International Conference on Renewable Energies and Power Quality (ICREPQ’14)

Cordoba (Spain), 8th to 10th April, 2014

Renewable Energy and Power Quality Journal (RE&PQJ)
ISSN 2172-038 X, No.12, April 2014

Active damping of torsional vibrations in the drive train of

a DFIG wind turbine

L. Chen1, H. Xu1,2 and J. Wenske1

1
Fraunhofer Institute for Wind Energy and Energy System Technology,
Am Seedeich 45, Bremerhaven, 27572, Germany
Phone/Fax number: +49 471 14290401, e-mail: [email protected], [email protected]
2
Institute of Electrical Engineering, Chinese Academy of Sciences,
No.6 Beiertiao, Zhongguancun, Beijing, 100190, China
e-mail: [email protected]

Abstract. Torsional vibrations in the drive train of the [2]-[3]. The generator speed is fed to the BPF for
doubly-fed induction generator (DFIG) based wind turbine can generating a torque ripple with the torsional frequency
cause large mechanical stress and reduce the life cycle of the and an appropriate phase. This ripple is added to the
components. They can be easily induced by sudden changes from torque command value and then the torsional vibration
the turbine rotor side or grid side. In this paper, a model-based can be compensated. However, if the vibration is induced
active damper of the torsional vibration designed with the linear- by the grid fault, which is much larger than the one
quadratic-Gaussian (LQG) algorithm is proposed. The modelling induced by wind variations, the BPF is not able to
of the drive train takes the flexibility of the rotor blades into
account and utilizes a three-mass model. A combination of
compensate it effectively [4].
different simulation packages, namely FAST (Fatigue, This paper gives an alternative solution, which works
Aerodynamics, Structure, Turbulence) and Matlab/Simulink even in the case of a grid fault. The main idea is that the
describing important dynamics of both the mechanical and torsional vibration can be suppressed by eliminating the
electrical side, is applied to analyse the vibrations in the drive speed difference between the turbine rotor and generator,
train and test the algorithm. Simulation results show that the if they are referred to the same side of the gearbox (e.g.
proposed active damping can suppress the torsional vibrations in in the low speed side). The LQG control algorithm with
the drive train effectively even if a gird fault occurs. the drive train model is employed for achieving better
control performance. To validate the method a
Key words sophisticated wind turbine model for simulation is built
in Matlab/Simulink. The aerodynamics and structure
Torsional vibration, Active damping, Drive train, DFIG, dynamics of the turbine is modelled with the FAST,
LQG control which has an interface to Matlab/Simulink. The electrical
parts of the turbine, including the generator, control
1. Introduction system, power electronics, protection systems and the
grid, are modelled directly with Simulink, as shown in
Wind turbines based on DFIG are the most installed wind Fig. 1. Then all details of the turbine in both mechanical
energy generation systems due to the high energy and electrical parts are simulated. Simulations show that
efficiency and low cost. Since the stator of the DFIG is the proposed method can effectively suppress torsional
directly connected to the grid, sudden changes from vibrations even in the worst-case, i.e., the turbine is
turbine rotor or gird side, such as wind gusts or voltage loaded with rated torque before a low voltage ride
sags, can induce torsional vibrations in the drive train. through (LVRT) event with a voltage sag of 100%.
They cause large mechanical stress and lead to reduction
gen DFIG (electrical part)
Grid
of lifetime of components. According to [1] gearbox- Drive
Train
related failures are responsible for over 20% of downtime Wind Te
Back-to-Back
of the wind turbines. They usually need replacement only Converter

after 6–8 years. It is necessary to take some measures to Crowbar

Control
System
mitigating torsional oscillations. For the wind turbines in
1,2,3
MW range it is difficult to increase enough damping
FAST Simulink Models
mechanically, in spite of the additional cost involved. A
common solution to this problem is bandpass filters (BPF) Fig. 1. The simulation structure of a DFIG

