This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Dr. M. ŞAHİN
AE 713 - Experimental Analysis of Vibrating Structures
Finite Element Model of the Cantilever Composite Beam(Top view)
0.55Lw
L
Damaged Area
Collocation Point
Damage Location
16
Dr. M. ŞAHİN
FINITE ELEMENT MODELLING
Dr. M. ŞAHİN
SENSITIVITY ANALYSES: Part I
Percentage Reduction in Natural Frequencies
Damage located at 0.20L Damage located at 0.35L
Damage Severity (% Reduction in Stiffness)
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
Damage Severity (% Reduction in Stiffness)
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
17
Dr. M. ŞAHİN
SENSITIVITY ANALYSES: Part I
Percentage Reduction in Natural Frequencies
Damage located at 0.45L Damage located at 0.55L
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
Damage Severity (% Reduction in Stiffness)
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
Damage Severity (% Reduction in Stiffness)
Dr. M. ŞAHİN
SENSITIVITY ANALYSES: Part I
Percentage Reduction in Natural Frequencies
Damage located at 0.65L Damage located at 0.80L
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
Damage Severity (% Reduction in Stiffness)
Per
cent
age
Red
uct
ion
in N
atur
al F
requ
ency
Damage Severity (% Reduction in Stiffness)
18
Dr. M. ŞAHİN
SENSITIVITY ANALYSES: Part II
Absolute Differences in Curvature Mode Shapes
along the beam
Damage located at 0.20LMode No: 2
Damage located at 0.35LMode No: 1
Mag
nitu
de
Measurement Locations [L]
Damage Severity
Mag
nitu
de
Measurement Locations [L]
Damage Severity
Dr. M. ŞAHİN
SENSITIVITY ANALYSES: Part II
Absolute Differences in Curvature Mode Shapes
along the beam
Damage located at 0.65LMode No: 1
Damage located at 0.80LMode No: 3
Mag
nitu
de
Measurement Locations [L]
Damage Severity
Mag
nitu
de
Measurement Locations [L]
Damage Severity
19
Dr. M. ŞAHİN
Pre-Processing
Dr. M. ŞAHİN
DAMAGE DETECTION ALGORITHM
- Pattern Associator -• Multi-Layer
– Input, output and hidden layers
• Feed-Forward Back-Propagation– Training iterations from input layer to output– To adjust the weights so that introducing of set of
inputs produces the desired set of outputs– Error values are calculated and fed in the backward
direction to minimise the error function
“Multi-Layer Feed-Forward Back-Propagation”
20
Dr. M. ŞAHİN
ARTIFICIAL NEURAL NETWORK
Input Data
• RNF : Reduction in (Normalised) Natural Frequency
• MADC : Maximum Absolute Differences in
Curvature Mode Shape
• LOC : Location where MADC occurs
Output Data
• DS :Severity of the Damage
• DL :Location of the Damage
• DS&DL : Both Severity and Location of the Damage
Architecture
a:b:cNumber of Input Number of Output
Number of Neuronsin the Hidden Layer
Dr. M. ŞAHİN
ARTIFICIAL NEURAL NETWORK
+
21
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Noise Free Data
RNF for
Damage Severity
RNF for
Damage Location
Tra
inin
g
Val
idat
ion
Number of Epochs
Mean Square Error
Number of Epochs Target
Target
Pre
dict
ed V
alu
esP
redi
cted
Val
ues
Tra
inin
g
Val
idat
ion
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Noise Free Data
MADCfor
Damage Location
MADC&LOC for
Damage Location
Target
Target
Pre
dict
ed V
alu
esP
redi
cted
Val
ues
Tra
inin
g
Val
idat
ion
Tra
inin
g
Val
idat
ion
Number of Epochs
Number of Epochs
Mean Square Error
22
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Noise Free Data
RNF&MADC&LOCfor
Damage Severity and Location
Damage LocationDamage SeverityNumber of Epochs
Mean Square Error
Tra
inin
g
Val
idat
ion
Target Target
Pre
dict
ed V
alu
es
Pre
dict
ed V
alu
es
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Data with Noise
RNF for
Damage Severity
0.5% Noise on RNF
1% Noise on RNF 2% Noise on RNF
Pre
dict
ed V
alu
es
Pre
dict
ed V
alu
es
TargetTarget
Pre
dict
ed V
alu
es
Target
23
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Data with Noise
