Applied Ocean Research 22 (2000) 169–182

www.elsevier.com/locate/apor

Development and validation of a lumped-mass dynamics model of a

deep-sea ROV system
F.R. Driscoll a,*, R.G. Lueck b, M. Nahon c
a
Department of Mechanical Engineering and School of Earth and Ocean Sciences, University of Victoria, P.O. Box 3055, Victoria, British Columbia,
Canada, V8W 3P6
b
Centre for Earth and Ocean Research, University of Victoria, P.O. Box 3055 Victoria, British Columbia, Canada, V8W 3P6
c
Department of Mechanical Engineering, University of Victoria,, P.O. Box 3055 Victoria, British Columbia, Canada, V8W 3P6
Received 23 March 1999; received in revised form 6 December 1999; accepted 13 December 1999

Abstract
A one-dimensional finite-element lumped-mass model of a vertically tethered caged ROV system subject to surface excitation is
presented. Data acquired during normal operation at sea are used with a least-squares technique to estimate the coefficients required by
the model. The model correctly predicts: (a) the motion of the cage and the tension in the tether at the ship; (b) the spectrum of cage
acceleration in the wave-band; (c) the transfer function between the vertical motion of the ship and cage; and (d) the natural frequency of the
system and its harmonics. A simple model of the wake of the cage was added to the model simulation and this reduced the error in the
calculated motion and tension by almost a factor of two and brought the calculated transfer function within the 95% confidence interval of the
measurements. By increasing ship motion slightly, the model accurately reproduced eight snap loads and their non-linear characteristics—a
regularly spaced series of rapid increases in the records of the acceleration of the cage and tension at the ship—that occurred during the
measurements used for validation. 䉷 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Tethered marine systems; Finite element analysis; Offshore engineering; Remotely operated vehicle; Snap loading; Cable modelling

1. Introduction water sampling rosettes, tethered equipment being lowered

Caged remotely operated vehicles (ROVs) are robots that
effectively operate in the dangerous high-pressure environ-
ment of deep water while their human controllers remain
safely aboard the support ship (Fig. 1). The tether connect-
ing the ship and cage provides the ROV with the ability to
stay at the underwater work site indefinitely and allows real-
time communication with the ROV. However, the tether
also couples the ship motion to the cage, and, as a result,
these systems cannot operate in rough seas. Rough seas
cause slack tether, large snap loads and erratic cage motion
[1], which damage the tether and its internal electrical and
optical conductors. The resulting degradation reduces the
life of the tether and endangers the recovery of the ROV.
Many other vertically tethered systems are dynamically
similar to caged ROVs, including CTDs (conductivity
temperature and depth measuring platforms), drill strings,
* Corresponding author. Present address: Department of Ocean Engineer-
ing, Florida Atlantic University, 101 North Beach Road, Dania Beach,
Florida 33004, USA. Tel.: ⫹ 954-924-7221; fax: ⫹ 954-924-7233.
E-mail address: rdriscol@oe.fau.edu (F.R. Driscoll).
0141-1187/00/$ - see front matter 䉷 2000 Elsevier Science Ltd. All rights reserved.
PII: S0141-118 7(00)00002-X
to the ocean floor, and piston coring tools. Reliable models
are needed to predict the behaviour of vertically tethered
systems so that designers can minimise motions and
tensions and users can know which sea states will produce
snap loads.
Two tools are available to predict the response of verti-
cally tethered systems—continuous (closed-form analytic)
and discrete models. Continuous models are powerful tools
and they have been used to accurately calculate many of the
dynamic characteristics of tethered systems, including
motion and tension spectra, transfer functions, natural
frequency and its harmonics, and the onset of slack in the
tether [2–5]. However, continuous models have limitations.
Equivalent linearisation techniques are used to approximate
the quadratic drag for steady amplitudes, which enable
closed form solutions and frequency-domain analysis [3–
6]. Thus, these models are not appropriate during unsteady
wave conditions. Continuous models are also invalid for a
slack tether, which is the precursor to snap-loads, and it is
difficult (if not impossible) to solve the differential equa-
tions governing a tether when hydrodynamic forces and its
properties (stiffness, cross-section, material) change along
170 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
Fig. 1. Diagrammatic representation of the ROPOS ROV system consisting of a support ship (C.S.S. John P. Tully), winch, umbilical tether, cage and vehicle.
ROPOS is investigating the actively venting sulfide structure Godzilla [20].
its length. Discrete models are not constrained by the limita-
tions of continuous models and they are valid for a wider
range of operating conditions and system configurations.
Niedzwecki and Thampi [7] developed a discrete multi-
segment lumped-parameter representation of a drill string
connecting a ship to a subsea package and used it to inves-
tigate snap loading. They reproduced many of the snap load
and motion characteristics that were observed by Driscoll et
al. [1] for a real system, namely, a repeating pattern of
rapidly increasing tension during snap loads and vertical
motion at the platform that was greater than at the ship.
The predicted behaviour can be sensitive to the model
coefficients, such as added mass [7], and choosing represen-
tative values is difficult because many components are
complex and asymmetric (armoured/faired tether, ROV
cages and Conductivity Temperature and Depth measuring
platforms). For example, Yoerger et al. [8] found that the
normal drag coefficient of a long vertical tow cable is ampli-
fied well above the value for a rigid cylinder and that it
depends on operating conditions. Therefore, coefficients
must be calibrated using measured data to obtain confidence
in a model. Additionally, discrete models are approxima-
tions to real systems, and therefore, validation against real
data is essential in proving their accuracy and reliability. In
the literature search, we found one work that calibrates
coefficients of a deep towed vehicle using data [8] and
one that compares results of a discrete model of a towed
system against its actual motion [9]. However, no work was
found that uses data to calibrate and validate discrete models
of vertically tethered systems.
In this paper, a finite-element lumped-mass model of a
vertically tethered ROV system is developed by applying
Galerkin’s method to a continuous representation of the
tether. Motion and tension measurements of the ROPOS
ROV system [1] are used to accurately determine values
for the model coefficients. In particular, a least-squares tech-
nique is used to estimate the drag coefficient and the virtual
mass (including the system mass, the hydrodynamic added
and the entrained mass) of the ROV cage. The vertical
motion of the top element is prescribed here along with
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
Fig. 2. Diagrammatic representation of the finite-element, including the
reference and elastic vectors for node i.
the actual ship motion and comparing the calculated and
measured motion of the cage and tension in the tether vali-
dates the model. A simple representation of the wake of the
cage is also developed and its addition significantly
improves the model accuracy. Finally, we compare the
predicted and actual response during snap loads.
2. Discrete model
Experimental measurements have shown that when a
171
roll, pitch and yaw motions is small compared to the vertical
scale O(1:1000) of the tether and such motions are not
effectively transmitted by the tether. Therefore, a one-
dimensional model is sufficient to represent a caged ROV
system. For modelling purposes, the system is broken down
into three individual sub-systems: the support ship with its
A-frame, the tether, and the cage and ROV. A finite-element
lumped-mass model is used to represent the tether, a displa-
cement function to prescribe the ship motion and a second-
order differential equation to represent the cage and ROV
unit. The model is developed using a single inertial refer-
ence frame located at the mean sea surface, with downward
displacements defined as positive (Fig. 2).
2.1. Tether model
The tether is a long elastic structural member that can
only sustain tensile loads and may have length-varying
properties. We represent the tether as a long elastic cylinder
that is suspended directly below the ship’s A-frame. Defin-
ing s as the vertical unstretched co-ordinate along the tether,
the balance of forces (per unit length) at any point in the
tether is [10]:
2T
⫹ W ⫹ F ˆ ma …1†
2s
where
W ˆ …rc ⫺ rw †Ag …2†
is the in situ weight per unit length, T(s) is the tension in the
tether, F(s) is the hydrodynamic force per unit length, m is
the effective mass per unit length and includes the entrained
water, a is the inertial acceleration, g is the gravitational
acceleration, A(s) is the cross-sectional area of the tether,
rc …s† is the density of the tether and rw is the density of sea
water. The added mass forces are zero for flow tangent to a
cylinder [11], and therefore, F(s) is only composed of drag
forces. The hydrostatic pressure induces circumferential and
radial stress and a depth dependent axial strain in the amour
strands of the tether [12]. The ratio of the standard deviations
of the strain induced by the tension to the strain induced by the
pressure at any point in the tether has a numeric value of
approximately 1000 because the tether only oscillates short
vertical distances (⬍3 m). Thus, hydrostatically induced axial
strain is considered constant in time and it is neglected in the
analysis. The strain in the tether is less than 0.004 of the
manufacturer’s maximum recommended working load and
the armour is wrapped around an incompressible core; there-
fore, the dilation of the tether will be very small and its cross-
section is assumed to be constant. Further, assuming the water
is stationary in the far field, the hydrodynamic force per unit
length, F(s), can be expressed as [13]:
 
