Ocean Engineering 267 (2023) 113280

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Overcoming the added mass effect in FSI calculations relating to dynamic

squat
Emmanuel Lefrançois a ,∗, Bo Yang a , Sami Kaidi b , Nisrine Mohamad c
a
Roberval Laboratory, Sorbonne Universités - Université de Technologie de Compiègne, CS 60319, 60203 Compiègne, France
b
CEREMA, 60203 Compiègne, France
c
Marine Engineering Department, Faculty of Mechanical and Electrical Engineering, Tishreen University, Syrian Arab Republic

ARTICLE INFO ABSTRACT

Keywords: This paper presents a corrected partitioned scheme for investigating fluid–structure interaction (FSI) generated
Ship squat effect by ship squat in presence of transient effects. A theoretical example describing the motion of a piston enclosed
Dynamics in a U-tube gives rise to an analysis of the divergence effect resulting from added mass, and it has been
Fluid–structure interaction
concluded that divergence always occurs when the mass in motion is less than the added mass. An added-
Added mass
mass-corrected scheme is proposed to guarantee the convergence of the coupling scheme irrespective of the
Weak coupling
Energy conservation
mass ratio. This corrected scheme is successfully applied to the ship dynamics in an analysis of ship squat via
a non-stationary potential fluid flow approach. The results of our coupled calculations have been compared
to the squat stability models detailed in Alderf et al. (2010); Alderf et al. (2015) in which transient effects
were neglected in the fluid flow solver. Finally, energy transfers between the ship’s motion and the fluid flow
have been analyzed. We confirm the energy conservation capability of the proposed fluid–structure coupling
scheme in relation to the transient response of the ship.

1. Context and interest of this study environmental pollution, and in some cases loss of life and prop-
erty (Mazaheri et al., 2014; Zhang et al., 2019). A number of such
Water transportation, with its low environmental impact, is nowa- accidents have occurred in recent years (CIAIM, 2018, 2016), and
days being promoted as way of limiting greenhouse gas emissions represent a significant proportion of all accidents for inland ships
(Galieriková and Sosedová, 2016). Many countries have inherited an navigating in shallow rivers and canals. Ship squat and grounding
extensive network of navigation canals, but which needs to be adapted have become a research hotspot in relation to shipping and ship me-
to accommodate ships of increasingly high tonnage (Radmilović, 2011). chanics (Pedersen, 2010; Zipfel and Lehmann, 2010; Barrass, 2004).
This adaptation can mean widening and deepening existing canals, or Different methods have been used to predict ship squat. These in-
digging new canals with a wider gauge (such as the Seine-Nord Canal clude CFD methods (Jachowski, 2008; Härting et al., 2009; Kok et al.,
in France). 2020), double-body methods (McTaggart, 2018), slender-body meth-
However, issues relating to the maneuverability of ships remain ods (Gourlay, 2008), and experimental methods (Elsherbiny et al.,
omnipresent, with the close proximity of the canal banks calling for 2019; Lataire et al., 2012; Xu et al., 2016). None of these, however,
a high level of maneuvering capability. Proximity effects also mean offset the added mass effect.
that precautions must be taken when ships pass each other. For a At low speeds, oscillation amplitudes will remain low. However,
ship considered in isolation, changes of heading, sudden variations in it has been demonstrated that there is a so-called squat maximum
depth, and actions on the pilot control (rudder, propeller) can induce speed beyond which any oscillation on the vertical plane inevitably
a dynamic oscillatory response of the ship that may lead to an over- leads to instability of the vessel and consequently an increased risk of
sinking effect (squat). This may cause the vessel to strike the bottom if irreversible damage.
the under-keel clearance (UKC) is small in relation to the amplitude of Various studies have been undertaken (Alderf et al., 2010, 2015,
oscillation. 2011; Debaillon, 2010) concerning the quasi-static and dynamic re-
Grounding accidents can cause permanent deformations and dam- sponses of ships. The existence of two sinking solutions was identi-
age to hull structure, and lead to the leakage of dangerous chemicals, fied (Alderf et al., 2015), one of which is stable (the closest to the free

∗ Corresponding author.
E-mail address: emmanuel.lefrancois@utc.fr (E. Lefrançois).

https://doi.org/10.1016/j.oceaneng.2022.113280
Received 29 August 2022; Received in revised form 15 November 2022; Accepted 23 November 2022
Available online 5 December 2022
0029-8018/© 2022 Elsevier Ltd. All rights reserved.
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

