1558-1748 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2018.2872174, IEEE Sensors
Journal
1
Abstract— The use of UHF RFID passive tags for defect detection
is a promising application in structural health monitoring.
However, it’s a challenging task while most related information to
tag antenna design is not available as well it suffers from the
interference effect on wireless measurements. In this article, we
investigated and developed a new technique for crack depth
sensing by using a passive UHF RFID tag as a sensor which
interrogated by thingmagic M6e platform. Wireless power
transfer WPT level and the frequency sweeping are used to match
between tag impedance and metal induction effect. The distance
between the tag and reader is adjusted at 30cm which can achieve
high quality factor. As a result, the tag backscatter signal become
rich with maximum peak components. The proposed technique
called power peaks feature extraction (PPFE) which is used to
detect the artificial crack depth on the surface of the stainless steel
and ferromagnetic samples. Skewness is applied on PPFE to offer
a direct approximation procedure for the crack depth. A linear
relationship of skewness achieves high accuracy result with a
maximum estimation error of 0.1 mm for stainless steel sample,
the technique is validated and compared with the frequency
domain result, and it achieves all most the same accuracy for the
stainless steel sample.
Index Terms—passive RFID tag, crack depth, skewness.
I. INTRODUCTION
he Small cracks appear on the metal material surface, or
deep inside defiantly affect the performance of the
mechanical structure. The growth of the crack leads to decay
system performance or to complete damage of the material
which it may cause a severe disaster if it is not detected in
earlier time. Health monitoring and non-destructive testing
(NDT) systems were emerged to give a real-time report of the
monitored system without disturbance of the system operation
[1]. Radio frequency identification (RFID) system provides an
alternative solution for wireless sensing and real-time health
monitoring. RFID system is composed of reader and tag. The
tag is composed of antenna and radio frequency integrated
circuit RFIC. The tag scavenges its operating power from the
reader interrogation signal. The backscatter signal from the tag
includes a tag electronic product code (EPC) unique identifier
and some measurable parameters such as received signal
strength (RSSI) and phase.
Nowadays, RFID systems have been widely used in many
areas and have been developed for use in the area of sensor
system [2]. RFID system is classified into three groups due to
operating frequency, low-frequency LF, high-frequency HF,
and ultrahigh frequency UHF. Later UHF RFID is the most
popular used when it is compared with LF and HF RFID
systems because of its far distance reading range up to ten
meters [3], and it could be deployed to form a monitoring
WSNs. The disadvantage of using UHF RFID signals it cannot
penetrate to detect defect inside material while other frequency
ranges of RFID were used for that purpose [4].
The challenges of using on metal mounted UHF RFID tag for
defect sense, rely on the change of the tag antenna specification
due to the change of the tagged object material. The use of tag
antenna sensing capability is divided into two categories direct
and indirect measurement strategies [1][2]. The direct strategy
may include tag turn on power [5], backscattered power [6], and
phase [7]. The indirect strategy may include radar cross section
(RCS) [8], an analog identifier (AID) [3], and an in-
phase/quadrature IQ signal based sensing [4]. The indirect
measurement strategy may need additional hardware for
investigation or need more information about the tag antenna
specifications like antenna impedance, chip impedance, and
chip activation power. Most often, not all of this information is
available in vendor datasheet. Therefore, researchers deal to use
their own designed tag when they use it as a sensor. Commercial
tags unavailable information’s and the limitation of the
harvested power level and the attenuation of the transmitted
signal make the use of a passive antenna for defect detection
remains a challenge. However, regarding the use of the passive
tag, researchers are developing various techniques for defect
detection like strain [9][10][11], cracks [6][12][13], and
corrosion [3][14][15]. The limitation of the techniques used for
crack detection either it used short communication range like
LF or HF, nor it can detect the cracks that directly persist on tag
antenna instead of the cracks that occur in the monitored
substances.
