Journal of Sound and Vibration 333 (2014) 5004–5017

Contents lists available at ScienceDirect

Journal of Sound and Vibration

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

Added mass matrix estimation of beams partially immersed

in water using measured dynamic responses
Fushun Liu a,n, Huajun Li a, Hongde Qin b, Bingchen Liang a
a
Shandong Provincial Key Lab of Ocean Engineering, Ocean University of China, Qingdao 266100, China
b
College of Shipbuilding Engineering, Harbin Engineering University, Harbin 150001, China

a r t i c l e in f o abstract

Article history: An added mass matrix estimation method for beams partially immersed in water is proposed
Received 4 March 2014 that employs dynamic responses, which are measured when the structure is in water and in air.
Received in revised form Discrepancies such as mass and stiffness matrices between the finite element model (FEM) and
29 April 2014
real structure could be separated from the added mass of water by a series of correction factors,
Accepted 17 May 2014
which means that the mass and stiffness of the FEM and the added mass of water could be
Handling Editor: L.G. Tham
Available online 13 June 2014 estimated simultaneously. Compared with traditional methods, the estimated added mass
correction factors of our approach will not be limited to be constant when FEM or the
environment of the structure changed, meaning that the proposed method could reflect the
influence of changes such as water depth, current, and so on. The greatest improvement is that
the proposed method could estimate added mass of water without involving any water-related
assumptions because all water influences are reflected in measured dynamic responses of the
structure in water. A five degrees-of-freedom (dofs) mass-spring system is used to study the
performance of the proposed scheme. The numerical results indicate that mass, stiffness, and
added mass correction factors could be estimated accurately when noise-free measurements
are used. Even when the first two modes are measured under the 5 percent corruption level,
the added mass could be estimated properly. A steel cantilever beam with a rectangular section
in a water tank at Ocean University of China was also employed to study the added mass
influence on modal parameter identification and to investigate the performance of the
proposed method. The experimental results demonstrated that the first two modal frequencies
and mode shapes of the updated model match well with the measured values by combining
the estimated added mass in the initial FEM.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

A slender beam is considered to be a simple yet very frequently used engineering structure. The frequency of using the
beam “in air” is usually higher than “in water”, mainly due to the influence of added mass and the “contact with water”. One
can also conclude that the added mass will influence many other factors, such as area coefficients, beam/draft ratio,
boundary conditions of a water tank (or sea), slenderness ratio, environmental conditions, etc. For offshore structures,
environmental conditions such as waves and currents change all the time, which implies that the added mass will not be
constant throughout a structure's service life.

n
Corresponding author.

http://dx.doi.org/10.1016/j.jsv.2014.05.036
0022-460X/& 2014 Elsevier Ltd. All rights reserved.
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5005

When submerged in a fluid, the dynamic response of a solid body is altered by the effect of the added mass of the fluid
(Ma ). Consequently, the ratio between the natural frequency of a given mode of vibration in water (f w ) and in air (f v ) can be
approximated as follows:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
f w;i 1

(1)
f v;i ð1 þ ðMa =mi ÞÞ

where i denotes each particular mode shape and m is the corresponding modal mass.
From Eq. (1), an added mass coefficient, C m , can be defined with:
 2
f v;i
Cm ¼ 1 (2)
f w;i

