development of accelerated damage detection techniques using...

박 사 학 위 논 문

Ph. D. Dissertation

비접촉식 레이저 스캐닝을 통한

고속 손상 진단 기법 개발

Development of Accelerated Damage Detection Techniques

Using Noncontact Laser Ultrasonic Scanning

2017

박 병 진 (朴兵津 Park, Byeongjin)

한 국 과 학 기 술 원

Korea Advanced Institute of Science and Technology

박 사 학 위 논 문

Ph.D. Dissertation



Development of Accelerated Damage Detection Techniques

using Noncontact Laser Ultrasonic Scanning

2017

박 병 진

한 국 과 학 기 술 원

건설 및 환경공학과



박 병 진

위 논문은 한국과학기술원 박사학위논문으로

학위논문 심사위원회의 심사를 통과하였음

2016년 11월 18일

심사위원장 손훈훈 (인)

심 사 위 원 박기환 (인)

심 사 위 원 안윤규 (인)

심 사 위 원 정형조 (인)

심 사 위 원 홍정욱 (인)

Development of

Accelerated Damage Detection Techniques

Using Noncontact Laser Ultrasonic Scanning

Byeongjin Park

Advisor: Hoon Sohn

A dissertation submitted to the faculty of

Korea Advanced Institute of Science and Technology in

partial fulfillment of the requirements for the degree of

Doctor of Philosophy in Civil and Environmental Engineering

Daejeon, Korea

November 18, 2016

Approved by

Hoon Sohn

Professor of Civil and Environmental Engineering

The study was conducted in accordance with Code of Research Ethics1).

1) Declaration of Ethical Conduct in Research: I, as a graduate student of Korea Advanced Institute of Science and

Technology, hereby declare that I have not committed any act that may damage the credibility of my research. This

includes, but is not limited to, falsification, thesis written by someone else, distortion of research findings, and plagiarism.

I confirm that my dissertation contains honest conclusions based on my own careful research under the guidance of my

advisor.

초록

본 연구에서는 비접촉식 레이저 초음파 스캐닝을 통한 고속 손상 진단 기법들을 개발하고자

했다. 개발된 고속 손상 진단 기법들은 (1) 통합된 이진 검색과 압축 센싱, (2) 고정 거리 스캐닝

기반 이진 검색 기법이다. 이 기법들은 검사 영역의 모든 지점에 대해 스캐닝을 수행하는 것이

아니라 손상 검출에 최적화된 지점들에 대해서만 스캐닝을 수행하기에 기존 기법에 비해 필요한

스캐닝 횟수 및 검사 시간을 크게 줄일 수 있다. 필요한 스캐닝 지점들은 이진 검색 알고리즘에

기반하여 결정되며, 이 때 요구되는 스캐닝 횟수는 검사 영역의 크기를 𝑁 ∙ 𝑀이라 할 때 최악의

경우에도 4log2 𝑁 ⋅ log2 𝑀 에 불과하다.

이 때 이진 검색 알고리즘은 반사와 투과 등 초음파와 손상 간의 상호작용을 파악하여 작동한다.

본 연구에서는 시간 영역에서 측정된 레이저 초음파 신호를 공간 초음파 영역으로 변환하여

해석함으로써 시간 영역에서 해석하기 어려웠던 초음파-손상간 상호작용 정보를 더욱 효과적으로

확인할 수 있도록 했다. 이 때 변환 과정에서 기존에 널리 사용되던 정합추적 방식 대신 기저추적

방식을 적용함으로써 더욱 정확한 변환 및 손상 감지가 가능하다.

이와 같이 개발된 기법들의 성능과 적용성을 수치 시뮬레이션과 실험실 규모의 실험을 통해

확인했다. 다양한 구조물에 대해서 개발된 기법들이 효과적으로 손상 진단을 수행함을 확인하였다.

핵심낱말 비접촉식 검사, 레이저 초음파, 손상 감지, 기저추적, 이진검색

Abstract

This dissertation aims to accelerate laser based nondestructive testing with the developed techniques inspired by

binary search. Two accelerated damage detection techniques are developed in this dissertation. Inspired by binary

search, damage is localized and visualized with reduced scanning points and scanning time. The number of scan-

ning points that is necessary for damage localization and visualization is dramatically reduced from 𝑁 ∙ 𝑀 to

4log2 𝑁 ⋅ log2 𝑀 even for the worst case scenario. 𝑁 and 𝑀 represent the number of equally spaced scanning

points in the x and y directions, respectively, which are required to obtain full-field wave propagation images of

the target inspection region.

The binary search algorithm is based on examining the interactions between the ultrasonic waves and damage,

such as reflections and transmissions. A time-domain ultrasonic response is transformed into a spatial ultrasonic

domain to better identify these interactions. Instead of a traditional matching pursuit approach which tries to solve

an ℓ0 minimization problem using a greedy sequential algorithm, basis pursuit approach solves an ℓ1 minimiza-

tion problem to transform a time-domain response with a better resolution.

The feasibility of the developed damage detection techniques is validated in both numerical and experimental

ways. The developed techniques can reduce the number of scanning points and scanning time over 90%.

Keywords Noncontact inspection, Laser ultrasonics, Damage detection, Basis pursuit, Binary search

DCE

20127031

박병진. 비접촉식 레이저 초음파 스캐닝을 통한 고속 손상

진단 기법 개발. 건설 및 환경공학과. 2017년. 115+vii 쪽.

지도교수: 손훈. (영문 논문)

Byeongjin Park. Development of accelerated damage detection tech-

niques using noncontact laser ultrasonics. Department of Civil and

Environmental Engineering. 2017. 115+vii pages. Advisor: Hoon

Sohn. (Text in English)

i

Contents

Contents ························································································································ i

List of Figures and Tables ································································································ v

Chapter 1. Introduction 1

1.1 Motivation ·········································································································· 1

1.2 Current noncontact damage detection techniques ····························································· 3

1.2.1 Eddy current testing ······················································································· 3

1.2.2 Radiographic testing ······················································································ 4

1.2.3 Shearography ······························································································· 5

1.2.4 Infrared thermography ··················································································· 6

1.2.5 Terahertz technique ························································································ 7

1.3 Research objectives, scopes and uniqueness ··································································· 8

1.4 Dissertation organization ······················································································· 10

Chapter 2. Working Principles of Noncontact Laser Ultrasonic Scanning System and Literature Review on

Damage Detection Techniques 11

2.1 Laser ultrasonic generation ···················································································· 11

2.1.1 Working principle of laser ultrasonic wave generation ·············································· 11

2.1.2 Numerical modeling of laser ultrasonic wave generation ·········································· 13

2.2 Laser ultrasonic measurements ················································································ 16

2.3 Laser ultrasonic scanning system ············································································· 20

2.3.1 Macroscopic scanning system ··········································································· 20

2.3.2 Microscopic scanning system ········································································· 22

2.4 Scanning strategies ······························································································ 24

2.4.1 Fully noncontact scanning strategies ··································································· 24

2.4.2 Partially noncontact scanning with a contact transducer ··········································· 25

2.5. Literature review on damage detection techniques using noncontact laser ultrasonic scanning ····· 27

2.5.1 Wavefield imaging techniques ·········································································· 27

2.5.2 Linear ultrasonic techniques ··········································································· 28

ii

2.5.3 Nonlinear ultrasonic techniques ······································································· 29

2.6 Chapter summary ······························································································· 31

Chapter 3. Accelerated Damage Detection with Reduced Laser Ultrasonic Scanning Points 32

3.1 Combined binary search and compressed sensing ·························································· 32

3.1.1 Binary search for damage localization ································································· 32

3.1.2 Compressed sensing for damage quantification ····················································· 35

3.2 Binary search with fixed pitch-catch distance scanning ···················································· 36

3.3 Chapter summary ······························································································· 39

Chapter 4. Representation of Laser Ultrasonic Responses using a Spatial Ultrasonic Dictionary 40

4.1 Sparse representation of a signal ·············································································· 40

4.2 Interactions between ultrasonic waves and damage ························································ 44

4.2.1 Ultrasonic wave propagation ············································································ 44

4.2.2 Wave-damage interactions ············································································· 46

4.3 Representation of time domain ultrasonic responses using a spatial ultrasonic dictionary ············ 48

4.4 Estimation of spatial domain ultrasonic responses using a spatial ultrasonic dictionary ·············· 52

4.5 Chapter summary ······························································································· 53

Chapter 5. Numerical Validation of the Developed Technique 54

5.1 Simulation description ·························································································· 54

5.2 Spatial ultrasonic transformation ·············································································· 56

5.2.1 Spatial ultrasonic transformation for the excitation point in front of the damage and behind the

damage ·········································································································· 56

5.2.2 Transition point identification ··········································································· 57

5.2.3 Sensitivity of the spatial ultrasonic transformation to damages with various width, thickness and

depth ··············································································································· 58

5.3 Damage detection results using the developed techniques ················································· 60

5.3.1 Combined binary search and compressed sensing ···················································· 60

5.3.2 Binary search with fixed pitch-catch distance scanning ··········································· 62

5.4 Discussion on the practical applicability of the developed techniques ··································· 64

5.5 Chapter summary ······························································································· 66

iii

Chapter 6. Experimental Validation of the Developed Technique using a Macroscopic Laser Ultrasonic

Scanning System 67

6.1 Hardware setup ·································································································· 67

6.2 Hidden notch detection in an aluminum plate ······························································· 68

6.2.1 Experimental setup ····················································································· 68

6.2.2 Notch detection using combined binary search and compressed sensing ·························· 69

6.2.3 Notch detection using binary search with fixed pitch-catch distance scanning ··················· 71

6.3 Fatigue crack detection in an aluminum plate ······························································· 73


6.3.2 Crack detection using combined binary search and compressed sensing ·························· 74

6.3.3 Crack detection using binary search with fixed pitch-catch distance scanning ··················· 76

6.4 Delamination detection in a carbon fiber reinforced plastic (CFRP) plate ······························ 78


6.4.2 Delamination detection using combined binary search and compressed sensing ················· 79

6.4.3 Delamination detection using binary search with fixed pitch-catch distance scanning ·········· 81

6.5 Delamination detection in a 10 kW glass fiber reinforced plastic (GFRP) wind turbine blade ······· 83


6.5.2 Delamination detection using combined binary search and compressed sensing ················· 84

6.5.3 Delamination detection using binary search with fixed pitch-catch distance scanning ·········· 86

6.6 Chapter summary ······························································································· 88

Chapter 7. Experimental Validation of the Developed Technique using a Microscopic Laser Ultrasonic

Scanning System 89

7.1 Hardware setup ·································································································· 89

7.2 Defect detection in an epoxy molding compound (EMC) bar ············································· 90


7.2.2 Defect detection using binary search with fixed pitch-catch distance scanning ·················· 91

7.3 Crack detection in a semiconductor chip ····································································· 93

7.3.1 Experimental setup ··························································· 93

7.3.2 Crack detection using binary search with fixed pitch-catch distance scanning ··················· 94

7.4 Chapter summary ······························································································· 96

iv

Chapter 8. Concluding Remarks 97

8.1 Summary of the works ························································································· 97

8.2 Uniqueness of the works ······················································································· 99

8.3 Future works ···································································································· 100

Bibliography 101

Acknowledgements in Korean 108

Curriculum Vitae 109

v

List of Figures and Tables

Figure 1.1. Representative examples where effective damage detection is needed ································· 2

Figure 1.2. Working principle of eddy current testing ································································· 3

Figure 1.3. Working principle of radiographic testing and detection of a defect inside ··························· 4

Figure 1.4. Working principle of a shearography ······································································· 5

Figure 1.5. Schematic diagram of (a) a halogen thermography and (b) a laser thermography ···················· 6

Figure 2.1. Working principle of laser ultrasonic generation through thermal expansion ······················· 11

Figure 2.2. Examples of laser ablation ················································································· 13

Figure 2.3. Schematic of the 2D numerical model of laser ultrasonic wave generation ························· 14

Figure 2.4. Comparison between the simulated and experimentally obtained laser ultrasonic response ······ 15

Figure 2.5. Electrical field magnitude of the reflected laser beam, called as a measurement beam, and the reference

beam without any reflection ······························································································ 16

Figure 2.6. Schematic diagram of a typical LDV and its working principle ······································ 17

Figure 2.7. Schematic of the macroscopic noncontact laser ultrasonic scanning system ························ 20

Figure 2.8. Main components for the macroscopic laser ultrasonic scanning system ···························· 21

Figure 2.9. Schematic of the microscopic noncontact laser ultrasonic scanning system ························ 22

Figure 2.10. Main components for the microscopic laser ultrasonic scanning system ··························· 23

Figure 2.11. Fully noncontact scanning strategies ···································································· 24

Figure 2.12. Partially noncontact scanning strategies ································································ 26

Figure 3.1. Schematic flow of binary search for transition line detection ········································· 32

Figure 3.2. Schematic flow of compressed sensing for damage quantification. ·································· 35

Figure 3.3. Schematic flow of the binary search with fixed pitch-catch distance scanning ····················· 36

Figure 4.1. An example to find the sparsest representation ························································· 41

Figure 4.2. Graphical representation of the ℓ1 minimization ······················································ 42

Figure 4.3. Schematic diagram of a uniform bar and corresponding axial wave propagation ·················· 44

Figure 4.4. Schematic diagram of a uniform beam and corresponding flexural wave propagation ············ 45

Figure 4.5. Schematic representation of wave reflection and transmission at an interface ······················ 46

vi

Figure 4.6. Comparison of ultrasonic responses (a) when a damage is located between the excitation point and

sensing point and (b) outside the direct wave propagation path between the excitation point and sensing point 48

Figure 4.7. Comparison of the transformed time-domain ultrasonic responses generated at (a) intact, (b) in front

of the damage, and (c) behind the damage ············································································· 51

Figure 5.1. Schematic of a numerical model for numerical validations of the developed techniques ········· 54

Figure 5.2. Determination of the location of the laser excitation point with respect to damage using the proposed

spatial ultrasonic transformation ························································································· 56

Figure 5.3. Comparison of ultrasonic responses measured from an intact plate and a damage plate in the spatial

ultrasonic domain ·········································································································· 56

Figure 5.4. Variation of the basis pursuit (BP) index value as the excitation point moves from a 10 mm distance

from the sensing point to 50 mm ························································································ 57

Figure 5.5. Schematic of an aluminum plate model with a notch damage with thickness t, width w, and tip depth d 58

Figure 5.6. Sensitivity of spatial ultrasonic transformation for damage detection ································ 59

Figure 5.7. Binary search for a transition line detection ····························································· 60

Figure 5.8. Damage quantification using compressed sensing ······················································ 60

Figure 5.9. Comparison of the original full-field and the reconstructed wave propagation snapshots that corre-

spond to the dashed box in Figure 5.8 (a) ·············································································· 61

Figure 5.10. Binary search with fixed pitch-catch distance scanning ·············································· 62

Figure 5.11. Comparison of the number of (a) scanning points p and (b) the reduction rate R using the conventional

full-field wave propagation imaging, combined binary search and compressed sensing, and binary search with

fixed pitch-catch distance scanning ····················································································· 64

Figure 6.1. Overview of the macroscopic laser ultrasonic scanning system ······································ 67

Figure 6.2. An aluminum plate with a hidden notch ································································· 68


Figure 6.4. Damage quantification via compressed sensing ························································· 70

Figure 6.5. Reconstructed wave propagation at different time steps ··············································· 70

Figure 6.6. Binary search with fixed pitch-catch distance scanning ················································ 71

Figure 6.7. An aluminum plate with a fatigue crack ································································· 73


Figure 6.9. Damage quantification via compressed sensing ························································· 75

vii

Figure 6.10. Reconstructed wave propagation at different time steps ·············································· 75


Figure 6.12. A CFRP plate with a delamination ······································································ 78

Figure 6.13. Binary search for a transition line detection ···························································· 79

Figure 6.14. Damage quantification via compressed sensing ······················································· 80


Figure 6.16. Delamination visualized using conventional full-field wave propagation imaging ··············· 81


Figure 6.18. A GFRP wind turbine blade with a delamination ····················································· 83

Figure 6.19. Binary search for a transition line detection ···························································· 84

Figure 6.20. Damage quantification via compressed sensing ······················································· 85



Figure 7.1. Overview of the microscopic laser ultrasonic scanning system ······································· 89

Figure 7.2. An epoxy molding compound (EMC) bar with a void ················································· 90


Figure 7.4. A semiconductor chip with a crack ······································································· 93


Table 2.1. Material properties of aluminum used in this numerical simulation ··································· 15

Table 5.1. Material properties of aluminum used in this numerical simulation ··································· 55

- 1 -

Chapter 1. Introduction

1.1. Motivation

In these days, there is an increasing concern on structural safety issues. This concern is especially enlarged in

Korea as they have many deteriorated civil infrastructures which were built in 1970s, their economic golden age.

The collapse of Sungsoo Grand Bridge (Seoul, Korea, 1994) and the collapse of Sampoong Department Store

(Seoul, Korea, 1995) are representative shocks engraved to Koreans’ memory. These structural safety issues have

been continuously reported including the collapse of Mauna Resort (Gyeongju, Korea, 2014), the breakage of a

cable of the Seohae Grand Bridge (Pyeongtaek, Korea, 2015), and the collapse of a 2 MW wind turbine (Taebaek,

Korea, 2016). There also have been catastrophic accidents in oversea countries including the derailment of an ICE

train (Eschede, Germany, 1998) and the fall down of the I-35W Bridge (Minnesota, USA, 2007). As these failures

result not only economic loss but also number of victims and fatalities, various acts and regulations on structural

monitoring and inspection are established to prevent these accidents.

Meanwhile, there are also increasing issues in industrial quality assurance. Damage detection is im-

portant not only for civil structures but also industrial products to increase their reliability and reduce correspond-

ing expenses. For example, automobile industries are interested in debonding detection in adhesive joints of their

automobiles. Composite materials and low weight metals e.g. aluminum are widely used for their newest models

to reduce their weight and increase their fuel efficiency. As it is hard to weld these materials, most of the joints

are bonded using adhesives. But it is quite challenging to assure whether the adhesives are uniformly spread or

not, as the debondings are hidden inside the joints. Another examples are micro structures including semiconduc-

tor chips and printed circuit boards (PCBs). They are very complex products and even a small crack in a circuit

or a soldered joint may result a failure of the whole structure.

Therefore, there is a large demand for effective damage detection techniques. Followings are several

requirements of an effective damage detection techniques suggested from field engineers. First, early and sensitive

detection is very important as most structural failures are resulted from micro damages. Campbell [1] reported

that up to 90% of failures of in-service metallic structures are results of fatigue cracks. Second, noncontact meas-

urement is required to avoid the interference to operating structures. It is also important for micro structures with

their small structural dimensions. Fast detection and short inspection time is especially emphasized in quality

assurances to avoid interruptions and delays in production process.

- 2 -

(a) (b)

(c) (d)

Figure 1.1. Representative examples where effective damage detection is needed. In-service structural failures:

(a) the failure of a 2 MW wind turbine in Korea and (b) the derailment of an ICE train in Germany. Industrial

products: (c) Adhesive joints in automobiles and (d) cracked welded joints in printed circuit boards (PCBs).

Visual inspection by experts has been the most common damage detection technique until recently.

However, this is labor intensive, time consuming and unreliable as it is performed by trained engineers. Structural

monitoring techniques using permanently installed sensors are the next movement, which can continuously collect

data, automatically analyze them and report damage detection results [2-4]. But these monitoring approaches have

several limitations in field applications: (1) Labor intensive installation of multiple sensors is required to achieve

a high spatial resolution in damage detection; (2) corresponding cables should be connected for power and data

transmission, and the length of these cables is more than several kilometers for long span bridges; (3) these sensors

and cables themselves are becoming the weakest links in the structure and requiring additional inspection for them;

and (4) for small structures, it is impossible to install sensors as they affect structural properties of the target

structures. Hence an effective noncontact damage detection technique is what society and industry need.

- 3 -

1.2. Current noncontact damage detection techniques

Currently a number of noncontact damage detection techniques are available. Their operations are not fully auto-

mated yet and usually in a form of nondestructive testing (NDT). NDT requires human interruption, and trained

engineers inspect targets with portable devices and analyze the collected data. In this subchapter, five noncontact

damage detection techniques are introduced and their working principles, pros, and cons are discussed. Though

ultrasonic testing (UT) is one of the most popular NDT techniques, it is not considered in this subchapter as most

of commercial UT require couplant between targets and devices such as water [5].

1.2.1. Eddy current testing

Figure 1.2. Working principle of eddy current testing [6].

In electromagnetism, electric fields and magnetic fields are not independent but mutually dependent. Simply said,

an electric current generates a magnetic field and a magnetic field generates another electric current. Any change

in electric or magnetic fields creates new electromagnetic fields to oppose this change. If an electric current flows

through the sending coil in Figure 1.2, this will generate the primary magnetic field. Then the secondary magnetic

field is created in the target specimen to revert the primary field. An eddy current is generated in the target as a

resultant of this secondary magnetic field. As the eddy current and the secondary magnetic field changes with the

electromagnetic conductivity of the target, it is possible to detect a damage by measuring changes in the secondary

magnetic field [7; 8] in the receiving coil.

- 4 -

While this technique is very sensitive to small surface defects, it has following limitations. First of all,

this technique is only applicable to conductive metal structures as it measures electromagnetic fields. This limits

its applications to a quite narrow range. Secondly, a demagnetization process is required to remove original mag-

netic fields of the target structure, which are sources of false alarms in data interpretation process. Finally, the

working distance of a typical eddy current measurement system is usually limited up to several centimeters for

effective measurements of the secondary magnetic field.

1.2.2. Radiographic testing

Figure 1.3. Working principle of radiographic testing and detection of a defect inside.

Radiographic testing technique is one of the most popular and powerful techniques in NDT field [9; 10]. Damage

can be detected by interpreting a radiograph, a photographic image produced on an X-ray film by the passage of

X-ray electromagnetic radiation. X-ray has a wavelength ranging from 0.01 to 10 nanometers, which can transmit

most of materials. Their transmission ratio is dependent on how thick and dense the target structure is. Let us

assume that a structure is exposed by X-ray radiation as shown in Figure 1.3. The left and right region of the film

is more exposed by the radiation with less thickness. Less radiation exposure is shown in the central region, but

the defect region is more exposed with the changed density in this region. The defect can be identified through

the aforementioned photograph interpretation process.

- 5 -

Radiographic testing is applicable to most of materials. It is quite easy to interpret the measured data

from the radiographs. Precise and accurate damage quantification is also available. This technique is especially

widely used in medical fields with these powerful advantages. However, even with these advantages, its field

application is quite limited with its biggest drawback. As it uses radiographic radiation, it may harm users’ health

with a high possibility. Radiographic testing is usually performed in specially designed facilities to minimize

radiation exposure, and its portable field application is very challenging.

1.2.3. Shearography

(a)

(b) (c)

Figure 1.4. (a) Working principle of a shearography [11]; (b) Laboratory setup for shearography [12]; and (c)

A commercial shearography testing device.

Shearography is an optical NDT technique which can measure surface deformation of the target object. This

technique measures out-of-plane deformation of structures according to given loads [11; 13; 14]. Figure 1.4 rep-

resents a typical setup for shearography testing. Laser rays from a light source are reflected from two points 1 and

2 with a separation of δx. The reflected rays meet in an interferometer, and the resulted interference pattern is

recorded by a camera. If there is any surface deformation and corresponding change in δx, the interference pattern

- 6 -

of two rays will be changed as shown in the right graphs of Figure 1.4 (a). As the damaged region and the intact

region show different deformation behavior, damage can be distinguished by subtracting the non-deformed image

from the new image taken with the external load.

Shearography has many advantages including large scan area, 1/3 of inspection time compared to UT

C scan, and sub-micrometer deformation measurement sensitivity. But there are still several limitations in its

operation. (1) With its high sensitivity, it is extremely vulnerable to environmental variation and ambient vibration

which may cause false alarms; (2) it is challenging to evaluate micro cracks, which do not make large changes in

local elastic properties; (3) loading process may harm the target structure and stimulate micro cracks; and (4) as

this is an optical testing technique, it requires a high reflectivity of the target surface. The target surface need to

have a roughness of more than one wavelength of the light.

1.2.4. Infrared thermography

(a) (b)

Figure 1.5. Schematic diagram of (a) a halogen thermography and (b) a laser thermography.

Infrared (IR) thermography technique visualizes and quantifies damage by detecting changes in heat transfer char-

acteristics near defect. As thermal images can be taken over large area using an IR camera, this technique has

been popular for instantaneous nondestructive testing of mechanical systems [15]. Traditionally, an IR camera

measures heat transfer from an ambient heat source (e.g. sun light) or a passive heat source such as a lamp. Then

this technique identifies possible damage from regions with abnormal temperature in comparison with the rest of

the area [16]. However, this approach is susceptible to ambient noise as the temperature difference between the

- 7 -

intact and damage region is usually less than 1 °C. This leads to the development of active thermography tech-

niques [17], where controllable heat sources such as halogen lamps and acoustic transducers are used [18]. A laser

thermography technique is relatively new in the field of active thermography [19]. It uses a high power laser as a

heat source. Compared to traditional heat sources such as halogen lamps, its biggest advantage is its highly local-

ized nature of heating spot and resulted high spatial resolution in damage detection.

