Mechanical Systems and Signal Processing 168 (2022) 108652

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

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

Invited Review Paper

A comprehensive review of loosening detection methods for

threaded fasteners
Jiayu Huang , Jianhua Liu , Hao Gong *, Xinjian Deng
Beijing Institute of Technology, School of Mechanical Engineering, 5 South Zhongguancun Street, Haidian District, Beijing 100081, China

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

Communicated by John E. Mottershead Loosening of threaded fasteners can cause preload decline, induce bolt fatigue fracture, and
severely compromise the reliability of mechanical products. Loosening detection is an effective
Keywords: method for early prevention of severe loosening behaviour. This study classifies various detection
Thread fasteners methods into sensor-based, vision-based and percussion-based methods and systematically sum­
Loosening detection
marises their research progresses. The sensor-based method implants or sticks sensors on the
Contact characteristics
mechanical structure with bolted joints, and achieves loosening detection by exploiting the
Structural dynamic characteristics
Visual inspection variation on measurement parameters of sensors. It can be divided into explicit detection and
implicit detection. The former requires accurate experimental calibration whereas the latter re­
quires to extract sensitive loosening features. The percussion-based method applies a hammer to
knock the mechanical structure and receives the audio signal. Like implicit sensor-based methods,
loosening severity is evaluated by extracting sensitive features from the received audio signal. The
vision-based method captures the images of threaded fasteners and calculates the rotational angle
or the length of exposed bolt for loosening detection. The implicit sensor-based, percussion-based,
and vision-based methods can only detect several discrete loosening states and be applied mainly
to a single bolted joint. It is considered essential and significant to develop a self-powered sensor
capable of signal wireless transmission and to conduct precise preload detection by establishing
the quantitative relationship between loosening features and preloads using deep learning
algorithms.

1. Introduction

Threaded fasteners constitute the most widely used connection method owing to the low cost, good interchangeability, and
convenience of installation and disassembly [1,2]. A typical threaded connection form comprises a bolt, a nut, and clamped com­
ponents. When the bolt or nut is tightened, the bolt elongates, and a preload is generated. An adequate preload is essential to ensure
connection and fastening performance and to improve product reliability. However, threaded fasteners are susceptible to loosening
owing to severe service-environment exposure, particularly vibration environments. A survey of automobile dealer service managers
in the United States by Holmes indicated that 23% of all service problems were traced to loose fasteners. Furthermore, 12% of all new
cars were observed to have loose fasteners [3]. Loosening of threaded fasteners directly results in preload decrease and initiates
mechanical failure. For example, the leakage of water, oil, and gas in special vehicles is mainly attributed to loose fasteners. Moreover,

* Corresponding author.
E-mail addresses: JiayuH1211@gmail.com (J. Huang), jeffliu@bit.edu.cn (J. Liu), gongh0220@163.com (H. Gong), kevindxj@163.com
(X. Deng).

https://doi.org/10.1016/j.ymssp.2021.108652
Received 8 May 2021; Received in revised form 20 August 2021; Accepted 14 November 2021
Available online 1 December 2021
0888-3270/© 2021 Elsevier Ltd. All rights reserved.
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

insufficient preload may induce bolt fatigue fracture under vibration, thereby causing catastrophic accidents. For example, a railway
accident involving the derailment of a high-speed train in the United Kingdom in 2007 resulted from the detachment of nuts from bolts
owing to loosening [4].
In the past 80 years, a substantial number of scholars have researched the loosening mechanism to reveal the factors and processes
of loosening [5–11]. Recent research results showed that the local slippage accumulation on the surfaces of bearing and thread surfaces
under cyclic transverse vibration causes rotational loosening [12,13]. It was also observed recently that the cyclic component force
along the radial direction dominates local slippage accumulation of thread surface, whereas the other two cyclic component forces do
not affect it [14,15]. Inspired by the achievements in the research on the loosening mechanism, certain anti-loosening structures or
methods have been proposed to slow down or restrain loosening behaviour. These include spherical bolt with matched spherical gasket
[16], step-lock bolt [17,18], and super lock nut [19]. In engineering applications, it is infeasible to completely inhibit loosening by
anti-loosening products. Loosening may also occur owing to embedding loss, fretting wear, and stress redistribution of contact
interface; and creep, stress relaxation, and cyclic plastic deformation of bolts or nuts. Regular loosening detection for threaded fas­
teners is a feasible and effective method for ensuring connection reliability. When loose fasteners are detected or identified, we can
retighten the old fasteners or replace these with new ones to recover the initial fastening capability.
Various loosening detection methods have emerged with the rapid development of sensor, signal processing, and computer vision
technology [20–22]. A comprehensive review of the loosening detection methods for threaded fasteners has not been performed. This
is the motivation for this study. Herein, we have extensively reviewed the literature on loosening detection methods published over the
past 20 years (particularly the past 5 years), summarised these systematically, and finally developed this review article. It is considered
that our study would help beginner researchers to rapidly understand the current status of the research on the loosening detection of
threaded fasteners. In addition, it would enhance engineers’ understanding of this subject and enable them to select appropriate
detection devices for engineering applications.
In this study, various loosening detection methods were outlined systematically. The remainder of this paper is organised as fol­
lows: In Section 2, an overview of various detection methods is presented. These methods can be categorised into three groups: sensor-
based, percussion-based and vision-based methods. In Sections 3–5, the research progresses on above three types of detection methods
are concluded, respectively. Finally, certain perspectives and future research directions for loosening detection are summarised in
Section 6.

2. Overview

The purpose of loosening detection is to evaluate the changes of preloads generated in bolted joints. It can be classified into sensor-
based, percussion-based detection, and vision-based methods, based on the differences of detection devices, as shown in Fig. 1. The
sensor-based method is applied most widely, which implants or sticks sensors on the mechanical structure with bolted joints, and
achieves loosening detection by exploiting the variation on measurement parameters of sensors. Various sensor-based detection
methods are divided into two classifications, i.e., explicit detection and implicit detection. For the explicit loosening detection, some
measurement parameters obtained by sensors have an explicit linear relationship with preloads, thus the preloads can be measured
quantitatively and accurately after experimental calibration. The typical sensors include strain gauge, ultrasonic sensor, and FBG

Fig. 1. Overview of loosening detection methods.

2
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

sensor. The corresponding measurement parameters related to preload variation are strain, the time-of-flight (TOF) of a pulse-echo
propagating along the bolt axis, and the wavelength of reflected light, respectively. For the implicit loosening detection, a piezo­
electric ceramic (PZT) sensor is generally employed on the mechanical system with bolted joints. The relationship between the
response signal obtained by PZT sensor and preload variation is implicit and nonlinear. The loosening severity can be evaluated
qualitatively by extracting sensitive features from the response signal. Five typical implicit methods have been developed: ultrasonic
attenuation, time reversal, vibro-acoustic modulation, electromechanical impedance, and excitation analysis methods.
In the percussion-based method, a hammer is applied to knock the mechanical structure, and the audio signal is captured using a
microphone. Like the implicit sensor-based method, loosening severity is detected by extracting sensitive features from the received
audio signal. In the vision-based detection method, the images of threaded fasteners are captured by camera or smartphone. The
rotational angle or the length of exposed bolt can be calculated using image processing technologies and machine learning algorithms
for loosening detection.

3. Sensor-based loosening detection

3.1. Explicit detection

In the explicit detection, the linear relationship between measurement parameters obtained by sensors and preloads of a bolt can be
expressed mathematically. Therefore, it is easy to directly and accurately calculate the value of preload according to the magnitude of
measurement parameter after experimental calibration. In this subsection, three types of typical sensors used for explicit detection are
concluded and their measurement principles and research statuses are introduced in detail.

3.1.1. Strain gauge

A strain gauge is generally bonded on the surface of bolt shank without thread [21], or the wall surface of the hole inside a bolt [23],
as shown in Fig. 2(a). Fig. 3 shows the principle of loosening detection using strain gauge. A preload will result in the bolt elongation
and cause the attached strain gauge to extend accordingly. The electric resistance of the strain gauge will vary, which can be measured
accurately for representing the magnitude of preload. Moreover, a strain gauge can also be imbedded inside a washer [24], functioning
as a pressure transducer, as shown in Fig. 2(b). Similarly, the compression strain of the strain gauge occurs under the preload, and the
electric resistance will change thereby. A bridge circuit is usually built to measure the change of electric resistance for detecting

Fig. 2. Installation forms of strain gauge.

3
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 3. Principle of loosening detection using strain gauge.

loosening quantitatively.
The measurement accuracy, advantages, and disadvantages of three types of strain gauges are summarised in Table 1. The second
installation form of strain gauge is convenient to be implemented. However, it is straightforwardly affected by tightening torque and
non-uniform pressure distribution on the contact surface, which will cause the relatively low measurement accuracy (measurement
error is approximately 10%). The first installation form, wherein a strain gauge is bonded to the surface of bolt shank, does not damage
the bolt and has a high measurement accuracy (measurement error is approximately 5%). The disadvantages are that it is unsuitable
for bolts with full threads and is straightforwardly affected by the adhesive melting due to high temperature. In the first installation
form, wherein a strain is bonded to the wall surface of a hole inside the bolt, the drilled hole would damage the bolt and reduce the
strength and reliability. In addition, the process technology for bonding a strain gauge on the wall surface inside a bolt is complex,
requiring experienced workers to finish. However, it has the highest measurement accuracy (measurement error is approximately 1%).
At present, there are few studies on the measurement of preload by strain gauge. Vand et al. [25] designed a type of steel bush. A
strain gauge was imbedded into it to measure the preload of a single bolted joint. The preload could be calculated accurately from the
compression strain of the steel bush between the nut and plate, by applying Hooke’s law. Wang et al. [26] applied the finite element
method to analyse the stress–strain state of a strain gauge under different preloads. It was bonded on the surface of a bolt head.
Simulation results revealed a good linear relationship between the radial strain and bolt preload. Based on this, a preload sensor
bonded on the surface of a bolt head was designed and fabricated. It showed good application potential.

3.1.2. Ultrasonic sensor

Preload measurement using ultrasonic sensors is based on the acoustoelastic effect, which has been extensively researched
worldwide. Both the length of a bolt and propagation velocity of an ultrasonic wave vary as the axial stress of the bolt increases. This
alters the time of flight (TOF) of pulse-echoes along the bolt axis. Therefore, the axial stress of the bolt, which is proportional to the
preload, can be calculated accurately by measuring the variation in the TOF of pulse-echoes relative to that in the non-stress state. The
variation in TOF may be highly marginal, wherefore it important to measure it accurately. The existing research on preload mea­
surement using ultrasonic can be divided into two categories according to the difference in number and type of ultrasonic waves:
mono-wave method [27–29] and bi-wave method [30–35].
As the name implies, the mono-wave method uses only one longitudinal or transverse ultrasonic to measure preload. Longitudinal
wave is adopted widely because it is more sensitive to preload variation. Fig. 4 shows the fundamental principle of the mono-wave
method. The method accurately measures the TOF under preloaded and unstressed conditions. The relationship between the varia­
tion in TOF and preload is as follows:

2Le (A11 + E− 1 ) F
ΔT = t(σ ) − t(σ 0 ) − ε2 = × − ε1 − ε2 (1)
V0 Se

V0 Se
F= × × (ΔT + ε1 + ε2 ) (2)
2(A11 + E− 1 ) Le

where ΔT is the variation in TOF, F is the preload, Se is the effective cross-sectional area, Le is the uniaxial stress length, A11 is the
acoustoelastic constant (which depends on the Lame’s coefficients and Murnaghan’s coefficients), E is the Young’s modulus of the bolt
material, t(σ ) is the TOF of pulse-echo in the stressed state, t(σ 0) is the TOF of pulse-echo in the initial state, V0 is the ultrasonic velocity
in the uniaxial state, ε1 is the error of the first-order expansion, and ε2 is the measurement error. The TOF is generally measured in the
time domain. However, the time domain signal is straightforwardly affected by noise, which hinders the accurate determination of
TOF. Jhang et al. [27] proposed the method of TOF phase detection to eliminate the noise effect and improve the measurement ac­
curacy. Liu et al. [29] proposed a method for measuring preload by dry coupling ultrasonic wave. It was aimed at reducing the
limitation of the liquid coupling method and solving the problem of rust and corrosion on bolt surfaces. The new method is more

4
 J. Huang et al.
Table 1
Measurement accuracy, advantages and disadvantages of strain gauge applied to loosening detection.
Sensor Form Accuracy Advantages Disadvantages

Strain Bonded to the surface of a bolt Approximately 5% Does not damage bolt Unsuitable for bolts with full threads; straightforwardly affected by high
gauge temperature
5

Bonded to the wall surface of a hole inside the Approximately 1% Highest measurement accuracy A drilled hole damages bolt and reduces strength and reliability; complex process
bolt technology
Imbedded inside a washer Approximately Convenient to implement; does not damage Straightforwardly affected by tightening torque and non-uniform pressure
10–20% bolt distribution

Mechanical Systems and Signal Processing 168 (2022) 108652

J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 4. Fundamental principles of mono-wave method.

suitable for preload measurement in the aerospace field.

The bi-wave method applies both longitudinal and transverse ultrasonic waves to measure preload. It is also called the L-S method.
Longitudinal and transverse ultrasonic waves can be generated independently and input to the bolt [30,31,34,35]. These can also be
obtained simultaneously by mode conversion at the interface after starting with a single longitudinal or transverse wave [32,33]. Fig. 5
shows the fundamental principles of bi-wave method. The velocity ratio method is proposed for preload measurement due to the
different acoustic elastic coefficients of longitudinal and transverse ultrasonic waves. Herein, the ratio of the TOF of a transverse
ultrasonic wave to that of a longitudinal ultrasonic wave is calculated [31],
[ ]
tTσ VL0 Le F
σ ≈ 0 1− (AT − AL ) 3
tL VT Li Se

where tT /tLσ is the ratio of the TOF of the transverse ultrasonic wave to that of the longitudinal ultrasonic wave, VL0 /VT0 is the ratio of
the velocity of the longitudinal ultrasonic wave to that of the transverse ultrasonic wave, AT and AL are the longitudinal and transverse
acoustoelastic coefficients, and Li is the sum of the effective length Le and unstressed portion L0. We can find from Eq. (3) that the
preload can be calculated accurately for a given tTσ/tLσ after the approximate linear relationship curve between tTσ/tLσ and preload is
calibrated by experiment. Compared with the mono-wave method, the latter does not require the TOF at the unstressed state, which
substantially simplifies and facilitates the measurement process.
In practical applications, many factors (e.g. the non-uniform distribution of axial stress and the inaccurate measurement of
equivalent stressed length) may significantly affect the results of preload measurement. Pan et al. [34] proposed a model of the shape
factor to eliminate the influence of non-uniform axial stress on the measurement results and circumvent the temperature factor in the
process of temperature measurement. To solve the problems of distortion and aliasing of the converted mode wave in a single-source
ultrasonic signal, Chen et al. [35] proposed a time–frequency parameter identification method based on the Gabor transform to
accurately identify the TOF of a converted mode wave. It can detect the state of a bolt connection more effectively than the con­
ventional L-S method. In addition, Sun et al. [28] embedded a piezoelectric transducer in a predetermined position of a bolt head and
developed an intelligent piezoelectric bolt to measure the TOF of elastic waves inside the bolt. The smart bolt displays precision and
stability that are higher than those of the conventional ultrasonic measurement method.