surface), and the other is unstable. The studies mentioned are all based significant to be overlooked. The second is to show that the use of a
on a quasi-static resolution of the fluid calculation and therefore on the classical coupling scheme systematically leads to divergence, due to the
hydrodynamic resultant exerted on the hull. This quasi-static approach added mass effect. The third objective is to demonstrate the interest of
does not incorporate the effects of added mass exclusively linked to the an added-mass-corrected coupling scheme to guarantee convergence.
acceleration of the ship. However, Song et al. (2013), Lefrançois (2017) The paper is divided into three sections, the first being this introduc-
and Lefrançois et al. (2016) showed (for two very different application tion. In a second section the example of a piston moving in a U-tube and
cases) that these effects of added mass can lead to a systematic diver- interacting with a heavy fluid such as a liquid is fully developed so as to
gence in the calculations of fluid–structure coupling. These works also enable a numerical analysis. The limit value of the mass ratio between
demonstrate that the coupling scheme can be modified to counteract the fluid and the piston for avoiding the divergence effect is predicted
the added mass effect and to improve convergence to a solution. and numerically validated using a simple finite difference scheme ap-
All of these studies are based on the computation of fluid–structure proach. This section establishes that the added-mass-corrected coupling
interactions (FSI). This FSI approach is based on a partitioned coupling scheme is effective, and the same numerical model is used to show
with a dedicated solver for each of the two physics (namely fluid that convergence is always guaranteed. A third section presents a 2D
flow and ship structure dynamics). Exchanges (pressure, hull position, application used for investigating the added mass effect encountered by
etc.) between the two solvers occur regularly via a coupling scheme ship’s dynamic squat. The fluid model is based on 2D finite elements,
(Piperno et al., 1995; Felippa et al., 2001) that is based on successive and the finite element model for estimating the added mass resulting
solutions produced by the fluid and ship structure solvers. The coupling from the ship motion is described in detail. Section 4 draws upon the
is said to be loosely coupled partitioned if only one shot (that is to say literature to validate the added mass estimation and the response of
a single computation) per time step is required for each field, and ship dynamics under general fluid flow conditions. In particular, it is
strongly coupled partitioned if an iterative procedure is used to obtain noted that the added mass systematically exceeds the mass of the ship,
convergence of the coupled solution (Idelsohn et al., 2009). In an and that using a corrected coupling scheme is the only way to ensure
industrial context, the greatest advantage of partitioned coupling over convergence to a coupled solution. Finally, concluding remarks refer to
monolithic coupling (i.e., with a single solver) is the modularity of the work currently underway with a view to extending this approach to 3D
approach, which makes the different solvers much easier to implement cases in restricted waterways.
and allows distributed computation. However, the major drawback
2. Theoretical investigation
of the standard partitioned FSI coupling scheme is that where higher
density fluids are involved (meaning strong effects of added mass Idel-
2.1. Mathematical model for fluid-structure interaction
sohn et al., 2009; Kassiotis et al., 2010), convergence is no longer
guaranteed, and there will generally be divergence, regardless of the
The problem of a moving piston in a U-tube is illustrated in Fig. 1.
chosen time step for incompressible flows (Fernández et al., 2005; van
The []wall pressure on the upper and lower faces of the piston (mass
Brummelen, 2009). To counter this, a number of approaches have been
𝑚, area 𝑆) are respectively denoted by 𝑝𝑎𝑡𝑚 and 𝑝𝑝𝑖𝑠𝑡𝑜𝑛 . The other end
proposed (Hughes et al., 1981; Fernández et al., 2006; Tallec et al.,
of the fluid domain is open to external static pressure 𝑝𝑎𝑡𝑚 . The U-tube
2005), including the semi-implicit discretization (Causin et al., 2005),
is filled with an incompressible, inviscid heavy fluid of mass density 𝜌.
the adaptive under-relaxation by Aitken (Küttler and Wall, 2008), but
convergence is not always guaranteed or may be slow in cases of high- 2.1.1. Structure model
density fluids such as blood or water. An attractive technique for new The motion of the piston is governed by the Fundamental Principle
partitioned procedures based on Robin-type transmission conditions is of Dynamics (FPD). For a movable piston located at 𝑤(𝑡) with respect to
proposed in Badia et al. (2008) but requires adjustable parameters to its position at rest, its 𝑧-axis projected form may be written as:
fit. In Michler et al. (2005) an interface-GMRES method is described
which reuses information from previously solved similar problems with 𝑑2𝑤
𝑚 = (𝑝𝑝𝑖𝑠𝑡𝑜𝑛 (𝑡) − 𝑝𝑎𝑡𝑚 )𝑆 − 𝑚𝑔. (1)
using the Krylov space. The theoretical background required means that 𝑑𝑡2
this approach does not have the attractive simplicity of the subiteration Gravity is denoted by 𝑔, 𝑧𝐴 (𝑡) and 𝑧𝐵 (𝑡) denote the time-dependent
method. vertical positions of the fluid interface with the piston (𝐵) and the
It has been reported in Causin et al. (2005) that the divergence ef- external conditions (𝐴) respectively.
fect is clearly dependent on the size of the fluid domain, and in Förster
et al. (2007) it has been proved that for ratios between the mass 2.1.2. Fluid model
densities of the fluids and the structure, the coupling scheme diverges. The []wall pressure exerted on the lower face of the piston can be
Our own experience shows that the divergence effect is best explained computed exactly using the non stationary form of Bernoulli’s principle,
in terms of the ratio between the inertia of the structure and the applied along a streamline connecting the piston (B) and the external
integration of the pressure force acting on a movable structure in a boundary conditions (A):
𝐵 [ ]𝐵
fluid flow. This is denoted as the added mass effect. The higher this is, 𝜕𝑉 𝑉2
the better, and artificially introducing an additional mass term on both 𝜌 d𝑠 + 𝑝 + 𝜌𝑔𝑧 + 𝜌 =0
∫𝐴 𝜕𝑡 2 𝐴
sides of the FSI equation can then favor the convergence criterion. This
Although the velocity 𝑉 (𝑡) is unsteady, it is uniform (𝑉 ≠ 𝑉 (𝑠)) along
idea is not new and earlier attempts may already be found in Tezduyar
the streamline of abscissa 𝑠, of average length 𝐿, the section being
(2004) and Tezduyar et al. (2006) as a short-cut approach for improving
constant. In can be written as follows:
the convergence based on a modified mass technique, and in Connell
𝑑𝑉
and Yue (2007) where the authors propose modifying the structure- 𝜌 𝐿 + 𝑝𝐵 (𝑡) + 𝜌𝑔𝑧𝐵 (𝑡) − 𝑝𝑎𝑡𝑚 − 𝜌𝑔𝑧𝐴 (𝑡) = 0
𝑑𝑡
to-fluid mass ratio. A further analysis of the modified-mass approach
the two terms of dynamic pressure canceling each other out. It is then
and a comparison with other iterative methods is provided in van
possible to isolate the relative pressure exerted under the piston in 𝐵:
Brummelen (2011). But the corrections only concern scalar additional
terms without general methodology and do not take into account the 𝑑𝑉
𝑝𝐵 (𝑡) − 𝑝𝑎𝑡𝑚 = −𝜌 𝐿 − 𝜌𝑔(𝑧𝐵 (𝑡) − 𝑧𝐴 (𝑡)). (2)
multi-modal behavior of the flexible structure (except for the first 𝑑𝑡
mode). To complete the model, the mass flow conservation law should be
The present paper has three objectives. The first is to show that included to connect the two points 𝐴 and 𝐵 as follows:
in flow calculations to precisely estimate pressures, if non-stationary 𝑑
effects are taken into account then the effects of added mass are too 𝜌d𝑣 = 0 ⇒ 𝐕 ⋅ 𝐧d𝑠 = 0.
𝑑𝑡 ∬𝑉𝑚 (𝑡) ∮𝑆𝑚 (𝑡)

2
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 2. Partitioned and staggered coupling scheme.

to get the final relation:

( ) 𝑑2𝑤
𝑚 + 𝑚𝑎𝑑𝑑 + 2𝜌𝑔𝑆𝑤(𝑡) = 0 with 𝑚𝑎𝑑𝑑 = 𝜌𝐿𝑆. (4)
𝑑𝑡2
Eq. (4) defines the strong coupling equation of the fluid–structure
interaction. The term 𝑚𝑎𝑑𝑑 denotes an added mass that only results from
the non-stationary part of the pressure equation (see Eq. (2)).
Fig. 1. Moving piston in a U-tube filled with an incompressible fluid.

Analytical solution
The differential Eq. (4) is homogeneous and its integration does not
By detailing the fluxes at both ends of the domain: lead to any particular difficulty. It admits the exact solution form:
( ) √
𝑑𝑧 𝑑𝑧
𝑆𝑉𝐴 − 𝑆𝑉𝐵 = 0 ⇒ −𝑆 𝐴 − 𝑆 𝐵 = 0 2𝜌𝑔𝑆
𝑑𝑡 𝑑𝑡 𝑤(𝑡) = 𝑤𝑜 cos(𝜔𝑡) with 𝜔 = . (5)
𝑚 + 𝜌𝐿𝑆
it results:
The coupled pulsation of the fluid column is defined by 𝜔.
𝑑𝑧𝐴 𝑑𝑧
= − 𝐵 which, after integration, gives 𝑧𝐴 (𝑡) + 𝑧𝐵 (𝑡) = 𝑤𝑜
𝑑𝑡 𝑑𝑡 2.2. Strongly coupled partitioned scheme and convergence analysis
with 𝑤𝑜 = 𝑧𝑜𝐴 +𝑧𝑜𝐵 denoting the level difference between both interfaces
𝐴 and 𝐵 at time 𝑡 = 0. Eq. (1) is the fluid–structure interaction equation. The partitioned
approach consists in allocating each of the terms on the left and right of
2.1.3. Coupling boundary conditions the equals sign to its own dedicated solver, respectively for the fluid and
The coupling between the dynamics of the piston (Eq. (1)) and the the structure parts. The staggered nature of the coupling scheme means
pressure of the fluid exerted on 𝐵 (Eq. (2)) is based on the common that both terms are successively solved in 4 steps, as illustrated in Fig. 2.
expression of their respective speeds: Indices 𝑛 and 𝑛 + 1 correspond to time steps 𝑡 and 𝑡 + 𝛥𝑡 respectively.
𝑑𝑧 𝑑𝑤 𝑑𝑉 𝑑2𝑤 In order to reduce the discrepancy between a piston motion computed
𝐕= 𝐤= 𝐤 ⇒ = ∀𝑡. at 𝑛 + 1 from a pressure given at time 𝑛, and consequently to reinforce
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡2
This corresponds to the kinematic compatibility condition between the the coupling, an enclosed iterative procedure is then added inside the
fluid and structure at their common interface. The dynamic condition temporal loop. The general algorithm (Algorithm 1) is based on an
expresses the equality between the pressures exerted under the piston iterative loop (𝑘) enclosed in a time loop (𝑛). Each equation is then
and 𝑝𝐵 : modified, with 𝑛 + 1 replaced by 𝑘 + 1 to write the iterative form:

𝑝𝐵 (𝑡) = 𝑝𝑝𝑖𝑠𝑡𝑜𝑛 ∀𝑡. 𝑚𝑤̈ 𝑘+1 = (𝑝𝑘𝐵 − 𝑝𝑎𝑡𝑚 )𝑆 − 𝑚𝑔 (6)

and
2.1.4. Mathematical solution
Injecting the two boundary conditions into the dynamic equation: 𝑝𝑘𝐵 − 𝑝𝑎𝑡𝑚 = −𝜌𝑉̇ 𝑘 𝐿 − 𝜌𝑔(𝑧𝑘𝐵 − 𝑧𝑘𝐴 ), (7)
2 ( ) where 𝑉̇ 𝑘 = 𝑤̈ 𝑘 and 𝑧𝑘𝐵 − 𝑧𝑘𝐴 = 2𝑤𝑘 + 𝑧𝑜𝐵 − 𝑧𝑜𝐴 .
𝑑 𝑤 𝑑𝑉
𝑚 = −𝜌 𝐿 − 𝜌𝑔(𝑧𝐵 (𝑡) − 𝑧𝐴 (𝑡)) 𝑆 − 𝑚𝑔
𝑑𝑡 2 𝑑𝑡 The discretization in time to be considered is based on the second
it gives: time order scheme of Newmark–Wilson (Dhatt et al., 2012):

