Mechanism and Machine Theory 98 (2016) 2–20

Contents lists available at ScienceDirect

Mechanism and Machine Theory

journal homepage: www.elsevier.com/locate/mechmt

Wind turbine drive train dynamic characterization using

vibration and torque signals
P. Srikanth, A.S. Sekhar ⁎
Department of Mechanical Engineering, Indian Institute of Technology Madras, Chennai 600036, India
Department of Mechanical Engineering, IIT-Madras, Chennai 600036, India

a r t i c l e i n f o a b s t r a c t

Article history: Dynamic analysis of wind turbine drive train subjected to stochastic aerodynamic loads is car-
Received 2 June 2015 ried out in the present study. The longitudinal wind speed at the turbine site normally consists
Received in revised form 26 September 2015 of a mean value superimposed with ramp, gust and turbulence components. In the present
Accepted 25 November 2015
study, the aerodynamic torque is obtained by considering wind speed parameters of a typical
Available online xxxx
wind turbine site. The dynamic model accounts for time varying gear mesh stiffness, bearing
elasticity and torsional shaft stiffness. The dynamic analysis is done with stochastic aerodynam-
Keywords: ic loads and the vibration responses are obtained in time and frequency domains. It is observed
Wind turbine
that the entire spectral content of the vibration signals is confined to low frequency region,
Drive train
whereas higher frequencies are hidden. In order to capture the hidden frequency information
Nonlinear non-stationary signal
Turbulence from vibration signals, the wavelet decomposition technique is used. The dynamic analysis
Internal resistive torque using torque signals is also discussed. The present study shows that, from the internal resistive
Dynamic transmission error torque all the characteristic frequencies can be clearly observed.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

The installation capacity of wind turbines is increasing significantly year by year. A wind turbine drive train is a typical case of
rotating equipment operating at slow speeds and subjected to severe stochastic loads. The wind turbine components are always
affected by uncertain loads. Such kind of load spectrum always leads to vibration signals which are nonlinear and non-stationary
in nature [1]. The dynamic analysis and condition monitoring of most of the rotating machinery focus on rotating components
that are subjected to constant, periodic or transient load spectra. These are relatively easier for modeling and analysis, compared
to the wind turbine drive trains. Wind turbines are subjected to wind loads which are stochastic in nature. Due to these loads, the
chances of failures are higher for critical components, such as gearbox. The downtime and cost associated with the gearbox is
more, as handling of this component is difficult. Hence, detailed predictive condition monitoring strategies have to be developed
for such components. A huge step-up speed in the range of 40:1 to 135:1 can be achieved from rotor to generator in wind turbine
using epicyclic gearbox [2]. Qin et al. [3] carried out wind turbine gearbox dynamic analysis with flexible multi-body modeling
technique, taking into account the elastic strain energy due to gears and bearings. Abboudi and Walha [4] carried out dynamic
analysis of a two stage external gearbox of wind turbine using an empirical approach based on aerodynamic torque. Guo et al.
[5] modeled gravity effect on the vibration response of wind turbine planetary gears and it is concluded that the gravity plays
a vital role when compared to gear tooth meshing excitation alone. These results compared well with those of mathematical
and experimental models. Rigid multibody modeling with discrete flexibility approach is used by Todorov et al. [6] to assess
the torsional dynamic behavior of wind turbine drive train. The dynamic behavior of wind turbine gearbox is evaluated using

⁎ Corresponding author at: Department of Mechanical Engineering, IIT-Madras, Chennai 600036, India. Tel.: +91 44 2257 4709 (O); fax: +91 44 2257 4652.
E-mail addresses: srikanth.poludasu1985@gmail.com (P. Srikanth), as_sekhar@iitm.ac.in (A.S. Sekhar).

http://dx.doi.org/10.1016/j.mechmachtheory.2015.11.013
0094-114X/© 2015 Elsevier Ltd. All rights reserved.
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 3

three techniques namely pure torsional, rigid multi-body modeling and flexible multi-body modeling [7,8]. Here, finite element approach
and test rig experiments are used. By taking variable gear mesh stiffness into account, Shi et al. [9] studied the torsional dynamic behavior
of the entire wind turbine drive train subjected to constant aerodynamic torque excitation using Lagrange's approach.

Notations
Ar, Ag ramp and gust amplitudes
C1, C2, C3 low, intermediate and high speed shaft torsional stiffness.
Cp(λ, β) power coefficient
H1(s), H2(s) spatial and rotational sampling filter
I turbulence intensity factor of wind
Ji (i = 1 … 8) inertias of rotor, carrier, planets, sun, gear1, gear2, gear3, and generator
kri, ksi, kpg time varying mesh stiffness of ring–planet, planet–sun and gear–pinion.
Kbj (j = 1 … 6) stiffness values of bearings
L1, L2 length scales of the Kaimal and von Karman spectra
[M], [C(t)], [K(t)] mass, damping, and stiffness matrices
R radius of rotor
S(f) auto spectral density function
Ti, Ts initial and stop times
TR(t), TG(t) rotor and generator torque
u(t), U, ur(t), ug(t), ut(t) total, average, ramp, gust and turbulence components of wind speed
yi, zi, ys, zs, y1, z1, y2, z2, y3, z3 translational degrees of freedom in y & z directions of planets, sun, gear1, gear2, and gear3.
σ wind speed standard deviation
γ decay factor
λ tip speed ratio
ϕ, θ, β incident angle, angle of attack, and pitch angle
ωr, ωd rotor speed and rotor damping factor
ωM ðpÞ , ωM ðg12Þ , ωM ðg34Þ gear mesh frequency of ring–planet and planet–sun, gear1–2, gear2–3
δri, δsi, δpg dynamic transmission errors of ring–planet, planet–sun, gear–pinion
ϕi (i = 1 … 8) Absolute rotational angles of rotor, carrier, planets, sun, gear1, gear2, and gear3, and generator absolute rotational
angles.

Coupled torsional bending dynamic analysis is carried out for fixed speed wind turbine using Lagrange's approach in [10].
Here, constant gear mesh stiffness, support bearing elasticity and strain energy associated with shafts are used in the formulation.
Aerodynamic torque is modeled as a periodic signal based on the empirical relation. Harmonic balance method is used by Ji et al.
[11] to obtain the dynamic responses of the wind turbine gearbox. Techniques like finite element analysis and experimental
modal analysis are adopted by Haijun et al. [12] for estimating the natural frequencies and mode shapes. Lumped parameter dy-
namic model of the wind turbine gearbox is developed by Long Quan et al. [13] to estimate the vibration levels. Yongqian et al.
[14] elaborated the effect of design parameters on the sensitivity of the natural frequency and dynamic characteristics with re-
spect to gear mesh stiffness in epicyclic wind turbine gearbox.
The details of different wind components existing at wind turbine sites are revealed in [15]. Different wind speed models
(Weibull, extreme value distribution of type 1 & type 2 and Rayleigh) are proposed by Jang and Lee [16] for the Taiwan area.
The random nature of wind is normally represented by the Weibull distribution model, four component composite model, auto
regressive and moving average model and power spectral model [17]. Near the wind turbine site, the best approximation to rep-
resent wind is the Weibull distribution; this has been concluded based on wind data, over a period of 44 years [18]. There always
exist deterministic and stochastic components in wind loads [19]. The main reason for gearbox failures is the wind stochastic load
that always creates uneven stochastic aerodynamic torque on the rotor. These loads are distributed unevenly between the bear-
ings and gears [20].
Zhu et al. [21] using the commercial SIMPACK software carried out the dynamic analysis of the wind turbine drive train by
considering external excitations due to load spectrum and internal excitations due to the time varying mesh stiffness and trans-
mission errors. These results are compared with the experimentally measured vibration acceleration signals. A simplified method
is proposed to estimate the long-term extreme value of the gear transmitted load. This value is estimated based on the cumulative
Weibull distribution for one hour mean wind speed [22]. The need to consider the effects of mesh stiffness and impact stress on
the dynamic transmission error evaluation of the wind turbine planetary gear system is revealed by Zhao et al. [23]. Wei et al. [24]
analyzed the effects of uncertainty in the gearbox system parameters like mesh stiffness, damping and transmission error on the
dynamic response of the system. In literature, dynamic analysis of the wind turbine drive train is performed by subjecting the
drive train to constant or periodic load components only. Srikanth and Sekhar [25] modeled the wind turbine drive train by in-
cluding the coupled torsional bending dynamics. Here, the wind turbine is subjected to stochastic loads which are estimated
based on the Danish standard DS 472. In addition, time varying mesh stiffness of gear, stiffness of shaft and bearing are considered
in the dynamic formulation. Although, the standard model is used, the obtained signals from this method are not close to realistic
signals on site. Hence, the authors further improved the wind modeling and carried out the drive train analysis in the present
study.
4 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

