Journal of Nondestructive Evaluation (2023) 42:29

https://doi.org/10.1007/s10921-023-00940-2

Defect Characterization Method for Bridge Cables Based on Topology

of Dynamical Reconstruction of Magnetostrictive Guided Wave Testing
Signals
Shuyu Duan1 · Xinjun Wu1 · Junjie Wang1 · Yiqing Zou2 · Lijun Jiang2 · Yaolin Wei2

Received: 30 November 2022 / Accepted: 26 February 2023

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023

Abstract
Multi-wire cables are widely used in suspension bridges and cable-stayed bridges as primary load bearing structural elements.
Broken Wires in cables can lead to catastrophic accidents such as bridge collapse. Magnetostrictive guided wave testing
technology has been employed to detect the broken wire defects in multi-wire cables, and the defect size is estimated by
analyzing the defect echo signals. However, there are many studies on the guided wave testing for the seven-wire steel strands
but fewer for the bridge cables which have a large number of wires. Moreover, the relationship between the guided wave
testing signal features and the defect size of multi-wire structures is imprecise, which means the defect size estimated by the
features may deviate significantly from the real defect size. In this paper, large-scale topological features are extracted by using
persistent homology from the dynamical reconstruction topology of the guided wave testing signals to characterize broken
wire defects in the bridge cable. The broken wire experiments were performed on a 61-wire cable. The experimental results
show a good linear relationship (the goodness of fit 0.9946) between the large-scale topological features and the number of
broken wires in the cable. It indicates that it is feasible to extract topological features from the topological domain of the
testing signals to characterize the broken wire defects of bridge cables.

Keywords Magnetostrictive guided wave testing · Dynamical reconstruction · Phase space · Persistent homology · Bridge
cable

1 Introduction

The multi-wire cable has the advantages of high strength,

light weight, good durability and easy operation. Therefore, it
B Xinjun Wu is widely used as the primary load bearing structural elements
[email protected] in suspension bridges and cable-stayed bridges. Cables are
Shuyu Duan exposed to many harmful long-term effects, such as fatigue
[email protected] load and environmental erosion. As a result, cables in high
Junjie Wang stress states are prone to corrosion or even broken wires,
[email protected] which may cause bridge collapse. Therefore, nondestructive
Yiqing Zou testing is crucial to ensure the proper structural performance
[email protected] of suspension bridges and cable-stayed bridges.
Lijun Jiang Various nondestructive testing (NDT) methods have been
[email protected] used to inspect cables, such as visual inspection [1], mag-
Yaolin Wei netic flux leakage testing [2, 3], acoustic emission testing
[email protected] [4, 5], vibration testing [6] and guided wave testing [7–9].
1 The visual inspection method is most commonly used but
School of Mechanical Science & Engineering, Huazhong
University of Science and Technology, Wuhan 430074, China can only inspect cable surfaces. The magnetic flux leakage
2 testing has high testing accuracy for local faults and cross-
Liuzhou OVM Machinery Co., LTD, Liuzhou 545005,
Guangxi, China sectional area loss in cables. However, this localized testing

0123456789().: V,-vol 123

 29 Page 2 of 18 Journal of Nondestructive Evaluation (2023) 42:29

method is less efficient and can only inspects flaws in the free that there is a large error between the defect size evaluated
section of the cable but not in the anchorage zone. Acoustic by these features and the real size of the defect.
emission testing is a successful dynamic nondestructive test- Topology plays a particular role in dynamics, since many
ing technique, but it is not effective in detecting pre-existing properties of dynamics depend upon topology, such as the
defects in cables.In contrast to the passive nondestructive existence of periodic orbits, transitivity, recurrence, entropy,
testing method of acoustic emission, guided wave testing can etc. It is well known that broken wire defect in multi-wire
be called an active testing method. Especially, the magne- cable leads to changes in the waveform structure of guided
tostrictive guided wave testing method has been widely used wave testing signals. Specifically, the broken wire defect will
in cable inspection due to its capability of long range, high reflect the guided wave signal, and the stronger the reflected
efficiency and non-contact defect detection compared with guided wave signal as the damage increases. These changes
traditional electromagnetic nondestructive testing methods are reflected not only in the time and frequency domains of
[9–14]. the guided wave testing signal, but also in the topology of
It is important to investigate propagation characteristics its dynamics. However, in guided wave testing, reconstruct-
of magnetostrictive guided wave and the interaction with ing the full dynamics from the 1-dimensional testing signals
defects for quantitative testing. Related studies have been can be a real challenge. According to embedding theorems,
conducted first on steel strands [15–18]. Then, the appli- the full dynamics may be reconstructed via delay-coordinate
cation of magnetostrictive guided waves to detect broken reconstruction [23]. The full dynamics reconstruction means
wires in bridge cables was studied [2, 13, 14]. Many fea- that it needs to reconstruct the dynamics which is diffeo-
ture extraction techniques have been proposed. The defect morphic—not homeomorphic—to the underlying dynamics
signal features are obtained from time-domain, frequency- manifold for the guided wave testing signal. The success of
domain and joint time–frequency domain of the guided wave the reconstruction procedure depends on several free param-
testing signals. In the time domain, the amplitude/peak-to- eters, such as reconstruction dimension and time delay. There
peak amplitude of the defect echo signal is often used to are many creative strategies developed to select these param-
characterize defects. Experiments in [17] showed that the eters, but these methods require good data and input from
amplitude of the notch defect echo in the helical wire or in a human expert. Since the guided wave testing signals have
the central wire of the seven-wire steel strand was almost finite length, nonzero sampling interval, limited precision and
linear with the defect depth. The transmitted amplitude of may be contaminated by noise, it is hard to reconstruct the full
the L(0,1) mode guided wave in a single wire decreases dynamics. Therefore, dynamic invariants like the correlation
monotonically with increasing notch depth and the reflected dimension and Lyapunov exponent are difficult to calculate
amplitude increases monotonically with increasing notch exactly. Garland [24] had shown that if one is only interested
depth [8]. In [16], the increasing number of broken wires in in topological structure, it is not necessary to reconstruct the
the seven-wire steel prestressing strand has an almost linear full dynamics for the scalar time series. And it is possible to
relationship with the ratio of the increase in the peak-to-peak compute the topology of an invariant set of a dynamical sys-
amplitude of the defect signals. This means the peak-to-peak tem from a low-dimensional reconstruction of a scalar time
amplitude can be used to quantify the broken wires of the series. Therefore, this work reconstructs the 2-dimensional
prestressing strand. A series of studies on the application phase space with optimal geometry of the guided wave signal
of wavelet analysis in guided wave signals for multi-wire rather than perform a full dynamics reconstruction.
strands are conducted [15, 19–21]. The variance and the root Persistent homology [25, 26] is a most widely used tool
mean square (RMS) of wavelet coefficient as damage index in topological data analysis (TDA) [27], which is a field that
are functions of notch depth [15]. Furthermore, the mul- combines data analysis, algebraic topology, computational
tidimensional damage index computed from the variance, geometry, computer science, statistics, and other related
RMS, maximum amplitude and peak-to-peak amplitude of areas. Persistent homology has been applied successfully in
the wavelet coefficients are demonstrated to be more effective many different fields, such as image analysis [28], complex
than unidimensional analysis [21]. The wave energy trans- networks [29], large data analysis [30], shape recognition
mission coefficients calculated from wavelet coefficients are [31], time series analysis [32], etc. Here, persistent homol-
presented as a measure for the damage condition (broken wire ogy is employed to obtain a coarse-grained description of
and corrosion) of a seven-wire strand [22]. Most of the exist- the topology of the 2-dimensional phase space. That is, the
ing studies have focused on strands with simple multi-wire topological features – persistences—of the 2-dimensional
structure, and there are fewer studies on bridge cables with a phase space are extracted by persistent homology and dis-
large number of steel wires. In addition, the functional rela- played in the 0-th and 1-th persistent barcodes (PB) [33].
tionship between the extracted features and the defect size of The 0-th PB records the variation of the number of connected
multi-wire structures is imprecise, and in many cases is only components with the connected radius during the complex
a trend rather than an exact functional equation. This means