One problem in Eq. (2) is that all parts of the structure in fluid will be assumed to have the same value C m once Eq. (2) is
solved, which opposes practical situations. Most of the existing relevant works have been obtained from experiments in which
the most popular technique is evaluation of the added mass at the resonant frequency corresponding to the peak of a frequency
response curve obtained from the 'forced' vibration analysis. The semi-circular cylinder is often adopted to simplify an object's
cross-section in the analysis of added mass. The added masses of circular cylinders and cylinders with other cross-sectional
shapes in deep water are available (Newman [1]). De Tarso et al. [2] investigated the added mass and damping of rectangular
cylinders mounted at the seabed and presented the shallow water effect on added mass and damping. Clarke [3] used conformal
mapping and calculated the added mass of a circular cylinder in shallow water. He demonstrated the effect of water depth
through a comparison of results from different methods based on conformal mapping techniques and concluded that the
approach of using a row of distributed dipoles gave the best accuracy. Clarke [4] also provided the added mass for the complex
case of an elliptical cylinder in shallow water using a general mapping technique based on the Schwartz–Christoffel method.
Clarke [5] used a similar method to calculate the added mass of an elliptical cylinder with a vertical fin stabilizer in shallow
water. Lin and Liao [6] applied the fast multiple boundary element method (FMBEM) to calculate the added mass coefficients of
complicated three-dimensional (3D) underwater bodies calculated by the FMBEM. The FMBEM is much more computationally
efficient than the traditional boundary element method. Therefore, the FMBEM provides an effective numerical method to
predict added mass coefficients of complicated underwater bodies.
Recently, Benaouicha et al. [7] addressed a theoretical study of the added mass effect in cavitating flow. The cavitation is
considered to induce a strong time–space variation of the fluid density at the interface between an inviscid fluid and a three
degrees-of-freedom rigid section. The added mass coefficients decrease as the cavitation increases, which should induce an
increase of the natural structural frequencies. Torre et al. [8] used a non-intrusive excitation and measurement system based
on piezoelectric patches mounted on the hydrofoil surface to determine the natural frequencies of the fluid–structure
system. Kramer et al. [9] investigated the effects of material anisotropy and added mass on the free vibration response of
rectangular, cantilevered composite plates/beams via combined analytical and numerical modeling. The results show that
the natural frequencies of the composite plates are 50–70 percent lower in water than in air due to large added mass effects.
Torre et al. [10] studied the influence of the boundary conditions on the added mass of a NACA0009 cantilever hydrofoil,
including experiments. A detailed fluid–structure model has been built for both cases, and a modal analysis has been carried
out. The obtained results are in reasonably good agreement with experimental data.
In practice, most added mass determination techniques involve solving Laplace's equation, which governs the induced
water field; in other words, water-related assumptions are involved. In this paper, we try to estimate the added mass of
beam structures from the view of structure (i.e., using only dynamic responses of the structure in water and in air). A five
degrees-of-freedom (dofs) mass-spring system will be used to identify the performance of the proposed scheme, and a steel
cantilever beam with a rectangular section in a water tank will be employed to demonstrate the approach.

2. Transverse vibration of a cantilever beam

Overlooking shear, damping, and axial-force effects, the solution for free bending transverse vibration of a beam is
obtained by solving the differential equation of motion of a Bernoulli–Euler beam, which can be written as follows [11]:
  2 
∂2 v ∂2 ∂ v
m 2 þ 2 EI ¼0 (3)
∂t ∂x ∂x2
where m is the mass distribution of the unit of length, EI is the flexural rigidity, and v is the transverse displacement, which
is a function of the spatial coordinate (x) and time (t).
The solution of Eq. (3) is assumed in the form of a product of two functions:
vðx; tÞ ¼ wðxÞYðtÞ (4)
Substituting Eq. (4) into Eq. (3) results in
 2 
∂2 YðtÞ ∂2 ∂ wðxÞ
wðxÞm þ 2 EI YðtÞ ¼ 0 (5)
∂t 2
∂t ∂x2
5006 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

Then, the nth circular or natural frequency (f ) of the considered uniform beam is obtained in the form of
rffiffiffiffiffi
χ n π 2 EI
f¼ 2 (6)
L m
where χ n are constants that depend on actual boundary conditions and L is the length of the beam.

3. Added mass estimation using dynamic responses

The mass matrix of a real structure in water is expressed as

~ ¼ Mþ ΔMf þΔMv þ ΔMf
M (7)
as as a

where M is the mass matrix from the finite element model (FEM). and ΔMfas ΔMvas
represent the mass discrepancy of the
portion of the structure in water and in air, respectively. ΔMfa represents the added mass due to the presence of water.
Likewise, the stiffness matrix of the real structure in water could be written by
~ ¼ K þ ΔKf þΔKv
K (8)
as as

where K is the stiffness matrix from FEM. ΔKfas

and ΔKvas denote the stiffness discrepancy of the segment of the structure in
water and in air, respectively.
The main purpose of this paper is to the estimate inevitable modeling errors from FEM ΔMfas , ΔMvas , ΔKfas and ΔKvas , and
the added mass due to the presence of water ΔMfa simultaneously.
We assume that the added mass is a modification of the related mass matrix Mfasf via:
Nfasf
ΔMfas ¼ ∑ εfasf Mfasf (9)
asf ¼ 1

where N fasf is the number of elements in water and εfasf is the corresponding correction factors.
Similarly:
Nvas
ΔMvas ¼ ∑ εvasv Mvasv (10)
asv ¼ 1