But there are several limitations in IR thermography applications. As it is an optical testing technique

like shearography, its applications are confined to materials with high emissivity. Subsurface defect detection is

challenging, and surrounding random heat source leads false alarms. Even with these limitations, its fast inspection

speed with areal inspection and relatively long working distance up to 1-2 meters make this technique attractive.

1.2.5. Terahertz technique

Terahertz (THz) testing is another emerging testing field in last few years [20-22]. This is not limited to a certain

technique but indicating various techniques using THz range electric fields and radiations, wavelength ranging

from 0.1 millimeters to 1 millimeter. It is positioning on the longer wavelength side of the visible rays, while

radiographic radiations such as X-rays locates on the shorter wavelength side of them. Various techniques of

radiographic techniques, such as C-Scan [23], tomography [24], time-of-flight evaluation [25], and 3D imaging

[26], are also applicable with THz waves with some modifications and optimizations.

As its wavelength is much longer than X-rays, it is intrinsically safe technique. It is applicable to both

metal and composite materials, and showing high sensitivity especially in depth measurement. Its applications are

ranging from spectroscopy, material characterization, security and medical fields. While its applications are ex-

panding and this technique is getting higher interest, still complex and extremely expensive optical setup is re-

quired at this point.

- 8 -

1.3. Research objectives, scopes and uniqueness

This dissertation mainly focuses on accelerated damage detection techniques using noncontact laser ultrasonic

scanning. Laser ultrasonic techniques are emerging as an attractive noncontact testing technique in the field of

nondestructive testing with their noncontact nature and high spatial resolution. However, its practicality is limited

because scanning at a high spatial resolution demands a prohibitively long scanning time. The author aimed to

accelerate laser based nondestructive testing with the proposed novel search techniques. The objectives, scopes

and uniqueness of this research can be summarized as follows.

1) Accelerated damage detection algorithm

Conventional techniques have required a huge number of scanning points to achieve a high spatial resolution in

damage detection. These techniques quantify damage by repeatedly checking whether each inspection point is

intact or damage. In this dissertation, two accelerated damage detection techniques are proposed. Only a reduced

number of scanning points is required for damage detection, and these scanning points are optimally selected. The

optimal selection process is inspired by the binary search algorithm, also known as the half-interval search, a

famous data searching algorithm in computer science field.

The first technique, combined binary search and compressed sensing, approximates damage location

from sparse measurements and visualizes wave propagation around the damage. Damage is quantified by analyz-

ing the visualized wavefield. Here the excitation laser beam or the sensing laser beam is fixed while the other

scans the inspection region. The other technique, binary search with fixed pitch-catch distance scanning, directly

quantifies the damage by moving both the excitation and the sensing laser beams while keeping the distance

between them. They reduce the required number of measurements from 𝑁 ∙ 𝑀 to 4log2 𝑁 ⋅ log2 𝑀 in the worst

case, where 𝑁 and 𝑀 represent the number of equally spaced scanning points in the x and y directions, respec-

tively, which are required to obtain full-field wave propagation images of the target inspection region.

2) Improved damage locating with a spatial ultrasonic transformation

Conventionally, damage existence is identified by subtracting the baseline data that corresponds to the pristine

condition of a specimen from the measured signal in the time domain. But the measured signal significantly de-

viates from its pristine condition even when the damage is located outside the direct wave propagation path due

- 9 -

to reflections from the damage. The effect of the damage outside the direct wave path may be alleviated by time

truncation. However, the selection of the proper truncation time point can be challenging due to the multi-mode

and dispersive nature of Lamb waves.

In the proposed techniques, a time domain laser ultrasonic response is transformed into a spatial domain

using a spatial ultrasonic dictionary. Interactions between laser generated ultrasonic waves and a damage, e.g.

reflections, and transmissions can be better identified in the spatial ultrasonic domain. Damage presence in the

ultrasonic path is identified by tracking an incident basis peak in the spatial ultrasonic domain. As this basis peak

is shifted only if ultrasonic waves propagate through a damage, this is a sensitive feature in damage detection.

This transformation is obtained by basis pursuit approach. Instead of a traditional matching pursuit

approach which tries to solve an ℓ0 minimization problem using a greedy sequential algorithm, this basis pursuit

approach solves an ℓ1 minimization problem to find the solution with a better resolution.

3) Numerical and experimental validation of the developed techniques

The performance of the developed damage detection techniques is validated using a numerical simulation and

experiments. For the numerical validation, an aluminum plate with a crack is modeled using COMSOL Mul-

tiphysics. For experimental validations, a macroscopic laser ultrasonic scanning system and a microscopic laser

ultrasonic scanning system is used for macroscopic structures and microscopic structures, respectively. An alu-

minum plate with a notch, an aluminum plate with a fatigue crack, a carbon fiber reinforced plastic (CFRP) plate

with a delamination, and a 10 kW glass fiber reinforced plastic (GFRP) wind turbine blade are tested with the

macroscopic scanning system. An epoxy molding compound (EMC) bar with a defect and a semiconductor chip

with a crack are tested with the microscopic scanning system.

- 10 -

1.4. Dissertation organization

This dissertation is organized as follows with eight chapters.

Chapter 1: The motivation of this research is presented, and the research objectives, scopes, and uniqueness are

provided. Working principles and limitations of current noncontact damage detection techniques are also dis-

cussed.

Chapter 2: Working principles of noncontact laser ultrasonic generation and measurement are explained. A laser

ultrasonic scanning system for this research and various scanning strategies using the system are provided. Previ-

ous damage detection techniques using noncontact laser ultrasonic scanning are also reviewed.

Chapter 3: Two accelerated damage detection techniques are proposed in this chapter. Their working procedures

are provided, and their inspection speeds are compared with the one of the conventional full field wave propaga-

tion imaging technique.

Chapter 4: This chapter discusses how a time domain ultrasonic response can be represented with a spatial ultra-

sonic dictionary. Signal transformation using basis pursuit approach is explained. Interactions between ultrasonic

waves and a damage, and corresponding representation using a spatial ultrasonic dictionary are discussed.

Chapter 5: The developed techniques are validated through numerical simulations. An aluminum plate with a

crack is modeled for the validations. The results from each technique, their practical applicability, and their in-

spection speed are discussed.

Chapter 6: The developed techniques are experimentally validated with a macroscopic scanning system. An alu-

minum plate with a notch, an aluminum plate with a fatigue crack, a carbon fiber reinforced plastic (CFRP) plate

with a delamination, and a 10 kW glass fiber reinforced plastic (GFRP) wind turbine blade are used for the vali-

dations. The results from each technique and their inspection speed are discussed.

Chapter 7: The developed techniques are experimentally validated with a microscopic scanning system. An

epoxy molding compound (EMC) bar with a defect and a semiconductor chip with a crack are used for the vali-

dations. The results using the fixed pitch-catch distance scanning scheme and their inspection speed are discussed.

Chapter 8: This chapter concludes this dissertation with summary of the works, their uniqueness, and future

works to be conducted.

- 11 -

Chapter 2. Working Principles of a Noncontact Laser Ultrasonic Scanning

System and Literature Review on Damage Detection Techniques

2.1. Laser ultrasonic generation

2.1.1. Working principle of laser ultrasonic wave generation

When a laser beam is emitted onto an infinitesimal solid surface, three phenomena may occur: absorption, reflec-

tion, and transmission of laser radiation. Absorbed laser energy leads localized heating of the infinitesimal area,

which causes thermoelastic expansion of the material and generation of ultrasonic waves [27; 28]. This process is

represented in Figure 2.1. Here, the direction of the principal stress is parallel to the surface as there is no external

constraints normal to the surface.

Figure 2.1. Working principle of laser ultrasonic generation through thermal expansion.

Most of our target structures for nondestructive testing are thick enough to neglect transmission of laser

radiation. In the case of absorption, the laser intensity is exponentially attenuated as it penetrates into the sample.

The laser intensity 𝐼(𝑧) can be expressed as:

𝐼(𝑧) = 𝐼(0)𝑒−𝛾𝑧 (2.1)

- 12 -

where 𝛾 is the absorption coefficient and 𝑧 is the penetration depth. Then a skin depth, where most of the laser

energy takes place, is defined as the penetration depth where the laser intensity falls to 1/e of its initial intensity

and expressed as follows [29]:

𝛿 =1

√𝜋𝜎𝜇𝑟𝜇0𝜐= √

𝜆

𝜋𝜎𝜇𝑟𝜇0𝑐 (2.2)

where 𝜎, 𝜇𝑟 , 𝜇0, 𝜐, 𝑐, and 𝜆 denote the conductivity of the material, the relative permeability of the material, the

relative permeability of free space, the frequency of the laser beam, the speed of light in free space, and the

wavelength of the laser beam, respectively. Please note that the skin depth is proportional to the square root of the

laser beam wavelength.

With the law of conservation of energy, the sum of the absorbed intensity and the reflected intensity

should be identical to its initial intensity. The reflected intensity needs to be minimized to maximize laser absorp-

tion and increase generated ultrasonic amplitude. Classical electromagnetic theory approximates the reflectivity

𝑅 of common metals in the visible light range as follows [29].

𝑅 = 1 −4

𝜇0𝜎𝑐𝛿. (2.3)

From Equation (2.2), it can be found that 𝑅 may decrease with shorter wavelength laser beams. This implies that

a short wavelength laser is more effective in ultrasonic wave generation.

Then the radiated area expands with the absorbed laser energy and leads ultrasonic wave generation.

The expanded volume of the material, 𝛿𝑉, can be represented as [27]:

𝛿𝑉 =3𝛼

𝜌𝐶𝛿𝐸, (2.4)

where 𝛼, 𝜌, 𝐶, and 𝛿𝐸 indicate the coefficient of linear expansion, the density and the specific heat of the mate-

rial, and the absorbed energy respectively. Then the corresponding displacement at a distance 𝑟, caused by this

volumetric thermal expansion is given by [27]:

𝑢𝑟 =𝛼

2𝜋𝜌𝐶𝑐1𝑟

𝛿𝐸

𝛿𝑡, (2.5)

where 𝑐1 is the compression velocity. To generate an ultrasonic wave amplitude of 10-10 m at 100 mm distance

on an aluminum sample, a laser power of 0.1 MW is required.

- 13 -

To make this thermoelastic induced displacement, or wave, to be in the ultrasonic range, the energy

absorption duration should be very short as a pulse. Let us assume the pulse laser beam to be Gaussian, then its

Fourier transform falls to its half maximum at a frequency of

𝑓𝐻𝑀 =0.1874

𝜏. (2.6)

Here 𝜏 denotes the full width at half maximum (FWHM) of this pulse, or simply called as a pulse width [27].

This pulse width should be less than 20 ns to ensure enough energy in ultrasonic range under 𝑓𝐻𝑀 = 10 MHz.

Typical pulse lasers for ultrasonic wave generation have a pulse width of 20 ~ 30 ns and a laser power of 1 MW.

The amplitude of the generated ultrasonic waves increases with the peak power of a pulse laser. How-

ever, it may harm the target surface if the laser beam power per unit area exceeds certain threshold level. This

damage takes a form of surface melting, vaporization, ablation and plasma formation [30]. Ablation is the most

common form and this removes out a part of the irradiated surface [31]. Figure 2.2 (a) shows its typical behavior

[32] and (b) shows an ablation resulted crater with 40 μm width and 15 μm depth [33]. These phenomena are very

useful for welding, cutting, and treating the target structure, but they should be avoided in the case of nondestruc-

tive testing. Therefore, parameters for the laser ultrasonic generation, such as the peak power, the pulse width and

the beam size, should be carefully designed to avoid these surface damages.

(a) (b)

Figure 2.2. Examples of laser ablation: (a) Snapshots of ablation simulation [32] and (b) scanning electron

microscope (SEM) photograph of a laser ablation on a zircon sample [33].

2.1.2. Numerical modeling of laser ultrasonic wave generation

Ultrasonic wave propagation induced by a pulse laser is modelled and analyzed with a commercial finite element

software COMSOL Multiphysics. To represent phenomenon occurring after the incident of a pulse laser beam

onto a plate, the model is divided into two parts, thermal wave region and ultrasonic wave region (Figure 2.3 (a)).

In thermal wave region, one thermal degree of freedom and two mechanical degrees of freedom are solved for

- 14 -

each node of elements. In ultrasonic wave region, only two mechanical degrees of freedom are solved for each

node [34]. For more details, please refer to the following reference [35].

(a)

(b)

Figure 2.3. Schematic of the 2D numerical model of laser ultrasonic wave generation: (a) Geometry of the

simulated specimen and size of thermal wave / ultrasonic wave region, and (b) finite element meshes for ther-

mal wave / ultrasonic wave region.

The entire model is meshed with triangular elements and a multi-scale element length is used to reduce

computation burden. Before performing this analysis, appropriate types of physics need to be selected. In this

study, a multi-physics solved in time domain analysis is considered. First, thermoelasticity is used in the thermal

wave region for the generation of ultrasonic wave due to the thermal stress induced by the pulse laser. The size of

the thermal wave region is determined based on the thermal diffusion length of the target material. Aluminum is

considered in this numerical simulation with corresponding material properties given in Table 2.1. Then thermal

wave region size is set as 0.11 mm depth and 1.11 mm radius, while its element size is 5 μm maximum for precise

modeling of ultrasonic wave generation process.

Interaction between solid and acoustic is then used for solving the ultrasonic wave propagation within

the material. For ultrasonic wave region, the element size is 0.5 mm maximum as less precision is required in

- 15 -

ultrasonic wave propagation modeling. The simulated specimen is with 75 mm radius and 3 mm thickness, as

shown in Figure 2.3 (a). Corresponding triangular element meshes are represented in Figure 2.3. (b).

Table 2.1. Material properties of aluminum used in this numerical simulation.

Density

𝜌 (kg/m3)

Young’s

modulus

𝐸 (GPa)

Poisson’s

ratio

𝜈

Coefficient

of thermal

expansion

𝛼𝑡 (K-1)

Thermal

conductivity

𝐾 (W/(m·K))

Heat

capacity at

constant

pressure

𝐶𝑝

(J/(kg·K))

Reflection

coefficient

𝑅

2700 68.9 0.33 2.34×10-5 170 900 0.95

The designed model is validated in Figure 2.4. It compares the out-of-plane velocities numerically and

experimentally acquired at the sensing point of the aluminum plate. Though the velocity signals obtained from

the simulations are pre-processed with a low-pass filter (cut-off frequency: 350 kHz), as the LDV used in the

experiments holds the frequency components only up to 350 kHz, the higher frequency components cannot be

totally removed. Furthermore, due to the energy loss in the experiment, the laser power illuminated on the speci-

men is most likely not the same for the simulations and experiments. Therefore, the velocities shown in Figure 10

are amplitude normalized. However, the velocities acquired from the simulation and experiment show similar

waveforms.

Figure 2.4. Comparison between the simulated and experimentally obtained laser ultrasonic response. Though

they are not perfectly identical, they show similar waveforms.

- 16 -

2.2. Laser ultrasonic measurement

Laser based displacement measurement devices have been developed from 19th century in a form of laser inter-

ferometers including Michelson, Fabry-Perot and Mach-Zender interferometer [27]. A Michelson interferometer

is the very first form of the interferometers, which measures displacement of the target by measuring the phase

difference between the reference beam and the target reflected beam. The biggest limitation of this interferometer

is that a long travel path of the laser beam is required to have a high resolution in phase difference measurement

[36]. A Fabry-Perot interferometer uses a device called etalon to increase the travel path effectively. As an etalon

is consisted of two parallel mirrors, the travel path is increased by producing a large number of reflection inside

the etalon [37]. A Mach-Zender interferometer is usually used for transparent samples, which interferes laser

beams transmitted a test sample and a compensating sample [38].

A laser Doppler vibrometer (LDV) is a relatively recent form of laser based measurement devices,

introduced in 1960s [39]. This measures velocity of the target while aforementioned interferometers measure

displacement. In ultrasonic application, velocity measurement is usually more effective as this has higher ampli-

tude in comparison with the displacement.

𝑢(𝑡) = 𝐴sin(2𝜋𝑓𝑡 + 𝜑), (2.7)

�̇�(𝑡) = 𝐴𝜔cos(2𝜋𝑓𝑡 + 𝜑). (2.8)

Here, 𝑢 and �̇� are the ultrasonic displacement and the ultrasonic velocity, 𝐴 is the ultrasonic displacement

amplitude, 𝑓 is its frequency, and 𝜑 is its phase information, respectively.

Figure 2.5. Electrical field magnitude of the reflected laser beam (blue dashed line), called as a measurement

beam, and the reference beam (red solid line) without any reflection. The frequency of the measurement beam

is changed with the velocity of the target surface.

- 17 -

A LDV uses Doppler effect to measure the velocity of the target structure. Doppler effect implies that

the frequency of a laser beam is shifted when this beam is reflected from a moving surface. This effect is repre-

sented in Figure 2.5. As the target moves outward from the LDV, the wavelength of the reflected laser beam

lengthens and its frequency becomes lower. On the other hand, shorter wavelength and higher frequency laser

beam is measured when the target moves toward to the LDV. The amount of this frequency shift Δ𝑓 is expressed

as:

Δ𝑓 = 𝑓𝑟 − 𝑓𝑖 = (1 −𝑣

𝑐)

2

𝑓𝑖 − 𝑓𝑖 ≈ −2𝑣

𝑐𝑓𝑖 = −

2𝑣

𝜆𝑖

, (2.9)

where 𝑓𝑟 , 𝑓𝑖 , 𝜆𝑖 , 𝑣, and 𝑐 denotes the frequency of the reflected laser beam, the initial frequency of the laser beam,

the initial wavelength of the laser beam, the velocity of the target surface, and the speed of light. Therefore, the

velocity of the target can be obtained by measuring the frequency shift of the reflected laser beam. This should be

noted that the measured 𝑣 is the out-of-plane velocity which is parallel to the incident laser beam. Recently, a

3D LDV is also commercially available which uses three photodetectors to measure in-plane velocity components

[40].

Figure 2.6. Schematic diagram of a typical LDV and its working principle. This calculates the velocity of the

target by measuring Doppler shift induced phase difference 𝜑(𝑡). PBS: Partial beam splitter; AOM: Acousto-

optic modulator; PD: Photodetector.

Figure 2.6 represents a schematic composition of a typical LDV [41]. First, a continuous wave (CW)

laser beam is irradiated from a laser source. Class 2 He-Ne laser with 633 nm (red light) wavelength is the most

typical laser source for LDVs in these days. This laser beam is divided into two beams from the partial beam

- 18 -

splitter (PBS) #1. One half, the reference laser beam, is frequency shifted with 𝑓𝐴𝑂𝑀 by passing through an

acousto-optic modulator (AOM). Its typical value is 40 MHz for commercial LDVs, and its role will be explained

afterwards. The other half, the measurement laser beam, is reflected back from the target surface and interfered

with the reference beam. Before the interference, the electric field of each beam can be represented as:

𝐸𝑅 = 𝐴𝑅 𝑐𝑜𝑠(2𝜋(𝑓𝑖 + 𝑓𝐴𝑂𝑀)𝑡)), (2.10)

𝐸𝑀 = 𝐴𝑀 𝑐𝑜𝑠(2𝜋𝑓𝑟𝑡) = 𝐴𝑀𝑐𝑜 𝑠(2𝜋(𝑓𝑖 + 𝛥𝑓)𝑡), (2.11)

where 𝐸𝑅 and 𝐸𝑀 represents the electric field of the reference beam and the measurement beam, and 𝐴𝑅 and

𝐴𝑀 represents corresponding electric field amplitude. They interfere at two photodetectors (PDs) and results fol-

lowing interference signals are measured.

𝐸 = 𝐸𝑅 + 𝐸𝑀 = 𝐴𝑅 𝑐𝑜𝑠(2𝜋(𝑓𝑖 + 𝑓𝐴𝑂𝑀)𝑡)) + 𝐴𝑀𝑐𝑜 𝑠(2𝜋(𝑓𝑖 + 𝛥𝑓)𝑡), (2.12)

𝑆 = 𝐼𝑅0 + 𝐼𝑀0 + 2√𝐼𝑅0𝐼𝑀0 cos(2𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡), (2.13)

where 𝐸 and 𝑆 represents the electric field and the corresponding measured intensity of the interfered beam,

and 𝐼𝑅0 and 𝐼𝑀0 represents intensity amplitudes of the reference beam and the measurement beam. By applying

a high pass filter to Equation (2.13), only fluctuating term is obtained.

𝑆0 = 2√𝐼𝑅0𝐼𝑀0 cos(2𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡) = 𝐼0 cos(2𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡). (2.14)

Here, if there is no AOM, positive and negative frequency shift cannot be distinguished as the cosine is an even

function. The moving direction of the target surface is only obtained with a presence of 𝑓𝐴𝑂𝑀. In the first PD,

sin(2𝜋𝑓𝐴𝑂𝑀𝑡) is multiplied to Equation (2.14) to modulate the measured interference signal.

𝑆𝐼 = 𝑆0 ∙ sin(2𝜋𝑓𝐴𝑂𝑀𝑡) = 𝐼0 cos(2𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡) ∙ sin(2𝜋𝑓𝐴𝑂𝑀𝑡)

=𝐼0

2[sin(4𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡) + sin(2𝜋∆𝑓𝑡)].

(2.15)

In the other PD, cos(2𝜋𝑓𝐴𝑂𝑀𝑡) is multiplied to Equation (2.14) to modulate it.

𝑆𝑄 = 𝑆0 ∙ cos(2𝜋𝑓𝐴𝑂𝑀𝑡) = 𝐼0 cos(2𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡) ∙ cos(2𝜋𝑓𝐴𝑂𝑀𝑡)

=𝐼0

2[cos(4𝜋𝑓𝐴𝑂𝑀𝑡 − 2𝜋∆𝑓𝑡) + cos(2𝜋∆𝑓𝑡)].

(2.16)

- 19 -

As 𝑓𝐴𝑂𝑀 is much higher than ∆𝑓, the following two signals, called as I and Q signal respectively, can be obtained

by applying low pass filters. This process is called IQ demodulation [41].

𝑈𝐼 =𝐼0

2sin(2𝜋∆𝑓𝑡), (2.17)

𝑈𝑄 =𝐼0

2cos(2𝜋∆𝑓𝑡). (2.18)

Then the Doppler shift induced phase difference 𝜑(𝑡) can be calculated by taking an arctangent of the ratio

between I and Q signal.

𝜑(𝑡) = tan−1𝑈𝐼

𝑈𝑄

= 2𝜋∆𝑓𝑡. (2.19)

From Equations (2.9) and (2.19), the velocity of the target is measured.

𝑣 = −𝜆

2Δ𝑓 = −

𝜆

4𝜋

𝑑𝜑

𝑑𝑡. (2.20)

However, it should be noted that a LDV has its fundamental limitations in measurements. The first

limitation is its resolution limit [41]. High intensity of the reference and the measurement beams lead strong I/Q

signals and effective frequency shift measurements. But this generates shot noise, which is proportional to the

power of the laser source. Therefore, most of LDV manufacturers try to minimize shot noise while yielding enough

laser intensity for interference. The optimized point usually reaches at 1 mW, which is a common value for many

commercial LDVs.

The second one is the intermode beating [27]. As most of laser sources are not ideal single mode laser,

undesirable signals are generated from beatings between each laser mode. They fluctuate with the path length

difference between the reference and the measurement laser beam. Hence the optimal working distance of a LDV

is usually given in a repeating form of 𝑎 + 𝑛𝑏, where 𝑎 and 𝑏 are specific coefficients.

The last one, and usually considered as the most important one, is its speckle noise [42]. A speckle

pattern is an irregular distribution of scattered laser beam from a rough surface, and this reduces the amplitude of

the returning measurement beam. Thus, the incident angle of the laser beam should be carefully controlled to

maximize the returned beam intensity. Often a special surface treatment is necessary to improve the reflectivity

of the target surface if the surface is highly rough and having a poor reflectivity.

- 20 -

2.3. Laser ultrasonic scanning system

2.3.1. Macroscopic scanning system

Figure 2.7. Schematic of the macroscopic noncontact laser ultrasonic scanning system. All units are synchro-

nized and controlled by a personal computer in the control unit.

In this subchapter, detailed specifications of a macroscopic laser ultrasonic scanning system used in this disserta-

tion are given. This system is appropriate to scan macroscopic specimens and structures. Figure 2.7 shows an

overall schematic of the laser ultrasonic scanning system composed of an excitation, a sensing and a control unit.

All units are synchronized and controlled by a personal computer in the control unit. Main components of the

system are represented in Figure 2.8.

The excitation unit is consisted of a diode pumped Q-switched Nd:YAG pulse laser (Quantel Centu-

rion+), a galvanometer (Scanlab hurrySCAN 20), and a focal lens. The width of the laser pulse emitted from this

laser is 12 ns, while it emits laser pulses up to 100 times per a second. This pulse laser can radiate both 1064 nm

and 532 nm laser, where their peak pulse energy and power are 50 mJ / 4.2 MW and 25 mJ / 2.1 MW respectively.