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 3 of 18 29

filtration, and the 1-th PB records the variation of the num- set whose type is finite metric space, is also called a point-
ber of 1-dimentional holes. The length of the barcode in PB cloud in the TDA literature. The topological information of a
indicates the lifetime (or persistence) of the corresponding point cloud can be obtained by constructing simplicial com-
topology feature. Edelsbrunner [25] classified a topological plexes from the point cloud and investigating the topology
feature to a useful feature or noise depending on its lifetime of these complexes.
or persistence within the filtration. In PBs, usually the long- Given a point cloud X  {x1 , x2 , …, xM } with M sam-
lived barcodes represent significant topological features or ple points inR D , d(xi , xj ) denotes the Euclidean distance
large-scale features, and the short-lived barcodes indicates between xi and xj (xi , x j ∈R D ), for i, j  1, 2, …, M. A
insignificant topological features or noise. In applications, k-simplex σ is defined as the convex hull of k + 1 affinely
the barcodes in PBs can be used directly or indirectly to obtain independent points < x0 , x1 , ..., xk >. For instance, a 0-
valuable information, depending on the specific objectives simplex is a vertex, a 1-simplex is an edge, a 2-simplex
and data characteristics. For example, thirteen topological is a triangle, and a 3-simplex is a tetrahedron, and so on.
features are extracted from PBs as feature vectors for SVM Let r (r ∈R) denote the connected radius. In a point cloud,
to achieve protein classification in [34]. For more informa- a simplex is usually constructed by creating circles with r
tion on topological features based on PBs see [35]. In our around each point. If two circles intersect, the points form an
work, compared with the 2-dimensional phase space of the edge (1-simplex); if three circles intersect, the points form
Gaussian noise signal, a significant structural feature of the a triangle (2-simplex), etc. In this work, the simplices is
2-dimensional phase space geometry of the guided wave sig- combined to construct Vietoris-Rips (VR) complex.  If the
nal is the presence of some big 1-dimensional ‘holes’. This k-simplex < x0 , x1 , ..., xk > satisfies d xi , x j ≤ 2r ,
structural feature is clearly and quantitatively described in 0 < i < j < k, then Rr (X ) is the VR complex with
the 1-th PB. Specifically, each long barcode in 1-th PB repre- vertices in X. When the r gradually increases, the follow-
sents a hole in the 2-dimensional phase space, while the short ing VR complex is larger than all previous ones, that is,
barcodes are noises. The longest barcode in 1-th PB repre- assuming0  r0 < r1 < r2 < · · · < rm , all of the
sents the most significant topological feature and its length VR complexes form a nested sequence called a filtration
is selected to be the large-scale topological feature to char- stream:Rr0 (X ) ⊂ Rr1 (X ) ⊂ · · · ⊂ Rrm (X ).
acterize the defects. Recently, the topological features have The filtration of the VR complexes of X induces a homo-
been used to characterize the severity of steel wires corro- morphism on homology. If ri < rj , Rri (X ) ⊂ Rrj (X ), and a
sion [36]. In this paper, the experiments were performed on a simplicial map f : Rri (X ) → Rrj (X ) induces a homomor-
61-wire bridge cable. The experimental results show that the phic map f k : Hk (Rri ) → Hk (Rr j ), where Hk (·) is the
proposed topological feature increases exactly linearly with homology group. For the VR complex sequence Rr0 (X ) ⊂
increasing the number of broken wires. R-square (the good- Rr1(X )⊂ · · · ⊂ Rrm (X ), the homology
 group sequence is
ness of fit) is 0.9946, which is larger than that between other Hk Rr0 → Hk Rr1 → · · · → Hk Rrm . Informally, k-th
features obtained from the time domain, frequency domain homology group Hk (R) is often used to compute the k-th
and joint time–frequency domain and the number of broken Betti number βk , which is equal to the rank of Hk (R). The
wires. k-th Betti number is an important topological invariant to
The remainder of this paper is organized as follows: In describe the topological features of a space. In fact, β0 gives
Sect. 2, the method of obtaining dynamical reconstruction the number of connected components in R, β1 measures the
with optimal geometry of the guided wave signal is presented, number of 1-dimentional holes, and β2 counts the number of
and persistent homology theory is introduced to capture the 2-dimentional voids, etc. For more detailed information on
large-scale topological feature of the dynamical reconstruc- persistent homology, please refer to literature [25].
tion. Broken-wire experiments on a 61-wire bridge cable are One of the outputs of persistent homology is persistent
performed in Sect. 3. The experimental results and discus- barcode (PB) [33], which is a collection of horizontal line
sions are provided in Sect. 4. A conclusion is finally drawn segments in a plane where the horizontal axis is the connected
in Sect. 5. radius r and the vertical axis is the barcode intervals. For a
fixed r , the number of barcodes in the k-th PB is equal to the
k-th Betti number βk . For one barcode in k-th PB, its length
represents the lifetime of a topological feature, and the left
2 Method end and the right end denote the birth and the death of the
same topological feature. In general, long barcodes represent
2.1 Persistent Homology significant features, while short ones are regarded as noise or
details.
Persistent homology is a powerful tool in TDA that is used to As an example, Fig. 1 shows the VR complexes filtration
study the components and shapes of a discrete data set. A data process of one 2-dimensional point cloud. Obviously, the

123
 29 Page 4 of 18 Journal of Nondestructive Evaluation (2023) 42:29

(a) 6 (b) 6 (c) 6

5 r=0 5 r=0.5 5 r=0.9
4 4 4 1
3 3 3

2 2 2

1 1 1

0 0 0

-1
-1 -1

-2 -2
-2 -2 -1 0 1 2 3 4 5 6 -2 -1 0 1 2 3 4 5 6
-2 -1 0 1 2 3 4 5 6

(d) 6 (e) 6 (f) 6

5 r=1.1 5
r=1.4 5 r=1.8
4 1 4 1 4
3 2 3 3
2 2 2
3 3
1 1 1
0 0 0
-1 -1 -1
-2 -2 -2
-2 -1 0 1 2 3 4 5 6 -2 -1 0 1 2 3 4 5 6 -2 -1 0 1 2 3 4 5 6

(g) (h)
r=3.5
0-th PB

0 1 2 3 4
3

1-th PB 1
2

0 1 2 3 4
r
Fig. 1 An example of persistent homology for a point cloud: a the point barcode and 1-th persistent barcode obtained by persistent homology
cloud (with connected radius r  0); b–g the filtration of VR com- corresponding to the filtration
plexes when r gradually increases from 0.5 to 3.5; h the 0-th persistent

main topological feature of the point cloud is the two ‘holes’ in r creates another hole, as shown in Fig. 1d and marked
of different sizes (Fig. 1a). In fact, the point cloud is formed by ‘2’. But this ‘hole-2’ is persisting a short time and closed
by sampling two circles of radius 1 and 2 and then adding when r  1.192. When r  1.4, the third hole is generated
noise. When the connected radius r  0.5, some points are (marked by ‘3’ in Fig. 1e). Asr continues to increase to 1.8
connected into complexes, as shown in Fig. 1b. As r increases and 3.5, ‘hole-1’ and ‘hole-3’ are closed in turn (Fig. 1f–g).
to 0.9, more points are connected to each other and one hole This process is faithfully recorded by β0 and β1 , which are
is generated (marked by ‘1’ in Fig. 1c). A further increase visualized by 0-th PB and 1-th PB, as shown in Fig. 1h. The

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 5 of 18 29

0-th PB shows that when r increases gradually, the number of function [38] and the average mutual information [39] are
connected components gradually decreases and finally forms commonly used to choose the time delay τ . However, the
one connected component. There are three barcodes in 1-th guided wave testing signals have finite length, nonzero sam-
PB, representing the three holes formed during the complex pling interval, limited precision and may be contaminated
filtration, and the length of the barcodes indicates the dura- by noise. Therefore, it is impossible to obtain an accurate
tion of the holes. Corresponding to the topology of the point reconstruction (diffeomorphism between the true and recon-
cloud, the shortest barcode (marked as ‘2’) can be regarded structed phase space) of the guided wave signals that fully
as noise and the other two barcodes can be considered to be recovers the underlying dynamics. Moreover, the purpose of
topological features of the point cloud, namely two different accurate reconstruction is to calculate dynamic system invari-
size holes. ants such as correlation dimensions and Lyapunov exponents.
In this work, we will use the longest barcode in 1-th PB This work is interest in extracting the topological structure
to represent the large-scale topological feature of the guided of the phase space of the guided wave signal, so requiring
wave signals. The length of the longest barcode is defined as only a homeomorphism is natural and more efficient. Gar-
land [24] pointed out that m can be less than 2d + 1 if the
lmax  max |d − b| (1) large-scale topological features of the phase space are stud-
(b, d)∈B ied. In addition, a large dimension causes a huge number of
computational resources, which may make the topology cal-
where B denotes the 1-th PB, d and b denote the right end culation unfeasible. Therefore, the parameters m  2 in this
and the left end of one barcode. paper. Moreover, the time delay τ is determined by analyzing
the structure of the guided wave signal to obtain the opti-
2.2 Dynamical Reconstruction of Guided Wave mal geometry of the reconstructed phase space. The optimal
Signal geometry means that the phase space is properly unfolded
off the main diagonal. In the magnetostrictive guided wave
The guided wave propagates along the multi-wire cable and testing, a sinusoidal signal of several cycles is employed as
interacts with the cable’s boundary. When the cable’s bound- the excitation signal and the defect signal has a similar struc-
ary conditions change, such as broken wire and corrosion, the ture with that. Usually, it is considered as a sinusoidal signal
wave dynamic characteristics will change. In this work, the of several cycles windowed by a Hanning window in the
reconstructed phase space of guided wave dynamic system structure [40, 41], which can be expressed as
is studied from the point of view of topology to extract the
    
