Department of Marine Technology

Specialisation Project

Nonlinear Static Analysis and Eigenvalue

Analysis of a Lattice of Floating Wind
Turbines

Author: Claire MONIN

Supervisor: Thomas SAUDER

17 December 2021
Abstract

Offshore floating wind turbines have the potential to upscale wind energy production. The devel-
opment of this technology is nonetheless hindered by the costs of mooring systems. A proposition
to reduce these costs is to design farms of floating wind turbines connected to one another by
shared mooring lines, where only the platforms on the edges of the farm are anchored. Shared
mooring lines however introduce couplings in the motions of the wind turbines which can no longer
be studied independently. A 2D lumped mass-spring model of a lattice of floating wind turbines
is developed here to understand the coupling effects generated by the shared lines and estimate
resonance periods. When operating, wind turbines are subject to important mean loads, in par-
ticular due to wind loads on the rotor. Therefore, both a linear and a nonlinear static analysis are
conducted to take into account the offsets in the floating wind turbine positions that result from
this loading. Furthermore, to investigate the dynamic behaviour of a farm with shared moorings,
eigenvalue analyses are performed around the loaded and unloaded equilibrium configurations of
the lattice. While it is found that static offsets of the turbines are well captured by a linear model,
the nonlinear static analysis is necessary to evaluate coupling modes and natural periods of the
loaded lattice. Indeed, the mode shapes and natural periods are dependant on the loading con-
ditions on the floating wind turbines. Estimates of these natural periods indicate that it is low
frequency resonance that is of concern, nonetheless, the analysis of lattices of different size show
that natural frequencies tend to be higher for larger farms. The relevant excitation loads should
thus be identified in future work as well as the damping sources in the system to further evaluate
the response of the lattice to dynamic loading.

i
Table of Contents

1 Introduction 1

2 System Description and Method 2

2.1 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.2 Eigenvalue Analysis Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Linear Stiffness Matrix 5

3.1 Linear Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.2 Derivation of the Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4 Nonlinear Stiffness Matrix 9

4.1 Newton-Raphson Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 Derivation of the Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

5 Results and Discussion 13

5.1 Analysis of the linear system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5.1.1 (2x2) Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5.1.2 (3x3) Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Effect of a Mean External Force on a (2x2) lattice . . . . . . . . . . . . . . . . . . 19

5.2.1 Amplitude of the External Force . . . . . . . . . . . . . . . . . . . . . . . . 19

5.2.2 Direction of the External Force . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3 Effect of a Mean External Force on a (3x3) lattice . . . . . . . . . . . . . . . . . . 26

5.3.1 Amplitude of the External Force . . . . . . . . . . . . . . . . . . . . . . . . 26

5.3.2 Direction of the External Force . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.4 Effect of Increasing the Size of the Lattice . . . . . . . . . . . . . . . . . . . . . . . 35

5.4.1 Constant Stiffness Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . 36

5.4.2 Variable Stiffness Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6 Conclusion 40

References 41

Appendix 42

A Newton-Raphson Algorithm Validation 42

B Mode Shape Association for Different Loading Amplitudes 42

ii
C Effect of Force Amplitude On Natural Periods for Different Force Directions 44

D Line Stiffness Coefficients for (NxN) Lattices 48

iii
1 Introduction

Offshore wind turbines offer the opportunity to exploit locations with stronger and more stable
winds. While bottom-fixed designs are quite mature and have been implemented in several energy
production farms, floating offshore wind turbines are still in development. In contrast with bottom-
fixed structures, floating wind turbines (FWTs), which consist of floating platforms moored to the
seabed, can be installed in any water depth and are thus not restricted to shallow water. Creating
farms of FWTs will allow to upscale the production of energy from wind power, however the
development of such farms is hindered by the costs of installing such a large number of moored
structures.
One approach for reducing the cost of FWT farms is to use shared mooring lines. Currently, only
shared anchors have been developed at full scale, for the Hywind Tampen wind farm (Haslum 2019).
Shared anchors consist in connecting multiple turbines to the same anchor, however each turbine
still has the same number of lines connecting it to the seabed. Another approach is to anchor
only part of the FWTs to the seabed and to moor other turbines to the anchored ones. To have a
large number of turbines in the farm, the latter can be designed as a lattice of floating structures,
where the platforms are all connected to one another with shared lines but only some of them
are anchored. This concept would reduce the number of anchors and the number of anchor lines
used, hence reducing the costs associated with their installation. Nonetheless, connecting floating
platforms to one another will induce couplings in the motions of the FWTs. The behaviour of the
lattice needs to be characterized to design mooring lines and power cables in accordance with the
expected motions and loads in the farm. The dynamics of the lattice will depend both on farm
size and layout. The design of such farms is thus technically challenging and has yet to be tested.

Loads on mooring lines and displacements in a lattice of FWTs have been investigated by Gold-
schmidt and Muskulus (2015) who also provided an estimation of the cost savings obtained from
shared moorings. Studies of natural frequencies and mode shapes of a FWT lattice as well as time
and frequency domain analyses have been performed by Connolly (2018). Additionally, Connolly
and Hall (2019) and Wilson et al. (2021) have investigated different lattice layouts and proposed
an optimization procedure for the design of the mooring lines. Numerical simulations have been
performed on certain farm layouts, such as a two-FWT configuration modelled in Liang et al.
(2020). Hall and Connolly (2018) have also proposed the use of coupled numerical models that
are solved in parallel to simulate the overall behaviour of a square grid comprising four turbines.
Nonetheless, the design of FWT lattices still lacks fully integrated analyses of the whole system.
To further develop shared moorings in FWT farms and make them commercially feasible, both
validated numerical models and experimental tools still need to be developed to better characterize
and predict the behaviour of such farms with different layouts and sizes.
The scope of this study is limited to a simplified lattice model of a square grid of FWTs. The
goal is to obtain an estimate of the natural periods and associated mode shapes of the system
under different loading conditions. Estimating mode shapes and natural frequencies provides a
first insight on how the system will respond to dynamic excitation. It allows to evaluate resonance
in the lattice and associated displacement patterns. As all platforms are coupled by the mooring
lines, mode shapes and natural periods are expected to depend on the lattice layout and dimensions.
The study conducted here is based on the thesis of Connolly (2018) which is used as a starting
point to elaborate a FWT farm model. Both the initial square grid and deformed configurations of
the layout are considered. Indeed, operating wind turbines are subject to important mean loads,
caused by the wind loads on the rotor as well as other drift loads due to waves and current. A
wind turbine farm is thus seldom expected to be in its initial configuration and will more often
present a deformed lattice configuration where all FWTs have a mean offset. The effect of this
offset on the mode shapes and natural periods is investigated.
The first section of this report details the characteristics of the system and the method used
to obtain natural frequencies and mode shapes, namely the eigenvalue analysis method. In the
second part, the stiffness matrix of the lattice is derived within a linear approximation, while
in the third part, the stiffness matrix is established for a lattice in which FWTs present a mean
offset. Finally, the mode shapes and natural periods obtained from eigenvalue analyses for different
loading conditions and farm size are discussed.

1
2 System Description and Method

2.1 System Description

The wind turbine farm layout considered consists of a square grid in which only the turbines
situated on the edges of the farm are anchored. The lines on the outside of the farm are thus
anchor lines while all other moorings are shared lines that connect the platforms to one another.
An example (3x3) grid as well as the convention used for numbering the FWTs in the lattice
are presented in Figure 1. This layout presents the advantage of having rotational symmetry.
The number of wind turbines can also be changed easily and the model derived in the following
sections will be elaborated for a general (NxN) square grid. The x axis, also referred to as the
surge direction, is defined as the direction given by the vector going from FWT 0 to FWT 1. The
y axis, representing the sway direction, is defined by the direction of the vector going from FWT
0 to FWT N in a (NxN) lattice.

(a) (3x3) lattice (b) Numbering convention

Figure 1: Square grid layout considered and numbering convention. Illustrations are taken from
Connolly (2018). Black and white markers represent respectively the FWTs and the anchors.
Anchor lines are shown with dashed lines while shared lines are represented by full lines.