Nfksf
ΔKfas ¼ ∑ εfksf Kfksf (11)
ksf ¼ 1

N vksf
ΔKvas ¼ ∑ εvksv Kvksv (12)
ksv ¼ 1

Nfaf
ΔMfa ¼ ∑ εfaf Mfaf (13)
af ¼ 1

where Mfasf , Mvasv are the asf th and asvth mass matrices from FEM in global coordinate; and Kfksf , Kvksv are the ksf th and ksvth
stiffness matrices from FEM in global coordinate; and Mfaf is the af th mass matrix from FEM in global coordinate.
Substituting Eqs. (9)–(13) into Eqs. (7) and (8) yields:
N fasf N vas N faf
~ ¼ Mþ
M ∑ εfasf Mfasf þ ∑ εvasv Mvasv þ ∑ εfaf Mfaf (14)
asf ¼ 1 asv ¼ 1 af ¼ 1

N fksf Nvksf
~ ¼ Kþ
K ∑ εfksf Kfksf þ ∑ εvksv Kvksv (15)
ksf ¼ 1 ksv ¼ 1

If modal parameters of the structure without consideration for the added mass of water could be obtained as well as the
jth eigenvalues and eigenvectors denoted as ! !
λ j and Φ j , then their relationship can be expressed as
0 f v
1 0 f
1
N ksf N ksf N asf N vas
@K þ ∑ εfksf Kfksf þ ∑ εvksv Kvksv A!
Φ j  λ j @M þ
!
∑ εfasf Mfasf þ ∑ εvasv Mvasv A!
Φj ¼ 0 (16)
ksf ¼ 1 ksv ¼ 1 asf ¼ 1 asv ¼ 1

Obtaining the pth eigenvalues and eigenvectors of the structure in water, i.e., λ~ p and Φ
~ p , then:
0 f v
1 0 f f
1
N ksf Nksf Nasf Nvas N af
@K þ ∑ εfksf Kfksf þ ~ p  λ~ p @M þ
∑ εvksv Kvksv AΦ ∑ εfasf Mfasf þ ∑ εvasv Mvasv þ ∑ εfaf Mfaf AΦ
~ p ¼0 (17)
ksf ¼ 1 ksv ¼ 1 asf ¼ 1 asv ¼ 1 af ¼ 1
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5007

Pre-multiplying Eqs. (16) and (17) by ΦTi , respectively, yields:

0 f
1 0 1
N ksf v
Nksf Nfasf Nvas
ΦTi @K þ ∑ εfksf Kfksf þ ∑ εvksv Kvksv A!
Φ j  λ j Φ i @M þ
! T
∑ εfasf Mfasf þ ∑ εvasv Mvasv A!
Φj ¼0 (18)
ksf ¼ 1 ksv ¼ 1 asf ¼ 1 asv ¼ 1

and
0 1
N fksf N vksf
ΦTi @K þ ∑ εfksf Kfksf þ ∑ εvksv Kvksv AΦ
~p
ksf ¼ 1 ksv ¼ 1

0 1
Nfasf Nvas N faf
 λ~ p ΦTi @M þ ∑ εfasf Mfasf þ ∑ εvasv Mvasv þ ∑ εfaf Mfaf AΦ
~ p ¼0 (19)
asf ¼ 1 asv ¼ 1 af ¼ 1

Combining Eqs. (18) and (19), one obtains:

Nfasf N vas N fksf N vksv N faf
∑ εfasf Aijp þ ∑ εvasv Bijp þ ∑ εfksf C ijp þ ∑ εvksv Dijp þ ∑ εfaf Eip ¼ ðF ij þ F ip Þ (20)
asf ¼ 1 asv ¼ 1 ksf ¼ 1 ksv ¼ 1 af ¼ 1

where

Aijp ¼ ΦTi Mfasf ð! ! ~ pΦ

λ jΦj þλ ~ pÞ (21)

