 30 minutes.

Book a 30-min technical fit call →

← Back to homepage

Imprint

Privacy

Direct contact