Unlike in the literature, in the present work, the wind loading acting on the turbine is modeled in detail where the simulated
wind or aerodynamic torque signal almost represents the realistic one. Further the dynamic characterization of drive train with
torque signals is discussed. To the best of the authors' knowledge, drive train characterization using torque signals is not done
in literature. The methodology used in the present study for dynamic characterization of wind turbine drive train is discussed
in the following subsection.

1.1. Methodology

The wind speed at the hub level of rotor generally consists of average, ramp, gust and stochastic (turbulence) components. In
order to capture the wind speed variation over the rotor swept area, the wind speed signal at the hub level has to be passed
through aerodynamic filters. The signal at hub level first pass through spatial filter from which high frequency components pres-
ent in the wind are eliminated; later the output of wind speed from spatial filter has to be passed through a rotational sampling
filter. The phenomenon of rotational sampling in the case of wind turbines can be explained as follows. Typically, the wind at the
wind turbine site consists of mean and turbulence components in it. The apparent wind speed that the blade is subjected to al-
ways depends upon the position of the blade. This phenomenon is called ‘rotational sampling’, because the turbulence spectrum
of wind is amplified at multiples of rotor running frequency. The effect of this rotational sampling is very important for the wind
turbine drive train dynamic analysis. The aerodynamic torque is obtained from the output wind signal of the rotational sampling
filter. The coupled torsional bending model developed by the authors in [25] is considered in the present study. The dynamic anal-
ysis of this drive train model subjected to stochastic aerodynamic torque is carried out. The equations of motion are solved for
vibration responses in both time and frequency domains. It is observed that the entire frequency spectrum is confined to the
low frequency region only and all the frequency components are not seen clearly from the FFT. Hence, in order to extract the hid-
den high frequencies, the wavelet packets are used on the vibration signal from which all the characteristic frequencies are ex-
tracted. The methodology is summarized in the flow chart given in Fig. 1.
Later torque signals are also used for drive train dynamic characterization. However, the external torque (output torque) fol-
lows a similar trend as the vibration signal, hence one can use wavelet packets to extract all the hidden information. The internal
resistive torque signals are also used for the drive train dynamic characterization where all the characteristic frequencies are seen
from the FFT itself. The procedure of estimating the gear tooth contact force and internal resistive torque by estimating the dy-
namic transmission error is discussed.

Turbulence longitudinal wind speed component generation based on PSD

Obtain total wind speed by adding average, ramp, gust components to turbulence component

Pass the total wind speed through spatial filter

Pass the spatial filter output through rotational sampling filter

Obtain aerodynamic torque signal from rotational sampling filter output

Perform dynamic analysis

Obtain Vibration Signals Obtain Torque Signals

External torque Internal resistive torque

Use Wavelet packets to extract all characteristic All characteristic frequencies can be
frequencies observed clearly from FFT itself.

Fig. 1. Flow chart for the drive train dynamic characterization.

 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 5

0
10

Von karman model

Power spectral density

-1
10

-2
10 -2 -1 0 1
10 10 10 10
Frequency (Hz)

Fig. 2. Power spectral density of the von Karman model.

2. Aerodynamic torque modeling

Wind speed variations can be represented as long term, annual and seasonal, synoptic and diurnal, turbulence components in
wind resource at the wind turbine site. The long term variation in wind means the amount of variation in wind from one year to
another or even over periods of decades. It is difficult to predict such kind of wind variations at the wind turbine site. It is hard to
predict the year to year annual variations. However, during the year, wind speed variations can be represented well with a prob-
ability distribution. The hourly mean wind speed over the year follows the Weibull distribution at the wind turbine site. Repre-
sentation of hourly mean wind speed by a probability distribution clearly indicates the random variation in it. Wind speed
variations are more random and less predictable on a shorter time scale compared to seasonal variations. These variations have
frequency content peaks at around 4 days. Weather fronts associated with high and low pressure areas are the cause for such var-
iations. On the other hand, diurnal variations are associated with a frequency spectrum that has peak at a frequency of 24 h. This
is due to the heating and cooling of air in a day. It is the best way to represent wind as a time series with a mean value
superimposed by the turbulence fluctuations averaged over a 10 min period. The mean component is accounted for through sea-
sonal, synoptic and diurnal effects [26].
The longitudinal wind acting at the wind turbine site always consists of basically four components, as given in the following
equation,

uðt Þ ¼ U þ ur ðt Þ þ ug ðt Þ þ ut ðt Þ; ð1Þ

where u(t) = total longitudinal wind speed; U = mean wind speed; ur(t) = ramp component; ug(t) = gust component; and
ut(t) = turbulence component.
The ramp component is expressed in terms of initial time (Ti), stop time (Ts) and ramp amplitude (Ar) as given in [27]. The
ramp component does not have much physical meaning except that it always tries to increase the wind speed mean value.

0 1
0 t b Ti
B t−T i C
B Ti b t b Ts C
ur ðt Þ ¼ @ Ar
T s −T i A: ð2Þ
0 t NT s

Table 1
Wind speed parameters and blade data.

Parameter Value

Mean speed of wind (U) 6 m/s

Turbulence intensity (I) 0.10
Length Scale (L2) 64.5 m
Ramp amplitude (Ar) 4 m/s
Gust amplitude (Ag) 7 m/s
Initial time (Ti) 2.5 s
Stop time (Ts) 7.5 s
Rotor damping factor (ωd) 0.8765 rad/s
Rotor Radius (R) 28 m
Pitch Angle (β) 2°
6 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

4
Single-Sided Amplitude Spectrum
10

(a)

power spectral density

2
10

X: 0.02686
X: 0.05737 X: 0.1184
0 Y: 10.24
Y: 6.561
10 Y: 5.946

X: 0.2551
Y: 0.57

-2 X: 0.8997
10 Y: 0.03985 X: 1.798
Y: 0.02129

-4
10 -3 -2 -1 0 1
10 10 10 10 10
Frequency (Hz)
Single-Sided Amplitude Spectrum
4
10
Gain of nearly one
at all freqencies
(b)
Power spectral density

2
10 Amplification around rotor characteristic
X: 0.02686
Y: 10.21 X: 0.05737 X: 0.1184 frequency and harmonics
Y: 6.504 Y: 5.619

X: 0.2551
0 Y: 0.5498
10
X: 0.8997
Y: 0.06585 X: 1.798
Y: 0.0254

-2
10

-4
10 -3 -2 -1 0 1
10 10 10 10 10
Frequency (Hz)

Fig. 3. Power spectral density (PSD) of wind signal (a) before filtering and (b) after filtering.