Bijp ¼  ΦTi Mvasv ð! ! ~ pΦ

λ jΦj þλ ~ pÞ (22)

C ijp ¼ ΦTi Kfksf ð! ~ pÞ

Φj þΦ (23)

Dijp ¼ ΦTi Kvksv ð! ~ pÞ

Φj þΦ (24)

Eip ¼  λ~ p ΦTi Mfaf Φ

~p (25)

!
F ij ¼ ΦTi K! T !
Φ j  λ j Φi M Φ j (26)

~ p  λ~ p ΦT MΦ
F ip ¼ ΦTi KΦ ~p (27)
i

Eq. (20) can also be rewritten as

8 f 9
>
> εasf >>
>
> >
>
>
> ε v > >
> asv >
h i< f >
> =
Aijp Bijp C ijp Dijp Eip εksf ¼  ðF ij þF ip Þ (28)
>
> >
>
> εvksv >
> >
>
> >
>
>
> >
: εf > ;
af

For clarity, Eq. (28) can be rewritten in a matrix form:

ZΔ ¼ F (29)

where
h i
Z ¼ Aijp Bijp C ijp Dijp Eip (30)

and
8 f 9
>
> εasf >>
>
> >
>
>
> εv > >
>
> asv >
< f > =
Δ ¼ εksf (31)
>
> >
>
>
> εvksv >
>
>
> >
>
>
> >
:ε >f ;
af

A standard inverse operation could be used to solve Eq. (29). Detailed information has been discussed in reference (Li
[12]).
5008 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

From Eqs. (13) and (29), one can obtain the added mass distribution for each element in water. For example, for the nth
element, i.e., af ¼ n in Eq. (13), the added mass could be expressed as

ΔMna ¼ εfn Mfn (32)

The global added mass matrix can then be obtained:
N faf
ΔMa ¼ ∑ εfn Mfn (33)
n¼1

In general, the proposed method includes four sequential steps:

(1) Step 1: Measuring and obtaining the jth eigenvalues and eigenvectors ! !
λ j and Φ j , of the structure without consideration
for the added mass of water, and the pth eigenvalues and eigenvectors of the structure in water, λ~ p and Φ~ p , respectively.
(2) Step 2: One determines the mode shapes Φi of the FEM of the measured structure and computing Eqs. (21)–(27), then
solve Eq. (31) by substituting Eqs. (21)–(27) into Eq. (29).
(3) Step 3: Computing Eq. (13) in terms of Eq. (31), and then substituting it into Eq. (33), one can obtain the added mass
ΔMa .
(4) Step 4: Computing Eqs. (9)–(12) in terms of Eq. (31), and then substituting Eqs. (9)–(12) into Eqs. (14) and (15), one can
get the updated FEM and make a comparison with measured modal parameters.

4. Numerical study: a 5-dof mass-spring system

To illustrate the procedure and demonstrate the performance of the proposed scheme, a 5-dof mass-spring system was
used to represent a lumped mass beam fixed at the ground and some portion of it in water, as shown in Fig. 1. The uniform
mass and stiffness coefficients were taken to be Mn ¼ 50 kg and Kn ¼ 2:9  107 N/m, n ¼ 1; 2; 3; 4; 5, respectively. The
coordinates of the 5-dof model are denoted by xn , with x1 at the fixed end and x5 at the free end. Mass 1, mass 2, and mass 3
are assumed to be in water; mass 4 and mass 5 are assumed to be in air.
Considering the inevitable modeling errors from FEM, a series of quantities as defined from Eq. (9) to Eq. (13) are
employed to calculate these errors. Here, we randomly generate a series of numbers using the Matlab function, assuming
   