https://doi.org/10.24084/repqj12.309 270 RE&PQJ, Vol.1, No.12, April 2014

This paper is organized as follows: in section 2 the vqs  Rs iqs  qs  s ds (2)
complete wind turbine model is introduced in detail.
Section 3 introduces the LQG based active damping vdr  Rr idr  dr  (s  r )qr (3)
strategy. Then simulation results are shown in section 4. vqr  Rr iqr  qr  (s  r )dr (4)
At last the conclusion is conducted in section 5.
The flux and current relations are given as
2. Wind Turbine Model ds  Ls ids  Lmidr (5)
qs  Ls iqs  Lmiqr (6)
A. Mechanical model based on FAST dr  Lr idr  Lmids (7)
FAST is a comprehensive aeroelastic simulation code and qr  Lr iqr  Lmiqs (8)
is capable of predicting both extreme and fatigue loads of The generator torque is expressed as
two- and three-bladed horizontal-axis wind turbines [5]- Te  1.5 p(ds iqs  qs ids ),
[6], which is developed by NREL (National Renewable where
Energy Laboratory). The core of this code is an advanced d, q direct and quadrature axis;
model for flexible structure of the wind turbines, taking the s, r stator and rotor variables;
flexibility of the tower, blades and other components of the Rs, Rr stator, rotor resistance;
wind turbines into account. Ls, Lr, Lm stator, rotor and magnetizing inductances;
In this research, FAST has been used to model a wind  s , r , stator and rotor angular electrical speed;
turbine, considering 16 DOFs (degree of freedom)
including first and second flapwise blade mode, edgewise p, number of pole pairs.
blade mode, drive-train rotational flexibility, yaw angle, All rotor-side variables and parameters above are referred
first and second tower fore-aft bending mode, first and to the stator. Here the d-q frame is aligned with the stator
second tower side-side bending mode, and generator voltage. The stator resistance is very small and can be
azimuth angle. To define a typical large wind turbine, the ignored. Then the generator torque can be formulated by
NREL 5-MW baseline turbine [7] is utilized, which is 3 pLm
Te   idr vds . (9)
widely used and accepted for the research purpose. The 2s Ls
basic parameters are listed in Table I. The parameters of DFIG used in this study are given in
Table Ⅱ.
Table I. – NREL 5MW Wind Turbine
Table II. – DFIG Parameters
Power rating 5 MW
Rotor 3-bladed, upwind Rated voltage (line to line) 960V
Control variable speed, pitch Equivalent inertia of
Rotor diameter 126m 5.03×106kgm2
generator and gearbox (J3)
Hub height 90m Rs 2.1mΩ
Rated, cut-out wind speed 11.4m/s, 25m/s Rr 2.1mΩ
Rated rotor speed 12.1rpm Ls 4.413mH
Total rotor inertial (J1+ J2) 3.09×107kgm2 Lr 4.409mH
Gearbox ratio 97:1 Lm 4.26mH
Equivalent shaft stiffness (K23) 8.676×108Nm/rad p 3
Equivalent shaft damping (D23) 6.215×106Nms/rad
2) Control system
B. Electrical part model based on Simulink In this paper wind turbine control system includes pitch
controller and generator torque controller. The yaw
Although FAST contains detailed models in the aspects of control is ignored. The basic control objectives are to
aerodynamics and structural dynamics of a wind turbine. optimize the power production in the case of partial load
The models of electrical components and controller are (below the rated wind speed) and to limit the
very simplified. Because FAST offers an interface to aerodynamic power in the case of full load (above the
Matlab/Simulink, the electrical parts including controllers rated wind speed).
are modelled by Simulink blocks. If the turbine runs under rated power, the generator speed
is adjusted by regulating the generator torque to follow
1) DFIG subsystem an optimal torque versus speed curve. The optimal
The electrical model of the DFIG is basically a wound generator torque is proportional to the square of the
rotor induction machine. It is expressed in a d-q reference generator speed.
frame. The dynamics of the induction machine can be If the turbine runs with or above the rated wind speed, the
written as [8]. (Stator and rotor equation are both based on generator torque is held constant at the rated value and
motor convention, taking positive currents going into the the power and speed of the turbine are limited by
machine.) : controlling the pitch angle. For the safety and stability of
vds  Rs ids  ds  s qs (1) the turbine the pitch system usually has a time constant
Tservo between 0.2s and 0.3s, as shown in Fig. 2, in which

