Control of Wind Turbines: A data-driven approach dr.ir. Jan-Willem van Wingerden March 14, 2011 1
Outline General introduction Data driven control cycle Smart rotor Visit NREL Conclusions and outlook March 14, 2011 2
Outline General introduction Data driven control cycle Smart rotor Visit NREL Conclusions and outlook March 14, 2011 3
General introduction: overview Management Architecture Civil Mechanic Maritime Electrical Math Industrial Design Applied Science Aerospace Wind energy Delft Center for Systems and Control 60 fte, 35 PhD March 14, 2011 4
General introduction: Ph.D. within DUWIND experiments / Bijl, van Bussel analysis & design / advanced NavierStokes Analysis, feasibility, lab tests / small scale field-test / large scale field-test van Wingerden vankuik Bersee Multidisciplinary design methods, component and system design, electric power conversion, Zaaijer, vantooren, Tomiyama,Polinder, Willemse, vdtempel Support structures, access systems, Floating offshore (combined with wave energy?) Kling, Kunneke grid connection & integration, market & institutions, ensemble with other sust.energy sources in DRI-TUD March 14, 2011 5
General introduction: DCSC 2003 1994 2008 1989 Gregor Baars Control: Long history in the wind Bongers (PhD) Molenaar (PhD) Wingerden (PhD => assistant prof.) Steinbuch (PhD) Baars (PhD) Bosgra (Prof.) Verhaegen (Prof.) March 14, 2011 6
General introduction: My Ph.D. students Ivo Houtzager (4 th year Ph.D.): adaptive and learning control algorithms for wind turbines Gijs van der Veen (2 nd year Ph.D.): Data driven control of wind turbines with smart rotors Patricio Torres (1 st year Ph.D): Wind farm control (distributed computation) Two new Ph.D. positions (1. combined structural and control optimization, 2. Fault tolerant wind farm control) March 14, 2011 7
Outline General introduction Data driven control cycle Smart rotor Visit NREL Conclusions and outlook March 14, 2011 8
Data-driven control: Conventional Model for control based on first principles Controller design algorithm Traditional design approach "..direct validation of models describing wind energy conversion systems by a direct comparison with measured data is of very limited use. One of the few possible solutions to this problem is the application of system identification. (bongers, 1994) March 14, 2011 9
Data comes in Model for control based data bbbbb Controller design algorithm Data driven design approach March 14, 2011 10
Data driven approach: Sys. ID Obtain model from measurement data System operates in closed loop Multiple-Inputs and Outputs Required external perturbation signal Which model structures are relevant? March 14, 2011 11
Data driven approach: Model structure Nonlinear (pff, really hard) Linear Time Invariant (LTI) (using linearization) Linear Parameter Varying (LPV) Hammerstein Models Gijs van der Veen March 14, 2011 12
Data driven approach: Model structure Gijs van der Veen March 14, 2011 13
Outline General introduction Data driven control cycle Smart rotor Visit NREL Conclusions and outlook March 14, 2011 14
smart rotor Wind energy: Young technology Rapid growth Too expensive Variable speed turbine Current desire (again) increasing size: Offshore (cost foundation) Power (with the square) Solution: New control concepts (and design methodologies) March 14, 2011 15
smart rotor Using integrated flaps SMA Piezo March 14, 2011 16
WT experiments I: March 14, 2011 17
WT experiments I: Wind tunnel Blade Pitch system Trailing edge flap Sensors Real-time system March 14, 2011 18
WT experiments I: First Principles vs experimental modeling We applied traditional loop shaping March 14, 2011 19
WT experiments I: Experimental results Feedforward control Feedback control Periodic disturbance Random disturbance (turbulence) V= 30 m/s α= 6 degrees 3P excitation March 14, 2011 20
WT experiments I: Experimental results Feedforward control Feedback control Periodic disturbance Random disturbance (turbulence) V= 30 m/s α= 6 degrees Eigenfrequency March 14, 2011 21
WT experiments I: Experimental results Feedforward control Feedback control Periodic disturbance Random disturbance (turbulence) V= 30 m/s α= 6 degrees Eigenfrequency flap excitation March 14, 2011 22
WT experiments I: Experimental results Feedforward control Feedback control Periodic disturbance Random disturbance (turbulence) Input spectrum V= 30 m/s α= 6 degrees Output spectrum March 14, 2011 23
WT experiments II: Plans March 14, 2011 24
WT experiments II: Real March 14, 2011 25
Data-driven approach: Periodic disturb. Black without excitation Gray with March 14, 2011 26
Data-driven approach: Periodic disturb. Black id with period Gray id without March 14, 2011 27
Data-driven approach: Feed Forward Parameterize input: Lift the expressions over one period: March 14, 2011 28
WT experiments II: MIMO feedback March 14, 2011 29
WT experiments II: Time domain results March 14, 2011 30
WT experiments II: March 14, 2011 31
Outline General introduction Data driven control cycle Smart rotor Visit NREL Vibrations CART III Drive train damper CART II Conclusions and outlook March 14, 2011 32
NREL: instability I March 14, 2011 33
NREL: instability II March 14, 2011 34
NREL: Uhm, where is it coming from 1. Unstable control law Not likely 2. Negatively/badly damped mode 3. Aeroelastic instability 4. Stall operation 5. P loads on top of badly damped structural modes 6. The drive 7. Something else Maybe Maybe Not likely Probably a part of the problem Maybe We hope not March 14, 2011 35
NREL: Linear model (FAST new) 7p 6p 5p 4p 3p 2p 1p March 14, 2011 36
NREL: Linear model (FAST old) 7p 6p 5p 4p 3p 2p 1p March 14, 2011 37
NREL: FFT vs rotor speed March 14, 2011 38
NREL: FFT vs rotor speed March 14, 2011 39
NREL: conclusion Is it badly damped combined with an excitation (4P)? Solution limit max RPM!! Other arguments to do this => March 14, 2011 40
NREL: 55 HZ (HSS Torque) Frequency (Hz) LSS RPM March 14, 2011 41
NREL: 109 HZ IMU X Frequency (Hz) LSS RPM March 14, 2011 42
Outline General introduction Data driven control cycle Smart rotor Visit NREL Vibrations CART III Drive train damper CART II Conclusions and outlook March 14, 2011 43
NREL 2: CART II LQR two inputs (ss tower accel, HSS speed) March 14, 2011 44
NREL 2: Before/after drive train replacement Can we really talk about damping?? March 14, 2011 45
NREL 2: What is going on?? 1. The LQR works perfect in simulation (even with a different stiffness) 2. Typically you see this response if there is a time delay E-stops (sample time 100Hz) 5 samples (0.05 s) will make a difference March 14, 2011 46
NREL 2: Solution Compensate for the delay (Pade) Generator speed Tower ss acceleration Generator Toque (ampl.) Generator Toque (phase) March 14, 2011 47
NREL 2: Conclusions 1. Simulations seem to support robustness issue 2. Waiting for wind 3. Robust control the way to go?? March 14, 2011 48
Outline General introduction Data driven control cycle Smart rotor Visit NREL Conclusions and outlook March 14, 2011 49
Conclusions. data driven algorithms are widely applicable Hammerstein model structure the way to go?? the smart rotor has the potential to reduce costs Waterfall plots are a really strong tool Delays can destroy performance March 14, 2011 50
Outlook. smart rotor: do new wind tunnel experiments sys. id. quantify uncertainties (for robust controller design) incorporate my research in education activities waiting for wind to validate our NREL work etc. etc.. March 14, 2011 51
NREL 2: Other modifications 1. LQR with frequency weights 2. Hinf control 3. Robust control March 14, 2011 52