Professional Documents
Culture Documents
book Advances in Non Destructive Evaluation 2020
book Advances in Non Destructive Evaluation 2020
Shyamsunder Mandayam
Sarmishtha Palit Sagar Editors
Advances
in Non
Destructive
Evaluation
Proceedings of NDE 2020
Lecture Notes in Mechanical Engineering
Series Editors
Francisco Cavas-Martínez , Departamento de Estructuras,
Construcción y Expresión Gráfica Universidad Politécnica de Cartagena,
Cartagena, Murcia, Spain
Fakher Chaari, National School of Engineers, University of Sfax, Sfax, Tunisia
Francesca di Mare, Institute of Energy Technology, Ruhr-Universität Bochum,
Bochum, Nordrhein-Westfalen, Germany
Francesco Gherardini , Dipartimento di Ingegneria “Enzo Ferrari”,
Università di Modena e Reggio Emilia, Modena, Italy
Mohamed Haddar, National School of Engineers of Sfax (ENIS), Sfax, Tunisia
Vitalii Ivanov, Department of Manufacturing Engineering, Machines and Tools,
Sumy State University, Sumy, Ukraine
Young W. Kwon, Department of Manufacturing Engineering and Aerospace
Engineering, Graduate School of Engineering and Applied Science, Monterey,
CA, USA
Justyna Trojanowska, Poznan University of Technology, Poznan, Poland
Lecture Notes in Mechanical Engineering (LNME) publishes the latest develop-
ments in Mechanical Engineering—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core of
LNME. Volumes published in LNME embrace all aspects, subfields and new
challenges of mechanical engineering. Topics in the series include:
• Engineering Design
• Machinery and Machine Elements
• Mechanical Structures and Stress Analysis
• Automotive Engineering
• Engine Technology
• Aerospace Technology and Astronautics
• Nanotechnology and Microengineering
• Control, Robotics, Mechatronics
• MEMS
• Theoretical and Applied Mechanics
• Dynamical Systems, Control
• Fluid Mechanics
• Engineering Thermodynamics, Heat and Mass Transfer
• Manufacturing
• Precision Engineering, Instrumentation, Measurement
• Materials Engineering
• Tribology and Surface Technology
To submit a proposal or request further information, please contact the Springer
Editor of your location:
China: Ms. Ella Zhang at ella.zhang@springer.com
India: Priya Vyas at priya.vyas@springer.com
Rest of Asia, Australia, New Zealand: Swati Meherishi
at swati.meherishi@springer.com
All other countries: Dr. Leontina Di Cecco at Leontina.dicecco@springer.com
To submit a proposal for a monograph, please check our Springer Tracts in
Mechanical Engineering at https://link.springer.com/bookseries/11693 or contact
Leontina.dicecco@springer.com
Indexed by SCOPUS. All books published in the series are submitted for
consideration in Web of Science.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface for NDE 2020 Proceedings
v
vi Preface for NDE 2020 Proceedings
has been possible only due to the wholehearted support from the authors and for
their painstaking efforts to submit good quality content. The Editors express their
gratitude to all the reviewers for their diligent efforts to provide valuable inputs to
the authors for improving the quality and technical correctness of the publication.
The Editors also thank the Organizing Committee of NDE2020 for the opportunity
provided to bring out the proceedings in collaboration with Springer-Nature.
About ISNT
vii
viii Contents
Dr. Sarmishtha Palit Sagar is a Senior Principal Scientist and Group leader of
NDE & MM group at CSIR-National Metallurgical Laboratory, Jamshedpur, India.
She has over 23 years of experience in the area of Non-destructive Evaluation and
Sensor development. She pursued her B.Tech. and M.Tech. in Electronics & Commu-
nication from the Institute of Radio Physics & Electronics, University of Calcutta
and Ph.D. from Jadavpur University, India. Her research interest lies in ultrasonic &
advanced ultrasonic, signal & image processing, sensor & device development for
Industrial applications. Currently, she is involved in developing AI and IOT based
systems for life assessment of components. Dr. Sagar has worked as a visiting Scien-
tist at Fraunhofer Institute for Non-destructive Testing, Saarbreucken, Germany from
xi
xii About the Editors
2000–2002 and at Pennsylvania State University, USA in 2009. She has 70+ publi-
cations in journals of national and international repute, 13 patents and 7 copyrights
to her credit. She has guided several Master students and 3 Ph.D. students. She
is the recipient of several prestigious awards like National NDE Award for Inter-
national Recognition, National NDE award in R&D, Raman Research Fellowship
award by CSIR, Lady Engineer Award by Institute of Engineers, “Tata Innovista
2015” Regional award in “Promising Innovation in the Core processes” category.
Evaluation of Parallel Processing
Employed in Industrial Computed
Tomography
1 Introduction
2 FDK Algorithm
For cone-beam CT case [12], f (x, y, z) denotes the spatial function to be reconstructed
and it is distributed in the Cartesian (x, y, z) space, while the projection value is in
the (β, a, b) space. The source, represented by function S(β), moves along a circle
of radius R. At projection angle β it is positioned at the coordinate (–Rcosβ, –Rsinβ,
0). A planar detector is placed on the centre plane containing the axis of rotation,
referred to as a virtual detector because that place is occupied by the object under
scan. Coordinates of the actual detector placed further away from the source could
be obtained by scaling of (a, b). Based upon the above cone-beam geometry with the
planar detector, the reconstruction equation of FDK algorithm could be as given in
Eq. (1) [12]:
2π
1 1
f F D K (x, y, z) = P(β, a(x, y, β), b(x, y, β))dβ (1)
2 0 U (x, y, β)2
3 Implementation of Algorithm
The CPU algorithm involves steps where the projections are normalized first
converting the cone-beam projections into equivalent parallel beam data. These
projections are arranged in the form of sinograms. A sinogram is generated by
stacking of projections corresponding to a horizontal array of FPD pixels from all
angular orientations in a single file. Secondly, the projection is convolved with a high
pass filter function. In our case it is a Shepp-Logan filter. The Shepp-Logan filter
is a sinc function combined with a low pass ramp filter [11]. Thirdly, the filtered
data is back-projected. On a CPU these functions take place serially. The process is
diagrammatically represented in Fig. 1.
The GPU version requires defining the variables on the GPU memory for the
functions to access them. The sinogram is defined in single-point precision on the
4 R. Acharya et al.
Convolution
Normalising Projections Back-projection
CPU Processing GPU Processing
Sinogram
Data generation Volume
Acquisition Construction
Fig. 1 Block diagram of data flow representation for GPU-based processing [5]
GPU. Other variables like the filter function, number of angular projections, number
of samples per projection, source to object distance and other relevant geometrical
information pertaining to scanning are defined on GPU. The operations taking place
on GPU are done in parallel, e.g., typically the smearing back of data in back-
projection operation from all angles takes place in parallel to save computational
time.
4 Experimental Evaluation
The data was acquired for a cylindrical sample of aluminium consisting of a thin
cylindrical object with three fins inside. A multi-axis mechanical manipulator is
employed for data acquisition. The turntable of the manipulator is automatically
triggered using rotate-stop-acquire mode through a computer program. Thus during
the acquisition the sample is first moved to a pre-defined angular position using the
turntable. Once the manipulator reaches the position it sends a signal to the FPD
to acquire a radiographic projection. After successful completion of the acquisition,
the FPD sends a signal to the manipulator to move to the next angular position. This
entire cycle of events is automated through a central computerized program which
controls both the manipulator as well as the FPD. The same software application
also enables saving projection data to the desired location on the drive.
Data is acquired at 1 s of radiation exposure time from FPD. The angular interval
used is 1°. The size of the projection data matrix is 1367 × 1344 × 360 with 16-bit
data.
The reconstruction has been carried out on a workstation with Intel—CPU Xeon—
Gold 5118 CPU @ 2.30 GHz. RAM is 64 GB DDR-4. The GPU used is NVIDIA
Quadro P6000 with important specifications given in Table 1.
Evaluation of Parallel Processing Employed in Industrial … 5
Table 1 Technical
CUDA cores 3840
specifications of GPU
Peak single precision FP32 12.0 TFLOPS
performance
GPU memory 24 GB GDDR5X ECC
Memory bandwidth 432 GB/s
Memory interface 384-Bit
System interface PCI express 3.0 × 16
CT reconstruction was done on CPU and GPU using the FDK algorithm. Results
of the trial on cone-beam reconstruction of 1368 CT slices each of size 1344 ×
1344 have been recorded in Table 2. Here both the operations are done on GPU,
which shows speed improvement; however, the back-projection operation shows
significant time reduction, 74 times when compared to CPU. As observed, improve-
ment in total computation time is reduced due to time for transfer of data and only
a marginal improvement in convolution operation. However, the 53 times overall
speed improvement is still significant.
The graphical representation of computation time can be seen in Fig. 2. The time
axis is on a logarithmic scale in Fig. 2. This is done to show the difference in all
operations on a single graph. As anticipated, for higher resolution images and a large
number of CT slices GPU gives much better performance.
The total number of floating-point operations involved in the convolution and
back-projection operations may be considered as another means of evaluating the
enhancement in performance. From the data in Table 2, the convolution operation
is ignored because there is not a significant saving in time on GPU when compared
to back-projection. In our case, the total number of floating-point operations in a
back-projection operation involving 1344 × 1344 × 1367 CT images works out to
approximately 1,294,990,560 FLOP. Thus the speed achieved on CPU is approxi-
mately 46,285.62 FLOPS (46 KFLOPS), whereas that on GPU is 3473881.32 FLOPS
Table 2 Time comparison between CPU (without GPU) vs. CPU with GPU for low-resolution CT
image
Operation CPU Time (s) (a) CPU − GPU C PU T ime
C PU +G PU T ime Remark
performed Time (s) (b)
Transfer of data 0 28.735 – –
Convolution 125.783 119.24 1.0548 Faster by
5.48%
Back-projection 27978.247 372.779 75.0532 Faster by
7405.32%
Total 28104.03 520.754 53.9677 Faster by
5296.77%
6 R. Acharya et al.
100000
CPU-GPU Time comparison
10000
100
10
1
Transfer of data Convolution Back-projection Total
Fig. 2 Graphical representation of processing time on high-resolution CT images using CPU with
GPU and CPU without GPU
(3.5 MFLOPS). The reconstructed images from CPU and GPU are shown in Fig. 3.
As can be observed, there is no qualitative change in the image quality. The image
qualities are not explicitly compared quantitatively because in principle the same
programming routine has been employed in reconstruction operations with modi-
fications only in the way variables are called in individual functions. Thus the CT
images obtained may be considered identical.
A separate data set is acquired on gamma ray-based CT system with Co60 source
and 25 mm NaI (Tl) detector. The sample consists of four solid metal rods of different
sizes of which two are steel and two are aluminium. The diameter of the rods is
approximately 25, 32, 10, and 15 mm. A single slice of data has been acquired at an
angular interval of 1° each. The data covers 360 angular projections and 171 samples
per projection. Thus the size of projection data set is 171 × 360 with 8-bit data.
FBP algorithm is used for data reconstruction on CPU. For GPU implementation,
the algorithm has been tweaked by defining relevant variables in GPU memory
Fig. 3 CT image reconstructed on a CPU and b GPU both of 1344 × 1344 pixel size
Evaluation of Parallel Processing Employed in Industrial … 7
and carrying out convolution and back-projection operations on GPU. All other
operations have been done on the CPU. Here the CPU time is 0.0155719 s for
convolution, 0.1647011 s for back-projection and thus 0.180273 s in total which was
found to be much lower than CPU + GPU timings where the transfer of data took
0.0108601 s, convolution took 0.0470461 s and back-projection took 0.2601996 s,
thus total time for a single CT slice was 0.3181058 s. This clearly demonstrates the
effectiveness of the GPU only when confronted with large data sets.
6 Conclusion
The performance of NVIDIA Quadro P6000 GPU was evaluated using cone-beam
CT (3D) reconstruction using FDK algorithm. For a high-resolution cone-beam CT
case with 1367 slices each with 1344 × 1344 image size, there was approximately
5300% (53 times) increase in speed with CPU-GPU combination employed. Much
of the speed improvement was in back-projection operation (75 times), which is the
most computationally expensive task. In both cases the results show that there was
no change in image quality.
References
1. Feldkamp LA, Davis LC, Kress JW (1984) Practical cone-beam algorithm. J Opt Soc Am A
1(6):612
2. Liu D, Sumanaweera T (2005) GPU gems 2:765–784
3. Garcic A, Franchetti F, Püschel M, Moura JMF, Pipatsrisawat T (2005) Performance analysis
of the filtered backprojection image reconstruction algorithms. In: Proceedings.(ICASSP’05).
IEEE international conference on acoustics, speech, and signal processing, vol 5, pp v/153–
v/156
4. Song S Jin K (2013) FPGA-based forward and back-projection operators for tomographic
reconstruction. In: Medical imaging 2013: physics of medical imaging. International society
for optics and photonics, vol 8668, pp 866836
5. Ametova E, Kopmann A, Mirone A, Chilingaryan S (2019) Reviewing GPU architectures to
build efficient back projection for parallel geometries. J Real-Time Image Process 17(5):1331–
1373 (2020) https://doi.org/10.1007/s11554-019-00883-w
6. Blas JG, Abella M, Isaila F, Carretero J, Desco M (2014) Surfing the optimization space of a
multiple-GPU parallel implementation of a X-ray tomography reconstruction algorithm. J Syst
Softw 95:166–175
7. Biguri A et al (2020) Arbitrarily large tomography with iterative algorithms on multiple GPUs
using the TIGRE toolbox. J Parallel Distrib Comput
8. Flores LA, Vidal V, Mayo P, Rodenas F, Verdú G (2014) Parallel CT image reconstruction
based on GPUs. Radiat Phys Chem 95:247–250
9. Redler G, Epel B, Qian Y, Qiao HHZ (2015) Implementation of GPU-accelerated back
projection for EPR imaging. J Xray Sci Technol 23(4):423–433
8 R. Acharya et al.
10. Noël PB et al (2010) GPU-based cone beam computed tomography. Comput Methods Programs
Biomed 98(3):271–277
11. Kak AC, Slaney M (2001) Principles of computerized tomographic imaging. Society for
Industrial and Applied Mathematics
12. Tan B-z, Yang Q-g, Li Z-r, Wang C (2020) Filtering path variable FDK (v-FDK) reconstruction
algorithm for circular cone-beam CT. In: International conference on optical instruments and
technology: IRMMW-THz technologies and applications, vol 11441. International Society for
Optics and Photonics, vol 114410J, 2019
Towards Rapid, in Situ Monitoring
of Thermal Barrier Coating Degradation
Using Eddy Current NDE Technique
Abstract Aircraft components exposed to high temperature are usually coated with
an insulating layer to protect them from failure due to thermal fatigue. This thin layer,
known as thermal barrier coating (TBC), has to be monitored regularly in order
to check for its degradation and loss of thickness during operation. Inspection of
such components poses a challenge owing to the chemical composition, unevenness,
and porous nature of the coating. In this paper, the authors propose a methodology
for the feasible implementation of the eddy current technique in order to perform
a high-speed inspection of thermal barrier-coated components and evaluate their
condition. A set of reference specimen with predetermined TBC thickness values is
used for the trials. After preliminary simulation studies to select the optimum param-
eters, raster scanning is performed on the specimen using an experimental config-
uration assembled in-house. The results show a variation in TBC thickness within
each specimen, which are validated using measurements from the THz time-domain
spectroscopy (TDS) method. A thickness map of each specimen is generated after
cross-correlation. Furthermore, this eddy current-based approach is implemented for
quantitatively evaluating the condition of another batch of TBC samples that were
subjected to varying in-service conditions. With a robust setup and proper calibration
procedure, the proposed methodology has the potential to be implemented for the fast
and reliable inspection of aircraft engine components to ensure structural integrity.
1 Introduction
Thermal barrier coatings (TBCs) are widely used in aircraft engineering and gas
turbines for propulsion and power generation [1]. Gas turbine blades and flaps in
modern aircraft engines are exposed to high temperatures. Therefore, TBC is adopted
to ensure that they can work in high temperature and high strain environment by
adding the insulation between gas and the alloying body of the blades [2]. During
service of the system, thermal cyclic conditions cause microstructural changes inside
the TBC, resulting in the degradation of heat insulation through various modes such
as erosion and accelerated oxidation [3]. In addition, the local chemical composition
of the underlying protective layer also changes, especially in highly loaded areas
[4]. The residual stress concentration is followed by spallation of the TBC system,
which exposes the mechanical components to extremely high temperatures, leading
to thermal fatigue, which drastically reduces the life of the material. The performance
of the turbine blades significantly contributes to the thermal efficiency of the engine
while their degradation determines the operating period between overhauls [5]. Since
TBCs are used in the hot sections of the engine, it is practically difficult and uneco-
nomical to carry out destructive tests. Hence, it is imperative to non-destructively
evaluate the TBC to determine any delimitation caused by the degradation of the
TBC layer, in order to avoid failure and catastrophic accidents of gas turbines.
Owing to the inherent characteristics of TBCs, such as porosity and thinness, tradi-
tional NDT methods such as ultrasonic testing (UT) and thermal imaging present
certain limitations. UT can detect interfacial defects, but analyzing multiple signals
reflected from each of the layers comprising the TBC system is cumbersome [6].
Moreover, high frequencies are required in order to acquire distinct signals from the
TBC layer, whose thickness is of the order 75–150 μm [7]. This causes increased
attenuation due to a higher degree of scattering from the microstructure. Furthermore,
UT requires a couplant, and the distortion caused by the contact condition between
the transducer and TBC can increase the uncertainty of the measured signals [8].
The thermal imaging method can measure the surface temperature of the TBC using
an infrared camera [9]. Due to low emissivity of the coated material, this method
requires that the surface of the coating be painted black for uniform emissivity.
Hence, inspection is both costly and time-consuming [10]. Moreover, there is the
issue of difficulty in the removal of the paint due to porosity of the TBC surfaces.
In this paper, the authors describe an eddy current-based non-destructive evaluation
(NDE) technique for the inspection of superalloy specimens with a thermal barrier
coating, both in unloaded condition and afterload exposure at high temperatures.
Eddy current testing (ECT) is an inspection method utilized for the evaluation of
metallurgical variations, in addition to defect detection [11]. ECT offers numerous
advantages such as high sensitivity, contactless inspection, and versatility [12]. This
work focuses on the qualitative evaluation of degraded TBC specimens by accurate
determination of the thickness variation and surface mapping of the specimen area.
The remainder of the paper is organized as follows: Section 2 presents the theoretical
background regarding ECT and its implementation to TBC inspection. In Sect. 3,
Towards Rapid, in Situ Monitoring of Thermal Barrier Coating … 11
numerical modeling is conducted for determining the ECT parameters. The experi-
mental procedure is outlined in Sect. 4 and the results obtained from the inspection
are presented and analyzed in Sect. 5. The conclusions from the work are summarized
in Sect. 6.
2 Background
Thermal barrier coatings (TBCs) are ceramic coatings applied over the corro-
sion/oxidation metallic coating to insulate the blade substrate from the high-
temperature fuel mixture. A turbine blade coated with a TBC is a blade coated with
a TBC system. It usually consists of a superalloy substrate, a metallic bond coating,
and a ceramic topcoat/TBC [13]. TBCs are applied by means of vacuum plasma
spray (VPS), air plasma spray (APS), or more recently electron beam vapor phase
deposition (EBVPD) [5]. The coatings significantly vary in their physical, thermal,
and mechanical properties. The TBC has a thermal and electrical conductivity much
lower than that of the metal substrate and the bond coat. Eddy current testing (ECT)
is an inspection method usually applied to conductive materials. This method works
on the principle of Faraday’s law. The ECT coil in proximity of a conductive material
generates magnetic fields called proximity fields, and these fields coupling with the
material induce eddy currents to the surface of the material. Due to this, secondary
magnetic fields are generated inside the material and consequently, the eddy currents
penetrate inside the material. These two fields interact with each other, affecting the
reactance of the probe. This characteristic of the probe provides information such
as changes in electromagnetic properties of the material or discontinuities inside the
multi-layer [14]. The changes in impedance as a result of eddy current induction can
be determined by the following two electromagnetic equations:
Z= R02 + (ωL 0 )2 (1)
where Z and ∅ are the magnitude and phase angle of the impedance, respectively.
R0 is the resistance, ωL 0 is the inductive reactance, and ω is the angular frequency
(ω = 2π × frequency).
The coil impedance or induced voltage is different when the coil is placed on
different materials or the materials have any metallurgical change as compared to
the reference. This variation in induced voltage in different materials is conveniently
utilized for the quantitative characterization of microstructural features using eddy
current parameters, namely, magnitude and phase angle of impedance. Eddy current
concentrates near the surface adjacent to the coil and its strength coefficients decrease
with distance from the coil. The eddy current density decreases exponentially with
12 H. Dutta and K. Balasubramaniam
depth. This phenomenon is known as skin effect [15]. The skin depth or standard
depth of penetration of the eddy currents can be estimated using Eq. (3):
1
Skindepth, δ = √ (3)
2π f μσ
3 Modeling Approach
The eddy current testing of TBC samples is modeled in Comsol Multiphysics 5.3.
The TBC component is modeled as a three-layered structure as shown in Fig. 1. The
bottom layer is the substrate which is the conducting alloy. The middle layer is the
bond coating and the topmost layer is the thermal barrier coating. The ECT coil is
modeled as a cylindrical structure and is excited by a current of 1 A on a plane parallel
to the TBC component. The modeling parameters of the TBC component as well as
the coil are tabulated in Tables 1 and 2, respectively. The resistance and inductance
variation is observed with varying TBC thickness at three different frequencies: 350
kHz, 1 MHz, and 2 MHz.
The TBC component model is simulated at 350 kHz, 1 MHz, and 2 MHz frequencies.
The objective is to determine an optimum frequency for experimentation. The vari-
ation of resistance with TBC thickness for the three different frequencies is plotted
in Fig. 2a. Figure 2b shows the changes in inductance with variation in TBC thick-
ness. It can be inferred from the plots that the sensitivity of resistance increases
with the increase in frequency, whereas inductance sensitivity does not show appre-
ciable changes with the increase in frequency. The resistance and inductance values
converge for higher TBC thickness values and sensitivity to higher TBC thickness
decreases. However, as frequency increases, the depth of penetration decreases and
will result in higher experimental noise. Moreover, the TBC samples that needed to
be inspected had a maximum coating thickness of 130 μm and within that range of
thickness, a frequency of 350 kHz is significantly sensitive; hence a frequency in the
range of 350 kHz to 2 MHz is selected for conducting ECT on the available TBC
specimens.
4 Experimental Methodology
The TBC samples to be inspected were obtained from the Defence Metallurgical and
Research Laboratory (DMRL), Hyderabad. There are two sets of test specimen. The
first set consists of a pair of TBC-layered flap specimen: one in pristine condition
and the other subjected to high-temperature load for a duration of 1000 h. These are
referred to as ‘non-serviced’ and ‘serviced’ flap sample, respectively. The second set
Towards Rapid, in Situ Monitoring of Thermal Barrier Coating … 15
Fig. 3 ECT results for TBC thickness of serviced and non-serviced flap samples at 350 kHz
Fig. 4 ECT results for TBC thickness of standard samples at 350 kHz
Surface mapping is done for the four standard specimens: S1, S2, S3, and S4 with
reported average TBC thicknesses of 40, 70, 100, and 130 μm, respectively. The area
of each specimen is 77 mm × 78 mm.
The scans are conducted with an absolute ECT probe (Technofour Group)
by maintaining a minimum constant lift-off for all the samples. Similar to the
previous manual inspection, scans are performed for two cases: one by balancing the
impedance point with respect to air and the other by taking the reference impedance
Towards Rapid, in Situ Monitoring of Thermal Barrier Coating … 17
point with respect to the substrate. Change in impedance values from the corre-
sponding reference point is obtained along the entire area of the sample. The move-
ment of the probe over the sample is controlled by means of a three-axis manipu-
lator and motion controller (MID-7604, National Instruments). Digitization and data
acquisition are done by Eddyblu software (Technofour Group). The scan parameters
are listed in Table 3.
The impedance map for the entire area of the TBC specimen is shown in Fig. 6a,
b. Figure 6a is the impedance map obtained when the reference point is in the air
and Fig. 6b is the plot obtained when the reference point is in the substrate. The
scan results clearly depict the TBC thickness variation within each sample, thus
demonstrating the non-uniformity of the coating. As expected, plots in Fig. 6a, b are
complementary to each other since, as the TBC thickness increases, the impedance
will tend toward the impedance of air and will deviate away from that of the alloy
Fig. 6 Surface map of impedance with a reference point in a air and b substrate at 350 kHz
frequency
18 H. Dutta and K. Balasubramaniam
(substrate). In both the cases, a similar trend for impedance variation can be observed
for each of the samples, thus validating the experimental procedure.
The gradual increase in impedance value in Fig. 6a corresponds to the case where
the probe coil has a transition from air and is approaching toward the substrate, and
vice versa since the impedance is measured with the reference point in the air. As
a constant lift-off is maintained with respect to the top surface of the TBC sample,
this transition (impedance increase) indicates that the TBC is degraded in that area,
reducing the effective lift-off between the substrate and the probe. Stating it the other
way around, a decrease in the impedance value depicts that the thickness of the TBC
is higher in that region.
On the other hand, an increase in impedance in Fig. 6b represents the transition
of probe coil from the substrate and approaching toward the air, since the impedance
is measured with the reference point on the substrate. In this case, higher impedance
indicates a thicker TBC as compared to the region where the impedance value is low.
The low impedance values at the edges of the samples in Fig. 6b and high values in
Fig. 6b depict that the probe is at the boundary between the sample and the air.
The impedance values obtained from the surface scans of the TBC samples were
correlated to the actual thickness values of the top coating. This has been done by
fitting an equation showing the relation between the TBC thickness (in microns)
and impedance (in Ohms). The curve has been fitted using the data obtained from
the simulation studies (Fig. 7). The actual TBC thickness maps obtained for the
scanning using three different frequencies (350 kHz, 1 MHz, and 2 MHz) have been
demonstrated in Fig. 8a, b, and c, respectively.
Fig. 7 Curve fitting used for mapping impedance to thickness (350 kHz)
Towards Rapid, in Situ Monitoring of Thermal Barrier Coating … 19
Fig. 8 TBC thickness map for a frequency of a 350 kHz, b 1 MHz, and c 2 MHz
6 Conclusions
This work primarily focuses on the qualitative evaluation of the thickness of the
non-conductive thermal barrier coating layer on an alloy substrate. This is required
to understand the degradation of the TBC layer, which acts as a protective coating
in several industrial components. It is necessary to inspect the thickness of the TBC
and detect flaws developed in it regularly in order to maintain the structural integrity
of these components.
The principle of eddy current non-destructive method was used to compare the
thickness of TBC in different samples. Change in impedance as a result of eddy
current generation was used to qualitatively evaluate the layer thickness. While
20 H. Dutta and K. Balasubramaniam
inspecting the TBC samples, it was observed that there were drastic variations in
the TBC thickness within the same sample. Hence, eddy current-based scan was
performed on the specimens to determine the thickness profile of the thermal barrier
coating.
Surface maps of impedance for the entire area of the test specimen demonstrated
the variations in TBC thickness in each of the samples. Thickness variations between
each of the specimens were inferred from the changes along the impedance scale.
Furthermore, quantitative analysis of the TBC thickness can be determined from the
impedance to thickness conversion as demonstrated in the manuscript.
Acknowledgements This study was supported by Defense Metallurgical Research Lab (DMRL)
of Defense Research and Development Organisation (DRDO), India. The authors would like to
thank Dr. Sreedhar Unnikrishnakurup for helping us with the THz TDS measurements.
References
1. Khan AN, Khan S, Ali F, Iqbal M (2009) Evaluation of zro2–24mgo ceramic coating by eddy
current method. Comput Mater Sci 44(3):1007–1012
2. Yong L, Chen Z, Mao Y, Yong Q (2012) Quantitative evaluation of thermal barrier coating
based on eddy current technique. NDT e Int 50:29–35
3. Kim JM, Park J-H, Lee HG, Kim H-J, Song S-J, Seok C-S, Lee Y-Z (2018) An effective
electrical resonance-based method to detect delamination in thermal barrier coating. J Therm
Spray Technol 27(3):336–343
4. Bruchwald O, Frackowiak W, Reimche W, Maier HJ (2016) Applications of high frequency
eddy current technology for material characterization of thin coatings. J Mater Sci Eng A
6(7–8):185–191
5. Crowther P (2004) Non-destructive evaluation of coatings for land-based gas turbines using a
multi-frequency eddy current technique. Insight-Nondestr Test Condition Monit 46(9):547–549
6. Crutzen H, Lakestani F, Nicholls J (1996) Ultrasonic characterization of thermal barrier
coatings. In: 1996 IEEE ultrasonics symposium proceedings, vol 1, pp 731–734. IEEE
7. Zhang B (2010) Nondestructive evaluation of thermal barrier coating systems using ultrasonic
techniques. West Virginia University
8. Ma Z, Zhao Y, Luo Z, Lin L (2014) Ultrasonic characterization of thermally grown oxide in
thermal barrier coating by reflection coefficient amplitude spectrum. Ultrasonics 54(4):1005–
1009
9. Eldridge JI, Bencic TJ, Allison SW, Beshears DL (2004) Depth-penetrating temperature
measurements of thermal barrier coatings incorporating thermographic phosphors. J Therm
Spray Technol 13(1):44–50
10. Ptaszek G, Cawley P, Almond D, Pickering S (2013) Transient thermography testing of
unpainted thermal barrier coating (TBC) systems. NDT e Int 59:48–56
11. Khan S, Ali F, Khan AN, Iqbal M (2008) Pearlite determination in plain carbon steel by eddy
current method. J Mater Process Technol 200(1–3):316–318
12. Auld B, Moulder J (1999) Review of advances in quantitative eddy current nondestructive
evaluation. J Nondestr Eval 18(1):3–36
13. Bernstein HL, Allen J (1992) A review of high temperature coatings for combustion turbine
blades. In: Steam and combustion turbine blading conference
14. Bray DE, McBride D (1992) Nondestructive testing techniques. NASA STI/Recon technical
report A, vol 93
Towards Rapid, in Situ Monitoring of Thermal Barrier Coating … 21
Abstract The present study discusses the potential of eddy current thermography
in the detection of edge defects in metal plates. The focus is on detecting the defect’s
dimension and location for the defects available at the edge of metal plates. We have
used a slab of SA387 GR12 (Killed steel) for performing our study. SA387 GR12
(Killed steel) possesses tremendous tensile and yield strength. The combination of
chromium-molybdenum provides corrosion-combating skills along with improved
tensile strength at higher temperatures. It has wide applications in the field of power
(especially thermal power plants), oil and chemical industry, construction sector etc.
as heat exchangers, pressure vessels, industrial boilers, piping industry and in piping
support structures, sour service environments etc. Small cracks in any component
can cause catastrophic failures. So it becomes a necessity to detect the crack in the
initial stage itself, especially to avoid stress concentration in processes like rolling of
steel plates. The study is performed on a SA387 GR12 sample with defects created
at various angles. Similar cracks can also be found in rails, created because of rolled
contact fatigue. Hence the study is going to be useful for the detection of defects in
rails also. ECT offers further possibilities in the area of automation and has various
advantages in comparison to conventional methods and the same is also discussed
in the work.
N. Kumar (B)
Bharat Heavy Electricals Ltd, Ranipet, Tamilnadu, India
e-mail: nakulkumar@bhel.in
N. Kumar · R. J. Joseyphus
National Institute of Technology, Tiruchirappalli, Tamilnadu, India
e-mail: rjustinj@nitt.edu
1 Introduction
that the thermal contrast decreases with an increase in heating time, so the heating
time shall be kept least. Wilson et al. [10] did their work for imaging multiple cracks
from rolling contact fatigue in the edges of rails for ‘Gauge Corner Cracking’ (GCC)
and ‘Head Checking’ (HC). These edge cracks will cause a temperature distribution
along the edge which can be captured by the thermographic camera for defects
detection. These studies basically focused on finding the detectability of defects.
In this work, detection of the defect’s depth and size will be presented using the
eddy current thermography. In this study, we have worked upon A387 GR12 plate
(killed steel). It has a wide area of applications in various industries. The experiment
discusses the inspection and finding the dimension and location of angular edge
defects using eddy current thermography.
The ASME SA387 standard covers the availability of weldable chrome molybdenum
alloy steel plates for pressure vessels applications at elevated temperature [ASTM
standard]. Molybdenum and chromium are used as main alloying materials which
provide excellent resistance against corrosion, especially at high-temperature [ASME
BPVC.II.A]. For this study, the A387 GR12CL2 sample is used. As per the ASME
standard, it is engineered for use at elevated temperature service with applications
in weldable pressure vessels and industrial boilers. In power plants, high thickness
plates (thickness varying from 100 to 220 mm) are needed for weldable pressure
vessels while low thickness plates are utilized in supporting structures for superheater,
reheater, DeSH etc. In the petrochemical industry, inside the reactor, drums etc., it
is used as drum internals like internal support clips or brackets. Chromium as alloy
material provides excellent resistance for corrosion and oxidation which makes it
ideal for sour service applications in the oil and gas industry. ASME BPVC code
[11] specify the tensile requirements as (Table 1).
ECT is based on the concept of heating the sample/component using eddy currents
and noticing the thermal variation alongside the crack for defect detection. It uses
a high power and frequency coil for induction heating of the component and uses
a high-resolution infrared camera for observing the thermal response. This tech-
nique can be used in line with other thermographic evaluation techniques such as
laser spot thermography [1] and sonic thermography. Any open or closed cracks or
surface/subsurface flaws will respond with a thermal gradient around itself.
26 N. Kumar and R. J. Joseyphus
Induction Heating: Induction heating being an efficient, precise and fast non-
contact method for heating electroconductive materials is largely employed in manu-
facturing industries, especially for pre-heating and post-heating processes during
welding. It depends on the principle of electromagnetic induction and a coil that is
suitably dimensioned and conducts high or medium-frequency alternating current
is placed near the metal component to be heated. This induces eddy currents in the
material which cause heating of the specimen. This heat is distributed to the rest of
the material by conduction. Induction heating thus doesn’t need any physical contact
and is a very efficient way of heating without much loss of energy. Induction heating
is a combination of three principles:
(a) Electromagnetic induction
(b) Joule heating
(c) Thermal conduction.
(a) Electromagnetic induction: Faraday’s law of electromagnetic induction states
that an emf is induced in a coil when magnetic flux through the coil changes with
time and the direction of induced emf is given by Lenz law, that is,
ε = −d/dt (1)
This change of flux may be because of various reasons like relative movement
between the secondary and the primary coil (generating magnetic flux) or using a
time-varying magnetic flux (e.g. induction heaters).
Eddy currents: When a bulk piece of the conductor is placed in the vicinity of
changing magnetic flux, induced currents are produced in it. As their flow patterns
resemble swirling eddies in water they are so-called eddy currents. The effect was
discovered by physicist Foucault and these currents are also called Foucault currents.
The eddy currents intensity is specified as
Here, isurface is the current at the surface, z is a coordinate for depth, c is a constant,
μ is the magnetic permeability of the material and σ is the conductivity of the material
(Fig. 1).
As the magnetic field is penetrating the specimen, the eddy currents are going
to be concentric circles perpendicular to the magnetic flux, as shown in Fig. 2.
The temperature distribution along the depth will also be a similar profile as on
the surface of the specimen with intensity decreasing exponentially with depth. As
the current density depends on material properties like magnetic permeability and
conductivity of the material, thus the temperature profile is going to be a signature
profile for a particular material at a specific frequency and applied field for a given
time. The presence of any flaw/crack/defect in the specimen will disturb the flow
of eddy currents, so the temperature profile will change for the specimen and thus
the signature temperature profile is distorted. By analysing this distortion in the
temperature profile, we will be able to locate the defect and find out its size, as
explained in the subsequent portion of this chapter.
(b) Joule heating: Joule heating or resistance heating is the generation of heat when
an electric current is passed through a conductor. Joule’s first law states that the heat
for the defect area and the rest of the specimen create the thermal gradient around
the defect area and thus becomes the base of the thermal inspection.
Infrared Thermography: Infrared thermography is based on the detection of
infrared energy emitted by an object, converting it to temperature, and displaying
the image of temperature distribution. On the basis of the source for heating the
specimen or the method of establishing the heat flow, thermography can be divided
into two sections, i.e. passive thermography and active thermography.
Active thermography requires an external stimulus to create the necessary temper-
ature gradient between defective and non-defective zone. So, thermography can
be further categorized on the basis of excitation mode and heat source. Source of
stimulus can be:
• Optical excitation
• Electromagnetic excitation
• Mechanical excitation.
Electromagnetic excitation is being used in our experiment.
3.1 Specimen
A plate of Cr–Mo steel of grade SA387Gr12Cl2 is used as a specimen for the study.
Three different samples having dimensions 75 mm × 60 mm × 10 mm were cut from
this plate. Now to make it similar with a real-time edge defect, angular slots with
dimension 7 ± 0.2 mm × 1.25 0.25 mm are created using saw-cutting throughout
60 mm width at angles of 90°, 60° and 30° with constant slot length and thickness,
in all three samples, respectively. The surface of all specimens was painted black to
avoid reflection. The schematic diagram of the specimen is as shown in Fig 4.
Indutech makes induction heaters with a maximum power output of 1 kW, which are
used for induction heating of the specimen. The electrical energy is applied through a
copper coil which produces a precisely controlled magnetic field over the specimen.
The operating frequency range of the system is 500 kHz.
Coil design: The coil used in the experiment is shown in Fig. 5.
• Material: Copper
• Geometry: Helical (with a constant base diameter and 4 number of turns)
• Internal diameter: 42 ± 2 mm and OD: 50 ± 2 mm
30 N. Kumar and R. J. Joseyphus
The schematic diagram for the experiment is shown in Fig. 7 and the actual exper-
imental setup is displayed in Fig. 6. The copper induction coil will induce eddy
currents in the specimen. These eddy currents will try to take the shortest closed
path which will create uneven heating around the flawed region. This uneven heating
will be captured by the IR camera and this image can be processed further for defect
detection.
The coolant system was used for the coil to avoid overheating. IR camera was cali-
brated accordingly. The setting was done at full power of 1 kW for induction heater
so as to deliver a current of 400 A. Heating tripped after 1 s and the sample was
Eddy Current Thermography as a Tool for Detecting the Location … 31
observed for a time of 10 s. In theory, the excitation signal and the thermal response
is supposed to be like as in Fig. 8.
Now the thermograms recorded for different angles are shown in Fig. 9. With the
thermograms, it is evident that where the eddy currents are rushed/trapped more,
there the indication is strong enough to provide a thermo-visible indication of the
defect. It can be easily seen that if the angle of trapment (At ) or slot angle is less
Fig. 9 Thermograms at 300 ms for (clockwise) 1: 30° slot angle; 2: 60° slot angle; 3: 90° slot angle
(like defect angle of 30°), the temperature rises at the marked area, as indicated in
Fig. 10, will show a sudden spike due to concentrated eddy currents. For the larger
slot angles like 60° and 90°, the entrapped area is shifted to the bottom, as shown
in Fig. 10, and thus the thermal indication of a defect will be strong in transmission
mode.
During the cooling phase, the heat generated in the entrapped zone is going to be
diffused to the neighbouring area causing a temperature rise in the neighbouring area
during the cooling phase as well. The heating zone will also vary with the frequency
and standard depth of penetration (Oswald-Tranta and Wally [7]).
Now an edge defect can be defined having dimensions ‘a’ and ‘b’ as shown in
Fig. 11. Now, if a is greater than b then it will be like a shallow defect and more
eddy currents will rush under the defect and heating in this zone is going to be more
(Yang et al. [9]).
Fig. 10 Left: A trapped area for less slot angle (30°). Right: A trapped area for more slot angle
(60° and 90°)
As for our experiment, we have created a defect throughout the width of the
sample, as shown in Fig. 12; therefore all the eddy currents are being forcefully
diverted under the defect. Because of this the defects are deeper than the standard
depth of penetration and are also going to be visible from the side view. The same is
evident from the thermograms shown in Fig. 9.
Temperature profiles for side face
If the temperature profile is taken along a vertical line drawn on the side face of a
defect-free sample, then this profile is going to act as a reference for defect detection
which is explained in Sect. 2.2.1 (a). The reason for this is that the presence of a
defect/flaw in the specimen will distort the path of eddy currents and so the intensity
of eddy current will get disturbed in the defective region causing a deviation in the
temperature profile of the region from that of standard profile which will form the
basis of our defect detection.
As shown in Fig. 13, vertical lines are drawn to take the temperature profiles.
The temperature data was taken at 0.70 s of heating. The temperature profile for the
defect-free specimen will appear as shown in Fig. 14. This temperature profile is
going to be unique at a particular heating time and thus can be used as a reference
profile for studying the specimen for defects.
Now for studying the specimen for edge defects, we are going to analyse the
temperature profiles along the side face for all slot angles. For this, vertical lines are
drawn on the side face as shown in Fig. 13 and the temperature profile is recorded
along these lines.
Slot Angle 30°: The temperature profile along the reference vertical lines is anal-
ysed. Due to the presence of any defect in the specimen, the thermal profile is going to
be distorted as the path of eddy currents is disturbed. This will provide an indication
of the presence of a defect.
To study it, five vertical lines such as L1, L2, L3, L4 and L5 are drawn in the
defective zone for 30° slot angle at 0 mm, 1.5 mm, 3 mm, 5 mm and 7 mm distance
from the tip of the left edge, respectively. L0 denotes the vertical temperature profile
for the defect-free region. This is projected in Fig. 13.
In Fig. 15, the temperature profile along the reference line drawn in the defect-
free zone, which is going to act as a reference profile (i.e., along vertical line L0),
indicates the decrease of temperature with depth which is supposed to be a natural
profile for as eddy current intensity decreases exponentially with depth. Now coming
Fig. 15 Temperature profile along line L0 against lines L1 and L5—30° slot angle
Eddy Current Thermography as a Tool for Detecting the Location … 35
to temperatures profiles for the defective zone (defect angle 30°), the temperature
profile along line L5 shows that the temperature will be more in the zone where
trapped eddy current is more comparatively to the temperature in the defect-free
zone. Further to this, as shown in Fig. 9 (thermogram for 30º) the heat will diffuse
from the tip of the defect to the neighbouring area, thus the temperature slope for the
point farther from the defect mouth (i.e., Line L5 is leftmost and farthest from defect
mouth) is relatively less in the trapped zone as depicted by curve from O to A. We
see a sudden drop in temperature at point A which shows deviation from the profile
of a defect-free zone. This depicts the presence of a zone with material properties
different from the parent material, i.e., a crack open to the surface in this case and
the point A marks the beginning of the crack opening. Later on, we see a sudden
spike in temperature to point B and this marks the other end of the crack opening. So
the horizontal/abscissa distance between the two points A and B defines the width of
the crack and O to A distance will give us the depth of the defect from the surface
(Fig. 16).
Similarly, for line L1, as the line is drawn at the tip of the defect, the temperature
is very much more at the surface than the temperature in the defect-free zone. We
see a sudden drop in temperature at point A’ indicating the beginning of the crack.
Further to this, we again see a temperature rise to point B’ indicating the other end
of the defect. Thus the distance from A’ to B’ denotes the width/depth of the defect.
In a similar fashion, starting and endpoint for the defect can also be located along
lines L2, L3 and L4.
Surface temperature comparison for lines L2, L3 and L4 is as temperature for
L2 > L3 > L4 (The starting point of curves (at points O and O ) indicates the surface
Fig. 16 Temperature profile along line L0 against lines L2, L3 and L4—30° slot angle
36 N. Kumar and R. J. Joseyphus
temperature). This is due to the reason that for a very steep slot angle, the current
density will be more for the points nearer to the tip of the defect (i.e., inside edge).
For 60°: Similarly, four lines are drawn in the defective zone for 60° slot angle,
also named L1, L2, L3 and L4 at a distance of 0 mm, 1.5 mm, 3 mm and 5mm from
the inside tip of the edge defect, respectively. L0 denotes the vertical temperature
profile for the defect-free zone.
It is evident from Fig. 17 that, in contrast to the case of 30° slot angle as shown
in Fig. 15, the temperature difference at the surface for lines L1, L2 and L0, i.e., at
starting point of profile at point O is not meagre. Lesser trapped eddy currents due
to slant slot angle in the zone can be a reason here.
In Fig. 18, the temperature profile for lines L3, L4 and L0 is almost the same up to
point A. Here onwards to point A, the temperature rise is very gradual and not sudden
for L3. This shows that line L3 may not cross the defect but can be in the vicinity
of the same and highlights the presence of a nearby defect. Also, as the temperature
decrease initially then we see a gradual increase later at a certain depth, and this
point towards the presence of a nearby angular/horizontal defect.
For line L4 in Fig. 18, the profile traced is similar to the profile of L0 line with
a little less slope and more temperature. This can be due to heat diffusion from a
nearby trapped heat zone as shown in Fig. 19.
For 90°: Contradictory to the temperature profile for a defect-free sample, line L1
graph shows a very less temperature at the surface which indicates an open-to-surface
defect. Temperature increases gradually with depth and reaches a point of maximum
Fig. 17 Temperature profile along line L0 against line L1 and L2—60° slot angle
Eddy Current Thermography as a Tool for Detecting the Location … 37
Fig. 18 Temperature profile along line L0 against lines L3 and L4—60° slot angle
temperature and then decreases suddenly. This provides a hint that the defect may
be a vertical open to surface defect, depth of which is measured by abscissa of OA.
For lines L2 and L3, the vertical temperature profile is similar up to point Z followed
by a gradual increase in temperature. This indicates the presence of a defect in the
vicinity (Fig. 20).
Some important points which can be drawn from the above graphical comparisons
are:
• In all the above graphs: Points A, A , A mark the beginning of the crack.
• Points B, B , B mark the end of the crack opening.
• Distance AB, A B , A B shows the height of the crack.
• Distance OA, OA , OA shows the depth of the crack.
• If there is no point B/B then it shows point A is just in the vicinity of the crack.
Table 2 is made in accordance with the above data, comparing the theoretical
values with the experimental values for defect depth and height.
38 N. Kumar and R. J. Joseyphus
Fig. 20 Temperature profile along line L0 against lines L1, L2 and L3—90° slot angle
Table 2 Experimental values vs geometrical values of defect depth and defect height
Slot angle Location The Geometrical value Experimental value
(degrees) (line drawn distance of The depth Height/width The depth Height/width
for taking the line of defect of defect of defect of defect
temp from defect from the (mm) from the (mm)
profile) tip (mm) surface surface
(mm) (mm)
30 L1 0.00 0.00 1.44 0.00 1.50
L2 1.50 0.87 1.44 1.00 1.25
L3 3.00 1.73 1.44 1.50 1.25
L4 5.00 2.890 1.44 2.50 1.25
L5 7.00 4.040 1.44 3.50 1.25
60 L1 0.00 0.00 2.50 0.25 2.75
L2 1.50 2.60 2.50 3.75 2.75
L3 3.00 No defect No defect – –
L4 5.00 No defect No defect – –
90 L1 0.00 7.00 0.00 6.75
L2 1.50 No defect No defect – –
L3 3.00 No defect No defect – –
Eddy Current Thermography as a Tool for Detecting the Location … 39
Table 2 shows that the experimental values for defect height and depth are
matching with the theoretical values with a tolerance of 0.25 mm which can be
rectified by different image processing tools or using phase analysis [3].
5 Automated ECT
The possibilities for automation of the steps involved in ECT inspection makes
thermography an interesting alternative to other manual inspection techniques such as
fluorescent penetration inspection (FPI) [12]. Automation of ECT by doing robotics
in handling induction coils and components was discussed by Vrana et al. [6] and
Netzelmann et al. [13]. Standardization of induction thermography and automated
inspection of rails and train wheels for surface defects were also discussed. Though
automation by doing the automated movements of components or induction coils is
being discussed so far for the ease of inspection, exploring the area of automated
final decision-making after inspection is going to be more interesting. A reference
criterion similar to GO/NO-GO gauges can be set so as to establish an accept-reject
criterion. Basically, it is going to ask for a reference sample/charts/table similar to
other NDE techniques. The theory in this direction can be explained as hereunder.
With the help of other NDE techniques and our own past experience, a Refer-
ence/Trueflaw sample set can be prepared for a specific product or a production line.
This Trueflaw sample set shall have all the possible defects for that production line
based on our experience. Now, this Trueflaw sample set is going to act as the limit
for the accept-reject criterion for our production line. Thermograms can be obtained
by performing the eddy current thermography study for this sample set. These ther-
mograms are going to act as the reference sets for setting the accept-reject criterion.
From these thermograms (as explained earlier in this paper) temperature graphs,
amplitude and phase images can be generated. The Accept-reject criterion can be
set for the production line in accordance with amplitude and phase images [3]. With
the current advancement in image processing and software development, maximum
available readings with us can be linked/fed in the image processing software for
the standard specimen. Now, all the components manufactured in the production
line having the thermal readings, measured by a thermal camera, is not within the
set limits. Those components can be marked for further evaluation or straightaway
rejected. In this fashion, the complete production line, including inspection decision,
can also be automated.
References
1 Introduction
introduced into the market as a non-destructive inspection (NDI) tool [4]. In India,
especially in the automobile, aerospace and defense sector, a few such systems are
currently in use for regular NDT purpose. The price of these commercial shearog-
raphy NDT systems is significantly high. Therefore, Indian medium or small-scale
industries have difficulty in implementing such systems into their production line for
quality control or NDT purpose. Toward this, a shearography system development
was initiated by the authors at Vikram Sarabhai Space Center, ISRO, aiming at devel-
oping a cost-effective portable shearography system with an area of coverage up to 1
square meter. The system is mainly targeted for in situ and online NDT of metallic and
composite structures including adhesively bonded sandwich metallic/composites.
When an object surface, with the surface roughness of one order higher than the laser
wavelength, is illuminated with a coherent laser, grainy speckle pattern is formed
on the surface. When this speckle pattern is imaged through an optical lens, then it
is called a subjective speckle pattern. Such grainy patterns can be recorded in the
image plane of a digital camera. The speckle pattern generated on the object surface
is due to the interference of the diffusively scattering light from neighboring points
on its surface. The interference at a point carries random phase and amplitudes due to
the random nature of scattering. The size of the subjective speckle pattern primarily
depends on the (i) aperture of the imaging system, (ii) magnitude of surface roughness
and (iii) the distance between the object and imaging system among other factors.
Shearography is based on the principle of speckle interferometry where a reference
speckle pattern is made to interfere with another speckle pattern imaged with a slight
spatial shift. Therefore, two neighboring points on the object are made to interfere at
the image plane of a digital sensor through a shearing optics (Fig. 1). Figure 1 shows
an object illuminated with a laser source. The shearing optics as shown in Fig. 1
is a Michelson’s interferometer, wherein two images are formed through tilting the
Fig. 2 a Specklegram of a disc-shaped object after the interference of two speckle patterns and b
shearogram of the same object obtained after digital subtraction of a reference specklegram from a
deformed specklegram. Here, the shear amount is given in the diagonal direction
mirror in one of the arms of this interferometer. The angle of tilt of the mirror
creates the lateral shift called the shear amount. The interference of the two speckle
patterns at the image sensor is called a specklegram (Fig. 2a). Likewise, a deformed
specklegram can be recorded in the image sensor after loading the object. Point to
point digital subtraction of the two specklegrams would modulate the object intensity
with fringe patterns. These fringe patterns are the loci of the surface displacement
derivatives and this modulated intensity map is called shearogram (Fig. 2b).
where ()∗ represents the complex conjugate operation. Equation (1) after simplifica-
tion provides the following expression for Iu :
where δϕ(x, y) = φ2 − φ1 is the random interference phase between the two speckle
patterns, A = (a12 + a22 ) and B = 2a1 a2 , respectively.
44 D. Swain et al.
Let be the relative phase change between two neighboring points due to the
relative displacement of these two points due to loading. Considering I L as the
intensity of the specklegram of the object after loading, I L can be written as [3]
The relative phase difference for normal illumination and viewing direction
with horizontal lateral shift is given as [3]
4π dw
= δx (5)
λ dx
where dw/d x is derivative of the out of plane displacement along x-direction and
λ is the wavelength of laser source used. When /2 in sin(/2) becomes integral
multiple of 2π , the intensity becomes zero resulting in black fringes. Herein, only x-
direction shearing is illustrated for simplicity. However, such shearing in y-direction
or a combination of x and y can be provided. The shearogram in Fig. 2b shows such
a combined shearing.
The size of the object to be imaged using the optical setup shown in Fig. 2 is limited
by the size of the optical elements and the imaging lens. Hence to increase the
coverage area Michelson’s interferometer is modified with 4-f imaging optics [6],
as shown in Fig. 3. Here the imaging lens and image sensor are placed separately,
Fig. 3 Modified
Michelson’s interferometer
with 4-f imaging optics
Indian Make Portable and Real-Time Shearography System (IMPRESS) … 45
unlike the conventional digital shearography system. The imaging lens is the first
optical element in this modified setup, whereas the beam splitter is the first element in
the conventional Michelson’s setup. The imaging lens focuses the object at its focal
plane and the first 4-f imaging lens is kept at the focal plane of this imaging lens. The
second 4-f imaging lens is kept at 2-f distance from the first one as shown in Fig. 3.
Finally, the CCD/image sensor is kept at the focal plane of the second 4-f imaging
lens. Thus large objects can be viewed through this setup. The angle of view can be
further increased by replacing the lens with a smaller focal length. Moreover, this
setup facilitates the use of small beam splitter and mirrors. Hence, it can be utilized
for developing miniaturized systems.
3 Development of IMPRESS
We have initially developed a bread-board model for checking the shearograms gener-
ated from a trial 4-f setup. During this period, a LabVIEW code was developed for
visualizing the shearograms. Subsequently, a portable optical head was configured
and fabricated for in situ and portable utility. This section provides these details.
A bread-board setup using Michelson’s interferometer with 4-f imaging optics was
made on a vibration isolation table. Trial experiments were carried out as shown in
the setup in Fig. 4a. Here the imaging lens was integrated with the image sensor. A
6 W VERDI laser at a maximum power level of 1 W was used in all the experiments
with an optical fiber for guiding the beam. The resulting real-time shearogram for a
circular disc loaded at the center is shown in Fig. 4b.
Fig. 4 a Bread-board model 4-f optical setup for trial experiments and b resulting shearogram for
a disc loaded at the center
46 D. Swain et al.
A front-end software was developed in LabVIEW platform to (i) acquire the speck-
legram in real time, (ii) store reference specklegram at any stress state, (iii) generate
shearogram through pixel to pixel digital subtraction between the reference speckle-
gram and (iv) recording of live specklegram at current stress state. The front screen of
the software is shown in Fig. 5, where each window respectively displays live speck-
legram, chosen reference image and shearogram in real time. The reference image
can be changed in real time to obtain good quality shearograms. The software was
developed to acquire specklegrams using PixeLINK CMOS camera. The feature for
selecting any region of interest (ROI) and controls for acquisition speed and exposure
time are provided in the software. A full frame size of 3000 × 2208 pixels at 5 fps
can be recorded. The outputs shearograms can be saved in image format and video
formats.
A compact unit was later designed to accommodate 4-f imaging lenses, mirror with
tilt control on a translation stage, beam splitter plate, imaging lens and image sensor.
An enclosure was fabricated for assembling all the optical elements inside it as shown
Indian Make Portable and Real-Time Shearography System (IMPRESS) … 47
in Fig. 6a. A provision was made in the system to replace the fixed plane mirrors in
the two arms of Michelson’s interferometer with piezo actuated mirrors. One piezo
mirror is for computerized control for the shear amount and the other one is for
phase shifting of one arm at λ/4 steps. Therefore, four phase-shifted shearograms
can be recorded to generate a phase map for quantifying displacement derivative.
The shearing camera as mounted on a tripod is shown in Fig. 6b, which was used for
in situ tests.
In this section, salient results obtained from the fabricated prototype would be
presented to demonstrate IMPRESS. Initially, we present the results obtained with
vibration testing followed by the results obtained using thermal stressing.
Fig. 7 a Honeycomb specimen, b experimental setup with piezo shaker and c shearogram of the
defects during vibration stressing
Fig. 8 Programmed defects on a honeycomb sandwich panel of different sizes identified at different
frequencies
almost eight defects could be identified. Only the smallest defect could not be
identified with the setup used. The system response and sensitivity need further
improvement to detect such defects.
tile specimen also could be obtained by heating a sample using an infrared light
source (Fig. 9d). A skin debond in a honeycomb sandwich panel with carbon epoxy
skin sheets also could be identified under thermal stressing (Fig. 9c). The object in
Fig. 9b is a conical satellite adapter made of metallic honeycomb sandwich with thin
aluminum face sheets of size around 0.5 m with programmed face sheet debonds.
The skin debond also could be identified under minimal thermal stressing. In all of
the above examples, the region, size and shape of the defects are clearly identifiable
from the fringe patterns. Since fringes are confined to only defective regions, defect
identification is very easy compared to holography technique.
