Journal of Coastal Research 37 3 576–588 Coconut Creek, Florida May 2021

Damage Detection of Pile Foundations in Marine

Engineering Based on a Multidimensional Dynamic
Signature
Ruihu Zhu†‡, Qiming Wang†§*, Jinhai Zheng‡, Jingbo Su‡, and Ning Wang‡
† ‡
Key Laboratory of Coastal Disaster and Defence College of Harbour, Coastal and Offshore Engineering
(Hohai University) Hohai University
Ministry of Education Nanjing, Jiangsu, China
Nanjing, China

§
College of Science
Hohai University
Nanjing, Jiangsu, China

ABSTRACT
Zhu, R.; Wang, Q.; Zheng, J.; Su, J., and Wang, N., 2021. Damage detection of pile foundations in marine engineering
based on a multidimensional dynamic signature. Journal of Coastal Research, 37(3), 576–588. Coconut Creek (Florida),
ISSN 0749-0208.

Compared with land engineering, pile foundations in marine engineering have the characteristics of long, free lengths in
water and bad service environments. The traditional methods of pile foundation damage detection have considerable
limitations in the application in marine engineering. In this paper, a new method for pile foundation damage detection is
developed. The method constructs a multidimensional dynamic signature space by fusing multivariate information, uses
the Mahalanobis distance to define the difference between the damaged and undamaged units of the structure, and
detects damage based on a robust outlier identification method in multidimensional space to realize reliable and
automatic damage detection under high noise and complex conditions. The effectiveness and reliability of the method are
verified using a finite element model (FEM) of a high-pile wharf under different damage degrees and different intensities
of Gaussian white noise. The applicability of the method is then experimentally validated on a physical experimental
model of a high-pile wharf. Both the FEM and the experimental results show that the method is capable of detecting pile
foundation damage and has strong application potential.

ADDITIONAL INDEX WORDS: Outlier identification, marine construction, civil engineering.

INTRODUCTION and have low operation efficiency. Additionally, none of the

As an important component connecting foundations and above methods can realize the online monitoring of pile
superstructures, pile foundations are widely used in high-pile foundation damage.
wharfs, offshore platforms, and other marine engineering Damage detection based on dynamic signatures is considered
applications. Compared with land engineering, pile founda- to be the most promising method for nondestructive testing of
tions in marine engineering have the characteristics of long, structures and is widely used in damage detection of bridges,
free lengths in water and bad service environments (Huo et al., high-rise buildings, and other civil engineering structures
2015; Wang and He, 2017). Many pile foundations have been (Zhu, He, and Chen, 2005; Zhu, Yao, and Li, 2019). With the
damaged due to corrosion, fatigue, collision, and other reasons continuous exploration of scholars for decades, the dynamic
during service. Therefore, it is particularly important to test damage detection signatures have expanded from the initial
their health. At present, methods for damage detection of pile frequency and natural modal to modal curvature, modal
flexibility, modal strain, and modal strain energy, etc. (Cao
foundations mainly include high-strain, low-strain, and local
and Qiao, 2009; Maia et al., 2011). At the same time, many
methods (ultrasonic method, rebound method, etc.) (Ding et al.,
damage detection methods have been constructed based on the
2011; Liu, Ren, and Yang, 2012). The traditional damage
above dynamic signatures (Ahmad, Waleed, and Virk, 2019;
detection methods of pile foundations have considerable
Zhao, Noori, and Altabey, 2017), such as the modal assurance
limitations in the application of marine engineering in service.
criterion, square frequency method, and sensitivity analysis
For example, there is no test space for using high strain and low
method (Jassim et al., 2013; Qian et al., 2012).
strain because of the limitation of the connection between the
In recent years, with the development of underwater
pile top and the superstructure. The ultrasonic method and the
vibration testing technology and embedded optical fiber
rebound method have difficulty in detecting underwater parts technology, vibration response measurements of underwater
structures have been realized, which provides the technical
DOI: 10.2112/JCOASTRES-D-20-00048.1 received 16 April 2020; conditions for damage detection of underwater structures by
accepted in revision 9 October 2020; corrected proofs received
9 December 2020.
dynamic signatures. Many scholars have successively carried
*Corresponding author: [email protected] out research on dynamic response laws and dynamic damage
Ó
Coastal Education and Research Foundation, Inc. 2021 detection of underwater pile foundations. Han et al. (2015) and
 Damage Detection Based on a Multidimensional Dynamic Signature 577

