Some Essentials of Data Analysis with Wavelets
Slid i h l f h i d l i Th
Slides in the wavelet part of the course in data analysis at The Swedish National Graduate School of Space Technology
Lecture 2: The continuous wavelet transform
Niklas Grip Department of Mathematics L leå Uni ersit of technolog Niklas Grip, Department of Mathematics, Luleå University of technology
Last update:
2009-12-10
The theme of Joh. Seb. Bach’s Goldberg variations
Music
Source:http://www.ti6.tu-harburg.de/~rolf/Goldberg.html
Time frequency analysis
Time-frequency analysis
The cochlea
The Gabor transform
( ):
Short time Fourier transform continuous Gabor transform
( )
, ,
2
( , ) ( ) ( ) ( ) ( ) ,
h ( ) ( ) (2 ) i (2 ) ( )
g f x f x
i ft
V s x f s t g t dt s g d
t
pft i ft t
x x x
¥ ¥
-¥ -¥
= ò = ò Parseval’s relation
( )
, 2
where g
f x( ) t = e
i pftg(t-x)= cos(2 p ft ) + i sin(2 p ft g t x ) ( - ).
A wavelet is a bounded function for which y y( )t dt 0.
¥
ò
( ) =(Some extra technical contitions (MRA) must be satisfied for
getting the orthonormal wavelet bases discussed in previous slides )
y y
-¥
ò
Continuous wavelet transform (CWT)
getting the orthonormal wavelet bases discussed in previous slides.)
The : continuous wavelet transform Parseval’s
relation
( , ) ( ) a b ( ) ( ) a b ( ) ,
s a b s t t dt s d
y y x y x x
¥ ¥
= ò = ò
W
relation
( )
, ,
( , ) ( ) ( ) ( ) ( ) , ( ) 1
a b a b
t b
y y x y x x
-¥ -¥
-
ò ò
( )
,
where a b ( ) 1 t b .
t a a
y = y
The time-frequency
( )
localization of
1 -
a b
( ) t b
y t = y
Wavelet TF-localization
( )
,
( )
is completely described by and .
a b
a a
a b
y y
by a a d b
Heisenberg boxes
Wavelets: STFT:
Time - frequency localization of a function g
(
0)
2( )
2where t
0( )
2g
t t g t dt t g t dt
¥ ¥
D = ò -
TF-localization= ò
( )
( )
0 0
0
2 2 2
( ) ( )
( ) h ( )
g
g g
f f f df f f f df
-¥
¥ ¥
D
ò ò
ò (
0)
2
2 0ò
20
( ) where ( )
g
f f g f df f f g f df
-¥
D =
ò - = ò
Formulas "borrowed" from mechanics
Note
H i b t it
0 0
2
Formulas borrowed from mechanics ( , centre of mass) and probability
theory ( ( ) 1 expectation Note.
t f
g t dt t f
¥
«
= «
ò
Heisenberg uncertanity principle: the area
1
¥0 0
theory ( ( ) 1, , expectation and ,g standard deviation.)
g
g t dt t f
-¥
= «
D D «
ò
2 -
1 ( ) .
g g
4 g t dt
p
¥D D ³
ò
Bearing condition monitoring Bearing condition monitoring
•Bearing failures can cause both personal damages and economical loss
Bearing condition monitoringdamages and economical loss.
•Often not possible to stop production to check bearings
check bearings.
•Usual monitoring techniques today analyse
analyse
time domain signal or Fourier transform.
Vibration measurements with handheld device
Vibration measurements with handheld device
Main goal
Noise-free vibrations
CWT vibration
analysis
Example plot after further analysis Example plot after further analysis
Close up:
Full:
20 40 60 80 100 120 140 160
20 40 60 80
14 16 18
8 10 12
2 4 6
2 4 6 8 10 12 14
2