Applied Ocean Research 123 (2022) 103183

Contents lists available at ScienceDirect

Applied Ocean Research

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

Reconstruction of incident wave profiles based on short-time ship

response measurements
Tomoki Takami a, *, Ulrik Dam Nielsen b, c, Chen Xi a, Jørgen Juncher Jensen b, Masayoshi Oka a
a
National Maritime Research Institute, Japan
b
Technical University of Denmark, Denmark
c
Centre of Autonomous Marine Operations and Systems, NTNU AMOS, Norway

A R T I C L E I N F O A B S T R A C T

Keywords: This paper presents a new approach to attain estimates of the sea state based on short-time sequences of wave-
Sea state estimation induced ship responses. The present sea state estimation method aims at reconstructing the incident wave
Wave reconstruction profiles in time domain. In order to identify phase components of the incident waves, the Prolate Spheroidal
Prolate spheroidal wave functions
Wave Functions (PSWF) are employed. The use of PSWF offers an explicit expression of phase components in the
Short-time measurements
measured responses and incident waves, indicating that estimations can be efficiently attained. A method to
estimate the relative wave heading angle based on the response measurements and pre-computed transfer
functions of the responses is also proposed. The method is tested with numerical simulations and experimental
measurements of ship motions, i.e. heave, pitch, and roll, together with vertical bending moment and local
pressure in a post-panamax size containership. Validation is made by comparing the reconstructed wave profiles
with the incident waves. The accuracy and efficiency of the present approach are promising. At the same time, it
is shown that the use of responses, which are more broad-banded in their frequency characteristics, is an effective
means to cope with high frequency noise in reconstructed waves.

1. Introduction (Nielsen et al., 2020) or 1 hour in full scale (Chen et al., 2020), are
necessary to catch the (statistical) variance of waves accurately, on the
Knowing encountered sea states or wave sequences is of particular assumption that the encountered sea state is stationary. However, the
importance for designing ships and evaluating the safety and efficiency use of relatively long-time measurements for sea state estimations can be
of operations of, amongst others, the Dynamic Positioning control sys­ problematic and result in inaccurate wave spectra because of the
tem (A. Dallolio et al., 2021; Agdal, 2018). As a result thereof, the sea following reasons:
state estimation problems based on wave-induced ship responses have
been widely addressed by various researchers, see e.g. (Nielsen, 2006; - Container ships, for instance, are sailing relatively high speed, often
Nielsen, 2017; U.D. Nielsen et al., 2018; Nielsen et al., 2020; Brodtkorb 20+ knots, which means that they cover a fairly long distance in just
et al., 2016; A.H. Brodtkorb et al., 2018; A.H. Brodtkorb et al., 2018), 30–60 min. In turn, this may compromise the assumption about a
and (Chen et al., 2018; Chen et al., 2020). Although these researches stationary and homogeneous sea state (Nielsen, 2021; Nielsen et al.,
have some differences in terms of the ship responses or methodologies 2021).
used for the sea state estimation, etc., they were fundamentally - Ships are veering the course, which will lead to nonstationary
achieving the same goal: estimating wave spectra using the measured response sequences in the given time frame since the encountered
wave-induced responses in combination with corresponding transfer wave heading angle changes (Hong et al., 2019; T. Takami et al.,
functions. The frequency analysis, like the Fast Fourier Transform (FFT) 2021), even if the sea state itself is stationary and homogeneous.
or the Auto-Regressive (AR) model, was used as the fundamental tool in
these researches in order to derive the wave spectra. In these cases, taking short-time measurements for sea state esti­
In applying the frequency analysis to the sea state estimation, it is mation, say 2–5 min, is ideal to avoid inaccurate wave spectra, but the
also recognized that sufficiently long-time measurements, say 30 min aleatory uncertainty, i.e. the short-term variability of the response

* Corresponding author: Tomoki Takami. National Maritime Research Institute: Kaijo Gijutsu Anzen Kenkyujo, 6-38-1, Shinkawa, Mitaka, Tokyo, 1810004 Japan.
E-mail address: takami-t@m.mpat.go.jp (T. Takami).

https://doi.org/10.1016/j.apor.2022.103183
Received 14 January 2022; Received in revised form 30 March 2022; Accepted 13 April 2022
Available online 25 April 2022
0141-1187/© 2022 Elsevier Ltd. All rights reserved.
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Table. 1 ship motions, vertical bending moment (VBM), and local pressure on the
Principal particulars of the subject ship. hull surface in a containership under long-crested irregular waves are
Ship length Breadth Depth Draft Displacement used to demonstrate the method. Validations are first made by using
numerically generated time series of measurements. Further validation
283.8 m 42.8 m 24.0 m 14.0 m 109,480 ton
is then made by using model scale seakeeping test measurements with
and without forward speed of the ship.
signals, will have a significant effect on the estimated sea states (Hong
et al., 2019; T. Takami et al., 2021). 2. Methodology
One way to partly address these issues is to estimate incident wave
profiles in time domain, rather than estimating only the encountered In this section, a methodology for wave profile reconstruction mak­
wave spectra. The effectiveness of the time domain approach obviously ing use of PSWF is introduced. The present approach hypothesizes that a
relies not only on the accuracy of the amplitude components at each ship is sailing in long-crested irregular waves, i.e. the wave propagates
wave frequency, but also on the accuracy of the phase components. The in one direction.
Fast Fourier Transform (FFT) can provide the phase components in the
frequency domain; however, it is extremely difficult to accurately 2.1. Detection of phase components by PSWF
determine the phase components by which the incident wave elevations
can be reconstructed, as the frequency resolution of the FFT becomes Let Rk (t) be the wave-induced ship responses within a time span of −
coarse if the time length of the measurements is short. Alternatively, a T ≤ t ≤ T [s], where the subscript k, by definition k = {1, 2, ..., K}, refers
time domain approach making use of the Kalman filter could be devel­ to the type of measured response, e.g., k = 1: heave motion, k = 2: pitch
oped and adopted. In fact, promising results of the incident waves in motion, etc. Rk (t) can be expressed by a linear superposition of trigo­
terms of the wave amplitude components have been obtained in (Pas­ nometric components:
coal and Soares, 2009; Pascoal et al., 2017; Kim et al., 2019), but the
Kalman filter necessitates a run-up phase to attain convergence. ∑
M √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )
Rk (t) ≈ 2PSDk (ωi )dωcos ωi t − εi,k
The authors’ research group has recently explored the application of
(1)
i=1
Prolate Spheroidal Wave Functions (PSWF) (Slepian and Pollak, 1961) ∑
M √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
{ }
to ship response prediction problems, see T. Takami et al. (2021), T. = PSDk (ωi )dω ui,k cos(ωi t) − ui,k sin(ωi t) ; − T≤t≤T
Takami et al. (2021), Takami et al. (2020). It is known that every
i=1