Lu and Jeng (2010) studied the acceleration response law of a

single pile under wave force through numerical simulations
and model tests and analyzed the natural frequency based on
the dynamic response. Padron et al. (2010) studied the upper
dynamic response of inclined piles and pile groups and
analyzed the stiffness and damping of the structure. Sun,
Wang, and Han (2011) studied the vibration of pile foundations
in a high-pile wharf by numerical simulation and model tests
and used modal strain energy to detect the damage of pile
foundations. Asgarian, Aghaeidoost, and Shokrgozar (2016)
used the rate of signal energy using the wavelet packet
transform to detect damage of jacket-type offshore platforms.
However, due to the complexity of marine engineering, the
current damage detection methods generally have the follow-
ing shortcomings.
First, damage detection methods are mostly based on visual
inspection or the simple PauTa criterion, which cannot meet
the needs of automatic analysis of massive monitoring data in
large projects (Cui et al., 2018).
Second, each method has its own emphasis, but each
signature has some shortcomings. Wang, Ren, and Qiao
(2016) noted that detecting damage by the change in a single
signature parameter is not comprehensive and accurate.
Third, the methods have poor reliability. Marine engineering
is under the combined action of wind, wave, current, mechan-
ical, and other dynamic loads, which makes the test noise and
system model error relatively large. Due to the existence of Figure 1. High-pile wharf scene and local damage of pile foundation.
noise, sensor measurement error, and system model error, the
accuracy of dynamic tests is relatively low, so the existing and modal curvature are calculated based on the measured
methods often lead to misjudgment (Cao et al., 2014). modal. The different dynamic signatures are calculated
To address this deficiency, this study formulates a new pile mathematically according to different physical meanings. In
foundation damage detection method. The method constructs a the process of mathematical calculation, the changes of
multidimensional dynamic signature space by fusing tradi- dynamic signatures caused by damage can be reflected, but
tional dynamic signature information to comprehensively the abnormal changes of them caused by measurement error
evaluate the dynamic characteristics of structural units. will be reflected in different positions because of different
Considering that in practical engineering, the damaged units mathematical methods.
represent a small proportion of whole structure, the statistical Therefore, a multidimensional dynamic signature is con-
characteristics of them are distinguishable from most other structed in this paper, with the purpose of strengthening the
units, i.e. the damage units can be detected by finding outliers damage characteristic information and weakening the influ-
which are inconsistent with the overall data. Therefore, in this ence of measurement errors through multidimensional infor-
paper, a robust outlier identification method in the statistical mation fusion, so that the damage location can be detected
calculation is used to realize the automatic and reliable more easily (Guo and Xu, 2018).
detection of pile foundation damage in marine engineering.
The rest of the paper is organized as follows. The ‘‘Methods’’ Construction of a Multidimensional Dynamic Signature
section constructs a multidimensional dynamic signature Space
space. The ‘‘Results’’ section constructs an algorithm for the The damage of pile foundations in marine engineering
automatic damage detection of pile foundations. The ‘‘Analy- generally starts locally, as shown in Figure 1. The studied pile
sis’’ section verifies the effectiveness and reliability of the foundation is divided into n units and n þ 1 nodes, as shown in
method using a finite element model (FEM) of a high-pile wharf Figure 2. Let xi, {/i} (i ¼ 1, 2, ..., n þ 1) be the ith natural
under different damage degrees and different intensities of frequency and mass-normalized modal of the pile foundation,
Gaussian white noise. The ‘‘Discussion’’ section verifies the respectively. According to Jahangiri et al. (2019), modal strain
applicability of the method on a physical experimental model of energy, modal flexibility, and modal curvature analyze damage
a high-pile wharf. characteristics from different perspectives. Additionally, con-
sidering that the high-order modal of large civil engineering
METHODS structures is difficult to obtain and the signal-to-noise ratio
In actual damage detection applications, the existing (SNR) is low, a multidimensional dynamic signature is
dynamic signatures are mostly calculated by easily measured constructed by using the above three signatures based on the
indexes. For example, the square frequency difference is low-order modal. Notably, the signatures that can be used to
derived based on the measured frequency, the modal flexibility, construct a multidimensional dynamic signature are not

Journal of Coastal Research, Vol. 37, No. 3, 2021

578 Zhu et al.

matrix is mainly determined by the low-order modal; that is, as

long as the low-order modal of the structure is measured, the
flexibility matrix with higher accuracy can be obtained.
Therefore, in this paper, it is approximately as follows:

1 1
½F ’ f/ gf/1 gT þ 2 f/2 gf/2 gT : ð3Þ
x21 1 x2

Modal Strain Energy

The first-order modal strain energy of the ith unit is defined
as (Jahangiri et al., 2019):

Ei ¼ f/1 gT ½Ki f/1 g; i ¼ 1; 2; . . . ; n ð4Þ

where, [Ki] is the stiffness matrix of the ith unit.

Modal Curvature
The first-order modal curvature of the ith node is defined as
(Pandey, Biswas, and Samman, 1991):