5 Concluding Remarks
the area of coverage. The resulting system with its software is named the Indian
Make Portable and Real-time Shearography System (IMPRESS). The inclusion of
4-f optics enhanced the capability of IMPRESS to image 1 square meter objects for
identifying defects. However, depending upon the sensitivity and size of the defect,
the area of coverage must be optimized during in situ tests. At this stage, the system
is tuned to work under mechanical, vibration and thermal stressing methods. This
system can be used with vacuum stressing also. The NDT results obtained are very
encouraging. A lot of expertise is gained through the development of the system and
the technological know-how. The current capability of the system would be sufficient
for NDT of various engineering components. Moreover, the system can be explored
further for various defects in metallic and composite structures such as weld defects,
brazing anomalies and other honeycomb defects. Furthermore, the system can be
used for investigating the structural integrity of pressure vessels, thermal protection
systems and other engineering components. Further work to automate the system is
in progress so that sophistications at par with commercial systems can be built into
the system.
References
1. Hung YY (1982) Shearography a new optical method for strain measurement and non-destructive
testing. Opt Eng 21(3):213391
2. Steinchen W, Yang LX, Schuth M (1996) TV-shearography for measuring 3D-strains. Strain
32(2):49–58
3. Steinchen W, Yang LX (2003) Digital shearography: theory and application of digital speckle
pattern shearing interferometry. In: SPIE Press monograph Vol PM100, SPIE Press, Bellingham,
Washington, USA
4. ASTM standard E 2581-07 (2007) Standard practice for shearography of polymer matrix
composites, sandwich core materials and filament wound pressure vessels in aerospace
applications
5. Yang LX, Xie X (2016) Digital shearography: new developments and applications. In: SPIE
Press monograph Vol PM267, SPIE Press, Bellingham, Washington, USA
6. Wu S, He X, Yang LX (2011) Enlarging the angle of view in Michelson interferometer based
shearography by embedding a 4f system. Appl Opt 50(21):3789–3794
Real-Time Detection of Defects
on a Honeycomb Composite Sandwich
Structure Using Digital Image
Correlation (DIC)
Abstract Digital image correlation (DIC) has been regularly used as a quantitative
structural integrity evaluation tool. Gradually, it is emerging as a quantitative NDE
(QNDE) tool for many structural engineering applications. Earlier the authors have
reported real-time NDE of metallic structures using DIC which detected defects
under continuous loading. This paper reports a unique application of DIC wherein
debonds on a honeycomb-composite sandwich structure could be detected online
and in situ during ground static structural testing with intermittent loading steps.
Defect response in terms of local displacement and strain fields at one location on the
hardware could be obtained both online and offline as well. However, it is noticed that
the sensitivity of the defects depends upon the reference image taken. Further work
is in progress to understand this behavior. Nevertheless, this paper demonstrates yet
another case study on the simultaneous use of DIC as a structural integrity evaluation
tool in tandem with a real-time quantitative NDE (QNDE) tool.
1 Introduction
The fabrication of composite structures depends heavily upon many process vari-
ables; hence they are always prone to incur flaws at the production stage. More-
over, these structures have a tendency to develop new defects/flaws while they are
in operation. Therefore, composite launch vehicle structures require rigorous eval-
uation of structural integrity to understand their flightworthiness. Non-destructive
evaluation (NDE) is one of the most critical activities in any composite product
fabrication process. NDE plays a central role in ensuring the quality and reliability
of composite structures thereby guaranteeing its in-service life. Many offline NDT
techniques are currently available to find the number and location of existing defects
object surface can be used if a good contrast between the background and the texture
exists. A typical random pattern is shown in Fig. 1a.
A digital image pair is necessary for extracting motion-related information in DIC
referred to as the reference and deformed images. The reference image is grabbed
before loading and the deformed image after loading. Since the random pattern is
available everywhere on the imaging area, relatively smaller areas can be chosen for
pattern matching known as subsets. Subsets are selected on the reference (zeroth)
image. Now, the subsets selected in the reference image act as unique fingerprints
which have to be searched in deformed images using a correlation algorithm. There-
fore the name digital image correlation (DIC). During a mechanical test, a sequence
of images is often taken at different load/time intervals to capture the time and load-
related mechanics of the targeted object using a certain correlation algorithm. The
algorithm used in the current study is known as normalized sum of squared difference
criterion (NSSD). For template matching, a correlation coefficient (R) is used for
estimating the degree of matching expressed as [2, 13, 14]
Fi G i
2
R =
2
2 G i − Fi , (1)
Gi
where Fi and G i are the grey values of each pixel inside the reference subset and the
deformed subsets.
Three versions of DIC have been developed until now, namely 2-D DIC, 3-D DIC
and multi-DIC. 2-D DIC makes use of a single camera and the experiments are simple;
it suits for studying in-plane motions of objects. Whenever out-of-plane motion exists
on an object surface, 2-D DIC no longer produces accurate displacement and strain
fields. Therefore, 3-D DIC becomes mandatory. Multi-DIC uses a similar principle
of 3-D DIC, however, more than two cameras are arranged to cover larger areas
mostly on non-planar objects. A typical 3-D DIC setup is shown in Fig. 1b. The two
cameras are arranged in a stereo fashion for imaging the random patterns on an object
surface. The 3-D DIC or multi-DIC setups are more complex than a 2-D DIC system
54 D. Swain et al.
3 Experimental Details
A large honeycomb-sandwich composite structure, namely the ogive pay load fairing
(OPLF), made for a typical launch vehicle, had undergone ground qualification
testing. This structure has a conical portion at the bottom called the boat-tail, a cylin-
drical central part of diameter 4 m and a conical ogive section at the top (Fig. 2a).
The length of the OPLF was 8.4 m. The structure was subjected to internal pressure,
shear loads as well as aerodynamic compression at the tip of the ogive section. A
fully assembled test rig in Fig. 2b shows the external loading arrangements.
The face sheets of the OPLF sandwich panels were made of bi-directional carbon-
epoxy composite laminates which are adhesively joined to the honeycomb core on
either side. There were chances of minor defects/flaws creeping in the structure during
fabrication. During this structural test, five 3-D DIC systems were used for assessing
structural integrity at various locations, especially near cut-outs and at other critical
locations.
Fig. 2 a The 4 m diameter ogive pay load fairing (OPLF) showing the boat tail, cylinder and ogive
sections a before and b after the external loading assembly was erected
Real-Time Detection of Defects on a Honeycomb Composite Sandwich Structure … 55
Fig. 3 a A cut-out on the boat-tail of the OPLF structure and b the same cut-out as viewed through
a 3-D DIC setup after the random pattern generation
In this article, debonds identified from a 3-D DIC system employed for monitoring
a cut-out on the boat-tail are reported. The cut-out is shown in Fig. 3a and the 3-D
DIC setup imaging the cut-out spray-painted with random pattern is shown in Fig. 3b.
Initially, the dead-weight of the cap, which was used for applying aerodynamic load
at the tip, was acting on the structure. At this stage, the reference image was recorded
using VICSNAP software [15]. Subsequently, the dead-weight was off-loaded from
the hardware before the actual load application. Two more images were captured at
this off-loading condition again. Afterward, loads were applied in intermittent steps
and images were captured for DIC analysis at these load steps. In total, four loading
cases were simulated on the structure. After the first load case, the structure was
unloaded which was considered as the reference image for the second load case. The
same procedure was repeated for third and fourth load cases. For each load case,
the reference image was recorded at the corresponding no loading condition which
was not the same as the image taken before off-loading of the cap dead-weight. The
analysis of the images captured with VICSNAP was carried out using VIC3D [15]
following the guidelines provided in [14].
The displacement fields obtained above the boat-tail cut-out for the first loading case
are reported in Fig. 4. It is again emphasized here that the reference image for this load
case was taken when the structure was under the dead-weight load of the cap. In Fig. 4,
the displacement data for three image frames are shown, the maps in Fig. 4a1, a2, a3
and b1, b2, b3 correspond to the first and second frames after the dead-weight is off-
loaded in the first load case. Here, (a1, b1), (a2, b2) and (a3, b3) correspond to lateral,
axial and radial displacements, respectively. Signature for two defects in the area of
interest (AOI) can be seen in Fig. 4a1, a2 and a3, one at the top left corner of AOI and
the second one below it. The defects can be identified from all the displacement fields.
56 D. Swain et al.
Fig. 4 Displacement maps from DIC above the cut-out. a and b are the displacement maps of two
frames after dead-weight removal, whereas c represents the displacement map at maximum load
and d corresponds to the displacement map after rigid body removal of (c). Here, 1, 2, 3 digits
associated with the letters correspond to the lateral, axial and radial displacements, respectively.
These results are for the first case of loading
Signature for a third defect adjacent to the second defect can be seen in Fig. 4b1, b2
and b3 along with the first defect. However, the second defect was not seen in Fig. 4b.
The time gap between the first and the second frame would be less than one second.
This inference tells that the defects are more sensitive to the small fluctuations in
load between the first and the second frame. Therefore, it provides a clue about the
importance of the state of loading at the time of recording the reference image in DIC.
The defects could have been missed had the reference image been recorded after the
dead-weight off-loading condition. Actually, the defects were in a certain initial state
Real-Time Detection of Defects on a Honeycomb Composite Sandwich Structure … 57
under small compression loading of the cap which changed once the dead-weight
was off-loaded. These changes are exactly captured in DIC displacements online and
in real time. The displacements corresponding to this minor change in load are a few
microns only. Thanks to the excellent displacement resolution of DIC, such micron
level changes could be captured. This ability of DIC qualifies it as an in situ, online
and real-time quantitative NDE (QNDE) tool. The defects also were identified while
the structure was in service.
In order to comprehend the signature of the defects at larger loads, displacement
maps at the maximum load of the first load case are reported in Fig. 4c1, c2 and
c3, respectively. In these contour maps, no conclusive defect signature could be
identified. Therefore, the probable rigid body displacements were removed to look
for the defect signatures in Fig. 4d1, d2 and d3. This exercise also did not provide
any conclusive defect signatures. The strain analysis for all the cases reported above
would be carried out in a forthcoming paragraph to understand the defect signatures.
Before discussing the strain results, here the mechanical behavior of the defects is
characterized in Fig. 5. This is solely possible due to the quantitative characterization
ability of DIC. The a-series of graphs in Fig. 5 plot the displacements across the first
defect seen in Fig. 4. Similarly, the b-series of graphs in Fig. 5 correspond to the
second and third defects at the bottom of DIC AOI. In each plot, the magnitude
of displacement for the two frames after dead-weight removal (primary y-axis) and
at the maximum load (secondary y-axis) for the first loading case are compared.
The radial displacements for the two frames after dead-weight removal in Fig. 5a3,
b3 indicate an undulation kind of behavior which reflected in both lateral as well
as axial displacements. Such undulations are generally seen during the buckling of
shells. Moreover, butterfly patterns as seen in shearography correspond to such out-
of-plane displacement signatures. These defect signatures indicate debonding of the
skin sheets from the core. The signatures for the first defect at the top were just the
reverse of the two defects at the bottom. Moreover, at the maximum load, the defects
at the bottom completely vanished, whereas the first defect at the top only showed
some signature in the lateral displacement map. This signature was submerged in
the full-field maps in Fig. 4c1, d1. The displacements correspond to the two frames
after dead-weight off-loading were less than 20 μm.
Strain results for the two frames after dead-weight removal are shown in Fig. 6 to
understand the signature of the defects in various strain contours. The defects were
clearly identifiable in the lateral strain and principal strain maps. However, it was not
clearly distinguishable in the axial strain maps, unlike the displacement contours.
The displacement contours at maximum load did not show any signature of defect.
The corresponding strain signatures are shown in Fig. 7, wherein only lateral and
principal strains are shown. In these strain maps, defect signatures can be clearly
identified. All these discussions lead to the conclusion that defect signature may get
reflected in the displacement and/or strain contours depending upon the nature of the
defect. Therefore, during in situ and online analysis, one may have to look for all the
full-field outputs from DIC. In the present context, the defect is best visible in the
lateral strain maps.
58 D. Swain et al.
Fig. 5 a1, a2 and a3 are the lateral, axial and radial displacements plotted along a line across the
defect at the top. Similarly, b1, b2 and b3 show the plots of lateral, axial and radial displacement for
the defect at the bottom in Fig. 4a1, b1. These results correspond to the images analyzed in Fig. 4
In order to compare the magnitude of strains at the sites of the identified defects,
line graphs plotting the strains across the three defects are shown in Fig. 8. In Fig. 8,
only lateral and major principal strains are shown, since the defect signatures were
more discernible in these contour maps. The lateral strain plot in Fig. 8a for the first
defect at the top of the AOI showed a peak tensile strain of 350 με for the two frames
after dead-weight removal, which further increases by only 100 με at the maximum
load. Meanwhile, the plot in Fig. 8b for the two defects at the bottom of the AOI
showed a peak compressive strain of around 400 με. Moreover, at the maximum
load, the compressive strains decrease at the sites of the second and third defect.
A sinusoidal lateral strain trend was seen near all the defects due to the undulating
Real-Time Detection of Defects on a Honeycomb Composite Sandwich Structure … 59
Fig. 6 Lateral strain (a, b), axial strain (c, d), major principal strains (e, f) and minor principal
strains (g, h) for the two frames after dead-weight off-loading condition, wherein a, c, e and g are
for the first frame and b, d, f and h are for the second frame. These results are for the first loading
case alone
Fig. 7 a The axial strain, b major and c minor principal strain maps for the image recorded at the
maximum load for the first loading case
Fig. 8 a and b are the lateral strains plotted along a line across the defect at the top and the defects
at the bottom in Fig. 6a. These results correspond to the images analyzed in Fig. 4 and for the first
loading case
60 D. Swain et al.
Fig. 9 The lateral strain maps at the maximum load for a second, b third and c fourth loading
cases, respectively. Here, the reference is the corresponding zero loading images prior to the test
nature of the radial displacements. The observations in the strain plots are in line
with the displacement data.
At this juncture, the defect locations and sizes are almost known. Therefore, an
investigation was made in the other load cases to identify the debonds. Toward this,
only the lateral strain maps at the maximum load for the second, third and fourth
load cases are reported in Fig. 9. In the present context, it is understood that lateral
strain contour provides clear defect signatures. However, in Fig. 9, no defects were
identifiable for any of the load cases. Only strain concentrations could be seen near
the corners. It may be recalled that the reference images in these load cases were not
similar to that of the first load case. It seems that the physical state of the defect at the
time of recording of the reference image in these load cases was such that the defect
is not getting sensitized by further application of load. This observation reiterates
the importance of the reference image in detecting defects in composite sandwich
structures using DIC.
5 Concluding Remarks
In this paper, DIC is successfully utilized for in situ, online and real-time identifi-
cation of debonds in a large honeycomb composite structure. Three debonds were
located near a cut-out on the boat-tail region of the OPLF. The mechanical behavior
of the defects was also characterized in terms of displacement and strain magni-
tudes. Thereby, DIC is yet again proven as a QNDE tool. The defects in the hardware
were identified while the structure was intermittently loaded. Therefore, here DIC
is demonstrated as an in-service defect identification tool also. Moreover, the NDE-
related information is obtained while the structural behavior of the AOI was assessed.
Therefore, DIC has served the purpose of structural integrity evaluation and QNDE
simultaneously. The most notable observation in this study was the importance of
reference image in detecting the debonds. DIC has the ability to detect existing as
well as new defects generated during operation. This paper provides further stim-
ulus to devise specific procedures for detecting and characterizing such defects in
composite structures. Work is under progress to generate shearography signatures
Real-Time Detection of Defects on a Honeycomb Composite Sandwich Structure … 61
at the identified defect locations. In the future, the numerical simulation would be
carried out to understand the sensitivity of the defects.
Acknowledgements The authors wish to acknowledge Shri S. Somanath, Director, VSSC, Shri
M. Premdas, Deputy Director/STR, VSSC and Shri N. Shyam Mohan, Group Director/STG, VSSC
for their encouragement. The authors thank Mr. S. N. Suresh, Sr. Technician B, EXMD/VSSC for
his relentless support during the tests.
References
1. Wang B, Zhong S, Lee T-L, Fancey KS, Mi J (2020) Non-destructive testing and evaluation
of composite materials/structures: a state-of-the-art review. Adv Mech Eng. https://doi.org/10.
1177/1687814020913761
2. Swain D, Philip J, Selvan SK, Pillai SA (2014) Digital Image Correlation (DIC)—an invaluable
tool for non-contact and full-field assessment of critical regions on launch vehicle structures.
In: Proceedings of ASET-2014, Trivandrum, India
3. Swain D, Philip J, Pillai SA (2014) Assessment of a complex aerospace design through optical
techniques. Appl Mech Mater 592–594:1006–1010
4. Swain D et al (2015) Characterization of different adhesives for bonding thermal protection
system (TPS) on a space crew capsule. Mater Sci Forum 830–831:407–410
5. Swain D, Thomas BP, Philip J, Pillai SA (2014) Quantitative non-destructive evaluation
(Q-NDE) of a composite over-wrapped pressure vessel (COPV) liner using Digital Image
Correlation (DIC). In: Proceedings of ASET-2014, Trivandrum, India
6. Swain D, Thomas BP, Selvan SK, Philip J (2016) Real-time detection and mechanical char-
acterization of brazing anomalies in a cryogenic engine nozzle divergent using 3-D DIC. In:
Proceedings of NDE-2016, Trivandrum, India
7. Swain D et al (2019) Performance characterization of a flexible nozzle system (FNS) of a large
solid rocket booster using 3-D DIC. Exp Tech 43(4):429–443
8. Tripathi S, Swain D, Muthukumar R, Anup SN (2020) Investigation on snap-through buckling
behavior of dished shells under uniform external pressure. J Appl Mech 87(12):121001(1–11)
9. Swain D et al (2020) A full-range stress-strain model for metallic materials depicting non-
linear strain-hardening behaviour. J Strain Anal Eng Des. https://doi.org/10.1177/030932472
0957798
10. Swain D et al (2019) Use of full-field measurements for understanding acoustic emission (AE)
signals from 15CDV6 ESR steel specimens. Proc Struct Integr 14:207–214
11. Niezrecki C, Baqersad J, Sabato A (2018) Digital image correlation techniques for NDE
and SHM. In: Ida N, Meyendorf N (eds) Handbook of advanced non-destructive evaluation.
Springer, Cham. https://doi.org/10.1007/978-3-319-30050-4_47-1
12. Thomas BP, Swain D, Selvan SK, Philip J (2016) Defect detection in composite sandwich struc-
ture with face sheet of varying thickness using time average digital holography. In: Proceedings
of NDE-2016, Trivandrum, India
13. Sutton MA, Orteu J, Schreier WH (2009) Image correlation for shape, motion and deformation
measurement: basic concepts, theory and applications, 1st edn. Springer, New York
14. Swain D, Philip J, Pillai SA (2013) Optimized Digital Image Correlation (DIC) for measuring
high strain gradient regions. In: Proceedings of 1st Indian conference on applied mechanics
(INCAM), IIT Madras, India
15. VIC-3D and VIC-SNAP supplied by correlated solutions, Inc., USA
Novel Data Processing Approaches
for Testing and Evaluation of Mild Steel
Sample Using Frequency-Modulated
Thermal Wave Imaging
1 Introduction
Various active IRT techniques like pulse thermography (PT), pulse phase ther-
mography (PPT) and lock-in thermography (LT) have been developed in the liter-
ature to quantify defects located inside the sample [2–6]. However, these primary
techniques limit their applicability due to high power requirements, non-uniform
heating and thermal effusivity changes for PT and PPT, respectively. LT on the other
hand requires repetition in experiments due to mono-frequency excitation in order
to evaluate defects in the entire sample thickness. Therefore, a frequency-modulated
thermal excitation scheme was introduced for improved depth resolution of defects
lying inside the sample at different depth locations with low peak power requirements
and less experimentation time.
Frequency-modulated thermal wave imaging (FMTWI) allows modulating heat
sources within a band of frequencies with significant magnitude resulting in better
test resolution. This paper presents the FMTWI approach for defect detection in mild
steel sample using conventional phase analysis and recently proposed time-domain-
matched filtering data processing approach. Matched filtering (MF) approach concen-
trates the energy of non-stationary signals with time using the cross-correlation (CC)
technique. CC enables to determine defect depth by evaluating peak value and its
corresponding time lag for matched filtering approach. Further, comparative analysis
between discussed data processing approaches has been presented for the FMTWI
technique.
2 Theory
In the FMTWI technique, heat sources modulated with significant peak power within
a band of frequencies are projected over a sample surface to diffuse thermal waves
from surface to interior of the sample given by Eq. (1) [7, 8]. These thermal waves
in the presence of any defect lead to a change in surface temperature which is
recorded during the heating cycle. The surface of the sample is illuminated with
linear frequency modulated (LFM) heat flux Q(t) while the other end of the sample
is insulated to prevent any heat transfer to the surroundings given by Eqs. (2) and
(3).
∂ 2 T (a, t) ρm Cm ∂ T (a, t)
= ; 0≤a≤d (1)
∂a 2 Km ∂t
∂ T (0, t) B 2
−K m = Q(t) = Q m ∗ 1 + sin 2π f m t + t (2)
∂a 2T
∂ T (d, t)
−K m =0 (3)
∂a
where T (a,t) is the temperature variation across spatial location a at time instant t,
K m is the thermal diffusivity, ρ m is the density and C m is the specific heat capacity
Novel Data Processing Approaches for Testing and Evaluation of Mild Steel … 65
of the sample, respectively. Qm is the peak power of input heat flux, f m is the initial
frequency, B is the bandwidth, T is the total duration of input heat flux and d is the
sample thickness. A general solution for surface temperature under given boundary
conditions, keeping sample initially at room temperature T o can be computed by
putting a = 0 in the form of
∞
Mi cos[βa]e−t ρm Cm β
Km 2
Tm (a, t) = (5)
i=0
∞
t − (t−τ )K m
β2
Tn (a, t) = ∫ Ni (τ ) cos[βa]e ρm C m
dτ (6)
0
i=0
The CC output compresses total energy of long duration input modulated signal
into short duration sinc-shaped pulse with maximum energy in the main lobe along
with time delay corresponding to defect depth.
66 A. Rani et al.
The fast-Fourier transform was applied to the zero mean thermal distributions for
each pixel T (a) in the region of interest for extracting the amplitude and phase
information as given below [13, 14]
N −1
1 −i2πva
F(v) = T (a)e[ N ] = Re(v) + iIm(v) (10)
N n=0
where Re(v) and Im(v) represent the real and imaginary components of the computed
fast-Fourier transform F(v). Further, the amplitude and phase images can be extracted
in the form of
Amp = Re2 (v) + Im2 (v) ; (11)
Im(v)
ϕ = a tan (12)
Re(v)
The phase does not change due to variation in the surface geometry and non-
uniform surface heating; therefore, it can be utilized as an important factor for the
quantitative analysis of the defect depths.
3 Numerical Modeling
In this section, the performance of FMTWI for inspection of mild steel sample has
been evaluated based on the three-dimensional finite-element modeling and simula-
tion (COMSOL Multiphysics 5.2) approach. Figure 1 presents a mild steel sample
having air defects located from a distance of 0.3 mm from the front surface: (a) six
Fig. 1 Layout of the mild steel sample with six defects having: a diameter (φ) 10 mm present at
different depth locations b different diameters (φ) in the range of 3–5.5 mm present at 0.3 mm depth
from the front surface (all dimensions in mm)
Novel Data Processing Approaches for Testing and Evaluation of Mild Steel … 67
Table 1 Defect dimensions for mild steel sample (a) and (b)
Defects Sample a Sample b
Diameter (mm) Depth (mm) Diameter (mm) Depth (mm)
1 10 0.5 5.5 1.5
2 10 0.8 5 1.5
3 10 1.1 4.5 1.5
4 10 1.4 4 1.5
5 10 1.7 3.5 1.5
6 10 2 3 1.5
Fig. 2 Thermal response obtained for the air defects having i the same diameter at different depth
locations, ii different diameters at same depth locations
68 A. Rani et al.
Fig. 3 Mean removed thermal response obtained for the air defects having i the same diameter at
different depth locations ii different diameters at the same depth locations
Fig. 4 Correlation coefficient profiles with sound as reference obtained for air defects having i the
same diameter at different depth locations, ii different diameters at the same depth locations
Fig. 5 Air defects of the same diameter at different depth locations, i Lag, ii Correlation coefficient
Novel Data Processing Approaches for Testing and Evaluation of Mild Steel … 69
Fig. 6 Air defects of different diameters at same depth location, i Lag, ii Correlation coefficient
the blind hole depth decreases from 2 mm (highest depth) to 0.5 mm (lowest depth),
the lag decreases. A similar trend was observed for the correlation coefficient in
Fig. 5ii; as the depth of the blind hole decreases the correlation coefficient decreases.
Figure 6i, ii depicts the trend observed for the time lag and correlation coefficients
for blind hole defects located at 0.3 mm from the front surface having diameters in
the range of 5.5–3 mm each having 1.5 mm depth. As the diameter of the blind hole
defect decreases, the time lag as well as the correlation coefficient decrease.
The depth scanning performance can be obtained from the time-domain and
Fourier-domain phase analysis approach shown in Figs. 7 and 8, respectively. A
number of phase images were obtained from both the approaches in which optimum
frames are chosen. In the case of time-domain phase analysis approach, the best phase
image which can be used to visualize deeper as well as the shallowest defect was
obtained at 0.4 s, as shown in Fig. 7. Similarly, the Fourier-domain phase analysis
approach involves phase images at various frequencies out of which the phase image
obtained at 0.04 Hz possesses the best visualization of all the defects, as shown in
Fig. 8. Here, time-domain phase analysis approach performs better in defect detec-
tion as compared to the Fourier-domain phase analysis approach. Correlation images
shown in Fig. 9, have been observed from the delayed cross-correlation profiles for
each pixel of the compressed pulse. It was observed that the deepest defects are
clearly visible and show more contrast than deeper defects both in cases of mild steel
sample.
Fig. 8 Fourier-domain
phase observed at 0.04 Hz
for air defects of i same
diameter at different depth
locations, ii different
diameters at same depth
locations
5 Conclusion
The paper presented a high accuracy method for defect depth detection located at
varying depths and diameters in a mild steel sample with artificially inserted blind
hole defects. This has been achieved by studying the linear frequency-modulated
thermal wave imaging approach. The paper initially discusses thermal response
observed for finite thickness heat diffusion model under adiabatic boundary condi-
tions. The thermal response recorded was fitted using a linear fit operation to obtain
zero mean thermal responses for different defect observations. The thermal response
was post-processed using the time-domain phase analysis and frequency-domain
phase analysis approach. Further, correlation coefficients and time lag were computed
for estimating the defect depths of different diameters located at different depth
locations.
Novel Data Processing Approaches for Testing and Evaluation of Mild Steel … 71
References
1. Carlslaw HS, Jaeger JC (1959) Conduction of heat in solids, 2nd edn. Oxford University Press,
London
2. Almond DP, Patel PM (1996) Photothermal science and techniques. Chapman-Hall
3. Busse G, Wu D, Karpen W (1992) Thermal wave imaging with phase sensitive modulated
thermography. J Appl Phys 71:3962–3965. https://doi.org/10.1063/1.351366
4. Almond D, Pickering S (2012) An analytical study of the pulsed thermography defect detection
limit. J Appl Phys 111:093510. https://doi.org/10.1063/1.4704684
5. Mulaveesala R, Tuli S (2006) Theory of frequency modulated thermal wave imaging for nonde-
structive subsurface defect detection. Appl Phys Lett 89:191913. https://doi.org/10.1063/1.238
2738
6. Tuli S, Mulaveesala R (2005) Defect detection by pulse compression in frequency modulated
thermal wave imaging. Quantitative InfraRed Thermogr J 2:41–54. https://doi.org/10.3166/
qirt.2.41-54
7. Cook CE, Paolillo J (1964) A pulse compression predistortion function for efficient sidelobe
reduction in high-power radar. Proc IEEE 52(4):377–389. https://doi.org/10.1109/PROC.1964.
2927
8. Rani A, Mulaveesala R (2020) Investigations on pulse compression favourable thermal imaging
approaches for characterization of glass fibre reinforce polymers. Electron Lett 56(19):995–
998. https://doi.org/10.1049/el.2020.0789
9. Ramp HO, Wingrove ER (1961) Principles of pulse compression. IRE Trans Mil Electron
5(2):109–116. https://doi.org/10.1109/IRET-MIL.1961.5008328
10. Rani A, Mulaveesala R (2020) Depth resolved pulse compression favourable frequency modu-
lated thermal wave imaging for quantitative characterization of glass fibre reinforced polymer.
Infrared Phys Technol 110:103441. https://doi.org/10.1016/j.infrared.2020.103441
11. Rani A, Mulaveesala R (2021) Pulse compression favorable frequency modulated thermal
wave imaging for non-destructive testing and evaluation: an analytical study. IOPSciNotes
2(2):024401. https://doi.org/10.1088/2633-1357/ac049a
12. Dua G, Mulaveesala R, Kher V, Rani A (2019) Gaussian windowed frequency modulated
thermal wave imaging for non-destructive testing and evaluation of carbon fibre reinforced
polymers. Infrared Phys Technol 98:125–131. https://doi.org/10.1016/j.infrared.2019.03.007
13. Vavilov VP (2017) Thermal nondestructive testing of materials and products: a review. Russ J
Nondestr Test 53(10):707–730. https://doi.org/10.1134/S1061830917100072
14. Tabatabaei N, Mandelis A (2009) Thermal-wave radar: a novel subsurface imaging modality
with extended depth-resolution dynamic range. Rev Sci Instru 80(3):34902. https://doi.org/10.
1063/1.3095560
Barker-Coded Thermal Wave Imaging
for Testing and Evaluation of Mild Steel
Abstract The estimation of defect depth is a crucial parameter in the field of non-
destructive testing. The paper presents the simulation aspects for three-dimensional
heat transfer equation for a finite thickness steel sample having six blind holes as
defects located at different locations using a promising Barker-coded thermal wave
imaging technique (BCTWI). The surface of the test sample is subjected to Barker-
coded modulated thermal excitation, and the analysis has been carried out using the
time domain phase and correlation analysis post-processing approach on the captured
data during the active heating. Finally, various adopted processing approaches have
been compared to evaluate defect detection capabilities.
1 Introduction
2 Theory
The thermal wave propagation in finite thickness solid can be represented by one-
dimensional heat diffusion equation in the absence of any external heat source or
sink given by Mulaveesala and Tuli [5]:
∂ 2 W (z, t) 1 ∂ W (z, t)
− = 0; 0 ≤ z ≤ L (1)
∂z 2 α ∂t
∂ W (0, t)
−k = Q BC T W I (t) (2)
∂z
The other end of test sample is kept insulted as per adiabatic boundary conditions,
and the sample is kept initially at room temperature T o represented by
∂ W (L , t)
−k =0 (3)
∂z
W (z, 0) = To (4)
Barker-Coded Thermal Wave Imaging for Testing and Evaluation … 75
Barker code is a simple and easily generated binary code to provide better compres-
sion of side lobes in comparison to other modulated schemes. Also, the compression
ratio in BCTWI is proportional to the code length in its auto-correlation. The code
length is chosen based on the requirement of experimentation time and reduction in
side lobes. Short experimentation time requires high peak power heat sources while
long experimentation reduces the advantages associated with BCTWI technique [6–
8]. Thus, an appropriate code length and experimentation time should be chosen to
obtain better thermal response. Therefore, 7-bit Barker code has been used to provide
a desired thermal response.
In 7-bit BCTWI, a Barker-coded thermal heat flux given by Eq. (5) is illuminated
over the front surface of the test sample, and the thermal response is obtained at the
surface W (0,t) given by
4
Q BC T W I (t) = Q o (−1)ni u(t − ai T ); n i = 0, 1, 2, 3; ai = 0, 3, 5, 6 (5)
i=1
where Qo —peak amplitude of input heat flux, T —total duration of input excitation.
The thermal profile at the front surface of the test sample can be computed by:
⎡ ⎤
b L pπ z pπ −αtλ2
∞ ⎢ W (z, 0) cos dz cos z e + ⎥
⎢ L 0 L L ⎥
W (z, t) = ⎢ t L ⎥ (6)
⎣ b ∂ W (0, τ ) pπ z −[α(t−τ )λ ]
2 ⎦
p=0 × dz cos e dz
0 L 0 ∂z L
pπ 1 p=0
where; λ = f or p = 0, 1, 2, ...., ∞ and b = f or
L 2 p = 1, 2, 3...∞
(7)
The resultant correlated profile C(τ ) compresses the long duration signal W (t) of
finite duration T s into a compressed pulse of duration 1/B, where B (Hz) is band-
width of waveform. Here, compression ratio TB is defined as ratio of input thermal
signal and output compressed signal obtained for correlation approach. Further, finite
attenuation and time delay in correlation profiles obtained for each defect location
are used for defect detection in test sample.
3 Numerical Modeling
In this section, simulation studies were carried out using three-dimensional finite
element model (FEM) for heat transfer in two mild steel samples each of size 280
× 40 × 2.3 mm shown in Fig. 2. Figure 2i shows mild steel sample of 2.3 mm
thickness having six blind holes as defects located at different depths each having
10 mm diameter. Figure 2ii depicts mild steel sample of 2.3 mm thickness having
six blind holes as defects having different diameters in the range of 3–5.5 mm. All
Barker-Coded Thermal Wave Imaging for Testing and Evaluation … 77
Fig. 1 Time domain data processing approach for phase and CC images
Fig. 2 i Defects of same diameter at different depths, ii Defects of different diameters at same
depth. (All defects present at 0.3 mm from front surface)
the six defects are located at a depth of 0.3 mm from the front surface of the test
samples shown in Fig. 2i, ii, respectively. A 7-bit Barker-coded input heat flux is
illuminated over the front surface of the samples for 100 s duration while keeping
the other end insulated. The test sample is assumed initially at room temperature, and
thermal response was observed on the front surface of the sample during the active
heating cycle for mild steel samples.
To validate the defect detectability of the proposed BCTWI technique, FEM simula-
tion and analysis has been carried out over two mild steel samples of thickness 2.3 mm
illustrated in Fig. 2 The test samples contain six blind holes as defects located at a
depth of 0.3 mm from front surface illustrated in Table 1. 7-bit Barker-coded input
heat flux with peak power 1 kW/m2 is exposed over the sample surface for 100 s as
depicted in Fig. 3 The thermal profile generated in accordance to defect depth for
78 A. Rani et al.
Table 1 Defect dimensions for mild steel sample (i) and (ii)
Defects Mild steel sample (i) Mild steel sample (ii)
Diameter (mm) Depth (mm) Diameter (mm) Depth (mm)
a 10 0.5 5.5 1.5
b 10 0.8 5 1.5
c 10 1.1 4.5 1.5
d 10 1.4 4 1.5
e 10 1.7 3.5 1.5
f 10 2 3 1.5
(i) (ii)
Fig. 4 Temperature response: i defects of same diameter at different depths, ii defects of different
diameters at same depth
Barker-Coded Thermal Wave Imaging for Testing and Evaluation … 79
(i) (ii)
Fig. 5 Zero mean temperature response: i defects of same diameter at different depths, ii defects
of different diameters at same depth
with reference thermal profile. Here, thermal profile obtained for sample thickness
is considered as reference signal. Further, Barker-coded thermal profiles with corre-
lation approach provide smaller side lobes by concentrating maximum energy in the
main lobe as observed in Fig. 6i, ii respectively. Deeper defects and large diame-
ters show high correlation with similarity among defect and reference profile. Also,
results show reshaping of spectra for barker code with improved compression prop-
erties. The correlation attenuation and time delay can be observed clearly in Fig. 6.
Mild steel sample with varying diameters observed a prominent time lag/delay in
comparison to varying defect depths.
Figures 7 and 8 illustrate the defect detection ability of time domain-based phase
and correlation data processing approach using 7-bit BCTWI technique. Figure 7i, ii
represents the phasegrams obtained from time domain data processing approach for
two mild steel samples using BCTWI at time instant of 1.92 s and 1.99 s, respectively.
Further, depth scanning performance of mild steel sample with varying defect depths
and diameters using CC analysis at time instant of 58.1 s and 61.5 s is presented in
Fig. 8i, ii respectively. Results show prominent improvement in defect contrast using
correlation approach in comparison to time domain phase image data processing
approach.
(i) (ii)
Fig. 6 Correlation coefficients: i defects of same diameter at different depths, ii defects of different
diameters at same depth
80 A. Rani et al.
(i) (ii)
(i) (ii)
5 Conclusion
Defect detection in two mild steel sample having six artificial blind holes of varying
depth and diameters has been carried out using 7-bit Barker-coded excitation scheme.
Further, time domain phase analysis and correlation data processing approach have
Barker-Coded Thermal Wave Imaging for Testing and Evaluation … 81
been applied to the recorded thermal profiles for defect analysis. Barker-coded exci-
tation scheme along with correlation approach has been applied for a reduction in side
lobes to extract sub-surface hidden defects. Results for the proposed scheme show
improved defect detection ability in time domain correlated images in comparison
to phase images.
References
1. Carlslaw HS, Jaeger JC (1959) Conduction of heat in solids, 2nd edn. Oxford University Press,
London
2. Almond DP, Patel PM (1996) Photothermal science and techniques. Chapman-Hall
3. Busse G, Wu D, Karpen W (1992) Thermal wave imaging with phase sensitive modulated
thermography. J Appl Phys 71:3962–3965. https://doi.org/10.1063/1.351366
4. Almond D, Pickering S (2012) An analytical study of the pulsed thermography defect detection
limit. J Appl Phys 111:093510. https://doi.org/10.1063/1.4704684
5. Mulaveesala R, Tuli S (2006) Theory of frequency modulated thermal wave imaging for nonde-
structive subsurface defect detection. Appl Phys Lett 89:191913. https://doi.org/10.1063/1.238
2738
6. Ahmad J, Akula A, Mulaveesala R, Sardana HK (2019) Barker coded thermal wave imaging
for nondestructive testing and evaluation of steel material. IEEE Sens J 19(2):735–742. https://
doi.org/10.1109/JSEN.2018.2877726
7. Dua G, Mulaveesala R (2013) Applications of Barker coded infrared imaging method for
characterisation of glass fibre reinforced plastic materials. Electron Lett 49(17):1071–1073.
https://doi.org/10.1049/el.2013.1661
8. Ghali VS, Panda SSB, Mulaveesala R (2011) Barker coded thermal wave imaging for
defect detection in carbon fibre-reinforced plastics. Insight Nondestr Testing Condition Monit
53(11):621–624. https://doi.org/10.1784/insi.2011.53.11.621
9. Rani A, Mulaveesala R (2020) Investigations on pulse compression favourable thermal imaging
approaches for characterization of glass fibre reinforce polymers. Electron Lett 56(19):995–
998. https://doi.org/10.1049/el.2020.0789
10. Rani A, Mulaveesala R (2020) Depth resolved pulse compression favourable frequency modu-
lated thermal wave imaging for quantitative characterization of glass fibre reinforced polymer.
Infrared Phys Technol 110:103441. https://doi.org/10.1016/j.infrared.2020.103441
11. Tuli S, Mulaveesala R (2005) Defect detection by pulse compression in frequency modulated
thermal wave imaging. Quant Infrared Thermogr J 2:41–54. https://doi.org/10.3166/qirt.2.
41-54
12. Cook CE, Paolillo J (1964) A pulse compression predistortion function for efficient sidelobe
reduction in high-power radar. Proc IEEE 52(4):377–389. https://doi.org/10.1109/PROC.1964.
2927
13. Ramp HO, Wingrove ER (1961) Principles of pulse compression. IRE Trans Mil Electron
5(2):109–116. https://doi.org/10.1109/IRET-MIL.1961.5008328
14. Rani A, Mulaveesala R (2021) Pulse compression favorable frequency modulated thermal
wave imaging for non-destructive testing and evaluation: an analytical study. IOPSciNotes
2(2):024401. https://doi.org/10.1088/2633-1357/ac049a
15. Dua G, Mulaveesala R, Kher V, Rani A (2019) Gaussian windowed frequency modulated
thermal wave imaging for non-destructive testing and evaluation of carbon fibre reinforced
polymers. Infrared Phys Technol 98:125–131. https://doi.org/10.1016/j.infrared.2019.03.007
82 A. Rani et al.
16. Vavilov VP (2017) Thermal nondestructive testing of materials and products: a review. Russ J
Nondestr Test 53(10):707–730. https://doi.org/10.1134/S1061830917100072
17. Tabatabaei N, Mandelis A (2009) Thermal-wave radar: a novel subsurface imaging modality
with extended depth-resolution dynamic range. Rev Sci Instru 80(3):34902. https://doi.org/10.
1063/1.3095560
Proximity: An Automatic Approach
for Defect Detection and Depth
Estimation in Infrared Non-destructive
Testing
thermographic and machine learning metrics are concerned to qualify the reliability
of the proposed methodology.
1 Introduction
Active infrared thermography uses contrast in temporal thermal maps over the object
surface to classify defects under a controlled optical stimulus. It is known that the
thermal profiles at the sound region possess similar attributes, whereas the thermal
profiles at the defective region present different characteristics depending on the
size and depth of the defect [19]. Similarly, a small defect generated in subsurface
layers is covered by significantly fewer pixel profiles than its non-defective region,
where the temporal thermal profiles covering this small defect region can be treated
as outliers. However, the score at which the given thermal profile is an outlier to
the sound region, thermal profile is dependent on the depth and size of the defect. In
the present case, material inspection through quadratic frequency-modulated thermal
wave imaging is focused.
In QFMTWI, a band of low frequencies modulates the low peak power optical
stimulus from halogen lamps, and the respective thermal map over the surface is
observed using an infrared camera [5, 6]. The experimental schematic of QFMTWI is
shown in Fig. 1. The observed thermal response is pre-processed and post-processed
to extract the finer subsurface details in the test sample. The subsurface anomalies
are identified by observing the respected contrast map from the processed thermal
86 G. T. Vesala et al.
response. On the other hand, the depth quantification of detected defects requires
complex and non-linear theoretical models [14]. However, deeper defects present
less contrast compared with the shallowest defects, which made them hard to detect
and prone to errors on human intervention. Hence, proximity-based automatic defect
detection and depth estimation method is introduced in this manuscript to avoid
human intervention.
Proximity means closeness, nearness or relation between some objects or data. LOF
is a density-based technique that uses local densities or proximity to detect anomalies
or outliers in the given data, where the anomalies or outliers are the rare instances
that deviate from most of the data [24]. However, based on their existence in the
feature space, the outliers are classified as local and global outliers, whereas the
global outliers are easily detected compared with the local outliers. In the present
case of infrared thermography, the shallowest and bigger defect presents high contrast
than the deeper and smallest defects; hence, they can be treated as global outliers
that can be easily detected by observing the contrast map of the thermogram. On
the other hand, the deeper and smaller defects present close variations of the sound
region treated as local outliers and are tough to detect. Detecting these local outliers
is a crucial task in infrared thermography. In the present case, LOF uses the score
at which given thermal response characteristics are close to or nearer to the sound
region thermal profiles to provide automatic defect detection.
At first, for a given temporal thermal profile, the k-nearest neighbors are computed
with respect to the local density as local reachability density (LRD), which defines
the proximity. Finally, a degree to which the given thermal response is an outlier
Proximity: An Automatic Approach for Defect Detection … 87
is calculated by comparing the LRD of a thermal profile with all the LRDs of its k
neighbors. LOF can be expressed as [25]
lr d Min Pts (o)
o∈N Min Pts ( p) lr d Min Pts ( p)
L O FMin Pts ( p) = (1)
|N Min Pts ( p)|
where p and o are two thermal profiles, MinPts specifies the number of k-nearest
neighbors, and LRD gives local reachability distance. MinPts decides the decision
function, i.e., the fraction of a given thermal profile to be an outlier responsible for
classifying the defect and non-defective regions. In the present case, LOF computes
the factor for a given thermal response at which it is an outlier corresponding to its
neighbors. Algorithm 1 presents the complete operation of LOF.
1
k
f (
x) = y N N (
x , i) (2)
k i=1
88 G. T. Vesala et al.
Furthermore, the estimated depth is rearranged at the classified defect region and
visualized as a depth map of the test sample.
Fig. 2 a Schematic layout (dimensions in cm) and b Optical photo of rear view of the CFRP test
specimen
Proximity: An Automatic Approach for Defect Detection … 89
Fig. 3 a Dynamic thermal profile extraction and, b Select thermal profiles from the defect region
for regression analysis
The observed temporal thermal response consists of static and dynamic components
that correspond to the DC offset and the excitation, respectively. The static part
is a mean raise in the thermal response, which can be removed by a proper fitting
technique and further mean removed to extract the dynamic component. The raw and
dynamic temporal thermal responses correspond to the defective and sound regions
are presented in Fig. 3a. The dynamic temporal thermal profile of 2500 samples is
down-sampled with a factor of 4 to obtain 625 samples to reduce the computation time
of classification and regression models [18]. Furthermore, the 3D thermal response
is converted to a 2D vector by arranging the thermal map of each pixel in rows and
their respective temporal variations in columns. Meanwhile, each thermal profile
is assigned with its ground truth values (+1 and −1) for classification with respect
to sound and defect region thermal profiles and their depth for regression analysis.
Since LOF is intended to learn the normal instances or positive class data, 1000
thermal profiles from the sound region are randomly selected to form training data.
Similarly, five-pixel profiles from each defect at selected locations are extracted to
form a regression analysis training set. The five locations in the defect are center,
right, left, top and bottom pixels as shown in Fig. 3b. The training set consists of 125
thermal profiles corresponding to five depths.
The defect classification and depth estimation are implemented on an Intel i3 CPU
with 2 GHz clock speed supported with 8 GB memory and 2 TB storage in python
90 G. T. Vesala et al.
3.8.5 environments. The training data and testing data are normalized with respect
to the standard deviation of each thermal profile. LOF is initiated with an outlier
factor of 0.01 and a min-pts of 35. And the regression model is initialized with five
neighbors. These hyperparameters are fixed after multiple trials. These classification
and regression models are available in the sci-kit learn library [27].
Fig. 4 Automatic defect detection in CFRP sample using a Isolation forest, b Proposed local outlier
factor and c presents the AUC curves of both the classifiers
Proximity: An Automatic Approach for Defect Detection … 91
Fig. 5 Decision thresholds generated over each thermal response by a Isolation forest and b Local
outlier factor
Where μ is the mean and σ is the standard deviation of the given region in
the decision function. The defect size is estimated by calculating the full width at
half maxima of the spatial profile passing through the defect center [19]. Figure 6a,
b presents the SNR and size estimation of defects in the CFRP sample from the
decision thresholds of IF and LOF. The figure suggests that LOF presents high
SNR’s compared with IF, which is due to the low variance observed in the sound
region, i.e., 0.0004, whereas IF produced 0.2270. On the other hand, LOF presents
good agreement with the actual sizes of defects compared with IF; this is because of
the more local operations performed by LOF than IF over a given thermal profile.
6 Conclusion
In the present article, automatic defect detection and depth estimation through
proximity-based classification and a regression model are discussed for composite
inspection in QFMTWI. The defect detection in infrared thermography is mapped to
the anomaly detection in machine learning by considering the defective thermal
profiles as rare instances compared with sound region thermal profiles. Experi-
mentation is conducted over a carbon fiber-reinforced polymer sample with arti-
ficially drilled flat bottom holes of different sizes at different depths as defects. From
Proximity: An Automatic Approach for Defect Detection … 93
the observations, it is concluded that the proposed methodology meets the present
revolution in NDT by employing proximity-based machine learning techniques for
automatic defect detection and characterization without human intervention.
Acknowledgements This work is partially supported by Naval Research Board, India under the
grant number NRB-423/MAT/18-19.
References
1. Maldague XPV (2001) Theory and practice of infrared technology for non-destructive testing.
Wiley, New York
2. Chung Y, Shrestha R, Lee S, Kim W (2020) Thermographic inspection of internal defects
in steel structures: analysis of signal processing techniques in pulsed thermography. Sensors
20(21):6015
3. Waugh RC, Dulieu-Barton JM, Quinn S (2014) Modelling and evaluation of pulsed and pulse
phase thermography through application of composite and metallic case studies. Ndt&E Int
66:52–66
4. Shrestha R, Park J, Kim W (2016) Application of thermal wave imaging and phase shifting
method for defect detection in stainless steel. Infrared Phys Technol 76:676–683
5. Mulaveesala R, Tuli S (2006) Theory of frequency modulated thermal wave imaging for
nondestructive subsurface defect detection. Appl Phys Lett 89(19):191913
6. Subbarao GV, Mulaveesala R (2012) Quadratic frequency modulated thermal wave imaging
for non-destructive testing. Progr Electromagn Res 26:11–22
7. Ghali VS, Panda SSB, Mulaveesala R (2011) Barker coded thermal wave imaging for
defect detection in carbon fibre-reinforced plastics. Insight-Nondestr Test Condition Monit
53(11):621–624
8. Mulaveesala R, Arora V, Dua G (2020) Pulse compression favorable thermal wave imaging
techniques for non-destructive testing and evaluation of materials. IEEE Sens J
9. Subhani SK, Suresh B, Ghali VS (2018) Quantitative subsurface analysis using non stationary
thermal wave imaging. Infrared Phys Tech 88:41–47
10. Tabatabaei N, Mandelis A (2009) Thermal-wave radar: a novel subsurface imaging modality
with extended depth-resolution dynamic range. Rev Sci Instrum 80(3):034902
11. Meier J, Tsalicoglou I, Mennicke R (2017) The future of NDT with wireless sensors, AI and
IoT. In: Proceedings 15th Asia Pacific conference for non-destructive testing, Singapore, pp
1–11
12. Fang Q, Nguyen BD, Castanedo CI, Duan Y, Maldague II X (2020) Automatic defect detec-
tion in infrared thermography by deep learning algorithm. In: Thermosense: thermal infrared
applications XLII, vol 11409, p 114090T. International Society for Optics and Photonics
13. Luo Q, Gao B, Woo WL, Yang Y (2019) Temporal and spatial deep learning network for
infrared thermal defect detection. NDT & E Int 108:102164
14. Darabi A, Maldague X (2002) Neural network based defect detection and depth estimation in
TNDE. Ndt& E Int 35(3):165–175
15. Saeed N, Abdulrahman Y, Amer S, Omar MA (2019) Experimentally validated defect depth
estimation using artificial neural network in pulsed thermography. Infrared Phys Technol
98:192–200
16. Dudzik S, Dudek G (2020) Detection of thinning of homogeneous material using active
thermography and classification trees. Metrol Meas Syst 28(1)
17. Fang Q, Maldague X (2020) A method of defect depth estimation for simulated infrared
thermography data with deep learning. Appl Sci 10(19):6819
94 G. T. Vesala et al.
18. Cao Y, Dong Y, Cao Y, Yang J, Yang MY (2020) Two-stream convolutional neural network for
non-destructive subsurface defect detection via similarity comparison of lock-in thermography
signals. NDT&E Int 112:102246
19. Vijaya Lakshmi A, Gopi Tilak V, Parvez MM, Subhani SK, Ghali VS (2019) Artificial
neural networks based quantitative evaluation of subsurface anomalies in quadratic frequency
modulated thermal wave imaging. Infrared Phys Technol 97:108–115
20. Parvez MM, Shanmugam J, Ghali VS (2020) Decision tree-based subsurface analysis using
Barker coded thermal wave imaging. Infrared Phys Technol 109:103380
21. Vijaya Lakshmi A, Ghali VS, Subhani SK, Baloji NR (2020) Automated quantitative subsurface
evaluation of fiber reinforced polymers. Infrared Phys Technol 110:103456
22. Gopi Tilak V, Ghali VS, Vijaya Lakshmi A, Suresh B, Naik RB (2021) Proximity based
automatic defect detection in quadratic frequency modulated thermal wave imaging. Infrared
Phys Technol 114:103674
23. Vesala GT, Ghali VS, Vijaya Lakshmi A, Naik RB (2021) Deep and handcrafted feature fusion
for automatic defect detection in quadratic frequency modulated thermal wave imaging. Rus J
Nondestr Testing 57(6):476–485
24. Munir M, Chattha MA, Dengel A, Ahmed S (2019) A comparative analysis of traditional
and deep learning-based anomaly detection methods for streaming data. In: 2019 18th IEEE
international conference on machine learning and applications (ICMLA), pp 561–566. IEEE
25. Breunig MM, Kriegel H-P, Ng RT, Sander J (2000) LOF: identifying density-based local
outliers. In: Proceedings of the 2000 ACM SIGMOD international conference on management
of data, pp 93–104
26. Altman NS (1992) An introduction to kernel and nearest-neighbor nonparametric regression.
J Am Stat Assoc 46:175–185
27. Pedregosa et al (2011) Scikit-learn: machine learning in python. JMLR 12:2825–2830
28. Liu FT, Ting KM, Zhou Z-H (2008) Isolation forest. In: 2008 eighth IEEE international
conference on data mining, pp 413–422. IEEE
A Comparative Evaluation of Two
Scanning Modalities in Industrial
Cone-Beam Computed Tomography
Abstract This paper discusses two scanning modalities for Industrial Cone-Beam
Computed Tomography (ICBCT) and their comparative evaluation. The first conven-
tional mode scans an object when it is stationary then it moves by an angular step
and stops for scanning (Start-Stop mode). The second mode acquires projections
on the fly while the object is in a steady motion. The second mode of acquisition
of projection data is aimed at reducing overall scanning time with an acceptable
level of systemic artifacts. Data generated in both modalities are digital radiographs,
which serve as inputs for the generation of CT images. Evaluation of 2D projections
acquired in both modalities was analyzed quantitatively to determine their compar-
ative quality. Cone-beam CT has been implemented using the FDK algorithm on all
acquired data sets. Subsequently, CT images from both modalities are analytically
assimilated in terms of statistical parameters like line profile and SNR for compara-
tive quality analysis. Reduction in scanning time from 3600 to 362 s was successfully
achieved, however, system artifact generated due to asynchronous acquisition in the
second modality was observed. The method to minimize systemic artifact generated
in the second modality is discussed and implemented. Results before system arti-
fact correction and after corrections are shown in the paper. The minimum time of
2π rotation motion for data acquisition without deteriorating reconstructed image
quality for defined acquisition parameter has been estimated.
1 Introduction
FDK algorithm. In continuous rotate scanning mode, the angular step is not prede-
fined and needs to be derived from available data. In a perfectly synchronized system,
it can be assumed that the first image and last image represent the 0° image and 360°
image, respectively. This assumption can be used to calculate the step angle. In CT
system developed for continuous rotation scanning, initially, the object starts rotating
and after 5 s detector starts acquiring data continuously. When the object completes
rotation of 2π then it triggers data acquisition to stop. This method is inherently asyn-
chronous due to the latency of transmission protocol. In this case, it is not mandatory
that the last image corresponds to 360°, therefore, some other methods should be
used to derive step angle. Theoretically, 0°th image and 360°th image will be iden-
tical but due to the statistical nature of X-ray and detector components, it will not be
an exact match but very comparable [22]. There are four reported methods [11, 12,
22, 23] based on 0°th image and 360°th image comparison to derive step angle. This
paper demonstrates the application of mean square error (MSE) method [22] for the
determination of 360°th image.
Amorphous silicon (a–Si) flat panel (FP) detectors can exhibit hysteresis that
affects projection images in fluoroscopy and cone-beam CT (CBCT) reconstructions
[24]. Step-by-step rotation scan has one more advantage over continuous scan that
it can have the provision to compensate hysteresis effect by normalization [24, 25].
The above-stated factors are the reasons why industrial CT scan generally uses step-
by-step rotation scan but is there any possibility to use continuous scanning mode in
industrial CT.
This paper analyzes the feasibility of continuous rotation scanning modality in
industrial cone-beam computed tomography. It describes the development of two
applications for the evaluation of two scanning modalities as elaborated above.
Multiple sets of two-dimensional (2D) projection images were acquired by the flat-
panel detector with both scanning modalities. Frame exposure time, X-ray voltage
and current were kept identical to get similar image qualities for comparative evalu-
ation. The primary concern of this paper is to identify the minimum rotation motion
time limit for a defined X-ray energy and intensity and single frame acquisition time.
This paper also discussed the necessity of sampling to reduce data size associated
with continuous rotation scanning in industrial CBCT.
2 Development
An integrated computed tomography system has been developed for the step-by-step
rotation scanning and continuous rotation scanning of an object, which is shown in
Fig. 1. Figures 2 and 3 show the user interface developed for step-by-step rotation
scanning and continuous-rotation scanning, respectively, to capacitate user to acquire
data in different modes with the different variable of parameters (Table 1).
In step-by-step mode, the motor rotates by a given angular step, and the detector
acquires projections when the object is stationary. This process continued till the
object rotates to user-defined final projection angle. The application developed as
98 A. Mitra et al.
shown in Fig. 2 facilitates the user to set the required initial projection angle, step
angle and final projection angle to accomplice step-by-step acquisition. User can set
single-frame acquisition time as per the requirement.