532 nm laser is used for this dissertation to have a higher absorption as predicted in Equation (2.3). The galva-

nometer has a maximum rotating speed of 70 rad/s, an angular resolution of 11.6 μrad, and an allowable scanning

angle of ±0.38 rad. The initial laser beam size is 3 mm in diameter, but it is focused to 0.5 mm diameter at 1 m

focal length with a specially designed focal lens. It achieves high spatial resolution in laser scanning.

- 21 -

The sensing unit is basically a commercial scanning LDV (Polytec PSV-400), which includes a LDV,

a built-in galvanometer and an auto focal lens. The maximum measurable out-of-plane velocity is in the range of

0.01 μm/s to 10 m/s, and can be adjusted by a user to maximize its vertical resolution. A Class 2 He-Ne laser with

633 nm wavelength is used as the laser source, which has a maximum power of 1 mW. Its optimal working

distance is 99 + 204𝑛 mm to avoid intermode beating phenomenon. The galvanometer has a maximum rotating

speed of 35 rad/s and an allowable scanning angle of ±0.35 rad. The auto focal lens can focus the laser beam to

0.5 mm diameter in its working range from 0.3 m to 10 m.

The control unit includes two parts: a velocity decoder and a personal computer. The velocity decoder

is a device which can extract velocity information from vibrometer measured signals and digitize it. It has a ve-

locity sensitivity of 20 mm/s/V with a maximum measurable frequency of 350 kHz, while its maximum sampling

frequency and digitizing resolution is 2.56 MHz and 14 bit respectively. The personal computer includes a main

control software of the whole system, which is programmed by LabVIEW and Visual Basic. This software can

detect a trigger signal from the excitation unit, order measurements and collect measured data from the sensing

unit, and adjust galvanometers to control excitation and sensing laser beams to the desired target points.

(a) (b)

(c) (d)

Figure 2.8. Main components for the macroscopic laser ultrasonic scanning system: (a) A pulse laser in the

excitation unit, (b) a galvanometer with a focal lens in the excitation unit, (c) a LDV in the sensing unit, and

(d) the control unit.

- 22 -

2.3.2. Microscopic scanning system

Figure 2.9. Schematic of the microscopic noncontact laser ultrasonic scanning system. All units are synchro-

nized and controlled by a personal computer in the control unit.

As discussed in Chapter 2.2, the performance of LDV is highly dependent on the intensity of the returned laser

beam. One easy way of increasing this intensity is to capture the returned laser beams at a close distance from the

reflected surface. Figure 2.9 shows an overall schematic of the microscopic laser ultrasonic scanning system com-

posed of an excitation, a sensing and a control unit. All units are synchronized and controlled by a personal com-

puter in the control unit. Main components of the system are represented in Figure 2.10.

The excitation unit is consisted of a diode pumped Q-switched Nd:YAG pulse laser (Quantel Centu-

rion+), and a fiber guide (Qbic Laser Ltd.). The specification of the pulse laser is identical to the one of the

macroscopic scanning system. Instead of a galvanometer, a fiber guide system guides the pulse laser beam to the

target structure. The laser beam is focused to 0.5 mm diameter at 90 mm focal length with the fiber guide system.

The sensing unit is basically a commercial microscopic LDV (Polytec MSA-100-3D), which includes

a LDV and a scanning stage. The maximum measurable out-of-plane velocity is in the range of 0.01 μm/s to 10

m/s, and can be adjusted by a user to maximize its vertical resolution. A Class 2M He-Ne laser with 532 nm

- 23 -

wavelength is used as the laser source, which has a maximum power of 5 mW. Its optimal working distance is

48.9 mm.

The control unit includes two parts: a velocity decoder and a personal computer. The velocity decoder

is a device which can extract velocity information from vibrometer measured signals and digitize it. It has a ve-

locity sensitivity of 500 mm/s/V with a maximum measurable frequency of 12.5 MHz, while its maximum sam-

pling frequency and digitizing resolution is 31.25 MHz and 14 bit respectively. The personal computer includes

a main control software of the whole system, which is programmed by LabVIEW and Visual Basic. This software

can detect a trigger signal from the excitation unit, order measurements and collect measured data from the sensing

unit, and adjust the scanning stage to move the target structure to the desired location.

(a) (b)

(c) (d)

Figure 2.10. Main components for the microscopic laser ultrasonic scanning system: (a) A pulse laser in the

excitation unit, (b) a fiber guide in the excitation unit, (c) a LDV in the sensing unit, and (d) the control unit.

- 24 -

2.4. Scanning strategies

In this subchapter, various scanning strategies using the laser ultrasonic scanning system are discussed. These

scanning strategies can be divided into two categories according to its use of contact transducers, (1) fully non-

contact scanning and (2) partially noncontact scanning. The proposed scanning system is fully noncontact, but

can work in the form of partially noncontact ones according to some needs.

2.4.1. Fully noncontact scanning strategies

(a)

(b)

(c)

Figure 2.11. Fully noncontact scanning strategies: (a) Fixed excitation and scanning sensing (FE/SS), (b) fixed

sensing and scanning excitation (FS/SE), and (c) fixed pitch-catch distance scanning (FDS).

For fully noncontact scanning, three different scanning strategies are represented in Figure 2.11. The first one

generates ultrasonic waves at a fixed point, and measures the corresponding responses at various points. This one

is called as fixed excitation and scanning sensing (FE/SS) strategy. The second one is working in the opposite

way to FE/SS. Here ultrasonic waves are consequently generated at various points and corresponding responses

are measured at a fixed sensing point. This is called as fixed sensing and scanning excitation (FS/SE) strategy. As

- 25 -

they fix the sensing point or the excitation point, they are referred to as fixed origin strategies. The last one controls

both lasers to keep the distance between the ultrasonic generation and measurement points, and referred to as the

fixed pitch-catch distance scanning strategy (FDS).

FE/SS and FS/SE strategies are generally used to measure and visualize ultrasonic waves propagation

in the scanned region. Theoretically, FE/SS and FS/SE give identical results according to the linear reciprocity of

ultrasonic waves [43]. This principle implies that a response 𝐻𝐴𝐵, corresponding to the excitation from an arbi-

trary location A and measured at another location B, is identical to 𝐻𝐵𝐴, measured at A with an identical excitation

from B, with an assumption of a linear system. In other words, the impulse response function of the path A to B,

𝑓(𝑡), is identical to the one of the path B to A, 𝑔(𝑡). But FS/SE is preferred with following practical limitations

of FE/SS [44]. When a LDV scans large area in FS/SE, it is challenging to assure same level of sensitivity for

every sensing point. As discussed in Section 2.2., the performance of a LDV is highly dependent on the surface

condition and the incident angle of the LDV beam. This is particularly more serious in curved structures, and

special surface treatments for large area might be required. Another limitation is in surface damage. Repetitive

radiation of high energy pulses to a fixed point may induce ablation on the target surface, as discussed in Section

2.1. And this should be avoided in the point of non-destructive testing.

FDS is a different sort of strategy compared to FE/SS or FS/SE. Its objective is not visualizing wave

propagation. This can be considered as an extended version of pitch-catch techniques in ultrasonic wave analysis.

Although this falls into same limitation with FE/SS in sensitivity variation as the sensing point should be moved,

the FDS may offer higher signal to noise ratio (SNR). Fixed short distance between the excitation and sensing

point assures high amplitude of the measured ultrasonic waves. This also helps in signal processing for damage

detection as small change due to a damage can be easily identified as every measurement should be identical for

an intact homogeneous structure.

2.4.2. Partially noncontact scanning with a contact transducer

For partially noncontact scanning, two different scanning strategies are represented in Figure 2.12. They are ba-

sically substituting a fixed excitation / sensing point into a fixed actuator / sensor. The first one generates ultra-

sonic waves using a fixed actuator, and measures the corresponding responses at various points using a LDV. This

one is called as fixed transducer actuation and scanning sensing (FT/SS) strategy. In the other one, ultrasonic

waves are consequently generated at various points using a pulse laser and corresponding responses are measured

using a fixed sensor. This is fixed transducer sensing and scanning excitation (FT/SE) strategy.

- 26 -

(a)

(b)

Figure 2.12. Partially noncontact scanning strategies: (a) Fixed transducer actuation and scanning sensing

(FT/SS), and (b) Fixed transducer sensing and scanning excitation (FT/SE).

These strategies are generally undesired in practical situations. First, they often require a number of

transducer installations to cover large area. It is especially a daunting task to install these transducers on curved

structures, though several flexible transducers are developed such as a macro fiber composite (MFC). Second,

corresponding cables should be connected to the transducers for power and data transmission. Wireless transduc-

ers have been proposed to solve this issue, but they still require large amounts of electrical energy and batteries.

Finally, transducer installations affect to structural properties and may deteriorate their value. This effect is espe-

cially significant for micro structures including semiconductors and MEMS. The transducer itself is already bigger

than these micro structures and they may not work with installed transducers.

Even with these limitations, partially noncontact scanning strategies are advantageous for some appli-

cations. First, it is possible to generate stronger ultrasonic waves in FT/SS strategy. In noncontact strategies, the

maximum amplitude of ultrasonic wave generation is quite limited to avoid ablation. A fixed actuator such as a

PZT can induce large strain and corresponding ultrasonic wave propagation by applying strong electrical inputs.

Second, ultrasonic generation of an arbitrary waveform is available. Laser-based ultrasonic generation is generally

limited to a pulse excitation though several special techniques have been proposed for narrowband excitations

[45-47]. A fixed actuator can generate arbitrary waveform ultrasonic waves and lead to effective signal filtering

and processing. Finally, laser-based ultrasonic generation and measurement are available only when the desired

locations are in their line-of-sight. For hidden locations where laser beams cannot reach, these transducers can

take a role of ultrasonic generation and sensing.

- 27 -

2.5. Literature review on damage detection techniques using noncontact laser ultrasonic

scanning

2.5.1. Wave propagation imaging techniques

Among various damage detection techniques using noncontact laser ultrasonic scanning, wave propagation imag-

ing techniques are the most popular and widely used techniques in these days. They visualize full-field ultrasonic

wave propagation in a scanned area. The biggest advantage is in intuitive analysis of the measured wavefield in

damage detection. Wave reflections, weak transmissions, and scatters from structural discontinuities and wave

speed change due to material property changes are easily visible in the constructed wavefield images. Most of the

works have used a contact transducer as their ultrasonic wave source and an LDV for scanning, which is defined

as FT/SS strategy in Chapter 2.4.2. Stazewski’s group is one of the earliest pioneers in this field [40; 48-50]. They

visualized ultrasonic wavefield by scanning PZT-generated ultrasonic waves with a LDV, and localized a notch

and a fatigue crack location in an aluminum plate by visualizing reflected ultrasonic waves from damages.

As this approach still requires human interruptions for damage detection, many groups have aimed to

automate the damage detection from the measured wavefield. Sohn et al. [51; 52] applied image processing tech-

niques to visualize delaminations in composite structures using FT/SS strategy. Higher wave energy is measured

near the damage as there are more wave interactions compared to the other locations. By applying a Laplacian

filter, which detects a sudden change in a target image, only damage locations are visualized where sudden in-

crease in wave energy occurs. Park et al. [53] applied this technique with FT/SE strategy for rotating metal blades.

As it is quite challenging to track the fixed sensing point in a rotating structure using a LDV, a fixed sensor is

used for ultrasonic measurements. It is also challenging to generate ultrasonic waves using a pulse laser at desired

locations. However, the ultrasonic generation locations can be estimated using an impact localization technique

[54]. Then the ultrasonic wavefield can be reconstructed from random excitations, and a notch in a steel blade is

visualized. Lee et al. [55] proposed an adjacent wave subtraction method for instantaneous damage detection using

FT/SE strategy. The measured ultrasonic signals show large changes near the damage while adjacent ultrasonic

signals are very similar in the intact region. By subtracting measured adjacent ultrasonic signals, large differences

near a hole in a steel plate and delaminations in a composite aircraft wing can be visualized.

Ruzzene et al. [56] and Michaels et al. [57] proposed a frequency-wavenumber domain analysis to

remove source waves and highlight weak reflection and scatters from damages in aluminum and composite plates.

Since the frequency and the wavenumber of source waves are known in FT/SS strategy, as they are generated

from predefined electric signals using a PZT, only damage reflections and scatters can be extracted by applying

appropriate windows. For delaminations in composite structures, Mesnil et al. [58] measured its instantaneous

- 28 -

wavenumber to estimate the delamination depth. A delamination reduces the effectiveness thickness of the plates

which guided ultrasonic waves propagate and changes their wavenumber. Then it is possible to localize a delam-

ination by measuring the wavenumber at each spatial point, and estimate the delamination depth by measuring

how much the wavenumber change is. While they used FT/SS strategy, similar approach is also done in FT/SE

strategy by Flynn et al. [59]. They visualized milled region in an aluminum plate, wall thinning in a pipe, and a

delamination in a composite aircraft wing from wavenumber changes.

The aforementioned examples used partially noncontact scanning strategies while there are also wave-

field imaging results using fully noncontact scanning strategies. An et al. [44] and Park et al. [60] used a fully

noncontact scanning with FS/SE strategy to visualize wavefield in an aluminum plate and composite structures,

respectively. Here they extract damage-ultrasonic interactions from standing waves generated in the damages.

Local standing waves are generated in the damage as ultrasonic waves are trapped inside from their multiple

reflections at damage boundaries. An et al. [61] also visualized a fatigue crack in an aluminum plate by extracting

and visualizing non propagating Lamb wave modes. These modes are usually ignored as they are evanescent

modes and exponentially decaying from its source. In other words, these modes generated from crack scatters are

localized near the crack and used as an attractive feature for crack identification and localization. Lee et al. [62]

visualized ultrasonic wavefield propagation for an immersed aluminum plate.

However, there is a need for accelerated inspection as full-field wave propagation imaging techniques

require a large number of scanning to cover inspection area with a high spatial resolution. One of the rising solu-

tions is compressed sensing [63]. This is originally proposed in the statistics field and used for image processing.

This assumes a signal, or an image, can be represented in a sparse form in another domain with appropriate bases.

With limited measurements, there is no unique solution in unmeasured data estimation. Then the solution which

maximizes the sparsity in the basis domain is considered to be the most appropriate solution, according to the

assumption. Its performance has been validated for an aluminum [64] and a composite [65] plate with FT/SS

strategy, and damages are detected with less than 50% of data compared to the original measurements. Flynn

proposed another way for accelerated damage detection, which uses a continuous LDV for measurements [66].

Here the target structure is excited with a number of powerful actuators to generate vibration or standing waves

on the structure. Then local changes in vibration behavior due to the damage can be extracted. This shows a rapid

inspection performance for wide region, but powerful actuator installations are required.

2.5.2. Linear ultrasonic techniques

There have also been several early studies to identify damage presence, which are known as linear ultrasonic

techniques. These techniques usually measure reflections, transmissions, and mode conversions of ultrasonic

- 29 -

waves from damage [67; 68]. Early wavefield imaging techniques can be considered as expanded versions of

these linear ultrasonic techniques, but they are out of interest in these days as they don’t have damage quantifica-

tion and localization capabilities. One of the most early studies is done by Tanaka and Izawa [69]. They generated

ultrasonic waves using a pulse laser and measure the response with a Fabry-Perot interferometer, and related the

amplitude of the reflected ultrasonic waves with a hole defect diameter in a carbon steel specimen. Park et al. tried

to localize delaminations using a time-of-flight triangulation of delamination reflected ultrasonic waves [70].

Several researches are focused to effective laser ultrasonic generation, as damage-ultrasonic interac-

tions are relatively small compared to the incident waves. Surface crack depth is estimated by measuring trans-

mitted ultrasonic waves through a crack, which are generated using a line laser [71]. Sohn and Krishnaswamy [72]

were inspired from waveform changes near a damage due to damage-ultrasonic interaction. If a pulse laser beam

is irradiated near a damage, the stronger ultrasonic waves are generated as the damage reflected ultrasonic waves

are added to the incident ultrasonic waves. A surface breaking flaw in an aluminum plate is successfully localized

by measuring generated ultrasonic wave amplitude at each spatial point. Kim et al. [73] used a specially designed

arrayed arc slit to focus laser generated ultrasonic waves. A pulse laser beam is illuminated on the target surface

through an arc-shaped slit, and the generated ultrasonic waves are concentrated at a focus of an arc. This not only

increases generated ultrasonic wave power at the focus but also enhances damage detection spatial resolution to

the focus.

2.5.3. Nonlinear ultrasonic techniques

It has been reported that over 90% of failures in in-service metallic structures are caused by fatigue cracks [1].

This recently leads a high interest to effective crack detection techniques. Nonlinear ultrasonic techniques are

considered as one of the most promising solutions [74]. These techniques measure nonlinear structural behavior

resulted from crack opening and closing, which occur as ultrasonic waves propagate through the crack. These

nonlinear behaviors are usually categorized into three: (1) sub-harmonics, responses appear at submultiples of the

input frequencies [75]; (2) harmonics, responses appear at multiples of the input frequencies [76]; and (3) modu-

lations, responses appear at linear combinations of the input frequencies [77].

There have been several works using narrowband actuators to measure nonlinear behaviors from fa-

tigue cracks. Bermes et al. [78] measured harmonics of the wedge transducer generated ultrasonic waves using a

laser interferometer, and Lim et al. [79] measured modulations of the air coupled transducer generated ultrasonic

waves using a 3D LDV. Especially Lim et al. shows that nonlinear modulations occur in a plate only if at least

one of two input ultrasonic waves is a symmetric Lamb mode.

- 30 -

Though nonlinear changes in laser generated ultrasonic waves have been numerically modeled and

experimentally measured [80], measuring nonlinear behaviors from laser generated ultrasounds is relatively chal-

lenging as a pulse laser is a broadband excitation. Sub-harmonics, harmonics, and modulations are hard to be

distinguished from its original response. But these nonlinear behaviors increase the number of peaks of the meas-

ured response in frequency domain compared to the one of the intact response. Fatigue cracks in an aluminum

plate and an aircraft fuselage are detected by counting these peaks [81]. Choi et al. [82] solved this problem by

generating narrowband ultrasonic waves using a special line arrayed slit and a pulse laser beam. They related the

amplitude of a harmonic response and plastic deformations of an aluminum alloy specimen.

Liu et al. [83] proposed a nonlinear damage detection technique using state space attractors [84]. Instead

of tracking sub-harmonic, harmonic, and modulation frequency peaks, this state space attractor approach com-

pares the measured response and the reference response to identify any changes in dynamic system characteristics.

This technique is also expanded into damage visualization using scanned ultrasonic measurements [85]. By using

adjacent measurements as reference responses, baseline-free nonlinear damage visualization is available for a

fatigue crack in an aluminum plate and a delamination in a composite wind turbine blade.

- 31 -

2.6. Chapter summary

In this chapter, the working principles of noncontact laser ultrasonics and a laser ultrasonic scanning system is

presented. When a laser beam is emitted onto an infinitesimal solid surface, three phenomena may occur: absorp-

tion, reflection, and transmission of laser radiation. Absorbed laser energy leads a localized heating of the infini-

tesimal area, which causes thermoelastic expansion of the material and generates ultrasonic waves. This ultrasonic

wave generation process can be numerically modeled by considering its thermal wave propagations and ultrasonic

wave propagations. Corresponding ultrasonic responses are measured by a laser Doppler vibrometer. This device

measure the frequency shift of a reflected laser beam from a moving surface. Ultrasonic response, or velocity of

this moving surface, is obtained from the relationship between the measured frequency shift and the velocity.

Then the schematic diagrams of the laser ultrasonic scanning systems are provided, and possible scanning strate-

gies are discussed. These scanning strategies can be divided into two categories according to its use of contact

transducers, (1) fully noncontact scanning and (2) partially noncontact scanning, while the proposed scanning

systems are fully noncontact.

The noncontact laser ultrasonic scanning system has following advantages: (1) No sensor installation

is required for target structures; (2) high spatial resolution in damage detection can be achieved with its small

beam size; and (3) it can be applicable to harsh environments with its noncontact nature. However, it should be

noted that this system has several challenges in field applications. These may include (1) eye safety issues of high

power pulse laser beams and (2) sensitivity limitation of LDVs. These limitations can be solved by introducing a

microscopic laser ultrasonic scanning system.

Previous damage detection works using noncontact laser ultrasonics are also discussed. Linear ultra-

sonic techniques are one of the earliest works, based on measuring reflections and transmissions of ultrasonic

waves from damages. Several researches are focused to narrowband laser ultrasonic generation to enhance their

damage detection performances. Nonlinear techniques are relatively recently emerging to detect fatigue cracks in

their early stage. They measure nonlinear structural behaviors including subharmonics, harmonics, and modula-

tions. But these techniques are limited in terms of damage localization with a high spatial resolution.

Wave propagation imaging techniques are the most popular in this field with its intuitive damage de-

tection availability. Wave propagation behaviors in an inspection region is obtained and its interactions with dam-

age are visualized. Though these techniques can localize and quantify damages effectively, they require a large

number of measurements to cover inspection area with a high spatial resolution. Many researchers are interested

in accelerating damage quantification with a high spatial resolution.

- 32 -

Chapter 3. Accelerated Damage Detection with

Reduced Laser Ultrasonic Scanning Points

In this chapter, two damage detection techniques are proposed. They are inspired by a binary search algorithm

[86] to reduce the required number of laser ultrasonic scanning points. The first one is combined binary search

and compressed sensing. This uses the fixed origin scanning strategies. The other technique is binary search with

fixed pitch-catch distance scanning strategy. Detailed working procedures and effectiveness of each technique are

discussed in following subchapters.

3.1. Combined binary search and compressed sensing

3.1.1. Binary search for damage localization

Figure 3.1. Schematic flow of binary search for transition line detection.

In this subchapter, a binary search is adopted to reduce the number of excitation scanning points and to identify

an approximate region of damage, which is referred to as the transition line. Specifically, the transition line is

- 33 -

defined as a single side damage border that faces the sensing points. Figure 3.1 provides an overview of the

proposed binary search.

Step 1: Assume that we have an inspection region of size 𝑁 (width) × 𝑀 (height) with a predefined

spatial resolution. Each point in this 𝑁 × 𝑀 grid is referred to as an inspection point, and the mth row of this

region is referred to as the mth inspection line. The middle inspection line, ⌈𝑀/2⌉th row is selected and a sensing

point is located to the left of this inspection line.

Step 2: Initially, the ultrasonic waves are generated at the center of the inspection line using a pulse

laser and measured at the sensing point using a LDV. Then, the measured ultrasonic response is transformed into

the spatial ultrasonic domain to verify whether the excitation point is ‘in front of the damage’ or ‘behind the

damage’. Details on this checking procedure are provided in Chapter 4.3. If the excitation point is in front of the

damage, the excitation point is marked in green. Then only the inspection points after the current excitation point

to the very last (far right) inspection point in the given inspection line are considered for the next excitation point

selection. If the excitation point is behind the damage, the excitation point is marked in yellow. In this case, the

inspection points before the current excitation point to the very first (far left) inspection point are considered for

the next excitation point selection. The next excitation point is moved to the center of the inspection points. This

binary search is repeated until no considerable inspection point exists for the next excitation point selection.

Step 3: The closest ‘behind the damage’ point from the sensing point is defined as the transition point

and is marked on the inspection region as a crossed yellow point. If no ‘behind the damage’ point in the inspection

line exists, the corresponding inspection line is named as an intact inspection line. On the other hand, if any

‘behind the damage’ point in the inspection line exists, it is named as a damage inspection line.

Step 4: Steps 1 to 3 are repeated for the next inspection line. The next inspection line is selected using

the binary search concept, and Steps 1 to 3 are repeated until the uppermost and lowermost transition points have

been identified.

Step 5: The approximate damage location can be identified by connecting all transition points, and this

connection of the transition points constitutes the transition line. Because the sensing points are located on the left

side of the inspection region, the transition line corresponds to the left border of the damage.

This binary search process requires a fewer number of scanning points to find the transition line. Let

𝑝𝑛 be the number of scanning points that are required by the binary search for a single inspection line. Then, 𝑝𝑛

can be obtained as [86]

- 34 -

𝑝𝑛 = log2 𝑁 (3.1)

where N is the size of the inspection region in the widthwise direction, as defined in Step 1. 𝑝𝑛 is independent to

whether the inspection line is intact or damaged. After 𝑝𝑛 scanning, the damage existence and the transition point

in an inspection line can be identified.

This process is repeated for other inspection lines. The binary search algorithm is adopted to select the

next inspection line. This algorithm is extended to the vertical direction by considering the intact inspection line

and the damage inspection line as an ‘in front of the damage’ line and a ‘behind the damage’ line, respectively.