typical deep-sea ROV system operates in the absence of F…s† ˆ ⫺ 12 …e ⫹ 1†rw DCDT U_ U_  …3†
time varying currents or large horizontal excursions of the
support vessels, the ship and cage are only coupled verti- where e is the local strain in the tether, D(s) is the diameter of
cally [1]. The displacement of the cage due to surge, sway, the tether, CDT is the tangential drag coefficient, ‘·’ overhead
172 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182

represents its time derivative and U_ is the velocity of the tether. methods, such as collocation, least squares and least squares
The relationship between tension and axial strain is: collocation, because it is more prevalent in literature [16].
Knowing that rc and A are uniform or change very slowly
T…s† ˆ EAe …4†
[…li =rc †…2rc =2s† Ⰶ 1 or … li =A†…2A=2s† Ⰶ 1] along an element
where E is the effective Young’s modulus of the tether. and that the strain in the tether is small …e Ⰶ 1†; the elemen-
A finite-element implementation was chosen because it is tal equations of element i are:
versatile and can model the complicated non-linear charac- " # " #2 …1† 3
teristics of vertically tethered systems. For example, one can mi li Z 1=3 1=6 u
4 i 5
⫹ mi li
quickly assemble complex models of multi-component 2 Z 1=6 1=3 u …2†
i
systems connected by tethers with different and length depen-
" #2 …1† 3 2 …1† 3 2 …1† 3
dent properties because the elemental tether equations are EA 1 ⫺1 u f b
developed from a generalised linear-elastic tether subject to ⫹ i i 4 i 5 ˆ 4 i 5 ⫹ 4 i 5 …8†
unspecified boundary conditions. Components, such as ROV
li ⫺1 1 u…2†i f …2† i b…2† i
cages, weights/floats, and intermediate platforms can be easily where Ei is the effective Young’s modulus of the tether and
added because they are each represented as a point mass and s_i ˆ si ˆ 0: The consistent hydrodynamic, gravitational and
the hydrodynamic, gravitational and buoyancy forces acting buoyancy forces at upper and lower nodes are:
on them are described by one differential equation.
The tether is discretised by first dividing it into N fi…1† ˆ 1
2 …mi ⫺ rw Ai †gli ⫺ 1
2 rw Di CDT
i
li
segments (elements) which have the same properties as
the continuous tether. The end points of each element are  ‰ 13 …Z_ ⫹ u…1† _ _ …1†
i †jZ ⫹ u i j⫹
1
6 …Z_ ⫹ u_ …2† _ _…2†
i †jZ ⫹ u i j
the nodes of the system. Each element i is of length li and
has an associated body-fixed reference frame attached to the ⫺ 1
12 …u_…1†
i ⫺u_ …2† _…1†
i †ju _ …2†
i ⫺u i jŠ …9a†
upper node with the z-axis pointing downward along the
element. The top of the tether is attached to the ship’s A- fi…2† ˆ 1
2 …mi ⫺ rw Ai †gli ⫺ 1
2 rw Di CDT
i
li
frame. To facilitate the finite-element derivation, the loca-
tion of any point p in element i is defined using a coupled set  ‰ 16 …Z_ ⫹ u_ …1† _ _ …1†
i †jZ ⫹ u i j⫹
1
3 …Z_ ⫹ u_ …2† _ _…2†
i †jZ ⫹ u i j
of reference and elastic co-ordinates [14] (Fig. 2):
⫺ 1
12 …u_…1†
i ⫺u_ …2† _…1†
i †ju _ …2†
i ⫺u i j …9b†
zip ˆ Z ⫹ sip ⫹ uip …5†
where Di is the diameter of the tether and the boundary
where Z is the distance from the mean sea surface to the top conditions for the continuity of strain between nodes of
of the tether at node 1, in the inertial frame, sip is the connecting elements at the upper and lower nodes, respec-
constant length position vector from node 1 to a point p tively, are:
along the undeformed tether and uip is the cumulative elastic
2u~
displacement of the tether up to point p. An elemental trial b…1†
i ˆ Ei A i f…s…1†
i †i;1 …10a†
solution, u~ ip [15], is used to approximate the elastic displa- 2s
cement, uip : 2u~  …2† 
b…2†
i ˆ Ei A i f si …10b†
X
n
2s i;2
uip ⬇ u~ ip …s; t; d† ˆ d…i j† …t†fi; j …s† …6†
jˆ1
2.2. Ship model
where fi; j …s† is a shape function, n ⫺ 1 is the order of the
shape function and di… j† …t† is a time-dependent coefficient The vertical motion of the ship is assumed to be known
(yet to be defined). Each tether element has only one degree and is treated as a prescribed displacement function applied
of freedom and a linear shape function is sufficient to to the upper end (first node) of the tether, Z ˆ ZS …t†:
approximate the elastic displacement of any point p in Because the motion of the first node is prescribed, measured
element i, that is n ˆ 2 and (actual) ship motion data can be used as input. Thus, the
model can be validated by directly comparing the calculated
s…2†
i ⫺s s ⫺ s…1† cage motion and tension in the tether against the actual
fi;1 …s† ˆ fi;2 …s† ˆ i
si ⫺ s…1†
…2†
i s…2†
i ⫺ s…1†
i …7† measured values for the same ship motion.