In continuous scanning mode, object rotation and data acquisition are done simul-
taneously. The application as shown in Fig. 3 provides the user to set only object
rotation speed to increase or decrease total acquisition time. Single frame acquisition
time is set on the detector.
A Comparative Evaluation of Two Scanning Modalities … 99
3 Experimental Data
One dataset with start-Stop (SS) mode and ten sets of data with continuous scanning
modalities by varying speed of rotations have been acquired. X-ray kV and amp
along with single frame acquisition time are given in Table 2. Radiographs acquired
at 0°, 90°, 180° and 270° are shown for all scanning modalities (Figs. 4, 5, 6, 7, 8, 9,
10, 11, 12, 13 and 14).
SNR at the center and four corners of each radiograph have been measured. These
are given in Table 3. A line profile tool is used to assess the variation of intensity
value along the width of radiograph as shown in Fig. 15. Minimum, maximum,
mean, median and standard deviation along the line profile were measured and given
in Table 3.
From Table 3, it is evident that statistical values in radiographic data of all modal-
ities are very similar. This can be further confirmed by the ratio of standard deviation
(σ) and mean value (μ) of each parameter. Ratios of standard deviation and mean
value are given in the last row of Table 3.
Quantitative analysis shows that SNR of radiographs acquired through all
modality is at par with the minimum SNR required for tomography. Line profile
comparison indicates that all these radiographs are having the same motion effects
and can be used for tomographic reconstruction.
Determination of 360° image: Minima of Mean square error (MSE) method have
been used to determine 360°th image. Mean square error method takes the mean of
squared value of the pixel to pixel difference of 2 images. It is mentioned in the
introduction that 0°th and 360°th image are very similar image. This implies that if
the 0°th image is compared with the rest of the images than the image whose mean
square error with the 0°th image is minimum will correspond to the 360°th image.
Using this method, last 90–300 images have been compared through 0°th degree
image to find 360°th image. Figures 16, 17, 18, 19, 20, 21, 22, 23, 24 and 25 show
the graph of MSE value of different images acquired at a total acquisition time of
48 s, 63 s, 92 s, 183 s, 362 s, 900 s, 1029 s, 1200 s, 1440 s and 1800 s, respectively.
104
Fig. 16 Mean square error of 0th image and last 90 images of continuous scan acquired in 48 s
360°th image was identified with MSE. Table 4 indicates the estimated 360°th
image and their respective mean square error.
Sampling of data: The datasets generated by continuous scans have images
around 200–7500. Except for few datasets, most of these datasets if fed directly
to reconstruction software either take a lot of processing time or maybe outside its
capacity due to hardware and software limitations. It further increases the reconstruc-
tion time, which will kill the main purpose of reducing scanning time as scanning
106 A. Mitra et al.
Fig. 17 Mean square error of 0th image and last 90 images of continuous scan acquired in 63 s
Fig. 18 Mean square error of 0th image and last 90 images of continuous scan acquired in 92 s
time plus reconstruction time is the total time in the CT system. Therefore, a limited
set of images have been extracted at regular interval from the stack generated in
the continuous scan. Sampling interval was so chosen that the final sampled image
should be close to the estimated 360°th image to accumulate 2π data. This provides
A Comparative Evaluation of Two Scanning Modalities … 107
Fig. 19 Mean square error of 0th image and last 90 images of continuous scan acquired in 183 s
Fig. 20 Mean square error of 0th image and last 90 images of continuous scan acquired in 362 s
a set of around 360–370 images. Table 3 contains the details of sampling interval
and the total number of images after sampling (Table 5).
Sinogram generation and back projection: Sinograms were prepared after
few pre-processing on reduced number of datasets without any correction and with
360°th angle correction. These sinograms are then back-projected to get tomographs.
108 A. Mitra et al.
Fig. 21 Mean square error of 0th image and last 300 images of continuous scan acquired in 900 s
Fig. 22 Mean square error of 0th image and last 300 images of continuous scan acquired in 1029 s
Fig. 23 Mean square error of 0th image and last 300 images of continuous scan acquired in 1200 s
Fig. 24 Mean square error of 0th image and last 300 images of continuous scan acquired in 1440 s
5 Results
See Figs. 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 and 47.
110 A. Mitra et al.
Fig. 25 Mean square error of 0th image and last 300 images of continuous scan acquired in 1800s
Table 5 Sampling interval, reduced number of images and final image angle
Scan Total number of Sampling Final image Total number of Final image
modalities images after interval images after angle
360°th sampling
correction
Continuous 186 Not 186 186 360
48 s required
Continuous 245 Not 245 245 360
63 s required
Continuous 362 Not 362 362 360
92 s required
Continuous 726 2 363 363 360
183 s
Continuous 1443 4 360 360 358
362 s
Continuous 3607 10 3600 3600 359.301
900 s
Continuous 4122 11 4114 374 359.214
1029 s
Continuous 4809 13 4797 369 359.027
1200 s
Continuous 5770 16 5760 360 359.314
1440 s
Continuous 7213 20 7200 360 359.301
1800s
Fig. 30 Continuous scan @ 183 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
Fig. 31 Continuous scan @ 362 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
A Comparative Evaluation of Two Scanning Modalities … 113
Fig. 32 Continuous scan @ 900 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
Fig. 33 Continuous scan @ 1029 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
Fig. 34 Continuous scan @ 1200 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
Fig. 35 Continuous scan @ 1440 s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
114 A. Mitra et al.
Fig. 36 Continuous scan @ 1800s images a Incorrect sinogram b Corrected sinogram c Flawed
tomograph d Corrected tomograph
Fig. 41 Continuous scan @ 183 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
Fig. 42 Continuous scan @ 362 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
Fig. 43 Continuous scan @ 900 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
Fig. 44 Continuous scan @ 1029 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
116 A. Mitra et al.
Fig. 45 Continuous scan @ 1200 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
Fig. 46 Continuous scan @ 1400 s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
Fig. 47 Continuous scan @ 1800s scan images a Radiograph b Sinogram c Tomograph d 3D front
view e 3D inclined view f 3D top view
6 Conclusion
The result shown in Figures 38, 39 and 40 shows shading artifact due to motion
blurriness. It can be concluded from this result that for X-ray energy and intensity of
200 kV and 3.2 mA, respectively, and single frame acquisition time of 249.884 ms, if
the time of 2π rotation motion is lesser than or equal to 92 s than it is not feasible to
acquire data through continuous scan. Figure 41 doesn’t have a significant shading
artifact but it has double wall artifact; therefore, this 2π rotation motion time cannot
be recommended for continuous scanning mode. Result from 42 to 47 precisely indi-
cates that continuous scan modality can be used in industrial tomography to reduce
total acquisition time without deteriorating reconstructed image quality. The above
statement is true provided X-ray energy and intensity are 200 kV and 3.2 mA, respec-
tively, single frame acquisition time should be 249.884 ms. It can be concluded that
for the above-mentioned acquisition parameter, application of continuous scanning
mode is feasible if the time of 2π rotation motion is higher than 360 s. Total acqui-
sition time, in this case, has been reduced from 3600 to 362 s, which is roughly 10
times. The scanning time can be further reduced with a high-intensity X-ray system
and high-speed manipulators. Some custom filters may be required for the reduction
of motion blurriness if further attempts were made to reduce scanning time with the
currently available systems.
References
Abstract A novel transient eddy current NDE system has already been developed
with double “D” excitation coil and fluxgate as a receiver. By using this system, an
artificially engineered subsurface defect in an aluminium plate has been detected by
shifting the defect from a depth of 2 mm to 20 mm and by maintaining the total
thickness of the plate of 21.5 mm. The defect size is 70 mm long, 1 mm wide and
1.5 mm high. The feature maximum diffusion time, t m that is the time at which
the diffusion of eddy current is maximum has been obtained and the same has been
correlated with the defect depth. Subsequently, an additional novel technique has been
established to measure the thickness of the conducting materials by combining the
transient eddy current NDE measurements and apparent conductivity calculations.
Here, two features such as maximum diffusion time, t m , and maximum apparent
conductivity, σ m app , have been obtained. From this, another parameter maximum
diffusion depth, δ m , has been calculated and the same has been correlated with the
thickness of the aluminium plates The parameters such as t m , σ m app and δ m are related
to the fundamental expression of diffusion depth (δ = (2t/σ μ)1/2 ) in time domain
eddy current measurements. This work has been further extended to detect the defects
in aluminium plates by combining transient eddy current NDE measurements and
apparent conductivity calculations. From this work, it has been shown that the depth
of the defect is proportional to the maximum diffusion depth, δ m .
1 Introduction
2 Principle of Operation
In transient eddy current NDE, the excitation current in the form of trapezoid is
applied to the excitation coil. During the ON time, the current reaches a steady
value. This current produces a primary magnetic field. The schematic of the tran-
sient waveforms is shown in Fig. 1. When this current is abruptly switched off,
Subsurface Defect Detection in Conducting Materials … 125
The NDE probe comprises an excitation coil in the form of absolute circular and
pickup coil in the form of differential. Both are fabricated from fibre reinforced
plastic material. The mean diameter and the height of the excitation coil are 32 mm
and 15 mm, respectively. The excitation coil has been wound with a copper wire
with a diameter of 0.38 mm and number of turns 260. The pickup coil has been
wound in differential configuration and arranged in such a way that the net primary
magnetic field coupled to the pickup coil is zero. With this arrangement, one can
apply a large excitation current of more than 1 A to induce more eddy currents in the
material without saturating the amplifier. The mean diameter and the baseline of the
pickup coil are 14 mm and 11 mm, respectively. The diameter of the wire used for
winding the pickup coil is 0.274 mm, and the number of turns in each coil is 250.
The cross-sectional view of the NDE probe is shown in Fig. 2.
The experimental setup for transient eddy current NDE measurement is shown in
Fig. 3. The excitation current of 1 A in the form of trapezoid with pulse repetition
of 400 ms (equivalent base frequency of 2.5 Hz) and duty cycle of 50% is applied to
the excitation coil by using a transmitter (ref. Fig. 1). The transmitter is controlled
by a transmitter controller, which controls the parameters such as base frequency,
duty cycle, etc. The transmitter controller is equipped with a crystal clock, which is
14 mm
32 mm
Data acquisition
Synchronization
Amp
System
Transmitter
Transmitter
controller
24 V Battery
Fig. 3 The schematic view of the transient eddy current NDE experimental setup
Subsurface Defect Detection in Conducting Materials … 127
synchronized with the data acquisition system before commencing the experiments.
The output voltage of the pickup coil is amplified and given to the data acquisition
system to record the decay transients. The recorded raw data are processed by the
built-in software available with the data acquisition system, which includes stacking,
averaging, window averaging, and finally, the decay of the output voltage with respect
to time is displayed. The transmitter, transmitter controller and the data acquisition
system used in this work are the same ones used for geophysical applications and
have been procured from Zonge International and EMIT, respectively.
In this work, transient eddy current NDE measurements were carried out to detect
the artificially engineered defect in aluminium plates by shifting the defect in the
stack from surface to depth of 12 mm in steps of 2 mm and by maintaining the total
plate thickness of 13.5 mm. The stack consisted of six identical aluminium plates
of length, width and thickness of 200 mm, 160 mm and 2 mm, respectively. Along
with these plates, an aluminium plate of thickness 1.5 mm with a defect of length
70 mm and 1 mm width has been used. The schematic view of the plate with defect
(the top and front views) and the stacked view when the defect is at 6 mm below
the surface are shown in Fig. 4a, b. The NDE probe was placed above the stack
with lift-off distance 1 mm. An excitation current of 1 A with base frequency of
200 mm
70 mm 160 mm
Top view 1 mm
200 mm
13.5 mm
(b)
Fig. 4 The schematic view of the stacked Aluminium plates with defects a The top and front view
of the plate with defect and b Stacked view when the defect is at 6 mm below the surface
128 L. Bisht et al.
2.5 Hz was passed through the excitation coil. The decay of the secondary magnetic
field during the off time of the transmitter current is recorded by the data acquisition
system with respect to time. The defect plate located on upper surface of the stack
was scanned from one end to other end with step size of 2 mm and total scanning
length of 160 mm (80 mm on each side of the defect). The same experiments were
repeated by shifting the defect to different depths in the stack. The voltage transients
recorded at times for a homogeneous half-space with conductivity, σ for the upper
layer are given by Spies [12]
5/2
I σ 3/2 μ0 a 2
V = √ (1)
20 π t 5/2
Here, a is the radius of the excitation coil, I is the excitation current, μ0 and
σ are the magnetic permeability of the free space and electrical conductivity of
the metal plates, respectively. The apparent conductivity for the recorded voltage
transients at each location in the scanning line, inversion has been done by using the
expression (1). The recorded decay transient at a defect free region and just above
the centre of the defect and the corresponding apparent conductivity are given in
Fig. 5a, b, respectively, for the defect located at the surface. From this (Fig. 5b),
the maximum apparent conductivity, σ m app , and the corresponding time known as
maximum diffusion time, t m , has been obtained at each location on the scanning
line. The maximum diffusion time, t m , and the corresponding maximum apparent
conductivity, σ m app , for the defects located at different depths ranging from surface
to 12 mm are given in Fig. 6a, b, respectively.
From Fig. 6a, b, one can observe that there are large fluctuations for the defect
located near the surface (say surface, 2–4 mm), which is due to short baseline of the
pickup coil. For the defect located near the surface, the strength of the differential
signal coupled to the pickup coil is low as compared to the defect located in deep
surface (say 6–12 mm). In addition to this, the electrical conductivity of the defect
plate is different from the rest of the other plates in the stack. Due to this, a substan-
tial offset has appeared in maximum diffusion time, t m , and maximum apparent
conductivity, σ m app , plots in the defect-free region.
Finally, the maximum values of the two parameters t m and σ m app obtained from
the plots (Fig. 6a, b) for defects at different depth are used to calculate the maximum
diffusion depth, δ m using the standard expression (2) and same has been given in
Fig. 7. From this, it is observed that.
2t
δ = (2)
σμ
for the defects located below 6–12 mm, the peak values of the δ m are proportional to
the defect depth. The baseline of the present pickup coil used in this experiment is not
suitable for the defects located at surface as well as near the surface. Therefore, the
Subsurface Defect Detection in Conducting Materials … 129
Fig. 5 The recorded decay transient at a defect free region and just above the centre of the defect
and the corresponding apparent conductivity are shown in (a) and (b) respectively for the defect
located at the surface. The inset shows the enlarged view of the transient at earlier time
optimization of baseline for the pickup coil used for this kind of study is inevitable.
Further work is underway to optimize the baseline of the pickup coil in order to
achieve linear relationship between δ m and defect depth for surface defects as well
as deep-lying defects in the conducting material.
5 Conclusion
A method that had been established earlier for thickness measurement in conducting
material by using the combination of transient eddy current NDE and apparent
conductivity calculations provides a linear relationship between the plate thickness
130 L. Bisht et al.
Fig. 6 The maximum diffusion time, t m and the corresponding maximum apparent conductivity,
σ m app for the defects located at different depths from surface to 12 mm are shown in (a) and
(b) respectively
Fig. 7 The maximum diffusion depth, δ m obtained for the defect located at different depths from
surface to 12 mm. The dashed red line indicates the linear relationship between the peak values of
δ m and defect depth
and the maximum diffusion depth, δ m [11]. Such technique has been implemented
for the present defect detection in conducting materials. Experiments have been
performed with circular excitation coil and differential pickup coil in order to apply
large excitation current to induce large eddy currents in the conducting material. This
may enhance the signal to noise ratio of the system. With the present design of NDE
probe, it has been shown that the defect depth is proportional to the peak values of
the maximum diffusion depth for the defect depth of 6–12 mm. This method will be
useful for defect detection and characterization in conducting materials with further
improvements.
Acknowledgements The authors would like to thank Shri R. Baskaran, Dr. N. V. Chandra Shekar
and Dr. Shaju K. Albert for their continuous support and encouragement. The authors would also
like to thank Shri Ram and Shri Gautam for their continuous support throughout this work.
Subsurface Defect Detection in Conducting Materials … 131
References
1. Sophian A, Tian GY, Taylor D, Rudlin J (2001) Electromagnetic and eddy current NDT: a
review. Insight 43:302–306
2. Yang HC, Tai CC (2002) Pulsed eddy-current measurement of a conducting coating on a
magnetic metal plate. Meas Sci Technol 13:1259–1265. https://doi.org/10.1088/0957-0233/
13/8/313
3. Zhang K, He Y and Dong Z (2018) Pulsed eddy current nondestructive testing for defect
evaluation and imaging of automotive lightweight alloy materials. J Sens 1–11. https://doi.org/
10.1155/2018/1639387
4. Bieber JA, Shaligram SK, Rose JH, Moulder JC (1997) Time-gating of pulsed eddycur-
rent signals for defect characterization and discrimination in aircraft lap-joints. Rev Progr
QNDE16B:1915–21
5. He Y, Luo F, Pan M, Weng F, Hu X, Gao J, Liu B (2010) Pulsed eddy current technique for
defect detection in aircraft riveted structures. NDT&E Int 43(2):176–181. https://doi.org/10.
1016/j.ndteint.2009.10.010
6. Tian GY, Sopian A (2005) Defect classification using a new feature for pulsed eddycurrent
sensors. NDT&E Int 38:77–82. https://doi.org/10.1016/j.ndteint.2004.06.001
7. Tai CC, Rose JH, Moulder JC (1996) Thickness and conductivity of metallic layers from
pulsed eddy current measurements. Rev Sci Instrum 67:3965–3972. https://doi.org/10.1063/1.
1147300
8. Kral J, Smid R, Ramos HMG, Ribeiro AL (2011) Thickness measurement using transient eddy
current techniques. IEEE Int Instrum Measur Technol Conf. https://doi.org/10.1109/imtc.2011.
5944293
9. Mohanty I, Nagendran R, ThanikaiArasu AV, Baskaran R, Mani A (2018) Correlation of defect
depth with diffusion time of eddy currents for the defects in conducting materials by using
transient eddy current NDE. Meas Sci Technol. https://doi.org/10.1088/1361-6501/aad613
10. Panaitov G, Krause H-J, Zhang Y (2002) Pulsed eddy current transient technique with HTS
SQUID magnetometer for non-destructive evaluation. Phys C 278–281. https://doi.org/10.
1016/s0921-4534(02)00696-2
11. Nagendran R, Bisht L, Mohanty I (2021) A novel method for thickness measurement in
conducting materials by apparent conductivity calculation through transient eddy current NDE.
J Nondestr Eval
12. Spies BR (1989) Depth of investigation in electromagnetic sounding methods. Geophysics
54(7):872–888. https://doi.org/10.1190/1.1442716
13. Nagendran R, Mohanty I (2020) Characterization of TDEM system with SQUID and fluxgate
magnetometers for geophysical applications. IEEE Trans Appl Supercond 30(2):1–8. https://
doi.org/10.1109/tasc.2019.2957198
Crack Localization Using Improved
Continuous Wavelet Transform Features
Abstract In this paper, an enhanced damage diagnostic strategy for the localization
of subtle cracks is proposed using modal parameters. The proposed work applies
continuous wavelet transform (CWT) on the traditionally followed mode shape
difference between the structure with and without crack as crack localization index
(CLI). The CLI is further enhanced by augmentation of weighting function involving
natural frequencies. This weighting function emphasizes the modes that are sensitive
to crack, which improves the crack diagnosis process. Numerical studies and lab-
level experiments have been performed on structures with varied crack parameters
to validate the improved CWT-based proposed approach.
1 Introduction
The damage diagnostic method based on vibration measurements has the advan-
tage of non-destructive, and it can carry out effective damage identification for
various engineering applications, namely, damage identification in airplane body and
engine, bridge cracks detection, etc. The vibration-based techniques are preferred for
damage diagnosis as they are global methods, and it does not need apriori informa-
tion about the damage including its spatial location [1–4]. Generally, the process of
vibration-based structural damage detection consists of recording structural vibra-
tion, extracting modal properties, and identifying damages based on changes in struc-
tural dynamic properties. In vibration data recording, the source of excitation can
be generated from a specific vibrating machine, ambient loading, or operational
loading [5]. At present, identification methods based on vibration characteristics
P. Madankumar (B)
Cochin University of Science and Technology, Kochi, Kerala, India
J. Prawin
CSIR-Structural Engineering Research Center, CSIR Campus, Chennai, Tamil Nadu, India
can be divided into natural frequency, mode shape and its curvature, residual force
vector flexibility matrix, frequency response function, and modal strain energy. These
features are directly related to changes in mass, stiffness, and damping matrices asso-
ciated with physics-based models (correlated with the original system). A comparison
of the features extracted from the modal parameters of the structure in a pristine state
to the features obtained from the current data provides information about the pres-
ence of damage, its corresponding spatial location, and severity [6]. The change of
structural vibration mode contains more damage information, which has a significant
advantage in determining crack location. However, the traditional crack localization
index based on the mode shape difference approach between the measurements of
the healthy and damaged structure gives false results by showing more than one peak
apart from the damage location due to its sensitivity to noise, difficulty in identi-
fying subtle cracks, and crack location coinciding with the zero strain energy node
of the corresponding selected mode. The traditional crack localization index based
on CWT coefficient of mode shape difference is augmented by weighting functions
involving natural frequencies to locate more reliably and robustly the damage present
at any location in the beam-like structures overcoming the above-said limitations.
Numerical studies and experimental validation have been performed for the proposed
approach.
2 Proposed Strategy
Natural Frequencies and mode shapes are calculated in both the cases
Difference in mode shape is calculated between the structures with and without
cracks
E I0
E I (x) =
1 + C exp(−2a|x − x j |/d )
3 3
I0 − Icj w d − dcj w d − dcj
C= ; Icj = ; Icj =
Icj 12 12
d
lc = = 1.5d, (1)
α
where I 0 , I cj indicate the moment of inertia of the beam without and with crack.
Similarly, d, d cj indicate the depth of the beam without and with crack at the section
‘X’ (particular element of the Finite element model). The parameters x j and d repre-
sent the position and depth of the crack. The coefficient α is a constant, and lc is the
effective length.
The basic step of the frequency domain decomposition (FDD) approach [8] is
the estimation of the power spectral density (PSD) Gyy ( jω) observed at discrete
frequencies ω = ωi and then decomposition of the PSD matrix using singular value
decomposition:
136 P. Madankumar and J. Prawin
where Gyy and Gxx represent the PSD of output and input, respectively. The symbols
H and Ĥ indicate the frequency response function and its corresponding complex
conjugate. The parameter U i is the singular vectors unitary matrix, and Si is a singular
value diagonal matrix.
The natural frequencies and mode shapes are computed, respectively, from the
singular values and the singular vectors (i.e., U i matrix). The singular vectors with
the largest Modal Assurance Criterion (MAC) values are computed as modes. On
the other hand, if none of the singular vectors has a MAC larger than a threshold, the
search for matching parts of the spectral density is terminated.
The power spectral density of the response is related to modal parameters by
dk φk φ T T
d k φk φk
G yy ( jω) = k
+ (4)
k∈ω
jω − λk jω − λk
where φk and λk indicate the mode shape and modal participation factors of the ‘k’th
mode. The parameter d k is another scalar constant.
Wavelet transformation is the process of dividing the signal into shifted and scaled
forms of the mother wavelet. CWT transforms the signal from time domain to time
scale domain and better suited for handling nonstationary signals [9–11]. CWT
coefficients for a signal x(t) are mathematically given by Eq. (5)
(5)
Crack Localization Using Improved Continuous Wavelet Transform … 137
The transformation by CWT and the sample sine signal, its corresponding mother
wavelet, is shown in Figs. 2 and 3, respectively. The process of obtaining CWT
coefficients for a signal x(t) is presented below in the form of step-by-step procedure.
1. A mother wavelet shall be taken, and it shall be compared to a section at the
beginning part of the original signal.
2. C—a number, which represents the similarity of the wavelet to the section of
the signal is calculated. The similarity is more if the value of C is higher.
3. The whole signal shall be covered by shifting the wavelet to the right. The steps
1 and 2 shall be repeated.
4. Scale (stretch) the wavelet and steps 1 through 3 shall be repeated.
5. The steps 1–4 shall be repeated for all scales.
In the present work for CWT coefficient calculation, the mother wavelet chosen
is Daubechies mother wavelet, and total number of scales considered is 128. The
process of CWT coefficients computation is presented in Fig. 4.
The weighting function involving the ratio of the natural frequency of the structure
with and without crack is mathematically given by
2
λwithout crack
Weighting Function Wi = 1 − i with crack (6)
λi
138 P. Madankumar and J. Prawin
The modes that show higher variation in frequency are highly sensitive to
crack/damage and hence only considered. The mode shapes that do not show any
variation in their natural frequencies are almost neglected in the weighting function.
n
C L Ii = Ai Wi (7)
i=1
Ai = C W T (∅i )
undamaged damaged
∅i = ∅i − ∅i (8)
6 Numerical Investigations
7 Experimental Validation
are D—100 mm; flange W (Width)—60 mm; flange T (Thick)—7.2 mm; web T
(thick)—4 mm; area A—14.6 cm2 ; I xx = 257.5 cm4 ; I yy = 40.8 cm4 . The experi-
mental arrangement of the beam specimen is shown in Fig. 10. The first four natural
frequencies of the beam without damage or crack are measured as 36.37, 146.35,
328.65, 580.92 Hz.
The beam is instrumented with 15 PZT accelerometers and the traditional modal
shaker, the data acquisition system is used. 2400 Hz is the considered sampling
frequency. A total of 100 realizations (with varied environmental and operational
variability) with each of 5 s is measured.
Two different damage scenarios furnished in Table 1 are simulated. The natural
frequencies corresponding to damage scenarios D1 and D2 are also presented in
Table 1. The traditional crack localization index based on the difference of mode
shapes of the beam with and without damage estimated for damage scenarios D1
and D2 is presented in Fig. 11a, b corresponding to the third mode.
Similar to numerical observation, it can be concluded from Fig. 11 that the mode
shape difference fails to locate the damage.
The result of the CWT-based crack localization index for the damage scenarios
D1 and D2 is presented in Fig. 12a, b.
It is observed from Fig. 12a, b that the maximum value of the CLI for the D1
damage scenario is at the location node 4. This is same as the geometric location
of the crack. Similar to this, for the scenario D2, the crack localization index peaks
at both node 4 and node 10 because damages induced at two different locations.
Hence, it is concluded from the experimental investigation that the proposed crack
localization strategy can locate multiple damages or cracks located anywhere in the
structure.
8 Conclusion
In the proposed work, the traditional crack localization index based on CWT coef-
ficient of mode shape difference is augmented by weighting functions involving
144 P. Madankumar and J. Prawin
natural frequencies to locate more reliably and robustly the damage present at any
location in the beam-like structures. From the detailed studies, it is evident that
the crack location can be easily identified at the sensor node corresponding to the
maximum CLI value. The proposed CWT-based technique is capable of localizing
more than one crack present at different special coordinates along the beam span.
The values of the CWT-based CLI are also related to the severity of the crack. The
proposed CWT-based technique can identify very subtle cracks even using polluted
measurements.
Acknowledgements The first author acknowledges the support of professional and technical help
received including lab-level experiment from SERC—CSIR, Chennai.
References
1. Prawin J, Lakshmi K, Rama Mohan Rao A (2020) Structural damage diagnosis under varying
environmental conditions with very limited measurements. J Intell Mater Syst Struct 31(5):665–
686
2. Farahani RV, Penumadu D (2016) Damage identification of a full-scale five-girder bridge using
time-series analysis of vibration data. Eng Struct 115:129–139
3. Husain NA, Ouyang H (2011) Detection of damage in welded structure using experimental
modal data. J Phys Conf Ser 305(1):012120
4. Shabbir F, Khan MI, Ahmad N, Tahir MF, Ejaz N, Hussain J (2017) Structural damage detection
with different objective functions in noisy conditions using an evolutionary algorithm. Appl
Sci 7(12):1245
5. Prawin J, Rama Mohan Rao A (2019) A method for detecting damage-induced nonlinearity
in structures using weighting function augmented curvature approach. Struct Health Monit
18(4):1154–1167
6. Prawin J, Rama Mohan Rao A (2019) Reference-free breathing crack identification of beam-
like structures using an enhanced spatial Fourier power spectrum with exponential weighting
functions. Int J Struc Stab Dyn 19(02):1950017
7. Sinha JK, Friswell MI, Edwards S (2002) Simplified models for the location of cracks in beam
structures using measured vibration data. J Sound Vib 251(1):13–38
8. Brincker R, Zhang L, Andersen P (2001) Modal identification of output-only systems using
frequency domain decomposition. Smart Mater Struct 10(3):441–451
9. Solís M, Algaba M, Galvín P (2013) Continuous wavelet analysis of mode shapes differences
for damage detection. Mech Syst Signal Process 40(2):645–666
10. Taha MR, Noureldin A, Lucero JL, Baca TJ (2006) Wavelet transform for structural health
monitoring: a compendium of uses and features. Struct Health Monit 5(3):267–295
11. Kim H, Melhem H (2004) Damage detection of structures by wavelet analysis. Eng Struct
26(3):347–362
An Enhanced Vibration-Based Damage
Diagnostic Strategy Under Changing
Environmental Conditions
1 Introduction
recording, the source of excitation can be generated from a specific vibrating machine,
ambient loading, or operational loading [5]. At present, identification methods based
on vibration characteristics can be divided into natural frequency, mode shape, mode
curvature, residual force vector flexibility matrix, frequency response function, and
modal strain energy. These features are directly related to changes in mass, stiffness,
and damping matrices associated with physics-based models (correlated with the
original system). Comparison of the features extracted from the modal parameters of
the structure in a pristine state to the features obtained from the current data provides
information about the presence of damage, its corresponding spatial location, and
severity [6]. The change of structural vibration mode contains more damage informa-
tion, which has a significant advantage in determining crack location. However, the
conventional damage localization index (DLI) based on the mode shape difference
approach gives false results by showing more than one peak apart from the damage
location due to its sensitivity to noise, difficulty in identifying subtle cracks, and crack
location coinciding with the zero strain energy node of the corresponding selected
mode. Distinguishing the variations in the modal parameters is difficult to conclude
as due to damage or due to confounding factors such as temperature, humidity, and
operational variations in loadings.
In the proposed enhanced vibration-based damage diagnostic strategy, the pres-
ence of the damage in the structure is confirmed by the deviation of the damage index
from one. The damage index for detection is based on the ratio of the frequency of
the structure with localized damage to the undamaged structure between the two
modes. The proposed damage index is highly reliable for damage detection as it is
insensitive to variations due to changing environmental and operational conditions.
The traditional mode shape difference, enhanced with the Katz fractal dimension, is
proposed as a damage localization index (DLI) to find the geometric location of the
damage. The magnitude of the Katz Fractal Dimension (KFD)-based DLI is related
to the relative severity of the damage.
The organization of the paper is given below: Frequency Domain Decomposition
method and Katz Fractal Dimension are presented in Sects. 2 and 3. Section 4 outlines
the detailed step by step procedure of the proposed technique. The validation of
the vibration-based approach presented in this article is performed using numerical
investigations on simple beam-like structures in Sect. 5, and the conclusions are
presented in Sect. 6.
The basic step of the Frequency Domain Decomposition (FDD) approach [7] is the
estimation of the power spectral density (PSD) Gyy ( jω) at discrete frequencies ω =
ωi and then decomposition of the PSD matrix using singular value decomposition:
where Gyy and Gxx represent the PSD of output and input, respectively. The symbols
H and Ĥ indicate the frequency response function and its corresponding complex
conjugate. The parameter U i is the singular vector unitary matrix, and S i is a singular
value diagonal matrix.
The natural frequencies and mode shapes are computed, respectively, from the
singular values and the singular vectors (i.e., U i matrix). The singular vectors with
the largest Modal Assurance Criterion (MAC) values are computed as modes. On
the other hand, if none of the singular vectors has a MAC larger than a threshold, the
search for matching parts of the spectral density is terminated.
The power spectral density of the response is related to modal parameters by
dk φk φ T T
d k φk φk
G yy ( jω) = k
+ (3)
k∈ω
jω − λk jω − λk
where φk and λk indicate the mode shape and modal participation factors of the ‘k’th
mode. The parameter d k is scalar constant.
log10 (n)
F Dn (xi ) = (4)
d(xi )
log10 L(x i)
+ log10 (n)
where n and L indicate the total number of sampling points and the length of the
beam or the size of the mode shape. L(x i ) represents the length of the sliding window.
The parameter d(x i ) indicates the diameter or the distance between the first sampling
point to the ‘i’th sampling point with the farthest distance. Even though there are
many types of fractal dimension techniques, the present work considers Katz’s fractal
dimension as it is stated to give consistent results.
KFD is applied to the mode shape to show the local variation of the geometric
intricacy and as well to magnify the irregularity induced by the crack at the damage
location. The KFD magnitude at the damage location is expected to be greater than
one, and its magnitude provides a relative indication of the severity of the damage
[8, 9].
148 V. Sri Harika and J. Prawin
4 Proposed Strategy
2. The modal features such as natural frequencies and mode shapes are computed
by any of the established operational modal analysis techniques. The popular
frequency domain decomposition is considered in the present investigations.
These details are presented in earlier Sect. 2.
3. Estimate the following Damage index:
(ficurrent /fireference )
Damage index = (5)
(fcurrent
j /freference
j )
where ficurrent , fireference represent the natural frequency corresponding to the ‘i’
th mode estimated from normal and current (may or may not be damaged) state
of the structure.
4. If the damage index estimated from Step 3 is equal to 1, then the structure does
not undergo any damage. If the magnitude deviates from 1, then the structure is
in the damaged state.
5. The mode shape difference is computed as follows
undamaged damaged
φi = φi − φi (6)
DL Ik = KFD(φi (k : k + 2));
DLI = Damage Localization Index;
KFD = katz Fractal Dimension;
i = mode number;
k = sensor node (7)
The details related to the Katz Fractal Dimension are already presented in
earlier Sect. 3.
7. The position of the maximum value of DLI (damage localization index) esti-
mated from previous step 6 [i.e., through Eq. (7)] gives the spatial location of
the damage.
8. The magnitude of the DLI at the identified damage location quantifies the
severity of the damage.
The numerical investigation has been executed on a simply supported beam to verify
and validate the proposed algorithm. The properties of the beam are length = 2.5 m;
cross-section = 0.05 × 0.1 m; elastic modulus = 20.5 GPa, and density = 7850 kg/m3 .
150 V. Sri Harika and J. Prawin
The beam is modeled using 1D Euler beam elements. In the present work, the 2.5 m
beam is idealized with 40 elements (axial, translational, and rotational degrees of
freedom per node) where each element length is 0.0625 m. The acceleration time
history measurements at varied locations are computed using the Newmark time
integration algorithm. For sensitivity analysis of the damage indices, presented in
this paper, to noise, the computed acceleration time history responses are corrupted
with SNR of 40 before processing by frequency domain decomposition technique to
compute modal frequencies and mode shapes. Both the results of the damage indices
using noise-free and polluted measurements are presented for comparison and for
verifying the robustness and efficiency of the algorithm.
Several modeling techniques are available in the literature to simulate crack in the
beam-type structure, the present work uses the popular Sinha damage model. The
Sinha damage-related modeling is only briefly reported, and more details can be
found in [10]. The Flexural rigidity (EI) at each section ‘x’ through Sinha damage
modeling is given by
E I0
E I (x) =
1 + C exp(−2a|x − x j |/d )
3 3
I0 − Icj w d − dcj w d − dcj
C= ; Icj = ; Icj =
Icj 12 12
d
lc = = 1.5d, (8)
α
where I 0 , I cj indicate the moment of inertia of the healthy beam and the beam with
simulated localized damage. Similarly, d, d cj indicate the depth at the section ‘X’ (at
a particular element of the Finite element model). The parameters x j and d represent
the position and depth of the crack. The coefficient α is a constant, and lc is the
effective length.
To verify and validate the proposed damage diagnostic strategy, the following four
damage scenarios are considered
D1—0.5 m (element affected are 7–10) from left-hand support with a 5% reduction
in moment of inertia;
An Enhanced Vibration-Based Damage Diagnostic … 151
D2—0.5 m (elements affected are 7–10) from left-hand support with a 20%
reduction in moment of inertia;
D3—0.95 m (elements affected are 14–17) from left-hand support with a 20%
reduction in moment of inertia;
D4—2 m (elements affected are 31–34) from left-hand support with a 20%
reduction in moment of inertia.
The cracked moment of inertia for the four above damage scenarios is presented
in Fig. 2 for easier visualization.
The Fourier spectrum of the 10th node response of the considered simply supported
beam without any damage is presented in Fig. 3. The computed Fourier spectrum
plotted in Fig. 3 shows a peak at 36.62, 148.3, 331.4, and 585.3 Hz coinciding with
the true modal frequencies of the healthy simply supported beam.
The damage indices estimated using Eq. (5) for the four damage scenarios D1, D2,
D3, and D4 are presented in Table 1. Table 1 depicts that the damage index magnitude
deviates away from 1 for all the damage scenarios. It can be noticed from Table 1
that the results corresponding to D2 and D4 are the same due to symmetry of damage
and the same amount of reduction in moment of inertia. Table 1 demonstrates that
the deviation of the damage index based on frequency ratios of damage scenario D2
is higher than D1 due to large stiffness reduction at the same location.
152 V. Sri Harika and J. Prawin
The traditional damage localization index based on the conventional mode shape
difference formula estimated using Eq. (6) for damage scenarios D1 and D2 utilizing
polluted acceleration measurements is presented in Fig. 4 corresponding to the second
mode. Similarly, the second mode results corresponding to damage scenarios D3 and
D4 are shown in Figs. 5 and 6. It should be mentioned here that the proposed algorithm
works effectively irrespective of the selected mode, however, the results of the second
mode are only presented in this paper for illustration.
It can be observed from Figs. 4, 5 and 6, even though peaks occur at the location of
the damage, there exist other peaks at other locations. This results in misinterpretation
of the damage. This is due to measurement noise, zero strain energy nodes of the
selected mode, and low intensity of the damage. Hence, the mode shape difference
between the two states of the structure cannot be employed for locating the damage.
To overcome this, the present work uses Katz fractal dimension to magnify the
discontinuities due to the damage in the vibration measurements even in the case of
a noisy environment. KFD is applied to the difference of undamaged and damaged
mode shape values for each mode to improve the damage localization process.
An Enhanced Vibration-Based Damage Diagnostic … 153
Fig. 4 Damage
scenario—D1 and
D2—mode shape
difference-based index of the
beam having damage at
0.5 m with 5% damage and
20% damage
Fig. 5 Damage
scenario—D3—mode shape
difference-based index of the
beam having at 0.95 m with
20% damage
The improved results by Katz Fractal dimension for the first two damage scenarios
corresponding to the second mode are given in Fig. 7. The second mode results for
the third and fourth damage scenarios are given in Figs. 8 and 9. The plots presented
in Figs. 7, 8 and 9 utilize noise-free measurements.
It can be observed from the results presented in Figs. 7, 8 and 9 that the proposed
KFD-based DLI shows a peak only at damage location for all the damage scenarios.
It can also be observed from Figs. 7 and 9 corresponding to damage scenarios of
symmetry damage that KFD plots are complementary to one another. This coincides
with the true physical behavior of the simply supported beam. Therefore, the results
based on KFD conclude that the proposed enhanced vibration-based damage diag-
nostic strategy can find the geometric location of the damage very accurately. Figure 7
also demonstrates that the magnitude of damage localization index increases with
154 V. Sri Harika and J. Prawin
Fig. 6 Damage
scenario—D4 mode shape
difference-based index of the
beam having at 2 m with
20% damage
Fig. 10 Damage
localization (D1 and D2) by
Katz’s fractal dimension for
5% damage and 20%
damage at 0.5 m with noise
(SNR = 40 dB)
156 V. Sri Harika and J. Prawin
Fig. 11 Damage
localization (D3 damage
scenario) by Katz’s fractal
dimension for 20% damage
at 0.95 m with noise (SNR =
40 dB)
Fig. 12 Damage
localization (D4 Scenario) by
Katz’s fractal dimension for
mode 2 20% damage at 2 m
with noise (SNR = 40 dB)
6 Conclusions
References
Abstract Thermal Non-destructive Testing and Evaluation is one of the widely used
non-destructive inspection methods due to its inherent merits such as remote, full
field, safe, and quantitative inspection capabilities for evaluation of wide variety of
materials. Among the various thermal non-destructive testing modalities recently
proposed matched filter based non-periodic infrared thermographic approaches
gained their importance due to their higher test resolution and sensitivity for detec-
tion of hidden defects in the test material. Further, feasibility to implement with
moderate peak power heat sources in a limited span of time in comparison with
conventional thermographic techniques makes these pulse compression favorable
techniques more economical and reliable. The present manuscript demonstrates the
advantages of pulse compression favorable digitized frequency modulated thermal
wave imaging approach for identification of defects in a steel material. The obtained
results have been compared with widely used conventional thermographic post
processing approaches by taking the signal to noise ratio as a figure of merit.
R. Mulaveesala (B)
Centre for Sensors, iNstrumentation and Cyber-Physical Systems Engineering (SeNSE), Indian
Institute of Technology Delhi, Hauz Khas, New Delhi 110016, India
e-mail: mulaveesala@sense.iitd.ac.in
G. Dua
Thapar Institute of Engineering & Technology, Patiala, Punjab, India
V. Arora
Indian Institute of Information Technology Una, Una, Himachal Pradesh, India
1 Introduction
In this work a 3D finite element analysis (FEA) has been carried out on a steel
sample using COMSOL Multiphysics. The mild steel sample with six inclusions
{Air, Calcium Fluoride (CaF2 ), Calcium Oxide (CaO), Titanium dioxide (TiO2 ),
Aluminium Oxide (Al2 O3 ), and Magnesium Oxide (MgO)} (shown in Fig. 1) has
been modeled with a finer mesh using 3D tetrahedral elements.
Digitized Frequency Modulated Thermal Wave … 161
The FEA was carried out by imposing a DFM heat flux over the surface of the test–
material, and the resultant surface thermal response has been captured at a frequency
of 25 Hz. The simulations were carried out under adiabatic boundary conditions,
with the sample at an ambient temperature of 300 K. The thermal properties of the
sample and the introduced inclusions are as given in Table 1.
The simulated data has been further processed using frequency domain analysis
and pulse compression for the detection of inclusions.
The captured thermal distribution profile after mean removal is further analyzed in
frequency domain using discrete Fourier transform (DFT) to extract the phase and
amplitude information from its complex components. One-dimensional DFT T (k) is
computed for each pixel in the field of view using a well-known formula [5–8]:
162 R. Mulaveesala et al.
N −1
j2πkn
T (k) = T (x, t)e− N = Re(T (k)) + jImg(T (k)), (1)
n=0
where T (x, t) denotes the captured thermal profile Re(T (k)) and Img(T (k)) are the
real and imaginary parts of T (k) respectively.
The amplitude images are then reconstructed from the computed real and
imaginary components as:
|T (k)| = (Re(T (k)))2 + (Img(T (k)))2 . (2)
Further, the phase images are reconstructed by using the real and imaginary
components as follows:
−1 Img(T (k))
∠T (k) = tan . (3)
Re(T (k))
Pulse compression P(τ ) is computed from the circular convolution between the
recorded thermal distribution response (T (x, t)) for each pixel in the field of view
with that of thermal profile of a chosen reference pixel (Ref (x i,j , t)). It is obtained as
[8–19]:
where (Ref (f )) and (T (f )) are the Fourier transforms of the reference temperature
signal and the captured temperature response. −1 and ∗ denote the inverse Fourier
transform and complex conjugate operators, respectively.
Pulse compression has the advantage of concentrating the total supplied energy
into a localized time instant which improves depth resolution with enhanced signal
to noise ratio.
The present work highlights the capability of DFMTWI approach in visualizing the
inclusions present in a structural mild steel sample. In this approach, the test material
is allowed to undergo a known controlled digitized version of frequency modulated
Digitized Frequency Modulated Thermal Wave … 163
thermal stimulation sweeping their fundamental frequency range from 0.01 to 0.1 Hz
in 100 s, and the corresponding thermal response over the surface is captured. The
temporal mean raise in captured thermal response is removed using an appropriate
fit. Above mentioned time and frequency domain post processing techniques have
been implemented onto the mean removed thermal response, and the results obtained
are shown in Fig. 2. The conventional frequency domain amplitude image is shown
in Fig. 2a, whereas the frequency domain phase image is shown in Fig. 2b. The
resultant pulse compressed image is as shown in Fig. 2c.
Results obtained clearly shows that material with different thermal properties
gives different thermal contrasts therefore can be easily utilized for identification of
sub-surface feature characteristic and its properties.
Further, thermal contrasts have been quantified in terms of the materials SNR as
computed using:
Mean of region of inclusion − Mean of region of sound material
SNR = 20log .
Standard deviation of the sound region
(5)
SNRs have been computed for different inclusions in the mild steel sample for
images obtained with different analysis methods as presented in Table 2.
It is proved that the proposed time domain pulse compression analysis scheme is
having a better noise intensive detection due to the adopted correlation based matched
filter approach which leads to the higher SNRs in comparison to that of frequency
domain analysis.
4 Conclusion
Fig. 2 Results obtained for DFMTWI technique for frequency and time domain transform tech-
niques a frequency domain amplitude image, b frequency domain phase image, and c pulse
compressed image
Digitized Frequency Modulated Thermal Wave … 165
References
1. Maldague XPV (2001) Theory and practice of infrared thermography for nondestructive testing.
Wiley, New York
2. Almond DP, Lau SK (1994) Defect sizing by transient thermography I: an analytical treatment.
J Phys D Appl Phys 27(5):1063–1069
3. Busse G (1979) Optoacoustic phase angle measurement for probing a metal. Appl Phys Lett
35(10):759–760
4. Maldague X, Marinetti S (2003) Pulse phase infrared thermography. Rev Sci Instrum 74(1
II):417–419
5. Mulaveesala R, Tuli S (2006) Theory of frequency modulated thermal wave imaging for
nondestructive subsurface defect detection. Appl Phys Lett 89(19), art.no. 91913
6. Mulaveesala R, Pal P, Tuli S (2006) Interface study of bonded wafers by digitized linear
frequency modulated thermal wave imaging. Sens Actuators A 128(1):209–216
7. Mulaveesala R, Tuli S (2005) Digitized frequency modulated thermal wave imaging for
nondestructive testing. Mater Eval 63(10):1046–1050
8. Mulaveesala R, Vaddi JS, Singh P (2008) Pulse compression approach to infrared nondestruc-
tive characterization. Rev Sci Instrum 79(9), art. no. 094901
9. Ghali VS, Jonnalagadda N, Mulaveesala R (2009) Three-dimensional pulse compression for
infrared nondestructive testing. IEEE Sens J 9(7):832–833
10. Tabatabaei N, Mandelis A, Amaechi BT (2011) Thermophotonic radar imaging: an emissivity-
normalized modality with advantages over phase lock-in thermograph. Appl Phys Lett 98(16),
art. no. 163706
11. Ghali VS, Mulaveesala R (2011) Comparative data processing approaches for thermal wave
imaging techniques for non-destructive testing. Sens Imaging 12(1–2):15–33
12. Dua G, Mulaveesala R (2013) Applications of Barker coded infrared imaging method for
characterisation of glass fibre reinforced plastic materials. Electron Lett 49(17):1071–1073
13. Ghali VS, Mulaveesala R (2011) Barker coded thermal wave imaging for defect detection in
carbon fibre-reinforced plastics. Insight: Non-Destruct Test Cond Monit 53(11):621–624
14. Mulaveesala R, Ghali VS (2011) Coded excitation for infrared non-destructive testing of carbon
fiber reinforced plastics. Rev Sci Instrum 82(5):054902
15. Mulaveesala R, Muniyappa A, Siddiqui JA, Arora V (2014) Numerical approach to binary
complementary Golay coded infrared thermal wave imaging. Proc SPIE 9105:91050T
16. Siddiqui JA, Arora V, Mulaveesala R, Muniyappa A (2015) Infrared thermal wave imaging for
nondestructive testing of fibre reinforced polymers. Exp Mech 55(7):1239–1245
17. Arora V, Mulaveesala R (2014) Pulse compression with Gaussian weighted chirp modulated
excitation for infrared thermal wave imaging. Prog Electromagn Res Lett 44:133–137
166 R. Mulaveesala et al.
18. Arora V, Mulaveesala R, Bison P (2016) Effect of spectral reshaping on frequency modulated
thermal wave imaging for non-destructive testing and evaluation of steel material. J Nondestruct
Eval 35(1):1–7, art. no. 15
19. Ghali VS, Mulaveesala R (2010) Frequency modulated thermal wave imaging techniques for
non-destructive testing. Insight: Non-Destruct Test Cond Monit 52(9):475–480
In-situ Detection and Quantification
of Oxide Scale Exfoliation in Stainless
Steel Boiler Tubes by Magnetic Coercive
Force (MCF) Measurement
Anil Kumar Das, Avijit Mondal, Satish Chand, and Amit Chhabra
Abstract Supercritical boilers use austenitic stainless steel (SS) boiler tubes in the
high temperature regions (mainly as superheater and reheater tubes) and oxide layers
(mainly magnetite) are formed on the internal surface of these tubes during operation.
When there are load changes or fluctuations in temperature during operation, there is
possibility of exfoliation of the internal scale due to large difference in coefficients of
thermal expansion of the base SS metal and the oxide. These exfoliated oxides accu-
mulate at the lower bends of the tubes and may cause blockage and failure of the boiler
tubes in steam path. While these tubes are non-magnetic, their oxides are magnetic.
Some methods based on estimation of only amplitude of magnetic field are available
for qualitative information on blockage. A few equipment analyzing amplitude and
phase both, claim for quantification of blockage, but they are rather expensive. Hence,
there is need for some simple, economic, easily deployable non-destructive method
for detection as well as quantification of such blockage. Quantification helps power
utilities in taking timely decision for cleaning or replacement of the blocked tube
bends. Use of Radiographic Testing is inconvenient as it hampers parallel working in
nearby areas. A method based on the estimation of Magnetic Coercive Force (MCF)
has been explored. Experimentation shows that the method gives good detectability
as well as quantitative correlation with the exfoliation build up inside the austenitic
stainless steel tubes. The method is non-destructive, fast and easily field deployable
with no hampering of any parallel activities during overhauling of boiler.
1 Introduction
Coal-fired power plants provide majority of the country’s energy demand and plant
maintenance outages are often planned events that have a known financial impact on
the utility company. Unplanned outages cause economic loss and reduces availability
of units for generation. One cause for extended, unplanned plant outages is excessive
steam-side oxidation of superheater tubes and subsequent exfoliation of iron-based
oxides [1, 2]. Exfoliated scale collects in lower tube bends and circumferential welds
of superheater tubes. The scale collects at these transitions and creates blockages. The
blockages can go undetected until short-term overheating from restricted steam flow
results in unexpected tube failures. As the temperatures and pressures in commercial
power plant have increased, the industry has turned to the use of austenitic stainless
steels for steam service, especially in supercritical power plants [3, 4] and there is need
for development of non-destructive methods for early detection and quantification
of such blockages for timely detection of such tubes for taking remedial action to
reduce the outage due to tube leakage.
Ferritic steels, such as T91 grade steels, are commonly used in the superheaters
and reheaters of subcritical power plants. Austenitic stainless steels tubes are more
appropriate than ferritic tubes in supercritical steam because of their improved steam-
side oxidation resistance and their strength at higher temperatures and pressures. The
shift in material choice has occurred to address the rapid oxidation of ferritic steels
that are commonly used in subcritical power plants when the steam temperature is
at or above 565 °C. Ferritic steel superheater tube alloys rely on a spinel oxide layer
that is 50–100 µm thick to protect the alloy from oxidation [2, 4]. Austenitic steels
rely on the spinel oxide as well, but they also have the capability of forming a thinner,
more protective Chromia layer [4–7].
Decreasing the oxidation rate of an alloy exposed to steam is directly related
to slowing the diffusion of iron through the protective oxide layers on the steam-
touched surfaces [2, 7]. Austenitic stainless steels contain higher concentrations
of the alloying elements that are responsible for improving oxidation resistance
compared with their ferritic steel counterparts. The most important alloying elements
in austenitic stainless steels for oxidation resistance in steam service are chromium
and silicon. Silicon is normally present at concentrations less than one percent by
weight. The minimum chromium concentration found in austenitic stainless steel
tubes is thirteen percent. Chromium and silicon form Chromia and Silica when
oxidized in steam. Iron can be considered an important alloying element in the
spinel oxide layer, but pure iron oxides, such as magnetite, hematite, and wustite,
afford little or no resistance to oxidation above 565 °C [2–5] due to their porosity and
tendency to exfoliate. Iron oxides are loosely adhered to the alloy, and they exfoliate
during thermal cycles and when their thickness approaches 20 µm [2–4]. Exfoliation
is driven by internal stress built up in the scale that is relieved by exfoliation [8–12].
Chromia is far more effective at decreasing the oxidation rate, followed by spinel
oxides that contain iron, nickel, and chromium.
In-situ Detection and Quantification of Oxide Scale Exfoliation … 169
The paper discusses the oxidation phenomena in the SS tubes and the use of
magnetic coercive force methodology as effective tools for their identification and
quantification.
The oxidation of the steel in the steam is a natural process, starting with the quick
formation of the initial oxide scale with slower formation of subsequent oxide layer.
Certain adverse operating conditions will accelerate the formation of the oxide scale,
such as over-temperature or fluctuations in temperature and pressure [12]. As per the
study by IG Wright, and BA Pint on the formation mechanism of the oxide scale and
its influencing factors, during the high-temperature oxidation, the thickness of the
oxide scale formation has the following law [13]:
D = K p · t n = A · e−Q/RT · t n
where K: Oxidation rate constant; Q: Metal oxide activation energy (J/mol); T: Metal
temperature (K); t: Running time (h); A: Arrhenius constants; R: Gas constant; n:
Index.
Scale exfoliation in steam plants is recognized as a problem worldwide and is
expected to assume increasing importance as plants are run for longer times at
maximum load (maximum temperature) and as steam temperatures increase with
new plant designs. Thus, there is a need to improve our understanding and determine
what measures can be taken to minimize the impact of exfoliation on plant operation.
This need has been recognized and addressed several times in the past [8–12], but
the remedial measures devised obviously have not removed the problem. Overall, it
appears that the basic causes of scale exfoliation are understood and that some of the
practices employed address some of the factors involved.
(a) (b)
Fig. 1 Structure of oxide scales in boiler tubes of a T91 and b TP347 respectively
uniform thickness with separation of the main inner and outer layers. In case of
TP347, the inner layer has somewhat irregular thickness with complete exfoliation
of outer layer.
Figure 2 shows relative thickness and location of oxide layers in scales formed on the
different classes of Fe–Cr alloys (after equivalent times and temperature). Vertical
axis in the diagram indicates original alloy surface. Left side of it indicates grown
oxides thickness on top the alloying surface and right side of it indicates the formation
of FeCr Spinel inside the alloys. Growth of exfoliation of scales from the ferritic and
austenitic steels of interest overwhelmingly involves loss of flakes of magnetite with
haematite, Fe2 O3 (outer layer), while the spinel (and/or chromia) remains adherent
to the alloy [2]. It can be observed that relatively thicker oxide scales are formed
for the ferritic steels T22 and T91 compared to TP347. However, as explained in the
next section, the exfoliation tendency is more in case of austenitic steel, like TP347.
There is difference in values of coefficient of thermal expansion for steel and its
oxide. The oxide scale will be squeezed or stretched during the cooling and heating
of boiler tubes, respectively. Since the oxide scale is a brittle material with much
smaller allowable stress than the steel, it easily wrinkles, cracks, or flakes in the
process of heating or cooling. Table 1 shows the coefficient of expansion values
for the ferritic steels (T22 and T91 grades), austenitic stainless steels (TP304 and
TP347), Haematite (Fe2 O3 ), and magnetite (Fe3 O4 ) oxides at different temperature.
In-situ Detection and Quantification of Oxide Scale Exfoliation … 171
Table 1 Thermal expansion coefficient of the steel and its oxide (X 10–6 K−1 )
Steel Temperature (°C)
100 300 500 600
T22 12.1 13.3 14.4 14.6
T91 10.9 11.7 12.3 12.6
TP304 17.1 17.8 18.9 19.1
TP347 17.3 17.7 18.6 18.9
Fe3 O4 8.9 9.1 10.9
Fe2 O3 11.3 11.9 13.8
During the operation of boiler, material temperature of the TP347 tubes in boiler
might experience a temperature of about 600 °C or even higher depending on the
design. It can be noted from Table 1 that the difference in coefficient of thermal
expansion between the tube and oxide is more pronounced at higher temperature for
TP347 tubes. This clearly shows that the tendency of exfoliation would be more for
the austenitic steel tubes compared to the ferritic steel. This also explains the thicker
oxide scales observed in the ferritic steel boiler tubes compared to the austenitic steel
boiler tubes in service.