εfasf ¼  0:0087 0:0333 0:0025 and εvasv ¼ 0:0058  0:0230 . Thus, the second mass modeling error is
2 3
0 0 0 0 0
6 0 50 0 0 0 7
6 7
6 7
 0:0333  M2 ¼ 6 6 0 0 0 0 0 7
7 (34)
6 7
40 0 0 0 05
0 0 0 0 0
f    
Likewise, εksf ¼ 0:0119 0:0119 0:0004 and εksv ¼ 0:0033 0:0017 are used to represent stiffness modeling
v

errors for elements in water and in air, respectively. For simulating added mass of water, we assume
εfaf ¼ 0:15 0:25 0:20 . The purpose of the following work is to study whether the added mass caused by water could
be estimated properly, including modeling errors and using measured modal parameters. By implementing Eigen analysis,
one can obtain the frequencies of the FEM: 34.5, 100.7, 158.75, 203.93, and 232.60 Hz. The frequencies of the model, without
taking into account the added mass of water, are 35.006, 102.98, 160.12, 206.5, and 235.66 Hz. When the added mass is
considered, the frequencies changed to 32.687, 92.635, 145.96, 190.00, and 212.77, respectively.
From Eq. (28) and Fig. 1, one can see that there are five mass, five stiffness, and three added mass coefficients that should
be estimated, which means that at least 13 equations are needed to solve Eq. (28). Therefore, we assume two modes from
the model that do not consider the added mass of water and two that do, and all analytical modes from FEM will be used.
As such, 20 equations could be constructed, which are sufficient to solve the above 13 coefficients. However, though 20
equations were constructed, the rank of coefficient matrix Z in Eq. (29) was 12, which means that the 13 unknowns could
not be solved uniquely. Therefore, we can assume that mass 4 could be modeled precisely and take it as a constraint. Fig. 2
shows the comparison of the estimated and preset coefficients plotted against the number of unknowns. Fig. 2 clearly
indicates that the added mass of water could be estimated accurately, including modeling errors, when the measured modes
from dynamic testing are noise free.
Water depth may have an influence on the added mass of a structure in water, and that added mass on the real structure
will be variable with the change in environmental conditions such as water depth, marine growth, and so on. Thus, we will
study whether our approach could properly estimate the added mass when environmental conditions change. Here, we take
the change in water depth as an example, assuming that masses 1 and 2 are in water and that masses 3 through 5 are in air,
which means that 12 unknowns should be calculated in this scenario. Based on the same considerations that yielded Fig. 2,
one can obtain a comparison of the estimated and preset coefficients plotted against the number of unknowns as shown in
Fig. 3. The numerical results demonstrated that our approach could identify and estimate added masses when operational
conditions of the structure changed.
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5009

Fig. 1. A 5-dof mass-spring system: three in water and the other two in air.

Another problem to note is that measured modes usually include noise; thus, the remaining numerical study focused on
implementing the proposed method with corrupted modes. Values of corrupted modes were generated by multiplying the
true value by a factor ð1 þNlÞ, where Nl is called a corruption level. Fig. 4 shows a comparison of the estimated and preset
coefficients when the first two modes were measured under a 5 percent corruption level. One can conclude from Fig. 4 that
the added mass could be properly estimated with the exception of the estimation at mass 1, which had relatively larger
errors. One can also conclude that many errors will be introduced as the corruption level increases; thus, the following is to
investigate the performance of added mass estimation with increasing corruption levels.
Here, we define the relative difference between the estimated (ε af ) and corresponding preset (εaf ) values as an indicator,
i.e.:

jε af  εaf j
Rm ¼  100% (35)
εaf

Fig. 5 shows the indicator Rm with a corruption level increase from zero to 10 percent. Numerical results indicate that the
approach could estimate added mass properly when the corruption level is below 5 percent, as shown in Fig. 4 and the
average line in Fig. 5. As predicted, estimation errors of added masses will increase when the noise becomes more
5010 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

Fig. 2. Comparison of estimated coefficients and preset values when mass 1 to mass 3 is in water.

Fig. 3. Comparison of estimated coefficients and preset values when mass 1 to mass 2 is in water.

significant, and the estimation of added mass at mass 1 is the most sensitive to noise compared with masses 2 and 3. All of
these factors imply that accurate modal parameters are expected for the approach.