The contribution of the proposed article can be drawn from
its ability to detect the under tag crack depth by using UHF
reader platform, while most of the similar research focus on
designing tags for sensing [3], or they may use high-cost
apparatus such as vector network analyzer (VNA) [14].
Mugahid Omer
1,2
, Gui Yun Tian
1, 3*
, Bin Gao
1
, Dongming Su
1
1 School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, People’s Republic of China
2 Faculty of Engineering and Technology, Nile Valley University, Atbara, Sudan
3 School of Electrical and Electronic Engineering, Newcastle University, Newcastle upon Tyne, NE1 7RU, United Kingdom
* Corresponding author: g.y.tian@uestc.edu.cn
Passive UHF RFID tag as a sensor for crack
depths
T
1558-1748 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2018.2872174, IEEE Sensors
Journal
2
Alternatively, a new reliable methodology is used, which we
call it power peak feature extraction (PPFE), and it achieves
high accuracy result with a direct relationship between the crack
depth increment and the change of the skewness function when
compared with the accuracy achieved in [7][16].
II. UHF RFID TAG SENSING METHODOLOGY
Researchers investigated the use of RFID systems for defect
detection and characterization for many reasons related to tag
features such as low cost, small profile, has a unique identifier,
easy to deploy in a wide area and remotely accessible. These
features, encourage researchers to propose a potential use of
tags as a distributed sensor network. Fig. 1 illustrates the
sensing mechanism and potential applications.
Fig. 1. Potential Distribution of passive RFID tag sensor network for defect
detection.
This article focuses on the study of detecting under tag
surface crack and the corresponding impact due to the
increment of the crack depth. However, to implement the
sensing technique, the power level is swept gradually, and the
received signal is analyzed to detect the crack and follow up the
changes of the crack depth.
A. RFID tag signal capturing principle
Backscatter signal from tag has different information such as
tag ID, RSSI, frequency, and phase. This information is
extracted to identify, track, and sensing. Many strategies are
followed to implement the desired target. Therefore, to achieve
the target of this article, a sweeping power technique is used. In
more details, for each value of the frequency band (902–
928MHz) the transmitter power is increased until the tag
receives sufficient power to activate the chip. Thus, as a result,
the tag backscatter the signal to the reader which start to record
all information corresponding to the received signal. In turn, the
similar scenario is repeated for the new frequency.
Charge pump rectifier circuit, which is used to provide DC
power to RFIC, produces more efficient power at different
peaks level of power optimization waveform POW when it is
compared with continues wave (CW) form [17]. As well it
seems like, pulse eddy current which it can reduce
environmental interference and increase transient response
measurement sensitivity [18]. In our case, the reliability of
RSSI measurement is increased. The conducting and
magnetizing properties of objects could be characterized by
transient response [19]. Therefore, all of this information
inspiring to infer a new technique depending on power peaks.
The transient response of power peaks are used to detect the
defect of the material while tag is frequency and power
dependent.
Maximum wireless power transfer (WPT) between tag and
reader is affected by the tag quality factor, which it is frequency
dependent. Equation (1) represents the quality factor without
metallic object effect [20]
2
r T ag
Tag
fL
Q
R
(1)
Where L
tag
and R
tag
refer to tag inductance and resistance
respectively. When the tag attached with a metallic object, the
mutual inductance between the tag and object should be
considered, and it could be represented by metal equivalent
resistance and inductance R and L respectively. The resistance
R depends inversely on metal conductivity and, L depends on
metal permeability, where both R and L depend on the eddy
current path. To simplify the effect of the metal attached or
placed near the tag, only the metal inductance effects could be
added in parallel with the tag circuit as shown in Fig. 2.
The new technique is called power peaks feature extraction
PPFE which it focuses on the peak points of the received power.
The proposed technique claims that the dominant extracted
features are accompanied with the transient response of the
peak points which can achieve high quality factor due to
impedance matching. The crack changes the induction behavior
of the metallic materials. on-object antenna impedance can
achieve both maximum radiation and chip impedance matching
due to power and frequency sweeping [1] as shown in
equivalent circuit Fig. 2.