bandlimited signal can be represented by a linear superposition of where M denotes the number of discretized frequencies, ωi the discrete
PSWF, see e.g. Xiao et al. (2001), while the phase components of the frequencies, dω the equidistant increment between discrete frequencies,
signal can be explicitly obtained, thanks to the Fourier invariance of and PSDk the power spectrum density of the response within −
PSWF (Moore and Cada, 2004). Finer frequency resolution than the FFT T ≤ t ≤ T [s]. The phase components εi,k are associated with {ui,k , ui,k }
can be attained even if the response time length is short, by estimating
according to the additive theorem. PSDk is calculated according to the
numerical values in PSWF together with the eigenvalues of PSWF in
Wiener-Khinchin theorem, given as:
advance, e.g. by using the Legendre polynomials-based approach (Xiao
∫
et al., 2001). Since global wave-induced ship responses are usually 2 T
PSDk (ω) = ACFk (t)(ωt) dt (2)
bandlimited signals where the significant frequency range is limited, π 0
PSWF may be applied in an approach for time domain sea state esti­
mation, where the outcome is the actual encountered wave profile with where ACFk denotes the autocorrelation of the response Rk (t). The nu­
accurate information on the phase components. merical derivation of ACFk can be made according to a procedure by e.g.
As outlined in the preceding, in this study, a new approach for time Kitagawa (Kitagawa, 2010).
domain wave reconstruction is presented. Fundamentally, the method is PSWF is then exploited to derive the phase component {ui,k , ui,k }
based on short-time response sequences and incorporates PSWF to from the measurements. According to (T. Takami et al., 2021), the time
evaluate the phase components of the wave-induced responses and series measurements Rk (t) within − T ≤ t ≤ T [s] can be represented by a
incident waves. The (relative) wave heading angle is derived through a linear superposition of PSWF:
time series analysis using the response measurements and pre-estimated (t )
∑
Ne
transfer functions of the responses. The present approach facilitates es­ Rk (t) ≈ aj,k ψ j ,c ; − T≤t≤T (3)
timations with small computational burdens, owing to an explicit solu­ j=0
T
tion of the phase components by PSWF. Time series measurements of the
Here ψ j denote the PSWF, c the Slepian frequency (Slepian and

Fig. 1. Location of VBM and pressure measurements in the containership.

2
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 2. Transfer functions of the ship responses calculated by the strip theory (NMRIW-II). Fn = 0.

where Ω is the upper frequency limit of the signal expressed by PSWF,

Table. 2
Parameters for estimations.
thus ωi ≤ Ω. One can determine c value as per Eq. (5) and then calculate
numerical values in PSWF with respect to c by using e.g. the Legendre
2×T Slepian frequency c Ne
polynomials-based approach (Xiao et al., 2001). Since the frequency
Ω

Case A 102.3 [s] 100 1.95 [rad/s] 63 band of PSWF is limited to ωi ≤ Ω [rad/s], the frequency band of the
Case B 204.7 [s] 200 1.95 [rad/s] 126
approximated response Rk by Eq. (3) is also limited to ωi ≤ Ω [rad/s]. If a
Case C 600.0 [s] 600 2.00 [rad/s] 380
broad-banded response which has significant contributions in ωi > Ω
[rad/s] is used, the accuracy of response fitting by Eq. (3) and resulting
Pollak, 1961), and Ne the number of PSWF to be considered. The wave profiles at ωi > Ω [rad/s] range will be deteriorated. Hence, one
orthogonality of PSWF (Moore and Cada, 2004) offers rapid derivation should choose not too small Ω value, based on prior understanding of the
of the coefficients aj,k as follows: transfer functions of the responses to be used.
The Fourier invariance of PSWF (Moore and Cada, 2004) indicates
∫1 (t ) (t) t the following relationships:
aj,k = ψj , c Rk d (4)
T T T
− 1

The Slepian frequency c and the time range T hold on to the following
relationship:
c = TΩ (5)

3
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 3. Heading angle estimation results in comparison with the frequency domain approach by Chen et al. (2020). Short-time span of 2 × T = 102.3 [s] (Case A) and
ship motion measurements (heave, pitch, and roll) with Fn = 0 are used.

(t ) 2 ∫1 (ω ) ( t ω) ω tained, the next step is to estimate the relative wave heading angle be­
ψj ,c = ψ j , c cos c d ; for even j tween the ship and waves (from hereafter referred to as ‘heading angle’)
T λj Ω T Ω Ω
0 by referring to the other responses. Provided that the phase components
(6) of k = l case, i.e. {ui,l , ui,l }, are obtained, the other responses (k ∕
= l) can
(t ) ∫1 (ω ) ( t ω) ω
ψj , c =
2
ψ j , c sin c d ; for odd j be expressed as follows:
T λj Ω T Ω Ω
0 ∑M √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅{ }
u cos(ωi t − Tl (ωi , χ ) + Tk (ωi , χ ))
Rek (t) ≈ PSDk (ωi )dω i,l ; k
− ui,l sin(ωi t − Tl (ωi , χ ) + Tk (ωi , χ ))
where λj denote the scaled eigenvalues of PSWF Moore and Cada, 2004). i=1

Eq. (6) in combination with Eqs. (1) and ((3) yields the following = l, ωi ≤ Ω
∕
analytical solution of {ui,k , ui,k }: (8)
√̅̅̅̅̅̅
2 dω ∑
Ne
aj,k (ωi ) where Tk (ωi , χ ) is the phase component of a transfer function of the kth
ui,k = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ψj , c
Ω PSDk (ωi ) j = 0 λj Ω response, which is to be prepared in advance. The heading angle χ is
then determined by referring to the time series measurements of each
even j
(7) response, so that the errors between Rek (t) by Eq. (8) and the measure­
√̅̅̅̅̅̅
∑ aj,k (ωi ) ments are minimized. In the present study, the Pearson correlation co­
Ne
2 dω
ui,k = − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ψj , c
Ω PSDk (ωi ) j = 0 λj Ω efficients between estimation (Rek ) and measurements (Rk) are employed
to evaluate the errors. Thus, the heading angle estimation can be made
odd j
as
for PSDk (ωi ) > 0. It should be also noted that the scaled eigenvalues λj {
∑ K
}
become close to zero if Ne > 2c/π (Moore and Cada, 2004), which can χ e = arg max ρk (χ )
χ ∈[0,2π ]
lead to inaccurate {ui,k , ui,k } by Eq. (7). In the present study, Ne is set to k=1, k∕
=l
(9)
an integer value close to 2c/π but not larger than 2c/π to circumvent cov(Rk e (t, χ ), Rk )
this. ρk (χ ) =
σ (Rk e (t, χ )) ∗ σ (Rk )

here cov(Rk e , Rk ) means the covariance of Rek and Rk, and σ (•) indicates
2.2. Relative wave heading angle estimation
the standard deviation. In the present study, the heading angle χ is
discretized into 10 [deg] increments and then χ e is determined based on
Once the phase components {ui,k , ui,k } of a certain response are ob­

4
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 4. Heading angle estimation results in comparison with the frequency domain approach by Chen et al. (2020). Short-time span of 2 × T = 204.7 [s] (Case B) and
ship motion measurements (heave, pitch, and roll) with Fn = 0 are used.

Fig. 5. Heading angle estimation results based on all the measurements (heave, pitch, roll, VBM, and pressure). Head sea (180 [deg]) case with Fn = 0.