d…1†
i ˆ u…1†
1 di
…2†
ˆ u…2†
i
2.3. Cage and ROV model

where … †…1† …2†

i and … †i represent the upper and lower nodes of
element i, respectively. Galerkin’s method of weighted resi-
duals [15] is applied directly to the elemental force balance
Eq. (1) to yield the discrete, finite-element, equations.
Galerkin’s method was chosen over other weighted residual
The ratio of tether length to cage height is large, O(1000),
and thus, the cage can be represented as a single point mass
that is attached to the lower end of the tether and subject to
the forces of gravity, buoyancy, drag and tension. When the
ROV is in the cage, its mass, weight and buoyancy are
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182 173

added to those of the cage. When the ROV is out of the cage,
the tether connecting the ROV and cage is slack and MROV
and WROV are set to zero. Thus, the force balance for the
heave motion at the cage is:
…MCG ⫹ MROV †…Z ⫹ u …2†
N †⫹
1
2 rw AC CD …Z_ ⫹ u_…2† _
N †jZ ⫹ u N j
ˆ WCG ⫹ WROV ⫺ T…L†
where MCG and MROV are the virtual mass (including
entrained and added mass) of the cage and ROV,
respectively, AC is the effective horizontal cross-
sectional area of the cage, CD is the drag coefficient
of the cage, WCG and WROV are the weight less buoyancy
end points; therefore, the boundary forces, b…2† i and b…1†
i⫹1 ;
cancel when the elemental equations are assembled. The
cage and ROV are included by the addition of their mass
to that of node N ⫹ 1 and applying the forces acting on them
to that node. The boundary condition at the bottom end of
_ …2†
the tether is given by Eq. (11). The consistent-mass matrix is
replaced with a lumped-mass matrix to uncouple the accel-
…11†
eration terms [17]. The set of N second-order differential
equations that govern the motion of the vertically tethered
system are:
Mz ⫹ Ku ˆ f …12†
ÿ

where the lumped-mass matrix, M 僆 RN× N⫹1
; is:

2 3
0 m1 l1 ⫹ m2 l2 0 … 0 0
6 7
60 m2 l2 ⫹ m3 l3 … 7
6 0 0 0 7
6 7
16 .. .. .. .. .. 7
Mˆ 6 7 …13†
26
6
. . . ] . . 7
7
6 7
60 0 0 … mN⫺1 lN⫺1 ⫹ mN lN 0 7
4 5
0 0 0 … 0 mN ⫹ 2MCG ⫹ 2MROV
ÿ

the stiffness matrix, K 僆 R N× N⫹1
; is:
2 3
E 1 A1 E 1 A1 E A E2 A2
6⫺ l ⫹ 2 2 ⫺ … 0 0 7
6 l l2 l2 7
6 1 1
7
6 7
6 E A E2 A 2 E A 7
6 0 ⫺ 2 2 ⫹ 3 3 … 0 0 7
6 l2 l2 l3 7
6 7
6 7
6 .. .. .. .. .. 7
Kˆ6 . . . . . 7 …14†
6 ] 7
6 7
6 EN⫺1 AN⫺1 E A EN AN 7
6 … ⫹ N N ⫺ 7
6 0 0 0 7
6 lN⫺1 lN lN 7
6 7
6 7
4 EN AN EN AN 5
0 0 0 … ⫺
lN lN

of the cage and ROV, respectively, T…L† is the tension in the the external force matrix, f 僆 RN×1 ; is:
tether at the cage termination, u_ …2†
N is the elastic displacement
of the tether at the cage termination, and here ‘·’ and ‘··’ 2 3
overhead represent first and second time derivatives, f1…2† ⫹ f2…1†
6 7
respectively. 6 …2† 7
6 f2 ⫹ f3…1† 7
6 7
6 7
6 . 7
fˆ6 .. 7
2.4. Model assembly 6 7
6 7 …15†
6 …2† …1† 7
6 fN⫺1 ⫹ fN 7
The N tether elements are assembled end-to-end and 4 5
joined at their node points as shown in Fig. 2. Nodes are fN…2† ⫹ fCG
numbered consecutively from the top node, 1, to the bottom  
node, N ⫹ 1: The heave acceleration at the ship is _ …2† 
fCG ˆ WCG ⫹ WROV ⫺ 1
2 rw AC CD …Z_ ⫹ u…2†
N †  Z ⫹ u N 
prescribed as the acceleration of node 1. The tether is
contiguous at each node point ranging between 2 and N;
therefore, the trial functions of connected elements must the acceleration of the top node, 1, is:
be identically equal at their connecting nodes. Additionally,
continuity requires the strain boundary conditions of
connected elements to be equal and opposite at coincident Z ˆ z1 ˆ Z S …t† …16†
174 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182

