Rene S 24 00900
Rene S 24 00900
Rene S 24 00900
Manuscript Number:
Keywords: Robustness analysis; dynamic modelling; wind turbine drivetrain; Polynomial chaos;
Monte Carlo
Nabih Feki
Abstract: For the wind turbine system, several research studies have shown the great sensitivity
of drivetrain parameters on their dynamic behaviour. This paper presents a robustness
analysis of a complex nonlinear system, focusing on the wind turbine drivetrain. The
investigation employs a comprehensive approach to evaluate the system's dynamic
responses under uncertainty in key parameters such as torsional stiffness, wind speed,
rotor inertia, and generator inertia. The efficiency of uncertainty quantification is
evaluated by comparing the proposed Generalized Polynomial Chaos method (GPC)
with the Monte Carlo simulations (MC). Additionally, the results obtained from these
two approaches are juxtaposed with the deterministic Runge-Kutta 5th-order method
(without uncertainty). Notably, the study emphasizes the importance of evaluating
extreme parameter values to gain insights into their influence on uncertainty
simulations. The findings contribute to a better understanding of the robustness of wind
turbine drivetrain system under complex multi-variable uncertainty.
Oussama Graja, Dr
University of Sfax
grajaoussama@gmail.com
Ahmed guerine, Dr
University of Burgundy
ahmedguerine@gmail.com
Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Title Page
Title page
Cover letter
29/01/2024
Dear Editor,
This paper investigates the dynamic behavior of wind turbine drivetrain systems, focusing on the
sensitivity of key parameters under uncertainty. Our robustness analysis employs a
comprehensive approach, evaluating the system's responses to variations in torsional stiffness,
wind speed, rotor inertia, and generator inertia.
The study contributes to a better understanding of the robustness of wind turbine drivetrain
systems under the complexities of multivariable uncertainty. The outcomes offer valuable
insights for designing and operating these systems in real-world conditions.
Moreover, assessing the extreme values of parameters is crucial for comprehending their impact
on uncertainty simulations, particularly when considering interactions.
Sincerely,
Highlights
1. Focus on Nonlinear Dynamic Response:
• Emphasis on exploring the nonlinear dynamic response of wind turbine
drivetrain system, considering multi-variable uncertain parameters.
2. Two Approaches Employed:
• Utilization of two distinct approaches to investigate the influence of the
uncertainty: the fundamental Monte Carlo method and the Probabilistic
Polynomial Chaos method with a regression approach.
3. Decoupled Form Analysis:
• Examination of uncertainty through a decoupled form, isolating a single
uncertain parameter to understand its individual impact on dynamic response.
4. Coupled Form Exploration:
• Transition to the coupled form involving two uncertain parameters, investigating
scenarios with torsional stiffness and wind speed, as well as rotor inertia and
generator inertia.
5. Efficiency Comparison of Methods:
• Thorough comparison of the efficiency of the proposed Polynomial Chaos
method with both the Monte Carlo approach and the deterministic Runge-Kutta
5th-order method.
6. Importance of Parameter Interactions:
• Emphasis on the significance of studying interactions between uncertain
parameters for a comprehensive understanding of uncertainty.
7. Crucial Role of Extreme Parameter Values:
• Recognition of the essential role of assessing extreme parameter values in
comprehending their impact on uncertainty simulations, especially within the
context of parameter interactions.
8. Contributions to Robustness Understanding:
• Overall, the research significantly contributes to advancing the understanding of
wind turbine drivetrain system robustness under the challenges of multivariable
uncertainties.
Manuscript Click here to view linked References
Abstract. For the wind turbine system, several research studies have shown the great sensitivity of
drivetrain parameters on their dynamic behaviour. This paper presents a robustness analysis of a
complex nonlinear system, focusing on the wind turbine drivetrain. The investigation employs a
comprehensive approach to evaluate the system's dynamic responses under uncertainty in key
parameters such as torsional stiffness, wind speed, rotor inertia, and generator inertia. The efficiency of
uncertainty quantification is evaluated by comparing the proposed Generalized Polynomial Chaos
method (GPC) with the Monte Carlo simulations (MC). Additionally, the results obtained from these
two approaches are juxtaposed with the deterministic Runge-Kutta 5th-order method (without
uncertainty). Notably, the study emphasizes the importance of evaluating extreme parameter values to
gain insights into their influence on uncertainty simulations. The findings contribute to a better
understanding of the robustness of wind turbine drivetrain system under complex multi-variable
uncertainty.
Keywords: Robustness analysis; dynamic modelling; wind turbine drivetrain; Polynomial chaos; Monte
Carlo
1. Introduction
With the increase in global demand for electricity, driven by technological developments and
progress [1], we are witnessing a growing orientation towards renewable energy systems. Among these
sources, we can cite wind energy which has been growing rapidly in the last decade. The utilization of
wind turbines harnessing wind energy has witnessed remarkable growth, emerging as a pivotal
sustainable energy source. The representation of wind turbines encapsulates their intricate engineering,
characterized by aerodynamic efficiency and power conversion mechanisms [2,3].