172 A. K. Das et al.
When oxide scales fail (crack), separate, and exfoliate (come out) from the boiler
tube surfaces, they fall down and accumulate at the bends (Fig. 3a, b). Figure 3c
shows the oxide accumulation at the bends of the tube in the radiography image
of the tubes bend. The accumulated oxide at bends gradually obstructs the flow of
steam. Inadequate flow of steam in the bend and nearby region of tube would cause
overheating of tube metal. Overheating results because there is insufficient steam
inside the tube to pick-up the heat from the hot air outside causing rise in local metal
temperature of the tube. This would result in failure of the tubes due to short term
overheating (see Fig. 3d), causing forced outage of the unit. Hence there is a need
for detection of such situation by periodic inspection of such regions.
The exfoliated oxide is magnetic in nature, while the austenitic tubes are non-
magnetic. Hence some equipment based on characterizing magnetic behavior is
available in the market. Most of them are based on estimation of amplitude of the
magnetic field near the bends to estimate the magnetic exfoliation build-up. However,
they do not give reliable information on the quantity of the exfoliation required for
any remedial action. Some other equipment work on the Low Frequency Electromag-
netic Testing (LEFT) principle analyzing the phase behavior also for estimation of
these exfoliations. The methodology used by us is based on measurement of magnetic
coercive force (MCF) by a portable equipment which give fast and reliable readings,
which can be used for the reliable estimation of exfoliation.
In material science, Magnetic Coercive Force (MCF), most popularly called coer-
civity, is a measure of the ability of a ferromagnetic material to withstand an external
magnetic field without becoming demagnetized. For ferromagnetic material, the coer-
civity is the intensity of the applied magnetic field required to reduce the magnetiza-
tion of that material to zero after the magnetization of the sample has been driven to
saturation. Thus, coercivity measures the resistance of a ferromagnetic material to
Fig. 3 a Superheater tubes bend area. b Tube filled with exfoliated oxide. c Digital RT images of
the bend with accumulated oxide. d Short term overheating (fish mouth opening) in superheater
tubes due to blockage
In-situ Detection and Quantification of Oxide Scale Exfoliation … 173
Keeping the length of exfoliation distribution constant, fill factor was varied and
measurements of MCF were carried out keeping the probe normal to tube length.
Measurements were carried out for various length of exfoliation distribution. Probe
was kept in centre of the oxide layer under the tube at 0° and 90°position. Figure 7
shows the variation of MCF with fill factor corresponding to fixed exfoliation distri-
bution length to 32 mm. It can be observed that for both 0° and 90° positions, the
MCF values gradually increases to value of ~15 A/cm corresponding to ~70% fill
176 A. K. Das et al.
factor. For exfoliation distribution length fixed to 64, 84, and 104 mm, variation
in MCF values with fill factor has been shown in Figs. 8, 9, and 10, respectively.
It can be observed that the MCF values for 70% fill factor increased to ~23 A/cm.
During the experimentation, it was observed that with the small probe, the MCF value
increases with distribution length upto ~40 mm, when the probe is placed centrally.
This length is interpreted to be the maximum sensing length for small probe. This
value was observed to be about 104 mm for bigger probe. Hence, during the experi-
ment, maximum length of exfoliation was kept limited to 104 mm. The experimental
results have been provided with measurement with large probe.
Fig. 11 Effect on MCF values of variable exfoliation quantity and variable distribution length
accumulation is more than 104 mm, big probe is preferable. For lesser length than
this, small probe should be used. Depending on the accepted fill factor or blockage,
which is not alarming, a cut off value of MCF could be decided based on the above
plots. This would be helpful in identification of severely affected tubes needing to
be cleaned or replaced as per the requirement or availability of time and spare bends
for replacement. It is also to be kept in mind that practically MCF measurement
with different types of equipment and probes shows variation in MCF values due
to difference in area/volumetric coverage of the magnetisation. Hence, one need
to develop the plots as per their requirement with available equipment and probe
sets. However, the experimentation show feasibility of use of the methodology for
qualitative and quantitative characterisation of exfoliation.
5 Conclusion
Acknowledgements Authors would like to express their gratitude to NTPC management for their
guidance and support to undertake this developmental project. They would also like to express their
sincere thanks to NTPC stations for providing exfoliation samples for the experimentation.
References
1. Steam, its generation and uses, 39th edn. Babcock and Wilcox (1989)
2. EPRI report 1013666, Program on technology innovation: oxide growth and exfoliation on
alloys exposed to steam, June 2007
3. U.S. program on materials technology for ultra supercritical coal-fired boilers, Advances in
materials technology for fossil power plants: proceedings of the 5th international conference
(#05226G) 2008
4. Viswanathan R (1989) Damage mechanisms and life assessment of high temperature compo-
nents. ASM Int, p 392
5. Otsuka N, Fujikawa H (1991) Scaling of austenitic stainless steels and nickel-based alloys in
high temperature steam at 973K. Corrosion 47(4):240–248
6. Otsuka N, Shida Y, Fujikawa H (1989) Internal-external transition for the oxidation of Fe-Cr-Ni
austenitic stainless steels in steam. Oxidat Metals 32:13–45
7. Birks N, Meier GH, Pettit FS (2006) Introduction to the high temperature oxidation of metals,
2nd edn
8. Wright IG, Tortorelli PF, Schütze M (2007) Oxide growth and exfoliation on alloys exposed
to steam. EPRI Report No. 1013666 (2007)
9. Wright IG, Dooley RB (2010) A review of the oxidation behavior of structural alloys in steam.
Int Mater Rev 55(3):129–167
10. Wright IG, Howe JY, Sabau AS (2009) Morphological evolution of oxide scales grown on
ferritic steels in steam. Mater High Temp 26(2):105–111
11. Wright IG, Sabau AS (2008) Development of an integrated model to predict and control oxide
scale exfoliation: overview & progress for 2008. EPRI Rept. No. 1014193, Nov 2008
12. Wright IG, Sabau AS (2009) Development of an integrated model to predict and control oxide
scale exfoliation: exfoliation calculations for ferritic, austenitic, and fine-grained austenitic
alloys-technical update. EPRI Rept No 1017625, Nov 2009
13. Pint BA, Shingledecker JP, Wright IG (2012) Characterization of steam oxidation products
from field-exposed tubes. In: Advances in condition and remaining life assessment for fossil
power plants, EPRI, Charlotte, NC, presented Hilton Head, S.C., Oct 2012
14. Sudharsanam V, Senthilkumar V, Raju N, Vetriselvan R (2015) Evaluation of post weld
heat treatment quality of modified 9Cr-1Mo (P91) steel weld by magnetic coercive force
measurements. Arch Civil Mech Eng 15:847–853
Case Studies in the Use of Digital
Radiographic Imaging for Structural
Integrity Assessment and Failure
Analysis
1 Introduction
Even though advances have been made in digital imaging, majority of industrial
radiography is still performed with conventional NDT films due to limitation in spatial
resolution of digital radiographic systems. To take advantage of digital imaging and
to minimize dust or dirt influence, film digitization has been performed [6, 7].
CR technology in NDT offer film like image quality, less dose and shielding
requirements, availability of results in short time, elimination of chemicals and dark-
room processing, significant reduction in consumables and maintenance. High reso-
lution CR imaging enables to meet the critical requirements of industries like nuclear,
aerospace and defense [8, 9].
In DDR mode, digital images are obtained by employing Digital Detector Array
(DDA) like Flat Panel Detector (FPD). It can be upgraded to provide a Computed
Tomography (CT) capability by adding a precision manipulator and a workstation
computer with image acquisition cards, motor control capabilities and software
for data acquisition, processing and display. CT provides cross-sectional images
through the specimen from the measured transmitted intensities using a mathematical
procedure referred to as ‘image reconstruction’ [10].
This paper presents case studies to extend the usefulness of radiographic imaging
employing CR and DR modalities for structural or weld integrity and failure analysis
of typical specimens as an augmentation to digital industrial data.
The experimental details and results obtained for the three specimens investigated in
the present study are given in the following sections.
A specimen having titanium-titanium welded joint that was formed using electron-
beam welding has been investigated. Figure 1 shows the specimen having Outer
Diameter (OD) of 18 and 2.5 mm wall thickness. As OD is less than 90 mm, experi-
ments were carried out using X-ray computed radiography employing Double Wall
Fig. 1 Electron-beam butt-welded joint of titanium pipes (18 mm OD × 2.5 mm wall thickness)
Case Studies in the Use of Digital Radiographic Imaging … 183
Double Image (DWDI) elliptical technique. Exposures were performed using a phos-
phor Imaging Plate (IP) of 6 × 24 mm at X-ray source settings of 150 kV, 4.0 mA,
0.4 mm focal spot and 1.3 m source-to-detector distance. Total three exposures at
different orientations (120° apart each) were taken to cover entire circumference
of the weld joint. They are categorized as A, B and C. Images were read out from
the exposed IP using an industrial NDT laser-scanner. Contrast sensitivity of X-ray
computed radiographs has been determined using hole-type IQI by the following
relationship [11–14]
100 TIQI dhole
EPS (% ) = ,
tmaterial 2T
(a)
(b)
Porosity
Non-uniform
weld profile
Concavity
Fig. 2 a X-ray computed radiograph of the specimen b zoomed portion showing achieved image
quality values (4T and 2T holes of IQI) and detected defects highlighted with yellow ROIs
184 L. Yenumula et al.
and detected defects (porosity, concavity and non-uniform weld profile) highlighted
with yellow ROIs. The experimentally achieved EPS is about 2%. Defects with size
of greater than 50 µm have been detected.
An electrical heater pin was fabricated with central heater pin inside SS tube (10 mm
OD × 6 mm ID) which has failed during operation. Figure 3 shows photograph of
the heater pin used in the investigation. X-ray digital radiography has been carried
out to identify failure point non-destructively. DR system having constant potential
X-ray equipment and an indirect FPD based on a-Si with 200 µm resolution has been
used. Exposures were taken with X-ray source settings of 160 kV, 0.8 mA, 0.4 mm
focal spot and 1.6 m source-to-detector distance. X-ray digital radiograph (Fig. 4a)
revealing failure point (black circle) of heater wire inside the specimen (Fig. 4b).
Quantification of failure points has been carried out using line profile method as
shown in Fig. 4c.
3 Conclusion
Case studies along with methodologies adopted for evaluation of weld structural
integrity and failure analysis of typical nuclear component has been discussed. Weld
defects with size of greater than 50 µm have been detected in X-ray computed
radiography of electron-beam welded titanium specimen. Failure point of heater wire
inside electrical heater has been identified using digital radiography performed with
flat panel detector having 200 µm resolution. The case studies extend the usefulness
of digital radiographic modalities and demonstrate enhanced visualisation and data
analysis.
Case Studies in the Use of Digital Radiographic Imaging … 185
(a)
(b)
(c) Failure
points
(5 mm)
Fig. 4 X-ray digital radiograph revealing failure point (black circle) of heater wire inside the
specimen
Acknowledgements The authors are grateful to Dr. P. K. Pujari, Director, Radiochemistry and
Isotope Group (RC&IG), Bhabha Atomic Research Centre (BARC), Mumbai, for his valuable
support and constant encouragement. Thanks are also due to the technical staff of Industrial Tomog-
raphy and Instrumentation Section, RC&IG, BARC, Mumbai for extending help in the experimental
work. Technical help and healthy long discussion with Shri Dipak Patel and Shri Abhimanyu
Chowdhury of Reactor Control Division (RCnD), BARC and Shri S. Kumar and Shri Saurin Bhatt
of Reactor Projects Division (RPD), BARC are thankfully acknowledged.
References
1. Halmshaw R (2012) Industrial radiology: theory and practice. Springer Science & Business
Media
2. Raj B, Venkataraman B (2014) Practical radiography. Narosa Publishing House
3. Ewert U, Zscherpel U, Bavendiek K (2007) Strategies for film replacement in radiography. In:
IV Conferencia Panamericana de END
186 L. Yenumula et al.
Abstract Delamination is a common defect that affects the durability and perfor-
mance of reinforced concrete structures. Major causes of delamination in concrete are
rebar corrosion, moisture ingression, damage due to reversal of loading etc. Impact
echo is a relatively simple non-destructive method that works based on propaga-
tion of stress waves at the impact point of specimen surface to other locations. The
recorded signals represented as a waveform in voltage and its corresponding fast
Fourier transform in amplitude spectrum essentially carries qualitative and quanti-
tative information about the flaw. This paper investigates the ability of impact echo
to image and quantify predefined flaws of different acoustic impedance and esti-
mate their physical dimensions. Impact response were recorded for six apertures
intersecting the defect locations at a fine interval of 10 mm. The echogram or B-scan
images of signal waveform showed the relative acoustic impedance of the flaws based
on the intensity of wave reverberations exhibited during the scans. Furthermore, B-
scan images of the fast Fourier transform provided the quantification of the flaws
viz. dimensions or depth. Steel rebars could not be detected in any of the B-scan
images. The findings reported in this study can be implemented in qualitative and
quantitative investigation of delamination in concrete structures.
1 Introduction
Major civil structures are built with an anticipated life span up to 100 years, and
structures which were built decades back are deteriorating. Unfortunately, the rate of
degradation of structure is accelerated by external influences directly or indirectly.
M. Yumnam (B)
Vellore Institute of Technology, Vellore, India
H. Gupta · D. Ghosh
CSIR-Central Building Research Institute, Roorkee, India
From past encounters, main factors affecting the strength and durability of structures
are mainly attributed to (i) environmental causes like moisture ingression leading
to rebar corrosion, (ii) seismic activity such as earthquake and (iii) human induced
causes like impact damage and overloading. The consequences of these undesirable
events in concrete structures are formation of cracks and delaminations. Therefore,
methods for early detection and characterization of defects are inevitable to assure
overall safety of structures through appropriate repair and retrofitting measures.
Impact Echo (IE) is a relatively simple and common method for non-destructive
testing developed in mid-1980s by researchers at National Bureau of Standards and
Cornell University in the United States [1]. Since its introduction in civil engineering
applications, successful detection of different types of defects has been reported in
concrete and masonry structures. IE is an improved alternative to ultrasonic tech-
nique and modal analysis. In case of impact echo, the frequency range of stress wave
generated due to mechanical impact compromises with the dimension of flaws and
at the same time avoids scattering and attenuation of waves due to heterogeneous
compositions in concrete. Furthermore, the additional feature of IE to produce simul-
taneously both time and frequency domain of the recorded signal offers an improved
data interpretation than either of the modal analysis and ultrasonic technique alone.
It is important to ascertain the advantages of IE to understand the practicality of the
method for any inspection procedure. The general advantages of IE can be listed
as ability to detect deep defects (up to 1 m), not subjective as in visual inspec-
tion and acoustic sounding, not exhaustive since light weight instrument are used,
and no interference on the data from external factors. However, like any other non-
destructive testing method, its effectiveness depends on the understanding of the
technicality involved in the operation of the method. For IE operations the prerequi-
sites are consistency in the impact of steel ball on the specimen, following uniform
interval of recording point and setting of appropriate acquisition parameters.
Detection of delamination in concrete plate is by far the most acknowledged appli-
cations of IE [2–5]. This is largely because the interference of boundary reflection is
significantly reduced in plate like structures where lateral dimension is much larger
than depth. Study on the characteristics of impact response on concrete cubes have
also been presented in [6]. In addition, IE can successfully characterize bonding
capacity in masonry structures [7]. IE is a reliable method for assessment of cylin-
drical ducts in concrete structures [8]. Successful development of B-scan images in
frequency domain has been reported that reveals delamination in concrete bridge deck
[9] and tendons ducts in concrete [10]. A comparative investigation between impact
echo and non-contact video based measurement reported that impact echo yields
overestimated and non-uniform depth of cracks in a reinforced concrete beam [11].
Successful applications on detection and prediction of corrosion in reinforcement
bars using impact echo have also been reported [12]. Air-coupled impact echo has
been used to produce accurate planar feature of shallow delamination in the form of
contour images [13, 14]. Furthermore, depth of delamination can be estimated from
the flexural resonance frequency [14]. Other applications of IE on a wide variety of
investigations are discussed in [1, 2].
Imaging of Delamination in Concrete Slab Using Impact Echo 189
This paper explores the feasibility of impact echo to image, quantify and differ-
entiate flaws having different physical dimensions and acoustic impedances. Experi-
mental study has been carried out on a reinforced concrete slab embedded with flaws
of different material compositions. In the subsequent sections the experimental set-
up and testing procedure are described. Further, the results illustrated as B-scan for
both time and frequency domain are discussed.
2 Background
2.1 Principle
Impact echo works based on excitation of stress wave on the specimen surface by an
impact source. The impact excites in the body, dilatational wave (P), distortional wave
(S) and surface wave (R). The body wave propagates as a hemispherical wavefront
into the body, and surface wave propagates along the air-surface interface away
from the point of impact. Theoretically, P wave is more important than S and R
waves since particle displacement caused by P wave is far significant that the other
two. Also, P wave propagates twice as fast as the S wave [1, 15]. However, if the
transducer is placed too far or too near the impact source the significance of P-wave
in the waveform will be reduced. P wave propagates periodically between the surface
and the interface of backwall or internal defect, forming multiple reflection which
can be considered a transient resonance. This phenomenon is apparent as multiple
peaks in voltage–time graph known as waveform. The reflection of wave between
the external boundary and defect interface can be understood as a consequence of
contrasting acoustic impedance of the adjoining materials. Acoustic impedance is a
function of material density and wave velocity given as
Z = ρv, (1)
where
ρ = density of the material (kg/m3 ) and
v = wave velocity in the material (m/s).
Whenever stress wave encounters a change in acoustic impedance of materials,
a portion of incidence wave penetrates and the rest reflected. This characteristic is
determined by a term called reflection coefficient R,
Z2 − Z1
R= , (2)
Z2 + Z1
190 M. Yumnam et al.
where
Z 1 = Acoustic impedance of material 1 and,
Z 2 = Acoustic impedance of material 2.
When R is a negative value, it indicates that the acoustic impedance of second
material is less than the first, that is, Z 2 < Z 1 . In such case, the reflected stress
wave will have an opposite sign to that of the incident stress wave. In other words,
incident wave with compressive stress will produce a reflected wave with tensile
stress and vice-versa. This theory can be related to the reflection at concrete-air
interface or between concrete and any material having lesser acoustic impedance.
Contrarily, when Z 2 > Z 1 the coefficient reflection gives a positive value. This means
that an incident wave having compressive stress will result in a reflected wave with
compressive stress, that is, no change in sign of the stress wave. This is the case with
concrete-steel interface [16].
Interpretation of data in impact echo requires an established knowledge on the
characteristics of signals in time and frequency domain. In the waveform obtained
from the surface displacement, a dip in the signature represents the Rayleigh wave
which reaches the transducer first followed by multiple lower peaks representing
multiple reflection of P wave between two interface boundaries. However, the wave-
form cannot depict much information about the targets encountered by the waves.
Hence, a transformation algorithm, especially Fast Fourier Transform (FFT) trans-
forms time domain waveform to a frequency domain spectrum. In the amplitude-
frequency plot or spectrum, the wave corresponding to the reflection from final
depth of a perfect solid body is called the thickness frequency, and it occurs as
a fundamental frequency. In this case, the fundamental frequency is the dominant
frequency in the spectrum. For a specimen with delamination at an intermediate
depth, a shift in dominant frequency given by f d = 1/td will be observed. Here, td is
the time between adjacent peaks in waveform. Note that the position of the dominant
frequency varies according to the depth and lateral extend of defect. The peaks that
appear in the spectrum can be related to the depth of targets encountered along the
wave path in the specimen. Therefore, for a P-wave velocity C p , the frequency value
f from the spectrum is used to determine the depth of target T,
CP
T = , (3)
2f
E(1 − ϑ)
CP = . (4)
ρ(1 + ϑ)(1 − 2ϑ)
the impact response on the geometric features of the specimen [6]. Experimental and
numerical studies have recommended β = 0.96 for a concrete plate. Therefore, for
a concrete plate of known thickness, Eq. (3) can be modified to calculate the depth
of targets in a specimen as
0.96C P
T = . (5)
2f
2.2 Instrumentation
3 Experiment
A reinforced concrete slab was casted with planar dimension 600 × 600 mm and
thickness 100 mm of grade M30 having reinforcement mat at 50 mm depth from
the surface. Six predefined flaws comprising different material types and dimensions
were placed at 50 mm depth, at the same depth as reinforcement mat. All the flaws
were provided with wired support to inhibit transverse and lateral dispositioning
during the pouring and vibration of the concrete at the time of casting. This is impor-
tant since flaws have their density below or above the wet concrete mix. The thin steel
wire was chosen as wired support to avoid its detection in the impact echo readings.
Table 1 shows the details of the flaws embedded in the slab. The inclusions of flaws
of varying dimensions and acoustic impedances imitate in a way delamination in
concrete subsurface that alters the acoustic characteristics of the inner structures in
the specimen. Figure 1a shows the planar view of flaws placed in the test specimen.
Imaging of Delamination in Concrete Slab Using Impact Echo 193
Fig. 1 Test specimen. a Flaws placed prior to casting of the specimen. b Schematic diagram of the
specifications and positions of the flaws. c Actual test specimen after casting
The flaws have been identified with a ‘letter-number’ combination where the initial
letter represents the material type and the number for the position of the flaw in row-
wise placement from left to right. F, P, W and R indicates that the flaw has a material
type of polystyrene foam, polyethylene foam, ply wood and steel rebar respectively.
194 M. Yumnam et al.
Fig. 2 a Experimental setup for impact echo testing. b Experimental testing of specimen being
carried out along scanline 2
The number 1–9 indicates the position of the flaw. For example, in Fig. 1b, ‘W4’
indicates that the flaw is a ply wood material positioned at fourth position in the
array.
For the impact echo measurements, six gridlines were chosen presumably to
intersect the position of flaws. The testing was carried out in NDT lab at CSIR-
CBRI, Roorkee, India. At the time of measurement any form of vibration due to
machinery at the proximity of testing area was avoided to ensure inception of the
least noise in the recorded response.
The conical tip of the transducer unit was pressed against the specimen for every
10 mm interval acquiring a total of 60 data points along each scanline. Spherical
steel ball having 10 mm diameter employed an instant and light tapping at an average
distance of 7–15 cm around the transducer head as shown in Fig. 2b. Prior to impact
echo measurement, wave speed measurement was performed using a dual head trans-
ducer. Five data points were recorded at randomly picked positions on the specimen
giving an average wave velocity of 4100 m/s. This value is important to describe the
structure of the specimen prior to actual impact echo measurement. The entire instru-
mentation suitably works for a single operator to carry out the task. However, two
operators each dedicated to control impact-transducer unit and output-storage unit
increased the accuracy and efficiency of the recording process. For every point along
the gridline the impact response is produced as a waveform and its corresponding
Fast Fourier transform (FFT) in the output unit.
Changes in the acoustic impedances due to flaws were reflected on the frequency
spectrum as a shift in predominant frequency. Therefore, the output screen in a way
provides an idea about the presence or absence of flaws at the point of measurement.
Every recorded point needs to be manually saved in a dedicated file. Regular errors
in the records were encountered due to insufficient contact of the transducer due
to dirt at transducer tips or improper contact. Also, prolonged monotonous testing
procedure may lead to loss of concentration resulting in reading points off-tracking
or multiple data saving at same point. These factors need to be taken care to avoid
error during data acquisition process. The data were recorded sequentially in desired
order of scanline (aperture) thus saving a lot of time in data sorting and sampling
post-recording.
Figures 3a and 4a shows the first three scanline (aperture) along which impact echo
measurement was carried out at an interval of 10 mm. The impact response or A-
scan of each measurement points were stacked to obtain the B-scan waveform for
six scanlines (SL) in Fig. 3b–d and 4b–d. MATLAB software was used to plot and
interpret the impact response in both time and frequency domain. From the B-scan of
the six scanlines, the amplitude variations of recorded signals are indicated in RGB
image representation depicting the materials encountered along the path of stress
waves having different acoustic impedances. It is evident from observation that the
196 M. Yumnam et al.
Fig. 3 B-scan image of the time-domain waveform obtained for three scanlines on the specimen.
a SL-1. b SL-2. c SL-3
Fig. 4 B-scan image of the time-domain waveform obtained for three scanlines on the specimen.
b SL-4. c SL-5. d SL-6
confirms the presence of flaws having lowest acoustic impedance in these positions.
This presence of aggravated or higher intensity coloured pixels can be attributed
to higher reverberations of transient stress energy due to higher negative coefficient
reflection at the interface. In Table 1, polystyrene foam (F) and polyethylene foam (P)
have the lowest value of acoustic impedance (Z). Therefore, these two flaws have the
highest negative reflection coefficient as in Eq. (1). This implies that there is more
reflection of stress wave producing a transient reverberating resonance of higher
energy than rest of the flaw. Hence, it is confirmed that these flaws are composed
of either F or P flaw. Furthermore, from the relative positions of flaws, since three
such similar patterns of aggravated pixels can be found in positions 1, 3 and 5, it
can be concluded that F flaw type is present in these locations. Second flaws are the
polyethylene foam (P) which have similar acoustic impedance to that of polystyrene
foam (F) with half the flaw thickness. Relative to the positions of F found, it is obvious
that positions 7 and 9 should be occupied with F. Tough P and F have near similar
acoustic impedance, the wave reverberations representing P is much lesser in terms
of intensity and width in the B-scan images. A possible explanation for this can be
due to inclination of the interface surface incurred during the pouring and vibration of
concrete at the time of casting causing reduction in amplitude of reflected wave due
to mode changing. Note that the amplitude of reflected wave depends on the angle
of incidence of stress wave. The amplitude of reflection is highest for a wave with
198 M. Yumnam et al.
normal incidence on the interface [18]. Wooden plates have its acoustic impedance
closer to that of concrete compared to the first two materials. Its relative positions in 2
and 4 labelled as W2 and W4 in Figs. 3 and 4 is clearly visible as a wave reverberation
of very low intensity and width than that of F and P. This is justifiable based on the
lower coefficient of reflection resulting due to proximity of the acoustic impedance
of wood to that of concrete. Another important observation from the B-scan images
in time domain is the absence of wave reverberation in position 6 and 8 where steel
rebars are placed. In B-scan images shown in Figs. 3 and 4, no aggravation of colour
intensity in the B-scan images are found in regions labelled as R6 and R8. This is
because the coefficient of reflection for stress waves at the interface of concrete-steel
is very low since the acoustic impedance of concrete and steel rebar is very close.
Impact-E software executes fast Fourier transform (FFT) in itself to provide signal
in both waveform and spectrum simultaneously. In Figs. 5 and 6, a prominent feature
in the B-scan images are the high intensity or amplitude pixels clustered at certain
repeating frequency in a rather unconvincing pattern. Nevertheless, relative to the
Fig. 5 B-scan image of the frequency-domain waveform obtained for three scanlines on the
specimen. a SL-1. b SL-2 and c SL-3
Imaging of Delamination in Concrete Slab Using Impact Echo 199
Fig. 6 B-scan image of the frequency-domain waveform obtained for three scanlines on the
specimen. a SL-4. b SL-5 and c SL-6
frequency at with the high intensity cluster of pixels occur indicating a probable
flaw, the vertical depth of the flaw in the specimen has been estimated using Eq. (5)
with the average wave velocity already obtained for the specimen using a dual-head
transducer. With reference to Figs. 5 and 6, a similarity in the depth of clusters in
scanline-1, 2 and 3 and that of scanline-4, 5 and 6 confirms the presence of flaw at
these depths.
In all the B-scan images of Figs. 5 and 6, prospective flaws are encircled with black
dashed ellipse and captioned with the flaw identity and its corresponding vertical
depth estimated using Eq. (5). From the figure, taking into account F1, F3 and F5, it
is apparent that all the flaws have near similar depth. At this point, some variations
are observed especially the relatively sparse cluster of F3 in scanline-1 against that
of F3 in scanline-6; see Figs. 5b and 6d. Similar trend can also be pointed out from
the low reverberations of F3 position in scanline-6; see Fig. 4d. This may be due to
many factors which have not been specifically addressed in this study. But a possible
explanation can be related to noise interruptions or a compromise in the geometric
properties and acoustic impedance of the flaws (Table 1).
Polyethelene foam P7 and P9 are clearly identified and located in the scanlines in
Fig. 5d, Fig. 6b, d. Similar to what was noticed in the B-scan images in time domain,
the presence of steel rebar is not evident in B-scan images in frequency domain.
200 M. Yumnam et al.
Table 2 Comparsion and remarks on the detectability and accuracy of flaw depth estimation in
frequency domain B-scan images
Flaw ID P-wave velocity Average dominant Average estimated Remarks on
(m/s) freq (kHz) depth (mm) detectability
F1 4100 80 24.6 Good
W2 100 19.7 Good
F3 85 23.2 Fair
W4 105 18.7 Good
F5 95 20.7 Good
R6 – – Poor
P7 100 19.7 Good
R8 – – Poor
P9 105 18.7 Good
One final and most important observation here is the indication of total depth of
specimen which is the concrete-air interface specifying total wave reflection. This
is indicated in all the B-scan in Figs. 5 and 6, a red horizontal arrow is indicated at
a close frequency range of 20–25 kHz. A band of irregular high amplitude pixels
spanning the entire length is a clear representation of backwall in the specimen.
Another interpretation from the B-scan images in frequency domain is the lateral
dimensions of flaws. When flaws are detected along the scan line, a predominant
frequency emerges that corresponds to a constant amplitude extending upto the last
point over the flaw. This constant amplitude appears for a certain length in the B-
scan images representing the lateral dimension of flaw. Deviation in the observed
dimension from the actual dimension is due to resolution of impact measurement,
that is, lateral dimension of flaws exhibited in the B-scan images must differ at most
by 10% from the actual dimension (10 mm) since the impact response is recorded
at an interval of 10 mm throughout the span. Table 3 gives the % error of the lateral
dimensions observed in frequency domain B-scan for all flaws except for R6 and
R8 which could not be detected. The error in the estimation of lateral dimension of
Table 3 % error of estimated and length and width of the three detectable flaw types as obtained
in first three scanlines and last three scanlines respectively
Flaw ID Estimated length (mm) % error Estimated width (mm) % error
F1 90 10 85 15
W2 90 10 70 30
F3 90 10 55 45
W4 70 30 95 5
F5 75 25 90 10
P7 70 30 90 10
P9 70 30 80 20
Imaging of Delamination in Concrete Slab Using Impact Echo 201
flaw lies in the range 10–45%. This contradicts the ideal range of error supposedly
conceived due to insufficient resolution.
5 Conclusion
References
1. Sansalone M, Carino NJ (1986) Impact-echo: a method for flaw detection in concrete using
transient stress waves. National Technical Reports Library—NTIS
202 M. Yumnam et al.
2. Carino NJ, Sansalone M (1990) Flaw detection in concrete using the impact-echo method. In:
Bridge evaluation, repair and rehabilitation, Springer Netherlands, pp 101–118. https://doi.org/
10.1007/978-94-009-2153-5_8
3. Cheng C, Sansalone M (1993) The impact-echo response of concrete plates containing delam-
inations: numerical, experimental and field studies. Mater Struct 26:274–285. https://doi.org/
10.1007/BF02472949
4. Wouters J, Poston RW (2001) Applications of impact-echo for flaw detection. In: Structures
2001, American Society of Civil Engineers, Reston, VA, pp 1–10. https://doi.org/10.1061/405
58(2001)16
5. Yeh PL, Liu PL (2008) Application of the wavelet transform and the enhanced Fourier spectrum
in the impact echo test. NDT and E Int 41:382–394. https://doi.org/10.1016/j.ndteint.2008.
01.002
6. Hsiao C, Cheng CC, Liou T, Juang Y (2008) Detecting flaws in concrete blocks using the
impact-echo method. NDT and E Int 41:98–107. https://doi.org/10.1016/j.ndteint.2007.08.008
7. Sadri A (2003) Application of impact-echo technique in diagnoses and repair of stone masonry
structures. In: NDT and E international. Elsevier, pp 195–202. https://doi.org/10.1016/S0963-
8695(02)00064-6
8. Chaudhary MTA (2013) Effectiveness of impact echo testing in detecting flaws in prestressed
concrete slabs. Constr Build Mater 47:753–759. https://doi.org/10.1016/j.conbuildmat.2013.
05.021
9. Scherr JF, Grosse CU (2020) Delamination detection on a concrete bridge deck using impact
echo scanning. Struct Concrete. https://doi.org/10.1002/suco.202000415
10. Algernon D, Ernst H, Dressler K (2012) Signal processing for air-coupled impact-echo using
microphone arrays. In: 18th world conference on nondestrutive testing, Durban, South Africa,
NDT.net
11. Sun Y, Huang P, Su J, Wang T (2018) Depth estimation of surface-opening crack in concrete
beams using impact-echo and non-contact video-based methods. Eurasip J Image Video Process
2018:1–10. https://doi.org/10.1186/s13640-018-0382-7
12. Te Liang M, Su PJ (2001) Detection of the corrosion damage of rebar in concrete using impact-
echo method. Cem Concr Res 31:1427–1436. https://doi.org/10.1016/S0008-8846(01)00569-5
13. Kee S-H, Gucunski N (2016) Interpretation of flexural vibration modes from impact-echo
testing. J Infrastruct Syst 22:04016009. https://doi.org/10.1061/(asce)is.1943-555x.0000291
14. Zhu J, Popovics JS (2007) Imaging concrete structures using air-coupled impact-echo. J Eng
Mech 133:628–640. https://doi.org/10.1061/(asce)0733-9399(2007)133:6(628)
15. Ghomi MT, Mahmoudi J, Darabi M (2013) Concrete plate thickness measurement using
the indirect impact-echo method. Nondestruct Testing Evaluat 28:119–144. https://doi.org/
10.1080/10589759.2012.711329
16. Carino NJ (2001) The impact-echo method: an overview. In: Structures 2001, American Society
of Civil Engineers, Reston, VA, pp 1–18. https://doi.org/10.1061/40558(2001)15.
17. Kee S-H, Oh T, Popovics JS, Arndt RW, Zhu J (2012) Nondestructive bridge deck testing with
air-coupled impact-echo and infrared thermography. J Bridg Eng 17:928–939. https://doi.org/
10.1061/(asce)be.1943-5592.0000350
18. Krautkrämer J, Krautkrämer H (1983) Ultrasonic testing of materials. Springer, Berlin
Heidelberg. https://doi.org/10.1007/978-3-662-02357-0
Development of Smart Cementitious
Composite Sensors for Ambient
Vibration-Based Continuous Health
Monitoring of Structures
1 Introduction
All structures, including mechanical, aerospace, and civil infrastructure facilities, are
complex engineered system that ensure the economy of society and industrial pros-
perity. However, due to severe loading scenarios (e.g., repeated traffic loads, earth-
quakes, hurricanes) and long-term degradation (e.g., aging, corrosion, and fatigue),
self-sensing capacity [6, 9–15]. Toward this, Chun et al. [18] incorporated carbon
fiber into cement matrix to evaluate the strain sensibility of cementitious compos-
ites. It was reported that developed sensors are capable of detecting the strain, stress
or damage initiation themselves. Yoo et al. [19] investigated the effect of incorpo-
ration of various nanomaterials on the piezoresistive sensing capacity of cement-
based composites and observed that 1 wt.% of MWCNT exhibits higher sensitivity
and can be used for monitoring the stress/strain generated in concrete structures.
Lee et al. [20] evaluated the electrical and piezoresistive behaviour of cementitious
composites by using hybrid CNT and carbon fiber and reported that the composites
with 0.50 wt.% MWCNTs and 0.10 wt.% carbon fibers can be optimally used as
cement-based sensors. Similarly, Azhari et al. [10] developed cement-based sensors
and evaluated their performance under cyclic compression load. They obtained a
linear relationship between the change in electrical resistance and applied compres-
sive load. The cement-based sensors show the increase in electrical resistance under
tensile loading while decrease in electrical resistance under compression loading. It
is due to increase in separation distance under tension and decrease in separation
distance under compression loading [10, 21]. Konsta-Gdoutos et al. [22] reported
that the appropriate concentration of CNTs into cement matrix reflect considerable
effect on the electrical resistivity and piezoresistive behaviour due to involvement
of various dispersion efficiencies of the conductive particles. They achieved stabi-
lized and reversible responses form the cement-based sensors within elastic stress
region. Li et al. [23] conducted a comparative study with treated (acid) and untreated
CNTs for the electrical resistivity and piezoresistivity of cement-based sensors. Study
reveals that treated-CNTs have a stronger effect in enhancing the piezoresistive prop-
erty, whereas the untreated-CNTs have more forceful effect on reducing electrical
resistivity. Cha et al. [24] observed that the functionalization of CNT with carboxyl
has enhanced the sensitivity behaviour of the CNT/cement composites because the
dispersibility of CNT is enhanced when it is functionalized with carboxyl. Recently
Rao et al. [25] reported that the performances of smart cementitious composites are
strongly dependent on the selection of CNTs type, concentration of CNTs and the
types of electrodes.
It is worth noting that the use of SCC is primarily based on strain sensing
under static or quasi-static loading scenarios for monitoring the health of structures.
However, very limited studies have been done based on vibration sensing capability of
large concrete structures. For example, CNT/cement composites were developed and
used by Han et al. [26] as embedded sensors for traffic monitoring. Under repeated
compression and impulse loading, the developed sensors show very good response.
Saafi et al. [27] have embedded the CNT/cement sensors in concrete structure for
damage detection. A strong correlation was also observed between the change in
electrical resistance and the initiation of damages. The dynamic behavior of long rein-
forced concrete (RC) beam was assessed by Ubertini et al. [13] using CNT/cement
composites-based sensors. Experimental study shows that sensors are capable of
capturing the fundamental frequencies of vibrating beam under impulse loading.
Similarly, Sasmal et al. [12] have demonstrated the dynamic sensing techniques using
CNT/cement composites for long RC beam. They reported that sensors are capable of
206 R. K. Rao and S. Sasmal
capturing the fundamental frequencies (at least first three flexural modes) of vibrating
beam. Meoni et al. [17] and D’Alessandro et al. [28] recently observed that the SCC
have potential for structural health monitoring of large structures. However, dosages
of MWCNTs, the consistency of dispersibility of MWCNTs, and measurement tech-
niques are the important parameters to achieve the high-sensitivity and consistent
responses form developed sensors.
The above limited literature review suggests that the SCC based sensors exhibit
excellent correlation between the change in resistivity and applied load. They exhibit
higher sensitivity with good reversible responses within the elastic region. Such
unique properties of SCC sensors bring out the scope for wide range of applications,
especially for vibration-based techniques for structural health monitoring. However,
use of these sensors for vibration monitoring is very limited. Therefore, the present
study is mainly focused on the development of smart cementitious composites-based
sensors that can be used for health monitoring of large RC structures under static
and dynamic loading scenarios. Three different types of CNTs including Pristine-
MWCNT, COOH-MWCNT, and OH-MWCNTs are used to fabricate the sensors.
Strain sensing capacity is evaluated under static cyclic loading condition. The vibra-
tion sensing capacity of developed sensors is established through determining the
fundamental frequencies of vibrating reinforced concrete (RC) beam captured using
the SCC and the same is also compared with conventional accelerometer sensors.
The developed sensors show very good piezoresistive behavior with high strain- and
vibration-sensibility.
SCC were developed by reinforcing the CNTs into Ordinary Portland Cement of
grade 53 (density—3159 kg/m3 ). Three types of CNTs such as Pristine-MWCNTs (P-
MWCNT), OH-MWCNTs, and COOH-MWCNTs were chosen. Their properties are
listed in Table 1. Sodium Dodecyl Benzene Sulfonate (SDBS) and Tributyl Phosphate
were used as the dispersing agent and defoamer, respectively. Water to cement ratio
was 0.4, while surfactant to CNTs ratio was 0.55. Copper wire mesh was used as
electrodes for measuring the output response form sensors.
High aspect ratio and hydrophobic nature of CNTs increase the possibility of entan-
glement and close packing of CNTs and hence lead to certain difficulties for incor-
poration in any medium. On the other hand, due to strong van der Waals force
pristine CNTs have low dispersibility and high tendency to agglomeration. There-
fore, to reduce the agglomeration and ensure the uniform dispersion of CNTs, both
the physical- and chemical-methods were used. Chemical method involves using the
chemical solvent to improve the dispersibility of CNTs. The surfactant SDBS was
mixed with water using magnetic stirrer for 5 min at a rate of 250 ± 5 rpm. Then CNTs
were poured into the solvent and mixed for another 10 min using magnetic stirrer.
After that water-surfactant-CNTs solution was subjected to the sonication for 45 min
using probe ultrasonicator (Hielscher, ultrasound technology). Then the solution was
mixed thoroughly into cement matrix. During the mixing of CNTs/cement paste,
Tributyl Phosphate defoamer of 0.25 vol.% was added to reduce the air bubble forma-
tion. Well mixed CNTs/cement pasted was then poured into cubic mold of 50 mm ×
50 mm × 50 mm and mechanically vibrated to achieve the desire compaction. Two
copper wire mesh electrodes of size 50 mm × 40 mm were immediately inserted
into the specimens. The embedment depth of electrodes was 35 mm form top surface
of samples. The samples were de-molded after 24 h and kept for 28 days for curing
purpose. Finally, samples were kept in oven at 600 C for 3 days to eliminate the
access of water content. Four dosages viz 0.10, 0.25, 0.50, and 0.75 wt.% from each
group of CNTs were synthesized. Fabrication procedure and schematic diagram for
electrodes placement are shown in Figs. 1 and 2, respectively.
The experimental studies were conducted to explore the strain and vibration sensi-
tivity of SCC sensors under long cyclic compression- and dynamic-loading (using
impulse hammer), respectively. DC measurement technique with two probes was
adopted to measure the electrical response from SCC through embedded electrodes.
Although two probes method is affected by the polarization effect, it is easier for
installation of electrodes inside the sensors and convenient for measurement with high
accuracy in comparison with four-probes method. Two strain gauges were attached
to opposite sides of the samples to measure the strain. The samples were subjected
to long cyclic compression load of amplitude between 2 and 10 kN at a constant rate
of 50 N/s. Vibration monitoring test was conducted on the 4500 mm long simply
supported reinforced concrete (RC) beam with a T-shaped cross-section. A SCC
sensor was mounted on the top surface of RC beam using anchor device made of iron
as shown in Fig. 3b. Three conventional accelerometers of sensitivity 1000 mV/g were
mounted on the bottom of RC beam. An impulse hammer of sensitivity 2.248 mV/N
was used for random hitting on the RC beam at different locations to generate the
desire vibration. The sampling rate for capturing the response signals was set to be
1200 Hz. A high pass filter with cut-off frequency 5 Hz was applied to reduce the
noise from the response signals. The experimental setups for both tests are shown in
Fig. 3, and schematic diagram for vibration sensing is shown in Fig. 4.
Strain sensing capability was explored under long repetitive cyclic compression load
using compression testing machine. Before applying the long-range repetitive cyclic
load, the specimens were subjected to polarization with the constant electric field
at least 1 h. During the application of load, the microstructures of nanocomposites
changed resulting in change in electric potential difference (in term of resistance)
between measuring electrodes. This change in resistance can take place (a) either
due to the electron hopping/tunneling, (b) or due to contact conductivity inside the
cement matrix, (c) or due to both, depending on the concentration of CNTs in cement
matrix. Adequate amounts of CNTs form strong conductive networks within cement
Development of Smart Cementitious Composite Sensors … 209
Fig. 3 Experimental setups for a strain sensing b vibration sensing of SCC-sensor c detail view
of SCC-senor
matrix and these networks will be altered according to the loading direction. At
very low CNTs concentration, tunneling effect dominates and responsible for the
electrical conductivity. However, SCC-sensors have high electrical resistivity and
high sensitivity at low concentration, but close enough to the percolation threshold.
During cyclic compression loading, the CNT particles come closer to each other,
while unloading CNT particles separate from each other resulting an increase or
210 R. K. Rao and S. Sasmal
R Rt − R0
λ= /ε = /ε, (1)
R0 R0
where R0 is the initial electrical resistance, Rt is the real time resistance, and ε is
the applied strain. The experimental studies have been conducted with four different
concentrations of CNTs viz. 0.1, 0.25, 0.5, and 0.75 wt.%. The SCC-sensors with low
CNT concentrations like 0.10 and 0.25 wt.% showed inconsistency and irreversibility
due to discontinuity in conductive networks within the cementitious matrix. The
insufficient amount of CNT particles forms weaker conductive networks and hence
reflect the irregularities in responses. However, SCC-sensors with 0.25 wt.% CNTs
show better sensitivity than that of 0.10 wt.% of CNTs. Further increase in the CNT
concentration exhibits very good reversibility and stability under cyclic loading. The
change in electrical resistance under applied load for SCC-sensors incorporated with
0.50 wt.% CNTs is shown in Figs. 5 and 6. The results show very good agreement
for the change in resistance with applied load. It is also observed that change in resis-
tance with respect to applied load is consistently linear and reversible for all types
of SCC-sensors. A relationship between fractional change in electrical resistance
(FCR) and applied strain is established and is shown in Fig. 7. The results reveal a
good linear relationship between these two parameters, and the slope of the curve
represents the sensitivity or gauge factor of SCC-sensors. The gauge factor obtained
from the slope of curve is presented in Table 2. The experimental study indicates
that the cementitious composites containing 0.50 wt.% MWCNTs are capable of
Development of Smart Cementitious Composite Sensors … 211
Fig. 5 Strain response captured from conventional strain gauge pasted on MWCNT a Pristine b
COOH- and c OH-SCC sensors
Fig. 6 FCR (%) of SCC-sensors reinforced with 0.50 wt.% of MWCNT a Pristine b COOH- and
c OH-SCC sensors
Fig. 7 Relationship between FCR and applied strain of SCC-sensors with 0.50 wt.% MWCNT a
Pristine b COOH- c OH-SCS sensors
forming strong conductive networks and exhibit good sensitivity. However, from
Table 2 it can be seen that the cementitious composites with 0.50 wt.% COOH-
MWCNTs exhibit highest gauge factor among all three group of MWCNTs. These
results indicate that the functionalization of CNTs with carboxyl group has beneficial
effects on the piezoresistive response of the composites because the dispersion of
CNTs is improved. On the other side, functionalized CNTs with a COOH-group are
more hydrophilic than non-functionalized CNTs and are likely to create a stronger
interfacial bonding with the cementitious composites. The bond of interfacial inter-
action between COOH-MWCNTs and the cementitious composites is strengthened
that leads to enhance in the load transfer efficiency. It was also found that further
addition of dosages of MWCNTs did not affect the reversibility and stability of the
strain under load application, but it was found that the strain sensing capacity or
piezoresistive behaviour was adversely affected. Therefore, in the present paper, the
piezoresistive behavior of SCC sensors with only 0.50 wt.% MWCNTs are reported.
From the piezoresistive test, it is clear that the SCC-sensors with 0.50 wt.% MWCNTs
show very good reversibility and high strain sensitivity due to the strong formation
of conductive networks. Among the three types of MWCNTs, the SCC incorporated
with COOH-MWCNTs exhibits the highest gauge factor or piezoresistive behavior.
Therefore, the vibration sensing capability of the cementitious composites incorpo-
rated with COOH-MWCNTs (0.50 and 0.75 wt.%) have been tested and reported
here. The long RC beam was excited by random impulse hitting using a hammer.
The application of impulse force on the RC beam causes the vibration of the struc-
ture, and the vibration induced stresses are transmitted into the SCC-sensor through
the mechanical anchor system. This stress transmission results in the generation of
electric charge and change in voltage between embedded electrodes of the sensors.
The change in voltage in the SCC-sensors occurs due to disturbance in the conduc-
tive networks. This changes in voltage between electrodes can be measured using a
data acquisition system. The recorded signals contain the information of vibratory
response of RC beam. The response signals in the vertical direction captured from
SCC-sensors and conventional accelerometers are shown in Fig. 8a–c, respectively.
During the experiment, at-least three sets of response signals were captured and
averaged considering the appropriate and consistent results. From Fig. 8, it can be
observed that the captured response signals from SCC-sensors for vibrating RC beam
contain the effect of hammer hitting in term of multiple peaks. The same peaks can be
observed in the response signals obtained from conventional accelerometers. Hence,
it is clear that SCC-sensors exhibit good vibration sensitivity, and the responses are
comparable with the conventional accelerometers. However, the obtained signals
from SCC-sensors exhibit very good response of impulse hitting but showed a small
drift in the time domain and is asymmetric around the steady-state value as shown
in Fig. 8a.
Development of Smart Cementitious Composite Sensors … 213
Fig. 8 Response obtained from a SCC-sensor b sensor output in selected portion c output from
the conventional accelerometer
In order to obtain the characteristic behavior of the response signals and modal
information of the vibrating beam, Fast Fourier Transform (FFT) and Power Spec-
tral Density (PSD) signal processing tools were used. The output responses of the
vibrating RC beam captured from the conventional accelerometers are shown in
Fig. 9. It is observed that there are three visible peaks at frequency 8.72, 34.69, and
86.12 Hz within 100 Hz. These peaks and their corresponding frequency(ies) repre-
sent the first, second, and third vibration mode of RC beam, respectively. Figures 10
and 11 show the responses obtained from SCC-sensors using FFT and PSD methods,
respectively. It can be seen in Fig. 10 that SCC-sensors incorporated with 0.50 wt.%
COOH-CNT have some peaks in their response signal indicating the capability of
vibration sensing of developed sensors. Though, these SCC-sensors contain the vibra-
tion modes corresponding to the frequency peaks, but unable to clearly distinguish
these peaks. This can be attributed due to insufficient voltage generation due to lack
of sufficient CNTs particles within the cementitious matrix. However, it is revealing
that increase in the dosage of CNTs can increase the sensitivity of SCC-sensors. The
same can be observed in Fig. 11. The results obtained from the 0.75 wt.% CNTs
shows the excellent vibration sensitivity with multiple voltage peaks suggesting that
sufficient number of CNT particles must be available to obtain and capture the actual
Fig. 9 Output response obtained from conventional accelerometers a FFT, b PSD method
214 R. K. Rao and S. Sasmal
Fig. 10 Output response obtained from SCC with 0.50 wt.% COOH- a FFT b PSD method
Fig. 11 Output response obtained from SCC with 0.75 wt.% COOH- a FFT b PSD method
vibration modes. It can be found that the developed SCC-sensors are able to iden-
tify the frequencies of the first three fundamental frequency modes in the range of
0–100 Hz. However, there was no clear peaks detected above 100 Hz. Within the
frequency range 0–100 Hz, the output responses obtained from SCC-sensors show
very good agreement with the responses obtained from the conventional accelerome-
ters. In addition, there is a sudden diminish in the voltage amplitude of SCC-sensors.
This sudden diminish in the voltage amplitude can be attributed due to polarization
phenomenon and disturbance in the conductive networks within the cementitious
matrix. The similar observation has also been reported by Ubertini et al. [13] in their
study on vibration monitoring of long RC beam using CNT/cement-based sensors.
The frequencies of the fundamental vibration modes of RC beam obtained from
conventional accelerometers and developed SCC-sensors are listed in Table 3. Table
3 shows that the deviation in the frequencies in all three modes of vibration obtained
from the SCC-sensor from the conventional ones is less than 2%, indicating the
Development of Smart Cementitious Composite Sensors … 215
Experimental study reveals that the developed sensors are able to capture the frequen-
cies of first three vibration modes (predominant global modes) of large RC beam.
Based on these results their sensitivity can be calculated using the peak-to-peak
amplitude of their response signals. Since both sensors are attached on the same RC
beam but opposite surfaces, therefore applied dynamic force on the RC beam will
produce same input excitation for both sensors. Therefore, peak to peak amplitudes
of respected signals represent their output responses for same input signal (shown in
Fig. 12). For sensitivity calculation first twenty peaks in the response signals obtained
from both sensors were identified and processed. The vibration sensing capacity of
SCC-based sensor can be evaluated as the ratio of electrical responses captured from
SCC-sensors in term of voltage (mV) to the responses obtained from conventional
accelerometers in term of ‘g’. The sensitivity can be expressed as Eq. 2 [29]. The
vibration sensitivity of developed sensors with 0.75 wt.% of COOH-MWCNTs was
found to be 34.24 mV/g.
Fig. 12 Output responses captured from a conventional sensors b SSC with 0.75 wt.% COOH
216 R. K. Rao and S. Sasmal
VSCC, pk− pk
SSCC = . (2)
gC A, pk− pk
6 Conclusion
Acknowledgements Authors would like to acknowledge the support received from the Special
and Multifunctional Structures Laboratory (SMSL) of CSIR-SERC.
References
1. Brownjohn JM (2007) Structural health monitoring of civil infrastructure. Philos Trans R Soc
A Math Phys Eng Sci 365(1851):589–622. https://doi.org/10.1098/rsta.2006.1925
2. Sun M, Staszewski WJ, Swamy RN (2010) Smart sensing technologies for structural health
monitoring of civil engineering structures. Adv Civil Eng
3. Grammatikos SA, Paipetis AS (2012) On the electrical properties of multi scale reinforced
composites for damage accumulation monitoring. Compos B Eng 43(6):2687–2696. https://
doi.org/10.1016/j.compositesb.2012.01.077
4. Alokita S, Rahul V, Jayakrishna K, Kar VR, Rajesh M, Thirumalini S, Manikandan M (2019)
Recent advances and trends in structural health monitoring. In: Structural health monitoring of
biocomposites, fibre-reinforced composites and hybrid composites. Woodhead Publishing, pp
53–73
5. Güemes A, Fernandez-Lopez A, Pozo AR, Sierra-Pérez J (2020) Structural health monitoring
for advanced composite structures: a review. J Compos Sci 4(1):13. https://doi.org/10.3390/jcs
4010013
6. Zheng Q, Han B, Ou J (2019) NanoComposites for structural health monitoring. In: Nanotech-
nology in eco-efficient construction. Woodhead Publishing, pp 227–259. https://doi.org/10.
1016/B978-0-08-102641-0.00011-6
7. Rana S, Subramani P, Fangueiro R, Correia AG (2016) A review on smart self-sensing
composite materials for civil engineering applications. AIMS Mater Sci 3(2):357–379. https://
doi.org/10.3934/matersci.2016.2.357
8. Sanati M, Sandwell A, Mostaghimi H, Park SS (2018) Development of nanocomposite-based
strain sensor with piezoelectric and piezoresistive properties. Sensors 18(11):3789. https://doi.
org/10.3390/s18113789
9. Ubertini F, Laflamme S, Ceylan H, Materazzi AL, Cerni G, Saleem H, et al (2014)
Novel nanocomposite technologies for dynamic monitoring of structures: a comparison
between cement-based embeddable and soft elastomeric surface sensors. Smart Mater Struct
23(4):045023
10. Azhari F, Banthia N (2012) Cement-based sensors with carbon fibers and carbon nanotubes
for piezoresistive sensing. Cem Concr Compos 34(7):866–873. https://doi.org/10.1016/j.cem
concomp.2012.04.007
11. Donnini J, Bellezze T, Corinaldesi V (2018) Mechanical, electrical and self-sensing properties
of cementitious mortars containing short carbon fibers. J Build Eng 20:8–14. https://doi.org/
10.1016/j.jobe.2018.06.011
12. Sasmal S, Ravivarman N, Sindu BS, Vignesh K (2017) Electrical conductivity and piezo-
resistive characteristics of CNT and CNF incorporated cementitious nanocomposites under
static and dynamic loading. Compos Part A Appl Sci Manuf 100:227–243. https://doi.org/10.
1016/j.compositesa.2017.05.018
13. Ubertini F, Materazzi AL, D’Alessandro A, Laflamme S (2014) Natural frequencies identifi-
cation of a reinforced concrete beam using carbon nanotube cement-based sensors. Eng Struct
60:265–275. https://doi.org/10.1016/j.engstruct.2013.12.036
14. Naeem F, Lee HK, Kim HK, Nam IW (2017) Flexural stress and crack sensing capabilities
of MWNT/cement composites. Compos Struct 175:86–100. https://doi.org/10.1016/j.compst
ruct.2017.04.078
15. Kim HK (2015) Chloride penetration monitoring in reinforced concrete structure using carbon
nanotube/cement composite. Constr Build Mater 96:29–36. https://doi.org/10.1016/j.conbui
ldmat.2015.07.190
218 R. K. Rao and S. Sasmal
16. Dassios KG, Alafogianni P, Antiohos SK, Leptokaridis C, Barkoula NM, Matikas TE (2015)
Optimization of sonication parameters for homogeneous surfactant-assisted dispersion of
multiwalled carbon nanotubes in aqueous solutions. J Phys Chem C 119(13):7506–7516.
https://doi.org/10.1021/acs.jpcc.5b01349
17. Meoni A, D’Alessandro A, Downey A, García-Macías E, Rallini M, Materazzi AL et al (2018)
An experimental study on static and dynamic strain sensitivity of embeddable smart concrete
sensors doped with carbon nanotubes for SHM of large structures. Sensors 18(3):831. https://
doi.org/10.3390/s18030831
18. Chung DDL (2002) Piezoresistive cement-based materials for strain sensing. J Intell Mater
Syst Struct 13(9):599–609. https://doi.org/10.1106/104538902031861
19. Yoo DY, You I, Youn H, Lee SJ (2018) Electrical and piezoresistive properties of cement
composites with carbon nanomaterials. J Compos Mater 52(24):3325–3340. https://doi.org/
10.1177/0021998318764809
20. Lee SJ, You I, Zi G, Yoo DY (2017) Experimental investigation of the piezoresistive properties
of cement composites with hybrid carbon fibers and nanotubes. Sensors 17(11):2516. https://
doi.org/10.3390/s17112516
21. Andrawes B, Chan LY (2012) Compression and tension stress-sensing of carbon nanotube-
reinforced cement. Mag Concr Res 64(3):253–258. https://doi.org/10.1680/macr.10.00182
22. Konsta-Gdoutos MS, Aza CA (2014) Self-sensing carbon nanotube (CNT) and nanofiber (CNF)
cementitious composites for real time damage assessment in smart structures. Cem Concr
Compos 53:162–169. https://doi.org/10.1016/j.cemconcomp.2014.07.003
23. Li GY, Wang PM, Zhao X (2007) Pressure-sensitive properties and microstructure of carbon
nanotube reinforced cement composites. Cem Concr Compos 29(5):377–382. https://doi.org/
10.1016/j.cemconcomp.2006.12.011
24. Cha SW, Song C, Cho YH, Choi S (2014) Piezoresistive properties of CNT reinforced cemen-
titious composites. Mater Res Innov 18(sup2):S2-716. https://doi.org/10.1179/1432891714Z.