5. Experimental study of a steel cantilever beam in still water

A steel cantilever beam with a rectangular section in a water tank at Ocean University of China was used to study the
added mass influence on modal parameter identification and to investigate the performance of the proposed method
employing measured dynamic signals. The experimental setup in which the cantilever beam was fixed on the ground by
four bolts is shown in Fig. 6. The section area of the steel beam is 8:8  10  4 m2, with a height of 1.9 m. Ten accelerometers
(Model 4803A-0002), termed Sensors 1 to 10, were mounted for response measurement and were equally distributed from
the free end to the fixed end of the beam. A measurement system (PL64-DCB8, Integrated Measurement & Control
Cooperation, Germany) was used for data acquisition.
In this experiment, an impulsive load was used to excite the beam, whether it was in water or not. The water depth was set to
be 1.5 m, and a sampling rate of 500 Hz was not changed throughout the whole experiment. Fig. 7 is a comparison of the measured
accelerations of Sensor 1 in the time domain. By employing the Eigensystem Realization Algorithm (ERA) [13] for modal parameter
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5011

Fig. 4. Comparison of estimated coefficients and preset ones when the first two modes were measured under a 5% corruption level.

Fig. 5. The indicator Rm changes with corruption level increases from zero to 10%.

identification, the first two frequencies and damping ratios extracted from measured signals in water and in air, respectively,
including the corresponding values from FEM, are listed in Table 1. Following the standard formulation for a uniform beam element
in a plane considering axial coordinates, one computes the element stiffness matrix kn and element (consistent) mass matrix mn of
a size 6  6. There are 10 nodal points and each nodal point has 3 DOFs, thus matrices K and M both are with a size 30  30. Node
and element number could be found in Fig. 8 (Model A). Table 1 clearly demonstrates that the frequencies decreased significantly,
while the damping ratios improved because of the influence of surrounded water.
In this experiment, two challenges should be properly handled: one is the spatial incompleteness of measured mode
shapes, and the other is the supplementation of added constraints to assure the unique solutions of the approach. To address
the first challenge, the direct mode shape expansion method (Liu[14,15]) was used to improve the accuracy of mode shape
expansion. Fig. 5 shows that noises that come from dealing with spatial incompleteness, environment, equipment, etc., will
influence the performance of our approach, so we firstly study the case that only added mass correction factors εfaf were
estimated. Using the first two modes measured when the beam was in water and in air and all modes from the FEM of the
beam, one can obtain correction factors εfaf as shown in Fig. 9, which indicates that the added mass correction factors were
not constants for each element of the beam in water. One should also note that these correction factors were estimated by
taking the FEM as a baseline model, which implies that these factors will change if the baseline model is changed.
Though Fig. 9 provides a series of added mass correction factors on each element of the cantilever beam in water, one
may suspect these factors' correctness. One way to demonstrate their accuracy is by studying modal parameters of the
updated FEM, which takes into account the added mass of the water. Table 2 lists the frequencies of the FEM, the beam in
5012 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

Fig. 6. A steel cantilever beam with rectangle section in a water tank of OUC.

Fig. 7. Measured accelerations of Sensor 1 in time domain: (a) in water, and (b) in air.

Table 1
The first two frequencies and damping ratios from FEM when the cantilever beam is in water and in air, respectively.

Frequency (Hz) Damping ratio

FEM In air In water FEM In air In water

1 7.82 7.26 7.07 0 0.0018 0.0075

2 49.03 49.55 45.51 0 0.0053 0.0066

water, and the updated FEM (Model A). The corresponding first two mode shapes are shown in Figs. 10 and 11, respectively.
Results demonstrated that the first two frequencies of the updated FEM considering added mass matched well to measured
values, and the first two mode shapes of the updated FEM were improved.
As discussed above, each estimated added mass correction factor is corresponding to a predetermined element in water;
one may conclude that the estimation of added mass maybe different when the immersed part of the beam was modeled
using different number of elements. Here we will further study how the estimated added mass correction factors change
with element numbers increasing. In this experiment, four scenarios were studied, i.e., the immersed part of the beam was
discrete into 1, 2, 4, and 6 elements, and the corresponding FEM were named Model E, D, C, and B, respectively, as shown in
Fig. 12. To compare results of above different scenarios, a sensitivity indicator (Sa) was defined,
Nf f
∑n asf
¼ 1 εasf ;n
Sa ¼ (36)
Nfasf
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5013

Fig. 8. Finite element model when the immersed part modeled using 8 elements: Model A.