/i1
1 þ /1iþ1  2/i1
Ci ¼ ; i ¼ 2; . . . n ð5Þ
h2
where, h is the unit length and /i1 is the ith component of the
first-order natural modal {/1}.
Before and after damage, the absolute changes of modal
flexibility, modal strain energy, and modal curvature are DUi ,
DEi , and DCi , respectively:
     D 
DUi ¼ UiD  UiN ; DEi ¼ ED N 
i  Ei ; DCi ¼ Ci  Ci
N
ð6Þ
Figure 2. Schematic diagram of units and nodes in pile foundation.
where, the superscript N represents the signature before the
limited to the above three signatures, and the actual damage and the superscript D represents the signature after
engineering structure can choose certain signatures to con- the damage.
struct the corresponding multidimensional signature accord- In traditional damage detection, the maximum change of
ing to the practical situation. each signature corresponds to the possible damage location. A
multidimensional dynamic signature is constructed by com-
Modal Flexibility bining the three signatures to comprehensively measure the
The structural flexibility matrix [F] is expressed by modal as dynamic characteristics of units. Due to the different magni-
follows: tude of each signature, the maximum and minimum standard-
ization method is adopted:
X
nþ1
1
½F ¼ 2
f/i gf/i gT : ð1Þ DUi  DUmin DEi  DEmin
x
i¼1 i DUi0 ¼ ; DE0i ¼ ; DC0i
DUmax  DUmin DEmax  DEmin
Since the modal flexibility matrix has many elements, to DCi  DCmin
¼ ð7Þ
conveniently study the modal flexibility changes before and DCmax  DCmin
after damage, many scholars have selected some representa- where, the subscript max, min is the maximum and minimum
tive values of the flexibility matrix to detect damage, among of the corresponding signature, respectively. The multidimen-
which the most commonly used are ‘‘diagonal element’’ and sional dynamic signature (MDS) of the ith unit is defined as
‘‘uniform load surface (ULS)’’. In this paper, the ULS method is follows:
adopted, which was proposed by Zhang and Aktan (1995).
 T
According to Zhang’s research (Zhang and Aktan, 1998), the fMDSi g ¼ DUi0 ; DE0i ; DC0i ; i ¼ 2; 3; . . . ; n: ð8Þ
ULS method is less error-sensitive and more reliable. For pile
foundations, the so-called ULS refers to the cross section It should be noted that modal strain energy is a unit
containing the node, and the ULS of the ith node is defined as signature, and modal flexibility and modal curvature are node
follows: signatures. After fusion, they are unified as unit signatures, i.e.
the modal flexibility and modal curvature of the ith node are
X
nþ1 used as the dynamic signature of the ith unit. Because the
Ui ¼ fij ; i ¼ 1; 2; . . . ; n þ 1 ð2Þ division of units and nodes is artificial, the ith node and the ith
j¼1
unit are usually close in the pile foundation position (as shown
where, fij is the ith row and jth column element of the flexibility in Figure 2). For large structures, the purpose of damage
matrix. It can be seen from Equation (1) that the flexibility detection is to find the suspected damaged area and then field

Journal of Coastal Research, Vol. 37, No. 3, 2021

 Damage Detection Based on a Multidimensional Dynamic Signature 579

Figure 3. Multidimensional dynamic damage detection flow chart.

surveys, ultrasonic detection, and other tests can be carried out Considering fMDSi g is the measurement result influ-
to verify the form and size of the damage. Therefore, selecting enced by a variety of relatively small noises, according to
units or nodes for uniform metering will not affect the damage Soong (2004), fMDSi g, i ¼ 2; 3; . . . ; n obeys a three-dimen-
detection practical effect. sional normal distribution. Then, the Mahalanobis distance
di has the following property (Balamurali and Melkumyan,
Damage Detection Based on Outlier Identification 2018):
To find the damaged units, it is necessary to determine
whether there are differences among the features of each unit, d2i ; v2 ð3Þ; i ¼ 2; 3; . . . ; n ð12Þ
which means that the distance between data points in the 2 2
multidimensional space needs to be defined. In this paper, the where, v (3) is the v -distribution with parameter 3.
Mahalanobis distance is used to calculate the degree of Based on classical probability and statistics theory, if given a
difference between multidimensional, which can eliminate significance level a, the 1a confidence interval is obtained as
the coherence interference among variables and is sensitive to follows:
small changes. Let {l}, [R] be the sample mean and sample d2i  v2a ð3Þ; i ¼ 2; 3; . . . ; n: ð13Þ
covariance matrix of fMDSi g, i ¼ 2; 3; . . . ; n, i.e.:
where, v2a ð3Þ is the upper a quantile of the v2-distribution
1 X n
with parameter 3. This interval is the expected fluctuation
flg ¼ fMDSi g; ½R
n  1 i¼2 range of normal data under certain random disturbances. If
1 X n
d2i is beyond the interval, the ith data point is considered an
¼ ðfMDSi g  flgÞðfMDSi g  flgÞT : ð9Þ
n  1 i¼2 outlier, i.e. the judgement basis of damaged unit is as
  follows:
The distance between MDSj ,fMDSk g is expressed as
follows: d2i . v2a ð3Þ; i ¼ 2; 3; . . . ; n: ð14Þ
  
Dist MDSj ; fMDSk g Thus, the v2a ð3Þ is called the ‘‘damage threshold’’.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Leys et al. (2018) showed that the Mahalanobis distance
  T   
¼ MDSj  fMDSk g ½R1 MDSj  fMDSk g : ð10Þ calculation depends on the mean and covariance matrix. The
mean and covariance matrix obtained by using Equation (9) is
According to Leys et al. (2018), the sample mean {l} is the greatly influenced by outliers and seriously lacks robustness,
center of the sample, and outliers are a few points away from i.e. the outliers do not necessarily have a large Mahalanobis
the data center, so whether the data point is an outlier is distance. Therefore, many scholars use the Minimum Covari-
measured by the distance between the data point and the ance Determinant (MCD) method to calculate the mean and
sample center. Based on Equation (10), the distance of the the covariance matrix, and then identify the outliers with
multidimensional signature between the ith unit and the Mahalanobis distance. The MCD method is a robust method
center {l} is obtained by: proposed by Rousseeuw, Driessen, and Abbasidoust (1999) to
di ¼ DistðfMDSi g; flgÞ identify outliers in multidimensional space, which has been
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi widely used in different fields (Balamurali and Melkumyan,
¼ ðfMDSi g  flgÞT ½R1 ðfMDSi g  flgÞ: ð11Þ 2018; Yeager et al., 2019).