3.1.3. Fibre Bragg grating (FBG) sensor

The FBG sensor has many advantages such as anti-electromagnetic interference capability, low weight, and high sensitivity. It has

Fig. 5. Fundamental principles of bi-wave method.

6
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

attracted substantial attention for strain measurement because this type of sensor can be conveniently embedded into structures and
can measure the host’s internal strain without significantly altering its inherent characteristics. In 1999, Hay [36] first proposed a
design to permanently embed an FBG sensor into a bolt for measuring its preload. The research results showed that the sensor used,
which had a small transverse dimension, could measure the strain of the bolt accurately without affecting its performance. In the linear
elastic range, the bolt preload could be read directly after the FBG sensor was calibrated experimentally.
The fundamental principle using FBG for preload measurement is shown in Fig. 6. FBG is generally embedded in an optical fibre and
the optical fibre is contacted with a bolt. FRG is a type of sandwich-like distributed reflector with a refractive index that oscillates
periodically. This type of sensor resembles a filter, transmitting the entire spectrum of the light source and reflecting the resonant
Bragg wavelength. The Bragg wavelength λB is given as [37]
λB = 2neff Λ (4)

where neff is the effective refractive index of the FBG and Λ is the grating period. The Bragg wavelength shifts when both the length and
width of an FBG vary. Eq. (5) shows the relationship between the relative shift of Bragg wavelength (indicated by ΔλB) and the strain ε
along the fibre axis under preload [38]:
ΔλB = (Cε ε + CT ΔT)λB (5)

where Cε and CT are material constants and ΔT is the temperature variation. We observe that for constant temperature and Bragg
wavelength, ΔλB is proportional to ε. Eq. (6) presents the relationship between the preload of a bolt F and ε:

de2 π
F= Eε (6)
4

where E is the elastic modulus and de is the equivalent diameter. Therefore, the preload can be measured based on the relative shift of
the Bragg wavelength.
Three methods of sensor implantation are applied in the research on the measurement of bolt preload using an FBG sensor. The first
method embeds the sensor inside a bolt body to produce a ‘smart bolt’ [39–41]. Meanwhile, the second one combines the sensor with a
washer for producing a ‘smart washer’ [42,43]. The final method bonds the FBG sensor on the surface of the bolt shank [44–46]. In
addition, the FBG sensor has also been applied to detect the loosening of atypical threaded fasteners. For example, Wang et al. [47]
proposed a ‘smart clamp’ to detect the clamping force of pipe locking bolts. In the literature [48–50], an FBG sensor instrumented on an
anchor plate was also used to measure the axial force of anchors in civil structures.
Eq. (5) shows that the temperature affects the reflection wavelength of an FBG sensor [51] and thereby, influences the detection
accuracy. Therefore, the influence of temperature on the relative shift of Bragg wavelength must be compensated for an accurate
measurement. An FBG temperature sensor, which has a good thermal contact with the FBG strain sensor, can be applied to realise the
temperature compensation. The process is simple. The wavelength shift measured by the FBG temperature sensor is subtracted from
that measured by the FBG strain sensor after the testing results are obtained [47]. However, the conventional FBG sensor cannot
measure the bolt preload at a very high temperature because of the following reasons. First, the conventional FBG (the first type) shows
a strong attenuation at high temperature. This limits measurement above 300 ℃ in principle [52]. Second, the FBG sensor is generally
embedded in a bolt body through epoxy resin. Most of the epoxy resin degrades when exposed to a high temperature (>400 ◦ C). Huang
et al. [44,46] successfully measured bolt preload at a high temperature using a metal-packaged regenerated fibre Bragg grating (RFBG)

Fig. 6. Principle of FBG sensor in bolt.

7
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

strain sensor.

3.2. Implicit detection

In the implicit detection, PZT sensor is usually preference owing to its good Piezoelectric effect, high sensitivity, and small size.
Different from the explicit detection, measurement parameters obtained by the PZT sensor in the implicit detection do not have a
pronounced relationship with preloads, which cannot be depicted mathematically. In general, some sensitive feature parameters are
extracted from the response signal. Loosening detection is conducted by establishing the relationship between these feature parameters
and preloads experimentally. Based on the difference of detection principles and feature parameters, five typical methods have been
developed, i.e., ultrasonic attenuation, time reversal, vibro-acoustic modulation, electromechanical impedance and excitation analysis
methods. We will introduce these implicit detection methods in detail.

3.2.1. Ultrasonic attenuation method

The ultrasonic attenuation method used to detect loosening contains two devices: actuator and sensor, as shown in Fig. 7. Take a
single lap joint as an example, the actuator is installed on one side of the bottom plate to generate an ultrasonic signal, whereas the
sensor is installed on one side of the upper plate to receive the response signal passing through the joint interface. At the microscale, the
joint interface is composed of many protrusions and depressions and is significantly rough. It is established that the actual contact area
of the joint interface is significantly affected by the preload of the bolted joint. The larger the preload, the larger is the actual contact
area. On the other hand, an increase in the contact area would reduce the attenuation of ultrasonic wave propagating through the joint
interface. Therefore, the attenuation magnitude of ultrasonic wave can be used to represent loosening severity. This is the fundamental
principle of ultrasonic attenuation method for loosening detection.
The process of detecting bolt loosening using the ultrasonic attenuation method mainly comprises two steps: the first step is the
generation of an ultrasonic signal using an actuator; and the second step is to the indication of ultrasonic attenuation. We have
summarised the relevant literature on the ultrasonic attenuation method for loosening detection in Table 2. Most of the actuators used
PZT transducers in the first step. A function generator or NI DAQ [53–69] was linked with the PZT transducer to generate an ultrasonic
signal exploiting the inverse piezoelectric effect. A few scholars also used pulsed lasers [70–72] to generate ultrasonic signals, taking
their advantages of high sensitivity (sub-nanometre level) and high detection bandwidth (GHz). Furthermore, this type of loosening
detection method was not limited to the application of single lap-joint structure, but also truss arch structures [61], girders [62], and
composite bolted joints [63].
The first key technical issue using the ultrasonic attenuation method for loosening detection is how to represent the ultrasonic
attenuation. Energy loss is generally used to represent the ultrasonic attenuation [53,56,57,60,61,65–69]. Based on the Hertz contact
theory [73], Liu et al. [74] analysed mathematically that the transmitted energy of an ultrasonic wave across the joint interface. As
shown in Fig. 8, the joint interface is simplified as a sine wave surface. When a pressure loading F is applied on the contact surface, the
actual contact area becomes less than the nominal contact area, as shown in Fig. 8(b). The relationship between the actual contact area
and the contact pressure can be expressed as,
√̅̅̅̅̅̅̅̅̅
Sa 1 F
= (7)
Sn π p0 Sn

Fig. 7. Fundamental principle of ultrasonic attenuation method for loosening detection.

8
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Table 2
Summary of ultrasonic attenuation for loosening detection.
Actuator Input signal Indication of ultrasonic attenuation Type of threaded fastener Key references

PZT Pulse signal Coda wave energy Single lap joint [53]
Peak amplitude [55]
[65]
[73]
[75]
Swept-frequency wave MMFE Lap joint with three bolts and nuts [54]
MPE [58]
Wavelet packet energy Subsea bolted connections [56]
Steel truss arch structure [61]
Signal energy Single lap joint [67]
Lamb wave Lamb wave energy [57]
Amplitude Composite bolted joints [63]
Multi-peak tone burst signal Correlation coefficient Multi-bolt lap joint [59]
Signal energy ratio Steel grid structures [62]
Power spectral moments Single lap joint [64]
Signal energy Thermal protection systems [66]
First five segments of 100 kHz components Composite bolted joints [68]
Square wave signal Wavelet energy entropy Bolt-jointed rotor [60]
– Signal energy Composite thermal protection panels [69]
Laser transmitter Pulse signal Power Single lap joint [70]
Q-switched Nd:YAG Pulse signal Time–distance–wave amplitude graph Aluminium box girder specimen [71]

Fig. 8. Sine wave of surface model: (a) unloaded; (b) loaded.

where Sa denotes the actual contact area, Sn is the nominal contact area, F is the total load, and p0 is the maximum Hertzian pressure.
The above formula shows that Sa is proportional to the square root of F. Owing to the ultrasonic scattering attenuation at the joint
interface, the energy of the received response signal would dissipate. The energy of the transmitted signal received by sensors can be
expressed as follows:
√̅̅̅̅ √̅̅̅̅
G0 ∝C1 Sa ∝C2 F∝C3 T (8)

where G0 represents the transmitted energy; C1, C2, and C3 are three constants; and T is the tightening torque. Therefore, the trans­
mitted energy has the positive correlation with the tightening degree, confirming the reliability of the above-mentioned detection
principle.
In addition, the decrease of amplitude [55,63,65,75], power [70], and power spectral moments [64] of the response signal in the
frequency domain were also adopted for representing the ultrasonic attenuation. An approximate linear relationship between the
indication of ultrasonic attenuation and tightening torque indicates the superior loosening detection ability. However, many re­
searchers found that when the preload is high, it is difficult to effectively detect a slight variation in preload, as shown in Fig. 9. This is
known as saturation phenomenon. Saturation phenomenon occurs because the actual contact area is almost unchanged when the large

Fig. 9. Saturation effect of ultrasonic attenuation method.

9
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

preload is changed slightly. It means that the early loosening detection cannot be conducted effectively by the ultrasonic attenuation
method. This is the second key technical issue required to be addressed. To solve it, Wang et al. [58] developed a new damage index
based on entropy by using the multiscale permeability entropy (MPE) algorithm. It eliminated the saturation phenomenon and
improved the performance of the ultrasonic attenuation method in the early detection of bolt loosening. In addition, it is also
impossible to use the ultrasonic attenuation to detect the loosening of multiple bolted joints. Recently, Wang et al. [54] proposed a
damage index based on multivariate multiscale fuzzy entropy (MMFE) to construct a data set and used genetic algorithm-based least
square support vector machine (GA-based LSSVM) to train a classifier for loosening detection. Their method could monitor the multi-
bolt connection looseness.

3.2.2. Time reversal method

Time reversal theory [76-79] reveals that an input signal can be reconstructed at the excitation point when the output signal
recorded at another point in the structure is reversed in the time domain and retransmitted back to the excitation point. The process
and principle of time reversal method used to detect loosening are shown in Fig. 10 and the relevant literature on the time reversal
method for loosening detection have been summarised in Table 3. Taking a single lap joint as an example, an input pulse signal is
generated at the excitation point and then propagates through the mechanical system with bolted joints. The response signal is ob­
tained at the record point and reversed in the time domain. Finally, the reverse signal is retransmitted back to the excitation point and
the reconstructed signal is obtained here. In general, PZT patches are bonded on the excitation and record points functioning as an
actuator and sensor simultaneously for generating and receiving signals [80,81]. It is known that a large preload will cause the me­
chanical system with bolted joints more compact; thus, the reconstructed signal is more intensive and approaches the original input
signal due to the less dissipation. That is to say, the magnitude of the reconstructed signal is closely related to the preload level and we
can evaluate loosening severity based on the magnitude of the reconstructed signal [82,83]. Parvasi et al. [84] and Huo et al. [85]
extracted the contact areas under different preloads by finite element modelling and fractal contact theory, respectively. It was
observed that the increase in the magnitude of the reconstructed signal was mainly attributed to the increase in the actual contact area
of the steel plates. Ruan et al. [86] verified the anti-interference capability of the time reversal method in the loosening detection of lap
joint.
In practical engineering applications, threaded connection structures may be not the same with the single lap joint shown in Fig. 10.
For different forms of threaded connection structures, how to install PZT sensors on the structure and effectively conduct loosening
detection exploiting time reversal method is the first technical issue required to be addressed. Typical threaded connection structures
including pipe thread [87], tapered thread [88], rock bolt [89], and bolted spherical joint [90] have been explored using time reversal
method for loosening detection. For example, Huo et al. [89] proposed that a PZT patch transducer, working as an actuator to generate
stress waves, was bonded onto the anchor plate of the rock bolt. A smart washer, fabricated by sandwiching a PZT patch between two
metal rings, was installed between the hex nut and the anchor plate along the rock bolt. This was more suitable for practical appli­
cation. Xu et al. [90] proposed that for the loosening detection of a bolted spherical joint connection, one PZT patch mounted on the
central bolted sphere at the upper chord was used as an actuator to generate a stress wave. Another PZT patch mounted on the bar was
used as a sensor to detect the propagated waves. In addition, the PZT transducers were installed on some non-threaded joints, such as
pin-connected structure [91], cuplok scaffold joint [92] and anchorage connection [93], and their loosening detection using time
reversal method were also researched. Like the ultrasonic attenuation method, saturation problem is the second technical issue to be
considered. To address it, Yin et al. [94] designed a smart washer composed of two annular disks with contact surfaces that were
machined into convex and concave respectively to eliminate the complete flat contact surfaces and to reduce the saturation effect.
The above time reversal method used for loosening detection may be inconvenient because it requires the time reversal of the signal
during each detection. Cai et al. [95] modified the traditional time reversal method and proposed the virtual time reversal (VTR)

Fig. 10. Fundamental principle of time reversal method for loosening detection.

10
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Table 3
Summary of the time reversal method used for loosening detection.
Method Structure form Fastening indicator Main research Key
reference

Time Lap joint Axial force The fractal contact theory was introduced to explain the detection [85]
reversal principle.
Studied the effect of surface roughness on the saturation effect. [82]
The non-planar contact washer eliminated the saturation effect. [94]
Finite element analysis was introduced to explain the detection [84]
principle.
Torque Studied the capability of resisting sound interference [86]
Application in bolted connection of wood structure [84]
Replaced PZT patch with smart washer [81]
Cuplok scaffold joint Tightening Circles/ Application to cuplok scaffold joint [92]
Tapered thread angles Application to tapered thread [88]
Pipe thread Application to pipe thread [87]
Rock bolt Pressure Application to rock bolt [89]
Anchorage connection Application to anchorage connection [93]
Pin-connected structure Axial force Application to pin connection [91]
Bolted spherical joint Torque Application to bolted spherical joint [90]
VTR Thermal protection – Realised the identification and positioning of looseness. [96]
system
L-shaped bolted Torque Studied the advantages of the normalised amplitude. [97]
structures Studied the influence of ultrasonic frequency on the result.
Proposed a novel index independent of baseline state. [98]
Studied the effect of roughness and temperature on the detection
results.
Lap joint Comparative experiment of wave energy dissipation, OTR, and VTR [99]
methods.

method for loosening detection. In the VTR method, the time reversal is performed only one time for the response signal in the healthy
state. The process of bolt loosening detection by VTR method is shown in Fig. 11. An input pulse signal generated at the excitation point
is applied to the structure of bolted joints in the healthy state. A sensor at the record point collects the response signal and subsequently
the response signal is reversed in the time domain. The reversed signal obtained in the healthy state is recorded as the reference
reemitting signal (RRS). The RRS at the excitation point will be input to the same structure with different loosening states. Finally, the
magnitude of the reconstructed signal at the record point can be used to represent loosening states. In the existing research on VTR for
loosening detection, the magnitude of the reconstructed signal can be indicated by the normalised value of peak amplitude [96,97], the
proportion of the focused wave packet energy in the entire signal energy [98], and the phase shift of the focused signal [99]. These
sensitive features (also called damage indicators) can be used to represent different loosening states by establishing the corresponding
relationship with the tightening torque.
The main contribution of the VTR method is the loosening detection of multi-bolt structures, e.g. thermal protection system (TPS)

Fig. 11. Process of bolt loosening detection by virtual time reversal method.