features related to cable damage. 1 2π t 2π t
Phase space is a domain used to represent all the pos- g(t)  A 1 − cos sin (2)
2 NT T
sible states of a dynamic system which is reconstructed
from a small number of observables. The phase space of where T is the period of the sinusoidal signals, A is the ampli-
1-dimentional guided wave testing signal preserves the topo- tude and N is the number of sinusoidal signal cycles. After
logical properties of the underlying dynamical system of triangular transformation, Eq. (2) can be rewritten as the fol-
the guided wave testing system. Many properties of phase lowing form
space—the existence of periodic orbits, transitivity, recur-
rence, entropy, etc.—depend only upon topology. Therefore,    
topology is of great significance in dynamics. For the 1- 1 2π t 1 N + 1 2π t
g (t)  A sin − A sin
deminsional guided wave testing signals, we must begin by 2 T 4 N T
 
reconstructing the full dynamics via delay-coordinate recon- 1 N − 1 2π t
struction. Takens time-delay embedding theory [23] shows − A sin .
4 N T (3)
that success of this reconstruction procedure depends on two
free parameters, the embedding dimension m and the time g(t) can be seen as the superposition of three sinusoidal
delay τ . A number of creative strategies have been devel- signals
oped to choose an appropriate m and τ to reconstruct the  t with the different frequency and the main signal is
sin 2πT . According to Takens time-delay embedding theory,
phase space of time series, which is diffeomorphic—not the parametric equation of the 2-dimensional phase space of
homeomorphic—to the true phase space. Takens time-delay g(t) can be expressed as
embedding theorem states that m should not less than 2d + 1,
where d is the dimension of the motive force system. How-      2π t   2π t 
2 A 1 − cos N T sin T
1
ever, d is difficult to determine for the real-world data sets. g(t)
P  1 2π (t+τ ) .
The false near neighbor algorithm [37] is most popular to g(t + τ ) 2 A 1 − cos NT sin 2π (t+τ
T
)

estimate the embedding dimension m. The autocorrelation (4)

123
 29 Page 6 of 18 Journal of Nondestructive Evaluation (2023) 42:29

(a) 1

0.5

Amplitude/V
0

-0.5

-1
0 1 2 3 4 5
Time/s 10-3

(b)
T/20 T/10 3T/20 T/5 T/4 3T/10 7T/20 4T/5 9T/20 T/2
= 1 1 1 1 1 1 1 1 1 1

g(t+ )0 0 0 0 0 0 0 0 0 0

-1 -1 -1 -1 -1 -1 -1 -1 -1 -1
-1
g(t)
0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1

11T/20 3T/5 13T/20 7T/10 3T/4 4T/5 17T/20 9T/10 19T/20 T

1 1 1 1 1 1 1 1 1 1

g(t+ )0 0 0 0 0 0 0 0 0 0

-1 -1 -1 -1 -1 -1 -1 -1 -1 -1
-1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1
g(t)
-1 0 1 -1 0 1 -1 0 1 -1 0 1

Fig. 2 A schematic showing a a simulated 5-cycle guided wave signal and b the various geometry structures of the 2-dimensional phase space of
this guided wave signal with varying time delay τ (from 20
T T
to T in steps of 20 )

As an example, Fig. 2 shows the reconstructed 2- which is an observable topological structure feature. More-
dimensional phase space of a 5-cycle guided wave signal with over, the geometric characteristics of the guided waves are
A  1 and different τ (τ ranges from 20 T T
to T in steps of 20 ). represented by the distribution and size of the holes. There-
Here it can be observed that, when τ < 4 and 4 < τ ≤ T,
T 3T fore, we focus on the 1-th Betti number (described in 1-th
the phase trajectories lie close to the secondary diagonal of PB) to extract the quantitative feature for broken wire defects.
the reconstruction space; when T4 < τ ≤ T2 and T2 < τ < 3T 4 ,
Persistent homology deals with discrete data sets, so the dis-
the phase trajectories lie close to the principal diagonal of the crete 2-dimensional phase space is required. Therefore, to
reconstruction space. The phase trajectories are in the center reconstruct discrete 2-dimensional phase space with optimal
of the reconstruction space when τ  T4 or 3T 4 (marked in red
geometry, the discrete time delay τ can be written as
boxes in Fig. 2b). Because the guided wave signal is a short-

duration tone burst signal, the phase space reconstructed by fs
τ  T4 is more complete and has a higher density than the τ̂  , (5)
4f
phase space reconstructed by τ  3T 4 . Therefore, for guided
wave signal phase space reconstruction, τ  T4 is the optimal where the symbol · means getting the maximum integer not
option. It’s worth noting that this holds true for other N-cycle more than the operated data, f s is the sampling frequency and
guided wave signals as well. f is the center frequency of the guided wave signal. Here, τ
is dimensionless.
2.3 The Topology of the Dynamical Reconstruction As an example, Fig. 3 shows the difference in the topolog-
ical features of the defect echo signal S(n) in the experiment
Persistent homology is employed to analyze the 2- and the simulated Gaussian noise signal N (n). The defect
dimensional phase space of guided wave signal to extract echo signal shown in Fig. 3a is extracted from the guided
the topological features. It is known that 1-th Betti number wave testing signals from the experiment discussed in Sect. 3.
(β1 ) measures the number of 1-dimenstional holes, where Figure 3d depicts a white noise signal with zero mean and
the longest barcode length represents the largest ‘hole’ in variance of 0.035. These two signals have approximately
the data set. As shown in Fig. 2b, there are many holes equal maximum amplitude. Their discrete phase spaces are
in the 2-dimensional phase space of a guided wave signal, shown in Fig. 3b and e with m 2 and τ  12, where f s 

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 7 of 18 29

(a)
Amplitude(V) 0.04 (b) 0.04 (c)

0.02 0.02

S(n+ )
0 0

-0.02 -0.02

-0.04 -0.04
0 50 100 150 200 -0.04 -0.02 0 0.02 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03

n S(n) r
(d) 0.04
(e) 0.04 (f)
Amplitude(V)

0.02 0.02
N(n+ )
0 0

-0.02 -0.02

-0.04 -0.04
0 50 100 150 200 -0.04 -0.02 0 0.02 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03

n N(n) r
Fig. 3 The difference in the topological features of the defect echo sig- c the 1-th PB of b computed by persistent homology, d a white noise
nal in the experiment and the simulated Gaussian noise signal. a An signal with zero mean and variance of 0.035, e the point cloud of the
experimental guided wave signal from experiment discussed in Sect. 3, noise signal, f the 1-th PB of e computed by persistent homology
b the point cloud of this guided wave signal with m  2 and τ  12,

Fig. 4 The workflow of the

proposed defect characterization Guided Dynamics Phase Persistent Large-scale
method of bridge cables Wave Reconstruction Space Homology Topological Feature

