You are on page 1of 190

Advanced NDE

OVERVIEW
Krishnan Balasubramaniam

MOTIVATION

COURSE OUTLINE

Week 1

Relevance of NDT in industries, Basics methods of NDT,


Statistics and Reliability for NDT

Week 2

Digital Signals and Data Analysis for NDT.

Week 3

Basics of acoustic wave modes, Wave equation in acoustic


and elastic media, Christoffel equations, reflection,
refraction, mode conversion, diffraction and scattering of
ultrasonic waves in isotropic and anisotropic media.

COURSE OUTLINE

Week 4

Methods of contact and non-contact generation and


reception of ultrasound.
Recent techniques in ultrasonic testing like Guided Waves
with applications.

Week 5

Time of Flight Diffraction, Synthetic Aperture Focusing


Technique, Phased and Sampled Array Ultrasound,.
PROJECT TITLES TO BE DECIDED AND GROUPS FORMED

Week 6

Basics of Electromagnetic NDT, Maxwells Equations,


Derivation of flux potential for eddy current inspection
technique, skin depth effects, lift off effects and
compensation,.

COURSE OUTLINE

Week 7

Low frequency and Remote Field Eddy Current


Techniques, Pulsed Eddy Current Technique, Microwave
field equations, reflection, refraction, mode conversion,
diffraction and scattering of electromagnetic waves

Week 8

Microwave instrumentation, applications of Eddy


Current and Microwave NDT.
MID SEMESTER EXAM

Week 9

Basics of Thermal Imaging, active and passive methods,


diffusion of thermal waves in isotropic and anisotropic
materials. Infra-red detectors and systems.
Applications.

COURSE OUTLINE

Week 10

Fundamentals of Radiograpy, Types of


sources and detectors

Week 11

Tomography and Laminography Algorithms


for slice and volume reconstruction.
Automatic Defect Recognition algorithms.

Week 12

Optical Methods of NDT, Hybrid techniques


in NDT, Sono-therm, Eddy-therm. NDE
sensors for process monitoring.

COURSE OUTLINE

Week 13

Application of NDT Methods


to Nuclear Power Plants

Week 14

PROJECT PRESENTATIONS.

Week 15

PROJECT PRESENTATIONS

ADVANCED NDE
WEEK 1
Relevance of NDT in industries, Basic methods of NDE
INSTRUCTOR: Krishnan Balasubramaniam

CAN WE DO THIS???....Always??

Non-Destructive Evaluation (NDE)


Non-destructive means of making
measurements that has relevance to the
material state and/or process parameters.
Without decreasing the performance
capability.
Safety

Reliability

Productivity

When are NDE Methods Used?


DESIGN, PRODUCTION, DEPLOYMENT, INSERVICE

To assist in product development


To screen or sort incoming materials
To monitor, improve or control
manufacturing processes
To verify proper processing such as heat
treating
To verify proper assembly
To inspect for in-service damage

Expanding Role of NDE

What are Some Uses


of NDE Methods?
Flaw Detection and Evaluation
Leak Detection
Location Determination

Dimensional Measurements

Fluorescent penetrant indication

Structure and Microstructure Characterization


Estimation of Mechanical and Physical Properties
Stress (Strain) and Dynamic Response Measurements
Material Sorting and Chemical Composition
Determination

What

Flaws
Material Properties
Process Properties
Structural Properties

&

WHY?
Improve Safety
Increase Performance
Residual Life
Assessment
Enhance Productivity
Re-engineering
Security Information

NDE for Flaws

Cracks
Voids
Corrosion
Delamination
Disbonds
Material Variations

Detect
Locate
Characterize
Size
Evaluate Criticality.

Quantitative and Verifiable Information

Nondestructive Evaluation
Excitation
Source
Input Transducer

Test
Specimen
Output Transducer

Signal
Conditioning

Inverse
Model

Methods of NDT
Visual

Visual Inspection
Most basic and common
inspection method.
Tools include
fiberscopes,
borescopes, magnifying
glasses and mirrors.

Portable video inspection


unit with zoom allows
inspection of large tanks
and vessels, railroad tank
cars, sewer lines.
Robotic crawlers permit
observation in hazardous or
tight areas, such as air
ducts, reactors, pipelines.

Liquid Penetrant Inspection


A liquid with high surface wetting characteristics
is applied to the surface of the part and allowed
time to seep into surface breaking defects.
The excess liquid is removed from the surface
of the part.
A developer (powder) is applied to pull the
trapped penetrant out the defect and spread it
on the surface where it can be seen.
Visual inspection is the final step in the
process. The penetrant used is often loaded
with a fluorescent dye and the inspection is
done under UV light to increase test
sensitivity.

Magnetic Particle Inspection


The part is magnetized. Finely milled iron particles coated with a dye
pigment are then applied to the specimen. These particles are
attracted to magnetic flux leakage fields and will cluster to form an
indication directly over the discontinuity. This indication can be
visually detected under proper lighting conditions.

Magnetic Particle Crack Indications

Radiography
The radiation used in radiography
testing is a higher energy (shorter
wavelength) version of the
electromagnetic waves that we
see as visible light. The radiation can
come from an X-ray generator or a
radioactive source.

High Electrical Potential


Electrons
+

X-ray Generator
or Radioactive
Source Creates
Radiation

Radiation
Penetrate
the Sample
Exposure Recording Device

Eddy Current Testing


Coil

Coil's
magnetic field

Eddy current's
magnetic field
Eddy
currents
Conductive
material

Eddy Current Testing


Eddy current testing is particularly well suited for detecting surface
cracks but can also be used to make electrical conductivity and
coating thickness measurements. Here a small surface probe is
scanned over the part surface in an attempt to detect a crack.

Ultrasonic Imaging
High resolution images can be produced by plotting signal strength or
time-of-flight using a computer-controlled scanning system.