The FWT grid is modelled as a lattice of lumped masses connected by linear springs. The turbines
are modelled as point masses with two degrees of freedom in the horizontal plane. The motion of
each turbine is described by its displacements in surge and in sway which are respectively denoted
as xj and yj for a turbine j. The displacement vector characterizing the position of each turbine
in a lattice of (n + 1) turbines is defined as:
   
r0 x0
 r1   y0 
   
 r2   x1 
   
r =  r3  =  y1  (1)
   
 ..   .. 
 .   . 
   
 r2n  xn 
r2n+1 yn
This implies that for any turbine j, xj is also described by the system degree of freedom (DOF)
r2j and yj is described by r2j+1 .
The anchor lines and shared lines are modelled as linear springs which are characterized by an
unstretched length and a constant stiffness coefficient, both projected on the horizontal plane. In

2
general, each line could have a different length and stiffness. As explained in Wilson et al. (2021),
it could be interesting to have different restoring properties depending on how far a shared line
is from an anchor. Even so, because of the symmetry of the lattice, the sequence of stiffness
coefficients and line lengths encountered on a row will be the same for every row, and will be the
same as the values encountered on any column. For convenience, a vector k of stiffness values is
defined. k represents the consecutive stiffness coefficients of each line on a row or a column of the
lattice. For a (NxN) lattice, k is thus of length N+1. This allows to easily access the stiffness
coefficients of the lines connected to any turbine. If a FWT j is considered, the stiffness of the
line in -x direction and in +x directions are respectively ki and ki+1 where i is the remainder of
the division of j by N. The stiffness of the lines in -y and +y directions can be obtained as ki and
ki+1 with i equal to the quotient of the division of j by N. Similarly, a vector l of length values is
defined to represent the line lengths.

In the model considered in the following analyses, anchor lines have a stiffness coefficient k1 and
a length l1 , and for simplicity, all shared lines have the same stiffness coefficient k2 and the same
length l2 . The lines are assumed to be all pre-tensioned with the same tension magnitude T0 .
Values of the various parameters used are summed up in Table 1.

Parameter Symbol Value Unit

Mass of one FWT (turbine + platform) M 13.47 · 106 kg
Anchor lines stiffness coefficient k1 120 kN/m
Shared lines stiffness coefficient k2 60 kN/m
Anchor lines length on waterplane l1 726 m
Shared lines length on waterplane l2 1260 m
Pre-tension in each line T0 950 kN

Table 1: Parameters of the model

Mass, stiffness coefficients and line lengths have been chosen according to Connolly (2018) which
takes the DeepCwind semisubmersible as reference platform. The pre-tension has been chosen in
accordance with Hall and Goupee (2015) which investigates a mooring line model for the Deep-
Cwind semisubmersible. The rated thrust of the wind turbines is assumed to be 1500 kN.

2.2 Eigenvalue Analysis Method

The eigenvalue analysis method will be used to determine the natural frequencies of the system and
the associated mode shapes that describe how the system deforms when oscillating at each natural
frequency. When performing an eigenvalue analysis it is assumed that the system is linear. The
method will thus be used to study small oscillations around different equilibrium configurations
that depend on the mean forces applied to the FWTs. The theory presented here is based on the
lecture notes from the Marine Dynamics course from Bachynski et al. (2019).

As described in Section 2.1, the lattice of FWTs is represented by an undamped mass-spring system.
The lattice is characterized by a mass matrix M and a stiffness matrix K that are independent of
the displacement vector r. Free oscillations of the system are then described by:
M r̈ + Kr = 0 (2)

The explicit shape of the mass and stiffness matrices is described in equation 3. Each matrix
coefficient Mij and Kij relates the force in degree of freedom (DOF) i with a displacement in DOF
j.
r
Krr00 Krr01 . . . Krr02n Kr02n+1
 r   
Mr00 0 ... 0 0
r
 0 Mrr11 . . . 0 0   Krr0 Krr11 . . . Krr12n Kr1 2n+1
1
. . .. .. 
   
 .. .
.. . .. .
.. .
..  K =  .. .. ..
M = . (3)
 
. . . 

r
  r
 Kr r r r 2n+1

 0 0 . . . Mr2n 2n
0  0
2n
Kr2n 1
. . . Kr2n 2n
Kr2n 
r2n+1 r2n+1
0 0 ... 0 Mr2n+1 Krr2n+1
0
Krr2n+1
1
. . . Krr2n+1
2n
Kr2n+1

3
It should be noted that the representation of the mass matrix as a diagonal matrix is correct for
the system considered in the following analyses but does not represent the general case. On the
other hand, in a FWT lattice, the off-diagonal terms in the stiffness matrix are generally non-zero
due to the coupling between the different platform motions introduced by the shared lines.
For a linear system oscillating at the radial frequency ω, equation 2 can be rewritten in frequency
domain as:
(−ω 2 M + K) · r = 0 (4)

An eigenvalue analysis consists in finding the values of ω, and the corresponding displacement
vectors, that satisfy the eigenvalue problem:

M −1 Kr = ω 2 r f or r 6= 0 (5)

Solutions of this problem will be the natural frequencies of the system and their associated mode
shapes. Natural frequencies give the excitation frequencies that will trigger resonance and can lead
to large motions in the system. Mode shapes represent the patterns of displacement associated
with each eigenfrequency.
For equation 5 to have a solution, the determinant of the matrix −ω 2 M + K must be zero.
det(−ω 2 M + K) = 0 is called the characteristic equation and gives a polynomial equation in ω 2
of order equal to the number of DOF in the system. For a lattice of (NxN) wind turbines, the
characteristic equation will give a polynomial of order 2N 2 . Additionally, ω should be non negative
thus 2N 2 natural frequencies are found when solving the eigenvalue problem. For each natural
frequency ωj , a mode shape φj can be derived by inserting ωj in equation 4. In the following
analyses, the eigenvalue problem is solved using the function linalg.eig from the python numpy
library. Each mode shape is normalized by setting its 2-norm to 1.

The stiffness matrix of the system will be derived within a linear approximation to perform a
first eigenvalue analysis. A nonlinear static analysis will also be conducted to obtain the stiffness
matrix of the lattice when it is loaded by mean external forces. Eigenvalue analyses will then be
performed using this nonlinear stiffness matrix.

4
3 Linear Stiffness Matrix

As a starting point, properties of the undisturbed initial farm configuration are studied. The
stiffness and mass matrix of the system are established within a linear approximation.

3.1 Linear Approximation

The displacements of each turbine around the unloaded equilibrium configuration are considered
to be small.
To understand the consequences of this assumption, a small displacement in surge of a FWT j is
considered as shown on Figure 2.

Figure 2: Small angle approximation for the linear analysis of a lattice. α is considered small such
that dL ≈ 0

While springs positioned in x direction will generate restoring forces on the turbine, it can be
shown that the force generated by the springs in y direction is negligible. The elongated length of
a spring in y direction verifies:
L
= cos(α)
L + dL
where notations are defined in Figure 2. For a small displacement r2j , α is small and cos(α) ≈ 1.
From this approximation:
L ≈ L + dL
The springs in y direction are not elongated and thus generate no force on turbine j.
Generally, for any small displacement in x or y directions, the elongation of the springs in the
orthogonal direction is approximately zero and the restoring force generated by these springs is
considered to be zero. The only turbines that are affected by the displacement are the neighbouring
turbines that are along the same axis as the displacement. Taking these considerations into account
will simplify the derivation of the stiffness matrix as a lot of degrees of freedom will not be coupled.
The effect of the linear assumption can thus be summed up as:

1. forces in surge direction will not affect degrees of freedom in sway and forces in sway will not
affect degrees of freedom in surge,

5
 2. two turbines that are not directly connected by a line will not be coupled: a motion in one
of the two will not affect the other.