Journal of Coastal Research, Vol. 37, No. 3, 2021

580 Zhu et al.

Specifically, the subsamples with sample size h is selected, and

the mean and covariance matrix are calculated based on the
subsample. This process is repeated for all possible subsamples
with the same sample size and finally, the minimum covariance
determinant subsample is selected as the subsample without
outliers. Therefore, the MCD method has strong robustness.

RESULTS
As a summary, a new damage detection method is construct-
ed in this section. To express more conciseness, dMCD is
defined as the Mahalanobis distance based on the MCD
method. The specific damage detection process is as follows:
Step 1: Randomly select subsamples with h points:
  
Hold ¼ fMDSi1 g; fMDSi2 g; . . . ; MDSih ; ð15Þ
Figure 4. The FEM of a high-pile wharf.
and, calculate subsample mean {lold}, covariance matrix [Rold]
by formulas:
The main idea of the MCD method is to find a subsample set
1X h  
H without outliers. The elements in H are used to calculate the flold g ¼ MDSij ; ð16:1Þ
h j¼1
mean and covariance to eliminate the influence of outliers.

Figure 5. Elevation view of first order modal of the FEM.

Journal of Coastal Research, Vol. 37, No. 3, 2021

 Damage Detection Based on a Multidimensional Dynamic Signature 581

Figure 6. MSE difference of No. 3 pile under different damage degree based Figure 7. ULS difference of No. 3 pile under different damage degree based
on the FEM calculation. on the FEM calculation.

The flow chart of damage detection is shown in Figure 3.

1X h     T
½Rold  ¼ MDSij  flold g MDSij  flold g : ð16:2Þ
h j¼1
ANALYSIS
Step 2: Calculate the Mahalanobis distance of the ith unit by The threshold selection of the judgement basis and the
Equation (11): correctness of the method based on FEM are analyzed in this
section.
di ¼ DistðfMDSi g; flold gÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Choice of Parameters
¼ ðfMDSi g  flold gÞT ½Rold 1 ðfMDSi g  flold gÞ: ð17Þ It can be seen from the construction process of the damage
detection method based on the multidimensional dynamic
The Mahalanobis distances are sorted from small to large, let
signature that the subsample size h and significance level a
dk1, dk2, ..., dkh be the first h minimum distance points, and
    need to be given, which determine the value of dMCDi2 and the
construct a new subsample: Hnew ¼ MDSk1 ; MDSk2 ;
  detection result. The two parameters should be selected
. . . ; MDSkh g:
according to statistical theory, structural health status, and a
The mean {lnew} and covariance matrix [Rnew] are calculated
large number of engineering test results. Different structures
based on Hnew by formula as follows:
may have different parameters, just as in a medical examina-
1X h   tion, people of different ages have different health judgement
flnew g ¼ MDSkj ; ð18:1Þ values (Mo et al., 2004). In the absence of a large number of
h j¼1
engineering cases to determine the selection of two parameters,

1X h     T
½Rnew  ¼ MDSkj  flnew g MDSkj  flnew g :
h j¼1
ð18:2Þ

Step 3: Compare det([Rnew]) with det([Rold]), where det() is

the determinant operation. If det([Rnew]) , det([Rold]), let {lold}
¼ {lnew}, [Rold] ¼ [Rnew], and then go to Step 2; otherwise, the
covariance matrix with the minimum determinant is found, let
{lMCD} ¼ {lnew}, [RMCD] ¼ [Rnew], and go to Step 4.
Step 4: Calculate the final Mahalanobis distance of each data
point based on {lMCD} and [RMCD]:

dMCDi ¼ DistðfMDSi g; flMCD gÞ

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ ðfMDSi g  flMCD gÞT ½RMCD 1 ðfMDSi g  flMCD gÞ:
ð19Þ

According to Equation (14), the judgement basis is as follows: Figure 8. MC difference of No. 3 pile under different damage degree based
on the FEM calculation.
dMCD2i . v2a ð3Þ: ð20Þ

Journal of Coastal Research, Vol. 37, No. 3, 2021

582 Zhu et al.

Figure 9. Multidimensional dynamic signature damage detection results of No. 3 pile under different damage degree based on the FEM calculation.

damage detection should aim to identify the suspected

damaged area and then conduct on-site detection and confir-
mation. According to statistical theory and the structure
considered here, a is selected as 0.05 and h is selected as the
rounding of 0.8n for the numerical simulation and model test.