The gust component of the wind is a function of initial, stop time and the gust amplitude which is expressed in terms of a
deterministic function as given in [27],
0 1
 0  t b Ti
B Ag t−T C
ug ðt Þ ¼ B
@ 2 1− cos 2π T −T
i
Ti b t b Ts C
A: ð3Þ
s i
0 t N Ts

However, the gust component is in general, stochastic in nature at the wind turbine site under severe loading conditions.
In general, the turbulence component of wind is expressed in terms of power spectral density. The turbulence component of
wind follows either the Kaimal or von Karman spectrum at the wind turbine site. The power spectral density associated with the
longitudinal wind speed component which is represented in the Kaimal model in terms of frequency, length scale and average
wind speed is given as [27],

Sð f Þ 4L1 =U
¼ ; ð4Þ
σ2 ð1 þ 6 f L1 =U Þ5=3

Fig. 4. Cross-section of wind turbine blade.

 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 7

where S(f), L1, U, f and σ represent the auto spectral density function, length scale, average wind speed, frequency and standard
deviation respectively.
The length scale depends upon surface roughness and height above the ground. The longitudinal wind speed component
standard deviation (σ) is constant with respect to height above the ground. However, this standard deviation depends upon
the turbulence intensity (I) and average wind speed (U). The turbulence intensity, I is given by

σ
I¼ : ð5Þ
U

The wind turbulence can also be represented by the von Karman spectrum especially for complex terrain, whose power
spectral density can be represented in the following form as given in [27],

Sð f Þ 4L2 =U
¼ 5=6 : ð6Þ
σ2 1 þ 70:8ð f L2 =U Þ2

4
Ramp component of wind
3
(a)
velocity (m/s)

0
0 1 2 3 4 5 6 7 8 9 10
Time (S)

6
Gust component
velocity (m/s)

5 of wind
(b)
4

0
0 1 2 3 4 5 6 7 8 9 10
Time (S)

20

(c)
Wind speed (m/s)

15

10

0
0 1 2 3 4 5 6 7 8 9 10
Time (S)
6
x 10
2
Aerodynamic Torque ( N-m)

1.5
(d)

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Time (S)

Fig. 5. Components of wind and obtained aerodynamic torque (a) ramp component, (b) gust component, (c) total wind speed (after filtering) and (d) aerodynamic
torque.
8 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

Fig. 6. Schematic diagram of wind turbine drive train.

Table 2
Drive train data.

J1 — inertia of the rotor (kg m−2) 4.18 × 106

J2 — inertia of the carrier (kg m−2) 57.72
mp — mass of the planet (kg) 57.79
J3 — inertia of the planet (kg m−2) 1.12
J4 — inertia of the sun (kg m−2) 0.86
J5 — inertia of gear1 (kg m−2) 14.32
J6 — inertia of the gear2 (kg m−2) 1.62
J7 — inertia of the gear3 (kg m−2) 0.2
J8 — generator inertia (kg m−2) 93.22
C1 — low speed shaft stiffness 7.19 × 107
C2 — intermediate shaft stiffness 1.4 × 107
C3 — stiffness of the high speed shaft 0.15 × 107
Kri — ring–planet gear mean mesh stiffness 0.73 × 108
Ksi — sun–planet gear mean mesh stiffness 0.73 × 108
Kg12 — gear1–2 mean mesh stiffness 2.02 × 109
Kg23 — gear2–3 mean mesh stiffness 0.11 × 108
K1 — stiffness of bearing1 8.04 × 109
K2 — stiffness of bearing2 4.08 × 109
K3 — stiffness of bearing3 4.08 × 109
K4 — stiffness of bearing4 2.16 × 109
K5 — stiffness of bearing5 2.16 × 109
K6 — stiffness of bearing6 2.16 × 109
Crp — contact ratio of ring–planet gear pair 1.9342
Csp — contact ratio of sun–planet gear pair 1.6242
Cg12 — gear1–2 contact ratio 1.6616
Cg23 — gear 2–3 contact ratio 1.5984
rc — radius of carrier (mm) 270
r3 — radius of planet (mm) 160
r4 — radius of sun (mm) 110
r5 — radius of gear1 (mm) 290
r61 — 2nd stage gear2 radius (mm) 95
r62 — 3rd stage gear2 radius (mm) 185
r7 — gear3 radius (mm) 80
α — pressure angle 20
Gear mesh frequency of ring–planet and planet–sun gears (Hz) 25.2
Gear1–2 gear mesh frequency (Hz) 83.8
Gear2–3 gear mesh frequency (Hz) 326.5
Gear ratio 34.654
Cr — root chord (m); Ct — tip chord (m) 3.3; 0.9
Ω = speed of rotor (rpm) 18 (0.3 Hz)
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 9

Table 3
Characteristic frequencies of wind turbine drive train.

Characteristic frequency component Frequency (Hz)

Gust frequency 0.2

First torsional natural frequency 0.4
Rotor characteristic frequency (three times the rotor rotational speed) 0.9
Second natural frequency 4.7
Gear mesh frequency of ring–planet and planet–sun gear 25.2
Gear mesh frequency of gear1–2 83.8
Gear mesh frequency of gear 2–3. 326.5

With the Kaimal model, three dimensional spatial wind components can be well represented and matched with experimental data.
However, the von Karman model fits better than the Kaimal model in the sense of power levels. Hence, the von Karman model is used in
the present study. Fig. 2 represents the power spectral density of the von Karman model used for the generation of stochastic process.

100

(a)
50
Response (rad/s2)

-50

-100
0 1 2 3 4 5 6 7 8 9 10
Time (S)

300

200 (b)
Response (rad/s2)

100

-100

-200

-300
0 1 2 3 4 5 6 7 8 9 10
Time (S)

800

600 (C)
Response (rad/s2)

400

200

-200

-400

-600
0 1 2 3 4 5 6 7 8 9 10
Time (S)

800

600 (d)
Response (rad/s2)

400

200

-200

-400

-600
0 1 2 3 4 5 6 7 8 9 10
Time (S)

Fig. 7. Time domain representation of vibration acceleration response: (a) gear1, (b) gear2, (c) gear3 and (d) generator.
10 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

The standard method of generating a stochastic process, the Shinozuka method [28] is used to obtain the wind stochastic time
series from the turbulence spectrum of the power spectral density. In this method, at each point in the time series, the wind
speed can be obtained by using weighed cosine series and a random phase angle, as given by

XN qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ut ðt Þ ¼ f
n¼1
Sð f n ÞΔf cosð2π f n t þ ϕn þ ΔϕÞ; ð7Þ

where ϕn is the uniformly distributed random phase angle, Δf is the incremental frequency, and fn is the frequency in an array
which is formed by dividing the frequency range of power spectral density model into Nf number of incremental frequencies.
The above approach takes into account the turbulence component of wind speed only; however, the ramp and gust compo-
nents of wind are also taken into account in the present approach by adding average, gust and ramp components (see

Single-Sided Amplitude Spectrum of Gear1

20

(a)
15
Amplitude

10

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)
Single-Sided Amplitude Spectrum of Gear2
50

40 (b)
Amplitude

30

20

10

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)
Single-Sided Amplitude Spectrum of Gear3
140
X: 5.127
Y: 135
120
(c)
100
Amplitude

80

60

40

20

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Single-Sided Amplitude Spectrum of Generator

140

120 (d)
100
Amplitude

80

60

40

20

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Fig. 8. Frequency domain representation of vibration acceleration response: (a) gear1, (b) gear2, (c) gear3 and (d) generator.
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 11

Single-Sided Amplitude Spectrum of Gear3

140

120

100

Amplitude 80

60

40

20

0
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)

Fig. 9. Vibration signal of gear3 at a frequency range of 0–10 Hz.