3.2 Derivation of the Stiffness Matrix

The stiffness matrix coefficient Kij relates the force in DOF i with a displacement in DOF j through
the equation:
Fi = −Kij · rj (6)
Equilibrium of the forces acting on each FWT is analysed to establish the stiffness matrix. Restor-
ing forces will be induced both by the elongation of the lines as the turbines displace and by the
pre-tension in the lines. First, equations of motions will be established for any turbine that is con-
nected only to other turbines, with shared lines. Secondly, the case of turbines that are connected
to one or more anchors will be examined.
Restoring generated by the line elongation
A turbine j connected to four other turbines is considered. The degrees of freedom and spring
constants affecting the equations of motion of FWT j are presented in Figure 3.

Figure 3: Spring configuration and DOF notation for a ”middle” turbine j only connected to shared
lines.

As a consequence of the linear assumption described previously, equilibrium in surge for a turbine
j can be expressed as:

mj · r̈2j = −(ki + ki+1 ) · r2j + ki · r2(j−1) + ki+1 · r2(j+1) (7)

where i is the remainder of j divided by N.

Similarly, equilibrium in sway for a turbine j can be written as:

mj · r̈2j+1 = −(ki + ki+1 ) · r2j+1 + ki · r2(j−N )+1 + ki+1 · r2(j+N )+1 (8)

where i is the quotient of j divided by N.

Restoring generated by the pre-tension

A pre-tension of magnitude T0 is applied by each line on each FWT. When a turbine is displaced
in the surge direction, the elongation of the springs in sway direction is considered to be zero as a

6
result of the linear assumption, however pre-tension of the springs in sway direction will provide a
restoring force in the surge direction.
To calculate this restoring force, the effect of a small displacement in surge r2j on the pre-tension
of a line initially in the y direction is quantified. Figure 4 shows the system before and after the
displacement. Only the forces due to pre-tension are represented. The restoring force due to the
line in -y direction is derived. The effect of the line in +y direction will be deduced by symmetry.

(a) (b)

Figure 4: Pre-tension forces applied on a turbine. (a) shows the initial configuration and (b) shows
the configuration after a small displacement in surge of the turbine.

When FWT j is in its initial undisplaced position, the load generated by the pre-tension of the line
in -y direction is:  
(0) 0
F = (9)
−T0
After a displacement r2j , this load is expressed as:
   −T0 
−T0 · sin(dθ) · r2j
F (1) = ≈ li (10)
−T0 · cos(dθ) −T0
Hence, the restoring due to the pre-tension is:
 −T0 
(1) (0) li · r2j
∆F = F −F = (11)
0
Similarly, the restoring due to the pre-tension in +y direction is given by:
 −T0 
· r2j
∆F = F (1) − F (0) = li+1 (12)
0
A displacement in surge will thus generate a restoring force in surge due to the pre-tension of the
lines but none in sway. The equation of motion in surge for a turbine j will be modified and will
include two more terms accounting for the pretension in the two springs in y direction. Equation
7 becomes:
T0 T0
mj · r̈2j = −(ki + ki+1 ) · r2j + ki · r2(j−1) + ki+1 · r2(j+1) − ( + ) · r2j (13)
li li+1
where i is the remainder of j divided by N.
Similarly, equation 8 in sway becomes:
T0 T0
mj · r̈2j+1 = −(ki + ki+1 ) · r2j+1 + ki · r2(j−N )+1 + ki+1 · r2(j+N )+1 − ( + ) · r2j+1 (14)
li li+1

7
where i is the quotient of j divided by N.
All the diagonal terms of the stiffness matrix are thus increased by an amount ( Tli0 + T0
li+1 ).

Anchored turbines
The established equations are valid for turbines that are surrounded by other turbines. In the case
of platforms that are connected to one or two anchor lines, the equilibrium equations 13 and 14 are
modified by setting to zero the degrees of freedom that are fixed by the anchor. As an example,
Figure 5 shows the configuration for a turbine that is anchored in the -x and in the -y directions.
In that configuration the degrees of freedom r2(j−1) and r2(j−N )+1 are set to zero in the equations
of motion.

Figure 5: Spring configuration and DOF notation for an anchored turbine.

Four cases can be distinguished:

• If the turbine is anchored in the -x direction, then j is a multiple of N and r2(j−1) is set to 0
in equation 13.
• If the turbine is anchored on the +x direction, then j+1 is a multiple of N and r2(j+1) is set
to 0 in equation 13.
• If the turbine is anchored in the -y direction, then j is strictly inferior to N and r2(j−N )+1 is
set to 0 in equation 14.
• If the turbine is anchored in the +y direction, then j is strictly superior to (N 2 − 1) − N and
r2(j+N )+1 is set to 0 in equation 14.

8
4 Nonlinear Stiffness Matrix

Loads on the rotors as well as mean drift loads from waves and current will induce an offset in the
equilibrium position of each FWT of the lattice. It is therefore important to study the dynamics
of the system around an equilibrium configuration with non zero offsets. To perform eigenvalue
analyses around the displaced farm configurations, the stiffness matrix of the deformed lattice must
be obtained.

A nonlinear static analysis is performed to determine the loaded equilibrium configuration of the
lattice. The linear dynamic behaviour of the system around this equilibrium is studied using the
stiffness matrix found from the nonlinear analysis. To derive this matrix, the assumption that
displacements and angles are small can no longer be considered valid and for each wind turbine,
the restoring from all four lines has to be taken into account. For a mean external force applied
on every turbine, the stiffness matrix and the displacements in the lattice are calculated through
an iterative Newton-Raphson method.

4.1 Newton-Raphson Method

When considering large displacements of a FWT, the restoring forces from its four mooring lines
are no longer linear. The displacements of the FWTs come into account in the stiffness matrix and
introduce geometric nonlinearities. When an external force Fext is applied on the system, both
the displacements r and the stiffness matrix which is a function of the displacements are unknown.
Solving for the displacements would require solving the nonlinear equations obtained when setting
the external forces equal to the internal forces, which is not in general possible to do analytically.
Hence, a numerical iterative method is used.
The Newton-Raphson method is applied and is presented here as described in the lecture notes
from the Non-Linear Finite Element Analysis course from NTNU (n.d.). The residual force which
exists when the system is out of balance is introduced.

R(r) = Fext + Fint (r) (15)

Fext is the constant external force vector applied on the lattice. Fint is the internal force vector,
which can be expressed as a function of displacement and which satisfies Fint = −Fext at equilib-
rium. The Newton-Raphson method aims at driving the residual force to zero. It searches for a
displacement r such that:
R(r) = 0
A Taylor expansion of R around an initial guess r0 gives:
 
dR(r0 )
R(r) ' R(r0 ) + (r0 ) · (r − r0 ) (16)
dr

Equation 16 can be solved for ∆r = r − r0 :

 −1
dR(r0 )
∆r = − · R(r0 ) (17)
dr

An improvement of the initial guess is then obtained as r1 = r0 + ∆r. By iteratively solving this
equation, the residual can be driven to zero. The final displacement is obtained as a result. The
stiffness of the system is obtained from the residual and is defined as:
     
dR d(Fext + Fint ) dFint
K=− =− =− (18)
dr dr dr

A coefficient (K)ij of the stiffness matrix is thus given as:

∂(Fint )i
(K)ij = − (19)
∂(r)j

9
Equation 17 which gives the improvement on the displacement at each step can be rewritten in a
more compact form:
∆r = K −1 · R (20)
where K and R are updated at each iteration.
The steps taken by a Newton-Raphson algorithm can be summed up as the following. For a given
external force Fext :

1. An initial displacement is calculated using an initial stiffness matrix: r0 = K0−1 · Fext . K0 is

often derived from a linear analysis of the system.
2. Internal forces Fint,0 due to the displacement r0 are calculated and the residual R0 is evalu-
ated.
3. The stiffness matrix is calculated from Fint,0 using equation 19.