Gray scale image produced using


the sound reflected from the front
surface of the coin

Gray scale image produced using the


sound reflected from the back surface
of the coin (inspected from heads side)

Common Application of NDT


Inspection of Raw Products
Inspection Following
Secondary Processing
In-Services Damage Inspection

Inspection of Raw Products


Forgings,
Castings,
Extrusions,
etc.

Inspection Following
Secondary Processing

Machining
Welding
Grinding
Heat treating
Plating
etc.

Inspection For
In-Service Damage
Cracking
Corrosion
Erosion/Wear
Heat Damage
etc.

Power Plant Inspection


Periodically, power plants are
shutdown for inspection.
Inspectors feed eddy current
probes into heat exchanger
tubes to check for corrosion
damage.

Pipe with damage

Probe
Signals produced
by various
amounts of
corrosion
thinning.

Wire Rope Inspection


Electromagnetic devices and visual
inspections are used to find broken
wires and other damage to the wire
rope that is used in chairlifts, cranes
and other lifting devices.

Storage Tank Inspection


Robotic crawlers use
ultrasound to inspect the
walls of large above
ground tanks for signs of
thinning due to corrosion.

Cameras on long
articulating arms
are used to inspect
underground
storage tanks for
damage.

Aircraft Inspection
Nondestructive testing is used
extensively during the
manufacturing of aircraft.
NDT is also used to find cracks
and corrosion damage during
operation of the aircraft.
A fatigue crack that started at
the site of a lightning strike is
shown below.

Jet Engine Inspection


Aircraft engines are overhauled
after being in service for a period
of time.
They are completely disassembled,
cleaned, inspected and then
reassembled.
Fluorescent penetrant inspection
is used to check many of the parts
for cracking.

Crash of United Flight 232


Sioux City, Iowa, July 19, 1989
A defect that went
undetected in an
engine disk was
responsible for
the crash of
United Flight 232.

Pressure Vessel Inspection


The failure of a pressure vessel
can result in the rapid release of
a large amount of energy. To
protect against this dangerous
event, the tanks are inspected
using radiography and
ultrasonic testing.

Rail Inspection
Special cars are used to inspect
thousands of miles of rail to find
cracks that could lead to a
derailment.

Bridge Inspection
The US has 578,000
highway bridges.

Corrosion, cracking and


other damage can all
affect a bridges
performance.
The collapse of the Silver
Bridge in 1967 resulted in
loss of 47 lives.
Bridges get a visual
inspection about every 2
years.
Some bridges are fitted
with acoustic emission
sensors that listen for
sounds of cracks growing.

Pipeline Inspection
NDT is used to inspect pipelines
to prevent leaks that could
damage the environment. Visual
inspection, radiography and
electromagnetic testing are some
of the NDT methods used.

Remote visual inspection using


a robotic crawler.

Magnetic flux leakage inspection.


This device, known as a pig, is
placed in the pipeline and collects
data on the condition of the pipe as it
is pushed along by whatever is being
transported.

Radiography of weld joints.

Special Measurements
Boeing employees in Philadelphia were given the
privilege of evaluating the Liberty Bell for damage using
NDT techniques. Eddy current methods were used to
measure the electrical conductivity of the Bell's bronze
casing at a various points to evaluate its uniformity.

New World
Affordable
and
Reliable
Inspection

Rapid
Inspection

IMPROVED
FLAW
DETECTION,
LOCATION,
SIZING

FLAW
DETECTION,
LOCATION,
SIZING
NDE for
Manufacturing

Structural
Health
Monitoring

NDE for
Measurements

Advanced NDE
Week 1
Mechanics, Statistics, and Probability
Basics for NDE

MOTIVATION, RELEVANCE, BENEFITS


You Cant Inspect In Quality

Need to make it right in the beginning

NDE techniques can


Provide insight during material development cycle
Provide feedback with respect to process window

Transitions from Safe Life to Fail Safe/Damage Tolerance

USAF programs on Retirement for Cause and Engine Rotor Life


Extension
Flaw detection
Material characterization

New Business Models of Power-by-the-Hour

Enhanced OEM interest in life extension

REQUIRES INTIMATE UNDERSTANDING


OF MATERIALS SCIENCE
Four Elements of Materials Science and Engineering

THE FUNDAMENTAL CHALLENGE

NDE is sensitive to microstructural parameters

Microstructural parameters control properties and performance

But the combination of microstructural parameters sensed by a given NDE


technique may not be the same as is needed to uniquely predict the
material property of interest

PREDICTION OF REMAINING LIFE

Stress Levels
Material Properties
(Crack Growth Parameters,
Fracture Toughness)

Fracture
Mechanics

Flaw Size

Need better ways to


Detect and size defects
Determine material properties
Determine stress

Remaining Life

A FEW EXAMPLES OF FAILURE MODES OF


OTHER MATERIALS
Ceramics
Used as ballistic armor
Fracture must occur in a controlled way to protect
soldier
Need NDE of microstructure
Composites
Impact damage can produce hidden damage
Need NDE to detect delaminations
Improper curing can degrade mechanical properties
Need NDE providing chemical information

SOME PARAMETERS FOR WHICH


MEASUREMENT TECHNIQUES ARE NEEDED
Plastic Deformation

y , y
u , uts
f , % RA
n
r, r

Fracture
KIC
CVN

Fatigue
C
m
Remaining life

Creep
Creep strain parameters

Other Failure Modes

CAUTION

Relationships will not, in general, be unique

Combination of microstructural parameters sensed by a given NDE technique may


not be the same as is needed to uniquely predict the material properties of interest
Must develop applications

With a good understanding of mechanisms

With a specific application space in mind

Statistics and Probability


Definitions
Moments of Data
SD
Variance
CV

Distributions
Probability
Bayes Theorm

Statistics is...
a standard method for...
- collecting, organizing, summarizing,
presenting, and analyzing data
- drawing conclusions
- making decisions based upon the
analyses of these data.
used extensively by engineers (e.g.,
quality control)
53

Populations and Samples


Population - complete set of all of the
possible instances of a particular
object
e.g., the entire class

Sample - subset of the population


e.g., a team

We use samples to draw conclusions


about the parent population.
54

Why use samples?