Fig. 3. ROPOS in the cage during a launch from the R/V Sonne, 1997.

and the position, z 僆 RN×1 ; and elastic displacement, u 僆 method takes small time steps during periods of rapid
RN×1 ; vectors are, respectively: change and large time steps when changes are slow [18].
2 3 2 3 Thus, good accuracy is achieved with only moderate
z1 Z
6 7 6 7 computational effort and errors and instabilities introduced
6 z 7 6 Z ⫹ s…1† ⫹ u…1† 7
6 2 7 6 2 2 7 by the previous time step information are significantly
6 7 6 7
6 7 6 …1† 7 reduced.
6 z3 7 6 Z ⫹ s…1† ⫹ u 3 7
6 7 6 3 7
zˆ6 6 . 7 6
7ˆ6
.
7
7 …17†
6 .. 7 6 .. 7
6 7 6 7 3. Data
6 7 6 7
6 7 6 …1† …1† 7
6 zN 7 6 Z ⫹ sN ⫹ uN 7
4 5 4 5 Motion data were collected for the ROPOS ROV system
zN⫹1 Z ⫹ s…2†
N ⫹ u …2†
N
(Fig. 3) to accurately determine the unknown model coeffi-
cients, to verify manufacturer’s specifications and to vali-
2 3 date the numerical model. ROPOS is a large ROV that dives
0
6 7 to 5000 m, and is attached to the cage by an umbilical tether.
6 z2 ⫺ s…1† ⫺ Z 7
6 2 7 The cage itself weighs 49 000 N in air and has dimensions
6 7
6 …1† 7 of 2.1 × 3.4 × 4.2 m (WLH) and is connected to the ship by
6 z3 ⫺ s3 ⫺ Z 7
6 7
uˆ6
6 .
7
7 …18† a triple armoured tether that is 0.03 m in diameter, 4200 m
6 .. 7 long and weighs 30.5 N m ⫺1 in air (Fig. 4).
6 7
6 7 Data were collected while ROPOS operated at depths of
6 …1† 7
6 zN ⫺ sN ⫺ Z 7 up to 1765 m. Sea conditions ranged from relatively calm to
4 5
…2† very rough with sheave displacements less than 1 m and
zN⫹1 ⫺ sN ⫺ Z
where s…1† …1†
1 and u1 are zero and R
m×n
denotes an m × n
matrix.

2.5. Numerical implementation

The N second-order differential equations given by Eqs.