Fig. 2. The equivalent circuit for the tag attached to a metallic object
The PPFE technique is applied in the time domain, then it
validated with the frequency domain analysis, and it achieves
high accuracy for under tag crack detection. One of the
advantages of the proposed testing techniques, only the reader
system is used for testing, and there is no need for more
additional apparatus. As well, all measurements are
independent of tag unknown parameters; these features will
expand the use of the commercial UHF RFID tags for defect
Perception
Programmable
Monitoring system
Tags
Crack
Bridge
Train
Pipeline
Reader antenna
1558-1748 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2018.2872174, IEEE Sensors
Journal
3
sensing. The concept and implementation of PPFE are
described in more details in section C.
B. RFID response and feature extraction for crack
characterization
RFID and PEC have the same behavior when they respond to
the pulsed signal. The signal which contains multiple
frequencies has different penetration depth capabilities. The
direct relationship between the frequency and the penetration
depth δ on the metallic material described in equation (2)
1 f
(2)
Where f is the pulsed signal frequency, σ and μ represent the
conductivity and permeability of the material. It’s obvious from
equation (2) that the skin depth has inversed relationship with
the frequency. as well the defects of the material can effect on
conductivity and permeability. Extensive studies have been
proposed to observe the change of material conductivity and
permeability over the corroded layer [20][21]. Same like
corrosion, cracks can be detected by exposing the material to
different frequency components and then, extract the features
that effected due to conductivity and permeability changes. The
change of signal penetration depth accompanied with the
change of material properties which caused as a result of defect
persistence, lead to extract different features like signal
maximum peak value, change of peaks during a period of time,
and the difference between the maximum and a minimum peak
of the signal. After the implementation of the PPFE technique,
we observe that the increment of the under tag crack depth
makes the stainless steel sample behave like the healthy
ferromagnetic sample and vice versa.
C. PPFE Implementation in the time domain
The PPFE is applied to the RSSI signal which is represented
in the time domain. The main idea behind the PPFE is to extract
and monitor the health status of the under tag material in a novel
and straightforward relationship. A set of steps should be
followed
i. The interrogation reader code sequence is shown in table 1.
TABLE 1.
READER PSEUDO CODE.
while(1) {
For frequency = 902 : step 0.5 : 928
For power= 5: step 0. 5: 30
Reader sent request
If tag respond
Save the received data
Exit power loop
End if
Next power
Next frequency
}
ii. The received data for 500 seconds is saved and is analyzed
in Matlab
iii. The total period is divided into short interval periods, the
length of the short period T is calculated as shown in Eq.
(1).
FP FR
T T T
(3)
Where T
FP
is the time for the first peak, and T
FR
is the time
for the first response. The transient response of the peaks is
calculated within each period of time T, which could be
defined as the count of peaks per each period time T. This
feature used to detect the variation of the crack depth. Fig.3
shows the representation of T period in the received signal
graph.
iv. Number of peaks is calculated for each interval time T
Fig. 3. Representation of the period time T on the received power signal.
D. Skewness feature extraction for PPFE
The skewness feature is used to test the bias of the PPFE
readings for each one of the test samples. The main role of this
statistical feature is to evaluate asymmetry of the data [4]. For
a set of data X, the skewness feature is given by Eq. (2).
3
E
Xm
SD
(4)
Where X is the data, m is the mean; SD is the standard
deviation. The skewness has zero value for the normally
distributed data. A negative value or positive value for the
skewness indicates that the left tail has long relative to the right
tail and vice versa.