4. The new displacement r1 = r0 + ∆r is evaluated using equation 20.

5. The steps are repeated from step 2 until all residual forces are close enough to zero.

4.2 Derivation of the Stiffness Matrix

To apply the Newton-Raphson method, we must be able to calculate the internal forces for any
displacement in the lattice, namely, the vector Fint should be expressed as a function of the
displacements r.

The forces acting on a turbine j are analysed. The turbine is assumed to be connected to four
other platforms. The case of anchored turbines is derived from the general case as was done in the
derivation of the linear stiffness matrix (Section 3). For simplicity, the degrees of freedom in surge
and in sway will be referred to as xj and yj respectively rather than r2j and r2j+1 .

All notations are defined in Figure 6. Four forces due to the mooring lines act on the turbine and
are referred to as Fa , Fb , Fc and Fd . Corresponding angles αa , αb , αc and αd are defined as the
angles between the original line axis and the line axis in the displaced configuration. Angles are
taken positive in the anti-clockwise direction.

10
Figure 6: Notations for the calculation of restoring forces on a FWT j when all turbines in the
lattice present an offset compared to their initial position.

Through geometric considerations, the restoring forces applied on FWT j due to the elongation
and pre-tension of each line can be expressed as:
Force a:
 
−1 yj − yj−1
αa = tan
li − xj−1 + xj
 
yj − yj−1
Fa = −ki · − li − T0 (21)
sin(αa )
Fa,x = −ki · (li − xj−1 + xj − li · cos(αa )) − T0 · cos(αa )
Fa,y = −ki · (yj − yj−1 − li · sin(αa )) − T0 · sin(αa )

Force b:
 
xj−N − xj
αb = tan−1
li − yj−N + yj
 
xj−N − xj
Fb = −ki · − li − T0 (22)
sin(αb )
Fb,x = ki · (xj−N − xj − li · sin(αb )) + T0 · sin(αb )
Fb,y = −ki · (li − yj−N + yj − li · cos(αb )) − T0 · cos(αb )

11
Force c:
 
yj+1 − yj
αc = tan−1
li+1 − xj + xj+1
 
yj+1 − yj
Fc = ki+1 · − li+1 + T0 (23)
sin(αc )
Fc,x = ki+1 · (li+1 − xj + xj+1 − li+1 · cos(αc )) + T0 · cos(αc )
Fc,y = ki+1 · (yj+1 − yj − li+1 · sin(αc )) + T0 · sin(αc )

Force d:
 
xj − xj+N
αd = tan−1
li+1 − yj + yj+N
 
xj − xj+N
Fd = −ki · − li+1 + T0 (24)
sin(αd )
Fd,x = −ki+1 · (xj − xj+N − li+1 · sin(αd )) − T0 · sin(αd )
Fd,y = ki+1 · (li+1 − yj + yj+N − li+1 · cos(αd )) + T0 · cos(αd )

where i is the remainder of the division of j by N for forces a and c and is the quotient of the
division of j by N for forces b and d.
The forces in surge and sway on turbine j are then obtained respectively as:

Fint,2j = Fa,x + Fb,x + Fc,x + Fd,x (25)

Fint,2j+1 = Fa,y + Fb,y + Fc,y + Fd,y (26)

Once the internal forces are expressed as a function of the displacements, the Newton-Raphson
method can be applied to obtain the stiffness matrix and the displacements of the system subjected
to an external force at equilibrium. The initial stiffness matrix K0 used to start the iterative
procedure is the stiffness matrix obtained from the linear analysis of the system around its original
unloaded configuration. At each iteration, the stiffness coefficients are determined by calculating
the derivatives of the internal forces as expressed by Equation 19. These derivatives are computed
using a finite difference method. Each column j of the stiffness matrix is calculated as:

−(fint (r + δrj ) − fint (r))

(K)j = (27)
h
r is the displacement at the current iteration. fint is the function giving the internal force vector
for a displacement vector. h is the step taken in the finite difference formula and δrj is a vector
with a value h on its j th coefficient and zero everywhere else, that represents a small variation in
DOF j.

Validation of the stiffness matrix obtained through the numerical method is not straightforward as
the stiffness matrix of the nonlinear system cannot be obtained analytically. The implementation
of the nonlinear static analysis has thus been tested on a simpler two degree of freedom system
which is detailed in Appendix A.
As a result from the Newton-Raphson method, the static offset of each FWT when a mean external
force is applied is obtained, as well as the stiffness matrix of the loaded lattice configuration that
will be used for an eigenvalue analysis around the new equilibrium configuration.

12
5 Results and Discussion

In the following section, results for both the linear unloaded and the nonlinear loaded system are
discussed. A (2x2) and a (3x3) lattice are first studied in details. Natural frequencies of a more
general (NxN) lattice are then investigated.

5.1 Analysis of the linear system

The eigenvalue analysis method is first applied to the linear system using the stiffness matrix
derived in Section 3. The results are presented for a (2x2) lattice and a (3x3) lattice of FWTs.

5.1.1 (2x2) Lattice

The analysis of a (2x2) grid of point masses with two degrees of freedom each gives 8 eigenvalues.
However, when applying the eigenvalue analysis method to the FWT lattice only two different
eigenvalues are found which are each repeated four times. The two natural periods of the system
have the following values:

T1 46.87s
T2 66.00s

Table 2: Natural periods obtained for a (2x2) lattice

The associated mode shapes are presented in Figure 7 and Figure 8. It can be seen that for each
period, there are four degenerate mode shapes. This means there are four mode shapes associated
with T1 and four associated with T2 . When the system oscillates at one of these two natural
frequencies, the motion will be a linear combination of the four degenerate modes. The mode
shapes at T2 , thus at the lowest frequency, show a pattern where neighbouring turbines move
together. On the other hand, the modes oscillating at higher frequency, at T1 , present a pattern
where neighbouring turbines move in opposite direction.

13
Figure 7: Mode shapes for a (2x2) lattice with linear assumption, T1 = 46.87s.

14
 Figure 8: Mode shapes for a (2x2) lattice with linear assumption, T2 = 66.00s.

5.1.2 (3x3) Lattice

For a (3x3) grid, three different natural periods are found, as presented in Table 3, each with 6
corresponding degenerate mode shapes.

T1 46.89s
T2 54.05s
T3 92.84s

Table 3: Natural periods obtained for a (3x3) lattice

Figures 9, 10 and 11 show the mode shapes associated respectively with T1 , T2 and T3 . As it
was observed for the (2x2) lattice, the lowest frequency modes for the (3x3) lattice, oscillating
at T3 , present a pattern where all turbines move together. The modes observed at T2 show a
pattern where turbines alternately move outwards and inwards. Finally, the pattern appearing at
the highest frequency is one where the motion of neighbouring turbines are out of phase.

15
Figure 9: Mode shapes for a (3x3) lattice with linear assumption, T1 = 46.89s.

16
Figure 10: Mode shapes for a (3x3) lattice with linear assumption, T2 = 54.05s.

17
Figure 11: Mode shapes for a (3x3) lattice with linear assumption, T3 = 92.84s.

18
The results obtained by the linear analysis can be compared to the results obtained by Connolly
(2018) who also performs an eigenvalue analysis of a linear model of the FWT lattice. The natural
periods found by Connolly (2018) on a (2x2) lattice correspond well, with values of 47.07s and
66.56s. The values found on a (3x3) lattice in his thesis are approximately of 35s, 42s and 74s
which differs with the values found here. This is explained by the fact that the stiffness coefficients
used for the lines were not the same. It is also noted that the mode shapes found by Connolly
(2018) present the same degeneracy as is observed here.
The analysis of the linear system will be used as a reference case for the following nonlinear analyses
of the loaded (2x2) and (3x3) lattices.

5.2 Effect of a Mean External Force on a (2x2) lattice

The effect of applying a mean external force Fext on the lattice is studied. Results are first
detailed for a (2x2) lattice where it is easier to visualize the behaviour of the system, and will then
be presented for a (3x3) lattice.