Analytical Verification Based on FEM

The numerical simulation and physical test adopt the same
model. The high-pile wharf model is 2.05 m long, 0.9 m wide,
and 1.65 m high. The model has three spans and the span
spacing is 0.65 m. The front and middle of the wharf consist of
single straight piles, and the rear consists of a pair of inclined
piles. The pile body is made of steel pipe piles, each with a
Table 1. Success rate of damage detection under different SNRs and
damage conditions.

SNR 5% Damage 10% Damage 20% Damage 30% Damage

70 dB 61.6% 75.5% 80.1% 92.1%
Figure 10. MSE difference of No. 3 pile under different damage degree based
80 dB 73.5% 83.4% 85.4% 93.4%
on the FEM calculation under noise.
90 dB 74.2% 96.7% 98.0% 99.3%

Journal of Coastal Research, Vol. 37, No. 3, 2021

 Damage Detection Based on a Multidimensional Dynamic Signature 583

Figure 11. ULS difference of No. 3 pile under different damage degree based Figure 12. MC difference of No. 3 pile under different damage degree based
on the FEM calculation under noise. on the FEM calculation under noise.

diameter of 0.06 m and a thickness of 2 mm. Concrete is poured two endpoints of unit 5, so it can be determined that unit 5 is
on the cross beams, longitudinal beams, and slabs of the wharf. damaged. The modal strain energy of unit 5 also has the largest
In the high-pile wharf FEM, all components adopt solid change, and unit 5 can also be determined as a damaged unit. It
elements. The material parameters are as follows: for the can also be seen that the greater the degree of damage, the
concrete, the density is 1950 kg/m3, the elastic modulus is 3.6 3 greater the differences of the three dynamic signatures, and
1010 Pa, and Poisson’s ratio is 0.2; for the steel pipe piles, the the comparison of dynamic signature differences can reflect the
density is 7850 kg/m3, the elastic modulus is 2.1 3 1011Pa, and degree of structural damage.
Poisson’s ratio is 0.3. The finite element mesh division and pile Figure 9 shows the results of the multidimensional dynamic
number in the front row are shown in Figure 4. This paper signature damage detection under different damage degrees.
focuses on the damage detection of the pile body above the mud Under different damage degrees, the dMCD2 values of unit 5
surface. Considering that the actual displacement under the are all greater than the threshold, so unit 5 is automatically
consolidation depth of the pile body is zero in actual detected as a damaged unit. This is consistent with the results
engineering, and referring to the method in Su et al. (2018), of the FEM calculation, which verifies the correctness of the
the constraint of soil above the pile foundation consolidation multidimensional dynamic signature damage detection meth-
point is simplified to that of 0.1 m concrete. The concrete and od. It can also be seen that the dMCD2 value of the damaged
the bottom of the pile foundation are completely consolidated. unit is much larger than those of undamaged units, and the
The No. 3 pile in the front row is used to simulate damage. dMCD2 values of units near the damaged unit are greater than
The free pile length between the pier panel and the bottom those far away from it. For example, under 5% damage, the
constraint of No. 3 pile is 1.3 m, which is divided into 13 units, dMCD2 value of unit 5 is 26.71, the dMCD2 values of unit 4 and
each 0.1 m in length. The damage is set in unit 5, and the unit 6 are 7.02 and 7.43, and the dMCD2 values of units far
degree of damage is set as 5%, 10%, 20%, and 30%. The damage away from unit 5 are approximately 1. It should be noted that
to the unit is realized by reducing its stiffness. for each damage degree, due to the normalization of multidi-
The first-order modal of the FEM is shown in Figure 5. It can mensional dynamic signature damage detection process, the
be seen that the first-order modal of the wharf is mainly in the dMCD2 values of different damage degrees have no significant
X-direction. The X-direction component of the first-order modal meaning for comparison. For example, the dMCD2 values of
is used to calculate the modal strain energy and modal unit 5 are all approximately 27 under 5%, 10%, 20%, and 30%
curvature. The X-direction components of the first-order modal damage. The multidimensional dynamic signature damage
and second-order modal are used to approximately calculate detection realizes the automation detection of damage location,
the modal flexibility. The purpose of using the main component but it cannot reflect the damage degree of the structure
of modal is to facilitate the sensor arrangement and modal significantly.
analysis in the experimental model, and to promote the
application of the research results. The changes of the three DISCUSSION
dynamic signatures are shown in Figures 6–8. The new method is discussed in this section about the
From Figures 6–8, it can be seen that the ULS, modal strain
sensitivity to noise and the application in practical engineer-
energy, and modal curvature identify the damage position of
ing.
the pile foundation well. The differences of the three dynamic
signatures all have a ‘‘peak’’ in the damage location. The Sensitivity to Noise
differences of the ULS and modal curvature are the largest at To verify the reliability of the new method, the MATLAB
node 5 and the second largest at node 6. Nodes 5 and 6 are the function Wn ¼ awgn(W0, SNR) was used to incorporate 90 dB,

Journal of Coastal Research, Vol. 37, No. 3, 2021

584 Zhu et al.

Figure 13. Multidimensional dynamic signature damage detection results of No. 3 pile under different damage degree based on the FEM calculation under noise.