Eqs. (2) & (3)) to the wind signal obtained in Eq. (7). Hence, the longitudinal wind speed component (Eq. (1)) is the algebraic
sum of the average, ramp, gust and turbulence components.
The obtained wind signal based on the above approach represents closely, the realistic wind speed variation with time at the
wind turbine site. However, the above wind signal represents wind speed time series at the hub level only. The longitudinal wind
speed is not constant along the blade length and over the rotor swept area. Hence, the influence of wind speed variation over the
swept area has to be accounted for in evaluating the mechanical aerodynamic torque. The procedure of accounting this is sum-
marized below. The wind parameters selected for the aerodynamic torque simulation are given in Table 1.
To account for the effect of wind speed variation over the rotor swept area, aerodynamic filters are used in the present work
based on literature [27]. The obtained wind speed signal at the hub level is the input to these aerodynamic filters and the output
signal is the wind signal that accounts for the effect of rotor swept area. The wind speed signal obtained at the hub level is passed
first through the spatial filter by which the high frequency components present in the wind are dampened. Then the signal has to
be passed through rotational sampling filter by which the effect of wind variation over the rotor swept area can be captured.
The spatial filter is represented by the Laplace transfer function as given by,
pffiffiffi
2 þ bs
H1 ðsÞ ¼ pffiffiffi pffiffiffi pffiffiffi  ; ð8Þ
2 þ bs a 1 þ b= as

where a = 0.55, b = γR/U, R is the rotor radius, U is the average speed of wind at the hub height and γ is the disc decay factor (in
general γ = 1.3). The rotational sampling filter can be represented with the Laplace transfer function as given by,

2 2 2
ωr þ ωd ðs þ ωr Þ
H2 ðsÞ ¼ ; ð9Þ
ωr 2 ðs þ ωd þ jωr Þðs þ ωd −jωr Þ

where ωr = 2πnN/60 and ωd is the rotor damping factor at a rotor speed of ωr, N is number of blades and n is the speed of the
rotor.
The procedure for selecting the damping factor for rotational sampling filter as given in [29] is summarized here. Petru [29]
used shaft torque obtained from an advanced and simplified approach through aerodynamic filters to estimate the damping factor
in the rotational sampling filter of a typical wind turbine. He observed that in a frequency range of 0.5 to 2.5 Hz, the wind speed
amplification by the filter is higher. The variation of damping factor with mean wind speed has been plotted. In the present study,

Fig. 10. Schematic representation of wavelet decomposition of vibration signal up to the 3rd level.
12 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

-3
x 10 Single-Sided Amplitude Spectrum
2
X: 326.7
Y: 0.001696
Gear mesh frequency of gear 2-3
1.5

Amplitude
1

0.5

0
0 50 100 150 200 250 300 350 400
Frequency (Hz)

Fig. 11. FFT of the decomposed signal at the 3rd level in frequency band 312–375 Hz.

an approximate damping factor is assumed from the aforesaid plot for simulation and is indexed in Table 1. The significance of the
rotational sampling frequency is explained in the following paragraph.
Fig. 3 represents the wind speed signal power spectral density (PSD) obtained at the fixed point (preferably hub level) and
PSD of the wind speed signal after passing through aerodynamic filters. It can be clearly observed that the filter gain is nearly
equal to one at the lower frequencies and around the high frequency region, frequency content is clearly dampened. The ampli-
fication in PSD value is observed at around rotor characteristic frequency and its higher harmonics.
In literature, the more detailed way of modeling rotational wind speed turbulence for large wind turbines is discussed and
some of which is summarized here. Rotational wind speed turbulence through shaping filters and correlation techniques is pro-
posed for large scale wind turbines in [30]. From the numerical results, it is observed that good approximation is found between
the rotational sampling spectrum by correlation technique and synthesized shaping filter. The effect of location along the blade on
the rotational sampling effect is also captured. It is also illustrated that for the calculation of aerodynamic torque the use of PSD at
four fifths of the rotor radius is recommended.
The rotor aerodynamic power Pa is given by [27],

1 2 3
Pa ¼ πρR u C p ðλ; βÞ: ð10Þ
2

The power coefficient (Cp(λ, β)) is specific to a given turbine that can be defined as the ability of the turbine to convert its
kinetic energy into the mechanical energy. The power coefficient is a function of pitch angle β and the tip speed ratio λ. Fig. 4
shows a typical cross-section of the wind turbine blade, where ϕ, θ, and β represent incident angle, angle of attack and pitch
angle respectively. The maximum achievable value of Cp for wind turbine is 0.59. Generally in practice, the maximum achievable
Cp is 0.5 for two bladed high speed turbines and 0.2–0.4 for low speed multi-bladed turbines [31].
The coefficient of performance Cp can be estimated using the following empirical relation as given in [31].

1 2

−0:17λ
C p ðλ; βÞ ¼ λ−0:022β −5:6 e : ð11Þ
2

Single-Sided Amplitude Spectrum

0.14

0.12 Gear mesh freqeuncy ring-planet,

planet-sun (25.2 Hz)
0.1
Amplitude

0.08

0.06

0.04

0.02

0
0 5 10 15 20 25 30 35
Frequency (Hz)

Fig. 12. FFT of decomposed signal at the 5th level at a frequency band of 15.625–31.25 Hz.
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 13

-3
x 10 Single-Sided Amplitude Spectrum
1.5

Gear mesh freqeuncy of gear1-2 (83.8 Hz)

1

Amplitude

0.5

0
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)

Fig. 13. FFT of decomposed signal at the 6th level at a frequency band of 85.94–93.75 Hz.

The tip speed ratio (λ) can be defined as,

ωR R
λ¼ ; ð12Þ
u

where R is the rotor radius, ωR is the rotational speed of the rotor and u is the total wind speed.
The rotor aerodynamic torque is the ratio of rotor aerodynamic power Pa to the shaft angular velocity. Hence by using Pa in
Eq. (10), the torque is given by,

1 2 3
TR ¼ πρR u C p ðλ; βÞ: ð13Þ
2ωR

The total wind speed is obtained using parameters in Table 1. The power coefficient is estimated using the blade data given in
Table 1. For the estimation of aerodynamic torque, the total wind speed and power coefficient estimated from Eq. (11) are used.
The aerodynamic torque obtained from the above approach is used for the simulation of wind turbine dynamics. Fig. 5 represents
the ramp, gust, total wind speed and the aerodynamic torque signal in time domain obtained from the above approach. This ob-
tained wind speed signal follows the same trend as the measured longitudinal wind speed signal at the site as given in [32,33].