Numerous studies have delved into the understanding and optimization of wind turbine performance,
employing various approaches such as Computational Fluid Dynamics (CFD) [4], fatigue [5], structural
[6], and dynamic analyses [7,8]. These investigations aim to enhance our comprehension of the complex
interactions between aerodynamic forces, structural dynamics, and power conversion mechanisms. The
dynamic behavior of a wind turbine is influenced by numerous factors, which can be categorized into
two families: external factors, including wind speed and blade size, and internal factors, such as
drivetrain and gearbox design. The drivetrain system plays a crucial role as it converts the kinetic energy
from the wind's rotation into electrical energy. Consequently, it stands out as one of the most critical
components in the overall operation of the wind turbine [9]. The modelling and analysis of the linear or
non-linear drivetrain system, especially the gearbox, have been subjects of extensive research [10-13].
As with any dynamic system, the wind turbine drivetrain is subject to uncertainties arising from
variations in its parameters [14]. Understanding and quantifying these uncertainties is essential for
robust design and reliable performance [15-17]. In the literature, we can find various approaches dealing
with uncertainty which can be divided mainly into three categories. Among these categories, the Monte
Carlo (MC) approach is mentioned [18, 19]. It serves as a common reference due to its ability to produce
1
results closely resembling reality. While the Monte Carlo method is reliable, it comes with the drawback
of consuming too much simulation time. Therefore, to reduce the time required for simulations involving
uncertainties, alternative approaches come into play. These methods may take a probabilistic form, such
as the perturbation method [20], the meta-model method [21] and the polynomial Chaos method [22,
23]. Additionally, we mention the possibilistic approach, exemplified by the interval computing method
[24–25].
Most research studies that deal with robustness and reliability focus on a single uncertain parameter
to identify the influence of this variable on the dynamic response of a system. Examining several
uncertain parameters at the same time (multi-variable uncertainty) allows us to study the interaction of
these parameters and get closer to reality in our modelling. the dynamic behaviour of the gear system
was studied by Mouna et al. [26] taking into consideration the uncertainty with two forms of analysis:
uncoupled form and coupled form.
This article delves into the complexities of parameter uncertainties within the wind turbine drivetrain
system, distinguishing between uni-variable and multi-variable approaches. By comprehensively
exploring uncertainties, this research contributes to the broader understanding of the robustness of wind
turbine systems in the face of multifaceted uncertainties.
This paper is organized as follows. In the second section, a new non-linear dynamic model of the
wind turbine drivetrain is presented. The motion equations, the nonlinear function and the excitation
sources are detailed. The third section is dedicated to presenting the tools used for incorporating
uncertainties in the simulation. Following that, the fourth section discusses the suggested simulations
strategy and investigates the simulations and the numerical results obtained. Finally, conclusions are
drawn based on these results.
2. Dynamic modeling
2
Rotor blade Gearbox Generator
2−𝜀𝛼𝑛
𝑘𝑚𝑎𝑥, 𝑛 = 𝑘𝑚𝑜𝑦,𝑛 (1 + 2.𝜀 ) (2)
𝛼𝑛 (𝜀𝛼𝑛 −1)
3
The first subsystem, which ensures the conversion of wind kinetic energy into mechanical energy
using a rotor, depends on various parameters such as the rotor diameter, the number of blades, and the
pitch angle, which is influenced by the wind flow velocity.
The power produced by the turbine Pwt(t) is a function of the wind speed, gradually increasing up to
the rated speed, and remaining constant until the wind speed reaches the cut-out value, including the
cut-out point. This power primarily depends on two factors: the amount of wind flow associated with
the current wind speed and the power coefficient Cp related to the blade characteristics [11].