11
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

steel plate structures [96] and L-shaped structures with multiple bolts [97,98]. For TPS steel plate structures, Wu et al. [96] con­
structed a PZT array with four sensors. They applied principal component analysis and neural network to identify and locate loosening
of multiple bolts. For L-shaped structures, Du et al. [97] observed that the normalised peak of a focused signal can reflect bolt loosening
sensitively and accurately. They also discussed the influence of ultrasonic frequency on detection sensitivity. Du et al. [98] proposed a
novel damage index that effectively prevents the dependence on the baseline state. In addition, the influence of temperature and
interface roughness on the detection results was analysed. The VTR method also was highly effective for detecting the loosening of
single-bolted joints. For example, Xu et al. [99] carried out experiments on a steel beam overlapped by a single bolt and proposed the
phase shift of the focused signal as the damage index. Three methods (the wave energy dissipation, TR, and VTR methods) were
compared. It was concluded that the VTR method is most effective for loosening detection of single-bolted joints.

3.2.3. Vibro-acoustic modulation method

Vibro-acoustic modulation (VAM) method uses contact acoustic nonlinearity [100] to detect loosening for solving the issue of
saturation effect. The process and principle of VAM-based loosening detection are shown in Fig. 12. Taking a single lap joint as an
example, two excitation signals are generated and fed into the structure: one is the low-frequency (LF) sinusoidal vibration (pumping
vibration) and the other is the high-frequency (HF) ultrasonic wave (probing wave). It is known that the joint interface is rough with
many protrusions and depressions at the microscale. When the pumping vibration and probing wave are mixed and cross the joint
interface, the low-frequency pumping vibration causes the “imperfect” joint interface to generate “close” and “open” motions (i.e.,
“breathing” effect), which will modulate the propagation characteristics of the high-frequency probing wave. On the other side, the
response signal containing nonlinearities is captured by a sensor (generally PZT transducer). Fast Fourier Transform (FFT) can be used
to obtain the response spectrum, where two nonlinear features, i.e., the left sideband (LS) and right sideband (RS) appear. A reduced
preload results in the actual contact area being small and the large contact nonlinearity; thus the amplitudes of the LS and RS will be
increased. Therefore, the amplitudes of the LS and RS can be used to indicate the loosening severity. This is the fundamental principle
of VAM-based loosening detection.
Some scholars validated theoretically the VAM method for bolt loosening detection based on the contact acoustic nonlinearity
(CAN) theory [100]. By simplifying a bolted joint as a single-degree-of-freedom system (as in Fig. 13), we can link the nonlinear
contact stiffness (which is caused by micro-scale interface roughness) to the bolt preload. The motion equation of the bolted joint
adopts the form shown in Eq. (9) when the joint is subjected to a mixed excitation of LF pumping vibration (denoted as F1cosω1t) and
HF probing ultrasonic wave (denoted as F2cosω2t), which are independent of each other [101]:
M ẍ + K1 x = F1 cosω1 t + F2 cosω2 t (9)

where M is the mass; t is the time; x1 and x2 are the frequencies of the LF pumping vibration and HF probing ultrasonic wave,
respectively; K1 and K2 are the linear and nonlinear contact stiffness, respectively; and ε denotes a marginal quantity to scale the
minimum perturbation. Subsequently, the solution to Eq. (9) can be expressed as follows according to the perturbation theory:
x = x1 + x2 (10)

where x1 and x2 represent the linear and nonlinear dynamic response, respectively, (caused by sidebands) of the joint.
Furthermore, after substituting Eq. (10) into Eq. (9), we equate and then eliminate the ε-related terms on both sides of the equation.
The following equation is obtained:

Fig. 12. Process and principle of VAM-based loosening detection.

12
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 13. Simplified model of VAM-based method for a bolted joint.

Mx1 = K11 F1 cosω1 t + F2 cosω2 t

(11)
Mx2 + K1 x2 = K2 X12

Finally, we can derive the magnitude of the sideband. It is proportional to the nonlinear contact stiffness K2 (as well as the bolt
preload) in the spectrum as given in Eq. (12) [101]:
G1 G2 G1 G2
xsidebands = K2 cos(ω1 + ω2 )t + K2 (ω2 - ω1 )t (12)
K1 − M(ω1 + ω2 )2 K1 - M(ω2 - ω1 )2

where G1 = K1 − FM
1
ω2
and G2 = K1 − FM
2
ω2
. It is apparent that the left sideband (LS) and right sideband (RS) are positioned at the frequencies
1 2

of ω2 ∓ ω1. For different preloads, the linear and nonlinear contact stiffness K1 and K2 are changed accordingly, causing the magnitudes
of LS and RS to vary. It demonstrates the VAM-based method for loosening detection.
Vibro-acoustic modulation for bolt loosening detection comprises the following three steps (see Fig. 14). 1) Input the generated
high-frequency probing wave and low-frequency pumping vibration into the structure of bolted joint. 2) Extract sensitive loosening
features from the collected modulation response signal. 3) Establish the relationship between the loosening feature and preload/
tightening torque to realise loosening detection of the bolted joint.
The research on vibro-acoustic modulation for bolt loosening detection is summarised in Table 4. Initially, many research results
showed that both amplitude and frequency of HF and LF vibration signals have a significant impact on the modulation response signal
and loosening detection. Therefore, the first key technical issue to be considered is how to select appropriate LF and HF vibration
signals to realise the optimum detection performance. The LF vibration signal can be divided into two categories: impact signal and
harmonic vibration signal. The former is a pulse signal generated by an impact hammer [102,103]. This form of vibro-acoustic
modulation is also called impact modulation (IM) in certain literature. The latter mainly generates a sine wave signal through a
direct current (DC) driver [104], shaker [105–108], or PZT patch [109–113]. The frequency of the sine wave used varies across the
related research. Zhang et al. [106–108] determined the natural frequency of the bolted joint by frequency sweep first. Then, they
considered the strongest natural frequency in the second-order response signal as the frequency of the LF vibration signal. Zhou et al.
[105] directly applied the first three natural frequencies of the tested structure as the frequency of the LF vibration signal. Wang et al.
[111] adopted the natural mode with the largest vibration intensity. Based on the actual vibrational environment of the bolted joint,
this vibration signal was tested and directly adopted as the LF vibration signal [104,112]. In addition to single vibration frequency,
Fierro et al. [109], Zhao et al. [110] and Wang et al. [113] adopted a certain range of sweep frequency as the input frequency of LF
pump vibration. The HF vibration is essentially harmonic vibration produced by PZT. Furthermore, the frequency of HF vibration

Fig. 14. Three steps for bolt loosening detection using VAM method.

13
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Table 4
Summary of vibro-acoustic modulation method for loosening detection.
Frequency of LF signal Frequency of HF signal Loosening index Tightness amount Key reference

100 Hz–2 kHz 30–50 kHz Gnome entropy Tightness cases [113]
3 Hz 150 kHz Eq. (13) Torque [112]
1 Hz 120 kHz Eq. (13) Tightness (%) [104]
7918 Hz 10 kHz Eq. (17) Torque loss [103]
100 Hz–2 kHz 100–300 kHz Eq. (16) Torque [111]
100 Hz–2 kHz 100–300 kHz NAMEMD and MMSE Torque [110]
100–200 kHz 100–200 kHz Zero-order acoustic moment Three bolt states [109]
992 Hz and 758 Hz 14.24 kHz and 14.99 kHz Eq. (16) Residual torque remained [108]
758 Hz 14.89 kHz Eq. (15) Residual torque remained [107]
992 Hz, 1145 Hz, and 1303 Hz 14.240 kHz, 14.777 kHz, and 15.049 kHz Eq. (14) Torque [106]
1–150 Hz 200 Hz Energy-based damage index Tightness cases [105]
7500 Hz 10510 Hz Eq. (17) Residual torque remained [102]
2100 Hz 100 kHz Eq. (14) Torque [114]

signal selected also varies across studies in the existing literature. For example, Amerini et al. [114] selected a certain frequency
according to past research experience. Zhao et al. [110] and Wang et al. [113] determined the vibration frequency according to an
experiment on the strongest response signal. Zhang et al. [106–108] and Wang et al. [111] selected the sweep signal in a certain range
as the HF vibration signal.
In the second step, the modulation response signal is generally collected by PZT [102,110–114], accelerometer [103–108], and
acoustic emission sensor [109]. The second key technical issue for the VAM-based method for loosening detection is the extraction of
sensitive loosening features from the response signal. For HF and LF signals in the sine wave form, most scholars put forward various
loosening indexes (LIs) (i.e. loosening features) based on the amplitudes of the left and right sidebands and those of the high- and low-
frequency vibration signals. Four typical LIs in the previous research are given [104,106–108,112]:
A+ + A−
LI1 = (13)
2 × AH

A+ + A−
LI2 = − AH (14)
2

A+ + A−
LI3 = − A (15)
2

A+ + A−
LI4 = − (AH + AL ) (16)
2

where A− and A+ represent the amplitudes of the left and right sidebands, respectively; AH and AL represent the amplitudes of the high-
frequency and low-frequency vibration signals, respectively. For the IM method, the frequency of the impact signal varies. Scholars
quantified the intensity of the modulated response signal in the form of an integral [102,103]:
∫ f2
Xdf
(17)
f
I = ∫ ω1 2
ω1 Xdf

where f1 and f2 are the modulation frequencies of interest; X is the response amplitude; f is the frequency variable; and ω1 and ω2 are
the high and low frequencies, respectively, produced by impact.

Fig. 15. Relationship between loosening features and torque. a) curve b) histogram.

14
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

In addition, other loosening features were also constructed for the HF and LF signal in the swept form. For example, Zhou et al.
[105] proposed an energy-based damage index according to high-frequency natural mode decomposition. Fierro et al. [109] adopted
the zero-order acoustic moment as the quantitative LI. Zhao et al. [110] established a new damage index for early bolt loosening
detection by using noise-assisted multivariate empirical mode decomposition (NAEMD) and multi-scale multiple sample entropy
(MMSE). Wang et al. [113] proposed the extraction of loosening feature sets by quantifying the dynamic characteristics of compressed
signals through the Gnome entropy (gEn).
In the third step, the relationship between the loosening features and the degree of loosening is established for loosening detection.
Most scholars directly depicted the relationship curve or histogram between LI and tightening torque [102–111,114] to achieve
loosening detection, as shown in Fig. 15. Other scholars input the loosening features into a classifier such as random forest classifier
[113] and artificial neural network [112]. Loosening detection is realised by classifier training.
It is noteworthy that the above studies focused mainly on loosening detection for a single lap joint with one bolt
[102,106–108,110,111,114]. How to detect the loosening of multiple bolted joints is the third key technical issue. A few scholars
applied the VAM method to a mechanical structure composed of multiple bolted joints, e.g. a flange structure [104,112], frame
structure [105], and satellite structure [103]. However, only one bolt can be detected each time. Fierro et al. [109] applied the VAM
method to an L-shaped structure with three bolted joints. The proposed method could detect only three states: complete tightening,
tightening, and complete loosening. Wang et al. [113] applied the VAM method to a lap joint structure with three bolts. The exper­
imental results showed that the method used could detect only the tightening and loosening states of the three bolts. Therefore, further
research would be directed toward the application of the VAM method to detect the loosening of multiple bolted joints simultaneously.

3.2.4. Electromechanical impedance method

The EMI method has apparent advantages compared with other SHM methods, such as good real-time performance and low re­
quirements of sensor and data processing [115–123]. Fig. 16 shows the process and principle of electromechanical impedance (EMI)-
based loosening detection. It is known that the mechanical impedance of a bolted connection structure would vary when loosening
occurs [124]. However, the actual mechanical impedance is difficult to determine. The EMI-based method bonds a PZT patch on the
bolted connection structure and utilizes the electromechanical coupling characteristics of the piezoelectric material. The electrical
impedance of PZT is directly related to the mechanical impedance of the bolted connection structure. Bolt loosening can be determined
by measuring the electrical impedance using an impedance analyser and comparing it with the reference value. Liang et al. [125] first
proposed a one-dimensional theoretical model to explain the working principle of the EMI method for loosening detection. Fig. 17
shows the simplified theoretical model with an integrated PZT and a host structure system. In this model, the PZT is fixed at one end,
and the host system is regarded as a mass–spring–damper system with one degree of freedom. Under an input V = vsin(ωt), the
impedance of the coupled system is affected by the dynamic characteristic of the PZT and the adjacent region of the host structure. In
this system, the electrical admittance Y(ω) of the PZT patch, which is the inverse of the electrical impedance, is expressed as,
{ ( ) ( )}
υl − T Za (ω) tanκl
Y(ω) = jω 2
ε33 − d31 Y− E + 2
d31 Y− E (18)
h ZS (ω) + Za (ω) κl

where Za(ω) and ZS(ω) represent the mechanical impedance of the PZT patch and host structure, respectively, at an angular frequency
ω; v, l, and h represent the width, length, and thickness, respectively, of the PZT patch; k is the wave number; d31 presents the
piezoelectric constant in the x-direction at zero stress; ε33 is the complex dielectric constant of the piezoelectric material under zero
stress; and Y− E is the complex modulus in the x-direction under zero stress. This model indicates that the electrical impedance of the
PZT patch has a quantitative relationship with the mechanical impedance of the host structure. This theoretically supports EMI-based
loosening detection.

Fig. 16. Process and principle of EMI-based loosening detection.

15
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 17. One-dimensional theoretical model of electromechanical impedance.

We summarise that bolt loosening detection using the EMI method comprises three steps (see Table 5): 1) The PZT patch is bonded
on the host structure, and a high-frequency sweep signal is input into the PZT patch. 2) The impedance signal measured by impedance
analyser is analysed to determine the damage sensitive frequency band, and loosening features. 3) The relationship between the
loosening features (also called evaluation indexes) and preloads (or tightening torques) is established to realise loosening detection.
For the first step, the key technical issue is the combination of a PZT patch and the host structure. The configurations according to
the placement of PZT patches can be classified into the following four types (see Fig. 18). In the first type, two PZT patches are bonded
to the host structure that is clamped by bolt [126] (see Fig. 18(a)). In the second type, a PZT patch is bonded to the outer surface of a
smart washer under a bolt head [127–130] (see Fig. 18(b)). In the third type, a PZT interface [131–135] is combined with the host
structure (see Fig. 18(c)). In the fourth type, two PZT patches are bonded to a wearable sensor device, which is combined with the host
structure [136,137] (see Fig. 18(d)). In addition, Ryu et al. [137] combined wearable sensor device technology with PZT interface
technology and developed a wearable PZT interface to realise real-time monitoring of cable axial load. There are two PZT configu­
rations for the process of measuring impedance, as shown in Fig. 19. In one of these, two PZT patches are used to generate vibration
excitation and collect structural response signals. This configuration enables the measurement of frequency response in the vibration
propagation path [127,138,139]. In the other configuration, a PZT functions as an actuator and a sensor simultaneously. It utilizes the
positive and negative piezoelectric effects of piezoelectric materials to measure the frequency response of a point [52,140–146]. An
impedance analyser [147–149] was applied to measure the impedance of PZT in certain research. The high cost, bulk, and high power

Table 5
Summary of electromechanical impedance method used to detect loosening.
Equipment Combination form PZT Loosening index Host structure Sweep range Key
configuration reference

Impedance analyser Washer Both RMSD and CC Pipe 88–100 kHz; [127]
28–52 kHz
Single 3 dB RMSD Plate 30–100 kHz [130]
RMSD Lap joint 10 kHz–1 MHz [128]
Specimen 10–1000 kHz [129]
PZT interface Beam 10–50 kHz [133]
[135]
CCD, RMSD [131]
Wearable Sensor Peak frequency shift Cable 0.2–15.5 kHz [137]
Device Normalised RMSD Flange 100–180 kHz [136]
PZT patch RMSD, RMSCR, MAPD, Flange 350–650 kHz [141]
CCD, COV
RMSD Bolted spherical joints 100–300 kHz [145]
Pin-connection 40–300 kHz [144]
Threaded pipe 100–300 kHz [147]
Plate 100–250 kHz [148]
Peak frequency shift Bolt clamp 2 kHz–5 MHz [52]
CC Beam, Pipe, Bridge 10–100 kHz [142]
– Lap joint 10–1 MHz [140]
Aluminium plate 500–2500 Hz [149]
AD5933 evaluation Double RMSD, CC Beam 50–80 kHz [139]
board Single RMSD Aircraft wing segment 60–70 kHz [146]
model
Plate 50–52.6 kHz [150]
PZT interface Beam 10–50 kHz [132]
CCD, RMSD Beam 10–50 kHz [134]
– PZT patch CC Beam 10–20 kHz [158]
MAPD Tail beam of Mi-8 200–1000 kHz [143]
helicopter
Double IIG Lap joint 15–30 kHz [138]

16
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 18. Four methods of combining PZT patches and host structure.