2 MHz and f  40 kHz. Their 1-th PBs computed by persis- 3 Experiment

tent homology algorithm are displayed in Fig. 3c and f. For
the guided wave signal, the longest barcode in red in Fig. 3c Approximately 4950-mm-long parallel multi-wire bridge
represents the large-scale topological feature of Fig. 3b, the cable with a 77 mm diameter were used in this study. The
‘hole’ marked by the red curve with an arrow. For the noise bridge cable is made of 61 epoxy steel wires with diameter
signal, all the barcodes are relatively short and there is no 7 mm bundled together in hexagonal shape and twisted con-
◦ ◦
obvious dominant barcode in the 1-th PB (Fig. 3f) because centrically in the same direction (twisting angle 2 ∼ 4 ),
the points are disordered and scattered in the phase space. then wrapped with high strength polyester tape and tied
In conclusion, the 1-th PB captures the topological struc- tightly, and finally sheathed in a direct hot extruded high
ture features of the phase space and the longest barcode density polyethylene plastic (HDPE).
represents the large-scale topological feature. In this work, A magnetostrictive guided wave testing system [16] was
the length of the longest barcode of the 2-dimensional phase used in the experiments. The excitation and receiving trans-
space of the guided wave testing signals is employed to char- ducers consist of bias magnetizers and coils. The excitation
acterize the broken wire damage in the bridge cables. The transducer operates under the magnetostrictive effect and the
workflow of the proposed method is shown in Fig. 4. receiving transducer operates under inverse magnetostrictive
effect. Both the coils of the exciter and receiver have a layer
and 40 turns of AWG No. 26 enameled wire approximately
50 mm long and 78 mm diameter. The permanent bias magne-
tizer consists of an armature and two magnets. The exciting

123
 29 Page 8 of 18 Journal of Nondestructive Evaluation (2023) 42:29

Fig. 5 Experimental setup: (a) 4950mm

a broken wires and transducer
layout in the multi-wire steel 1000mm 1000mm 1500mm
cable, b Photo of the multi-wire
steel cable experimental setup

Broken wires Excitation Receiving

transducer transducer

(b)

Receiving
Excitation transducer transducer
61-wire cable Broken wires

Magnetostrictive guided wave testing instrument

coil induced an alternate axial magnetic field by means of the 61-wire cable. The nine wires in the outermost layer of
an alternating current, and the bias magnetic field provided the cable were cut off one by one and the order of cutting
a static axial magnetic field. The longitudinal mode guided these nine wires is shown in Fig. 6a, where Num denotes the
wave is generated with the interaction of these two fields. number of broken wires. A sample (Num  9) is depicted in
The working flow of the testing system is described as fol- Fig. 6b. Table 1 lists the corresponding cross-sectional area
lows. The computer controls the signal generator to generate loss ratios (denoted by Alost ). Every time a wire was cut, a
a sinusoidal tone burst by program. The sinusoidal tone burst guided wave test is carried out according to the experimental
is amplified by a power amplifier and then loaded on the exci- layout in Fig. 5.
tation coil. The amplified sinusoidal tone burst excites the However, there is no closed-form analytical solution that
excitation coil to induce an alternating magnetic field in the describes wave propagation in a multi-wire steel cable.
cable. This alternating magnetic field interacts with the static Therefore, the optimal exciting frequency is determined by
axial magnetic field provided by the bias magnetic field to sweeping frequency method [42]. The longitudinal mode
generate longitudinal mode guided waves propagating along with 50 kHz is chosen, which is sensitive to defects and has
the cable. The receiving coil can capture the changes of the acceptable dispersion. In this experiment, 3-cycle toneburst
magnetic induction of the cable induced by the strain (or at 50 kHz was generated as the excitation signal. Typically,
stress) associated with the elastic wave. This changing mag- 200 successive signals were captured for each measurement
netic induction induces an electrical signal. The receiving and these were averaged to improve the SNR. The testing
signal is amplified by the preamplifier, and then converted to signals are sampled at 2 MHz.
the digital signal by the analog-to-digital converter (ADC) to
the computer.
The experimental setup is displayed in Fig. 5. In order to
obtain separate defect echo signals, the receiving and excita- 4 Results
tion transducers were clamped to the cable at 1500 mm and
2500 mm from the right end, respectively. And at a distance 4.1 Guided Wave Testing Signals
of 1000 mm from the excitation transducer on the left side of
the cable, broken wire defects were machined by saw-cutting. According to the experimental setup shown in Fig. 5, the first
Thus, the defect echo signal is prevented from overlapping four propagation paths of the guided wave signals in the 61-
with the electromagnetic pulse signal or left end and right wire bridge cable are shown in Fig. 7. The travel distances are
end echo signals. It is worth noting that before making the 1000 mm, 3000 mm, 4000 mm and 5900 mm, respectively.
defects, the polyester tape and the HDPE sheath at the cor- The guided wave testing signal for the defect-free cable is
responding location were removed first by cutter (showed in displayed in Fig. 8. It is noted that the testing signal is fil-
Fig. 6b). Figure 6a displays the cross-sectional schematic of tered based on a bandpass filter using the Parks-McClellan
algorithm [43]. The signal EM is the electromagnetic pulse

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 9 of 18 29

Table 1 The number of broken wires and the corresponding cross-sectional area loss ratio

N um 1 2 3 4 5 6 7 8 9

Alost (%) 1.64 3.28 4.92 6.56 8.20 9.84 11.45 13.11 14.75

(a) The outer PE Num=1 Num=2 Num=3

The
steel wires
The inner PE

Num=4 Num=5 Num=6 Num=7 Num=8 Num=9

(b) The inner PE The outer PE

The broken wires

Fig. 6 The diagram of the broken wire defects: a the cross-sectional schematic of the 61-wire cable, and the order of cutting nine wires and b the
photo of 9 wires cut

Fig. 7 The propagation paths of

the guided wave signal in the
cable

123
 29 Page 10 of 18 Journal of Nondestructive Evaluation (2023) 42:29

15
4.2 Propagation Properties of the Guided Wave
10 Testing Signals
Amplitude(V)

0 In the experiment, each wire was cut, a guided wave test is

-5 performed on the cable. Figure 10 displays some normalized
-10
guided wave testing signals (excluding the electromagnetic
EM P End1 End2 pulse signal) for the defect-free and different number of bro-
-15
0 0.5 1 1.5 ken wires cable. For the defect-free case, the signal interval
Time(s) 10-3
between P and End1 is regarded as the background noise.
Fig. 8 The guided wave testing signal of defect-free cable When there is one broken wire in the cable, the defect echo
signal is very weak. One broken wire in the cable is equiv-
alent to a loss of 1.64% of the cable’s cross-sectional area.
However, the designed transducer is insensitive to the dam-
15
age when the cross-sectional area loss rate is less than 3%
10
[44]. When there are more than two broken wires in the
Amplitude(V)

5 cable, i.e., the cross-sectional area loss rate is larger than

0 3.28%, the defect echo signal is clearly observed in the guided
-5 wave testing signal. As the number of broken wires increases
-10
D (from Num  2–9), the amplitude of the defect echo sig-
EM P End1 End2 nal increases. In order to show more clearly the variation of
-15
0 0.5 1 1.5 the testing signals with the number of broken wires, Fig. 11
Time(s) 10-3
shows the guided wave testing signals for all broken wire
Fig. 9 The guided wave testing signal of 7 broken wires in the cable cases and zooms in to show defect echo signal regions. It can
be seen that these defect echo signals have distinct difference
in amplitude, phase or waveform structure.
This information can also be clearly observed in corre-
signal, which is caused by electrically leaking from the air. sponding STFT of the guided wave signals shown in Fig. 12.
The signal P is the first passing signal, the signal End1 and The spectrograms are estimated and compared with the the-
End2 are the right end and the left end echo signals, respec- oretical results of dispersion curves of the φ7 steel wire [36],
tively. The distance between the excitation transducer and the where the dispersion curves are changed to the form of time
receiving transducer is 1000 mm, and the first passing signal versus frequency using the method of dividing group veloc-
appears at 0.194 ms. Therefore, the group velocity can be ity dispersion curves by the propagation distance. Here, the
calculated as 5154 m/s. End1 and End2 appear at 0.783 ms L(0,1) mode is observed clearly in the spectrograms. It can
and 1.152 ms, and hence the corresponding travel distances be seen that the longitudinal wave mode arrival time shows
are 4035 mm and 5971 mm. The relative errors are 0.88% a good agreement with peaks in the spectrogram. Theoreti-
and 1.20%, respectively. Figure 9 shows an example guided cally, both L(0,1) and L(0,2) mode guided waves are present
wave testing signal for seven broken wires in the cable. Obvi- in the received signal, but since the L(0,2) mode is more
ously, there is a defect echo signal D appear at about 0.590 ms attenuated in the cable, the L(0,1) mode guided wave is
(marked by a dashed red box). According to the group veloc- predominant in the received signal. It indicates that the longi-
ity, the travel distance of the defect echo signal is 3040 mm, tudinal mode guided wave propagates along individual wires
and the relative error is 1.33%. The inherent characteristics of the cable, and the dispersion and multi-mode effects are
of guided waves—multi-mode and dispersion—are the main not obvious in the first travel period. This facilitates further
cause of the errors. In addition, there are errors in the local- analysis of the relationship between the defect echo signal
ization of the guided wave signals. It is beyond the scope and the number of broken wires in the cable.
of this study to discuss how to overcome multi-mode and
dispersion (e.g., mode separation, dispersion compensation) 4.3 Topological Features of the Guided Wave Testing
and how to improve the localization accuracy of guided wave Signals
signals (e.g., accurate localization algorithms), and therefore
will not be discussed here. The relative error will guide the In this work, one time window is selected for the defect
selection of signal intervals for later analysis. Furthermore, echo signal. Theoretically, according to the group velocity
the signal intervals between P and End2 is normalized to − 1 of the guided wave (about 5154 m/s) and the travel distance
~ 1 to ensure that changes in transducer and cable installation of the defect echo signal (1000 mm showed in Fig. 7), the
conditions have no effect on the analysis results. 1000mm
defect echo signal should appear at 5154m/s  0.582 ms.