A static external force is applied on each FWT of the system to represent the effect of mean loads
from wind, waves and current. The offset of each turbine due to this force is calculated using a
nonlinear static analysis as detailed in Section 4. From this analysis, a stiffness matrix for the
deformed grid configuration is obtained, which is used to conduct an eigenvalue analysis.

The maximum offset generated by Fext as well as the effect of Fext on the natural periods and
mode shapes are investigated. The turbines are assumed to have a rated thrust of 1500kN . The
amplitude of Fext will thus be varied between 0 and 2000kN . Due to the symmetry of the layout,
the direction of the external force only needs to be varied between 0 and 45 degrees to represent
all possible directions, where the angle characterising the force direction is defined as the angle
between the x axis and −Fext , that is, an angle of 0 degrees corresponds to a force in -x direction
and an angle of 45 degrees corresponds to a force with equal positive components in -x and -y
directions.

5.2.1 Amplitude of the External Force

The effect of increasing the external force amplitude on a (2x2) lattice is first presented. Fext is
applied with a 15 degrees angle. This force direction is chosen in order to keep the loading case as
general as possible and avoid load cases that preserve the symmetry of the system, such as 0 and
45 degrees.
Maximum offset

The maximum offset in the lattice obtained through the nonlinear static analysis is presented as a
function of the amplitude of Fext as the blue curve in Figure 12. The orange curve represents the
maximum offset obtained if the linear approximation is assumed valid. The displacements of the
linear system are obtained by multiplying the external force vector by K0−1 where K0 is the linear
stiffness matrix as defined in Section 3.

19
Figure 12: Maximum displacement (maximum of the norm of the displacement of each FWT)
in a (2x2) lattice for different loading amplitudes. The blue curve represents the displacement
computed through a nonlinear static analysis, and the orange curve represents the displacement
calculated using a linear static analysis.

The results show the maximum offset curve from the nonlinear analysis is very close to the linear
analysis curve. The system nonlinearities do not appear to be relevant when considering the static
offset of the FWTs in the farm. However, the results of the eigenvalue analysis show the geometric
nonlinearity of the system influences the natural periods and mode shapes, and will thus influence
the dynamic behaviour of the lattice.
Natural periods

Figure 13 shows the natural periods calculated through the eigenvalue analysis for each loading
amplitude. The dashed lines represent the eigenvalues obtained from the analysis on the linear
system. It can be seen that using the nonlinear stiffness matrix that takes into account the
geometric nonlinearity of the system rather than the linear stiffness matrix affects the natural
periods that are found.

20
Figure 13: Natural periods for a (2x2) lattice for different loading amplitudes. The blue dashed
lines represent the periods that are found when performing an eigenvalue analysis with the stiffness
matrix from the linear analysis.

When a force is applied on the system, less degenerate mode shapes are observed. It can be seen
that some eigenvalues which are equal at Fext = 0 split into differentiated eigenvalues when a force
is applied. The natural period values spread when the amplitude of the load increases.
Mode shapes
From the curves presented in Figure 13, each eigenvalue at a given amplitude of Fext can be assigned
a predecessor among the eigenvalues at the previous loading amplitude, that is, an eigenvalue at the
previous amplitude from which it stems. For each eigenvalue of the system, the successive values it
takes as Fext is increased can be determined to reconstruct the path taken by each eigenvalue. This
path represents how a natural period is modified when Fext is increased. Since each eigenvalue is
associated with a mode shape, the evolution of the latter as Fext increases can be determined.
Eigenvectors are thus associated to one another for different loading amplitudes in order to visualize
how a mode shape of the system evolves with variations in the external force amplitude. This is
accomplished using the Modal Assurance Criterion (MAC), which is a number between 0 and 1,
expressing how correlated two mode shapes are.
The mode shapes at the largest loading amplitude Fext = 2000kN are used as starting points
and are first paired with mode shapes at 1500kN. For a given eigenvector from the analysis at
2000kN, the MAC of that vector with each mode shape at 1500kN is calculated. A MAC close
to 1 means the two vectors considered have a very similar shape, while a MAC close to 0 means
the two vectors represent very different patterns. Each mode shape at 2000kN is paired with the
mode shape from the set at 1500kN that yields the highest MAC number. This procedure can

21
be repeated for the comparison of sets of mode shapes at lower force amplitudes, for example to
compare analyses at 1000kN and at 1500kN, and then to compare results at 500kN with results
at 1000kN, until the whole force amplitude range is covered. In this manner, starting from a
mode shape at 2000kN, it can be determined from which mode shapes it is derived at 1500kN,
500kN, 1kN and the modification of the displacement patterns when the external force magnitude
is increased can be represented. More details on the method used to associate mode shapes are
presented in Appendix B.
The mode shapes association for Fext ranging from 0 to 2000kN is presented in Figure 14 and
Figure 15. The case in which a force of 1kN is applied is used to represent the unloaded system.
To be exact, the unloaded system behaviour should be obtained from the linear analysis where
Fext = 0, however, the linear system presents much more degenerate modes than a loaded system
and it is harder to associate mode shapes when a lot are degenerate. Indeed, many modes will
be represented by the same eigenvalue at the unloaded configuration and thus will all come from
the same four degenerate modes that in reality do not occur independently as they correspond
to the same natural period. A loading of 1kN is thus applied to differentiate the eigenvalues and
more easily associate mode shapes between 1kN and 500kN. Moreover, 1kN remains a low force
magnitude compared to the expected loads on the FWTs when in operation. It will thus generate
very small offsets and can be representative of an unloaded system.
In all the figures representing mode shape associations, the static displacement of the FWTs is
represented and eigenvectors are plotted at the displaced position of the FWTs. The MAC number
given for each force amplitude corresponds to the MAC between the mode shape selected at this
amplitude, and the one selected at the higher amplitude. Hence, the MAC number specified for
Fext = 1kN corresponds to an association between the set at 1kN and the set at 500kN.

22
Figure 14: Mode shapes association for a (2x2) lattice for a varying load magnitude. Higher
frequency modes are grouped here. Fext is applied with a 15 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher.

23
Figure 15: Mode shapes association for a (2x2) lattice for a varying load magnitude. Lower
frequency modes are grouped here. Fext is applied with a 15 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher.

Even though natural frequencies are no longer degenerate, mode shapes can still be separated into
higher frequency modes, presented in Figure 14 and lower frequency modes, presented in Figure 15.
For some of the lower frequency modes the overall shape of the pattern is not modified much and
the modal assurance criterion is very close to 1. For other modes, both at high and low frequency,
the pattern seems to change progressively as Fext is increased. Finally, for part of the higher
frequency modes, the mode shapes are changed as soon as the system is subjected to a significant
load, but are then not much modified when the loading amplitude is increased. This is observed
for the two last high frequency modes presented. For these, the MAC between the modes at 1kN
and at 500kN is of 0.09 and of 0.5, whereas the MAC for larger force amplitudes is above 0.9.

24
5.2.2 Direction of the External Force

The effect of changing the direction of the external force is investigated. The force amplitude is
taken equal to 1500kN and the direction of the force is varied between 0 and 45 degrees with steps
of 15 degrees.

Maximum offset
The maximum offset in the lattice, that is the maximum displacement norm amongst all turbines, is
calculated using the nonlinear static analysis described in Section 4 and is represented in Figure 16.
It is observed that the changes in maximum displacement norm when the direction of Fext is
modified are negligible. The difference between the highest and the lowest maximum displacement
is only of approximately 0.5% of the lowest maximum displacement.

Figure 16: Maximum offset norm in a (2x2) lattice for an external force of 1500kN applied with
different angles, and applied on each FWT

Natural periods
Figure 17 presents the natural periods obtained from the eigenvalue analysis on the lattice with a
force applied with different angles. The force direction does not seem to have a significant effect
on the eigenfrequencies.