𝑃𝑤𝑡 (𝑡) = 𝐶𝑝 . 𝑃𝑤𝑖𝑛 (𝑡) (3)
The amount of energy available for transfer to the rotor is influenced by the rotor diameter R and the
current wind speed Vwin and can be expressed by:
1 3
𝑃𝑤𝑖𝑛 (𝑡) = 2 . 𝜋. 𝜌𝑎𝑖𝑟 . 𝑅 2 . 𝑉𝑤𝑖𝑛 (4)
The wind turbine speed is considered a fixed value corresponding to a moderate wind speed, which
varies over time and represents a sinusoidal variation of the wind speed as indicated below:
𝑉𝑤𝑖𝑛 (𝑡) = 𝑉𝑚𝑜𝑦 [1 − 0.2 𝑐𝑜𝑠( 𝜔1 . 𝑡) − 0.05 𝑐𝑜𝑠(𝜔2 . 𝑡)] (6)
Q(t ) = x1, y1, x 2, y 2, x 3, y 3, (1,1), (1,2), (2,1), (2,2), (3,2), (3,3)
T
(8)
The equations of motion are composed of the general mass matrix [M] associated with the
acceleration {𝑄̈ }, the damping matrix [C] related to the velocity of displacements {𝑄̇ }, and the stiffness
matrix [K] related to displacements. The sum is equal to the external forces’ matrix, leading to the
equation of motion expressed as follows:
𝑥̈ 1 𝑥̇ 1 0
𝑥1
𝑦̈1 𝑦̇1 0
𝑦1
𝑥̈ 2 𝑥̇ 2 0
𝑥2
𝑦̈ 2 𝑦̇ 2 𝑦2 0
𝑥̈ 3 𝑥̇ 3 𝑥3 0
𝑦̈ 3 𝑦̇ 3 𝑦3 0
[𝑀] 𝜃̈11 + [𝐶(𝑡)] 𝜃̇11 + ([𝐾𝑆 ] + ⟨𝐿𝛿𝑛 ⟩𝑇 ⟨𝐿𝛿𝑛 ⟩[𝐾(𝑡)]) 𝜃 = 𝐶𝑎𝑒𝑟𝑜 (𝑡) (9)
11
𝜃̈12 𝜃̇12 𝜃12 0
𝜃̈21 𝜃̇21 𝜃21 0
𝜃22 0
𝜃̈22 𝜃̇22
𝜃32 0
𝜃̈32 𝜃̇32
{𝜃33 } {𝐶𝑒𝑙−𝑚 (𝑡, 𝑄̇ (𝑡), 𝑄̈ (𝑡))}
{𝜃̈33 } {𝜃̇33 }
4
The general mass matrix is composed of two main elements: mass matrix [ML] and inertia matrix
[MA].
𝑀𝐿 𝑍
[𝑀] = [ ] (10)
𝑍 𝑀𝐴
Indeed, the mass matrix [ML] is constituted from the set of masses corresponding respectively to
each block j:
𝑚1 0 0 0 0 0
0 𝑚1 0 0 0 0
0 0 𝑚2 0 0 0
𝑀𝐿 = (11)
0 0 0 𝑚2 0 0
0 0 0 0 𝑚3 0
[0 0 0 0 0 𝑚3 ]
Beyond the mass matrix of the blocks, the general mass matrix also comprises the inertial matrix,
which in turn is constituted by the grouping of the inertias of the various elements, which make up each
block, and it expressed as follows:
𝐼𝑟𝑜𝑡 0 0 0 0 0
0 𝐼12 0 0 0 0
0 0 𝐼21 0 0 0
𝑀𝐴 = 0 0 0 𝐼22 0 0 (12)
0 0 0 0 𝐼32 0
[ 0 0 0 0 0 𝐼𝑔𝑒𝑛 ]
Furthermore, a generalized damping of the system is considered, which is associated with the general
mass matrix in combination with the structural stiffness matrix, in addition to the time-dependent mesh
stiffness matrix. It is expressed by the following formulation:
In this article, we take the damping matrix coefficients with the following values: ψ=10-4 and μ=10-5.
The associated stiffness matrix [K] is made up of the mean stiffness matrix of the structure [Ks] and
the meshing stiffness matrix variable over time [Km(t)]. The latter typically reflects the progression of
the number of engaged teeth, which can vary between (1 and 2). From which we can distinguish an
average matrix [KC] and a variable matrix over time [𝐾𝑣 (𝑡)], as explained below:
Lδ1 and Lδ2 are the teeth deflections vector and can be written as follows:
= − sin (1 ) , cos (1 ) ,sin (1 ) , − cos (1 ) , 0, 0, 0, r 12, r 21, 0, 0, 0
b b
L (15)
1
5
𝑘𝜃1 −𝑘𝜃1 0 0 0 0
−𝑘𝜃1 𝑘𝜃1 0 0 0 0
0 0 𝑘𝜃2 −𝑘𝜃2 0 0
𝐾𝜃 = (17)
0 0 −𝑘𝜃2 𝑘𝜃2 0 0
0 0 0 0 𝑘𝜃3 −𝑘𝜃3
[ 0 0 0 0 −𝑘𝜃3 𝑘𝜃3 ]
where 𝐶𝑔𝑒𝑛 (𝑡) represents the mechanical torque of the generator, 𝐶𝑒𝑙−𝑚 (𝑡) describes the
electromagnetic torque, and the parameter fv defines the coefficient of viscous friction. Note also that
we consider that the rotation speed of the rotor and the rotation speed of the rotating elements of the
generator are constant speeds, from which we remove only the gain (the multiplication ratio) G(t):
𝛺
𝛺 = 𝑔𝑒𝑛
{ 𝑟𝑜𝑡 𝐺 (19)
𝐶𝑟𝑜𝑡 (𝑡) = 𝐺 . 𝐶𝑔𝑒𝑛 (𝑡)
By integrating equation (20) of the dynamic transmission ratio G(t) in the derived part of the last
expression, we obtain:
6
or 𝜃̇11 = 𝜃̇11 (𝑡) ; 𝜃̇33 = 𝜃̇33 (𝑡) ; 𝜃̈11 = 𝜃̈11 (𝑡) ; 𝜃̈33 = 𝜃̈33 (𝑡) (22)
Consequently, the electromagnetic torque 𝐶𝑒𝑙−𝑚 (𝑡, 𝑄̇ (𝑡), 𝑄̈ (𝑡))is based on the fluctuations over
time of the aerodynamic torque, as well as on the angular velocities (input-output) in steady state, and
on the variations over time of the velocities and their accelerations.
