NON-DESTRUCTIVE TESTING AND EVALUATION TECHNIQUES ? My Past, Present, and Future Research

1
NON-DESTRUCTIVE TESTING AND EVALUATION
TECHNIQUES ? My Past, Present, and Future
Research

• Dr. Guang-Ming Zhang
• General Engineering Research Institute, Liverpool
  JMU
• October 10, 2008

2
Outline of Research Experience

• April 2003 - present
• Research Fellow in General
  Engineering Research Institute, Liverpool John
  Moores University, Liverpool, UK
• March 2001 March 2003
• Researcher in Signals and Systems
  Group, Uppsala University, Uppsala, Sweden
• Sep. 1999 March 2001
• Postdoc in Institute of Acoustics,
  Nanjing University, Nanjing, P.R.China.
  (Promoted as an Associate Professor in Nov. 2000)
• Sep. 1996 June 1999
• Ph.D in Laser and Infrared Research
  Institute, Xian Jiaotong University, Xian,
  P.R.China, Full time
• Sep. 1993 June 1996
• MSc in Laser and Infrared Research
  Institute, Xian Jiaotong University, Xian,
  P.R.China, Full time

3
Development of JTUIS Ultrasonic Imaging Testing
System
MSc Research supported by industry. Xian
Jiaotong University

• My main contributions include initiating
  the system plan and structure, development of NDT
  techniques, designing a PC based data acquisition
  and control circuit board and software
  programming.
• This product has been patented and
  commercialized. It has been sold in more than
  eighteen companies including big state-owned
  companies in China and multi-national companies
  for example ABB, and companies in Indian.

4
Development of JTUIS Ultrasonic Imaging Testing
System
5
Applications
Low Voltage Switch
High Voltage Vacuum Switch
High Voltage Switch Sintered Contact
Ultrasonic inspection
6
Theory and Application Study on Time-Frequency
Analysis Technology for Ultrasonic NDT
PhD Research. Xian Jiaotong University

• Integrating neural networks techniques for
  recognition and classification of ultrasonic
  signals and defects.
• Continuous wavelet transform and parameter
  optimization in ultrasonic non-destructive
  evaluation.
• Adaptive time-frequency decomposition
• Flaw identification and noise suppressing using
  Wavelet Ridge.
• 2D wavelet packet transform for image
  enhancement.
• Ultrasonic non-destructive evaluation of weld
  defects.

7
Infrared Thermography
Xian Jiaotong University

• Developed an infrared image collecting and
  processing system. It was designed for AGEMA
  series thermovision. This product has been
  commercialized and was awarded a science and
  technology progress prize by National Education
  Ministry of P.R.China. It has been used in
  universities and petrochemical companies
• Investigated pulse-heating infrared thermography.
  High power ultrasound was used as the heating
  source.
• Carried out composite material non-destructive
  testing using the infrared thermal imaging
  technique.

8
Acoustical Sensors and Actuators Based on MEMS
Techniques
Nanjing University

• Supported by the Natural Science Foundation of
  China (equivalent to EPSRC).
• Developed a surface acoustic wave rotation motor
  (size at millimetrelevel).
• A prototype MEMS (Micro Electro-Mechanical
  System) motor has been designed and made using
  photolithography, and its basic characteristics
  have been studied by both experiments and
  computer simulation.
• Fabrication and characterisation of ferroelectric
  thin film using magnetron sputtering.

9
Acoustical Sensors and Actuators Based on MEMS
Techniques
Schematic diagram of the miniaturized SAW rotary
motor
Time evolution of the angular velocity both in
theoretical simulation and in experiment
The stator is made by a 128 rotated y-cut
x-propagation LiNbO3 substrate, on which two
pairs of IDTs (Inter-digital transducer) are
arranged in parallel in the direction of
propagation. The rotor is composed of an
aluminium disk with several steel balls around
the circumference. The size of the stator is
17110.4 mm3, and the size of the rotor is
?90.7 mm3. With the operating frequency of 30
MHz and the driving voltage of 120 Vpp, the
motor can rotate at a speed of 270 rpm.

10
Ultrasonic Image Compression

• Part of the European 5th Framework
  Programme project of Signal processing and
  improved qualification for non-destructive
  testing of aging reactors . Uppsala University

• Investigated transform and subband coding for
  ultrasonic Image compression.

0.125 bpp
Original signal
JPEG
Wavelet transform (JPEG200)
A MATLAB toolbox
Karhunen-Loève transform
These RF signals were extracted from original and
decompressed ultrasonic B-scan images
11
Ultrasonic Image Compression

• Carried out research on simultaneous denoising
  and compression of ultrasonic images using vector
  quantization of image subbands.