Fig. 19. Two configurations of PZT patches.

of impedance analysers limit the large-scale use of the EMI technology. A few scholars applied the impedance measurement platform
[150–155] integrated with an AD5933 impedance chip to realise a low-cost, flexible, and convenient impedance method to detect bolt
loosening.
In the second step, the key technical issue is the selection of the optimal frequency range in the impedance signal, related to the
detection capability of the EMI method [156]. Researchers determine the appropriate frequency range mainly by the following three
methods: 1) trial-and-error method to investigate several frequencies with 20–30 peaks to select the appropriate frequency range, 2)
application of neural network to identify the most appropriate frequency range [157], and 3) generate sensitive impedance response in
a pre-defined frequency band by designing the geometry and material characteristics of the PZT interface [134]. In general, the
frequency impedance response signal would reveal the peak variation and frequency shift, with variation in bolt preload. How to
construct sensitive loosening features (i.e. damage indicators) based on the optimal frequency range is another key technical issue.
Various damage indicators employed to characterise bolt loosening contained the cross-correlation coefficient (CC) [127,139,142],
mean absolute percentage deviation (MAPD) [143], root mean square deviation (RMSD) [144–150], cross-correlation deviation (CCD)
[131,141,158], and peak frequency shift [52,137]. They are illustrated in Eqs. (19)–(22).
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑[ ( ) ]2̅ ∑ [ ( 0 ) ]2
RMSD = Re(Yi ) − Re Y 0i / Re Yi (19)
N N

∑ [ ( )] ( )
MAPD = | Re(Yi ) − Re Yi0 /Re Yi0 | (20)
N

1 ∑{ [ ( ) ( )]}
CCD = 1 − [Re(Yi ) − Re(Y) ] ⋅ Re Yi0 − Re Y 0 (21)
N σY σY 0 N

[ ( ) ]
1 ∑ [Re(Yi ) − Y ] ⋅ Re Yi0 − Y 0
CC = (22)
N N σY σY0

The experimental results showed that the RMSD and CCD could effectively reflect the degree of bolt loosening. The RMSD is
sensitive to the vertical displacement (i.e. amplitude variation) and horizontal displacement (i.e. frequency displacement) of
impedance signals. Meanwhile, the CCD is sensitive only to the horizontal displacement of impedance signals. By monitoring the PZT
admittance attached to a flange surface, Jiang et al. [141] compared the effect of different damage indicators on loosening detection
for flanged bolts. The RMSD was observed to display a higher noise resistance than the CCD.
However, the EMI method used to detect loosening is straightforwardly affected by environmental factors such as temperature and
humidity. Under an identical damage condition, different temperatures may cause substantial variations in impedance amplitude and
peak frequency [159,160]. Huynh and Nguyen [161] applied the EFS-algorithm to minimise the influence of temperature on the
impedance response of a prestressed tendon anchorage, and analysed the influence of temperature and preload loss by using the finite

17
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

element method. An et al. [138] proposed a new damage index IIG based on impedance and guided wave technology. It considers the
influence of temperature on impedance and guided wave. The monitoring of different damage types (such as loosening) under different
temperatures is realised. Koo et al. [162] proposed the effective frequency shift (EFS) method. By this method, a new damage index
was used to compensate for the effect of temperature on the EMI method. Recently, Na [163] combined EMI technology with prob­
abilistic neural networks (PNN) to predict the torque loss of the bolt structure. Zhou et al. [164] proposed a new hybrid model to realize
the multi bolt loosening detection. They used multiple PZTs to form sensor networks, and input the EMI information and sensor
location information into graph revolutionary networks (GCN) for training. Their method can detect the state of multiple bolted
connections at the same time and give quantitative torque loss, which is an improvement of the traditional EMI method.

3.2.5. Excitation analysis method

For a bolted joint structure, it is known that bolt loosening will influence its whole dynamic characteristics. When an actuator
attached applies an excitation to the structure, the response signal received by another sensor will be changed thereby in time, fre­
quency or time–frequency domain because of the change of dynamic characteristics. Loosening can be evaluated based on the analysis
of the response signal in time, frequency or time–frequency domains. How to find sensitive features from the time, frequency and
time–frequency signal is the key technical issue for the loosening detection based on excitation analysis method. This is the funda­
mental principle of excitation analysis method for loosening detection, as shown in Fig. 20. Subsequently, we will summarise the
research on signal analysis and feature extraction for loosening detection in time, frequency or time–frequency domains, respectively.
1) Time domain analysis
The relevant research on time domain analysis for loosening detection is summarised in Table 6. Different loosening models and
damage indices were established from the response signals in the time domain for loosening detection. Hasni et al. [165] proposed a
new sensor fusion model to effectively detect and locate bolt loosening and crack in steel frame structures. It used the histogram fitting
Gaussian mixture model (GMM) output from a hybrid network of piezoelectric strain gauge and acceleration sensor to define the initial
damage index features. Li et al. [166] proposed two novel damage indices (damage detection index and damage localisation index)
based on the peak coherence function of short-time temporal coherence (STC). With these two indices, the free vibration response
signals in the loose and tight states were analysed to detect and locate bolt loosening. Based on the concept of virtual beam-like
structures, Wang et al. [167] carried out an effective method of fault feature extraction and fault diagnosis for bolt loosening and
its surrounding cracks with limited prior knowledge.
In addition, a few scholars extracted certain characteristics of response signals in the time domain and input them into machine
learning models for intelligent detection of bolt loosening. For example, Olson et al. [168] extracted root-mean-square (RMS) values
for the segments of response signals recorded by each sensor in a square aluminium plate of a thermal protection system. Then the RMS
values were imported to the classification system based on statistical pattern recognition methods for estimating the tightening state of
fasteners. However, this classification system showed low performance in terms of detection of severity of damage. Liang et al. [169]
considered the peak variation in strain in the time domain as the loosening feature. Then, the loosening feature was input into the
fusion decision system, based on which the bolt loosening could be recognised rapidly and accurately. Zhang et al. [170] directly
obtained the acceleration response curves of a structure and regarded these as input for a one-dimensional convolution neural network.
The detection of local damage caused by bolt loosening in a multi-scale beam structure was realised using the proposed method.
2) Frequency domain analysis
The relevant research on frequency domain analysis for loosening detection is summarised in Table 7. New methods were proposed
to analyse the frequency domain signal. Li et al. [171–173] applied the second-order output spectrum (SOOS) method. The SOOS

Fig. 20. Process and principle of excitation analysis method for loosening detection.

18
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Table 6
Summary of time-domain analysis.
Excitation signal Signal analysis method Loosening feature Classifier Application Key
structure reference

Periodic vibration Gaussian mixture model (GMM) Sensor voltage output – Steel frame [165]
Pulse Analytic mode decomposition (AMD), Peak coherence Flange [166]
Short-time temporal coherence (STC)
Frequency- Fuzzy decision method Peak-to-peak value Complex beam [167]
sweeping et al. structure
excitations
Pulse – Time domain CNN, SVM Steel beam [170]
acceleration response
Sine wave Decision fusion system Peak change of strain SVM et al. Aluminium plate [169]
Swept frequency Signal segmentation Root-mean-square Statistical pattern recognition Square [168]
sinusoid (RMS) classification system aluminium plate

method was combined with other methods to describe the dynamic characteristics of a system and realise the detection and location of
bolt loosening in a complex structure. Pal et al. [174] proposed a fixity factor to represent the tightening state of a bolt using the
frequency data. The actual position and fixation coefficient of a bolt under different loosening levels were obtained by adopting the
location-determined fixity factor as a variable to update the model continuously. The nonlinear characteristics of excitation response
signals differ statistically across loosening states. Therefore, Nichols et al. [175] compared and analysed the differences to detect
whether bolt loosening occurred.
Modal analysis is also an important type of frequency domain analysis of structural dynamics. The critical modal parameters
include natural frequency, mode shape, and damping ratio. Natural frequency is an index that is widely used to estimate the degree of
loosening. Okugawa [176] employed the sub-space state space identification algorithm (4SID) to identify the natural frequency of a
smart washer. It was observed that the natural frequency of the smart washer was related to the tightening force of the bolted joint.
Nakahara et al. [177] focused on the natural mode of a thin plate sensor and studied the relationship between its 12th mode and
preload. Huda et al. [178] researched the high-frequency vibration response of a six-bolt cantilever beam. The Taguchi method was
used to evaluate the frequency response data, and a damage index for detecting and identifying bolt loosening was proposed. Hosoya
[179] regarded the variation in the local modal frequency of a bolt head in a high frequency region as a function of bolt locking force.
The bolt locking force was determined by analysing the first four local natural frequencies of the bolt head. Sah et al. [180] investigated
the relationship between the transverse natural frequency (particularly the first transverse natural frequency) and bolt tightening state.
Based on Bayesian operational modal analysis and the modal strain energy analysis method, Hu et al. [181] established the rela­
tionship between damage and the loosening index of a beam structure under environmental excitation. Hosoya et al. [182] studied the
natural frequency of the bending mode of a threaded part protruding from a nut. They succeeded in measuring the axial force of the
bolt. In addition, through comparative experiments, Caccee et al. [183] and Anginthaya et al. [184] determined that high frequency
transmission functions between pairs of sensors and transfer functions were more reliable than the low-order modal parameters of
mechanical system in terms of loosening detection.
3) Time–frequency analysis

Table 7
Summary of frequency-domain analysis.
Excitation signal Signal analysis method Modal parameters Application structure Key
reference

Symmetric harmonic Second-order output spectrum (SOOS) – Satellite-like structure [171–173]

Pulse Particle swarm optimisation (PSO) – Steel frame [174]
Ambient ‘wave’ forcing Iterative amplitude adjusted Fourier transform method – Composite beam [175]
(IAAFT)
Controlled vibration Low frequency modal analysis, High frequency transfer Low order Composite plate and [183]
functions and transmittance functions analysis steel frame
Pulse Modal analysis, Transmissibility function analysis Low order Stainless steel beams [184]
Pseudo random binary Sub-space state space identification algorithm (4SID) 4th order L-type angle iron [176]
signal
Pseudo random signal Local vibration mode analysis 12th order Bolted joint with thin [177]
plate sensor
Nd:YAG pulse laser Recognition: Taguchi method 16th order Six-bolt joint cantilever [178]
Pulse Mode characteristic analysis 4th order Simple-shaped integral [179]
structure
Pulse Mode characteristic analysis First transverse Test rig [180]
natural frequency
Pulse Mode characteristic analysis Bending mode Single-piece cube [182]
frequency
Environmental excitation Bayesian operational modal analysis Modal strain energy Cantilever beam [181]
and pulse

19
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

The relevant research on time–frequency analysis for loosening detection is summarised in Table 8. On the one hand, the fusion of
time domain and frequency domain features was used for loosening detection. Chen et al. [185] used the manifold learning algorithm
to reduce the dimension of the mixed domain feature set obtained from a fusion response signal. The mixed domain feature set includes
time-domain feature, frequency-domain feature, and frequency response function. The reduced feature set was input into the weighted
k-nearest neighbour classifier (WKNNC). Wei et al. [186] directly analysed the acceleration response signal of a rail by wavelet packet.
They proposed two damage indexes to determine the position and degree of loosening of the rail fastener.
In addition, empirical mode decomposition (EMD) is a new adaptive method for processing time–frequency signal. It is particularly
suitable for non-linear and non-stationary signals. Scholars decomposed response signals by EMD, calculated the intrinsic mode
function (IMF), and established different bolt loosening damage indexes [187–189]. Xu et al. [190] proposed a state-of-the-art
nonlinear system identification method based on EMD. It applied the dynamic response signal of a structure to construct a
nonlinear reduced-order model. The damage index calculated by the reduced-order model could sensitively indicate bolt loosening.
Recently, Wang [191] adopted the method of EMD and correlation analysis to effectively identify bolt preload. The proposed method is
based on the frictional effect of the connect interface and the values of the IMFs.
The advantages and disadvantages of various implicit loosening detection methods have been summarised in Table 9. For the
ultrasonic attenuation method and time reversal method, linear features are extracted for loosening detection. Therefore, they both
have apparent saturation phenomenon and are insufficient to recognise slight preload variation. In comparison, vibro-acoustic
modulation method is more sensitive to slight loosening due to exploiting nonlinear features, such as sideband. However, many tri­
als should be performed for selecting appropriate HF and LF signals for better performance. The electromechanical impedance method
has significant advantages and disadvantages. The advantages are good real-time performance, low requirements of sensor and data
processing while the disadvantages show that the detection accuracy is straightforwardly affected by environmental factors such as
temperature and humidity. The last excitation analysis method has wide engineering applications due to the convenient implement;
but the detection results are disturbed easily by noise and advanced signal processing methods are required for features extraction and
analysis.