3. Modeling of dynamics of drive train

Fig. 6 shows a schematic diagram of a wind turbine drive train. The main components of the drive train are rotor, gearbox and
generator. Generally, the wind turbine is equipped with an epicyclic gearbox which contains two external and one epicyclic spur
gear stages. The coupled torsional bending dynamic model of wind turbine gearbox is developed based on the Lagrange's formu-
lation and the dynamic equations of motion are formulated in [25]. The same modeling is used in the present study, hence, the
details of the dynamic model are summarized briefly as follows.
The time varying gear mesh stiffness, torsional shaft stiffness and bearings elasticity are accounted for in the dynamic model.
Rigid multibody modeling with discrete flexibility approach is used for the development of equations of motion. In this model,
there exist eight rotational and ten transverse degrees of freedom.

Single-Sided Amplitude Spectrum

140

120

100 Peak at first torsional

Gust natural freqeuncy
Amplitude

freqeuncy
80

60

40
Rotor Characteristic freqeuncy
20
( 1X and 2X components)
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency (Hz)

Fig. 14. Frequency domain representation of gust, peak at first natural frequency and rotor characteristic load frequency and its harmonics (8th level
decomposition).
14 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

4
x 10
6

Torque (N-m)
2

-2

-4

-6
0 1 2 3 4 5 6 7 8 9 10
Time (S)

Fig. 15. Time domain representation of external output torque.

The generalized coordinates to represent the degrees of freedom are as follows,

T
fqg ¼ fϕ1 ϕ2 ϕ3 ϕ4 ϕ5 ϕ6 ϕ7 ϕ8 yi zi ys zs y1 z1 y2 z2 y3 z3 g : ð14Þ

where ϕi (i = 1, 2, … ,8) are the rotor, carrier, planets, sun, gear1, gear2, gear3 and generator rotational degrees of freedom re-
spectively and yi, zi, ys, zs, y1, z1, y2, z2, y3, and z3 are the planets, sun, gear1, gear2 and gear3 transverse degrees of freedom in the y
and z directions respectively.
Lagrange's technique is used for developing the equation of motion where kinetic and potential energies of rotating and
orbiting gears are considered. The kinetic energy of orbiting and rotating gears can be represented as,

1 _2 1  1 1 1 1 1 1 3 3 1
J 2 þ 3mi r c1 ϕ_ 2 þ ð3 J 3 Þϕ_ 3 þ J 4 ϕ_ 4 þ J 5 ϕ_ 5 þ J 6 ϕ_ 6 þ J 7 ϕ_ 7 þ J 8 ϕ_ 8 þ mi y_ i þ mi z_ i þ ms y_ s
2 2 2 2 2 2 2 2 2 2 2
K:E: ¼ J 1 ϕ1 þ
2 2 2 2 2 2 2 2 2 2 2
1 2 1 2 1 2 1 2 1 2 1 2 1 2
þ ms z_ s þ m1 y_ 1 þ m1 z_ 1 þ m2 y_ 2 þ m2 z_ 2 þ m3 y_ 3 þ m3 z_ 3 : ð15Þ
2 2 2 2 2 2 2

Mainly the gear mesh stiffness, bearings and shaft stiffness contribute to the elastic potential energy.
The potential energy due to the time varying gear mesh stiffness is represented as,

1X n
2 1X n
2 1 pg 2
P:E:1 ¼ kri δri þ k δ þ kpg δ pg ; ð16Þ
2 i¼1 2 i¼1 si si 2

where kri, ksi, and kpg are the ring–planet, planet–sun and external gear pinion mesh stiffness and δri, δsi, and δpg are the fixed
ring–planet, sun–planet and external gear–pinion dynamic transmission errors respectively. The time varying gear mesh stiffness
is represented in terms of Fourier series whose details along with dynamic transmission error are discussed in [25].
Bearings associated with elastic potential energy, whose contribution to the total elastic potential energy is represented
as,

1X 6  
2 2
P:E:2 ¼ k y þ zb j ; ð17Þ
2 j¼1 b j b j

Single-Sided Amplitude Spectrum of torque

14000

12000
Amplitude (N-m)

10000

8000

6000

4000

2000

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Fig. 16. Frequency domain representation of external output torque.

 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 15

4
x 10
5

4
(a)

Carrier internal torque (N-m)

3

-1

-2

-3
0 1 2 3 4 5 6 7 8 9 10
Time (S)

4
x 10
4
Sun Internal torque (N-m)

2 (b)

-2

-4

-6
0 1 2 3 4 5 6 7 8 9 10
Time (S)

4
x 10
6
Gear 1 internal torque (N-m)

(c)
4

-2

-4
0 1 2 3 4 5 6 7 8 9 10
Time (S)

4
x 10
1
Gear2 internal torque (N-m)

(d)
0.5

-0.5

-1
0 1 2 3 4 5 6 7 8 9 10
Time (S)

4000
Gear3 internal torque ( N-m)

2000 (e)
0

-2000

-4000

-6000

-8000
0 1 2 3 4 5 6 7 8 9 10
Time (S)

Fig. 17. Internal resistive torque due to time varying gear mesh stiffness (a) carrier, (b) sun, (c) gear1, (d) gear2 and (e) gear3.
16 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

where kbj represents bearing stiffness, and ybj and zbj represent transverse displacements of bearings in the y and z directions
respectively.
The elastic potential energy due to the torsional shaft stiffness is represented as

1 2 1 2 1 2
P:E:3 ¼ C ðϕ −ϕ1 Þ þ C 2 ðϕ5 −ϕ4 Þ þ C 3 ðϕ8 −ϕ7 Þ ; ð18Þ
2 1 2 2 2

where C1, C2, and C3 represent the low speed, intermediate and high speed shaft stiffness respectively. The details of evaluating
strain energy associated with bearings and shafts are discussed in [25].
Lagrange's formulation is used to extract the coupled torsional bending dynamic equations of motion for the gearbox. The
equations of motion associated with rotational and translational coordinates are indexed in Appendix A.
The generator torque is represented in the form of,

T R ðt Þ
T G ðt Þ ¼ ; ð19Þ
G

where G is the epicyclic gearbox gear ratio.

The damping is included in the dynamic system using the Rayleigh's method of proportional damping. The damping ratio of 5%
is considered in this study, the justification of this value is discussed in detail in [34].
The generalized equations of motion are represented in the matrix form as,

€g þ ½C ðt Þfq_ g þ ½K ðt Þfqg ¼ fT 0 ðt Þg;

½Mfq ð20Þ

where [M], [C(t)] and [K(t)] represent the mass, time varying damping and stiffness matrices respectively. Newmark time integra-
tion algorithm is used to solve the equation of motion and the vibration responses are obtained in time and frequency domains.

4. Results and discussion

The wind turbine drive train parameters as considered in [25] are used in the present study and are tabulated in Table 2. The
generation of the aerodynamic torque time series is discussed in Section 2. The vibration responses in the time and frequency do-
mains are obtained from these signals.
In the present study, vibration, external torque and internal resistive torques are used for the dynamic characterization of drive
train. Table 3 represents the characteristic frequencies that can be expected from the vibration signal of a typical wind turbine
drive train, for the data considered in Table 2.