To select the next inspection line and find the uppermost transition point, the inspection lines that are higher than

the ‘behind the damage’ line or lower than the ‘in front of the damage’ line are considered for the next inspection

line selection. To select the next inspection line to find the lowermost transition point, the opposite process is

employed. The required number of inspection lines is presented as follows, where M is the size of the inspection

region in the heightwise direction, as defined in Step 1:

𝑝𝑚 = 2 log2 𝑀 (damaged),

(3.2)

𝑝𝑚 = 𝑀 (intact).

Note that the logarithm term is multiplied by two because the uppermost and lowermost transition points are

required by this approach. If there is no damage, every inspection line needs to be searched to ensure an intact

condition. Then, the number of scanning points 𝑝 required by the proposed binary search is calculated as follows:

𝑝 = 𝑝𝑛 ⋅ 𝑝𝑚 = 2 log2 𝑁 ⋅ log2 𝑀 (damage),

(3.3)

𝑝 = 𝑝𝑛 ⋅ 𝑝𝑚 = log2 𝑁 ⋅ 𝑀 (intact).

Then, the reduction rate R is defined as follows:

𝑅 = (1 −𝑝𝑛 ⋅ 𝑝𝑚

𝑁 ⋅ 𝑀) × 100% = (1 −

2 log2 𝑁 ⋅ log2 𝑀

𝑁 ⋅ 𝑀) × 100% (damage),

(3.4)

𝑅 = (1 −𝑝𝑛 ⋅ 𝑝𝑚

𝑁 ⋅ 𝑀) × 100% = (1 −

log2 𝑁

𝑁) × 100% (intact).

- 35 -

3.1.2. Compressed sensing for damage quantification

Once the approximate damage region (transition line) is identified using the proposed binary search, the wavefield

images around the transition line are solely reconstructed based on the previous scanning points without any ad-

ditional scanning. Figure 3.2 provides an overview of the proposed damage quantification technique based on

compressed sensing.

Figure 3.2. Schematic flow of compressed sensing for damage quantification.

Step 1: The imaging region is defined as the smallest rectangular region, including all scanning points

in the damage inspection lines, as represented by a dashed box in Step 1 of Figure 3.2.

Step 2: Within the imaging region, responses at unscanned inspection points are reconstructed from the

scanning points using compressed sensing (CS) [63-65]. Details on this reconstruction procedure are provided in

Chapter 4.4.

Step 3: Wavefield images are created for the imaging region from the scanned responses and recon-

structed unscanned responses.

Step 4: The actual damage points are identified by extracting standing wave energy generated by wave-

damage interactions [44]. Local standing waves are generated in the damage as the ultrasonic waves are trapped

inside from their multiple reflections at the damage boundaries. Then, the damage is visualized and quantified.

- 36 -

3.2. Binary search with fixed pitch-catch distance scanning

Figure 3.3. Schematic flow of the binary search with fixed pitch-catch distance scanning.

In this chapter, the fixed pitch-catch distance scanning strategy is adopted to reduce the number of scanning points

and to quantify the damage. There are two notable advantages of this technique to the aforementioned one: (1)

High SNR is guaranteed even for larger inspection regions as the distance between the excitation and the sensing

point is fixed; and (2) additional signal processing e.g. compressed sensing is not required for damage quantifica-

tion. Figure 3.3 provides an overview of the proposed damage quantification technique.

Step 1: Assume that we have an inspection region of size 𝑁 (width) × 𝑀 (height) with a predefined

spatial resolution. Each point in this 𝑁 × 𝑀 grid is referred to as an inspection point, and the mth row of this

region is referred to as the mth inspection line. The middle inspection line, ⌈𝑀/2⌉th row is selected.

- 37 -

Step 2: The selected inspection line is divided into equally spaced divisions. The size K of the division

is determined to be as large as possible while maintaining high SNR. Initially, the ultrasonic waves are generated

at the right end of the center division using a pulse laser and measured at the left end of the division. Then, the

measured ultrasonic response is transformed into the spatial ultrasonic domain to verify whether the excitation

point is ‘in front of the damage’ or ‘behind the damage’, as in Chapter 3.1.1. If the excitation point is in front of

the damage, the excitation point is marked in green and the corresponding division is named as an intact division.

On the other hand, if the excitation point is behind the damage, the excitation point is marked in yellow and the

corresponding division is named as a damage division. This process is repeated for other divisions. The binary

search algorithm is adopted to select the next division by considering the intact division and the damage division

as an ‘in front of the damage’ division and a ‘behind the damage’ division, respectively.

Step 3: Approximate damage region is identified from the leftmost damage division to the rightmost

damage division, and marked with a dashed box. If there is no damage region, Step 4 to 6 are unrequired as there

is no damage point to identify.

Step 4: The right transition point is identified by adopting the binary search algorithm to the rightmost

damage division. If the excitation point is in front of the damage, the sensing point is marked in green while the

sensing point is marked in yellow if the excitation point is behind the damage. The rightmost yellow point is

marked as a crossed yellow point.

Step 5: Step 4 is repeated to identify the left transition point. Here, the excitation points are marked in

yellow or in green. The leftmost yellow point is marked as a crossed yellow point.

Step 6: The left and the right transition points are identified. These two points indicate the left and the

right border of the damage in this inspection line, respectively.

Step 7: Steps 1 to 6 are repeated for the next inspection line. The next inspection line is selected using

the binary search concept, and Steps 1 to 6 are repeated until the uppermost and lowermost transition points have

been identified.

Step 8: The damage can be quantified by connecting all transition points, and the region inside this

connection is considered to be damage.

This fixed pitch-catch distance scanning requires a fewer number of scanning points to quantify the

damage. Let 𝑝𝑎 the number of scanning points that are required for damage region approximation (Step 2 to 3).

Then, 𝑝𝑎 can be obtained as

- 38 -

𝑝𝑎 = 2log2

𝑁

𝐾 (damage),

(3.5)

𝑝𝑎 =𝑁

𝐾 (intact),

where N is the size of the inspection region in the widthwise direction, and K is the size of the division. 𝑝𝑎 is

dependent on whether the inspection line is intact or damaged. If damage region exists, additional number of

scanning points 𝑝𝑏 are required for transition point identification (Steps 4 to 6).

𝑝𝑏 = 2 log2 𝐾. (3.6)

Then the number of scanning points 𝑝𝑛 that are required by the fixed pitch-catch distance scanning

for a single inspection line is:

𝑝𝑛 = 2log2

𝑁

𝐾+ 2 log2 𝐾 = 2 log2 𝑁 (damage),

(3.7)

𝑝𝑛 =𝑁

𝐾 (intact).

Note that the logarithm terms are multiplied by 2, as this approach searches both left and right boundary.

This process is repeated for other inspection lines with the same manner to Equation (3.2)

𝑝𝑚 = 2 log2 𝑀 (damage),

(3.8)

𝑝𝑚 = 𝑀 (intact).

Then, the number of scanning points 𝑝 and the reduction rate R is calculated as follows:

𝑝 = 𝑝𝑛 ⋅ 𝑝𝑚 ≤ 4 log2 𝑁 ⋅ log2 𝑀 (damage),

(3.9)

𝑝 = 𝑝𝑛 ⋅ 𝑝𝑚 =𝑁 ⋅ 𝑀

𝐾 (intact),

𝑅 = (1 −𝑝𝑛 ⋅ 𝑝𝑚

𝑁 ⋅ 𝑀) × 100% ≤ (1 −

4 log2 𝑁 ⋅ log2 𝑀

𝑁 ⋅ 𝑀) × 100% (damage),

(3.10)

𝑅 = (1 −𝑝𝑛 ⋅ 𝑝𝑚

𝑁 ⋅ 𝑀) × 100% = (1 −

1

𝐾) × 100% (intact).

- 39 -


In this chapter, two damage detection techniques are proposed. They are inspired by a binary search algorithm to

reduce the required number of laser ultrasonic scanning. The (1) combined binary search and compressed sensing

and (2) binary search with fixed pitch-catch distance scanning are based on the fixed origin scanning and the fixed

pitch-catch distance scanning strategy, respectively.

In the first technique, the approximate damage location is identified by examining the interactions be-

tween the ultrasonic waves and damage at the sparse scanning points that are selected by the binary search algo-

rithm. Then, wavefield images around the damage are reconstructed from the previously selected scanning points

using compressed sensing. The number of scanning points that is necessary for damage localization and visuali-

zation is dramatically reduced from 𝑁 ∙ 𝑀 to 2 log2 𝑁 ⋅ log2 𝑀. 𝑁 and 𝑀 represent the number of equally

spaced scanning points in the x and y directions, respectively, which are required to obtain full-field wave propa-

gation images of the target inspection region.

The other technique approximates damage location by identifying damage divisions in each inspection

line. Then, the exact left and right boundary of the damage are identified by scanning excitation and sensing laser

beams with a fixed distance between them. The divisions and the scanning points are selected by the binary search

algorithm. The number of scanning points that is necessary for damage quantification is dramatically reduced

from 𝑁 ∙ 𝑀 to 4 log2 𝑁 ⋅ log2 𝑀 in the worst case. There are two notable advantages of this technique in com-

parison with the aforementioned two: (1) High SNR is guaranteed even for larger inspection regions as the dis-

tance between the excitation and sensing point is fixed; and (2) no additional signal processing such as compressed

sensing is required for damage quantification.

- 40 -

Chapter 4. Representation of a Laser Ultrasonic Response

using a Spatial Ultrasonic Dictionary

4.1. Sparse representation of a signal

A transform 𝐹 is a function mapping a vector set 𝐒 to another vector set 𝐀 [87].

𝛂 = 𝐹(𝐬),

(4.1)

𝛂 ∈ 𝐀, 𝐬 ∈ 𝐒.

Here 𝐬 is referred to as a signal in a signal space ℂ𝑇 (𝑇 × 1 vector) and 𝛂 is its representation in the transform

domain ℂ𝐿 and has a dimension of 𝐿 × 1. 𝐹 is referred to as a linear transform if the signal 𝐬 is represented

as a weighted linear combination of bases via following matrix operation [87].

𝐬 = 𝐃𝛂,

(4.2)

𝐃 = {𝐝1, 𝐝2, … , 𝐝𝐿}.

D is a 𝑇 × 𝐿 matrix, which is referred to as a dictionary and consists of 𝐝𝑖 bases (i = 1, …, L). Each basis 𝐝𝑖 is

a vector with the same dimensions as 𝐬. The Fourier transform is one of the most famous examples of linear

transforms. In this case, 𝐬 and 𝛂 can be considered as a time domain signal and its frequency domain represen-

tation respectively, and 𝐃 is a set of sinusoids with various frequencies.

As a large L value is preferred to achieve a high resolution in the transform domain, Equation (4.2)

typically represents an underdetermined system of equations (L > T), which creates non-unique solutions for 𝛂.

A unique 𝛂 can be obtained via the assumption that the actual solution has the sparsest representation, which

produces the minimum number of nonzero entities in 𝛂 [88]. For example, in acoustic fields, Fourier transform

is very preferred as this can represent a complicated time domain acoustic signal into a sum of several frequency

components. By selecting an appropriate dictionary, the signal can be represented in a sparse form and its repre-

sentation can be obtained by solving the following problem 𝑃0.

𝑃0: min‖𝛂‖0 s. t. 𝐬 = 𝐃𝛂, (4.3)

where ‖𝛂‖0 is defined as the ℓ0 norm of 𝛂 which represent the number of nonzero elements in 𝛂.

- 41 -

But it is quite challenging to solve 𝑃0 both in theoretically and numerically. Instead, Mallat and Zhang

proposed a sub-optimal greedy sequential algorithm called matching pursuit [89]. This algorithm works in fol-

lowing steps: (1) Set �̅� = 𝐬 and 𝛂 as a 𝐿 × 1 zero vector; (2) find 𝐝𝒊 which is the most correlated to �̅� with

a correlation of 𝑐𝑖; (3) add 𝑐𝑖 to 𝛼𝑖 , and substitute �̅� − 𝑐𝑖𝐝𝒊 to �̅�; (4) repeat (2)~(3) until the residual �̅� is

smaller than the predefined threshold. This algorithm can find the sparsest representation for many cases, but it

shows unsatisfactory results for several examples as shown in Figure 4.1. Here the dictionary contains 200 bases,

first 100 bases for sinusoidal signals and the other half for spikes. The signal (Figure 4.1 (a)) is composed of two

sinusoidal signals (basis #20 and #100) and three spikes (basis #133, #164 and #194). As shown in Figure 4.1 (b),

unexpected sinusoidal bases are included in the obtained solution.

(a)

(b)

(c)

Figure 4.1. An example to find the sparsest representation: (a) a signal to be transformed, (b) a matching pursuit

representation, and (c) a basis pursuit representation. The original signal is consisted of basis #20, #100, #133,

#164, and #194.

- 42 -

To acquire a better representation, Chen and Donoho proposed the basis pursuit approach to solve the

underdetermined system in Equation (4.2) [90]. The sparsest representation is obtained by solving the following

problem with a given dictionary D.

𝑃1: min‖𝛂‖1 s. t. 𝐬 = 𝐃𝛂, (4.4)

where ‖𝛂‖1 denotes the ℓ1 norm of 𝛂. Time-domain ultrasonic signals can be sparsely represented using the

designed spatial ultrasonic dictionary.

Figure 4.2. Graphical representation of the ℓ1 minimization [91].

As it is in a form of linear programming, this problem can be numerically solved. Then how can we

assure that the sparsest solution can be obtained by solving 𝑃1? It is shown that it successfully finds the sparsest

representation in [90] with detailed explanations and proofs while Figure 4.2 provides a schematic idea. This

shows solutions for 𝑃𝑝 with N=1 and L=2 in three cases, (1) 0 ≤ 𝑝 < 1; (2) p= 1; and (3) 𝑝 > 1.

𝑃P: min‖𝛂‖𝑝 s. t. 𝐬 = 𝐃𝛂. (4.5)

In the first two cases, the signal (red dotted line) and the solution (blue solid line) meets on the axis at the black

dots. This is the sparsest solution as the signal is represented with a single basis. But in the third case the signal

and the solution cannot meet on the axis and the signal is now represented with two bases. Therefore, the sparsest

representation can be obtained by solving 𝑃𝑝 for 0 ≤ 𝑝 ≤ 1, and 𝑝 = 1 can take the full benefits from the linear

programming. Figure 4.1 (c) shows that the desired sparse representation is obtained through the basis pursuit

approach. Gribonval and Nielsen [88] also proved that basis pursuit is guaranteed to find the sparsest solution

under the following condition:

- 43 -

‖𝛂‖0 < 0.5 (1 +1

𝑀(𝐃)), (4.6)

where 𝑀(𝐃) is the coherence of the dictionary.

But 𝑃1 is not a practical problem formulation as there should be undesired noise in the signal. Basis

pursuit denoising (BPDN) is a solving approach with a consideration to the noise components [92].

min‖𝛂‖1 s. t. ‖𝐬 − 𝐃𝛂‖22 < 𝛿. (4.7)

Equation (4.7) is a common expression for BPDN problem, where 𝛿 is a predefined threshold value. And Equa-

tion (4.8) is its alternative formulation, where 𝜆 is a parameter controlling trade-off between the sparsity and the

reconstruction fidelity.

min 𝜆‖𝐬‖1 +1

2‖𝐬 − 𝐃𝛂‖2

2. (4.8)

This finds the sparsest representation with a consideration of residual noise components. Many numerical algo-

rithms to solve this problem have been proposed, including the in-crowd algorithm [92], the homotopy continua-

tion [93], and the fixed point continuation [94]. In this paper, a dual active-set quadratic programming method

and corresponding MATLAB open source solver is used, which is developed by a research team in Stanford

University in 2012 [95].

- 44 -

4.2. Interactions between ultrasonic waves and damage

Ultrasonic waves can propagate through a solid media in a various form [96]. This may include pressure waves,

shear waves, torsional waves, surface Rayleigh waves, bulk waves, and guided waves. But these terms can be

considered as a result of interactions between axial waves and flexural waves in the corresponding media. In this

subchapter, wave equations and corresponding solutions for axial and flexural waves are presented. Also interac-

tions between ultrasonic waves and damage are discussed.

4.2.1. Ultrasonic wave propagation

Figure 4.3. Schematic diagram of a uniform bar and corresponding axial wave propagation.

Axial wave in a bar is the simplest conceptualization of ultrasonic wave propagation [96]. The resultant displace-

ment is parallel to the wave propagation direction. Let us assume a uniform bar of elastic modulus E, cross sec-

tional area A, and mass m. Then the equation of motion for axial waves in the bar is [97]

𝐸𝐴𝑢′′ = 𝑚�̈�, (4.9)

where 𝑢(𝑥, 𝑡) indicates longitudinal displacement at space x and time t. Equation (4.9) can be represented in a

form of wave equation by dividing both sides by m:

𝑐𝐴2𝑢′′ = 𝑐𝐴

2𝜕2𝑢

𝜕𝑥2=

𝜕2𝑢

𝜕𝑡2= �̈�, (4.10)

with the wave speed 𝑐𝐴 = √𝐸𝐴 𝑚⁄ = √𝐸 𝜌⁄ where 𝜌 is the density of this bar. As E and 𝜌 are uniform in this

bar, the wave speed is constant and independent to the frequency of this axial wave. Then the corresponding

solution for this wave equation is

𝑢(𝑥, 𝑡) = 𝐴1𝑒−𝑗𝑘𝐴(𝑥−𝑐𝐴𝑡) + 𝐴2𝑒𝑗𝑘𝐴(𝑥+𝑐𝐴𝑡), (4.11)

- 45 -

where 𝑘𝐴 = 𝜔 𝑐𝐴⁄ is the wavenumber of this wave at radial frequency of 𝜔. Here, the first term represents ‘for-

ward’ propagating waves which propagate into positive x direction, while the second term represents ‘backward’

propagating waves.

Figure 4.4. Schematic diagram of a uniform beam and corresponding flexural wave propagation.

Flexural wave is another important concept in ultrasonic wave motion [97]. Here the resultant displace-

ment is corresponding to the bending motion. Let us assume a uniform beam of elastic modulus E, moment of

inertia I, and mass m. Then the equation of motion for flexural waves in a beam is

𝐸𝐼𝑤′′′′ + 𝑚�̈� = 0, (4.12)

where 𝑤(𝑥, 𝑡) indicates transverse displacement at space x and time t. Equation (4.12) can be represented in a

form of wave equation as Equation (4.13) by dividing both sides by m and the corresponding solution for this

wave equation is:

𝑎4𝑤′′′′ + �̈� = 0, (4.13)

𝑤(𝑥, 𝑡) = 𝐴1𝑒−𝑗𝑘𝐹(𝑥−𝑐𝐹𝑡) + 𝐴2𝑒𝑗𝑘𝐹(𝑥+𝑐𝐹𝑡) + 𝐵1𝑒−𝑘𝐹𝑥𝑒𝑗𝜔𝑡 + 𝐵2𝑒𝑘𝐹𝑥𝑒𝑗𝜔𝑡 , (4.14)

with the wave speed 𝑐𝐹 = 𝑎√𝜔 = (𝐸𝐼 𝑚⁄ )1/4√𝜔 and wavenumber 𝑘𝐹 = 𝜔 𝑐𝐹⁄ . It should be noted that now the

wave speed is depending on the wave frequency and not a constant anymore. This leads to a dispersive phenom-

enon, which waves of different frequencies travel at different speeds. This is especially important in laser ultra-

sonics as the laser generated ultrasonic waves are usually broadband. Other things to notice are the third and the

fourth term of Equation (4.14). These terms are known as evanescent terms as their amplitude decreases exponen-

tially as they propagate. They are only meaningful near the wave source and can be neglected for most of cases.

- 46 -

4.2.2. Wave-damage interactions

When ultrasonic waves meet a damage during its propagation, wave-damage interactions occur. Two types of

wave-damage interactions are discussed in this subchapter: reflection and transmission.

Figure 4.5. Schematic representation of wave reflection and transmission at an interface.

When waves encounter an interface between two materials, some energy is transmitted through the

other material while the other is reflected [98]. Damage can be considered as a discontinuity, an air as the second

material in its extreme case scenario. Damage detection techniques in Chapter 3 are based on these linear changes.

Both axial waves and flexural waves solutions, Equations (4.11) and (4.14), can be represented as follows.

𝑢(𝑥, 𝑡) = 𝐴1𝑒−𝑗𝑘𝑖(𝑥−𝑐𝑖𝑡) + 𝐴2𝑒𝑗𝑘𝑖(𝑥+𝑐𝑖𝑡). (4.15)

Here i can be A or F, representing axial and flexural waves respectively. Let us represent the incident, reflected,

and transmitted waves as follows:

𝑢𝐼(𝑥, 𝑡) = 𝐴𝐼𝑒−𝑗𝑘𝑖(1)

(𝑥−𝑐𝑖(1)

𝑡), (4.16)

𝑢𝑅(𝑥, 𝑡) = 𝐴𝑅𝑒𝑗𝑘𝑖(1)

(𝑥+𝑐𝑖(1)

𝑡), (4.17)

𝑢𝑇(𝑥, 𝑡) = 𝐴𝑇𝑒−𝑗𝑘𝑖(2)

(𝑥−𝑐𝑖(2)

𝑡), (4.18)

where 𝐴𝐼, 𝐴𝑅, and 𝐴𝑇 indicate the amplitude of the incident, reflected, and transmitted waves respectively, and

𝑘𝑖(1)

, 𝑘𝑖(2)

, 𝑐𝑖(1)

, and 𝑐𝑖(2)

indicate the wavenumbers and the wave speeds in the material 1 and 2 respectively.

By considering the boundary condition, 𝐴𝑅 and 𝐴𝑇 can be represented in terms of 𝐴𝐼.

- 47 -

𝑢𝑅(𝑥, 𝑡) =𝜌1𝑐𝑖

(1)− 𝜌2𝑐𝑖

(2)

𝜌1𝑐𝑖

(1)+ 𝜌2𝑐𝑖

(2)𝐴𝐼𝑒𝑗𝑘𝑖

(1)(𝑥+𝑐𝑖

(1)𝑡) = 𝐶𝑅𝐴𝐼𝑒𝑗𝑘𝑖

(1)(𝑥+𝑐𝑖

(1)𝑡), (4.19)

𝑢𝑇(𝑥, 𝑡) =2𝜌1𝑐𝑖

(1)

𝜌1𝑐𝑖

(1)+ 𝜌2𝑐

𝑖

(2)𝐴𝐼𝑒−𝑗𝑘𝑖

(2)(𝑥−𝑐𝑖

(2)𝑡) = 𝐶𝑇𝐴𝐼𝑒−𝑗𝑘𝑖

(2)(𝑥−𝑐𝑖

(2)𝑡), (4.20)

where 𝐶𝑅 and 𝐶𝑇 are known as the reflection and the transmission coefficient, respectively. They are deter-

mined by the material property difference between two materials, and the term 𝜌𝑛𝑐𝑖(𝑛)

is called as the acoustic

impedance of the material n.

- 48 -

4.3. Representation of time domain ultrasonic responses using a spatial ultrasonic dic-

tionary

The binary search algorithm in Chapter 3.1 is based on whether the current excitation point is ‘in front of the

damage’ or ‘behind the damage’. This problem entails locating damage with respect to the sensing point and the

current excitation point.

(a) (b)

Figure 4.6. Comparison of ultrasonic responses (a) when a damage is located between the excitation point and

sensing point and (b) outside the direct wave propagation path between the excitation point and sensing point.

Blue solid line and red dashed line denote the measured signals from the intact condition and damage condition

of the specimen, respectively. The bottom subfigures represent the difference between the intact signal and the

damage signal.

Figure 4.6 illustrates the limitation of the conventional ultrasonic approach in terms of locating damage

with respect to the excitation and sensing points. Conventionally, damage existence is identified by subtracting

the baseline data that corresponds to the pristine condition of a specimen from the measured signal in the time

domain. Figure 4.6 (a) shows the baseline signal from an intact condition (blue solid line) and the measured signal

from a damage condition (red dashed line) when the damage is located between the excitation point and the sens-

ing point. In addition, the difference between the two signals is revealed in the bottom subfigure. In a similar

manner, Figure 4.6 (b) shows the baseline and measured signals when the damage is introduced near but outside

- 49 -

the direct path. Figure 4.6 (b) reveals that the measured signal significantly deviates from its pristine condition

even when the damage is located outside the direct wave propagation path due to reflections from the damage.

The effect of the damage outside the direct wave path may be alleviated by time truncation [67]. However, the

selection of the proper truncation time point can be challenging due to the multi-mode and dispersive nature of

Lamb waves.

To determine a damage location with respect to the excitation and sensing points, the measured time-

domain ultrasonic signal 𝐬 (𝑇 × 1 vector) is represented as a weighted linear combination of bases via the fol-

lowing transformation [87].

𝐬 = 𝐃𝛂,

(4.21)

𝐃 = {𝐝1, 𝐝2, … , 𝐝𝐿},

where 𝛂 is a representation of s in the transformed domain and has a dimension of 𝐿 × 1. D, a dictionary, is a

𝑇 × 𝐿 matrix and consists of 𝐝𝑖 bases (i = 1, …, L). Each basis 𝐝𝑖 is a time-domain signal with the same di-

mensions as 𝐬.