𝑑2𝑤 𝛥𝑡2 𝑛 𝛥𝑡
(𝑚 + 𝜌𝐿𝑆) + 𝜌𝑔𝑆(𝑧𝐵 (𝑡) − 𝑧𝐴 (𝑡)) + 𝑚𝑔 = 0. 𝑤𝑘+1 = 𝑤𝑛 + 𝛥𝑡𝑤̇ 𝑛 + (𝑤̈ + 𝑤̈ 𝑘+1 ), 𝑤̇ 𝑘+1 = 𝑤̇ 𝑛 + (𝑤̈ 𝑛 + 𝑤̈ 𝑘+1 ) (8)
4 2
𝑑𝑡2
to extract the second derivative in time:
The equilibrium position 𝑧𝑜𝐵 of the mass 𝑚 in the U-tube of constant
4 𝑘+1 4
section 𝑆 = 𝑅𝑏 is governed by the law of hydrostatics: 𝑤̈ 𝑘+1 = 𝑤 − (𝑤𝑛 + 𝛥𝑡𝑤̇ 𝑛 ) − 𝑤̈ 𝑛 (9)
𝛥𝑡2 𝛥𝑡2
𝑚𝑔
𝑝𝐵 + 𝜌𝑔𝑧𝑜𝐵 = 𝑝𝐴 + 𝜌𝑔𝑧𝑜𝐴 with 𝑝𝐵 = 𝑝𝑎𝑡𝑚 + and 𝑝𝐴 = 𝑝𝑎𝑡𝑚 Combining Eqs. (6) and (9), and adopting an incremental approach
𝑆
such as 𝛥𝑤 = 𝑤𝑘+1 −𝑤𝑛 , the iterative scheme for the staggered coupling
to write:
scheme can be written as:
( )
𝑚𝑔 = 𝜌𝑔(𝑧𝑜𝐴 − 𝑧𝑜𝐵 )𝑆 (3) 𝛥𝑡2 ( 𝑘 ) 𝛥𝑡2 𝑛
𝑚𝛥𝑤 = (𝑝𝐵 − 𝑝𝑎𝑡𝑚 )𝑆 − 𝑚𝑔 + 𝑚 𝛥𝑡𝑤̇ 𝑛 + 𝑤̈ (10)
Injecting Eq. (3) into the Fundamental Principle expression gives: 4 4

𝑑2𝑤
(𝑚 + 𝜌𝐿𝑆) + 𝜌𝑔𝑆(𝑧𝐵 (𝑡) − 𝑧𝑜𝐵 ) − 𝜌𝑔𝑆(𝑧𝐴 (𝑡) − 𝑧𝑜𝐴 ) = 0. The coupling scheme is based on data passing, so as to update
𝑑𝑡2
variables common to the two physics. The piston transmits its position
The following equation can be written:
𝑤𝑘+1 and its velocity 𝑤̇ 𝑘+1 , whereas the fluid transmits its pressure field
𝑧𝐵 (𝑡) − 𝑧𝑜𝐵 = 𝑤(𝑡) and 𝑧𝑜𝐴 − 𝑧𝐴 (𝑡) = 𝑤(𝑡), 𝑝𝑘+1 acting on the piston.

3
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Algorithm 1 General partitioned algorithm Table 1

Numerical parameters.
Require: 𝑤(0), 𝑤(0)̇ ̈
→ 𝑤(0)
𝐿 𝑆 𝜌 𝑤(0) ̇
𝑤(0) 𝑁
1. 𝑤𝑛 ← 𝑤(0), 𝑤̇ 𝑛 ← 𝑤(0),̇ 𝑤̈ 𝑛 ← 𝑤(0)
̈ [m] [m2 ] kg/m3 [m] [m/s] –
Require: 𝑝(0)
1 0.1 1000 0.01 0 70
2. for 𝑛 = 1 to 𝑛𝑠𝑡𝑒𝑝 do
3. 𝑤𝑘 = 𝑤𝑛
4. while 𝜖 > 10−12 do Table 2
Average number of iterations required for convergence vs. mass ratio (𝑚𝑎𝑑𝑑 = 𝑚𝑓 ).
5. —————— Dynamic piston part
6. 𝑤𝑘+1 ← Eq. (10) 𝑚𝑓 ∕𝑚 0.2 0.5 0.66 0.8 0.9 0.95 0.98 0.99 1 2 10
7. 𝑤̇ 𝑘+1 ← Eq. (8) Standard 14 29 49 89 210 427 495 1404 ∞ – –
8. —————— Fluid part Corrected 5 5 5 5 5 5 5 5 5 5 5

9. 𝑝𝑘 ← Eq. (7)
10. 𝜖 = |𝑤𝑘+1 − 𝑤𝑘 |
11. end while By definition, the pressure force term and the added mass term are
12. 𝑤𝑛+1 = 𝑤𝑘+1 opposite signs such as 𝑓𝑝 (𝑡) = −𝑚𝑒𝑥𝑎𝑐𝑡 𝑒𝑥𝑎𝑐𝑡
𝑎𝑑𝑑 𝑤 with 𝑚𝑎𝑑𝑑 > 0 : it explains
̈
13. 𝑤̇ 𝑛+1 , 𝑤̈ 𝑛+1 ← Eq. (8)
how this added term helps to decrease the impact of the force term
14. 𝑤𝑛 ← 𝑤𝑛+1 , 𝑤̇ 𝑛 ← 𝑤̇ 𝑛+1 , 𝑤̈ 𝑛 ← 𝑤̈ 𝑛+1
and to increase the inertia term.
15. end for
Using the same approach, a corrected version for the iterative
incremental solution can be written as follows:
( )
𝛥𝑡2 𝛥𝑡2 𝑛
Convergence analysis (𝑚 + 𝑚𝑎𝑑𝑑 )𝛥𝑤 = 𝑓𝑝 (𝑝𝑘 , 𝑚𝑎𝑑𝑑 𝑤̈ 𝑘 ) + (𝑚 + 𝑚𝑎𝑑𝑑 ) 𝛥𝑡𝑤̇ 𝑛 + 𝑤̈ (13)
4 4
The convergence analysis for the staggered coupling scheme (Eq.
(10)) involves injecting the fluid term equation Eq. (7). Retaining only The convergence analysis needs to be combined with Eq. (7); hence the
the iterative terms indexed with 𝑘 and 𝑘 + 1, we may write: following relation can be written:

𝑚
4 𝑘+1
𝑤 + ⋯ = −𝜌𝐿𝑆
4 𝑘
𝑤 + ⋯ − 𝜌𝑔𝑆(2𝑤𝑘 + 𝑧𝑜𝐵 − 𝑧𝑜𝐴 ) − 𝑚𝑔 𝑚𝑎𝑑𝑑 𝑤̈ 𝑘+1 + 𝑚𝑤̈ 𝑘+1 = (−𝜌𝑤̈ 𝑘 𝐿 − 𝜌𝑔(2𝑤𝑘 + 𝑧𝑜𝐵 − 𝑧𝑜𝐴 ))𝑆 − 𝑚𝑔 + 𝑚𝑎𝑑𝑑 𝑤̈ 𝑘
𝛥𝑡2 𝛥𝑡2
After factorization, the following iterative form is obtained: It is then straightforward to show that the amplification factor of the
( ) iterative scheme becomes:
𝜌𝐿𝑆 𝛥𝑡2
𝑤𝑘+1 = − + 𝜌𝑔2𝑆 𝑤𝑘 + ⋯ 2(𝑚𝑎𝑑𝑑 − 𝑚𝑓 ) − 𝛥𝑡2 𝜌𝑔𝑆
𝑚 4𝑚 𝐺𝑐𝑜𝑟𝑟 = such that 𝑤𝑘+1 = 𝐺𝑐𝑜𝑟𝑟 𝑤𝑘 + ⋯
2(𝑚𝑎𝑑𝑑 + 𝑚)
After simplification:
Whatever the mass ratio between the fluid and the piston, convergence
2𝑚𝑓 + 𝛥𝑡2 𝜌𝑔𝑆
𝑤𝑘+1 = − 𝑤𝑘 with 𝑚𝑓 = 𝜌𝐿𝑆. is now unconditionally guaranteed. Finally, at convergence, 𝑤𝑘+1 = 𝑤𝑘
2𝑚
means that the two corrective terms cancel each other out, yielding the
We can thus extract the coefficient of amplification 𝐺𝑁.𝑊 . controlling
original equation.
the convergence of this scheme:
2𝑚𝑓 + 𝛥𝑡2 𝜌𝑔𝑆
𝐺𝑁.𝑊 . = − such that 𝑤𝑘+1 = 𝐺𝑁.𝑊 . 𝑤𝑘 + ⋯ (11) 2.4. Numerical validation
2𝑚
Convergence is guaranteed for |𝐺𝑁.𝑊 . | < 1. In this case, if a sufficiently The example given here is useful for two reasons. First, it highlights
small 𝛥𝑡 time increment for the second term is assumed to be neglected, the limitations regarding the effects of added mass in the classical
convergence is ensured if the mass ratio is such that:
fluid–structure coupling scheme. Second, it confirms the significant
𝑚 > 𝑚𝑓 . contribution that the improved model has to offer. Table 1 gives the
numerical parameters retained for the example. The parameter 𝑁 is the
It is clear that for general cases this scheme is seriously limited by
sampling value of the coupled period 𝜔 (see Eq. (5)) used to calculate
its stringent convergence restriction. For a more classical temporal
the time step 𝛥𝑡 between two successive steps. The total time considered
scheme, such as an explicit second-order centered difference, a slightly
different 𝐺𝑒𝑥 coefficient is obtained: is equal to three coupled periods (𝑇 = 2𝜋∕𝜔). The coupling is based on
( ) Algorithm 1 given above. Eq. (10) in the standard scheme is replaced
𝑚𝑓 + 𝛥𝑡2 2𝜌𝑔𝑆 by Eq. (13) for the corrected scheme.
𝐺𝑒𝑥 = − .
𝑚 The results correspond to the number of iterations required to
The convergence condition, however, remains identical. satisfy the convergence criterion 𝜖 = |𝑤𝑘+1 − 𝑤𝑘 | < 10−14 . Numbers of
iterations for different mass ratios 𝑚∕𝑚𝑓 are shown in Table 2, for both
2.3. Added mass effect correction for FSI coupling scheme the standard and the corrected schemes. The added mass term 𝑚𝑎𝑑𝑑 is
taken to be equal to the fluid mass 𝑚𝑓 .
The objective here is to slightly modify the current coupling scheme The standard scheme seems to be unsuitable for general cases, with
in order to ensure the convergence criterion independently of the fluid no convergence for mass ratios greater than one, thus confirming our
mass density. Eq. (11) implies that the higher the force term (function convergence analysis above. For the corrected scheme, however, the
of 𝑚𝑓 ), the poorer the convergence. Moreover, it implies that the higher average number of iterations for convergence is small and remains
the inertia term (𝑚𝑤̈ 𝑖+1 ), the better the convergence. constant for all mass ratios considered, which confirms the interest of
The correction that we introduce to improve the convergence is the added mass correction.
based on the observation that increasing the inertial term and decreas- The choice 𝑚𝑎𝑑𝑑 = 𝑚𝑓 as in Table 2 is optimum, but not mandatory.
ing the force term will favor the iterative process. An estimated added
A sensitivity analysis of the corrected scheme with regard to the added
mass term (𝑚𝑎𝑑𝑑 𝑤)̈ is then added on both sides but shifted with respect
mass value has been done and results are shown in Table 3. It would
to the iterative index, namely:
appear less penalizing to overestimate the added mass term than to
𝑚𝑎𝑑𝑑 𝑤̈ 𝑘+1 + 𝑚𝑤̈ 𝑘+1 ≈ (𝑝𝑘𝐵 − 𝑝𝑎𝑡𝑚 )𝑆 − 𝑚𝑔 + 𝑚𝑎𝑑𝑑 𝑤̈ 𝑘 (12) underestimate it.

4
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 3. Fluid domain  for ship motion resulting from FSI.

Fig. 4. Iterative loop of the partitioned iterative coupling scheme between two time steps.

Table 3 Algorithm 2 General partitioned algorithm for dynamic squat

Sensitivity analysis of corrected scheme with 𝑚𝑎𝑑𝑑 . investigation
𝑚𝑓 ∕𝑚 0.2 0.5 0.66 0.8 0.9 0.95 0.98 1 2 10
Require: Initialization for all variables. . .
𝑚𝑎𝑑𝑑 = 0.5𝑚𝑓 10 14 15 17 18 19 19 19 29 109 1. for 𝑛 = 1 to 𝑛𝑠𝑡𝑒𝑝 do
𝑚𝑎𝑑𝑑 = 2𝑚𝑓 12 15 16 17 18 18 18 18 21 26
2. while 𝜖 > 10−4 do
𝑚𝑎𝑑𝑑 = 5𝑚𝑓 21 33 38 41 43 44 44 44 56 71
𝑚𝑎𝑑𝑑 = 10𝑚𝑓 36 61 70 78 81 83 84 85 108 142 3. —————— Fluid part
4. 𝜑𝑘𝑖 ← Eqs. (18), (19)
5. 𝜑̇ 𝑘𝑖 ← Eq. (21)
6. 𝐕(𝐱𝑖 ) = ∇𝜑𝑘𝑖
3. Dynamic squat investigation with added mass effect 7. 𝑝𝑘 (𝐱𝑖 ) ← Eq. (20)
8. —————— Ships’s dynamic part
3.1. Numerical fluid–structure coupling model 9. 𝐅𝑘𝑝 ← Eq. (15)
10. 𝑤𝑘+1 , 𝑤̇ 𝑘+1 , 𝑤̈ 𝑘+1 ← Eq. (17)
The dynamic equilibrium of the ship is governed by the fundamental 11. Mesh coordinates update
principle of dynamics. Let us denote the vertical position and velocity 𝐱𝑘+1 − 𝐱𝑛−1
12. 𝐕𝑘+1 =
of the ship as 𝐰(𝑡) = 𝑤(𝑡)𝐤 and 𝐰(𝑡)
̇ = 𝑤(𝑡)𝐤
̇ respectively, as illustrated 𝑚𝑒𝑠ℎ 2𝛥𝑡
in Fig. 3: 13. 𝜖 = |𝑤𝑘+1 − 𝑤𝑘 |
We introduce the notations In, Ship, Out, Botm and FreeS respectively 14. end while
for the segments 𝐴𝐹 , 𝐷𝐸, 𝐵𝐶, 𝐴𝐵 and 𝐶𝐷∪𝐸𝐹 . The fluid domain is , 15. 𝜑𝑛−1
𝑖 ← 𝜑𝑛𝑖 , 𝜑𝑛𝑖 ← 𝜑𝑘+1 𝑖
16. 𝑤 𝑛−1 ← 𝑤𝑛 , 𝑤𝑛 ← 𝑤𝑘+1 , 𝑤̇ 𝑛 ← 𝑤̇ 𝑘+1 , 𝑤̈ 𝑛 ← 𝑤̈ 𝑘+1
and  denotes its closed frontier. The notations 𝑇 , 𝐻, and 𝑈 𝐾𝐶 stand
respectively for the draft, the water depth, and the under-keel clearance 17. 𝐱𝑛−1 ← 𝐱𝑛 , 𝐱𝑛 ← 𝐱𝑛+1
(UKC), where 𝑈 𝐾𝐶 = 𝐻 − 𝑇 + 𝑤. 18. end for
Limited to the vertical degree of freedom (𝐰(𝑡)) the following rela-
tion can be obtained:
𝑑2𝐰 The following sections describe the dynamic ship model, the fluid
𝑚 = 𝑚𝐠 + (𝑝 − 𝑝𝑎𝑡𝑚 )𝐧𝑑𝑆 (14)
𝑑𝑡2 ∫𝑆ℎ𝑖𝑝 𝑠ℎ𝑖𝑝 model based on the calculation of a velocity potential with triangular
where the bold notation is here used to denote a vector quantity. Let finite elements, and the added mass estimation obtained by solving
𝑚 denote the mass of the ship, 𝐠 the gravity constant, 𝐕∞ the relative Poisson’s equation. The mesh deformation is based on the Moving
fluid flow due to horizontal ship motion, 𝑝𝑎𝑡𝑚 the atmospheric pressure, Submesh Approach (MSA) for fast fluid mesh deformation resulting from
and 𝑝𝑠ℎ𝑖𝑝 the absolute pressure exerted by the 2D water flow under the ship motion.
elementary surface 𝑑𝑆 of the ship. The unit vector 𝐧 is here external to
the fluid domain. 3.1.1. Dynamic ship model with added mass correction terms
Here, the coupling is based on an iterative strongly-coupled parti- The equilibrium relation of the ship is governed by Eq. (14). The
tioned scheme, as illustrated in Fig. 4. pressure force acting under the ship is calculated from the pressure field
Each of the physics has its own dedicated solver based here on in 𝑝𝑠ℎ𝑖𝑝 transmitted by the fluid solver at the previous iterative step 𝑘:
house Matlab scripts, the coupling algorithm is described in (Algorithm
2). 𝐅𝑘𝑝 = (𝑝𝑘𝑠ℎ𝑖𝑝 − 𝑝𝑎𝑡𝑚 )𝐧𝑑𝑆 (15)
∫𝑆ℎ𝑖𝑝