III. EXPERIMENTAL SETUP
A specimen of a ferromagnetic sample is 140×60×10mm,
and a sample of stainless steel is 120 × 60 × 5mm are shown in
Fig. 4. These samples attached with the RFID tag for crack
detection. The ferromagnetic sample has four artificial cracks
with 8 × 0.2 mm for length and width respectively, while it is
prepared with different depths 8, 8.5,9 and 9.5mm. The
stainless steel sample has three artificial cracks prepared with
similar length and width 13 × 0.5 mm respectively and with
different depth 0.5, 1 and 1.3mm. The samples are tested from
30cm far from the reader by using the Thingmagic M6e
platform with 6 dBi reader antenna gain to monitor and observe
the changes in the received signal. The received signal from
healthy and cracked sample is analyzed in the frequency
domain and time domain. The maximum reading distance for
the tag to respond is 40cm for the healthy samples and 35cm for
0 50 100 150 200 250 300 350 400 450 500
-75
-70
-65
-60
-55
-50
-45
Time (s)
T
RSSI (dBm)
Peak points
1558-1748 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2018.2872174, IEEE Sensors
Journal
4
the cracked samples.
Fig. 4. Reader measurement platform and the sample under test.
IV. RESULTS AND DISCUSSION
A. Time domain analysis for stainless steel
The received signal peaks distribution, as shown in Fig. 5, 6,
7, and 8, have clear differences. Healthy sample as shown in
Fig. 5 has no peaks because the height difference between the
adjacent peaks is not sufficient to pass the threshold value,
which it has been adjusted to be at least more than two. Thus,
the period T as defined in Eq. (1) devolves to zero and the
corresponding bar plot will be zero. Fig. 6 and Fig. 7 show the
received signal for the cracked sample with 0.5 mm, 1 mm and
1.3 mm, respectively. Fig. 6 represents the received signal with
the maximum peaks while Fig .7 represents the peak counts in
each period T. If the existence of peaks within the period, is
represented by hit state, and the absence of peaks within T
period represented by miss state. Therefore, all crack states for
stainless steel as shown in the bar plot of Fig. 7, could be
encoded in the form of [hit miss hit], while for healthy sample
only miss state is available. These features could be used to
distinguish between healthy and cracked sample also it could
be coded in binary form.
Fig. 5. Received power signal for the stainless steel healthy sample measured
from different distances (a) 30 cm far from the reader (b) 35 cm far from the
reader (c) 40 cm far from the reader.
Fig. 6. Received power signal for stainless steel cracked samples measured from
30 cm (a) 0.5 mm crack depth (b) 1 mm crack depth (c) 1.3 mm crack depth.
Fig. 7. Peaks count extracted from the received power at each T interval time
for stainless steel cracked samples measured from 30 cm (a) 0.5 mm crack depth
(b) 1 mm crack depth (c) 1.3 mm crack depth.
After crack detection, it is required to go deep and derive a
relationship between the number of peaks and crack depth
estimation; Skewness is used to build this relationship.
Therefore, as shown in Fig.7, when the crack depth is increased
the skewness is decreased as shown in Eq. (3).
1
d
sk
c
(5)
Where sk is the skewness for the extracted power peaks, the
cd is the crack depth. In Fig. 7.b, this linear curve fitting has a
maximum error of 0.1mm; this means the crack depth
estimation is achieved by a simple and direct linear relationship.
This relationship is affected by the separation distance between
reader and test sample.
Fig. 8. The skewness for different crack depth on stainless steel (a) skewness
with the curve fitting (b) residuals.
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
-20
received power in time domain uncracked
0 50 100 150 200 250 300 350 400 450 500
-70
-60
-50
-40
0 50 100 150 200 250 300 350 400 450 500
-65
-60
-55
-50
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
received power in time domain cracked 30cm
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
1 2 3 4 5
0
5
10
15
1 2 3 4 5 6
0
5
10
15
1 2 3 4 5 6
0
10
20
-0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
-0.1
0
0.1
residuals
-0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0.4
0.6
0.8
1
1.2
1.4
skewness
linear fitting
Antenna
Stainless steel
Cracks
Cracks
Ferromagnetic
Reader platform
a
b
c
a
c
b
a
b
c
a
b
Time (s)
RSSI (dBm)
Time (s)
RSSI (dBm)
No of T periods
Peaks count
Skewness
Crack depth
Residual
s
Stai
nles
s
steel
Tag
Monitoring system
1558-1748 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2018.2872174, IEEE Sensors
Journal
5
To test the occurrence of the peaks phenomenon even if the
measurement distance is changed. The cracked stainless steel
sample tested from 35cm, and the received signal is shown in
Fig. 9, where Fig. 10 represents PPFE of the received signal.
