Robust Control of a Wind Turbine Using Third

Generation CRONE Control

Francisco Ravasco Rui Melicio, Nelson Batista Duarte Valério
IDMEC, Instituto Superior Técnico ICT, Dep. Fı́sica, Escola de IDMEC, Instituto Superior Técnico
Universidade de Lisboa Ciências e Tecnologia, Universidade de Évora Universidade de Lisboa
Lisboa, Portugal IDMEC, Instituto Superior Técnico Lisboa, Portugal
[email protected] Universidade de Lisboa [email protected]
Évora and Lisboa, Portugal; [email protected]

Abstract—This paper presents the identification, development These problems can be minimised if, instead of using
and simulation of a complex order, third generation CRONE Horizontal Axis Wind Turbines (HAWT), to which class the
controller for a new vertical-axis wind turbine prototype of the vast majority of wind turbines installed belong nowadays,
Darrieus type. The prototype was tested in a wind tunnel and the
data used to develop an accurate model, validated experimentally. Vertical Axis Wind Turbines (VAWT) are used instead —
The CRONE control system design methodology was then used to such as the turbine studied and controlled in this paper. It
develop a robust controller, which was then tested in simulation. is possible to have a larger number of smaller VAWTs, with
Simulations allow for parameter uncertainty and different wind reduced noise and visual impact, spread over a city, producing
conditions, with and without the presence of turbulence. This decentralised electricity in a Smart Grid context. Even though
control system can be easily implemented using an Arduino
platform. they are behind HAWTs in development and investment,
Index Terms—Wind Energy, Darrieus Wind Turbines, Pro- VAWTs present several advantages [2]: insensitivity to yaw
totype, Control Strategy, Fractional Control, Third Generation wind direction; a smaller number of components; low sound
CRONE Control emission; ability to produce energy from wind in skewed
flows; capability of operating closer to the ground level; a
I. I NTRODUCTION generator position closer to the ground level, allowing for an
easier maintenance.
Wind energy has some disadvantages, such as long periods
A VAWT can be of two different types: Darrieus and
during which energy conversion is impossible (because of
Savonius. The prototype studied in this paper is a curved blade
either low wind speeds or excessive wind that could damage
Darrieus VAWT, developed and tested in [2]–[5]. Unlike a
the turbine), or the need of energy storage and some com-
Savonius VAWT, the Darrieus wind turbine usually requires a
plementary alternative form of energy production (caused by
start-up force to begin its movement, but the prototype herein
the main production hours tending to be at night time, when
addressed was developed featuring a new blade design that
consumption is smaller). But these are compensated by its
ensures its self-start capabilities [3], for wind speeds starting
many advantages, such as being an infinite, sustainable and re-
at 1.25 m/s, and exhibiting a stable behaviour under stress for
newable source of energy, widely distributed all over the world
winds up to 25 m/s, while maintaining a good performance.
(unlike e.g. oil), and presenting a low payback period (5 to 8
While a PID controller and an LQR have already been
months on 2 MW wind turbines) [1]. Its ecological impact is
developed and tested in simulation using previously developed
minimal, but noise pollution prevents energy production close
models [5], in this paper a model with an improved fit to
to the final consumer; remote locations are thus preferred.
real data is analysed, and then used for controller synthe-
While on the one hand this has the advantage of reducing
sis and simulation tests, using the third generation CRONE
the risk of disasters (because there is no one around), there
methodology. CRONE is the French acronym for Commande
are accrued difficulties of installation (because components are
Robuste d’Ordre Non Entier. The CRONE Control System
large, requiring exceptional transport and complex logistics),
Design (CSD) theory is a frequency domain approach of robust
there is a strong visual impact on rural landscapes, and wind
control, that makes use of fractional (or non-integer) order
farms must then be connected to the grid by long lines.
derivatives [6], and aims at ensuring robustness to plant pa-
This work is funded by: European Union through the European Re-
rameter uncertainty or variation [7]. Third generation CRONE
gional Development Fund, included in the COMPETE 2020 (Operational control [8] is a generalisation of the first two generations, and
Program Competitiveness and Internationalization) through the ICT project is based on complex order derivatives. These are used to define
(UID/GEO/04683/2019) with the reference POCI010145FEDER007690; FCT,
through IDMEC, under LAETA, Project UID/EMS/50022/2019.
a template for the desired open-loop frequency response, that
ensures the robustness of the controller within its uncertainty
domain. In other words, the main objective is to ensure that
978-1-7281-0653-3/19/$31.00 ©2019 IEEE the closed-loop gain, or the closed-loop damping coefficient,
Fig. 1. Experimental Prototype Data, Wind Speed vs Angular Speed, from
[2].
Fig. 2. Power coefficient Cp vs. tip speed ratio (TSR) λ proposed in [2].

