condition monitoring of wind turbine blades

University of Calgary

PRISM: University of Calgary's Digital Repository

Graduate Studies The Vault: Electronic Theses and Dissertations

2018-03-01

Condition Monitoring of Wind Turbine Blades

Sanati, Hadi

Sanati, H. (2018). Condition Monitoring of Wind Turbine Blades. (Unpublished master's thesis).

University of Calgary, Calgary, AB. doi:10.11575/PRISM/10689

http://hdl.handle.net/1880/106410

master thesis

University of Calgary graduate students retain copyright ownership and moral rights for their

thesis. You may use this material in any way that is permitted by the Copyright Act or through

licensing that has been assigned to the document. For uses that are not allowable under

copyright legislation or licensing, you are required to seek permission.

Downloaded from PRISM: https://prism.ucalgary.ca

UNIVERSITY OF CALGARY

Condition Monitoring of Wind Turbine Blades

by

Hadi Sanati

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE

DEGREE OF MASTER OF SCIENCE

GRADUATE PROGRAM IN MECHANICAL ENGINEERING

CALGARY, ALBERTA

MARCH, 2018

© Hadi Sanati 2018

ii

Abstract

The failure of wind turbine blades is a major concern in the wind power industry due to the

resulting high cost. Leading-edge erosion is another issue of blades which may affect their

performance and result in energy loss. It is therefore crucial to develop methods to improve the

integrity of wind turbine blades and to detect surface and subsurface defects before they can result

in blade failure.

This research employed laser scanning to reconstruct the surface of the blade to measure

leading-edge erosion. Computational Fluid Dynamics simulations were used to determine the

deterioration of aerodynamic characteristics resulting from leading-edge erosion. The results

suggest that it is possible to successfully evaluate the aerodynamic characteristics of eroded and

clean airfoils.

Different methods are available to detect subsurface damage in blades but most require close

proximity between the sensor and the blade. To address this limitation, the use of thermography

as a non-contact method was developed in this study. Both passive and active thermography

techniques were investigated for different purposes. Passive thermography can be used to detect

internal defects on an operating blade, whereas active thermography is restricted to pre-delivery

blade inspection and site inspections when the blades are removed from the turbine. Pulsed and

step heating as active thermography methods were studied.

The raw thermal images captured by both active and passive thermography demonstrated

that image processing was required to improve the quality of thermal data. Different image

processing algorithms, including Thermal Signal Reconstruction, Principal Component

Thermography, Matched Filters, and Pulsed Phase Thermography, were used to increase the

iii

thermal contrasts of subsurface defects in thermal images obtained by active thermography. A

method called “Step Heating Phase and Amplitude Thermography”, which applies a transform-

based algorithm on step heating data, was developed. This method was also applied to passive

thermography results. The outcomes of image processing on both active and passive thermography

indicated that the techniques employed could considerably increase the quality of the images and

the visibility of internal defects.

iv

Acknowledgment

I would like to acknowledge those who extended out their hand and helped me during this

project.

I would like to express my deepest sense of appreciation to my supervisor and co-

supervisor, Dr. David Wood and Dr. Qiao Sun for their great support and invaluable guidance

throughout this project. I am greatly indebted to them for their vital role in going above and beyond

to equip me with all the resources and priceless directions required for my research. Without their

indispensable advices and support, this work would not have been possible.

I am very grateful to Dr. Derek Lichti and Dr. Simon Park, who kindly allowed me to use

some of their equipment for different parts of my research.

Special thanks to my many friends that have shown me endless support and encouragement

in all my endeavors.

Last but not least, I would like to express my heartfelt and limitless gratitude to my parents

and family, especially Mehdi, Zahra, Hossein, and Diana, for their boundless love, support and

sacrifices all throughout my life. Without them, I have not been able to go through this journey.

v

Dedication

“To my parents”

vi

Table of Contents

ABSTRACT ....................................................................................................................... ii

ACKNOWLEDGMENT ................................................................................................... iv

DEDICATION ....................................................................................................................v

TABLE OF CONTENTS .................................................................................................. vi

LIST OF TABLES .......................................................................................................... viii

LIST OF FIGURES ............................................................................................................x

LIST OF SYMBOLS, ABBREVIATION AND NOMENCLATURE ......................... xvii

INTRODUCTION .........................................................................................1

1.1 IMPORTANCE OF WIND ENERGY AND CONDITION MONITORING OF WIND

TURBINE BLADES ...................................................................................................1 1.2 OBJECTIVES ................................................................................................................6 1.3 THESIS OUTLINE ........................................................................................................7

LITERATURE REVIEW .............................................................................8 2.1 BACKGROUND ...........................................................................................................8

2.1.1 Leading-Edge Erosion ............................................................................................8 2.1.2 Reconstruction of Eroded Blade Surface ..............................................................10

2.1.3 Wind Turbine Blade Condition Monitoring Techniques ......................................16 2.2 THERMOGRAPHY ....................................................................................................22

2.2.1 Active Thermography ...........................................................................................24

2.2.2 Passive Thermography ..........................................................................................30

THEORY OF THERMOGRAPHY AND IMAGE PROCESSING.........31 3.1 PULSED THERMOGRAPHY THEORY ...................................................................31 3.2 STEP HEATING THERMOGRAPHY THEORY ......................................................34

3.3 IMAGE PROCESSING ALGORITHMS IN THERMOGRAPHY ............................34

3.3.1 Thermal Signal Reconstruction (TSR) .................................................................35 3.3.2 Matched Filters (MF) ............................................................................................36 3.3.3 Principal Component Thermography (PCT) .........................................................38 3.3.4 Pulsed Phase Thermography (PPT) ......................................................................40 3.3.5 Quantitative evaluation .........................................................................................43

EXPERIMENTAL PROCEDURES AND SETUPS .................................45

4.1 MATERIAL .................................................................................................................45 4.2 EXPERIMENTAL SETUPS .......................................................................................47

4.2.1 3D Laser Scanning ................................................................................................47 4.2.2 Passive Thermography ..........................................................................................49

4.2.3 Active Thermography ...........................................................................................52

vii

RESULTS AND DISCUSSION .................................................................55

5.1 AERODYNAMIC CHARACTERISTICS ..................................................................55

5.1.1 Reconstruction of the Blade Section Surface .......................................................55 5.1.2 CFD Simulation Using ANSYS Fluent 16.2 ........................................................57 5.1.3 Models validation and CFD simulation results .....................................................58

5.2 ACTIVE THERMOGRAPHY.....................................................................................64 5.2.1 Raw Pulsed and Step Heating Thermography Data ..............................................64

5.2.2 Thermal Image Processing ....................................................................................72 5.3 PASSIVE THERMOGRAPHY ...................................................................................98

5.3.1 Day time experiment .............................................................................................98 5.3.2 Monitoring during heating and cooling ..............................................................101 5.3.3 Passive Thermal Image Processing .....................................................................103

CONCLUSION AND FUTURE WORKS .................................................106 6.1 CONCLUSION ..........................................................................................................106 6.2 FUTURE WORKS.....................................................................................................109

APPENDIX A: SUMMARY OF INSTRUMENTS SPECIFICATIONS ......................123

A.1. Laser Scanner .......................................................................................................123

A.2. IR Camera ............................................................................................................124 A.3. Power Supply .......................................................................................................125 A.4. Flash Lamp ...........................................................................................................126

A.5. Halogen Lamps ....................................................................................................126

A.6. Computer System and Software ...........................................................................128

APPENDIX B: Copyright Permissions ..........................................................................129

viii

List of Tables

Table 1.1. Typical configurations of wind turbines (Ciang, Lee, and Bang 2008). ....................... 3

Table 2.1. Summary of main NDT methods for monitoring the condition of blades ................... 20

Table 2.2. Sub-division of infrared radiation according to temperature range (Byrnes 2008). .... 22

Table 5.1. Aerodynamic characteristic comparison of both reference and clean airfoils ............. 64

Table 5.2. SNR related to data captured by application of SAM on pulsed thermal data ............ 79

Table 5.3. SNR related to data captured by application of SAM on step heating thermal data ... 79

Table 5.4. SNR related to data captured by application of ACE on step heating thermal data .... 79

Table 5.5. SNR related to data captured by application of F statistic on step heating thermal data

....................................................................................................................................................... 79


....................................................................................................................................................... 79

Table 5.7. SNR of 3th orthogonal function of thermograms captured at cooling after 10s heating

....................................................................................................................................................... 84


....................................................................................................................................................... 84


....................................................................................................................................................... 84


....................................................................................................................................................... 84

Table 5.11. SNR of 2th orthogonal function of thermograms captured at 20s heating ................. 87



ix




Table 5.17. SNR of phase image obtained by application of PPT on pulsed thermal data .......... 90

Table 5.18. SNR of amplitude image of thermograms captured during cooling after 75s heating

....................................................................................................................................................... 97

Table 5.19. SNR of phase image of thermograms captured during 75s heating .......................... 98

Table 5.20. SNR of amplitude image of thermograms captured during 75s heating .................... 98

Table A.1. Technical specifications of Creaform HandyScan laser scanning

(“https://www.creaform3d.com” 2018) ...................................................................................... 123

Table A.2. Specifications of FLIR T1030Sc infrared Camera (“Http://www.flir.ca/home/” 2018)

..................................................................................................................................................... 124

Table A.3. Specifications of Speedotron 4803CX power supply (“Http://www.speedotron.com/”

2018) ........................................................................................................................................... 125

x

List of Figures

Figure 1.1. Global incremental installation of wind energy until the end of 2013 (“GWEC. Global

Wind Report Annual Market Update” 2016). ................................................................................. 1

Figure 1.2. Estimation of cumulative wind power supply between 2015 and 2018 (de Azevedo,

Araújo, and Bouchonneau 2016). ................................................................................................... 2

Figure 1.3. Development of wind turbine size and captured power over time (Molina and Mercado

2011). .............................................................................................................................................. 4

Figure 1.4. The percentage of annual failure rate of wind turbines for different rated powers with

respect to the operational year (Echavarria et al. 2008). ................................................................. 5

Figure 2.1. Formation of pits, gouges and delamination on the leading-edge area after (a) 1 year

(b) 2 years (c)10 years and (d) more than 10 years in service (Keegan, Nash, and Stack 2013).

Copyright 2013, Used with permission from Journal of Physics D: Applied Physics. .................. 9

Figure 2.2. Non-contact measurement of a surface (Summers et al. 2016). Copyright 2016, Used

with permission from Robotic and Computer-Integrated Manufacturing. Elsevier. .................... 12

Figure 2.3. Inspection of blade using LR through scanning from different locations(Summers et

al. 2016). Copyright 2016, Used with permission from Robotic and Computer-Integrated

Manufacturing. Elsevier. ............................................................................................................... 12

Figure 2.4. Frequency deference between reflected and generated signals (Talbot et al. 2016). . 13

Figure 2.5. The principle of phase shift (Rohrbaugh 2015). ......................................................... 14

Figure 2.6. Schematic of the triangulation principle (Malhorta, Gupta, and Kant 2011) ............. 15

Figure 2.7. Schematic of laser tracker principle (Ouyang, Liang, and Zhang 2006). ................... 16

xi

Figure 2.8. A cross section of a wind turbine blade consisting of (1) leading edge, (2) trailing edge,

(3) shear webs glued to (4) suction and pressure sides, (A) composite material, and (B) wood or

plastic foam. .................................................................................................................................. 17

Figure 2.9. Inspection of wind turbine blade using an industrial climber (Marsh 2011). Copyright

2011, Used with permission from Reinforced Plastics. Elsevier. ................................................. 19

Figure 2.10. Schematic of thermal image acquisition and processing using an IR camera

(Meinlschmidt and Aderhold 2006). ............................................................................................. 23

Figure 2.11. Summary of thermography techniques (Matovu 2015)............................................ 24

Figure 2.12. A robotic in-situ rotor blade inspection schematic (Chatzakos, Avdelidis, and

Hrissagis 2010). Copyright 2011, Used with permission from Robotics Automation and

Mechatronics (RAM), IEEE. ........................................................................................................ 25

Figure 2.13. Schematic of pulse thermography setup (C. Ibarra-Castanedo et al. 2007). ............ 26

Figure 2.14. Schematic of experimental setup of lock-in thermography (Pawar 2012). .............. 28

Figure 2.15. Thermal wave propagation and thermal wave phase difference and amplitude in lock-

in thermography (Pawar 2012). .................................................................................................... 29

Figure 2.16. Experimental lock-in thermography for a wind turbine blade (Manohar and Lanza di

Scalea 2013). Copyright 2013, Used with permission from Structural Health Monitoring. SAGE

Publications. .................................................................................................................................. 29

Figure 3.1. (a) Thermograms recorded as a 3D matrix in time domain, and (b) temperature variation

for a pixel in a sound area (Clemente Ibarra-Castanedo and Maldague 2004). Copyright 2004,

Used with permission from Journal of Research in Non-destructive Evaluation. Taylor & Francis.

....................................................................................................................................................... 32

xii

Figure 3.2. (a) A typical temperature variation of defective and non-defective areas, and (b) ATC

(Manohar 2012). ........................................................................................................................... 33

Figure 3.3. Decomposition of square pulse and thermal decay pulse (Castanedo 2005). ............ 41

Figure 3.4. Transformation of step heating thermal response during heating and cooling. ......... 42

Figure 4.1. The damaged blade section used in passive thermography experiment. .................... 46

Figure 4.2. Geometry and pattern of blind holes in the defect plate. (a) shows the rear of the plate

(b) gives the size and depth of the blind holes. The blind holes in each row, labeled A-D have the

same diameter but varying depth. ................................................................................................. 48

Figure 4.3. Experimental setup of laser scanning by Leica HDS 6100. ....................................... 50

Figure 4.4. HandyScan laser scanning experiment. ...................................................................... 50

Figure 4.5. Passive thermography experimental set-up. ............................................................... 52

Figure 4.6. Pulse thermography experimental set-up. .................................................................. 53

Figure 4.7. Step heating thermography experimental set-up. ....................................................... 54

Figure 5.1. (a) Reconstructed blade section, (b) an image of the blade section with small surface

defects, and (c) reconstruction of blade section presented in (b). ................................................. 56

Figure 5.2. The leading-edge region of (a) clean and (b) rough airfoils obtained by HandyScan

laser scanner. ................................................................................................................................. 57

Figure 5.3. (a) Multiblock structure and (b) generated mesh around the airfoil for CFD simulation.

....................................................................................................................................................... 59

Figure 5.4. Comparison of clean and DU 96-W-180 airfoils. ...................................................... 60

Figure 5.5. (a) Experimental values of Cl/Cd at different Re values for reference airfoil (Sareen,

Sapre, and Selig 2014), (Timmer and Van Rooij 2003) and (b) numerical value of Cl/Cd at

Re=3,000,000 for clean and rough airfoils. .................................................................................. 61

xiii

Figure 5.6. Numerical (lines) and experimental (blue dots) Cl versus of the reference airfoil at

Re=3,000,000 (Timmer and Van Rooij 2003). ............................................................................. 62

Figure 5.7. Numerical Cl at Re=3,000,000 for clean (solid lines) and reference airfoils (dotted

lines) at varying . ...................................................................................................................... 62

Figure 5.8. (a) Cl and (b) Cd at Re=3,000,000 of clean and rough airfoils at varying . ............ 63

Figure 5.9. (a) A thermogram immediately after pulse, and (b) a thermogram after 5s of cooling,

(c) temperature decay of sample in marked defect (“D” in part (b)) and sound (“S”) positions. . 65

Figure 5.10. A thermogram after (a) 40s heating and (b) 15s cooling. Time history of temperature

at defect and sound points during (c) heating and (d) cooling. ..................................................... 66

Figure 5.11. Step heating thermograms after (a) 15s, (b) 20s, (c) 30s and (d) 75s of heating. .... 67

Figure 5.12. (a) Positions of temperature distribution profiles. (b) Temperature distribution profiles

using flash thermography. ............................................................................................................. 69

Figure 5.13. (a) Temperature distribution after 75s of heating (b) temperature distribution after 6

s of cooling. ................................................................................................................................... 70

Figure 5.14. Thermal profiles for different heating time of step heating thermography at (a) first

row (b) second row (c) third row. ................................................................................................. 71

Figure 5.15. Effect of depth and size of defect on the temperature distribution in sample heated by

halogen lamps for 75s. .................................................................................................................. 72

Figure 5.16. Absolute thermal contrast of defect plate under pulse thermography technique at

defects located in (a) first row (b)second row (c) third row and (d) fourth row. .......................... 74

Figure 5.17. ATC of thermograms obtained by step heating thermography after 75s of heating at

(a) first row (b) second row (c) third row and (d) fourth row. ...................................................... 75

xiv

Figure 5.18. Four MF algorithms results of thermograms obtained by flash thermography (a) SAM,

(b) ACE, (c) t-statistic and (d) F statistic ...................................................................................... 76

Figure 5.19. Four matched filters including (a) SAM, (b) ACE, (c) t-statistic and (d) F statistic

when the specimen is under step heating thermography. ............................................................. 77

Figure 5.20. (a) Signal values over the lines crossing the defects and (b) background noise. ..... 78

Figure 5.21. (a) First (b) second (c) third and (d) fourth orthogonal functions obtained from pulsed

thermograms. ................................................................................................................................ 81

Figure 5.22. (a) First (b) second (c) third and (d) fourth orthogonal functions extracted from

thermograms recorded during cooling after 20s of heating. ......................................................... 82

Figure 5.23. Third orthogonal functions extracted from thermograms recorded during cooling after

(a) 10s (b) 30s (c) 40s and (d) 75s of heating. .............................................................................. 83


thermograms recorded during 30s heating. ................................................................................... 85

Figure 5.25. Second orthogonal function of thermograms captured at (a) 20s (c) 40s and (e) 75s

heating and third orthogonal function extracted from thermograms recorded at (b) 20s (d) 40s and

(f) 75s heating. .............................................................................................................................. 86

Figure 5.26. (a) Raw thermogram and (b) phase image (acquisition time =53.2 s) obtained from

thermograms recorded at cooling period after flashing the surface. ............................................. 90

Figure 5.27. Application of FFT on pulsed thermograms and obtained phase images of specimen

with minimum frequency of a) 0.149Hz (acquisition time = 6.66s), b) 0.075Hz (acquisition time

= 13.33s) c) 0.05 Hz (acquisition time = 20s) d) 0.027 Hz (acquisition time = 36.66s). ............. 92

Figure 5.28. Application of FFT on step heating thermograms and amplitude images of specimen

captured at cooling process after (a) 10s (fmin=0.018Hz) (c) 20s (fmin=0.012Hz) (e) 30s

xv

(fmin=0.0154Hz) of heating and phase images of specimen obtained at cooling process after (b) 10s

(d) 20s (f) 30s of heating. .............................................................................................................. 93


captured at cooling process after (a) 40s (fmin=0.0144Hz) and (c) 75s (fmin=0.015Hz) of heating

and phase images of specimen obtained at cooling process after (b) 40s and (d) 75s of heating. 94

Figure 5.30. Amplitude images of specimen captured at heating after (a) 20s (fmin=0.075Hz) (c)

40s (fmin=0.0292) and (e) 75s (fmin=0.0227Hz) of heating and phase images of specimen obtained

at heating period after (b) 20s (d) 40s and (f) 75s of heating. ...................................................... 95

Figure 5.31. Normalized amplitude value distribution of defects with minimum frequency of

0.015Hz as thermograms obtained during cooling after 75s of heating. ...................................... 96

Figure 5.32. Normalized phase value distribution of defects with minimum frequency of 0.015Hz

as thermograms obtained during 75s of heating. .......................................................................... 97

Figure 5.33. Thermographic results of experiment at morning around 9 am. The vertical arrows

indicate the shear webs. ................................................................................................................ 99

Figure 5.34. Raw thermograms at (a) noon and (b) around 6 pm (sunrise and sunset were around

5.53 am and 9.30 pm, respectively). The vertical arrows and dashed lines indicate the shear webs.