In this study, a dictionary coined as a spatial ultrasonic dictionary is created for effective representations

of ultrasonic responses. For a given specimen, each basis in the spatial ultrasonic dictionary represents an ultra-

sonic response that corresponds to a specific wave propagation distance from the fixed sensing points. For exam-

ple, 𝐝𝑖 denotes the ultrasonic response generated at 𝑥𝑖 (= 𝑥1 + (𝑖 − 1)∆𝑥) distance from the sensing point. 𝑥1

denotes the distance from the sensing point to the closest excitation point, and ∆𝑥 is the spacing between two

excitation points.

𝐝𝑖 = 𝐴𝑒−𝑗𝑘(𝑥𝑖−𝑐𝑡) = 𝑓(𝑥𝑖 − 𝑐𝑡), (4.22)

where A, k and c represent the amplitude, wavenumber and wave speed, respectively, of the propagating ultrasonic

waves in the specimen, and t denotes time. A smaller ∆x generates a higher spatial resolution in the spatial ultra-

sonic domain, and increases the dimension L of the dictionary D. Then the sparsest representation 𝛂 is obtained

by solving the problem in Equation (4.4) with the given dictionary.

If no damage is introduced to the specimen after the dictionary is composed (Figure 4.7 (a)), the ultra-

sonic response 𝐬 generated at 𝑥𝑎 distance from the sensing point can be represented with only a single basis.

This is only one nonzero element in 𝛂, and a sparse representation of the time-domain signal is possible in the

spatial ultrasonic domain. Then, 𝐬 can be represented as

- 50 -

𝐬 = 𝑓(𝑥𝑎 − 𝑐𝑡) = 𝐝𝑎 = 𝐝𝑎𝛼𝑎 , (4.23)

where 𝛼𝑎 is the amplitude of the basis 𝐝𝑎 in the transformed domain and its value is equal to one when no

damage in the specimen exists.

Next, the time-domain responses obtained from a damage specimen are represented in the spatial ul-

trasonic domain. First, assume that the damage is located outside the direct wave propagation path (Figure 4.7

(b)). The excitation point is ‘in front of the damage’. Second, the measured signal 𝐬 can be represented as the

superposition of the direct incident waves and the damage reflected waves.

𝐬 = 𝑓(𝑥𝑎 − 𝑐𝑡) + 𝑔(𝑥𝑎 + Δ𝑥𝐷 − 𝑐(𝑡 − Δ𝑡𝐷))

= 𝑓(𝑥𝑎 − 𝑐𝑡) + 𝑔(𝑥𝑎 + Δ𝑥𝐷 − 𝑐𝑡 + 𝑐Δ𝑡𝐷))

= 𝑓(𝑥𝑎 − 𝑐𝑡) + 𝑔((𝑥𝑎 + 2Δ𝑥𝐷) − 𝑐𝑡))

= 𝑓(𝑥𝑎 − 𝑐𝑡) + 𝐶𝑅𝑓((𝑥𝑎 + 2Δ𝑥𝐷) − 𝑐𝑡))

= 𝑓(𝑥𝑎 − 𝑐𝑡) + 𝐶𝑅𝑓(𝑥𝑏 − 𝑐𝑡)) = 𝐝𝑎 + 𝐶𝑅𝐝𝑏 = 𝐝𝑎𝛼𝑎 + 𝐝𝑏𝛼𝑏 .

(4.24)

The first term f represents the direct incident waves generated at 𝑥𝑎 distance from the sensing point, and the

second term 𝑔 represents the reflection from the damage. The term Δ𝑡𝐷 is included in 𝑔 as the waves are re-

flected from the damage Δ𝑡𝐷 after the original ultrasonic generation. Δ𝑡𝐷 = Δ𝑥𝐷/𝑐 is the time-of-flight of the

ultrasonic waves from the excitation point to the damage, where Δ𝑥𝐷 is the distance between the excitation point

and the damage. The signal is represented with two bases: 𝐝𝑎 and 𝐝𝑏. The first base corresponds to the incident

waves with the ultrasonic travel distance 𝑥𝑎, and the second base corresponds to the reflected waves with travel

distance 𝑥𝑏 = 𝑥𝑎 + 2Δ𝑥𝐷. 𝛼𝑏 is the amplitude of the basis 𝐝𝑏 in the transformed domain and its value is equal

to the reflection coefficient of the damage 𝐶𝑅.

A similar analysis can be performed for the other case, where the ultrasonic waves pass through the

damage (Figure 4.7 (c)). Then, the excitation point is ‘behind the damage’, and the measured signal 𝐬 can be

represented as

𝐬 = ℎ((𝑥𝑎 − Δ𝑥𝐷 − Δ𝑥𝑑) − 𝑐(𝑡 − Δ𝑡𝐷 − Δ𝑡𝑑))

= ℎ ((𝑥𝑎 − Δ𝑥𝐷 − Δ𝑥𝑑) − 𝑐𝑡 + Δ𝑥𝐷 +𝑐

𝑐𝑑

Δ𝑥𝑑))

= ℎ ((𝑥𝑎 + (𝑐

𝑐𝑑

− 1)Δ𝑥𝑑) − 𝑐𝑡) = 𝐶𝑇𝑓((𝑥𝑎 + Δ𝑥𝑐) − 𝑐𝑡)

= 𝐶𝑇𝑓(𝑥𝑐 − 𝑐𝑡) = 𝐶𝑇𝐝𝑐 = 𝐝𝑐𝛼𝑐,

(4.25)

- 51 -

where the function h represents the damage transmitted ultrasonic waves. Δ𝑡𝑑(= Δ𝑥𝑑/𝑐𝑑) is the transmission

time of the waves through the damage, Δ𝑥𝑑 represents the width of the damage, and 𝑐𝑑 (< 𝑐) is the wave speed

in the damage. From Equation (4.20), h can be represented by 𝐶𝑇 and f, which are the transmission coefficient

and the direct incident waves, respectively, without any damage transmission. The basis 𝐝𝑐 will have an ampli-

tude of 𝛼𝑐 = 𝐶𝑇. Note that the ultrasonic waves are delayed while transmitting through the damage because 𝑥𝑐

is larger than 𝑥𝑎.

An important observation in Figure 4.7 is that the measured response in the spatial ultrasonic domain

includes the basis 𝐝𝑎, which corresponds to the intact incident waves, if no damage is observed between the

excitation point and sensing point (either no damage in the specimen or the excitation point is ‘in front of the

damage’). However, this basis is shifted if the excitation point is ‘behind the damage’. Therefore, the damage

presence within the direct wave propagation path can be easily identified by performing spatial ultrasonic trans-

formation and verifying the shift of the basis that corresponds to the intact incident waves.

Figure 4.7. Comparison of the transformed time-domain ultrasonic responses generated at (a) intact, (b) in

front of the damage, and (c) behind the damage.

- 52 -

4.4. Estimation of spatial domain ultrasonic responses using a spatial ultrasonic diction-

ary

In Chapter 3.1.2, we need to obtain a wavefield image signal 𝐬𝐈 (𝑁I ∙ 𝑀I × 1 vector), which represents a spatial

distribution of the ultrasonic waves in the imaging region at a specific time step. 𝐬𝐈 is the vectorization result of

a wavefield image for the imaging region of size 𝑁I (width) × 𝑀I (height). Note that this wavefield image

signal represents a spatial information of wave propagation instead of a time-domain ultrasonic response. As

shown in Equation (4.26), the wavefield image signal can be represented as a linear combination of bases.

𝐬𝐈 = 𝐃CS𝛂𝐈,

(4.26)

𝐃CS = {𝐝1CS, 𝐝2

CS, … , 𝐝𝑁I∙𝑀ICS }.

Designing a dictionary is also an important task for Equation (14). 𝐃CS (𝑁I ∙ 𝑀I × 𝑇 matrix) is a dic-

tionary for this transformation, which consists of 𝐝𝑖CS

bases (𝑖 = 1, … , 𝑇). Each basis 𝐝𝑖CS has the same dimen-

sions as 𝐬𝐈 and corresponds to the spatial wave distributions in the imaging region at each time step 𝑡𝑖 (𝑖 =

1, … , 𝑇). The bases for 𝐃CS is reconstructed from D (𝑇 × 𝐿 matrix) in Equations (4.21) and (4.22) as

𝐝𝑖,𝑛CS = 𝐴𝑒−𝑗𝑘(𝑥𝑛−𝑐𝑡𝑖) = 𝑓(𝑥𝑛 − 𝑐𝑡𝑖) = 𝑓(𝑥𝑙 − 𝑐𝑡𝑖) = 𝐝𝑙,𝑖 , (4.27)

where 𝑥𝑛 is the distance between the nth scanning point in the imaging region and the sensing point. The nth

element of 𝐝𝑖CS is obtained from the lth basis of D, which corresponds to the ultrasonic response generated at

𝑥𝑙 = 𝑥𝑛. 𝐝𝑖,𝑛CS represents the ultrasonic response of the nth scanning point in the imaging region at time step 𝑡𝑖.,

This is identical to 𝐝𝑙,𝑖, which is the ultrasonic response at time step 𝑡𝑖 generated at the distance 𝑥𝑙 = 𝑥𝑛 from

the sensing point.

As in Equation (4.4), the following problem is solved to obtain 𝛂𝐼

min‖𝛂I‖1 𝑠. 𝑡. 𝐬�̅� = �̅�CS𝛂I = 𝚲𝐃CS𝛂I, (4.28)

where 𝐬�̅� (𝑃 × 1 vector) is the scanned part of 𝐬𝐈 and P is the number of scanning points in the imaging region.

𝚲 (𝑃 × 𝑁I ∙ 𝑀I matrix) is a measurement matrix that indicates the scanning point locations, e.g., yellow and green

points in the imaging region in Figure 6. Because 𝐬�̅�, 𝚲, 𝐃𝐂𝐒 are known, 𝛂𝐈 is obtained using the basis pursuit

approach. Then, 𝑠𝐼 can be obtained from Equation (4.28) with the designed 𝐃CS and obtained 𝛂𝐈.

By repeating this process for every time step, the ultrasonic wave propagation data in the imaging

regions are obtained. The estimated unscanned inspection points are represented as purple points in Step 2 of

Figure 3.2. Then, wavefield images are created for the imaging region by matricizing 𝐬𝐈 at each time step.

- 53 -


In this chapter, ultrasonic wave propagations are theoretically presented and they are represented using a spatial

ultrasonic dictionary. The basis pursuit approach is used for transformation of a time domain ultrasonic signal to

a spatial domain signal. This approach minimizes the sum of absolute values of nonzero elements in the trans-

formed representation. This approach shows better performance in transformation in comparison with matching

pursuit approach, which is more widely known in the ultrasonics field.

For a given structure, a sparse ultrasonic dictionary is constructed by collecting ultrasonic responses

from a structure. This approach assures sparse representations of an arbitrary ultrasonic measurement naturally as

the dictionary itself is a set of ultrasonic signals from the structure. Even after introducing a damage, the corre-

sponding responses can be represented as a sum of bases from the dictionary.

An important observation is that the measured response in the spatial ultrasonic domain includes the

basis which corresponds to the intact incident waves, if no damage is observed between the excitation point and

sensing point (either no damage in the specimen or the excitation point is ‘in front of the damage’). However, this

basis is shifted if the excitation point is ‘behind the damage’. Therefore, the damage presence within the direct

wave propagation path can be easily identified by performing spatial ultrasonic transformation and verifying the

shift of the basis that corresponds to the intact incident waves.

The spatial ultrasonic transformation can represent not only time domain responses but also spatial do-

main responses as a weight sum of bases. This is used to estimate the unscanned responses of the spatial ultrasonic

wave distribution, known as compressed sensing. The ultrasonic wave propagation images around the damage

region can be reconstructed using the compressed sensing.

- 54 -

Chapter 5. Numerical Validation of the Developed Techniques

5.1. Simulation description

(a)

(b)

Figure 5.1. Schematic of a numerical model for numerical validations of the developed techniques: (a) Geom-

etry of the simulated aluminum plate with a crack. A crack with 1 cm height, 0.01 cm width, 0.15 cm depth is

introduced to the plate. (b) Finite element meshes for thermal wave / ultrasonic wave region.

The proposed techniques in Chapter 3 are validated via a numerical simulation that is performed with the com-

mercial finite element software program COMSOL Multiphysics. The simulated aluminum plate model has di-

mensions of 15 × 15 × 0.3 cm3, as displayed in Figure 5.1. A notch, with length, width and depth dimensions of 1

cm, 0.01 cm, and 0.15 cm, respectively, is introduced on the opposite side of the scanned surface. A square region

of 4 cm around the notch is established as an inspection region. Detailed material properties for this aluminum

plate model are given in Table 5.1.

- 55 -

To represent the phenomenon that occurs after the incident of a laser beam pulsed onto the plate, the

model is divided into two parts: the thermal wave region and the ultrasonic wave region. First, thermoelasticity is

used to model ultrasonic wave generations due to the thermal stress induced by a pulse laser. Second, an interac-

tion between solid and acoustic is used for solving the ultrasonic wave propagation within a material. The entire

model is meshed with triangular elements, and a multi-scale element length is used to reduce the computation

burden. In this study, a multi-physics simulation in the time-domain analysis is considered. Please refer to Chapter

2.1.2 for more details about the numerical modelling of ultrasonic wave propagation generated using a pulse laser.

Table 5.1. Material properties of aluminum used in this numerical simulation.

Density

𝜌 (kg/m3)

Young’s

modulus

𝐸

(GPa)

Poisson’s

ratio

𝜈

Coefficient

of thermal

expansion

𝛼𝑡

(K-1)

Thermal

conductivity

𝐾

(W/(m·K))

Heat

capacity at

constant

pressure

𝐶𝑝

(J/(kg·K))

Reflection

coefficient

𝑅

2700 68.9 0.33 2.34×10-5 170 900 0.95

A spatial ultrasonic dictionary was constructed from an intact aluminum plate model. Bases in the

spatial ultrasonic dictionary were generated by moving the excitation point from a 0.4 cm distance to the sensing

point to 7.5 cm with 0.1 cm resolution to create 72 bases. By spline interpolations between two adjacent bases,

the dictionary was constructed from a total of 721 bases with a 0.01 cm spatial resolution. Each base was sampled

for 25 μs with a 5.12 MHz sampling frequency. A bandpass filter with a low cutoff frequency of 50 kHz and a

high cutoff frequency of 450 kHz was used to capture ultrasonic information in the interested frequency bandwidth

in the measured signals.

For a binary search, each scanning was performed with identical parameters in the basis construction

process. The spatial scanning resolution for a binary search was 0.2 cm for the x and y directions, which divides

the inspection region into 21×21 grids. For combined binary search and compressed sensing, the sensing laser

beam vertically moved according to the vertical location of the current scanning point while the excitation laser

beam scanned the inspection region. The horizontal distance was fixed to 1 cm from the sensing points to the left

edge of the inspection region. For fixed pitch-catch distance scanning, the pitch-catch distance was fixed to 2 cm,

leading K=10.

- 56 -

5.2. Spatial ultrasonic transformation

5.2.1. Spatial ultrasonic transformation for the excitation point in front of the damage and behind the dam-

age

(a) (b)

Figure 5.2. Determination of the location of the laser excitation point with respect to damage using the pro-

posed spatial ultrasonic transformation: (a) The excitation point is in front of the damage, and (b) the excitation

point is behind the damage.

First, the simulated time response is transformed into the spatial ultrasonic domain to determine whether the laser

excitation point is positioned either ‘in front of’ or ‘behind’ the damage, as shown in Figure 5.2.

(a)

(b)

Figure 5.3. Comparison of ultrasonic responses measured from an intact plate (blue solid line) and a damage

plate (red dashed line) in the spatial ultrasonic domain: (a) When the excitation point is in front of the damage

(Figure 5.2 (a)), and (b) when the excitation point is behind the damage (Figure 5.2 (b)).

Figure 5.3 compares the ultrasonic responses obtained from intact (blue solid line) and damage (red

dashed line) conditions of the plate in the proposed spatial ultrasonic domain. Figure 5.3 (a) compares the intact

and damage conditions when the laser excitation point is placed in front of the damage. When the excitation is

- 57 -

placed in front of the damage, the amplitude of the initial basis that corresponds to the incident waves does not

change and only additional bases with non-zero amplitudes appear. When the excitation is positioned behind the

damage (Figure 5.3 (b)), the location of the maximum peak amplitude is shifted away from the initial basis. By

examining the amplitude of the basis that corresponds to the incident waves, the position of the laser excitation

point with respect to the damage is determined, which enables a repeated binary search along a single inspection

line, as described in the following subchapter.

5.2.2. Transition point identification

(a)

(b)

Figure 5.4. Variation of the basis pursuit (BP) index value as the excitation point moves from a 10 mm distance

from the sensing point to 50 mm: (a) when there is no damage between the moving excitation point and the

fixed sensing point, and (b) when there is a notch at 24 mm distance from the fixed sensing point.

To find the transition point in each inspection line, a basis pursuit (BP) index is defined. The BP index is the

amplitude of the basis that corresponds to the initial incident waves. Figure 5.4 maps the BP index values as the

excitation point moves from a 10 mm distance from the sensing point to 50 mm. When there is no damage between

the excitation point and the sensing points, the BP index values are similarly large for all scanning points. When

a notch damage is observed, the BP index value significantly decreases as the excitation point passes through the

damage location, that is, when the excitation point is in front of the damage, the BP index that corresponds to the

excitation point is large. When the excitation point moves behind the damage, the BP index value becomes prac-

tically zero. In Figure 5.4, all 21 scanning points in a single inspection line are scanned for illustration. In practice,

only five (⌈log2 21⌉) scanning points are required to identify the transition point using the proposed binary search.

- 58 -

5.2.3. Sensitivity of the spatial ultrasonic transformation to damages with various width, thickness and

depth

Figure 5.5. Schematic of an aluminum plate model with a notch damage with thickness t, width w, and tip

depth d.

The sensitivity of the spatial ultrasonic transformation to damages with various width, thickness and depth is

discussed in this subchapter. An aluminum plate model with a notch damage is modeled as Figure 5.5. The simu-

lated time response is transformed into the spatial ultrasonic domain. Then it is checked whether the laser excita-

tion point is determined as ‘behind’ the damage (the notch is identified) with various combinations of notch ge-

ometry.

Figure 5.6 presents the notch identification results. Each figure reveals the notch identification results

for various combinations of notch thickness and notch tip depth with given notch width. Green, red, and white

square indicates the notch identified combination, the notch unidentified combination and the unavailable combi-

nation, respectively. The spatial ultrasonic transformation identifies the notch easier as it becomes thicker. The

notch closer to the scanning surface is easily identified even with a small thickness, and the notch on the other

surface is also identified while the notch inside the specimen is harder to identify. This process is repeated for six

cases of notch width from 0.001 to 0.5. For a 5 mm thickness aluminum plate with 200 kHz A0 Lamb wave

propagation (wavelength 𝜆 =10 mm), 𝑤

𝜆= 0.001 is corresponding to 10 μm width notch and the notch combi-

nations thicker than 2 mm are identified. 𝑤

𝜆= 0.1 is corresponding to 1 mm width notch and the notch combina-

tions thicker than 1 mm are identified. For the same plate with 1000 kHz A0 Lamb wave propagation (𝜆 =3 mm),

𝑤

𝜆= 0.001 is corresponding to 3 μm width notch.

An important observation in Figure 5.6 is that the spatial ultrasonic transformation can identify a notch

more effectively (1) as the notch is thicker; (2) as the notch is wider; (3) as the notch is closer to the surface and

(4) as the probing ultrasonic frequency is higher.

- 59 -

(a) (b)

(c) (d)

(e) (f)

Figure 5.6. Sensitivity of spatial ultrasonic transformation for damage detection. Notch identification results

for various combinations of notch thickness and notch tip depth are presented with six cases of notch width:

(a) 𝑤

𝜆= 0.001, (b)

𝑤

𝜆= 0.005, (c)

𝑤

𝜆= 0.01, (d)

𝑤

𝜆= 0.05, (e)

𝑤

𝜆= 0.1, and (f)

𝑤

𝜆= 0.5.

- 60 -

5.3. Damage detection results using the developed techniques

5.3.1. Combined binary search and compressed sensing

(a) (b)

Figure 5.7. Binary search for a transition line detection: (a) layout of scanning points selected by the binary

search and (b) the detected transition line.

(a) (b)

Figure 5.8. Damage quantification via compressed sensing. (a) The imaging region is represented as a dashed

box and (b) the corresponding notch quantification result is presented.

Figure 5.7 (a) illustrates the outcome of the binary search. The 40 mm by 40 mm square area represents the

inspection region, whereas the sensing points had been located at a 10 mm distance from the left edge of the

square area. When the scanning point is located behind the damage, the scanning point is colored in yellow. The

- 61 -

scanning point in front of the damage is marked in green. The transition points are colored in yellow and marked

with crosses. When the 40 mm by 40 mm area is scanned with a 2 mm resolution, a total of 441 (21 x 21) inspection

points are generated. Only 32 scanning points (7.3%) are needed to identify the transition line. The transition line

in Figure 5.7 (b) is obtained by connecting all transition points in Figure 5.7 (a). This reduction rate of 92.7% is

close to the theoretical reduction rate of 91.6% in Equation (3.4).

(a)

(b)

Figure 5.9. Comparison of the original full-field and the reconstructed wave propagation snapshots that corre-

spond to the dashed box in Figure 5.8 (a): (a) original full-field wave propagation images and (b) reconstructed

wave propagation images

Next, compressed sensing is performed for detailed damage quantification. Ultrasonic wave propaga-

tion is visualized by estimating the responses in unmeasured points using compressed sensing. The imaging region

is indicated with a dashed box in Figure 5.8 (a). Reconstructed wave propagation snapshots are visualized in

Figure 5.9 (b) and compared with the original full-field wave propagation snapshots (Figure 5.9 (a)).

As represented in Figure 5.9, the reconstructive wave propagation images reveal satisfactory agreement

with the original full-field wave propagation images. Waves propagate from the left edge of the region towards

the right, reflected from the notch, and propagate back to the left side. The actual damage quantification is

achieved by applying a standing wave energy analysis [16]. By this analysis, the notch is identified and visualized

in Figure 5.8 (b). The red color in Figure 5.8 (b) represents the damage points, where their damage-induced stand-

ing wave energy is greater than the threshold value obtained from the standing wave energy analysis. The numer-

ical simulation that is presented in this chapter validates that the proposed damage detection technique using

- 62 -

combined binary search and compressed sensing can detect and visualize damage with a significantly reduced

number of scanning points.

5.3.2. Binary search with fixed pitch-catch distance scanning

Figure 5.10 illustrates the outcome of the binary search with fixed pitch-catch distance scanning. The 40 mm by

40 mm square area represents the inspection region. Each inspection line is divided into two divisions for damage

region identification. Only 14 scanning points are needed to identify the damage regions, indicated with dashed

boxes in Figure 5.10 (a).

(a) (b)

(c) (d)

Figure 5.10. Binary search with fixed pitch-catch distance scanning: (a) identification of damage regions, (b)

layout of scanning points for right transition point detection, (c) layout of scanning points for left transition

point detection, and (d) the corresponding notch quantification result.

- 63 -

eand left transition point detection, respectively. In Figure 5.10 (c), it can be observed that there is no

left transition point as the notch in this example is corresponding to a single inspection point in widthwise. 18

additional scanning points are required to identify the transition points.

The notch is identified and visualized in Figure 5.10 (d) from the identified transition points. The red

color in Figure 5.10 (d) represents the damage points, where they are between the left and right transition points.

When the 40 mm by 40 mm area is scanned with a 2 mm resolution, a total of 441 (21 x 21) inspection points are

generated. Only 32 scanning points (7.3%) are needed to quantify the notch. This reduction rate of 92.7% is higher

than the theoretical reduction rate of 83.2% in Equation (3.10) as this equation considers the worst case. The

numerical simulation that is presented in this chapter validates that the proposed damage detection technique using

binary search with fixed pitch-catch distance scanning can detect and visualize damage with a significantly re-

duced number of scanning points.

- 64 -

5.4. Discussion on the practical applicability of the developed techniques

(a)

(b)

Figure 5.11. Comparison of the number of (a) scanning points p and (b) the reduction rate R using the conven-

tional full-field wave propagation imaging (black solid line), combined binary search and compressed sensing

(blue dashed line), and binary search with fixed pitch-catch distance scanning (red dotted line). The reduction

rate R increases as the size of the inspection region increases.

Figure 5.11 shows the number of scanning points p and the reduction rate R for increasing N. Black solid line,

blue dashed line, and red dotted line represent the conventional full-field wave propagation imaging, combined

binary search and compressed sensing, and binary search with fixed pitch-catch distance scanning, respectively.

Note that the reduction rate increases as the inspection region increases or the spatial resolution is improved.

Please note that the represented numbers are considering the worst case, and the actual p would be smaller and

the actual R would be higher than them as shown in Chapter 5.3.