4. Percussion-based loosening detection

Considering that the dynamic characteristic of a bolted connection structure would vary when loosening occurs, the percussion-
based method (also called the tapping method) is applied in the detection of bolt loosening. The fundamental principle of
percussion-based loosening detection is that the variation in dynamic property caused by loose bolts will influence the audio signal
generated by knocking. An experienced inspector can assess whether the bolt loosens by hearing the audio signal [192–195]. However,
this method depends substantially on the aural capability and previous experience of the inspector. It is difficult for an inexperienced
person to accurately distinguish the loosening state of the bolt from the audio signal produced by knocking.
To overcome the above-mentioned disadvantage, in recent years, an amalgamation of the tapping-based method and artificial
intelligence technology [196] has been proposed (See Fig. 21 and Table 10). Supervised learning based on machine learning algo­
rithms was generally used. The difficulty is how to extract representative features from the audio signal and select an appropriate
classifier. Initially, researchers [197,198] adopted some principal components in time domain and Mel-scale frequency cepstral co­
efficients in frequency domain as signal features, and the classical support vector machine was mainly used to train and test the dataset.
Subsequently, frequency domain features became preferences. For example, Kong et al. [199] employed the power spectrum density
(PSD) of the recorded signal with selected frequency segment as signal features and fed them into a decision tree; Yuan et al. [200]
adopted the improved multiscale sample entropy (IMSE) values of certain intrinsic mode functions (IMFs) decomposed from the
percussion-caused audio signals as condition indicators in bolt loosening detection. Above feature selections were all manual based on
prior knowledge. Considering this, Wang et al. [201] applied the multifractal analysis and the joint mutual information maximization
method to select automatically the feature sets and fed them into a gradient boosting decision tree for avoiding to highly depend on
experienced personnel. In addition, they also proposed an analytical model [202] to describe the acoustic behaviour of bolted con­
nections under impact excitation. Combined with the layer-wise theory and acoustic radiation mode method, analytical modelling of
impact-induced radiation sound pressure level was carried out. This has guiding significance for the selection of appropriate feature
sets in model training. In another paper of Wang et al. [203], the loosening detection of multiple bolts by the tapping method was
explored for the first time using the Mel-frequency cepstral coefficient and the memory-augmented neural network.

5. Vision-based loosening detection

When loosening of a bolted joint occurs, the nut will rotate a certain angle against the bolt, causing a difference with the initial
tightening state. The process of the vision-based loosening detection method is shown in Fig. 22. We can find that this type of detection
method first captures images containing target bolts at the tightening and loosening states using digital cameras, mobile phones, or
other equipment. The images are then analysed by exploiting the technologies of image processing, machine vision and others. We can
identify whether the target bolts are tightened or loose, or evaluate the loosening severity. Compared with sensor-based and
percussion-based loosening detection methods, the vision-based method has the following advantages. First, it displays higher mea­
surement accuracy and efficiency and adapts to a variety of complex engineering environments. Second, it is less affected by envi­
ronmental factors such as temperature and humidity.
According to the detection results, the existing vision-based loosening detection methods are divided mainly into the following four
categories (see Fig. 23). 1) Bolt missing detection (shown in Fig. 23(a)): It is the most severe form of loosening of bolted joint because

20
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Table 8
Summary of time–frequency analysis.
Excitation signal Signal analysis method Loosening feature Classifier Application structure Key reference

Pulse ONPE Root mean square frequency et al. WKNNC Fan foundation [185]
Wavelet packet analysis WPT component energy – Rail [186]
EMD IMF Flanged pipe assembly, Plates [187–189]
EMD Peak frequencies Beam [190]
EMD IMF Beam [191]

Table 9
Advantages and disadvantages of five implicit loosening detection methods.
Implicit detection methods Advantages Disadvantages

Ultrasonic attenuation method Convenient to implement; Simple feature extraction Saturation phenomenon; Not sensitive to slight preload
variation
Time reversal method Strong anti-noise and anti-interference abilities Saturation phenomenon; Complicated implement process
Vibro-acoustic modulation Nonlinear features; Sensitive to slight preload variation Many trials for selecting HF and LF signals
method
Electromechanical impedance Good real-time performance and low requirements of sensor Detection accuracy is affected easily by temperature and
method and data processing humidity
Excitation analysis method Convenient to implement; wide applications Disturbed easily by noise; Requirement of advanced signal
processing methods

Fig. 21. Fundamental principle of percussion-based loosening detection.

Table 10
Summary of tapping method used for loosening detection.
Recording Feature extraction Classifier Structure Key
equipment reference

Smartphone Principal components and Mel-scale frequency Support vector machine Iron plates installed nine standard [197,198]
cepstral coefficients bolts and nuts
Smartphone Power spectrum density Decision tree 12-bolt subsea flange [199]
Smartphone Improved multiscale sample entropy of certain BP neural network 12-bolt subsea flange [200]
intrinsic mode functions
Microphone Joint mutual information Maximization Gradient boosting decision 12-bolt subsea flange [201]
Microphone Mel-frequency cepstral coefficient Memory-augmented neural Spatial bolt-ball joint [203]
network

21
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 22. Process of vision-based loosening detection.

Fig. 23. Four categories of vision-based loosening detection methods.

the bolt and nut fall off completely. 2) Qualitative detection of whether the bolt is loose (as shown in Fig. 23(b)): The bolt is considered
to be loose when the rotational angle between the bolt and nut varies or the length of exposed bolt increases. 3) Quantitative method
where the rotational angle of bolt head relative to the initial state is calculated for loosening detection for bolted joints (shown in

Table 11
Summary of bolt-missing detection.
Feature extraction Classifier Key reference

wavelets transform and independent component analysis Support vector machine [204]
Wavelets transform and principal component analysis Artificial neural network [205]
Wavelet transform Artificial neural network [206]
Gabor wavelet transform Support vector machine [207]
Jones algorithm AdaBoost [208,209]
Gradient orientation Support vector machine [210]
Improved edge orientation histograms and macroscopic local binary pattern Bayesian compressive sensing [211]
Discrete wavelet transforms Two multilayer perceptron neural classifiers [212]
Local binary pattern encoding Support vector machine [213]

22
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

Fig. 23(c)). 4) Quantitative method where the length of exposed bolt is measured for loosening detection for bolted joint (shown in
Fig. 23(d)). We will summarise in detail the research on these four types of vision-based loosening detection methods.
Bolt missing detection in the existing research mainly focuses on the detection of the absence/presence state of bolts that secure the
rails to the sleepers. The images are acquired by a digital line scan camera installed under a train. The task aims to determine auto­
matically and accurately the absence/presence state of bolts based on the acquired images for traffic safety issues and preventing
dangerous situations. This process was widely regarded as a binary classification (see Table 11). Therefore, the popular solution was to
extract some pixel features from the images and feed them into a classifier. Based on the supervised learning, bolt missing detection
was conducted. The choices of pixel features and classifiers are the key to high detection accuracy. The associations of typical tech­
nologies (such as wavelets transform (WT) [204–206], Gabor wavelet transform [207] independent component analysis (ICA) [204],
principal component analysis (PCA) [205], Viola-Jones algorithm [208,209] and gradient orientation (GO)) [210] for image feature
extraction, and classical classification algorithms (including support vector machine (SVM), neural network (BP), multi-layer per­
ceptron (MLP), AdaBoost and Bayesian compression sensing) have been validated with excellent detection accuracy. The robustness
[211] and real time [212,213] were also significant issues and should be noted in the bolt missing detection due to the changes in
external environments and requirements of high-speed running of trains in practical applications. For example, Feng et al. [214]
applied a probabilistic topic model to simultaneously model diverse types of fasteners with different orientations and illumination
conditions, and Zhang et al. [215] proposed a novel structured light method based on motion image (SLMMI) for moving object in­
spection. Moreover, considering the trained classifiers being hard to generalize from a specific railway to the others, an unsupervised
learning method, fast template matching-based algorithm was proposed by Dou et al. [216] to locate the bolts with constrains of the
railway geometric structure. Then, a nearest neighbour classifier was used to determine whether a bolt was in position or not.
The second category involves the qualitative detection of the bolt tightening/loosening state. This implies that the output has only
two feasible states. The length information of the target bolt, particularly the length of the exposed bolt, is a key indicator that rep­
resents the tightening/loosening state. A few researchers applied Canny edge detection and Hough transform [217], or Viola–Jones
detection algorithm and Hough transform [218] to extract three main length features: the horizontal and vertical lengths of the bolt
head and the vertical length between the exposed thread and connect surface. These were then input into an SVM classifier to realise an
accurate classification of the fastening state. The rotational angle of bolt relative to its initial state is another important indicator for
representing the loosening state. Kong et al. [219] collected images of bolt heads in the initial state and in vibration conditions. The
two types of images were then aligned through image registration process. When bolt loosening occurs during service, the bolt would
rotate by an angle. The difference feature would be introduced into the registration error as a good indicator for bolt loosening
detection. The loose bolts can be visualised clearly after strengthening. In addition, a few scholars used deep learning to automatically
extract multi-level loosening features. These methods generally regard loosening and tightening states as two targets. The target
detection algorithm is used to realise loosening detection. For example, Zhang et al. [220] employed the fast region-based convolu­
tional neural network to directly output the target bolt and its tightening state.
The third category is the quantitative detection of loosening angle (see Table 12). The key process of loosening angle detection
comprises mainly of the following three steps: 1) Image preprocessing such as denoising or distortion correction is applied. 2) The
region of interest including the target bolt is framed. 3) The rotational angle is calculated. Park et al. [221–224] from Pukyong National
University were the first researchers to carry out the relevant research. In 2015–2016, they published several conference papers to
illustrate the concept of rotational angle detection. First, distortion correction is performed for bolt images involving different shooting
angles. Second, the boundary region of the inner hole in the bolt is identified, and its ellipse parameters are determined by the Canny
method and circular Hough transform. Finally, a boundary detection algorithm and line Hough transform are used to identify the
boundary of the hexagon nut, determine the line parameters, and calculate the loosening angle. However, the above process is tedious
for detecting the ROI. A few scholars, e.g. Huynh et al. [225] and Wang et al. [226] used deep learning algorithm to realise the
automatic detection of ROI. Pham et al. [227] also used RCNN, but his detection method used more geometric information, breaking
through the limit of small sample size. All the above methods used a Canny edge detector and line Hough transform to calculate the
loosening angle. In addition, Zhao et al. [228] proposed a new method to calculate loosening angle. In this method, the classes ‘bolt’
and ‘num’ in the images were identified and located. Then, the centre point coordinates of the two classes could be obtained. The angle
of the bolt loosening state was calculated according to the coordinates of the two points.
Few researchers have evaluated the degree of loosening by quantitatively detecting the length of exposed bolt. Xie et al. [229] and
Sun et al. [230] proposed a fast bolt-loosening detection method for the key components of a running train based on binocular vision.
In their research, the sub-pixel edges of the bolt cap and mounting surface in the localised regions were extracted and segmented,

Table 12
Summary of quantitative detection of loosening angle.
Preprocessing ROI Rotation angle Key reference

Denoise and correction of perspective distortion Circular Hough transform Sobel/Canny edge detector and Line Hough transform [221-224]
Perspective correction R-CNN Canny edge detection [225]
Line Hough transform
– Single Shot MultiBox Detector. A vector formed by two centre point coordinates [228]
Perspective transformation CNN Canny edge detector [226]
Line Hough transform
Perspective rectification R-CNN model Canny method [227]
Line Hough transform

23
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

respectively, in combination with the convolutional neural network (CNN). Based on stereo matching and the binocular vision model,
the 3D data of these edges were obtained to calculate the distance between the bolt cap and mounting surface. In addition, this method
could be used for multi-bolt loosening detection simultaneously.

6. Conclusions and prospects

In this paper, we have summarised various loosening detection methods for threaded fasteners. Loosening detection methods are
divided into sensor-based, percussion-based and vision-based methods, according to the differences of detection devices. The sensor-
based method contains explicit detection and implicit detection based on the relationship between measurement parameters by sensor
and preload. The former indicates that there exists a linear mathematical relationship between measurement parameters and preloads.
The typical sensors used for explicit detection include strain gauge, ultrasonic sensor, and FBG sensor. The corresponding measurement
parameters related to preload variation are strain, the TOF of a pulse-echo propagating along the bolt axis, and the wavelength of
reflected light, respectively. The key for explicit detection is the accurate experimental calibration. For the latter sensor-based loos­
ening detection, the relationship between measurement parameters and preloads is nonlinear and cannot be described by definite
mathematic formulae. For the implicit loosening detection, a PZT sensor is generally employed on the mechanical system with bolted
joints. The relationship between the response signal obtained by PZT sensor and preload variation is implicit and nonlinear. The key of
implicit loosening detection is the extraction of sensitive features from the response signal. Five typical implicit methods have been
developed and summarised: ultrasonic attenuation, time reversal, vibro-acoustic modulation, electromechanical impedance and
excitation analysis methods. Typical loosening features contain energy, amplitude, sideband, high-order harmonics and etc. The
percussion-based method exploits the variation in dynamic property caused by loose bolts, which will influence the audio signal
generated by knocking. An amalgamation of percussion-based method and artificial intelligence technology becomes popular
currently. Finally, the vision-based loosening detection analysed the captured images by exploiting the technologies of image pro­
cessing, and machine vision. The key is how to extract sensitive pixel features and select appropriate classifiers or directly apply deep
learning networks.
Although various loosening detection methods have been formulated and advanced in recent years with the development of signal
processing, sensing technology, and computer technology, certain issues need to be addressed further. First, the sensors used to directly
measure preloads require electric wires to supply power and transmit signal. Therefore, it is challenging for these sensors to measure
preloads of threaded fasteners in service in real time. This limits their practical applications. Most sensors can detect loosening only
before a machine works or after it is shut down. In addition, it is difficult to implement sensors to internal threaded fasteners for
loosening detection. Therefore, in the future, self-powered sensors with the capability of signal wireless transmission would display a
broad market potential. These should be investigated extensively. Second, for indirect loosening detection, most of the existing
methods can only detect several discrete loosening states of threaded fasteners. Thereby, a deviation exists between the detection result
and actual loosening state. In particular, for marginal decreases in preload (e.g. a preload loss of 5% of the initial preload), it is
significantly difficult for the existing indirect methods to realise accurate loosening detection. The saturation phenomenon can occur
straightforwardly. Therefore, it is essential and significant to achieve precise preload detection by establishing the quantitative
relationship between loosening features and preloads. The significant challenge is the extraction of representative and robust loosening
features, which can adapt to various scenarios. In addition, the existing indirect detection methods are applied mainly to single bolted
joints. A little research on multi-bolt loosening detection can only output several loosening states simultaneously. Quantitative multi-
bolt loosening detection by combining deep learning algorithm is another important research direction for the future.

CRediT authorship contribution statement

Jiayu Huang: Methodology, Writing – original draft. Jianhua Liu: Funding acquisition, Project administration, Supervision. Hao
Gong: Formal analysis, Writing – review & editing. Xinjian Deng: Data curation, Formal analysis.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Acknowledgements

This work was support by the National Natural Science Foundation of China [grant numbers 51935003 and 52105503] and China
Postdoctoral Science Foundation [2021M690396].

References

[1] J.H. Bickford, An Introduction to Design and Behavior of Bolted Joints, Marcel Dekker Inc, New York, 1995.
[2] D.P. Hess, Vibration-and shock-induced loosening, in: J.H. Bickford, S.A. Nassar (Eds.), Handbook of Bolts and Bolted Joints, Marcel Dekker Inc, New York,
1999, pp. 757–824.
[3] H. Holmes, Seeking the Perfect Locking Method for Threaded Fasteners, US Automation, 1998.
[4] H. Gong, J. Liu, H. Feng, Review on anti-loosening methods for threaded fasteners, Chin. J. Aeronaut. (2021).