Fig. 9. Received power signal for stainless steel cracked samples measured from
35 cm (a) 0.5 mm crack depth (b) 1 mm crack depth (c) 1.3 mm crack depth.
Fig. 10. A number of peaks extracted from the received power at each T interval
time for stainless steel cracked samples measured from 35 cm (a) 0.5 mm crack
depth (b) 1 mm crack depth (c) 1.3 mm crack depth.
It is obvious from Fig. 9.a the signal is increased and
decreased gradually without making any variation in high peaks
as a result in Fig. 10.a the number of peaks is zero which mean
the small depth cracks could not be detected at the maximum
reading distance, In other words, the change of the separating
distance between the test sample and the reader affects the
under tag material behavior. Thus, the separating distance 30cm
give a good match to detect small crack depth for the on-object
tag.
B. Time domain analysis for ferromagnetic sample
The distribution of the received signal peaks is shown in
Fig.11. It consists of clear differences. Healthy sample as
shown in Fig. 11.a, and cracked sample as shown in Fig. 11.e,
they have clear peaks. While Fig. 11.b, Fig. 11.c and Fig. 11.d,
have smooth curves. Thus, the technique PPFE could not be
used in this case because it has no enough peaks to be linked
with the skewness function to fit a direct relationship.
Fig. 11. Received power signal from ferromagnetic sample measured from
30cm (a) healthy sample (b) 8mm crack depth (c) 8.5mm crack depth (d) 9mm
crack depth (e) 9.5mm crack depth.
C. Frequency domain analysis for stainless steel sample
The received signal and the transmitted power are measured
three times for each frequency within the range 902–928 MHz
with the capturing procedure illustrated in table 1. The average
value of the transmitted power and the ratio of the (received
power/transmitted power) are shown in Fig. 12 and Fig. 13,
respectively.
Fig. 12. Transmitted power for stainless steel sample.
Fig. 13. (Received power/transmitted power) signal ratio for stainless steel
sample.
Healthy sample signal, as shown in Fig. 12 and Fig. 13, has
an obvious difference compared with the cracked sample. Thus,
to create a linear relationship for crack depth estimation, the
mean (m) of the (received power/ transmitted power) is
calculated in Eq. 4.
0 50 100 150 200 250 300 350 400 450 500
-70
-65
-60
-55
received power in time domain cracked 35cm
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
0 50 100 150 200 250 300 350 400 450 500
-80
-70
-60
-50
1
-1
0
1
0 5 10 15 20 25 30 35 40
0
2
4
1 2 3 4 5 6 7
0
5
10
15
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
received power in time domain cracked 30cm
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
0 50 100 150 200 250 300 350 400 450 500
-70
-60
-50
0 50 100 150 200 250 300 350 400 450 500
-70
-60
-50
0 50 100 150 200 250 300 350 400 450 500
-80
-60
-40
900 905 910 915 920 925 930
5
10
15
20
25
30
35
AVERAGE OF THE SENT POWER FOR CRACKED SAMPLE 30cm
healthy
0.5mm
1mm
1.3mm
900 905 910 915 920 925 930
0
0.5
1
1.5
x 10
-6
915 920 925
1
2
3
4
x 10
-8
healthy sample
0.5mm crack
1mm crack
1.3mm crack
a
b
c
a
b
c
Time (s)
RSSI (dBm)
No of T periods
Peaks count
e
d
a
b
c
Time (s)
RSSI (dBm)
Frequency (MHz)
Transmitted power
Frequency (MHz)
Received power / transmitted power