Assuming a 50 mm square inspection region with a 1 mm spatial resolution, a full-field wave propaga-

tion imaging requires 2500 scanning points, which is identical to the number of inspection points in this inspection

- 65 -

region. The combined binary search and compressed sensing requires only 64 scanning points for a damage case

achieving a 97.5% reduction in scanning points. The reduction rate decreases to 88% for the intact case. The

binary search with fixed pitch-catch distance scanning requires 128 scanning points for a damage case achieving

a 95% reduction in scanning points.

However, the actual scanning time using binary search with fixed pitch-catch distance scanning would

be further reduced as the number of averaging is smaller for this technique. This technique does not require a large

number of averaging as the distance between the excitation point and the sensing point is small enough to have a

high signal-to-noise ratio (SNR). But this falls into a limitation in sensitivity variation as the sensing point should

be moved during inspection. As discussed in Section 2.2, the performance of a LDV is highly dependent on surface

conditions and an incident angle of the LDV beam. Additional surface treatment for large region might be required

to use this technique in field. This technique works more appropriately with the microscopic scanning system in

Chapter 2.3.2 therefore.

- 66 -


In this chapter, the developed damage detection techniques are numerically validated with a cracked aluminum

plate model. The numerical model is analyzed with a commercial finite element software COMSOL Multiphysics.

To represent phenomenon occurring after the incident of laser beam pulsed onto the plate, the model is divided

into two parts, thermal wave region and ultrasonic wave region.

Damage detection is effectively performed for the tested model. With this example, it is validated that

(1) the proposed damage detection techniques can identify and quantify damage with a reduced number of scan-

ning points, less than 10% of conventional wavefield imaging techniques; and (2) even a 0.01 cm width notch can

be identified and quantified.

The inspection speed of the developed techniques is also numerically discussed. The required number

of scanning points of the developed techniques is significantly reduced from the one of the conventional full-field

wave propagation imaging technique. The reduction rate increases as the inspection region increases or the spatial

resolution is improved. Though the theoretical inspection speed of the binary search using fixed pitch-catch dis-

tance scanning is slower than the combined binary search and compressed sensing, the actual number of meas-

urements are similar.

- 67 -

Chapter 6. Experimental Validation of the Developed Techniques

using a Macroscopic Laser Ultrasonic Scanning System

6.1. Hardware setup

Figure 6.1. Overview of the macroscopic laser ultrasonic scanning system.

Figure 6.1 shows an overview of the macroscopic laser ultrasonic scanning system used in the experiments. A

diode pumped Q-switched Nd:YAG pulse laser (Quantel Centurion+) radiates 532 nm wavelength laser pulses.

The width of the laser pulse is 12 ns, and this device emits 3 mm diameter laser pulses up to 100 times per second.

Optical modules, which are covered by a black box, guide radiated pulse laser beams into the galvanometer (Scan-

lab hurrySCAN 20). This black box prevents an accidental reflection and scattering of high-power pulse laser

beams outside the module. The laser beam was aimed at the desired points by controlling the galvanometer. The

lens in front of the galvanometer focuses the laser beam diameter to 0.5 mm on the specimen surface. The ultra-

sonic responses were measured by a commercial scanning LDV (Polytec PSV-400). The distance from the target

specimen to the galvanometer and the LDV was 1 m and 1.1 m, respectively. The LDV distance is set to avoid

the intermode beating aforementioned in Chapter 2.2.

- 68 -

6.2. Hidden notch detection in an aluminum plate

6.2.1. Experimental setup

Figure 6.2. An aluminum plate with a hidden notch. A notch with 2 cm height, 0.2 cm width, 0.15 cm depth is

introduced to the plate. This hidden notch is introduced on the opposite side of the scanning surface.

The first tested specimen is an aluminum plate with a notch. This plate has dimensions of 30 × 30 × 0.3 cm3, as

displayed in Figure 6.2. A 2 cm long, 0.2 cm wide and 0.15 cm deep notch is introduced on the opposite side of

the scanning surface. A square region of 5 cm around the notch is established as an inspection region.

A spatial ultrasonic dictionary was constructed from an intact aluminum plate. Bases in the spatial

ultrasonic dictionary were generated by moving the excitation point from a 1 cm distance to the sensing point to

10 cm with 0.1 cm resolution to create 91 measurements. By spline interpolations between the two adjacent bases,

the dictionary was constructed from a total of 901 bases with a spatial resolution of 0.01 cm. The peak energy and

power of the pulse laser was 3 mJ and 0.3 MW, respectively. Each basis was sampled for 37.5 μs with a 2.56 MHz

sampling frequency. A bandpass filter with a low cutoff frequency of 50 kHz and a high cutoff frequency of 450

kHz was used to capture ultrasonic information in the interested frequency bandwidth. Each basis collection was

averaged 100 times to improve their signal-to-noise ratios.






edge of the inspection region. For fixed pitch-catch distance scanning, the pitch-catch distance was fixed to 2.5

cm, leading K=25. Each measurement was averaged only 30 times for this scanning strategy.

- 69 -

6.2.2. Notch detection using combined binary search and compressed sensing

(a) (b)

Figure 6.3. Binary search for a transition line detection: (a) layout of scanning points selected by the binary

search and (b) the detected transition line.

Figure 6.3 illustrates the outcome of a binary search. The 50 mm × 50 mm square area represents the inspection

region, whereas the sensing points have been located at a 10 mm distance from the left edge of the square area.

When the 50 mm × 50 mm area is scanned with a 1 mm resolution, a total of 2601 (51 × 51) inspection points are

generated. Only 55 scanning points (2.1%) are required to identify the transition line. The transition line in Figure

6.3 (b) is obtained by connecting all transition points in Figure 6.3 (a). This result identifies the left border of the

notch perfectly.

To obtain the results shown in Figure 6.3, 0.9 minutes of scanning time is required. Note that this

scanning time is obtained with (1) averaging 100 times for each scanning, (2) 100 Hz repetition rate of a pulse

laser, and (3) 0.1 cm scanning resolution for a 5 cm square region. If (1) less averaging was employed, (2) a faster

pulse laser was available, or (3) a lower scanning resolution was acceptable, the scanning time could have been

further reduced.

Next, compressed sensing is performed for detailed damage quantification. The imaging region is in-

dicated with a dashed box in Figure 6.4 (a). Reconstructed wave propagation snapshots are visualized in Figure

6.5 (b) and compared with the original full-field wave propagation snapshots (Figure 6.5 (a)).

- 70 -

(a) (b)


box and (b) the corresponding crack quantification result is presented.

(a)

(b)

Figure 6.5. Reconstructed wave propagation at different time steps. The image is corresponding to the dashed

box in Figure 6.4 (a): (a) Reconstructed wave propagation. (b) Original wave propagation.

As represented in Figure 6.5, the reconstructed wave propagation images show satisfactory agreement


- 71 -

the right edge of the region, reflected from the notch, and propagate back to the left side. From a standing wave

analysis, the notch is identified and visualized in Figure 6.4 (b) using red points. This result visualizes the notch

perfectly. The experiment presented in this subchapter validates that the proposed damage detection technique

using combined binary search and compressed sensing can detect and visualize a damage on the opposite side of

the scanning surface.

6.2.3. Notch detection using binary search with fixed pitch-catch distance scanning

(a) (b)

(c) (d)



point detection, and (d) the corresponding notch quantification result.

- 72 -

Figure 6.6 illustrates the outcome of the binary search with fixed pitch-catch distance scanning. The 50 mm by 50

mm square area represents the inspection region. Each inspection line is divided into two divisions for damage



Figures 6.6 (b) and 6.6 (c) represents the layout of scanning points for right transition point detection

and left transition point detection, respectively. 50 additional scanning points are required to identify the transition

points.

The notch is identified and visualized in Figure 6.6 (d) from the identified transition points. The red


When the 50 mm by 50 mm area is scanned with a 1 mm resolution, a total of 2601 (51 × 51) inspection points

are generated. Only 72 scanning points (2.8%) are needed to quantify the notch. This reduction rate of 97.2% is

lower than Chapter 6.2.2, but the actual scanning time is only 0.4 minutes as a small number of averagings is

applied. Note that this scanning time is obtained with (1) averaging 30 times for each scanning, (2) 100 Hz repe-

tition rate of a pulse laser, and (3) 0.1 cm scanning resolution for a 5 cm square region. If (1) less averaging was

employed, (2) a faster pulse laser was available, or (3) a lower scanning resolution was acceptable, the scanning

time could have been further reduced. The experiment presented in this subchapter validates that the proposed

damage detection technique using binary search with fixed pitch-catch distance scanning can detect and visualize

a damage on the opposite side of the scanning surface.

- 73 -

6.3. Fatigue crack detection in an aluminum plate


Figure 6.7. An aluminum plate with a fatigue crack. A crack with a height of 1.8 cm and width of 10 μm is

introduced to the plate by repetitive tensile loadings.

The second tested specimen is an aluminum plate with a fatigue crack. This plate has dimensions of 30 × 12 × 0.3

cm3, as displayed in Figure 6.7. A through-the-thickness fatigue crack with a length of 1.8 and width of 10 μm

was introduced by repetitive tensile loadings with a universal testing machine. A square region of 2 cm around

the crack is established as an inspection region.

A spatial ultrasonic dictionary was constructed from an intact aluminum plate. Bases in the spatial

ultrasonic dictionary were generated by moving the excitation point from a 1 cm distance to the sensing point to

7 cm with 0.1 cm resolution to create 61 measurements. By spline interpolations between the two adjacent bases,

the dictionary was constructed from a total of 601 bases with a spatial resolution of 0.01 cm. The peak energy and

power of the pulse laser was 3 mJ and 0.3 MW, respectively. Each basis was sampled for 37.5 μs with a 2.56 MHz

sampling frequency. A bandpass filter with a low cutoff frequency of 50 kHz and a high cutoff frequency of 450

kHz was used to capture ultrasonic information in the interested frequency bandwidth. Each basis collection was

averaged 100 times to improve their signal-to-noise ratios.






- 74 -

edge of the inspection region. For fixed pitch-catch distance scanning, the pitch-catch distance was fixed to 2 cm,

leading K=20. Each measurement was averaged only 30 times for this scanning strategy.

6.3.2. Crack detection using combined binary search and compressed sensing

(a) (b)

Figure 6.8. Binary search for a transition line detection: (a) layout of scanning points that were selected by the

binary search and (b) the detected transition line.


region, whereas the sensing points have been located at a 10 mm distance from the left edge of the square area.


generated. Only 37 scanning points (8.4%) are required to identify the transition line. The transition line in Figure

6.8 (b) is obtained by connecting all transition points in Figure 6.8 (a). This result identifies the left border of the

crack with high accuracy. The discrepancies between the actual left border of the crack and the transitional line

are less than the scanning resolution of 1 mm. To obtain the results shown in Figure 6.8, 0.6 minutes of scanning

time is required. This scanning time can be reduced as mentioned in Chapter 6.2.



6.10 (b) and compared with the original full-field wave propagation snapshots (Figure 6.10 (a)).

- 75 -

(a) (b)


box and (b) the corresponding crack quantification result is presented.

(a)

(b)

Figure 6.10. Reconstructed wave propagation at different time steps. The image is corresponding to the dashed

box in Figure 6.9 (a): (a) Reconstructed wave propagation. (b) Original wave propagation.

- 76 -

As represented in Figure 6.10, the reconstructed wave propagation images show satisfactory agreement


the right edge of the region, reflected from the crack, and propagate back to the left side. From a standing wave

analysis, the crack is identified and visualized in Figure 6.9 (b) using red points. This result visualizes the crack

with a high accuracy. The experiment presented in this subchapter validates that the proposed damage detection

technique using combined binary search and compressed sensing can detect and visualize any damage with a

significantly reduced number of scanning points, even for a fatigue crack.

6.3.3. Crack detection result using binary search with fixed pitch-catch distance scanning

(a) (b)

(c) (d)



point detection, and (d) the corresponding crack quantification result.

- 77 -






and left transition point detection, respectively. In Figure 6.11 (c), it can be observed that there is no left transition

point as the crack in this example is corresponding to a single inspection point in widthwise. 56 additional scan-

ning points are required to identify the transition points.

The crack is identified and visualized in Figure 6.11 (d) from the identified transition points. The red


When the 20 mm by 20 mm area is scanned with a 1 mm resolution, a total of 441 (21 x 21) inspection points are

generated. Only 65 scanning points (14.7%) are needed to quantify the notch. This reduction rate of 85.3% is

lower than Chapter 6.3.2, but the actual scanning time is only 0.3 minutes. This scanning time can be reduced as

mentioned in Chapter 6.2. The experiment presented in this subchapter validates that the proposed damage detec-

tion technique using binary search with fixed pitch-catch distance scanning can detect and visualize any damage

with a significantly reduced number of scanning points, even for a fatigue crack.

- 78 -

6.4. Delamination detection in a carbon fiber reinforced plastic (CFRP) plate


Figure 6.12. A CFRP plate with a delamination. A delamination with 1 cm diameter is introduced to the plate

with an impact.

The third test specimen for experimental validation is a carbon fiber-reinforced plastic (CFRP) plate. This plate

has dimensions of 27.5 × 27.5 × 0.3 cm3, as displayed in Figure 6.12. The plate is composed of IM7 graphite

fibers with 977-3 resin material and 12 plies with a layup of [0/±45/0/±45]S. A 1 cm diameter delamination was

introduced at the center of the plate by an impact. A square region of 5 cm around the delamination is established

as an inspection region. The delamination is located in the top-left corner of the inspection region.

A challenge of the ultrasonic wave analysis of an anisotropic material is that the ultrasonic wave speeds

and propagation characteristics vary with their propagation direction. To prevent this variance, the inspection lines

are oriented in the same direction with each other. A dictionary only for this single direction is sufficient to apply

the proposed accelerated laser scanning technique to anisotropic specimens.

Because no intact CFRP plate is available, a spatial ultrasonic dictionary was constructed from an intact

part of the plate. Bases in the spatial ultrasonic dictionary were generated by moving the excitation point from a

1 cm distance to the sensing point to 10 cm with 0.1 cm resolution to create 91 measurements. Using spline

interpolations between the two adjacent bases, the dictionary was constructed from a total of 901 bases with a

0.01 cm spatial resolution. The peak energy and power of the pulse laser was 3 mJ and 0.3 MW, respectively.

Each basis was sampled for 37.5 μs with a 2.56 MHz sampling frequency. A bandpass filter with a low cutoff

frequency of 50 kHz and a high cutoff frequency of 450 kHz was employed to capture ultrasonic information in

- 79 -

the interested frequency bandwidth. Each basis collection was averaged 100 times to improve their signal-to-noise

ratios.



the inspection region into 51× 51 grids. For combined binary search and compressed sensing, the sensing laser





6.4.2. Delamination detection using combined binary search and compressed sensing


region, whereas the sensing points had been located at a 10 mm distance from the left edge of the square area.


generated. Only 75 scanning points (2.9%) are needed to identify the transition line, which corresponds to 1.3

minutes of scanning time. This scanning time can be reduced as mentioned in Chapter 6.2. The transition line in

Figure 6.13 (b) is obtained by connecting all transition points in Figure 6.13 (a).

(a) (b)

Figure 6.13. Binary search for a transition line detection: (a) layout of the scanning points selected by the


- 80 -

(a) (b)

Figure 6.14. Damage quantification through the compressed sensing. (a) The imaging region is represented as

a dashed box and (b) the corresponding delamination quantification result is presented.

(a)

(b)

Figure 6.15. Comparison of the original full-field and the reconstructed wave propagation snapshots that cor-

respond to the dashed box in Figure 6.14 (a): (a) Original full-field wave propagation and (b) reconstructed

wave propagation.



6.15 (b) and compared with the original full-field wave propagation snapshots (Figure 6.15 (a)). As represented

in Figure 6.15, the reconstructed wave propagation images show satisfactory agreement with the original full-

field wave propagation images. Waves propagate from the left edge of the region toward the right and interact

inside the delamination. In the standing wave analysis, the delamination is identified and visualized in Figure 6.14

(b) with red points. The visualized delamination is comparable to the conventional laser ultrasonic imaging results,

as shown in Figure 6.16. The experiment in this subchapter validates that the proposed damage detection technique

- 81 -

using combined binary search and compressed sensing can detect and visualize damage, even for anisotropic

specimens.

Figure 6.16. Delamination visualized using conventional full-field wave propagation imaging.

6.4.3. Delamination detection using binary search with fixed pitch-catch distance scanning







points.

The delamination is identified and visualized in Figure 6.17 (d) from the identified transition points.

The red color in Figure 6.17 (d) represents the damage points, where they are between the left and right transition

points. When the 50 mm by 50 mm area is scanned with a 1 mm resolution, a total of 2601 (51 x 51) inspection

points are generated. Only 71 scanning points (2.8%) are needed to quantify the delamination, corresponding to

1.2 minutes of scanning time. This scanning time can be reduced as mentioned in Chapter 6.2. The experiment

presented in this subchapter validates that the proposed damage detection technique using binary search with fixed

pitch-catch distance scanning can detect and visualize damage, even for anisotropic specimens.

- 82 -

(a) (b)

(c) (d)



point detection, and (d) the corresponding delamination quantification result.

- 83 -

6.5. Delamination detection in a 10 kW glass fiber reinforced plastic (GFRP) wind tur-

bine blade


Figure 6.18. A GFRP wind turbine blade with a delamination. A delamination with 1.5 cm diameter is simu-

lated by inserting a Teflon tape during fabrication.

The last test specimen for experimental validation is a 10 kW wind turbine blade. This is manufactured by Human

Composites Ltd., a Korean wind turbine blade manufacturer. This blade has a dimension of 3.5 m × 0.45 m with

the maximum thickness of 0.3 m, and is composed of six plies of glass fiber reinforced plastic (GFRP), with a

layup of [0/±45]S. The elastic modulus E12, shear modulus G12, and poisson ratio ν12 are 24.65 GPa, 8.52 GPa

and 0.476, respectively. A 1.5 cm diameter Teflon tape was inserted between the third and fourth ply during

fabrication of the blade to simulate internal delamination. A square region of 5 cm around the delamination is

established as an inspection region. The delamination is located in the bottom-left corner of the inspection region.

Because no intact blade is available, a spatial ultrasonic dictionary was constructed from an intact part

of the blade. Bases in the spatial ultrasonic dictionary were generated by moving the excitation point from a 1 cm

distance to the sensing point to 10 cm with 0.1 cm resolution to create 91 measurements. Using spline interpola-

tions between the two adjacent bases, the dictionary was constructed from a total of 901 bases with a 0.01 cm

spatial resolution. The peak energy and power of the pulse laser was 3 mJ and 0.3 MW, respectively. Each basis

was sampled for 37.5 μs with a 2.56 MHz sampling frequency. A bandpass filter with a low cutoff frequency of

50 kHz and a high cutoff frequency of 450 kHz was employed to capture ultrasonic information in the interested

frequency bandwidth. Each basis collection was averaged 100 times to improve their signal-to-noise ratios.




- 84 -





6.5.2. Delamination detection using combined binary search and compressed sensing


region, whereas the sensing points had been located at a 10 mm distance from the left edge of the square area.


generated. Only 38 scanning points (5%) are needed to identify the transition line, which corresponds to 0.6

minutes of scanning time. This scanning time can be reduced as mentioned in Chapter 6.2. The transition line in

Figure 6.19 (b) is obtained by connecting all transition points in Figure 6.13 (a).

(a) (b)

Figure 6.19. Binary search for a transition line detection: (a) layout of the scanning points selected by the




6.21 (b) and compared with the original full-field wave propagation snapshots (Figure 6.21 (a)). As represented

in Figure 6.21, the reconstructed wave propagation images show satisfactory agreement with the original full-

field wave propagation images. Waves propagate from the left edge of the region toward the right and interact

inside the delamination. In the standing wave analysis, the delamination is identified and visualized in Figure 6.20

(b) with red points. The experiment in this subchapter validates that the proposed damage detection technique

- 85 -

using combined binary search and compressed sensing can detect and visualize damage, even for realistic aniso-

tropic specimens.

(a) (b)

Figure 6.20. Damage quantification through the compressed sensing. (a) The imaging region is represented as

a dashed box and (b) the corresponding delamination quantification result is presented.

(a)

(b)

Figure 6.21. Comparison of the original full-field and the reconstructed wave propagation snapshots that cor-

respond to the dashed box in Figure 6.20 (a): (a) Original full-field wave propagation and (b) reconstructed

wave propagation.

- 86 -

6.5.3. Delamination detection using binary search with fixed pitch-catch distance scanning

(a) (b)

(c) (d)



point detection, and (d) the corresponding delamination quantification result.





- 87 -



points.

The delamination is identified and visualized in Figure 6.22 (d) from the identified transition points.

The red color in Figure 6.22 (d) represents the damage points, where they are between the left and right transition

points. When the 50 mm by 50 mm area is scanned with a 1 mm resolution, a total of 676 (26 × 26) inspection

points are generated. Only 48 scanning points (7.1%) are needed to quantify the delamination, corresponding to

0.8 minutes of scanning time. This scanning time can be reduced as mentioned in Chapter 6.2. The experiment

presented in this subchapter validates that the proposed damage detection technique using binary search with fixed

pitch-catch distance scanning can detect and visualize damage, even for realistic anisotropic specimens.

- 88 -


In this chapter, the developed damage detection techniques are experimentally validated with four specimens, an

aluminum plate with a notch, an aluminum plate with a fatigue crack, a carbon fiber reinforced plastic (CFRP)

plate with a delamination, and a 10 kW glass fiber reinforced plastic (GFRP) wind turbine blade with a delami-

nation. The macroscopic laser ultrasonic scanning system is configured to perform experimental validations with

these specimens.

Damage detection is effectively performed for all four examples. With these examples, it is validated

that (1) the proposed damage detection techniques can identify and quantify damage with a reduced number of

measurements and inspection time, less than 10% of conventional wavefield imaging techniques; (2) not only

large notches but also incipient fatigue cracks can be identified; and (3) these techniques are applicable to aniso-

tropic composite structures.

However, still there are several technical challenges for field applications: (1) A special treatment of a

target surface is necessary to achieve ultrasonic measurements with a high signal-to-noise ratio; and (2) there is

an eye safety issue with the excitation pulse laser beams. These challenges will be handled in the final dissertation

with a microscopic based laser ultrasonic scanning system. In addition, their feasibility of these techniques with

structural complexities should be investigated. In Chapter 7, additional examples will be presented overcoming

these limitations using a microscopic laser ultrasonic scanning system.

- 89 -

Chapter 7. Experimental Validation of the Developed Techniques

using a Microscopic Laser Ultrasonic Scanning System

7.1. Hardware setup

Figure 7.1. Overview of the microscopic laser ultrasonic scanning system.

Figure 7.1 shows an overview of the macroscopic laser ultrasonic scanning system used in the experiments. A

diode pumped Q-switched Nd:YAG pulse laser (Quantel Centurion+) radiates 532 nm wavelength laser pulses.

The width of the laser pulse is 12 ns, and this device emits 3 mm diameter laser pulses up to 100 times per second.

A fiber guide system guides radiated pulse laser beams to the target structure and focuses the laser beam diameter

to 0.5 mm on the specimen surface. The ultrasonic responses were measured by a commercial microscopic LDV

(Polytec MSA-100-3D). The location of the target structure is controlled using a scanning stage synchronized

with the LDV. As the excitation laser beam location and the sensing laser beam location are not controlled inde-

pendently, only binary search with fixed pitch-catch distance scanning technique is applied in this chapter. The

distance from the target specimen to the fiber guide system and the LDV was 9 cm and 4 cm, respectively.

- 90 -

7.2. Defect detection in an epoxy molding compound (EMC) bar


(a)

(b)

Figure 7.2. An epoxy molding compound (EMC) bar with a defect. (a) A packaged chip requires EMC to

encapsulates the semiconductor chip inside. (b) There is a circular defect of 1 mm diameter on the backside of

the scanning surface of the specimen.

The first tested specimen is an epoxy molding compound (EMC) bar with a defect. EMC is a molding compound

used to encapsulate a semiconductor chip, as represented in Figure 7.2 (a). Any void inside EMC may expand due

to heat from the semiconductor chip. This expansion induces sudden explosion of EMC and breakage of the inner

semiconductor chip. Tested EMC bar has dimensions of 125 × 13 × 3 mm3, as displayed in Figure 7.2 (b). A 1

mm diameter defect is introduced in the backside of the chip. The 20 mm by 10 mm rectangular area around the

defect is established as an inspection region.

A spatial ultrasonic dictionary was constructed from the intact part of the EMC bar. Bases in the spatial

ultrasonic dictionary were generated by moving the excitation point from a 5 mm distance to the sensing point to

20 mm with 500 μm to create 31 measurements. By spline interpolations between the two adjacent bases, the

- 91 -

dictionary was constructed from a total of 301 bases with a spatial resolution of 50 μm. The peak energy and

power of the pulse laser was 1 mJ and 0.08 MW, respectively. The pulse laser energy is limited to low to prevent

any damage on the semiconductor chip. Each basis was sampled for 8 μs with a 31.25 MHz sampling frequency.

A bandpass filter with a low cutoff frequency of 3 MHz and a high cutoff frequency of 6 MHz was used to capture

ultrasonic information in the interested frequency bandwidth. Each basis collection was averaged 100 times to

improve their signal-to-noise ratios.