5
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 5. Illustration of the two concepts of the MSA.

The added-mass-corrected form results from Eq. (14) modified accord- the fluid mesh for the deformation. The fluid mesh is then updated on
ing to: the coarse deformed mesh (Fig. 5(a)) with an interpolation technique
𝑑 2 𝐰𝑘+1 𝑑 2 𝐰𝑘 similar to the finite element method (Fig. 5(b))
(𝑚 + 𝑚𝑎𝑑𝑑 ) ≈ 𝑚𝐠 + 𝐅𝑘𝑝 + 𝑚𝑎𝑑𝑑 (16) The MSA methodology may be separated into four successive steps
𝑑𝑡 2 𝑑𝑡2
illustrated in Fig. 6:
The acceleration term is here based on a classical centered differ-
ence scheme, which leads to the iterative scheme: Step 1: Fine mesh generation;
( )
−𝑚𝑔 + 𝐹𝑝𝑘 + 𝑚𝑎𝑑𝑑 𝑤̈ 𝑘 Step 2: Triangular ‘zones’ (coarse mesh) definition (or generation)
𝑤𝑘+1 = 2𝑤𝑛 − 𝑤𝑛−1 + 𝛥𝑡2 (17) with:
𝑚 + 𝑚𝑎𝑑𝑑
(a) identification of the zone corner nodes;
where 𝑛 − 1 and 𝑛 denote respectively the time stations 𝑡 − 𝛥𝑡 and 𝑡, and
(b) writing of the zone connectivity (such as finite element).
𝑘, 𝑘 + 1 are two successive iterative steps with 𝛥𝑡 the time step.
Step 3: Deformation of the coarse mesh (by extraction or computation
3.1.2. 2D-fluid model of the coordinates of the corners of the new zone);
The fluid is assumed to be incompressible and inviscid in order Step 4: Loop on all the nodes (indiced 𝑝 ) of the fine mesh:
to consider lighter meshes and faster calculations with no lack of
generalities. Indeed, viscosity has no impact on added mass effect. A (a) Find (or extract, if already stored) the corresponding
potential flow 𝜑 can therefore by defined by 𝐕(𝐱) = ∇𝜑, with 𝐕(𝐱) the zone;
local fluid velocity, governed by Poisson’s equation: (b) Extract the updated coordinates 𝑋𝑖 , 𝑌𝑖 of the corners of
the zone, and the corresponding displacement compo-
𝛥𝜑(𝐱) = 0 ∀𝐱 ∈ . (18)
nents 𝑈𝑖𝑧𝑜𝑛𝑒 , 𝑉𝑖𝑧𝑜𝑛𝑒 ;
This equation is completed by the following boundary conditions (see (c) Extract the nodal coordinates 𝑥𝑝 and 𝑦𝑝 ;
Fig. 3 for boundary spotting): (d) Calculate the nodal values of the three approximation
𝜕𝜑 || 𝜕𝜑 || 𝜕𝜑 || functions 𝑁𝑖 (𝑥𝑝 , 𝑦𝑝 );
𝜑|𝐼𝑛 = 0, = 𝐕∞ ⋅ 𝐧, = 0, ̇
= 𝑤. (19)
𝜕𝑛 ||𝑂𝑢𝑡 𝜕𝑛 ||𝐵𝑜𝑡𝑚, 𝐹 𝑟𝑒𝑒𝑆 𝜕𝑛 ||𝑆ℎ𝑖𝑝 (e) Interpolate the nodal displacement components:
∑
3 ∑
3
A classical finite element (Dhatt et al., 2012) with 2D linear triangular 𝑢(𝑥𝑝 , 𝑦𝑝 ) = 𝑁𝑖 (𝑥𝑝 , 𝑦𝑝 )×𝑈𝑖𝑧𝑜𝑛𝑒 ; 𝑣(𝑥𝑝 , 𝑦𝑝 ) = 𝑁𝑖 (𝑥𝑝 , 𝑦𝑝 )×𝑉𝑖𝑧𝑜𝑛𝑒 .
element is used to solve the Laplacian equation (Dhatt et al., 2012). The 𝑖=1 𝑖=1
potential flow solution needs to be completed using the general form
of Bernoulli’s equation: (f) Save the nodal parameters to avoid unnecessary compu-
tations for the next mesh updates.
𝑉 2 (𝐱) 𝜕𝜑(𝐱, 𝑡) 𝑉∞2 (g) Update the fluid mesh nodal coordinates.
𝑝(𝐱, 𝑡) + 𝜌𝑔𝑧 + 𝜌 +𝜌 = 𝑝𝑎𝑡𝑚 + 𝜌𝑔𝐻 + 𝜌 (20)
2 𝜕𝑡 2
with 𝜌 = 1000 kg∕m3 the fluid mass density (water) and where the 3.2. Added mass estimation
𝜕𝜑
derivative for is deduced from the material derivative:
𝜕𝑡
The calculation of the added mass matrix [𝑀𝑎𝑑𝑑 ] is based on the
𝑑𝜑 𝜕𝜑
= + 𝐕𝑚𝑒𝑠ℎ ⋅ 𝐕. (21) assumptions that the fluid flow is inviscid and that convective effects
𝑑𝑡 𝜕𝑡
can be neglected with regard to the pressure gradient field (Brennen,
This may easily be computed from the moving node velocity 𝐕𝑚𝑒𝑠ℎ .
1982). This leads to the classical Poisson equation (Polyanin, 2002)
It is straightforward to calculate the material derivative using a finite
for the relative pressure 𝑝 on the domain , completed by boundary
difference scheme, and the precision order is seen to have a significative
conditions relating to the body []wall accelerations:
impact on the result (second order selected). As explained above, this
enables to connect the fluid velocity to the pressure and consequently 𝛥𝑝 = 0 ∀ 𝐱 ∈ , (22)
to ensure the coupling between the fluid and the structure.
with:
3.1.3. Mesh deformation technique: the Moving Submesh Approach (MSA) 𝜕𝑝 𝜕𝑝
At each time step, the fluid mesh is deformed to reflect the latest = 𝜌𝑓 𝐰̈ ⋅ 𝐧 on 𝑆ℎ𝑖𝑝, = 0 on 𝐵𝑜𝑡𝑚 and 𝑝 = 0 on 𝐼𝑛, 𝑂𝑢𝑡, 𝐹 𝑟𝑒𝑒𝑆.
𝜕𝑛 𝜕𝑛
ship position. The mesh adaptation is based on the Moving Submesh (23)
Approach (MSA) developed by the main author (Lefrançois, 2008). MSA
dramatically reduces the CPU time required for the mesh deformation The vector 𝐧 denotes the normal vector oriented external to the flow.
process. This technique considers a triangular coarse mesh (submesh) of The calculation of [𝑀𝑎𝑑𝑑 ] first requires the computation of the global

6
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 6. Illustration of the MSA technique step-by-step.