25
Figure 17: Natural periods of a (2x2) lattice for an external force of 1500kN applied with different
angles, and applied on each FWT

5.3 Effect of a Mean External Force on a (3x3) lattice

The same analysis as was conducted on a (2x2) lattice is performed for a (3x3) lattice, which is
of interest as it includes FWTs that are connected to only one anchor as well as a FWT that is
connected only to shared lines.

5.3.1 Amplitude of the External Force

The effect of increasing the loading amplitude is studied for a force applied on each FWT with a
direction of 15 degrees.

Maximum offset
Figure 18 presents the maximum offset obtained from the nonlinear static analysis compared to
the maximum offset obtained from a linear analysis when Fext is increased. It is observed again
that the result from the nonlinear analysis is close to the linear analysis result.

26
Figure 18: Maximum displacement (maximum of the norm of the displacement of each FWT)
in a (3x3) lattice for different loading amplitudes. The blue curve represents the displacement
computed through a nonlinear static analysis, and the orange curve represents the displacement
calculated using a linear static analysis.

Natural periods
Natural periods of the system for different loading amplitudes are presented in Figure 19. Similarly
as for a (2x2) lattice, the natural periods spread around the initial period values of the unloaded
system. However, compared to the (2x2) lattice, the way eigenvalues change as Fext is increased is
more complex. Figure 20 presents a detail of the curves presented in Figure 19. It can be seen that
some degenerate natural periods split into eigenvalues that later become degenerate with another
value. This shows the need to define a criterion to associate mode shapes and is the reason why
the modal assurance criterion has been used to reconstruct the evolution of each mode shape when
the force is increased.

27
Figure 19: Natural periods for a (3x3) lattice for different loading amplitudes. The blue dashed
lines represent the periods that are found when performing an eigenvalue analysis with the stiffness
matrix from the linear analysis.

Figure 20: Detail of the natural periods curves for a (3x3) lattice for different loading amplitudes
shown in Figure 19.

28
Mode shapes
Using the same method as explained for a (2x2) lattice and as detailed in Appendix B, the set
of mode shapes at 2000kN are associated with the mode shapes at the previous amplitudes to
visualize how a mode shape is modified when Fext increases. Three groups of mode shapes can
be distinguished as the eigenvalues presented in Figure 19 show. Higher frequency modes are
presented in Figure 21, modes at a slightly lower frequency are presented in Figure 22, and finally,
significantly lower frequency modes are presented in Figure 23. Compared to a (2x2) lattice, it can
be seen that more mode shapes association have a low MAC which means eigenvectors are subject
to more modifications when Fext is increased.

29
Figure 21: Mode shapes association for a (3x3) lattice for a varying load magnitude. Highest
frequency modes are grouped here. Fext is applied with a 15 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher.

30
Figure 22: Mode shapes association for a (3x3) lattice for a varying load magnitude. Second
highest frequency modes are grouped here: frequency is slightly lower than for the group of highest
frequency modes but is still closer to the highest frequencies than to the lowest frequencies. Fext
is applied with a 15 degrees angle. The MAC is given for the association with the mode shape at
the force amplitude that is one step higher.

31
Figure 23: Mode shapes association for a (3x3) lattice for a varying load magnitude. Lower
frequency modes are grouped here. Fext is applied with a 15 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher.

32
Conclusion on the effect of increasing the external force amplitude
From both the analyses on the (2x2) and the (3x3) lattice, it can be seen that increasing the
amplitude of the external force on each FWT of the lattice has the following effects. The maximum
displacement obtained by the nonlinear static analysis increases with Fext and is very close to the
displacement that would be obtained from a linear analysis. However, the geometric nonlinearities
of the system are of importance when studying its dynamic behaviour. As the amplitude of the
loading is increased, the number of differentiated natural periods increases. Eigenvalues spread
around the initial unloaded system natural periods. There are thus less degenerate mode shapes.
It can be deduced from the mode shapes association of the (2x2) lattice that applying a significant
load on the system, such as an external force of 500kN on each FWT, modifies significantly some
of the mode shapes compared to an unloaded system. For part of the mode shapes, the pattern
is greatly modified when a force is applied, but does not change much when the amplitude of the
force is increased from 500kN to 2000kN. On the other hand, for other mode shapes, when Fext
is increased, the pattern is progressively modified. For these, when two loading amplitudes are
close to one another the pattern is similar, however for large differences in loading amplitude, the
difference in the mode shapes becomes significant. The resulting mode shapes when the system is
loaded with 2000kN is thus different from the mode shapes of the unloaded system. For a (3x3)
lattice, the effect of increasing Fext on the evolution of the mode shapes is more ambiguous. It can
be that from 1500kN to 2000kN the mode shapes are modified substantially, while from 1000kN
to 1500kN only minor changes are observed.

5.3.2 Direction of the External Force

The effect of the external force direction is also studied on the (3x3) lattice.
Maximum offset
Figure 24 shows the maximum displacement norm in a (3x3) FWT grid. As it was concluded for a
(2x2) lattice, it can be seen that the force direction does not have a significant effect on the norm
of the maximum offset.

33
Figure 24: Maximum offset norm in a (3x3) lattice for an external force of 1500kN applied with
different angles, and applied on each FWT

Natural periods
The natural periods of a (3x3) lattice of FWTs for different force directions are represented in
Figure 25. For each group of eigenvalues (eigenvalues around 46s, around 54s and around 92s),
the maximum and minimum natural period are not significantly affected by the changes in force
direction. However, the distribution of the periods between those minima and maxima change
and the degeneracy of the eigenvalues is modified. The way in which the mode shapes evolve as
the amplitude of Fext is increased is thus expected to present differences depending on the angle
with which Fext is applied. The plots presenting the way eigenvalues are modified when the load
amplitude is increased for different force directions and the mode shape association for a different
force direction can be found in Appendix C. It is observed that for a given force amplitude, the
mode shapes depend on the force direction.

34
Figure 25: Natural periods of a (3x3) lattice for an external force of 1500kN applied with different
angles, and applied on each FWT

Conclusion on the effect of the loading direction

From the analyses conducted on the (2x2) and the (3x3) lattice it appears that the force direction
does not have a significant effect neither on the maximum offset in the grid nor on the natural
period values. However, the angle with which the force is applied influences the degeneracy of the
eigenvalues and affects the mode shapes which will thus depend both on force amplitude and force
direction.

5.4 Effect of Increasing the Size of the Lattice

The effect of an increased lattice size on the natural periods of the system is presented. For a
(NxN) square grid, N is varied between 2 and 6. It has been seen that loading of the lattice affects
the natural period values. Loaded conditions are more representative of a FWT in operation,
therefore, a mean external force of 1500kN with a 15 degrees angle is applied on each FWT in the
following analyses. A first analysis is conducted by varying N and keeping all other parameters
constant and equal to what has been defined in Section 2.1. However, when designing a FWT farm,
the stiffness coefficients of the mooring lines would most likely be adapted to the size of the lattice,
with stiffer lines for larger lattices to prevent too large displacements. A second analysis is thus
conducted in which the stiffness coefficients of the lines are varied to constrain the displacements
to not exceed a certain maximum value.

35
5.4.1 Constant Stiffness Coefficients

As the lattice size is increased, the number of degrees of freedom of the system is increased. Since
the number of eigenvalues given by the eigenvalue analysis method is equal to the number of
degrees of freedom, the total number of natural periods increases with N. From Figure 26, which
presents the eigenvalues for each lattice size, it can be seen that new values appear as N increases.
The lowest natural periods remain around the same values, however larger natural periods are
observed.

Figure 26: Natural periods for different lattice size. Fext = 1500kN is applied on each FWT with
an angle of 15 degrees. Mooring line parameters are the same for all N.

All natural periods are above 40s which is larger than typical wave periods. Resonance due to wave
frequency loads is thus not of concern, however nonlinear effects lead to higher order loads that
can trigger low frequency resonance. Even though these loads are generally of smaller amplitude
than first order wave frequency loads, the higher order responses can be large due to resonance and
can be problematic for the power cables and the anchoring system, and can also lead two FWTs
to become too close to each other as they oscillate in surge and sway.
Nevertheless, results for the larger lattice size with the line stiffness coefficients unchanged do
not represent realistic designs. The maximum offset in FWT lattices of different size is shown in
Figure 27. For N=6, a static offset of more than 100m is observed. A realistic FWT design would
limit the maximum offset, in particular due to constraints on the installation of power cables.