4.1. Vibration signal

Typically, wind turbine is an example of slow speed rotating machinery, rotating at a constant speed and subjected to stochas-
tic loads at the power production stage (in the case of both fixed and variable speed wind turbines). Over a finite period of time,
the obtained vibration signal due to stochastic loads can be assumed to be continuous and aperiodic, hence, in the present study,
FFT is applied. Figs. 7 and 8 represent the time and frequency domain (FFT) representation respectively of the vibration acceler-
ation signals of gear1, gear2, gear3 and the generator. From the time domain signals, it is clearly evident that the vibration signals
of wind turbine drive train are nonlinear and non-stationary in nature due to the type of impressed load. From the frequency do-
main representation, a clear peak is observed at a frequency close to the second natural frequency (see Table 3) and all the energy
content of the vibration signal is confined to the low frequency region only. Further, the vibration signal is zoomed at low fre-
quencies to extract the characteristic frequencies of the drive train. Fig. 9 represents the zoomed vibration signal from 0–10 Hz
of gear3, where no clear information is available except the peaks close to the second natural frequency.
There exist a number of hidden frequencies such as, gear mesh frequency, gust amplitude frequency, peak at the first torsional
natural frequency and characteristic load frequencies in the frequency spectrum, all of which are to be identified. From the pre-
vious work of authors, it is found that gear3 (see Fig. 1) is subjected to more vibration acceleration of all gears in the wind turbine
gearbox [25]. Hence, in the present work, wavelet decomposition is applied to the vibration signal of gear3 to identify all the ex-
pected characteristic frequencies. A sym8 wavelet is used with a sampling frequency of 1000 Hz for decomposition. The signal is
decomposed up to the 8th level with the use of commercial software package MATLAB 7.10.0. Fig. 10 shows a typical wavelet de-
composition up to the 3rd level.
From the FFT of gear3 vibration signal, no clear information is found about the characteristic frequencies of the gearbox. How-
ever, FFTs are applied to the decomposed signals from which all the characteristic peaks are seen. The gear mesh frequency of
gear2–3 (326.5 Hz) is seen clearly in the 3rd level decomposition at a frequency band of 312–375 Hz (see Fig. 11). This gear
mesh frequency is clearly seen from wavelet decomposition from the fact that no frequencies exist in the gearbox except that
at a range of 250–500 Hz.

Fig. 18. Frequency domain representation of internal resistive torque (a) carrier, (b) sun, (c) gear1, (d) gear2 and (e) gear3.
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 17

Single-Sided Amplitude Spectrum

5000

Peak near second

4000 natural frequency (a)

Amplitude (N-m)
Ring-planet, planet-sun gear mesh
3000 freqeuncy and higher harmonics (25.2 Hz)

2000

1000

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

4
x 10 Single-Sided Amplitude Spectrum
2

(b)
1.5
Amplitude (N-m)

peak near second natural

freqeuency
1
Sun-planet gear mesh frequency

0.5

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Single-Sided Amplitude Spectrum

15000

(c)
Amplitude (N-m)

10000

Gear1-2 gear mesh frequency (83.8 Hz)

5000

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Single-Sided Amplitude Spectrum

1200

1000 (d)
Amplitude (N-m)

Gear 2-3 gear mesh freqeuncy (326.5 Hz)

800
Gear 1-2 gear mesh frequency (83.8 Hz)

600

400

200

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)

Single-Sided Amplitude Spectrum

1600

1400
(e)
1200
Amplitude (N-m)

1000

800 Gear 2-3 gear mesh freqeuncy (326.5 Hz)

600

400

200

0
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)
18 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

Single-Sided Amplitude Spectrum

500
Gust amplitude frqeuncy

400

Amplitude (N-m)
Peak near first torsional
natural freqeuncy. Peak at charecterisitc load freqeuncy
( 1X, 2X, 3X and 4X)
300

200

100

0
0.5 1 1.5 2 2.5 3 3.5 4
Frequency (Hz)

Fig. 19. Enlarged view of frequency spectrum of gear3 internal torque.

Figs. 12 & 13 bring out gear mesh frequencies of ring–planet, planet–sun and gear1–2 (25.2 and 83.5 Hz) through wavelet de-
composition. These gear mesh frequencies are observed at 5th and 6th levels of decomposition. The wavelet packet at the 8th
level is used to capture the characteristic load frequency of the rotor, the first natural frequency and gust frequency (see
Fig. 14). Due to the rotational sampling effect (as discussed earlier with respect to Fig. 3), 1×, and 2× peaks are observed near
the rotor characteristic load frequency within the specified frequency limits. Hence, by wavelet decomposition it is possible to ex-
tract the hidden frequency information which exists in the gearbox.

4.2. Torque signals

The possibility of understanding the dynamics of drive train through torque signals is discussed in this section. Generally, there
exists internal resistive torque in addition to external torque in the gearbox. It is clearly evident from equation of motion (see
Eq. (A1(h)) in Appendix A) that the output external torque can be obtained from the algebraic sum of generator inertia torque
and internal resistive torque due to high speed shaft stiffness. Hence, this signal depends upon the vibration acceleration of gen-
erator, vibration displacements of gear3 and generator. Figs. 15 & 16 show the time and frequency representation respectively of
external output torque signal obtained from vibration.
It can be noticed that the external torque signal also behaves similar to the vibration signal and if one needs a detailed study of
it, wavelet decomposition is the best strategy for dynamic characterization of such signals. The possibility of characterizing the
drive train with internal resistive torque signals is also discussed as in the following.
The internal resistive torques exist due to gear mesh stiffness, damping present in the system and inertia effects. These types
of torques are represented in equations of motion (Eqs. (A1(a–h)) in Appendix A). For example, from the equation of motion
(Eq. (A1(h))), the internal resistive torque due to stiffness is presented in the form of modulation of time varying gear mesh stiff-
ness, dynamic transmission error and radius of gear3. Similar kind of forms can also be observed for internal resistive torques of
other gears also from Eqs. (A1(a–g)). Figs. 17 and 18 represent the internal resistive torque in time and frequency domains re-
spectively that exist between gears, due to time varying gear mesh stiffness.
In Fig. 17, it can be noticed that in gear tooth contact force time domain representation, gear mesh frequencies are overlapped
over the dynamic transmission error. From frequency domain representation, it is evident that the gear mesh frequencies are
equally dominating with low frequency region peaks. It is also observed that frequency spectrum corresponding to gear2 internal
resistive torque contains gear mesh frequencies of both gear1–2 and gear2–3. From the frequency spectra all the gear mesh fre-
quencies and their higher harmonics are also clearly observed.
Fig. 19 represents the enlarged view of internal resistive torque of gear3 in the low frequency region. From this it is evident
that clear peaks are observed at gust frequency, first torsional natural frequency and multiples of rotor characteristic frequency
(1×, 2×, 3× and 4×) from FFT itself besides the peaks at the second natural frequency and gear mesh frequency.
Hence, one can notice that FFT of internal resistive torque signal itself reveals all the characteristic frequencies of the drive
train.
It may be observed from the above discussion that, it is difficult to measure the internal resistive torques directly in the wind
turbine drive train. Hence, a strategy of measuring such signals is discussed in the present study. From the equation of motion
(Eq. (A1) in Appendix A) it can be understood that the internal resistive torque due to stiffness is a function of time varying
gear mesh stiffness and dynamic transmission error. Hence, internal resistive torque can be measured indirectly by measuring
the above two parameters. It is easy to measure the dynamic transmission error which exists at any gear stage. The detailed pro-
cedure of dynamic transmission measurement is discussed in [35]. Similarly, the gear mesh stiffness depends upon parameters
like load and rotational position of gears. It is difficult to measure the gear mesh stiffness directly, however one can measure
this indirectly by measuring the dynamic transmission error over the angular position by subjecting the gearbox to a constant
torque. Theoretically it is possible to obtain vibration responses individually for each and every gear by rigid multi body modeling
with discrete flexibility approach. However in practice, the torsional vibration exists in the coupled mode and the dynamic trans-
mission error itself is the torsional vibration that exists between the two gears [36]. Hence, dynamic transmission error is the in-
dication of torsional vibration in the case of gears. Thus, in practice it is possible to measure time varying gear mesh stiffness of
two gear pairs.
 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20 19

Since internal resistive torque due to stiffness is a function of measurable parameters such as, dynamic transmission error and
time varying mesh stiffness, it is recommended to use these two parameters to extract the frequency information.