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 11 of 18 29

(a) Num=0 (b) Num=1

1.5 1.5

P End1 End2 P End1 End2

Amplitude(V) 1 1

Amplitude(V)
0.5 0.5

0 0

-0.5 -0.5
D
-1 -1

-1.5 -1.5
2 4 6 8 10 12 14 2 4 6 8 10 12 14
-4
10 10-4
Time(s) Time(s)

(c) 1.5
Num=3 (d) 1.5
Num=5
P End1 End2 P End1 End2
1 1
Amplitude(V)

Amplitude(V)
0.5 0.5

0 0

-0.5 -0.5
D D
-1 -1

-1.5 -1.5
2 4 6 8 10 12 14 2 4 6 8 10 12 14
10-4 10-4
Time(s) Time(s)

(e) 1.5
Num=7 (f) 1.5 Num=9
P End1 End2 P End1 End2
1 1
Amplitude(V)

Amplitude(V)

0.5 0.5

0 0

-0.5 -0.5
D D
-1 -1

-1.5 -1.5
2 4 6 8 10 12 14 2 4 6 8 10 12 14
-4
10 10-4
Time(s) Time(s)

Fig. 10 The normalized guided wave testing signals of the broken wire cable, where the number of broken wires is: a Num  0, b Num  1, c Num
 3, d Num  5, e Num  7, f Num  9, respectively

123
 29 Page 12 of 18 Journal of Nondestructive Evaluation (2023) 42:29

1.5
Num=0
Num=1
Num=2
1 Num=3
Num=4
Num=5
Num=6
0.5
Num=7
Amplitude(V)

Num=8
Num=9
0

-0.5

-1

-1.5
2 4 6 8 10 12 14
Time(s) 10-4

Fig. 11 The normalized guided wave testing signals with the number of broken wires from 0 to 9

In theory, the lasting time of the guided wave signal is barcode is marked in red. The longest barcode indicates the
N × T  N × 1f  3 ∗ 50×10 1
3  0.06 ms, where N is the persistence of the largest hole in the phase space, which is
number of excited sinusoidal signal cycles, T is the period of the large-scale topological feature, while other short barcodes
the sinusoidal signal and f is the corresponding frequency. are considered as noise.
That is, the theoretical defect echo window can be fixed at It can be observed from Fig. 13 that the amplitude
0.582 ms to 0.642 ms. However, due to the complexity of the of the defect echo signal and the longest barcode length
61-wire cable structure, non-single frequency of the excita- both increase with the number of broken wires increasing.
tion signal and the characteristic dispersion of the guided Therefore, in addition to the proposed topological feature,
wave, it is difficult to precisely obtain the actual lasting time we selected other three features from the time domain,
and the arrival time of the defect echo signal. This is corrob- frequency domain and joint time–frequency domain (e.g.,
orated by the relative error in the travel distance of the right wavelet domain) of the original guided wave testing signals
and left end echo signals mentioned in Sect. 4.1. Thus, the to characterize the broken wires defect. In the time domain,
theoretical defect arrival time was combined with the obser- peak-to-peak ratio of the defect echo signal D and the passing
vation of the guided wave testing signal to determine the signal P is selected. In the frequency domain, the spectrum
defect echo window. That is, it extends about 10.5% before peak ratio of the defect echo signal D and the passing sig-
and after the theoretical defect arrival time (0.582 ms), and nal P at propagating frequency range is chosen. The wavelet
the defect echo window is finally determined to be fixed at coefficient variance ratio of defect echo signal D and the
0.519 ms to 0.644 ms. passing signal P is employed in the wavelet domain. Here,
The phase space with optimal geometry of the defect the Daubechies wavelet of order 40 (db40) is used for dis-
echo signal is reconstructed using the technique described in crete wavelet transform (DWT), and the wavelet coefficient
Sect. 2.2, where dimension m  2 and the time delay τ  T4 vector cD5 at level 5 is employed to compute the ratio, due to
 
(continuous case) or τ  4fsf (discrete case). In this work, its similarity to guided wave signals. Notably, another time
window is needed for the passing signal, which was fixed
the sampling rate f s is 2 MHz and the center frequency f of
between 0.132 and 0.257 ms.
the received signal is 40 kHz, so the discrete time delay τ
The relationship between these features and the number of
is 12. It is noted that, the center frequency is not same as
broken wires is shown in Fig. 14, where the scatter blue dots
the exciting frequency (50 kHz) due to the center frequency
are experimental data and the solid red line is the best fitting
shift (CFS) caused by the width of the coil [45]. Some typical
curve. It can be seen that these four features all have good
defect echo signals and their 2-dimensional phase spaces with
functional relationships with the number of broken wires.
optimal geometry are shown in the left and middle columns
The first three features are linear functions of the number of
of Fig. 13: (a) Num  3, (b) Num  5, (c) Num  7, (d) Num
broken wires, while the wavelet coefficient covariance ratio
 9. The topological feature, 1-dimensional ‘hole’, can be
is non-linear function of the number of broken wires. For the
observed clearly in each phase space. Then, persistent homol-
computation, the proposed method is larger than the other
ogy is employed to capture the topological information of the
methods. However, the experimental topological feature has
2-dimensional phase space. JavaPlex library [46] is selected
only a slight deviation from the best fitting curve when the
to compute persistent homology. The corresponding 1-th PBs
number of broken wires is less than 3, which is better than
are displayed in the right column of Fig. 13, where the longest

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 13 of 18 29

Num=0 Num=1
(a) 200 20
(b) 200 20

15 15
150 L(0,1) L(0,1) L(0,1) L(0,1) 150 L(0,1) L(0,1) L(0,1) L(0,1)
Frequency(kHz)

Frequency(kHz)
10 10

100 5 100 5

0 0
50 50
-5 -5

0
P End1 End2 -10 0
P D End1 End2 -10
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time(s) -4 Time(s) -4
10 10

Num=3 Num=5
(c) 200 20
(d) 200 20

15 15
150 L(0,1) L(0,1) L(0,1) L(0,1) 150 L(0,1) L(0,1) L(0,1) L(0,1)
Frequency(kHz)

Frequency(kHz)
10 10

100 5 100 5

0 0
50 50
-5 -5

0
P D End1 End2 -10 0
P D End1 End2 -10
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time(s) 10-4 Time(s) 10-4

Num=7 Num=9
(e) 200 20
(f) 200 20

15 15
150 L(0,1) L(0,1) L(0,1) L(0,1) 150 L(0,1) L(0,1) L(0,1) L(0,1)
Frequency(kHz)

Frequency(kHz)

10 10

100 5 100 5

0 0
50 50
-5 -5

0
P D End1 End2 -10 0
P D End1 End2 -10
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time(s) 10-4 Time(s) 10-4

Fig. 12 The spectrograms of the guided wave testing signals of the bro- solid lines are the theoretical curves for the longitudinal modes for the
ken wire cable, where the number of broken wires is: a Num  0, b Num φ7 mm steel wire
 1, c Num  3, d Num  5, e Num  7, f Num  9, respectively. The

