Structural Health Monitoring using Apodized Pi-

Phase Shifted FBG: Decoupling Strain and

Temperature Effects
Farinaz Kouhrangiha*, Mojtaba Kahrizi, Khashayar Khorasani
Department of Electrical and Computer Engineering (ECE), Concordia University, Montreal, Canada
Emails: *[email protected], [email protected], [email protected]

Abstract— Fiber Bragg Gratings (FBGs) as tools for cross-sensitivity effects in the FBG, temperature can also
structural health monitoring are broadly used for assessing the affect strain measurement at the same time. Thus, it is
composite structures behavior under different scenarios to important to separate the temperature effects from the strain
secure the system in a reliable and accurate way. Due to cross- measurements concurrently to have an accurate
sensitivity effects in optical sensors, one way to enhance the representation of the strain that is experienced by the
accuracy is to disentangle the strain from other affecting structure [9].
parameters such as temperature or vibration for recognizing
the real strain that is experienced by the composite structure. In the present paper, in order to enhance the FBG sensor
In this work, design and numerical simulations of Apodized Pi- performance and increase its accuracy the problem of
Phase Shifted FBG (π-PSFBG) are presented to evaluate the compensation is addressed by utilizing NNs. This method
performance of non-uniform FBG for simultaneous strain and attempts to compensate the strain from the other affected
temperature monitoring. Due to specific accuracy and spectral parameters by predicting the physical gauges according to
characteristics of the π-PSFBG, it is selected as an optical the sensor reflection signals. This intelligent behavior is
sensor to enhance the sensitivity of the measurements. Sensor implanted to the sensor to measure the strain effects
signals are designed and simulated by solving coupled mode independent of changes in the temperature by training a NNs
equations using the transfer matrix to represent the reflection yielding quite a satisfactory performance.
spectrum of π-PSFBG. To accomplish spectral improvement,
the Gaussian apodization function is applied to the FBG Towards this end, the performance of non-uniform FBG(
reflected spectrum in order to optimize its spectra by π-phase shifted FBG) have been investigated under various
suppressing side lobes. Moreover, we have developed π-PSFBG quantities of strain and temperature. Aassessment of the
sensor using Neural Networks (NNs) approach to sense and device performance was carried out using software to
discriminate strain from other affecting gauges such as simulate the sensor numerically. The simulation represent the
temperature. The proposed neural networks is trained to learn reflected spectrum of the FBG considering all affecting
the relationship between the reflection spectrum and the parameters simultaneously. In order to increase the
external parameters such as strain and temperature. Our sensitivity of the device, a π-phase shifted FBG (as an
investigations not only characterize the performance of an advanced optical element) was applied along with the
apodized π-PSFBG to simultaneously measure two parameters
Gaussian apodization function. Our framework has two NNs
with high sensitivity, but also yield the minimum error in
compensation of strain from temperature. that are capable of learning the relationship between physical
parameters and the reflection spectrum to predict expected
Keywords— pi-phase shifted FBG; apodization; neural parameters with minimum error. Our goal was to develop a
networks (NNs); fiber Bragg grating (FBG); optical sensor sensor with high sensitivity that can detect and measure the
affecting parameters (such as strain and temperature) on the
system structure. This sensor would be able to compensate
I. INTRODUCTION
for the effect of temperature changes and consequently
Fiber Bragg Gratings (FBGs) as effective monitoring provide a strain that is experienced by the composite
tools are used in Structural Health Monitoring (SHM) structure. The results confirm the effectiveness of our
application to monitor, detect, and measure the present proposed methodology.
anomalies such as strain, which have affected the system [1].
Simultaneous multi-parameter measurement along with other II. MODELING AND SIMULATIONS
numerous advantages do make FBG sensors as optical
devices valuable for health condition monitoring of a system A. Design of Apodized π-Phase Shifted FBG
[2], [3]. The embedded FBG sensors in composite materials
can measure the change of various external phenomena such In this section, we propose the spectral characteristics
as strain, temperature, vibration, and other parameters to and spectral response of the highly sensitive π-PSFBG
monitor and enhance reliability and accuracy of the system sensor that is based on coupled mode behavior using the
[4], [5]. External forces on a fiber cause a shift in the Bragg transfer matrix method. Due to accuracy and spectral
wavelength according to the change in its effective refractive characteristics of the π-PSFBG, it is selected as an optical
index. Therefore, by measuring the wavelength change the sensor to enhance sensitivity measurements. In our design,
strain and temperature could be measured [5], [6], [7], [8]. the π-phase shift discontinuity is considered in the middle of
Due to the correlation dependency of the parameters and the FBG in order to generate two separate gratings. The shift
in the reflection spectrum shows a very narrow transmission