24
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

[5] G.H. Junker, New criteria for self-loosening of fasteners under vibration, Soc Automot Eng. 78 (1969) 314–335.
[6] A. Yamamoto, S. Kasei, A Solution for self-loosening mechanism of threaded fasteners under transverse vibration, J. Jpn. Soc. Precis. Eng. 18 (3) (1984)
261–266.
[7] S. Izumi, T. Yokoyama, A. Iwasaki, Three-dimensional finite element analysis of tightening and loosening mechanism of threaded fastener, Eng. Fail Anal. 12
(4) (2005) 604–615.
[8] G. Dinger, C. Friedrich, Avoiding self-loosening failure of bolted joints with numerical assessment of local contact state, Eng. Fail Anal. 18 (8) (2011)
2188–2200.
[9] H. Gong, J. Liu, Some factors affecting the loosening failure of bolted joints under vibration using finite element analysis, P. I. Mech. Eng. C-J. Mec. 232 (12)
(2018) 3942–3953.
[10] H. Gong, J. Liu, X. Ding, Study on the mechanism of preload decrease of bolted joints subjected to transversal vibration loading, P. I. Mech. Eng. C-J. Mec. 233
(12) (2019) 2320–2329.
[11] H. Gong, J. Liu, X. Ding, Study on the critical loosening condition toward a new design guideline for bolted joints, P. I. Mech. Eng. C-J. Mec. 233 (9) (2019)
3302–3316.
[12] N.G. Pai, D.P. Hess, Experimental study of loosening of threaded fasteners due to dynamic shear loads, J. Sound Vib. 253 (3) (2002) 585–602.
[13] N.G. Pai, D.P. Hess, Three-dimensional finite element analysis of threaded fastener loosening due to dynamic shear load, Eng. Fail Anal. 9 (4) (2002) 383–402.
[14] H. Gong, J. Liu, X. Ding, Study on local slippage accumulation between thread surfaces and novel anti-loosening structure designs under transversal vibration,
Tribol. Int. 153 (2021), 106558.
[15] H. Gong, J. Liu, X. Ding, Thorough understanding on the mechanism of vibration-induced loosening of threaded fasteners based on modified Iwan model,
J. Sound Vib. 473 (2020), 115238.
[16] Q. Sun, Q. Lin, B. Yang, Mechanism and quantitative evaluation model of slip-induced loosening for bolted joints, Assembly Autom. 40 (4) (2020) 577–588.
[17] N. Sase, K. Nishioka, S. Koga, An anti-loosening screw-fastener innovation and its evaluation, J. Mater. Process. Tech. 77 (1–3) (1998) 209–215.
[18] N. Sase, H. Fujii, Optimizing study of slbs for higher anti-loosening performace, J. Mater. Process. Tech. 119 (1–3) (2001) 174–179.
[19] N.A. Noda, Y. Xiao, M. Kuhara, Optimum design of thin walled tube on the mechanical performance of super lock nut, J. Solid Mech. Mater. Eng. 2 (6) (2008)
780–791.
[20] T. Wang, G. Song, Review of bolted connection monitoring, Int. J. Distrib. Sens. N. 9 (12) (2013), 871213.
[21] N.S.M. Yadavar, G. Masoud, A review paper on looseness detection methods in bolted structures, Lat. Am. J. Solids Stru. 14 (12) (2017) 2153–2176.
[22] R. Miao, R. Shen, S. Zhang, S. Xue, A review of bolt tightening force measurement and loosening detection, Sensors. 20 (2020) 3165.
[23] Z. Zhang, J. Wang, X. Shi, K. Liu, Q. Zhang, Research and application of inserted intelligent bolt pre-tightening force measurement system implanted with
strain gauge, Qual. Reliab. 3 (2020) 31–34, in Chinese.
[24] J.M. Ren, An analysis and experiment investigation on pre-stress of spiral wound gasket, Lub. Eng. 3 (2003) 41–42, in Chinese.
[25] E. Vand, R. Hemmati, H. Oskouei, T.N. Chakherlou, An experimental method for measuring clamping force in bolted connections and effect of bolt threads
lubrication on its value, World Acad. Sci. Eng. Technol. 22 (2008) 457–460.
[26] T. Wang, B. Tan, G. Lu, Bolt pretightening force measurement based on strain distribution of bolt head surface, J. Aerospace Eng. 33 (4) (2020) 04020034.
[27] K.Y. Jhang, H.H. Quan, J. Ha, Estimation of clamping force in high-tension bolts through ultrasonic velocity measurement, Ultrasonics 1 (8) (2006)
1339–1342.
[28] Q. Sun, B. Yuan, X. Mu, Bolt preload measurement based on the acoustoelastic effect using smart piezoelectric bolt, Smart Mater. Struct. 28 (5) (2019),
055005.
[29] Y. Liu, E. Liu, Y. Chen, Measurement of fastening force using dry-coupled ultrasonic waves, Ultrasonics 108 (2020), 106178.
[30] H. Yasui, K. Kawashima, Acoustoelastic measurement of bolt axial load with velocity ratio method, in 2000 Rome, Italy.
[31] S. Chaki, G. Corneloup, I. Lillamand, Combination of longitudinal and transverse ultrasonic waves for in situ control of the tightening of bolts, J. Press. Vess.-T.
ASME. 129 (3) (2007) 94–110.
[32] N. Kim, M. Hong, Measurement of axial stress using mode-converted ultrasound, NDT&E. Int. 42 (3) (2009) 164–169.
[33] X. Ding, X. Wu, Y. Wang, Bolt axial stress measurement based on a mode-converted ultrasound method using an electromagnetic acoustic transducer,
Ultrasonics 54 (3) (2014) 914–920.
[34] Q. Pan, R. Pan, M. Chang, A shape factor based ultrasonic measurement method for determination of bolt preload, NDT&E. Int. 111 (2020), 102210.
[35] P. Chen, X. He, W. Song, Parameter recognition of mode-converted wave in single-source ultrasound using Gabor transform for bolt axial stress evaluation,
J. Sensors. 111 (5) (2020) 1–11.
[36] A.D. Hay, Bolt, stud or fastener having an embedded fiber optic Bragg grating sensor for sensing tensioning strain, USA. Patent. (1999).
[37] Q. Wu, Y. Okabe, F. Yu, Ultrasonic structural health monitoring using fiber Bragg grating, Sensors 18 (10) (2018) 3395.
[38] A. Othonos, Fiber Bragg gratings, Rev. Sci. Instrum. 68 (12) (1997) 4309–4341.
[39] A. Khomenko, E.G. Koricho, M. Haq, G.L. Cloud, Bolt tension monitoring with reusable fiber Bragg-grating sensors, J. Strain Anal. Eng. Design. 51 (2) (2016)
101–108.
[40] R. Liang, T. Feng, M. Ho, A smart “shear sensing” bolt based on FBG sensors, Measurement 122 (2018) 240–246.
[41] C. Duan, H. Zhang, Z. Li, Y. Tian, Q. Chai, J. Yang, Z. Lv, FBG smart bolts and their application in power grids, IEEE T. Instrum. Meas. 69 (5) (2019)
2515–2521.
[42] D. Chen, L. Huo, H. Li, A fiber Bragg grating (FBG)-enabled smart washer for bolt pre-load measurement: design, analysis, calibration, and experimental
validation, Sensors 18 (8) (2018) 2586.
[43] M. Yeager, A. Whitaker, M. Todd, A method for monitoring bolt torque in a composite connection using an embedded fiber Bragg grating sensor, J. Intel. Mat.
Syst. Str. 29 (3) (2018) 335–344.
[44] Y.K. Huang, Y. Tu, S.T. Tu, Monitoring of bolt clamping force at high temperatures using metal-packaged regenerated fiber Bragg grating strain sensors, AOPC
2017: Fiber Optic Sensing and Optical Communications Vol. 10464 (2017).
[45] Z. Wang, M. Liu, Z. Zhu, Y. Qu, Q. Wei, Z. Zhou, F. Yang, Clamp looseness detection using modal strain estimated from FBG based operational modal analysis,
Measurement 137 (2019) 82–97.
[46] Y.K. Huang, Y. Tu, S.T. Tu, Monitoring of bolt clamping force at high temperatures using metal-packaged regenerated fiber Bragg grating strain sensors, in:
AOPC 2017: Fiber Optic Sensing and Optical Communications, Vol. 10464, International Society for Optics and Photonics, 2017, p. 1046412.
[47] Z. Wang, M. Liu, Z. Zhou, Y. Qu, Q. Wei, L. Hong, F. Yang, An FBG based smart clamp for the detection of incipient clamp looseness in industrial piping system,
Measurement 140 (2019) 416–426.
[48] P. Wang, N. Zhang, J. Kan, Z. Xie, Q. Wei, W. Yao, Fiber Bragg grating monitoring of full-bolt axial force of the bolt in the deep strong mining roadway, Sensors
20 (15) (2020) 4242.
[49] S.C.M. Ho, W. Li, B. Wang, G. Song, A load measuring anchor plate for rock bolt using fiber optic sensor, Smart Mater. Struct. 26 (5) (2017), 057003.
[50] T. Wang, J. Chang, P. Gong, W. Shi, N. Li, S. Cheng, The experimental instrumented bolt with fiber Bragg grating force sensors, Arch. Min. Sci. 65 (1) (2020).
[51] D.L. Weidman, G.H. Beall, K.C. Chyung, G.L. Francis, R.A. Modavis, R.M. Morena, A novel negative expansion substrate material for athermalizing fiber Bragg
gratings, in: Proceedings of European Conference on Optical Communication, Vol. 1, IEEE, 1996, pp. 61–64.
[52] J. Shao, T. Wang, H. Yin, D. Yang, Y. Li, Bolt looseness detection based on piezoelectric impedance frequency shift, Appl. Sci. 6 (10) (2016) 298.
[53] C. Hei, M. Luo, P. Gong, G. Song, Quantitative evaluation of bolt connection using a single piezoceramic transducer and ultrasonic coda wave energy with the
consideration of the piezoceramic aging effect, Smart Mater. Struct. 29 (2) (2020), 027001.
[54] F. Wang, Z. Chen, G. Song, Monitoring of multi-bolt connection looseness using entropy-based active sensing and genetic algorithm-based least square support
vector machine, Mech. Syst. Signal. Pr. 136 (2020), 106507.

25
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

[55] N. Li, F. Wang, G. Song, Monitoring of bolt looseness using piezoelectric transducers: three-dimensional numerical modeling with experimental verification,
J. Intel. Mat. Syst. Str. 31 (6) (2020) 911–918.
[56] J. Jiang, S.C.M. Ho, T. Tippitt, Feasibility study of a touch-enabled active sensing approach to inspecting subsea bolted connections using piezoceramic
transducers, Smart Mater. Struct. 29 (8) (2020), 085038.
[57] R. Kedra, M. Rucka, Preload monitoring in a bolted joint using Lamb wave energy, B. Pol. Acad. Sci-Tech. 67 (6) (2019) 1161–1169.
[58] F. Wang, M. Ho, G. Song, Monitoring of early looseness of multi-bolt connection: a new entropy-based active sensing method without saturation, Smart Mater.
Struct. 28 (10) (2019) 10LT01.
[59] Y. Zhang, D. Li, X. Zheng, Detection and location of bolt group looseness using ultrasonic guided wave, Smart Struct. Syst. 24 (3) (2019) 293–301.
[60] X. Zhang, Q. Yan, J. Yang, J. Zhao, Y. Shen, An assembly tightness detection method for bolt-jointed rotor with wavelet energy entropy, Measurement 136
(2019) 212–224.
[61] T. Jiang, Q. Wu, L. Wang, L. Huo, G. Song, Monitoring of bolt looseness-induced damage in steel truss arch structure using piezoceramic transducers, IEEE
Sens. J. 18 (16) (2018) 6677–6685.
[62] S. Yan, W. Liu, G. Song, Connection looseness detection of steel grid structures using piezoceramic transducers, Int. J. Distrib. Sens. N. 14 (2) (2018),
1550147718759234.
[63] B. Yang, F.Z. Xuan, Y. Xiang, Lamb wave-based structural health monitoring on composite bolted joints under tensile load, Materials. 10 (6) (2017).
[64] R. Kędra, M. Rucka, Damage detection in a bolted lap joint using guided waves, Procedia Eng. 199 (2017) 2114–2119.
[65] F. Wang, L.S. Huo, G. Song, A piezoelectric active sensing method for quantitative monitoring of bolt loosening using energy dissipation caused by tangential
damping based on the fractal contact theory, Smart Mater. Struct. 27 (2017), 015023.
[66] W. Xie, S. Meng, J. Han, A new diagnostic method of bolt loosening detection for thermal protection systems, Proc. SPIE – Int. Soc. Optical Eng. 7493 (2009)
111.
[67] R. Kędra, M. Rucka, Research on assessment of bolted joint state using elastic wave propagation, J. Physics Conference. 628 (2015), 012025.
[68] R.A. Esmaeel, F. Taheri, Application of a Simple and Cost-effective Method for Detection of Bolt Self-Loosening in Single Lap Joints, Nondestruct Test, Eva,
2013.
[69] J. Yang, F.K. Chang, Detection of bolt loosening in C-C composite thermal protection panels: II. Experimental verification, Smart Mater. Struct. 15 (2) (2006)
591–599.
[70] Y. Zhou, C. Yuan, X. Sun, Monitoring the looseness of a bolt through laser ultrasonic, Smart Mater. Struct. 29 (11) (2020) 1–11.
[71] D. Quoc, J.W. Kim, K.D. Tola, Artificial intelligence-based bolt loosening diagnosis using deep learning algorithms for laser ultrasonic wave propagation data,
Sensors 20 (18) (2020) 5329.
[72] T. Kundu, C. Haynes, M. Yeager, Monitoring bolt torque levels through signal processing of full-field ultrasonic data. Spie Smart Structures & Materials
Nondestructive Evaluation & Health Monitoring, International Society for Optics and Photonics, 2014.
[73] K.L. Johnson, K.L. Johnson, Contact Mechanics, Cambridge University Press, 1985.
[74] S. Liu, Y. Li, T. Wang, A piezoelectric active sensing method for detection of bolt load loss, Sensor Rev. 34 (4) (2014) 337–342.
[75] T. Wang, G. Song, Z. Wang, Proof-of-concept study of monitoring bolt connection status using a piezoelectric based active sensing method, Smart Mater. Struct.
22 (8) (2013), 087001.
[76] M. Fink, D. Cassereau, A. Derode, Time-reversed acoustics, Rep. Prog. Phys. 63 (12) (2000) 1933.
[77] M. Fink, Time reversal of ultrasonic fields. I. Basic principles, IEEE Trans. Ultrason. Ferr. 39 (5) (1992) 555–566.
[78] R.K. Ing, M. Fink, Time-reversed Lamb waves, IEEE T. Ultrason. Ferr. 45 (4) (1998) 1032–1043.
[79] M. Fink, G. Montaldo, M. Tanter, Time-reversal acoustics in biomedical engineering, Annu. Rev. Biomed. Eng. 5 (1) (2003) 465.
[80] G. Du, Q. Kong, F. Wu, An experimental feasibility study of pipeline corrosion pit detection using a piezoceramic time reversal mirror, Smart Mater. Struct. 25
(3) (2016), 037002.
[81] L. Huo, D. Chen, Q. Kong, Smart washer – a piezoceramic-based transducer to monitor looseness of bolted connection, Smart Mater. Struct. 26 (2) (2017),
025033.
[82] T. Wang, Health monitoring of bolted joints using the time reversal method and piezoelectric transducers, Smart Mater. Struct. 25 (2) (2016), 025010.
[83] Z. Zhao, P. Chen, E. Zhang, Health monitoring of bolt looseness in timber structures using PZT-enabled time-reversal method, J. Sensors. 2019 (2019) 1–8.
[84] S.M. Parvasi, S.C.M. Ho, Q. Kong, Real time bolt preload monitoring using piezoceramic transducers and time reversal technique—a numerical study with
experimental verification, Smart Mater. Struct. 25 (8) (2016), 085015.
[85] L. Huo, F. Wang, H. Li, A fractal contact theory based model for bolted connection looseness monitoring using piezoceramic transducers, Smart Mater. Struct.
26 (10) (2017), 104010.
[86] J. Ruan, Z. Zhang, T. Wang, An anti-noise real-time cross-correlation method for bolted joint monitoring using piezoceramic transducers, Smart Struct. Syst. 16
(2) (2015) 281–294.
[87] Y. Liang, Q. Feng, D. Li, Loosening monitoring of the threaded pipe connection using time reversal technique and piezoceramic transducers, Sensors. 18 (7)
(2018) 2280.
[88] X. Hong, G. Song, J. Ruan, Active monitoring of pipeline tapered thread connection based on time reversal using piezoceramic transducers, Smart Struct. Syst.
18 (4) (2016) 643–662.
[89] L. Huo, B. Wang, D. Chen, Monitoring of pre-load on rock bolt using piezoceramic-transducer enabled time reversal method, Sensors. 17 (11) (2017) 2467.
[90] J. Xu, C. Wang, H. Li, Health monitoring of bolted spherical joint connection based on active sensing technique using piezoceramic transducers, Sensors 18 (6)
(2018) 1727.
[91] Y. Liang, D. Li, Q. Kong, Load monitoring of the pin-connected structure using time reversal technique and piezoceramic transducers—a feasibility study, IEEE
Sens. J. 16 (22) (2016) 7958–7966.
[92] L. Zhang, C. Wang, L. Huo, Health monitoring of cuplok scaffold joint connection using piezoceramic transducers and time reversal method, Smart Mater.
Struct. 25 (3) (2016), 035010.
[93] X. Zhang, L. Zhang, L. Liu, Prestress monitoring of a steel strand in an anchorage connection using piezoceramic transducers and time reversal method, Sensors
18 (11) (2018) 4018.
[94] H. Yin, T. Wang, D. Yang, A Smart Washer for Bolt Looseness Monitoring Based on Piezoelectric Active Sensing Method, Appl. Sci. 6 (11) (2016) 320.
[95] J. Cai, L. Shi, S. Yuan, High spatial resolution imaging for structural health monitoring based on virtual time reversal, Smart Mater. Struct. 20 (5) (2011)
55018–55028.
[96] G. Wu, C. Xu, F. Du, W. Zhu, A modified time reversal method for guided wave detection of bolt loosening in simulated thermal protection system panels,
Complexity 2018 (2018) 1–12.
[97] F. Du, C. Xu, G. Wu, Preload monitoring of bolted L-shaped lap joints using virtual time reversal method, Sensors. 18 (6) (2018) 1928.
[98] F. Du, C. Xu, J. Zhang, A bolt preload monitoring method based on the refocusing capability of virtual time reversal, Struct. Control Health. 26 (8) (2019)
e2370.1–e2370.16.
[99] C. Xu, G. Wu, F. Du, A modified time reversal method for guided wave based bolt loosening monitoring in a lap joint, J. Nondestruct. Eval. 38 (4) (2019) 1–13.
[100] U. Andreaus, P. Casini, F. Vestroni, Non-linear dynamics of a cracked cantilever beam under harmonic excitation, Int. J. Non-linear. Mech. 42 (3) (2007)
566–575.
[101] Z. Zhang, M. Liu, Z. Su, Quantitative evaluation of residual torque of a loose bolt based on wave energy dissipation and vibro-acoustic modulation: a
comparative study, J. Sound Vib. 383 (2016) 156–170.
[102] J.J. Meyer, D.E. Adams, Theoretical and experimental evidence for using impact modulation to assess bolted joints, Nonlinear Dynam. 81 (1–2) (2015)
103–117.