5. Conclusions

The dynamic analysis of wind turbine drive train is carried out. The dynamic model as developed by the authors using Lagrange's
formulation is used in the present study. The aerodynamic torque signal is obtained by taking into account the average, ramp, gust and
turbulence components of the wind. This signal represents the almost typical wind signal at the site. The vibration responses and ex-
ternal toque responses are obtained to get the dynamic characterization of the wind turbine gearbox. It is revealed that these signals
are very less sensitive to loads and require a detailed signal processing to get the full information. Hence, wavelet packets are used in
the present approach from which all the characteristic frequencies are clearly seen in the decomposed signal at different levels. The
study shows that dynamic characterization can be done through internal resistive torque signals also. It is noted that, there exists a
number of hidden characteristic frequencies of the drive train in vibration signal, which needs detailed signal processing. However,
if internal resistive torque is used all such frequencies can be clearly observed in FFT itself.
Further, the possibility of indirect measurement of internal resistive torque signals by dynamic transmission error and time
varying gear mesh stiffness measurement is elaborated. The study envisages the suitability of internal resistive torque measure-
ment for wind turbine drive train condition monitoring.

Appendix A

The equation of motion in rotational coordinates are given by

€ þ C ðϕ −ϕ Þ ¼ T ðt Þ
J1 ϕ ðaÞ
1 1 1  2 R
2 € 2 2
J 2 þ 3m3 r c1 ϕ 2 þ C 1 ðϕ2 −ϕ1 Þ þ 3K ri ðt Þr c1 cosðα Þϕ2 þ 3K si ðt Þr c1 cosðα Þϕ2 ðbÞ
−3K si ðt Þr 3 r c1 cosðα Þ þ 3K ri ðt Þr 3 r c1 cosðα Þ−3K si ðt Þr c1 r 4 cosðα Þϕ4 ¼ 0
€ þ 3K ðt Þr 2 ϕ þ 3K ðt Þr 2 ϕ þ 3r r K ðt Þϕ
3J 3 ϕ 3 ri 3 3 si 3 3 s i si 4 ðcÞ
þ3r c1 r i cosðα ÞK si ðt Þϕ2 þ 3r c1 ri cosðα ÞK ri ðt Þϕ2 ¼ 0 
J4 ϕ€ þ K ðt Þr r cosðα Þϕ þ K ðt Þr r ϕ þ C þ K ðt Þr 2 ϕ −C ϕ ¼ 0 ðdÞ
4 si c1 4 2 si 3 4 3 2 si 4 4 2 5
ðA1Þ
J5 ϕ€ þ C ðϕ −ϕ Þ þ K ðt Þr 2 ϕ þ K ðt Þr r ϕ þ K ðt Þr sinðα Þy
5 2 5 4 g12 5 5 g12 5 61 6 g12 5 1 ðeÞ
þK g12 ðt Þr 5 cosðα Þz1 −K g12 ðt Þr 5 sinðα Þy2 −K g12 ðt Þr 5 cosðα Þz2 ¼ 0
€
J 6 ϕ6 þ K g23 ððy2 −y3 Þ sinðα Þ þ ðz2 −z3 Þ cosðα Þ þ r 62 ϕ6 þ r7 ϕ7 Þr 62
ðfÞ
þK g12 ððy1 −y2 Þ sinðα Þ þ ðz1 −z2 Þ cosðα Þ þ r5 ϕ5 þ r 61 ϕ6 Þr 61 ¼ 0
J7 ϕ€ þ C ðϕ −ϕ Þ þ K ðt Þððy −y Þ sinðα Þ þ ðz −z Þ cosðα Þ þ r ϕ þ r ϕ Þr ¼ 0 ðgÞ
7 3 7 8 g23 2 3 2 3 62 6 7 7 7
J8 ϕ€ þ C ðϕ −ϕ Þ ¼ −T ðt Þ ðhÞ
8 3 8 7 G

where TR(t), and TG(t) are the rotor and generator torques respectively.
The equations of motion associated with translational coordinates are given by

3my €i þ 
k1 yi −1:5K si ðt Þys ¼ 0  ðaÞ
si
3m€zi þ k1 −1:5K ri ðt Þ−1:5K si ðt Þ zi þ 1:5K si ðt Þzs ¼ 0 ðbÞ
   
k þ 9k3 k −3k3
ms y€s −ð1:5K si ðt ÞÞyi þ 2 −1:5K ri ðt Þ ys þ 2 y ¼0 ðcÞ
 4   4  1
k þ 3k3 k −3k3
ms €zs −ð1:5K si ðt ÞÞzi þ 2 þ 1:5K si ðt Þ zs þ 2 z1 ¼ 0 ðdÞ
   4  4
k −3k3 k þ k3
m1 y€1 þ 2 y þ 2 y þ K g12 ðt Þððy1 −y2 Þ sinðα Þ þ ðz1 −z2 Þ cosðα Þ þ r 5 ϕ5 þ r61 ϕ6 Þ sinðα Þ ¼ 0 ðeÞ
 4  s  4  1 : ðA2Þ
k −3k3 k þ k3
m1 €z1 þ 2 zs þ 2 z1 þ K g12 ðt Þððy1 −y2 Þ sinðα Þ þ ðz1 −z2 Þ cosðα Þ þ r 5 ϕ5 þ r 61 ϕ6 Þ cosðα Þ ¼ 0 ð f Þ
4 4  
m2 y€2 þ ðk4 þ k5 Þy2 −K g12 ðt Þððy1 −y2 Þ sinðα Þ þ ðz1 −z2 Þ cosðα Þ þ r 5 ϕ5 þ r 61 ϕ6 Þ sin α g ðgÞ
þK g23 ðt Þððy2 −y3 Þ sinðα Þ þ ðz2 −z3 Þ cosðα Þ þ r 62 ϕ6 þ r 7 ϕ7 Þ sinðα Þ ¼ 0
m2 €z2 þ ðk4 þ k5 Þz2 −K g12 ðt Þððy1 −y2 Þ sinðα Þ þ ðz1 −z2 Þ cosðα Þ þ r 5 ϕ5 þ r 61 ϕ6 Þ cosðα Þ
þK g23 ðt Þððy2 −y3 Þ sinðα Þ þ ðz2 −z3 Þ cosðα Þ þ r 62 ϕ6 þ r 7 ϕ7 Þ cosðα Þ ¼ 0ðhÞ
m3 y€3 þ k6 y3 −K g23 ðt Þððy2 −y3 Þ sinðα Þ þ ðz2 −z3 Þ cosðα Þ þ r 62 ϕ6 þ r 7 ϕ7 Þ sinðα Þ ¼ 0 ðiÞ
m3 €z3 þ k6 z3 −K g23 ðt Þððy2 −y3 Þ sinðα Þ þ ðz2 −z3 Þ cosðα Þ þ r 62 ϕ6 þ r 7 ϕ7 Þ cosðα Þ ¼ 0 ð jÞ