The population may be large
all people on earth, all stars in the sky.

The population may be dangerous to observe

automobile wrecks, explosions, etc.


The population may be difficult to measure
subatomic particles.
Measurement may destroy sample
bolt strength
55

Measures of Central Tendency


If you wish to describe a population (or a
sample) with a single number, what do you
use?

Mean - the arithmetic average


Mode - most likely (most common) value.
Median - middle of the data set.

56

Sample Mean
1
x
n

x
i 1

Where:
x is the sample mean
xi are the data points
n is the sample size

57

Population Mean
1

x
i 1

Where:
is the population mean
xi are the data points
N is the total number of observations in the
population
58

What is the Mode?


mode - the value that occurs the most
often in discrete data (or data that
have been grouped into discrete
intervals)

Example, students in this class are


most likely to get a grade of B.
59

Mode continued
Example of a grade distribution with
mean C, mode B
25
20
15
10
5
0
F

60

What is the Median?


Median - for sorted data, the median is
the middle value (for an odd number of
points) or the average of the two middle
values (for an even number of points).
useful to characterize data sets
with a few extreme values that
would distort the mean (e.g., house
price,family incomes).
61

What Is the Range?


Range - the difference between the
lowest and highest values in the set.
Example, driving time to Houston is 2
hours +/- 15 minutes. Therefore...
Minimum = 105 min
Maximum = 135 minutes
Range = 30 minutes

62

Standard Deviation

Gives a unique and unbiased


estimate of the scatter in the
data.

63

Standard Deviation

Population

Sample

1
N

2
(
x

)
i

Variance = 2

i 1

Deviation
n
1
2
s
( xi x )

(n 1) i 1

Variance = s2

64

The Subtle Difference Between and

N versus n-1
n-1 is needed to get a better estimate of the
population from the sample s.

Note: for large n, the difference is trivial.

65

A Valuable Tool
Gauss invented standard deviation circa 1700
to explain the error observed in measured star
positions.
Today it is used in everything from quality
control to measuring financial risk.

66

Probability Density Function


p=f(x) frequency function or
probability density function (PDF)
Sum of two numbers
obtained when two
dice are tossed

Frequency Distribution
67

Cumulative Frequency Distribution

F (x j )

f (x )

x j xi

F ( x) f(x)dx
-

f(x)dx

dF ( x)
f ( x)
dx
68

For
continuous
random
variable

N
1
sx
( xi x ) 2

( N 1) i 1

Note

i
N
( x 2 ) Nx 2 i 1

N
i 1
N

N
1
2
2
sx
(
x

2
x
x

x
)

i
i
( N 1) i 1

i
N
N
(2 xi x ) 2 x xi 2 Nx 2 2 i 1

N
i 1
i 1
N

i
n
2
x
i 1

i
N
s x i 1

N 1
N

Nx

i 1

N 1

69

Easier form to use


with calculators

Distribution of tensile properties of hot-rolled


UNS G10350 steel

Discrete Frequency Histogram with


occurrences or class frequencies fi
70

k=21, w=1 kpsi

1
x
N

( f x )
i 1

sx

i 1

f i xi
2

f i xi i 1
N

N 1

i 1

f i xi Nx 2
2

N 1

Cumulative Density Function is


wi represents the class
width at xi

f i wi i 1
Fi
( f j wj )
2
j 1
71

Stochastic variable and variate are also used to mean


a random variable
Deterministic quantity is something that has a single
specific value
Stochastic variable can be described by the standard
deviation and the mean or by the coefficient of
variation Cx
The variate x can be expressed as follows

Sx
Cx
x
x X ( x , S x ) x X (1, C x )
72

Example 1
Five tons of 2-in round rod of 1030 hot-rolled steel
has been received for work piece stock. Nine
standard-geometry specimens have been
machined from random locations in various rods.
In the report, the ultimate tensile strength was
given in kpsi.
Find the mean, the standard deviation, and the
coefficient from the sample, such that these are
best estimates of the parent population (the stock
your plant will convert to product)
73

Sut

x2

62.8

3 943.84

64.4

4 147.36

65.8

4 329.64

66.3

4 395.69

68.1

4 637.61

69.1

4 774.81

69.8

4 872.04

71.5

5 112.25

74

5 476.00

S611.8

S41 689.24

1 9
611.9
x
xi

N i 1
9
x 67.98kpsi
n

2
x

N
x
i
2

sx

i 1

N 1

41689.24 9 * 67.982
sx
3.543kpsi
9 1
S x 3.543
Cx

0.0521
x 67.98

74

Example 2
The data in Eg. 1 have come to the designer in
the histographic form. Using the data in this
form, find the mean, standard deviation and
the coefficient of variation

75

Class
Class
f.x
midpoint frequency f
x
63.5
2
127

f.x2

66.5

133

8 844.50

69.5

208.5 14 480.75

72.5

145

Total

613.5 41 912.25
76

8 064.50

10 513.50

1
x
N

613.5
( f i xi )
68.17 kpsi

9
i 1
2

f i xi
n
2
2
i 1

613
.
5
f i xi
41912.25

N
9
i 1
sx

N 1
9 1
s x 3.391kpsi
N

S x 3.391
Cx

0.0497
x 68.17
77

Normal (Gaussian) Distribution


Transformation
Variate

z x / x
Area = 1.00

1
z2
2

e
f ( x)
x 2

0.5
0.4
0.3
0.2
0.1
0.0
-4.0
78

-3.0

-2.0

-1.0

0.0

1.0

2.0

3.0

4.0

Effect of SD on the PDF

Normal Distribution Curves with small


and large
79

Mircrosoft Excel and Stats


http://office.microsoft.com/training/Training.
aspx?AssetID=RP010919241033&CTT=6&Origi
n=RC010919231033