Eq. (9) among the discretized χ . √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ { }

∑
M
PSDk (ωi )dω ui,k cos(ωi t − Tk (ωi , χ e ))
ηe (t) = ; − T ≤ t ≤ T, ωi
2.3. Incident wave reconstruction i=1
Ak (ωi , χ e ) − ui,k sin(ω i t − T ω
k( i , χ e ))

≤Ω
Using the estimated heading angle χ e and the phase components of (10)
the kth response {ui,k ,ui,k }, the incident wave elevation ηe (t) at midship,
centerline, can be reconstructed in combination with the transfer func­ where Ak (ωi , χ e ) is the amplitude component of a transfer function of the
tion and the power spectrum density: kth response. Note that ωi can be read as the encountered wave fre­
quencies if the subject ship has a forward speed.
The wave reconstruction by Eq. (10) may not function in the fre­
quency range where Ak (ωi , χ e ) is close to zero. An effective way to avoid
this issue is to introduce multiple responses of which transfer functions

5
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 6. Heading angle estimation results with the short-time span of 2 × T = 204.7 [s] (Case B) and 2 × T = 600.0 [s] (Case C) with Fn = 0.078. Estimation results
based on only the ship motions and all the responses are compared.

cover different frequency bands into ηe estimation. To this end, an value is made, and then α = 0.1 is found to be appropriate. In Section
introduction of the weighting factors together with normalisations of 3.4, the wave reconstruction results for α = 0 (i.e. no criterion) and α =
transfer functions (Nielsen et al., 2019) is further made. In the present 0.1 will be compared and discussed.
study, a sum of squares of the normalised amplitude components of
transfer functions are used. It means: 3. Numerical investigation using simulated measurements
M √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
{ }
∑K ∑
PSDk (ωi )d ω Ak (ωi ,χ e ) ui,k cos(ωi t − Tk (ωi , χ e ))
Ak,max Ak,max − ui,k sin(ωi t − Tk (ωi , χ e )) 3.1. Subject ship, response, and environment
ηe,wf (t) = k=1 i=1 K { }2 ; − T≤t
∑
A container ship of 6600TEU size (post-panamax size) is used for
Ak (ωi ,χ e )
Ak,max
validating the developed method. The principal particulars are given in
k=1

≤ T, ωi ≤ Ω Table 1. Five distinct responses are used for a series of wave re­
(11) constructions. The responses are numbered as follows:
k = 1: Heave motion
where Ak,max = max{Ak (ωi , χ e )}. It should be noted that the transfer k = 2: Pitch motion
function of roll motion in head seas or following seas may indicate near k = 3: Roll motion
zero values of Ak (ωi , χ ) in all the frequency ranges, which in turn the k = 4: VBM at SS5.5
derivation of ηe,wf fails as Ak,max also becomes close to zero. Hence in the k = 5: Pressure on the ship bottom surface at SS4.25, center line
present study, the k value corresponds to the roll motion will be skipped Fig. 1 indicates the location of the VBM and pressure measurement
in computing Eq. (11). points. Note that the above five responses have been also measured
The reconstructed waves by Eq. (11) may include numerical errors at during the model experiment presented in Section 4.
the frequency range if the denominator in the equation, i.e. The strip theory based code, NMRIW-II (Matsui et al., 2017), is used
∑K {Ak (ωi ,χe )}2
k=1 Ak,max , are too small. Therefore, a criterion of the lower limit to compute the transfer functions of the responses. The transfer func­
tions in zero forward speed, i.e. the Froude number Fn = 0, are calculated
of these values is further introduced. The frequency components which
in 30 [deg] increments. By way of example, the amplitude components
meet the following equation are cut out in calculating the reconstructed
of calculated transfer functions of 90 [deg], 120 [deg], and 180 [deg]
waves by Eq. (11):
are shown in Fig. 2, noting that the heading angles in {0, 180} [deg]
K { }2
∑ Ak (ωi , χ e ) correspond to the port seas between the following seas (0 [deg]) and the
< α2 (12) head seas (180 [deg]). The roll motion is anti-symmetric w.r.t. X-axis
k=1
Ak,max
(see Fig. 1), implying that the phase angles of the transfer functions in
where α indicates the criterion. A small sensitivity study in terms of α the starboard seas ({180, 360} [deg]) are shifted 180 [deg] from the port
seas, while the other four responses are symmetric w.r.t. X-axis. One can

6
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 7. Estimated wave spectra smoothed by WAFO using all reconstructed short-time sequences from the entire time span 0–5000 [s]. The results follow from Eq.
(11) with Fn = 0. Estimation setting of Case B and the criterion α = 0.0 are used.

notice from Fig. 2 that the amplitudes in the transfer functions of the
VBM and pressure have significant non-zero values even in the high
ωi 2 Vcosχ
ωei = ωi − (14)
frequency range, say, ω > 1.0 [rad/s]. On the other hand, the heave, g
pitch, and roll motions are less sensitive to such a high frequency range.
where V denotes the forward speed of the ship and g the gravitational
The Pierson-Moskowitz wave spectrum with significant wave height
acceleration. All the time series are generated at the sampling frequency
Hs = 8 [m] and zero-crossing wave period Tz = 12 [s] is considered as the
of 10 [Hz].
encountered sea state, assuming long-crested waves. Responses are
simulated for stationary conditions, where the heading angle and for­
3.2. Case studies and estimation setting
ward speed are constant within 0 ≤ t ≤ 5000 [s], and thus time series of
the responses are generated making use of the transfer functions as
Three cases with different short-time length T, i.e. 2 × T = 102.3 [s]
follows:
(Case A), 204.7 [s] (Case B), and 600.0 [s] (Case C) are considered. The
Mw∑
=1001 {
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ v cos(ωei t − Tk (ωei , χ ))
} relevant parameters used in the estimations are listed in Table 2. The
Rk (t) = Ak (ωei , χ ) S(ωi )Δωi i ; 0≤t upper limit frequency Ω is determined so that the frequency range ω ≤ Ω
− vi sin(ωei t − Tk (ωei , χ ))
covers the dominant frequency range of the responses and waves.
i=1

≤ 5000 [s] Therefore, PSWF of Slepian frequencies c = 100, 200, and 600 are used
(13) for detecting the phase components; noting that the numerical values of
PSWF can be derived in the same way as ref. (T. Takami et al., 2021).
where S(ωi) denotes the wave spectrum, {vi , vi } are the independent and The number Ne is decided so that it becomes Ne ≈ 2c/π. In the present
Gaussian distributed stochastic variables, and Δωi are the non- study, the number of discretized frequencies, M in Eq. (1), is set M =
equidistant increment between discrete absolute wave frequencies ωi. ( )
1001, thereby discretizing PSWF ψ j ωΩi , c in 1001 components within
{vi , vi } are generated according to the Box-Muller algorithm. For the
0 ≤ ωΩi ≤ 1, according to the relationship in Eq. (7). The short-time es­
transfer functions, i.e. Ak and Tk, the zero forward speed ones shown in
Fig. 2 are used both for generating response time series with zero and timations within 2 × T [s] are repeated until 5000 [s] without over­
non-zero forward speed. The frequency range of 0 ≤ ωi ≤ 2.0 [rad /s] is lapping of the time window, i.e. 48, 24, and 8 estimations are made by
discretized by Mw = 1001 components. The encounter frequency ωei is Case A, Case B, and Case C, respectively.
defined by the Doppler shift:

7
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 8. Correlation coefficients between incident waves and the reconstructed waves by Eq. (11) with Fn = 0. Estimation setting of Case B is used.