or both, will never get beyond a certain value, even if some The corresponding maximum TSR is
parameter of the plant varies within a known range.
λmax = 2.04 · r ≈ 0.353. (3)
This paper is organised as follows. In section II the pro-
totype is presented; the selected model is presented and Experimental tests were conducted for this paper showing that,
analysed; the turbulent wind model to be used in simulation with the actuators installed, the range of the prototype’s TSR
is presented; and third generation CRONE control presented. would be reduced to approximately [0.1; 0.2]. The upper limit
In section III there is an analysis of the model with the of this range is expected to be improved with better actuators,
uncertainty domains; the controller objectives are stated; and and so, in what follows, an optimum value λopt = 32 λmax ≈
the controller synthesis explained. In section IV the results 0.2353 was assumed.
obtained in simulation are presented and the controller perfor- The power coefficient function CP is defined as the ratio
mance is evaluated. Finally, section V outlines the conclusion. between the extracted power and the available wind power.
It depends on λ and on the blade’s pitch angle β; this last
II. BACKGROUND dependence can be neglected. It can be shown that the power
A. The prototype and its model coefficient never surpasses the value of 0.593; in other words,
no wind turbine can capture more than 59.3% of the winds
The prototype’s technical specifications are given in Table
kinetic energy: this limit is known as the Betz limit [9]. The
I, and Figure 1 presents experimental data used for model
CP (λ) shown in Figure 2 was proposed in [2], and can be
identification [2] collected with the device’s permanent magnet
fitted as
synchronous generator (PMSG), responsible for converting the
mechanical power captured by the rotor into electric power, CP (λ) = −0.007365 λ2 + 0.1015 λ + 0.002052. (4)
in open circuit. Therefore this data represents the highest ratio Actually the simulations that originated this theoretical re-
between the rotor’s angular speed and the incoming wind lationship considered different dimensions for the turbine;
speed ωv [rad/s]
[m/s] of which this this prototype is capable. therefore, it is expected that (4) does not rigorously describe
this prototype.
TABLE I The identified transfer function model for the turbine devel-
T URBINE ’ S S PECIFICATIONS oped in [5] is
PARAMETERS S YMBOL VALUE 7131 s + 2879.9
Number of Blades Z 5 Δω(s) = Δv(s)+
Rotor Height [mm] H 480 s2+ 13200 s + 1861
Rotor Radius [mm] r 173 952 s + 61.41
Blade Body Height [mm] - 360 ΔRc(s), (5)
s + 13200 s + 1861
2
Blade Profile Chord [mm] c 53
Moment of Inertia [kg m2 ] Jt 0.1 where Δv(s) and ΔRc(s) are the wind speed variation and
the PMSG circuit load variation, respectively, and Δω(s) is
As this type of VAWT has a fixed pitch, the rotors angular the rotor’s angular speed variation.
speed is expected to be directly proportional to the incoming
B. Turbulent Wind Model
wind speed. Figure 1 corresponds to
The study of the wind behaviour around buildings and other
ω
= 2.04. (1) man made formations assumes a role of extreme importance
v
[10]. The model chosen to simulate wind turbulence is [11]:
The tip speed ratio (TSR) is given by  

ωr v = v0 1 + Ak sin(ωk t) , (6)
λ= . (2)
v k
where k is an index associated with the mechanical oscillation
excited, Ak is the amplitude of oscillation k, and ωk is
its frequency. Mechanical oscillations are modelled in the
[0.1, 10] Hz frequency range with the parameters in Table II
[11].