(or partial) set of 𝑁 eigenvectors of the structures denoted 𝐕𝑖 such that: and rotation modes of a ship (square topology) are shown in Fig. 7.
Solid lines and dashed lines represent respectively positive and negative
∑
𝑁 ∑
𝑁
such as 𝐕⊤ isovalues of the pressure.
𝐰= 𝑎𝑖 𝐕𝑖 , 𝑖 𝐕𝑖 = 1 and 𝑝 = 𝑎̈𝑖 𝑝𝑖 , (24)
𝑖=1 𝑖=1 Finally, the component 𝑀𝑎𝑑𝑑 (𝑖, 𝑗) of the modal pressure field 𝑝𝑗 is
deduced from the projection form of the pressure force 𝐅𝑝 on the modal
where 𝑝𝑖 is the modal component of the pressure field related to the
vector 𝐕𝑖 :
𝑖th eigenvector of the structure. This, together with Eq. (22), enable
the solution of 𝑁 Poisson equations: {𝑓𝑝 } = 𝑎̈𝑗 𝑝𝑗 𝐕𝑖 ⋅ 𝐧𝑑𝑠
∮𝑆ℎ𝑖𝑝
𝛥𝑝𝑖 = 0 ∀ 𝐱 ∈  with ∇𝑝𝑖 ⋅ 𝐧 = 𝜌𝑓 𝐕𝐢 ⋅ 𝐧 on 𝑆ℎ𝑖𝑝 and ∇𝑝𝑖 ⋅ 𝐧 = 0
to deduce:
on 𝐵𝑜𝑡𝑚. (25)
𝑚𝑎𝑑𝑑 (𝑖, 𝑗) = 𝑝𝑗 𝐕𝐢 ⋅ 𝐧𝑑𝑠 such that [𝑚𝑎𝑑𝑑 ] = [𝑋]𝑇 [𝑀𝑎𝑑𝑑 ][𝑋], (26)
The pressure 𝑝𝑖 being relative to the hydrostatic state, a zero pressure is ∮𝑆ℎ𝑖𝑝
imposed on the other boundaries. For illustrative purposes, the pressure where [𝑚𝑎𝑑𝑑 ] is the projection of the added mass matrix on the eigen-
field (isovalues) corresponding respectively to the horizontal, vertical vector base [𝑋] = [{𝑉1 } … {𝑉𝑁 }]. All calculations are here based

7
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 7. Pressure fields 𝑝𝑖 for three rigid body motions : 𝐿 = 26.66 m, 𝐻 = 20 m; 𝑇 = 13.34 m.

Fig. 8. Topology and dimensions.

on the same finite element approach used for solving the potential The equilibrium positions (ie 𝑤̇ = 𝑤̈ = 0) are the roots of the third-
velocity (Dhatt et al., 2012). order polynomial function. Since the original model does not take
non-stationary effects into account in the pressure calculation, the
3.3. Analytical prediction of stable and unstable ship positions added mass contribution 𝑚𝑎𝑑𝑑 has here been added to the equation.

For given fluid flow conditions, Alderf et al. (2010, 2011) and Alderf 3.4. Results
et al. (2015) showed that there exist three vertical positions for the
ship, which may be stable or unstable. Their analytical model will not The following sections provide a validation of the procedure for
be detailed here, but simply recalled and completed. It can be rewritten estimating the added mass, with reference to the literature and inves-
in the form of a polynomial fraction with a numerator of third order. tigations of dynamic squat concerning general flow cases.
It admits three roots 𝑤1 , 𝑤2 and 𝑤3 such that:
3.5. Sensitivity analysis of the added mass with geometric parameters
( ) 𝑑2𝑤  𝑤3 +  2 𝑤2 +  3 𝑤 +  4
𝑚 + 𝑚𝑎𝑑𝑑 = 1 ,
𝑑𝑡 2 2 We first look at the case where the length/draft ratio is equal to one.
(𝑤 − 𝑤1 )(𝑤 − 𝑤2 )(𝑤 − 𝑤3) To validate the added mass estimation in this particular case we take
= 𝜌𝑔𝑆 (27)
2 the results detailed in Wendel (1956)(page 64). Different UKC values
where: are included. With only one degree of freedom, the eigenvector is here
reduced to a scalar unity. The ship has a rectangular shape (Fig. 8) with
2
1 = −𝜌𝑆𝑔, 2 = −𝜌𝑆(2(𝐻 − 𝑇 )𝑔 − 𝑈∞ ∕2), a length 𝐿 = 20 m. The water depth is 𝐻 = 20 m and the In and Out
3 = −𝜌𝑆(𝑔(𝐻 − 𝑇 )2 − 𝑉∞2 (𝐻 − 𝑇 )), 4 = 𝜌𝑆𝑉∞2 ∕2((𝐻 − 𝑇 )2 − 𝐻 2 ) boundaries are given as 50 m from the bow and the stern of the ship.

8
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Table 4
Validation for added mass estimation for the particular case 𝑎∕𝑇 = 1.
𝐻∕𝑇 1.25 1.5 2.5 2.8 3.6 ∞
Inertia coefficient 𝑖 from Wendel (1956) 2 1.35 1 0.89 0.83 0.75
Inertia coefficient estimated Eq. (26) 1.93 1.34 0.9 0.87 0.81 0.75

Table 5 Table 6
Inertia coefficient predicted (Eq. (26)) for general cases. Geometric, physical parameters and initial conditions.
2 𝐻 𝐻∕𝑇 𝐿∕𝑇 𝑉∞ [m/s] 𝑤(0) ̇
𝑤(0) 𝑖 (Eq. (26)) 𝑚𝑎𝑑𝑑 ∕𝑚
𝐻∕𝑇 1.25 1.5 1.75 2 2.5 3 4 6 10
𝜋(𝑎∕𝑇 )
𝑎∕𝑇 ≤ 20 m 2 4 [0–4] −0.1 m 0 m/s 1.14 3.6

0.5 1.54 1.19 1.06 0.95 0.93 0.9 0.87 0.87 0.84 1.27
1 1.93 1.34 1.12 1.01 0.9 0.85 0.81 0.77 0.76 0.63
Table 7
1.5 2.35 1.53 1.23 1.08 0.93 0.85 0.78 0.72 0.69 0.42
Root values for different current velocities.
2 2.76 1.73 1.36 1.16 0.97 0.87 0.78 0.73 0.7 0.31
3 3.61 2.14 1.62 1.35 1.08 0.94 0.81 0.72 0.67 0.21 𝑉∞ [m/s] 0 1 2 2.25 2.5
4 4.45 2.56 1.89 1.55 1.21 1.03 0.86 0.73 0.66 0.16 𝜔1 (stable) [m] 0 −0.16 −0.75 −1.02 −1.4
5 5.3 2.98 2.17 1.76 1.34 1.12 0.91 0.75 0.66 0.12 𝜔2 (unstable) [m] −10 −8.46 −6.52 −5.9 −5.19

with low UKC: the longer the ship and the closer the canal bottom, the
higher the added mass effect.
The mass and added mass ratio can be deduced from the inertia
coefficient (Eq. (28)):
𝑚 2𝑇
= .
𝑚𝑎𝑑𝑑 𝜋𝑎𝑖
Ensuring a ratio greater than one to guarantee convergence for classical
coupling scheme may only be possible if:
2
𝑖 ≤ .
𝜋(𝑎∕𝑇 )
From the last column of Table 5, it is clear that this criterion is rarely
satisfied, which confirms that a corrected scheme should always be
used where there is a transient fluid flow effect.