(12) and (16) are rewritten as 2N first-order differential
equations, thus allowing the use of a conventional ODE
integration routine. This approach adds very little computa-
tional overhead because the additional N terms are simple
equality statements; however, previous time step informa-
tion is needed. A Runge–Kutta fourth/fifth-order numerical Fig. 4. A cross-section of the main umbilical tether used in the ROPOS
integration routine with adaptive step size is used. This ROV system.
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
Table 1
Values of the model coefficients estimated using motion and tension
measurements of the ROPOS ROV system
Coefficient Calculated value Manufacturer’s or
empirical estimates
⫺1 ⫺1
m 3.01 ^ 0.06 kg m 3.11 kg m
W 25.4 ^ 0.5 N m ⫺1 25.5 N m ⫺1
WCG 43 200 ^ 1000 N 42 000 N
EA 45.5 ^ 1.4 × 10 6 N 33.4 × 10 6 N
CDT 0.02 0.02
CVCG 8150 ^ 170 kg m ⫺1 7300 kg m ⫺1
MCG 7750 ^ 160 kg 9200 kg
greater the 4 m peak-to-peak, respectively. Two sites were
instrumented: the top of the ship’s A-frame (directly above
the sheave that supports the tether and thus, at node 1), and
the top of the ROV cage (node N ⫹ 1). Tri-axial, orthogonal
accelerometers and rate gyros were used to measure all six
degrees of motion of the ship and cage. A pressure sensor
was used to measure the depth of the cage and a load cell
was used to measure the tension in the tether at the A-frame.
Data were sampled at 256 Hz to a resolution of 16 bits. A
detailed description of the instrumentation, data reduction
and data analysis is given in Ref. [1]. For all measurements
used in the calibration and validation, the ROV was out of
the cage, the tether connecting the ROV and cage was slack
and the ROV and cage were horizontally separated by at
least 5 m. Thus, MROV and WROV are set to zero.
4. Calibration
The ability of any model to accurately predict the
response to specific input requires the numeric values of
the coefficients to be well known. However, values for
most of the model’s coefficients are difficult to determine.
The cage is a three-dimensional bluff body with two planes
of symmetry and complex geometry. Its drag coefficient, CD
and added mass (included in MCG † cannot be accurately
determined using empirical values of coefficients for bodies
of simple geometry. The linear stiffness, EA; of the tether
can be difficult to accurately determine using analytical
equations and empirical data because the tether consists of
an inner core of several different materials surrounded by
layers of armour strands that are counter-helically wound.
Thus, experimental data was used to determine accurate
values of the coefficients.
For several different deployments, the weight less buoy-
ancy of the tether, W (per unit length), and cage, WCG ; were
estimated from a least-squares linear fit of tension to depth
for the range of 218 to 1730 m. Tether deployment was
halted to allow a steady length for each measurement. The
slope W and offset WCG (Table 1) agreed well with the
manufacturer’s specifications. The mass of the tether is
approximated by its dry mass plus the mass of the water
entrained into the void space between the strands of armour.
175
The wet mass, m; is 2% larger than the dry mass and its
numeric value is given in Table 1.
The mass and drag coefficients of the cage were deter-
mined by expressing Eq. (11) as:
WCG ⫺ T…L† ˆ MCG …Z ⫹ u…2†
N †⫹
1
2 rw AC CD …Z_ ⫹ u_…2† _ _ …2†
N †jZ ⫹ u N j
…19†
Since Eq. (19) is linear in MCG and CD ; a least-squares fit of
tension in the tether to the velocity times its modulus and the
acceleration of the cage yields values for the unknown coef-
ficients. However,
ÿ
tension was not measured at the cage
termination,ÿ T
L ; but it was assumed that it is well approxi-
mated by T 0 for ÿ
L ⱕ 50ÿ m.
For L ⱖ 50 m, the phase and
magnitude of T L and T 0 are different [1] and therefore
records for these greater depths cannot be used.
Records from 28 and 32 m depth, measured on separate
days with different sea states, provide values and 95% con-
fidence intervals of MCG ˆ 7750 ^ 160 kg and CVCG ˆ
1=2rw AC CD ˆ 8150 ^ 170 kg m⫺1 : The statistical error is
less than 1% but the different values for the two depths
imply a bias and systematic error of 2% and this larger
value is used as the confidence interval. Agreement is within
10% for values of MCG and CVCG calculated from the data
and values estimated using empirical tables (Table 1).
The axial stiffness, EA; is given by [4]:
EA ˆ mc2 …20†
where c is the speed of strain waves along the tether. As will
be shown in Section 7 (see Fig. 9), echoes of a snap load are
spaced at constant intervals equal to the travel time of a
strain wave through the length of tether, and thus they
provide a very precise measure c for large L. A second
measure of c is provided by the higher order natural frequen-
cies, which are spaced at intervals of 0:5c=L Hz: Both meth-
ods yield identical values of c ˆ 3870 m s⫺1 : Using this
value in Eq. (20), the value of the axial stiffness was
found to be EA ˆ 45:5 ^ 1:4 × 106 N: The manufacturer’s
estimated value of EA is approximately 25% less (Table 1),
indicating a large error in their calculations.
5. Validation
The model was validated by directly comparing the real
and simulated motion of the cage and tension in the tether
for the same forcing at the top node. This was done for
several depths, but for brevity this paper will focus on
operation at 1730 m. Numerical results were calculated
using 50 elements and the estimated coefficients (Table 1).
The heave acceleration of node 1 was prescribed using a
5000 s time series of measured acceleration of the A-frame
(ship). Since data were collected during a one-month period,
which included a large range of sea states and operating
depths, different data were used for the coefficient estima-
tion and model validation. For example, MCG and CD were
estimated from two data sets collected two weeks before,
176 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
Fig. 5. Short time series of calculated (thick line) and measured (thin line) cage motion and tension in the tether for the system at 1730 m. The ship motion
(dotted line) is included for comparison.
and at an operating depth 1700 m shallower than the data
used for validation in this report.
The model provides a very good approximation of the
actual displacement of the cage as suggested by Fig. 5.
The ship displacement was included in Fig. 5a to highlight
the difference between input and response. Indeed, the
model correctly predicts that the displacement of the cage
lags behind and is greater than that of the ship. This parti-
cular section of data has a large range of amplitudes and
includes a snap load which starts at t ˆ 627 s: The larger
amplitudes are underestimated by the model and the snap
behaviour of the system, that is the sudden change of accel-
eration at the cage, was not reproduced. Quantitatively, the
standard deviation of the difference between the simulated
andC measured position of the cage (tension at the Cship) s is
Ts
s diff
Z
ˆ 0:11 m …s diff ˆ 4000 N† and the ratio of s diff
Z
…s diff
T
†
to the rms of the measured values is 0.12 (0.28).
The spectra of the real and simulated vertical cage accel-
eration are compared for the same 5000 s record of ship
motion in Fig. 6. The model predictions that include the
wake effects are addressed later in the Section 7. The
frequency domain analyses focuses upon the waveband
that extends from 0.1 to 0.25 Hz, because this band contains
90% of the variance of ship motion. The measured and
simulated acceleration spectra agree well within the wave-
band at 1730 m. The largest difference is 13% at 0.14 Hz
and the spectra are equal at 0.175 Hz. For L ˆ 975 m (not
shown) the real and simulated spectra are almost identical
and the largest difference is only 7% at 0.14 Hz. Above
1 Hz, significant excitation is only produced by the impul-
sive snap loads and no wave excitation is transferred to the
tether because the large inertia and length of the ship act as a
low-pass filter for short (high-frequency) waves.
The cage is a bluff body and is subject to quadratic drag
which is well approximated by an odd Talyor series. Thus,
variance is produced at odd harmonics of the fundamental
frequency of excitation, 0.15 Hz (the peak of the spectrum
in Fig. 6). The third and fifth harmonics are evident at 0.44
and 0.73 Hz (Fig. 6). The model over predicts the variance
at those frequencies; however, the system remains weakly
non-linear [1] because the harmonics only contributed a
small fraction to the total variance. For example, at
1730 m, the variance at the harmonics provided only 4.6
and 0.3% of the total variance of the cage acceleration.
The transfer function estimate (TFE) provides a measure
of the motion amplification and phase difference between
the ship and cage and it is estimated using:
CZ S Z C
HZ S Z C ˆ …21†
CZ S Z S
where CZ S Z C is the cross-spectrum of ship-displacement, Z S ;
and cage-displacement, Z C ; and CZ S Z S is the auto-spectrum
of Z S : The analysis of the TFE is restricted to the waveband
(0.1–0.25 Hz) because the error in the real TFE is large and
the modelled cage response is not significant at higher
frequencies. At 1730 m, agreement between the real and
simulated TFE is good up to 0.2 Hz and the largest
difference in magnitude is only 8% at 0.13 Hz (Fig.
7). The model over predicts the cage response at the
higher frequencies within the wavebound. The simulated
phase is always slightly larger than the real phase and
the maximum difference is 15⬚ at 0.2 Hz (near the
middle of the waveband). These discrepancies are attrib-
uted, in part, to the assumption of a constant drag coef-
ficient and a constant added mass of the cage which is not
completely accurate for oscillatory flow around a bluff
body. This issue is addressed in Section 7. At 975 m, the
magnitude and phase of the TFE (not shown here) agree
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182 177