000000000564
25. Rao R, Sindu BS, Sasmal S (2020) Synthesis, design and piezo-resistive characteristics of
cementitious smart nanocomposites with different types of functionalised MWCNTs under
long cyclic loading. Cem Concr Compos 103517. https://doi.org/10.1016/j.cemconcomp.2020.
103517
26. Han B, Yu X, Kwon E (2009) A self-sensing carbon nanotube/cement composite for
traffic monitoring. Nanotechnology 20(44):445501. https://doi.org/10.1088/0957-4484/20/44/
445501
27. Saafi M (2009) Wireless and embedded carbon nanotube networks for damage detection in
concrete structures. Nanotechnology 20(39):395502. https://doi.org/10.1088/0957-4484/20/
39/395502
28. D’Alessandro A, Ubertini F, Materazzi AL, Porfiri M (2014) Electrical modelling of carbon
nanotube cement-based sensors for structural dynamic monitoring. In: AIP conference
proceedings (vol 1603, No. 1, pp 23–30). American Institute of Physics
29. Coskun MB, Qiu L, Arefin MS, Neild A, Yuce M, Li D, Alan T (2017) Detecting subtle
vibrations using graphene-based cellular elastomers. ACS Appl Mater Interf 9(13):11345–
11349. https://doi.org/10.1021/acsami.7b01207
Detection of Debonds
in Reinforced Concrete Using Ground
Penetrating Radar
1 Introduction
Reinforced Concrete (RC) is the most widely used construction material for infras-
tructure development around the world. Internal discontinuities are among the
common flaws in RC structures due to various reasons like chemical attack, poor
2 Theoretical Background
J = σ.E (1)
D = ε.E (2)
222 S. T. Kuchipudi et al.
In the above complex equation, the real part εc indicates energy storage capacity
of concrete and the imaginary part represents the losses in electromagnetic energy.
The permittivity εe is to be divided by permittivity of air ε0 (a constant) to obtain
relative permittivity εr .
Velocity of EM wave being denoted as c and εr and εr are the real and imaginary
components of the relative permittivity εr respectively. As concrete has low electrical
conductivity which yields low losses, and the equation eventually can be rounded
off as
c
υ= (6)
εr
Thus Eq. (6) concludes that different media with diverse dielectric constants
εr have distinct wave velocities and wave scattering characteristics. The dielectric
constant of steel rebars is infinite as metals possess zero net electric field inside them
[16]. Hence there will be a huge gap between pulse velocities in concrete and steel.
Detection of Debonds in Reinforced Concrete Using Ground … 223
Generally, GPR has three working components: (a) control unit, generates and
collects pulses, (b) antenna, a transducer and receiver set for radiating signals into
specimen and collecting them back, and (c) a recording unit, collects digital output
from control unit, stores and feeds the user interface. Short packets of electromag-
netic energy are emitted into the testing media through transducer. The waves undergo
reflection and scattering upon striking obstacles with distinct dielectric constants in
their path. The reflected energy is gathered and displayed as waveforms with which
essential characteristics are assessed.
The position of GPR antenna also plays an important role in the speed and
resolution-depth of data acquire. There are two categories of antennas based on their
relative position with working specimen, either a ground-coupled or an air-coupled
antenna. Compared to air-contact antenna, the clear advantage with ground-coupled
one is its superior resolution and high depth of penetration.
The time domain data obtained after collection at the receiver is to be represented in
required interpretable form. The pulse characteristics over time positioned at a single
point is the A-scan. Although A-scan presents all required details focused on a single
point, it is not a choice for inspection of large surfaces. Contrarily B-scan assembles
all the one-dimensional data from A-scan over scanning length axis and converts it
into a visually appealing 2D images. Amplitude of signals is represented with pixel
intensities on a color scale. These images are most prominently used to depict results
of a non-destructive survey owing to continuity of data. In this investigation B-scans
are presented for the comfort of the reader to take away the key annotations in a
moment.
The reflected waves collected by the receiver undergoes several stages of processing
to produce an indicative output of the subsurface characteristics in the form of images.
The raw signal needs to be corrected for the surface reflections at the air-concrete
interface by offsetting the reference position to a specified break point. This offset
is said to be time zero correction, and it depends on the dielectric constant of the
medium as well as frequency of the waves used.
The waves reflected experiences attenuation due to losses, and they fail to present
details of deep elements. Multiplicative gain functions need to be applied after noise
reduction to strengthen the signal. There are several techniques which are generally
224 S. T. Kuchipudi et al.
used for enhancement of SNR. These are helpful to reveal the actual rebar and defect
signatures. Time reversal, Reverse time Migration (RTM) [17, 18] etc. are some of
the used filtering/data processing techniques for the GPR and ultrasound wave based
NDE. However, many of the techniques which require detailed processing and are
therefore, time consuming. Migration in time domain is one of very effective to the
kind of pitch catch radar data in concern. It was first incorporated for positioning
subsurface targets using GPR data for geophysical applications [18]. It can detect
smaller reflectors along with the main scatterer as evident in [19]. The time domain
migration technique has been adopted in this investigation for increasing the SNR
of the rebar/defect images.
For an original radar signal X (t), Hilbert transform Y (t) produces the analytic signal,
1 X (t )
Y (t) = P dt (7)
π t − t
where a(t) is the amplitude and θ (t) is the phase component. Each component as a
function of time gives the instantaneous characteristics of the radar signal thus, offers
an improved contrast of reflection and absorption of wavelets in B-scan image.
4 Experimental Investigation
Surface of the slab was smoothened to ensure comfortable ground coupling of the
antenna. Gridlines were marked for every 10 mm intervals along and across the
specimen as shown in Fig. 2b. Antenna was connected to the control unit and a
power source through cables. A computing device, which stores and accesses pulse
data, is digitally connected to the setup.
Ground-coupled configuration has been adopted for an antenna of 900 MHz center
frequency to get healthier readings. This antenna was first calibrated for pulse veloc-
ities in the specimen, appropriate module for this frequency has been initiated, and
reference coordinates were set. Radar waves were radiated into specimen by running
Fig. 1 (a) Casting the specimen, (b) Schematic diagram with plan view (c) Sectional view of slab
226 S. T. Kuchipudi et al.
Fig. 2 Laboratory setup (a) Complete GPR equipment (b) Representation of rebar alignment and
position of three debonds
the transducer of 900 MHz frequency on each gridline one after the other sequentially
along the major axis (in ) of rebars as presented in the schematic (Fig. 3). The scan-
ning procedure was carried out with the geometric center of antenna coinciding with
the gridline. Each gridline was scanned along its complete length, and the reflected
pulse data was saved for post processing. The similar operation was repeated for every
gridline across the major axis (jn ) of rebars, and 59 + 59 sets of data were recorded
in total. This recorded information is time-domain response of scattered waves in
rebar-debond embedded concrete media. The same process is repeated with antenna
of frequency 2 GHz.
The pulse data received from reflections is processed and analyzed to get corre-
sponding B-scan images of slab detailing the rebars and delaminations.
The image B-scan results upon scanning across the slab are discussed here. Figure 4
represents the processed images after gain and corrections obtained from the wave
data of 900 MHz transducer plotted between transducer position (m) and depth of
vertical study (m). Although depth of slab being 200 mm, results are presented
for depth up to 300 mm to show bottom reflection profile of waves. The image
based on transverse scan of the specimen across aperture 1 shown in Fig. 4a clearly
shows the presence of rebars in the form of three bright spots at 100 mm level. The
reflections at bottom edge of slab can be noticed near 0.2 m depth range. Slight
fading of color can be noticed for the first rebar in Fig. 4a which may be attributed to
weakened diminished signal strength due to the presence of intermediate aggregate
obstructions. There is no clear indication of the presence of delamination around
rebar in any of the images.
To obtain the accurate position and details of debond, 2 GHz antenna scanning results
of aperture-1 over the 12 mm rebar are presented in Fig. 5. The series of arcs and other
random bright spots in Fig. 5b corresponds to noise because of aggregate scattering.
Undulation around 0.2 m depth range can be accounted for bottom edge of slab,
Fig. 4 B-scan images with depth of slab on y-axis for 900 MHz scan after processing along
(a) aperture 1 (b) aperture 2 and (c) aperture 3
228 S. T. Kuchipudi et al.
Fig. 5 Scanning results of 2 GHz antenna (a) Schematic figure of slab scanning on aperture 1 and
B-scans after (b) Smoothened Gain (c) Migration in Time domain (d) Hilbert Transformation
Fig. 6 Scanning results of 2 GHz antenna 1 (a) Schematic figure of slab scanning on aperture 2
and B-scans after (b) Smoothened Gain (c) Migration in Time domain
Fig. 7 Scanning results of 2 GHz antenna 1 (a) Schematic figure of slab scanning on aperture 3
and B-scans after (b) Smoothened Gain (c) Migration in Time domain
slope due to its weight in concrete ambience, as a clear and gradual drift in its position
is noticeable from all sets of images.
The scan results of a gridline corresponding to a location without any predefined
debonds are presented in Fig. 8. The image after applying smoothened gain exposes
the rebar positions. Polarity of the reflected signal is similar for all rebars in this image
compared to that of in Figs. 5, 6, and 7, where an additional polarity inversion for the
bar with debond installed can be seen. This contrast in characteristics exhibited by
the dual configured rebars makes debonds distinguishable in images. The presence
of continuous, intense artifacts visible between 0.2 and 0.3 m range are due to strong
reflections by the metallic carriers of forklift placed on top of concrete flooring.
Apart from the desirable aspects, extreme fluctuations in amplitudes of the
reflected radar waves in the B-scan images are visible in Figs. 5, 6, 7 and 8. Migration
in time domain significantly increases the SNR of B-scans where sources of noise
are multiple reflections from the boundaries and aggregate scattering. However, the
230 S. T. Kuchipudi et al.
Fig. 8 Scanning with 2 GHz antenna (a) Schematic figure of slab scanning on random gridline and
B-scans after (b) Smoothened gain (c) Migration in Time domain
migrated images suffer some artifacts which cannot be removed. Sometime bound-
aries, weak reflections from other sources, get mapped at the wrong location and
results in increase in background intensities of the image. However, in the present
case the migrated time domain processing successfully distinguishes between rebar
and delaminated rebar over the background noises present in the images.
6 Conclusion
This paper presented a study on the investigation of rebar debond detectability using
GPR. Artificial flaws have been embedded at a specified depth around rebars of
different diameters. Though high frequency radar waves experiences attenuation,
GPR proved to be effective in detection of rebar. Some of the key highlights from
the study are as follows:
• GPR study is capable of positioning rebars accurately. Defects around the rebar
have also been detected in the form of a change in polarity. Several post-processing
techniques have been explored for the image quality enhancement. The time-
domain migration is found to be most suitable for the rebar and defect detection.
• Comparative analysis of 900 MHz and 2 GHz GPR antenna shows that improved
resolution of subsurface features can be obtained with the high-frequency radar
waves at the shallower depths.
• Although the proposed technique successfully detects delamination, the analysis
did not quantify the debond specifications like its thickness and length.
• Both migration in time domain and Hilbert transform are not successful in
suppressing the uninvited artifacts manifested in images.
• Being computationally faster, the implemented algorithm in this study, i.e., Migra-
tion in Time Domain, has an edge over the other techniques like Time Reversal
and similar algorithms for field inspections.
Detection of Debonds in Reinforced Concrete Using Ground … 231
Future studies can focus on exploration of newer and robust signal processing
techniques for the accurate detection and quantification of defects in the reinforced
concrete structure. Statistical methods can also be employed to quantify the severity
of the defect.
References
18. Beniwal S, Ganguli A (2015) Defect detection around rebars in concrete using focused ultra-
sound and reverse time migration. Ultrasonics 62:112–125. https://doi.org/10.1016/J.ULTRAS.
2015.05.008
19. Qin Y, Wang Q (2012) Kirchoff migration algorithm for ground penetrating radar data. In:
Proceedings 2012 international conference on computer science and electronics engineering,
ICCSEE 2012. vol 2, pp 396–398. https://doi.org/10.1109/ICCSEE.2012.256
20. Dinh K, Gucunski N, Duong TH (2018) Migration-based automated rebar picking for condition
assessment of concrete bridge decks with ground penetrating radar. NDT E Int 98:45–54. https://
doi.org/10.1016/J.NDTEINT.2018.04.009
Highly Depth Resolved Coded Thermal
Wave Imaging Technique for Infrared
Non-destructive Testing and Evaluation
1 Introduction
R. Mulaveesala (B)
Centre for Sensors, Instrumentation and Cyber-Physical Systems Engineering (SeNSE), Indian
Institute of Technology Delhi, Hauz Khas, New Delhi 110016, India
e-mail: mulaveesala@sense.iitd.ac.in
V. Arora
Indian Institute of Information Technology Una, Distt., Una, Shimla, Himachal Pradesh, India
G. Dua
Thapar Institute of Engineering and Technology, Patiala, Punjab, India
2 Theory
Fig. 2 Auto-correlation functions of complementary coded pair and their resultant sum
236 R. Mulaveesala et al.
b 0.4
c 0.6
d 0.8
e 1.0
f 1.2
Table 1 Specimen
Parameters Mild steel
parameters
Thermal conductivity 60.5
k [W/(m K)]
Heal capacity 434
Cp [J/(Kg K)]
Density 7854
ρ [Kg/m3 ]
For the present study, a mild steel specimen having blind holes as defects of 10 mm
diameter located at depths ranging from 0.2 mm to 1.2 mm in increments of
0.2 mm from the front surface of specimen is numerically developed using FEA
modeling. The specimen thickness is 2 mm. Figure 3 shows the schematic diagram
of the modeled specimen. The model is meshed with finer tetrahedral elements. The
simulation parameters of the specimen are given in Table 1.
4 FEA Results
In order to verify the enhanced defect revealing capabilities of the proposed coded
excitation scheme, a pair of complementary coded thermal stimulus of 50 W/m2 heat
flux is imposed over the specimen for duration of 100 s. The temporal thermal history
over the specimen is acquired at a frame rate of 20 Hz. For further processing is carried
out by appropriate polynomial fit onto the captured simulated thermal response data
to reconstruct the zero mean thermal profile.
For each one of the complementary coded sequence, correlation coefficient
between the temporal thermal map of each pixel in the field of view with a chosen
Highly Depth Resolved Coded Thermal Wave Imaging … 237
reference non-defective pixel is then computed. The obtained correlation profiles are
then added to achieve maximum peak with suppressed sidelobes.
Figure 4a, b shows the depth scanning capability obtained from the sequence ga
(Fig. 1 (a)) and the sequence gb (Fig. 1b), respectively, whereas Fig. 4c shows obtained
depth scanning from their correlation sum. It is clear from Fig. 4c that detection capa-
bility of correlation sum leads to improved sensitivity and resolution due to reduction
of side lobes and increased peak height along with its improved compression prop-
erties while adding the correlations of complementary codes. Further Fig. 4d shows
the frequency domain phase stack reconstructed from the compressed resultant pulse
obtained from the auto-correlation sum of the individual correlation sequences of
each complimentary code. It is clear from the reconstructed phase sequence that the
detectability is far superior to the individual correlated sequence stacks. The obtained
performance of correlation sum shows better depth scanning properties than any one
of the individual code sequences due to its high signal to noise ration obtained from
the proposed approach.
It is further observed that the correlation peak of the deepest defect is close to
the peak of reference non-defective pixel. With the increase in depths from 0.2 mm
to 1.2 mm, shift in the peaks of correlation profiles decreases. Depth dependence of
peak delay is shown in Fig. 5. It is also clear from this figure that defects located near
to the front surface of the sample exhibit more delay as compared to deeper defects
of same diameter.
5 Conclusion
In this work, a finite element analysis modeling is proposed to inspect a mild steel
specimen having flat bottom hole defects using 8-bit Golay Coded Thermal Wave
Imaging Technique (GCTWI). The results show that addition of correlation profiles
of complementary coded pair sequences leads to elimination of sidelobes with most
of the power concentrated in the main lobe. This technique has great potential to
make defect determination with higher test resolution and sensitivity for detection
of subsurface defects.
238 R. Mulaveesala et al.
Fig. 4 (continued)
240 R. Mulaveesala et al.
References
1. Maldague XPV (2001) Theory and practice of infrared thermography for nondestructive testing.
Wiley, New York
2. Almond DP, Lau SK (1994) Defect sizing by transient thermography I: an analytical treatment.
J Phys D Appl Phys 27(5):1063–1069
3. Busse G (1979) Optoacoustic phase angle measurement for probing a metal. Appl Phys Lett
35(10):759–760
4. Maldague X, Marinetti S (2003) Pulse phase infrared thermography. Rev Scient Instrum 74(1
II):417–419
5. Mulaveesala R, Tuli S (2006) Theory of frequency modulated thermal wave imaging for
nondestructive subsurface defect detection. Appl Phys Lett 89(19):91913
6. Mulaveesala R, Pal P, Tuli S (2006) Interface study of bonded wafers by digitized linear
frequency modulated thermal wave imaging. Sens Actuators A 128(1):209–216
7. Mulaveesala R, Tuli S (2005) Digitized frequency modulated thermal wave imaging for
nondestructive testing. Mater Eval 63(10):1046–1050
8. Mulaveesala R, Vaddi JS, Singh P (2008) Pulse compression approach to infrared nondestruc-
tive characterization. Rev Scient Instrum 79(9):094901
9. Kher V, Mulaveesala R (2021) Probability of defect detection in glass fibre reinforced polymers
using pulse compression favourable frequency modulated thermal wave imaging. Infrared Phys
Technol 113(103616)
10. Rani A, Mulaveesala R (2020) Depth resolved pulse compression favourable frequency modu-
lated thermal wave imaging for quantitative characterization of glass fibre reinforced polymer.
Infrared Phys Technol 110(103441)
11. Kher V, Mulaveesala R (2020) Probability of defect detection in pulse compression favourable
thermal excitation schemes for infra-red non-destructive testing. Electron Lett 56(19):998–
1000
12. Kaur K, Mulaveesala R (2020) Efficient selection of independent components for inspection
of mild steel sample using infrared thermography. Electron Lett 56(19):990–993
13. Arora V, Mulaveesala R, Dua G, Sharma A (2020) Thermal non-destructive testing and evalua-
tion for subsurface slag detection. Numer Model Insight Non Destruct Test Condition Monitor
62(5):264–268
14. Dua G, Arora V, Mulaveesala R (2020) Defect detection capabilities of pulse compression
based infrared non-destructive testing and evaluation. IEEE Sensors J
15. Kher V, Mulaveesala R (2019) Probability of defect detection in pulse compression favourable
frequency modulated thermal wave imaging. Electron Lett 55(14):789–791
Highly Depth Resolved Coded Thermal Wave Imaging … 241
16. Mulaveesala R, Arora V, Rani A (2019) Coded thermal wave imaging technique for infrared
non-destructive testing and evaluation. Nondestruct Testing Eval 34(3):243–253
17. Dua G, Mulaveesala R, Kher V, Rani A (2019) Gaussian windowed frequency modulated
thermal wave imaging for non-destructive testing and evaluation of carbon fibre reinforced
polymers. Infrared Phys Technol 98:125–131
18. Ahmad J, Akula A, Mulaveesala R, Sardana HK (2019) Barker-coded thermal wave imaging for
non-destructive testing and evaluation of steel material. IEEE Sensors J 19(2)(8502804):735–
742
19. Arora V, Mulaveesala R, Rani A, Sharma A (2019) Digitised frequency modulated thermal
wave imaging for non-destructive testing and evaluation of glass fibre reinforced polymers.
Nondestruct Testing Eval 34(1):23–32
20. Dua G, Mulaveesala R (2018) Thermal wave imaging for non-destructive testing and evaluation
of reinforced concrete structures. Insight Non Destruct Testing Condition Monitor 60(5):252–
256
21. Suresh B, Subhani SK, Ghali VS, Mulaveesala R (2017) Subsurface detail fusion for anomaly
detection in non-stationary thermal wave imaging. Insight Non Destruct Testing Condition
Monitor 59(10):553–558
22. Arora V, Mulaveesala R (2017) Application of Golay complementary coded excitation schemes
for non-destructive testing of sandwich structures. Opt Lasers Eng 93:36–39
23. Mulaveesala R, Dua G, Arora V, Siddiqui JA, Muniyappa A (2017) Pulse compression
favourable aperiodic infrared imaging approach for non-destructive testing and evaluation of
bio-materials. In: Proceedings of SPIE the international society for optical engineering, vol
10214, no. 102140G
24. Arora V, Mulaveesala R, Bison P (2016) Effect of spectral reshaping on frequency modulated
thermal wave imaging for non-destructive testing and evaluation of steel material. J Nondestruct
Eval 35(1)(15):1–7
25. Siddiqui JA, Arora V, Mulaveesala R, Muniyappa A (2015) Infrared thermal wave imaging for
nondestructive testing of fibre reinforced polymers. Exp Mech 55(7):1239–1245
26. Dua G, Mulaveesala R, Siddique JA (2015) Effect of spectral shaping on defect detection in
frequency modulated thermal wave imaging. J Optics (United Kingdom) 17 (2)(025604)
27. Arora V, Siddiqui JA, Mulaveesala R, Muniyappa A (2015) Pulse compression approach
to nonstationary infrared thermal wave imaging for nondestructive testing of carbon fiber
reinforced polymers. IEEE Sensors J 15(2)(6936841):663–664
28. Mulaveesala R, Arora V (2017) Complementary coded thermal wave imaging scheme for
thermal non-destructive testing and evaluation. Quant InfraRed Thermogr J 14(1):44–53
29. Rani A, Mulaveesala R (2020) Investigations on pulse compression favourable thermal imaging
approaches for characterisation of glass fibre-reinforce polymers. Electron Lett 56(19):995–998
30. Ahmad J, Akula A, Mulaveesala R, Sardana HK (2020) Probability of detection of deep
defects in steel samples using barker coded independent component thermography. Electron
Lett 56(19):1005–1007
Microstructural Phase Detection in Steel
Components by Magnetic NDE Sensor
Abstract Steel is the most widely used material in structural components due to its
superior strength, ductility, and corrosion properties both at room temperature and
elevated temperatures. The microstructural phase detection by NDE techniques is
advantageous for the identification of any structural damage in steel structures and
components so as to avoid any catastrophic failure. With this viewpoint, a magnetic
NDE sensor, MagStar, is developed by CSIR-NML in collaboration with M/s, Tech-
nofour, Pune (India). The sensing head of this device is placed in contact with steel
components and important soft magnetic parameters like coercivity and Barkhausen
voltage are measured. Two important steels, interstitial free (IF) and rail steels, are
chosen for this research and different annealing treatment is done to generate different
microstructural phases. The annealing treatment generates strain-free recrystallized
microstructure in cold rolled IF steel and decarburized ferrite layer in rail steel.
These structural changes are related to increasing magnetic softness in IF and rail
steel, monitored by MagStar. Therefore, the electromagnetic NDE sensor is found
a potential tool for monitoring microstructural phases of steel during its processing
and service exposure as the components.
1 Introduction
The steel is the most widely used material in structural and service components of
the modern era due to its superior strength, ductility, and corrosion properties both at
room temperature and elevated temperatures [1]. The wide application and superior
properties of steel occur for its stable and suitable microstructures in different envi-
ronments. However, the allotropic transformation in steel possesses varying phases
2 Principle of EM Technique
Any ferromagnetic materials, like steel, consist of many small magnetic domains
which are distributed at different orientations under zero magnetic field (Fig. 1a).
With the application of magnetic field by a low carbon iron core (C-shaped) along any
steel components (tube, as shown in Fig. 1b), the magnetic fluxes are passed through
the material and the domains are aligned along the applied field direction with the
effect of domain wall movement. When the applied field is cyclic, the reversible
domain wall movement takes place, and the material becomes to its initial magne-
tized state. However, the reversion of this field does not occur through a similar path
and a magnetic hysteresis loop (MHL) is generated along the x–y axis (Fig. 1c).
Many magnetic parameters, e.g., coercivity (Hc), remanence (Br), magnetic satura-
tion induction (Bs), etc., are derived from this MHL curve. Although the MHL curve
line seems smooth, the magnetic induction at the microscopic scale (yellow circle
in Fig. 1c) is found a discontinuous change, forming as many individual steps with
a continuous magnetizing field. It is caused by the sudden jump of the domain wall
from one position to another during the magnetization process. The signal generated
due to these irreversible magnetization changes is known as magnetic Barkhausen
emission (MBE), which occurs due to the irreversible motion of 180° domain walls.
The MBE voltage is explained by r.m.s. voltage and pulse height distribution.
Microstructural Phase Detection in Steel Components … 245
Fig. 1 Magnetic behaviours of steel explained by a magnetic domain positions, b magnetic flux
lines along tube sample, and c generation of magnetic parameters
3 Experimental Techniques
Two types steels, viz., interstitial free (IF) and rail, are investigated in this research
and their compositions are explained in Table 1. These steels are heat treated at
different conditions for monitoring different microstructural behaviours, which are
described in Table 2.
The hot rolled IF steel plate is industrially processed through several cold rolling
and annealing steps. In this research, the industrially 80% cold rolled IF steel of 1 mm
thickness was cut as a rectangular shape (70 × 20 × 1 mm3 ) machined and annealed
at 700 °C for 5 and 405 min holding to obtain the partially and completely recrystal-
lized microstructures, respectively. The rail steel plate was cut as a rectangular shape
(70 × 30 × 10 mm3 ) machined and annealed at 1050 °C for 3 and 16 h to increase
the decarburization layer on the outer layer of the steel. The magnetic properties of
annealed samples were evaluated by an electromagnetic sensor Magstar described
elsewhere [12]. The applied magnetic field and frequency of the MHL curve were
1500Oe and 50 MHz, respectively. The MBE voltage was measured applying sinu-
soidal magnetic field strength of about 400–500Oe at the frequency of 40 Hz. The
longitudinal section of samples (10 × 5 × 1 mm3 ) was polished, etched by 6% Nital
solution, and followed by optical microscopy of Leica (DM2500M) and scanning
electron microscopy (SEM) of FEI (NovaNanoSEM430).
Microstructural Evolution
The IF steel is widely used in packaging and automobile industries, and during
its processing, the annealing treatment is effective for controlling its microstruc-
tural behaviour and resultant mechanical properties. Therefore, the monitoring of
annealing behaviour is important during the processing of this steel. The cold rolled
IF steel comprises elongated ferrite grains along the rolling direction (Fig. 2a). The
fine (~145 nm) titanium carbosulphide particles (point 1) are distributed in the matrix,
examined by SEM image (Fig. 2b).
The post-annealing treatment causes the stress relieving through annihilation and
rearrangement of dislocations at recovery and subsequent strain-free grain genera-
tion at recrystallization. A small annealing time (5 min) at 700 ºC generates a large
number of strain- free grains in the deformed matrix, indicating the partially recrys-
tallized state (Fig. 3a), while a long annealing time (405 min) results in the complete
recrystallized microstructure with a distribution of equiaxed ferrite grains (Fig. 3b).
Microstructural Phase Detection in Steel Components … 247
Fig. 3 Optical microstructures of IF steel annealed at 700 ºC for a 5 min and b 405 min
400
20000
Annealing temp: 700oC 200
As-rolled
15000 & 0
Magnetic Induction (Gauss)
Annealing Time
10000 -200
As-rolled
5 min -400
Amplitude (mV)
5000 Annealing time: 5 min
405 min 200
0 0
-200
-5000
-400
Annealing time: 405 min
-10000 Lowest 200
coercivity
0
-15000
-200
-20000 -400
-1500 -1000 -500 0 500 1000 1500 0 1000 2000 3000 4000
Magnetizing field (Oe) Counts (No)
(a) (b)
Fig. 4 Magnetic behaviour of as-rolled and annealed IF steel, explained by a MHL and b MBE
Table 3 Magnetic
Heat treatment condition Coercivity (Oe) MBE voltage (mV)
parameters of IF steel
(Temp./Time)
As-rolled 85.3 0.263
700 °C/5 min 59.7 0.157
700 °C/405 min 52.6 0.149
sites like grain boundaries, dislocations, precipitates, etc. Accordingly, the cold rolled
steel shows high MBE voltage than annealed steel due to high dislocation density in
the former condition. In small grains, the grain boundary area is more and the domain
wall consequently needs to overcome more pinning sites of dislocations associated
with grain boundary. Therefore, the MBE voltage decreases with increasing grain
size, interacting less grain boundary areas. Therefore, the monitoring of magnetic
parameters during steel processing shows an indication of microstructural change due
to the annealing of cold rolled steel. It can be helpful for controlling the microstructure
and resultant mechanical properties of the steel.
Microstructural Evolution
The as-received rail steel comprises a nearly fully pearlite microstructure, and the
decarburization from the outer layer takes place during industrial processing at high
temperature and long time exposure in rail service. Figure 5 describes the ferrite
layer formation over the pearlite structure. The degradation of microstructure causes
three different layers, viz., pearlite (core), ferrite-pearlite combined structure, and
ferrite. The thin iron oxide layer formed at the outer surface is removed by acid
rinsing treatment. The core is mostly a pearlite structure with a thin ferrite layer
Microstructural Phase Detection in Steel Components … 249
20 400
200
15 As Received As-received
1050C/3h 0
Magnetic Induction (kG)
10 1050C/16h -200
-400
Amplitude (mV)
5 Annealing time: 3h
200
0 0
5
As Received -200
-5
3h -400
Annealing time: 16h
200
-10 16h
-15 -200
0
-400
-20 0 1000 2000 3000 4000
-1500 -1000 -500 0 500 1000 1500 Counts (No)
Magnetizing field (Oe)
(b)
(a)
Fig. 6 Magnetic behaviour of rail steel (as-received and annealed), explained by a MHL and b
MBE
250 R. K. Roy et al.
3 h (64.0 Oe) and 16 h (58.2Oe). Similarly, the MBE envelopes are also changed
with decarburization (Fig. 6b), justified by lower MBE voltage in as-received steel
compared to annealed one (Table 4). Under the decarburized state, the pearlite phase
(which is magnetically harder) exists under the ferrite surface layer and magnetic
domain wall (DW) movement is easier in individual ferrite structures than ferrite-
pearlite combined structure. Accordingly, the coercivity and MBE voltage decreases
with increasing ferrite layer thickness at higher annealing time. In other words, the
decarburization is responsible for increasing ferrite layer thickness and increasing
soft magnetic properties of rail steel. Therefore, the coercivity is a potential magnetic
parameter for the estimation of decarburization depth of steel during processing and
service exposure, monitoring by electromagnetic NDE sensor.
5 Conclusions
In this research, two types of steels, interstitial free (IF) and rail, are investigated for
microstructural phase detection by an electromagnetic sensor and the conclusions
are as follows:
(i) The IF steel shows lowering of coercivity and MBE voltage with increasing
annealing time, justifying the enhancement of soft magnetic properties with
recrystallized strain-free microstructures.
(ii) The decarburization in rail steel decreases coercivity and increases MBE
voltage with increasing ferrite layer thickness.
(iii) The electromagnetic NDE sensor is found a potential tool for monitoring
microstructural phases of steel during its processing and service exposure as
the components.
References
3. Kwun H (1985) Effects of plastic deformation on magnetically induced ultrasonic wave velocity
changes in steel. J Appl Phys 57:1397–1399
4. Swartzendrube LJ (1989) Quantitative problems in magnetic particle inspection. Springer,
Boston, New York, pp 2133–2140
5. Ranjan R, Jiles DC, Rastogi PK (1987) Magnetic properties of decarburized steels: an
investigation of the effects of grain size and carbon content. IEEE Trans Magn 23:1869
6. Moorthy V, Vaidyanathan S, Baldev R, Jayakumar T, Kashyap BP (2000) Insight into the
microstructural characterization of ferritic steels using micromagnetic parameters. Metall
Mater Trans A 31A:1053–1064
7. Chen H, Xie S, Chen Z, Takagi T, Uchimoto T, Yoshihara K (2014) Quantitative nondestructive
evaluation of plastic deformation in carbon steel based on electromagnetic methods. Mater
Trans 55:1806–1815
8. Mitra A, Mohapatra JN, Swaminathan J, Ghosh M, Panda AK, Ghosh RN (2007) Magnetic
evaluation of creep in modified 9Cr–1Mo steel. Scripta Mater 57:813–816. https://doi.org/10.
1016/j.scriptamat.2007.07.004
9. Martinez-de-Guerenu A, Arizti FA, Dıaz-Fuentes M, Gutierrez I (2004) Recovery during
annealing in a cold rolled low carbon steel. Part I: Kinetics and microstructural characterization.
Acta Mater 52:3657–3664. https://doi.org/10.1016/j.actamat.2004.04.019
10. Zhou L, Liu J, Hao XJ, Strangwood M, Peyton AJ, Davis CL (2014) Quantification of the
phase fraction in steel using an electromagnetic sensor NDT. Int 67:31–35. https://doi.org/10.
1016/j.ndteint.2014.06.007
11. Roy RK, Dutta S, Panda AK, Rajinikanth V, Das SK, Mitra A, Strangwood M, Davis CL
(2018) Assessment of recovery and recrystallization behaviours of cold rolled IF steel through
nondestructive electromagnetic characterization Phil. Mag 98:1933–1944. https://doi.org/10.
1080/14786435.2018.1465241
12. Roy RK, Panda AK, Mitra A (2012) An electromagnetic sensing device for microstruc-
tural phase determination of steels through non-destructive evaluation. IEEE Explore 226–
229. https://doi.org/10.1109/ICSensT.2012.6461675
Application of NDE in Damage Tolerant
Analysis and Fracture Control
of Fracture Critical Metallic Structures:
An Overview
M. Anver Ali
M. A. Ali (B)
Liquid Propulsion Systems Centre, Indian Space Research Organisation, Thiruvananthapuram
695547, Kerala, India
1 Introduction
Fracture and failure processes are applicable to all ranges of length scales, ranging
from nanoscopic atomic bonds to macroscopic engineering structures. With the
advent of structures loaded in tension, unprecedented failures were observed in many
steel structures at loads well below the perceived tensile strength of the materials.
This sensitized lacunae in the prevailed understanding about the mechanical behavior
of materials. Two representative examples of such failures are given in Figs. 1 and 2.
The former figure corresponds to the Molasses tank failure that happened at Boston
in 1919. More than 2 million gallons of molasses were spilled, resulting in 12 deaths,
40 injuries, and massive property damage. This failure emphasized the importance
of Structural Health Monitoring (SHM) [1]. The latter figure corresponds to fuse-
lage failure in Aloha airlines (1988), which suffered an explosive decompression,
resulting in tearing of the ceiling of the Boeing 737. This catastrophe triggered the
awareness about the accelerated aging of structures due to corrosion [2].
The commonality in all such failures is that the failure of structure made of ductile
materials occurred in a brittle manner, without any perceptible plastic deformation.
It could be attributed to any of the factors such as the presence of a tri-axial state
of stress at the tip of inherent flaws and consequent flaw growth to a critical size,
environmental effects such as low temperature or corrosion and high loading rates.
Any engineering component may contain inherent flaws, either from the raw mate-
rial stage or introduced during fabrication, with associated complex stress regimes in
their vicinity, which may be either linear-elastic (for brittle materials), elastic–plastic
(for ductile materials with stable crack growth), or plastic collapse regimes (for mate-
rials with high toughness). LEFM and EPFM principles are applicable for the first
two regimes; whereas, the failure is due to tensile or stiffness instability across the
thickness of the component in case of plastic collapse, for which conventional design
approaches based on strength or stiffness or other failure modes such as buckling,
fatigue, creep, environmental degradation, and wear are adopted [3]. Whether, under
cyclic or sustained tensile stress, any of the such flaws does or does not propagate
depends on the following [4]:
• How the material behaves in the presence of crack.
• What is the initial size and geometry of the crack.
• Whether any aggressive environment is present or not.
• What is the geometry of the structure or component containing that flaw.
• What is the stress amplitude and the number of loading cycles.
• What is the duration of sustained load.
• What is the temperature to which the material is being exposed, and so on.
The advancements in fracture mechanics concepts improved the understanding about
the failure behavior of materials by recognizing the role of inherent flaws. Fracture
mechanics quantifies the critical combinations of three variables: the applied stress,
the flaw size, and fracture toughness of the material, which is the amount of energy
256 M. A. Ali
required to create fracture surfaces. Figure 3 illustrates the broad domains of fracture
mechanics, which collaborates areas such as Stress Analysis and Design, Material
Science, and Non-Destructive Testing [3].
The attribute “fracture critical” is designated to a part when its failure due to the
initiation and/or propagation of a crack may lead to catastrophic hazards and associ-
ated loss of life and/or mission. Fracture Control is the combination of approaches to
ensure safe operations without catastrophic failure by a fracture due to crack growth
from inherent flaws. The extent of the fracture control measures depends upon the
criticality of the structure, economic impact due to the breakdown of the structure as
well as the damage caused by failure. The process of fracture control involves a wide
spectrum of activities such as Damage Tolerance Analysis, material selection, design
optimization, structural testing, and scheduled maintenance/inspection/replacement.
Application of NDE in Damage Tolerant Analysis and Fracture Control … 257
Fracture control requires the determination of critical crack size and its influence
on structural strength as well as the time required for an initial flaw to grow to its
critical size using Damage Tolerant Analysis (DTA). DTA is an Integrity Management
Procedure, which predicts the remaining life of structures that are susceptible to
time-dependent flaw growth such as Fatigue, Environmentally Assisted Cracking,
and Creep and establish inspection intervals [1]. LEFM establishes the relationship
between crack growth per cycle and the applied stress intensity range through fatigue
crack growth laws. The crack growth rate and stress intensity factor can be correlated
to compute critical crack size for failure using the known value of fracture toughness.
The first step in DTA is the estimation of the critical flaw size (ac ). Then a tolerable
flaw size (at ) can be obtained either by incorporating a suitable safety factor on ac
(to compensate for uncertainties) or choosing a value of at , at which crack growth
rate (da/dt) is relatively small and hence the time for the flaw to grow from at to ac
during service is sufficiently long to preclude any catastrophe. DTA demands reliable
detection of manufacturing as well as service-induced flaws using appropriate NDI
techniques, based on which inspection intervals can be decided. If no significant
flaws are detected, the initial flaw size (ao ) is set as aNDE , which is the largest flaw
that might be missed (not the minimum size detected under favorable conditions) by
the NDE. As illustrated in Fig. 4, the useful service life (H) is the time for the flaw to
grow from ao to at , which is estimated using fracture mechanics concepts. The first
inspection interval (I1 ) decided shall be less than H (usually 50% of H), as shown in
Fig. 5.
258 M. A. Ali
After the first inspection, if no flaws greater than ao are detected, the second
inspection interval, I2 , will be equal to I1 . During subsequent inspections, if any flaw
a1 > ao is detected, the next inspection interval, I3 , must be less than I2 , as determined
through fracture analysis. Thus, the inspection intervals would become progressively
shorter, as shown in Fig. 6. It must be repaired or retired when either the flaw size
reaches the maximum tolerable size, or the inspections become too frequent to justify
continued operation [1].
Application of NDE in Damage Tolerant Analysis and Fracture Control … 259
If it can be proved that a flaw can grow in a stable manner through the wall of
a component without causing a catastrophic failure; hence, a leak can be detected
as an early warning and immediate remedial actions can be taken. This type of
260 M. A. Ali
NASA fracture control protocols demand that all human-rated aerospace flight
systems be subjected to fracture control measures to impede catastrophic events.
These protocols call for NDE to confirm that the likely failure initiation of relevant
crack-like flaws is absent in the critical areas. Accurate predictions of any structural
performance require characterizing the flaws. Non-Destructive Evaluation (NDE)
methods applied should be able to resolve these flaws as well as define its geometry
with reasonable accuracy. Different NDT methods, ranging from simple visual tests
to very elaborate techniques utilizing automated systems, are adopted in the industry.
The role of NDE is crucial in characterizing the “initial flaw size”, at the beginning
of the structure’s service life, which is the starting point for safe life analyses of
fracture critical parts as well as to identify and quantify the “detectable flaw size”
for consecutive “fitness-for-service” audits [10].
NDE method is selected based on its capability to detect and quantify the flaw, in
terms of their depth, length, angle with the part surface, the location from the surface,
and the spacing between the cracks in case of multiple cracks. An appropriate NDE
technique for retrospective inspection may be selected according to the limiting flaw
size to be determined, which corresponds to the assessment point on the FAD curve.
Usually, in order to detect most of the surface and internal flaws, multiple techniques,
as complementary to each other, are essential.
Generally, the following misunderstandings and assumptions about NDE appli-
cations prevails [10]:
• “After application and discrimination by an NDE procedure, no flaws exist”:
whereas, there are flaws smaller than the detection threshold is always present.
• “NDE capability is at the smallest flaw detected”: whereas, the real NDE
capability is the “Largest flaw missed”.
• “The detection and discrimination capability of an NDE procedure is at the cali-
bration level”: The detection capabilities are rarely at the level of “calibration”
threshold used to set the NDE response, since the increase in amplifier gain also
increases the “noise” and thus increases the “false call” level.
• “Responses from Cracks and slots are equal”: In fact, they are never equivalent
even to the cracks of an equivalent size.
• “All identically generated cracks are equal in response”: whereas they show
variation in NDE responses.
• “Cracks are equally detectable under both Laboratory and Field conditions”:
Responses will vary with testing conditions.
• “Critical crack size applicable everywhere”; however, from an economic stand-
point, different critical flaw sizes must be considered at different locations and
flaw orientations.
• “All NDE personnel perform at the same level”: In fact, the skills of personnel
involved are never similar.
Application of NDE in Damage Tolerant Analysis and Fracture Control … 263
Several factors influence the NDT results, such as Operator skill, Instrument quality,
Environment, Geometries of flaws and parts, and Material characteristics. Thus, flaws
are not detected and characterized in a deterministic manner and there always exists
an inherent probabilistic nature.
The Probability of Detection (POD) is defined as “the probability that, using a given
inspection procedure/system/sensor, an inspector will detect a flaw that is present in
the component”. As illustrated in Fig. 10, an idealized POD curve is a step function;
whereas, a realistic POD curve represents the NDE capability accurately. The POD
function provides an indication of how reliably flaws of all relevant sizes can be
found [11].
The POD concept was first pronounced in 1973. NASA evaluated NDE reliability
based on the maximum size flaw that can be missed instead of the smallest flaw
that can be detected. The NASA POD development project was a benchmarking
contribution in quantifying the NDE capabilities and was recognized as a standard
method. The NASA methodology of quantifying NDE capability was unified in
the standard NASA-STD-5009B [12]. For high reliability demanding applications,
the capabilities of the NDE techniques are quantitatively assessed and ensured that
permissible flaw sizes are detected with high probabilities and required confidence
level. The confidence level indicates the capability to estimate the POD from a definite
set of samples.
Military specification, MIL-A-83444 (USAF), “Airplane Damage Tolerance
Requirements” indicates that aNDE is that crack length for which it can be shown
there is 95 percent confidence that 90 percent of all cracks will be found. This
implies that there is a 5% probability that 90% is an overvalue of the true POD. This
crack size is denoted as a90/95 . aNDE is the longest crack that could pass undetected
through NDE [13].
Initially, the POD function involved the use of Binomial statistics. The results
are grouped into intervals of crack length with approximately the same POD esti-
mated as the number of detections divided by the number of opportunities. The first
analytical function fitted to estimate the POD function was during the “Have Cracks
Will Travel” program (1978). Under this USAF program fatigue damaged structural
samples were inspected using standard NDE methods. The NDE results and subse-
quent tear-down measurement results of these cracks are documented in a database
commonly known as “Have Cracks Will Travel” data. This database has results of
almost 22,000 inspections made on 174 cracks, each of which was inspected by 2 or
3 suitable NDE procedures by 107 different inspectors. The results were categorized
into 13 sets based on structure type and inspection method [14].
Inspection reliability and inspection sensitivity are employed to quantify results
from NDT methods. The sensitivity indicates the smallest flaw which is detected
and the largest one that is missed by the technique. Reliability in NDT is graphi-
cally expressed by POD, as a function of flaw length, which shows an increase in
probability with crack length (Fig. 11). For values of crack sizes between a1 and a2 ,
the probability is between zero and one. Any flaw size larger than a2 can always be
detected and those smaller than a1 never detected [15].
Berens et al. used the Log-linear logistics or log-odds model for extrapolating
the data into a POD function, given in Eq. (1), through regression analysis. The
regression curve with POD (a) function suggested by Berens et al. is illustrated in
Fig. 12.
exp(α + β ln ai )
P O D(ai ) = (1)
1 + exp(α + β ln ai )
where
POD (ai ) = probability of detection in ith interval;
ai = flaw size;
α and β = location and slope parameters (constants), respectively.
α and β can be estimated using maximum likelihood or log-odds method as explained
by Berens et al. [14]. The above equation was shown to fit the “Have Cracks” data
and may be an acceptable model for other NDE applications.
266 M. A. Ali
An event that has only two possible outcomes is referred to as a binomial event.
Let
N and n = Total no. of specimens with identical flaws and No. of flaws detected,
respectively;
P and q = Probability of detection and missing, respectively;
P (≥ n) = Probability of detecting n or more flaws (where, n can be 0, 1, 2….N);
G and p1 = Confidence Level and lower confidence bound, respectively.
Then
N
N
P (≥ n) = pi q N −i (2)
i
i=n
And
N
N
G =1− p1i (1 − p1 ) N −i (3)
i
i=n
The sample size required to estimate the lower confidence limit (p1 ) and confi-
dence level (G) can be determined by utilizing Eq. (3) A set of values (which must
be integers) can be computed for N and n, each pair of which indicates the number
of inspections and the number of detections required to achieve the specified POD
at the stated confidence level. For example, 90% POD at 95% confidence level can
be achieved as in Eq. (4)
N
N ˙
0.95 = 1 − (0.9)i (0.1) N −l (4)
i
i=n
Two of such combinations of N and n are 29/29 and 46/45 [17]. Table 1 below
summarizes combinations of N and n for different values of G and p1 [18].
In order to obtain how many additional NDE trials are required to improve the
reliability estimates of a particular batch of data, the following Eq. (5) can be used:
N +δ
N +δ
1−G = p1i (1 − p1 ) N +δ−1 (5)
i
i=n+δ−
where
δ and ε = Required no. of additional tests and maximum no. of non-detections,
respectively [17].
Application of NDE in Damage Tolerant Analysis and Fracture Control … 267
Table 1 Total number of samples for a given number of misses at a specified confidence level and
POD [18]
Level of confidence Number of misses Probability of detection
90% 95% 99%
90% 0 22 45 230
1 38 77 388
2 52 105 531
3 65 132 667
4 78 158 798
5 91 184 926
10 152 306 1,000+
20 267 538 1,000+
95% 0 29 59 299
1 46 93 473
2 61 124 628
3 76 153 773
4 89 181 913
5 103 208 1,000+
10 167 336 1,000+
20 286 577 1,000+
99% 0 44 89 458
1 64 130 662
2 81 165 838
3 97 198 1,000+
4 113 229 1,000+
5 127 259 1,000+
10 197 398 1,000+
20 325 656 1,000+
There are two methodologies: one is signal response (â) Vs. flaw sizes (a), and other
is binary data Vs. flaw sizes (hit/miss versus a).
The former is applicable when the results of the inspection are quantifiable and
recordable, such as in ultrasonic or eddy-current methods. In such cases, the POD
(a) function can be obtained from the statistical scatter in the signal response (â)
as a function of crack size (a) (Fig. 13). On the other hand, the hit-miss approach
is applicable to those NDI systems where the response is not quantifiable (such as:
visual, penetrant, magnetic particle, or film radiography testing) and gives results
as either discontinuity is present (hit → 1) or discontinuity is absent (miss → 0)
268 M. A. Ali
Two methods are typically used: in the first, discontinuities having known size and
shape are artificially seeded in the components. Using the NDT technique, these are
inspected, and, hit/miss data is created. In the latter method, various natural flaws
present in the components are characterized by destructive evaluations, and hit/miss
data is created in the same manner. Subsequently, the entire range of flaw lengths is
grouped into a number of intervals. The number of hits to the total number of flaws
in each interval gives the probability of detection in the mentioned intervals (Pi ). The
obtained POD values and mid-point of the respective intervals (ai ) are candidates of
the respective intervals. The hit/miss information obtained yields a number of (ai ,
Pi ) pairs.
Specimen Preparation. Specimens, with and without flaws, as well as with metal-
lurgical indications (i.e., with poor SNR) are prepared. Specimens are made from
material and heat treatment conditions similar to the one used in actual applications.
Defective regions shall be discarded, whereas regions with metallurgical indications
may be identified and included for study.
Number and Distribution of Flaws. MIL-HDBK-1823 recommends 60 flaws
distributed in each size interval. The allowable flaw size stipulated by the designer
is taken as the target value. To demonstrate 90% POD at 95% confidence level by
binomial analysis, a minimum of 29 out of 29 flaws with target size shall be detected
(Refer Table 1). The maximum number of specimens prepared shall have sizes that
of the target value, and also the specimens with cracks larger and smaller than the
target size shall be included, to establish the POD for different crack sizes.
Defect Insertion. Surface cracks are grown by fatigue loads (tensile or bending) from
the EDM starter notches, (ASTM E740) as shown in Fig. 15. Aspect ratio and length
of cracks shall be validated by break-open sufficient number of trial specimens. If
there is a lack of repeatability, the crack generation procedure shall be fine-tuned
by optimizing the parameters. The optimized values of parameters: starter notch
size, fatigue load levels, measured crack size and machining stock to be removed to
eliminate the starter notch, etc. shall be documented. Shape and size control of the
cracks can be done by varying the starter notch size and shape or the stress field or
both. In axial tension, the crack grows to nearly semicircular, whereas in bending, it
grows to an elliptical shape.
Internal Targets. These are the simulated voids and inclusions which can be incor-
porated by milling holes with appropriate size and morphology (and filling with
suitable materials for inclusions) on one of the faces of mating blocks, followed by
diffusion bonding and finish machining.
Unflawed Inspection Sites. At least twice of flawed sites are suggested to facilitate
the estimation of false call rate.
270 M. A. Ali
5 Conclusion
Any engineering materials adopted for various applications may contain flaws. Due to
the detectability limitations of common NDT techniques, it is assumed that discon-
tinuities of minimum detectable sizes are invariably present in every component,
even after the successful application of one or more NDE methods. This signifies the
importance of understanding the behavior of structures in the presence of flaws. The
Application of NDE in Damage Tolerant Analysis and Fracture Control … 271
process of fracture control involves a wide spectrum of activities. The role of NDE
is crucial in characterizing the “initial flaw size” which is the starting point for safe
life analyses of fracture critical parts. Also, it quantifies the “detectable flaw size”
for consecutive FFS analyses. This paper summarizes an overview about the key
concepts of fracture control and NDE system reliability assessment. The review of
these concepts is highly relevant to the contemporary context, since the Indian NDT
community has to play a vital role in the human-rating of Indian space systems.
References
1. Anderson TL (2017) Fracture mechanics: fundamentals and applications, (4th edn). CRC Press
2. ASM Hand Book. Fatigue and Fracture; ASM International, vol 19. USA
3. Ramesh K (2007) E-Book on engineering fracture mechanics. IIT Madras
4. Space Engineering: Fracture Control−ECSS-E-ST-32–01C (2009) European cooperation for
space standardization. The Netherlands
5. NASA Space Vehicles Design Criteria (Structures) (1970) Fracture control of metallic pressure
vessel−NASA SP-80 40. National Aeronautics and Space Administration, USA
6. API Recommended Practice 579−Fitness-for-Service (2000) 1st edn. American Petroleum
Institute
7. Barsom JM, Rolfe ST (1977) Fracture and fatigue control in structures: applications of fracture
mechanics, 3rd edn. ASTM, USA
8. BS 7910–2015 (2015) Guide to methods for assessing the acceptability of flaws in metallic
structures
9. Broek D (1988) The practical use of fracture mechanics. Kluwer Academic Publishers, USA
10. Rummel WD, Matzkanin GA (1997) Non-destructive evaluation (NDE) capabilities data book,
3rd edn. NTIAC Texas Research Institute Austin, Inc., USA
11. Hovey PW, Berens AP (1991) Statistical evaluation of NDE reliability in the aerospace industry.
University of Dayton Research Institute, USA
12. NASA Technical Standard (2019) Nondestructive evaluation requirements for fracture-critical
metallic components−NASA-STD-5009B. National Aeronautics and Space Administration,
USA
13. Lewis WH, Dodd BD, Sproat WH, Hamilton JM (1978) Final report−reliability of nondestruc-
tive inspections. Lockheed-Georgia Company, USA
14. Berens AP, Hovey PW (1981) Technical report: evaluation of NDE reliability characterization,
AFWAL-TR-81–4160, vol I. Air Force Wright Aeronautical Laboratories, USA
15. Zolfaghari A (2017) Farhad kolahan; technical paper: reliability and sensitivity of visible liquid
penetrant ndt for inspection of welded components. Material Testing 59(3):290–294
16. Brausch J, Butkus L, Campbell D, Mullis T, Paulk M (2008) Technical report: recommended
processes and best practices for non-destructive inspection of safety-of-flight structures, AFRL-
RX-WP-TR-2008–4373. Air Force Research Laboratory, USA.
17. Yee BGW, Chang FH, Couchman JC, Lemon GH, Packman PF (1975) Assessment of NDE
reliability data, NASA CR- 134991. National Aeronautics and Space Administration, USA
18. ASME Boiler & Pressure Vessel Code; Section V-Non-Destructive Examination; 2015th edn
19. Hand Book-NDE System Reliability Assessment; MIL-HDBK-1823A; 2009; Dept. of Defense,
USA
20. ASTM E 740 03; Standard Practice for Fracture Testing with Surface-Crack Tension Specimens
A Novel Analytical Approach
for Nondestructive Testing
and Evaluation of Bone Implants Using
Frequency Modulated Thermal Wave
Imaging
Abstract In the recent decade, infrared thermography has gained its importance in
nondestructive testing and the evaluation of solid materials due to its whole field,
fast, noncontact, and in-service testing abilities. In infrared thermography, active
infrared thermography has been widely adopted as the most promising technique for
the structural health monitoring of various solid materials. The most frequently used
active infrared techniques for nondestructive testing and evaluation of various solid
structures are pulse thermography and lock-in thermography. But due to the innate
limitations of these techniques, their usage for health monitoring of solid materials is
limited. To overcome all these limitations of existing techniques, this work proposes
an aperiodic pulse compression favorable thermal wave imaging approach for subsur-
face features detection in bone implants. An analytical model has been developed
with the help of Green’s Function method for linear frequency modulated thermal
wave imaging for the detection of subsurface features in the bone implant. Titanium
alloys are commonly used materials for the manufacturing of bone implants. In this
1 Introduction
are evaluated based on the heat sink and thermal variation distribution, which can
be efficiently screened through the infrared-sensitive camera or detection unit [1,
8]. In this way, the heated material could clearly project the existence of micro or
macro-structural anomalies which appear as hot or cold regions in the screening.
The present work proposes an analytical approach for non-periodic heating to
demonstrate the detection ability for subsurface inclusions qualitatively and quan-
titively in modeled Titanium alloy-based bone implant. To evaluate the detection
ability of this approach, correlation coefficient-based data processing approach has
been implemented to the acquired thermal sequence for the heat flux with a frequency
sweep from 0.01 to 1 Hz for a total time span of 100 s. Then correlation coefficient has
been plotted with depths of inclusions and the behavior of graphs has been studied.
Further, all the solutions obtained from the analytical approach have been validated
with the finite element-based model solved with the help of COMSOL Multiphysics.
