Context
Individual first-year master's (MESI) lab project (Renewable Energy module) at the Faculty of Physics & Engineering, University of Strasbourg, supervised by Prof. Dominique Knittel. The goal: model in MATLAB/Simulink a wind energy conversion chain, a turbine driving a DC generator (PMDC) feeding a resistive load, then compare three increasingly 'intelligent' control strategies to see which extracts the most power, in steady state and during transients.
Approach / Solution
I built and studied three configurations, each with a static study (parameter sweeps) and a dynamic one (step responses): • Direct connection, the load Ro is wired straight onto the generator. Sweeping Ro, the pitch angle β and the wind speed, I recovered the turbine's aerodynamic signature (the Cp(λ) bell curve, peak Cp ≈ 0.40 at λ ≈ 12) and its maximum power point (MPP ≈ 7.5 kW). The limitation it exposed: the output voltage Vo is uncontrolled, a load step made it swing by a factor of 31. • Boost converter + PI regulation. I inserted a step-up converter between the generator and the load, with a PI loop regulating Vo. Vo then becomes independent of load and wind. Tuning took care: an over-aggressive PI that oscillated (gains brought down to P = 1e-4, I = 0.05), a duty-cycle saturation [0.01; 0.99], and a switch to a fixed-step ode4 solver (1 µs) to cope with the 10 kHz switching. I also hit the Boost’s intrinsic limit: it can only step the voltage up, never down. • Perturb & Observe MPPT. I replaced the fixed setpoint with a P&O algorithm (a MATLAB Function block) that finds the top of the power curve on its own by perturbing the duty cycle and observing the effect on P. A tricky detail: I had to flip the step signs versus the original photovoltaic code, because on a wind turbine with a Boost the duty-cycle-to-voltage relationship is reversed.
Results
The MPPT is clearly the most accomplished solution: it reaches Cp ≈ 0.44 (close to the turbine's theoretical maximum) and captures ~2.4× more power than regulation alone, with no manual tuning beforehand. It adapts on its own to wind and load steps, and even stays robust when a disturbance knocks it off a sub-optimal plateau. My takeaway: the more intelligence you put into the control, the more power you extract, without touching the hardware. I also got a feel for the system's physical limits: at very high power the model diverges (Vo running away beyond 700 to 850 V), which ties directly to real component sizing (output capacitor, MOSFET). Avenues I could still explore: a Buck-Boost converter (able to step the voltage both down and up, unlike the Boost), a finer MPPT (variable step, Incremental Conductance) to cut the oscillations around the MPP, component sizing for high power, and eventually experimental validation on a real test bench rather than in simulation.