36
Figure 27: Maximum offset norm in a (NxN) FWT lattice, where N is increased from 2 to 6.
Fext = 1500kN is applied on each FWT with an angle of 15 degrees. Mooring line parameters are
the same for all N.

5.4.2 Variable Stiffness Coefficients

The maximum offset in the FWT farm is limited to 30m. For each value of N, the stiffness coeffi-
cients of the anchored and shared lines k1 and k2 are adjusted so that the maximum displacement
in the lattice is of 30m. Details on how k1 and k2 are chosen and on the resulting maximum offset
in the lattices are presented in Appendix D.

The natural periods obtained as the farm size is increased are shown in Figure 28.

37
Figure 28: Natural periods for different lattice size. Fext = 1500kN is applied on each FWT with
an angle of 15 degrees. The stiffness coefficients of the lines are increased as N increases to limit
the maximum offset to 30m.

It is still observed that the natural period value range increases as the lattice size becomes larger,
however, the natural periods now tend to decrease. This can be explained by the fact that k1
and k2 are increased with N and thus the system becomes stiffer when FWTs are added to the
lattice. Natural periods become very close to wave periods for larger lattices. For the (6x6) lattice,
the lowest natural period is of 25s. Wave frequency resonance should be avoided as it is at these
frequencies that the excitation loads are the largest. This means a challenge for the design of large
FWT farms with shared lines will be to find a compromise between limiting the static offset of the
FWTs when loaded and designing a system that is soft enough for the resonance periods to be
larger than the wave periods.

5.5 Future Work

The eigenvalue analysis method has been applied to loaded lattices of FWTs and provides insight
on the dynamic behaviour of these lattices. Nevertheless, there remains complications with the
methodology used due to the fact that the eigenvalue analysis method is originally developed
for linear systems and not nonlinear ones. The calculation of the nonlinear stiffness matrix does
not yield a symmetric matrix as opposed to the linear case. This can be explained by looking
at the forces acting on each FWT. A displacement in sway of a FWT will generate a restoring
force in surge on its neighbour, while a displacement in surge of its neighbour will not generate
any restoring force in sway on it. For a non symmetric matrix, the eigenvalue analysis will in
general give complex eigenvalues and eigenvectors. In the analysis of the (2x2) lattice, two of the
mode shapes obtained are complex, and four are complex for the (3x3) lattice. In the previous

38
discussion, the imaginary parts have been discarded and only the real parts have been considered.
The imaginary part of the eigenvalues is small compared to the real part, however, for the mode
shapes, the imaginary part of some components are comparable to their real part and are thus not
negligible. It should therefore be looked into the meaning of these complex mode shapes and it
should be assessed whether they reflect any physical behaviour.

In addition, several improvements can be made to the model to better represent the behaviour of a
real system. A simplification that has been made is to model the mooring lines with linear springs.
This implies that for any elongation ∆l of a line, the tension in the line is expressed as T0 + k · ∆l.
The magnitude of the restoring is thus the same whether the line is compressed or elongated. In
reality, a mooring line pulls more than it pushes. This will affect the offsets of the FWTs when
a static external force is applied and the model could be improved to take this into account. To
perform the nonlinear static analysis, the distance between two FWTs in the lattice has been
computed. Instead of applying a linear spring model, the tension in a line could be obtained using
this distance, by calculating the tension as a cubic function of the length between the two points
of attachment of the line. The restoring force would thus be larger when the distance between the
turbines is increased than when the distance is reduced.

The following step would then be to identify relevant sources of damping to include them in the
analysis. Contributions to damping will come both from the platform and the drag forces on the
mooring lines.
A more complex model could also take into account the yaw motions of the FWTs. In this case the
spatial extension of each platform would have to be considered to account for the position of the
attachment points of the lines. The restoring forces provided by the mooring lines are expected to
be affected by the FWT yaw motions.
Finally, further investigations of the system could include frequency and time domain analyses and
in parallel, the relevant excitation loads acting on the lattice should be assessed.

39
6 Conclusion

A FWT farm lattice model has been developed to assess coupled modes of oscillations in farms with
shared moorings. A nonlinear static analysis was applied on the system to observe the effect of a
mean external loading and was compared to a linear analysis of the unloaded lattice. Results have
been obtained from the eigenvalue analysis of both the unloaded and the loaded system. Although
the imaginary part of certain mode shapes still lacks an interpretation, several conclusions can be
made from these analyses. It was found that when considering the mean displacements of the wind
turbines, the nonlinear analysis does not provide significantly different information than the linear
analysis. Nonetheless, taking into account the geometric nonlinearities of the loaded lattice does
affect the natural frequencies and mode shapes. Eigenvalue analyses of loaded lattices show the
coupling modes are affected by both the external force amplitude and direction, while the natural
periods depend mainly on the loading amplitude. These resonance periods lie in the low frequency
range. Even so, it was seen that when increasing the size of the lattice, the natural periods spread
and tend to decrease. An important task will thus be to assess the relevant excitation loads on
the FWT farm, as well as to take into account damping of the system to evaluate its response
to dynamic loading. Furthermore, an improvement can be made to the model by using a more
realistic expression for the tension in the lines.

40
References

Bachynski, Erin et al. (2019). TMR4182 Marine Dynamics lecture notes. Norwegian University of
Science and Technology.
Connolly, Patrick (2018). Resonance in shared mooring floating offshore wind turbine farms. (Hon-
ours thesis). University of Prince Edward Island.
Connolly, Patrick and Matthew Hall (2019). ‘Comparison of pilot-scale floating offshore wind farms
with shared moorings’. In: Ocean Engineering 171, pp. 172–180.
Goldschmidt, Marek and Michael Muskulus (2015). ‘Coupled Mooring Systems for Floating Wind
Farms’. In: Energy Procedia 80, pp. 255–262.
Hall, Matthew and Patrick Connolly (June 2018). ‘Coupled Dynamics Modelling of a Floating
Wind Farm With Shared Mooring Lines’. In: International Conference on Offshore Mechanics
and Arctic Engineering Volume 10: Ocean Renewable Energy.
Hall, Matthew and Andrew Goupee (2015). ‘Validation of a lumped-mass mooring line model with
DeepCwind semisubmersible model test data’. In: Ocean Engineering 104, pp. 590–603.
Haslum, Herbjorn (2019). TMR4225 Marine Operations guest lecture from Equinor. Norwegian
University of Science and Technology.
Liang, Guodong, Karl Merz and Zhiyu Jiang (Aug. 2020). ‘Modeling of a Shared Mooring System
for a Dual-Spar Configuration’. In: International Conference on Offshore Mechanics and Arctic
Engineering Volume 9: Ocean Renewable Energy.
NTNU (n.d.). TKT4197 Non-Linear Finite Element Analysis lecture notes. Norwegian University
of Science and Technology.
Wilson, Samuel et al. (2021). ‘Linearized modeling and optimization of shared mooring systems’.
In: Ocean Engineering 241, p. 110009.

41
Appendix

A Newton-Raphson Algorithm Validation

A simple fictive two degrees of freedom system is used to test the algorithm developed to implement
the Newton-Raphson method. The system consists of two independent degrees of freedom x1 and
x2 which could represent two point masses connected to two independent springs. Quadratic
restoring forces are assumed and the internal forces are given by:
 2
−x1
Fint =
−x22
The theoretical stiffness matrix of the system is defined by:
h i 2 · x 0

1
K = ∂F int
∂x1
∂Fint
∂x2 =
0 2 · x2
An external force Fext is applied. The analytical solution can be found by solving Fint = −Fext :
  p 
x1 (Fext )1
= p
x2 (Fext )2
The solutions given by the numerical implementation of the Newton-Raphson method is compared
to the analytic solution to validate the numerical calculation.