..................................................................................................................................................... 100

Figure 5.35. Thermographic results at (a) heating and (b) cooling. ........................................... 102

Figure 5.36. (a) Phase images of passive thermograms captured during the morning at a frequency

of 0.00184 Hz and (b) amplitude image of passive thermograms recorded during the morning at a

frequency of 0.0165 Hz............................................................................................................... 103

Figure 5.37. (a) Amplitude image at a frequency of 0.0318 and (b) phase image at a frequency of

0.0053 extracted from raw passive thermal images captured at noon. ....................................... 104

xvi

Figure A.1. (a) HandyScan laser scanner (“https://www.creaform3d.com” 2018), (b) T1030 SC

IR camera (“Http://www.flir.ca/home/” 2018) ........................................................................... 125

Figure A.2. Speedotron 4803CX LV power supply .................................................................. 126

Figure A.3. Speedotron 206VF strobes...................................................................................... 127

Figure A.4. Portable halogen lamps with the power of each 500W ........................................... 127

xvii

List of Symbols, Abbreviations, Nomenclature

Abbreviations Definition

ACE Adaptive Coherence Estimator

AE Acoustic Emission

AOA Angles of Attack

ATC Absolute Thermal Contrast

Az Azimuth

CFD Computational Fluid Dynamics

CLR Coherent Laser Radar

CMF Clutter Matched Filter

COE Cost of Energy

DFT Discrete Fourier Transform

El Elevation

ET Eddy Current Thermography

FBG Fibre Bragg Grating

FFT Fast Fourier Transform

FPA Focal Plane Arrays

GAEP Gross Annual Energy Production

IR Infrared

LIDAR Light Detection and Ranging

LT Lock-in Thermography

LR Laser Radar

MF Matched Filters

MSP Mirrored Spherical Probe

NDE Non-Destructive Evaluation

NDT Non-Destructive Testing

PCA Principal Component Analysis

PCT Principal Component Thermography

xviii

PPT Pulsed phase thermography

PT Pulsed Thermography

Rg Range

SAM Spectral Angle Map

SHPAT Step Heating Phase and Amplitude Thermography

SMF Simple Matched Filters

SNR Signal to Noise Ratio

SVD Singular Value Decomposition

TSR Thermal Signal Reconstruction

UAS Unmanned Aerial Systems

UT Ultrasonic Testing

VT Vibro-Thermography

Symbols Definition

Cl Lift Coefficient

Cd Drag Coefficient

Angle of Attack

Re Reynolds Number

Q Input thermal energy

T Temperature distribution

t Time

z Depth

thermal diffusivity

e emissivity

k Thermal conductivity

Phase Shift

c Speed of Light in Air

Mass density

pC Specific heat capacity

xix

Td Temperature at defective area

TSa Temperature at sound area

L Thickness of object

ierfc( )x First integral of the complementary error function

Tobs Observed temperature

Tref Reflected temperature from defective area

Tideal Ideal temperature at sound area

C Covariance matrix

A Data cube of thermal images sequence

Nx and Ny Pixel dimensions of image

Nt Number of frames

Diagonal matrix of singular values of matrix A

U and VT Singular vectors of a matrix A

f Frequency increment

N Number of Frames

Re Real part of the transformation function

Im Imaginary part of the transformation function

t Time interval

Standard Deviation

1

INTRODUCTION

1.1 Importance of Wind Energy and Condition Monitoring of Wind Turbine Blades

Over the last century, development in world’s industry and population growth have led to

increased global energy consumption. Fossil fuels are the primary source of energy, which has led

to various environmental issues such as air pollution and global warming. Since the world energy

crisis in the 1970s, renewable energy has experienced fast growth (Ciang, Lee, and Bang 2008).

Wind energy has unique potentials, where its low cost of operation, extensive availability, and

excellent safety make it an attractive source of energy. Compared to all other types of available

renewable energies, installed wind power capacity is expanding the most quickly (Nie and Wang

2013). According to a report by the Global Wind Energy Council (GWEC), the yearly incremental

installation of wind energy has increased by 25.31% between 1999 and 2013 (Irwanto et al. 2014).

Another report by GWEC shows that the incremental installed wind capacity in 2016 was 486790

MW, which is almost 12.5% more than the installed capacity in 2015. This growth is shown in

Figure 1.1 (“GWEC. Global Wind Report Annual Market Update” 2016).

Figure 1.1. Global incremental installation of wind energy until the end of 2013 (“GWEC.

Global Wind Report Annual Market Update” 2016).

2

Wind farms supply about 2.3% of the world’s electricity (Soua et al. 2013). Figure 1.2 shows

that wind power capacity is expected to increase by around 43% between 2015 and 2018 (de

Azevedo, Araújo, and Bouchonneau 2016). Another forecast predicts that wind energy supply may

reach 2.6 TWh by 2020, at which point it will provide about 11.5-12.3% of the world’s total

electrical power. This value is projected to continue to grow to 21.8% by 2030 (Soua et al. 2013).

Figure 1.2. Estimation of cumulative wind power supply between 2015 and 2018 (de Azevedo,

Araújo, and Bouchonneau 2016).

Due to rapid growth in the wind energy sector, it is necessary to develop new designs and

condition monitoring techniques to fulfill the power demand of customers (Tippmann 2014). To

accomplish this, different challenges must be addressed. The general designed lifetime of wind

turbines is around 20 years. During this time, they are subjected to variable loads, which may

impose severe fatigue and mechanical stress on them (J Ribrant 2006; Tchakoua, Wamkeue, and

Ouhrouche 2014). Consequently, different types of failures may occur in different components of

3

wind turbines which may result in a shutdown rate of up to 3% of their lifespan (Echavarria et al.

2008; Hahn and Durstewitz 2007; Tchakoua, Wamkeue, and Ouhrouche 2014).

Wind turbine blades are also in contact with airborne particles which result in the loss of

material from their surface around the leading edge known as “leading-edge erosion”. Leading-

edge erosion is a real threat to the integrity of blade’s surface and depending on the severity of

erosion it may increase the drag coefficient around 6-500% (Sareen, Sapre, and Selig 2014). Drag

increase has a negative impact on annual power generation, for example 80% increase of drag,

which is a result of light leading-edge erosion, may lead to almost 5% drop in annual captured

energy (Sareen, Sapre, and Selig 2014).

The size of the wind turbine is another factor that may affect the frequency of failure. In

recent years, larger wind turbines have been developed to capture more energy.

Table 1.1 lists typical configurations of horizontal wind turbines (Ciang, Lee, and Bang

2008). Figure 1.3 shows the progression of wind turbine size and the amount of power they capture

over time (Molina and Mercado 2011).

Table 1.1. Typical configurations of wind turbines (Ciang, Lee, and Bang 2008).

Rated power (kW) Range of hub height (m) Blade length (m) Price excluding

foundation (1000 US$)a

150 35-60 12-13 -

600 45-80 16-23 510-670

1000 50-90 26-29 960-1410

1500 60-110 31-37 1530-2030

2000 60-100 34-39 1860-2200

a Based on equivalent currency between Euro and US dollar in year 2003 (1 Euro=1.13 US dollars)

4

Research on onshore wind turbines demonstrates that larger wind turbines experience greater

rates of failure (Tchakoua, Wamkeue, and Ouhrouche 2014). The results of a 13-year study on the

failure rate of wind turbines with different power ranges over several years is shown in Figure 1.4;

as size increases, so does the annual failure rate, which is the amount of incidents for each wind

turbine during a year (Echavarria et al. 2008).

The results of a study on 1500 wind turbines over 15 years demonstrate that 67% of failures

occur in five major components: electrical system, hydraulic systems, control system, sensors, and

blades (Tchakoua, Wamkeue, and Ouhrouche 2014). Research conducted by Sandia National

Laboratories shows that blade failure accounts for a higher percentage of wind turbine shutdowns

than other components (Tippmann 2014).

Figure 1.3. Development of wind turbine size and captured power over time (Molina and

Mercado 2011).

A preventative maintenance schedule is a major step towards preventing failure in the wind

energy industry (Walford 2006; Tchakoua et al. 2013). Based on one estimate, the operation and

5

maintenance costs of wind turbines are more than twice those of natural gas production (Greco et

al. 2013).

Figure 1.4. The percentage of annual failure rate of wind turbines for different rated powers with

respect to the operational year (Echavarria et al. 2008).

Another survey estimates the operation and maintenance cost of onshore wind turbines at

almost 10% of the total energy cost (Blanco 2009). This increases to 30% for offshore wind

turbines (Blanco 2009). In general, 25-35% of the maintenance of offshore wind turbines is

considered “preventative”, to reduce downtime, and 65-75% is “corrective” to fix faults. In

contrast, the ratio of preventative and corrective maintenance strategies for onshore wind turbines

is split almost evenly (Verbruggen 2003). Unscheduled maintenance of wind turbines is expensive,

so developing a preventative maintenance strategy is essential (Adams et al. 2011). A preventative

maintenance schedule employs different condition monitoring techniques to evaluate the health

condition of wind turbine components and to determine the most efficient time between corrective

and preventative maintenance schedules (Besnard and Bertling 2010; Amayri, Tian, and Jin 2011;

McMillan and Generation 2008; Fischer, Besnard, and Bertling 2012).

6

1.2 Objectives

The overall objective of this research is to develop new methods to monitor the condition of

wind turbine blades with the goals of predicting their failure under harsh weather conditions and

cyclic loading and allowing better scheduling of blade repair. This research pursues the following

objectives to fill the knowledge gaps in condition monitoring of wind turbine blades:

Objective 1: Developing a method to estimate the aerodynamic performance deterioration

of wind turbine blades due to leading-edge erosion. This common phenomenon may result in

energy loss, followed by structural failure. Introducing a method to evaluate the degradation of

aerodynamic characteristics due to changes in the blade’s shape is essential.

Objective 2: Developing a non-destructive testing method to precisely detect subsurface

defects, even small defects with a size of some millimeters, of the blade before installation.

Different flaws may be introduced to and propagated across blades during either the manufacturing

process or operation. It is therefore necessary to develop a Non-Destructive Testing (NDT)

technique that is capable of precise fault detection before installation or while a blade is on the

ground for major maintenance.

Objective 3: Developing a new method to monitor the condition of operational wind turbine

blades. Blades may experience different kinds of surface and subsurface damages during their

operation as a result of environmental conditions and the dynamic nature of loads on the blades.

Since accessing the blade is not easy while it is in service, a new method must be developed to

remotely inspect it. This new method will eliminate the limitations associated with conventional

condition monitoring techniques of wind turbine blades, such as the requirement of sensor

attachment and high amount of both inspection time and downtime.

7

1.3 Thesis Outline

This thesis consists of six chapters. The motivations and objectives of this research are

presented in chapter 1. Chapter 2 reviews the literature regarding leading edge erosion,

reconstruction of an object, NDT techniques, and thermal imaging methods. This chapter also

discusses the principles of passive and active thermography methods, while the theories behind

active thermal imaging and different image processing algorithms are covered in chapter 3. The

experimental procedures are outlined in chapter 4. Chapter 5 contains the results and discussion

regarding experimental thermography for inspection of subsurface defects and Computational

Fluid Dynamics (CFD) simulations for aerodynamic performance evaluation of reconstructed

airfoils. Finally, chapter 6 provides a summary and conclusion of the research as well as

recommendations and comments about future works.

8

LITERATURE REVIEW

2.1 Background

The most crucial components of wind turbines, the blades, are susceptible to different types

of damage during their operation. The failure of one blade may damage nearby blades and wind

turbines, increasing the total damage cost. Different surface and subsurface defects, including

delamination, cracks, air inclusion, fibre-matrix debonding, and others, may be introduced to the

blade during manufacturing or operation (Tao et al. 2011). Once the wind turbine is in operation,

harsh environmental conditions and airborne particles such as storms, hail, snow, rain, ice, and

dirt, expose wind turbine blades to more potential harm. These environmental hazards, in

conjunction with the increasing size of wind turbines, make the integrity of blades a critical and

challenging task. Leading-edge erosion and subsurface damage and their propagation are the most

common blade problems (Keegan, Nash, and Stack 2013).

2.1.1 Leading-Edge Erosion

Wind turbine blade erosion usually begins with the creation of pits near the blade’s leading-

edge, especially around the tip. The erosion then propagates toward the root of the blade and over

increasing portions of the suction and pressure sides. As the number of pits grows over time, they

combine and create gouges. Under extreme conditions, gouges propagate more which can result

in delamination in the leading-edge area as illustrated in Figure 2.1 (Sareen, Sapre, and Selig

2014).

Generated rough surfaces may also have negative effects on the blade’s aerodynamic

performance, as flow can be affected by roughness on the blade’s surface. The blade element

9

suffers a decrease in the lift coefficient, while the drag increases (Mendez, Munoz, and Munduate

2015). The lift-to-drag ratio of the blade element is therefore reduced at all angles of attack (AOA),

which decreases the Gross Annual Energy Production (GAEP) and increases the Cost of Energy

(COE) (Sareen, Sapre, and Selig 2012; Mayor, Moreira, and Muñoz 2013; Mendez, Munoz, and

Munduate 2015).

(a)

(b)

(c)

(d)

Figure 2.1. Formation of pits, gouges and delamination on the leading-edge area after (a) 1 year

(b) 2 years (c)10 years and (d) more than 10 years in service (Keegan, Nash, and Stack 2013).

Copyright 2013, Used with permission from Journal of Physics D: Applied Physics.

Despite the importance of the leading-edge erosion effect on the performance of wind turbine

blades, few investigations have been carried out in this area. Former studies have mostly focused

on the effect of leading-edge roughness due to the accumulation of insects and dust and the

formation of ice, rather than on the impacts of erosion (Somers 2004; Khalfallah and Koliub 2007).

This research sets out to develop an effective method of determining the aerodynamic

characteristics of clean and eroded airfoils. For this purpose, the eroded blade shape is first

10

reconstructed, and then analyzed by CFD using ANSYS Fluent to assess its aerodynamic

characteristics deterioration.

2.1.2 Reconstruction of Eroded Blade Surface

It is essential to use and develop Non-Destructive Evaluation (NDE) methods to assess the

integrity of wind turbine blades and subsequently predict their failure. Non-contact methods are

the most efficient and practical approaches to inspecting the surface of the blade. It is mainly

because access to the blade’s surface is difficult. A variety of non-contact methods for the

inspection and 3D reconstruction of objects exist, with laser scanning and photogrammetry among

their most common. In order to capture small variations in the surface resulting from erosion

submillimeter accuracy is required. Also, the method employed must be usable on an operating

blade.

Photogrammetry is a feature detection method where a set of digital stereo images of an

object is recorded. Matching algorithms are employed to find any overlap between the images.

Camera parameters are then calibrated to calculate each point’s co-ordinate, which ultimately

results in the extraction of a 3D model of the object (Remondino and El-Hakim 2006).

The most important consideration for photogrammetry is that random texture patterns on the

surface must be used as references to compare images captured from different angles (D’Apuzzo

1998). Since the surfaces of wind turbine blades are smooth without any random texture, this

method cannot be relied upon to accurately reconstruct their surface. It is therefore not accurate

enough for the submillimeter accuracy that is required to capture the roughness and shape variation

created by erosion on the blades.

11

Another common technique for creating 3D models of objects is 3D laser scanning. This

method is employed across a wide range of applications, including 3D reverse engineering,

measurement, and quality control (Bogue 2010; Xu et al. 2011). Different sets of measurement

points, known as “point clouds”, can be captured by scanning an object. Once a set of point clouds

that covers the object’s entire surface is obtained, it can be employed to extract a 3D model of the

object (Chen et al. 2013).

The complete point cloud of the object’s entire surface can be created by combining localized

point clouds. It should be noted that each point cloud obtained from a scan has a local co-ordinate

system representing a section of the object, which must be transformed into a global co-ordinate

system to generate a uniform point cloud of the object (Chen et al. 2013). The process of

transforming a single point cloud to a unique global co-ordinate system is known as “registration”

which is one of the main challenges when pursuing 3D reconstruction with laser scanning (Chen

et al. 2013). A large number of points must be measured to reconstruct a surface with high accuracy

(Savio, Chiffre, and Schmitt 2007).

Several companies such as Creaform and Nikon are active in the production of commercial

precise laser scanners. Each of these companies provides a range of instruments with different

accuracy ranges, prices, and applications.

Recent scanners by Nikon Metrology U.K, Coherent Laser Radar (CLR) can precisely

measure blade surfaces (Talbot et al. 2016). This type of laser scanner can achieve high accuracy

from a large distance, and can measure and inspect large and complex geometries. Figure 2.2

illustrates a schematic of a Laser Radar (LR) scanning system (Summers et al. 2016). LR is

currently employed by Vestas Winds System A/S to measure the surfaces of blades (Talbot et al.

2016). Multiple laser scanner positions are used to cover the entire surface. Figure 2.3 shows a

12

schematic of a blade surface measurement using LR where six different locations are used to

inspect the entire surface of a 60 m wind turbine blade (Summers et al. 2016).

LRs can generally work in any lighting within a Range (Rg) of 50 m. The laser beam

generated by this devise can cover an Azimuth (Az) of 360o and an Elevation (El) of 120o, which

is depicted in Figure 2.2.

Figure 2.2. Non-contact measurement of a surface (Summers et al. 2016). Copyright 2016, Used

with permission from Robotic and Computer-Integrated Manufacturing. Elsevier.

Figure 2.3. Inspection of blade using LR through scanning from different locations(Summers et

al. 2016). Copyright 2016, Used with permission from Robotic and Computer-Integrated

Manufacturing. Elsevier.

13

LRs use a frequency-based method to measure the position of a point on the surface of an

object. A LR emits a laser beam towards the object and captures the return beam. As illustrated in

Figure 2.4, the frequency difference between the emitted beam and the reflected beam are used to

calculate the range between the scanner and the object as (Talbot et al. 2016; Summers et al. 2016):

0.667

fRg

(in microns) (2.1)

Figure 2.4. Frequency deference between reflected and generated signals (Talbot et al. 2016).

This system can measure Rg with an accuracy of 10 m (Summers et al. 2016). The precise

positions of points on the surface can be obtained by calculating Rg, Az and El.

Terrestrial laser scanners are another device that can be used to reconstruct the surfaces of

objects. In this method, an amplitude-modulated light beam with a frequency f is directed toward

the object and bounced back to the scanner. The phase of the reflected light changes, which can be

used to determine the time-of-flight. The principle of this method is illustrated in Figure 2.5.

14

Figure 2.5. The principle of phase shift (Rohrbaugh 2015).

The scanner contains a sensor that can capture the phase shift between the emitted signals

and the received signals. Then, the time-of-flight can be calculated as (Rohrbaugh 2015):

2Time of flight

f

(2.2)

Where is phase shift between the emitted and reflected beams and f is the modulated

frequency. The range between the object and the scanner can be calculated using :

2

c tR

(2.3)

where R is the range, c is the speed of light in air ( 83 10 ), and t is the time-of-flight.