3.3. Heading angle estimation results is improved.

Next, the non-zero forward speed cases are investigated. The
Using the numerically generated response time series, cf. Eq. (13), response time series with a forward speed of 8 [kt] (Fn = 0.078) are
the (relative) heading angles are estimated from Eq. (9) and the results generated according to Eq. (13) for the heading angles of 30, 120, and
will be examined in the following. The cases with zero-forward speed (V 180 [deg], and then the present time domain approach is applied. Fig. 6
= 0 [kt], Fn = 0) and heading angles of 90, 120, and 180 [deg] are first shows the estimation results of Case B when the ship motion measure­
discussed. In selecting the heading angles, by the optimization in Eq. (9), ments (k ≤ 3) and all the measurements, respectively, are used. From
the heave motion is used as the lth response in Eq. (8). The frequency the plots of the 120 and 180 [deg] cases, the estimations indicate the
domain approach presented by (Chen et al., 2020) are also applied to true angles, whereas small fluctuations are observed in the 30 [deg]
estimate the heading angles of each short-time span for comparison. In case. The deviation relative to the true value is due to the fact that only
the frequency domain approach, the directional distribution function is the low frequency components appear in the responses for the oblique
ignored because long-crested waves are assumed in this particular sea case with forward speed. Therefore, a longer time span is needed to
simulation study. accurately capture the heading angle. This is illustrated in Fig. 6a, where
By using only the ship motion measurements, i.e. k ≤ 3, the heading the results of Case C, with the short-time length of 2 × T = 600.0 [s], are
angles at each short-time sequence are estimated, and the results are plotted; noting that all responses are used for estimation. One can
plotted in Fig. 3 (2 × T = 102.3 [s], Case A) and Fig. 4 (2 × T = 204.7 ascertain from Fig. 6a that the estimation setting of Case C lead to ac­
[s], Case B). The frequency domain approach results plotted in the fig­ curate results.
ures are also based only on the ship motion measurements. For beam and
oblique seas, i.e. 90 and 120 [deg] cases, the present approach, hereafter 3.4. Wave reconstruction results
referred to as the time domain approach, provides perfect estimations in
both Case A and Case B. The frequency domain approach also provides The wave reconstruction by Eq. (11) for the cases with Fn = 0 are first
plausible estimations; however, the results indicate small fluctuations. examined. In order to visualize the frequency characteristic of the
The fluctuations are mitigated by increasing the short-time length, see reconstructed waves, the wave spectra are calculated by applying the
Fig. 4a and Fig. 4b, which supports the well-known fact that some Fast Fourier Transform (FFT) to the sequence of reconstructed wave
minimum duration (“time window”) is necessary to give reliable results profiles, from 0 up to 5000 [s]. The estimated wave spectra are shown in
in the frequency domain approach (Nielsen et al., 2020). For head sea, i. Fig. 7, where those based on different number of responses included in
e. 180 [deg], the estimation results show fluctuations by both the fre­ Eq. (11) (k ≤ 3, k ≤ 4, and all responses) with the criterion α = 0.0 are
quency domain approach and the present time domain approach, since plotted. The estimation setting of Case B is used throughout. The esti­
the roll motion is zero in this case. Re-estimations by the use of other mated results are compared to the (true) generating wave spectrum.
responses than ship motions, for instance, the VBM and pressure, have Note that the spectra in the plots have been smoothed by using WAFO
also been carried out. In this case, the time domain approach yields package (The Wafo Group, 2000), where a consistent smoothing
heading estimations as plotted in Fig. 5, and it is noted that the accuracy parameter is applied. From the plots, significant high frequency noises

8
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 9. Example of a comparison of incident wave and reconstructed wave profiles by Eq. (11) in 180 [deg] case with Fn = 0. Estimation setting of Case B is used.

appear in particular the 120 and 180 [deg] cases. The inclusion of are considered. Moreover, α = 0.1 is applied when k ≤ 3 or k ≤ 4.
multiple responses, on the other hand, significantly improves the accu­ Generally, high values of ρη, close to 1.0, are observed, implying that the
racy in the spectrum, even without the criterion α. It is evident that the present method is capable of reconstructing encountered wave profiles
introduction of the VBM can be used to mitigate the high frequency also for non-zero forward speed cases. Note that the results presented
noise, and, clearly, even better results are obtained by further intro­ here are only for low forward speed. Applicability of the present method
ducing the pressure (legend, ‘Case B - all response’). to higher forward speeds will be discussed in the experimental investi­
To validate the time domain accuracy in the wave profile recon­ gation in Section 4.
struction given by Eq. (11), the correlation coefficients between the A further investigation into the absolute errors between the incident
reconstructed waves and the true incident waves η(t) are calculated wave and reconstructed wave profiles is made, focusing on one short-
within each short-time span. It means, time span. By taking the short-time span of 0 to 102.3 [s], 0 to 204.7
[s], and 0 to 600.0 [s] for Case A, Case B, and Case C, respectively, the
cov(ηe,wf (t), η(t))
ρη = ; − T≤t≤T (15) absolute errors between the incident wave and ηe,wf (t) for the 90 [deg]
σ(ηe,wf (t)) ∗ σ(η(t))
(Fn = 0) and 120 [deg] (Fn = 0.078) cases are shown in Fig. 11, with
Computed values of ρη are plotted in Fig. 8. It can be seen that ηe,wf (t) respect to the non-dimensional time t/T. Note that all the responses are
based on all responses yield high ρη values, close to 1.0, indicating that used for reconstructions. The figure shows that the absolute errors are
the wave profiles are reconstructed with high accuracies. The values of fairly small within − 0.6 ≤ t/T ≤ 0.6, whereas the errors take larger
ρη decrease as the number of responses decrease, notably for the 120 and values when 0.8 ≤ t/T, which applies to all the cases. For Fn = 0.078
180 [deg] cases. This deterioration follows because the amplitude case, the absolute errors have more pronounced high-frequency com­
components of the transfer functions for ship motions, i.e. k ≤ 3, can ponents than in the Fn = 0 case. This is due to the appearance of high-
cover only the low frequency range, see Fig. 2. This fact can be seen from frequency components in the incident waves with forward speed. The
Fig. 9a that shows a comparison of the reconstructed wave profile with errors are the largest at t/T = 1.0, which leads to a discontinuity of the
the corresponding incident one for the 180 [deg] case. The result from reconstructed waves between individual short-time spans. Hence, if one
k ≤ 3 or k ≤ 4 cases can be improved by introducing the criterion α = aims to reproduce continuous wave profiles over a long-time period, it
0.1 to Eq. (11), see Fig. 9b. This is also illustrated in Fig. 8d, where will be necessary to overlap each short-time span, while employing the
values of ρη for k ≤ 3 and k ≤ 4, with α = 0.1, are plotted for the 180 reconstructed wave profiles at the time range (t/T) of, say, −
[deg] case. Note that when all the responses are considered, introduc­ 0.6 ≤ t/T ≤ 0.6. The time range of reconstructed waves by Eq. (11) is
tion of α = 0.1 does not change the results, as no frequency component limited within − T ≤ t ≤ T [s], since PSWF is bandlimited in both time
satisfies Eq. (12) at ωi ≤ Ω. and frequency domains (Slepian and Pollak, 1961). It means, if t > T [s]
Fig. 10 presents the results for ρη when situations with forward speed waves are estimated (predicted) based on Eq. (11), i.e. the waves at t > T
(Fn = 0.078) and heading angles of 30, 120, and 180 [deg], respectively, [s] are calculated by substituting the obtained {ui,k , ui,k } directly into the

9
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 10. Correlation coefficients between incident waves and the reconstructed waves by Eq. (11) with Fn = 0.078. Estimation setting of Case B is used.