123
 29 Page 14 of 18 Journal of Nondestructive Evaluation (2023) 42:29

0.05
0.05
Amplitude(V)

S(t+ )
0 0

-0.05 -0.05
5.5 6 6.5 5 -0.05 0 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03
(a) Time(s) 10-4 S(t) r

0.05
Amplitude(V)

0.05

S(t+ )
0 0

-0.05 - -0.05
5.5 6 6.5 -0.05 0 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03
(b) Time(s) 10
-4
S(t) r

0.05
Amplitude(V)

0.05
S(t+ )

0
0

-0.05 -0.05
5.5 6 6.5 -0.05 0 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03
(c) Time(s) 10-4 S(t) r

0.05 0.05
Amplitude(V)

S(t+ )

0 0

-0.05 -0.05
5.5 6 6.5 -0.05 0 0.05 0 0.005 0.01 0.015 0.02 0.025 0.03
(d) Time(s) 10-4 S(t) r
Fig. 13 The typical defect echo signals (the left column), the corre- (the middle column) and the 1-th PBs of the phase spaces for differ-
sponding discrete 2-dimensional phase spaces with optimal geometry ent number of broken wires in the cable: a N um  3, b N um  5,
c N um  7, d N um  9, where the red barcode is the longest one

that of the other three defect features. That is, the proposed In conclusion, the proposed topological feature—the
topological feature can better characterize small defects. Fur- longest barcode length in 1-th PB—extracted from the topo-
thermore, the linear relationship between the topological logical domain shows an excellent linear relationship with
feature and the number of broken wires is best due to the the number of broken wires in the cable, which better
largest R-square, 0.9946, where the other three R-squares are characterizes the broken wire defects and outperforms the
0.9845, 0.9888 and 0.9614, respectively. It is a forward prob- conventional defect features extracted from the time domain,
lem in the experimental study, but it is an inverse problem frequency domain and joint time–frequency domain.
in practice. This means that in practice, the number of bro-
ken wires predicted from the topological feature will have
4.4 The Case of Defect Location Not Fixed
the minimum error with the true number of broken wires,
especially when the number of broken wires is less than 3.
In the experiments, the location of the broken wires is fixed.
Therefore, the defect echo signal is known in the guided wave

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 15 of 18 29

(a) (b)
0.02
R2=0.9946 0.05 R2=0.9845
Longest barcode

Peak-to-Peak
0.04
0.015

0.03

0.01
0.02

0.005
0.01

0 2 4 6 8 10 0 2 4 6 8 10
Num Num
10-3
(c) (d) 4
R2=0.9614

DWT-coef Variance Ratio

0.12 R2=0.9888
Spectrum Peak Ratio

3
0.1

0.08
2
0.06

0.04 1

0.02

0 0
0 2 4 6 8 10 0 2 4 6 8 10
Num Num
Fig. 14 Defect features extracted from the topology domain, time domain peak-to-peak ratio, c the frequency spectrum peak ratio, d the
domain, frequency domain and joint time–frequency domain of the variance of the DWT coefficient. The scatter blue dots are experimental
guided wave testing signals for the broken wire cable as a function data and the solid red line is the best fitting curve
of the number of broken wires: a longest barcode length, b the time

the serial number of the sliding time window, respectively.

Sliding time window Signal The full signal can be scanned by the sliding time window
when the window slides from left to right. Then, the sub-
Amplitude

signals intercepted by the sliding time window are analyzed

n=i n=i+1 by our proposed topological method, and the spectrum of the
topological features of the signal is obtained. As an exam-
ple, there are three simulated guided wave testing signals
with different SNRs (− 5 dB, 0 dB, 5 dB), as shown in the
w left column of Fig. 16. The guided wave signal is simulated
according to Eq. (9) and is located in the middle of the sim-
Time
ulated testing signals (marked by the red dashed box). Here,
Fig. 15 The diagrammatic sketch of the sliding time window the length of the simulated signals is 500 and the length of
the guided wave signal is 100. The width of the sliding time
window is set to be the length of the guided wave signal. The
testing signal. However, the location of the broken wires of time window slides along the testing signal from left to right.
the real bridge cables is not fixed and the defect echo signal The sub-signals extracted by the sliding time window are
is unknown. Furthermore, the SNRs of the testing signals in analyzed by our proposed topological method, and then the
the real field are poor. In this section, a sliding time window spectrum of the topological features of the simulated testing
is used to solve this problem. signal is obtained. The corresponding spectra are shown in
Figure 15 depicts the diagrammatic sketch of the rectan- the right column of Fig. 16, where the horizontal axis n is
gle sliding time window, where w and n are the length and the serial number of the sliding time window and the vertical

123
 29 Page 16 of 18 Journal of Nondestructive Evaluation (2023) 42:29

1.5 0.25
Guided wave Range of the time window

Longest barcode length

SNR=-5dB
1 signal 0.2
containing the guided wave signal
Amplitude(V)
0.5
0.15
0
0.1
-0.5

0.05
-1

-1.5 0
0 100 200 300 400 500 0 100 200 300 400
N n
1.5 0.25

SNR=0dB

Longest barcode length

1
0.2
Amplitude(V)

0.5
0.15
0
0.1
-0.5

0.05
-1

-1.5 0
0 100 200 300 400 500 0 100 200 300 400
N n
1.5 0.25
Longest barcode length

1
SNR=5dB
0.2
Amplitude(V)

0.5
0.15
0
0.1
-0.5

0.05
-1

-1.5 0
0 100 200 300 400 500 0 100 200 300 400
N n
Fig. 16 The topological spectra of the simulated guided wave testing signals with different SNRs

axis is the length of the longest barcode of the 1-th PB of its topological feature spectrum. Then, the peak value of the
the sub-signal. Here, the blue dashed box marks the range of spectrum can be extracted to characterize the defect size.
the sliding window containing the guided wave signal (n 
150~50). It is clear that for all the simulated testing signals
with different SNRs, the peak value of the spectrum is in the 5 Discussion
blue dashed box. The peak position is not fixed due to noise.
This means that in the guided wave testing of the bridge cable In this work, we designed nine different defect cases (as
in service, without knowing the exact location of the defect, shown in Fig. 6) and all the broken wires were on the outer
the guide wave testing signal is processed by the proposed ring one by one. Such defect cases are designed based on
method with the help of the sliding time window to calculate following three reasons. Firstly, from the appearances of a