For fixed pitch-catch distance scanning, each scanning was performed with identical parameters in the

basis construction process. The spatial scanning resolution for a binary search was 500 μm for the x and y direc-

tions, which divides the inspection region into 41×21 grids. The pitch-catch distance between the excitation laser

beam and the sensing laser beam was fixed to 10 mm, leading K=20.

7.2.2. Defect detection using binary search with fixed pitch-catch distance scanning


mm rectangular area represents the inspection region. Each inspection line is divided into two divisions for dam-

age region identification. Only 16 scanning points are needed to identify the damage regions, indicated with a

dashed box in Figure 7.3 (a).


and left transition point detection, respectively. Both left and right transition points are identified as this defect is

a wide damage. 24 additional scanning points are required to identify the transition points.

The defect is identified and visualized in Figure 7.3 (d) from the identified transition points. The red


The identified size of the defect is slightly larger than the actual defect, but quite similar to the exact one. When

the 20 mm by 10 mm area is scanned with a 0.5 mm resolution, a total of 861 (41 × 21) inspection points are

generated. Only 40 scanning points (4.7%) are needed to quantify the defect. This is corresponding to 0.7 minute

of scanning time. Note that this scanning time is obtained with (1) averaging 100 times for each scanning, (2) 100

Hz repetition rate of a pulse laser, and (3) 500 μm scanning resolution for a 20 mm by 10 mm rectangular region.

If (1) less averaging was employed, (2) a faster pulse laser was available, or (3) a lower scanning resolution was

acceptable, the scanning time could have been further reduced. The experiment presented in this subchapter vali-

dates that the proposed damage detection technique using binary search with fixed pitch-catch distance scanning

can detect and visualize a damage on the backside of a specimen using the developed microscopic laser ultrasonic

scanning system.

- 92 -

(a) (b)

(c) (d)

Figure 7.3. Binary search with fixed pitch-catch distance scanning: (a) identification of damage regions, (b) layout of scanning points for right transition point detection,

(c) layout of scanning points for left transition point detection, and (d) the corresponding defect quantification result.

- 93 -

7.3. Crack detection in a semiconductor chip


(a)

(b)

Figure 7.4. A semiconductor chip with a crack. (a) A semiconductor chip can be cracked while grinding the

semiconductor wafer. (b) A crack with 10 mm height, 50 μm width, approximately 40 μm depth is introduced

to the chip.

The second tested specimen is a semiconductor chip with a crack. A semiconductor chip can be cracked while

grinding the semiconductor wafer to multiple semiconductor chips. This chip has dimensions of 15 × 10 × 0.05

mm3, as displayed in Figure 7.4 (b). A 10 mm long, 50 μm wide and approximately 40 μm deep crack is introduced

to the chip. The whole area of the chip, 15 mm by 10 mm rectangular area, is established as an inspection region.

A spatial ultrasonic dictionary was constructed from an intact semiconductor chip. Bases in the spatial

ultrasonic dictionary were generated by moving the excitation point from a 2 mm distance to the sensing point to

8 mm with 200 μm to create 31 measurements. By spline interpolations between the two adjacent bases, the

dictionary was constructed from a total of 301 bases with a spatial resolution of 20 μm. The peak energy and

power of the pulse laser was 0.6 mJ and 0.05 MW, respectively. The pulse laser energy is limited to low to prevent

any damage on the semiconductor chip. Each basis was sampled for 4 μs with a 31.25 MHz sampling frequency.

- 94 -

A bandpass filter with a low cutoff frequency of 3 MHz and a high cutoff frequency of 6 MHz was used to capture

ultrasonic information in the interested frequency bandwidth. Each basis collection was averaged 100 times to

improve their signal-to-noise ratios.

For fixed pitch-catch distance scanning, each scanning was performed with identical parameters in the

basis construction process. The spatial scanning resolution for a binary search was 200 μm for the x and y direc-

tions, which divides the inspection region into 77×51 grids. The pitch-catch distance between the excitation laser

beam and the sensing laser beam was fixed to 3.8 mm, leading K=19.

7.3.2. Crack detection using binary search with fixed pitch-catch distance scanning


mm rectangular area represents the inspection region. Each inspection line is divided into four divisions for dam-

age region identification. Only 44 scanning points are needed to identify the damage regions, indicated with

dashed boxes in Figure 7.5 (a).


and left transition point detection, respectively. In Figure 7.5 (c), it can be observed that there is no left transition

point as the crack in this example is corresponding to a single inspection point in widthwise. 66 additional scan-

ning points are required to identify the transition points.

The crack is identified and visualized in Figure 7.5 (d) from the identified transition points. The red


When the 15 mm by 10 mm area is scanned with a 0.2 mm resolution, a total of 3927 (77 × 51) inspection points

are generated. Only 110 scanning points (2.8%) are needed to quantify the crack. This is corresponding to 1.8

minutes of scanning time. Note that this scanning time is obtained with (1) averaging 100 times for each scanning,

(2) 100 Hz repetition rate of a pulse laser, and (3) 200 μm scanning resolution for a 15 mm by 10 mm rectangular

region. This scanning time can be reduced as mentioned in Chapter 7.2. The experiment presented in this sub-

chapter validates that the proposed damage detection technique using binary search with fixed pitch-catch distance

scanning can detect and visualize a micro crack on a microscopic structure.

- 95 -

(a) (b)

(c) (d)

Figure 7.5. Binary search with fixed pitch-catch distance scanning: (a) identification of damage regions, (b) layout of scanning points for right transition point detection,

(c) layout of scanning points for left transition point detection, and (d) the corresponding crack quantification result.

- 96 -


In this chapter, the developed binary search with fixed pitch-catch distance scanning technique is experimentally

validated with two specimens, an epoxy molding compound (EMC) bar with a defect and a semiconductor chip

with a crack. The microscopic laser ultrasonic scanning system is configured to perform experimental validations

with these specimens. As the specimen location is controlled using the scanning stage, rather than controlling the

excitation laser beam and the sensing laser beam, only fixed pitch-catch distance scanning scheme is available

with the microscopic scanning system.

Damage detection is effectively performed for both examples. With these examples, it is validated that

(1) the proposed damage detection technique can identify and quantify damage with a reduced number of meas-

urements and inspection time, less than 5% of conventional wavefield imaging techniques; (2) the technique can

be applied not only to large structures but also to microscopic specimens; (3) the technique is still working with

complex patterns in the target specimens.

- 97 -

Chapter 8. Concluding Remarks

8.1. Summary of the works

This dissertation focuses on accelerated damage detection techniques using noncontact laser ultrasonic scanning.

Laser ultrasonic scanning is attractive for damage detection due to its noncontact nature, sensitivity to local dam-

age, and high spatial resolution. However, its practicality is limited because scanning at a high spatial resolution

demands a prohibitively long scanning time. The author aims to accelerate laser based nondestructive testing with

the developed techniques inspired by binary search. This dissertation can be summarized as follows.

1) Development of accelerated damage detection techniques

Two accelerated damage detection techniques are developed in this dissertation. Inspired by binary search, dam-

age is localized and visualized with reduced scanning points and scanning time. However, it should be noted that

a spatial ultrasonic dictionary is required for the following algorithms. Baseline responses need to be collected to

construct the dictionary.

The first technique, combined binary search and compressed sensing, identifies the approximate dam-

age location from measurements at the sparse scanning points that are selected by the binary search algorithm.

Then damage is visualized by reconstructing wavefield image around damage using compressed sensing. The

number of scanning points that is necessary for damage localization and visualization is dramatically reduced

from 𝑁 ∙ 𝑀 to 2log2 𝑁 ⋅ log2 𝑀. 𝑁 and 𝑀 represent the number of equally spaced scanning points in the x and

y directions, respectively, which are required to obtain full-field wave propagation images of the target inspection

region. This technique is more appropriate to scan a large inspection region as the location of the sensing laser

beam is fixed and surface treatment is minimized.

The other technique, binary search with fixed pitch-catch distance scanning, directly quantifies the

damage using fixed pitch-catch distance scanning strategy from the sparse scanning points that are selected by the

binary search algorithm. The approximate damage region is identified first, then the left and right damage bound-

ary are detected. The number of scanning points that is necessary for damage localization and visualization is

dramatically reduced from 𝑁 ∙ 𝑀 to 4log2 𝑁 ⋅ log2 𝑀 even for the worst case scenario. This technique is usually

faster than the first technique as it requires less number of averagings. However, this technique may require large

- 98 -

surface treatment to move the sensing laser beam location. In that sense, this technique is more appropriate to scan

small structures using the microscopic laser ultrasonic scanning system.

2) Binary search based on locating a damage with a spatial ultrasonic transformation

The binary search algorithm is based on examining the interactions between the ultrasonic waves and damage,

such as reflections and transmissions. A time-domain ultrasonic response is transformed into a spatial ultrasonic

domain to better identify these interactions. An important observation is that the measured response in the spatial

ultrasonic domain includes the basis which corresponds to the intact incident waves, if no damage is observed

between the excitation point and sensing point (either no damage in the specimen or the excitation point is ‘in

front of the damage’). However, this basis is shifted if the excitation point is ‘behind the damage’. Therefore, the

damage presence within the direct wave propagation path can be easily identified by performing spatial ultrasonic

transformation and verifying the shift of the basis that corresponds to the intact incident waves.

An optimal, or sparse, representation after the transformation is obtained by basis pursuit approach.

Instead of a traditional matching pursuit approach which tries to solve an ℓ0 minimization problem, basis pursuit

approach solves an ℓ1 minimization problem to transform a time-domain response with a better resolution.


The feasibility of the developed damage detection techniques is validated in both numerical and experimental

ways. For a numerical validation, an aluminum plate with a crack is modeled using COMSOL Multiphysics. For

experimental validations, an aluminum plate with a notch, an aluminum plate with a fatigue crack, a carbon fiber

reinforced plastic (CFRP) plate with a delamination, and a 10 kW glass fiber reinforced plastic (GFRP) plate with

a delamination are tested. An epoxy molding compound (EMC) bar with a defect and a cracked semiconductor

chip are also evaluated.

The developed techniques identified and quantified tested damages with a high accuracy. With the

developed techniques, more than 90% of required scanning points and scanning time are reduced from the ones

of conventional ultrasonic imaging techniques. Though the second technique, binary search with fixed pitch-catch

distance scanning, requires more number of scanning points than the first technique, combined binary search and

compressed sensing, the scanning times is further reduced with a less number of averagings.

- 99 -

8.2. Uniqueness of the works

The unique contributions of this dissertation are summarized as follows.

1) Improved damage detection sensitivity

Conventionally, damage existence is identified by subtracting the baseline data that corresponds to the pristine

condition of a specimen from the measured signal in the time domain. However, it is revealed that the measured

signal significantly deviates from its pristine condition even when the damage is located outside the direct wave

propagation path due to reflections from the damage. In the developed techniques, damage existence is identified

by performing spatial ultrasonic transformation and verifying the shift of the basis that corresponds to the intact

incident waves. They have shown outstanding performance in damage detection with a high spatial sensitivity.

2) Accelerated search algorithm

Conventional techniques have required a huge number of measurements to achieve a high spatial resolution in

damage detection. It is as they quantify a damage by repeatedly checking whether each point is intact or damaged.

In the developed techniques, only a reduced number of scanning points is required by searching the damage from

optimally selected scanning points. The selection of these optimized locations is inspired by a binary search algo-

rithm. The number of scanning points that is necessary for damage localization and visualization is dramatically

reduced from 𝑁 ∙ 𝑀 to 4log2 𝑁 ⋅ log2 𝑀 even for the worst case scenario 𝑁 and 𝑀 represent the number of

equally spaced scanning points in the x and y directions, respectively, which are required to obtain full-field wave

propagation images of the target inspection region.


The performances of the developed damage detection techniques are validated with a numerical and various real-

istic structures. It is validated that the developed techniques can quantify even an incipient crack and be applicable

for large realistic anisotropic structures. Microscopic structures including a semiconductor chip are also evaluated.

More than 90% of required scanning points and scanning time are reduced from the ones of conventional ultra-

sonic imaging techniques.

- 100 -

8.3. Future works

It is envisioned that the developed laser ultrasonic scanning techniques can be applied to various nondestructive

testing applications thanks to their uniqueness and promising advantages. However, there are still several technical

challenges that have to be tackled for their field applications.

1) Further validations of the developed laser ultrasonic scanning technique

Though the developed scanning techniques are validated for many applications, their practical applicability should

be validated with more examples. As the developed techniques are more appropriate for the quality assurance of

products, various microscopic specimens and samples will be tested with the developed techniques and the devel-

oped techniques are improved accordingly.

2) System improvement

To improve the applicability and the damage sensitivity of the developed scanning techniques to the micro struc-

tures, ultrasonic waves with higher frequency bandwidth are required. The proposed laser ultrasonic scanning

system with a nanosecond pulse laser and a LDV is not applicable for ultrasonic wave generation and measure-

ment over 50 MHz. Recently, a new laser ultrasonic technique known as picosecond ultrasonics [99] is being

studied for high frequency ultrasonic generation and measurement using picosecond pulse lasers. Various ap-

proaches will be considered to improve the applicability and the damage sensitivity of the developed scanning

techniques.

3) Improvement of the damage detection algorithms

The developed scanning techniques are based on the spatial ultrasonic transformation of ultrasonic waves to iden-

tify wave-damage interactions and damage existence. As this transformation is based on linear ultrasonic interac-

tions e.g. reflections and transmissions, a high frequency ultrasonic measurable device is required to detect wave-

damage interactions from small damages. Identification of nonlinear wave-damage interactions [77] and corre-

sponding scanning strategies are being studied. Also, the possibility of damage existence identification without

basis construction will be studied to improve the field applicability.

- 101 -

Bibliography

[1] Campbell, F.C. (2008). Elements of metallurgy and engineering alloys. ASM International.

[2] Balageas, D., Fritzen, C.-P., and Güemes, A. (2006). Structural health monitoring, Vol 493. Wiley Online

Library.

[3] Chang, F.K. et al. (2001). "Structural health monitoring." System Health Management: With Aerospace

Applications, pp. 419-428.

[4] Boller, C., and Staszewski, W. (2004). "Structural health monitoring." Proceedings of the second European

workshop on structural health monitoring, München: pp. 7-9.

[5] Krautkrämer, J., and Krautkrämer, H. (2013). Ultrasonic testing of materials. Springer Science & Business

Media.

[6] Rosen-Group. "Measurement technology - Eddy current." http://www.rosen-group.com/global/company/

explore/we-can/technologies/measurement/eddy-current.html, March 22nd, 2016.

[7] Libby, H.L. (1971). "Introduction to electromagnetic nondestructive test methods." pp.

[8] Blitz, J. (2012). Electrical and magnetic methods of non-destructive testing, Vol 3. Springer Science &

Business Media.

[9] Harding, G., and Kosanetzky, J. (1989). "Scattered X-ray beam nondestructive testing." Nuclear Instruments

and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,

280, pp. 517-528.

[10] Hanke, R., Fuchs, T., and Uhlmann, N. (2008). "X-ray based methods for non-destructive testing and material

characterization." Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers,

Detectors and Associated Equipment, 591, pp. 14-18.

[11] Zhu, L. et al. (2013). "Real-time monitoring of phase maps of digital shearography." Optical Engineering,

52, pp. 101902-101902.

[12] Xie, X. et al. (2014). "Spatial Phase-Shift Digital Shearography for Out-of-Plane Deformation Measurement."

SAE Int J Mater Manf, 7, pp. 402-405.

[13] Hung, Y. (1982). "Shearography: a new optical method for strain measurement and nondestructive testing."

Optical Engineering, 21, pp. 213391-213391-.

[14] Steinchen, W., and Yang, L. (2003). Digital shearography: theory and application of digital speckle pattern

shearing interferometry. SPIE press Bellingham.

[15] Maldague, X. (2001). "Theory and practice of infrared technology for nondestructive testing." pp.

- 102 -

[16] Clark, M., McCann, D., and Forde, M. (2003). "Application of infrared thermography to the non-destructive

testing of concrete and masonry bridges." NDT & E International, 36, pp. 265-275.

[17] Maldague, X.P. (2002). "Introduction to NDT by active infrared thermography." Materials Evaluation, 60,

pp. 1060-1073.

[18] Maierhofer, C. et al. (2014). "Characterizing damage in CFRP structures using flash thermography in

reflection and transmission configurations." Composites Part B: Engineering, 57, pp. 35-46.

[19] An, Y.-K., Kim, J.M., and Sohn, H. (2014). "Laser lock-in thermography for detection of surface-breaking

fatigue cracks on uncoated steel structures." NDT & E International, 65, pp. 54-63.

[20] Mittleman, D.M. et al. "Recent advances in terahertz imaging." Applied Physics B, 68, pp. 1085-1094.

[21] Amenabar, I., Lopez, F., and Mendikute, A. (2013). "In introductory review to THz non-destructive testing

of composite mater." Journal of Infrared, Millimeter, and Terahertz Waves, 34, pp. 152-169.

[22] Guillet, J.P. et al. (2014). "Review of Terahertz tomography techniques." Journal of Infrared, Millimeter, and

Terahertz Waves, 35, pp. 382-411.

[23] Palka, N. et al. (2016). "Non-destructive evaluation of puncture region in polyethylene composite by terahertz

and X-ray radiation." Composites Part B: Engineering, 92, pp. 315-325.

[24] Wang, S., and Zhang, X. (2004). "Pulsed terahertz tomography." Journal of Physics D: Applied Physics, 37,

pp. R1.

[25] Takayanagi, J. et al. (2009). "High-resolution time-of-flight terahertz tomography using a femtosecond fiber

laser." Optics express, 17, pp. 7533-7539.

[26] Rothbart, N. et al. (2013). "Fast 2-D and 3-D terahertz imaging with a quantum-cascade laser and a scanning

mirror." Terahertz Science and Technology, IEEE Transactions on, 3, pp. 617-624.

[27] Scruby, C.B., and Drain, L.E. (1990). Laser ultrasonics techniques and applications. CRC Press.

[28] Davies, S. et al. (1993). "Laser-generated ultrasound: its properties, mechanisms and multifarious

applications." Journal of Physics D: Applied Physics, 26, pp. 329.

[29] Bleaney, B.I., Bleaney, B.I., and Bleaney, B. (2013). Electricity and Magnetism, Volume 2, Vol 2. Oxford

University Press.

[30] Hutchins, D. (1986). "Mechanisms of pulsed photoacoustic generation." Canadian Journal of Physics, 64,

pp. 1247-1264.

[31] Ready, J. (2012). Effects of high-power laser radiation. Elsevier.

[32] Roth, J. et al. (2012). "Laser Ablation of Metals: Generation of Drops and Bubbles." Innovatives

Supercomputing in Deutschland, 10, pp. 30-35.

[33] Simonetti, A. et al. (2005). "U–Pb zircon dating by laser ablation-MC-ICP-MS using a new multiple ion

- 103 -

counting Faraday collector array." Journal of Analytical Atomic Spectrometry, 20, pp. 677-686.

[34] Liu, W., and Hong, J.-W. (2015). "Modeling of three-dimensional Lamb wave propagation excited by laser

pulses." Ultrasonics, 55, pp. 113-122.

[35] Liu, P., Wahab, N.A., and Sohn, H. (2015). "NUMERICAL AND EXPERIMENTAL ANALYSES OF PULSE

LASER-GENERATED ULTRASOUND." CANSMART 2015 Vancouver, BC, Canda.

[36] Hariharan, P. (2010). Basics of interferometry. Academic Press.

[37] Hernández, G. (1988). Fabry-perot interferometers, Vol 3. Cambridge University Press.

[38] Zetie, K., Adams, S., and Tocknell, R. (2000). "How does a Mach-Zehnder interferometer work?" Physics

Education, 35, pp. 46.

[39] Castellini, P., Martarelli, M., and Tomasini, E.P. (2006). "Laser Doppler vibrometry: development of

advanced solutions answering to technology's needs." Mechanical Systems and Signal Processing, 20, pp. 1265-

1285.

[40] Staszewski, W., Lee, B., and Traynor, R. (2007). "Fatigue crack detection in metallic structures with Lamb

waves and 3D laser vibrometry." Measurement Science and Technology, 18, pp. 727.

[41] Johansmann, M., Siegmund, G., and Pineda, M. (2005). "Targeting the limits of laser Doppler vibrometry."

Proc IDEMA, pp. 1-12.

[42] Martin, P., and Rothberg, S. (2009). "Introducing speckle noise maps for laser vibrometry." Optics and Lasers

in Engineering, 47, pp. 431-442.

[43] Achenbach, J. (2012). Wave propagation in elastic solids, Vol 16. Elsevier.

[44] An, Y.-K., Park, B., and Sohn, H. (2013). "Complete noncontact laser ultrasonic imaging for automated crack

visualization in a plate." Smart Materials and Structures, 22, pp. 025022.

[45] Swacek, C., Kim, J.-Y., and Jacobs, L. (2013). "Optical excitation of narrowband Rayleigh surface waves for

second harmonic generation." REVIEW OF PROGRESS IN QUANTITATIVE NONDESTRUCTIVE

EVALUATION: VOLUME 32: pp. 375-381.

[46] Huang, J., Krishnaswamy, S., and Achenbach, J.D. (1992). "Laser generation of narrow‐band surface waves."

The Journal of the Acoustical Society of America, 92, pp. 2527-2531.

[47] Choi, S., and Jhang, K.-Y. (2013). "Influence of slit width on harmonic generation in ultrasonic surface waves

excited by masking a laser beam with a line arrayed slit." NDT & E International, 57, pp. 1-6.

[48] Staszewski, W. et al. (2004). "Structural health monitoring using scanning laser vibrometry: I. Lamb wave

sensing." Smart Materials and Structures, 13, pp. 251.

[49] Mallet, L. et al. (2004). "Structural health monitoring using scanning laser vibrometry: II. Lamb waves for

damage detection." Smart Materials and Structures, 13, pp. 261.

- 104 -

[50] Leong, W. et al. (2005). "Structural health monitoring using scanning laser vibrometry: III. Lamb waves for

fatigue crack detection." Smart Materials and Structures, 14, pp. 1387.

[51] Sohn, H. et al. (2011). "Automated detection of delamination and disbond from wavefield images obtained

using a scanning laser vibrometer." Smart Materials and Structures, 20, pp. 045017.

[52] Sohn, H. et al. (2011). "Delamination detection in composites through guided wave field image processing."

Composites science and technology, 71, pp. 1250-1256.

[53] Park, B. et al. (2013). "Laser ultrasonic imaging and damage detection for a rotating structure." Structural

Health Monitoring, 12, pp. 494-506.

[54] Park, B. et al. (2012). "Impact localization in complex structures using laser-based time reversal." Structural

Health Monitoring, pp. 1475921712449508.

[55] Lee, J.-R. et al. (2012). "Laser ultrasonic anomalous wave propagation imaging method with adjacent wave

subtraction: Algorithm." Optics & Laser Technology, 44, pp. 1507-1515.

[56] Ruzzene, M. (2007). "Frequency–wavenumber domain filtering for improved damage visualization." Smart

Materials and Structures, 16, pp. 2116.

[57] Michaels, T.E., Michaels, J.E., and Ruzzene, M. (2011). "Frequency–wavenumber domain analysis of guided

wavefields." Ultrasonics, 51, pp. 452-466.

[58] Mesnil, O. et al. (2014). "Frequency Domain Instantaneous Wavenumber Estimation for Damage

Quantification in Layered Plate Structures." EWSHM-7th European Workshop on Structural Health Monitoring.

[59] Flynn, E.B. et al. (2013). "Structural imaging through local wavenumber estimation of guided waves." NDT

& E International, 59, pp. 1-10.

[60] Park, B., An, Y.-K., and Sohn, H. (2014). "Visualization of hidden delamination and debonding in composites

through noncontact laser ultrasonic scanning." Composites science and technology, 100, pp. 10-18.

[61] An, Y.-K., and Sohn, H. (2015). "Visualization of non-propagating Lamb wave modes for fatigue crack

evaluation." Journal of Applied Physics, 117, pp. 114904.

[62] Lee, J.-R., Jang, J.-K., and Kong, C.-W. (2014). "Fully noncontact wave propagation imaging in an immersed

metallic plate with a crack." Shock and Vibration, 2014, pp. 895693.

[63] Donoho, D.L. (2006). "Compressed sensing." Information Theory, IEEE Transactions on, 52, pp. 1289-1306.

[64] Di Ianni, T. et al. (2015). "Compressive sensing of full wave field data for structural health monitoring

applications." Ultrasonics, Ferroelectrics, and Frequency Control, IEEE Transactions on, 62, pp. 1373-1383.

[65] Mesnil, O., and Ruzzene, M. (2016). "Sparse wavefield reconstruction and source detection using

Compressed Sensing." Ultrasonics, pp.

[66] Flynn, E.B. (2014). "Embedded Multi-Tone Ultrasonic Excitation and Continuous-Scanning Laser Doppler

- 105 -

Vibrometry for Rapid and Remote Imaging of Structural Defects." EWSHM-7th European Workshop on

Structural Health Monitoring.