Handheld laser scanners can reconstruct objects with high accuracy. These devices use

triangulation to reconstruct the surface. Triangulation laser scanners emit a laser beam toward the

object. The projected laser point on the surface is monitored by a camera. Since the lines

connecting the camera, the laser scanner, and the projected laser point on the surface make a

15

triangle, this method of reconstruction is known as triangulation. Figure 2.6 shows a schematic of

the triangulation principle.

Figure 2.6. Schematic of the triangulation principle (Malhorta, Gupta, and Kant 2011)

The distance between the scanner and the object is obtained by calculating the length of the

side of triangle that lies between them. We know the distance between the laser scanner and the

camera, so the length of one side of triangle is known. The angle of the corner of scanner is also

known. The angle of the camera corner can be calculated using the focal length and the field of

view of camera. By simple mathematical calculations, the remaining parameters of the triangle,

including the distance between scanner and projected laser point on the surface, can be determined

(Malhorta, Gupta, and Kant 2011). After capturing the co-ordinates of each point on the object

surface a 3D model of object can be reconstructed.

A laser tacker is a precise measurement device that can be used for large objects such as

wind turbine blades with micrometer accuracy. This method tracks a mirrored spherical probe

(MSP) using a laser beam. Figure 2.7 shows how the surfaces of objects are measured by the laser

16

tracker. As illustrated in this figure, a laser beam is emitted from the laser tracker to track the MSP.

The reflected light is received by the laser tracker, which uses an interferometer and angular

encoders to determine the range and angles between laser tracker and the MSP. This information

can be used to measure the co-ordinates of the point on the object’s surface (Ouyang, Liang, and

Zhang 2006).

Figure 2.7. Schematic of laser tracker principle (Ouyang, Liang, and Zhang 2006).

2.1.3 Wind Turbine Blade Condition Monitoring Techniques

Subsurface damage initiation and propagation is another common source of failure in wind

turbine blades in addition to leading-edge erosion, which causes energy loss before structural

degradation. Most blades consist of two halves made of fiberglass composite and shear webs,

which are glued together with strong adhesive materials (Jüngert 2008). The main function of the

shear webs is to increase the strength of the structure. These bonded zones are potential sites for

17

damage initiation and propagation (Meinlschmidt and Aderhold 2006). Most defects are initiated

by manufacturing flaws such as the use of low-quality adhesives in bonded areas, which include

the trailing and leading edges and the connected areas between the shells and shear webs as shown

in Figure 2.8. Defective blades are rarely replaced because wind turbine blades are expensive to

manufacture. To prevent failure, blades need to be continuously monitored through Non-

Destructive Testing (NDT) methods (S. M. Habali and I. A. Saleh 2000). Monitoring the condition

of blades during both manufacturing and operation minimizes the possibility of failure later and

increases the blades’ longevity.

Figure 2.8. A cross section of a wind turbine blade consisting of (1) leading edge, (2) trailing

edge, (3) shear webs glued to (4) suction and pressure sides, (A) composite material, and (B)

wood or plastic foam.

Different NDT techniques have been employed to inspect the integrity of wind turbine

blades. Most of these methods are only usable when the blade is stationary on the ground. The

Acoustic Emission (AE) method has been widely used to inspect internal damages such as cracks,

delamination, and fibre-matrix debonding in wind turbine blades (Ciang, Lee, and Bang 2008). In

AE, stress waves emitted from the defective area are captured by piezoelectric sensors as the

damage propagates inside the blade (S. M. Habali and I. A. Saleh 2000; A. Beattie 1997). The

18

stress waves are generated only if fracture is progressing, and therefore loading is required for AE

to work. In-service blades are subjected to variable loads applied by wind, while the external

loading is required for the blades on the ground. The position of sensors with respect to the

defective area affects the accuracy of this method, so many sensors are necessary to produce

accurate and reliable results (Myrent 2013).

Fibre Bragg Grating (FBG) strain sensors are an effective method of monitoring the

condition of blades. They can be used to detect delamination and debonding of adhesive joints in

blades. Since FBGs are light and small, they can be embedded inside blades without disturbing the

blade’s integrity (Krebbera and Habela 2005).

The next NDT method that can monitor the health condition of wind turbine blades is

vibration analysis. In this method, the modal parameters of blades are evaluated (G. Larsen et al.

2002; White, Adams, and Rumsey 2011). The assumption of this method is that a defect in a blade

will alter its physical properties, leading to detectable variations in the modal parameters of the

blade (Zhang et al. 1999).

There are currently different methods used in the industry to monitor installed blades, with

visual inspection and tap tests among the most common. Visual inspection often involves high

resolution cameras that are used to inspect the blades (Liu, Tang, and Jiang 2010). A tap test is

appropriate for the inspection of internal defects near the surface. The main idea behind a tap test

is that the blade’s thickness and material variation, and the presence of defects result in different

sounds produced by the hammer impulse (Liu, Tang, and Jiang 2010; Juengert and Grosse 2009).

There are three primary types of tap test: manual hammer tapping, “Woodpecker” portable

bondtester, and the Computer Aided Tap Tester system (Liu, Tang, and Jiang 2010). Local

resonance spectroscopy is a tap test variant that uses a microphone to record the sound of a hammer

19

impulse, and subsequently inspect defects near the surface (Juengert and Grosse 2009). The wind

turbine industry also employs Ultrasonic Testing (UT) as an NDT technique to detect different

types of defects such as cracks and delamination. UT is used to inspect wind turbine blades in both

wind farms and research centers (Lading et al. 2002). This method allows defects embedded in

deep layers to be inspected (Juengert and Grosse 2009; Lading et al. 2002; Márquez et al. 2012).

Table 2.1 summarizes the main techniques used for monitoring the condition of blades.

Conventional NDT techniques generally require close proximity between the sensor and the

blade (Borum and McGugan 2006). Since access to a blade is difficult and requires an industrial

climber or crane, Figure 2.9, which can be dangerous and time consuming, the practical

implementation of conventional methods sometimes requires blade removal. Developing new

NDT techniques that are capable of detecting faults in the blades from larger distances is essential.

Infrared (IR) thermography is a non-contact, long-distance NDT technique that can inspect

extensive areas quickly by capturing thermal images of the object’s surface. In general, defective

areas cause different temperature distributions on the surface that are measured by IR cameras.

Figure 2.9. Inspection of wind turbine blade using an industrial climber (Marsh 2011). Copyright

2011, Used with permission from Reinforced Plastics. Elsevier.

20

Table 2.1. Summary of main NDT methods for monitoring the condition of blades

NDT method Methodology Application status

Acoustic Emission (AE)

(Joosse and Blanch 2002;

Rumsey and Paquette 2008)

AE Records elastic waves emitted from

defective areas

Research

Fibre Bragg Grating (FBG)

(Schroeder et al. 2006;

Krebbera and Habela 2005)

FBG is wavelength multiplexed and its

central wavelength alter if any stain

change occurs

Research

Strain gauge (Schubel,

Crossley, and Boateng 2013;

Rumsey and Paquette 2008)

Determine the strain distribution on the

surface of the blade

Research

Vibration and Modal

analysis (White, Adams, and

Rumsey 2011; G. Larsen et

al. 2002)

Measure modal parameters and

dynamic response using accelerometer

Research

Visual inspection (Liu, Tang,

and Jiang 2010)

Inspection using high resolution

cameras

Wind farm

Tap tests (Juengert and

Grosse 2009; Liu, Tang, and

Jiang 2010)

Variation of sound emitted during

knocking test due to thickness/material

change or the presence of defect

Research & Wind

farm

Ultrasonic testing (UT)

(Lading et al. 2002; Márquez

et al. 2012)

A pulse of ultrasound is emitted and

propagate into material and the

presence of defect is evaluated by

analysis the reflected waves

Research & Wind

farm

21

Different studies have used thermography to detect faults in wind turbine blades.

Meinlischmidt and Aderhold (Meinlschmidt and Aderhold 2006) employed passive thermography

to detect internal structural features and subsurface defects such as poor bonding, and

delamination. Beattie and Rumsey (Alan Beattie and Rumsey 1999) employed thermography to

inspect blades during fatigue tests of a 13.1 m blade made from wood-epoxy-composite and a 4.25

m fiberglass blade. This experiment allowed them to identify the root region of the blade as a

defective area. Shi-bin et al. (Zhao, Zhang, and Wu 2009) employed infrared thermal wave testing

to detect subsurface faults such as foreign matter and air inclusions at various depths of a blade

section. In 2011, Galleguillos et al. (Galleguillos et al. 2015) conducted a new experiment to

inspect an installed wind turbine blade. They mounted an IR camera on an unmanned aerial vehicle

(UAV) and captured thermograms of installed blades while the blades were stationary,

demonstrating the capability for fast data acquisition and inspection with this setup. Other research

evaluated the suitability of different weather conditions for revealing the internal features of a

blade section with thermal imaging (Worzewski, Krankenhagen, and Doroshtnasir 2016;

Worzewski and Krankenhagen 2016). Doroshtnasir et al. (Doroshtnasir et al. 2016) proposed a

new passive thermography technique that can inspect operating blades from the ground. This

experiment developed a new image processing technique to improve the thermal contrast quality

by removing the effect of disruptive factors such as environmental reflections.

22

2.2 Thermography

Thermography uses infrared radiation emitted by objects to produce thermograms. All

materials with temperatures above absolute zero emit energy from their surfaces. This energy is

mostly in the infrared range when the temperature of the material is equal to or higher than the

ambient temperature (Bagavathiappan and Saravanan 2008). Infrared thermal devices are sensitive

to wavelengths between 0.75 μm and 25 μm (Bagavathiappan and Saravanan 2008). Infrared

radiation is divided into different categories depending on the wavelength. Table 2.2 lists the

different categories of infrared radiation based on various temperature ranges.

Table 2.2. Sub-division of infrared radiation according to temperature range (Byrnes 2008).

Division name Wavelength Temperature range

Near-infrared 0.75-1.4 μm 1797-3591 oC

Short-wavelength infrared 1.4-3 μm 693-1797 oC

Mid-wavelength infrared 3-8 μm 89-693 oC

Long-wavelength infrared 5-15 μm -80-89 oC

Early development of infrared detectors occurred between 1930-1950 when military

organizations began to use the technology for object detection (Gavrilov 2014). Modern

commercial IR cameras use focal plane array (FPA) technology (Gavrilov 2014). Figure 2.10

shows that modern IR cameras can capture the infrared radiation emitted from an object and

display them as thermograms (Meinlschmidt and Aderhold 2006). All thermograms captured by

IR cameras are grayscale images, where light and dark spots correspond to hot and cold

temperature areas, respectively. Each grayscale pixel can be displayed as a false color

representation to provide better visualization of the temperature distribution on a surface

(Meinlschmidt and Aderhold 2006).

23

Figure 2.10. Schematic of thermal image acquisition and processing using an IR camera

(Meinlschmidt and Aderhold 2006).

Infrared thermography has been employed in different applications to evaluate the health

condition of structures. Examples of the application of thermography as an NDT method can be

found in aerospace, civil structures, and electrical devices. It includes inspection of welding joints,

composite materials, and monitoring of plastic deformation (Bagavathiappan, Lahiri, and

Saravanan 2013). This method has also been used to monitor the condition of wind turbine blades

(Alan Beattie and Rumsey 1999; Meinlschmidt and Aderhold 2006; Zhao, Zhang, and Wu 2009;

Bagavathiappan, Lahiri, and Saravanan 2013; Galleguillos et al. 2015).

Thermographic inspection is typically divided into two categories: active and passive. Active

thermography needs close access to the object, which makes it useful in initial inspection of blades

to evaluate the integrity of their structure prior to installation or during maintenance, but less useful

as a tool for monitoring wind turbine blades in operation. In this study, active thermography is

utilized to allow comparison with passive thermography, which is mostly used for qualitative

applications (Bagavathiappan and Saravanan 2008). Passive thermography has recently been

introduced for monitoring the condition of wind turbine blades and therefore, is not a common

24

approach in this industry. More development is required to adapt passive thermography for this

application (Ciang, Lee, and Bang 2008).

2.2.1 Active Thermography

Active thermography requires the surface of the object to be heated using different methods.

The different heating and cooling rates of defective and non-defective regions generate dark or

light spots associated with damaged areas on the thermograms (Meinlschmidt and Aderhold 2006).

Different methodologies such as mechanical (vibro-thermography) (Holland 2011), electrical

(eddy current thermography) (Biju, Ganesan, and Krishnamurthy 2009), or thermal (step heating,

flash or pulsed, and lock-in thermography) (Lizaranzu, Lario, and Chiminelli 2015) can be used.

Several factors should be considered while selecting an appropriate method, including the object

material, its application, availability, and features to be inspected. Figure 2.11 presents a summary

of the most frequently-used thermography techniques (Matovu 2015). Among these methods,

pulsed, step heating, and lock-in thermography have been used for fault detection of blades

(Amenabar, Mendikute, and López-Arraiza 2011; Tao et al. 2011; Manohar and Tippmann 2012;

Manohar and Lanza di Scalea 2013; Doroshtnasir et al. 2016).

Figure 2.11. Summary of thermography techniques (Matovu 2015).

25

It can be difficult and sometimes dangerous to mount equipment near rotating blades.

However, (Amenabar, Mendikute, and López-Arraiza 2011) proposed a robotic system to move a

thermography system along a blade, providing access to every part of the blade to ensure complete

monitoring. An experimental setup of the proposed method which is a multi-axis inspection system

and active thermography implementation is illustrated in Figure 2.12 (Chatzakos, Avdelidis, and

Hrissagis 2010). The active thermography system is mounted on the end of a movable scanning

arm, which is mounted on an aluminum truss that has the capability of vertical movement to cover

all parts of a blade. This robotic inspection system has been designed and tested on a small-scale

blade section at ZENON S.A. Robotic & Informatics laboratory in Athens, and still needs more

development to be used on installed blades in wind farms (Chatzakos, Avdelidis, and Hrissagis

2010).

Figure 2.12. A robotic in-situ rotor blade inspection schematic (Chatzakos, Avdelidis, and

Hrissagis 2010). Copyright 2011, Used with permission from Robotics Automation and

Mechatronics (RAM), IEEE.

The following sections discuss the principles of the three main techniques of active thermography.

http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=5508191

http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=5508191

26

2.2.1.1 Pulsed Thermography

The straightforward experimental setup and full-field defect detection capability of pulsed

thermography make it a favoured technique (Manohar and Tippmann 2012). Figure 2.13 shows a

pulsed thermography setup schematic where the surface of a specimen is heated by high-power

flash lamps and the thermal variation on the surface is recorded by an IR camera. High-power flash

lamps are used to heat an object suddenly and to generate a temperature gradient in the object.

Weak signals may be generated due to a short heating period, preventing the detection of deeper

defects. In order to preserve the contrast, it is necessary to increase the power of the pulse heating

source (Bainbridge 2010).

Figure 2.13. Schematic of pulse thermography setup (C. Ibarra-Castanedo et al. 2007).

Pulsed thermography operates in two modes: reflective and transmissive. Since the inner

part of the blade is not accessible, only reflective mode is applicable for inspection of wind turbine

blades. The reflective mode captures thermal data from the same side as the lamps (Pawar 2012).

27

2.2.1.2 Step Heating Thermography

Step heating thermography, also known as “long-pulse thermography”, is an attractive NDT

method. In this method, the surface of the object is heated using long-pulse heating sources such

as halogen lamps. The transient temperature variation is also recorded as the object’s temperature

either increases or decreases, whereas pulsed thermography just records data during cooling

(Balageas et al. 2000; Maldague 2002).

Deeper defects can be monitored using this method because it heats the material for a longer

time (Badghaish and Fleming 2008). A single step heating experiment can cover a large area of

the specimen. The experimental setup and data acquisition procedure of this method is the same

as pulsed thermography except that flash lamps are replaced with halogen lamps. Low-excitation

power application on the object is the primary advantage of step heating thermography. Another

advantage is the cost efficiency compared to pulsed or lock-in thermography since the excitation

is by halogen lamps, which cost less than the excitation units used for the other methods (Matovu

2015).

2.2.1.3 Lock-In Thermography

Contrary to pulsed thermography, lock-in thermography can detect deeper defects by

continuously heating the surface using a periodic heat source (Chatterjee et al. 2011) such as a

modulated halogen lamp (Montanini 2010). A typical schematic of lock-in thermography is

depicted in Figure 2.14. Thermal waves pass through the object and reflect when they reach a

defect. The IR camera records the surface thermal pattern that results from the interference of

absorbed and reflected thermal waves (Pawar 2012). As illustrated in Figure 2.15, the principle

28

behind the lock-in thermography for subsurface fault detection is that internal damage alters the

phase difference between reflected and modulated thermal waves (Manohar and Tippmann 2012;

Pawar 2012). After recording the thermal data at a certain excitation frequency, Fourier transforms

are employed to calculate the phase and amplitude images of recorded thermograms (Montanini

and Freni 2012). A lock-in thermography experiment for a wind turbine blade is shown in Figure

2.16, where a function generator is employed to modulate the output of the halogen lamp. The

thermal data is first captured by an IR camera, and then stored on a PC for further processing

(Manohar 2012).

Figure 2.14. Schematic of experimental setup of lock-in thermography (Pawar 2012).

The major advantage of lock-in thermography is that the phase images obtained by this

method are less sensitive to non-uniform heating. On the other hand, it may be time consuming to

use this method to detect all internal defects since it requires the use of various excitation

frequencies to inspect the material at different depths (Montanini and Freni 2012).

29

Figure 2.15. Thermal wave propagation and thermal wave phase difference and amplitude in

lock-in thermography (Pawar 2012).

Figure 2.16. Experimental lock-in thermography for a wind turbine blade (Manohar and Lanza di

Scalea 2013). Copyright 2013, Used with permission from Structural Health Monitoring. SAGE

Publications.

30

2.2.2 Passive Thermography

The simple data acquisition and analysis of passive thermography make this method an

appropriate technique for condition monitoring (Bagavathiappan, Lahiri, and Saravanan 2013).

When passive thermography is used for monitoring the condition of wind turbine blades, different

excitation sources can be considered (Worzewski, Krankenhagen, and Doroshtnasir 2016;

Worzewski and Krankenhagen 2016; Meinlschmidt and Aderhold 2006). High temperature

differentials between day and night can reveal internal defects of the blade. Solar heating can also

generate a temperature difference between the front of the blade and the back, which can be

exploited for passive thermography (Meinlschmidt and Aderhold 2006).

31

THEORY OF THERMOGRAPHY AND IMAGE PROCESSING

Different factors such as environmental reflection, non-uniform heating, surface condition,

and geometry are among the main sources that may deteriorate the quality of thermograms

(Kretzmann 2016). Thermal signatures associated with subsurface defects are not therefore

sufficiently distinguishable from the background. Different image processing algorithms must be

used to increase the quality of thermal contrast. Various image processing algorithms have been

developed to improve the quality of thermal images. Thermal Signal Reconstruction (TSR)

(Shepard, Ahmed, and Rubadeux 2001), Matched Filters (MF) (C. Larsen 2011), Principal

Component Thermography (PCT) (Castanedo 2005), and Pulsed Phase Thermography (PPT)

(Maldague and Marinetti 1996) are among the most efficient techniques.

In this chapter, the theory and formulations associated with pulsed and step heating

thermography are reviewed. The theoretical backgrounds of each of the most common image

processing techniques are then discussed.