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 17 of 18 29

large number of the decommissioned cables (collected by cables by pushing the steel wires closer together, thus allow-
Liuzhou OVM Machinery Co., LTD), the damages (such ing for ultrasonic energy transfer between more steel wires
as corrosion) of the outer wires in the cable are generally [47–49]. Moreover, in the field bridge cable testing, the SNR
more serious than that of the inner ones. This is because that of the guided wave testing signals are very low. Under these
once the cable sheath is damaged, harmful substances cor- practical conditions, the characterization and detection of
roded the wires from the outer to the inner. We know that the broken wires of the bridge cables using magnetostric-
further development of corrosion may result in wire break- tive guided waves based on the topological features of the
age. The more severe the corrosion of the wire, the greater testing signals is exactly what we are gonging to working on
the possibility of wire breakage. Secondly, the bridge cable is in the future.
made of multiple steel wires bundled together in hexagonal
shape, wrapped with high strength polyester tape tightly, and
sheathed in a direct hot extruded high density polyethylene 6 Conclusion
plastic (HDPE). Thus, it is very difficult to make artificial
defects in the inner wires of the cable. Considering the dam- In this study, large-scale topological features are proposed
age of the real bridge cables, as well as the efficiency and to characterize the broken wire defects in the multi-wire
cost of the experiments, only the broken wire defects in outer bridge cable, which are extracted by persistent homology
ring of the cable were made in this work. Thirdly, the HDPE from the topology of the dynamical reconstruction of guided
sheath of the cable cracks due to stress, and harmful sub- wave signal. Based on Takens time-delay embedding theory,
stances enter the inside of the cable from the crack, causing the dynamical reconstructions with optimal geometry of the
damage to the steel wires. The cracking of HDPE sheath is guided wave signals are carried out by selecting appropri-
gradually expanded from local area to whole ring, so the steel ate time delay parameter. The reconstructed 2-dimensional
wires of cable will be damaged one by one. Thus, the steel phase space is properly unfolded off the main diagonal. Per-
wires of the cable are designed to be cut right next to each sistent homology is used to capture the topological features
other in our work. For the defect cases of broken wires in of the 2-dimensional phase space and shows the results in
inner ring or not close to each other, the propagation char- 1-th PB. The large-scale topological feature, the longest bar-
acteristics of the guided wave are different from the cases code length in 1-th PB, shows a strong correlation with the
of the our designed. This is due to the contact and friction number of broken wires in the multi-wire bridge cable. A
between the wires of the cable under the binding of the HDPE 61-wire bridge cable was tested in our experiment. Nine
sheath that allows the guided wave energy to be transmitted wires of the cable were cut one by one and the correspond-
between the wires. Thus, the properties of the guided wave ing cross-sectional area loss is from 1.64 to 14.75%. The
signal and its phase space for these defect cases need to be testing signals were obtained by a magnetostrictive guided
verified experimentally in the future. wave testing instrument. The analysis results show that the
Moreover, this paper only focuses on the multiple defects proposed topological feature increases exactly linearly with
in the same location in a multi-wire cable and it found that increasing the number of broken wires and the goodness of
the proposed topological feature increases linearly with the fit for the fitting curve is 0.9946.
number of broken wires. For the case of defects at multiple Furthermore, for the guided wave testing of the bridge
locations, guided wave reflection and transmission between cables in service, without knowing the exact location of the
multiple defect spots make the signal very complex. Due defect, the sliding time window should be used to calculate
to the overlapping of the defect echo signals, the received the spectrum of the topological features of the testing signal.
defect echo signals may not increase with the number of the The peak value of the spectrum is then used to characterize the
broken wires. In our previous work [18], it investigated the defect size of the cables. It is worth noting that this method is
multiple broken-wire flaws in the same wire of strands, and also applicable to other size of parallel wire cables depending
obtained the relationship between the energy exchange and on the propagation characteristics of guided waves and the
the frequency of the exciting signal along the same propa- interaction with the broken wires.
gating distance by analyzing the reflection and transmission However, only the most common cases of broken wire
coefficients at the three different flaws. Additional experi- defects were investigated in this work. From the comprehen-
ments are needed in the future to investigate the relationship sive consideration of the damage of the real bridge cables
between topological features and the defect level at multiple as well as the experimental efficiency and cost, this paper
locations. only focuses on the defect case of broken wires at the outer
In addition, the multi-wire bridge cable was not subject to ring. In addition, based on the fact that the cracking of HDPE
tension loading in this work. In reality, the bridge cables sheath is gradually expanded from local area to whole ring,
would be under high tension loading. It has been shown the steel wires of the cable are designed to be cut right next
that tension can change the wave propagation on multi-wire to each other. Moreover, the experiments were conducted

123
 29 Page 18 of 18 Journal of Nondestructive Evaluation (2023) 42:29

in an unloaded cable. In the future, we will investigate the 7. Zhang, P.F., Tang, Z.F., Lv, F.Z., Yang, K.J.: Numerical and
topological properties of the guided wave testing signals for experimental investigation of guided wave propagation in a multi-
wire cable. Appl. Sci. 9, 1028 (2019). https://doi.org/10.3390/
more complex defect cases, such as broken wires in inner app9051028
ring, multi-location broken wire defects, and defects in high 8. Schaal, C., Bischoff, S., Gaul, L.: Damage detection in multi-wire
tension loading cables, corrosion, and so on. cables using guided ultrasonic waves. Struct. Health Monit. Int. J.
15, 279–288 (2016). https://doi.org/10.1177/1475921716642747
Acknowledgements Thanks to OVM Machinery Co., Ltd. for provid- 9. Pavlakovic, B., Lowe, M., Cawley, P.: Guided ultrasonic waves
ing the experimental materials and the experimental site. for the inspection of post-tensioned bridges. In: Thompson, D.O.,
Chimenti, D.E. (eds.) Review of Progress in Quantitative Nonde-
Author Contributions SD: Writing—Original Draft, Supervision. XW: structive Evaluation, pp. 1557–1564. Springer, Boston (1998)
Writing—review & editing, Project administration. YZ: Resources. JW: 10. Kwun, H., Teller, C.M.: Detection of fractured wires in steel cables
Formal analysis. LJ: Investigation, Data curation. YW: Software. using magnetorestrictive sensors. Mater. Eval. 52, 503–507 (1994).
https://doi.org/10.2749/222137815815775790
Funding No funding. 11. Rizzo, P., di Scalea, F.L.: Monitoring steel strands via ultrasonic
measurements. In: Proceedings of the smart structures and materi-
als 2002 conference, vol. 4696, pp. 62–73. San Diego, CA. https://
Data Availability The results/data/figures in this manuscript have not
doi.org/10.1117/12.472578 (2002)
been published elsewhere, nor are they under consideration (from you
12. Liu, Z.H., Zhao, J.C., Wu, B., Zhang, Y.N., He, C.F.: Configura-
or one of your Contributing Authors) by another publisher.
tion optimization of magnetostrictive transducers for longitudinal
guided wave inspection in seven-wire steel strands. NDT E Int. 43,
Declarations 484–492 (2010). https://doi.org/10.1016/j.ndteint.2010.05.003
13. Xu, J., Wu, X.J.: Nondestructive testing of bridge cables using
Competing interests The authors declare that they have no competing magnetostrictive guided wave technique. Appl. Mech. Mater.
interests as defined by Springer, or other interests that might be per- 130, 2015–2018 (2011). https://doi.org/10.4028/www.scientific.
ceived to influence the results and/or discussion reported in this paper. net/amm.130-134.2015
14. Zhang, P.F., Tang, Z.F., Duan, Y.F., Yun, C.B., Lv, F.Z.: Ultra-
Ethical Approval and Consent to Participate The authors have read sonic guided wave approach incorporating SAFE for detecting wire
and understood the publishing policy, and submit this manuscript in breakage in bridge cable. Smart Struct. Syst. 22, 481–493 (2018).
accordance with this policy. https://doi.org/10.12989/sss.2018.22.4.481
15. Rizzo, P., di Scalea, F.L.: Feature extraction for defect detection
Consent for Publication The authors confirm that they understand jour- in strands by guided ultrasonic waves. Struct. Health Monit. 5,
nal Journal of Nondestructive Evaluation is a transformative journal. 297–308 (2006). https://doi.org/10.1177/1475921706067742
When research is accepted for publication, there is a choice to publish 16. Jiang, X., Xinjun, W., Liangyun, W., Chen, H., Kang, Y.H.: Detect-
using either immediate gold open access or the traditional publishing ing the flaws in prestressing strands using guided waves based on
route. the magnetostrictive effect. Insight-Non-Destr. Test. Cond. Monit.
49, 647–650 (2007). https://doi.org/10.1784/insi.2007.49.11.647
17. Liu, Z.H., Zhang, Y.N., He, C.F., Wu, B.: Defect detection in helical
and central wires of steel strands using advanced ultrasonic guided
wave technique with new type magnetostrictive transducers. IEEE
Ultrason. Symp. 2008, 832–835 (2008). https://doi.org/10.1109/
ultsym.2008.0200
References 18. Xu, J., Wu, X.J., Sun, P.F.: Detecting broken-wire flaws at multiple
locations in the same wire of prestressing strands using guided
1. Elliott, M.E., Heymsfield, E.: Inspection of Luling bridge cable waves. Ultrasonics 53, 150–156 (2013). https://doi.org/10.1016/j.
stays: case study. J. Constr. Eng. Manag. 129, 226–230 (2003). ultras.2012.05.003
https://doi.org/10.1061/(asce)0733-9364(2003)129:2(226) 19. Rizzo, P., di Scalea, F.L.: Ultrasonic inspection of multi-wire steel
2. Xu, J., Wu, X.J., Cheng, C., Ben, A.R.: A magnetic flux leakage strands with the aid of the wavelet transform. Smart Mater. Struct.
and magnetostrictive guided wave hybrid transducer for detect- 14, 685–695 (2005). https://doi.org/10.1088/0964-1726/14/4/027
ing bridge cables. Sensors 12, 518–533 (2012). https://doi.org/10. 20. Rizzo, P., di Scalea, F.L.: Wavelet-based feature extraction for
3390/s120100518 automatic defect classification in strands by ultrasonic structural
3. Christen, R., Bergamini, A., Motavalli, M.: Three-dimensional monitoring. Smart Struct. Syst. 2, 253–274 (2006). https://doi.org/
localization of defects in stay cables using magnetic flux leakage 10.12989/sss.2006.2.3.253
methods. J. Nondestr. Eval. 22, 93–101 (2003). https://doi.org/10. 21. Rizzo, P., Sorrivi, E., di Scalea, F.L., Viola, E.: Wavelet-based out-
1023/b:jone.0000010736.74285.b6 lier analysis for guided wave structural monitoring: application to
4. Rizzo, P., di Scalea, F.L.: Acoustic emission monitoring of multi-wire strands. J. Sound Vib. 307, 52–68 (2007). https://doi.
carbon-fiber-reinforced-polymer bridge stay cables in large-scale org/10.1016/j.jsv.2007.06.058
testing. Exp. Mech. 41, 282–290 (2001). https://doi.org/10.1007/ 22. Tang, Z.F., Sui, X.D., Duan, Y.F., Zhang, P.F., Yun, C.B.: Guided
bf02323146 wave-based cable damage detection using wave energy transmis-
5. Zejli, H., Laksimi, A., Tessier, C., Gaillet, L., Benmedakhene, S.: sion and reflection. Struct. Control Health Monit. 28, e2688 (2021).
Detection of the broken wires in the cables’ hidden parts (anchor- https://doi.org/10.1002/stc.2688
ings) by acoustic emission. Adv. Mater. Res. 13, 345–350 (2006). 23. Takens, F.: detecting strange attractors in turbulence. Dyn. Syst.
https://doi.org/10.4028/0-87849-420-0.345 Turbul. (1980). https://doi.org/10.1007/bfb0091924
6. Mehrabi, A.B., Tabatabai, H.: Unified finite difference formulation 24. Garland, J., Bradley, E., Meiss, J.D.: Exploring the topology of
for free vibration of cables. J. Struct. Eng. 124, 1313–1322 (1998). dynamical reconstructions. Physica D 334, 49–59 (2016). https://
https://doi.org/10.1061/(asce)0733-9445(1998)124:11(1313) doi.org/10.1016/j.physd.2016.03.006