26
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

[103] J.J. Meyer, D.E. Adams, Using impact modulation to quantify nonlinearities associated with bolt loosening with applications to satellite structures, Mech. Syst.
Signal. Pr. 116 (2019) 787–795.
[104] S. M. Y. Nikravesh, M. Goudarzi, Experimental and numerical looseness detection and assessment in flanged joints using vibro-acoustic modulation method,
Mech. Based Des. Struct. (2020) 1-17. Online.
[105] W. Zhou, Y. Shen, L. Xiao, W. Qu, Application of nonlinear-modulation technique for the detection of bolt loosening in frame structure, J. Test. Eval. 44 (2)
(2016) 967–975.
[106] Z. Zhang, M. Liu, Z. Su, Y. Xiao, Quantitative evaluation of residual torque of a loose bolt based on wave energy dissipation and vibro-acoustic modulation: a
comparative study, J. Sound Vib. 383 (2016) 156–170.
[107] Z. Zhang, H. Xu, Y. Liao, Vibro-acoustic modulation (VAM)-inspired structural integrity monitoring and its applications to bolted composite joints, Compos.
Struct. 176 (2017) 505–515.
[108] Z. Zhang, M. Liu, Y. Liao, Contact acoustic nonlinearity (CAN)-based continuous monitoring of bolt loosening: Hybrid use of high-order harmonics and spectral
sidebands, Mech. Syst. Signal. Pr. 103 (2018) 280–294.
[109] G.P.M. Fierro, M. Meo, IWSHM 2017: structural health monitoring of the loosening in a multi-bolt structure using linear and modulated nonlinear ultrasound
acoustic moments approach, Struct Health Monit. 17 (6) (2018) 1349–1364.
[110] N. Zhao, L. Huo, G. Song, A nonlinear ultrasonic method for real-time bolt looseness monitoring using PZT transducer–enabled vibro-acoustic modulation,
J. Intel. Mat. Syst. Str. 31 (3) (2020) 364–376.
[111] F. Wang, G. Song, Bolt early looseness monitoring using modified vibro-acoustic modulation by time-reversal, Mech. Syst. Signal. Pr. 130 (2019) 349–360.
[112] S.M.Y. Nikravesh, M. Goudarzi, Looseness detection and assessment of flange type joints using vibroacoustic modulation method, Russ. J. Nondestruct+. 56 (6)
(2020) 490–505.
[113] F. Wang, G. Song, Monitoring of multi-bolt connection looseness using a novel vibro-acoustic method, Nonlinear Dynam. 1–12 (2020).
[114] F. Amerini, M. Meo, Structural health monitoring of bolted joints using linear and nonlinear acoustic/ultrasound methods, Struct Health Monit. 10 (6) (2010)
659–672.
[115] H. Sohn, C.R. Farrar, D.J. Inman, Overview of piezoelectric impedance-based health monitoring and path forward, Shock Vib. 35 (6) (2003) 451–463.
[116] V. Giurgiutiu, A.N. Zagrai, J.J. Bao, J.M. Redmond, D. Roach, K. Rackow, Active sensors for health monitoring of aging aerospace structures, Inter. J.
Comadem. 6 (1) (2003) 3–21.
[117] G. Mook, J. Pohl, M. Michel, Non-destructive characterization of smart CFRP structures, Smart Mater. Struct. 12 (2003) 997–1004.
[118] K.K. Tseng, M.L. Tinker, J.O. Lassiter, D.M. Peairs, Temperature dependency of impedance-based nondestructive testing, Exp. Techniques. 27 (5) (2003)
33–36.
[119] S. Bhalla, C.K. Soh, Structural impedance-based damage diagnosis by piezo-transducers, Earthq. Eng. Struct. D. 32 (12) (2003) 1897–1916.
[120] S. Bhalla, C.K. Soh, High frequency piezoelectric signatures for diagnosis of seismic/blast induced structural damages, NDT&E Int. 37 (1) (2004) 23–33.
[121] C. Bois, C. Hochard, Monitoring of laminated composites delamination based on electro-mechanical impedance measurement, J. Intel. Mat. Syst. Str. 15 (1)
(2004) 59–67.
[122] D. Peairs, G. Park, D.J. Inman, Improving accessibility of the impedance-based structural health monitoring method, J. Intel. Mat. Syst. Str. 15 (2) (2004)
129–140.
[123] M. Abe, Y. Fujino, T. Miyashita, Quantitative health monitoring of bolted joints using a piezoceramic actuator-sensor, Smart Mater. Struct. 13 (1) (2004)
20–29.
[124] F.P. Sun, Z. Chaudhry, C. Liang, Truss structure integrity identification using PZT sensor-actuator, J. Intel. Mat. Syst. Str. 6 (1) (1995) 134–139.
[125] C. Liang, F. Sun, C. Rogers, Coupled electro-mechanical analysis of adaptive material systems—determination of the actuator power consumption and system
energy transfer, J. Intell. Mater. Syst. Struct. 5 (1994) 12–20.
[126] G. Park, H.H. Cudney, D.J. Inman, Feasibility of using impedance-based damage assessment for pipeline structures, Earthq. Eng. Struct. D. 30 (10) (2010)
1463–1474.
[127] A. Martowicz, A. Sendecki, M. Salamon, Application of electromechanical impedance-based SHM for damage detection in bolted pipeline connection,
Nondestruct. Test. Eva. 31 (1) (2016) 1–28.
[128] L. Huo, D. Chen, Y. Liang, Impedance based bolt pre-load monitoring using piezoceramic smart washer, Smart Mater. Struct. 26 (5) (2017), 057004.
[129] W. Bo, L. Huo, D. Chen, Impedance-based pre-stress monitoring of rock bolts using a Piezoceramic-based smart washer—a feasibility study, Sensors 17 (2)
(2017) 250.
[130] D. Chen, L. Huo, G. Song, EMI based multi-bolt looseness detection using series/parallel multi-sensing technique, Smart Struct. Syst. 25 (2020) 423–432.
[131] T.C. Huynh, J.T. Kim, Quantification of temperature effect on impedance monitoring via PZT interface for prestressed tendon anchorage, Smart Mater. Struct.
26 (12) (2017) 1–19.
[132] K. D. Nguyen, S. Y. Lee, P. Y. Lee, Wireless SHM for bolted connections via multiple PZT-interfaces and Imote2-platformed impedance sensor node, in:
Proceedings of the 6th International Workshop on Advanced Smart Materials and Smart Structures Technology, 2011, Dalian, China.
[133] T.C. Huynh, T.D. Nguyen, D.D. Ho, Sensor fault diagnosis for impedance monitoring using a piezoelectric-based smart interface technique, Sensors 20 (2)
(2020) 510.
[134] T.C. Huynh, N.L. Dang, J.T. Kim, Preload monitoring in bolted connection using piezoelectric-based smart interface, Sensors 18 (9) (2018) 2276.
[135] T.C. Huynh, D.D. Ho, N.L. Dang, Sensitivity of Piezoelectric-based smart interfaces to structural damage in bolted connections, Sensors. 19 (17) (2019).
[136] C. Wang, N. Wang, S.C. Ho, Design of a novel wearable sensor device for real-time bolted joints health monitoring, IEEE Internet Things. 5 (6) (2018)
5307–5316.
[137] J.Y. Ryu, T.C. Huynh, J.T. Kim, Tension force estimation in axially loaded members using wearable piezoelectric interface technique, Sensors. 19 (1) (2018)
47.
[138] Y.K. An, H. Sohn, Integrated impedance and guided wave based damage detection, Mech. Syst. Signal Pr. 28 (2012) 50–62.
[139] S.K. Samantaray, S.K. Mittal, P. Mahapatra, S. Kumar, An impedance-based structural health monitoring approach for looseness identification in bolted joint
structure, J. Civ. Struct. Health. 8 (5) (2018) 809–822.
[140] F. Wang, S.C.M. Ho, L. Huo, G. Song, A novel fractal contact-electromechanical impedance model for quantitative monitoring of bolted joint looseness, IEEE
Access 6 (2018) 40212–40220.
[141] X. Jiang, X. Zhang, Y. Zhang, Evaluation of characterization indexes and minor looseness identification of flange bolt under noise influence, IEEE Access 8
(2020) 157691–157702.
[142] J. Min, S. Park, C.B. Yun, Impedance-based structural health monitoring incorporating neural network technique for identification of damage type and
severity, Eng. Struct. 39 (2012) 210–220.
[143] I. Pavelko, V. Pavelko, S. Kuznetsov, Bolt-joint structural health monitoring by the method of electromechanical impedance, Aircr. Eng. Aerosp. Tec. 86 (3)
(2014) 207–214.
[144] S. Fan, W. Li, Q. Kong, Q. Feng, G. Song, Monitoring of pin connection loosening using eletromechanical impedance: Numerical simulation with experimental
verification, J. Intell. Mater. Syst. Struct. 29 (9) (2018) 1964–1973.
[145] J. Xu, J. Dong, H. Li, C. Zhang, S.C. Ho, Looseness monitoring of bolted spherical joint connection using electro-mechanical impedance technique and BP
neural networks, Sensors. 19 (8) (2019) 1906.
[146] H.J. Lim, M.K. Kim, H. Sohn, Impedance based damage detection under varying temperature and loading conditions, NDT & E Int. 44 (8) (2011) 740–750.
[147] Y. Liang, Q. Feng, D. Li, Loosening monitoring of a threaded pipe connection using the electro-mechanical impedance technique—Experimental and numerical
studies, Sensors 18 (11) (2018) 3699.
[148] M. Chen, B. Xu, Bolted joint looseness damage detection using electromechanical impedance measurements by PZT sensors. Third International Conference on
Smart Materials and Nanotechnology in Engineering, International Society for Optics and Photonics, 2012.