When the immersed part was modeled just using one element, as shown in Fig. 12(d), this FEM has 9 dofs, i.e., 3 master
dofs and 6 slave dofs. The first two modes were employed in implementing the proposed method, then one can get the
added mass correction factor to be 0.72, which means the added mass for each element in water in Fig. 9 is a constant, as
shown in Fig. 13(a). By substituting this correction factor back into the FEM, the updated frequencies were 6.90 and 47.83 Hz.
Results indicated there were much errors for the second mode if only the immerse part had been modeled using one
element and regarded as a baseline for added mass estimation. Likewise, added mass correction factors from other scenarios
were plotted in Fig. 13(a), and the sensitivity indicator was plotted in Fig. 13(b). Table 2 lists comparisons of frequencies from
the FEM, the measured values when the beam in water, and the updated FEM (Models A–E), respectively. From Fig. 13, one
can find that the sensitive indicator becomes stable when the immersed part is modeled more than 6 elements.
If substituting the obtained added mass correction factors into the FEM (Model C) and recalculate the first two modal
parameters, one can get the two frequencies 7.0473 and 44.0753 Hz, which also demonstrated the correctness of Fig. 9 from
another view point.
In this experiment, we also tried to simultaneously estimate discrepancies of mass and stiffness from FEM along with
added mass. Based on the same considerations yielding Fig. 9, we used the first two modes measured when the beam was in
water and in air and all modes from the FEM of the beam in implementing the proposed method for Model A, in which
28 unknowns should be estimated and results are shown in Fig. 14(a). One can conclude that the discrepancy between the
5014 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

0.8

0.7

0.6

0.5

0.4
ε
0.3

0.2

0.1

0
1 2 3 4 5 6 7 8
Number of Unknowns
Fig. 9. Estimation of added mass correction factors using Model A.

Table 2
Comparison of frequencies from FEM, beam in water, and updated FEM (Models A-E).

Modal order Frequency (Hz)

Updated
FEM In water
Model A Model B Model C Model D Model E

1 7.82 7.07 7.09 7.05 7.02 7.09 6.90

2 49.03 45.51 45.03 44.08 43.28 42.72 47.83

Fig. 10. Comparison of the first mode shape between FEM, measured, and updated FEM.

FEM and the experimental beam mainly comes from the mass modeling errors and the influence of added mass. Fig. 14(b)
and (c) shows the mass correction factors εas ¼ ½εfasf ; εvasv  and added mass correction factors εfaf , respectively. Comparing
Fig. 14 and Fig. 9, we can find that the sixth correction factor of added mass is  0.1179, and the average correction factor is
estimated to be 0.1319 for this experiment. If we employ Eq. (2) to compute the added mass factor using the measured
frequencies, one obtains 0.0545 and 0.1854, respectively; and the average value is 0.1199, which is close but smaller than our
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5015

Fig. 11. Comparison of the second mode shape between FEM, measured, and updated FEM.

Fig. 12. Finite element model when the immersed part modeled using: (a) 6 elements (Model B), (b) 4 elements (Model C), (c) 2 elements (Model D), (d) 1
element (Model E).

Fig. 13. Estimation of added mass correction factors: (a) Model A–E, and (b) sensitivity indicator Sa.
5016 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017

7
x 10
2

ε
0

−2
0 5 10 15 20 25

Number of Unknowns

0.5

0
εas
−0.5

−1
1 2 3 4 5 6 7 8 9 10

Number of Unknowns

0.5
εaf

0
f

−0.5
1 2 3 4 5 6 7 8

Number of Unknowns
Fig. 14. Estimation of correction factors of Model A: (a) mass, stiffness and added mass correction factors when they are estimated simultaneously, (b) mass
correction factors when mass and added mass are estimated simultaneously, and (c) added mass correction factors when mass and added mass are
estimated simultaneously.

FEM
0.6 Measured
Mode Shape Value

0.5 Updated

0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10 11
Node Number

0.4
Mode Shape Value

0.2

−0.2 FEM
Measured
−0.4 Updated

−0.6
0 1 2 3 4 5 6 7 8 9 10 11
Node Number

Fig. 15. Comparison of the first and the second mode shape between FEM, measured, and updated FEM when mass and added mass are estimated
simultaneously: (a) the first order, and (b) the second order.

estimated value 0.1319. To further approve the rightness of these estimated correction factors, the first two frequencies and
corresponding mode shapes (see Fig. 15) of the updated model are studied. So, one can get the first two frequencies 7.0499
and 45.9486 Hz of the updated model, respectively. We can find the first two frequencies of the updated FEM are very close
to the measured ones, and the corresponding mode shapes are improved significantly as well.