3.1 Pulsed Thermography Theory

Pulsed thermography is the simplest version of active thermography, where flash lamps are

employed to heat the surface by creating a short square pulse of thermal energy (Q). By assuming

a material has homogeneous properties and a simple geometry, and by using the Fourier heat

diffusion equation, the propagation of thermal waves inside the specimen that is subject to thermal

pulse can be estimated as shown in (Ravichandran 2015; Pawar 2012; Matovu 2015):

2 10

TT

t

(3.1)

32

where T is temperature on the surface of the object, is thermal diffusivity, t is time, and 2 is

the Laplacian operator. By assuming a semi-infinite body with infinite depth in the z-direction, the

one-dimensional solution of Equation (3.1) is (Maldague and Marinetti 1996; Balageas et al. 2000;

Carslaw and Jaeger 1959):

)4

exp(),(2

0t

z

te

QTtzT

(3.2)

where T0 is the initial temperature, Q is input thermal energy, z is depth, and e is the effusivity of

material, defined as:

pCke (3.3)

where k , , and pC are thermal conductivity, mass density, and specific heat capacity of the

specimen, respectively.

As Figure 3.1 (a) shows, thermograms are recorded over time (t) as a 3D matrix where x and

y are the co-ordinates of each pixel. The temperature variation profile of a sound pixel, illustrated

in Figure 3.1 (b), can be approximated using Equation (3.2); however, it may not follow the

theoretical distribution exactly (Castanedo 2005).

(a)

(b)

Figure 3.1. (a) Thermograms recorded as a 3D matrix in time domain, and (b) temperature

variation for a pixel in a sound area (Clemente Ibarra-Castanedo and Maldague 2004). Copyright

2004, Used with permission from Journal of Research in Non-destructive Evaluation. Taylor &

Francis.

33

When heating starts, thermal waves begin propagating through the thickness of a material.

During this process, the thermal waves can be distorted when they reach defects with different

thermal properties, so the temperature distribution associated with defects will be different than

the non-defective area (Gavrilov 2014). Equation (3.2) shows that the temperature variation of a

sound area is a function of time, and similar to the variation in a defective area immediately after

excitation. Over longer times when heat reaches the defect, the temperature profile may increase

or decrease as illustrated in Figure 3.2 (Manohar 2012), (Castanedo 2005). Absolute Thermal

Contrast (ATC) is the temperature difference between sound and defective areas at the same time,

and can be calculated as:

d sT T T (3.4)

where Td and Ts are temperature at the defective and non-defective regions, respectively (Manohar

2012). Figure 3.2 shows a typical example of temperature variation in defective and non-defective

areas, as well as the ATC. This method cannot eliminate noise, which may affect the quality of

results (Manohar 2012).

(a)

(b)

Figure 3.2. (a) A typical temperature variation of defective and non-defective areas, and (b) ATC

(Manohar 2012).

34

3.2 Step Heating Thermography Theory

By considering a material with a simple geometry, homogeneous properties, an adiabatic

boundary condition at the back surface, and a uniform heating at the front surface, the temperature

distribution can be estimated as (Ghadermazi and Khozeimeh 2015; Badghaish 2008; Matovu

2015):

0

2 (1 2 ) (1 2 )( , ) [ierfc( ) ierfc( )]

2 2n

tQ L n x L n xT x t

k t t

(3.5)

where L is the thickness of sample. In this equation, ierfc(x) is the first integral of the

complementary error function, and defined as (Badghaish 2008):

1ierfc( ) .erfc( )xx e x x

(3.6)

By considering that x=L and substituting in Equation 3.5, the temperature at the surface of sample

is (Badghaish 2008):

1

2( , ) [1 2ierfc( )]

n

tQ nLT L t

k t

(3.7)

Finding the time at which deviation in T occurs between defective and non-defective areas is an

appropriate criterion for fault detection using step heating thermography.

3.3 Image Processing Algorithms in Thermography

The noise in raw thermograms can make it difficult to detect subsurface defects. The

application of image processing on raw thermal images to reduce noise and improve the quality of

thermal contrast is therefore essential. Without the use of appropriate image processing techniques,

only near-surface defects whose thermal properties differ significantly from the background can

35

be detected. Several post-processing methods have been developed. This section summarizes the

fundamentals of the most effective and popular thermal image processing algorithms, including

Thermal Signal Reconstruction (TSR) (Shepard, Ahmed, and Rubadeux 2001), Matched Filters

(MF) (C. Larsen 2011), Principal Component Thermography (PCT) (Castanedo 2005), and Pulsed

Phase Thermography (PPT) (Maldague and Marinetti 1996).

3.3.1 Thermal Signal Reconstruction (TSR)

Thermal Signal Reconstruction (TSR), an effective thermal image processing method, was

proposed by Shepard (Shepard 2003). This method increases the quality of the thermal signatures

associated with internal defects by improving the signal-to-noise ratio (SNR), while at the same

time reducing image blurring and increasing the sensitivity (C. Larsen 2011). Each pixel of the

image sequence is compared in a logarithmic domain instead of analyzing individual frames,

resulting in more visible deviation between defective and sound areas (C. Larsen 2011). The

temperature response is transferred to a logarithmic domain, which is then fitted by a polynomial

curve using the least square fitting approach (Shepard 2003; C. Larsen 2011). The fit is:

N

on

n

n

n

nsurf tatatataatT ))(ln())(ln(...))(ln()ln())(ln( 2

210 (3.8)

where n is the order of polynomial. It was identified that the choice n = 5 or 6 gives good resolution

and acts as a low-pass filter to reduce the noise level (Shepard 2003; C. Larsen 2011). After

reconstructing the temperature distribution in the logarithmic domain, the temperature response at

the surface can be determined as (Matovu 2015; C. Larsen 2011):

nN

n

nsurf taT ))(ln(exp0

(3.9)

36

This is the expected thermal response behaviour. Subsurface defects can be detected

whenever the response deviates from the fitted behaviour. The cooling rate is proportional to the

first derivative of Equation (3.9) (Matovu 2015).

3.3.2 Matched Filters (MF)

Matched Filters (MF) have been proposed to improve image contrast of subsurface defects

by increasing the contrast of defective areas and reducing the signals from sound areas (Foy 2009;

Kretzmann 2016; C. Larsen 2011). Different types of MF algorithms have been developed, all of

which are based on the assumption that (Kretzmann 2016; C. Larsen 2011):

idealreflobs TTT (3.10)

at any time. Tobs is the temperature recorded by the IR camera, Tref is the temperature response

reflected from defective area, and Tideal is the ideal temperature variation response of the sound

area. Equation 3.10 can be represented in vector form:

X=S+W (3.11)

Where these vectors collect all recordings over time and X obsT , S reflT , W idealT . Equation

(3.11) is multiplied by a vector q that maximizes the visibility of the reflected temperature from

the defective area and minimizes the response of non-defective areas. The vector q can therefore

be calculated by (Kretzmann 2016; C. Larsen 2011):

22minmax qWtosubjectSq

q

T

q (3.12)

where qT is q transposed and methods of finding this q vector result in the generation of different

types of MF algorithms.

37

Both reflection and ideal temperatures should be known in each MF algorithm. The ideal

temperature can be calculated by manually selecting several pixels in the defect-free area.

Determining the temperature response due to reflection from defective areas is challenging, but

can be accomplished using Equation (3.10) (Kretzmann 2016; C. Larsen 2011). The reflectance

temperature with maximum SNR can be determined if several target pixels are selected and the

average of results are considered (Kretzmann 2016; C. Larsen 2011).

The simplest type of MF algorithm that considers Trefl as the q vector is Simple Matched

Filters (SMF). The selection of q leads to a correlation image, which can be described based on

(Kretzmann 2016; C. Larsen 2011):

ij

T XSSMF (3.14)

where i and j imply that the calculation is repeated for all pixels of all thermal images to provide a

single correlation image, and ST is S transposed.

Spectral Angle Map (SAM) is a type of MF that is based on the SMF type. The SAM is

obtained as the vector magnitudes of reflected and observed temperature responses are normalized.

The resulting image can be described as (Kretzmann 2016; C. Larsen 2011):

ij

T

ij

T

ij

T

XXSS

XSSAM (3.15)

The SAM algorithm also correlates the reflected and observed temperatures so that the final image

is the cosine of the angle between the vectors of X and S.

The Clutter Matched Filter (CMF) algorithm uses the covariance matrix C to combine the

structural information obtained from the object (Kretzmann 2016; C. Larsen 2011). The resulting

image in this method is defined by:

ij

T XCSCMF 1 (3.16)

38

where C is covariance matrix of the ideal temperature vector defined as:

m

i

TWWm

C1

1 (3.17)

The Adaptive Coherence Estimator (ACE) method is based on CMF and normalized by the

magnitude vector in the same way that SAM was determined through SMF (Kretzmann 2016; C.

Larsen 2011):

ij

T

ij

T

ij

T

XCXSCS

XCSACE

11

1

(3.18)

The t-statistic and F statistic methods are widely used in regression (Kretzmann 2016; Foy

2009; C. Larsen 2011). The F statistic can be obtained using:

)1()(

)(2

2121

21

dXRSXRX

xRSFstat

ij

T

ij

T

ij

ij

T

(3.19)

where )1

(1SRS T

and the t-statistic is the square root of the F statistic, defined as:

)1()( 2121

1

dXRSXRX

xRStstat

ij

T

ij

T

ij

ij

T

(3.20)

3.3.3 Principal Component Thermography (PCT)

Principal Component Thermography (PCT) is a statistical processing technique developed

by Rajic (Rajic 2002) that uses the Principal Component Analysis (PCA) developed by Pearson

(Pearson 1901). PCT employs Singular Value Decomposition (SVD) to model the data by

converting the 3D thermal image sequences to 2D data through a raster-like operation. The initial

3D data are rearranged into a data-cube A with the dimension M×Nt where M=Nx×Ny, Nx, Ny are

39

pixel dimensions of each thermogram and Nt is the number of frames (Gavrilov 2014; C. Larsen

2011). The columns of matrix A are standardized as:

n

nmnAmnA

),(),( (3.21)

where:

tN

nt

n mnAN 1

),(1

(3.22)

2

1

2 )),((1

1n

N

nt

m

t

mnAN

(3.23)

By employing SVD, the matrix A can be decomposed into different components (Kretzmann

2016; Manohar 2012; C. Larsen 2011):

TVUA (3.24)

where is a diagonal matrix with M×M dimensions of eigenvalues (singular values) of matrix A,

and where U and VT consist of left and right singular vectors of matrix A. VT contains temporal

variation components that can provide information regarding the depth of defects (Gavrilov 2014;

C. Larsen 2011). Matrix U consists of a set of orthogonal column vectors that describe all thermal

images through certain combination of matrix columns.

Orthogonal functions are created by applying the reverse raster transformation that was

used to form matrix A on matrix U. The first few orthogonal functions contain the most spatial

variations (Gavrilov 2014; Matovu 2015; Manohar 2012; C. Larsen 2011). In general, first five

modes possess more than 95% of the total variance (Matovu 2015; Marinetti et al. 2004).

One negative point about this method is that it may enhance the contrast of some defects

while degrading the visibility of others (C. Larsen 2011). Another drawback of PCT is that there

40

are no general ways to interpret the results. One certain defect may generate cold or hot spots

depending on experimental conditions or other factors (Gavrilov 2014).

3.3.4 Pulsed Phase Thermography (PPT)

One of the most popular thermographic image processing techniques is Pulsed Phase

Thermography (PPT), which was developed by Maldague and Marinetti (Maldague and Marinetti

1996). This method can eliminate the sensitivity of pulsed thermography to non-uniform

stimulation. It can also be employed for quantitative evaluation of defects (Castanedo 2005).

PPT is a combination of lock-in and pulsed thermography, so it can deliver the advantages

of both of these methods, (Shin 2013; Castanedo 2005; Maldague 2002; Maldague and Marinetti

1996). Lock-in thermography can detect deeper defects, but requires more time to detect defects

located at different depths. It is also less sensitive to non-uniform heating and environmental

reflections. Data acquisition and processing of pulsed thermography is quick, but the results are

influenced by non-uniform heating and reflections from the surroundings (Castanedo 2005; Pawar

2012).

PPT uses a transform algorithm such as the Fast Fourier Transform (FFT) to convert time

domain data to frequency components. The application of the Fourier transform on a square pulse

and thermal pulse response is illustrated in Figure 3.3 (Castanedo 2005). The experimental setup

and data acquisition process in this method is similar to pulsed thermography. After recording the

data, the Fourier transform is applied to the thermograms to obtain the phase and amplitude data

(Shin 2013; Pawar 2012; Castanedo 2005).

Discrete Fourier Transform (DFT) of recorded thermal data at each pixel of thermogram can

be represented as (Pawar 2012; Castanedo 2005):

41

nn

N

k

nN

nkjtkTtF ImRe)

2exp()(

1

0

(3.25)

where T is temperature of each pixel, t is the time interval between image sequences, n is the

frequency increment (n=0,1,2,…,N), k is the image index, and Re and Im are the real and imaginary

part of the transformation, respectively. The real and imaginary parts can be used to calculate the

phase and amplitude of the captured signal at each pixel (Pawar 2012; Castanedo 2005):

2 2( ) Re ( ) Im ( )A n n n (3.27)

Im( )( ) arctan

Re( )

nn

n

(3.28)

Figure 3.3. Decomposition of square pulse and thermal decay pulse (Castanedo 2005).

The phase images are useful for NDT applications because they are less affected by non-uniform

heating, environmental reflections, and surface condition and geometry (Maldague and Marinetti

1996). The frequency resolution ( f ) that is achieved by N frames can be controlled via the

42

sampling interval as f =1/(Nt). It should be noted that lower frequencies can capture the defects

embedded in deeper layers (Castanedo 2005; Pawar 2012; Shin 2013).

A new method, Step-Heating Phase and Amplitude Thermography (SHPAT), was developed

in this study to improve the quality of the raw thermograms. In this method, step heating

thermograms recorded during either heating or cooling are transformed from a time domain to a

frequency domain. Figure 3.4 shows the application of the Fourier transform on step heating data.

The processing algorithm in this method is similar to that used in PPT where an FFT transform is

applied to active thermal data. The main difference is that the SHPAT applies FFT to step heating

thermal data while PPT uses pulsed thermal data.

Figure 3.4. Transformation of step heating thermal response during heating and cooling.

43

Since amplitude data are sensitive to non-uniform heating, it is not possible to obtain enough

useful information when amplitude images of the pulsed thermography experiment are evaluated.

Step heating thermography uses long-pulse to heat the surface almost uniformly, so amplitude

images obtained from step heating thermal data can provide useful information about subsurface

defects. Both phase and amplitude images obtained by Fourier transform can therefore increase

the thermal contrast associated with internal defects. The frequency resolution, which is a function

of sampling frequency and acquisition time, can significantly affect the quality of results. The time

of heating is another important parameter that can greatly affect the quality of the phase and

amplitude images captured by the application of FFT on step heating data.

3.3.5 Quantitative Evaluation

The signal to noise ratio (SNR) of images is useful tool to quantitatively evaluate the quality

of data. The traditional definition of SNR is the ratio of average power of signal values over the

power of background or noise. Based on this definition, the SNR can be written as (Xia 1998):

21

0

2

( )N

n

x n

SNRN

(3.29)

where x(n) is the power of signal and 2 is the variance of background noise. Since most signals

have wide dynamic ranges, they are represented by a logarithmic decibel scale. By considering the

definition of a decibel (dB), the SNR can be defined as (Jain and Jain 1981):

2

210log( )SNR

(3.30)

where is the peak value of signal. Equation (3.30) can be simplified as:

44

20log( )SNR

(3.31)

where is the standard deviation of background noise.

This chapter discussed the theoretical basis of the most common and effective thermal image

processing methods. The typical procedure of thermographic data acquisition and processing is to

first record raw thermal images by using both pulse and step heating, and to then apply different

image processing algorithms on the raw thermographic data. The results of application of these

methods will be discussed and compared in the next chapters.

45

EXPERIMENTAL PROCEDURES AND SETUPS

This chapter details the experimental procedures and setups. The experiments consist of

reconstructions of both clean and rough wind turbine blade sections using laser scanning, and

condition monitoring of the damaged blade section using both passive and active thermography.

The material properties and geometry of the blade used for the experiments will also be discussed.

A summary of instrument specifications employed for different experiments is given in appendix

A.

4.1 Material

All samples used in this experiment came from a 50 m long wind turbine blade made of

fibreglass composite obtained after it had been damaged in transit to a wind farm. The blade was

never installed or operated. Figure 4.1 shows the 3 m long blade section with significant surface

damage that was used for the passive thermography experiments. The chord length and thickness

were approximately 1 m and 19 cm, respectively. A separate section was used for the laser

scanning experiments.

The sandwich part of the blade near the trailing edge and a narrow band near the leading

edge are made of thermoplastic foam. This foam possesses thermal diffusion properties that are

different from the fibreglass composite forming the remainder of the skin. These properties will

lead to the generation of thermograms with different intensities in the affected regions. Since

fibreglass composites are composed of different layers, different types of defects may be generated

between them (Marinetti, Muscio, and Bison 2000).

46

The yellow/orange regions in Figure 4.1 are the exposed sandwich core on the rear section

of the suction surface where there is very little laminate. Some patches of glue on the suction side

results in different effusivity than the background, and therefore generates spots with different

brightness on the thermograms. Coloured fibres are embedded in the blade at three different spots,

which will give rise to different effusivity and energy absorption at these specific spots. The glued

sections and coloured fibres are marked by red boxes in Figure 4.1.

Figure 4.1. The damaged blade section used in passive thermography experiment.

47

A “defect plate” with dimensions of 170 mm 195 mm 8 mm was cut from the laminate

skin of another section of the blade. Flat-bottomed holes with different diameters and depths were

drilled in the defect plate from the rear to produce a range of known “defects”. The holes had

diameters ranging from 4 mm to 20 mm with the depths between 0.5 mm and 3 mm. Figure 4.2

(b) is a schematic of the plate that illustrates the geometry and pattern of the defects.

The defect plate was attached to the surface of the damaged blade section during passive

thermography experimentation. It was also the only blade material tested with active

thermography. The holes were used to evaluate the minimum defect size-to-depth ratio that can be

inspected using passive and active thermography.

4.2 Experimental Setups

The experimental procedures included both laser scanning and thermography experiments.

In the thermography section, both active and passive thermography are discussed. Two separate

experiments used active thermography, one with pulsed heating and the other with step heating.

4.2.1 3D Laser Scanning

A Leica HDS 6100, a Surphaser 100 hsx, and a Creaform HandyScan were the laser scanners

employed for this research. The Leica and Surphaser scanners do not provide the submillimeter

accuracy required to capture roughness and shape variation due to erosion. The complex

experimental setup required of both of those scanners also detracted from their utility for field

measurements. The setup for laser scanning with the Leica HDS is shown in Figure 4.3.

48

(a) (b)

Figure 4.2. Geometry and pattern of blind holes in the defect plate. (a) shows the rear of the plate

(b) gives the size and depth of the blind holes. The blind holes in each row, labeled A-D have

the same diameter but varying depth.

To obtain a complete point cloud for the entire surface of a blade, the scans must be repeated

from different angles with respect to the blade. A registration algorithm is then used to transfer

data from local co-ordinates to a unique global co-ordinate system. Figure 4.3 shows that the

experimental setup used different targets around the blade. Local co-ordinates of targets were

employed for registration purposes. There are many different types of targets available, including

paper, paddle, and sphere (Becerik-Gerber, Jazizadeh, and Kavulya 2011). Paddle targets were

used in this research. Targets should be placed at different angles and heights instead of in a straight

line to avoid registration problems and to provide enough information for registration. One source

49

of error in registration is the time-consuming manual selection and measurement of targets in a

point cloud, known as “labeling”.