3.6. Dynamic response for general fluid flow conditions

This section looks at the dynamic response of a ship with regard to

Fig. 9. Estimated inertia coefficient 𝑖 (𝐻∕𝑇 , 𝑎∕𝑇 ).
squat conditions. The geometric and physical parameters as illustrated
in Fig. 8 are summarized in Table 6.
Inlet and outlet boundaries are at a distance of 𝐿 from the bow and
The ship length is deduced from the 𝐻∕𝑇 ratio. A sensitivity analysis
1.5𝐿 from the stern of the ship. The finite element mesh is composed of
with the mesh size (not detailed here) was done, and the retained
3295 nodes for 6588 linear triangular elements. The mass ratio 𝑚𝑎𝑑𝑑 ∕𝑚
element size was 0.1 m near the ship, progressively increasing to 2 m
is greater than unity, with the corrected FSI coupling scheme being used
far from the ship. This gives an overall mesh size of approximately 7200
to guarantee convergence.
triangular elements for 3500 nodes. The location of the inlet and outlet
boundaries were also determined by a sensitivity analysis with regard The equilibrium positions 𝑤𝑖 are computed from Eq. (27) and shown
to their distances from the ship. in Fig. 10. The horizontal axis is the relative velocity 𝑉∞ , and the
Results are given in Wendel (1956) via an inertia coefficient 𝑖 as vertical axis the vertical position. The dashed line represents the static
follows: position (draft) and the three roots 𝑤1 (stable), 𝑤2 (unstable) and 𝑤3
(non-physical because lower than the canal bottom) are also shown.
𝑚𝑎𝑑𝑑 = 𝑖 𝜌𝜋𝑎2 with 𝑎 = 𝐿∕2. (28) The first two roots are complex and conjugate for a squat velocity
higher than 𝑉𝑠𝑞𝑢𝑎𝑡 = 2.9 m∕s. The first root corresponds to a stable equi-
The results are summarized in Table 4. There last one shows a very
good agreement between theoretical and predicted values. librium position, which is obtained from static calculation, whereas the
The table can then be completed with values for the parame- second unstable root is usually reached during an oscillatory response
ters 𝐻∕𝑇 and 𝑎∕𝑇 that are different from those considered in Wen- of the ship.
del (1956) and Brennen (1982). The estimated inertia coefficients Five different relative velocities were considered. Their respective
𝑖 (𝐻∕𝑇 , 𝑎∕𝑇 ) are given in Table 5 and plotted in Fig. 9. We point out root values (relative to the static draft position) are given in Table 7.
that the Eq. (26) slightly underestimates the added mass values ex- Results (other than for the no-flow case) are reported in Fig. 11(a)–
tracted from Wendel (1956), this may be due to the no flow assumption (d). The boxes on the left indicate the ship’s vertical motion 𝑤(𝑡), the
and the inlet/outlet boundaries that are not carried to infinity. two roots 𝜔1 and 𝜔2 are shown as colored dashed lines. The boxes
In Fig. 9, the thick red line with circle symbols relates to values on the right show the number of iterations required to satisfy the
referenced in the literature. It can be seen that the highest values of convergence criterion 𝜖 < 10−7 . The 𝑥-axis corresponds to the time
the inertia coefficient correspond to the lowest values of the 𝐻∕𝑇 ratio multiplied by the coupled frequency (see Eq. (5)).
and the highest values of the 𝑎∕𝑇 ratio. These cases concern long ships There is a number of points to note:

9
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 10. Velocity dependency for all three roots.

1. Obtaining coupling results for the non-stationary approach re- of generality, this study is voluntarily limited to a two-dimensional
quires the use of the added-mass-corrected FSI scheme. With no analysis, with no viscous effects and with only one degree of freedom
added mass correction, calculations invariably diverge. (vertical motion).
2. In the three first cases, all of which are stable, it is observed A theoretical analogy with the motion of a piston enclosed in a
that the ship oscillates around an average value higher than that U-tube was the basis for a mathematical analysis of the divergence
predicted by the stability model. This is because the prediction effect resulting from added mass when the fluid flow model is subject
model is based entirely on a 1D analysis, whereas the coupling to non-stationary effects: the conclusion of this work lies in proving
results derive from 2D calculations. The pressure profile exerted that the divergence systematically occurs in cases where the mass in
under the hull is not uniform and tends to diminish at the two motion is smaller than the added mass. A corrected scheme similar to
extremities, unlike in the prediction model (see Fig. 12). that proposed by the main author in Song et al. (2013), Lefrançois
3. The number of iterations required for convergence is clearly et al. (2016) and Lefrançois (2017) can systematically ensure the
dependent on the stability of the ship’s response. For the extreme convergence of the coupling scheme via a simple modification of the
unstable case obtained for 𝑉∞ = 2.5 m∕s, it is remarked that the two terms in the iterative form of the dynamics equation.
number of iterations grows exponentially, reflecting the squat For a numerical investigation of the phenomenon of ship squat,
phenomenon encountered by the ship. a fluid–structure model coupling a ship dynamics solver and a non-
4. The observed period is greater than the linear predicted period stationary potential fluid flow solver with added mass effect compen-
one. This is a consequence of the non-linear, non-stationary sation was presented in detail and successfully applied. A moving mesh
aspect of the pressure calculation in the 2D FEM model. technique based on the MSA approach (Lefrançois, 2008) was used
5. Irrespective of the stability, the energy conservation between to ensure kinematic compatibility between the rectangular ship’s hull
the fluid and the ship’s motion is strictly ensured, as shown in and the fluid boundary. It would appear from an initial study that
Fig. 13, where what is lost by the structure (ship) is gained by the added mass resulting from the ship’s motion and relative to its
dimensions generally exceeds the mass of the ship, which confirms the
the fluid flow, with:
need to a corrected scheme. Coupled calculations have been success-
𝑡
1 fully performed and their results have been compared to those given by
𝑐 (𝑡) = ̇ 2,
𝑚𝑤(𝑡) 𝑝 (𝑡) = 𝑚𝑔(𝑇 +𝑤(𝑡)), (𝑡) = − 𝑤(𝑡)𝑝(𝑠, 𝑡)d𝑠d𝑡
2 ∫0 ∫𝑆ℎ𝑖𝑝 squat stability models proposed in Alderf et al. (2010) and Alderf et al.
𝑐 and 𝑝 denote the kinetic and potential energies calculated by (2015), that is to say in studies where transient effects were neglected
the ship solver,  denotes the fluid exchange work calculated in the fluid flow solver. Finally, energy transfers have been analyzed
by the fluid solver, and 𝑜 is the initial energy level used to between the ship’s motion and the fluid flow, and the energy conserva-
normalize all curves. The solid and dashed lines correspond tion capability of the proposed fluid–structure coupling scheme could
respectively to the 𝑉∞ = 2 m∕s and 𝑉∞ = 2.5 m∕s cases. be confirmed with respect to the ship’s transient response.
The next steps, already begun, are the incorporation of a 3D analysis
that is easily extendable due to the fact that no strictly 2D restricted
4. Conclusion and prospects assumptions have been considered for both solvers, the added mass
estimation and the MSA. The aim is to study the effect of moving
This paper looks at the effect of added mass on the convergence from open sea configurations to restricted waterways and to improve
capabilities of fluid–structure coupling schemes in the numerical in- the behavior of a current navigation simulator for which ship dynamic
vestigation of ship squat. A strongly coupled scheme for FSI that response and maneuverability are important. The use of more realistic
significantly reduces the dependency of convergence on fluid density ship’s hull is making it possible to add new degrees of freedom such as
has been detailed. The proposed method offers the advantage of a direct roll and pitch angles.
physical interpretation related to the added mass effects and for a given At least, this coupling model is however dependent on the estima-
limit, its efficiency is proportional to the fluid density. Without loss tion of the mass-added effects matrix currently calculated under the

10
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 11. Dynamic ship’s response function of the current velocity.

Fig. 12. Pressure distribution under hull at 𝑉∞ = 1 m∕s.

assumption of negligible convective effects. A track currently being estimated, integrating thus the time evolution as a function of the
analyzed is to enhance at each time step an added mass matrix initially geometry of the ship and the flow conditions.

11
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

Fig. 13. Energy conservation: solid line [𝑉∞ = 2 m∕s], dashed line [𝑉∞ = 2.5 m∕s].