80 dB, 70 dB Gaussian white noise into the modals of the FEM

under each damage degree, where Wn and W0 denote noisy and
clean modals, respectively. SNR is the signal-to-noise ratio.
Under each damage degree and each SNR, 1000 modals were
randomly generated and the multidimensional dynamic signa-
ture damage detection was carried out. The success rate of
damage detection was obtained, as presented in Table 1. It can
be seen from Table 1 that under the influence of the same SNR
noise, the greater the damage degree, the higher the success
rate. That is because the smaller the damage degree, the
smaller the change of dynamic signature will be. If the noise is
too large, the dynamic signature change will be easily drowned
by noise so as to reduce the success rate of damage detection
due to misjudgment. On the contrary, the greater the damage
degree, the greater the dynamic signature change; under the
same noise, the dynamic signature change is not easy to be
drowned so as to increase the success rate of damage detection.
When the damage degrees were larger than 10% and SNRs
were larger than 80 dB, the success rates were greater than
80%. That is, the new method has strong reliability.
Figure 14. Physical model of a high-pile wharf. A successful case of damage detection under 70 dB noise is
randomly selected to demonstrate the superiority of the new

Journal of Coastal Research, Vol. 37, No. 3, 2021

 Damage Detection Based on a Multidimensional Dynamic Signature 585

Figure 16. Comparison of intact modal between the experiment and the
FEM of No. 3 pile.

realized by reducing the section size. The frequencies and

modals are obtained by analyzing the vibration response of the
pile foundation acquired by the vibration pick-ups. Thirteen
acceleration vibration pick-ups were arranged at 1–13 nodes.
The arrangement of vibration pick-ups is shown in Figure 15.
The frequency comparison between the experiment and the
FEM is shown in Table 2. The experimental frequencies are
very close to the FEM simulation, and the maximum relative
error is less than 4%. The similarity between the experiment
mode and the FEM mode is shown in Figure 16 and Figure 17
(due to space limitations, only the first modal comparison of
intact and 10% damage were presented). The experiment
Figure 15. Acceleration vibration pick-up arrangement of No. 3 pile.
modes and the FEM modes are very similar, i.e. the dynamic
signatures of the experiment and the FEM are consistent.
method. Figures 10–12 show the three dynamic signature After obtaining the modals under intact and damaged
damage detection diagrams. As seen from the figures, under degrees, the difference of ULS, modal strain energy, and modal
the influence of noise, a single signature is prone to misjudg- curvature under each damage degree were calculated. The
ment, such as unit 4 under 5% and 20% damage in Figure 11, results are shown in Figures 18–20. It can be seen that the
unit 6 under 10% damage, unit 12 under 30% damage in Figure differences of the three dynamic signatures all have a ‘‘spike’’
11, and unit 10 under 30% damage in Figure 12. Figure 13 in the damage location. Additionally, due to the existence of
shows the results of the multidimensional dynamic signature experimental noise, boundary conditions, modal analysis
damage detection under different damage degrees. Unit 5 was errors, etc., the dynamic signature difference curve is dis-
accurately detected as a damaged unit under each damage turbed, which causes interference with the detection results
degree. and even leads to misjudgment. In Figure 18, the modal strain
energy difference of unit 4 in the 30% damage is greater than
Practical Application
that of unit 5, so it will be misjudged that the damage in unit 4
The physical model of a high-pile wharf has the same
is more serious than that of unit 5. There are several ‘‘spikes’’
geometric dimensions, unit numbers, and node numbers of the
in the single signature detection diagram, such as unit 10
pile body as the FEM. The upper beam, carling, and panel are
under 10% damage in Figure 19 and unit 8 under the 20%
poured with C60 concrete, and the pile body is made of Q235
damage in Figure 18, which can also be detected as damaged
steel tube pile. The bottom of pile is fixed with 0.1 m concrete,
units by visual inspection. Abrupt changes occurred at some
and the bottom concrete is consolidated on the ground. The
boundaries, such as unit 12 under 30% damage in Figure 18
physical model is shown in Figure 14. The unit damage is
and unit 12 under 30% damage in Figure 20, which can
Table 2. Comparison of the experiment and the numerical simulation significantly interfere with damage detection results. There-
frequency (Hz).
fore, the traditional damage detection method based on a single
dynamic signature will cause misjudgment due to interferences
Order FEM Experiment Error Relative Error
in practical application. At the same time, in actual engineer-
1st 10.35 10.74 0.39 3.77%
ing, the dynamic signature cannot qualitatively reflect the
2nd 31.22 31.25 0.03 0.10%
damage degree of the structure. For example, in Figure 19, the

Journal of Coastal Research, Vol. 37, No. 3, 2021

586 Zhu et al.

Figure 17. Comparison of modal under 10% damage degree between the Figure 19. ULS difference of No. 3 pile under different damage degree based
experiment and the FEM of No. 3 pile. on the experiment.