B Mode Shape Association for Different Loading Amplitudes

To visualize how mode shapes are modified when the external loading is increased, the eigenvectors
at each external force amplitude are associated with the eigenvectors at a lower force amplitude
using the Modal Assurance Criterion (MAC). The MAC characterizes how much two mode shapes
are similar to one another. The use of this criterion is explained using the example of a (2x2) grid
of FWTs.
For a (2x2) FWT lattice, eight mode shapes are calculated for each value of Fext . Let us consider
(2000)
one of the mode shapes calculated for Fext = 2000kN . It will be denoted Φm , with m ∈
(1500)
{1, 2, ..., 8}. For a mode shape Φn amongst the set of mode shapes at 1500kN, with n ∈
{1, 2, ..., 8}, the MAC number for these two vectors is defined as:
(2000) (1500) 2
hΦm |Φn i
M AC(Φ(2000)
m , Φ(1500)
n )= (2000) (1500)
(28)
kΦm k2 · kΦn k2
where k.k2 denotes the 2-norm of the vectors. Since the eigenvectors are normalized, the MAC
number reduces to the square of the scalar product between the two mode shapes.

This number is calculated for all association of mode shapes at 1500kN with mode shapes at
2000kN. A correlation matrix can be defined where each coefficient (n,m) represents the MAC
number between mode shape m from the set at 2000kN and mode shape n at 1500kN:
correlation(n, m)1500−2000 = hΦ(1500)
n , Φ(2000)
m i2 (29)

A MAC number close to 1 means the mode shapes are very similar, while a low MAC number
means the mode shapes have little correlation. For each mode shape at 2000kN we will thus select
the mode shape at 1500kN that has the highest MAC with it. This means that for each column m
of the correlation matrix, the row with the maximum MAC number is found to determine which
mode shapes are paired.
By defining this correlation matrix again for the sets of mode shapes at 1000kN and 1500kN, and
then at 500kN and 1000kN, and finally at 1kN and 500kN, each mode shape at 2000kN can be

42
assigned a corresponding vector at each of the previous force amplitudes by starting from the set
of mode shapes at 2000kN and iterating through the force amplitudes.
The correlation matrices for a (2x2) lattice between the set of mode shapes at 2000kN and the
set at 1500kN, and between the set at 500kN and the set at 1kN are presented in Figure 29a and
Figure 29b.

(a) Correlation matrix for the set of mode shapes (b) Correlation matrix for the set of mode shapes
at 2000kN and the set of mode shapes at 1500kN. at 500kN and the set of mode shapes at 1kN.

Figure 29: Correlation matrices for a (2x2) lattice. MAC numbers are rounded to two decimal
places.

Between a loading of 1500kN and a loading of 2000kN, identification of the mode shape to associate
with each vector from the set at 2000kN is straightforward as for each column and row there is one
and only one MAC number that is larger than 0.9. All mode shapes from the 2000kN set present
a strong correlation with a different mode shape from the 1500kN set. However, the association
of mode shapes between a loading of 1kN and a loading of 500kN is more complex. It can be seen
that mode shapes 2 and 3 at 500kN both are closest to mode shape 4 at 1kN (with numbering
of eigenvectors from 0 to 7 as presented on Figure 29b). Nonetheless, the goal is to show how
all mode shapes change as Fext is increased, thus all mode shapes should be represented and one
mode shape at a given amplitude should appear only once in the final mode shape association. The
criterion chosen to decide which mode shape should be associated with which predecessor in the
situation described above is that the mode shape with the largest MAC number will be assigned
first. In the matrix presented in Figure 29b mode shape number 3 at 500kN will thus be associated
to mode shape 4 at 1kN, and mode shape number 2 at 500kN will be associated with mode shape
3 at 1kN. This criterion is chosen so that the mode shapes that are the most likely to be related
are assigned first. For example, if two coefficients on the same row n of the correlation matrix
correspond to the maximum MAC number for two different columns m1 and m2 , but one MAC
number is 0.9 and one is 0.5, it can be deduced that the vector that has a max correlation of 0.5
with any of the previous mode shapes has little correlation with any of them. With the method
explained above, it is ensured that the mode shape with the MAC number of 0.9 is associated with
the mode shape n rather than the mode shape that has a poor correlation.

The algorithm performing the mode shape association is thus such that all mode shapes appear
once and only once in the mode shape association, and in case of conflict, the pair that gives the
largest MAC number is always selected. As a test case, if at a given force amplitude, the correlation
matrix is given as:  
0.6 0.1 0.7
0.7 0.95 0.9
0.4 0.9 0.2

43
then the algorithm will assign row 3 to column 1, row 2 to column 2, and row 1 to column 3.

C Effect of Force Amplitude On Natural Periods for Differ-

ent Force Directions

Figure 30 shows the way eigenvalues are modified when Fext is increased in amplitude for directions
of 0, 15, 30 and 45 degrees. The maximum and minimum values of natural periods for each
eigenvalue group are not significantly changed. However, the degeneracy of each eigenvalue is
influenced by the loading direction. Changes in the way mode shapes evolve when the loading
amplitude is increased can thus be expected. As an example, the mode shape association for a
force applied with a 45 degrees angle is presented in figures 31, 32 and 33. The displacement
patterns appear different to the ones observed at 15 degrees.

(a) 0 degrees (b) 15 degrees

(c) 30 degrees (d) 45 degrees

Figure 30: Natural periods for a (2x2) lattice for different loading amplitudes. Curves are plotted
for four different force directions: 0, 15, 30 and 45 degrees. The blue dashed lines represent the
periods that are found when performing an eigenvalue analysis with the stiffness matrix from the
linear analysis.

44
Figure 31: Mode shapes association for a (3x3) lattice for a varying load magnitude. Highest
frequency modes are grouped here. Fext is applied with a 45 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher

45
Figure 32: Mode shapes association for a (3x3) lattice for a varying load magnitude. Second
highest frequency modes are grouped here: frequency is slightly lower than for the group of highest
frequency modes but is still closer to the highest frequencies than to the lowest frequencies. Fext
is applied with a 45 degrees angle. The MAC is given for the association with the mode shape at
the force amplitude that is one step higher

46
Figure 33: Mode shapes association for a (3x3) lattice for a varying load magnitude. Lower
frequency modes are grouped here. Fext is applied with a 45 degrees angle. The MAC is given for
the association with the mode shape at the force amplitude that is one step higher

47
D Line Stiffness Coefficients for (NxN) Lattices

To study the effect of the size of the lattice on the natural periods, the stiffness coefficients of the
anchoring lines k1 and of the shared lines k2 are adjusted to limit the maximum offset in the lattice
to 30m. The values of k1 and k2 used are presented in Table 4.

N k1 [kN/m] k2 [kN/m]
2 48 25
3 120 65
4 200 100
5 320 160
6 450 225

Table 4: Anchoring line and shared line stiffness coefficients used for the natural periods analysis
of (NxN) lattices.

The resulting maximum offset in each lattice is shown in Figure 34 and is between 30m and 30.5m
for all lattices.

Figure 34: Maximum offset in lattices of different size, where stiffness coefficients have been ad-
justed to limit the maximum displacement.

For each lattice (NxN), other pairs (k1 , k2 ) are possible to obtain a maximum displacement of
30m. Other sets of stiffness coefficients, for example with k1 = k2 for each N, have been tested.
The natural periods found with k1 = k2 adjusted to have a maximum offset of 30m are plotted in
Figure 35. The values of the natural periods are modified by the choice of the stiffness coefficients
however, the overall trend and the shape of the curves presenting the natural periods against N
remain the same.

48
Figure 35: Natural periods for different lattice size. Fext = 1500kN is applied on each FWT with
an angle of 15 degrees. The stiffness coefficients of the lines are increased as N increases to limit
the maximum offset to 30m keeping k1 = k2

49