equation, resulting waves will rapidly decay to zero for t > T [s]. Ap­ measured via a servo type wave height meter, which was connected to a
pendix A provides an example. If the present method is to be used for towing carriage together with the ship model.
predictions ahead of actual time, one possible way would be to repeat Long-crested irregular waves defined by the ISSC-type wave spec­
the estimations based on Eq. (11) by shifting the time span − T ≤ t ≤ T [s] trum were generated in the tank. The significant wave height, zero-
in order to prepare the (past) time series of incident waves over a crossing wave period, heading angle, and forward speed conditions
long-time period, and then apply the existing wave forecasting method are summarized in Table 3. For zero forward speed case, the responses in
such as (Ma et al., 2018; U.D. Nielsen et al., 2018), and (T. Takami et al., oblique (60, 120 [deg]) and beam (90 [deg]) seas are used, while for the
2021). non-zero forward speed case, the responses in the heading angles of 60
and 120 [deg] with the respective Froude numbers of Fn = 0.060 and Fn
4. Experimental investigation = 0.179 are used in the subsequent sections. Fn = 0.060 and Fn = 0.179
are equivalent to 25% and 75% of service speed of the present ship (Fn =
4.1. Experimental condition 0.239), respectively. All the time series data were stored at the sampling
frequency of 50 [Hz] (5.1 [Hz] in full scale) and are converted to the full-
The developed method has also been assessed using experimental scale based on the Froude similarity law. Re-sampling to 10 [Hz] in full
data. The data originates from a seakeeping test with a scaled ship scale is made by applying the linear interpolation to the measured time
model, and the test was conducted at National Maritime Research series so that the same estimation settings used in Section 3.2 are
Institute (NMRI) in 2010 (Oka et al., 2010). The model is a 1/94.6 scale applied. The present experiment was not carried out only for validating
ship of the 6600TEU size container ship used in the numerical investi­ the present method, but for reproducing extreme responses like whip­
gation in Section 3, see also Table 1. A schematic of the model is illus­ ping and slamming. Thus, only experimental data for severe sea condi­
trated in Fig. 12. Five responses (k = 1 to 5), as listed in Section 3.1, were tions in zero forward speed cases are available. Nonetheless, the data are
measured during the experiment. The model was with an aluminum expected to be useful in terms of demonstrating the effectiveness of the
backbone used for reproducing the vertical bending flexibility of the hull present method in severe conditions, where nonlinear ship responses
girder. The pressure time series on the hull surface at SS4.25, center line, may appear. The duration of the measurements is 3000 [s] for the zero
were measured by a pressure gage. A low-pass filter with cut-off fre­ forward speed cases (State 1, State 2, and State 3) and 2000 [s] for the
quency 19.5 [rad/s] (corresponding to 2.0 [rad/s] in full scale) is non-zero forward speed cases (State 4 and State 5).
applied to the measured VBM and pressure time series. As the natural
frequency of the 2-noded vertical bending vibration was measured at 4.2. Heading angle estimation results
42.7 [rad/s] (4.4 [rad/s] in full scale) in the backbone, the
high-frequency vibrations have been eliminated from the measurements Estimations are made using the settings as per Case B, see Section 3, i.
via the low-pass filter. The natural frequency of the roll motion in calm e. the short-time length is 204.7 [s]. The transfer functions calculated by
water was found at 2.0 [rad/s] (0.21 [rad/s] in full scale). The NMRIW-II with zero forward speed (Fn = 0), see Fig. 2, are used for all
encountered wave elevations alongside amidships of the model were the estimations, which means that the forward speed effect is not taken

10
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 11. Examples of absolute errors between incident waves and reconstructed waves by Eq. (11). 0 to 102.3 [s] (Case A), 0 to 204.7 [s] (Case B), and 0 to 600.0 [s]
(Case C) measurements are used.

Fig. 12. Schematic of a scaled model of 6600TEU container ship.

responses considered in the estimations (k ≤ 3, k ≤ 4, and all responses).

Table 3 The dotted lines in the figures indicate the starboard side angles which
Experimental conditions. are symmetrical to the true heading angles with respect to the center line
Heading angle Froude number Fn Hs (full scale) Tz (full scale) of the ship. From the figures, although the results occasionally indicate
State 1 60 [deg] 0 15.0 [m] 11.5 [s]
sudden jumps from port seas to starboard seas with the same relative
State 2 90 [deg] 0 15.0 [m] 11.5 [s] angles, the estimation results using all responses produce accurate
State 3 120 [deg] 0 15.0 [m] 11.5 [s] heading angles with Fn = 0, whereas the results based on only the ship
State 4 60 [deg] 0.060 12.0 [m] 10.5 [s] motions (k ≤ 3) indicate small deviations, notably in 60 [deg] case (State
1). The deviations in the 90 [deg] case (State 2) based on the k ≤ 4 re­
State 5 120 [deg] 0.179 6.0 [m] 9.5 [s]
sponses are attributed to inaccurate transfer function of the VBM (k = 4)
in beam seas. Inclusion of pressure (k = 5) has apparently further
into account in the transfer functions. improved the estimation results. The heading angle estimation results in
The estimated heading angles for the cases with Fn = 0 are plotted in the non-zero forward speed cases are shown in Fig. 14, where the results
Fig. 13, where the results are compared for different numbers of are compared by changing the number of responses used in the

11
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 13. Heading angle estimation results with the short-time span of 2 × T = 204.7 [s] (Case B) from experimental measurements. Zero forward speed cases.

Fig. 14. Heading angle estimation results with the short-time span of 2 × T = 204.7 [s] (Case B) from experimental measurements. Non-zero forward speed cases.

estimations as well. The results on the non-zero forward speed cases starboard-related decisions when the reconstructed wave profiles are
indicate the same trend as in Fig. 13: the introduction of VBM and utilized to evaluate local responses in the ship, for instance. In the next
pressure has improved the accuracy in the estimations. section, a complementary method to solve this problem based on the
In the numerical investigation in which the response time series are cross spectrum analysis will be shown.
generated as per Eq. (13), the roll motion time series are assumed to be
fully linear w.r.t. waves, thus the port/starboard decision has been
4.3. Port/Starboard side detection
correctly made by referring to ρk values in Eq. (9). However, in the
experimental cases, the roll motion might have contained nonlinearity
The port/starboard decision may be possible through an additional
that cannot be followed by using the linear transfer function. In this case,
cross spectrum analysis making use of the heave (k = 1) and roll (k = 3)
the present time series analysis based heading angle estimation by Eq.
motions. The basic concept of the cross spectrum analysis adopted in this
(9) does not function well to detect port/starboard side. The jumps of the
study follows ref. (A.H. Brodtkorb et al., 2018), and a conceptual
heading angles between port side and starboard side will not influence
example is illustrated in the following. Let the complex-valued transfer
the reconstructed wave profiles, but it will be needed to make port- or
functions of heave and roll be Φ1 (ω, χ ) and Φ3 (ω, χ ), respectively, then

12
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 15. Real and imaginary parts of analytical cross spectrum of heave and roll motion at χ = 60, 90, and 120 [deg]. The transfer functions based on NMRIW-II and
the Pierson-Moskowitz wave spectrum of Hs = 8 [m] and Tz = 12 [s] is used.

the analytical solution of the cross spectrum in the case of heading angle The cross spectrum from the time series measurements can be esti­
χ can be derived as: mated by exploiting PSWF. Owing to the Fourier invariance of PSWF, see
Eq. (6), the Fourier transform of heave (X1) and roll (X3) can be given as
G13 (ω, χ ) = Sw (ω) ∗ Φ1 (ω, χ ) ∗ Φ3 (ω, χ ) (16) (T. Takami et al., 2021):
∑ (ω ) ∑ (ω )
where the overline denotes the complex conjugate, and Sw the wave X1 (ω) = N N

spectrum for which a standard wave spectrum, e.g. the Pierson- j = 0 aj,1 Ψ Ω, c − i j = 0 aj,1 Ψ Ω, c
Moskowitz type, can be used. Using the transfer functions Φ1 (ω, χ ) even odd
(17)
∑ (ω ) ∑ (ω )
and Φ3 (ω, χ ) at χ = 60, 90, and 120 based on the strip theory, while using X3 (ω) = N N
j = 0 aj,3 Ψ Ω, c − i j = 0 aj,3 Ψ Ω, c
the Pierson-Moskowitz wave spectrum with, for instance, Hs = 8 [m] and
even odd
Tz = 12 [s] as Sw, the real and imaginary parts of the cross spectrum G13
are shown in Fig. 15. It is seen from the figure that the sign of the global where aj,1 = aj,1 /Ωλj , aj,3 = aj,3 /Ωλj . Then, the measured cross spectrum
peaks in the imaginary parts of G13 appears to be on the positive side density of heave and roll, C13, can be written
when χ = 60, 90, 120 [deg]. Therefore, the port/starboard detection can
be made by looking into the sign of the global peaks in the imaginary
parts of the measured cross spectra.