The HandyScan laser scanner uses a simple experimental setup, data acquisition and

processing. With an accuracy of 0.03 mm and a resolution of 0.05 mm, the HandyScan laser

achieves the desired minimum accuracy. The 3D reconstruction experiment using the HandyScan

is illustrated in Figure 4.4. The device requires that optical reflectors are attached to the object’s

surface, which is one of the drawbacks of this system. A section of blade with a thickness of 19

cm, a chord length of 1050 cm, and a span of 50 cm was used for reconstruction through

HandyScan laser scanner in this study.

This device’s software has auto-triangulation that uses the optical reflectors to connect the

information from each scan to the others to generate a unique and comprehensive point cloud of

the entire object. The distances between the optical reflectors are important to achieve a continuous

point cloud of the object. At least four or five reflectors should be visible to the scanner in each

scan.

4.2.2 Passive Thermography

In the first thermography experiment, the suction side of the blade was monitored outdoors

during a sunny day from morning until afternoon. The blade’s position was not changed relative

to the IR camera during this time. This experiment, whose setup is depicted in Figure 4.5, sought

to determine the most favorable conditions to reveal the most defects and to evaluate the fault

detection capability of thermography when the blade is heated by the solar radiation.

50

Figure 4.3. Experimental setup of laser scanning by Leica HDS 6100.

(a)

(b)

Figure 4.4. HandyScan laser scanning experiment.

Wind turbine

blade section

Target

Leica laser

scanner

Laptop for

recording data

datadatadata

51

The experiment was conducted on a sunny day in July 2017 from 9:00 am to 7:30 pm.

Sunrise and sunset on this day were 5:53 am and 9:30 pm, respectively. During the experiment,

the sky was clear, the humidity was 36%, and the temperature varied between 16 °C and 26 °C.

The T1030Sc IR camera made by FLIR Systems was located 4 m from the blade section and

equipped with a 21.2 mm lens, resulting in a special resolution of 4 mm per pixel.

ResearchIR, a software package developed by FLIR Systems that provides high speed data

recording and image analyzing capabilities, was used to record thermograms at a frequency of 1

Hz. ResearchIR can manipulate recording options, including the start and end of the recording, the

frequency of the image recording, the temperature range, and the appearance of the thermal image.

This software was installed on a PC and then connected to the IR camera using a USB cable to

capture thermal images or movies from the heated objects.

The second experiment was designed to evaluate the effects of heating and cooling on the

detection of subsurface defects during a short period of time. Similar to the first experiment, the

blade section’s surface was heated by the sun. The suction side of the blade was initially exposed

to the sunlight for around 30 minutes. The surface temperature variation was recorded by the IR

camera and ResearchIR during the heating period.

After 30 minutes, the blade section that was mounted on a pallet was rotated approximately

180° using a pallet jack to shade the surface and record thermograms while the blade cooled for

about 30 minutes. The general setup information, including the distance between the camera and

the blade section, the recording frequency, and the environmental conditions, was identical to the

first experiment. The average temperature during the experiment was 25 °C with a humidity of

36%.

52

Figure 4.5. Passive thermography experimental set-up.

4.2.3 Active Thermography

In this section, the details of two different active thermography methods are discussed. The

first method is pulsed thermography, and the second is step heating thermography. These

techniques were used only on the defect plate, which was mounted approximately 1.5 m from the

IR camera. The IR camera was equipped with an 83.4mm lens to provide thermograms with a

resolution of mmmm 2.02.0 . The process of heating and data collection are different for each of

these techniques, as discussed below.

4.2.3.1 Pulsed Thermography

Conventional pulsed thermography methods employ high power xenon lamps for heating in

millisecond pulses (Maldague 2002; Pawar 2012). In this study, the defect plate was heated by a

2400 W flash lamp, and thermal images were recorded at a frequency of 15 Hz immediately after

flashing the sample. To provide the input energy for triggering the flash lamp, a 4800 W power

53

pack was utilized. To heat the surface uniformly, the flash lamp was around 0.3 m from the object

with the angle of around 150 with respect to the normal of the defect plate. The pulsed

thermography experiment is depicted in the Figure 4.6.

The ResearchIR software was used to record thermal images at a frequency of 15 Hz. The

image acquisition process began as the plate cooled after the surface was flashed. Specially-written

MATLAB codes, including MFs, PCT, TSR and transformed based algorithms processed the raw

captured thermograms to increase the visibility of thermograms using different image processing

algorithms.

Figure 4.6. Pulse thermography experimental set-up.

4.2.3.2 Step Heating Thermography

The step heating thermography setup, shown in Figure 4.7, used two 500W halogen lamps

to continuously heat the surface. The sample was heated between 10 and 75 s, and the thermal

evolution on the surface of the specimen was recorded at a frequency of 15 Hz. Once heating was

54

finished, thermal decay was recorded with the same frequency. The room temperature was

23 2 0C during the experiment. The thermal contrast associated with the defects could be

observed after few seconds of heating.

Figure 4.7. Step heating thermography experimental set-up.

55

RESULTS AND DISCUSSION

The experimental and numerical results, including the results of aerodynamic characteristics

of airfoil, passive thermography, and active thermography, are discussed in this chapter. ANSYS

Fluent was employed to determine the aerodynamic characteristics of both clean and rough airfoils,

which have been extracted using laser scanning. Different image processing algorithms were used

to improve the quality of active thermography results. The most effective thermal image

processing technique was determined and applied to the raw thermograms captured by passive

thermography to improve the quality of results.

5.1 Aerodynamic Characteristics

5.1.1 Reconstruction of the Blade Section Surface

Sandblasting was used to simulate leading-edge erosion by generating roughness on the

leading-edge area of the blade’s surface. The HandyScan laser scanner was used to reconstruct

both clean and rough surfaces. Figure 5.1 (a) shows the 3D model of the blade section that was

obtained. Real and reconstructed surfaces, illustrated in Figures 5.1 (b) and (c), demonstrate this

method’s ability to capture small surface defects. The extracted airfoil profiles of the clean and

rough blade sections are depicted in Figure 5.2.

56

(a)

(b)

(c)

Figure 5.1. (a) Reconstructed blade section, (b) an image of the blade section with small surface

defects, and (c) reconstruction of blade section presented in (b).

57

(a)

(b)

Figure 5.2. The leading-edge region of (a) clean and (b) rough airfoils obtained by HandyScan

laser scanner.

5.1.2 CFD Simulation Using ANSYS Fluent 16.2

ANSYS Fluent 16.2 was used to determine the aerodynamic characteristics of both clean

and rough airfoils. A 2D double precision pressure based and steady state solver was used in this

simulation. Second order formulation and second order upwind scheme were selected for pressure

and discretization of convection terms, respectively.

A calculation was conducted for angle of attack, , ranging from zero to a value that provides

the maximum wind turbine blade efficiency by maximizing the lift:drag ratio. Three different

58

models, including k-kl-, Transition SST, and k- SST, were employed to calculate aerodynamic

properties of airfoils. Since experimental data are available in high Reynolds numbers of 1 million

and 3 million, these values were considered for numerical modeling in this study. The obtained

results were compared with available experimental data (Sareen, Sapre, and Selig 2014), (Timmer

and Van Rooij 2003).

H grid topology was used in this study. The domain size was 10 chords upstream and around

the airfoil to prevent the influence of any boundary conditions on the developing flow, and 20

chords downstream to allow the full development of the wake. Multiblock structured mesh with

quadrilateral elements was used to generate fine grids around the airfoil. Y+ values were

considered below one, leading to the minimum distance of 6105.8 between the first cell and the

wall to appropriately resolve turbulence models. The aerodynamic characteristics of airfoil and

blade sections with four different numbers of cells ranging from 24,000 to 120,000 were calculated

to assess grid independence. The relative lift coefficient difference between 78,000 and 120,000

cells was less than 10-3. A reference value of 99,000 cells was therefore selected. This value meets

the requisite accuracy while simultaneously decreasing the time and computational costs. The

multiblock structure and final resulting mesh are shown in Figures 5.3 (a) and (b), respectively.

5.1.3 Models validation and CFD simulation results

The clean airfoil that was reconstructed by laser scanning was compared with different

available airfoils. It was concluded that it is a modified version of the DU 96-W-180 airfoil

developed at Delft University of Technology (Timmer and Van Rooij 2003). Therefore, we used

DU 96-W-180 as a reference and its published experimental results to conduct model validation.

Figure 5.4 shows the comparison of the clean airfoil and the DU 96-W-180 airfoil (reference

59

airfoil). As illustrated, the main difference between reference and clean airfoils happen around the

trailing-edge. The aerodynamic characteristics including lift coefficient (Cl), drag coefficient (Cd),

and Cl/Cd for different were therefore calculated for both the DU 96-W-180 airfoil and the clean

airfoil. The results of the rough airfoil were also computed and compared against those of the clean

surface.

(a)

(b)

Figure 5.3. (a) Multiblock structure and (b) generated mesh around the airfoil for CFD

simulation.

60

Figure 5.4. Comparison of clean and DU 96-W-180 airfoils.

The experimental values of Cl /Cd for the reference airfoil up to an that provides the

maximum value is depicted in Figure 5.5 (a) (Sareen, Sapre, and Selig 2014), (Timmer and Van

Rooij 2003). The that yields the peak value of Cl/Cd decreases as Re increases. Experimental

and numerical Cl for the DU 96-W-180 airfoil at Re=3,000,000 are illustrated in Figure 5.6. It can

be observed that the numerical results using three different turbulence models are well matched

with the experimental data. The results of Cl/Cd were also well matched with the experimental

data.

After validating three different models, they were used to determine the aerodynamic

characteristic of the clean and rough airfoils. The Cl /Cd for the clean and rough airfoils are shown

in Figure 5.5 (b). Figure 5.7 depicts the numerical results of Cl for clean and reference airfoils at

Re=3,000,000. Modification of the airfoil shape results in a jump in Cl while increasing Cd.

Figure 5.8 shows Cl and Cd of the clean and rough surfaces at various values at

Re=3,000,000 to evaluate the effect of leading-edge roughness on the aerodynamic behavior. The

rough airfoil shows a lower Cl and an increase in Cd for all . Figure 5.5 (b) depicts Cl/Cd versus

61

to demonstrate that at which Cl/Cd peaks is almost the same for both rough and clean airfoils.

However, the rough airfoil presents a lower peak value than the clean surface.

(a)

(b)

Figure 5.5. (a) Experimental values of Cl/Cd at different Re values for reference airfoil (Sareen,

Sapre, and Selig 2014), (Timmer and Van Rooij 2003) and (b) numerical value of Cl/Cd at

Re=3,000,000 for clean and rough airfoils.

62

Figure 5.6. Numerical (lines) and experimental (blue dots) Cl versus of the reference airfoil at

Re=3,000,000 (Timmer and Van Rooij 2003).

Figure 5.7. Numerical Cl at Re=3,000,000 for clean (solid lines) and reference airfoils (dotted

lines) at varying .

63

(a)

(b)

Figure 5.8. (a) Cl and (b) Cd at Re=3,000,000 of clean and rough airfoils at varying .

64

Table 5.1 shows that the airfoil shape modification has more influence on Cl than on Cd. As

Cl/Cd increases, the aerodynamic performance of the airfoil improves.

Table 5.1. Aerodynamic characteristic comparison of both reference and clean airfoils

AOA %Cl

increase

%Cd

increase

Cl/Cd reference

airfoil

Cl/Cd clean airfoil %Cl/Cd

increase

0.42 89 6 57 102 77

1.44 65 10 78 116 49

2.46 51 12 95 128 34

3.5 41 14 110 136 24

4.55 35 16 123 142 16

5.57 31 12 125 145 16

6.59 27 14 129 143 11

It was demonstrated that changes to the airfoil shape resulting from erosion degrade

aerodynamic properties. Another common challenge is the detection of internal defects in blades.

The following section provides a solution for this problem.

5.2 Active thermography

5.2.1 Raw Pulsed and Step Heating Thermography Data

The pulsed thermal images and temperature decay curve for the defect plate are shown in Figure

5.9. Defective regions had higher temperatures that formed brighter spots in thermograms. The

sizes of these spots can provide useful information regarding the size and depth of subsurface

defects. Maximum thermal contrast occurred approximately 1-3 s after cooling began for near-

surface defects. Deeper defects took longer to achieve maximum thermal contrast. Figure 5.9 (b)

65

shows the results recorded 5 s after cooling started. It can be concluded that deeper defects are

detected after longer periods of cooling. The signature of lines in Figures 5.9 (a) and (b) are

because of the fiberglass within the composite material.

(a)

(b)

(c)

Figure 5.9. (a) A thermogram immediately after pulse, and (b) a thermogram after 5s of cooling,

(c) temperature decay of sample in marked defect (“D” in part (b)) and sound (“S”) positions.

66

Typical step heating thermograms obtained during heating and cooling are shown in Figures

5.10 (a) and (b), respectively. The thermal evolution and decay curves are shown in Figures 5.10

(c) and (d).

(a)

(b)

(c)

(d)

Figure 5.10. A thermogram after (a) 40s heating and (b) 15s cooling. Time history of temperature

at defect and sound points during (c) heating and (d) cooling.

67

Figure 5.10 (a) was captured after 40 s of heating. Defects near the surface are readily

apparent, but deeper defects can only be observed after several seconds of cooling as depicted in

Figure 5.10 (b), which shows a typical thermogram recorded 15 s after cooling of the defect plate

that has been heated for 40 s. The bright signatures in middle sections of Figures 5.10 (a) and (b)

are caused by environmental reflection.

All step heating thermograms based on different heating periods are shown in Figure 5.11.

The contrast associated with defects at different depths reach their peaks at various times.

(a)

(b)

(c)

(d)

Figure 5.11. Step heating thermograms after (a) 15s, (b) 20s, (c) 30s and (d) 75s of heating.

68

The temperature distribution profiles depicted in Figure 5.12 (b) are plotted along with the

rows identified in Figure 5.12 (a). The raw thermograms do not show smaller defects located deep

within the plate. Defects with a diameter of 4 mm were barely detected. The signals associated

with defects in the middle part of the plate dominate the signals associated with defects of the same

size and shallower depth. This is primarily caused by the non-uniform heat distribution generated

by a flash lamp, where the middle part of the sample receives more thermal energy than the sections

near the boundaries.

Most of the defects located in the first three rows are visible in raw thermograms since the

thermal variations of these areas are between 0.6 and 3.5 °C and the employed IR camera has a

thermal sensitivity of 20 mK. Smaller or deeper defects display low thermal variation or low

temperatures, which cannot be detected.

Temperature distribution profiles during the step heating thermography for all pixels along

the lines shown in Figure 5.12 (a) are illustrated in Figure 5.13. Figure 5.13 (a) was recorded after

heating for 75 s, and Figure 5.13 (b) was recorded after cooling for 6 s.

Figure 5.13 shows that most defects, especially those with higher diameters, generate

distinguishable thermal variation signals that are easily detectable by an IR camera. Signals

generated by the defects located in the last row, which have the smallest diameters, are not strong

enough to be easily detected. Contrary to pulse thermography, Figure 5.13 shows that the surface

of the object has been uniformly heated during step heating thermography which increases the

efficiency of this method in detecting internal defects.

The heating period affects the efficiency of step heating thermography, and is therefore one

of this method’s important parameters (Ghadermazi and Khozeimeh 2015; Daryabor and

Safizadeh 2016). Figure 5.11 demonstrates that smaller and deeper defects can be detected by

69

increasing the heating time. The same conclusion can be drawn by analyzing the profiles depicted

in Figure 5.14, where the temperature profiles of the first three rows of defects for different heating

periods are illustrated. Increasing the heating time increases the temperature variation between the

defective and sound areas, a finding that is corroborated by previous research (Daryabor and

Safizadeh 2016). Smaller or deeper defects can therefore be revealed with a longer heating period.

(a)

(b)

Figure 5.12. (a) Positions of temperature distribution profiles. (b) Temperature distribution

profiles using flash thermography.

The data presented in Figures 5.13 (a) and (b) demonstrate that the deeper the defect, the less

detectable it is. Moreover, the size of the defect is important.

70

(a)

(b)

Figure 5.13. (a) Temperature distribution after 75s of heating (b) temperature distribution after

6 s of cooling.

71

(a)

(b)

(c)

Figure 5.14. Thermal profiles for different heating time of step heating thermography at (a) first

row (b) second row (c) third row.

72

Temperatures in the areas of the defects were measured and analyzed to better evaluate the

effects of the defect’s depth and size on the temperature variation. The temperatures measured at

the locations of the flat-bottomed holes with various depths are illustrated in Figure 5.15. The

sample was heated by two halogen lamps for 75 s, during which time data were recorded. It can

be concluded from this figure that an increase in a defect’s depth reduces its temperature variation,

which makes detection more difficult.

Figure 5.15. Effect of depth and size of defect on the temperature distribution in sample heated

by halogen lamps for 75s.

5.2.2 Thermal Image Processing

5.2.2.1 Absolute Thermal Contrast (ATC)

The raw data were not of a sufficient quality to reveal the presence of deeper defects. Several

image processing techniques were used to increase the visibility and contrast of internal damages.

Figure 5.16 shows the results of applying Absolute Thermal Contrast (ATC) to pulsed

73

thermography data. ATC yields results that are a simple subtraction of temperatures in defective

and sound areas in C. Figure 5.16 shows that defects near the surface are the hottest points after

flash, and that those areas provide a greater contrast than deeper defects. Defects closer to the

surface also cool faster, which means that their contrasts fade away within several seconds of the

pulse. The first and second columns of defects, which are closer to the surface, are an exception

primarily due to non-uniform heating of the surface resulting in an uneven energy absorption.

Although it is an effective technique, ATC is still unable to detect some defects, especially the

deeper ones.

The results of ATC applied to the step heating thermograms taken after 75 s of heating are

shown in Figure 5.17. By comparing the results in Figures 5.16 and 5.17 it can be observed that

step heating has been more uniform than flash thermography. The rate of the temperature variation

is reduced as the defect’s depth increases. Another remarkable observation is that there is a

significant variation in thermal contrast at the early stages, which decreases as time increases until

it reaches almost steady state. This observation implies that after a certain point, increasing the

heating period will not result in the detection of more defects (Matovu 2015). The Absolut contrast

variations are not significant after 75 s heating and therefore this time was deemed to be an

appropriate heating time for the object. The results shown in Figures 5.16 and 5.17 demonstrate

that defects in the deepest areas of the blade and most of smaller defects could not generate enough

visible thermal contrast. Further image processing is required to improve the thermal contrast of

the poorly-visible defects.

74

(a)

(b)

(c)

(d)

Figure 5.16. Absolute thermal contrast of defect plate under pulse thermography technique at

defects located in (a) first row (b)second row (c) third row and (d) fourth row.

75

(a)

(b)

(c)

(d)

Figure 5.17. ATC of thermograms obtained by step heating thermography after 75s of heating at

(a) first row (b) second row (c) third row and (d) fourth row.

Different algorithms, including PCT, MF, and a combination of PPT as a transform-based

technique and TSR were employed to increase the SNR, which then increase the visibility of the

defects. The newly-developed method, Step Heating Phase and Amplitude Thermography

(SHPAT), was applied to step heating thermograms and successfully used for passive thermal data

processing.

5.2.2.2 Matched Filters (MFs)

MFs processing can successfully increase the thermal contrast in composite materials. In this

study, the MFs algorithms were applied to thermograms obtained from both pulsed and step

76

heating thermography. Figure 5.18 shows four MFs analyses of pulsed thermograms, including

the SAM, the ACE, the t-statistic, and the F statistic. The SAM provided the best results with high

visibility of all defects except D5.