Fig. 6. Spectra of actual (thin solid line) and model calculated (dotted line—without wake, thick solid line—with wake) vertical acceleration records of the
cage at a depth of 1730 m. The arrows are the 3rd and 5th harmonic of the peak cage motion and the natural frequency, f1. Both axes have logarithmic scales.

exceptionally well. The largest TFE differences are 0.05
(4%) at 0.19 Hz and 10⬚ at 0.25 Hz.
The depth dependent natural frequencies, fn ; provide
another characteristic to validate the model. The predicted
and observed first natural frequency, indicated by the first
90⬚ phase shift of HZ S Z C shown in Fig. 7, agree within 7%.
Although very little excitation occurs above 1 Hz, the model
exactly predicts the second, third, and fourth natural frequen-
cies at 1.17, 2.24 and 3.36 Hz. This is shown in Fig. 6.
6. Sensitivity
During the design process, the model coefficients are not
accurately known and thus, differences will exist between
estimated values and actual values as shown in Table 1.
Therefore, the model output must be reasonably insensitive
to variations of its coefficients to be a useful design tool. If
not, slight errors in the values of the estimated coefficients
could produce erroneous predictions. Nine coefficients are
used in the model: N, L, MCG, m, CVCG ; CDT, WCG, W, and EA.
The tether drag can be ignored because the ratio of tether-to-
cage drag force is 0.06. Gravitational and buoyancy forces
only affect the static solution, and the dynamic response of
the system in the waveband is independent of the number of
elements used when N ⬎ 5 (the effects of N on the predicted
response will be addressed later in Section 7). Thus, only

Fig. 7. Magnitude (a) and phase (b) of the actual (stars) and calculated (dotted line—without wake, solid line—with wake) transfer function estimate between
the ship and cage motion at a depth of 1730 m.
178 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182

Fig. 8. Sensitivities of the magnitude (a) and the phase (b) of the transfer function between ship and cage motion to changes in MCG (thin solid line), CVCG (dashed
line), and EA (thick solid line) for operating depth of 1730 m. The dotted line is the difference between the actual and simulated TFE.