https://doi.org/10.24084/repqj12.309 271 RE&PQJ, Vol.1, No.12, April 2014

β stands for the pitch angel. Thus, even if the frequency of Therefore, it is necessary to include the crowbar device
the torsional vibration is slow as 1Hz it is difficult for the into a DFIG model for the vibration analysis.
pitch system to handle the torsional vibration. In this paper The crowbar modelled as a symmetric three phase y-
the pitch actuator is modelled by a first order lag with a connected resistance Rcrow. If the crowbar is activated, the
time constant of 0.2s with limiters of both pitch angle and rotor of the DFIG is switched from the rotor-side
change rate. converter to the external resistors Rcrow. The rotor-side
converter is bypassed and can be protected. If the grid
gen+  ref+ d fault is cleaned, the crowbar is switched off and the rotor-
 PI
1 1
side converter is connected with generator again.
Tservo dt s According to [13], small resistance values lead to high
ref current and torque transient peaks at the fault moment,
while a very high crowbar resistance can imply a risk of
K PI Gain excessive rotor current, torque and reactive power
scheduling transients when crowbar is removed. In this paper, a
Fig. 2. The pitch system constant resistance value of 80Rr will be used as an
appropriate trade-off.
3) Converter system The crowbar cannot be switched arbitrarily. The switch
A back-to-back converter system is used to connect the rules are very complicated and are out of the focus of this
DFIG rotor to the grid, as shown in the Fig. 3. The rotor- paper. In our simulation the operation of the crowbar is
side and grid-side converter have a common DC-bus. A simply controlled by the time.
DC-bus capacitor is employed to stabilize the DC voltage.
A chopper is also added to eliminate small fluctuations of 5) Grid Model
the DC voltage. The DFIG is controlled by the rotor-side The grid is represented by a Thevenin equivalent model,
converter. Meanwhile the active and reactive power of the which consists of three voltage sources connected with
stator is controlled indirectly by means of the inner rotor their impedance in series. To simulate the voltage dip, a
current control loop. The control is performed field voltage step with duration of 150ms, is applied to the
oriented, where the rotor current loop is stator voltage or voltage source.
flux orientated. The grid-side converter controls the DC
link voltage to be constant and may also control the active 3. Model-based Active Damping
and reactive power taken from or transferred to the grid. In
this paper the converter system is simulated by the average Large torsional torque will be expected due to weak
model with the space vector PWM modulation for structure damping and sudden changes from turbine rotor
investigating the impact of the electrical components on and grid side. Here a model-based active damper is
the drive train, especially during and after a grid event. designed with the LQG algorithm to adjust the generator
The detailed modelling principle can be found in [9]. torque for mitigating torsional variations.

A. Equivalent model of the drive train

In earlier work, a two-mass model was usually used to

represent the drive train mechanical system, which takes
into account only the shafts flexibilities and assumes that
the blades are rigid. The drive train is represented by two
inertias, i.e., one represents the turbine rotor and the other
stands for the generator and gearbox. The rest of the
drive train is considered as massless. However, the rotor
of modern wind turbine is large and long and cannot be
Fig. 3. The structure of the converter system [10]
simply considered as a rigid body. In the study of
electrical transient performance, it is also important to
4) Crowbar
consider the turbine rotor dynamics [14]-[15]. The reason
Nowadays the fault ride through ability becomes a basic
is that the blade edgewise symmetrical mode couples
requirement for a modern DFIG wind turbine. The new
directly to the drive train so if this mode is excited it will
grid codes force wind turbines to stay connected to the grid
also lead to the torsional vibration.
during and after a voltage dip, i.e., low voltage ride
through (LVRT). The conventional way to protect a DFIG Rigid blades
Flexible blades Generator
+hub
wind turbine from a grid fault is by installing crowbar K23
Twin K12
devices [11]-[12]. The crowbar is capable of limiting the Te
overvoltage in the DC-Link and overcurrent in the rotor- J1 J2 J3
side converter. However, the rotor circuit of the DFIG has D12 D23
to be shorted by the crowbar during the grid fault which 1 & 1 2 &  2 3 & 3
turns the DFIG into a conventional induction machine.
During generator mode transition, the generator is not Fig. 4. The equivalent model of the drive train
controllable and the torque begins to oscillate. Furthermore Hence a three-mass model, which considers both
this oscillation will be transferred to the drive train. flexibilities of drive train shaft and rotor edgewise, is

https://doi.org/10.24084/repqj12.309 272 RE&PQJ, Vol.1, No.12, April 2014

employed as shown in Fig. 4. Where J1 represents the model would be about 2.2Hz. It also implies that it is
inertia of the effective flexible part of the blades; J2 necessary to model the drive train with a three-mass
represents the inertia of the hub and the rigid part of the model.
blades; J3 stands for the equivalent inertia of the generator 80
and gearbox. K12 represents the effective blade stiffness.
K23 is the equivalent shaft stiffness referred to the low 60

speed side. D12, D23 are the damping coefficients. θ1,2,3 and
ω1,2,3 represent the rotation angle and rotation angular 40