(a)

(b)

(c)

(d)

Figure 5.18. Four MF algorithms results of thermograms obtained by flash thermography (a)

SAM, (b) ACE, (c) t-statistic and (d) F statistic

Applying MF to step heating thermograms improves the quality of results, as shown in

Figure 5.19. All four versions of MF increased the visibility and contrast of defects. Three of these

filters, including the SAM, the ACE and the F statistic, showed very promising results where all

defects could be detected.

77

(a)

(b)

(c)

(d)

Figure 5.19. Four matched filters including (a) SAM, (b) ACE, (c) t-statistic and (d) F statistic

when the specimen is under step heating thermography.

To quantitatively evaluate the results obtained by the application of MFs to either pulsed or

step heating thermography, the SNRs for defective areas are determined. The signal values of

defects are measured first. Then, by calculating the standard deviation of background noise and

using equation (3.31), the SNR of each defect can be calculated. Figure 5.20 (a) illustrates the

signals of F statistic results obtained from step heating data. The background noise is also shown

in Figure 5.20 (b).

78

(a)

(b)

Figure 5.20. (a) Signal values over the lines crossing the defects and (b) background noise.

SNR values associated with data obtained by the application of MFs to pulsed and step

heating thermograms are presented in Tables 5.2-5.6. Defect position names used in these tables

are defined in Figure 4.2 (b).

As these tables show higher SNR values can be obtained when MFs are applied to the step

heating data. It can therefore be concluded that the application of MFs to step heating data is

capable of revealing more details regarding internal defects.

79

Table 5.2. SNR related to data captured by application of SAM on pulsed thermal data

1 2 3 4 5

A 28.38 22.80 25.30 17.23 19.39

B 28.25 20.53 25.58 16.30 16.47

C 28.89 25.44 17.51 14.49 5.63

D 10.43 14.23 10.85 5.43 NA

Table 5.3. SNR related to data captured by application of SAM on step heating thermal data

1 2 3 4 5

A 36.78 35.33 31.83 26.98 15.08

B 33.45 33.69 31.15 27.86 17.02

C 30.74 29.33 29.09 26.31 18.83

D 16.12 18.68 18.64 13.94 5.483

Table 5.4. SNR related to data captured by application of ACE on step heating thermal data

1 2 3 4 5

A 33.18 31.97 29.14 24.33 11.64

B 30.60 29.96 28.46 25.40 15.25

C 26.42 27.08 26.45 24.03 16.69

D 14.71 16.88 16.92 11.74 8.47


1 2 3 4 5

A 44.28 41.65 34.97 28.46 15.69

B 35.59 37.04 34.51 38.77 22.74

C 28.24 30.16 30.29 27.64 19.35

D 13.50 18.39 16.0 11.33 6.58


1 2 3 4 5

A 40.18 37.61 32.74 27.00 13.82

B 33.04 33.84 32.17 28.26 17.82

C 27.16 28.96 28.62 26.07 18.46

D 13.55 17.23 16.53 11.63 2.69

80

Tables 5.2-5.6 also reveal that larger defects closer to the surface generate higher SNR

values. There are some exceptions to this general conclusion, however, primarily due to the type

of heating.

When MFs are used on the step heating and pulsed thermograms, they improve the quality

of the results associated with subsurface defects. As described in Chapter 3, however, the manual

selection of pixels from sound and target areas makes this method challenging since the contrast

in the results depends on appropriate points selection. Another effective and productive thermal

image processing method is transform based processing, which is discussed in the next section.

5.2.2.3 Principal Component Thermography (PCT)

As discussed in chapter 3, the output of PCT is orthogonal functions that contain spatial

variation components that can be used to detect internal defects (Matovu 2015; Manohar 2012).

Only the first few orthogonal functions have interpretable information associated with subsurface

defects, so there is no need to study all possible functions (Marinetti et al. 2004; Rajic 2002;

Matovu 2015).

Figure 5.21 shows the first four orthogonal functions of the thermal image sequence obtained

by using flash thermography technique. To achieve better visualization of defects, all the images

are presented in Jet colormap. The first orthogonal function was affected by non-uniform heating

of the surface and can capture any spatial variation within the data (Matovu 2015; Manohar 2012).

The third orthogonal function presented the highest visibility of the defects, and demonstrated the

ability to detect defects at greater depths. The fourth orthogonal function does not contain any

useful information, presumably because of low contrast on it. Based on the results presented here,

81

the second and third orthogonal functions provide the highest contrast related to the subsurface

defects.

The same image processing algorithm was used for step heating thermography for different

time steps ranging from 10 s to 75 s. Thermograms were recorded during both heating and cooling

periods. Figure 5.22 illustrates the first four orthogonal functions of thermograms recorded during

cooling after 20 s of heating by two halogen lamps. Similar to flash thermography, the third

orthogonal function provides the most useful information regarding subsurface defects while other

orthogonal functions do not yield enough information. Only the results of this variation component

are therefore presented for other time steps.

(a)

(b)

(c)

(d)

Figure 5.21. (a) First (b) second (c) third and (d) fourth orthogonal functions obtained from

pulsed thermograms.

82

Figure 5.23 shows the third orthogonal functions of thermograms captured during the

cooling after heating for various time steps. The SNR values related to third orthogonal functions

shown in Figure 5.23 are presented in Tables 5.7-5.10.

(a)

(b)

(c)

(d)


thermograms recorded during cooling after 20s of heating.

83

It can be concluded from the SNR values shown in Tables 5.7-5.10 that PCT processing of

the thermograms that were recorded during cooling after 40 s of heating provide the best outcomes.

Heating the surface for a longer time degrades most SNR values of defective areas. Similar to the

SNR results of MFs, the larger and closer defects to the surface have higher SNR values. In this

research, 40 s can therefore be considered as the maximum heating time for PCT processing of

thermograms captured during cooling to reveal the most information about the defects.

(a)

(b)

(c)

(d)

Figure 5.23. Third orthogonal functions extracted from thermograms recorded during cooling

after (a) 10s (b) 30s (c) 40s and (d) 75s of heating.

84


1 2 3 4 5

A 36.70 32.33 27.63 22.66 NA

B 30.91 30.37 28.38 22.78 9.60

C 28.21 13.40 24.75 23.12 16.34

D 28.04 15.00 NA NA NA


1 2 3 4 5

A 38.63 33.40 29.61 26.70 19.93

B 34.95 24.38 26.74 26.23 20.03

C 30.51 25.13 10.86 20.24 18.71

D 21.63 17.67 4.30 NA NA


1 2 3 4 5

A 38.40 35.98 31.27 28.29 22.71

B 35.48 26.24 26.95 26.73 24.14

C 30.49 26.63 13.72 19.03 19.02

D 23.56 19.42 11.13 NA NA


1 2 3 4 5

A 31.28 25.48 30.40 29.76 25.78

B 29.66 16.93 22.88 25.71 23.55

C 30.47 25.42 19.62 11.11 17.92

D 11.62 11.04 8.13 NA NA

The first four orthogonal functions associated with the thermograms captured during 30 s of

heating are depicted in Figure 5.24 . Both the second and third orthogonal functions contain

information associated with subsurface defects. Most flaws can be detected by considering a

combination of these two functions, so both are evaluated during different heating periods to assess

the effectiveness of PCT. The results are presented in Figure 5.25.

85

(a)

(b)

(c)

(d)


thermograms recorded during 30s heating.

86

(a)

(b)

(c)

(d)

(e)

(f)

Figure 5.25. Second orthogonal function of thermograms captured at (a) 20s (c) 40s and (e) 75s

heating and third orthogonal function extracted from thermograms recorded at (b) 20s (d) 40s

and (f) 75s heating.

87

Tables 5.11-5.16 list the SNR values related to subsurface defects as a quantitative

evaluation of the results depicted in Figure 5.25. The results presented in Tables 5.11-5.13 show

that the second orthogonal function obtained from thermograms captured during the 40 s heating

can provide the maximum SNR values of all defective areas. The SNR values of the third

orthogonal function presented in Tables 5.14-5.16 also demonstrate that the 40 s heating period

delivers higher values in most of defective positions. The results presented in Tables 5.11-5.16

show that 40 s is the best heating time for detecting internal defects using PCT processing of the

thermograms recorded during heating. Further increasing the heating period does not help to detect

more defects. As illustrated in Figures 5.25 (c) and (d), all defects except D5 (shown in Figure 4.2)

are highly visible.

Table 5.11. SNR of 2th orthogonal function of thermograms captured at 20s heating

1 2 3 4 5

A 29.6 25.93 17.43 14.39 NA

B 28.97 25.48 18.38 6.91 NA

C 28.18 26.94 21.15 12.55 NA

D 19.81 16.68 7.068 NA NA


1 2 3 4 5

A 35.56 36.28 33.46 30.18 20.63

B 35.345 35.59 31.90 28.61 20.82

C 30.25 35.51 31.09 27.99 19.43

D 19.92 23.61 18.49 17.56 NA

88


1 2 3 4 5

A 29.04 31.51 29.99 26.36 19.67

B 26.16 28 27.65 24.23 18.05

C 3.48 21.84 22.95 21.2 16.5

D 5.85 7.41 11.03 6.21 NA


1 2 3 4 5

A 29.42 27.9 23.4 18.21 7.14

B 18.31 26.15 22.98 18.08 4.73

C 32.92 20.82 21.82 18.2 9.82

D 11.67 6.4 10.47 NA NA


1 2 3 4 5

A 38.81 25.34 28.44 27.08 19.60

B 37.34 18.37 26.37 25.12 26.19

C 38.94 23.36 17.51 20.66 17.66

D 20.23 11.80 8.634 NA NA


1 2 3 4 5

A 41.27 13.3 27.68 28.57 23.79

B 39.59 24.53 20.87 23.95 20.74

C 36.25 27.74 17.51 10.1 16.56

D 19.74 17.61 9.64 NA NA

This section demonstrated that PCT is an effective method of increasing the visibility of

subsurface defects. As discussed in chapter 3, however, this method has some drawbacks that were

documented. The next thermal image processing employed in this study is MF.

89

5.2.2.4 Transform Based Processing Techniques (PPT, SHPAT)

Two thermography techniques, pulse and long pulse or step heating, were employed in this

study to heat the defect plate. The heating period for step heating varied from 10 to 75 s. The

results of phase images and amplitude images obtained from step heating thermography, and PPT

applied on pulsed thermography data are explored in this section.

Section 3.3.4 discussed that amplitude images are more sensitive to non-uniform heating

than phase images, and are therefore more reliable when the surface is heated uniformly (Maldague

and Marinetti 1996).

In pulsed thermography, the thermal decay acquisition time is 54.2 s with a sampling

frequency of 15 Hz. For this recording time and sampling frequency, a time interval of 0.066 s can

be achieved. Based on the relationship between frequency interval and acquisition time

( f =1/N t ) (Castanedo 2005; Pawar 2012), the minimum frequency of 0.0187 Hz can be

achieved. Since the flash lamp was close to the plate and the effect of non-uniform heating is high

during the first second of cooling, thermograms captured in the initial second of cooling are not

included in the analysis to increase the quality of results. Figures 5.26 (a) and (b) show a

comparison of raw thermograms and normalized phase contrast obtained using FFT. PPT can

significantly increase the contrast so that all defects, except D5, can be detected. The random

contrast in Figure 5.26 (b) are caused by non-uniform heat distribution of flash lamps.

The SNR values related to the phase image depicted in Figure 5.26 are presented in Table

5.17, which shows that larger defects that are closer to the surface generate higher SNR values.

The first column of defects is an exception to this observation, mainly due to non-uniform heating

of the surface.

90

(a) (b)

Figure 5.26. (a) Raw thermogram and (b) phase image (acquisition time =53.2 s) obtained from

thermograms recorded at cooling period after flashing the surface.

Table 5.17. SNR of phase image obtained by application of PPT on pulsed thermal data

1 2 3 4 5

A 25.48 26.80 24.94 23.22 20.73

B 23.12 26.52 26.97 23.39 19.91

C 19.59 20.76 19.87 14.24 12.90

D 13.87 16.41 12.81 7.01 NA

The possibility of defect detection in terms of minimum diameter-to-depth ratio is an

important parameter in the PPT technique (Couturier and Maldague 1997). Figure 5.26 (b)

illustrates that the smallest defect size that could be detected by the PPT setup that was employed

in this experiment was 4 mm embedded at a depth of 2 mm. This results in a minimum diameter-

to-depth ratio of 4/2=2.

The frequency window is another important parameter in PPT that can affect the detection

of defects at different depths (Castanedo 2005; Pawar 2012). Figure 5.27 shows the effect of the

minimum frequency on the thermal contrast quality where the sampling rate was kept constant and

the minimum frequency was changed by manipulating the recording time. By increasing the

91

minimum frequency, the noise increases and the contrast associated with the defects decreases.

Increasing the minimum frequency therefore decreases the quality of results obtained with PPT.

In other words, as the minimum frequency increases, the detectable depth of defects decreases

(Pawar 2012). There is an optimal value of minimum frequency that provides the best contrast

between defective and sound areas, so it is important to choose optimal sampling parameters

including acquisition time and sampling frequency to produce the most visible results for all

defects embedded in different layers.

The object was heated for different periods of time in step heating thermography. The

sampling frequency for this experiment was identical to that of the pulsed thermography

experiment. Data were recorded during both heating and cooling. The phase and amplitude images

are discussed here. Figures 5.28 and 5.29 show the phase and amplitude images at the minimum

frequency where FFT was applied to the thermograms captured during the cooling after heating

for different periods of time.

92

(a)

(b)

(c)

(d)

Figure 5.27. Application of FFT on pulsed thermograms and obtained phase images of specimen

with minimum frequency of a) 0.149Hz (acquisition time = 6.66s), b) 0.075Hz (acquisition time

= 13.33s) c) 0.05 Hz (acquisition time = 20s) d) 0.027 Hz (acquisition time = 36.66s).

93

(a)

(b)

(c)

(d)

(e)

(f)


captured at cooling process after (a) 10s (fmin=0.018Hz) (c) 20s (fmin=0.012Hz) (e) 30s

(fmin=0.0154Hz) of heating and phase images of specimen obtained at cooling process after (b)

10s (d) 20s (f) 30s of heating.

94

(a)

(b)

(c)

(d)


captured at cooling process after (a) 40s (fmin=0.0144Hz) and (c) 75s (fmin=0.015Hz) of heating

and phase images of specimen obtained at cooling process after (b) 40s and (d) 75s of heating.

Phase images can reveal defects with better contrast visibility than amplitude images in some

specific areas. This demonstrates that a reliable inspection can be achieved by evaluating the

combination of both results. Figure 5.30 shows the results of an FFT transform that was applied to

thermograms captured during different heating periods.

95

(a)

(b)

(c)

(d)

(e)

(f)

Figure 5.30. Amplitude images of specimen captured at heating after (a) 20s (fmin=0.075Hz) (c)

40s (fmin=0.0292) and (e) 75s (fmin=0.0227Hz) of heating and phase images of specimen obtained

at heating period after (b) 20s (d) 40s and (f) 75s of heating.

96

The phase image of thermograms captured during 75 s of heating could capture all defects,

demonstrating that phase images extracted from heating data provide more visibility than data

obtained from cooling. Amplitude images obtained during the cooling period are less noisy and

contain more detail, which allows the shape of defects to be determined.

Figure 5.31 presents a further analysis of the normalized amplitude contrast distribution of

thermograms captured during cooling after 75 s of heating. The curves in this figure are related to

the amplitude variation along the lines shown in Figure 5.12 (a). A significant change occurs

between sound and defective areas in the amplitudes, leading to a sharp boundary around the

defects. Deeper defects generally show lower amplitudes with a smoother transition from sound

to defective areas. As the size of the defect decreases, so does the amplitude.

Figure 5.31. Normalized amplitude value distribution of defects with minimum frequency of

0.015Hz as thermograms obtained during cooling after 75s of heating.

The normalized phase contrast distribution of thermograms obtained at 75 s of heating is

depicted in Figure 5.32. Bigger defects closer to the surface generate higher phase contrasts and

97

are subsequently more visible in the phase image. All defects could be detected by analyzing these

plots, but D5 is not seen clearly.

Figure 5.32. Normalized phase value distribution of defects with minimum frequency of 0.015Hz

as thermograms obtained during 75s of heating.

The SNR values for the phase and amplitude images captured by the application of FFT on

step heating data are listed in Tables 5.18-5.20. These results demonstrate that amplitude images

extracted from thermal data captured during cooling have higher SNR values and reveal more

details of subsurface defects. The SNRs of both phase and amplitude images show that larger

defects closer to the surface provide higher values.

Table 5.18. SNR of amplitude image of thermograms captured during cooling after 75s heating

1 2 3 4 5

A 46.18 45.09 42.23 38.43 29.66

B 41.02 41.34 39.57 36.74 28.70

C 33.05 33.74 34.60 33.40 25.90

D 14.86 23.46 21.24 17.02 13.56

98

Table 5.19. SNR of phase image of thermograms captured during 75s heating

1 2 3 4 5

A 38.54 30.94 29.23 27.57 23.61

B 36.02 32.00 29.21 27.55 23.22

C 21.47 26.30 27.77 26.59 22.97

D 19.01 13.29 18.30 17.69 12.69

Table 5.20. SNR of amplitude image of thermograms captured during 75s heating

1 2 3 4 5

A 33.63 41.97 40.86 38.31 31.65

B 26.91 38.47 37.59 34.72 29.22

C 20.32 26.21 30.10 28.67 24.21

D 8.02 13.06 12.70 10.83 9.425

It can be concluded from the results presented in Tables 5.18-5.20 that the application of the

FFT transform to step heating thermography is more efficient than its application to flash

thermography. This is especially true for amplitude data, where higher SNR values are achieved.

The amplitude images extracted from step heating thermography are an effective means of

revealing subsurface damages, as they can detect even small and deep defects.

5.3 Passive thermography

This section presents the results of raw and processed passive thermography.

5.3.1 Day time experiment

Passive thermal imaging of the damaged blade section was conducted at different times of

day. Typical results are shown in Figures 5.33 and 5.34. The results demonstrate that passive

thermography is capable of capturing cracks, delamination, and internal features of the blade

section.

99

Early morning experiments provided visible contrast of the flaws on the defect plate attached

to the damaged blade section primarily due to the considerable temperature change on the blade

during this period (Worzewski and Krankenhagen 2016; Worzewski, Krankenhagen, and

Doroshtnasir 2016; Meinlschmidt and Aderhold 2006). All defects of the defect plate, except the

smallest 4 mm ones, were detected during this period. Less useful information about the defects

was obtained when the experiment was performed around noon. None of the defects were visible

during the evening (around 6 pm), which is mainly due to the balanced temperature on the blade

surface after several hours of heating.

Figure 5.33. Thermographic results of experiment at morning around 9 am. The vertical arrows

indicate the shear webs.

100

(b)

(c)

Figure 5.34. Raw thermograms at (a) noon and (b) around 6 pm (sunrise and sunset were around

5.53 am and 9.30 pm, respectively). The vertical arrows and dashed lines indicate the shear

webs.

101

Thermal contrasts associated with dirt (glue on the surface), within the blue box in Figure

5.34, were most pronounced at noon. These contrasts faded during the afternoon. It is difficult to

distinguish between signatures associated with dirt and other features without image processing.

The blade’s internal features such as shear webs were detected as cold regions during the morning

and at noon, but became hot signatures during the evening after several hours of heating.