123
Journal of Nondestructive Evaluation (2023) 42:29 Page 19 of 18 29

25. Edelsbrunner, L., Zomorodian: Topological persistence and sim- 41. Alleyne, D., Cawley, P.: A two-dimensional Fourier transform
plification. Discret. Comput. Geom. 28, 511–533 (2002). https:// method for the measurement of propagating multimode signals. J.
doi.org/10.1109/sfcs.2000.892133 Acoust. Soc. Am. 89, 1159–1168 (1991). https://doi.org/10.1121/
26. Edelsbrunner, H., Harer, J.: Persistent homology—a survey. Sur- 1.400530
veys on discrete and computational geometry, pp. 257–82. https:// 42. Ervin, B.L., Bernhard, J.T., Kuchma, D.A., Reis, H.: Estimation
doi.org/10.1090/conm/453/08802 (2008) of general corrosion damage to steel reinforced mortar using fre-
27. Wasserman, L.: Topological data analysis. Annu. Rev. Stat. Appl. quency sweeps of guided mechanical waves. Insight-Non-Dest.
5, 501–532 (2018). https://doi.org/10.1146/annurev-statistics- Test. Cond. Monit. 48, 682–692 (2006). https://doi.org/10.1784/
031017-100045 insi.2006.48.11.682
28. Bendich, P., Edelsbrunner, H., Kerber, M.: Computing robustness 43. Filip, S.I.: A robust and scalable implementation of the Parks-
and persistence for images. IEEE Trans. Vis. Comput. Graph. 16, McClellan algorithm for designing FIR filters. ACM Trans. Math.
1251–1260 (2010). https://doi.org/10.1109/tvcg.2010.139 Softw. 43, 1–24 (2016). https://doi.org/10.1145/2904902
29. Horak, D., Maletić, S., Rajković, M.: Persistent homology of com- 44. Xu, J., Tang, H., Wu, X.J., Ben, A.R., Kang, Y.H.: Broken wire
plex networks. J Stat Mech. (2009). https://doi.org/10.1088/1742- detection in stay cables based on guided waves. In: Proceedings of
5468/2009/03/p03034 the International Conference on Mechanic Automation and Control
30. Liu, X.C., Xie, Z., Yi, D.Y.: A fast algorithm for constructing topo- Engineering, pp. 1207–1210. IEEE. https://doi.org/10.1109/mace.
logical structure in large data. Homol. Homot. Appl. 14, 221–238 2010.5536416 (2010)
(2012). https://doi.org/10.4310/hha.2012.v14.n1.a11 45. Hu, C.Y., Xu, J.: Center frequency shift in pipe inspection using
31. Lum, P.Y., Singh, G., Lehman, A., Ishkanov, T., Vejdemo- magnetostrictive guided waves. Sens. Actuators A Phys. 298,
Johansson, M., Alagappan, M., et al.: Extracting insights from the 111583 (2019). https://doi.org/10.1016/j.sna.2019.111583
shape of complex data using topology. Sci. Rep. 3, 1–8 (2013). 46. Adams, H., Tausz, A., Vejdemo-Johansson, M.: javaPlex:
https://doi.org/10.1038/srep01236 a research software package for persistent (Co)Homology,
32. Umeda, Y.: Time series classification via topological data analysis. pp. 129–36. https://doi.org/10.1007/978-3-662-44199-2_23
Inf. Media Technol. 12, 228–239 (2017). https://doi.org/10.1527/ (2014)
tjsai.d-g72 47. Liu, X.C., Wu, B., Qin, F., He, C.F., Han, Q.: Observation of
33. Ghrist, R.: Barcodes: the persistent topology of data. Bull. Am. ultrasonic guided wave propagation behaviours in pre-stressed
Math. Soc. 45, 61–75 (2007). https://doi.org/10.1090/s0273-0979- multi-wire structures. Ultrasonics 73, 196–205 (2017). https://doi.
07-01191-3 org/10.1016/j.ultras.2016.08.014
34. Chang, Z.X., Mu, L., Wu, K.D., Opron, K., Xia, K.L., Wei, G.W.: 48. Treyssède, F.: Investigation of the interwire energy transfer of elas-
A topological approach for protein classification. Comput. Math. tic guided waves inside prestressed cables. J. Acoust. Soc. Am. 140,
Biophys. 3, 140–162 (2015). https://doi.org/10.1515/mlbmb-2015- 498–509 (2016). https://doi.org/10.1121/1.4955080
0009 49. Rizzo, P.: Ultrasonic wave propagation in progressively loaded
35. Pun, C.S., Xia, K., Lee, S.X.: Persistent-homology-based machine multi-wire strands. Exp. Mech. 46, 297–306 (2006). https://doi.
learning and its applications--a survey. arXiv:181100252v1. org/10.1007/s11340-006-7248-0
https://doi.org/10.2139/ssrn.3275996 (2018)
36. Duan, S.Y., Wu, X.J., Zou, Y.Q., Jiang, L.J.: Corrosion charac-
terization of steel wires based on persistent homology theory for
Publisher’s Note Springer Nature remains neutral with regard to juris-
magnetostrictive guided wave testing signal. Struct. Health Monit.
dictional claims in published maps and institutional affiliations.
(2022). https://doi.org/10.1177/14759217221119790
37. Kennel, M.B., Brown, R., Abarbanel, H.D.: Determining embed-
Springer Nature or its licensor (e.g. a society or other partner) holds
ding dimension for phase-space reconstruction using a geometrical
exclusive rights to this article under a publishing agreement with the
construction. Phys. Rev. A 45, 3403 (1992). https://doi.org/10.
author(s) or other rightsholder(s); author self-archiving of the accepted
1103/physreva.45.3403
manuscript version of this article is solely governed by the terms of such
38. Kantz, H., Schreiber, T.: Nonlinear Time Series Analysis. Cam-
publishing agreement and applicable law.
bridge University Press, Cambridge (2004)
39. Fraser, A.M., Swinney, H.L.: Independent coordinates for strange
attractors from mutual information. Phys. Rev. A 33, 1134–1140
(1986). https://doi.org/10.1103/physreva.33.1134
40. Wu, J., Wang, Y., Zhang, W.W., Nie, Z.H., Lin, R., Ma, H.W.:
Defect detection of pipes using Lyapunov dimension of Duffing
oscillator based on ultrasonic guided waves. Mech. Syst. Sig-
nal Process. 82, 130–147 (2017). https://doi.org/10.1016/j.ymssp.
2016.05.012

123