PSD (dB)
speed of the three inertias respectively. Twin and Te
20
represent the aerodynamic and generator torque. The
dynamics referred to the low-speed shaft (LSS) is given by 0
equations (10)-(14).
-20

J11  Twin  K12 (1  2 )  D12 (1  2 ) (10)

-40
J 22  K12 (1   2 )  D12 (1  2 ) 
0 1 2
10 10 10
Frequency (Hz)
(11)
K 23 ( 2  3 )  D23 (2  3 ) Fig. 6. PSD of the LSS torque

J 33  K23 (2  3 )  D23 (2  3 )  N  Te (12) B. Active Damping with LQG
1  2  1  2 (13)
Fig. 7 shows the schematic of the proposed control
2  3  2  3 . (14)
strategy. The controller is designed to compensate the
Where N is the gearbox ratio. torsional vibration in the drive train by the modification
The natural frequencies of torsional vibration f1, f2 of the of reference of the generator torque.
three-mass model are given by (15): Te*  Tbc  Tda . (16)
1 b b2  4c 0.5 Where Tbc is the basic torque control reference calculated
f1,2  (  ) , (15)
2 2 2 from the torque-speed curve. Tda is the damping torque
where derived by the LQG, which is simply the combination of
1 1 1 1 a linear-quadratic regulator (LQR) and a Kalman filter.
b  [ K12 (  )  K 23 (  )]
J1 J 2 J 2 J3
Speed-Torque Low-pass
and curve Filter
J1  J 2  J 3
c  K12 K 23 (
). Twin
J1 J 2 J 3 Drive train
Tbc  Te* Converter and Te gen
In this three-mass model the parameters J1, J2 and K12 are and Rotor
T generator (mechanical)
unknown. But they can be dreived if f1 and f2 are known. da

Simulations are carried out to identifiy f1 and f2. The

worst-case scenario is considered here. Three dimensional
Damping
turbulent wind has been used for simulating inflow
turbulence environments which is produced by Turbsim
torque
LQR
xˆk Kalman
Controller Filter
[16]. The three phase 100% voltage dip occurs at t=7s Tˆwin
with the duration of 150ms simulated by a voltage step of Model-based damper
the voltage source in the grid model. The crowbar is Fig. 7. The generator control with the LQG damper
activated at t=7.002s and switched off at t=7.152s. The
load in the drive train, which is represented by the LSS The linear drive train model developed from (10-14) can
torque, is shown in Fig. 5. be expressed in the state space form as:
 x  Ax  Bu  Bd ud

7000

 . (17)
 y  Cx
6000


5000
x is a vector consisting of five system states:
LSS Torque (kNm)

x  [1  2 1  2 2  3 2  3 3 ]T .
4000

3000
u represents the control input Tda and ud is the disturbance
2000
input Twin. 3 is the generator speed, which is the only
1000
measurement of the system (17). The other matrices and
0
0 5 10 15 20 25
vectors are shown in the following:
Time (sec.)
T
 N N
Fig. 5. The LSS torque with a LVRT event B  0 0 0   ,
 J3 J3 
According to the power spectral density (PSD) analysis of T
1 
LSS torque, shown in Fig. 6, two dominant oscillation Bd   0 0 0 0 ,
frequencies can be extracted at about 1.7Hz and 4Hz,  J1 
while the torsional frequency according to the two-mass