T
C13 (ω) = X1 (ω)X∗3 (ω)
π
⎡⎛ ⎞⎛ ⎞ ⎛ ⎞⎛ ⎞⎤
⎢⎜ ∑ ( ω )⎟ ⎜ (ω )⎟ ⎜ (ω )⎟ ⎜ ( ω )⎟ ⎥
⎢⎜ ⎟⎜ ∑ N ⎟ ⎜∑N ⎟⎜ ∑ N ⎟⎥
=⎢⎜
⎢⎜
N ⎟⎜
j = 0 aj,1 Ψj Ω, c ⎟⎜
⎟ ⎜
j = 0 aj,3 Ψj Ω, c ⎟ + ⎜
⎟⎜
j = 0 aj,1 Ψj Ω, c ⎟⎜
⎟⎥
j = 0 aj,3 Ψj Ω, c ⎟⎥−
⎣⎝ ⎠⎝ ⎠ ⎝ ⎠⎝ ⎠⎦
even even odd odd (18)
⎡⎛ ⎞⎛ ⎞ ⎛ ⎞⎛ ⎞⎤
⎢⎜ ∑ ( ω )⎟ ⎜ (ω )⎟ ⎜∑ (ω )⎟ ⎜ ( ω )⎟ ⎥
⎢⎜ ⎟⎜ ∑ N ⎟ ⎜ ⎟⎜ ∑ N ⎟⎥
i⎢⎜
⎢⎜
N
a Ψ ,
j = 0 j,1 j Ω ⎟⎜c ⎟ ⎜ a Ψ ⎟
j = 0 j,3 j Ω, c ⎟ −
⎜
⎜
N
a Ψ , c
j = 0 j,1 j Ω ⎟⎜
⎟⎜ a Ψ ,
j = 0 j,3 j Ω ⎟⎥c ⎟ ⎥
⎣⎝ ⎠⎝ ⎠ ⎝ ⎠⎝ ⎠⎦
even odd odd even

13
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 16. Global peaks in the imaginary parts of measured cross spectra between heave and roll at each short-time span. Experimental measurements with zero
forward speed are used.

Fig. 17. Global peaks in the imaginary parts of measured cross spectra between heave and roll at each short-time span. Experimental measurements with non-zero
forward speed are used.

The global peaks in the measured cross spectra C13 in the zero for­ Case C, whereas still the fluctuations have appeared in State 4. The
ward speed cases (State 1, State 2, and State 3) obtained by Case B and required time length for the accurate detection would be dependent on
Case C within each short-time span are summarized in Fig. 16. Case B the number of the zero-up crossings included in the measured response
results occasionally indicate fluctuations of the sign of the global peaks, within a time span. Hence, for State 4 where the ship being operated in
prominently in the 60 [deg] case (State 1), while Case C results consis­ the following seas with forward speed, longer time span than 600 [s]
tently indicate the same signs, meaning that all the heading angles in might be needed. The present cross-spectrum analysis approach is ex­
each short-time span can be judged as the port seas. The global peaks of pected to be also applicable to directional wave cases (short-crested
C13 in the non-zero forward speed cases (State 4 and State 5) obtained by waves) (A.H. Brodtkorb et al., 2018). Other directional wave estimation
Case B and Case C within each short-time span are summarized in methods which have been actively studied in recent years, see e.g.
Fig. 17. For State 5, all the global peaks indicate the positive values by (Chen et al., 2021), may also be useful in this case.

14
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 18. Correlation coefficients between experimentally measured incident waves and the reconstructed waves by Eq. (11). Experimental measurements with zero
forward speed are used. Estimation setting of Case B and the criterion α = 0.1 are adopted.

Fig. 19. Correlation coefficients between experimentally measured incident waves and the reconstructed waves by Eq. (11). Experimental measurements with non-
zero forward speed are used. Estimation setting of Case B and the criterion α = 0.1 are adopted.

4.4. Wave reconstruction results State 2, and State 3) are computed as per Eq. (15) using Case B estimation
settings for each short-time span and are plotted in Fig. 18, where the
The reconstructed wave profiles by Eq. (11), ηe,wf (t), are validated by criterion α = 0.1 is applied. ρη based on different numbers of responses
comparing with the experimentally measured incident wave elevations. considered in the estimations (k ≤ 3, k ≤ 4, and all responses) are
The correlation coefficients ρη in the zero forward speed cases (State 1, compared. In the figures, sudden drops of ρη values are observed in the

15
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 20. Examples of comparisons between incident wave and reconstructed wave profiles by Eq. (11). Estimation setting of Case B and the criterion α = 0.1
are adopted.

k ≤ 3 and k ≤ 4 based results, especially for 60 and 90 [deg] cases (State theory, are plausible in the zero forward speed cases.
1 and State 2), which are due to the wrong heading angle estimations, ρη values in the case of non-zero speed cases (State 4 and State 5) are
see Fig. 13a and Fig. 13b. With the accurate heading estimation results, then calculated by using Case B and α = 0.1 as well, and are plotted in
regardless of the port or starboard side, the resulting ρη reaches values Fig. 19. For State 4, see Fig. 19a, although a sudden drop of ρη occurs in
fairly close to 1 in all the cases, meaning that the incident wave profiles the k ≤ 3 case due to the low accuracy in the estimated heading angle,
can be accurately reconstructed. The achieved values of ρη (close to 1) see Fig. 15a, all the values of ρη are relatively close to 1. For State 5, on
imply that not only the amplitude components but also the phase the other hand, all results have low values of ρη, despite the fact that the
components of the transfer functions, calculated by the present strip heading angles have been accurately estimated, see Fig. 14b.

16
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 20. (continued).