6. Conclusions

This article tried to estimate the added mass of water from the view of structure using dynamic responses. A series of
correction factors was employed to correct and calculate mass and stiffness discrepancies in FEM and the added mass of the
 F. Liu et al. / Journal of Sound and Vibration 333 (2014) 5004–5017 5017

water simultaneously. One theoretical development is that the approach could reflect the influence of changes such as water
depth, current, etc., and provide refined added mass estimation. The other development is that the proposed method does
not involve much water assumptions because all influences of water can be reflected in measured dynamic responses of the
structure in water. The numerical results show that the mass, stiffness, and added mass correction factors could be
accurately estimated when noise-free measurements are used, even the measured modes with a certain noise. One should
note that some constraints may be required for obtaining unique solutions, and these constraints can be easily imposed in
terms of some inspections. Experimental results also demonstrated that spatial incompleteness is a key issue and that a
more accurate and stable mode shape expansion method is preferred when applying the approach to real problems.

Acknowledgments

The authors wish to acknowledge financial support from the 973 project (Grant no. 2011CB013704), the 111 Project
(B14028) and the National Natural Science Foundation of China (Grant nos. 51279188 and 51379197).

References

[1] J.N. Newman, Marine Hydrodynamics, third ed. MIT Press, Cambridge, Massachusetts and London, England, 1985.
[2] P. De Tarso, T. Esperanca, S. Hamilton Sphaier, Added mass and damping on bottom-mounted rectangular cylinders, Journal of the Brazilian Society of
Mechanical Sciences 18 (4) (1996) 355–373.
[3] D. Clarke, Calculation of the added mass of circular cylinders in shallow water, Ocean Engineering 28 (9) (2001) 1265–1294.
[4] D. Clarke, Calculation of the added mass of elliptical cylinders in shallow water, Ocean Engineering 28 (10) (2001) 1361–1381.
[5] D. Clarke, Calculation of the added mass of elliptical cylinders with vertical fins in shallow water, Ocean Engineering 30 (1) (2003) 1–22.
[6] Z.L. Lin, S.J. Liao, Calculation of added mass coefficients of 3D complicated underwater bodies by FMBEM, Communications in Nonlinear Science and
Numerical Simulation 16 (1) (2011) 187–194.
[7] M. Benaouicha, J.-A. Astolfi, Analysis of added mass in cavitating flow, Journal of Fluids and Structures 31 (2012) 30–48.
[8] O. De La Torre, X. Escaler, E. Egusquiza, M. Farhat, Experimental investigation of added mass effects on a hydrofoil under cavitation conditions, Journal
of Fluids and Structures 39 (2013) 173–187.
[9] M.R. Kramer, Z. Liu, Y.L. Young, Free vibration of cantilevered composite plates in air and in water, Composite Structures 95 (2013) 254–263.
[10] O. De La Torre, X. Escaler, E. Egusquiza, M. Farhat, Numerical and experimental study of a nearby solid boundary and partial submergence effects on
hydrofoil added mass, Computers & Fluids 91 (5) (2014) 1–9.
[11] M. Skrinar, A. Umek, Structural Dynamics, Chapman and Hall, New York, 1997.
[12] H.J. Li, F.S. Liu, S.-L.J. Hu, Employing incomplete complex modes for model updating and damage detection of damped structures, Science in China
Series E: Technological Sciences 51 (12) (2008) 2254–2268.
[13] J.N Huang, R.S. Pappa, An eigensystem realization algorithm (ERA) for modal parameter identification and model reduction, Journal of Guidance,
Control, and Dynamics 8 (5) (1985) 620–627.
[14] F.S. Liu, Direct mode-shape expansion of a spatially incomplete measured mode by a hybrid-vector modification, Journal of Sound and Vibration 330
(18–19) (2011) 4633–4645.
[15] F.S. Liu, H.J. Li, A two-step mode shape expansion method for offshore jacket structures with physical meaningful modeling errors, Ocean Engineering
63 (1) (2013) 26–34.