80

The Standard Normal Distribution

The Standard Normal Distribution


81

Z-transform
Excel

z x /

=STANDARDIZE(x,mean,stddev)
Example:
=STANDARDIZE(85,75,10) gives 1.0
82

Standard Normal Cumulative


Distribution
0.5
0.4

area from minus


infinity to z

0.3
0.2
0.1
0.0
-4.0

-3.0

-2.0

-1.0

0.0

1.0

Excel
=NORMSDIST(z)

2.0

3.0

4.0

Example:
=NORMSDIST(1.0)
=0.8413
83

Normal Data in Excel


To avoid Z transform, use:
=NORMDIST(x,mean,stddev,true)

Example

=NORMDIST(85,75,10,true)
= 0.8413

84

Inverse Problem
Given , and probability, find x
=NORMINV(prob,mean,stddev)
Given probability, find z
=NORMSINV(prob)
Note: The probability is the area under the curve
from minus infinity to x (or z)
85

Inverse Problem:
Example 1
A batch of bolts have length =5.00 mm, =0.20 mm.
99% of the bolts are shorter than what length?

Solution 1:
=NORMINV(0.99,5,0.2) gives 5.47 mm
Solution 2:
=NORMSINV(0.99) = 2.33
5.00+0.20*2.33 = 5.47 mm

86

Example
In a shipment of 250 connecting rods, the
mean tensile strength is found to be 45 kpsi
and the standard deviation 5 kpsi.
Assuming a normal distribution, how many
rods can be expected to have a strength less
than 39.5 kpsi?
How many are expected to have a strength
between 39.5 and 59.5 kpsi?
87

z39.5 x / x 39.5 45 / 5 1.10


F(z)=f(-1.10), from NORMSDIST=0.1357
250* f(-1.10)=33.9; about 34 rods has
smaller than 39.5 kpsi
88

z59.5 x / x 59.5 45 / 5 2.90


F(z)=f(2.90)=1- f(-2.90)=1-0.00187=0.99813

p= f(z59.5)- f(z39.5)=0.99813-0.1357=0.86243
250* 0.862 43=215.5; about 215 rods are
between 39.5 and 59.5 kpsi
89

The Lognormal Distribution


The distribution is valid if the log (base n) of
the variable is a normal distribution.
The variables have only positive values and
asymmetrical about the mean
Used when life is involved; fatigue life under
stress or wear life of rolling bearings

x LN x , x
y ln x

y N y , y

90

x has a lognormal
distribution, after
transforming to y, y
has normal
distribution

The Lognormal Distribution


1 ln x y

2 y

e
g(x)
x y 2

x0

CHECK THE MATH

g(x) 0, x 0
1 2
y ln x ln 1 C ln x C x
2
2
x

y ln 1 C x2 C x
91

92

Example 2.4
One thousand specimens of 1020 steel were
tested to rupture and the ultimate tensile
strengths were reported as grouped in Table

etc

Class
Frequ xi*fi
midpoint ency
fi
xi

xi2*fi

Observ Normal
ed PDF density
fi/(Nw) f(x)

Lognor
mal
density
g(x)

56.5

113.0

6 384.5

0.002

0.0026

1 000

63 625

4 054 864

1.00

93

0.0035

1
x
N

63625
( f i xi )
63.625kpsi

1000
i 1
2

f i xi
n
2
2
i 1

63625
f i xi
4054864

N
1000
i 1
sx

N 1
1000 1
s x 2.594kpsi
N

Sx
2.594
Cx

0.0408
x 63.625
94

Treated as normal
distribution

Example 5
1 x 63.625 2
exp

2 2.594245
f(x)
x0
2.594245 2

y ln x ln 1 C x2
y ln 63.625 ln 1 0.040782 4.1522

y ln 1 C x2 ln 1 0.040782 0.0408
1 ln x 4.1522 2
exp

2 0.0408
Lognormal
g(x)
x 0 distribution
0.0408 x 2
95

normal [f(x)] and lognormal [g(x)]


distribution
96

Distribution Parameters

Scale Parameter
The scale parameter is the most common type of parameter. All distributions in this
reference have a scale parameter. In the case of one-parameter distributions, the sole
parameter is the scale parameter. The scale parameter defines where the bulk of the
distribution lies or how stretched out the distribution is. In the case of the normal
distribution, the scale parameter is the standard deviation.
Shape Parameter
The shape parameter, as the name implies, helps define the shape of a distribution.
Some distributions, such as the exponential or normal, do not have a shape parameter
since they have a predefined shape that does not change. In the case of the normal
distribution, the shape is always the familiar bell shape. The effect of the shape
parameter on a distribution is reflected in the shape of the pdf.
Location Parameter
The location parameter is used to shift a distribution in one direction or another. The
location parameter, usually defines the location of the origin of a distribution and can be
either positive or negative. In terms of lifetime distributions, the location parameter
represents a time shift.

97

Location Parameter
Realistically, the calculation
of a negative location
parameter is indicative of
quiescent failures (failures
that occur before a product
is used for the first time) or
of problems with the
manufacturing, packaging
or shipping processes.