978-1-7281-1634-1/19/$31.00
Authorized licensed use limited to: Indian©2019
InstituteIEEE
of Technology (ISM) Dhanbad. Downloaded on August 20,2024 at 09:51:23 UTC from IEEE Xplore. Restrictions apply.
 window that increases the precision of the parameter
measurements [10], [11]. In the designed FBG, the length of
the fiber is kept constant at 25 mm with an effective
refractive index of 1.458, and the index modulation variation
of 1.1×10-4 with a designed wavelength of 1550 nm. The
mathematical model of the π-phase shifted FBG spectrum
can be defined according to the following expression [12]:

where denotes the “dc” index change (change over the

grating period), denotes the fringe visibility of the index
change, denotes the grating period, z denotes the
coordinate of light propagation, and represents the
grating chirp which is zero for a uniform grating. The
reflection spectrum of the π-PSFBG is observed by invoking
the coupled mode theory using the transfer matrix approach
that was proposed by [13].
To improve sensitivity of a non-uniform FBG, the
Gaussian apodization profile is applied to the simulated
reflectivity. Large side lobes are observed due to presence of
multiple reflections at the end of the grating. There are
clearly undesirable for FBG sensing and make the approach
not suitable for high-performance applications. Hence, we
propose to utilize the apodization profile to act as a filter, Fig. 1. (a) The π-PSFBG, and (b) Apodized π-PSFBG
will suppress the side lobes while maintain a high reflectivity
having a narrow bandwidth. The apodization function is TABLE 1
given by [11], [13], [14]: Pi-PSFBG vs Apodization Profile Evaluation
FEATURES REFLECTIVITY BANDWIDTH
PS-FBG 0.9483 522.134 pm
where denotes the grating length and denotes an Apodized PSFBG 0.9261 49.096 pm
apodized factor. Figures. 1.(a) and 1.(b) display the
simulation results before and after applying the Gaussian
apodization approach, respectively. Comparisons of the π-
PSFBG reflectivity, the width of the signal (FWHM), and the
side lobes with an apodized π-PSFBG show that by applying
the Gaussian apodization approach more than 92% of the
side lobes have been suppressed, the bandwidth (FWHM) of
the reflection spectrum has decreased to a narrower width,
although; the reflectivity has almost preserved having the
same height (Table 1). Consequently, improvements in the
spectral signal describe the prominent role of apodization
profile in enhancing the sensor sensitivity.
External parameters such as the strain and temperature
Fig. 2. Apodized π-PSFBG Strain and Temperature Measurements
will cause a shift in the Bragg wavelength. This may be
computed as follow [5]:
Temperature( ):

where denotes the thermal

expansion coefficient (for Silica) and the thermo-optic
coefficient (for Germania doped Silica) is given by:
.
Fig. 3. Changes in the Center Wavelength of π-PSFBG vs. (a) Temperature
Strain ( ): (b) Strain