difference of ULS under 5% damage is greater than that under following advantages over traditional single signature dynamic
10% damage and 20% damage; in Figure 18 and Figure 19, the damage detection.
differences of the dynamic signatures under 5% damage and First, the automation of damage detection is realized.
10% damage are similar. Because the threshold is set in the multidimensional dynamic
Figure 21 shows the damage detection results of multidi- signature damage detection, the damage locations exceeding
mensional dynamic signature under different damage degrees. the threshold value are automatically detected. Second, the
Table 3 lists the comparison of damage detection results damage location feature is more obvious, and it is easier to
distinguish the damaged unit from the undamaged unit. Third,
between traditional methods and the multidimensional dy-
through information fusion, the interferences of random error,
namic signature method. It can be clearly found that the
model error, and boundary conditions are weakened, and the
damage locations of each damage degree can be accurately
multidimensional dynamic signature damage detection has
detected. It should be noted that under the 5% damage, unit 6 is
strong reliability in actual damage detection.
also detected as damage. Due to the large change in dynamic
characteristics of unit 6 as the adjacent position of the damage, CONCLUSIONS
under the influence of experimental errors, it is acceptable that The traditional methods of pile foundation damage detection
unit 6 is detected as the suspected damaged area for the field have considerable limitations in the application of marine
survey. engineering. In this paper, a new method for pile foundation
From Figures 18–21 and Table 3, it can be seen that the damage detection is formulated. The method constructs a
multidimensional dynamic signature damage detection has the multidimensional dynamic signature space by fusing multi-

Figure 18. MSE difference of No. 3 pile under different damage degree based Figure 20. MC difference of No. 3 pile under different damage degree based
on the experiment. on the experiment.

Journal of Coastal Research, Vol. 37, No. 3, 2021

 Damage Detection Based on a Multidimensional Dynamic Signature 587

Figure 21. Multidimensional dynamic signature damage detection results of No. 3 pile under different damage degree based on the experiment.

variate information and detects damage in the multidimen- weakened in the information fusion. This is a new attempt
sional space by using the Mahalanobis distance based on the and exploration of the damage detection of pile foundations in
MCD method. The effectiveness and reliability of the method marine engineering. Future research should focus on the
are verified by the FEM of a high-pile wharf under different subsample size of h and the reasonable selection of a threshold
damage degrees and different intensities of Gaussian white for the damage detection of pile foundations with different
noise. The applicability of the method is then experimentally health degrees and forms of damage and apply the method to
validated by the physical experimental model of a high-pile the detection of multiple damage degrees.
wharf. This method constructs the multidimensional dynamic
signature by fusing a lot of signature information so that the ACKNOWLEDGMENTS
features of damaged units are strengthened and the influences This work was funded by the National Natural Science
of the measurement error in the undamaged area are Foundation of China (Grant Nos. 51709093 and No.
Table 3. Comparison of damage detection results between traditional methods and multidimensional dynamic signature method.

Multidimensional
Damage Actual Modal Flexibility Modal Strain Energy Modal Curvature Dynamic Signature
Degree Damage Position Detection Result Detection Result Detection Result Detection Result
5% Unit 5 Unit 3, 5, 11 Unit 5, 11 Unit 5, 11 Unit 5, 6
10% Unit 5 Unit 5, 7, 10 Unit 5, 8, 11 Unit 5, 9 Unit 5
20% Unit 5 Unit 5, 8 Unit 5, 8, 11 Unit 5, 9 Unit 5
30% Unit 5 Unit 5, 9 Unit 4, 9 Unit 5, 10 Unit 5