27
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

[149] M. Abe, T. Miyashita, S. Ritdumrongkul, Quantitative health monitoring of bolted joints using piezoceramics actuator-sensor, J. Appl. Mech. 6 (2003)
1033–1042.
[150] S. Park, H.H. Shin, C.B. Yun, Wireless impedance sensor nodes for functions of structural damage identification and sensor self-diagnosis, Smart Mater. Struct.
18 (5) (2009), 055001.
[151] K. D. Nguyen, S. Y. Lee, P. Y. Lee, J. T. Kim Wireless SHM for bolted connections via multiple PZT-interfaces and Imote2-platformed impedance sensor node, in
Proceedings of the 6th International Workshop on Advanced Smart Materials and Smart Structures Technology, 2011, Dalian, China.
[152] S. Park, J.J. Lee, C.B. Yun, D.J. Inman, Electro-mechanical impedance-based wireless structural health monitoring using PCA-data compression and k-means
clustering algorithms, J. Intell. Mater. Syst. Struct. 19 (2008) 509–520.
[153] S.M. Parvasi, S.C.M. Ho, Q. Kong, R. Mousavi, G. Song, Real time bolt preload monitoring using piezoceramic transducers and time reversal technique—a
numerical study with experimental verification, Smart Mater. Struct. 25 (2016), 085015.
[154] S. Na, H. Lee, Resonant frequency range utilized electromechanical impedance method for damage detection performance enhancement on composite
structures, Compos. Struct. 94 (2012) 2383–2389.
[155] S.G. Taylor, K.M. Farinholt, G. Park, M.D. Todd, C.R. Farrar, Multi-scale wireless sensor node for health monitoring of civil infrastructure and mechanical
systems, Smart Struct. Syst. 6 (2010) 661–673.
[156] J. Min, S. Park, C.B. Yun, Impedance-based structural health monitoring using neural networks for autonomous frequency range selection, Smart Mater. Struct.
19 (12) (2010).
[157] G. Park, H. Sohn, C.R. Farrar, D.J. Inman, Overview of piezoelectric impedance based health monitoring and path forward, Shock Vib. 35 (6) (2003) 451–463.
[158] A.C. Rutherford, G. Park, C.R. Farrar, Non-linear feature identifications based on self-sensing impedance measurements for structural health assessment, Mech.
Syst. Signal. Pr. 21 (1) (2007) 322–333.
[159] J.H. Park, T.C. Huynh, J.T. Kim, Temperature effect on wireless impedance monitoring in tendon anchorage of prestressed concrete girder, Smart Struct. Syst.
15 (4) (2015) 1159–1175.
[160] T.C. Huynh, J.T. Kim, Compensation of temperature effect on impedance responses of PZT interface for prestress-loss monitoring in PSC girders, Smart Struct.
Syst. 17 (6) (2016) 881–901.
[161] T.C. Huynh, T.C. Nguyen, S.H. Choi, Impedance monitoring at tendon-anchorage via mountable PZT interface and temperature-effect compensation.
Proceedings of SPIE-The International Society for Optical Engineering, 2016.
[162] K.Y. Koo, S. Park, J.J. Lee, C.B. Yun, Automated impedance-based structural health monitoring incorporating effective frequency shift for compensating
temperature effects, J. Intell. Mater. Syst. Struct. 20 (2009) 367–377.
[163] W.S. Na, Bolt loosening detection using impedance based non-destructive method and probabilistic neural network technique with minimal training data, Eng.
Struct. 226 (2021), 111228.
[164] L. Zhou, S.X. Chen, Y.Q. Ni, A.W.H. Choy, EMI-GCN: a hybrid model for real-time monitoring of multiple bolt looseness using electromechanical impedance
and graph convolutional networks, Smart Mater. Struct. 30 (3) (2021), 035032.
[165] H. Hasni, P. Jiao, A.H. Alavi, Structural health monitoring of steel frames using a network of self-powered strain and acceleration sensors: A numerical study,
Automat. Constr. 85 (2018) 344–357.
[166] Y. Li, R. Jiang, J. Tapia, S. Wang, W. Sun, Structural damage identification based on short-time temporal coherence using free-vibration response signals,
Measurement 151 (2020), 107209.
[167] H. Wang, X. Jing, Vibration signal–based fault diagnosis in complex structures: a beam-like structure approach, Struct. Health. Monit. 17 (3) (2018) 472–493.
[168] S.E. Olson, M.P. DeSimio, M.M. Derriso, Fastener damage estimation in a square aluminum plate, Struct. Health. Monit. 5 (2) (2006) 173–183.
[169] D. Liang, S. Yuan, Decision fusion system for bolted joint monitoring, Shock Vib. 2015 (2015), 592043.
[170] Y. Zhang, Y. Miyamori, S. Mikami, T. Saito, Vibration-based structural state identification by a 1-dimensional convolutional neural network, Comput-Aided
Civ. Inf. 34 (9) (2019) 822–839.
[171] Q. Li, X. Jing, A novel second-order output spectrum based local tuning method for locating bolt-loosening faults, Mech. Syst. Signal. Pr. 147 (2021), 107104.
[172] Q. Li, X. Jing, A systematic second-order output spectrum based method for fault diagnosis with a local tuning approach, J. Sound Vib. 475 (2020), 115283.
[173] Q. Li, X. Jing, A second-order output spectrum approach for fault detection of bolt loosening in a satellite-like structure with a sensor chain, Nonlinear Dynam.
89 (1) (2017) 587–606.
[174] J. Pal, S. Banerjee, S. Chikermane, Estimation of fixity factors of bolted joints in a steel frame structure using a vibration-based health monitoring technique,
Int. J. Steel Struct. 17 (2) (2017) 593–607.
[175] J.M. Nichols, S.T. Trickey, M. Seaver, Using ambient vibrations to detect loosening of a composite-to-metal bolted joint in the presence of strong temperature
fluctuations, J. Vib. Acoust. 129 (6) (2007) 710–717.
[176] M. Okugawa, Bolt loosening detection method by using smart washer adopted 4SID, in: 45th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics &
Materials Conference, 2004.
[177] T. Nakahara, M. Yamamoto, Y. Ohya, Bolt loosening detection using vibration characteristics of thin plate with piezoelectric elements. Smart Structures and
Materials 2004: Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems, International Society for Optics and Photonics, 2004.
[178] F. Huda, I. Kajiwara, N. Hosoya, S. Kawamura, Bolt loosening analysis and diagnosis by non-contact laser excitation vibration tests, Mech. Syst. Signal. Pr. 40
(2) (2013) 589–604.
[179] N. Hosoya, T. Hosokawa, I. Kajiwara, S. Hashimura, F. Huda, Evaluation of the clamping force of bolted joints using local mode characteristics of a bolt head,
J. Nondestruct. Eval. 37 (4) (2018) 1–8.
[180] S.M. Sah, J.J. Thomsen, M. Brøns, A. Fidlin, D. Tcherniak, Estimating bolt tightness using transverse natural frequencies, J. Sound Vib. 431 (2018) 137–149.
[181] Y.J. Hu, W.G. Guo, C. Jiang, Y.L. Zhou, W. Zhu, Looseness localization for bolted joints using Bayesian operational modal analysis and modal strain energy,
Adv. Mech. Eng. 10 (11) (2018), 1687814018808698.
[182] N. Hosoya, T. Niikura, S. Hashimura, I. Kajiwara, F. Giorgio-Serchi, Axial force measurement of the bolt/nut assemblies based on the bending mode shape
frequency of the protruding thread part using ultrasonic modal analysis, Measurement 162 (2020), 107914.
[183] V. Caccese, R. Mewer, S.S. Vel, Detection of bolt load loss using frequency domain techniques, in: Proceeding of the 15th international conference on adaptive
structures and technologies, 2004, Bar Harbor, ME, USA.
[184] K. Anginthaya, S.A.R. Kuchibhatla, K.V. Gangadharan, A. Kishan, A comparative study on the effectiveness of system parameters in monitoring pre-load loss in
bolted joints, J. Braz. Soc. Mech. Sci. 40 (8) (2018) 1–15.
[185] R. Chen, S. Chen, L. Yang, J. Wang, X. Xu, T. Luo, Looseness diagnosis method for connecting bolt of fan foundation based on sensitive mixed-domain features
of excitation-response and manifold learning, Neurocomputing. 219 (2017) 376–388.
[186] J. Wei, C. Liu, T. Ren, H. Liu, W. Zhou, Online condition monitoring of a rail fastening system on high-speed railways based on wavelet packet analysis,
Sensors. 17 (2) (2017) 318.
[187] R.A. Esmaeel, J. Briand, F. Taheri, Computational simulation and experimental verification of a new vibration-based structural health monitoring approach
using piezoelectric sensors, Struct. Health. Monit. 11 (2) (2012) 237–250.
[188] P. Razi, R.A. Esmaeel, F. Taheri, Improvement of a vibration-based damage detection approach for health monitoring of bolted flange joints in pipelines, Struct.
Health. Monit. 12 (3) (2013) 207–224.
[189] R.A. Esmaeel, F. Taheri, Application of a simple and cost-effective method for detection of bolt self-loosening in single lap joints, Nondestructive Test. Eva. 28
(3) (2013) 208–225.
[190] C. Xu, C.C. Huang, W.D. Zhu, Bolt loosening detection in a jointed beam using empirical mode decomposition–based nonlinear system identification method,
Int. J. Distrib. Sens. N. 15 (9) (2019).
[191] D. Wang, Identification for joint interfaces with correlation analysis of instantaneous dynamics, Arch. Appl. Mech. 90 (1) (2020) 187–198.
[192] H.D. Wu, M. Siegel, Correlation of accelerometer and microphone data in the “coin tap test”, IEEE T. Instrum. Meas. 49 (2000) 493–497.

28
J. Huang et al. Mechanical Systems and Signal Processing 168 (2022) 108652

[193] S. Baglio, N. Savalli, “Fuzzy tap-testing” sensors for material health-state characterization, IEEE T. Instrum. Meas. 55 (2006) 761–770.
[194] S. Yella, N.K. Gupta, M.S. Dougherty, Comparison of pattern recognition techniques for the classification of impact acoustic emissions, Transport. Res. C-Emer.
15 (2007) 345–360.
[195] F. Tong, X.M. Xu, B.L. Luk, K.P. Liu, Evaluation of tile-wall bonding integrity based on impact acoustics and support vector machine, Sensor Actuat. A-Phys.
144 (2008) 97–104.
[196] F. Wang, G. Song, Y.L. Mo, Shear loading detection of through bolts in bridge structures using a percussion-based one-dimensional memory-augmented
convolutional neural network, Comput.-Aided Civ. Inf. 36 (3) (2021) 289–301.
[197] Y. Zhang, X. Zhao, X. Sun, Bolt loosening detection based on audio classification, Adv. Struct. Eng. 22 (13) (2019) 2882–2891.
[198] Y. Zhang, X. Zhao, W. Su, Audio-based bolt-loosening detection technique of bolt joint, in: Nondestructive Characterization and Monitoring of Advanced
Materials, Aerospace, Civil Infrastructure, and Transportation XII, International Society for Optics and Photonics, 2018.
[199] Q. Kong, J. Zhu, S.C.M. Ho, Tapping and listening: a new approach to bolt looseness monitoring, Smart Mater. Struct. 27 (7) (2018) 07LT02.
[200] R. Yuan, Y. Lv, Q. Kong, Percussion-based bolt looseness monitoring using intrinsic multiscale entropy analysis and BP neural network, Smart Mater. Struct. 28
(12) (2019), 125001.
[201] F. Wang, G. Song, Bolt-looseness detection by a new percussion-based method using multifractal analysis and gradient boosting decision tree, Struct Health
Monit. 19 (6) (2020) 2023–2032.
[202] F. Wang, S.C.M. Ho, G. Song, Modeling and analysis of an impact-acoustic method for bolt looseness identification, Mech. Syst. Signal Pr. 133 (2019), 106249.
[203] F. Wang, A. Mobiny, H. V. Nguyen, If structure can exclaim: a novel robotic-assisted percussion method for spatial bolt-ball joint looseness detection, Struct
Health Monit. (2020). Online.
[204] P.L. Mazzeo, N. Ancona, E. Stella, Visual recognition of hexagonal headed bolts by comparing ICA to wavelets. Proceedings of the 2003 IEEE International
Symposium on Intelligent Control, 2003.
[205] P.L. Mazzeo, M. Nitti, E. Stella, A. Distante, Visual recognition of fastening bolts for railroad maintenance, Pattern Recogn. Lett. 25 (6) (2004) 669–677.
[206] P.L. Mazzeo, E. Stella, M. Nitti, Visual recognition of fastening bolt in railway maintenance context by using wavelet transform, Int. J. Graphics, Vision Image
Process SI 1 (2005) 25–32.
[207] H. Zhang, H. Ping, X. Yang, Fault detection of train center plate bolts loss using modified LBP and optimization algorithm, Open Automat. Control Syst. J. 7 (1)
(2015) 1916–1921.
[208] Y. Rubinsztejn, Automatic Detection of Objects of Interest from Rail Track Images, University of Manchester, Faculty of Engineering and Physical Science,
2011.
[209] Y. Xia, F. Xie, Z. Jiang, Broken railway fastener detection based on adaboost algorithm. International Conference on Optoelectronics & Image Processing, 2011.
[210] L. Liu, F. Zhou, Y. He, Automated status inspection of fastening bolts on freight trains using a machine vision approach, Proceedings of the Institution of
Mechanical Engineers, P. I. Mech. Eng. F-J. Rai. 230 (7) (2016) 1629–1641.
[211] J. Liu, Y. Xiong, L.I. Bailin, Research on automatic inspection algorithm for railway fastener defects based on computer vision, J. China Railway Soc. (2016).
[212] F. Marino, A. Distante, P.L. Mazzeo, A real-time visual inspection system for railway maintenance: automatic hexagonal-headed bolts detection, IEEE T. Syst.
Man. Cy. C. 37 (3) (2007) 418–428.
[213] H. Fan, P.C. Cosman, Y. Hou, High-speed railway fastener detection based on a line local binary pattern, IEEE Signal Proce. Let. 25 (6) (2018) 788–792.
[214] H. Feng, Z. Jiang, F. Xie, Automatic fastener classification and defect detection in vision-based railway inspection systems, IEEE T. Instrum. Meas. 63 (4)
(2014) 877–888.
[215] H. Zhang, J. Yang, T. Wei, Vision method of inspecting missing fastening components in high-speed railway, Appl. Optics. 50 (20) (2011) 3658.
[216] Y. Dou, Y. Huang, Q. Li, A fast template matching-based algorithm for railway bolts detection, Int. J. Mach. Learn. Cyb. 5 (6) (2014) 835–844.
[217] Y.J. Cha, K. You, W. Choi, Vision-based detection of loosened bolts using the Hough transform and support vector machines, Automat. Constr. 71 (2016)
181–188.
[218] L. Ramana, W. Choi, Y.J. Cha, Fully automated vision-based loosened bolt detection using the Viola-Jones algorithm, Struct. Health. Monit. 18 (2) (2019)
422–434.
[219] X. Kong, J. Li, Image registration-based bolt loosening detection of steel joints, Sensors. 18 (4) (2018) 1000.
[220] Y. Zhang, X. Sun, K.J. Loh, Autonomous bolt loosening detection using deep learning, Struct. Health. Monit. 19 (1) (2020) 105–122.
[221] J. Park, T. Kim, J. Kim, Image-based bolt-loosening detection technique of bolt joint in steel bridges. 6th international conference on advances in experimental
structural engineering, University of Illinois, Urbana-Champaign, 2015.
[222] J. Park, T. Kim, K. Lee, Novel bolt-loosening detection technique using image processing for bolt joints in steel bridges, Incheon, Korea, sn. (2015).
[223] J.H. Park, T.C. Huynh, S.H. Choi, Vision-based technique for bolt-loosening detection in wind turbine tower, Wind Struct. 21 (6) (2015) 709–726.
[224] P.J. Shull, T.C. Nguyen, T.C. Huynh, Bolt-loosening identification of bolt connections by vision image-based technique, P. SPIE-Int. Society Opt. Eng. 9804
(2016), 980413.
[225] T.C. Huynh, J.H. Park, H.J. Jung, Quasi-autonomous bolt-loosening detection method using vision-based deep learning and image processing, Automat. Constr.
105 (2019), 102844.
[226] C. Wang, N. Wang, S.C. Ho, Design of a new vision-based method for the bolts looseness detection in flange connections, IEEE T. Ind. Electron. 67 (2) (2019)
1366–1375.
[227] H.C. Pham, Q.B. Ta, J.T. Kim, Bolt-loosening monitoring framework using an image-based deep learning and graphical model, Sensors. 20 (12) (2020) 3382.
[228] X. Zhao, Y. Zhang, N. Wang, Bolt loosening angle detection technology using deep learning, Struct. Health. Monit. 26 (1) (2019).
[229] Y. Xie, J. Sun, On-Line Bolt-Loosening Detection Method of Key Components of Running Trains Using Binocular Vision, Society of Photo-optical
Instrumentation Engineers, 2017.
[230] J. Sun, Y. Xie, X. Cheng, A fast bolt-loosening detection method of running train’s key components based on binocular vision, IEEE Access 7 (2019)
32227–32239.

29