3. Uncertainty method
In this paper, we examine the influence of random variables on the dynamic behaviour of our system
using two approaches: the probabilistic Generalized Polynomial Chaos (GPC) method and the Monte
Carlo (MC) sampling method, as deterministic tools employed for result verification and the analysis of
extreme values.
3.1. Generalized polynomial chaos method
The probabilistic approach entails the mapping of random variables onto the probability space
through the application of Wiener's homogeneous chaos theory. This involves utilizing a spectral
expression referred to as generalized polynomial chaos (GPC).
In accordance with Askey's scheme [22], employing Legendre polynomials is particularly effective
in capturing uncertainties associated with uniform distributions. Therefore, the uncertainly of the
random eigenvalues is given by:
𝑁𝑝
𝑉𝑖 (𝑥, ξ) = ∑𝑖=0 𝑉̅ 𝑖, 𝑗 (𝑥)𝛷𝑗(ξ); ξ=( ξ1, ξ2, …) (23)
where 𝛷𝑗(ξ) define the orthogonal polynomials inside the unrelated random factors ξ.
The terms number denoted as Np is employed, and its value depend on the order of the polynomial
P=max (rj) and the stochastic dimension which represents the number of the uncertain parameters. More
description of the method used in this article can be found in the following references [26, 30].
3.2. Monte Carlo method
The Monte Carlo method is introduced through the simultaneous solving of a deterministic model
that incorporates uncertain parameters represented by random variables [31]. This approach expressed
by the following function:
A = M (B) (24)
where M defines the considering model, A is the vector of the estimated outputs that will be a random
vector and B= [B1, B2, …, Bn] represents the vector of uncertain input parameters.
7
The present study performs uncertainty accounting studies based on a single uncertain parameter and
then a coupled-parameter investigation that considers two uncertain parameters to progressively reveal
the interaction between the parameters that contribute to system investigation. The work strategy was
presented in the flowchart (Fig. 2).
8
Fig. 3. Mean value of velocities of the third bloc (j=3) for σKθ = 5%
Figure 3 depicts the dynamic response fluctuations through velocity curves of the third block (j=3)
or the shaft linked to the generator. Figure 3(a) describes the fluctuations of 𝒙̇ 3, and this curve does not
show any significant difference between the deterministic curve and the superimposed curves of the
average values of GPC and MC. The Min-Max value curves resulting from the application of MC are
plotted with a similar pattern to the other curves, exhibiting alternating amplitudes that slightly exceed
the amplitude of the deterministic curve. The curves in Figure 3(b) illustrate the velocity fluctuation of
𝒚̇ 3. The superimposed curves of the average values of GPC and MC do not closely follow the
deterministic curve, and even their maximum fluctuation amplitudes remain incomplete. The Min-Max
values from MC exhibit alternating maximum amplitude values, but the MC's Min curve represents a
maximum amplitude. Fig. 3(c) describes the angular velocity curves using GPC and MC superimposed,
resulting in a period and amplitude that are minimal compared to the deterministic curve. The curves
representing the dynamic responses of the variables at MC's minimum and maximum cases follow the
average value curve with the same velocity fluctuation amplitude.
To recapitulate, torsional stiffness plays a significant role in the rate of fluctuations, the periodicity
of the response, and the amplitudes as well. A higher maximum value of torsional stiffness indicates
greater stability concerning these criteria.
9
• Uncertainty of the generator inertia
This parameter characterizes the inertia of the components in the revolution of the generator, and it
plays a crucial role in accounting for the non-linearity arising from the electromagnetic torque of the
generator. Consequently, we considered the uncertainty associated with this parameter to understand its
influence on the dynamic behavior of the system.
In the context of the uncertainty approach, numerous simulations were conducted to determine the
standard deviation of the generator's inertia, aiming to gain insights into the dynamic behavior of the
displacement velocities.