Journal of Coastal Research, Vol. 37, No. 3, 2021

588 Zhu et al.

51679081), Fujian Provincial Department of Transportation Maia, N.M.M.; Almeida, R.A.B.; Urgueira, A.P.V., and Sampaio,
Science and Technology Development Project 201708, and R.P.C., 2011. Damage detection and quantification using trans-
missibility. Mechanical Systems and Signal Processing, 25(7),
Hohai University Student Innovation and Entrepreneurship
2475–2483.
Training Project 201910294014Z. Mo, F.; Choi, B.C.K.; Li, F.C.K., and Merrick, J., 2004. Using health
utility index (HUI) for measuring the impact on health-related
LITERATURE CITED quality of life (HRQL) among individuals with chronic diseases.
Asgarian, B.; Aghaeidoost, V., and Shokrgozar, H.R., 2016. Damage The Scientific World Journal, 4, 746–757.
detection of jacket type offshore platforms using rate of signal Padron, L.A.; Aznarez, J.J.; Maeso, O., and Santana, A., 2010.
energy using wavelet packet transform. Marine Structures, 45, 1– Dynamic stiffness of deep foundations with inclined piles. Earth-
21. quake Engineering and Structural Dynamics, 39(12), 1343–1367.
Ahmad, S.; Waleed, A., and Virk, U.S., 2019. Multiple damage Pandey, A.K.; Biswas, M., and Samman, M.M., 1991. Damage
detections in plate-like structures using curvature mode shapes detection from changes in curvature mode shapes. Journal of
and gapped smoothing method. Advances in Mechanical Engineer- Sound and Vibration, 145(2), 321–332.
ing, 11(5), 1687814019848921. Qian, X.; Cao, M.; Su, Z., and Chen, J., 2012. A hybrid particle swarm
Balamurali, M. and Melkumyan, A., 2018. Detection of outliers in optimization (PSO)-simplex algorithm for damage identification of
geochemical data using ensembles of subsets of variables. Math- delaminated beams. Mathematical Problems in Engineering, 2012,
ematical Geosciences, 50(4), 369–380. 1–11.
Cao, M. and Qiao, P., 2009. Novel Laplacian scheme and multi- Rousseeuw, P.J.; Driessen, K.V., and Abbasidoust, F., 1999. A fast
resolution modal curvatures for structural damage identification. algorithm for the minimum covariance determinant estimator.
Mechanical Systems and Signal Processing, 23(4), 1223–1242. Technometrics, 41(3), 212–223.
Cao, M.; Radzieński, M.; Xu, W., and Ostachowicz, W., 2014. Soong, T.T., 2004. Fundamentals of Probability and Statistics for
Identification of multiple damage in beams based on robust Engineers. New York: Wiley, pp. 142–150.
curvature mode shapes. Mechanical Systems and Signal Process- Su, J.; Luan, S.; Zhang, L.; Zhu, R., and Qin, W., 2019. Partitioned
ing, 46(2), 468–480. genetic algorithm strategy for optimal sensor placement based on
Cui, H.; Xu, X.; Peng, W.; Zhou, Z., and Hong, M., 2018. A damage structure features of a high-piled wharf. Structural Control and
detection method based on strain modes for structures under Health Monitoring, 26(1), e2289.
ambient excitation. Measurement, 125, 438–446. Sun, X.; Wang, Y., and Han, Y., 2011. Research on the dynamic
Ding, X.; Liu, H.; Liu, J., and Chen, Y., 2011. Wave propagation in a characteristics of the high-piled wharf considering soil-structure
pipe pile for low-strain integrity testing. Journal of Engineering interaction. Proceedings of the 2nd International Conference on
Mechanics, 137(9), 598–609. Mechanic Automation and Control Engineering (Mongolia, China),
Guo, T. and Xu, Z., 2018. Data fusion of multi-scale representations pp. 6791–6794.
for structural damage detection. Mechanical Systems and Signal Wang, S.; Ren, Q., and Qiao, P., 2016. Structural damage detection
Processing, 98, 1020–1033. using local damage factor. Journal of Vibration and Control, 12(9),
Han, Y.; Zhan, J.; Su, W.; Li, Y.S., and Zhou, Q., 2015. Numerical 955–973.
simulation of in-line response of a vertical cylinder in regular
Wang, Y.-z. and He, L.-l., 2017. Simplified calculation methods for all-
waves. Journal of Coastal Research, 31(4), 879–891.
vertical-piled wharf in offshore deep water. China Ocean Engi-
Huo, S.; Chao, Y.; Dai, G., and Gong, W., 2015. Field test research of
neering, 31(2), 182–191.
inclined large-scale steel pipe pile foundation for offshore wind
Yeager, M.; Gregory, B.; Key, C., and Todd, M., 2019. On using robust
farms. Journal of Coastal Research, 73, 132–138.
Mahalanobis distance estimations for feature discrimination in a
Jahangiri, M.; Najafgholipour, M.A.; Dehghan, S.M., and Hadianfard,
damage detection scenario. Structural Health Monitoring, 18(1),
M.A., 2019. The efficiency of a novel identification method for
structural damage assessment using the first vibration mode data. 245–253.
Journal of Sound and Vibration, 458, 1–16. Zhang, Z. and Aktan, A.E., 1995. The damage indices for the
Jassim, Z.A.; Ali, N.N.; Mustapha, F., and Abdul Jalil, N.A., 2013. A constructed facilities. Proceedings of the 13th International Modal
review on the vibration analysis for a damage occurrence of a Analysis Conference (Nashville, Tennessee), pp. 1519–1529.
cantilever beam. Engineering Failure Analysis, 31, 442–461. Zhang, Z. and Aktan, A.E., 1998. Application of modal flexibility and
Leys, C.; Klein, O.; Dominicy, Y., and Ley, C., 2018. Detecting its derivatives in structural identification. Research in Nonde-
multivariate outliers: Use a robust variant of the Mahalanobis structive Evaluation, 10(1), 43–61.
distance. Journal of Experimental Social Psychology, 74, 150–156. Zhao, Y.; Noori, M., and Altabey, W.A., 2017. Damage detection for a
Liu, D.Z.; Ren, J.M., and Yang, Y., 2012. Ultrasonic sector scanning beam under transient excitation via three different algorithms.
method in the application of the accurate determination of the Structural Engineering and Mechanics, 6(1), 803–817.
concrete bored pile defect position. Advanced Materials Research, Zhu, H.; He, B., and Chen, X.-q., 2005. Detection of structural damage
430–432, 1128–1132. through changes in frequency. Wuhan University Journal of
Lu, J. and Jeng, D., 2010. Dynamic response of an offshore pile to Natural Sciences, 6(1), 128–132.
pseudo-Stoneley waves along the interface between a poroelastic Zhu, H.; Yao, L., and Li, J., 2019. Influence factors on the seismic
seabed and seawater. Soil Dynamics and Earthquake Engineering, behavior and deformation modes of gravity retaining walls.
30(4), 184–201. Journal of Mountain Science, 16(1), 168–178.

Journal of Coastal Research, Vol. 37, No. 3, 2021