five coefficients affect the dynamic response of the model: L,
MCG, m, CVCG ; and EA. In previous work [4], it was shown
that the non-dimensional transfer function between ship and
cage motion is only a function of two non-dimensional para-
meters, namely:
CVCG MCG
; the signs of the sensitivities are preserved between the
mc mL
when the frequency is non-dimensionalised with:
r
0 m L
v ˆL vˆ v
EA c able values are contained within the ^15% sensitivities. At
where v 0 is the non-dimensional frequency and v is the
dimensional frequency in rad s ⫺1. Therefore, there are
only three independent parameters that characterise the
dimensional response and only the dependence of the
model on CVCG ; MCG, and EA need be examined. This
is done by calculating the relative changes in the
magnitude and the absolute changes of the phase of
the transfer function between vertical motions of the
ship and cage for variations of CVCG ; MCG, and EA, of
^15% about their nominal values (Fig. 8). The sensi-
tivities of the TFE to EA look similar to those of MCG
but the magnitude and phase do not scale. The dotted
line is the difference between the real and calculated TFE
(using the nominal coefficient values) and it is addressed in
Section 7.
In the mid to upper wave band …0:15 ⱕ f ⱕ 0:25 Hz†; the
changes in the TFE are, at most, of the same order as the
changes in the coefficients and the magnitude is most sensi-
tive to CVCG while the phase is most sensitive to EA. In the
lower waveband, … f ⬍ 0:15 Hz† the magnitude and phase of
HZ S Z C are not sensitive to EA, MCG, and CVCG : Therefore,
reasonably accurate predictions of the magnitude and
phase of the transfer function can be obtained using
“rough” estimates of EA, MCG, and CVCG : As in the contin-
uous model [4], the sensitivities of the phase to MCG are zero
at 0.25 Hz and the phase is independent of MCG; and
between 0.22 and 0.25 Hz the magnitude of HZ S Z C is only
slightly sensitive to variations in CVCG : Also notable is that
continuous model and present finite-element model used
in this study.
The bounds of uncertainty for the estimated values of
MCG, CVCG and EA are less than 5% and thus all their prob-
most frequencies in the waveband, the difference between
the real and calculated TFE (the dotted line in Fig. 8) is
much larger than the ^15% sensitivities; therefore, the
model results cannot be significantly improved with realistic
changes in MCG, CVCG and EA. Additionally, because the
sensitivities change sign (either in magnitude and phase)
at different frequencies (not coincident with the error) one
cannot expect that changes in MCG, CVCG and EA will
improve the predictions at all frequencies within the
waveband.
7. Discussion
Replacing the consistent-mass matrix with the lumped-
mass matrix to uncouple the acceleration terms in the
elemental equations eliminated the inversion of a N × N ⫺
1 matrix and 6N 2 operations. This simplification signifi-
cantly reduced computation time, especially during periods
of rapid system change and very small time stepping. In
particular, simulation times were reduced from hours to
minutes for a 5000 s run. For comparison with the lumped
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
Fig. 9. Measured (thin line) and calculated (thick line) records of the tension in the tether at the A-frame (a) and acceleration of the cage (b) during a snap load.
mass model, the motion of the cage and tension at the ship
are calculated using the consistent-mass representation with
50 elements and the same 5000 s record of ship motion.
Results from the two representations are identical. Notice-
able differences only occur when a small number of
elements (⬍5) is used. For example, predictions of the
second harmonic of the natural frequency differ by 0.26
and 0.06 Hz for a two-element and a three-element model,
respectively.
The run time of the simulation increases quadratically
with the number of elements and, therefore, it is advanta-
geous to use as few elements as possible. In the model, the
mass of each element is “lumped” at its nodes and, as a
result, the mass distribution and (therefore) the accuracy
of the static results increase with increasing number of
elements. However, the dynamic motion of the cage and
tension at the ship were found to be nearly independent of
the number of elements for predictions within the wave-
band. Thus, the practical considerations for the lower
bound of N are as follows. The prediction of zero tension
(snap loads) depends on the total (static ⫹ dynamic) tension
in the tether and, therefore, a fine element break-up is
179
required. An N-element model can only predict up to N
natural frequencies and the accuracy of the predicted
value increases by increasing N.
The hydrodynamic forces on the cage have been charac-
terised by a drag coefficient multiplied by the velocity of the
cage. Although this formulation is common in practice, it is
based upon constant speed flow over a body. For steady
external (far from the body) flows, the drag coefficient
changes with changing speed because of variations in the
point of boundary layer separation and the pattern of recir-
culation over the body and in its wake. The boundary layer
and wake of an accelerating body that has reached a given
speed are never the same as those over the same but non-
accelerating body at that speed. In addition, the boundary
layer during acceleration is not the same as during decelera-
tion (for the same speed) [19].
The inertial forces on the cage have been represented by
its acceleration multiplied by its effective mass. This, too, is
common practice but this formulation obscures the physics
of unsteady flow around bluff bodies and only makes sense
if the relative rate of change of velocity is small and the drag
coefficient is not sensitive to small changes in Reynolds
180 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
number. The notion of added mass becomes simplistic when
the relative change of velocity is large and when the velocity
changes direction. For example, the added mass can
momentarily become negative when vortices detach from
the body during acceleration and very large when they reat-
tach during deceleration.
The flow history of water near the cage has been ignored.
Although this is also common in practice, it is based upon a
body travelling in a constant direction. For an oscillating
body, the effects of a disturbed flow become important when
significant momentum is transferred to the water and the
period of oscillation is less than the decay time of the
disturbed flow (wake). The heave motion of the cage is
unsteady, it oscillates between positive and negative values
and the flow around the cage has been disturbed by previous
oscillations. Thus, one cannot expect hydrodynamic forces
on the cage to be represented, with complete accuracy, by a
constant drag coefficient, a constant added mass and the
exclusion of flow history.
The sensitivity analysis indicates that differences between
measured and modeled motions cannot be explained by
parameter uncertainty. The differences may be due to the
limitations of the concept of a constant drag coefficient, a
constant added mass and excluding flow history. To test this
notion, a simple wake model was added that takes some
account of the history of motion. Only directional dependent
drag coefficients and added mass values were used because
estimates as functions of velocity and acceleration could not
be calculated from the data. The objective is to motivate
additional work on the unsteady forces on bluff bodies
attached to tethers rather than to promote this particular
model. It is assumed that, at each time step, the water
directly behind the cage (on its lee side) has the same velo-
city as the cage. The position of these water parcels is
tracked and it is assumed that their momentum decays expo-
nentially with time. When a water parcel overtakes the cage
(typically when the cage decelerates) an additional force is
added to the lea side that is proportional to the velocity
difference between the cage and this parcel. The decay
time for momentum is 2.9 s and was chosen by trial and
error. The corresponding length scale is 2.9 m for typical
speeds and is comparable to both the height of the cage and
the maximum amplitude of motion.
Including the wake model significantly improves the
results of the model for operation at 1730 m. The standard
deviation of the difference between the calculated and
measured position of the cage (tension in the tether)
is C improved by S a factor of 1.6 (1.8) and it is
s diff
Z
ˆ 0.07 m …s diff
T
ˆ 2220 N† vs 0.11 mC (4000 N) for the
TS
model without wake. The new ratio of s diff Z
…s diff † to the rms
of the measured values is 0.08 (0.16). The calculated and
real spectrum and transfer function (both magnitude and
phase) between ship and cage motion are nearly identical,
and the agreement in the upper waveband is notably
improved as seen in Figs. 6 and 7, respectively. The error
in the magnitude (phase) of the TFE is reduced from 0.12
(15⬚) to 0.07 (7⬚), respectively. Surprisingly, addition of the
wake model reduces the non-linearity of the predicted
response and brings the predicted spectrum of acceleration
at the third and fifth harmonic of the fundamental frequency
of excitation into close agreement with the measured spec-
trum. Similar improvements were found for an operation