Fig. 4. Mean value of velocities of the third bloc (j=3) for 𝝈𝑰𝒈𝒆𝒏 = 3%
Figure 4(a) illustrates the average velocity values obtained along the output shaft along 𝒙̇ 3. For this
particular value of the standard deviation of the generator's inertia, the average values resulting from the
application of GPC and MC are depicted with curves that closely match and lead to a minimization of
fluctuation amplitude compared to the deterministic curve at the same time frequency. For the Min-Max
curves of MC, they follow the trend of the deterministic curve, where the minimum value leads to an
increase in fluctuation amplitude and response frequency over time.
For 𝒚̇ 3 curve (Fig. 4-b), the average values of GPC and MC are depicted with curves that closely
match and have a lower fluctuation amplitude compared to the deterministic curve. For the Min-Max
curves of MC, they also follow the pattern of the deterministic curve with alternating maximum
fluctuation amplitude and response frequency over time. The minimum value described in the enlarged
part of the figure shows that this minimum value leads to a minimization of perturbation. Regarding the
10
average angular velocity value, the curves are represented in Figure 4(c), where the four curves
describing the average values of GPC and MC, along with the Min-Max curves of MC, are well
superimposed on the deterministic curve. The zoomed-in part of the curve reveals a slight increase in
fluctuation amplitude for the Min-MC value compared to the maximum.
Fig. 5. Mean value of velocities of the third bloc (j=3) for 𝝈𝑰𝒓𝒐𝒕 =5%
Figure 4 depicts the dynamic velocity response of the third block (j=3) under the influence of rotor
inertia uncertainty within a 1.5-second interval. The graph in Figure 5(a) illustrates five well-
superimposed curves, including the average values of GPC and MC, the Min-Max curves of MC, and
the deterministic response curve. From this, we can conclude that the uncertainty in rotor inertia has no
significant influence along x3 of the third block. The dynamic velocity response of the fluctuations along
y3 is shown in Figure 5(b), where the five curves, representing the average values of GPC and MC, the
Min-Max curves of MC, and the deterministic response, are well superimposed. Beyond the first second,
the maximum fluctuation amplitude alternates between taking the value of Min-MC and Max-MC. The
curves of angular velocity fluctuations of the third block (j=3) represented in Figure 5(c) are well
overlapped with each other. The Min-Max curves with MC method show minor fluctuations for these
extreme values compared to the deterministic curve. According to the graph, the Min-MC value of the
rotor's inertia leads to a minimization of the fluctuation period over time.
To summarize, the uncertainty in the rotor's inertia has no significant influence on the vibrational
behavior of the system.
11
The wind speed is a parameter present in the gear system, and it plays a role in characterizing the
aerodynamic torque, thus affecting the electromagnetic torque CEG. This parameter is in strong fluid-
structure coupling, indirectly influencing the gear system. The uncertainty of this parameter was
considered in our nonlinear dynamic model.
Furthermore, the uncertainty study is conducted using the nominal value of the wind speed
Vwind=12m/s. The uncertainty is performed with a standard deviation Vwind =5% for which the average
value curves resulting from the implementation of GPC and MC show compliance with the second-order
polynomial.
Figure 6 depicts the dynamic displacement velocity response of the third block under the influence
of wind speed uncertainty. The graph in Figure 6(a) illustrates five well-superimposed curves, including
the average values of GPC and MC, the Min-Max curves of MC, and the deterministic response curve.
From this, we can conclude that the wind speed uncertainty has no significant influence along x3 of the
third block.
Fig. 6. Mean value of velocities of the third bloc (j=3) for 𝝈𝑽𝑾𝒊𝒏𝒅 = 5%
The dynamic behavior of the velocity variations 𝒚̇ 3 is visualized in Figure 6(b), where the five curves,
representing the average values of GPC and MC, the Min-Max values of MC, and the deterministic
response, perfectly overlap. From the zoomed-in part, it can be observed that the maximum amplitude
alternates between taking the value of Max-MC, while the value of Min-MC leads to the minimization
of this amplitude.
Furthermore, the fluctuations of the angular velocity 33 are shown in Figure 6(c). It can be observed
that the curves of the average values of GPC and MC, along with the deterministic response, are perfectly
overlapping. The Min-Max curves of MC reveal the presence of significant fluctuations compared to
12
the deterministic curve, and according to the graph, the Min-MC value of the average wind speed leads
to a minimization of the fluctuation amplitude, in contrast to the maximum value.
From the preceding results, we can conclude that the uncertainty associated with the average wind
speed results in a very slight increase in the amplitude of fluctuations in the x and y directions. However,
this increase is significant when considering the fluctuations of angular velocities 33.
Fig. 7. STD of the velocities mean value of the third Fig. 8. Random variable cloud for 𝛔𝑲𝜽 =
bloc (j=3) 𝜎𝐾𝜃 = 4% and 𝜎𝑉𝑣𝑒𝑛𝑡 = 10% 𝟒% 𝐚𝐧𝐝 𝛔𝑽𝒗𝒆𝒏𝒕 = 𝟏𝟎%
13
MC, along with the deterministic curve, exhibit moderate compatibility regardless of whether it's in Fig.