where denotes an effective photoelastic coefficient and is

equal to . In this Figure. 2, depicts, the sensor reflectivity under different
relation = 0.113, = 0.252, and = 0.16. parameter changes. The linear dependence of the Bragg
wavelength with respect to strain and temperature are shown
Applying equations (3) and (4), the Bragg wavelength shift in Figures: 3(a) and 3(b), respectively where the π-PSFBG
due to simultaneous variations of strain and temperature can shows good linearity with respect to both temperature and
be computed according to: strain responses.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on August 20,2024 at 09:51:23 UTC from IEEE Xplore. Restrictions apply.
 TABLE 2
B. Neural Networks (NNs) Actual and predicted temperature and strain measured by NNs
FBG sensors are capable of sensing the parameters NN Temperature
variations such as strain and temperature by changing Spectrum Index Target (actual) Values Predicted Values
wavelengths which appears in its reflection spectrum [5], 5821 -10.303 -10.362
[15]. We have proposed an intelligent sensor approach by 2516 -17.374 -17.299
using NNs to obtain strain values that are independent of 4282 75.960 75.834
temperature measurements as far as the reflection spectrum 4173 63.232 63.097
in concerned. Therefore, the relationship between reflection 255 37.778 37.607
spectra and its corresponding physical parameters is expected MAE: 0.7940
to be learned by utilizing the designed NNs. NN Strain
We have developed two NNs that are operating in Spectrum Index Target (actual) Values Predicted Values
parallel. The output targets for NNs are the temperature and 5821 0.0003 0.0003
the strain, respectively and the input for both is a matrix of 2516 0.0002 0.0002
the reflected spectrum features. 4282 0.0012 0.0011
4173 0.0011 0.0011
In order to create a dataset for our artificial neural
network, it is necessary to extract the features of reflection 255 0.0008 0.0008
spectra and apply them as an input to the model. To have the MAE: 0.0067
precise results in ANN, the normalized input data should be
encoded carefully because of the high dependency of
reflection spectrum on both strain and temperature
values. After creating a dataset, the algorithms should train
our dataset and estimate the output values.
Neural networks utilized in this paper are feed forward
networks. The back-propagation (BP) algorithm is used to
train the two networks. The extracted features from the
reflected spectrum act as an input matrix of X and the
temperature and strain variations as the output vector Y are
fed into the network. Fig. 4. NN Temperature error with respect to its validation.

Our dataset is divided into training and testing sets with

the ratio of 70% and 30%, respectively. The training dataset
is obtained from the reflection spectrum characterization of
the π-PSFBG with respect to concurrent changes in both
strain (with range of 0 με- 1000 με) and temperature (with
range of -40 ºC- 100 ºC). It can be seen from Table 2 that for
the pi-phase shifted FBG, very close strain and temperature
results are obtained by the NNs when compared to the actual
ones based on their reflected spectrum features. It shows that
the NNs model are capable of predicting the corresponding
values quite correctly. Fig. 5. NN Strain error with respect to its validation.