References

[1] Y. Li, Discussion on the principles of wind turbine condition monitoring system, International Conference on Materials for Renewable Energy and Environment
ICMREE-2011, 20–22 May 2011, Shanghai, China, 2011.
[2] U. Giger, K. Arnaudov, High efficiency high torque gearbox for multi megaWatt wind turbines, Mach. Technol. Mater. (1) (2010) 19–24.
20 P. Srikanth, A.S. Sekhar / Mechanism and Machine Theory 98 (2016) 2–20

[3] D. Qin, J. Wang, T.C. Lim, Flexible multibody dynamic modeling of a horizontal wind turbine drive train system, J. Mech. Des. ASME 131 (2009)
114501(1)–114501(6).
[4] K. Abboudi, L. Walha, Dynamic behavior of a two stage gear train used in a fixed speed wind turbine, Mech. Mach. Theory 46 (2011) 1888–1900.
[5] Y. Guo, J. Keller, R.G. Parker, Nonlinear dynamics and stability of wind turbine planetary gear sets under gravity effects, Eur. J. Mech. A Solids 47 (2014) 45–57.
[6] M. Todorov, Dobrev, F. Massouh, Analysis of torsional oscillation of the drive train in horizontal-axis wind turbine, Proceedings of 2009 ELECTROMOTION EPE
Chapter Electric Drives Joint Symposium, 1–3 July 2009, Lille, France, 2009.
[7] J. Helsen, F. Vonhollebeke, Multibody modeling of varying complexity for modal behavior analysis of wind turbine gearboxes, Renew. Energy 36 (2011)
3098–3113.
[8] J. Helsen, F. Vonhollebeke, Insights in wind turbine drive train gathered by validating advanced models on a newly developed 13.2 MW dynamically controlled
test rig, Mechatronics 21 (2011) 737–752.
[9] W. Shi, C.W. Kim, C.W. Chung, H.C. Park, Dynamic modeling and analysis of wind turbine drive train using the torsional dynamic model, Int. J. Precis. Eng. Manuf.
14 (2013) 153–159.
[10] P. Srikanth, A.S. Sekhar, Dynamic analysis of fixed speed wind turbine drive train, Proceedings of the 2013 International Congress on Sound and Vibration, 7–11
July 2013, Bangkok, Thailand, 2013.
[11] D.D. Ji, Y.M. Song, J. Zhan, Dynamics of gear train set in wind turbine, Mater. Sci. Forum 698 (2012) 701–705.
[12] W. Haijun, D.U. Xuesong, X.U. Xiangyang, Dynamic analysis and experimental study of MW wind turbine gearbox, Appl. Mech. Mater. 86 (2011) 739–742.
[13] L. Quan, L. Yongqian, Y. Yongping, Vibration Response Analysis of Gear Driven System of Wind Turbine, 3, IEEE, 2010 380–383.
[14] L. Yongqian, L. Quan, Y. Yongping, Sensitivity analysis of natural frequency of gear driven system of wind turbine, Proceedings of the CMCE 2010 Conference, 24–
26 Aug 2010, IEEE 2010, pp. 559–562.
[15] H. Sipeng, Z. Yangpei, X. Li, Y. Yue, Equivalent Wind Speed Model in Wind Farm Dynamic Analysis, IEEE, 2011 1751–1755.
[16] J.J. Jang, J.R. Lee, Analysis of design wind speed distribution of Taiwan area, J. Mar. Sci. Technol. 5 (1997) 55–63.
[17] Z. Xuesong, L. Ji, M. Youjie, Review on wind speed model research in wind power systems dynamic analysis, International Conference on Sustainable Power Gen-
eration and Supply SUPERGEN '09, April 2009, Nanjing, China, 2009.
[18] W. Dong, Y. Zing, T. Moan, Z. Gao, Time domain based gear contact fatigue analysis of a wind turbine drive train under dynamic conditions, Int. J. Fatigue 48
(2013) 133–146.
[19] S. Ahmed, Wind Energy Theory and Practice, PHI Learning Private Limited, New Delhi, India, 2010.
[20] A.M. Ragheb, M. Ragheb, Wind Turbine Gearbox Technologies, 1st ed. Intech Europe, Croatia, 2011 189–206.
[21] C. Zhu, S. Chen, H. Liu, H. Haung, G. Li, F. Ma, Dynamic analysis of the drive train of a wind turbine based upon the measured load spectrum, J. Mech. Sci. Technol.
28 (6) (2014) 2033–2040.
[22] A.R. Nejad, Z. Gao, T. Moan, Long term analysis of gear loads in fixed offshore wind turbines considering ultimate operating conditions, Energy Procedia 35 (2013)
187–197.
[23] Y. Zhao, M. Ahmat, K. Bari, Nonlinear dynamics modeling and analysis of transmission error of wind turbine planetary gear system, J. Multi-body Dyn. 228 (2014)
438–448.
[24] S. Wei, J. Zhao, Q. Han, F. Chu, Dynamic response analysis on torsional vibrations of wind turbine geared transmission with uncertainty, Renew. Energy 78 (2015)
60–67.
[25] P. Srikanth, A.S. Sekhar, Dynamic analysis of wind turbine drive train subjected to nonstationary wind load excitation, Proc. Inst. Mech. Eng. C J. Mech. Eng. Sci.
229 (3) (2015) 429–446.
[26] T. Burton, D. Sharpe, M. Jenkins, E. Bossanyi, Wind Energy Handbook, Wiley, Chichester, UK, 2001.
[27] A.P. Martins, P.C. Costa, A.S. Carvalho, Coherence and wakes in wind models for electro mechanical and power systems standard simulations, European Wind
Energy Conference and Exhibition EWEC 2006, 27 February–2 March 2006, Athens, 2006.
[28] M. Shinozuka, C.M. Jan, Digital simulation of random process, J. Sound Vib. 25 (1) (1972) 111–128.
[29] T. Petru, Modeling of Wind Turbine System for Power System Studies(Phd thesis) , 2003, ISBN 91-7291-306-1.
[30] A. Burlibasa, E. Cenga, Rotationally sampled spectrum approach for simulation of wind speed turbulence in large wind turbines, Appl. Energy 111 (2013)
624–635.
[31] P.O. Ochieng, A.W. Manyonge, A.O. Oduor, Mathematical analysis of tip speed ratio of a wind turbine and its effect on power coefficient, 4 (1) (2014) 61–66.
[32] T. Barszcz, M. Bielexka, A. Bielecki, M. Wojcik, Wind speed modeling using Weierstrass function fitted by a genetic algorithm, J. Wind Eng. Ind. Aerodyn. 109
(2012) 68–78.
[33] A. Bielecki, T. Barszcz, M. Wojcik, Modeling of a chaotic load of wind turbine drive train, Mech. Syst. Signal Process. 55 (2015) 491–505.
[34] S. Chauhan, M.H. Hansen, D. Tcherniak, Application of operational modal analysis and blind source separation/independent component analysis techniques to
wind turbines, Conference nd Exposition on Structural Dynamics IMAC-XXVII, 9–12 February 2009, Orlando, Florida, USA, 2009.
[35] D.R. Houter, G.W. Blankenship, Methods for measuring gear transmission error under load and at operating speeds, SAE 8918691989.
[36] Z. Hameed, Y.S. Hong, Y.M. Cho, S.H. Ahn, C.K. Song, Condition monitoring and fault detection of wind turbines and related algorithms: a review, Renew. Sust.
Energ. Rev. 13 (2009) 1–39.