TABLE II
M ECHANICAL O SCILLATIONS

Source k Ak ωk
Asymmetry 1 0.01 ωk
Tower Shadow 2 0.08 3 ωk
Blades 3 0.15 9π

C. Third Generation CRONE Control

In third generation CRONE control, the Nichols plot of the
open-loop is set to match a template, which is a straight-
line segment around the gain cross-over frequency ωcg for
the nominal parametric state of the plant [8]:
β(s) = sa+jb ≈ (7)
  π sign(b)  ω a   
s
−sign(b)
cg
≈ cosh b cos b ln .
2 s ωcg
Fig. 3. Third generation CRONE control. Top: template in the Nichols plot;
The real part a determines the location of the phase of the bottom: three generalised templates tangent to a contour corresponding to a
generalised template at frequency ωcg , that is −a π2 , and the particular value of Mt (figures adapted from [7]).
imaginary part b determines its angle to the vertical, as seen
in Figure 3. When considering the uncertainty of the plant,
one has to account for the variations such uncertainties will
cause in the desired resonant peak Mr , and minimise the
influence of the plant’s uncertainties to ensure stability. There
is an indefinite number of templates tangent to the desired M -
contour for the nominal parametric state of the plant, as seen
in Figure 3; from all these, the optimal template is the one
that minimises variations of Mr , i.e. the overlapping between
the possible Nichols plots, when the different uncertainties are
considered, and the M -contour [7].
III. C ONTROLLER D EVELOPMENT
This system has two inputs: the PMSG resistance Rc is a
controllable input, the wind velocity v is a disturbance. Thus
only transfer function
s Fig. 4. Bode diagram of Δω(s)
.
Δω(s) 0.033 ( 0.0645 + 1) ΔRc(s)
= s s , (8)
ΔRc(s) ( 13200 + 1)( 0.141 + 1)
with the Bode diagram of Figure 4, will be analysed for
controller synthesis. The main objective of the control strategy is to maintain
The system is stable in closed loop for positive gains, the turbine’s efficiency at its maximum. This translates into
and has infinite gain and phase margins. The uncertainty trying to maintain λ in a value that maximises Cp , making
was assumed to be of ±20% in the plant’s coefficients. The use of the influence Rc has on ω, whatever the wind speed
nominal plant’s Nichols plot, with the uncertainty domains and passing through the turbine. The specifications proposed in
the extreme plants, is shown in Figure 5. The extreme plants, [5] for the system’s closed-loop step response were a 20%
useful to simulate the system’s response in the extreme cases, maximum overshoot and a 2% settling time of 10 s or less.
are For the design of the controller, the CRONE CSD Toolbox for
Matlab was used [12]. The Toolbox allows synthesizing any
Δω(s) 1142 s + 40.94
= 2 (9) CRONE controller, from first to third generation, by setting
ΔRc(s) s + 8800 s + 1861
the specifications in the software and adjusting the different
Δω(s) 761.6 s + 92.11 sensitivity functions that characterise the system. After several
= 2 (10)
ΔRc(s) s + 19800 s + 1861 modifications of the parameters and sensitivity functions, the
 Fig. 6. Closed-loop Response (C = 1).