https://doi.org/10.24084/repqj12.309 273 RE&PQJ, Vol.1, No.12, April 2014

 C  [0 0 0 0 1], sampling times are applied. The smallest sampling time
is 0.25ms, which is the sampling time of the space vector
 K12 K12 K 23 
0  J  J 0 0 PWM modulation. The measurement and controllers
J2 work with 1ms and 5ms, respectively. The FAST has two
 1 2 
1 0 0 0 0 time constants for integrations, 10ms for the
  aerodynamics and 5ms for the structure dynamics. Then
K12 K 23 K 23
A  0 0   0 . good balance between the accuracy and simulation time
 J2 J2 J3  can be achieved. The damping torque Tda is limited by
  10% of the rated torque. Furthermore, the proposed LQG
0 0 1 0 0
damper is also compared with the conventional two-mass
 K 23  model based LQG.
0 0 0 0 Fig. 8 shows the speed of the generator. The speed with
 J3  active damping oscillated much less. There are not so
much differences of damping performance of two-mass
Because the damping factors are very small and usually and three-mass mass model.
unknown, they are assumed to be zero here. It actually
makes the active damping of the model is more difficult 1280
no damping
than the real plant. 1260 LQG three-mass
LQG two-mass
The LQR algorithm usually uses a performance index J to 1240

Generator speed (rpm)

define the controller objectives, which is a quadratic 1220
function and can be expressed as:
1200

J    xT (t )Qx(t )  uT (t ) Ru (t )  dt. (18) 1180
0
Where Q is a symmetric, positive semi-definite weighting 1160

matrix on the states that satisfies the algebraic Riccati 1140

equation (ARE) and R is a symmetric, positive definite 1120
weighting on the control input. The matrices are adjusted
1100
based on a trade-off between high control performance 0 5 10
Time (sec.)
15 20 25

(large Q) and low control output (large R). For the Fig. 8. Generator speeds with/without damping
simplicity they are usually set as diagonal. With the LQR
theory if the optimal input can be found, the index function The characteristics of the loads in the drive train in time
J can be minimized, which means all state variables will domain and frequency domain are shown in Fig 9. and
tend to zero. In our case only the states x1 and x3 should be Fig.10., respectively. It can be observed that the LSS
controlled to zero so they have much larger weighting torque also oscillated significantly less if the damping
factors then others. The appropriate Q and R can be torque was activated. But according to the PSD in Fig. 10
obtained by running simulations.
7000
The LQR needs all information of the system. A discrete no damping
Kalman filter is used here, which is robust to measurement 6000 LQG three-mass
LQG two-mass
and system noises. It makes a one-step-ahead prediction of
5000
the states, and also estimates what the measured output
LSS Torque (kNm)

would be. A correction updates the state estimates, taking 4000

into account the prediction error. The algorithm is
3000
iterative and the correction factor, Kalman gain Kk, updates
at every iteration. Assuming the stochastic distances 2000
acting on the system are Gaussian, Kk is derived from the
1000
system dynamics to minimize the expected sum of squares
of the prediction error. For reducing the online 0
0 5 10 15 20 25
computational cost the stationary Kalman filter is applied, Time (sec.)
which means the Kalman gain is constant. Thus the Fig. 9. LSS torques with/without damping
estimator has the structure as shown in (19)
xˆk 1  xˆk  Kk ( yk  yˆk ). (19) 80

Where yk stands for the measurement. no damping

60 LQG three-mass
The system is subject to the unknown input Twin. It is LQG two-mass
approximated by 40
Tˆwin  Te  ( J1  J 2 )ˆ gen , (20)
PSD (dB)

20
where ˆ gen is the numerical differentiation of the generator
0
speed filtered by a first order lowpass filter.
-20
4. Simulation
-40 0 1 2
10 10 10
The proposed active damper is tested by simulations. The Frequency (Hz)

test scenario is same as described in the section 3. Multiple Fig. 10. PSD of LSS torques with/without damping