Comparisons of wave elevations for States 1 to 5 within inclusion of VBM (k ≤ 4) and further introduction of pressure (all re­
0 ≤ t ≤ 200 [s], i.e. the first short-time span, are shown in Fig. 20. For sponses) measurements significantly improve the accuracy. Up to this
the States 1 to 4 cases, although small high-frequency oscillations appear point, all the estimations are made based on the zero-speed transfer
in the reconstructed waves, the reconstructed wave elevations are in functions. Despite this, Fig. 19 supports a fact that even the use of the
good agreement with the experimental measurements, resulting in high zero-speed transfer functions is applicable to reconstructing low-
ρη values as plotted in Fig. 18 and Fig. 19a. For the State 5 case, the frequency waves in high forward speed, suggesting that any unmea­
deviation from the experimental measurements appears to originate sured responses that are primarily excited by low-frequency waves will
from the high frequency components. Therefore, it can be inferred that be effectively estimated by the present method. If one aims to recon­
the assumed transfer functions may have included discrepancies from struct the high frequency incident wave components, it will be necessary
the experimental model in the high frequency range when the ship speed to use response measurements sensitive to such a high frequency range,
is fast and is running in oblique head seas as in State 5. The low- together with the transfer functions being reasonably accurate at high
frequency wave components are then examined by applying a low- forward speed.
pass filter to the incident wave elevations. Note that the low-pass Even when the low-pass filter is applied to the reconstructed waves
filtering of the reconstructed waves can be easily made by simply by the present method based on all responses, overestimations of wave
limiting the frequency range in Eq. (11). Fig. 21a shows a comparison of heights are occasionally observed, see Fig. 21a. A possible cause for the
low-pass filtered wave profiles with a cut-off frequency of 1.0 [rad/s]. It overestimations may be that the amplitude components of the zero-
is apparent from the figure that when k ≤ 4 or all responses are included, forward speed transfer functions (i.e. Ak in Eq. (11)) are too small. A
the reconstructed wave profiles well capture the experimental mea­ further investigation is made by using the transfer functions with for­
surements. In the k ≤ 3 case, the high frequency components have ward speed effect (Fn = 0.179) calculated by the present strip method
already been cut off by introducing α = 0.1, thus no improvement can be (NMRIW-II). Calculated transfer functions are shown in the Appendix B.
made. The correlation coefficients ρη are re-calculated based on the low- A comparison of low-pass filtered wave elevations (cut-off frequency 1.0
pass filtered experimentally measured incident wave and reconstructed [rad/s]) within 0 ≤ t ≤ 200 [s] is shown in Fig. 22a, while Fig. 22b
waves, and are plotted in Fig. 21b. It is evident from Fig. 21b that the provides the correlation coefficients ρη of the low-pass filtered

17
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 21. Comparisons of reconstructed wave elevations and correlation coefficients in State 5 case by Eq. (11) with an application of a low-pass filter with cut-off
frequency 1.0 [rad/s]. Estimation setting of Case B and the criterion α = 0.1 are adopted.

experimentally measured incident wave and reconstructed waves based approach, heading angle and wave profile estimations can be made
on the transfer functions with forward speed effect. The overestimations efficiently at small computational burden, owing to an explicit solution
have been mitigated by using the transfer functions with forward speed of the phase components by PSWF. The present approach was validated
effect, see Fig. 22a, since the amplitude components of the transfer by using simulated and experimentally measured time series of the ship
function are overall smaller than the zero-speed transfer functions, as motions, vertical bending moment (VBM), and local pressure on the hull
can be seen from Fig. A1 in the Appendix B. However, no significant surface in a containership under long-crested irregular waves, making
improvement in terms of the correlation coefficients has been found as use of the transfer functions based on the strip theory. The following
seen in Fig. 22b. This may be due to the difficulty of appropriately conclusions were drawn.
capturing forward speed effect by the strip theory, thus it will be needed
to use the transfer functions being reasonably accurate at higher forward 1 The present approach attained accurate heading angle estimations
speed by application of other seakeeping codes, e.g. the three- even based on the short-time responses, in contrast to the frequency
dimensional panel method, for further improvement. domain approach. Inclusion of VBM and pressure for the estimation
further improved the estimation accuracy. For port or starboard side
5. Conclusions detection, an additional cross spectrum analysis using relatively
long-time measurements of heave and roll motions, in order to take
This paper proposed a new approach for time domain wave recon­ sufficient number of zero-up crossings, was found to be effective
struction based on short-time ship response measurements. The present through the experimental investigation.
method incorporated prolate spheroidal wave functions (PSWF) to 2 In reconstructing wave profiles, the weighting factors and a criterion
evaluate the phase components of the wave-induced responses and to the normalised transfer functions were introduced, with the
incident waves. The (relative) wave heading angle was derived through objective of integrating multiple response measurements into the
a time series analysis using the response measurements and pre- reconstructed waves and mitigating high frequency noise. The nu­
estimated transfer functions of the responses. By following this merical investigation revealed that the criterion is vital to mitigate

18
T. Takami et al. Applied Ocean Research 123 (2022) 103183

Fig. 22. Comparisons of reconstructed wave elevations and correlation coefficients in State 5 case by Eq. (11) with an application of a low-pass filter with cut-off
frequency 1.0 [rad/s]. Estimation setting of Case B and the criterion α = 0.1 are adopted. Transfer functions with forward speed effect of Fn = 0.179 are used.

Fig. A1. Example of a comparison of incident wave and reconstructed wave profiles by Eq. (11) in 90 [deg] case with Fn = 0 (at t > T [s] included). Estimation setting
of Case B using all responses is applied.

the high frequency noise which is notable when only the ship mo­ 3 The present approach, together with the weighting factors and cri­
tions are used for the reconstruction, while the introduction of VBM terion, was shown to be able to accurately reproduce the incident
and pressure fairly reduced the high frequency noise even without wave profiles measured in the seakeeping test with and without
the criterion. forward speed. Although the high frequency incident wave compo­
nents in oblique head seas with high forward speed case could not be

19
T. Takami et al. Applied Ocean Research 123 (2022) 103183

captured well, which might be due to the low accuracy of the transfer Validation, Formal analysis, Writing – original draft, Writing – review &
functions, the low frequency components of the incident waves were editing. Ulrik Dam Nielsen: Conceptualization, Methodology, Super­
accurately reconstructed, especially by utilizing VBM and pressure. vision, Writing – review & editing. Chen Xi: Validation, Formal analysis,
Writing – review & editing. Jørgen Juncher Jensen: Methodology,
As the present approach offers efficient reconstruction of the long- Supervision, Writing – review & editing. Masayoshi Oka: Data curation,
crested wave profiles with fair accuracy, it could provide useful infor­ Writing – review & editing.
mation for the operation of e.g. the Dynamic Positioning control system,
as the knowledge of wave encounter frequency, amplitude, and phase Declaration of Competing Interest
plays an important role on the operation, see e.g. (A. Dallolio et al.,
2021) and (Belleter et al., 2015). In addition, the present approach The authors declare that they have no known competing financial
would aid to estimate nonlinear ship responses which are hard to be interests or personal relationships that could have appeared to influence
measured in practice, by combining with time domain computation the work reported in this paper.
methods which use the incident wave profiles as input. Further appli­
cability of the present method to short-crested waves, in which wind Acknowledgements
waves and swell waves come from different directions, should be
demonstrated in the future and, a method to handle multi-directional The authors are grateful to Dr. Toichi Fukasawa for thorough dis­
waves is currently under investigation. cussion about the presented methodologies and results. The authors also
acknowledge Assoc. prof. Astrid H. Brodtkorb for instructive discussion
CRediT authorship contribution statement about the Dynamic Positioning control system.

Tomoki Takami: Conceptualization, Methodology, Investigation,

Appendix a: Reconstructed wave by Eq. (11) beyond the short-time span

The wave reconstruction by Eq. (11) at t > T [s] is checked by substituting the obtained {ui,k , ui,k } based on Eq. (7) directly into the equation. Fig A1
shows an example of the numerical case of 90 [deg], Fn = 0, where Case B using all responses is applied. The reconstructed wave at t > T [s] rapidly
decays to zero, indicating that Eq. (11) itself cannot be used for predicting waves ahead of time.

Appendix b: Transfer functions with forward speed effect

The transfer functions of heave, pitch, VBM, and pressure at the relative heading angle χ = 120 [deg] taking into account the forward speed effect
of Fn = 0.179 are calculated by NMRIW-II and are shown in Fig. A2.

Fig. A2. Comparisons of transfer functions (χ = 120 [deg]) with and without forward speed effect of Fn = 0.179.