9(a), (b), or (c).
In the direction of x3, the fluctuations are illustrated in Figure 9(a), where the four extreme value
curves are very well superimposed with the other curves. The curve of the 2nd case is slightly elevated
compared to the 1st case at the maximum amplitude, while the minimum amplitude is achieved with the
3rd and 4th cases.
For the displacement velocities along y3 in Figure 9(b), the extreme value curves are plotted in two
groups. The curve of the first case and the second case represents the first group, while the 2nd group
consists of the 3rd and 4th cases. These two groups are plotted in a way that each of them describes the
maximum amplitude, and they are plotted in opposition of phase. Furthermore, the angular displacement
velocity 33 is depicted in Figure 9(c), where the four curves, including the first and third cases, represent
the minimum amplitude. Conversely, the second and fourth cases lead to an elevation of the amplitude.
Fig. 9. Mean value of velocities of the third bloc (j=3) for 𝝈𝑲𝜽 = 𝟒% 𝐚𝐧𝐝 𝝈𝑽𝑾𝒊𝒏𝒅 = 𝟏𝟎%
During the simulations of univariable uncertainty, inertia parameters did not reveal significant
impacts on the fluctuations. Therefore, this study led to the dynamic analysis of a double-stage straight
transmission system using bivariable uncertainty to assess the effect of their interactions by increasing
the standard deviation of the two parameters compared to decoupled uncertainty simulations.
Therefore, several simulations are conducted to achieve better correspondence. For a combination of
standard deviations of 𝝈𝑰𝒓𝒐𝒕 = 𝟓% & 𝝈𝑰𝒈𝒆𝒏 = 𝟒%, an acceptable curve correspondence between GPC
14
and MC average value curves is found based on 2,000 simulations, using a fifth-order polynomial. These
results are well illustrated in Figure 10 of the standard deviation of average displacement velocity,
depicting the curve correspondence. This correspondence is a result of the correspondence in
displacement curves as well.
For 𝝈𝑰𝒓𝒐𝒕 = 𝟕% & 𝝈𝑰𝒈𝒆𝒏 = 𝟓%, Figure 11 of the standard deviation of average values reveals the
lack of correspondence between the displacement velocity curves of GPC and MC at the levels of third
bloc (j=3), with some correspondence evident in the rotation speed. This will be clearer in the subsequent
average value curves.
Additional simulations were conducted to achieve minimal correspondence between the curves of
GPC and MC for a third-order polynomial and a combination of standard deviation.
Fig. 10. STD of the velocities mean value of the third Fig. 11. STD of the velocities mean value of the third
bloc (j=3) for 𝝈𝑰𝒓𝒐𝒕 = 5% and 𝝈𝑰𝒈𝒆𝒏 = 4% bloc (j=3) for 𝝈𝑰𝒓𝒐𝒕 = 7% and 𝝈𝑰𝒈𝒆𝒏 = 5%
15
Like the other multivariable uncertainty simulation, we assess the extreme values resulting from the
interaction of random variables of uncertain parameters using the Monte Carlo method. Indeed, we rely
on the cloud of random variables shown in Figure 12 to define these values.
Furthermore, the consideration of multivariable uncertainty in the dynamic study of displacement
velocity at the third block level is illustrated in Figure 13.
Regarding the linear displacement velocities, the curve of 𝒙̇ 3 as illustrated in Figure 13(a), 𝒚̇ 3 as
illustrated in Figure 13(b), and along ̇ 33 as shown in Figure 13(c), it can be noted that there is
correspondence between the curves of the two methods, GPC and MC, but not with the deterministic
curve. This lack of compatibility evolves over time. Therefore, we can conclude that the uncertainty of
these two parameters holds significant implications for the evolution of linear displacement velocities
along x3 and y3.
Fig. 13. Mean value of velocities of the third bloc (j=3) for𝜎𝐼𝑟𝑜𝑡 = 7% & 𝜎𝐼𝑔𝑒𝑛 = 5%
For the extreme value curves, they result from the deterministic Monte Carlo method. Along the
direction of x3, the maximum amplitude is predominantly displayed by the curves of the first and third
cases, which exhibit a minimal periodicity compared to the curves of the second and fourth cases. Along
y3, the curves of the first and third cases, and those of the second and fourth cases, alternate at the
maximum amplitude value, surpassing the amplitude of the deterministic curve. Finally, For the ̇ 33, the
curve of the 1st case leads to a minimization of periodicity and increases the amplitude of fluctuations
in direct opposition to the curve of the fourth case, which results in a minimization of amplitude. The
curves of the second and third cases are intermediate curves.