CRediT authorship contribution statement Galieriková, A., Sosedová, J., 2016. Environmental aspects of transport in the context
of development of inland navigation. Ekológia (Bratislava) 35 (3), 279–288.
Emmanuel Lefrançois: Conceptualization, Methodology, Software, Gourlay, T., 2008. Slender-body methods for predicting ship squat. Ocean Eng. 35 (2),
191–200.
Writing – original draft. Bo Yang: Writing – original draft. Sami Kaidi:
Härting, A., Laupichler, A., Reinking, J., 2009. Considerations on the squat of unevenly
Writing – original draft. Nisrine Mohamad: Validation, Writing – trimmed ships. Ocean Eng. 36 (2), 193–201.
original draft. Hughes, T.J.R., Liu, W.K., Zimmermann, T.K., 1981. Lagrangian-eulerian finite element
formulation for incompressible viscous flows. Comput. Methods Appl. Mech. Engrg.
Declaration of competing interest 29, 329–349.
Idelsohn, S.R., Del Pin, F., Rossi, R., Oñate, E., 2009. Fluid–structure interaction
problems with strong added-mass effect. Internat. J. Numer. Methods Engrg. 80
The authors declare that they have no known competing finan- (10), 1261–1294.
cial interests or personal relationships that could have appeared to Jachowski, J., 2008. Assessment of ship squat in shallow water using CFD. Arch. Civ.
influence the work reported in this paper. Mech. Eng. 8 (1), 27–36.
Kassiotis, C., Ibrahimbegovic, A., Matthies, H., 2010. Partitioned solution to fluid–
Data availability structure interaction problem in application to free-surface flow. Eur. J. Mech. B:
Fluids 29 (6), 510–521.
Kok, Z., Duffy, J., Chai, S., Jin, Y., 2020. Multiple approaches to numerical modeling
No data was used for the research described in the article. of container ship squat in confined water. J. Waterw. Port Coast. Ocean Eng. 146
(4).
References Küttler, U., Wall, W., 2008. Fixed-point fluid–structure interaction solvers with dynamic
relaxation. Comput. Mech. 43 (1), 61–72.
Alderf, N., Lefrançois, E., Sergent, Ph., 2015. Virtual bottom for ships sailing in Lataire, E., Vantorre, M., Delefortrie, G., 2012. A prediction method for squat in
restricted waterways (unsteady squat). J. Ocean Eng. 110, 205–214. http://dx.doi. restricted and unrestricted rectangular fairways. Ocean Eng. 55, 71–80.
org/10.1016/j. Lefrançois, E., 2008. A simple mesh deformation technique for fluid–structure in-
Alderf, N., Lefrançois, E., Sergent, Ph., Debaillon, P., 2010. Transition effects on ship teraction based on a submesh approach. Int. J. Numer. Methods Eng. 75,
sinkage in highly restricted waterways. J. Eng. Marit. Environ. 224 (2), 141–153. 1085–1101.
Alderf, N., Lefrançois, E., Sergent, Ph., Debaillon, P., 2011. Dynamic ship response Lefrançois, E., 2017. How an added mass matrix estimation may dramatically improve
integration for numerical prediction of squat in restricted waterways. Int. J. Numer. FSI calculations for moving airfoils. Appl. Math. Model. 2017 (51), 655–668.
Methods Fluids 65 (7), 743–763. http://dx.doi.org/10.1002/fld.2194. Lefrançois, E., Brandely, A., Mottelet, S., 2016. Strongly coupling partitioned scheme
Badia, S., Nobile, F., Vergara, C., 2008. Fluid–structure partitioned procedures based for enhanced added mass computation in 2D fluid–structure interaction. Coupled
on robin transmission conditions. J. Comput. Phys. 227, 7027–7051. Syst. Mech. Int. J. 5 (3), 235–254.
Barrass, C.B.D., 2004. Ship squat in open water and in confined channels. Ship Des. Mazaheri, A., Montewka, J., P., Kujala, 2014. Modeling the risk of ship grounding-
Perform. Masters Mates 14, 8–163. a literature review from a risk management perspective. WMU J. Marit. Aff. 13,
Brennen, C.E., 1982. A Review of Added Mass and Fluid Inertial Forces. Department 269–297.
of the Navy, Port Hueneme, CA. McTaggart, K., 2018. Ship squat prediction using a potential flow Rankine source
Causin, P., Gerbeau, J., Nobile, F., 2005. Added-mass effect in the design of partitioned method. Ocean Eng. 148, 234–246.
algorithms for fluid–structure problems. Comput. Methods Appl. Mech. Engrg. 194 Michler, C., van Brummelen, E.H., de Borst, R., 2005. An interface Newton-Krylov solver
(42–44), 4506–4527. for fluid–structure interaction. Internat. J. Numer. Methods Fluids 47, 1189–1195.
CIAIM, 2016. Grounding of the Vessel TIDE NAVIGATOR at the Port of Vilanova I la
Pedersen, P.T., 2010. Review and application of ship collision and grounding analysis
GeltrÚ. CIAIM-08/2017 REPORT.
procedures. Mar. Struct..
CIAIM, 2018. Marine Inquiry MO-2018-203. Grounding of container ship Leda Maersk
Piperno, S., Farhat, C., Larrouturou, B., 1995. Partitioned procedures for the transient
Otago Lower Harbour.
solution of coupled aeroelastic problems, part I: Model problem, theory and
Connell, B.S.H., Yue, D.K.P., 2007. Flapping dynamics of a flag in a uniform stream.
two-dimensional application. Comput. Methods Appl. Mech. Engrg. 124, 79–112.
J. Fluid Mech. 581, 33–67.
Polyanin, A.D., 2002. Handbook of Linear Partial Differential Equations for Engineers
Debaillon, P., 2010. Numerical investigation to predict ship squat. J. Ship Res. 54
and Scientists. Chapman Hall-CRC Press.
((June) N.2), 133–140.
Dhatt, G., Touzot, G., Lefrançois, E., 2012. Finite Element Method. Wiley-ISTE. Radmilović, Z.V.M., 2011. Role of danube inland navigation in Europe. Int. J. Traffic
Elsherbiny, K., Tezdogan, T., Kotb, M., Incecik, A., Day, S., 2019. Experimental analysis Transp. Eng..
of the squat of ships advancing through the New Suez Canal. Ocean Eng. 178, Song, M., Lefrançois, E., Rachik, M., 2013. Development of a partitioned algorithm for
331–344. fluid–structure coupling with no fluid density dependency. Comput. Fluids 190–202.
Felippa, C.A., Park, K.C., Farhat, C., 2001. Partitioned analysis of coupled mechanical http://dx.doi.org/10.1016/j.compfluid.2013.05.022.
systems. Comput. Methods Appl. Mech. Engrg. 190 (24–25), 3247–3270. Tallec, P.L., Gerbeau, J.F., Hauret, P., Vidrascu, M., 2005. Fluid structure interaction
Fernández, M.A., Gerbeau, J.-F., Grandmont, C., 2005. A projection semi-implicit problems in large deformation. C. R. Mec. 333 (12), 910–922.
scheme for the coupling of an elastic structure with an incompressible fluid. Tezduyar, T.E., 2004. Finite element methods for fluid dynamics with moving bound-
Internat. J. Numer. Methods Engrg. 69 (4), 794–821. aries and interfaces. In: Stein, E., de Borst, R., Hughes, T.J.R. (Eds.), Encyclopedia
Fernández, M., Gerbeau, J.F., Grandmont, C., 2006. A projection algorithm for fluid– of Computational Mechanics, Vol. 3. John Wiley and Sons.
structure interaction problems with strong added-mass effect. C. R. Acad. Sci. Paris Tezduyar, T.E., Sathe, S., Keedy, R., Stein, K., 2006. Space–time finite element
342 (4), 279–284. techniques for computation of fluid–structure interactions. Comput. Methods Appl.
Förster, C., Wall, W., Ramm, E., 2007. Artificial added mass instabilities in sequential Mech. Engrg. 195, 2002–2027.
staggered coupling of nonlinear structures and incompressible viscous flows. van Brummelen, E.H., 2009. Added mass effects of compressible and incompressible
Comput. Methods Appl. Mech. Engrg. 196 (7), 1278–1293. flows in fluid–structure interaction. J. Appl. Mech. 76 (2), 021206-7.

12
E. Lefrançois et al. Ocean Engineering 267 (2023) 113280

van Brummelen, E.H., 2011. Partitioned iterative solution methods for fluid–structure Zhang, S., Pedersen, P.T., Villavicencio, R., 2019. Probability and mechanics of ship
interaction. Internat. J. Numer. Methods Fluids 65, 3–27. collision and grounding.
Wendel, K., 1956. Hydrodynamic Masses and Hydrodynamic Moments of Inertia. Navy Zipfel, B., Lehmann, E., 2010. Evaluation of critical grounding incidents. In: 5th
Department Report, Jahrb. d. STG, Volume 44. International Conference on Collision and Grounding of Ships. pp. 97–103.
Xu, J., Xuan, G., Li, Y., Li, Z., Hu, Y., Jin, Y., Huang, Y., 2016. Study on the squat of
extra- large scale ship in the three gorges ship lock. Ocean Eng. 123, 65–74.

13