98

The Uniform Distribution


A closed-interval distribution that arises when
the chance of an observation is the same as
the chance for any other observation
Observed in manufacturing; when a part is
mass-produced in an automatic operation and
the dimension gradually changes through tool
wear and increased tool forces between
setups
99

The Uniform Distribution


1
f ( x)
; axb
ba
F(x) 0; x a
x -a
F(x)
; axb
b-a
F(x) 1; x b

100

Exponential Distribution

f(x) e

( x )

101

x0

Example 7
A round bar subject to a bending load has a
diameter d=LN(2.000,0.002) inch
This equivalency states that the mean
diameter is 2.000 in and the standard
deviation is 0.0002
Find the mean and the standard deviation of
the second moment of area

102

Linear Regression
Regression
Obtaining a curve that best fits a set of data points
If the best fitting is a straight line, it is linear
regression

y mx b
yi mxi b i

i yi y
103

Linear Regression

104

i yi mxi b
2

d
d
0,
0
dm
db
N xi yi xi yi
m

2
2
N xi xi

m xi
N

y m x
105

x y Nx y
x Nx
i

i
2

Correlation Coefficient r
sx
r m
sy
sm

s y. x

x x

sb s y. x

1
x2

N xi x 2
y m x y
y

b
i i i i
2

s y. x

N 2

106

r=0, no
correlation,
r=1 or 1
perfect
correlation.
Negative
sign means
negative
slope

Example 8
A specimen of a medium carbon steel was tested in
tension. With an extensometer in place, the specimen
was loaded then unloaded, to see if the extensometer
reading returned to the no-load reading, then the next
higher load was applied.
The loads and extensometer elongations were reduced to
stress and strain, see the data
Find the mean Youngs modulus and its standard
deviation. Since the extensometer seems to have an
initial reading at no load, use y=mx+b

107

Data

5033

10 068

15 104

20 143

35 267

0.000 20

0.000 30

0.000 50

0.000 65

0.001 15

Worksheet

x2

xy

y2

5033

0.000 20

0.04E-6

1.006 600

25 330 089

0.130E-6

0.002 80

2.125E-6 65.229

2.004E9

0.556E-6

( x x )2

Summation

85615

108

N xi yi xi yi
N xi xi
2

565.229 0.002885615

2
50.000002215 0.0028
31.03E 6 psi E
y

m xi

N
254.69 psi

85615 31.03E 60.0028

109

sx
3162163E 4
r m
31.03E 6
0.998
sy
11601.11
s y. x

2
i

b yi m xi yi
N 2

2004328267 254.6985615 31.03E 665229

52
811.1 psi
sm

s y. x

x x

811.1

0.000000558

1.086 E 6 psi
110

111

What probability theory is for


Suppose youve already texted the characters
There in a minu
Youd like your mobile phone to guess the most likely
completion of minu rather than MINUET or MINUS
or MINUSCULE
In other words, youd like your mobile phone to
know that given what youve texted so far, MINUTE is
more likely than those other alternatives
PROBABILITY THEORY was developed to formalize
the notion of LIKELIHOOD
113

TRIALS (or EXPERIMENTS)


Anything that may have a certain OUTCOME (on
which you can make a bet, say)
Classic examples:
Throwing a die
A horse race

In LE:
Looking at the next word in a text
Having your system perform a certain task

114

(ELEMENTARY) OUTCOMES
The results of an experiment:
In a coin toss, HEAD or TAILS
In a race, the names of the horses involved
Or if we are only interested in whether a particular
horse wins: WIN and LOSE

In LE:
When looking at the next word: the possible
words
In the case of a system: RIGHT or WRONG
115

EVENTS
Often, we want to talk about the likelihood of
getting one of several outcomes:
E.g., with dice, the likelihood of getting an even
number, or a number greater than 3

An EVENT is a set of possible OUTCOMES


(possibly just a single elementary outcome):
E1 = {4}
E2 = {2,4,6}
E3 = {3,4,5,6}
116

SAMPLE SPACES
The SAMPLE SPACE is the set of all possible outcomes:
For the case of a dice, sample space S = {1,2,3,4,5,6}

For the texting case:


Texting a word is a TRIAL,
The word texted is an OUTCOME,
EVENTS which result from this trial are: texting the word
minute, texting a word that begins with minu, etc
The set of all possible words is the SAMPLE SPACE
(NB: the sample space may be very large, or even infinite)

117

Probability Functions
The likelihood of an event is indicated using a
PROBABILITY FUNCTION P
The probability of an event E is specified by a
function P(E), with values between 0 and 1
P(E) = 1: the event is CERTAIN to occur
P(E) = 0: the event is certain NOT to occur

Example: in the case of die casting,

P(E = getting as a result a number between 1 and 6) =


P({1,2,3,4,5,6}) = 1
P(E = getting as a result 7) = 0

The sum of the probabilities of all elementary


outcomes = 1
118

Probabilities and
relative frequencies
In the case of a die, we know all of the possible outcomes ahead of time,
and we also know a priori what the likelihood of a certain outcome is. But
in many other situations in which we would like to estimate the likelihood
of an event, this is not the case.
For example, suppose that we would like to bet on horses rather than on
dice. Harry is a race horse: we do not know ahead of time how likely it is
for Harry to win. The best we can do is to ESTIMATE P(WIN) using the
RELATIVE FREQUENCY of the outcome `Harry wins
Suppose Harry raced 100 times, and won 20 races overall. Then
P(WIN) = WIN/TOTAL NUMBER OF RACES = .2
P(LOSE) = .8

119

Joint probabilities
We are often interested in the probability of TWO events
happening:
When throwing a die TWICE, the probability of getting a 6 both
times
The probability of finding a sequence of two words: `the and
`car
We use the notation A&B to indicate the conjunction of two events,
and P(A&B) to indicate the probability of such conjunction
Because events are SETS, the probability is often also written as
We use the same notation with WORDS: P(the & car)

P( A B)
120

Other combinations of events


A B: either event A or event B happens
P(A B) = P(A) + P(B) P(AB)

A: event A does not happen


P( A) = 1 P(A)