The general VQ-based coding scheme
The proposed coding scheme
0.47bpp
12
Inspection of Copper Canisters for Spent Nuclear
Fuel
Supported by SKB (the Swedish Nuclear Fuel and
Waste Management Co). Uppsala University

• Studied nonlinear acoustical imaging techniques.
  Phased array Allin ultrasonic imaging system, and
  RITEC advanced ultrasonic measurement system
  RAM-5000 were used to carried out experiments. A
  precision network analyzer from Agilent was used
  to analyze transducer characteristics.
• Performed modelling and computer simulation of
  nonlinear acoustical fields in immersed solids.
• Designed a software called DREAM for acoustic
  field simulation using discrete representation
  array modelling

13
Nonlinear acoustical imaging for NDE
A C-scan image of the electron beam welding test
piece obtained from the fundamental wave
The second harmonic image
14
Calibration
An ultrasonic pulse measured by a hydrophone at
31mm away from the transducer in water, and its
spectrum. It can be seen that nonlinearity in
water can be neglected in our experiment.
An ultrasonic pulse reflected from a side-drilled
hole and its spectrum. Nolinearity is clearly
observed.
15
Modelling and computer simulation of nonlinear
acoustical fields in immersed solids
Acoustical field from the 2.3-MHz spherically
annular transducer operating in copper block
immersed in water with a 30-mm water path length,
with an initial source intensity 21W/cm2. Left
without accounting for the nonlinearity in water
Right with accounting for the nonlinearity in
water. Normalized second harmonic at the welded
interface 30.8854dB, 19.6513dB.
16
Acoustic field simulation using discrete
representation array modelling
http//www.signal.uu.se/Toolbox/dream/ (Free
Download)
17
Quality and reliability testing of BGA and
Flip-Chip solder bonds subjected to
mixed-environments testing using AMI
EPSRC ROPA project. Liverpool JMU

• Carried out a comparison study of X-ray
  inspection and acoustic micro imaging for the
  evaluation of modern microelectronic packages.
• AMI is an effective approach for detecting
  gap-type defects such as voids, delaminations and
  cracks due to the strong reflection of ultrasound
  in solid-air interfaces. These defects are
  difficult to be found by X-ray inspection owing
  to low contrast. The contrast of X-ray images
  relies on the thickness of internal
  structures/defects, and differences in the atomic
  mass and density. Thus, X-ray inspection is fit
  for volumetric defects for example broken wires.
  However, this kind of defect is very hard or
  impossible to be detected by AMI.
• Was assessed as tending to outstanding

18
Quality and reliability testing of BGA and
Flip-Chip solder bonds subjected to
mixed-environments testing using AMI
EPSRC ROPA project. Liverpool JMU
AMI
X-ray inspection
X-ray inspection
19
Signal model in AMI
(a)
(b)
AMI echoes at typical boundaries in Flip-Chip
package mounted on ceramic substrate (a), and an
example A-scan obtained using a 230MHz transducer
(b).
Ultrasonic signals reflected by defects possess
information about defect size and orientation.
20
Imaging modes of ultrasonic imaging
(a)
A-scan

C-scan
B-scan
21


Multiple slices of flip-chip solder joints by
AMI. From left to right top slice (silicon-bump
interface), middle slice (bump area), and bottom
slice (bump-PCB). Lower part corresponding
electronic gate

22
Imaging technology for x-ray inspection

• Conventional 2D transmission x-ray
• Oblique angled viewing
• 3D digital tomography
• 3D laminography

Principle of digital lexicography
23
Advanced Acoustic Micro-Imaging for the
Evaluation of Microelectronic Packages
Liverpool JMU

• Background
• Modern PCBs and semiconductor devices are
  difficult to inspect
• This will get harder as devices become smaller
  and more complex
• 3D packages as shown in the above picture are
  emerging, bringing new problems that need
  solutions
• The key acoustic challenges are axial resolution
  for delamination and cracks at closely-spaced
  interfaces and penetration through multiple
  interfaces.

24
Limitations of Conventional AMI Techniques
Left by a 230 MHz transducer Right by a 50 MHZ
transducer.
(Dimension of the flip-chip package is
8.19mm?8.30mm?0.86mm)
25
Advanced AMI Techniques

• Basic Principle
• Firstly, performing a-priori selection of a
  possibly over-complete signal dictionary (a
  collection of parameterised waveforms) in which
  the ultrasonic pulses are assumed to be sparsely
  representable.
• Secondly, separating the incident pulses by
  exploiting their sparse representability.
• Thirdly, selecting an appropriate echo and
  producing a C-scan output.