20
T. Takami et al. Applied Ocean Research 123 (2022) 103183

References Moore, I.C., Cada, M., 2004. Prolate spheroidal wave functions, an introduction to the
Slepian series and its properties. Appl. Comput. Harmon. Anal. 16, 208–230. https://
doi.org/10.1016/j.acha.2004.03.004.
Agdal, B.O., 2018. Design and Implementation of Control System For Green Unmanned
Nielsen, U.D., Brodtkorb, A.H., Sørensen, A.J., 2018a. A brute-force spectral approach for
Surface Vehicle Bendik Olai Agdal. Norwegian University of Science and
wave estimation using measured vessel motions. Mar. Struct. 60, 101–121. https://
Technology.
doi.org/10.1016/j.marstruc.2018.03.011.
Belleter, D.J.W., Galeazzi, R., Fossen, T.I., 2015. Experimental verification of a global
Nielsen, U.D., Brodtkorb, A.H., Jensen, J.J., 2018b. Response predictions using the
exponential stable nonlinear wave encounter frequency estimator. Ocean Eng 97,
observed autocorrelation function. Mar. Struct. 58, 31–52.
48–56. https://doi.org/10.1016/j.oceaneng.2014.12.030.
Nielsen, U.D., Brodtkorb, A.H., Sørensen, A.J., 2019. Sea state estimation using multiple
A.H. Brodtkorb, U.D. Nielsen, A.J. Sorensen, Sea state estimation using model-scale DP
ships simultaneously as sailing wave buoys. Appl. Ocean Res. 83, 65–76. https://doi.
measurements, ocean. 2015 - MTS/IEEE washingt. (2016). https://doi.org/
org/10.1016/j.apor.2018.12.004.
10.23919/oceans.2015.7404402.
Nielsen, U.D., Dietz, J., 2020. Estimation of sea state parameters by the wave buoy
Brodtkorb, A.H., Nielsen, U.D., Sørensen, A.J., 2018a. Sea state estimation using vessel
analogy with comparisons to third generation spectral wave models. Ocean Eng 216,
response in dynamic positioning. Appl. Ocean Res. 70, 76–86. https://doi.org/
107781. https://doi.org/10.1016/j.oceaneng.2020.107781.
10.1016/j.apor.2017.09.005.
Nielsen, U.D., Ikonomakis, A., 2021. Wave conditions encountered by ships—a report
A.H. Brodtkorb, U.D. Nielsen, A.J. Sørensen, Online wave estimation using vessel motion
from a larger shipping company based on ERA5. Ocean Eng 237, 109584. https://
measurements *, IFAC-PapersOnLine. 51 (2018) 244–249. https://doi.org/
doi.org/10.1016/j.oceaneng.2021.109584.
10.1016/j.ifacol.2018.09.510.
Nielsen, U.D., 2006. Estimations of on-site directional wave spectra from measured ship
Chen, X., Okada, T., Kawamura, Y., 2018. Sea state estimation using measured hull
responses. Mar. Struct. 19, 33–69. https://doi.org/10.1016/j.marstruc.2006.06.001.
responses on 14, 000TEU large container ships. In: Proc. 7th Int. Marit. Conf. Des.
Nielsen, U.D., 2017. A concise account of techniques available for shipboard sea state
Saf, pp. 160–168.
estimation. Ocean Eng 129, 352–362.
Chen, X., Okada, T., Kawamura, Y., Mitsuyuki, T., 2020. Estimation of on-site directional
Nielsen, U.D., 2021. Spatio-temporal variation in sea state parameters along virtual ship
wave spectra using measured hull stresses on 14,000 TEU large container ships.
route paths. J. Oper. Oceanogr. 0, 1–18. https://doi.org/10.1080/
J. Mar. Sci. Technol. 25, 690–706. https://doi.org/10.1007/s00773-019-00673-w.
1755876X.2021.1872894.
Chen, X., Okada, T., Kawamura, Y., Mitsuyuki, T., 2021. Estimation of directional wave
Oka, M., Hattori, K., Ogawa, Y., 2010. Reproduction of wave induced vibrations of a ship
spectra and hull structural responses based on measured hull data on 14,000 TEU
by using a new type backbone model. In: Proc. 10th Res. Present. Meet, pp. 297–298.
large container ships. Mar. Struct. 80, 103087 https://doi.org/10.1016/j.
Natl. Marit. Res. Inst.
marstruc.2021.103087.
Pascoal, R., Soares, C.Guedes, 2009. Kalman filtering of vessel motions for ocean wave
Dallolio, A., Alfredsen, J.A., Fossen, T.I., Johansen, T.A., 2021a. Experimental validation
directional spectrum estimation. Ocean Eng 36, 477–488. https://doi.org/10.1016/
of a nonlinear wave encounter frequency estimator onboard a wave-propelled USV.
j.oceaneng.2009.01.013.
IFAC-PapersOnLine 54, 188–194. https://doi.org/10.1016/j.ifacol.2021.10.092.
Pascoal, R., Perera, L.P., Soares, C.Guedes, 2017. Estimation of directional sea spectra
Dallolio, A., Alfredsen, J.A., Fossen, T.I., Johansen, T.A., 2021b. Experimental validation
from ship motions in sea trials. Ocean Eng 132, 126–137. https://doi.org/10.1016/j.
of a nonlinear wave encounter frequency estimator onboard a wave-propelled USV.
oceaneng.2017.01.020.
IFAC-PapersOnLine 54, 188–194. https://doi.org/10.1016/j.ifacol.2021.10.092.
Slepian, D., Pollak, H.O., 1961. Prolate spheroidal wave functions, fourier analysis and
Hong, Y., Iseki, T., Nielsen, U.D., 2019. The effect of short-term variability of cross-
uncertainty - I, Bell Syst. Tech. J. 40, 43–63.
spectral analysis on wave buoy analogy. In: Proc. Twenty-Ninth Int. Ocean Polar
Takami, T., Iijima, K., Jensen, J.J., 2020. Extreme value prediction of nonlinear ship
Eng. Conf, pp. 1311–1318.
loads by FORM using prolate spheroidal wave functions. Mar. Struct. 72, 102760
Kim, H., Kang, H., Kim, M., Choi, J., Lee, D., 2019. Real-time estimation of ocean wave
https://doi.org/10.1016/j.marstruc.2020.102760.
spectra from vessel motion using adaptive kalman filter. Appl. Sci. 9 https://doi.org/
Takami, T., Nielsen, U.D., Jensen, J.J., 2021a. Real-time deterministic prediction of
10.3390/app9142797.
wave-induced ship responses based on short-time measurements. Ocean Eng 221.
Kitagawa, G., 2010. Introduction to Time Series Modeling, 1st ed. Chapman and Hall/
https://doi.org/10.1016/j.oceaneng.2020.108503.
CRC.
Takami, T., Nielsen, U.D., Jensen, J.J., 2021b. Estimation of autocorrelation function and
Ma, Y., Sclavounos, P.D., Cross-Whiter, J., Arora, D., 2018. Wave forecast and its
spectrum density of wave-induced responses using prolate spheroidal wave
application to the optimal control of offshore floating wind turbine for load
functions. J. Mar. Sci. Technol. 26, 772–791.
mitigation. Renew. Energy. 128, 163–176.
The Wafo Group, WAFO a matlab toolbox for analysis of random waves and loads,
Matsui, S., Murakami, C., Oka, M., 2017. Validation of the nonlinear wave load analysis
(2000).
program NMRIW-II in comparison with experiments-ship responses in regular wave-,
Xiao, H., Rokhlin, V., Yarvin, N., 2001. Prolate spheroidal wavefunctions, quadrature
Pap. Natl. Marit. Res. Inst. 17, 83–166.
and interpolation. Inverse Probl 17, 805–838.

21