121

Prior probability vs. conditional


probability
The prior probability P(WIN) is the likelihood of an event occurring
irrespective of anything else we know about the world
Often however we DO have additional information, that can help us
making a more informed guess about the likelihood of a certain event
E.g, take again the case of Harry the horse. Suppose we know that it
was raining during 30 of the races that Harry raced, and that Harry won
15 of these races. Intuitively, the probability of Harry winning when its
raining is .5 - HIGHER than the probability of Harry winning overall
We can make a more informed guess
We indicate the probability of an event A happening given that we
know that event B happened as well the CONDITIONAL PROBABILITY
of A given B as P(A|B)

122

Conditional probability
Conditional probability is DEFINED as follows:

P( A & B)
P( A | B)
P( B)
Intuitively, you RESTRICT the range of trials in
consideration to those in which event B took place,
as well (most easily seen when thinking in terms of
relative frequency)

123

Example
Consider : Horse has won in rain 15 times. The
season is 100 days and it has rained on 30 days.
Where:

P(WIN & RAIN )


P(WIN | RAIN )
P( RAIN )

P(WIN&RAIN) = 15/100 = .15


P(RAIN) = 30/100 = .30

This gives:
0.15
P(WIN | RAIN )
0.5
0.3
124

The chain rule


The definition of conditional probability can we
rewritten as:
P(A&B) = P(A|B) P(B)
P(A&B) = P(B|A) P(A)

These equation generalize to the so-called CHAIN


RULE:
P(w1,w2,w3,.wn) = P(w1) P(w2|w1) P(w3|w1,w2) .
P(wn|w1 . wn-1)

The chain rule plays an important role in statistical


LE:
P(the big dog) = P(the) P(big|the) P(dog|the big)
125

Independence
Additional information does not always help. For example, knowing
the color of a dice usually doesnt help us predicting the result of a
throw; knowing the name of the jockeys girlfriend doesnt help
predicting how well the horse he rides will do in a race; etc. When
this is the case, we say that two events are INDEPENDENT
The notion of independence is defined in probability theory using
the definition of conditional probability
Consider again the basic form of the chain rule:
P(A&B) = P(A|B) P(B)
We say that two events are INDEPENDENT if:
P(A&B) = P(A) P(B)
P(A|B) = P(A)

126

Bayes theorem
Suppose youve developed an Information Retrieval
system for searching a big database (say, the Web)
Given any search, about 1/100,000 documents is
relevant (REL) i.e. P(REL) =0.00001
Suppose your system is pretty good:
P(YES|REL) = .95
P(YES| REL) = .005

What is the probability that the document is


relevant, when the system says YES?
P(REL|YES)?
127

Bayes Theorem
Bayes Theorem is a pretty trivial consequence of the
definition of conditional probability, but it is very useful
in that it allows us to use one conditional probability to
compute another
We already saw that the definition of conditional
probability can be rewritten equivalently as:
P(A&B) = P(A|B) P(B)
P(A&B) = P(B|A) P(A)

If we equate the two left sides, we get Bayes theorem

P( A | B) P( B)
P( B | A)
128 P ( A)

Application of Bayes theorem


P(YES | REL ) P( REL )
P( REL | YES )
P(YES )
P(YES | REL ) P( REL )

P(YES | REL ) P( REL ) P(YES | REL ) P(REL )

0.95 0.00001

0.002
0.95 0.00001 0.005 0.99999

129

Damage Tolerant Approach to Life


Assessment for Perfect NDE

Distribution of nucleating conditions


Broken curves: damage removed by perfect inspection

Short Course - QNDE for Pipes_July 2008

130

NDE is Not Perfect


All damage of the same size does not produce the
same signal strength
Detection occurs for that fraction of damage of a
given size whose signals exceed a fixed threshold
(POD)
Potential for false alarms influences this threshold

Short Course - QNDE for Pipes_July 2008

131

NDE is Not Perfect

Short Course - QNDE for Pipes_July 2008

132

Imperfection in NDE Leads To Probability of a


Miss

Short Course - QNDE for Pipes_July 2008

133

Key Inputs To Damage Tolerance Analysis


Initial damage distribution
Damage growth laws
Effectiveness of NDE (POD)

Short Course - QNDE for Pipes_July 2008

134

Outline

Overview of Damage Tolerance


Current AF Practice in Fatigue
Probabilistic Risk Assessment for Fatigue
Determination of POD
Influence of Various Factors
Implications for Pipeline Inspection

Short Course - QNDE for Pipes_July 2008

135

Safe Life Design


Assume structure initially unflawed
How long before crack initiates?
No credit for crack propagation

Design lives based on


Fatigue test mean life
Safety factor (e.g., 4) such that acceptably small
numbers of failure during design life (e.g., <0.1%).

Short Course - QNDE for Pipes_July 2008

136

Safe Life Design


Extensive component testing plus a safety factor

Reference: Grandt
Short Course - QNDE for Pipes_July 2008

137

Safe Life Design


A related but different view

Reference: Christodoulou and Larsen


Short Course - QNDE for Pipes_July 2008

138

Weaknesses of Safe-Life Approach


Economic
Many components reach end of design life with
much actual life left
Improved NDE implies more components removed
from service, irrespective of actual remaining life
Safety/Readiness
Unanticipated initial damage can greatly reduce
crack nucleation time

Short Course - QNDE for Pipes_July 2008

139

Some Past Problems with the Approach


Unanticipated initial damage in specific cases greatly
reduces crack nucleation portion of fatigue life
KC-135 Although the safe-life for this transport/tanker aircraft was
determined to be 13,000 flight hours, there were 14 cases of unstable
cracking in lower wing skins that occurred between 1800 and 5000
hours. These premature failures resulted in a costly wing modification
for the entire fleet at 8500 service hours
F-5 One of these fighters failed by fatigue cracking in the lower wing skin
after 1900 flight hours although the safe-life for this area had been
shown to be 4000 hours.
F111 The safe-life for this aircraft was 4000 flight hours, but one aircraft
was lost to fatigue after only 105 hours of service.
Reference: Grandt