26
Time-Frequency Domain AMI Techniques
Acoustic time-frequency domain imaging. (a)
Time-frequency representations of Fig.2a by CWT
(b) Wavelet transform modulus maxima (c)
Significant local maxima obtained by wavelet
thresholding.
27
Sparse Signal Representation Based AMI Approaches
What is sparse signal representation?
The sparse signal representation problem is
formulated as
Given a signal
,
and the overcomplete dictionary
seek the sparsest coefficient vector
.
under the linear model
This corresponds to solving the following
variational problem
Minimize
subject to
It is an NP-hard problem. So many approximation
solutions were proposed.
28
Sparse Signal Representation Based AMI Approaches

• What are the advantages of sparse signal
  representation?
• Recently sparse overcomplete representation is of
  great interest in many applications such as image
  compression, denoising of signals, and blind
  source separation because of its advantages.
• One is that there is greater flexibility in
  capturing structure in the data. Instead of a
  small set of general basis vectors, there is a
  larger set of more specialized basis vectors such
  that relatively few are required to represent any
  particular signal. These can form more compact
  representations, because each basis vector can
  describe a significant amount of structure in the
  data.
• The second is super-resolution. We can obtain
  a resolution of sparse objects that is much
  higher than that possible with traditional
  methods.
• An additional advantage is that overcomplete
  representations increase stability of the
  representation in response to small perturbations
  of the signal.
• In addition, the redundant representations
  have the desired shift invariance property.

29
Sparse Signal Representation Based AMI Approaches

• Sparse signal representation methods
• Basis Pursuit (BP), Matching Pursuit (MP),
  and Best Orthogonal Basis (BOB) (Wavelet Packet
  decomposition).
• 2) Over-complete dictionaries
• Wavelet packets dictionaries, Cosine packet
  dictionaries, and Gabor dictionaries.
• 3) SSRAMI
• BP base AMI, MP based AMI, and BOB based AMI.

30
Sparse Signal Representation Based AMI Approaches
The echo separation result of an A-scan shown in
phase plane and a succession of time-frequency
windows. The darkness of the time-frequency image
increases with the energy value, and each
time-frequency atom selected by the BP method is
represented by a Heisenberg box.
31
Sparse Signal Representation Based AMI Approaches
The AMI results with simulated A-scans by
different AMI techniques.
Parameters Parameters Parameters Parameters Aerror () Aerror () Aerror ()
A1/A2 v2-v1 u2-u1 s2-s1 SSRAMI TDAMI FDAMI
1.00 -0.08 48 0 4.42 0.00 47.41
1.00 0.10 48 0 4.39 0.00 55.04
1.00 0.15 36 0 1.83 0.42 11.63
1/0.8 0.2 32 0 0.83 0.44 4.58
1/0.8 -0.3 28 5 1.92 13.38 0.00
1/1.5 -0.3 28 10 1.16 42.85 0.00
1/1.5 -0.3 20 15 3.32 60.60 0.00
32
Learning Overcomplete Representation Based AMI
Approaches

• The success of SSRAMI greatly depends on the
  overcomplete dictionary. There are several ways
  to determine the overcomplete dictionary
• To choose a non-adapted overcomplete dictionary
  from the existing dictionaries, such as Gabor
  dictionary, and wavelet packet dictionaries.
• To learn an adapted overcomplete dictionary.
• Dictionary learning methods Principal
  Component Analysis (PCA), Independent Component
  Analysis (ICA), overcomplete ICA, and
  FOCUSS-based dictionary learning, and recently
  developed K-SVD.
• 3. To combine dictionaries to make bigger, more
  expressive dictionaries.

33
Learned Overcomplete Dictionaries
Learned basis vectors. Left 50MHz transducer
Right 230MHz transducer.
34
Learning Overcomplete Representations
Learning overcomplete representation techniques
Basis Pursuit (BP), Matching Pursuit (MP),
Overcomplete ICA, FOCUSS, and Sparse Bayesian
Learning (SBL).
Sparse representations of ultrasonic signals (a)
A simulated A-scan the sparse representations of
(a) by the overcomplete ICA (b), FOCUSS (c) and
SBL (d) algorithms (e) the original echo the
first echo in (a) (f) the recovered echo from
(b) (g) the recovered echo from (c) (h) the
recovered echo from (d).
35
Sparse Deconvolution

• Investigated sparse deconvolution of ultrasonic
  NDE Traces accounting for pulse variances by
  learning overcomplete representations.