2 Analytical Solution
∂u i (x, y, z, t)
ρi ci = ki ∇ 2 u i (x, y, z, t) (1)
∂t
Fig. 1 Schematic of frequency modulated thermal heat flux imposed over the implant surface
∂u 3 (x4 , y, z, t)
qx4 = −k3 =0 (3)
∂x
All other surfaces are considered to be insulated and their boundary conditions
are represented as Sharma et al. [9, 14, 15]
∂u i (x, y1 , z, t) ∂u i (x, y2 , z, t)
= 0; =0
∂y ∂y
Where : y1 ≤y≤y2 ; i = 1, 2, 3; t 0 (4)
∂u i (x, y, z 1 , t) ∂u i (x, y, z 2 , t)
= 0; =0
∂z ∂z
Where : z 1 ≤z≤z 2 ; i = 1, 2, 3; t 0 (5)
And the interfacial boundary conditions between the parent material and the inclu-
sion is considered as a perfect thermal contact and can be described as Sharma et al.
[9, 14, 15]
where, i = 1, 2, 3; n = 3.
Hence, to solve Eq. (9) an appropriate generalized Green function can be described
as Sharma et al. [9]
gi j (x, y, z, t; ξ, ζ, λ, τ )
∞ ∞ ∞
kj Xie (x)X je (ξ )ϒ f (y)ϒ f (ζ )Zg (z)Zg (λ)
= e−βe f g (t−τ ) (10)
g=0 f =0 e=0
αj ηxe η y f ηzg
Where; i, j = 1, 2, 3; n = 3;
gπ √ fπ√
βe f g = βe2 + β 2f + βg2 ; βg =αi ; β f = αi ;
z2 y2
βe βe
Xie (x) = ai cos √ x + bi sin √ x
αi αi
βf
ϒ f (y) = cos √ y ;
αi
βg z2 2
Zg (z) = cos √ z ; ηzg = ∫ Zg (λ) dλ
αi z1
y2 2
η y f = ∫ ϒ f (ζ ) dζ ;
y1
n
kj x j+1 2
ηxe = ∫ X je (ξ ) dξ
j=1
α j xj
The eigenvalues Xe (βe , x) are then calculated from the transcendental equation.
The transcendental equations have been formulated by substituting the appropriate
eigenfunction in the homogeneous part of equations Eqs. (2) and (3) and the compat-
ibility condition represented in Eqs. (7) and (8). All this produces a set of 2n
number of linear homogenous equations with the constant coefficients which can
be further represented in the matrix form as [R][Z] = [C], where [R] = coefficients
matrix, [Z] = constants matrix, and [C] = null matrix [14]. Here the constants
278 A. Sharma et al.
∞
q0 t q0 Xie (x) 1 − e−βe t
2
u i (x, y, z, t) = T0 + +
ηx0 e=1 ηxe βe2
q0 ε i ( f 2 ε 2 − π )
+ √ e 4 (Er f [ ] − Er f [ ])
2 1
ηx0 π
∞
q0 ε Xie (x) i ( f 2 ε2 −κ 2 − π )−( ε κ )
+ √ e 4 π (Er f [4 ] − Er f [3 ])
e=1
2 ηxe π
∞
1 −βe2 t
+ e Xie (x)T0 [T01 + T02 + T03 ] (11)
e=1
ηxe βe
where i = 1, 2, 3
√ √
π T bt
T01 = T02 = T03 = T04 = T0 = 310.15 K;ε = √ ; = 2π f + ;
b T
εβe2 1+i 1 + i ε
κ= ; 1 = √ f ε; 2 = √
2π 2 2 2π
1+i 1 + i ε
3 = √ ( f ε + iκ); 4 = √ + iκ ;
2 2 2π
βe βe ki α
θii = √ xi ; θi = √ x ; ψi = ; here = (i + 1)
αi αi k αi
T01 = k1 ρ1 c1 sin(θ12 );
1 2
T02 = k2 ρ2 c2 cos(θ12 ) sin(θ23 − θ22 ) − 2ψ12 sin(θ12 ) sin θ23 − θ22
2
T03 = k3 ρ3 c3 [ψ12 sin(θ12 ) sin(θ22 − θ23 ) + cos(θ12 ) cos(θ22 − θ23 )] tan(θ33 − θ34 )
A Novel Analytical Approach for Nondestructive Testing … 279
The analytical solution for step heat flux has been obtained by following up the same
procedure as mentioned in the previous Sect. 2.1, for frequency modulated thermal
heat flux. The intensity of step input heat flux is 3000 W/m2 for a time span of 100 s
Further, the solution for the step input is represented as Sharma et al. [9–11]
∞
q0 t q0 Xie (x) 1 − e−βe t
2
u i (x, y, z, t) = T0 + +
ηx0 e=1 ηxe βe2
∞
1 −βe2 t
+ e Xie (x) T0 [T01 + T02 + T03 ] (12)
e=1
η xe βe
Note: The nomenclature of all the symbols has been presented in Appendix 1
(Table 2).
The three-dimensional Titanium alloy-based implant has been considered for model-
based evaluation via COMSOL Multiphysics as shown in Fig. 2. This model has
been considered to incorporate different types of inclusions like Air, Titanium Oxide
(TiO2 ), and Aluminum Oxide (Al2 O3 ) at different depths (1 mm-2 mm-3 mm) from
the surface. Overall, nine rectangular shape inclusions have been modeled in the
current study, with size 10 mm × 10 mm × 1 mm (Length × Width × Height). To
solve the finite element-based model, the finer mesh has been used with tetrahedral
elements. The thermal properties of Titanium alloy (Ti6Al4V), and all embedded
inclusions are given in Table 1.
The finite element-based analysis has been performed by imposing the frequency
modulated thermal heat flux of intensity q0 = 3000 W/m2 over the Titanium alloy
implant by sweeping the frequency from 0.01 to 1 Hz in a time span of 100 s. The
modeled sample has been insulated from all the sides and has an initial condition of
293.15 K. Further, the resultant data captured from simulation has been processed
with a pulse compression-based data processing approach for the detection of inclu-
sions embedded in the modeled implant. This processed data is then compared
with the analytically acquired data to validate the analytical model explained in
the previous section.
280 A. Sharma et al.
Fig. 2 Schematic of modeled Titanium alloy-based bone implant with inclusions at different depths.
(where I-II-III represent Al2 O3 inclusion at 1 mm-2 mm-3 mm depths, respectively; IV-V-VI repre-
sent TiO2 inclusion at 1 mm-2 mm-3 mm depths, respectively; VII-VIII-IX represent Air as inclusion
at 1 mm-2 mm-3 mm depths, respectively)
Fig. 4 Schematic of analytically obtained (a) correlation coefficient response at different air
inclusion depths and (b) correlation coefficient response with different inclusion depths
inclusion from the implant surface. This has been observed because of the poor values
of thermal effusivity of air, due to which it does not allow the heat-wave to travel
through it repeatedly and raises implant surface thermal response drastically.
Fig. 5 Schematic of numerically obtained correlation coefficient response at different depths for
(a) Al2 O3 inclusion and (b) TiO2 inclusion
A Novel Analytical Approach for Nondestructive Testing … 283
Fig. 6 Schematic of numerically obtained (a) correlation coefficient response at different air
inclusion depths, (b) correlation coefficient response with different inclusion depths
numerically have followed the similar pattern as that of results obtained analytically
(Figs. 3a, b and 4a, b), except for the values of the correlation coefficient.
5 Conclusion
Appendix 1
References
1. Zhu QF, Sun ZM, Ma TD, Li P, Zhang DH, Vavilov VP (2013) Thermal non-destructive testing
for the titanium implants. In: Advanced materials research. Trans Tech Publ, vol 785, pp 52–57.
https://doi.org/10.4028/www.scientific.net/AMR.785-786.52
2. Teo AJ, Mishra A, Park I, Kim Y-J, Park W-T, Yoon Y-J (2016) Polymeric biomaterials for
medical implants and devices. ACS Biomater Sci Eng 2(4):454–472. https://doi.org/10.1021/
acsbiomaterials.5b00429
3. Fousová M, Vojtěch D, Doubrava K, Daniel M, Lin C-F (2018) Influence of inherent surface and
internal defects on mechanical properties of additively manufactured Ti6Al4V alloy: compar-
ison between selective laser melting and electron beam melting. Materials 11(4):537. https://
doi.org/10.3390/ma11040537
4. Vykhodets V, Kurennykh T, Tarenkova N (2012) genesis of gas containing defects in cast tita-
nium parts. Titanium alloys–towards achieving enhanced properties for diversified applications,
p 43
A Novel Analytical Approach for Nondestructive Testing … 285
5. Moghaddam NS et al (2016) Metals for bone implants: Safety, design, and efficacy.
Biomanufact Rev 1(1):1–16. https://doi.org/10.1007/s40898-016-0001-2
6. Tilton M, Lewis GS, Manogharan GP (2018) Additive manufacturing of orthopedic implants.
Orthopedic biomaterials. Springer, pp 21–55. https://doi.org/10.1007/978-3-319-89542-0
7. Rabel K et al (2020) Controlling osteoblast morphology and proliferation via surface micro-
topographies of implant biomaterials. Sci Rep 10(1):1–14. https://doi.org/10.1038/s41598-020-
69685-6
8. Mulaveesala R, Arora V (2017) Complementary coded thermal wave imaging scheme for
thermal non-destructive testing and evaluation. Quant InfraRed Thermogr J 14(1):44–53.
https://doi.org/10.1080/17686733.2016.1229329
9. Sharma A, Mulaveesala R, Dua G, Arora V, Kumar N (2020) Digitized frequency modulated
thermal wave imaging for detection and estimation of osteoporosis. IEEE Sens J. https://doi.
org/10.1109/JSEN.2020.3043282
10. Sharma A, Mulaveesala R, Arora V (2020) Novel analytical approach for estimation of thermal
diffusivity and effusivity for detection of osteoporosis. IEEE Sens J. https://doi.org/10.1109/
JSEN.2020.2973233
11. Sharma A, Mulaveesala R, Dua G, Kumar N (2020) Linear frequency modulated thermal wave
imaging for estimation of osteoporosis: an analytical approach. Electron Lett. https://doi.org/
10.1049/el.2020.0671
12. Arora V, Siddiqui JA, Mulaveesala R, Muniyappa A (2014) Pulse compression approach
to nonstationary infrared thermal wave imaging for nondestructive testing of carbon fiber
reinforced polymers. IEEE Sens J 15(2):663–664. https://doi.org/10.1109/JSEN.2014.2361391
13. Mulaveesala R, Dua G, Arora V, Siddiqui JA, Muniyappa A (2017) Pulse compression
favourable aperiodic infrared imaging approach for non-destructive testing and evaluation of
bio-materials. In: Thermosense: thermal infrared applications XXXIX, international society
for optics and photonics, vol 10214, p 102140G. https://doi.org/10.1117/12.2263924
14. Haji-sheikh A, Beck J (2002) Temperature solution in multi-dimensional multi-layer bodies.
Int J Heat Mass Transf 45:1865–1877. https://doi.org/10.1016/S0017-9310(01)00279-4
15. Carlslaw HS, Jaeger JC (1959) Conduction of heat in solids, 2nd edn. Oxford University Press,
London
Role of Collisional Interaction Between
Helium Molecules and Background Air
Molecules in the NDT Helium Leak
Testing Method in Vacuum
Abstract Response time and Sensitivity of leak detection are two important param-
eters that decide the efficiency and usefulness of a Leak detection Method in any
system. Mass Spectrometric Helium Leak Detection Method (MSLD) is a supreme
method since the response time is very low and the sensitivity is very high for leak
detection. Essentially, the rate and quantum of helium molecules that reach the Leak
Detector from the leakage location in the system decide these two parameters. In
a practical situation, these two parameters are invariably measured. Although it is
possible to approximately predict these two parameters for a particular leak detection
system, better theoretical and numerical attempts are always sought for the improve-
ment of this prediction. The finite element approach is a prospective method. In the
finite element approach, to evaluate these two parameters, one requires the knowl-
edge of the effective cross-section value for the collisions between helium molecules
and the background air molecules. In this paper, we present a method to evaluate the
effective cross-section from the basic collision data available in the literature. As a
proof of the usefulness of this effective cross-section value, we also present a way
to derive the response time and sensitivity of leak detection in a system.
1 Introduction
Vacuum and Pressure vessels are essential parts of any Petrochemical, Pharmaceu-
tical, Chemical, Aerospace and Nuclear plants. A vessel can never be entirely leak
tight, and occurrence of leak can lead to escape of confined medium to atmosphere
or ingress of ambient media into the vessel, depending on whether the vessel is
under pressure or vacuum, respectively. Leak testing methods play a crucial role in
detecting leaks [1, 2]. Helium leak testing method is widely used for detecting the
leak due to many reasons [3–7]. Primarily, due to its small atomic size, it passes
MSLD
(c)
Helium Leak
MSLD
(d)
Helium Leak
VALVE
VACUUM VESSEL HV PUMP LV PUMP
MSLD
the above two collisions. Wall reflections can be assumed to be purely reflective for
non-absorbing walls and isotropic for perfectly absorbing walls. The effective cross-
section between helium molecules and nitrogen/oxygen molecules in the volume
then becomes the most essential factor in the estimation.
290 G. J. Sharma and V. N. Ramani
Table 1 Comparison of response time and sensitivity of helium leak detection for the four different
configurations are shown in Fig. 1 V is the volume of the system
System Leak High Low Net Response Sensitivity
configuration detector vacuum vacuum pumping time
(Figure Ref.) (MSLD) (HV) (LV) speed of
Pumping Pumping Pumping system
speed speed speed
1a SL – – SL V / SL 1
1b SL SH SR SL + S H V / SL + SL / SL +
SH SH
1c SL SH SR SH V / SH SL / SL +
SR
1d SL SH SR SH V / SH 1
In this work, we present a method to evaluate the effective cross-section for helium
molecule interaction with background molecules and the usefulness of the evaluated
value to find out the response time and sensitivity.
Extensive works on the collisional interactions between charged and neutral particles
have been well summarized in the literature [11]. However, the available data do not
cover a very wide range of energy. In the following section, we evaluate the cross-
section values over the entire energy range of interest by extrapolation of available
data.
(a) 1.7E-19
1.6E-19
1.4E-19
1.3E-19
1.2E-19
1.1E-19
1.0E-19
4.0E+06 5.0E+06 6.0E+06 7.0E+06 8.0E+06 9.0E+06 1.0E+07
Rela ve velocity (m/s)
(b) 1.0E-17
1.0E+00 1.0E+01 1.0E+02 1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Total Cross-sec on (sq meter)
1.0E-18
1.0E-19
1.0E-20
1.0E-21
Rela ve velocity (m/s)
Fig. 2 a Typical Helium-Nitrogen elastic collision Cross-section data available in the literature—
values with cross marks are from Amdur et al. [12] and with star marks are from Newman et al.
[13]. b A plot of obtained Cross-section values by extrapolation
When the helium particles with a particular velocity Vi interact with a background gas
Tg whose velocity is characterized by a Three-Dimensional Maxwellian Distribution,
then the effective cross-section σeff can be written as
∞ ∞ ∞
σeff = (1/Vi ) < σ V > = (1/Vi ) f(V) σ (Vr ) Vr dV
−∞ −∞ −∞
292 G. J. Sharma and V. N. Ramani
where Vr is the relative velocity between the interacting particle and V is the velocity
space over which the integration is performed. Expressing the relative velocity in
terms of velocity components parallel and perpendicular to the incident beam, and
expressing the velocity in terms of components parallel and perpendicular to the
beam direction, the three-dimensional integration is reduced to the following one-
dimensional integral. This reduces the computation time and also reduces errors
introduced by numerical methods of integration.
< σV >
∞
= (Mg /2πKTg ) 1/2
e−(Mg /2πKTg )(Vi −V)2 −e−(Mg /2πKTg )(Vi + V)2 σ (V) V2 dV
0
The plot of σeff as a function of V is given in Fig. 3 for helium particles injected into
nitrogen, for various different injection velocities. The gas temperature considered
is 300 K. For a variation of temperature by 20 deg K, the estimated effective cross-
section was found to vary by <10%.
For estimation in air, similar evaluations in oxygen can be made and a proportional
addition of the estimations of nitrogen and oxygen must be done.
The obtained values of σeff as a function of V can be used to estimate the evolution
of the time profile of helium molecules at any location in the vacuum chamber and
thereby one can go on to derive the response time and sensitivity of the helium
leak detector. For example, we can consider the volume space of the chamber is
divided into J elements, using any finite element technique. When the helium gas
begins to enter the chamber through the leak, i.e. when the helium gas begins to get
sprayed or when a calibrated leak standard is exposed to the vacuum chamber, let
the time be considered as t = 0. At time t = 0, the helium concentration is present
only at the source element, which we denote by 1. At all other elements, the helium
concentration is zero. In the notation, the first suffix stands for the element number
and the second suffix stands for the time interval.
Role of Collisional Interaction Between Helium Molecules and Background … 293
2.0E-14
1.5E-14
1.0E-14
5.0E-15
0 5000 10000 15000 20000 25000 30000 35000 40000
Velcoity of Helium Injec on (m/s)
Fig. 3 Evaluated effective cross-section for nitrogen gas medium as a function of velocity of
injection of helium
At the end of first-time interval t, there will be an increase in the concentration
of helium at all elements, which will be the net distributed component of the helium
that would have remained in the vessel after a fraction of it getting pumped out of
the system. Hence, at the end of first-time interval t, the concentrations of helium
at all elements would be
and
This process of injection and pump down will continue till the steady state is
reached. In general, the concentration of helium at all elements, at the end of M time
intervals, can be written down as below.
Firstly, the wall collisions are totally neglected. Secondly, it is assumed that density
gradient-driven expansion of helium molecules is very fast compared to the pump
down time constant. In addition, the establishment of a density gradient from the
leak source point to detector location is not considered at all. All these aspects
are geometry dependent and have to be solved specifically for a given system and
situation. Hence, the above steps described are not claimed to be a complete solution
to the problem of finding response time or sensitivity and it is left to more involved
future works. Instead, it serves as proof for the claim that the evaluation of an effective
cross-section can successfully lead to the evaluation of response time and sensitivity
in a leak detection system.
4 Summary
The work presents a method to evaluate the effective cross-section of helium molecule
interaction with nitrogen molecule. It is also shown that the evaluated effective cross-
section can lead to the evaluation of sensitivity and response time in the helium leak
detection of a vacuum system.
References
1. Fitch RK (1989) Leaks and leak detection. In: Chambers et al (ed) Basic vacuum technology.
Adam Hilger, Bristol
2. Hucknall DJ (1991) Leak detection. In: Vacuum technology and application. Butterworth-
Heinemann
3. Weissler GL, Carlson RW (1979) Methods of experimental physics. In: Vacuum physics and
technology. Academic Press, vol 14
4. O’Hanlon JF (1980) A user’s guide to vacuum technology. John Wiley & Sons
5. Wutz M, Adam H, Walcher W (1989) Theory and practice of vacuum technology. Friedr.
Vieweg & Sohn, Braunschweig/Wiesbaden
6. Lawrence TL (1979) The fine art of leak detection and repair. In: Weissler GL, Carlson RW
(eds) Vacuum physics and technology. Academic Press
7. Dutta NG (2004) Bull Indian Vacuum Soc 7(1):5–19
8. Moore OP (2011) Non-destructive testing handbook. In: Leak testing (3rd edn), vol 1. ASNT,
USA
9. Rozanov LN (2002) Vacuum technique. In: Hablanian MH (ed). Taylor & Francis
10. Van Atta CM (1965) The nature and behavior of gases. In: Vacuum science and engineering.
McGraw-Hill Book Company, New York
11. Hasted JB (1964) Physics of atomic collisions. London Butterworths
12. Amdur I et al (1957) Scattering of high velocity neutral particles: He-N2, A-N2, N2–N2
interaction. J Chem Phys 57(2):527
13. Newman JH et al (1985) Differential scattering cross-section for collisions of 0.5, 1.5, and 5.0
Kev Helium atoms with He, H2, N2, And O2. J Geophys Res 90(A11):11045–11054
On the Application of Sniff Probe
Technique for Fugitive Emission
Detection Applications
Venkat N. Ramani
1 Introduction
Helium leak detection and methane leak detection are the two important leak detec-
tion techniques utilized for fugitive emission monitoring and control in the petroleum
industries as well as in chemical and nuclear industries. Both the leak detection
techniques are well developed and have good sensitivities for leak detection over a
wide range of leak rates. Typically, helium leak detection using a mass spectrometer
covers the range of 10−10 to 10−2 Pa m3 /s and methane leak detectors perform over
the range of 1–10,000 ppm with good calibration set ups and procedures in the labo-
ratories. Although helium leak detectors are, relatively, more sensitive and have a
faster response than the methane leak detectors, the methane leak detection is impor-
tant as it provides a more direct and typical corrosive and flammable environment,
particularly at high temperature. An important aspect of the leak detection technique
using a sniffer probe is the calibration of the sniffer probe with the leak detector,
at the site, to measure the leakage in PPM unit or leakage unit as required. This is
always a challenge.
The leakage specification is generally made in PPM or std cc/s units. The former
is a concentration unit and the latter is a leakage (or flow) rate unit. Generally, the
V. N. Ramani (B)
Plasma & Vacuum Technologies, Ahmedabad, Gujarat, India
e-mail: vnramani@plasvac.com
Qc (Qd-Qc)
Gas Air
Calibrated
Leak
Mix
Qd
Detector
(Ng/Na)= (Qc/Qd) (Concentration)
Qc<<Qd
Fig. 1 Schematic diagram for a leak detector measuring the leakage in concentration units. Qc is
the leak rate and Qd is the flow rate into the detector. The concentration of the leakage gas in an
ambient medium is given by Ng/Na
Qc
Detector
Gas (Leakage Rate)
Calibrated Leak
Fig. 2 Schematic diagram for a leak detector measuring the leakage in leak rate units. Qc is the
leak rate as well as flow rate into the detector
procedures and codes prescribe the leakage from fixed components (joints, valve
bodies, etc.) in concentration units and from dynamic components (valve gland
joints, pumps, etc.) in leakage rate units. The calibrated standard leaks are gener-
ally available for specified leak rates and hence it is straightforward to calibrate the
leakage measuring instrument in flow rate units at the site. It is not easy to do so
for concentration measuring detectors. To calibrate a leak detection in PPM units,
one requires a special set up. Additionally, one needs to take into account the fact
that the measured signal depends on the conductance of the tube attached to the
probe and detector. This work describes the construction of a simple standard leak
and the calibration set up required for serving the on-site calibration requirement,
over a wide range of leakage values. As an advantage, when an appropriate on-site
calibration exists, a single detector can be utilized to measure the leakage in either
PPM or Std cc/s. Figures 1 and 2 give the schematic diagrams for concentration and
leak rate measurements.
Figure 3 presents two geometries that can be easily adopted for the construction of
standard leaks. We constructed 2 standard leaks—one with a flat disc permeation
On the Application of Sniff Probe Technique for Fugitive Emission … 297
Fig. 3 Schematic view of the two constructed calibration leaks. using helium and methane leak
detectors
barrier and the other with a cylindrical shell permeation barrier. In both the cases, the
housing is made of stainless steel with standard KF25 end flanges. The length of the
housing is 75 mm. Inside the housing, at both inlet and outlet, two sintered porous
discs made of bronze, are introduced to avoid damage to the permeation barrier. All
joints to the permeation barrier with the housing are made leakproof with appropriate
mounting and further sealing with high vacuum epoxy. Both the permeation barriers
were constructed from clean flat layers of 25 micron thick PTFE material obtained
from standard commercially available rolls of 300 mm width. When the number of
layers varied, the leak rate was found to vary. We used 2 layers in both the geometries,
for our study.
The constructed standard leaks were checked against reference calibrated leaks, under
various pressure and temperature conditions, in a high vacuum set up fitted with
helium reference leaks calibrated in National Physical Laboratory, New Delhi, and
methane reference leaks calibrated at Vacuum Research Corporation, USA. Alcatel
make, ADIXEN ASM340 helium leak detector and Thermo Fisher Scientific Model
TVA2020 were used for helium and methane detection, respectively. The summary
of the results is shown in Fig. 4. The repeatability has been found to be very good.
A few trials conducted at various temperatures in the range from 15 to 50° Celsius
show that the temperature coefficient of variation of the leak rate is less than 1% per
degree Celsius.
298 V. N. Ramani
3.5E-04
3.0E-04
Leak Rate (std cc/s)
2.5E-04
2.0E-04
1.5E-04
1.0E-04
5.0E-05
0.0E+00
Flat Disc Flat Disc Cylindrical Shell Cylindrical Shell
(Helium) (Methane) (Helium) (Methane)
The schematic of a set up built for on-site calibrations of standard leaks is shown
in Fig. 5. The set up allows calibration over a wide range. The observed parameter
ranges in various chambers for obtaining leak rate in the range 1e-2 to 1e-6 std cc/s
are shown in Fig. 6.
5 Discussion
The calibrated leak described, although very simple, can be made very sophisti-
cated and improved in its performance by using more rugged and special materials,
like quartz or special polymers. The calibration set up described, incorporating the
described calibrated leak, in comparison to standard methods of mixed gas chambers
and capillaries, leads to significant advantages in simplicity, repeatability, and accu-
racy. Between the flat and cylindrical geometries, the latter showed better design
repeatability. This is attributed to better control and uniformity that are obtained
during the formation of layers in cylindrical geometry.
6 Summary
In this work, the construction of a simple calibrated leak for on-site calibration of
concentration measurement is described. A set up, which can be made easily portable,
that leads to a wide calibration range of helium and methane gases, is also described
and demonstrated.
On the Application of Sniff Probe Technique for Fugitive Emission … 299
Fig. 5 The schematic of the set up used for on-site calibration of leak detectors. 1-Methane Gas;
Cylinder; 2-Zero Air/Helium Cylinder; 3,4,5-Calibration Locations of Methane Leak Detector; 3
for Probe Calibration, 4 for EPA21 Calibration, 5 for Zeroing Leak Detector’ 6: Vacuum Pump; 7:
Reservoirs; 8-Pressure Gauges (0–280 barG); 9-Pressure Gauges (0–70 barG); 10 Pressure Gauges
(-1–2 barG); 11-Pressure Gauges (-1–2 barG); 12-Methane Calibrated Leak
300 V. N. Ramani
160
140
120
Pressure range (barG)
100
80
60
40
20
0
Chamber 1 Chamber 2 Chamber 3 Chamber 4 Chamber 5
Fig. 6 Ranges of operation pressures in the various chambers for calibration of methane detectors
in the range 1 to 10,000 ppm. Reference gas used was zero air
Investigation of CT Linearity
and Establishing the Density Calibration
Curve for the Gamma Ray CT System
Abstract Computed tomography (CT) is one of the measuring techniques for the
quantification of variation in material density. Variation in material density causes a
change in the attenuation coefficient of the reconstructed image. At RESD, IGCAR,
an indigenously developed single source and single HPGe detector based, first-
generation Gamma-ray computed tomography system has been used for nondestruc-
tive evaluation of materials and producing the scanned object’s 3D internal structure.
This system has the capability of scanning object of size up to 650 mm in diameter.
Using the gamma-ray CT system, we have studied defects in the high-density steel
drums, tanker wheels and concrete samples. Also, we have studied the variation in
the density profile from the reconstructed CT images. It is necessary to quantify the
performance of the system to authenticate the result. One of the parameters to eval-
uate the CT system is CT linearity. In a CT image, measured attenuation coefficients
for the material are represented in a dimensionless CT number. The relationship
between the material density and the CT number (in Hounsfield units) is represented
as CT linearity. CT number of material is exactly proportional to the density of the
material being imaged and this relation is used to characterize a particular system.
In this work, the CT linearity has been studied using a cylindrical test phantom
consisting of through-holes at constant pitch circle diameter. These through-holes
are filled with materials of different densities and diameters. Experiments have been
conducted to establish the CT linearity curve for different energies and to estimate
the variation in the measured density.
1 Introduction
the density calibration curve using the known material to be established. Then this
curve is used to find out the material density of the scanned object in the closed
container. In this paper, the correlation between CT numbers and density has been
established with the help of an in-house developed test phantom having materials of
various densities and is used for determining the unknown material density.
I = I0 exp(−μx) (1)
or
The linear attenuation coefficient μ mainly depends on the atomic number (Z) and
the incident energy E. Same material composition and having different density has
different linear attenuation coefficient. When considering the heterogeneous material,
contributions from each elemental of materials have been taken into account. So the
mass attenuation coefficient for compound material μm is calculated by taking the
weighted sum of its constituent elements.
μ
μm = = wi (μ/ρ)i (4)
ρ i
304 R. Ramar et al.
The (μ/ρ)i is the mass attenuation coefficient of the ith element, which is inde-
pendent of the density. Here, the wi is the Weight fraction of the ith element in the
composition. The mass attenuation coefficient for the individual elements for the
effective energy is determined from the NIST- X-com reference table.
In Computed Tomography, CT linearity [9] is the mathematical relationship
between the measured linear attenuation coefficient and the corresponding CT
number or Hounsfield Unit (HU), and it is calculated by the following equation
[10]:
μmaterial − μwater
HU = × 1000 (5)
μwater
μmaterial is the linear attenuation coefficient of the object and μwater is the linear
attenuation coefficient of water.
The linear relationship between the HU and the linear attenuation coefficient (µ)
is given by
HU = m · μ + c (6)
In the above linear equation, m and c represent the slope and the intercept respec-
tive. Since µ is directly proportional to the material’s density, the relationship between
the CT number and density of the material is also linear. In this study, the linear rela-
tion between the CT number and µ has been established and followed by the cali-
bration curve between CT number and the density of the material generated using
test phantom.
3 Test Phantom
In this study, we have used an in-house developed cylindrical test phantom with a
120 mm diameter and 30 mm height made of perspex (methyl methacrylate) material.
The phantom consists of 15 through-holes with varying diameters (1–15 mm in steps
of 1 mm), and each hole is placed at an equiangular interval of 22.5°. These holes
are filled with materials of different densities such as water (at 15 mm hole), brass
(12 mm), aluminium (11 mm), copper (9 mm), mild-steel (8 mm), stainless steel
(6 mm) and lead (4 mm). The remaining holes are kept empty to symbolize the air
voids. Figure 2 shows the schematic drawing of the in-house fabricated test phantom,
and Fig. 3 shows the photograph of the fabricated test phantom.
Investigation of CT Linearity and Establishing the Density … 305
4 Experimental Methods
In this work, CT linearities for different energies are studied using a 75 Se multi-
energy gamma-ray source with the activity of 10 Ci. The source is collimated by a
3 mm collimator. An HPGe detector with 100% relative efficiency with collimation
of 4 mm was used to receive the transmitted gamma-ray. Both the collimated source
and the detector were aligned in line and facing each other. A two-axis manipulator
system was placed in between source and detector to perform the CT movement.
The manipulator can move the loaded object in the direction perpendicular to the
collimated gamma-ray beamline and rotate the object about the Z-axis. The projection
profile of the particular angle is collected by moving the object perpendicular to the
beamline. The CT projection profile of the object was recorded for every 1° interval.
A 141 parallel gamma rays beam collected each projection profile by moving the
object in X-axis at 1 mm. There are 180 projection profiles collected to make a
sinogram. Figure 4 shows the collected Sinogram for the peak energy 260 keV.
Scanned Sinogram is linearly interpolated for 512 rays per projection and convo-
luted with Ram-Lak filter and back-projected using Convolution based Filter back
projection. The reconstructed image for 264 keV is shown in Fig. 5.
306 R. Ramar et al.
Fig. 5 Reconstructed CT
image of the test phantom at
264 keV energy
All the inserted material mean attenuation values are calculated by selecting a region
of interest (ROI) area corresponding to the material present in the reconstructed
image shown in Fig. 5. The mean pixel value of the selected ROI area gives the
mean attenuation value of the selected material. Similarly, the mean attenuation
value of water is calculated by selecting the ROI at 15 mm hole. The CT numbers or
Hounsfield Unit for every inserted material are calculated using the mean attenuation
value of the material and mean attenuation value of water using Eq. 5. Table 1 shows
the measured attenuation and calculated CT number for different materials measured
for primary energies (264, 280, 401 keV) of 75 Se.
Table 1 µ and CT number for the different materials for different energy
Material Measured attenuation value CT number
264 keV 280 keV 401 keV 264 keV 280 keV 401 keV
Brass 1.0478 0.9963 0.8052 7933 7511 5995
Aluminium 0.2683 0.2687 0.2313 1287 1308 1102
Copper 1.1072 1.0382 0.8432 8439 7868 6319
Mild steel 0.8598 0.8232 0.6594 6330 6035 4752
Stainless steel 0.88 0.8375 0.6615 6502 6157 4770
Lead 5.1584 4.3944 1.8179 42,976 36,480 14,628
Perspex 0.1345 0.1313 0.1141 147 136 103
Water 0.1173 0.1153 0.102 0 0 0
308 R. Ramar et al.
As per the Nyquist sampling theorem, since the scanning interval between rays
was 1 mm, object size below 2 mm cannot be captured with good contrast in the
CT image. Even though we could see the 1 mm diameter hole with low contrast, the
0.5 mm hole is not visible in the reconstructed CT image shown in Fig. 5. Artefacts
are seen in 4 mm hole having lead material and 12 mm hole having brass of the
reconstructed image shown Fig. 5, and it is caused due to the sudden density change.
Figures 6, 7 and 8 shows the curve for the calculated CT number against the linear
attenuation coefficient in the selected ROI area for 264, 280 and 401 keV. Further-
more, it is noted from Figs. 6, 7 and 8 that the linear relationship between attenuation
(µ) and CT number is linear, for which the slope and intercept are calculated is given
in Table 2.
With the help of the National Institute of Standard and Technology (NIST) web
database program called X-Com [11], the mass attenuation values for elemental,
compound and mixture alloys are calculated for different photopeak energies. Table
3 gives the obtained mass attenuation values for different energies from the X-com
database.
Density values of listed material in Table 1 are calculated using the measured
attenuation value µ in Table 1 and obtained mass attenuation coefficient µm in Table
3 using Eq. 4. Calculated density values are listed in Table 4 with the error percentage
for 264 keV energy. Figure 9 shows the calculated density values fitted against the CT
4
x 10 Calibration Curve CT Number(Vs)μ
4.5
Pb
4
264keV
3.5
3 HU=8525.5 × μ-1000
R2=0.999
CT Number →
2.5
1.5
1 Cu
SS
Brass
0.5 MS
Al
Perspex
0
0H2O 1 2 3 4 5 6
μ Cm-1 →
Fig. 6 CT number against the linear attenuation coefficient at 264 keV energy
Investigation of CT Linearity and Establishing the Density … 309
4
x 10 Calibration Curve CT Number(Vs)μ
Pb
3.5 280keV
2.5
CT Number →
1.5
1
Cu
SS Brass
0.5 MS
Al
Perspex
0H2O
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
-1
μ Cm →
Fig. 7 CT number against the linear attenuation coefficient at 280 keV energy
number for 264 keV. The fitted equation ρ = 9.51 × 10−4 × HU + 1.0763 g/cm3
gives the calibration curve for 264 keV. Similarly, for other peak energy of 75 Se,
the calibration curves fitted and are shown in Fig. 10. The fitted coefficient for the
density Calibration curve is given in Table 5, and it is obtained from Fig. 10. During
the fitting, the density of lead is avoided. The inclusion of density of lead during the
fitting of the linear equation affects the CT linearity. This is mainly due to statistical
error in the collected photon counts in the presence of lead in the scanning ray path.
In the presence of lead, the photon count reduces from 32,031 to 436. The statistical
error is more than 4%. Thus, in calculated attenuation, values of lead have more
error which is associated with counting statistics. Among the calibration curves for
different energy shown in Fig. 9, the variation between them is minimum at 1 g/cc.
The discrepancy becomes significant for higher density material and is due to the
increase of photoelectric attenuation for lower photon energy which leads to a higher
CT number. So, for the given density CT number increases with decreasing energy
[12].
The reconstructed CT image attenuation value provides the CT number for the
material. The fitted linear coefficient between CT number and density makes it easy
to determine the unknown material density if the CT number of material is known.
310 R. Ramar et al.
14000
401keV
12000 HU=8646 × μ-1000
R2=0.999
CT Number →
10000
8000
Cu
6000 Brass
SS
MS
4000
2000 Al
Perspex
0 H2O
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
-1
μ Cm →
Fig. 8 CT number against the linear attenuation coefficient at 401 keV energy
Density Vs CT Number
9
7 264keV
6
Density(ρ)→
0
-1000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000
CT number→
Fig. 9 CT linearity curve for 264 keV
312 R. Ramar et al.
6
260 keV
5
Density ρ
280 keV
4 401 keV
0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
CT Number
6 Conclusion
9.33%. The density values are fitted for different energies and the linearity coeffi-
cients have been arrived. With the help of the density calibration curve, the unknown
inclusion material’s density in the scanned object can be found using this CT system.
References
1. Mull RT (1984) Mass estimates by computed tomography: physical density from CT numbers.
Am J Roentgenol. https://doi.org/10.2214/ajr.143.5.1101
2. ASTM International (2019) Standard test method for calibrating and measuring CT density,
E1935–97. https://doi.org/10.1520/E1935-97R19
3. CT linearity, Radiol. Technol. Inf. Portal. (n.d.). https://www.radiology-tip.com/serv1.php?
type=db1&dbs=Linearity
4. Sharma D, Sharma S, Sanu K, Saju S, Deshpande D, Kannan S (2006) Performance evalua-
tion of a dedicated computed tomography scanner used for virtual simulation using in-house
fabricated CT phantoms. J Med Phys. https://doi.org/10.4103/0971-6203.25667
5. Yada N, Onishi H (2016) Validation of computed tomography-based attenuation correction of
deviation between theoretical and actual values in four computed tomography scanners, Asia
Ocean. J Nucl Med Biol. https://doi.org/10.7508/aojnmb.2016.02.004
6. Taina IA, Heck RJ, Elliot TR (2008) Application of X-ray computed tomography to soil science:
a literature review. Can J Soil Sci. https://doi.org/10.4141/cjss06027
7. Longuetaud F, Freyburger C, Mothe F, Constant T, Leban JM (2009) Measuring wood density
by means of X-ray computer tomography. Ann For Sci. https://doi.org/10.1051/forest/2009071
8. Du Plessis A, Meincken M, Seifert T (2013) Quantitative determination of density and mass
of polymeric materials using microfocus computed tomography. J Nondestruct Eval. https://
doi.org/10.1007/s10921-013-0195-7
9. Kriz RJ, Strauss KJ (1985) An investigation of computed tomography (CT) linearity. Med Imag
Instrum 85. https://doi.org/10.1117/12.949494
10. Bryant JA, Drage NA, Richmond S (2012) CT number definition. Radiat Phys Chem. https://
doi.org/10.1016/j.radphyschem.2011.12.026
11. Berger J, Hubbell MJ (1987) XCOM photon cross sections. https://physics.nist.gov/PhysRe
fData/Xcom/html/xcom1.html
12. Das IJ, Cheng CW, Cao M, Johnstone PAS (2016) Computed tomography imaging parameters
for inhomogeneity correction in radiation treatment planning. J Med Phys. https://doi.org/10.
4103/0971-6203.177277
Non-destructive Evaluation of Historic
Masonry Structures Using Infrared
Thermography and GPR
Abstract The Solani Aqueduct was built over the Solani River in Roorkee, Haridwar
District, Uttarakhand, to traverse the Upper Ganga Canal, which was built by the
Britishers for irrigation purposes. In the year 1854, the construction was completed.
Along with the Aqueduct, a road bridge or carriageway was constructed to cross the
river. This bridge carries a considerable amount of traffic; however, it is decrepit
in various places. Because of the constant storage of water in the Ganga canal
aqueduct, the structural characteristics inside the bridge are buried, and it is also
inaccessible from any other site. Over time, the structure has degraded, and visual
distresses can be seen. Therefore, it is imperative to assess the structural conditions to
understand this historic structure’s present load transfer mechanism. Non-destructive
testing is a feasible option for detecting these concealed structural characteristics.
Infrared thermal imaging and Ground Penetrating Radar (GPR) investigation have
been performed to assess the structure’s present condition. Infrared Thermography
(IRT) is used for the localization of defects, moisture seepage, and disintegration of
masonry at inaccessible locations. GPR inspection is carried out on the carriageway
to detect hidden structural features. The acquired radargrams are further processed to
remove noises in the image. The arch-like signatural system and traces of inspection
gallery is observed beneath the carriageway.
1 Introduction
2 Background
Tr e f l andTatm are the temperature of the ambient source and the object.
The amount of radiation emitted by a body increases with the increment in temper-
ature. Therefore, the thermal image generates a variation of temperature profiles
based on the object’s temperature. An uncooled long-wave (spectral range 7.5–
14 μm) thermal camera is used in the present study. The thermal wave-based inspec-
tions performed in this study can be categorized as (a) passive and (b) active tech-
niques. Passive Thermography has been used in the current study. The schematic
diagram of the passive thermal imaging technique is shown in Fig. 2. Where natural
thermal source, e.g., Sun, acts as an excitation heat source for the medium.
The use of thermal imaging technology can therefore play an important role
in many civil engineering applications [2]. The effect of moisture in buildings is
normally related to damage, which may occur due to the presence of moisture itself
or due to its evaporation. Researchers also use infrared thermography for the inves-
tigation of concrete structures, which includes investigating surface and subsurface
defects [3], delamination of surface layers [4], damage evaluation of steel-plate
concrete composite walls [5].
318 D. Ghosh et al.
Velocity of EM wave being denoted as c and εr & εr are the real and imaginary
components of the relative permittivity εr , respectively. As masonry has low electrical
conductivity, which yields low losses, the equation eventually can be rounded off as
c
υ= (4)
εr
Thus, Eq. 2 concludes that different media with diverse dielectric constants εr
have distinct wave velocities and wave scattering characteristics. In the operation
frequency range of radar waves, EM waves are affected by these two properties,
Non-destructive Evaluation of Historic Masonry Structures … 319
and it is difficult to distinguish their effects individually. The real component of the
relative permittivity equation is known as the dielectric constant, and the imaginary
part being energy losses due to attenuation. The dielectric constant gives the amount
of energy stored in the form of electric polarization. Table 1 shows the dielectric
constants and corresponding propagation velocities in the different materials.
The moisture content of the medium is correlated to the dielectric permittivity
(ε) and the electrical conductivity (σ), of the medium [6]. The velocity of propa-
gating waves is governed by the material’s dielectric constant, which is inversely
proportional to the square root of the dielectric constant. It’s worth noting that the
higher the medium’s dielectric constant, the sharper the hyperbolic profile’s reflec-
tions. Depending on the depth and required resolution of the item under inquiry,
several antenna frequencies are used. GPR antennas with a medium frequency (200–
600 MHz) are commonly used for imaging construction features because they can
generate a high-fidelity image of the targets [7]. If continuous depth information is
provided, 3D visualization of the photographed features may be conceivable. High-
frequency antennas (600 MHz–2 GHz) are often used in built heritage projects to map
the internal structure of a building [8] or to assess the state of preservation [9]. Rein-
forcing bars in the concrete medium are also detected using neural network-based
processing techniques. GPR is widely used for the feature identification of cultural
heritage sites [10–13]. In comparison to other NDT techniques, this methodology is
also particularly useful for assessing masonry structures [14]. We employed GPR as
a method to examine historic hydraulic structures in our investigation. The approach
has revealed structural characteristics that were previously buried beneath the water.
The investigations have been carried out using Thermal Imaging and GPR-based
inspection techniques. The results of the study are presented in the next section. As
illustrated in Fig. 3, the carriageway runs beside the Aqueduct and crosses the Solani
River. The carriageway’s location is shown by the red box.
320 D. Ghosh et al.
The Aqueduct is supported by masonry piers that may be seen from the river.
However, due to the ongoing storage of water in the Aqueduct, the supports beneath
the carriageway are not visible.
IRT of Solani aqueduct has been performed using Fluke Tix660 thermal imager. The
technical specification of the Camera is mentioned in Table 2. The images have been
captured from various distances ranging from 1 to 400 m range. Therefore, there
is a variation of resolutions in the images. However, the colour pattern for defect
Table 2 Technical
Spectral range of the Camera 7.5–14 μm
specifications of the Thermal
Camera IR resolution 640 × 480 pixel
Field of view (FOV) 30.9° × 23.1°
Spatial resolution (IFOV) 0.8 mrad
Thermal sensitivity (NETD) ≤0.03 °C @ 30 °C
Image frequency 60 Hz
Accuracy ± 1.5 °C or ± 1.5% of reading
Temperature range −40 °C to + 2000 °C
Non-destructive Evaluation of Historic Masonry Structures … 321
Fig. 4 a Solani aqueduct bridge side view b Thermogram of solani aqueduct bridge [15]
and intact locations are consistent. The images have been captured on a sunny day in
November in the evening (after 5 PM) time. Therefore, the structure received enough
sunlight exposure before carrying out the thermal imaging.
The visual and thermal image of the overall aqueduct is shown in Fig. 4a, b. The
defected areas of the aqueduct bridge are localized, as shown in the thermal image
(Fig. 4b). The dark stains in the visual image indicate the degradation of the structure.
The main cause of the degradation of the aqueduct bridge is, it is exposed to adverse
environmental conditions.
Some other locations of the aqueduct are also investigated. The entrance of the
north side base is also imaged with the thermal camera (Fig. 5).
Defected locations are marked as ellipses in the thermal image. Structural dete-
rioration and vegetation growth is observed. Deteriorations of masonry can also be
observed from one side of the Aqueduct (Fig. 6).
Fig. 7 Apertures
to as apertures. The carriageway road is 297 m long and 5.5 m wide. Two longitudinal
apertures are on either side of the road, with the third in the middle. Cross-sectional
scans are also done at an interval of 1 m across the six (6) cross-sectional apertures
(labelled a to f) in the middle of the aqueduct road, as shown in Fig. 7.
The investigation makes use of a 400 MHz antenna and a Georadar from IDS
Italy. Figure 8 depicts the components of the GPR apparatus. The installed wheel
encoder is used to collect data in distance mode.
K2 Fast Wave software is used to collect radargrams along the apertures. During
the data collection process, several structural traits are observed. Radargrams gath-
ered along the apertures are used to image subsurface structures beneath the
carriageway. Figure 8 shows the B-Scan of aperture 1. The scan data can be used
to estimate a depth of 3–4 m. Gain adjustment is used to equalize the power over
the course of a sweep. The B-Scan corresponding to linear gain is shown in Fig. 9a,
while the B-Scan corresponding to smoothed gain is shown in Fig. 9b. A better image
can be seen in B-Scan once smoothed gains have been applied. The presence of two
major layers in the pavement is evident.
Non-destructive Evaluation of Historic Masonry Structures … 323
Soil layer
Fig. 9 B-Scans of aperture 1 after (a) linear gain (b) smoothed gain
Arches
Arches
Fig. 10 B-Scans of aperture 2 after applying (a) linear gain (b) smoothed gain (c) time-domain
migration
the image. Pavement layers (two) are also found out to be at 0.2 m from the surface.
Cross-sectional scans (a–f) are carried out in the centre of the road, with B-scans for
each scan displayed in Fig. 11.
The arch can be seen up to 1/4 of the way across the span. However, in Fig. 11f,
the arch signature is missing, which could indicate degradation. At the centre of the
cross-section, there may be an inspection gallery. It can be seen on all cross-sectional
radargrams.
Non-destructive Evaluation of Historic Masonry Structures … 325
Inspection
Gallery
Arch
Gallery
3 Conclusions
Thermal imaging can be used for the inspection of degraded masonry, moisture
ingress in historic structures. The technique has successfully identified the dilapidated
portion of the structure. As a global NDE technique, thermal imaging also localized
the disintegrated locations of the hydraulic structure. This should be complemented
with other NDE and material characterization techniques to understand the degree
of deterioration and accordingly, suitable measures may be suggested.
To understand the structural features and structural system of the Solani Aqueduct,
GPR scanning on the carriageway has been performed. The radargrams acquired
along the various apertures are analysed to image subsurface structures beneath the
roadway. The GPR data reveals the following characteristics:
• A pattern of continuous arches may be seen beneath the road’s sub-layers.
According to the radargram, each arch spans roughly 12 feet. The base of the
road was constructed in layers. Each stratum has a depth of about 0.2 m.
326 D. Ghosh et al.
• When scanning the road from one side to the other, the arch part is only visible on
the canal side, while no similar pattern is evident on the other. The hollow arch
component extends up to around a quarter of the bridge’s span.
• At the road’s centre, traces of a continuous hollow cavity, most likely an inspection
gallery, can be seen.
The water level in the canal was dropped in the winter season owing to canal
maintenance, and some parts of the concealed hydraulic structure became visible. It
was possible to see the arches beneath the carriageway (Fig. 12). When measuring
the structure manually, the crowns of these arches are located at an interval of around
12fts, which was also detected from the radargram.
The research demonstrates a hybrid approach using Infrared thermography and
GPR-based method for locating subsurface characteristics of a historic hydraulic
construction. The techniques will be further advanced to create a 3D representation
of the various structures.
References
1. Water Highways The Aqueduct Wonders (2008) NBM&CW infra construction and equipment
magazine. www.nbmcw.com/article-report/infrastructure-construction/roads-and-pavements/
water-highwaysthe-aqueduct-wonders.html
2. Thusyanthan I, Blower T, Cleverly W (2017) Innovative uses of thermal imaging in civil
engineering. Proc Inst Civ Eng Civ Eng 170:81–87. https://doi.org/10.1680/jcien.16.00014
3. Hiasa S (2016) Investigation of infrared thermography for subsurface damage detection of
concrete structures. http://stars.library.ucf.edu/etd/5063
4. Wibowo H et al (2016) Delamination assessment of an ultra-high performance concretedeck
overlay using infrared imaging. In: First international interaction symposium UHPC, pp 1–8.
https://doi.org/10.21838/uhpc.2016.112
5. Matovu MJ, Farhidzadeh A, Salamone S (2016) Damage assessment of steel-plate concrete
composite walls by using infrared thermography: a preliminary study. J Civ Struct Heal Monit
6:303–313. https://doi.org/10.1007/s13349-016-0169-4
6. Pettinelli E, coauthors (2014) A controlled experiment to investigate the correlation between
early-time signal attributes of ground-coupled radar and soil dielectric properties. J Appl
Geophys 101:68–76
Non-destructive Evaluation of Historic Masonry Structures … 327
Abstract Masonry structures are vulnerable to damages due to their lower tensile
strengths. The near-surface zone of these structures got deteriorated due to moisture
ingress, non-favorable exposure conditions, etc. These structural damages become
a conduit for seepage of detrimental chemicals and aggravate the process of deteri-
oration. The detection and rehabilitation of these damages are essential before the
development of any damages. Thermal imaging has been used in this study as a
diagnostic tool for the identification of deteriorated locations in masonry structures.
Active and passive infrared imaging techniques have been used in this paper. While
performing the active technique, it is observed that the amplitude image shows better
detectability than the phase image. Thermal imaging of laboratory-based single leaf
masonry structures has been performed throughout a day (24 h) to obtain the best
suitable time for detecting a different defect in a masonry structure. A masonry wall
has been cast with several near-defects to simulate the different types of embedded
defects in a masonry wall. The finished masonry surface has been inspected using
a longwave thermal camera to detect several defects. The temperature profiles have
been drawn for a defect with the time (24 h.). The polarity changes of the defect
signature in the thermogram are also observed during the heating and cooling process
throughout the day. It is observed that near-surface defect detection using infrared
thermography is most suitable approximately at 5:00 P.M. and 7 A.M in the summer
season. For most types of defects in masonry.
1 Introduction
Major structures of importance like historical buildings and bridges are primarily
built with brick and stone masonries decades ago. Brick masonry is still one of the
most economical and reliable construction materials. However, these structures lack
the strength and design standards required to withstand external disturbances encoun-
tered during their lifespan. Common deterioration and decay of masonry structures
are exhibited in the form of moisture ingression, reinforcement corrosion, spalling,
micro-cracks, etc. [1, 2]. Routine in situ inspection of such defects in masonry struc-
tures is essential to adopt timely repair and restoration of the structures. Several
Nondestructive Testing (NDT) methods have been employed for the inspection of
masonry structures over the past few decades, especially for mortar-masonry bond
assessment [3–5]. However, these established NDT methods either failed to produce
the anticipated results or demanded exhaustive post-processing of the acquired data.
This problem is mainly attributed to the inhomogeneous composition of masonry
structures. The thermographic method that relies on the compositions’ thermal prop-
erties has shown its efficiency in assessing such inhomogeneous subsurface features.
Also, near-surface defects up to 10 cm can be efficiently imaged using the ther-
mographic method [6, 7]. This is especially important since some critical defects
like voids and honeycombing generally occur in the upper half of the reinforcement
concrete structures [7].
Infrared Thermography (IRT) based on non-contact imaging of surface has been
reported to overcome challenges associated with speed and size of investigation
[8]. IRT reproduces any near-surface flaws exhibiting temperature variation in the
recorded thermal image. Generally, two methods of thermography are practiced in
civil engineering applications—active and passive. Active heating based on an arti-
ficial heat source such as lamp, flash, and heat blanket is impractical for in situ
applications. This is because the heat source used has a tiny aperture and consumes
a lot of power, which is uneconomical for in situ applications. Passive heating is
economical, provides uniform heating, and heats large areas exposed to sunlight
over an extended period of a day. A challenge with implementing passive heating
is the daily and seasonal variation of solar radiation, limiting the thermal gradient
required to produce a clear thermal image. Unlike active thermography in a labora-
tory setup, the thermal images in passive thermography are affected by several factors
like wind, evaporation, and rain. However, studies have reported the applicability of
passive thermography even in the cold temperate climate of Europe to detect mois-
ture ingression in masonry arch [9]. Past research has reported achieving the best
result when passive thermography is adopted on a sunny day during the fall season
[10–12]. Also, frequent recording of the thermal image at a specified time interval for
24 h is desirable to find an optimum thermal gradient to account for the variation in
ambient conditions. These advantages of sunlight have been utilized to obtain appre-
ciable thermal images which can depict flaws in masonry and concrete structures.
IRT provides quantitative information about the crack depth based on the thermal
characteristics exhibited based on the heating and cooling rates of heated masonry
Identification of Defects in Masonry Structure Using … 331
structures [5]. The thermal imaging method has a major application in assessing
repair and restoration activities of cultural heritage buildings like surface cleaning,
stone consolidation, and mortar joints [13]. Ancient masonry buildings are generally
made from stone and brick layup with little or no consideration of seismic capacity.
Such vulnerability has been exploited by seismic events that propagate cracks and
structural failures. IRT can successfully track such propagation of defects over time
[14, 15]. The application of Passive IRT to the Indian cultural heritage situated in a
tropical climate has been studied by [16, 17] where most optimum and non-optimum
seasons have been identified. One of the most widespread applications of passive
thermography is the evaluation of large structures like bridges, building façade, and
road pavements [10, 11, 18]. Rigorous research has shown that IRT, coupled with the
active thermography technique, can detect the defect of lateral dimensions as small
as 3 × 3 cm [19]. This is far from true when the heating method with varying and
nearly low thermal intensities like sunlight is adopted. The efficacy of defect detec-
tion depends on the characteristics of flaws, especially the size of flaws [19]. Also,
the composition of flaw is essential since the thermal detectivity of the air-filled void
is much faster and clearer than air-filled void due to the difference in specific heat
[20]. These factors’ detrimental effects on the thermal image can be accounted for
when high intensity, uniform, and modulated heat source is implemented in active
thermography [21]. However, this is still a major challenge in the application of
passive thermography, where the lack of thermal gradient can give insufficient quan-
titative information such as imaging of indoor or confined portions of structures.
These problems have been dealt with a combined approach of adopting both passive
and active thermography to attain a sufficient amount of thermal contrast required
for quantification of flaws [5]. Nevertheless, studies have reported detecting delam-
ination in the concrete specimen at a minimum thermal gradient of 0.4 °C during
the day and 0.2 °C at night [12]. Therefore, IRT is a reliable technique for in situ
investigations of a defect, especially for imaging of large structures like concrete and
masonry buildings and bridges exposed to direct sunlight having anticipated the size
of the defect large enough to produce sufficient thermal gradient.
Regarding the other nondestructive evaluation techniques, ultrasonic wave-based
methods are also widely used to identify near-surface defects in concrete structures
[22–24]. However, these techniques are not suitable for brick masonry structures
due to their attenuation and scattering in the medium. Active and passive thermal
imaging techniques have been used in this study for the identification of defects in
single leaf masonry structures.
2 Background
Nondestructive evaluation. The MWIR and LWIR show the least absorption in air
particles, are suitable for camera sensors, and are the most ideal for IRT. The IR
thermal camera measures the IR radiation emitted from the surface of the structure
and converts the detected energy into temperature values.
The total radiation energy (E tot ) received by the camera is the sum of emissions
from the object (E obj ), the ambient source (Er e f l ), and the atmosphere (E atm ). This
can be expressed as
Tr e f l andTatm are the temperature of the ambient source and the object.
The amount of radiation emitted by a body increases with the increment in
temperature. Therefore, the thermal image generates a variation of temperature
profiles based on the object’s temperature. The temperature resolution of an image
is defined as the smallest difference in temperature in the field of view measured.