Short Course - QNDE for Pipes_July 2008

140

Undetected Forging Crack in F-111

This 1969 failure of a new F-111 aircraft was caused by an undetected forging defect
that quickly grew to failure by fatigue in the high-strength steel wing pivot fitting
structure.
Reference: Grandt
Short Course - QNDE for Pipes_July 2008

141

Damage Tolerant Design


Assume structure contains initial cracks
Structure designed to resist fracture from cracks of a
given size for a specified interval of time
Initial crack size often based on inspection limits

Short Course - QNDE for Pipes_July 2008

142

Fracture Mechanics Models for Evolution of


Fatigue Damage

Short Course - QNDE for Pipes_July 2008

143

Slow Crack Growth


Assume
structure initially contains
cracks
these grow in a controlled
way as predicted by fracture
mechanics
allowable life determined by
a detectable flaw size (aNDI) aNDI
and safety factor

Reference: Grandt
Short Course - QNDE for Pipes_July 2008

144

Retirement for Cause


Extension beyond initial design
life
Periodic inspection locate
damaged components that are
repaired or replaced
Fracture mechanics used to
determine the remaining service
life an undetected crack to grow
to failure
Requires reliable determination
of maximum crack size that could
go undetected

Reference: Grandt
Short Course - QNDE for Pipes_July 2008

145

Outline

Overview of Damage Tolerance


Current AF Practice in Fatigue
Probabilistic Risk Assessment for Fatigue
Determination of POD
Influence of Various Factors
Implications for Pipeline Inspection

Short Course - QNDE for Pipes_July 2008

146

Probabilistic, Risk Analysis Framework of


USAF

Gallagher, Babish, Malas, Damage Tolerant Risk Analysis Techniques for


Evaluating the Structural Integrity of Aircraft Structures, Presented at 11th
International Conference on Fracture, Turin, Italy (March, 2005).
Short Course - QNDE for Pipes_July 2008

147

Nature of Inputs
Probabilistic

Initial and repair flaw sizes


Fracture toughness
Probability of detection
POD
Maximum stress per flight

Deterministic
Normalized stress intensity function at critical locations
Fatigue crack growth curve

Short Course - QNDE for Pipes_July 2008

148

Enhanced Life Management Process of FAA

Damage Tolerance for High Energy Turbine Engine Rotors, Federal Aviation
Administration Advisory Circular AC 33.14-1, ANE-110 (January 8, 2001).
Short Course - QNDE for Pipes_July 2008

149

Typical Elements of a Titanium Melt Related


Anomaly Risk Assessment

Short Course - QNDE for Pipes_July 2008

150

Relative Risk Calculation for Single In-Service


Inspection

Short Course - QNDE for Pipes_July 2008

151

Outline

Overview of Damage Tolerance


Current AF Practice in Fatigue
Probabilistic Risk Assessment for Fatigue
Determination of POD
Influence of Various Factors
Implications for Pipeline Inspection

Short Course - QNDE for Pipes_July 2008

152

Determination of POD

Quantify, for a given flaw size, either


Details of signal distribution
Numbers of hits and misses
Short Course - QNDE for Pipes_July 2008

153

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

154

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

155

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

156

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

157

Breaking Up Into Size Ranges (Grouping)


Purely empirical analysis of
binary, hit/miss data

NumberDetected
POD
Total Number
Short Course - QNDE for Pipes_July 2008

158

Challenges
Want high confidence (resolution in POD)
How confident are you that a POD estimate based on particular
experiment is close to the correct answer?
Resolution in POD improves when number of samples
increases

Want high resolution in size


Implies groups have smaller size range
For a fixed total number of samples, fewer samples per group

Trade-offs must be made between resolution in POD and


resolution in size

Short Course - QNDE for Pipes_July 2008

159

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

160

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

161

A Parametric Approach
Assume a model between test POD and size
Use data to determine parameters of model

Two variations
Hit/miss data
Flaw response data ( versus a)
For each flaw, record response () and size (a)
More information than hit/miss
More precise estimates (narrower confidence bounds)

Short Course - QNDE for Pipes_July 2008

162

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

163

Pictorial Representation of Flaw Response POD


Determination ( versus a)

Obtain and plot data of log (flaw response) versus log (flaw size),
known as a a-hat versus a
Perform a linear regression
When distribution about regression line is normal, POD determined
by:
Mean
Standard deviation
Short Course - QNDE for Pipes_July 2008
164
Threshold

Similar Approach Exists for Binary Data

Short Course - QNDE for Pipes_July 2008

Reference: Annis

165

Directed Design Of Experiments For Validating Probability Of


Detection Capability Of NDE Systems (DOEPOD)

ABSTRACT. The capability of an inspection system is established by applications of


various methodologies to determine the probability of detection (POD). One
accepted metric of an adequate inspection system is that there is 95% confidence
that the POD is greater than 90% (90/95 POD). Directed design of experiments for
probability of detection (DOEPOD) has been developed to provide an efficient and
accurate methodology that yields observed POD and confidence bounds for both
Hit-Miss or signal amplitude testing. Specifically, DOEPOD demands utilization of
observance of occurrences. Directed DOEPOD does not assume prescribed POD
logarithmic or similar functions with assumed adequacy over a wide range of flaw
sizes and inspection system technologies, so that multi-parameter curve fitting or
model optimization approaches to generate a POD curve are not required.
E. R. Generazio, National Aeronautics and Space Administration
Review of Progress in Quantitative Nondestructive Evaluation,
Vol 27, pp. 1693-1700

Short Course - QNDE for Pipes_July 2008

166

Outline

Overview of Damage Tolerance


Current AF Practice in Fatigue
Probabilistic Risk Assessment for Fatigue
Determination of POD
Influence of Various Factors
Implications for Pipeline Inspection

Short Course - QNDE for Pipes_July 2008

167

Short Course - QNDE for Pipes_July 2008

Reference: Rummel

168

Dependence of EC POD on Feature for Aircraft


Engine (Test Object Variables)

Short Course - QNDE for Pipes_July 2008

169

In the real world flaw morphology and other factors can


make things more complex.