Deconvolution of (a) a synthetic non-stationary
trace segment (SNR8.6dB). (b) The true
reflectivity (b). The reflectivity recovered by
(c) the proposed method and (d) the IWM-BD method.
Upper true and estimated local pulses for the
first and middle echoes by the proposed method.
Solid lines devote the true pulses and dashed
lines devote the estimated pulses. Lower by the
IWM-BD method.
36
Ultrasonic Flaw Detection via Overcomplete and
Sparse Representations
1. Signal detection for signals with additive
white Gaussian noise
The proposed algorithm
WTSP
37
Ultrasonic Flaw Detection via Overcomplete and
Sparse Representations
2. Signal detection for signals with correlated
noise
Correlated noise and an original ultrasonic flaw
signal
The proposed algorithm
WTSP
38
Ultrasonic Flaw Detection via Overcomplete and
Sparse Representations
3. Estimate the pulse arrival time for defect
location, thickness measurement, and sound
velocity determination.
SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB) SNRin(dB)
15.62 6.97 1.54 0.22 -0.74 -1.61 -2.58 -3.23 -4.14 -4.72 -5.79
Error of time delay (in sample number) The proposed 0 0 0 0 0 0 1 0 1 1 2
Error of time delay (in sample number) WTSP 0 0 0 0 1 1 1 1 1 1 2
The error of measure delay time between two
pulses against the SNR. Results are presented in
sample number.
39
Ongoing EPSRC Project
Super-Resolution Acoustic Time-Frequency Domain
Imaging Techniques

• 1.Simulation of Acoustic Microimaging
• To elucidate the defect detection mechanism and
  predict ultrasonic pulses.
• To investigate the basic ultrasound limitation
  penetration vs. resolution in modern 3D packages.
• To optimize the parameters of transducer
  focusing.
• To study the ultrasound nonlinearity.
• 2.Super-resolution Acoustic Microimaging
  Approaches
• To achieve a sub-wavelength axial resolution.
• To study sparse representation based AMI
  approaches.
• To develop fast time-frequency domain AMI
  approaches.
• To develop deconvolution-based AMI
• To study 3D AMI

40
Other Related Work

• Worked on development of INTRANET network
  software under the IBM AS-400 computer operating
  system in Primax Manufacturing LTD, Guangdong,
  P.R.China in June-September, 1996.
• Involved in research of photoacoustic imaging and
  laser ultrasonics techniques at Nanjing
  University.
• Worked on the TELEPATH an EU funded
  e-learning project involved in partners from UK,
  Portugal and China at EDC.
• Provided consulting help to local (Merseyside)
  companies under the EU ERDF funding at EDC.
• Co-supervised postgraduate students and Ph.D
  students.

41
GRANTS AND PROPOSALS

• Super-resolution acoustic time-frequency domain
  imaging techniques for the evaluation of modern
  microelectronic packages. EPSRC responsive mode,
  Granted (200,000), 2008 July (Researcher Co-I).
• Micro-electrical-mechanical ultrasound phased
  array system. Supported by Postdoctoral Science
  Foundation of China, 2000 (PI).
• Ferroelectric thin film manufacturing and study
  on structure and performance. Supported by State
  Key Lab. Science Foundation of Nanjing
  University, 2000 (PI).
• Nanoscale tomographic imaging of buried
  structures. ERC Starting Independent Researcher
  Grant, not granted, April 2007 (PI) (Very good
  comments were received).
• Characterisation of future packages using
  advanced imaging techniques. EPSRC responsive
  mode, missed by one positions, March 2007 (Co-I).
  
• Advanced imaging approaches for diagnosis of
  electronic assemblies. EPSRC Advanced Research
  Fellowship, I was invited for interview and the
  proposal was ranked 11 in the final stage but
  unfortunately only 9 have been granted, Sep. 2006
  (PI).

42
Summary

• 1) I have developed significant expertise in
  a broad research area by working in different
  universities in different countries
• Ultrasonic non-destructive evaluation, ultrasonic
  engineering, signal and image processing, pattern
  recognition, information theory, instrumentation
  development and electronic design.
• Xian Jiaotong University - Nanjing University -
  Uppsala University LJMU
• China Sweden UK
• 2) As main researcher I have been involved in
  multiple research projects supported by Natural
  Science Foundation of China, European Framework
  Programme, Sweden, EPSRC, and industries.

43
Summary

• 3) Have gained experience in project management
  As principal investigator managed two projects.
  Co-managed 3 projects. Co-supervised several
  postgraduate students and PhD students.
• 4) Strong research and product development
  abilities
• Published 40 refereed journal papers (25 as first
  author), and 7 conference papers (6 as first
  author).
• 1 patent.
• 2 products commercialised.
• A number of awards have been awarded by Chinese
  government and other organisations.
• Two invited presentations. (Invited by the vice
  Chancellor of Xian Science and Technology
  University, and by the head of Institute of
  Acoustics, Nanjing University, in 2005.)

44
Thank You for Your Attention