16
5. Conclusion
This paper explores the nonlinear dynamic response of the wind turbine drivetrain system,
considering uncertain parameters. The investigation employs two approaches: the Monte Carlo method,
presenting fundamental outcomes, and the Probabilistic Polynomial Chaos method utilizing the
regression approach.
Initially, we examined uncertainty by taking a single uncertain parameter (decoupled form) to
determine the impact of each parameter on the dynamic response. Subsequently, we focus on the
coupled form involving two uncertain parameters. In the first scenario, we included torsional stiffness
and wind speed, while in the second scenario, we take rotor inertia and generator inertia. The efficiency
of the proposed Polynomial Chaos method was compared with the Monte Carlo approach and the
deterministic method of Runge Kutta 5th order. This study presented a good result of the response
curves, in a carefully studied combination of standard deviation of parameters.
Examining uncertainty through the decoupled form provides valuable insights into the individual
effects of each parameter. However, it is essential to note that the summation of these individual results
does not align with the consideration of multivariable uncertainty. This underscores the significance of
studying interactions between parameters for a comprehensive understanding of uncertainty.
Indeed, assessing the extreme values of parameters is crucial for comprehending their impact on
uncertainty simulations, particularly when considering interactions.
References
[1] International Energy Agency, Offshore Wind Outlook 2019: World Energy Outlook Special Report, 2019. Accessed: Mar.
04, 2022. [Online]. Available from www.iea.org/t&c/.
[2] Ghorbel, A., Graja, O., Hentati, T. et al. The Effect of the Brake Location and Gear Defects on the Dynamic Behavior of a
Wind Turbine. Arab J Sci Eng 45, 5421–5433 (2020). https://doi.org/10.1007/s13369-020-04439-0
[3] C. Bernuzzi, et al., Resonance of steel wind turbines: Problems and solutions, Structures, vol. 32, pp. 65–75, Aug. 2021.
DOI: 10.1016/j.istruc.2021.02.053.
[4] I. Bel Mabrouk, and A. El Hami, Effect of number of blades on the dynamic behavior of a Darrieus turbine geared
transmission system, Mech. Syst. Sig. Process., vol. 121, pp. 562–578, Apr. 2019. DOI: 10.1016/j.ymssp.2018.11.048.
[5] LIAO, Ding, ZHU, Shun-Peng, CORREIA, José AFO, et al. Fatigue reliability of wind turbines: historical perspectives,
recent developments and future prospects. Renewable Energy, 2022, vol. 200, p. 724-742.
[6] EL MOUHSINE, Sanaa, OUKASSOU, Karim, ICHENIAL, Mohammed Marouan, et al. Aerodynamics and structural
analysis of wind turbine blade. Procedia Manufacturing, 2018, vol. 22, p. 747-756.
[7] YANG, Yang, YE, Kehua, LI, Chun, et al. Dynamic behavior of wind turbines influenced by aerodynamic damping and
earthquake intensity. Wind Energy, 2018, vol. 21, no 5, p. 303-319.
[8] GUO, Shuangxi, LI, Yilun, et CHEN, Weimin. Analysis on dynamic interaction between flexible bodies of large-sized
wind turbine and its response to random wind loads. Renewable Energy, 2021, vol. 163, p. 123-137.
[9] GHORBEL, Ahmed, GRAJA, Oussama, BOUAZIZ, Amel, et al. Parametric Study of a Minimal Model of Wind Turbine
Drivetrain System. In : Mechatronics 4.0: Proceedings of the First International Workshop on Mechatronics 4.0, June 8–9,
2019, Mahdia, Tunisia 4.0. Springer International Publishing, 2020. p. 125-132.
[10] Mohammed, A.A.; et al.: Parametric study and comparison of aerodynamics momentum-based models for straight-bladed vertical
axis wind turbines. Arab. J. Sci. Eng. 45(2), 729–741 (2020)
[11] Abboudi, K.; Walha, L.; Driss, Y.; Maatar, M.; Fakhfakh, T.; Haddar, M.: Dynamic behavior of a two-stage gear train used in a
fixed-speed wind turbine. Mech. Mach. Theory 46, 1888–1900 (2011)
[12] Zhao,M.; Ji, J.C.: Nonlinear torsional vibrations of a wind turbine gearbox. Appl. Math. Model. 39, 4928–4950 (2015)
[13] Ghorbel, A.; et al.: Effect of the gear local damage and profile error of the gear on the drivetrain dynamic response. J. Theor.
Appl. Mech. 56, 765–779 (2018)
[14] M. Hadj Kacem, A. El Hami, H. Trabelsi, L. Walha, and M. Haddar, Considération de l’incertitude à l’étude du rendement
d’un multiplicateur biétage d’une éolienne, Incert Fia, vol. 5, no. 1, 2021. DOI: 10.21494/ISTE.OP. 2022.0778.