The camera temperature resolution depends on several experimental parameters like
object temperature, ambient environmental temperature, object to camera distance,
presence of filters, etc. The most commonly used parameters used as a measure of
temperature resolution are Noise Equivalent Temperature Difference (NETD). An
uncooled longwave (spectral range 7.5–14 µm) thermal camera is used in the present
Identification of Defects in Masonry Structure Using … 333
study. The IRT inspections can be of two types as (a) passive and (b) active tech-
niques. The schematic diagram of the active IRT imaging technique is shown in
Fig. 2a and passive IRT is demonstrated in Fig. 2b.
The external thermal excitation induced into the medium creates a thermal
gradient. In active IRT, artificial thermal sources, viz., lamps, blowers, or microwaves
act as thermal excitation. In contrast, a natural heat source like Sun acts as excitation
for the passive IRT technique.
The schematic diagram for the time–temperature curve of a pixel in the vicinity
of the defect is shown in Fig. 3. The medium without defect would have followed the
blue line. If there is any defect, change in the thermal material properties, deviation
in the time–temperature curve will be observed.
3 Experimental Investigations
The active and passive IRT experimental investigations have been performed to assess
near-surface defects’ detectability in single-leaf brick masonry walls. Fluke Tix 660
thermal cameras have been employed for this study.
The experimental investigation of masonry wall specimen for active thermal imaging
has been carried out as shown in Fig. 4 (left). After placement of different defects
in foam, thermocol, cardboard, and wood specimens, the samples are plastered with
mortar and the surface has been appropriately finished. The size of the specimen
is 600 mm × 600 mm × 100 mm. After the curing period, the wall specimen is
heated with 4 × 500 kW halogen lamps for 15 min, which are placed at a distance
of 500 mm from the specimen surface. After heating the sample, the thermal images
are captured in the cooling cycle (Fig. 4 right). The images have been captured for
2 min at an interval of 2 s. Similar trials have also been performed to ensure the
repeatability of the experiment.
Fast Fourier Transform (FFT) operation is carried out with the time–temperature
data for the temperature values corresponding to each pixel. The phase and amplitude
images have been prepared.
The specimen for passive thermal imaging is shown in Fig. 5. After placing the
Identification of Defects in Masonry Structure Using … 335
Fig. 5 Experimental specimen (a) before and (b) after placement of defect
several defects, the single-leaf masonry wall has been plastered and prepared for
inspection. Specimen size is 1.5 m × 1.5 m.
The sample has been monitored using the thermal imaging camera for 24 h to
identify the timeslot to perform passive IRT. The results are discussed in the next
section.
4 Results
The results of this study have been elaborated in this section. The IRT results of
active and passive techniques are discussed in Sects. 4.1 and 4.2, respectively.
After performing the FFT, the amplitude and phase images are plotted in Fig. 6.
MATLAB platform has been used for data processing.
Raw image shows three out of five defects can be identified. Four defects can
also be identified in the amplitude-based image, and the defect resolutions have also
been improved compared to the raw image. However, the edges become blurry in
this image. Edges are comparatively bright in the phase image, although the defects
are not visible. Therefore, amplitude images show better defect resolution compared
to phase and raw images.
336 H. Gupta et al.
Fig. 6 Thermal images: a) raw image b) amplitude image and c) phase image
The wall has been monitored for 24 h and thermograms at one-hour intervals have
been presented in Fig. 8. The experiment has been started at 10:19 am. The locations
of the thermal camera are kept in an approximately similar position. The emissivity
and humidity corrections are also incorporated in the images. The data acquisition is
carried out at a 1-h interval, and thermograms show a change in defect polarity with
the time. The polarity change is observed between 7 and 8 pm. The time–temperature
curve for a pixel within the defect, background temperature (object), and ambient
temperature is shown in Fig. 7.
The polarity changes of the defect signature in the thermogram are also observed
during the heating and cooling process. Here, the polarity means-change color inten-
sity (+/-) with respect to the mean or average temperature of the background. It
signifies, during daytime defects heated up faster as compared to the background and
similarly during nighttime, these also cooled down faster with respect to background
average sample temperature.
5 Conclusion
Fig. 7 The time–temperature curve of pixels located at defects and object and the ambient
temperature curve
References
1. Lourenço PB, Luso E, Almeida MG (2006) Defects and moisture problems in buildings from
historical city centres: a case study in Portugal. Build Environ 41:223–234. https://doi.org/10.
1016/j.buildenv.2005.01.001
2. Zampieri P, Zanini MA, Faleschini F (2016) Influence of damage on the seismic failure analysis
of masonry arches. Constr Build Mater 119:343–355. https://doi.org/10.1016/j.conbuildmat.
2016.05.024
3. Sadri A (2003) Application of impact-echo technique in diagnoses and repair of stone masonry
structures. In: NDT and E International. Elsevier, pp 195–202. https://doi.org/10.1016/S0963-
8695(02)00064-6
4. Diamanti N, Giannopoulos A, Forde MC (2008) Numerical modelling and experimental veri-
fication of GPR to investigate ring separation in brick masonry arch bridges. NDT and E Int
41:354–363. https://doi.org/10.1016/j.ndteint.2008.01.006
5. Tavukçuoğlu A, Akevren S, Grinzato E (2010) In situ examination of structural cracks at historic
masonry structures by quantitative infrared thermography and ultrasonic testing. J Mod Opt
57:1779–1789. https://doi.org/10.1080/09500340.2010.484553
6. Krause M, Müller W, Wiggenhauser H (1997) Ultrasonic inspection of tendon ducts in concrete
slabs using 3D-SAFT. Springer, Boston, MA, pp 433–439. https://doi.org/10.1007/978-1-4419-
8588-0_69
7. Maierhofer C, Röllig M, Schlichting J (2010) Active thermography for evaluation of rein-
forced concrete structures. In: Non-destructive evaluation of reinforced concrete structures:
non-destructive testing methods. Elsevier Inc., pp 370–402. https://doi.org/10.1533/978184
5699604.2.370
8. McCann DM, Forde MC (2001) Review of NDT methods in the assessment of concrete and
masonry structures. NDT and E Int 34:71–84. https://doi.org/10.1016/S0963-8695(00)00032-3
9. Clark MR, McCann DM, Forde MC (2003) Application of infrared thermography to the non-
destructive testing of concrete and masonry bridges. In: NDT and E International. Elsevier, pp
265–275. https://doi.org/10.1016/S0963-8695(02)00060-9
10. Hing CLC, Halabe UB (2010) Nondestructive testing of GFRP bridge decks using ground
penetrating radar and infrared thermography. J Bridg Eng 15:391–398. https://doi.org/10.1061/
(ASCE)BE.1943-5592.0000066
11. Halabe UB, Vasudevan A, Klinkhachorn P, Gangarao HVS (2007) Detection of subsurface
defects in fiber reinforced polymer composite bridge decks using digital infrared thermography.
Nondestr Testing Eval 22:155–175. https://doi.org/10.1080/10589750701448381
12. Rocha JHA, Póvoas YV, Santos CF (2018) Detection of delaminations in sunlight-unexposed
concrete elements of bridges using infrared thermography. J Nondestr Eval 38:1–12. https://
doi.org/10.1007/s10921-018-0546-5
13. Avdelidis NP, Moropoulou A (2004) Applications of infrared thermography for the inves-
tigation of historic structures. J Cult Herit 5:119–127. https://doi.org/10.1016/j.culher.2003.
07.002
14. Bisegna F, Ambrosini D, Paoletti D, Sfarra S, Gugliermetti F (2014) A qualitative method
for combining thermal imprints to emerging weak points of ancient wall structures by passive
infrared thermography-a case study. J Cult Herit 15:199–202. https://doi.org/10.1016/j.culher.
2013.03.006
15. Paoletti D, Ambrosini D, Sfarra S, Bisegna F (2013) Preventive thermographic diagnosis of
historical buildings for consolidation. J Cult Herit 14:116–121. https://doi.org/10.1016/j.cul
her.2012.05.005
16. Kavuru M, Rosina E (2021) IR Thermography for the restoration of colonial architecture in
india—case study of the british residency in Hyderabad, Telangana. J Cult Herit 48:24–28
17. Kavuru M, Rosina E (2020) Developing guidelines for the use of passive thermography on
cultural heritage in tropical climates. Appl Sci 10(23):8411
340 H. Gupta et al.
18. Sfarra S, Cicone A, Yousefi B, Ibarra-Castanedo C, Perilli S, Maldague X (2019) Improving the
detection of thermal bridges in buildings via on-site infrared thermography: the potentialities of
innovative mathematical tools. Energy Build 182:159–171. https://doi.org/10.1016/j.enbuild.
2018.10.017
19. Tran QH, Han D, Kang C, Haldar A, Huh J (2017) Effects of ambient temperature and relative
humidity on subsurface defect detection in concrete structures by active thermal imaging.
Sensors 17:1718. https://doi.org/10.3390/s17081718
20. Huh J, Tran QH, Lee JH, Han DY, Ahn JH, Yim S (2016) Experimental study on detection of
deterioration in concrete using infrared thermography technique. Adv Mater Sci Eng
21. Brown J, Hamilton H (2007) Heating methods and detection limits for infrared thermography
inspection of fiber-reinforced polymer composites. ACI Mater J 104:480–490
22. Ghosh D, Beniwal S, Ganguli A, Mukherjee A (2018) Reference free imaging of subsurface
cracks in concrete using R ayleigh waves. Struct Control Health Monitor 25(10):e2246
23. Ghosh D, Beniwal S, Ganguli A (2015) Detection of defect in concrete slab using Rayleigh
waves. In: IEEE international ultrasonics symposium (IUS). IEEE, pp 1–4
24. Ghosh D, Kumar R, Ganguli A, Mukherjee A (2020) Nondestructive evaluation of rebar
corrosion-induced damage in concrete through ultrasonic imaging. J Mater Civ Eng
32(10):04020294
25. Halabe UB, Dutta SS, GangaRao HVS (2010) Infrared thermographic and radar testing of
polymer-wrapped composites. Mater Eval 68:447–451
Application of Frequency Modulated
Thermal Wave Imaging for Bone
Diagnostics
1 Introduction
Fig. 1 Representation of frequency modulated thermal heat flux used for numerical studies for
osteoporotic bone model
The present numerical work highlights the ability of frequency modulated thermal
wave imaging for the detection of density variation in modeled bone samples as
shown in Fig. 2. The modeled bone skin surface has been exposed to frequency
modulated thermal heat flux, and temporal temperature variations are numerically
obtained as shown in Fig. 3. Further, in order to consequently process the obtained
numerical outcome with the pulse compression-based data processing approach, the
344 A. Sharma et al.
Table 1 Description of thermo-physical properties considered for modeled bone [9, 10, 13–15]
Materials Thermal Density Specific Blood Metabolic Thermal
conductivity (ρ) heat perfusion heat effusivity
(k) (Kg/m3 ) (cp ) rate generation
e = kρc p
(W/mK) (J/KgK) ml
ml s
(W/m3 )
(Ws1/2 /m2 K)
SKIN 0.37 1109 3391 0.002 368.1 1179.589
FAT 0.21 911 2348 0.0002 400 670.2208
MUSCLE 0.49 1090 3421 0.0009 700 1351.723
BONE-H 0.616 2420 1430 0 0 1460.044
BONE-1 0.25 1480 1200 0 0 666.3333
BONE-2 0.34 1200 2000 0 0 903.3272
BONE-3 0.504 1980 1170 0 0 1080.54
BONE-4 0.532 2090 1235 0 0 1171.824
BONE-5 0.56 2200 1300 0 0 1265.543
BONE-6 0.588 2310 1365 0 0 1361.636
BLOOD - 1055 3660 - - -
Fig. 2 Schematic of osteoporotic bone model used for numerical studies of bone density variations
(details are shown in Table 1)
Application of Frequency Modulated Thermal Wave … 345
Fig. 3 Temporal thermal response obtained over skin surface of modeled osteoporotic bone sample
(bone details are shown in Table 1)
mean rise in the acquired data needs to be removed by an appropriate data fitting
methodology to zero mean value. Considering this, a polynomial fitting approach
has been followed up to acquire zero mean fitting data.
Now, obtained mean-removed data has been processed with the coefficient of
correlation-based data processing approach and variation of correlation coefficient
with the density variation has been studied. In Fig. 4, it has been observed that a
healthy bone (depicting higher density) has a lower coefficient of correlation. This
is due to higher effusivity values associated with the dense bone. As bone density
decreased, the effusivity also decreased and the surface temperature increases over
the modeled bone sample which leads to lower values of correlation coefficient when
correlating with a healthy bone sample.
Consequently, it can be observed that a decrease in the bone density parameter is
in an indirect proportion with the correlation coefficient as observed with numerical
analysis. These can be readily observed from Fig. 5, in which the variation in density
is plotted with respect to the coefficient of correlation of modeled bone samples.
Thus, ruling out the bone sample 2 data value, all the modeled bone samples are
seen to follow a straight-line trend, where an increase in the correlation coefficient
values the bone density subsequently decreases. Additionally, it has been known that
density parameter has a direct relation with effusivity of bone. Therefore, it can be
conclusively stated from Fig. 6 that with the rising value of the correlation coefficient,
346 A. Sharma et al.
Fig. 4 Correlation coefficient response for different density variations (shown in Table 1) of
osteoporotic bone model
there is a linear decrease in the effusivity values of the modeled bone sample.
4 Conclusion
This work proposes a numerical study for “Frequency Modulated Thermal Wave
Imaging (FMTWI)” which could be an efficient active infrared thermographic-based
detection technique of osteoporosis. Different stages and severity of the osteoporotic
disorder have been analyzed by employing the coefficient of correlation approach as
a figure of merit. The current work conclusively states that a considerable variation
in the thermo-physical properties, i.e., density and thermal effusivity, of a modeled
bone surface has a strong relation to the coefficient of correlation. Hence, correlation
coefficient-based data processing approach is of major relevance for diagnosing the
osteoporotic disorder.
Application of Frequency Modulated Thermal Wave … 347
Fig. 5 Density variation with correlation coefficient for modeled osteoporotic bone sample
Fig. 6 Thermal effusivity variation with correlation coefficient for modeled osteoporotic bone
sample
348 A. Sharma et al.
References
1. Yang TL, Shen H, Liu A, Dong SS, Zhang L, Deng FY, Zhao Q, Deng HW (2019) A road map
for understanding molecular and genetic determinants of osteoporosis. Nat Rev Endocrinol
16(2):1–13. https://doi.org/10.1038/s41574-019-0282-7
2. Lorentzon M (2019) Treating osteoporosis to prevent fractures: current concepts and future
developments. J Intern Med 285(4):381–394. https://doi.org/10.1111/joim.12873
3. Adams AL, Fischer H, Kopperdahl DL, Lee DC, Black DM, Bouxsein ML, Fatemi S, Khosla
S, Orwoll ES, Siris ES, Keaveny TM (2018) Osteoporosis and hip fracture risk from routine
computed tomography scans: the fracture, osteoporosis, and CT utilization study (FOCUS). J
Bone Miner Res 33(7):1291–1301. https://doi.org/10.1002/jbmr.3423
4. Clynes MA, Harvey NC, Curtis EM, Fuggle NR, Dennison EM, Cooper C (2020) The
epidemiology of osteoporosis. Br Med Bull 133(1):105–117. https://doi.org/10.1093/bmb/lda
a005
5. Anouti FAL, Taha Z, Shamim S, Khalaf K, Kaabi AL, Alsafar H (2019) An insight into the
paradigms of osteoporosis: from genetics to biomechanics. Bone Reports 11:100216. https://
doi.org/10.1016/j.bonr.2019.100216
6. Pinheiro MB, Oliveira J, Bauman A, Fairhall N, Kwok W, Sherrington C (2020) Evidence on
physical activity and osteoporosis prevention for people aged 65+ years: a systematic review
to inform the WHO guidelines on physical activity and sedentary behaviour. Int J Behav Nutr
Phys Act 17(1):1–53. https://doi.org/10.1186/s12966-020-01040-4
7. Harvey N, Dennison E, Cooper C (2010) Osteoporosis: impact on health and economics. Nat
Rev Rheumatol 6(2):99–105. https://doi.org/10.1038/nrrheum.2009.260
8. Wahl DA, Cooper C, Ebeling PR, Eggersdorfer M, Hilger J, Hoffmann K, Josse R, Kanis JA,
Mithal A, Pierroz DD, Stenmark J, Stöcklin E, Dawson-Hughes B (2012) A global represen-
tation of vitamin D status in healthy populations. Arch Osteoporos 7(1):155–172. https://doi.
org/10.1007/s11657-012-0093-0
9. Sharma A, Mulaveesala R, Arora V (2020) Novel analytical approach for estimation of thermal
diffusivity and effusivity for detection of osteoporosis. IEEE Sens J. https://doi.org/10.1109/
JSEN.2020.2973233
10. Sharma A, Mulaveesala R, Dua G, Kumar N (2020) Linear frequency modulated thermal wave
imaging for estimation of osteoporosis: an analytical approach. Electron Lett. https://doi.org/
10.1049/el.2020.0671
11. Arora V, Siddiqui JA, Mulaveesala R, Muniyappa A (2014) Pulse compression approach
to nonstationary infrared thermal wave imaging for nondestructive testing of carbon fiber
reinforced polymers. IEEE Sens J 15(2):663–664. https://doi.org/10.1109/JSEN.2014.2361391
12. Mulaveesala R, Dua G, Arora V, Siddiqui JA, Muniyappa A (2017) Pulse compression
favourable aperiodic infrared imaging approach for non-destructive testing and evaluation of
bio-materials. Thermosense: thermal infrared applications XXXIX. Int Soc Optics Photon
10214:102140G. https://doi.org/10.1117/12.2263924
13. Mulaveesala R, Dua G (2016) Non-invasive and non-ionizing depth resolved infra-red imaging
for detection and evaluation of breast cancer: a numerical study. Biomed Phys Eng Express
2(5):055004. https://doi.org/10.1088/2057-1976/2/5/055004
14. Werner J, Buse M (1988) Temperature profiles with respect to inhomogeneity and geometry
of the human body. J Appl Physiol 65(3):1110–1118. https://doi.org/10.1152/jappl.1988.65.3.
1110
15. Williams LR, Leggett RW (1989) Reference values for resting blood flow to organs of man.
Clin Phys Physiol Meas 10(3):187–217. https://doi.org/10.1088/0143-0815/10/3/001
Calculation of X-ray Attenuation
Coefficient for Different Layers in a Case
Bonded Composite Solid Rocket Motor
Abstract A Case bonded Solid Rocket Motor (SRM) can be imagined as a multi-
layer cylinder in which the outermost layer is the casing which can be metallic or
made from composite materials. The casing is then lined with an insulation layer; the
insulation is generally of NBR or EPDM-based rubber. A thin layer of liner material
is then coated on the insulation surface which facilitates bonding with the composite
solid propellant. The inspection of the SRM mainly focuses on the grain inspection
for the presence of flaws such as voids, porosity, foreign intrusion and cracks and
interface inspection to ascertain the soundness of the bond between different layers.
The high-energy X-ray Radiographic Testing (RT) method is conventionally followed
for the inspection of large-sized SRM by Biswas et al (J Radiat Res Appl Sci 9:26–
34, 2016 [1]). The densities of the different layers of the SRM are so close to each
other that obtaining a good contrast image is not easy compared to other engineering
materials. Interpretation of such radiographic films is a challenging task considering
the contrast each material is producing owing to very close density values. The density
values of the composite casing, insulation rubber, liner material and propellant all
lie in the range of 1–1.7 g/cc with a density difference of 0.3–0.7 g/cc. In the present
work, the X-ray attenuation coefficient for different layers of a case bonded SRM is
calculated for different energies from the optical density values measured from the
radiographic films. The values are then compared and plotted to obtain radiographic
contrast. As the attenuation coefficient is a function of energy and nature of the
material, the calculated contrast values will divulge intricate details of the interfaces
at different layers.
1 Introduction
1.1 Brief
The mass attenuation coefficient values are generally calculated for developing
shielding materials and to comprehend the shielding effectiveness of these mate-
rials [1]. Apart from designing shielding materials, attenuation coefficients have
various other applications in X-ray science [2]. The attenuation coefficient values
for unknown materials when exposed to an X-ray source of known energy help in
distinguishing materials [3].
As the different layers of the SRM have closely lying density values, even a
slight difference in the attenuation coefficient (μ) of these materials when exposed
to different energies will generate varying contrast images. The contrast of the image
depends upon various factors such as source geometry, scattered photons and the
characteristics of the detector [4]. The radiographic measurement of the attenuation
coefficient depends upon the energy of the photon, atomic number and density of the
material under inspection. So, the selection of energy for inspection is decided upon
the attenuation of photons in the material. For ideal conditions, the optimum energy
is selected such that μl = 2, where l is the mean free path of the photon inside the
material. A range of energies will give a transmission of 10–30%; however, the most
suitable energy for a given combination of materials will yield a significantly high
contrast image due to differences in ‘μ’ values.
The NDE inspection of the SRM mainly focuses on the grain inspection for the
presence of flaws such as voids, porosity, foreign intrusion and cracks and interface
inspection to ascertain the soundness of the bond between different layers [5, 6].
The high-energy X-ray Radiographic Testing (RT) method is conventionally
followed for the inspection of large-sized SRM [7]. As the densities of the different
layers of the SRM are so close to each other, obtaining a good contrast image is not
easy as compared to other engineering materials. Interpretation of such radiographic
films is a challenging task considering the nearly same contrast each material is
producing owing to very close density values. The density values of the composite
casing, insulation rubber, liner material and propellant all lie within the range of
1–1.7 g/cc with a density difference of 0.3–0.7 g/cc.
2 Experimental Setup
The experimental setup consists of the SRM placed on a suitable object manipulator
which can provide translational and rotational motion to the object and high energy
Linear Accelerator (Linac) with energies (4/7/9 MeV) as a source of radiation (Fig.1).
Industrial radiographic film loaded in a suitable cassette flanked by lead screens on
both sides is used as a detector which records the latent image from the differential
Calculation of X-ray Attenuation Coefficient for Different … 351
Table 1 Details
Equipment Parameters Details
of Equipment and Parameters
used in the experiment Inspection Radiographic testing X-ray photons
technique
Machine used High-energy linac 4, 7, 9 MeV
Detector Industrial CEN class C2 and C3
radiographic film
Developer Automatic film Time: 3 min; Temp:
processor 35 °C
Densitometer Model: LH SP D –
absorption of the transmitted beam. The film is then developed in a darkroom using
an automatic film processor.
The optical density values are calculated using the densitometer (Table 1).
3 Procedure
Beer-Lambert’s law for X-ray attenuation can be used to calculate the attenuation
coefficient for an unknown material. The attenuation coefficient values for some
materials are available in NIST-126 data.
To validate the method for the determination of attenuation coefficient, radio-
graphic images of some known materials such as graphite, Perspex and steel were
taken. After plotting the ln (Io /I) versus thickness, the attenuation coefficient was
calculated using Beer-Lambert’s law for attenuation.
I = I oe−μx (1)
where,
352 K. Lubna et al.
This equation is valid for mono-energetic and narrow beam conditions for
homogenous material. However, the same was used in the present case also.
In the present work, the attenuation coefficient is calculated using the linear fit
method of optical density versus thickness. Tangential exposure was taken to obtain
interface information. SRM was kept in the horizontal position and the X-ray beam
was focused at tangent to the insulation-propellant interface at a suitable film to focus
distance (FFD). The object was exposed using three different energies (4, 7, 9 MeV)
for the same equivalent thickness.
The cassette was kept in contact with the object and the dose was calculated
considering the equivalent thickness coming in the path of X-rays.
The exposed film was then processed and viewed on a film viewer with an in-
built densitometer. Optical density was measured at different locations on the film.
The optical density was then plotted against the thickness under inspection and the
attenuation coefficient was calculated.
4 Observations
In order to compare the calculated attenuation coefficient values with the values
reported in the literature, some known materials such as Perspex and graphite were
used initially. These materials with variable thickness were exposed to 4 MeV X-
rays, and the image was recorded on the industrial radiographic film. The optical
densities were then measured (Fig. 2).
The calculated and reported values are listed below in Table 2.
where μm is μ/ρ, and ρ is the density of the test material.
As seen from Table 2, the values obtained from this method are in close agreement
with the values reported in the literature.
So, with the above validation, we calculated the absorption coefficient for different
layers in an SRM. The outermost layer is the composite casing, then insulation and
the inside layer is propellant.
The linear attenuation coefficient for each layer was calculated at three different
energies, namely 4, 7 and 9 MeV. The plot of optical density versus thickness for
different energies is as shown.
The attenuation coefficient was then calculated for composite casing, insulation
and propellant for 4, 7 and 9 MeV energies. The values are listed in the table below
(Table 3).
Calculation of X-ray Attenuation Coefficient for Different … 353
(a) Plot of optical density versus thickness for Graphite(L) and Perspex (R)
Composite Casing
Fig. 2 Plot of optical density versus thickness for Graphite (L) and Perspex (R)
Table 2 Summary of
Material Density μm μm
calculated and reported values
(g/cc) (Calculated, cm2 /g) (Reported*, cm2 /g)
Graphite 2.26 0.0345 0.03047
Perspex 1.18 0.0932 0.0949
354 K. Lubna et al.
Table 3 Attenuation
Materials Attenuation coefficient μ (mm−1 )
Coefficients for various
materials at different energies 4 MeV 7 MeV 9 MeV
Composite casing 0.00891 0.007 0.00337
Insulation 0.0043 0.00328 0.00116
Propellant 0.00515 0.00488 0.00245
(a) The attenuation coefficient values calculated for graphite and Perspex by this
method when compared with the values reported in National Institute of Stan-
dards and Technology 126 [8] are found close. This verifies the validity of the
proposed method for measuring the linear attenuation coefficient for materials.
(b) The X-ray beam from Linac is not mono-energetic. Due to the broad beam
nature, the first 2–3 HVL values are lower than the subsequent levels. After
the initial 2–3 HVLs, the less penetrating components of the beam are filtered
out. In the energy range where Compton scattering is predominant, the softer
component of the beam is filtered out but they are replaced by multiple Compton
scatter produced by absorption of shorter wavelength components. Thus, an
approximate equilibrium value of attenuation coefficient is reached which is
almost equal to the attenuation coefficient value of a mono-energetic beam [4].
(c) The HVL values in terms of propellant equivalent thickness are 116 mm,
145 mm and 149 mm for 4, 7 and 9 MeV energies used in the present study.
The attenuation coefficient is calculated at nearly 2-3 HVL thicknesses. The
marked difference in the linear attenuation coefficient values of 9 MeV from
4 and 7 MeV may be due to the reason that the equilibrium HVL value for 9
MeV was not reached for the given thickness measurement.
(d) The radiation contrast depends upon the material under inspection (density,
composition), radiation energy, thickness variations and scatter radiation [9].
As the radiation energy increases, the inherent un-sharpness due to scattering
also becomes prominent. The selection of a lower energy source may give
less scattering for a homogenous material. During the inspection of a combi-
nation of materials or composite layers, the varying attenuation within mate-
rials of different densities with varying thicknesses will yield high contrast for
relatively higher energy.
(e) The absorption of the X-ray photon within the material depends primarily on
the thickness and mass of the absorber and the energy of the photon [9]. The
spectrum of photon energies for Bremsstrahlung X-ray beam as in Linac is
wide. In such a heterogeneous beam, the probability of absorption of indi-
vidual photons depends on their energy. Low-energy photons are much more
likely to be absorbed than high-energy photons. Consequently, the superficial
layers of the SRM, i.e. the casing tends to remove the low-energy photons
and transmit the higher energy photons. Therefore, as an X-ray beam passes
Calculation of X-ray Attenuation Coefficient for Different … 355
Fig. 3 Variation of
attenuation coefficient with
different energies
through different layers of the SRM of varying thicknesses, the intensity of the
beam decreases but the mean energy of the resultant beam increases.
(f) In the case of the casing-insulation interface, 4 meV Linac beam gives suffi-
ciently high differential absorption for casing and insulation material thick-
nesses for generating a good contrast radiographic image whereas, for the
insulation-propellant interface, the 7 MeV Linac beam generates a good
contrast (Fig. 3).
6 Conclusion
(a) The attenuation coefficient decreases with increasing photon energy as seen
from the behaviour of attenuation coefficient with the energy of incident X-rays
plot below.
(b) The difference in absorption coefficient values for the range of closely lying
density materials under inspection is not more in 9 MeV energy. Low contrast
will be obtained for the images exposed with 9 MeV source at the interface
and intricate details may be missed.
(c) The contrast for the Propellant-Insulation interface will be more when 7 MeV
energy is used. However, for inspection of the casing, a 4 meV source will
be more useful as the difference in the absorption coefficient of casing and
insulation material is more.
(d) The calculation of attenuation coefficient values can be used to obtain infor-
mation about the contrast of the image to extract intricate details of closely
lying density materials apart from the selection of shielding materials.
(e) Suitable energy can also be selected for inspection based upon the attenuation
coefficient values.
356 K. Lubna et al.
7 Limitations
References
1. Biswas R, Sahadath H, Mollah AS, Huq Md. F (2016) Calculation of gamma-ray attenuation
parameters for locally developed shielding material: Polyboron. J Radiat Res Appl Sci 9:26–34
2. Mousa A, Kusminarto K, Suparta GB (2017) A new simple method to measure attenuation
coefficient of materials using micro-digital radiography machine. Int J Appl Eng Res 12(21):
10589–10594. ISSN 0973–4562
3. Sutton GP, Biblarz O (2011) Rocket propulsion elements, 8th edn. John Wiley & Sons
4. Halmshaw R (1995) Industrial radiology: theory and practice, 2nd edn. Chapman and Hall
5. Ramakanthan S, Moindenkully KK, Gunshekaran R, Thoamas C, Thomas CR (2015) Analysis
of defects in solid rocket motors using X-ray radiography. J Non Destruct Testing 20(6). www.
ndt.net
6. Ghose B (2016) Evaluation of probability of detection and minimum number of exposures
required for detection of planar flaws in cylindrical object by radiographic nde method. www.
ndt.net
7. Shull PJ (2002) Non-destructive evaluation: theory, techniques, and applications. Marcel Dekker
8. Hubbell JH, Seltzer SM (2010) NIST standard reference database 126
9. Serman N (1998) Production of X-ray and interaction of X-rays with Matter, pp 11–20. www.
columbia.edu.
A Data Fusion Technique Using
Acceleration and Strain Measurements
for Enhanced Damage Localization
in Civil Structures
K. Lakshmi
K. Lakshmi (B)
CSIR-Structural Engineering Research Centre, Taramani, Chennai 638113, Tamilnadu, India
e-mail: lakshmik@serc.res.in
1 Introduction
The structures built are susceptible to degradation in the course of their service life,
starting from their installation. The external factors like erosion, corrosion, wear,
fatigue and heavy loadings induce degradation and damage, increasing the risk of
becoming unfit for further use. The scale of degradation and damage needs to be
assessed in order to prevent the catastrophic failure of the structures, creating loss
of life and assets. Fortunately, the developments in the field of Structural Health
Monitoring (SHM) in the last two decades have provided hope and solutions to
tackle the above issues.
Structural Health Monitoring (SHM) is a process that measures and analyzes the
data from aerospace, civil, and mechanical structures to interpret the condition, on
a near real-time basis. When the changes in the form of material and/or geometric
properties are observed, the condition is described as damage. Damage is capable of
profoundly affecting the performance of the structures.
The damage can be identified by extracting and analyzing the damage-sensitive
features, from the measurements, acquired using sensors of various types. SHM
comprises three basic steps, namely sensing, signal processing and diagnosis.
Famously, statistical time series models,
Fourier transform, short-time Fourier transform, Cohen’s class, wavelet trans-
form and Hilbert–Huang transform are used for signal processing and artificial
neural networks, fuzzy logic, support vector machine, Bayesian classifiers and hybrid
classifiers are used for interpreting the data.
The damage identification can be classified into parametric methods and non-
parametric methods. The parametric methods profoundly use the extracted dynamic
properties of the structure such as natural frequency, mode shapes and their deriva-
tives like modal curvature, modal strain energy and modal flexibility matrices as
damage features to determine the health state of the structure [1–4]. The parametric
modal-based techniques suffer from a few disadvantages such as (i) the requirement
of the finite element model, (ii) inability to handle the effects of environmental and
operational variability, (iii) insensitivity to minor damage and (iv) infeasibility for
online monitoring. Due to the inherent disadvantages of the parametric techniques,
the non-parametric methods famously involving statistical pattern recognition using
time series analysis [5, 6] have gained popularity. The times series analyses using
signal processing tools are robust and highly amenable for implementation through
machine learning frameworks.
The majority of the damage detection techniques reported in the literature have
exploited the responses collected using the accelerometers while monitoring a struc-
ture at a global level. The global methods have shown success in identifying the
damage with higher intensities. However, the nature of damage in the structure is
highly local. Apart from this, the low levels of damage do not change the global
characteristics significantly, and so, the global level sensors may not be successful
in capturing the effect of the damage in the vibration responses. Recently, the use of
A Data Fusion Technique Using Acceleration and Strain … 359
strain sensors for damage detection [7–11] has gained popularity. The strain sensors
can capture local information and so they are sensitive to small intensity damage.
Nevertheless, the sensors famously used for the response measurements are
accelerometers and strain gages. The strain sensors are cheaper in cost when
compared to accelerometers. Also, a strain sensor can capture the damage if and
only if it occurs near the sensor’s proximity. As the strain sensor picks the local
information perfectly, even minor intensity damage, near the strain sensor, can be
captured.
Therefore, damage detection using strain sensors is a local approach, which is
capable of identifying even the minor damage, provided the damage lies in its vicinity.
However, in a spatial civil structure, it is highly impossible to install and maintain
many strain gages to obtain the point-to-point local information, due to damage.
Also, the change in the measured dynamic response depends upon the location and
the intensity of damage. These factors are seen as the shortcomings while using the
local approaches such as strain sensors for damage detection.
The shortcomings of both local and global methods can be overcome by employing
a viable solution through multi-metric sensing, where the different types of sensors
are employed and their measurements are fused.
2 Data Fusion
Data Fusion involves the process of combining the information from sources with
varied representations. This yields fused new data which possess detailed and reliable
information about the system. Normally, data fusion is carried out in three distinct
ways, namely data level, decision level and feature level. The fusion in the data level
combines the raw data from various sources into a single data without any post-
processing. The noise and errors are not propagated and magnified in this method
of data fusion. The fusion in the decision level combines the outputs of the various
algorithms to finally provide an accurate solution. However, the solution from this
method of data fusion is subjective which relies on the potential of the decision-
making authority. Data fusion in the feature-level performs the fusion of information,
which is obtained after the series of processes on the data. Therefore, the feature-level
data fusion can provide more precise information about the status of the system.
Zhu et al. [12] fused the commonly collected acceleration and strain measurements
of supertall structures, to propose an accurate and economical method of measuring
the dynamic displacement. They found that the displacement from the data fusion
approach is more accurate than the global position system-measured displacement
in the time and frequency domains using error analysis. The main advantage of data
fusion of acceleration and strain sensor measurements is that they act complementary
to each other and therefore can provide better information on the structure. This
can be explained with the following facts. While the modal strain energy is high
at the locations with the fixed boundaries, the displacement is zero. Similarly, the
displacement is high at the locations with free boundary conditions whereas the
360 K. Lakshmi
strain value is zero. Also, due to the presence of noise and the uncertainties in the
measurements, it is highly challenging to detect the damage using the capabilities of
a single type of sensor.
A feature-level data fusion is carried out by Guo et al. [13], using accelera-
tion data, to integrate the outputs of the cepstral metric conversion method (CMC)
using the AR model and an autoregressive conditional heteroskedasticity conver-
sion (ARCHC) method based on the ARCH model by applying the Dempster–
Shafer theory. Another feature-level data fusion is proposed by Soman et al. [14]
for damage identification with strain sensors and accelerometers. The authors used
the modal flexibility index (MFI) method [15] on the strain-based mode shapes (from
strain sensors) and displacement-based mode shapes (from accelerometers). A coor-
dinate transformation is used to combine the extracted features together for damage
detection.
Guo [16, 17] fused the changes of frequencies and mode shapes of two-
dimensional truss structures and were successful in detecting damage. Basir and
Yuan [18] employed an evidence theory on multi-sensor data to diagnose the engine
system. Bao and Li [19] minimized the uncertainty and improved the accuracy of
damage identification by employing the D-S evidence theory and Shannon entropy.
Yang and Kim [20] used the D-S theory to diagnose the fault in the motors. Guo and Li
[21] developed a two-stage method, using a weighted data fusion damage index and
genetic algorithm, to locate and quantify the extent of multiple structural damages.
They demonstrated the technique on a two-story concrete frame, instrumented with
piezoceramic-based smart aggregates.
3 Current Focus
The present paper proposes a new damage detection methodology based on multi-
metric data fusion. In this work, it is proposed to perform a feature extraction level
data fusion [22] using the measurements from strain sensors and accelerometers.
A transfer function is established between the acceleration and the strain data for a
group of sensors and used further to predict the responses by the autoregressive model
with exogenous inputs (ARX) model. Initially, the failure of the features extracted
by the single type of sensor measurements is illustrated using an example of a simply
supported beam. Later, the formulations of the proposed data fusion technique are
presented. It is followed by the methodology of the proposed technique and the
validation studies. Finally, the conclusions arrived from the validation studies are
presented.
A Data Fusion Technique Using Acceleration and Strain … 361
a
b
x(t) = αi x(t − i) + β j ex (t − j) + εx (t) (1)
i=1 j=1
where εx (t) is the residual error after fitting the ARX(a, b) model to x(t) using the
input signal. Equation (1) relates the current output x(t) to a finite number of past
outputs x(t − i) and inputs ex (t − j). The structure is thus entirely defined by the
integers a and b. The one-step prediction error εx (t) of an ARX model over a dataset
is generally referred to as model residuals. The model parameters are estimated by
methods based on least squares, maximum likelihood and Yule–Walker equation
estimates. In this method, the response time series from the adjacent sensors are
considered as exogenous Inputs while predicting an ARX model of a time series of
the interested sensor node.
The damage identification using a single type of measurement is carried out by
initially considering the acceleration data and later using strain data independently.
The outcomes of the diagnosis on the four damage scenarios (D1–D4) using accel-
eration measurements and four scenarios using strain measurements are tabulated in
Table 1. The successful detection of damage is indicated as Y and the failure of detec-
tion is indicated as N in Table 1. It can be seen that the acceleration measurement
failed to identify the minor damage cases D1 and D2. This is because the presence of
measurement noise in the acceleration response masks the effect of minor damage and
makes it difficult to realize its feature in the response. The major damage scenarios
D3 and D4 were detected by the ARX method using acceleration measurement as
the major damage creates a global phenomenon. Similarly, the strain measurement
captured both the minor and major damages D1 and D3, in the vicinity of the sensor
nodes and failed to identify the damage scenarios D2 and D4, located away from the
sensor nodes.
From the above study, it can be clearly seen that the single type of measurement by
using either acceleration or strain for damage diagnosis cannot cater to the detection
of damage with varied intensities and locations. In view of this, in this paper, a novel
technique is proposed, to fuse the capabilities of acceleration and strain measurements
for enhanced damage localization. The proposed data fusion technique establishes
a relationship between the strain and the acceleration response from the structure
and exploits this relation to predicting the damage location using the ARX model.
Precisely, one type of measurement is used as an input to the ARX model to predict
the other type of response collected from the structure of interest. The formulation
details of the proposed technique are presented in the next section.
Y1 Y2
Model
U System -
Y2 + Residual
of Y1 acting on the structure. Therefore, the method which requires the knowledge of
only the output arising from the structure is preferable for damage detection. In view
of this, it is proposed to use one type of measurement as an input to the model to
predict the other type of measurement. This can be explained with the help of Fig. 2.
In Fig. 2, Y1 and Y2 are the two types of measurements, namely strain and accel-
eration, respectively. The strain measurement is given as an input to the model and
the acceleration response is predicted as Y 2 . The residue is obtained using the orig-
inally measured acceleration and the predicted one. In this way, a relationship is
formed between the two types of measurements, which can be exploited to extract
the damage feature in the data fusion framework.
Y1
Considering the first block of Fig. 2, U is the input and Y = are the outputs.
Y2
The transfer function of the system is written as
G1
G= (2)
G2
Y = G ·U
that is,
Y1 G1
= ·U (3)
Y2 G2
where Y , G ∈ 2×1 .
In the case where G −1
1 exists, we have
Y2 = G 2 · G −1
1 · Y1 (4)
that is,
364 K. Lakshmi
Equation 5 shows that the output Y1 is related to the output Y2 through H21 .
Therefore, it is possible to use this relationship to extract the information related to
the structure. If there are a series of outputs Y 1 to Y n, they can be grouped into S1
and S2 such as
⎡ ⎤ ⎡ ⎤
Y1 Yi+1
⎢ Y2 ⎥ ⎢ Yi+2 ⎥
⎢ ⎥ ⎢ ⎥
S1 = ⎢ . ⎥ S2 = ⎢ . ⎥ (6)
⎣ .. ⎦ ⎣ .. ⎦
Yi and
Yn
S2 = T21 · S1 (7)
Y = (G + F) · U (8)
that is,
Y1 G 1 + F1
= ·U (9)
Y2 G 2 + F2
F1
where F = .
F2
If there is no damage, F = 0.
When (G 1 + F1 )−1 exists mathematically, similar to Eq. (4), it can be written as
Y2 = (G 2 + F2 ) · (G 1 + F1 )−1 · Y1 (10)
r = Y2 − Ỹ2 (11)
Na Nb
y 2 (n) = k=0 ak · y1 (n − k) − k=1 bk · y 2 (n − k), ∀n > N b
(12)
ỹ2 (n) = y2 (n), ∀n ≤ N b
Initially,
The residual is evaluated as
Na
Nb
r (n) = y2 (n) − ak · y1 (n − k) − bk · ỹ2 (n − k) ,
k=0 i=1
∀ max(N a, N b) ≤ n ≤ N
r (n) = 0, ∀n < max(N a, N b) (14)
The threshold can be fixed by observing the averages and standard deviation of
R.
obtained for every subset. After the residuals of all the subsets are obtained, a vector
R is formed by combining the residuals. The threshold value is evaluated as
After the threshold is evaluated, the technique is ready to assess new data from the
system in the testing phase. The following steps are performed in the testing phase:
1. Obtain the current measurements—Acceleration and strain measurements from
the current state are collected and formed into subsets.
2. Standardize and normalize the signal—After standardization, data normal-
ization is performed. Every current subset of acceleration and strain data is
matched with the corresponding baseline subsets to handle the operational and
environmental variability.
3. Autoregressive Models (ARX)—Fitting of the ARX model for the matched
subsets to predict the acceleration data of the compound nodes.
4. Damage index, H = Mean(R) for every compound node.
5. If H > threshold, the region of damage is identified.
6. Inference of damage location—The exact location of damage can be inferred by
considering the acceleration responses of the two compound nodes as input and
predicting the strain sensor node responses. The highest residual corresponds
to the location of the strain sensor near damage.
A brief flowchart of the basic procedure is shown in Fig. 3.
5 Validation Studies
The ability of the proposed technique for enhanced damage localization is validated
using a numerical example and an experimental study. A simply supported beam is
considered for the numerical example and a scaled-down bridge model of a span of
the in-use live RCC T-beam bridge is considered for the experimental verification.
To evaluate the proposed data fusion-based approach for damage diagnosis, a simply
supported beam girder with a span of 12 m, as shown in Fig. 4, is considered. The
cross-sectional dimensions of the girder are 0.45 m × 0.5 m. The girder is discretized
with 135 beam elements and 134 sensor nodes, out of which the sensor nodes 15, 30,
45, 60, 75, 90, 105 and 120 are the compound nodes carrying both the accelerometer
and strain sensor. The other nodes are simulated as strain sensor nodes and 35 strain
sensors are placed equidistant of 0.34 m to each other. The time history responses
of all the nodes are computed in the time domain. The beam is excited by a uniform
load which is assumed to be a band-limited white noise excitation. Newmark’s time
marching scheme is used to compute the displacement and acceleration time history
response at every node with a sampling speed of 1000 Hz. This simulates the data
collected on a bridge girder on several time instants with varied operational (i.e.,
traffic levels and wind) and environmental loads.
Considering the fact that the time history responses of a structure, during field
experiments, are often polluted with the measurement noise, in this numerical simu-
lation, the computed time history measurements are also contaminated with white
Gaussian noise as given in Eq. (18). This contaminated measurement is useful in
investigating the sensitivity of the proposed data fusion-based damage diagnostic
algorithm, to handle the measurement noise.
.. ..
..
x = x +ζp σ x Nnos (18)
m c c
0.10 8.0x10-8
Healthy Healthy
0.08 6.0x10-8
0.06
Acceleration m/s2
4.0x10-8
0.04
2.0x10-8
0.02
Strain
0.00 0.0
-0.02 -2.0x10-8
-0.04 -4.0x10-8
-0.06
-6.0x10-8
-0.08
-8.0x10-8
0 4000 8000 12000 16000 0 4000 8000 12000 16000
Time Time
(a) Acceleration time history response (b) Strain time history response
.. ..
where x refers to the noise polluted time history measurements, x are the noise-free
m c
computer-generated time history data, ζ p is a control parameter to indicate the noise
level added to measurements in terms of percentage, Nnos is the standard normal
..
distribution within [−1, 1] and σ x is the standard deviation of the computed time
c
history response. In the numerical simulations, the noise-free simulation results are
polluted by random Gaussian noise in the range of 0–8%.
During the time history computations, the structure is subjected to varying levels
of temperature. The reference temperature is varied from 0 °C to + 45 °C. These
variations are taken into consideration during finite element simulations to include the
effect of temperature variation in the acceleration time history calculations. Typical
acceleration and dynamic strain responses are shown in Fig. 5.
The initial phase of the proposed technique is executed. A group of 7 strain sensors
(3 on either side of a compound node and 1 on the compound node) are associated
with an accelerometer of every compound node. The residual R of the ARM model
and the coefficients are evaluated. The threshold value is also evaluated for future
diagnosis.
Similarly, two damage scenarios S1 and S2, near and away from the compound
node, respectively, are simulated. The damage scenario S1 is simulated by reducing
the stiffness of the element near sensor node 15 by 4%. Scenario S2 is simulated
with an 8% stiffness reduction in the middle of the sensor nodes 30 and 45 which are
also the compound nodes. Care is taken such that none of the damage is simulated
exactly on a strain sensor node. The damage scenarios are indicated pictorially in
Fig. 4.
The responses generated using acceleration and strain quantities are employed
for diagnosis in the testing phase. The step-by-step procedure indicated in Sect. 3 is
followed and the damage index is evaluated for every compound node. The group
of strain sensor responses are used as inputs for the ARX model to predict the
acceleration response in the compound node. The values of the damage indices (mean
of the residuals) of the compound nodes are shown in Fig. 6. It can be seen from
Fig. 6 that the region of damage is clearly indicated through the higher magnitudes of
A Data Fusion Technique Using Acceleration and Strain … 369
Damage Index, H
0.5
0.4
0.3
0.2
0.1
0.0
0 15 30 45 60 75 90 105 120
Sensor Node Number
the mean of the residuals obtained from the ARX model. Once the compound nodes
with damage indication are identified, the strain sensors in the region are subjected
to the ARX model predictions using the acceleration responses of the significant
compound nodes to identify the exact location of the damage. Figure 7 indicates the
exact location of damage simulated in scenarios S1 and S2. The number indicated
within the parenthesis in X-axis represents the compound node number.
From Figs. 6 and 7, it can be verified that the proposed data fusion-based damage
detection technique can locate the minor damage using acceleration and strain data
with the presence of measurement noise and temperature variability.
0.8 0.7
0.7 0.6
Damage Index, H
Damage Index, H
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2 0.2
0.1 0.1
0.0 0.0
1(15) 2 3 4 5(30) 1(30) 2 3 4 5 (45)
Sensor Nodes Sensor Nodes
(a) (b)
Fig. 7 Indices of the identified damage region: a Between compound nodes 15 and 30; b Between
compound nodes 30 and 45
370 K. Lakshmi
bridge model with a span of 7 m is given the simply supported boundary conditions.
The bridge model is instrumented with 15 strain sensors, 3 accelerometers and a
displacement transducer. The healthy responses are generated by exciting the model
with a long-stroke exciter as shown in Fig. 9. The longitudinal girders are the prime
load-carrying members of the bridge. Hence, the probability of damage occurring in
these girders is relatively higher. Accordingly, minor damage is induced in the middle
longitudinal girder of the bridge. A crack of length 4 mm and negligible width is
created by a cutting machine at the bottom of the web. There are 3 compound nodes
and 7 associated strain sensors for each compound node. There is no need to include
page numbers or running heads; this will be done at our end. If your paper title is
too long to serve as a running head, it will be shortened. Your suggestion as to how
to shorten it would be most welcome.
The dynamic strain and acceleration measurements collected before and after
damage are used in the proposed technique to identify the minor damage simulated
in this experimental model. The coefficients, residuals and the threshold are evaluated
in the initial phase as explained in Sect. 3. The testing phase evaluated the damage
indices of the compound nodes using the strain sensor responses as inputs to the
ARX model. The damage indices of the compound nodes are plotted in Fig. 9. It
can be seen that the second compound node is showing higher damage index and so
the sensor groups around that compound node are refitted using the ARX model, by
considering the response from the second compound node as input to the ARX model.
The damage indices of all the groups of sensors are shown in Fig. 10. The number
indicated within the parentheses in X-axis represents the compound node number.
It can be seen from Fig. 10 that the minor damage is indicated in the middle of the
sensor group, which is the mid-span location. From the above study, it is clear that
the fusion of the information from acceleration and strain sensors has successfully
identified the location of the minor damage, even from the noisy data.
372 K. Lakshmi
0.9 0.9
0.8 0.8
Damage Index, H
Damage Index, H
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
4 8 12 1 2 3 4 (4) 5 6 7 8
Sensor Node Number Sensor Node Number
(a) (b)
Fig. 10 a Damage indices at compound nodes; b Damage indices around the identified region
(compound node number 4)
Acknowledgements This paper is published with the permission of the Director, CSIR-SERC;
Chennai.
A Data Fusion Technique Using Acceleration and Strain … 373
References
1. Salawu OS, Williams C (1995) Bridge assessment using forced-vibration testing. J Struct Eng
121:161–173
2. Yamaguchi H, Matsumoto Y, Kawarai K, Dammika AJ, Shahzad S, Takanami R (2013) Damage
detection based on modal damping change in bridges. In: Proceedings of the ICSBE’12, Kandy,
Sri Lanka, pp 14–16
3. Zhao J, Zhang L (2012) Structural damage identification based on the modal data change. Int
J Eng Manuf 4:59–66
4. Pandey AK, Biswas M, Samman MM (199) Damage detection from changes in curvature mode
shapes. J Sound Vib 145:321–332
5. Amezquita-Sanchez JP, Adeli H (2014) Signal processing techniques for vibration-based health
monitoring of smart structures. Arch Comput Methods Eng. https://doi.org/10.1007/s11831-
014-9135-749
6. Sohn H, Farrar CR (2001) Damage diagnosis using time series analysis of vibration signals.
Smart Mater Struct 10(3):446
7. Chakraborty S, DeWolf JT (2006) Development and implementation of a continuous strain
monitoring system on a multi-girder composite steel bridge. J Bridg Eng 11:753–762
8. Benedetti M, Fontanari V, Zonta D (2011) Structural health monitoring of wind towers: remote
damage detection using strain sensors. Smart Mater Struct 20:055009
9. Adewuyi A, Wu Z, Serker N (2009) Assessment of vibration-based damage identification
methods using displacement and distributed strain measurements. Struct Health Monit 8:443–
461
10. Lan C, Zhou Z, Ou J (2014) Monitoring of structural prestress loss in rc beams by inner
distributed brillouin and fiber bragg grating sensors on a single optical fiber. Struct Control
Health Monit 21:317–330
11. Reynders E, Wursten G, De Roeck G (2014) Output-only structural health monitoring in
changing environmental conditions by means of nonlinear system identification. Struct Health
Monit 13:82–93
12. Zhu H, Gao K, Xia Y, Gao F, Weng S, Sun Y, Hu Q (2020) Multi-rate data fusion for dynamic
displacement measurement of beam-like supertall structures using acceleration and strain
sensors. Struct Health Monit 19(2):520–536
13. Guo H, Zhou R, Zhang F (2019) Structural nonlinear damage detection using improved
Dempster-Shafer theory and time domain model. J Vibroeng 21(6):1679–1693
14. Soman R, Kyriakides M, Onoufriou T, Ostachowicz W (2018) Numerical evaluation of multi-
metric data fusion based structural health monitoring of long span bridge structures. Struct
Infrastruct Eng 14(6):673–684
15. Talebinejad I, Fischer C, Ansari F (2011) Numerical evaluation of vibration-based methods for
damage assessment of cable-stayed bridges. Comput Aided Civil Infrastruct Eng 26:239–251
16. Guo H, Zhang L (2006) A weighted balance evidence theory for structural multiple damage
localization. Comput Methods Appl Mech Eng 195(44–47):6225–6238
17. Guo HY (2006) Structural damage detection using information fusion technique. Mech Syst
Signal Process 20(5):1173–1188
18. Basir O, Yuan X (2007) Engine fault diagnosis based on multi-sensor information fusion using
Dempster-Shafer evidence theory. Information Fusion 8(4):379–386
19. Bao Y, Li H (2007) Application of information fusion and Shannon entropy in structural damage
detection. In: Health monitoring of structural and biological systems, proceedings of SPIE, San
Diego, Calif, USA
20. Yang BS, Kim KJ (2006) Application of Dempster–Shafer theory in fault diagnosis of induction
motors using vibration and current signals. Mech Syst Sig Process 20(2):403–420
21. Guo HY, Li ZL (2009) A two-stage method to identify structural damage sites and extents
by using evidence theory and micro-search genetic algorithm. Mech Syst Signal Process
23(3):769–782
374 K. Lakshmi
22. He H, Yan W, Zhang A (2012) Structural damage information fusion based on soft computing.
Int J Distrib Sensor Netw 8:798714
23. Lakshmi K, Rao RM (2014) A robust damage-detection technique with environmental vari-
ability combining time-series models with principal components. Nondestructive Testing Eval
29(4):357–376
24. Lakshmi K, Rao A (2015) Damage identification technique based on time series models for
LANL and ASCE benchmark structures. Insight Non Destructive Testing Condition Monit
57(10):580–588
25. Lakshmi K, Rao ARM (2018) Output-only damage localization technique using time series
model. Sādhanā 43(9):147
Nonlinear Ultrasonic Wave Propagation
for Detection of Initial Stage of Corrosion
in Reinforced Concrete
Abstract Detection of structural cracks due to corrosion at early age is more impor-
tant to reduce the repair cost and to achieve the intended service life. The present paper
investigates the technique for early stage detection of corrosion-induced damage in
reinforced concrete elements by using nonlinear ultrasonic wave propagation tech-
nique. Accelerated corrosion test was conducted on the reinforced concrete spec-
imens. Current was measured continuously to record any change in intensity of
current during the entire experimental investigations. Ultrasonic tests were conducted
at specific intervals during the initiation of corrosion, with an excitation signal
frequency of 25–75 kHz. From the investigations, it is observed that nonlinear ultra-
sonic wave propagation techniques can be able to detect the micro-cracks due to
corrosion at initial stage in which the linear ultrasonic techniques could not iden-
tify. A distinct and consistent change in the nonlinear ultrasonic features with the
progression of corrosion-related damage in the reinforcement concrete element is
also found.
1 Introduction
Corrosion of reinforced steel is one of the most important problems to affect the
performance of reinforced concrete (RC) structural components, especially in the
case of bridge and offshore platform structures. When reinforcement steel is corroded
due to adverse environment, the corroded products occupy more volume than that of
the uncorroded steel. This expansion results in cracking and spalling of the concrete
cover that further accelerates the rate of corrosion. If corrective efforts are not taken
in time, a crack will accelerate its expansion and eventually lead to structural failure.
Therefore, the use of suitable evaluation technique for the detection or identification
of cracks at early stage is imperative for maintaining the integrity and safety of RC
structures.
Towards this, several methods have been developed in the recent years such as
electrochemical methods (potential measurement, resistivity measurement, polariza-
tion resistance measurement), electromagnetic methods (ground penetrating radar),
optical sensing methods (fibre Bragg grating), elastic wave methods (ultrasonic pulse
velocity, acoustic emission, and impact echo) and infrared thermography [1]. Among
them, ultrasonic measurements have been considered as one of the potential non-
destructive evaluation (NDE) techniques [2]. Most of the existing ultrasonic NDE
techniques use the linear parameters obtained from ultrasonic measurement (e.g.,
velocity variation, mode conversion, time of flight (TOF), attenuation coefficients and
frequency shift) to detect and monitor the corrosion-induced damage in RC structural
components. However, the technique based on the linear parameters is quite insensi-
tive to the early stage of corrosion-induced damage. Reason being is the small damage
or micro-crack formation doesn’t mark significant changes in the linear features to be
extracted from the temporal ultrasonic wave. To overcome this, nonlinear ultrasonic
signatures based on the frequency domain analysis of signals (Signal harmonic gener-
ation (SHG), nonlinear wave modulation spectroscopy (NWMS)) have been emerged
in recent years for corrosion monitoring as they can effectively detect micro-cracks.