Cracks and delamination on the suction side are apparent near the blade’s trailing edge. At

noon, with peak sunlight, cracks and delamination were detected. Delamination in the upper area,

identified by the green box in Figure 5.33, could not be detected clearly during the morning. The

evening thermograms did not provide any information regarding cracks and delamination, so

morning and noon were the best times for crack and delamination monitoring.

5.3.2 Monitoring during heating and cooling

Heating and cooling were both used in this study to monitor the damaged blade section.

Figure 5.35 shows that different internal features and shear webs were detected during both the

heating and cooling periods. Blue boxes indicate areas of dirt and glue on the surface and the

yellow box highlights coloured fibres inside the material. Green and purple boxes show cracks and

delamination.

102

(a)

(b)

Figure 5.35. Thermographic results at (a) heating and (b) cooling.

The shear webs generated cold signatures during heating of the blade, while these cold

signatures were replaced with warmer spots during cooling, as marked by red arrows in Figure

5.35. The temperature inversion was also noticeable on the defect plate as hot and cold spots during

heating and cooling, respectively. Thermograms recorded surface inhomogeneities such as dirt and

glue, which are highlighted by blue boxes in Figure 5.35. Coloured parts inside the material,

marked by the yellow box in Figure 5.35, possess different thermal properties, as they absorb more

energy. They created hot spots during heating, but were not apparent during the cooling. It can be

concluded that heating and cooling both provide useful information pertaining to the inspection of

cracks and delamination, but that heating is more effective and provides greater contrast. The

signatures of cracks and delamination obtained during heating and cooling are highlighted by green

and purple boxes in Figure 5.35.

103

5.3.3 Passive Thermal Image Processing

The transform based technique discussed in section 3.3.4 of chapter 3, SHPAT, was

employed for the first time to increase the quality of passive thermography results. Phase images

captured using this method are not sensitive to non-uniform heating.

FFT was applied to thermograms obtained at different times of day. The results are shown

in Figure 5.36. Part (b) illustrates that amplitude images could considerably increase the quality

and visibility of the visualized subsurface defects, as the visibility of cracks, delamination and a

large portion of the flat-bottomed hole defects was improved. The white boxes highlight cracks

and delamination and the red box show thermoplastic foam. Phase images could noticeably

increase the contrast of shear webs signatures, marked by red arrows in Figure 5.36 (a).

(a)

(b)

Figure 5.36. (a) Phase images of passive thermograms captured during the morning at a

frequency of 0.00184 Hz and (b) amplitude image of passive thermograms recorded during the

morning at a frequency of 0.0165 Hz.

104

Cracks and delamination are marked by white boxes in Figure 5.36 (b). The thermoplastic

foam near the leading edge was detected in the amplitude results and is highlighted by red box in

Figure 5.36 (b). This method not only increases the quality of thermal images and improves the

detectability of thermography, but also eliminates the false indications associated with

environmental reflections, dirt, and dust on the surface.

The results of the image processing of the passive thermograms obtained at the noon are

shown in Figure 5.37. The amplitude image revealed an acceptable visibility of the shear webs,

while just one of the shear webs was detected by the phase image. Other signatures associated with

cracks, delamination, blind holes, and thermoplastic foam were improved with almost the same

quality for both phase and amplitude images. These are marked with different colored boxes in

Figures 5.37 (a) and (b).

(a)

(b)

Figure 5.37. (a) Amplitude image at a frequency of 0.0318 and (b) phase image at a frequency of

0.0053 extracted from raw passive thermal images captured at noon.

105

It can be concluded from the results in this section that the visibility of different subsurface

features increases after applying the proposed thermal image processing technique, SHPAT, to the

raw passive thermography data. It was observed that both phase and amplitude images could

provide useful information for monitoring the condition of wind turbine blades.

106

CONCLUSION AND FUTURE WORKS

6.1 Conclusion

The most critical components of wind turbines, the blades, are susceptible to failure and

performance degradation due to the initiation and propagation of surface and subsurface damage

in many forms. The most common surface damage to wind turbine blades is leading-edge erosion

resulting from airborne particles. Manufacturing problems or operation under harsh weather

conditions may result in subsurface damage. The increasing size of wind turbine blades also makes

these issues more challenging. Monitoring the condition of wind turbine blades is of great

importance to minimize unwanted failures.

Non-Destructive Testing techniques can be used to increase the lifespan of wind turbine

blades while decreasing their loss of power production and the cost of failure. Chapter 2 discussed

the advantages and drawbacks of many different Non-Destructive Testing techniques. Because of

the structure of wind turbines and the continuous operation of the blades, new methods are required

to remotely inspect wind turbine blades. Thermography has the potential to detect internal damage

of operating wind turbine blades as a non-contact, full-field condition monitoring method.

Leading-edge erosion is an important phenomenon in the wind turbine industry which that

may result in a loss of energy. In this study, a method was developed to evaluate the degradation

of airfoil performance due to leading-edge erosion. To achieve this, a section of a wind turbine

blade that has never been used was sand-blasted to simulate severe leading-edge erosion. Then,

both clean and rough surfaces were reconstructed and Computational Fluid Dynamics simulations

were used to evaluate the aerodynamic characteristics of the airfoils. Among all available

reconstruction methods, laser scanning was the most suitable technique to reconstruct the surface

107

of the blade. The accuracy of reconstruction is the most important factor, so a laser scanner with

submillimeter accuracy was selected for this study to provide the capability of capturing roughness

and small shape variations on the surface due to erosion.

Passive and active thermography were used to detect subsurface defects. Active

thermography heated the surface using halogen and flash lamps. Passive thermography used

sunlight to heat the surface, particularly in the morning and noon. Since artificial lighting of

operational blades is difficult, active thermography cannot be employed to inspect blades in the

field. The technique is therefore limited to monitoring the condition of blades on the ground during

major maintenance or pre-delivery.

Active thermography was conducted on a specially-constructed defect plate to evaluate this

method’s potential for detecting subsurface defects. The 170195 mm plate was cut from the

laminate skin of a wind turbine blade. Flat-bottomed holes were drilled from the inside to produce

defects with known diameters and penetrations. Pulsed and step heating thermography were used

in this research. The major difference between these two active thermography techniques was the

heating source: pulsed thermography used a flash lamp, while halogen lamps were used for step

heating. The results demonstrated that active thermography is a powerful method for the

monitoring and fault detection of wind turbine blades, but that the signals generated by some small

defects could not be detected.

Passive thermography was conducted on the damaged blade section and the defect plate

attached to it. This experiment was conducted at different times of day to determine the most

favorable time for maximum defect detection. The results showed that certain times of the day

were ideal for detecting certain types of defects. Cracks, delamination, and surface dirt generated

108

the most visible signatures around noon, while defects such as flat-bottomed holes in the defect

plate were more visible in the morning as the sun heated the targets.

The thermograms obtained by both passive and active thermography must be processed to

reveal the small defects located deep in the laminate skin. Different image processing algorithms

including Absolute Thermal Contrast, Principal Component Thermography, Matched Filters,

Thermal Signal Reconstruction, and Pulsed Phase Thermography were used to increase the quality

of active thermograms. A method called “Step-Heating Phase and Amplitude Thermography”, was

developed to analyze the step heating and passive thermal images.

All the image processing algorithms that were employed improved the quality of active

thermograms, but some drawbacks were noted for the Matched Filters and Principal Component

Thermography methods. The Matched Filters method is not fully automated and requires manual

selection of points in a sound area of the damaged blade, which is time-consuming and affects the

quality of results. The Principal Component Thermography improves the visibility of some

defects, but it may adversely affect the quality of contrast associated with other subsurface

damages. The interpretation of results obtained by Principal Component Thermography is also

challenging, as each defect may generate either hot or cold spots depending on the experimental

conditions.

Pulsed Phase Thermography, and Step Heating Phase and Amplitude Thermography as

transform based techniques improved the visibility and contrast of the subsurface defects. Phase

images are not sensitive to non-uniform surface heating, while amplitude results could be affected

by it. Step Heating Phase and Amplitude Thermography could provide efficient results for step

heating thermography. Step Heating Phase and Amplitude Thermography, as a successful method

for improving active thermography results, was applied to passive thermal data. The quality of

109

passive thermograms was increased as a result. This method could also eliminate the false

indications associated with environmental reflections and dirt on the surface.

6.2 Future Works

The results of this study demonstrated that both laser scanning and thermography methods

are useful for monitoring the condition of wind turbine blades. More researches are required to

increase the efficiency of these methods. The following proposals can be considered to advance

the suggested methods for inspecting surface and subsurface defects of blades:

1. The method employed in this research to address the issue of leading-edge erosion is

limited to blades on the ground. It cannot be implemented for operating wind turbine

blades, primarily because the proposed laser scanning method is a close-range

application method which requires the attachment of several optical reflectors on the

blade’s surface. Development of a non-contact method to measure the blade’s

roughness and to determine its shape with high accuracy from long distance is

important. Research needs to be developed to solve these limitations and to make this

method more practical for blades in the field.

2. The proposed passive thermography method should be improved for use on operating

wind turbine blades. A flight campaign, using a drone-mounted IR camera, may be an

effective method of condition monitoring. A setup should be designed to inspect all

surfaces of the blade in an optimal time. Different factors such as minimum distance

between the drone and the blade, the angle of view, and environmental conditions

should be considered during this experiment’s design. The proposed image processing

110

algorithms are not applicable to this experiment because those methods need a constant

relative position between the blade and the IR camera during the experiment. A

development of other image processing algorithms to improve the quality of raw

passive thermograms is therefore required.

3. The design and development of a passive thermography system to inspect operating

wind turbine blades from the ground is another suggested future work. Since infrared

emissions can be captured from long distances, an experiment can be designed to

monitor the blade from the ground. The image processing algorithms suggested in this

study may be usable for this application. During image processing, an image correlation

algorithm as a preprocessing method may need to be conducted on captured

thermograms to eliminate the effect of small changes to the position of blades during

the experiment. Other passive thermography experiments and image processing

techniques can be developed to inspect the blades under operation.

111

References

Adams, Douglas, Jonathan White, Mark Rumsey, and Charles Farrar. 2011. “Structural Health

Monitoring of Wind Turbines: Method and Application to a HAWT.” Wind Energy 14 (4):

603–23.

Amayri, A, Z Tian, and T Jin. 2011. “Condition Based Maintenance of Wind Turbine Systems

Considering Different Turbine Types.” Quality, Reliability, Risk, Maintenance, and Safety

Engineering (ICQR2MSE), 2011 International Conference On. IEEE, 596–600.

Amenabar, I, A Mendikute, and A López-Arraiza. 2011. “Comparison and Analysis of Non-

Destructive Testing Techniques Suitable for Delamination Inspection in Wind Turbine

Blades.” Composites Part B: Engineering 42 (5): 1298–1305.

Badghaish, A. A. 2008. “Nondestructive Inspection of Thick Composites Using Step Heating

Thermography.” PhD Diss., Florida Institute of Technology.

Badghaish, A. A, and D. C Fleming. 2008. “Non-Destructive Inspection of Composites Using Step

Heating Thermography.” Journal of Composite Materials 42 (13): 1337–57.

Bagavathiappan, S, BB Lahiri, and T Saravanan. 2013. “Infrared Thermography for Condition

Monitoring–a Review.” Infrared Physics & Technology 60: 35–55.

Bagavathiappan, S, and T Saravanan. 2008. “Condition Monitoring of Exhaust System Blowers

Using Infrared Thermography.” Insight-Non-Destructive Testing and Condition Monitoring

50 (9): 512–15.

Bainbridge, BG. 2010. Optimization of Transient Thermography Inspection of Carbon Fiber

Reinfored Plastics. M.Sc., Southern Illinois University at Carbondale.

Balageas, D., S. Bourasseau, M. Dupont, E. Bocherens, V. Dewynter-Marty, and P. Ferdinand.

112

2000. “Comparison between Non-Destructive Evaluation Techniques and Integrated Fiber

Optic Health Monitoring Systems for Composite Sandwich Structures.” Journal of Intelligent

Material Systems and Structures 11 (6): 426–37.

Beattie, A. 1997. “Acoustic Emission Monitoring of a Wind Turbine Blade during a Fatigue Test.”

In 35th Aerospace Sciences Meeting and Exhibit, 239–48. Reston, Virigina: American

Institute of Aeronautics and Astronautics.

Beattie, Alan, and Mark Rumsey. 1999. “Non-Destructive Evaluation of Wind Turbine Blades

Using an Infrared Camera.” In 37th Aerospace Sciences Meeting and Exhibit. Reston,

Virigina: American Institute of Aeronautics and Astronautics.

Becerik-Gerber, B, F Jazizadeh, and G Kavulya. 2011. “Assessment of Target Types and Layouts

in 3D Laser Scanning for Registration Accuracy.” Automation in Construction 20 (5): 649–

58.

Besnard, F, and L Bertling. 2010. “An Approach for Condition-Based Maintenance Optimization

Applied to Wind Turbine Blades.” IEEE Transactions on Sustainable Energy 1 (2): 77–83.

Biju, N, N Ganesan, and CV Krishnamurthy. 2009. “Frequency Optimization for Eddy Current

Thermography.” NDT & E International 42 (5): 415–20.

Blanco, MI. 2009. “The Economics of Wind Energy.” Renewable and Sustainable Energy

Reviews, 1372–82.

Bogue, Robert. 2010. “Three‐dimensional Measurements: A Review of Technologies and

Applications.” Sensor Review 30 (2): 102–6.

Borum, KK, and M McGugan. 2006. “Condition Monitoring of Wind Turbine Blades.” 27th Risø

International Symposium on Materials Science. Risø National Laboratory,.

Byrnes, J. 2008. Unexploded Ordnance Detection and Mitigation. Springer Science & Business

113

Media.

Carslaw, HS, and JC Jaeger. 1959. “Heat in Solids.” Oxford: Clarendon Press, 2nd Ed.

Castanedo, CI. 2005. “Quantitative Subsurface Defect Evaluation by Pulsed Phase Thermography:

Depth Retrieval with the Phase.” PhD Diss., Université Laval.

Chatterjee, K, S Tuli, SG Pickering, and DP Almond. 2011. “A Comparison of the Pulsed, Lock-

in and Frequency Modulated Thermography Nondestructive Evaluation Techniques.” NDT

& E International 44 (7): 655–67.

Chatzakos, P, N Avdelidis, and K Hrissagis. 2010. “Autonomous Infrared (IR) Thermography

Based Inspection of Glass Reinforced Plastic (GRP) Wind Turbine Blades (WTBs).”

Robotics Automation and Mechatronics (RAM), IEEE Conference, 557–62.

Chen, J, X Wu, MY Wang, and X Li. 2013. “3D Shape Modeling Using a Self-Developed Hand-

Held 3D Laser Scanner and an Efficient HT-ICP Point Cloud Registration Algorithm.” Optics

& Laser Technology 45, 414–23.

Ciang, C. C, JR Lee, and HJ Bang. 2008. “Structural Health Monitoring for a Wind Turbine

System: A Review of Damage Detection Methods.” Measurement Science and Technology

19 (12): 122001.

Couturier, JP, and XP Maldague. 1997. “Pulsed Phase Thermography of Aluminum Specimens.”

International Society for Optics and Photonics 3056: 170–76.

D’Apuzzo, N. 1998. “Automated Photogrammetric Measurement of Human Faces.” In ISPRS

Commission V Symposium’Real-Time Imaging and Dynamic Analysis’. Swiss Federal

Institute of Technology, Institute of Geodesy and Photogrammetry (IGP).

Daryabor, P, and MS Safizadeh. 2016. “Investigation of Defect Characteristics and Heat Transfer

in Step Heating Thermography of Metal Plates Repaired with Composite Patches.” Infrared

114

Physics & Technology 76: 608–20.

de Azevedo, H. D. M, A. M Araújo, and N Bouchonneau. 2016. “A Review of Wind Turbine

Bearing Condition Monitoring: State of the Art and Challenges.” Renewable and Sustainable

Energy Reviews 56: 368–79.

Doroshtnasir, Manoucher, Tamara Worzewski, Rainer Krankenhagen, and Mathias Röllig. 2016.

“On-Site Inspection of Potential Defects in Wind Turbine Rotor Blades with Thermography.”

Wind Energy 19 (8): 1407–22.

Echavarria, E., B. Hahn, G. J. W. van Bussel, and T. Tomiyama. 2008. “Reliability of Wind

Turbine Technology Through Time.” Journal of Solar Energy Engineering 130 (3): 31005.

Fischer, K, F Besnard, and L Bertling. 2012. “Reliability-Centered Maintenance for Wind

Turbines Based on Statistical Analysis and Practical Experience.” IEEE Transactions on

Energy Conversion 27 (1): 184–95.

Foy, B. R. 2009. “Overview of Target Detection Algorithms for Hyperspectral Data.” Los Alamos

National Lab., Los Alamos, NM, Report to NNSA, Rep. No. LU-UR-09-00593.

Galleguillos, C., A. Zorrilla, A. Jimenez, L. Diaz, Á. L. Montiano, M. Barroso, A. Viguria, and F.

Lasagni. 2015. “Thermographic Non-Destructive Inspection of Wind Turbine Blades Using

Unmanned Aerial Systems.” Plastics, Rubber and Composites 44 (3): 98–103.

Gavrilov, DJ. 2014. “Development and Optimization of Thermographic Techniques for Non-

Destructive Evaluation of Multilayered Structures.” PhD Diss., University of Windsor.

Ghadermazi, K, and MA Khozeimeh. 2015. “Delamination Detection in Glass–epoxy Composites

Using Step-Phase Thermography (SPT).” Infrared Physics & Technology 72: 204–9.

Greco, A, S Sheng, J Keller, and A Erdemir. 2013. “Material Wear and Fatigue in Wind Turbine

Systems.” Wear 302 (1): 1583–91.

115

“GWEC. Global Wind Report Annual Market Update.” 2016.

Hahn, B, and M Durstewitz. 2007. “Reliability of Wind Turbines.” Wind Energy, Springer, Berlin,

Heidelberg, 329–32.

Holland, SD. 2011. “Thermographic Signal Reconstruction for Vibrothermography.” Infrared

Physics & Technology 54 (6): 503–11.

“Http://www.flir.ca/home/.” 2018. Accessed Online on 1/26/2018.

“Http://www.speedotron.com/.” 2018. Accessed Online on 1/26/2018.

“https://www.creaform3d.com.” 2018. Accessed Online on 1/26/2018.

Ibarra-Castanedo, C., M. Genest, P. Servais, X. P. V. Maldague, and A. Bendada. 2007.

“Qualitative and Quantitative Assessment of Aerospace Structures by Pulsed

Thermography.” Nondestructive Testing and Evaluation 22 (2–3): 199–215.

Ibarra-Castanedo, Clemente, and Xavier Maldague. 2004. “Pulsed Phase Thermography

Reviewed.” Quantitative InfraRed Thermography Journal 1 (1): 47–70.

Irwanto, M, N Gomesh, MR Mamat, and YM Yusoff. 2014. “Assessment of Wind Power

Generation Potential in Perlis, Malaysia.” Renewable and Sustainable Energy Reviews 38:

296–308.

Jain, J, and A Jain. 1981. “Displacement Measurement and Its Application in Interframe Image

Coding.” IEEE Transactions on Communications 29 (12): 1799–1808.

J Ribrant. 2006. “Reliability Performance and Maintenance—a Survey of Failures in Wind Power

Systems.” KTH School of Electrical Engineering, 59–72.

Joosse, PA, and MJ Blanch. 2002. “Acoustic Emission Monitoring of Small Wind Turbine

Blades.” TRANSACTIONS-AMERICAN SOCIETY OF MECHANICAL ENGINEERS

JOURNAL OF SOLAR ENERGY ENGINEERING 124 (4): 446–54.