[15] A. El Hami, and B. Radi, Uncertainty and optimization in structural mechanics. New York: John Wiley & Sons, 2013.
[16] Guerine A, El Hami A, Walha L, Fakhfakh T, Haddar M (2016) Dynamic response of a spur gear system with uncertain
parameters. J Theor Appl Mech 54(3):1039–1049
[17] Guerine, A., Abderezzak, B. & Hami, A.E. Effect of Wheel Position of Gear System Vibration with Variables
Parameters. J. Vib. Eng. Technol. (2023). https://doi.org/10.1007/s42417-023-01134
[18] Fishman GS (1996) Monte Carlo Concepts, algorithms and applications, 1st edn. Springer Verlag, New York
[19] ZHANG, Jiaxin. Modern Monte Carlo methods for efficient uncertainty quantification and propagation: A survey. Wiley
Interdisciplinary Reviews: Computational Statistics, 2021, vol. 13, no 5, p. e1539.
[20] A. K. Panda, and S. v Modak, An FRF-based perturbation approach for stochastic updating of mass, stiffness and damping
matrices, Mech. Syst. Sig. Process., vol. 166, p. 108416, Mar. 2022. DOI: 10.1016/j.ymssp.2021.108416.
[21] F. Abid, et al., Surrogate models for uncertainty analysis of micro-actuator, Microsyst. Technol., vol. 26, no. 8, pp. 2589–
2600, Aug. 2020. DOI: 10.1007/s00542-020-04799-5.
17
[22] K. Dammak, S. Koubaa, A. Elhami, L. Walha, and M. Haddar, Numerical modeling of uncertainty in acoustic propagation
via generalized polynomial chaos, J. Theor. Appl. Mech., vol. 57, no. 1, pp. 3–15, Jan. 2019. DOI: 10.15632/jtam-pl.57.1.3.
[23] A. Guerine, A. El Hami, T. Fakhfakh, and M. Haddar, A polynomial chaos method to the analysis of the dynamic behavior
of spur gear system, Struct. Eng. Mech., vol. 53, no. 4, pp. 819–831, Feb. 2015. DOI: 10. 12989/sem.2015.53.4.819.
[24] Hassen Trabelsi, Amir Guizani, Maher Barkallah, Moncef Hammadi, Ayman Hadrich, and Mohamed Haddar,
Consideration of the uncertainty in the dimensioning of a gearbox of a wind turbine, J. Theor. Appl. Mech., vol. 59, no. 1, pp.
67–79, 2020. DOI: 10.15632/jtam-pl/128854.
[25] A. Guerine, A. El Hami, L. Walha, T. Fakhfakh, and M. Haddar, Dynamic response of wind turbine gear system with
uncertain-but-bounded parameters using interval analysis method, Renewable Energy, vol. 113, pp. 679–687, Dec. 2017. DOI:
10.1016/j.renene.2017.06.028.
[26] M. Hadj Kacem, A. El Hami, K. Dammak, H. Trabelsi, L. Walha & M. Haddar (2022): Consideration of multi-variable
uncertainty using the GPC method for the dynamic study of a two-stage gearbox of a wind turbine, Mechanics of Advanced
Materials and Structures, DOI: 10.1080/15376494.2022.2138650
[27] PERDANA, Abram. Dynamic models of wind turbines. Chalmers University of Technology, 2008.
[28] Ghorbel, A., et al. "Modal analysis and dynamic behavior for analytical drivetrain model." Journal of Mechanics 34.4
(2018): 399-415.
[29] Brahmi, Jemaa, Lotfi Krichen, and Abderrazak Ouali. "A comparative study between three sensorless control strategies
for PMSG in wind energy conversion system." Applied energy 86.9 (2009): 1565-1573.
[30] X. Zeng, J. Red-Horse, and R. Ghanem, Accelerated basis adaptation in homogeneous chaos spaces, Comput. Methods
Appl. Mech. Eng., vol. 386, pp. 114109, Dec. 2021. DOI: 10.1016/j. cma.2021.114109.
[31] A. Makhloufi, Y. Aoues, and A. El Hami, Reliability based design optimization of wire bonding in power microelectronic
devices, Microsyst Technol., vol. 22, no. 12, pp. 2737–2748, Dec. 2016. DOI: 10.1007/s00542-016-3151-5.
18
Declaration of Interest Statement
Declaration of interests
☐The authors declare that they have no known competing financial interests or personal relationships
that could have appeared to influence the work reported in this paper.
☒The authors declare the following financial interests/personal relationships which may be considered
as potential competing interests:
Ahmed Ghorbel reports was provided by Laboratory of Mechanics, Modeling, and Production,
National School of Engineering of Sfax, University of Sfax, Tunisia. If there are other authors, they
declare that they have no known competing financial interests or personal relationships that could
have appeared to influence the work reported in this paper.