However, more common nature of the nonlinear response as obtained from these
methods is the low amplitude of the higher harmonics or sidebands. In most of
the situations, the presence of noise in the received ultrasonic signal hinders the
higher harmonic amplitude and that makes it difficult to detect the damage, unless
the receiving sensors are located very close to defect location or the input excitation
amplitude is sufficiently high. To overcome this issue with the nonlinear ultrasonic
parameters obtained in frequency domain analysis, Scalerandi et al. [3] proposed
the Scaling Subtraction Method (SSM) that captures nonlinear parameter from the
temporal signal. In spite of the promising features of nonlinear ultrasonic for damage
detection, investigation on the initial stage detection of corrosion-induced damage is
scanty [4–8]. Major challenge associated with ultrasonic investigations of concrete
structures is the presence of enormous attenuation, because of the absorption and scat-
tering phenomena by individual constituents such as cement paste and aggregates,
which limits the excitation frequency range to reduce noise. For a reliable detection
of micro-cracks in heterogeneous medium like concrete, comprehensive ultrasonic
signal interpretation is further needed both from theoretical and experimental point
of view.
Having known the potential of the nonlinear ultrasonic features in terms of early
stage detection of damage, comprehensive study on the evaluation of nonlinear
parameters at different stages of corrosion is worth demanding. For this purpose,
prismatic RC specimens were cast and subjected to accelerated corrosion tests
by applying defined electric field. Meanwhile, ultrasonic measurements were
performed. The time domain analyses are employed for early stage detection of
Nonlinear Ultrasonic Wave Propagation for Detection of Initial … 377
cracks induced by corrosion. The goal of this contribution is (a) to analyze the effects
of damage induced by corrosion on the linear and nonlinear ultrasonic parameters
in time domain; (b) to identify the suitable measurement protocol for the nonlinear
ultrasonic measurements for corrosion-induced concrete damage; (c) to bring out the
promising ultrasonic technique(s) for identifying the early stage damage detection
in large concrete structures.
In Sect. 2, brief recall on how the different linear, nonlinear parameters could be
obtained from time domain analysis will be presented. Section 3 discusses the exper-
imental procedure followed in the present study. In Sect. 4, results are presented
demonstrating the sensitivity of nonlinear parameter. Finally, major conclusions
arrived based on the present work are highlighted in Sect. 5.
2 Theoretical Background
Ultrasonic parameters such as time of flight (or in terms of ultrasonic wave speed),
attenuation obtained from time domain waveform are generally used for condi-
tion assessment. These parameters are linear and well documented. Apart from
this, Scaling Subtraction Method (SSM) developed by Scalerandi et al. [3] involves
nonlinear ultrasonic measurement in the time domain, and it has proven to an efficient
method to monitor damage. Brief description about each parameter is given below
for the sake of continuity and ready reference.
When the ultrasonic wave travels through a material, the wave speed, c can be
calculated from the distance ‘d’ between two transducers and its corresponding
time-of-flight ‘t’ (time to reach the receiver transducer location) as:
c = d/t (1)
In the present work TOF, t is determined using the auto-picking algorithm based
on the Akaike information criteria (AIC) [9].
Attenuation represents the decaying phenomenon of elastic mechanical radiation
energy of the ultrasonic waveform after experiencing absorption and scattering in its
propagation path. After travelling a distance (Z) through the medium, the amplitude
of the signal reduces to U from its initial amplitude U0 . Generally, attenuation is
expressed in terms of coefficient, ξ that relates the signal amplitude with its decay
along the travelling direction, as Antonaci et al. [10]:
U = U0 e−ξ Z (2)
378 A. Thirumalaiselvi and S. Sasmal
Nonlinearity of the system in time domain can be estimated using the SSM [3].
SSM assumes that for low amplitude of a harmonic excitation, the response of the
damaged specimen is almost equal to that of the undamaged case [3, 10]. Under this
assumption, the recorded signal for a low excitation amplitude can be considered
as the linear elastic response of the system, with the definition of a reference signal
vref (t) [10]. The reference signal at amplitude A = k × A low (A low = A linear)
can be defined in a post-processing procedure. In fact, if the system was linear, the
superposition principle would hold, and the output signal would be expected to be
SSM method consists of, exciting the system by a low harmonic signal (amplitude,
Alow ) and its corresponding response is vlow (t). With the scaled amplitude of the
excitation, and the corresponding output vA (t) acquired, scaled subtracted signal
w(t) is computed as:
While the scaled subtracted signal w(t) in undamaged stage is expected to vanish,
the one obtained in damaged stage contains the nonlinear contributions of the system
response [3].
Nonlinear parameter, θ is defined as the energy of the subtracted signal over a
time interval T:
1 T 2
θ (A) = w (t)dt (5)
T 0 A
Plotting of curve between θ and x for different damage stages indicates the pres-
ence of nonlinearity in the system. For each damage stage, the curve can be fitted by
power low as:
θ = ax b (7)
3 Experimental Procedure
Reinforced concrete prism specimen of size 100 × 100 × 500 mm3 was cast (see
Fig. 1) in steel moulds, with a single reinforcement steel bar of 16 mm diameter,
located approximately in the centre of the specimen, leaving an exposed steel area of
240 cm2 (approximately). Before placing the reinforcement into the empty mould,
its surface was cleaned to get rid of the inherent corrosion products. The concrete
specimens were mechanically compacted, surface finished, and left in the moulds
for about 24 h of time. After that specimens were demoulded and cured in water for
28 days.
Concrete mix constituents include Ordinary Portland Cement (OPC) of grade
53, fine aggregate (FA), coarse aggregate (CA) (of 10 mm and 20 mm size) and
water; the mix proportions by weight of cement, FA and CA were 1:2.25:2.35, with
a water-to-cement ratio of 0.5. Superplasticizer was added (100 ml per 50 kg of
cement).
The accelerated corrosion test was performed by applying an electric field between
the reinforcement bar (anode) and an external cathode, consisting of a steel plate.
The test, which lasted 14 days, was run by maintaining the constant voltage of 10 V
using a power supply unit. During the tests, the bottom of the specimen was kept over
the steel plate, in contact with water containing 3.5% NaCl solution and the height of
the contact between water and specimen is 66 mm, just above the reinforcement bar
level, in order to maintain an adequate level of electric conductivity for the material
(see Fig. 2).
Every day, damage (appearance of first crack and crack propagation) was visually
inspected and continuous current measurements are computer recorded using the
multimeter to know about the current flow and to further correlate the same with the
ultrasonic measurements as damage progresses. Due to the adopted corrosion test
set-up and specimen geometry, the cracks due to steel corrosion appeared only at the
end, lateral and bottom surface of the specimen, not at top surface.
Table 1 Sensor
Arrangement Sensors
arrangements considered
Transmitter Receiver
A1 T1 R1
A2 T2 R2
A3 T3 R3
A4 T3 R4
A5 T4 R4
A6 T4 R3
Nonlinear Ultrasonic Wave Propagation for Detection of Initial … 383
(intervals of 24 h) were taken daily, day 0 (before the start of accelerated corrosion
test) till the end of test. The test was continued for 14 days, up to the maximum crack
width of 2 mm on the lateral surface of the specimen.
A time window of 2.5 ms was used for waveform acquisition. One millisecond time
window around first arrivals for varying excitation amplitude (10–50 V after ampli-
fication) was considered for time domain-based SSM analysis. Waveform obtained
using 50 V excitation amplitude alone is used for other linear parameter analysis
other than SSM.
The current required to maintain the constant potential of 10 V is plotted against time
in Fig. 5. The current versus time plot shows almost a steady low rate of increase
in the current with minimal fluctuations till 8th day. After that a rapid jump in the
current value is noticed that coincides with the first cracking on the specimen surface
(end and lateral surfaces) noted visually and marked as D8 in Fig. 6a, b. Second jump
in the current value observed at 12th day, which may be due to generation of new
crack, it may be the propagation of pre-existed crack or formation of a new crack.
This is also noticed visually in the lateral surface of the specimen and marked as D12
in the crack pattern photo given in Fig. 6b. Test is terminated when the maximum
crack width of 3 mm is observed with a tremendous increase in the current value.
Fig. 6 Photo taken at the end of corrosion test (a) end surface; (b) lateral surface; bottom surface
TOF
The change in TOF, i.e. normalized with that of the TOF value before start of corrosion
test (t = 0) is plotted versus the time of exposure to corrosion in Fig. 7.
Change in TOF plotted in Fig. 7 does not show any significant trend for corrosion-
induced damage progression in the specimen considered here.
Attenuation
The change in attenuation, i.e. normalized with that of the attenuation obtained before
start of corrosion test (t = 0) is plotted versus the time of exposure to corrosion in
Fig. 8.
Fig. 8 Change in
attenuation in terms of
amplitude reduction
386 A. Thirumalaiselvi and S. Sasmal
From Fig. 8, it is seen that monotonic reduction in peak amplitude is seen from
day 1 to day 4 (day 0 to day 4 in A2 sensor arrangement) beyond that there is sudden
increment at day 7 (day 10 in A2 sensor arrangement), further no significant trend
could be observed.
Scaling Subtraction Method
The quantitative criterion that is adopted in this work is the change of the nonlinearity
parameter (θ ), relative to the initial value. Considering a small time window of the
recorded signal around the first arrivals (t = 1000 µs), nonlinear parameter θ
is obtained for fixed value of x0 = 0.2 v2 (see Theoretical background Sect. 2.2
for the steps followed). Figure 9 shows the measured nonlinearity of specimen as
corrosion-induced damage progresses in terms of measured change in θ (ax0 b ).
From the visual inspection during experiments, the first microscopic observation
of cracking at the side surface of the specimen near rod was at the sixth day after
starting the test. Also, first microscopic observation of cracking at the lateral surface
of the specimen was at the eighth day after the start of test. From Fig. 9, it is seen
that the nonlinearity of the specimen slightly increases from day 0 to day 2, beyond
that a considerable increase of the nonlinearity is observed till day 4 in A2 sensor
arrangement (day 5 in A1 sensor arrangement). This difference in the time at which
change in nonlinearity becomes predominant is reasonable that on the 4th day (at
which clear increase in θ in A2 sensor arrangement), 2 days before the appearance
of the first visible surface crack at rod end, the micro-cracks were initiated at the
concrete cover zone over the protruded reinforcement end and the same has been
replicated in A2 sensor arrangement. Whereas on the 5th day (at which clear increase
in θ in A1 sensor arrangement), 3 days before the observation of the visible crack
at lateral surface, the micro-cracks were actively developing at the concrete cover
region over the steel rebar throughout the length and the same has been replicated in
T1R1 measurement.
Nonlinear Ultrasonic Wave Propagation for Detection of Initial … 387
In most of the derived ultrasonic (linear and nonlinear) parameters in time domain,
it is observed that signals received in A2 sensor arrangement (direct transmission)
are highly sensitive towards monitoring the evolution of corrosion-induced micro-
cracking when compared to A1 sensor arrangement (indirect/surface transmission). It
may be due to the following reason: Signal received in surface mode of transmission
may contain contribution from both surface wave and waves reflected from the steel
bar. This would hinder the sensitivity if micro-cracking initiates far away from the
surface and that was the phenomenon exactly observed visually too, i.e. cracks were
initiated not initiated on the top surface till the end of test. Also, cracks appeared
below the reinforcement rod level at the specimen bottom surface, lateral surface and
rod end surface. This resulted in less sensitive ultrasonic parameters from A1 sensor
388 A. Thirumalaiselvi and S. Sasmal
5 Conclusions
From the present study, it is seen that it is feasible to detect micro-cracking at initial
stage of cracking mechanism due to the corrosion of reinforcements in RC structures.
The appearance of visible surface micro-cracks accompanies with the occurrence
of strong nonlinearity in the received signal: drastic peak in nonlinear parameter
obtained from Signal Subtraction Method (SSM) is clearly observed. The results
also suggest that the linear parameters (TOF and attenuation) are not as efficient as
the nonlinearity parameter (θ ), for detecting the nonlinear characteristics linked with
the micro-cracking events induced by corrosion in concrete.
Another important observation made in the present study is that after the signifi-
cant events of the micro-cracking phenomena, nonlinear parameter returned to values
typical to the previous situation. This may be due to filling of the void space created by
the crack with liquid containing steel corrosion products. However, further compre-
hensive experimental studies are needed to confirm this, and extensive works are
being carried to derive the potentiality of nonlinear ultrasonic features towards the
detection of corrosion-induced damage in large reinforced concrete structures.
Nonlinear Ultrasonic Wave Propagation for Detection of Initial … 389
Acknowledgements The authors would like to acknowledge the technical and project staff of
Special and Multifunctional Structures Laboratory (SMSL), CSIR-Structural Engineering and
Research Centre, India for their help during experimental investigations.
References
1. Zaki A, Chai HK, Aggelis DG, Alver N (2015) Non-destructive evaluation for corrosion
monitoring in concrete: a review and capability of acoustic emission technique. Sensors
15(8):19069–19101
2. Yeih W, Huang R (1998) Detection of the corrosion damage in reinforced concrete members
by ultrasonic testing. Cem Concr Res 28(7):1071–1083
3. Scalerandi M, Gliozzi AS, Bruno CLE, Masera D, Bocca P (2008) A scaling method to enhance
detection of a nonlinear elastic response. Appl Phys Lett 92(10):101912
4. Kwun H, Burkhardt GL, Fisher JL (1993) U.S. Patent No. 5,180,969. Washington, DC: U.S.
Patent and Trademark Office
5. Woodward C, Amin MN (2008) Evaluating rebar corrosion using nonlinear ultrasound. In: AIP
conference proceedings. American Institute of Physics. vol 975, no. 1, pp 1314–1319
6. Korenska M, Matysik M, Vyroubal P, Pospisil K (2009) Assessment of reinforcement corrosion
using nonlinear ultrasonic spectroscopy. In: Proceedings of the 5th NDT progress—interna-
tional workshop of NDT experts. Prague, Czech Republic, pp 12–14
7. Antonaci P, Bruno CLE, Scalerandi M, Tondolo F (2013) Effects of corrosion on linear and
nonlinear elastic properties of reinforced concrete. Cem Concr Res 51:96–103
8. Ramis J, Carbajo J, Poveda P, Ivorra S, López MP, Antón C, Climent MA (2020) Detection of
cracks caused by corrosion in simple structural elements using nonlinear ultrasonic techniques.
In: Proceedings of the 22th international congress of sound and vibration (ICSV22). Florencia
(Italia)
9. Zhang H, Thurber C, Rowe C (2003) Automatic P-wave arrival detection and picking
with multiscale wavelet analysis for single-component recordings. Bull Seismol Soc Am
93(5):1904–1912
10. Antonaci P, Bruno CLE, Gliozzi AS, Scalerandi M (2010) Monitoring evolution of compressive
damage in concrete with linear and nonlinear ultrasonic methods. Cem Concr Res 40(7):1106–
1113
Ultrasonic Sensor Developments
for Monitoring the Temperature
in the Long Region of Interest
Abstract This work reports an ultrasonic waveguide sensor for sensing the temper-
ature changes in a long region of interest using a long (length >20 m) waveguide
and fundamental longitudinal L (0, 1) wave mode. The long waveguide can cover
large (area/volume) region of interest and may also allow for an increased number
of sensors along the length of waveguide while measuring the properties of the
surrounding medium. The accessible length of the waveguide is dependent on the
velocities (longitudinal (VL) and torsional (VT)) of the ultrasonic waves. Thus, a
larger region of interest can be covered by a longer waveguide using L (0, 1) mode
as compared to torsional T (0, 1) wave mode in the same time scale. The waveguide
approaches presented in this work are generic in nature and can be applied using both
T (0, 1) as well as L (0, 1) modes. Hence, the longitudinal wave mode is more suitable
for designing the long-range waveguide sensor as compared to torsional mode due
to its faster velocity (4960 m/s). A pair of waveguides of similar lengths (Alumal
23 m and Chromel 24 m) is connected to a single transducer (shear/longitudinal) at
different orientations to study the possibilities of increasing the coverage length (for
example, 47 m) for the large monitoring region of interest.
1 Introduction
In steel, glass melting and aerospace industries, monitoring the temperature at various
points inside the melting furnace and structure is essential as per manufacturing stan-
dards. Thermocouples and radiation pyrometers are the common temperature sensors
used by industries. In these industries, thermocouples are placed at the required
locations throughout the job, and temperature uniformity should be proved during
processes such as soaking. The parameters that are important for monitoring include
temperature profiles of the glass melts, plenum off-gases, molten glass conductivity,
density, and viscosity (for example, see Woskov et al. [1]). However, these tempera-
ture diagnostic tools have accuracy problems due to sensor drift for long-term opera-
tion (as explained in [2]). Thermocouples, in particular, are prone to reliability issues
(at the hot junction) while being used in hostile elevated temperature environments.
Hence, the replacement of thermocouple-based temperature measurements with a
more robust and reliable temperature measurement approach is of much interest.
Several researchers have developed alternate sensors (contact and non-contact,
for example [3]) for online monitoring of temperature. In the work reported here,
our aim to develop the distributed ultrasonic temperature sensor using long (24 m
+ 23 m) flexible wires that can function as the ultrasonic waveguide. While ultra-
sonic approaches, in general, allow one to probe inside the region of interest, bulk
ultrasound-based methods [4–6] require transducer ruggedization, as the operating
temperatures impact them adversely. Lithium-Niobate and Bismuth Titanate crystals
have been explored to make ultrasonic high-temperature transducers [7, 8] that can
withstand the high temperatures with modest success. But, the ultrasonic waveguide-
based measurement techniques have been extensively used for developing sensors for
level, density, temperature, viscosity and rheology measurement of the surrounding
fluid among others [9–13]. The operating temperature doesn’t affect the transducer
when the transducer is connected with other ends of the long waveguide at room
temperature. The main drawback of the early ultrasonic waveguide approaches
reported in the literature is that they are aimed at the measurement of tempera-
ture at a single region of interest. Moreover, due to continuously varying tempera-
tures, measurement at a single region is also challenging, with straight waveguide
approaches due to the temperature gradient issues [14].
In the present paper, we consider two different types of materials for designing
the long ultrasonic sensors. The thermocouple wires such as Alumal (24 m long)
and Chromel (23 m long) wires are preferred for these sensor developments. Thin
wires with a circular cross-section, similar to those used in K-type thermocouples
are chosen as the waveguide embodiments. The studies are carried out using the
L(0,1) fundamental longitudinal ultrasonic guided wave mode of circular cylindrical
wires. From the ultrasonic guided wave signals reflected from the bend embodi-
ments, the time of flight (TOF due to change in elastic moduli) measurements can
be used for learning more about the temperatures of the environment surrounding
the waveguides. For this purpose, the material properties of the waveguide (density
and elastic moduli) are assumed to be known as a function of temperature. If this is
not available, an earlier reported technique [15–17] may be used to determine these
temperature-dependent moduli.
The change in TOF due to a change in temperature, henceforth called δTOF, was
used to monitor temperatures changes around the bend (horizontal portion) region
of the waveguide. A peak tracking algorithm [more detail elsewhere, [18–21] is
used to analyze the ultrasonic signals at different temperatures from each smooth
bend region of the waveguide. Each bend region of the waveguide is considered to
Ultrasonic Sensor Developments for Monitoring the Temperature … 393
Fig. 1 (a, b) Schematic of long waveguide sensors connected with longitudinal and shear transducer
respectively, (c). A-scan signals collected from two different bent waveguides (Chromel and Alumal)
at room temperature, (d) Reconstructed signal obtained using Time Frequency Analysis (TFA) from
original signal
and it was used for this experiment. In this work, used instruments were the same
as the previously reported works [14, 18, 21]. In order to locate the sensors along
the length of the waveguide, embodiments such as bends can be introduced in the
waveguide that allows the signals to be reflected. Here, periodically spaced bends
were used to reflect the signals from these sensors. A pair of bends can be considered
as one sensor, and a gage (sensor) length 100 mm was used. The A-scan signals were
acquired from the two long waveguides (Alumal-23 m and Chromel-24.2 m) as shown
in Fig. 1c, each waveguide secured to a single sensor at room temperature. Hence,
using this experimental setup, the temperature at least 23 m in either direction (from
the exiting/receiving source) can be monitored in any remote area. The joint Time–
Frequency algorithm (TFA) used in this work employs a discrete Gabor transform,
which is an invertible, linear time–frequency transform. The output of the discrete
Ultrasonic Sensor Developments for Monitoring the Temperature … 395
Gabor transform is called the Gabor coefficients. This concept is applied to develop
a TFA tool for signal reconstruction and improving the peak tracking in this work.
Hence, the TFA algorithm (available as a sub-vi tool in LabVIEW®) was used to
reconstruct the original signals in order to reduce noise, as shown in Fig. 1d. The
reconstructed signal had a similar shape (Hanning pulse and approximately 19–21 μs
pulse widths) to the obtained original signal. There were also amplitude variations
in the reconstructed signal when compared to the original signal (using a 10 MHz
sampling rate for long waveguide). This issue could not be observed when using a
shorter length of waveguide when using 100 MHz sampling rate [14, 18, 21]. Finally,
we considered the sensor development based on the TOF shift using 24 m length of
the Chromel waveguide.
Next, four pairs of bend embodiments were provided on the long (24.2 m) Chromel
waveguide, and each sensor had a length of 100 mm, all sensors distributed along
the length of the waveguide. The gage length was chosen for clearly identifying
multiple reflections from each bend embodiment, and it can be varied according
to application. Multiple reflections (original and TFA reconstructed signals) were
obtained from the distributed sensor waveguide at room temperature, as shown in
Fig. 2a, b. Here, a single 0.5 MHz longitudinal transducer and sampling rate of
10 MHz was used in this experiment. The reflection coefficient can vary based on
the bent radii, which are used along the length of the waveguide sensor (more details
refer; [22]). In this work, bent radii were randomly selected/modified for obtaining
the appropriate strength of reflected signals. The total number of sensors could be
Fig. 2 (a, b) Shows the original and the TFA reconstructed A-scan signals with multiple reflections
obtained respectively, from distributed sensors in a 24.2 m long ultrasonic Chromel waveguide
396 S. Periyannan and K. Balasubramaniyam
3 Conclusion
Fig. 3 (a, b) shows the reconstructed A-scan signals collected from sensor 1 and 3, respectively, at
different temperatures and the change in the TOF of the RF signals can be observed with change in
temperature, which can be utilized for measuring local temperatures, (c) shows TOF versus Temp
of sensor 1 at various temperatures
398 S. Periyannan and K. Balasubramaniyam
References
1. Woskov PP, Sundaram SK, Daniel WE (2008) Millimeter-Wave measurement of high level and
low level activity Glass melts. MIT, PSFC/RR-08–4, pp 1–28
2. Bentley RB (1990) Long-term drift in mineral-insulated nicrosil-sheathed type K thermocou-
ples. Sens Actuat A Phy 24:21–26
3. Carlson NM, Musick CA, Watkins AD, Tillotson RD, Mullenix PD (1999) Evaluation of a
novel temperature sensing probe for monitoring and controlling glass temperature in joule
heated glass melter, INEEL/CON–99–00609, pp 1–10
4. Tsai WY, Chen HC, Liao TL (2005) An ultrasonic air temperature measurement system with
self-correction function for humidity. Meas Sci Technol 16:548–555
5. Tsai WY, Chen HC, Liao TL (2006) High accuracy ultrasonic air temperature measurement
using multi-frequency continuous wave. Sens Actuat A Phy 132:526–532
6. Huang KN, Huang CF, Li YC, Young MS (2003) Temperature measurement system based on
ultrasonic phase-shift method. IEEE, pp 294–295
7. Baba A, Searfass CT, Tittmann BR (2010) High temperature ultrasonic transducer up to 1000
°C using lithium niobate single crystal. Appl Phy Lett 97:232901
8. Parks DA, Zhang S, Tittmann BR (2013) High-temperature (> 500 Degrees C) ultrasonic trans-
ducers: an experimental comparison among three candidate piezoelectric materials ultrasonics
ferroelectrics and frequency control. IEEE Trans 60:1010–1015
9. Lynnworth LC (1989) Ultrasonic measurements for process control: theory, techniques
applications. Academic Press Inc. New York, U.S.A
10. Kazys R, Sliteris R, Raisutis R, Zukauskas E, Vladisauskas A, Mazeika L (2013) Waveguide
sensor for measurement of viscosity of highly viscous fluids. Appl Phys Lett 103:204102
11. Nishanth R, Balasubramaniam K, Periyannan S (2019) Ultrasonic waveguide-based level
measurement using flexural mode F (1, 1) in addition to the fundamental modes. Rev Scient
Instrum 90(4):045108
12. Prasad VSK, Balasubramaniam K, Kannan E, Geisinger KL (2008) Viscosity measurements of
melts at high temperatures using ultrasonic guided waves. J Mater Process Technol 207:315–
320
13. Pandey JC, Raj M, Lenka SN, Periyannan S, Balasubramaniam K (2011) Measurement of
viscosity and melting characteristics of mould powder slags by ultrasonics. J Iron Making
Steel Making 38:74–79
14. Periyannan S, Balasubramanian K (2015) Multi-level temperature measurements using
ultrasonic waveguides. Measurement 61:185–191
15. Balasubramaniam K, Periyannan S (2020) Waveguide technique for the simultaneous measure-
ment of temperatures dependent properties of materials. US Patent no: US 10,794,870
B2
16. Periyannan S, Balasubramaniam K (2015) Simultaneous moduli measurement of elastic mate-
rials at elevated temperatures using an ultrasonic waveguide method. Rev Scient Instrum
86:114903
17. Periyannan S, Balasubramaniam K (2016) Moduli determination at different temperatures by
an ultrasonic waveguide method. Experim Mech 1–14. https://doi.org/10.1007/s11340-016-
0157-y
18. Periyannan S, Rajagopal P, Balasubramaniam K (2016) Re-configurable multi-level tempera-
ture sensing by ultrasonic “Spring-like” helical waveguide. J Appl Phys 119:144502
19. Periyannan S, Rajagopal P, Balasubramaniam K (2016) Torsional mode ultrasonic helical
waveguide sensor for re-configurable temperature measurement. AIP Adv 6(065116)
20. Balasubramaniam K, Periyannan S (2016) A novel ultrasonic waveguide technique for
distributed sensing and measurements of physical and chemical properties of surrounding
media. US Patent no: US 10520370B2
21. Periyannan S, Rajagopal P, Balasubramaniam K (2017) Multiple temperature sensors
embedded in an ultrasonic spiral-like waveguide. AIP Adv 7(035201)
Ultrasonic Sensor Developments for Monitoring the Temperature … 399
1 Introduction
Since PPM-EMATs are used in this study, the excitation wavelength is constrained.
The wavelength is equal to the periodicity of the magnets. A PPM EMAT of 6.35 mm
wavelength is used in this study. Figure 2a shows the phase velocity dispersion
curves of a 1 mm aluminium plate. The inclined line (or λ-line) overlaid has a slope
of 6.35 mm. The λ-line intersects the SH0 mode at 500 kHz. A 5 cycle Hanning
window is used for the excitation signal, Fig. 2b.
A boundary stiffness approach is used for modelling the interface, which is given by
the equations:
σ = K u (1)
⎡ ⎤
Kt 0 0
K = ⎣ 0 Kn 0 ⎦ (2)
0 0 Kt
Here, K is the boundary stiffness matrix. K t and K n denote the stiffness compo-
nent in the transverse and normal direction. A spring distribution here represents the
isotropic thin layer of young modulus and shear modulus. K n and K t Respectively
related to E and G by the following relation.
E G
Kn = Kt = (3)
d d
Fig. 2 a Phase velocity dispersion curve for SH wave propagation in aluminium 1 mm plate with
6.35 mm operating wavelength line overlaid. b Five-cycle Hanning window input signal
3 Modelling Results
are collected at a total of 150 points. Since the simulation is done in the frequency
domain, the collected data are a function of frequency and spatial coordinates. A
Fourier transform along the spatial dimension yields the ω-k plot. Further to generate
the time responses from the frequency response at a given location, an inverse Fourier
transform is performed [9].
To simulate the wave propagation in a perfectly bonded HSS, K t and K n are fixed
to be 1e17 [N/m3 ]. The time response generated from the FE simulation at a distance
of 110 mm from the excitation source is shown in Fig. 3c. Here multiple modes can
be observed. To identify the modes, ω-k plot is generated from the FE simulation,
and dispersion curve of 1 mm thick aluminium plate is overlid to identify the modes,
which is shown in Fig. 3d. The ω-k reveals that SH0 and S0 modes exist in a perfectly
bonded HSS. The S0 mode is generated as a result of reflection and mode conversion
of SH0 from the core of HSS.
Fig. 3 a Schematic of COMSOL model for perfect bond condition, K n = 1017 [N/m3 ] and K t =
1017 [N/m3 ], b A-scan result for monitoring point at 110 mm from excitation source c ω-k plot for
verifying mode generated in honeycomb sandwich structure
Detection of Interfacial Weakness (Kissing Bonds) in Honeycomb Sandwich … 407
Fig. 4 a Schematic of COMSOL model for perfect dis-bond condition, K n = 0[N/m3 ] and K t =
0[N/m3 ], b A-scan result for monitoring point at 110 mm from excitation source c ω-k plot for
verifying mode generated in honeycomb sandwich structure
To simulate a dis-bond at the interface, K n and K t are fixed to be zero. For perfect
bonding conditions, it was observed that the SH0 mode is converted to S0 and SH0
mode at the given frequency. However, when there is a disbond only SH0 mode is
present in the ω-k plot shown in Fig. 4d. This is because the honeycomb sandwich
structure’s top skin and bottom skin act like an aluminium plate structure. Due to
the perfect dis-bond condition, there is no leakage of wave energy into the core part.
Thus, at dis-bond case, there is no interaction between the SH0 mode with the core
region and hence no mode conversion to S0 mode.
4 Experiment
Fig. 5 Experimental setup for EMAT-based non-destructive detection for honeycomb sandwich
structure
bonded using epoxy adhesive of 0.2 mm thickness. The excitation pulse of the 5-
cycle sinusoid from the Ritec RPR pulsar receiver, fed to the transmitting EMAT. The
signal is collected using an oscilloscope and recorded for further signal processing in
MATLAB. For this experiment, innerspec device configuration, five cycles, 600 vpp,
500 kHz burst frequency, 50 MHz sampling frequency, and bandgap of 400–500 kHz
with 26.5 dB gain and 50 kHz PRF is chosen to generate the excitation signal. The
distance between transmitting and receiving EMAT is around 80 mm.
The experiment results are shown in Fig. 6. The experiments are performed on a
Fig. 6 Experimental result for honeycomb (a) time response monitored at a distance of 80 mm
from the excitation source (b) ω-k plot
Detection of Interfacial Weakness (Kissing Bonds) in Honeycomb Sandwich … 409
perfectly bonded region. It can be observed that for perfect bonding conditions, SH0
mode and S0 mode exist. However, as opposed to the FE results, the intensity of S0
mode is low. This might be due to the attenuation due to the adhesive viscoelasticity
being higher in the experiment. Also, the waveform is different from the FEM result,
but with the ω-k plot, we can conclude that both SH0 and S0 modes exist in a perfectly
bonded region.
5 Conclusion
The shear horizontal guided wave-based method is studied for detecting interface
weakness in the honeycomb sandwich structure. For modelling the different interface
adhesion levels, the spring stiffness boundary condition is modelled. The propagation
of SH0 in the honeycomb sandwich structure is studied using the ω-k plot. It was
observed that incident SH0 mode is converted to S0 mode in a perfect bonding
condition. However, only SH0 mode is present in perfect disbond condition. The
presence of S0 mode can be used as an indicator of bond condition.
For a more precise understanding of this technique, several areas need to be
investigated. The current result demonstrated the potential of SH waves for interfacial
weakness problems. A further detailed study is required for finding the threshold limit
of the mode conversion. For future work, experiments need to be investigated for the
honeycomb sandwich sample.
References
1 Introduction
Composites have been a promising material in almost every industry providing new
prospects for modern technologies [1]. They are made by combining a minimum of
two or more macroscopically identifiable materials, often with different properties
to accomplish the desired result. The materials used in the field of civil, aerospace
and automotive industries require excellent strength and stiffness to weight ratios
and great flexibility in design. The most commonly used composites are glass fiber-
laminated composites because of their cheap cost [2]. In laminated composite mate-
rials, due to unanticipated low-velocity impact and continuous loading, damages
may appear which are much more critical than other materials. The most frequent
damages in laminated composites are delaminations or de-bonds, fiber breakage,
matrix cracking, fiber–matrix debonding, in which delamination, where subsurfaces
are separated, is the most predominant mode of failure in laminated composites
[3–5]. Delamination reduces the flexural and in-plane compression strength signifi-
cantly and it cannot be detected through visual inspection. In such a case, the most
promising technique for detecting delamination is ultrasonic non-destructive testing
(NDT).
Lamb waves, also known as ultrasonic-guided waves, have gained more attention
in the context of offering a convenient approach to inspect composite laminates.
These waves propagate long distances within free boundaries and undergo little
energy loss while monitoring structures. The interaction of the Lamb waves with
delamination for facilitating damage detection in laminated structures has been the
focus of researchers for more than a decade and it still continues. Interaction of
fundamental Lamb modes linearly with delamination has been explored in a large
volume of literatures [6–9]. However, relatively a few researchers have addressed the
nonlinear interaction of Lamb waves with the damage [10–17]. Linear interaction
of Lamb wave-based damage detection methods incorporates the characteristics of
wave signals in the time domain such as reflection, transmission, mode conversion,
etc. [18–20]. Due to multimodal dispersion characteristics, temporal and geomet-
rical constraints modeling of Lamb wave for detection and characterization of the
incipient and small defects is a challenging problem. Nonlinear interaction of Lamb
wave-based detection methods is of great interest because these detection methods
having a large inspection range of Lamb waves are highly sensitive, making it a
perfect method for the detection and characterization of incipient and small defects.
Nonlinear interaction of Lamb waves with delamination in composite plates gives
rise to the generation of higher harmonics, subharmonics, frequency shifts, etc. [21–
24]. These harmonics can become the reliable precursor of the identification of even
small delamination at the early stage. From the perspective of determining the magni-
tude and position of delamination in composite plates, researchers, in the recent past,
have zeroed in on nonlinear harmonics generation in the frequency domain. Some of
the researchers have investigated the linear interaction of Lamb wave with multiple
delaminations [25–27]. However, the effect of nonlinear interaction with multiple
delaminations is yet to be explored. To this end, the fundamentals and approach for
utilizing the nonlinear interaction of Lamb waves for multiple delaminations in the
composite laminate are discussed in this work.
The present study investigates the nonlinear interactions of the fundamental anti-
symmetric Lamb wave mode (A0) with multiple through-width delaminations placed
at various interlaminar locations in a unidirectional glass fiber-reinforced polymer
Effects of Delamination on Higher Harmonics Generation … 413
2 Numerical Simulation
should be kept unmerged. The numerical model requires fine mesh discretization for
wave propagation of all the frequencies. In order to achieve reasonable accuracy, the
mesh density should have at least 10 element nodes per wavelength. Higher harmonics
are also expected to be generated during wave interaction with delamination. The
minimum wavelength in the waveguide will determine mesh size for simulations.
Semi-analytical finite element (SAFE) approach is utilized to achieve the phase
velocities along with their corresponding wavelengths at different frequencies. Phase
velocity dispersion curve obtained using the SAFE approach is shown in Fig. 2. A
mesh size of 0.25 mm has been chosen to accommodate up to third harmonics of the
fundamental frequency.
The optimum time step is also desired for the accuracy of the wave propagation in
numerical simulations. It can be quantified from respective frequencies’ phase veloc-
ities and the mesh size using Courant–Friedrichs–Lewy (CFL) condition. The very
small time step can also cost computational time. So the tradeoff between accuracy
and the computational cost was taken into consideration. 0.1 µs time step obtained
from the CFL condition has been used for numerical simulations. Contact delamina-
tion between two layers is modeled using contact pairs in these FE simulations. The
friction forces between the delamination faces were neglected in these simulations.
CONTA172 and TARGET169 surface to surface contact elements have been used as
contact and target surfaces to the delamination surfaces. There are many algorithms
available to detect contact interaction and generate contact forces to avoid any pene-
tration. The augmented Lagrange method has been used in this investigation for
the contact interaction as this method effectively has better penetration control and
achieves the contact equilibrium in a very less number of iterations.
A tone burst Gaussian window modulated force signal is applied over the top and
the bottom surface of laminate in in-plane direction at the actuation points in the
opposite sides to generate fundamental anti-symmetric (A0) Lamb mode, as shown
in Fig. 1. The total simulation time is 0.5 ms with the actuation time of 0.1 ms. The
excitation signal of 60 kHz frequency in time and frequency domain is shown in
Fig. 3. The selection of excitation frequency is based on higher harmonics generated
as no two peaks should merge into each other.
Numerical simulations have become one of the important tools, for characterizing
and evaluating delamination of laminated composite structures, through nonlinearity
trends, calculated from higher harmonics generated due to CAN. The selection of
416 A. S. Gangwar et al.
A2 + A3 + A4 + . . .
NLI = (1)
A1
FFT plots that the higher harmonics strength decreases when the location of second
delamination changes from 2nd–3rd layer to 4th–5th layer. Figure 5 shows the FFT
response for SD (delamination placed between two layers are 100% overlapped) at
different interfacial locations experimentally.
The experiments were carried out for stacked delaminations. Two composite lami-
nates having eight plies with [0]8 orientation and 450 mm long with a thickness of
4 mm are fabricated with glass/epoxy prepregs using the hand layup technique. Teflon
tape of thickness 15 µm and 12 mm wide is being used to create delamination in
the laminate. Delamination is placed in one plate between 1st–2nd layer and 4th–5th
layer and in other plates between 1st–2nd layer and 2nd–3rd layer. Properties were
calculated as ASTM tests and numerically as illustrated in Table 1. Rectangular-
shaped (25 mm × 8 mm × 1 mm) SP-5H PW transducers adhesively bonded to the
composite structure have been utilized to initiate the A0 Lamb mode. The excitation
signal conforming to Fig. 3 was generated with the help of a signal generator and
amplified by a high-speed bipolar amplifier (NFmade-BA4825). The PW actuator
receives the amplified signal to produce the required Lamb waves. A Polytec VIB-
E-400 scanning laser Doppler vibrometer is employed to sense the displacements.
The post-processing of output data was carried out using the laser. Figure 5 shows
the experimentally obtained results of the FFT response for SD on two composite
plates (one having delamination between 1st and 2nd and in between 2nd and 3rd
layer and other having delamination between 1st and 2nd and in between 4th and 5th
layer). Higher harmonics are observed and diminish when the interfacial location of
the second delamination moves from edge to center.
For investigating the relative strength of higher harmonics for different overlap-
ping conditions, nonlinearity index was calculated for all cases. Figure 8 shows the
trend of nonlinearity index with the degree of overlapping between multiple delam-
inations. It is noticed that the NLI is maximum for the cases of butted delamination
and it decreases as the overlapping between delaminations increases. This trend can
Effects of Delamination on Higher Harmonics Generation … 419
be explained by the various factors that affect the contact nonlinearity [17] in general,
like phase difference among the wave motion in the delamination surfaces. Phase
difference increases with the increase in the difference between the phase velocities
of wave in the delamination region. As the other delamination moves toward the
center, phase difference between sub-laminates also decreases owing to less differ-
ence between the sub-laminate thickness. The nonlinearity trend for different delam-
ination overlapping can also be explained using the same phenomena. Higher over-
lapping will create more sub-laminate region, thus it decreases the relative thickness
and phase difference between the sub-laminates.
In the present study, a numerical analysis based on higher harmonics for the investiga-
tion of multiple delaminations on a laminated composite structure is accomplished.
FE simulation study reveals that higher harmonics are generated due to the CAN
depending on multiple delaminations. Further, the strength of higher harmonics is
found to be dependent on the interfacial location and the degree of overlapping
between delaminations. Some basic experiments are performed, which are in line
with FE results. NLI is found to be maximum for butted delamination cases and the
numerical results show that the nonlinearity increases as the overlap percent between
two delaminations decreases.
Acknowledgements The authors wish to acknowledge the financial support from the Science and
Engineering Research Board (SERB), India under grant agreement No: ECR/2017/001171.
420 A. S. Gangwar et al.
References
23. Joglekar DM, Mitra M (2016) Analysis of flexural wave propagation through beams with
a breathing crack using wavelet spectral finite element method. Mech Syst Signal Process
76:576–591
24. Joglekar DM, Mitra M (2016) Time domain analysis of nonlinear frequency mixing in a slender
beam for localizing a breathing crack. Smart Mater Struct 26(2):025009
25. Chan WS, Chou CJ (1995) Effects of delamination and ply fiber waviness on effective axial
and bending stiffnesses in composite laminates. Compos Struct 30(3):299–306
26. Ng CT, Veidt M (2011) Scattering analysis of fundamental anti-symmetric lamb wave at
delaminations in composite laminates. Aust J Mech Eng 8(2):197–205
27. Nagaraj MH, Carrera E, Petrolo M (2020) Progressive damage analysis of composite laminates
subjected to low-velocity impact using 2d layer-wise structural models. Int J Non Linear Mech
127:103591
28. Shen Y (2017) Numerical investigation of nonlinear interactions between multimodal guided
waves and delamination in composite structures. In: Health monitoring of structural and
biological systems. International Society for Optics and Photonics, vol 10170, pp 101701Z
29. Joglekar DM (2020) Analysis of nonlinear frequency mixing in timoshenko beams with a
breathing crack using wavelet spectral finite element method. J Sound Vibrat 115532
Effect of Ply Orientation on Nonlinear
Lamb Wave Delamination Interactions
in GFRP Composite Laminates
Abstract Laminated composite plates are widely used in many engineering applica-
tions. In order to ensure safety and reliability of such laminated composite structure,
various nondestructive testing (NDT) techniques are used. Lately, Lamb waves-based
methods have been explored for their use in NDT of composite laminates, as they
offer an optimum combination of the area scanned and damage sensitivity. Tech-
niques based on nonlinear interactions of Lamb waves with damage further increase
the damage sensitivity when compared with their linear counterparts. However, most
of these studies are concerned with unidirectional laminates. Although the effect of
ply orientation on the linear interaction has been reported recently, the effect of ply
orientation on the nonlinear interaction of the Lamb waves with breathing behaviour
of delamination has not been addressed yet. In this regard, this study reports the effect
of ply orientation on the generation of higher harmonics. This work deals with the
nonlinear interactions of A0 Lamb mode with delamination in an eight-layer GFRP
composite plate, numerically. Further, a thorough parametric study is performed for
assessing the impact of ply orientations and inter-laminar position of the delamina-
tion on the nonlinearity index. The investigation also suggests that the ply orientation
plays an important role in determining the wave speed, the trend which proportionally
matches with the nonlinearity index. Therefore, this analysis can prove to be useful
in arriving at a strategy to characterize and localize the delamination in laminated
composite plate, with various layup sequences.
1 Introduction
The variation of ply orientations in the laminate is also being widely used as it has
better mechanical properties than uniformed composite laminate. So the wave-based
nonlinear damage inspection techniques need to consider the effect of ply orientation
on the contact nonlinearity as well. Researchers have studied the linear wave based on
the delamination characterization for the laminated composite having different ply
orientations [10]. The nonlinearity pattern and the factors affecting higher harmonics
generation and its strength have been studied only on uniformed composite laminate
[1, 6, 7]. However, a similar study for nonlinear characterization based on CAN for
different ply-oriented laminates is yet to be done. In this work, a delamination type
defect in a laminated composite structure with different ply layup configurations is
discussed, where nonlinearities are induced due to the contact acoustic nonlinearity
(CAN).
In this paper, numerical simulations were conducted to investigate the nonlinear
interaction of Lamb wave with delamination in eight-layered GFRP-laminated
composite plate having different ply layup configuration and interlaminar location
of delamination. These simulations were conducted with the variation of 45° lami-
nated composite plate. This study suggests that higher harmonics are generated due
to clapping between delamination interfaces and dependent on ply layup and inter-
laminar location of delamination. The remaining part of the paper is organized as
follows. First, the numerical simulations of the laminated composite specimen with
delamination, modelled using contact elements, are presented in Sect. 2. Then the
simulation results are discussed in Sect. 3, and finally, the work is concluded in
Sect. 4.
All the numerical simulations of wave propagation in the composite laminate have
been carried out in ANSYS. A 2D model was prepared instead of 3D model to avoid
high computational time. The plate of eight laminates each of 0.5 mm thickness
and 450 mm length has been modelled using PLANE82 elements. All the material
properties used in the FE modelling is shown in Table 1. Actuation and sensing
locations were selected in such a way so as to avoid the generation of additional
lobes from the reflections in the response signal. A0 waves have been generated in
all the simulations by applying force signals at top and bottom locations as shown
in Fig. 1.
All the laminates were meshed such that at least 10 nodes per wavelength exists.
60 kHz signal was used in this study and higher harmonics are also expected
For all simulations, 60 kHz tone-burst signal is chosen as the input excitation
frequency signal. The total time of simulation is 0.5 ms in which the excitation
time of 60 kHz signal is 0.1 ms. The input signal in time and frequency domain is
shown in Fig. 2. To study the effect of ply orientation of composite laminate on the
Effect of Ply Orientation on Nonlinear Lamb Wave … 427
1.2 1.0
Normalised amplitude
0.8
0.8
Normalised force
0.4
0.6
0.0
0.4
-0.4
0.2
-0.8
-1.2 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0 60 120 180 240 300
Time (ms) Frequency (kHz)
(a) (b)
Fig. 2 60 kHz tone-burst excitation input signal in a time and b frequency domain
contact nonlinearity, three ply orientation angles (0°, 45° and 90°) have been consid-
ered. Delamination divides the laminate into two sub-laminate regions (i.e., thicker
and thinner sub-laminates). The parametric study has been carried out in the numer-
ical analysis in which the ply angles of thinner sub-laminate have been altered. The
ply angles of thicker sub-laminate are kept 0°. In this parametric study, the location
of delamination was also changed with the variation of ply angles.
1.2 0.20
Normalised amplitude
0.8
Normalised force
0.15
0.4
0.10
0.0
0.05
-0.4
-0.8 0.00
0.0 0.1 0.2 0.3 0.4 0.5 0 60 120 180 240 300
Time (ms) Frequency (kHz)
(a) (b)
A2 + A3 + A4 + . . .
NLI = (1)
A1
The effect of ply orientation on the generation of higher harmonics has been studied
by comparing FFT plots on those laminate where only ply angles are changing while
keeping the interlaminar location of delamination the same. Figure 4 shows the FFT
response of the laminate having delamination in between the 1st and 2nd layer with
the variation of ply angles in thinner sub-laminate. It can also be observed from
Fig. 5 that the strength of higher harmonics are increasing when the ply angle of
thinner sub-laminate increases from 0° to 90° with the increment of 45°. Further,
the delamination location is changed to 2–3 layers. The thinner sub-laminate has
Effect of Ply Orientation on Nonlinear Lamb Wave … 429
0.10
00
0.05
Normalised amplitude
0.00
0 60 120 180 240 300
0.10
45o
0.05
0.00
0 60 120 180 240 300
0.10
90o
0.05
0.00
0 60 120 180 240 300
Frequency (kHz)
Fig. 4 FFT response for the laminate having delamination at 1–2 layer with different ply angles in
thinner sub-laminate
thinner sub-laminate
0.09
0.08
0.07
0 45 90
Ply orientation (degree)
two plies and the response has been obtained for each and every combination of ply
angles. Delamination located at 2–3 layers has been divided into two cases. In the
first case, both plies would have the same ply angle and in the second case, there
will be a combination of two out of the three ply angles. Figures 5 and 6 show the
FFT responses of the laminate with delamination at 2–3 layers with the same and
different ply angles in the thinner sub-laminate, respectively. The nonlinearity index
plots for FFT data of Figs. 8 & 9 respectively.
430 Y. Agrawal et al.
To further understand the effect of ply angles on the higher harmonics generation, the
wave properties affected by the ply orientations are studied. Different ply orientation
leads to variation in the phase speed of the wave. When the wave passes through
the delamination, it splits into two sub-laminates. The phase speeds of these waves
are different due to different laminate thickness and ply orientations. At the starting
edge of the delamination, there would not be any phase difference between the upper
and the lower laminate. However, as both waves propagate, it starts gaining phase
difference. These out-of-plane phase differences at the delamination surfaces cause
the collision between both sub-laminates and thus lead to the generation of higher
harmonics. Figure 7 depicts the displacement profile of the top and bottom sub-
laminate at the leading and trailing edge of the delamination. There is almost no
phase difference between both sub-laminates at the leading edge. However, there
is a significant phase difference at the trailing edge. Nonlinearity will be higher
for the case where the phase speed difference in sub-laminate is higher. The main
cause of the phase speed variation is the laminate thickness. However, when the
nonlinearity comparison is made, delamination location is kept the same and only
the ply orientation is varied. So, the phase speed of the laminate is varying with the
ply orientation. Table 2 refers to the phase speeds for different ply orientations and
ply thickness for 60 kHz signal. It can be observed from Table 2 that phase speed
decreases as the ply angle increases.
0.10
00/00
0.05
Normalised amplitude
0.00
0 60 120 180 240 300
0.10
45 /45o
o
0.05
0.00
0 60 120 180 240 300
0.10
90o/90o
0.05
0.00
0 60 120 180 240 300
Frequency (kHz)
Fig. 6 FFT response for the laminate having delamination at 2-3 layer with same ply angles in
thinner sub-laminate
Effect of Ply Orientation on Nonlinear Lamb Wave … 431
1.0
00/450
0.5
Normalised amplitude
0.0
0 60 120 180 240 300
1.0
0o/90o
0.5
0.0
0 60 120 180 240 300
1.0
45o/90o
0.5
0.0
0 60 120 180 240 300
Frequency (kHz)
Fig. 7 FFT response for the laminate having delamination at 2-3 layer with different ply angles in
thinner sub-laminate
0.14
0.13
0.12
0 45 90
Ply orientation (degree)
FE simulation-based numerical analysis has been carried out in this paper to investi-
gate the effect of ply orientations on the higher harmonics generation due to delam-
ination. Ply angles has been varied from 0° to 90° in the thinner sub-laminate. It
has been found that the nonlinearity increases with the ply angles. It has also been
discussed and concluded that nonlinearity induced from the delamination is mainly
432 Y. Agrawal et al.
Nonlinearity Index
thinner sub-laminate 1.0
0.7
0.4
0.1
45/90 0/90 0/45
Ply orientation (degree)
Normalised displacement
Normalised displacement
2 2
Leading edge 60 kHz Trailing edge 60 kHz
1 1
0 0
-1 -1
-2 -2
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (ms) Time (ms)
(a) (b)
Fig. 10 Displacement response of 60 kHz signal on the delamination surfaces at (a) leading and
(b) trailing edge of delamination
dependent on the phase difference and the laminate thickness. Further, it was also
observed that phase difference is directly dependent on the difference in the phase
speed of sub-laminates at the delamination region. This study concluded that the
nonlinearity patterns can be useful to devise a method to identify and characterize
the delamination as part of nondestructive evaluation.
Effect of Ply Orientation on Nonlinear Lamb Wave … 433
Acknowledgements The authors gratefully acknowledge the financial aid provided by the Science
and Engineering Research Board (SERB) under the Department of Science and Technology (DST),
Government of India under the Grant No.: ECR/2017/001171.
References
1. Agrawal Y, Gangwar AS, Joglekar DM (2021) Mixing of lamb waves at delamination defect
in a unidirectional gfrp laminate. In: Advances in Non-destructive evaluation. Springer, pp
317–328
2. Bovsunovsky A, Surace C (2015) Non-linearities in the vibrations of elastic structures with a
closing crack: a state of the art review. Mech Syst Signal Process 62:129–148
3. Broda D, Staszewski W, Martowicz A, Uhl T, Silberschmidt V (2014) Modelling of nonlinear
crack–wave interactions for damage detection based on ultrasound—a review. J Sound Vib
333(4):1097–1118
4. Della CN, Shu D, Zhao Y (2005) Vibration of composite beams with two overlapping
delaminations. Acta Mech Sin 21(1):47–55
5. Feng B, Ribeiro AL, Ramos HG (2018) Interaction of lamb waves with the edges of a delamina-
tion in cfrp composites and a reference-free localization method for delamination. Measurement
122:424–431
6. Gangwar AS, Agrawal Y, Joglekar DM (2020) Nonlinear frequency mixing in gfrp lami-
nate with a breathing delamination. In: European workshop on structural health monitoring.
Springer, pp 825–836
7. Gangwar AS, Agrawal Y, Joglekar D (2021) Nonlinear interactions of lamb waves with a
delamination in composite laminates. J Nondestr Eval Diagnost Prognost Eng Syst 4(3):031008
8. Giurgiutiu V (2015) Structural health monitoring of aerospace composites. Academic Press
9. Guo N, Cawley P (1993) The interaction of lamb waves with delaminations in composite
laminates. J Acoust Soc Am 94(4):2240–2246
10. Gupta S, Rajagopal P (2018) Effect of ply orientation and through-thickness position of delam-
ination on the reflection of fundamental symmetric s0 lamb mode in gfrp composite plate
structures. Ultrasonics 90:109–119
11. Hayashi T, Endoh S (2000) Calculation and visualization of lamb wave motion. Ultrasonics
38(1–8):770–773
434 Y. Agrawal et al.
12. Jhang KY (2009) Nonlinear ultrasonic techniques for nondestructive assessment of micro
damage in material: a review. Int J Precis Eng Manuf 10(1):123–135
13. Joglekar D (2020) Analysis of nonlinear frequency mixing in timoshenko beams with a
breathing crack using wavelet spectral finite element method. J Sound Vibrat 115532
14. Joglekar D, Mitra M (2015) Nonlinear analysis of flexural wave propagation through 1d
waveguides with a breathing crack. J Sound Vib 344:242–257
15. Joglekar D, Mitra M (2016) Analysis of flexural wave propagation through beams with a
breathing crack using wavelet spectral finite element method. Mech Syst Signal Process 76:576–
591
16. Joglekar D, Mitra M (2016) Time domain analysis of nonlinear frequency mixing in a slender
beam for localizing a breathing crack. Smart Mater Struct 26(2):025009
17. Khurana A, Kumar A, Raut SK, Sharma AK, Joglekar M (2021) Effect of viscoelasticity on the
nonlinear dynamic behavior of dielectric elastomer minimum energy structures. Int J Solids
Struct 208:141–153
18. Khurana A, Sharma AK, Joglekar M (2021) Nonlinear oscillations of electrically driven
aniso-visco-hyperelastic dielectric elastomer minimum energy structures. Nonlinear Dyn
104(3):1991–2013
19. Mitra M, Gopalakrishnan S (2006) Wavelet based spectral finite element modelling and
detection of de-lamination in composite beams. Proc Royal Soc A Math Phys Eng Sci
462(2070):1721–1740
20. Munian RK, Mahapatra DR, Gopalakrishnan S (2018) Lamb wave interaction with composite
delamination. Compos Struct 206:484–498
21. Ramadas C, Balasubramaniam K, Joshi M, Krishnamurthy C (2009) Interaction of the primary
anti-symmetric lamb mode (ao) with symmetric delaminations: numerical and experimental
studies. Smart Mater Struct 18(8):085011
22. Ramadas C, Balasubramaniam K, Joshi M, Krishnamurthy C (2010) Interaction of guided
lamb waves with an asymmetrically located delamination in a laminated composite plate.
Smart Mater Struct 19(6):065009
23. Rauter N, Lammering R, Kühnrich T (2016) On the detection of fatigue damage in composites
by use of second harmonic guided waves. Comp Struct 152:247–258
24. Rose JL (2014) Ultrasonic guided waves in solid media. Cambridge university press (2014)
25. Schaal C, Samajder H, Baid H, Mal A (2015) Rayleigh to lamb wave conversion at a
delamination-like crack. J Sound Vib 353:150–163
26. Shen Y (2017) Numerical investigation of nonlinear interactions between multimodal guided
waves and delamination in composite structures. In: Health monitoring of structural and
biological systems. International Society for Optics and Photonics, vol 10170, pp 101701Z
27. Soleimanpour R, Ng CT (2017) Locating delaminations in laminated composite beams using
nonlinear guided waves. Eng Struct 131:207–219
28. Yelve NP, Mitra M, Mujumdar P (2015) Detection of stiffener disbonding in a stiffened
aluminium panel using nonlinear lamb wave. Appl Acoust 89:267–272
29. Yelve NP, Mitra M, Mujumdar P, Ramadas C (2016) A hybrid method based upon nonlinear
lamb wave response for locating a delamination in composite laminates. Ultrasonics 70:12–17