116

Juengert, A, and CU Grosse. 2009. “Inspection Techniques for Wind Turbine Blades Using

Ultrasound and Sound Waves.” Proceedings of Non-Destructive Testing in Civil Engineering,

Nantes, France.

Jüngert, A. 2008. “Damage Detection in Wind Turbine Blades Using Two Different Acoustic

Techniques.” The NDT Database & Journal (NDT).

Keegan, M H, D H Nash, and M M Stack. 2013. “On Erosion Issues Associated with the Leading

Edge of Wind Turbine Blades.” Journal of Physics D: Applied Physics 46 (38): 383001.

Khalfallah, MG, and AM Koliub. 2007. “Effect of Dust on the Performance of Wind Turbines.”

Elsevier 209 (1–3): 209–20.

Krebbera, K, and W Habela. 2005. “Fiber Bragg Grating Sensors for Monitoring of Wind Turbine

Blades.” In Proc. of SPIE 5855: 1037.

Kretzmann, JE. 2016. “Evaluating the Industrial Application of Non-Destructive Inspection of

Composites Using Transient Thermography.” M.Sc. Diss., Stellenbosch: Stellenbosch

University.

Lading, Lars, Malcolm Mcgugan, Peder Sendrup, Jørgen Rheinländer, and Jens Rusborg. 2002.

“Fundamentals for Remote Structural Health Monitoring of Wind Turbine Blades – a

Preproject Annex B – Sensors and Non-Destructive Testing Methods for Damage Detection

in Wind Turbine Blades.” Risø National Laboratory, Roskilde, Denmark, Risø.

Larsen, CA. 2011. “Document Flash Thermography.” M. Sc. Utah University.

Larsen, GC, MH Hansen, A Baumgart, and I Carlén. 2002. “Modal Analysis of Wind Turbine

Blades.” Risø National Laboratory, Roskilde, Denmark.

Liu, W, B Tang, and Y Jiang. 2010. “Status and Problems of Wind Turbine Structural Health

Monitoring Techniques in China.” Renewable Energy 35 (7): 1414–18.

117

Lizaranzu, M, A Lario, and A Chiminelli. 2015. “Non-Destructive Testing of Composite Materials

by Means of Active Thermography-Based Tools.” Infrared Physics & Technology 71: 113–

20.

Maldague, X. 2002. “Introduction to NDT by Active Infrared Thermography.” Materials

Evaluation 60 (9): 1060–73.

Maldague, X., and S. Marinetti. 1996. “Pulse Phase Infrared Thermography.” Journal of Applied

Physics 79 (5): 2694–98.

Malhorta, A, K Gupta, and K Kant. 2011. “Laser Triangulation for 3D Profiling of Target.”

International Journal of Computer Applications 35 (8).

Manohar, A. 2012. “Quantitative Nondestructive Testing Using Infrared Thermography.” PhD

Diss., UC San Diego.

Manohar, A, and F Lanza di Scalea. 2013. “Detection of Defects in Wind Turbine Composite

Blades Using Statistically Enhanced Lock-In Thermography.” Structural Health Monitoring:

An International Journal 12 (5–6): 566–74.

Manohar, A, and J Tippmann. 2012. “Localization of Defects in Wind Turbine Blades and Defect

Depth Estimation Using Infrared Thermography.” Proc. SPIE 8345: 83451O.

Marinetti, S, E Grinzato, PG Bison, and E Bozzi. 2004. “Statistical Analysis of IR Thermographic

Sequences by PCA.” Infrared Physics & Technology 46 (1): 85–91.

Marinetti, S, A Muscio, and PG Bison. 2000. “Modeling of Thermal Nondestructive Evaluation

Techniques for Composite Materials.” Thermosense Xxii, International Society for Optics

and Photonics 4020: 164–74.

Márquez, FPG, AM Tobias, JMP Pérez, and Papaelias. M. 2012. “Condition Monitoring of Wind

Turbines: Techniques and Methods.” Renewable Energy 46: 169–78.

118

Marsh, G. 2011. “Meeting the Challenge of Wind Turbine Blade Repair.” Reinforced Plastics 55

(4): 32–36.

Matovu, M. 2015. “Damage Assessment of Steel-Plate Concrete Composite Walls by Using

Infrared Thermography.” PhD Diss., State University of New York at Buffalo.

Mayor, GS, A.B.B Moreira, and HC Muñoz. 2013. “Influence of Roughness in Protective Strips

of Leading Edge for Generating Wind Profiles.” 22nd International Congress of Mechanical

Engineering, COBEM.

McMillan, D, and GW Ault Generation. 2008. “Condition Monitoring Benefit for Onshore Wind

Turbines: Sensitivity to Operational Parameters.” IET Renewable Power Generation 2 (1):

60–72.

Meinlschmidt, P, and J Aderhold. 2006. “Thermographic Inspection of Rotor Blades.”

Proceedings of the 9th European Conference on NDT.

Mendez, B, A Munoz, and X Munduate. 2015. “Study of Distributed Roughness Effect over Wind

Turbine Airfoils Performance Using CFD.” In 33rd Wind Energy Symposium. Reston,

Virginia.

Molina, MG, and PE Mercado. 2011. “Modelling and Control Design of Pitch-Controlled Variable

Speed Wind Turbines.” Wind Turbines. InTech.

Montanini, R. 2010. “Quantitative Determination of Subsurface Defects in a Reference Specimen

Made of Plexiglas by Means of Lock-in and Pulse Phase Infrared Thermography.” Infrared

Physics & Technology 53 (5): 363–71.

Montanini, R, and F Freni. 2012. “Non-Destructive Evaluation of Thick Glass Fiber-Reinforced

Composites by Means of Optically Excited Lock-in Thermography.” Composites Part A:

Applied Science and Manufacturing 43 (11): 2075–82.

119

Myrent, NJ. 2013. “Rotor Blade Operational Data Analysis Methods and Applications for Health

Monitoring of Wind Turbines Using Integrated Blade Sensing.” M.Sc. Diss., Purdue

University.

Nie, M, and L Wang. 2013. “Review of Condition Monitoring and Fault Diagnosis Technologies

for Wind Turbine Gearbox.” Procedia Cirp 11: 287–90.

Ouyang, J, Z Liang, and H Zhang. 2006. “Research of Measuring Accuracy of Laser Tracker

System.” International Society for Optics and Photonics 6280: 62800T.

Pawar, SS. 2012. “Identification of Impact Damage in Composite Laminates through Integrated

Pulsed Phase Thermography and Embedded Thermal Sensors.” PhD Diss., North Carolina

State University.

Pearson, Karl. 1901. “On Lines and Planes of Closest Fit to Systems of Points in Space.”

Philosophical Magazine Series 6 2 (11): 559–72.

Rajic, N. 2002. “Principal Component Thermography for Flaw Contrast Enhancement and Flaw

Depth Characterisation in Composite Structures.” Composite Structures 58 (4): 521–28.

Ravichandran, A. 2015. “Spatial and Temporal Modulation of Heat Source Using Light Modulator

for Advanced Thermography.” M.Sc., Missouri University of Science and Technology.

Remondino, Fabio, and Sabry El-Hakim. 2006. “Image-Based 3D Modelling: A Review.” The

Photogrammetric Record 21 (115): 269–91. doi:10.1111/j.1477-9730.2006.00383.x.

Rohrbaugh, NB. 2015. “A New Technique for Modeling the Geomorphology of a Slow Moving,

Soft-Slope Landslide Using Terrestrial LiDAR.” M.Sc. Missouri University of Science and

Technology.

Rumsey, MA, and JA Paquette. 2008. “Structural Health Monitoring of Wind Turbine Blades.” In

Proc. SPIE 6933: 69330E.

120

S. M. Habali and I. A. Saleh. 2000. “Local Design, Testing and Manufacturing of Small Mixed

Airfoil Wind Turbine Blades of Glass Fiber Reinforced Plastics: Part II: Manufacturing of

the Blade and Rotor.” Energy Conversion and Management 41 (3). Pergamon: 281–98.

Sareen, Agrim, Chinmay A. Sapre, and Michael S. Selig. 2012. “Effects of Leading-Edge

Protection Tape on Wind Turbine Blade Performance.” Wind Engineering 36 (5): 525–34.

Sareen, Agrim, Chinmay A Sapre, and Michael S Selig. 2014. “Effects of Leading Edge Erosion

on Wind Turbine Blade Performance.” Wind Energy 17 (10): 1531–42.

Savio, E, L De Chiffre, and R Schmitt. 2007. “Metrology of Freeform Shaped Parts.” CIRP

Annals-Manufacturing Technology 56 (2): 810–35.

Schroeder, K, W Ecke, J Apitz, and E Lembke. 2006. “A Fibre Bragg Grating Sensor System

Monitors Operational Load in a Wind Turbine Rotor Blade.” Measurement Science and

Technology 17 (5): 1167.

Schubel, PJ, RJ Crossley, and EKG Boateng. 2013. “Review of Structural Health and Cure

Monitoring Techniques for Large Wind Turbine Blades.” Renewable Energy 51: 113–23.

Shepard, S. M. 2003. “Reconstruction and Enhancement of Active Thermographic Image

Sequences.” Optical Engineering 42 (5): 1337.

Shepard, S. M, T Ahmed, and B Rubadeux. 2001. “Synthetic Processing of Pulsed Thermographic

Data for Inspection of Turbine Components.” Insight 43 (9): 587–89.

Shin, PH. 2013. “Non-Destructive Inspection in Adhesively Bonded Joints Using Pulsed Phase

Thermography.” M.Sc., North Carolina State University.

Somers, DM. 2004. “S814 and S815 Airfoils: October 1991--July 1992.” National Renewable

Energy Lab., Golden, CO (US).

Soua, S, P Van Lieshout, A Perera, TH Gan, and B Bridge. 2013. “Determination of the Combined

121

Vibrational and Acoustic Emission Signature of a Wind Turbine Gearbox and Generator

Shaft in Service as a Pre-Requisite for Effective Condition Monitoring.” Renewable Energy

51: 175–81.

Summers, A, Q Wang, N Brady, and R Holden. 2016. “Investigating the Measurement of Offshore

Wind Turbine Blades Using Coherent Laser Radar.” Robotics and Computer-Integrated

Manufacturing 41: 43–52.

Talbot, J, Q Wang, N Brady, and R Holden. 2016. “Offshore Wind Turbine Blades Measurement

Using Coherent Laser Radar.” Measurement 79: 53–65.

Tao, Ning, Zhi Zeng, Lichun Feng, Xiaoli Li, Yeshu Li, and Cunlin Zhang. 2011. “The Application

of Pulsed Thermography in the Inspection of Wind Turbine Blades.” In International

Symposium on Photoelectronic Detection and Imaging 2011, 819319–819319.

Tchakoua, P, R Wamkeue, and M Ouhrouche. 2014. “Wind Turbine Condition Monitoring: State-

of-the-Art Review, New Trends, and Future Challenges.” Energies 7 (4): 2595–2630.

Tchakoua, P, R Wamkeue, T. A Tameghe, and G Ekemb. 2013. “A Review of Concepts and

Methods for Wind Turbines Condition Monitoring.” Computer and Information Technology

(WCCIT), 2013 World Congress On. IEEE, 1–9.

Timmer, WA, and R Van Rooij. 2003. “Summary of the Delft University Wind Turbine Dedicated

Airfoils.” TRANSACTIONS-AMERICAN SOCIETY OF MECHANICAL ENGINEERS

JOURNAL OF SOLAR Energy ENGINEERING 125 (4): 488–96.

Tippmann, JD. 2014. Applications of Passive Vibration Control and Damage Detection in

Composite Wind Turbine Blades. PhD diss., University of California, San Diego.

Verbruggen, TW. 2003. “Wind Turbine Operation & Maintenance Based on Condition Monitoring

WT-Ω.” Final Report.

122

Walford, CA. 2006. “Wind Turbine Reliability: Understanding and Minimizing Wind Turbine

Operation and Maintenance Costs.” Report. SAND2006-1100. Sandia National Laboratories.

White, J. R., D. E. Adams, and M. A. Rumsey. 2011. “Modal Analysis of CX-100 Rotor Blade

and Micon 65/13 Wind Turbine.” In Structural Dynamics and Renewable Energy, 15–27.

Worzewski, T, and R Krankenhagen. 2016. “Thermographic Inspection of Wind Turbine Rotor

Blade Segment Utilizing Natural Conditions as Excitation Source, Part II: The Effect of

Climatic Conditions on.” Infrared Physics & Technology 76: 767–76.

Worzewski, T, R Krankenhagen, and M Doroshtnasir. 2016. “Thermographic Inspection of a Wind

Turbine Rotor Blade Segment Utilizing Natural Conditions as Excitation Source, Part I: Solar

Excitation for Detecting Deep.” Infrared Physics & Technology 76: 756–66.

Xia, XG. 1998. “A Quantitative Analysis of SNR in the Short-Time Fourier Transform Domain

for Multicomponent Signals.” IEEE Transactions on Signal Processing 46 (1): 200–203.

Xu, J, N Xi, C Zhang, Q Shi, and J Gregory. 2011. “Real-Time 3D Shape Inspection System of

Automotive Parts Based on Structured Light Pattern.” Optics & Laser Technology 43 (1): 1–

8.

Zhang, H, MJ Schulz, A Naser, and F Ferguson. 1999. “Structural Health Monitoring Using

Transmittance Functions.” Mechanical Systems and Signal Processing 13 (5): 765–87.

Zhao, S bin, C Zhang, and N Wu. 2009. “Infrared Thermal Wave Nondestructive Testing for Rotor

Blades in Wind Turbine Generators Non-Destructive Evaluation and Damage Monitoring.”

International Symposium on Photoelectronic Detection and Imaging 2009.

123

Appendix A: Summary of instruments specifications

In this section, the specifications of all instruments for various experiments including 3D

reconstruction using laser scanning technique, passive and active thermography are provided.

A.1 Laser Scanner

HandyScan laser scanner as a portable laser scanner with a high accuracy was used in this

research to reconstruct the surface of the blade. The environmental conditions do not affect the

accuracy and performance of this scanner. The calibration process of this device takes about 2

minutes (“https://www.creaform3d.com” 2018). The technical specification of this devise is

summarized in Table A.1. A picture of the Creaform HandyScan laser scanner is shown in Figure

A.1 (a).

Table A.1. Technical specifications of Creaform HandyScan laser scanning

(“https://www.creaform3d.com” 2018).

Accuracy Up to 0.030 mm (0.0012 in.)

Volumetric accuracy 0.020 mm + 0.060 mm/m (0.0008 in. + 0.0007 in./ft)

Resolution 0.050 mm (0.0020 in.)

Measurement rate 480,000 measurements/s

Scanning area 275 x 250 mm (10.8 x 9.8 in.)

Stand-off distance 300 mm (11.8 in.)

Part size range (recommended) 0.1 – 4 m (0.3 – 13 ft)

Software VXelements

Output format .dae, .fbx, .ma, .obj, .ply, .stl, .txt, .wrl, .x3d, .x3dz, .zpr

Laser class 2M (eye-safe)

Weight 0.85 kg (1.9 lbs.)

Dimensions 77 x 122 x 294 mm (3.0 x 4.8 x 11.6 in.)

Operating temperature range 5-40 0C (41-104 0F)

124

A.2 IR Camera

T1030Sc, which is a high-resolution research and science infrared camera of FLIR systems,

was used to capture thermal images. This camera is portable, uncooled, long-wave infrared

(LWIR) and very sensitive to temperature variation with HD detector (“Http://www.flir.ca/home/”

2018). The camera was supported by a tripod, and an image of this camera is shown in Figure A.1.

(b). Some of specifications of this camera are listed in Table A.2.

Table A.2. Specifications of FLIR T1030Sc infrared Camera (“Http://www.flir.ca/home/” 2018).

IR Sensor resolution 1024 x 768 pixels

Thermal Sensitivity/NETD < 20 mK at +30°C (+86°F)

Lens Choices 12°, 28°, 45°, 50 µm Close-up

Field of View (FOV) 28° × 21°

Minimum Focus Distance 0.4 m (1.32 ft.)

Spatial Resolution/IFOV 0.47 mrad (standard lens)

Detector Type Focal Plane Array (FPA), uncooled

microbolometer

Image Frequency 30 Hz

Spectral Range 7.5 - 14 µm

Detector Pitch 17 μm

Measurement

Object Temperature Range –40°C to +150°C (–40°F to +302°F)

0 to +650°C (+32°F to +1202°F)

+300°C to +2000°C (+572°F to +3632°F)

Accuracy

±1°C (±1.8°F) or ±1% at 25°C for

temperatures between 5°C to 150°C.

Storage of Media

Storage Media Removable SD card (Class 10)

Image Storage Standard JPEG, including digital photo and

measurement data

Time Lapse 15 seconds to 24 hours

Video Recording/Streaming

Time Constant < 10 ms

125

(a)

(b)

Figure A.1. (a) HandyScan laser scanner (“https://www.creaform3d.com” 2018), (b) T1030 SC

IR camera (“Http://www.flir.ca/home/” 2018).

A.3 Power Supply

A Speedotron 4803CX LV power supply was used to provide the required power of a 206VF

flash lamp for heating the surface. This pack has a maximum power of 4800 W with 270 power

options. The specifications of this power supply are listed in Table A.3. The power pack used in

the experiment is depicted in Figure A.2.

Table A.3. Specifications of Speedotron 4803CX power supply (“Http://www.speedotron.com/”

2018).

Maximum Power 4800 Watt-secs

Recycle Time 4.0 seconds

Power Channels 3

Ratio Combinations 27

Total Output Variations 270

Special Features Ratio, Dial-down Power, Optional Remote

Size 9" x 14" x 14"

126

Figure A.2. Speedotron 4803CX LV power supply.

A.4 Flash Lamp

A single Speedotron 206VF light unit was used in the pulsed thermography experiment to

heat the surface of specimen. This light unit works in conjunction with a 4803CX LV power supply

where it can emit a power of 4800 W if it is connected to its special outlet. The light unit contains

a focusing reflector mount which provides the capability of controlling the flash light direction

and coverage. The employed 206VF flash lamp is illustrated in Figure A.3.

A.5 Halogen Lamps

Portable double-quartz halogen lights were used in the step heating thermography to apply

long pulse heat to the surface of the object. Two lamps were mounted on an adjustable stand. Each

lamp contained one single halogen bulb with the power of 500W. Also, aluminum fixtures were

embedded inside the housings to enhance the emissivity of emitted light. A typical picture of

halogen lamps used in this experiment is shown in Figure A.4.

127

Figure A.3. Speedotron 206VF strobes.

Figure A.4. Portable halogen lamps with the power of each 500W.

128

A.6 Computer System and Software

A Dell computer desktop with an Intel(R) Core(TM) i7-6700K CPU and 32 GB RAM was

used to record thermal data. In this regard, the ResearchIR software was installed on the desktop

while the camera was directly connected to it through a USB cable. The ResearchIR is a powerful

software that can easily acquire and analyze thermal data. Each thermal image or frame of data

can be easily exported into different formats such as JPG, CSV, 32bit TIFF for further analysis

and image enhancement through other software such as MATLAB (“Http://www.flir.ca/home/”

2018).

129

Appendix B: Copyright Permissions

Copyright permission of Figure 2.1.

130

Copyright permission of Figures 2.2 and 2.3.

131


132


133


Figure 3.1. copyright permission

condition monitoring of wind turbine blades

Documents