[67] Ihn, J.-B., and Chang, F.-K. (2008). "Pitch-catch active sensing methods in structural health monitoring for

aircraft structures." Structural Health Monitoring, 7, pp. 5-19.

[68] Sohn, H. et al. (2003). "Wavelet-based active sensing for delamination detection in composite structures."

Smart Materials and Structures, 13, pp. 153.

[69] Tanaka, T., and Izawa, Y. (2001). "Nondestructive detection of small internal defects in carbon steel by laser

ultrasonics." Japanese Journal of Applied Physics, 40, pp. 1477.

[70] Park, B. et al. (2016). "Delamination localization in wind turbine blades based on adaptive time-of-flight

analysis of noncontact laser ultrasonic signals." Nondestructive Testing and Evaluation, pp. 1-20.

[71] Scala, C., and Bowles, S. (2000). "Laser ultrasonics for surface-crack depth measurement using transmitted

near-field Rayleigh waves." REVIEW OF PROGRESS IN QUANTITATIVE NONDESTRUCTIVE

EVALUATION: Volume 19: pp. 327-334.

[72] Sohn, Y., and Krishnaswamy, S. (2004). "Interaction of a scanning laser-generated ultrasonic line source with

a surface-breaking flaw." The Journal of the Acoustical Society of America, 115, pp. 172-181.

[73] Kim, H. et al. (2006). "A noncontact NDE method using a laser generated focused-Lamb wave with enhanced

defect-detection ability and spatial resolution." NDT & E International, 39, pp. 312-319.

[74] Jhang, K.-Y. (2009). "Nonlinear ultrasonic techniques for nondestructive assessment of micro damage in

material: a review." International journal of precision engineering and manufacturing, 10, pp. 123-135.

[75] Solodov, I. et al. (2004). "Nonlinear self-modulation and subharmonic acoustic spectroscopyfor damage

detection and location." Applied physics letters, 84, pp. 5386-5388.

[76] Cantrell, J.H., and Yost, W.T. (2001). "Nonlinear ultrasonic characterization of fatigue microstructures."

International Journal of fatigue, 23, pp. 487-490.

[77] De Lima, W., and Hamilton, M. (2003). "Finite-amplitude waves in isotropic elastic plates." Journal of sound

and vibration, 265, pp. 819-839.

[78] Bermes, C. et al. (2007). "Experimental characterization of material nonlinearity using Lamb waves." Applied

physics letters, 90, pp. 021901.

[79] Lim, H.J. et al. (2015). "Noncontact fatigue crack visualization using nonlinear ultrasonic modulation." NDT

& E International, 73, pp. 8-14.

[80] Hess, P., Lomonosov, A.M., and Mayer, A.P. (2014). "Laser-based linear and nonlinear guided elastic waves

at surfaces (2D) and wedges (1D)." Ultrasonics, 54, pp. 39-55.

[81] Liu, P. et al. (2014). "Noncontact detection of fatigue cracks by laser nonlinear wave modulation spectroscopy

- 106 -

(LNWMS)." NDT & E International, 66, pp. 106-116.

[82] Choi, S., Seo, H., and Jhang, K.-Y. (2015). "Noncontact Evaluation of Acoustic Nonlinearity of a Laser-

Generated Surface Wave in a Plastically Deformed Aluminum Alloy." Research in Nondestructive Evaluation, 26,

pp. 13-22.

[83] Liu, P. et al. (2015). "Fatigue crack localization using noncontact laser ultrasonics and state space attractors."

The Journal of the Acoustical Society of America, 138, pp. 890-898.

[84] Overbey, L., Olson, C., and Todd, M. (2007). "A parametric investigation of state-space-based prediction

error methods with stochastic excitation for structural health monitoring." Smart Materials and Structures, 16, pp.

1621.

[85] Liu, P., Sohn, H., and Park, B. (2015). "Baseline-free damage visualization using noncontact laser nonlinear

ultrasonics and state space geometrical changes." Smart Materials and Structures, 24, pp. 065036.

[86] Neapolitan, R.E., and Naimipour, K. (2011). Foundations of algorithms. Jones & Bartlett Learning.

[87] Anton, H., and Busby, R.C. (2003). Contemporary linear algebra, Vol 170. Wiley New York.

[88] Gribonval, R., and Nielsen, M. (2003). "Sparse representations in unions of bases." Information Theory, IEEE

Transactions on, 49, pp. 3320-3325.

[89] Mallat, S.G., and Zhang, Z. (1993). "Matching pursuits with time-frequency dictionaries." Signal Processing,

IEEE Transactions on, 41, pp. 3397-3415.

[90] Chen, S., and Donoho, D. (1994). "Basis pursuit." Signals, Systems and Computers, 1994 1994 Conference

Record of the Twenty-Eighth Asilomar Conference on: pp. 41-44.

[91] Elad, M. (2004). "Sparse Representations of Signals: Theory and Applications " IPAM-MGA Program,

Powerpoint Presentation.

[92] Gill, P.R., Wang, A., and Molnar, A. (2011). "The in-crowd algorithm for fast basis pursuit denoising." Signal

Processing, IEEE Transactions on, 59, pp. 4595-4605.

[93] Malioutov, D.M., Cetin, M., and Willsky, A.S. (2005). "Homotopy continuation for sparse signal

representation." Acoustics, Speech, and Signal Processing, 2005 Proceedings(ICASSP'05) IEEE International

Conference on: pp. v/733-v/736 Vol. 735.

[94] Hale, E.T., Yin, W., and Zhang, Y. (2010). "Fixed-point continuation applied to compressed sensing:

implementation and numerical experiments." J Comput Math, 28, pp. 170-194.

[95] Friedlander, M.P., and Saunders, M.A. (2012). "A DUAL ACTIVE-SET QUADRATIC PROGRAMMING

METHOD FOR FINDING SPARSE LEAST-SQUARES SOLUTIONS." DRAFT Technical Report,Department

of Computer Science, University of British Columbia.

[96] Giurgiutiu, V., Zagrai, A., and Bao, J.J. (2002). "Piezoelectric wafer embedded active sensors for aging

- 107 -

aircraft structural health monitoring." Structural Health Monitoring, 1, pp. 41-61.

[97] Graff, K.F. (1975). Wave motion in elastic solids. Courier Corporation.

[98] Rose, J.L. (2004). Ultrasonic waves in solid media. Cambridge university press.

[99] Matsuda, O. et al. (2015). "Fundamentals of picosecond laser ultrasonics." Ultrasonics, 56, pp. 3-20.

- 108 -

Acknowledgements in Korean

그동안 수많은 분들의 도움이 있었기에 이 논문이 나올 수 있었습니다. 박사학위를 받기까지

도움 주신 분들께 진심으로 감사드립니다. 지면상의 제약으로 모든 분을 일일이 언급드리지

못하는 점 양해 부탁드립니다.

먼저 학부생 때부터 저를 지도해 주신 손훈 교수님께 감사의 말씀을 드립니다. 입학 면접

시험장에서 뵈었던 인연이 여기까지 이어지게 된 것을 보면 인연이란 참 신기한 것 같습니다.

사회에 나가서도 교수님의 이름을 부끄럽게 하지 않는 제자가 되도록 노력하겠습니다. 바쁘신

와중에도 논문을 심사해 주시고 귀한 조언들을 아끼지 않으신 박기환, 안윤규, 정형조, 홍정욱

교수님께도 다시 한 번 감사드립니다.

관정 이종환 교육재단과 이종환 회장님의 지원에도 감사드립니다. 저를 믿고 기회를 주신

덕분에 꿈을 이루고자 하는 용기를 얻을 수 있었습니다. 재단과 회장님의 믿음이 헛되이 되지

않도록 사회에 기여할 수 있는 연구자가 되도록 하겠습니다.

언제나 저의 연구를 응원해 주고 도와 주었던 연구실 구성원들이 없었더라면 이 연구를 끝까지

마치지 못했을 것입니다. 연구자의 자세를 가르쳐 주신 선배님들과 어려운 문제들을 함께

풀어나간 후배님들, 그리고 연구를 지원해 주신 사무원 분들이 있었기에 즐거운 연구실 생활을

할 수 있었습니다. 모두와의 즐거웠던 하루하루가 언제나 그리울 것입니다.

이외에도 언제나 제 편이 되어 주었던 고등학교 친구들, 즐거운 추억들을 만들었던 동아리

구성원들, 항상 함께였던 학과 선후배/동기들 등 많은 친구들이 있었기에 가끔씩 지친 마음을

위로받고 용기내어 앞으로 나아갈 수 있었습니다.

마지막으로 언제나 저를 믿어 주시는 부모님과 가족들에게 진심으로 고마움을 전합니다.

평생이 걸려도 주신 사랑에 보답할 수는 없겠지만 자랑스러운 아들이자 손자, 형이 될 수 있도록

언제나 노력하겠습니다.

다시 한번 모든 분들께 감사드립니다.

- 109 -

Curriculum Vitae

Byeongjin PARK

Ph.D.

Department of Civil and Environmental Engineering

Korea Advanced Institute of Science and Technology (KAIST)

Daehakro 291, Yuseong-gu, Daejeon 34141, Republic of Korea

E-mail: [email protected]

RESEARCH INTERESTS

Structural Health Monitoring & Damage Detection, Non-destructive Testing, Quality Control, Non-

contact Sensing Technologies, Laser Ultrasonics, Statistical Pattern Recognition, Machine Learning,

Image Processing, Probabilistic & Statistical Analysis

EDUCATION

2011 – 2017

Ph.D. in Civil and Environmental Engineering, KAIST, Korea

2008 – 2011 B.S. summa cum laude in Civil and Environmental Engineering, KAIST, Korea.

DISSERTATION

Byeongjin Park, “Development of Accelerated Damage Detection Techniques Using Noncontact

Laser Ultrasonic Scanning,” Doctoral Dissertation, Department of Civil and Environmental Engi-

neering, KAIST, Korea, 2017.

mailto:[email protected]

- 110 -

HONORS & AWARDS

2015

Award by the Minister of Land, Infrastructure and Transport of Korea

I won the Best Award by the Minister of Land, Infrastructure and Transport of

Korea by presenting an idea for 2015 U-City Idea Competition. My idea was eval-

uated as the best idea among 201 applicants.

2015

Most Cited Article in Smart Materials and Structures

The journal paper below is selected as one of the 20 most highly cited papers

published in 2013 in “Smart Materials and Structures” Journal: Yun-Kyu An,

Byeongjin Park and Hoon Sohn “Complete noncontact laser ultrasonic imaging

for automated crack visualization in a plate,” Smart Materials and Structures, Vol.

22(2), 025022, 2013.

2011 – 2012

Kwanjeong Scholarship for Graduates

This scholarship is granted by the Kwanjeong Educational Foundation, one of the

most prestigious and competitive scholarship available in Korea. This supports

each grantee 10 Million Won per year for two years.

2011

The Best Paper Award in KBIM Conference

The conference proceeding below is selected as the best paper in 2011 KBIM An-

nual Conference: Min Koo Kim, Byeongjin Park and Hoon Sohn, “Optimized pipe

damage detection using 3D geometry information,” KBIM Annual Conference,

May 20, 2011, Seoul, Korea.

2010

Undergraduate Research Program Award

I awarded the Second Best Paper Award during the 2010 winter/spring undergrad-

uate research program workshop at KAIST for my project entitled “Development

of a laser controlling system for non-contact structural health monitoring”.

2010

Kwanjeong Scholarship for Undergraduates

This scholarship is granted by the Kwanjeong Educational Foundation, one of the

most prestigious and competitive scholarship available in Korea. This supports

each grantee 10 Million Won per year until his/her graduation.

2009

Honor Student

This honor is given to the undergraduate students whose GPA is high in his/her

own major. Honor students are eligible to take any graduate course in KAIST

without any restriction.

2008

Representative of Freshmen in the Entrance Ceremony of KAIST

I took the oath as the representative of over 700 freshmen in the entrance ceremony

of KAIST. This honor is given to the student who passed KAIST’s admission pro-

cess with the best grade.

- 111 -

JOURNAL PUBLICATIONS

* 7 SCI(E) published, 1 submitted

* The corresponding author is underlined

1. Byeongjin Park, Hoon Sohn and Peipei Liu, “Accelerated noncontact laser ultrasonic scan-

ning for damage detection using combined binary search and compressed sensing,” Mechani-

cal Systems and Signal Processing, submitted.

2. Byeongjin Park, Hoon Sohn, Pawel Malinowski and Wieslaw Ostachowicz, "Delamination

localization in wind turbine blades based on adaptive time-of-flight analysis of noncontact

laser ultrasonic signals," Nondestructive Testing and Evaluation, Vol. 31, 2016.

3. Hyung Jin Lim, Byeongju Song, Byeongjin Park and Hoon Sohn, “Noncontact fatigue crack

visualization using nonlinear ultrasonic modulation,” NDT&E International, Vol. 73, pp. 8-

14, 2015.

4. Peipei Liu, Hoon Sohn and Byeongjin Park, “Baseline-free damage visualization using non-

contact laser nonlinear ultrasonics and state space geometrical changes,” Smart Materials and

Structures, Vol. 24, 065036, 2015.

5. Byeongjin Park, Yun-Kyu An and Hoon Sohn, “Visualization of hidden delamination and

debonding in composites through noncontact laser scanning,” Composites Science and Tech-

nology, Vol. 100(21), pp.10-18, 2014.

6. Byeongjin Park, Hoon Sohn, Chul Min Yeum and Truong Thanh Chung, “Laser ultrasonic

imaging and damage detection of a rotating structure,” Structural Health Monitoring An Inter-

national Journal, Vol. 12, No. 5-6, pp. 494-506, 2013.

7. Yun-Kyu An, Byeongjin Park and Hoon Sohn “Complete noncontact laser ultrasonic imaging

for automated crack visualization in a plate,” Smart Materials and Structures, Vol. 22, 025022,

2013.

8. Byeongjin Park, Hoon Sohn, Martin P. DeSimio, Steven E. Olson, Kevin Brown and Mark

Derisso, “Impact localization in complex structures using laser based time reversal,” Structural

Health Monitoring An International Journal, Vol. 11(5), pp. 577-588, 2012.

- 112 -

CONFERENCE PROCEEDINGS

* 13 international and 3 domestic

* The corresponding author is underlined

1. Byeongjin Park, Hoon Sohn, “Accelerated damage visualization using binary search with

fixed distance laser ultrasonic scanning,” SPIE International Symposium, Smart Structures and

Material Systems + Nondestructive Evaluation and Health Monitoring, Portland, USA, March

25-29, 2017.

2. Timotius Yonathan Sunarsa, Byeonjin Park, Peipei Liu and Hoon Sohn, “Noncontact debond-

ing identification in adhesive using guided waves,” The 2016 World Congress on Advances in

Civil, Environmental, and Materials Research, Jeju, Korea, August 28-September 1, 2016.

3. Byeongjin Park, Hoon Sohn, “Reconstruction of Laser Ultrasonic Wavefield Images from

Reduced Sparse Measurements using Compressed Sensing Aided Super-resolution,” 43rd Re-

view of Progress in Quantitative Nondestructive Evaluation, Atlanta, USA, July 16-22, 2016.

4. Byeongjin Park, Hoon Sohn, Peipei Liu, “Accelerated Laser Ultrasonic Scanning using Bi-

nary Search,” The 8th European Workshop on Structural Health Monitoring, Bilbao, Spain,

July 5-8, 2016.

5. Byeongjin Park, Dongkyu Kim, Hoon Sohn, Kihwan Park, “Development of a noncontact

laser ultrasonic system for in-situ inspection of wind turbine blades,” Composites Seoul 2015,

Seoul, Korea, October 21-23, 2015.

6. Hoon Sohn, Hyung Jin Lim, Byeongjin Park, Peipei Liu, Byeongju Song, Yongtak Kim,

"Nonlinear ultrasonic modulation for damage detection," International Conference Vibroengi-

neering 2015, Nanjing, China, September 26-28, 2015. (Invited Keynote Presentation)

7. Byeongju Song, Byeongjin Park, Hoon Sohn, Cheol-Woo Lim, Jae-Roung Park, “Detection

and localization of fatigue crack on a rotating steel shaft using air-coupled nonlinear ultrasonic

modulation,” SPIE International Symposium, Smart Structures & Materials and Nondestruc-

tive Evaluation for Health Monitoring and Diagnostics, San Diego, USA, March 8-12, 2015.

8. Byeongjin Park, Hoon Sohn, “Instantaneous damage identification and localization through

sparse laser ultrasonic scanning,” The 7th European Workshop on Structural Health Monitor-

ing, Nantes, France, July 8-11, 2014.

9. Byeongjin Park, Hoon Sohn, Pawel Malinowski, Wieslaw Ostachowics “Damage detection

in composites by noncontact laser ultrasonics,” The 7th European Workshop on Structural

Health Monitoring, Nantes, France, July 8-11, 2014.

10. Hyung Jin Lim, Byeongju Song, Byeongjin Park, Peipei Liu, Hoon Sohn, “Non-contact visu-

alization of nonlinear ultrasonic modulation for reference-free fatigue crack detection,” SPIE

International Symposia, Smart Structures & Materials and Nondestructive Evaluation for

Health Monitoring and Diagnostics, San Diego, USA, March 9-13, 2014.

- 113 -

11. Byeongjin Park, Truong Thanh Chung, Chul Min Yeum, Hoon Sohn, “Laser ultrasonic Im-

aging of a rotating blade,” SPIE International Symposia, Smart Structures & Materials and

Nondestructive Evaluation for Health Monitoring and Diagnostics, San Diego, USA, March

11-15, 2012.

12. Hoon Sohn, Yun Kyu An, Byeongjin Park, “Advancements and challenges in a Laser ultra-

sonic scanning system for structural health monitoring applications,” CIMTEC 4th Interna-

tional Conference on Smart Materials, Structures and Systems, Tuscany, Italy, June 10-15,

2012. (Invited Presentation)

13. Truong Thanh Chung, Byeongjin Park Chul Min Yeum, Hoon Sohn, “Laser ultrasonic imag-

ing of a rotating object with a dropout elimination technique,” The 24th KKCNN Symposium

on Civil Engineering, Hyogo, Japan, December 14-16, 2011.

14. Byeongjin Park, Hoon Sohn, “Localization of crack initiation in a pipe structure using a laser

based acoustic emission technique,” The 8h International Workshop on Structural Health Mon-

itoring, Stanford, USA, September 13-15, 2011.

15. Min Koo Kim, Byeongjin Park, Hoon Sohn, “Optimized pipe damage detection using 3D

geometry information,” KBIM Annual Conference, Seoul, Korea, May 20, 2011.

16. Byeongjin Park, Yun Kyu An, Hoon Sohn, “Noncontact laser ultrasonic imaging for auto-

mated damage detection,” COSEIK Annual Conference, Busan, Korea, April 14-15, 2011.

PATENTS

* 2 US patents (2 pending), 1 International PCT patent and 8 Korean patents (3 granted and 5 pending)

1. Hoon Sohn, Byeongjin Park, Peipei Liu, “손상 검출 방법 및 이를 수행하는 손상

검출 장치 Accelerated Damage Detection using Fixed-distance Noncontact Laser Ultrasonic

Scanning,” Korean Patent (Application # 10-2016-0119443), September 19th, 2016.

2. Hoon Sohn, Byeongju Song, Byeongjin Park, Peipei Liu, “구조물의 진단 방법 및 진단

시스템 Apparatus and technique for structural health monitoring based on nonlinear ultra-

sonic wave,” Korean Patent (Application # 10-2016-0066363), May 30th, 2016.

3. Hoon Sohn, Byeongjin Park, Byeongju Song, Chulwoo Lim, Jaeryong Park, “Development

of noncontact health monitoring system,” US Patent (Application # 14/933/910), December

1st, 2015.

4. Hoon Sohn, Byeongjin Park, “다지점 동시 레이저 가진을 통한 비접촉 결함

검사방법 및 장치(Accelerated Ultrasonic Imaging Technique and System via Multipoint La-

ser Excitation),” Korean Patent (Application # 10-2015-0121805), August 28th, 2015.

- 114 -

5. Hoon Sohn, Byeongjin Park, Byeongju Song, Chulwoo Lim, Jaeryong Park, “비접촉 내구

진단 장치 및 방법 CONTACTLESS DURABILITY DIAGNOSIS APPARATUS AND

METHOD,” Korean Patent (Application # 10-2015-0103773), July 22nd, 2015.

6. Hoon Sohn, Byeongjin Park, “다중 밀도 비접촉식 레이저 스캐닝을 통한 구조물

손상 진단 고속화 장치 및 이의 진단 방법 Accelerated structural damage visualization

through multi-level noncontact laser ultrasonic scanning and visualization method thereof,”

Korean Patent (Publication # 10-1539599-0000), July 21st, 2015.

7. Hoon Sohn, Byeongjin Park, Jinyeol Yang, Jaemook Choi, Soonkyu Hwang, Irl Kim,

Yeonjun Mun, Kyeongyoon Jang, Changkyu Chung, Sungwoo Choi, “검사 장치 및 그

제어 방법 Examining apparatus and examining method thereof,” Korean patent (Application

# 10-2015-0045358), March 31st, 2015.

8. Hoon Sohn, Byeongjin Park, Truong Thanh Chung, “Noncontact laser ultrasonic scanning

and damage visualization for rotating targets,” US Patent (Application # 14/418/427), January

29th, 2015.

9. Hoon Sohn, Byeongjin Park, Truong Thanh Chung, “회전 구조물의 레이저 초음파

영상화 방법 및 장치 Noncontact laser ultrasonic scanning and damage visualization for ro-

tating targets,” Korean Patent (Publication # 10-1410923-0000), June 17th, 2014.

10. Hoon Sohn, Byeongjin Park, Truong Thanh Chung, “회전 물체에 대한 비접촉식 레이저

초음파 스캐닝 및 손상 시각화 기법 Laser ultrasonic imaging of a rotating blade,” Ko-

rean Patent (Publication # 10-1369212-0000), February 25th, 2014.

11. Hoon Sohn, Byeongjin Park and Truong Thanh Chung, “LASER ULTRASONIC IMAGING

METHOD AND LASER ULTRASONIC IMAGING DEVICE FOR ROTATIONAL STRUC-

TURE,” International PCT Patent (PCT-KR2013-006195), February 6th, 2014.

BOOK & BOOK CHAPTERS

1. Hoon Sohn, Byeongjin Park, “Laser based Structural Health Monitoring,” a book chapter in

Encyclopedia of Earthquake Engineering (Editors: Michael Beer, Edorardo Patelli, Ioannis

Kougioumtzoglou and Ivan Siu-Kui Au), Springer, 2014.

2. Hyun Seok Lee, Jin Yeol Yang, Hoon Sohn, Byeongjin Park, “Sensing solutions for assessing

and monitoring of nuclear power plants (NPPs),” Chapter 20 in Sensor Technologies for Civil

Infrastructures, Volume 2/Part II: Case studies in assessing and monitoring specific structures,

Woodhead Publishing, 2014.

- 115 -

EDUCATIONAL EXPERIENCES

Spring 2015 Tutor, CE201 “Mechanics of Materials”, KAIST.

Spring 2014 Teaching Assistant, CE520 “Introduction to Smart Structure Technology,” KAIST.

Spring 2013 Teaching Assistant, CE553 “IT for U-Space,” KAIST.

Fall 2011 Teaching Assistant, CE207 “Elementary Structural Engineering and Laboratory,”

KAIST.

PARTICIPATED PROJECTS

1. Development of a noncontact monitoring system for online quality monitoring of indus-

trial products: Co-funded by EUROSTARS2 Program by the European Union Horizon 2020

Framework Programme and Korea Ministry of Trade, Industry and Energy (Funded:

750,000,000 KRW out of total 2,976,000,000 KRW for 11/01/15 to 10/31/18)

2. Development of an in-situ wind turbine blade inspection system using laser ultrasonics:

Climate Change Research Hub, KAIST (Funded: 160,000,000 KRW for 01/01/14 to 12/31/15)

3. Development of a facility monitoring system through laser scanning: Hyundai-Kia Auto-

mobile (Funded: 80,750,000 KRW for 05/01/13 to 08/31/14)

4. Development of a noncontact wind turbine blade monitoring system using laser ultrason-

ics: The Energy Technology Development Program at Ministry of Knowledge Economy in

Korea (Funded: 1,370,000,000 KRW for 11/01/12 to 10/31/15)

5. Characterization of guided wave propagation in aircraft structures: National Sciences and

Engineering Research Council of Canada (Funded: 50,000 USD for 09/01/13 to 12/31/15)

6. A Smart Scanning System for Green Energy Infrastructure: The National Research La-

boratory Program (NRL) at National Research Foundation of Korea (Funded: 1, 548,000,000

KRW for 05/01/10 to 04/30/15)

7. In-service Monitoring of Nuclear Power Plants: Energy, Environment, Water and Sustain-

ability (EEWS) Program at KAIST (Funded: 150,000,000 KRW for 04/01/10 to 03/31/13)

development of accelerated damage detection techniques using...

Documents