Upon fitting the NNs model, its accuracy can be detecting the anomaly location on the composite structure.
determined, and the loss recorded regarding both the true and The apodized π-phase shifted FBG sensor is proposed by
estimated values. As shown in Table 2 the Mean Absolute incorproating the coupled mode equations using the transfer
Error (MAE) is around 0. 7940 for the temperature and matrix method. The designed sensor can compensate for
0.0067 for the strain that are smaller than 1. As Figures. 4 measurements at the same time by using the NNs approach
and 5 display the loss with respect to epochs, it follows that to sense both strain and temperature concurrently. From the
as epoch increases the loss decreases, which show good simulation studies, it is observed that sensor characteristics
learning rate with no overfitting of the network to the and NNs model outputs match very closely which
training data for both NNs networks. The learning curves demonstrates excellent results with low losses.
show that the training and validation errors implying how
well our NNs train and learn. Using a reliable training Furthermore, simulation results confirm that the FBG
algorithm; such as the BP can fulfill this objective. reflection spectrums with respect to variations of strain and
Therefore, the loss would reduce after each optimization temperature show a good response according to their central
iteration. Also increasing the size of the input samples and wavelength. Moreover, investigation of the sensor sensitivity
reducing the batch size can make the model more accurate. shows that the optical element response to strain and
From these investigations, the effectiveness and benefits of temperature with respect to the Bragg wavelength is linear.
the NNs models are quite evident. Finally, it was concluded that a NNs intelligent sensor with
Apodized π-PSFBG is a reliable and valid candidate for
III. DISCUSSIONS AND CONCLUSIONS sensing two parameters of the strain and temperature
simultaneously. Compensation of the strain from affecting
In this paper, we have developed a NNs methodology for temperature has been verified by utilizing our proposed NNs
an intelligent strain and temperature sensor for the purpose of methodology.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on August 20,2024 at 09:51:23 UTC from IEEE Xplore. Restrictions apply.
 REFERENCES [8] M. Etezad, M. Kahrizi, K. Khorasani, “Detection of transverse load
using the high birefringence bragg grating and genetic algorithm,”
[1] M. Etezad, M. Kahrizi, K. Khorasani, “A novel multiplexed fiber InIECON 2012-38th Annual Conference on IEEE Industrial
Bragg grating sensor with application in structural health Electronics Society, IEEE, pp. 2579-2582, Oct 25, 2012,
monitoring,” physica status solidi c, vol. 8, no. 9, pp. 2953-6, Sep.
[9] G. Pereira, M. McGugan, and L. P. Mikkelsen, “Method for
2011.
independent strain and temperature measurement in polymeric tensile
[2] Q. Zhang, T. Zhu, J. Zhang, and K. S. Chiang, “Micro-fiber-based fbg test specimen using embedded fbg sensors,” Polymer Testing, vol. 50,
sensor for simultaneous measurement of vibration and temperature,” pp. 125-134, 2016.
IEEE Photonics Technology Letters, vol. 25, pp. 1751-1753, Sept
2013. [10] H. Wang et al., “Phase-shifted fiber-Bragg-grating-based humidity
sensor,” Proc. SPIE, vol. 8720, no. 1, pp. 872019, May 2013.
[3] J. Shan-chao, W. Jing, S. Qing-mei, Y. Qing-lin, and W. Li-jun,
[11] H. M. EL-Gammal, H. A. Fayed, A. Abd El-Aziz, M. H. Aly et al.,
“Study of three-component fbg vibration sensor for simultaneous
measurement of vibration, temperature, and verticality,” Journal of “Performance analysis & comparative study of uniform, apodized and
pi-phase shifted FBGs for array of high-performance temperature
Sensors, vol. 2015, 2015.
sensors,” Optoelectronics and Advanced Materials – Rapid
[4] H. Yi, H. Kim, S. Choi, and J. Pan, “Novel intensity-based fiber optic Communications, Vol. 9, No. 9-10, pp. 1251 -1259, September –
vibration sensor using mass-spring transducer,” in 2014 International October 2015.
Conference on Electronics, Information and Communications
(ICEIC), pp. 1-2, Jan 2014. [12] T. Erdogan, “Fiber grating spectra,” Journal of lightwave
technology, vol. 15, no. 8, pp.1277-1294, 1997.
[5] O. Hassoon, M. Tarfoui, and A. E. Malk, “Numerical simulation of
[13] K. Khalid, M. Zafrullah, S. Bilal, and M. Mirza, “Simulation and
fiber bragg grating spectrum for mode- delamination detection,”
International Journal of Mechanical, Aerospace, Industrial, analysis of gaussian apodized fiber bragg grating strain sensor,”
Journal of Optical Technology, vol. 79, no. 10, pp. 667-673, 2012.
Mechatronic and Manufacturing Engineering, vol. 9, no. 1, pp. 144 -
149, 2015. [14] F. Chaoui, O. Aghzout, M. Chakkour, and M. El Yakhloufi,
[6] A. Güemes, L. Fernández-López, P. Díaz-Maroto, A. Lozano, and J. “Apodization optimization of fbg strain sensor for quasi-distributed
sensing measurement applications,” Active and Passive Electronic
Sierra- Perez, “Structural health monitoring in composite structures
by fiber-optic sensors,” Sensors, vol. 18, no. 4, p. 1094, 2018. Components, vol. 2016, 2016.
[15] B. Mondal, Z. Hussain, P. Sharan, and S. Islam, “Modelling and
[7] S. Huang, J. Li, J. Zhou, and W. Gu, “Novel spectrum properties of
simulation of temperature sensitivity of bragg grating sensor for
the periodic pi-phase-shifted fiber bragg grating,” Optics
structural health monitoring application,” International Journal of
Communications, vol. 285, no. 6, pp. 1113-1117, 2012.
Advanced Computer Research, vol. 4, no. 1, p. 364, 2014.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on August 20,2024 at 09:51:23 UTC from IEEE Xplore. Restrictions apply.