Example of estimating POD for the ultrasonic


detection of hard-alpha inclusions in aircraft engine,
billet material

Short Course - QNDE for Pipes_July 2008

170

Two Systems for Detection of


Hard-Alpha Inclusions

Reference: Default POD Report


Short Course - QNDE for Pipes_July 2008

171

Knowledge of Misses and Flaw Morphology Important


in POD Determination

Reference: Default POD Report


Short Course - QNDE for Pipes_July 2008

172

Current POD Curves for Detection of HardAlpha Inclusions

Reference: Default POD Report


Short Course - QNDE for Pipes_July 2008

173

Model-Assisted probability of Detection


(MAPOD)
An approach to allow extending old POD results to new
situations
Extension based on information gained in controlled laboratory
experiments or from physics-based models
Allows extension to new flaw morphologies, geometries,
materials, etc.
Several demonstrations underway
Large potential savings in time and cost

http://www.cnde.iastate.edu/MAPOD/
Short Course - QNDE for Pipes_July 2008

174

MAPOD Strategy

R. B. Thompson, Materials Evaluation, Vol. 66,


No. 6, pp. 667-673, 2008.
Short Course - QNDE for Pipes_July 2008

175

MAPOD Example
mc

Lab data to control variability

CRACKS

crack response

Flat Plate

bc

NOTCHES

notch response

crack depth

Crack
Response

Notch
=
Response

10
10

mcxc + bc
mnxn + bn

mn

bn

Complex Geometry/Notch
nothc depth

POD

POD

Variability Data

Complex Geometry/Notch

Notch size

K. Smith, MAPOD Working Group, Palm Springs,


CA, February 4, 2005.
Short Course - QNDE for Pipes_July 2008

Crack size

176

Outline

Overview of Damage Tolerance


Current AF Practice in Fatigue
Probabilistic Risk Assessment for Fatigue
Determination of POD
Influence of Various Factors
Implications for Pipeline Inspection

Short Course - QNDE for Pipes_July 2008

177

Several Issues Requiring Attention


If want to do quantitative, probabilistic risk assessment,
need to have strategy to determine
Initial distribution of damage
Evolution of damage with time
POD of inspection

Determination of the latter will be influenced by

Geometry of the samples


Number of the samples
Flaws of realistic morphology
Independent knowledge of misses if analysis based on field data

Short Course - QNDE for Pipes_July 2008

178

Model-Assisted probability of Detection


(MAPOD)
An approach to allow extending old POD results to new
situations
Extension based on information gained in controlled laboratory
experiments or from physics-based models
Allows extension to new flaw morphologies, geometries,
materials, etc.
Several demonstrations underway
Large potential savings in time and cost

http://www.cnde.iastate.edu/MAPOD/
179

MAPOD Strategy

R. B. Thompson, Materials Evaluation, Vol. 66, No. 6, pp. 667-673, 2008.

180

Modeling Approach to POD


Validated models can replace costly
Experimental approach.

Damage Characterization of Material Systems

Overview of NDE at Difference Scales


Creep cracks

Scanning Optical Microscopy


Thermal wave microscopy
Acoustic microscopy
Micro focal-Radiography
Scanning electron microscope

Ultrasonic testing
Interferometry
Thermography
X-Ray Imaging
Eddy current

1 mm

Macroscopic defects
Cracks
Pores
Inclusions
Thickness loss
Adapted from Mayendorf et. Al.

Transmission electron microscope


Positron annihilation
Nonlinear acoustics
Near Field Imaging Nanoscopy

1 m

Micro defects
Initial corrosion pits
Micro cracks
Micro pores

1 nm

Microstructure
Grain boundaries
Precipitations
Voids

Crystal lattice defects


Dislocations
Vacancies
Micro voids
Interstitials
1
8

Materials Characterisation

Is Failure Predictable ?

Key science and technology disciplines


Coupled physics-based models of materials damage and behavior
Interaction of multiple damage/failure mechanisms
Multi-scale, mechanism-based
Microstructurally-based stochastic behavior
Integrated information from state-awareness tools
Interrogation of damage-state
Intelligently exploit existing sensors
Feature extraction from global sensors
Materials-damage-state interrogation techniques and recorders
Sensor signature analysis
Data management and fusion
Capability matched to mission
Component usage data
Component history and pedigree

Step 1: Understanding Mechanisms

Courtesy: Chirstoudoulou et. al, DARPA, USA

Step 2: Interrogation of State of the


Material/Structure

Courtesy: Chirstoudoulou et. al, DARPA, USA

Step 3: Prognosis

Courtesy: Chirstoudoulou et. al, DARPA, USA

DAMAGE MECHANICS GENERIC APPROACH


DAMAGE MODELING

FEA

- Develop damage model


- Identify model parameters
* Tensile, LCF, Creep

(Specimen level)

Damage Analysis

VALIDATION
Predicted Damage
Measured Damage

No
Application to
Component
Courtesy: Dr. Vikas Kumar, DMRL, Hyderabad

No

Yes

Good Comparison
?

Prognosis Benefits

Courtesy: Chirstoudoulou et. al, DARPA, USA

SUMMARY
NDE has a very WIDE and CRITICAL role in the industry.
NDE improves productivity and safety.
Understanding the physics, capabilities, and limitation
of different modalities on NDE is key to best practices
in the industry.
Probability and Statistics play a very key role in
providing realistic expectations from NDT.
Theoretical Models may go a long way.
NDE plays a key role in the Prognosis of Critical
Components, however, only when understood in
conjunction with other key partnering technologies
including Materials Science, Modeling, Mechanics, etc.

You might also like