plant and both extreme plants. The results are presented in

Figure 6. The difference between the nominal and extreme
plant’s response is small. This is due to the fact that the
uncertainty was only introduced in the transfer function from
ΔRc to Δω(s), and this transfer function has a low open
loop gain (Figure 5). The same test was performed for the
controlled nominal plant, and for both extreme plants; the
result is presented in the Figure 7. The angular velocity
variation follows the reference needed to maintain the TSR at
the optimum value; all three plants fulfil the specifications. The
control effort shows that the load imposed resistance reaches
Fig. 5. Nichols plots; top: without control; bottom: with CRONE control. perfectly acceptable values of around 200 Ω, for the extreme
Green: uncertainty domains; red: nominal plant; blue: extreme plants. plant 2.
The effect on the TSR of the abrupt increase of the wind
speed is also presented. The sudden change in the wind
Nichols locus obtained after applying the rational controller is
velocity causes a decrease in the TSR and consequently
also shown in Figure 5. After the several approximations and
a decrease in the turbine’s efficiency. The presence of the
simplifications of the CRONE CSD, the controller is finally
controller counteracts this effect, and in less than 10 s the
given by the rather simple transfer function
turbine’s efficiency is again at its maximum.
−23.524 −9.0679 32.5992
C(s) = + + (11)
s + 3.9295 s + 0.34063 s B. With Turbulence
IV. S IMULATION RESULTS AND CONTROLLER In order to test conditions as real as possible, the step
PERFORMANCE
input was substituted for a smoother signal, and the turbulence
A. Without Turbulence present in a real wind stream was considered. The wind value
First, the controller was tested using a step as wind per- from which the optimal angular speed is computed is still the
turbation, neglecting the turbulence present in the wind, to signal without turbulence. Thus, the signal with turbulence
assess the controlled closed-loop system’s behaviour and the will only be used as a disturbance, and not to compute the
uncontrolled response in closed-loop (C = 1). Of course, in reference.
nature, the wind does not have an abrupt, step-like behaviour. The wind input is modelled resorting to the turbulent wind
However, it is valid to admit that, if the system behaves model presented before. After the addition of the turbulence
accordingly with a step input disturbance, in the presence of model, the wind speed signal is as presented in Figure 8. This
smoother disturbances its performance should be of higher simulation is only shown for the nominal plant, as the results
quality. For this test, the wind is kept at its nominal value from the three plants overlap each other.
v = 6 m/s, and at t = 30 s the disturbance is simulated as a In Figure 9 it is shown that, despite the turbulence and the
step Δv = 4 m/s. For t ≥ 10 s the wind velocity is kept at its variation around the optimal angular speed, the system can
final value v = 10 m/s. track reasonably well the reference imposed by the variation
Firstly, for comparison purposes, this test was performed of 4 m/s in the wind speed. The circuit imposed load has its
in closed-loop without controller (C = 1), for the nominal maximum value of approximately 151 Ω for a ΔRc ≈ 15 Ω. It
 Fig. 8. Wind speed variation with turbulence.

• the integral of time multiplied by the absolute value of

f
the error, IT AE = 0 t · |e(t)| dt;
• integral of the square value of the control input, ISV =
f 2
0 u (t) dt.
ITAE is used as a numerical measure of tracking performance
for the entire error curve, and ISV shows the energy consump-
tion of the control effort.
Because in the case where turbulence is used it appears
only from 20 s on, the time frame taken into account is
from 20 s to 70 s for the simulation with turbulent wind
input. In the step input, only the simulation occurred after the
step will be analysed. Table III summarises the performance
assessment results for the different simulations. As expected,
the step input simulation without turbulence is the easiest test
to track, as the reference signal is simple and does not include
any disturbances other than the wind step itself. However,
this is the test that requires greater energy consumption, as
the high amplitude abrupt change in the reference has to be
accompanied by a higher control effort. In both tests, again
as expected, results for the ITAE are much better for the con-
trolled system when compared to the closed-loop uncontrolled
system. Comparing the results between plants shows that the
difference from the plant with the higher ITAE to the one
with the lower is small, within the same test and for the
controlled response. This particularity highlights the capability
Fig. 7. Results with the CRONE controller, without turbulence. Top: angular of the third generation CRONE controller in positioning the
speed variation; middle: control effort; bottom: TSR. uncertainty domains, in such a way that they have the smallest
influence in the system’s behaviour.

is obvious that there is an improvement when using CRONE V. C ONCLUSIONS

control. While the controlled response oscillates around the In this document, a Darrieus VAWT prototype is studied.
optimal TSR, the uncontrolled one is almost always far from A fractional third generation CRONE control was success-
the optimal point. fully developed resorting to the CRONE CSD Toolbox. Two
different disturbances were used to test the controller in a
C. Controller Performance Assessment simulation environment: a step and a (modelled) turbulent
wind. In general, the controlled system has an (ITAE) tracking
In order to evaluate the performance of the controller the performance twice as good as the system’s response without
following metrics were applied: controller (C = 1). The CRONE controller has a good per-
 TABLE III
S TEP I NPUT AND H ARMONIC T URBULENT W IND S IMULATION