depth of L ˆ 975 m.
An interesting phenomenon can be seen in the magnitude
of the transfer function between ship and cage motion (Fig.
7). The magnitudes of the TFE calculated for the models
with and without wake agree at 0.17 Hz. At lower (higher)
frequencies within the waveband, the magnitude of the TFE
predicted by the model with wake is larger (smaller) than
that for the model without wake. This phenomenon is easily
explained by the momentum exchange between the cage
and wake. When the cage slows from its maximum velocity,
the wake overtakes it and transfers momentum to the cage.
When the cage changes direction, it must pass back through
the wake and momentum is transferred from the cage to the
wake. At frequencies lower than 0.17 Hz, much of wake has
decayed by the time the cage reverses direction and thus,
there is a net momentum transferred to the cage. At higher
frequencies, the slowing down and reversal is more rapid
and the wake is stronger as the cage passes back through. As
a result, there is a net momentum transfer to the wake.
Motion during a snap load is non-linear and different
from that during normal operation. Large and rapid down-
ward displacements of the ship produce slack tether at the
cage. A snap load results when the tether is re-tensioned
because the velocities of the tether and cage are different.
The induced strain wave travels along the tether with speed
c and it is reflected at its ends (the ship and cage). This
results in a series of rapid changes of cage acceleration
(jerks) and tension at the ship and these changes are spaced
at intervals equal to the round trip travel time of strain waves
[1].
Snap loads occurred in less than 1% of the motion cycles
in the 5000 s record that was used to validate the model.
Such threshold conditions are very difficult to predict
because they require the numerical model to be very precise.
Without the wake model, the simulation underestimated the
large displacement and did not reproduce the slack tether—
the minimum tension was 10 000 N. With the inclusion of
the wake model, the calculated motion only slightly under-
estimated the large vertical displacement and the tension in
the tether at the cage approached zero. However, zero
tension was still not achieved and snap loads were not
predicted. Therefore the ship motion was increased by
10% so that slack tether would result and the response of
the model to this condition could be investigated. Under
these conditions, the model reproduced the observed snap
loads and the associated acceleration and tension profiles
(Fig. 9). The rapid changes of acceleration and tension
aligned with the measured values and agreed in magnitude.
Most notable is that the slope of the real and calculated
acceleration is nearly identical at the first jerk. It is therefore
 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182
concluded that the discrete model inherently has the ability
to produce characteristics associated with snap loads. A
more realistic representation of the drag, added mass and
wake could improve the prediction of slack tether so that
threshold conditions of snap loading could be better
predicted without artificially increasing the ship motion.
8. Conclusions
A one-dimensional finite-element lumped-mass model of
a caged remotely operated vehicle system using a coupled
set of reference and elastic co-ordinates was developed. The
elastic displacements were approximated with a linear trial
solution. Using Galerkin’s method, the equation for the
balance of forces at any point in the tether was discretised.
The resulting elemental equations were assembled to obtain
the finite-element equations of motion for a deep-sea ROV
system. The resulting model is a function of only nine coef-
ficients: N, L, MCG, m, CVCG CDT, WCG, W, and EA, when
tether has constant properties along its length.
Values of the model coefficients were estimated using
motion and tension measurements of the ROPOS ROV
system. Although many of the coefficients are readily esti-
mated, it is very difficult to determine accurate values of the
mass, MCG, and drag coefficient, CVCG ; of the cage and the
stiffness of the tether, EA. A least-squares method was used
to determine values for MCG and CVCG to an accuracy of 2% of
their real values. The stiffness, EA, was determined to 3%
accuracy using the speed of tensile waves in the tether, c,
which was estimated from the time of successive tension
maxima during a snap load. These methods are useful because
they only used data acquired during normal operation of the
ROV system and did not require any special procedures.
The vertical motion of the top node was prescribed based
upon measurements of ship motion. These measurements
were used to validate the model by comparing the output
against the corresponding measurements of cage motion and
tension at the ship. The model presented herein accurately
predicts (a) the motion of the cage and the tension in the
tether; (b) the spectrum of cage acceleration in the wave-
band; (c) the transfer function between the vertical motion
of the ship and cage; and (d) the natural frequency and
harmonics of the system. Since data used in the coefficient
estimation and model validation are different, the model
should be accurate for a larger range of ship motions than
presented in the validation.
Sensitivities to the three independent parameters of the
model indicate that the mismatch between the simulated and
real motion cannot be significantly improved by adjusting
the values of those parameters. However, inclusion of a
simple wake model improved the agreement between the
real and calculated results by almost a factor of 2. Eight
snap loads also occurred in the measurements used to vali-
date the model; however, these were only threshold occur-
rences. Using the wake model and increasing the ship
181
motion by 10%, the model predicted the observed snap
loads and accurately reproduced their characteristics.
These included rapid increases in the acceleration/tension
record that were spaced at regular intervals, equal to the
travel time of the strain wave along the length of deployed
tether.
Acknowledgements
We would like to thank Jim McFarlane and the staff of
International Submarine Engineering (ISE) for their contin-
ual support and assistance in this project. We would also
like to thank Inna Sharf for sharing her expertise in finite-
element modelling and Nedjib Djilali for sharing his exper-
tise in fluid dynamics. F.R. Driscoll was financially
supported by scholarships from the Natural Science and
Engineering Research Council (NSERC), the Science Coun-
cil of British Columbia (SCBC) and ISE. This work was
partially supported by the US Office of Naval Research
under contract N00014-93-1-0362 and NSERC.
References
[1] Driscoll FR, Lueck RG, Nahon M. The motion of a deep-sea remotely
operated vehicle system. Part 1: motion observations. Ocean Engi-
neering 2000;27:29–56.
[2] Hover FS, Grosenbaugh MA, Triantafyllou MS. Calculation of
dynamic motions and tensions in towed underwater cables. IEEE
Journal of Oceanic Engineering 1994;19:449–57.
[3] Grosenbaugh MA. On the dynamics of oceanographic surface moor-
ings. Ocean Engineering 1996;32:7–25.
[4] Driscoll FR, Lueck RG, Nahon M. The motion of a deep-sea remotely
operated vehicle system. Part 2: analytic model. Ocean Engineering
2000;27:57–76.
[5] Lueck RG, Driscoll FR, Nahon N. A wavelet for predicting the time-
domain response of vertically tethered systems. Ocean Engineering
2000 (in press).
[6] Caughey TK. Equivalent linearization techniques. Journal of the
Acoustic Society of America 1963;35:1706–11.
[7] Niedzwecki JM, Thampi SK. Snap loading of marine cable systems.
Applied Ocean Research 1991;13:210–8.
[8] Yoerger DR, Grosenbaugh MA, Triantafyllou MS, Burgess JJ. Drag
forces and flow-induced vibrations of a long vertical tow cable—Part
1: steady-state towing conditions. Journal of Offshore Mechanics and
Arctic Engineering 1991;113:117–27.
[9] Hover FS, Yoerger DR. Identification of low-order dynamics models
for deeply towed underwater vehicle systems. International Journal of
Offshore and Polar Engineering 1992;2:38–45.
[10] Ablow CM, Schechter S. Numerical simulation of undersea cable
dynamics. Ocean Engineering 1983;10:443–57.
[11] Newman JN. Marine hydrodynamics. Cambridge MA: MIT Press,
1989.
[12] Sparks CP. The influence of tension, pressure and weight on pipe and
riser deformations and stresses. Transactions of the ASME
1984;106:46–54.
[13] Folb R, Nelligan JJ. Hydrodynamic loading on armoured towcables.
DTNSRDC report 82/116, 1983.
[14] Shabana AA. Dynamics of multibody systems. New York: Wiley, 1989.
[15] Burnett DS. Finite element analysis from concepts to applications.
Don Mills, Ontario: Addison-Wesley, 1987.
182 F.R. Driscoll et al. / Applied Ocean Research 22 (2000) 169–182

[16] Finlayson BA. The method of weighted residuals and variational
principles. New York: Academic Press, 1972.
[17] Logan DL. A first course in the finite element method. Boston, MA:
PWS-KENT, 1992.
[18] Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numer-
ical recipes in C. 2nd ed. Cambridge: Cambridge University
Press, 1992.
[19] Sarpkaya T, Isaacson M. Mechanics of wave forces on offshore struc-
tures. New York: Van Nostrand Reinhold, 1981.
[20] Robigou V. Journal of Geophysical Research Letters 1993;20:1.