https://doi.org/10.24084/repqj12.309 274 RE&PQJ, Vol.1, No.12, April 2014

the three-mass model based damper can mitigate the both [4] F. Zhang, W. Leithead, and O. Anaya-Lara, "Wind turbine
torsional vibrations, while the two-mass model based control design to enhance the fault ride-through
damper can only suppress the vibration with the low capability," in Renewable Power Generation (RPG 2011),
frequency and intensifies the other one. IET Conference on, 2011, pp. 1-6.
[5] J. M. Jonkman and M. L. Buhl Jr, "FAST user’s guide,"
The price for the improved system dynamics is the slightly Golden, CO: National Renewable Energy Laboratory,
increased load at very low frequency area, because the 2005.
LQG is trying to force all system states tend to zero. Fig. [6] R. Fadaeinedjad, M. Moallem, and G. Moschopoulos,
11 shows the torque of the generator. Due to the additional "Simulation of a wind turbine with doubly fed induction
damping torque the generator is a little bit overloaded, generator by FAST and Simulink," Energy Conversion,
which forces the power electronics devices to be IEEE Transactions on, vol. 23, pp. 690-700, 2008.
overloaded. Usually the 10% is under the tolerance range [7] J. M. Jonkman, S. Butterfield, W. Musial, and G. Scott,
of the electric devices. Therefore, it will not bring extra Definition of a 5-MW reference wind turbine for offshore
hardware cost for the implementation. system development: National Renewable Energy
Laboratory Colorado, 2009.
[8] B. Wu, Y. Lang, N. Zargari, and S. Kouro, Power
no damping
LQG three-mass
conversion and control of wind energy systems: Wiley-
50
LQG two-mass IEEE Press, 2011.
Generator torque (kNm)

[9] F. Iov, A. D. Hansen, P. Sørensen and F. Blaabjerg,

45
“Wind Turbine Blockset in Matlab/Simulink”, Report,
40 Aalborg Univeristy, 2004.
[10] A. Perdama, “Dynamic Models of Wind Turbines”, Ph. D
35 thesis, Chalmers University of Technology, 2008.
[11] R. Lohde, S. Jensen, A. Knop and F. W. Fuchs, "Analysis
30 of three phase grid failure and doubly fed induction
generator ride-through using crowbars," in Power
25
Electronics and Applications, 2007 European Conference
8 9 10 11 12 13 14 15 16
Time (sec.) on, 2007, pp. 1-8.
Fig. 11. Generator torques with/without damping (zoomed) [12] L. G. Meegahapola, T. Littler, and D. Flynn, "Decoupled-
DFIG fault ride-through strategy for enhanced stability
performance during grid faults," Sustainable Energy,
5. Conclusion IEEE Transactions on, vol. 1, pp. 152-162, 2010.
[13] A. D. Hansen and G. Michalke, "Fault ride-through
This paper proposed a model-based damper of the capability of DFIG wind turbines," Renewable energy,
torsional vibrations of the drive train. The modelling of the vol. 32, pp. 1594-1610, 2007.
drive train takes the symmetrical mode of the blades into [14] G. Ramtharan, N. Jenkins, O. Anaya-Lara and E.
account and a three-mass model is applied. The unknown Bossanyi, "Influence of rotor structural dynamics
representations on the electrical transient performance of
parameters can be derived by the frequency analysis of the
FSIG and DFIG wind turbines," Wind Energy, vol. 10, pp.
load in the drive train. The LQG algorithm is employed to 293-301, 2007.
calculate the damping torque. For the active damping only [15] H. Li and Z. Chen, "Transient stability analysis of wind
needed measurement of the drive train system is the turbines with induction generators considering blades and
generator speed. The active damper is tested with a shaft flexibility," 33rd Annual Conference of the IEEE in
sophisticated wind turbine model realized by different Industrial Electronics Society, 2007, pp. 1604-1609.
software, in which most important dynamics in mechanical [16] B. J. Jonkman, TurbSim user's guide: Version 1.50:
parts and electrical parts are modelled. The results show National Renewable Energy Laboratory Colorado, 2009.
that the torsional vibrations of the drive train have two
dominant frequencies vibrations. The three-mass model
based LQG damper can supress the both effectively. If the
standard two-mass model is applied, there is the risk that
the torsion oscillation with another frequency can be
increased. The proposed damping method does not add
any additional hardware to the system and need less online
computation. Thus, it can be easily implemented in
practice.

References
[1] P. Asmus and M. Seitzler, “The wind energy operation &
maintenance report,” Nat. Wind Watch, Rowe, MA, Feb.
2010.
[2] E. Bossanyi, “Wind turbine control for load reduction,
Wind Energy, vol. 6, no. 3, pp. 229-244, 2003.
[3] A. Nasiri, E. Muljadi, G. Mandic, and F. Oyague, "Active
Torque Control for Gearbox Load Reduction in a Variable
Speed Wind Turbine," IEEE Transactions on Industry
Applications, vol. 48, no. 6, pp. 2424-2432, 2012.

https://doi.org/10.24084/repqj12.309 275 RE&PQJ, Vol.1, No.12, April 2014