Nominal Extreme 1 Extreme 2

Uncontrolled Controlled Uncontrolled Controlled Uncontrolled Controlled
Step Input ITAE 797.594 27.76 802.747 32.133 796.198 19.855
Simulation ISV 38.702 14731 35.853 10102 39.439 22716
Harmonic Turbulent ITAE 695.116 265.064 694.251 256.437 698.469 271.769
Wind Simulation ISV 17.129 6447.1 16.93 4680 17.368 8715.3

formance when it comes to overcome the differences between

the nominal plant and perturbed ones. The metrics used to
assess the controller performance do not present significant
difference between plants, within the same test, despite the
presence of the ±20% uncertainty in the controlled transfer
function. Experimental tests were performed to determine the
prototype’s TSR limits.
Future work includes obtaining a new experimental Cp(λ)
curve and testing experimentally the developed controller,
comparing the performance with that of the controllers in [5].
An Arduino platform has been selected for this purpose and
the work is under way.
R EFERENCES
[1] K. R. Haapala and P. Prempreeda, “Comparative life cycle assessment
of 2.0 MW wind turbines,” International Journal of Sustainable Manu-
facturing, vol. 3, no. 2, pp. 170–185, Jan 2014.
[2] N. Batista, R. Melicio, V. M. F. Mendes, M. Calderón, and A. Ramiro,
“On a self start Darrieus wind turbine; blade design and field tests,”
Renewable and Sustainable Energy Reviews, vol. 52, pp. 508–522, Dec
2015.
[3] N. Batista, R. Melicio, J. C. O. Matias, and J. P. S. C. ao, “New blade
profile for Darrieus wind turbines capable to self start.” IET Conference
on Renewable Power Generation (RPG 2011), Sept 2011, pp. 1–5.
[4] N. Batista, R. Melicio, and V. M. F. Mendes, “Darrieus-type vertical axis
rotary-wings with a new design approach grounded in double-multiple
streamtube performance prediction model,” AIMS Energy, vol. 6, no. 5,
pp. 673–694, 2018.
[5] T. Pereira, N. Batista, C. Cardeira, P. Oliveira, and R. Melicio, “Darrieus
wind turbine prototype control study,” no. 4. IEEE, 2018, pp. 853–858.
[6] A. Oustaloup, “La commande CRONE,” Commande Robuste d’Ordre
Non Entier, no. 6, pp. 130–135, 1991.
[7] J. Sabatier, P. Lanusse, P. Melchior, and A. Oustaloup, Fractional Order
Differentiation and Robust Control Design, ser. Intelligent Systems, Con-
trol and Automation: Science and Engineering. Springer Netherlands,
2015, vol. 77.
[8] P. Lanusse, A. Oustaloup, and B. Mathieu, “Third generation CRONE
control,” in Proceedings of IEEE Systems Man and Cybernetics Confer-
ence - SMC. IEEE, 1993, pp. 149–155.
[9] A. Betz, “Das Maximum der theoretisch möglichen Ausnützung des
Windes durch Windmotoren,” Zeitschrift für das gesamte Turbinenwe-
sen, vol. 26, pp. 307–309, 1920.
[10] F. Balduzzi, A. Bianchini, E. A. Carnevale, L. Ferrari, and S. Magnani,
“Feasibility analysis of a Darrieus vertical-axis wind turbine installation
in the rooftop of a building,” Applied Energy, vol. 97, pp. 921–929, Sept
2012.
[11] R. Melicio, V. M. F. Mendes, and J. P. S. C. ao, “A pitch con-
trol malfunction analysis for wind turbines with permanent magnet
synchronous generator and full-power converters: proportional integral
versus fractional-order controllers,” Electric Power Components and
Systems, vol. 38, no. 4, pp. 387–406, 2010.
[12] P. Lanusse, R. Malti, and P. Melchior, “CRONE control system design
toolbox for the control engineering community: tutorial and case study,”
Fig. 9. Results with turbulence. Top: angular speed variation; middle: control Philosophical Transactions of the Royal Society, vol. 1990, no. 371, Apr
effort; bottom: TSR. 2013.