MADC&LOC for
Damage Location
1% Noise on MADC
5% Noise on MADC3% Noise on MADC
Pre
dict
ed V
alu
es
Pre
dict
ed V
alu
es
Target
Target Target
Pre
dict
ed V
alu
es
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Data with Noise
Damage Severity
0.5% Noise on RNF and 1% Noise on MADC
Damage Location
RNF&MADC&LOCfor
Damage Severity and Location
Target Target
Pre
dict
ed V
alu
es
Pre
dict
ed V
alu
es
24
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Data with Noise
RNF&MADC&LOCfor
Damage Severity and Location
Pre
dict
ed V
alu
es
Target
Damage Severity
Target
Pre
dict
ed V
alu
es
Damage Location
1% Noise on RNF and 3% Noise on MADC
Dr. M. ŞAHİN
RESULTS: ANN Predictions- Data with Noise
RNF&MADC&LOCfor
Damage Severity and Location
Pre
dict
ed V
alu
es
Target
Damage Severity
Target
Pre
dict
ed V
alu
es
Damage Location
2% Noise on RNF and 5% Noise on MADC
25
Dr. M. ŞAHİN
CONCLUSIONS -Composite Beam
• Completely numerical study has been performed.
• Vibration-based feature extraction coupled with a trained ANN provides the basis for real time damage assessment.
Dr. M. ŞAHİN
CONCLUSIONS -Composite Beam
• Although reduction in natural frequencies is considered as an indicator for the existence of the damage and its severity, they did not provide any useful information about the location of the damage.
• Maximum absolute differences in curvature mode shapes and their corresponding locations along the beam served as better indicators for the location of the damage. Therefore, these features were used as separate input for the ANNs.
26
Dr. M. ŞAHİN
STRUCTURE
Cantilever Steel Beam
Dr. M. ŞAHİN
STRUCTURE- Beam Finite Element Model
Beam elements
Shell elements
Solid elements
27
Dr. M. ŞAHİN
STRUCTURE- Beam Finite Element Model
Dr. M. ŞAHİN
STRUCTURE- Beam Finite Element Model
28
Dr. M. ŞAHİN
STRUCTURE- Beam Finite Element Model
Dr. M. ŞAHİN
STRUCTURE- Specimen
29
Dr. M. ŞAHİN
EXPERIMENTAL SET-UP
PC
PC
PC
Digital Vibration Controller& Data Acquisition Unit
Power Amplifier
Electro-dynamic Vibration Generator
Clamp
Beam
Signal Conditioning
Amplifier
Data Acquisition Unit
Fibre Optic Strain Measurement System
Accelerometer(Control)
StrainGages
Fibre Optic Sensors
Dr. M. ŞAHİN
EXPERIMENTAL SET-UP
Strain gauge conditioning unit
30
Dr. M. ŞAHİN
EXPERIMENTAL SET-UP
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
31
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
32
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
33
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for intact steel beam for different frequency ranges1-20 Hz
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for intact steel beam for different frequency ranges70-90 Hz
34
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for intact steel beam for different frequency ranges205-225 Hz
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for damagedsteel beam for different frequency ranges1-20 Hz
35
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for damagedsteel beam for different frequency ranges70-90 Hz
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Strain values for damagedsteel beam for different frequency ranges205-225 Hz
36
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of intact steel beamMode 1
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of intact steel beamMode 2
37
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of intact steel beamMode 3
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of damagedsteel beamMode 1
38
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of damagedsteel beamMode 2
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Normalised absolute curvature mode shapes of damagedsteel beamMode 3
• Different damage scenarios have been created by reducing the local thickness of the selected elements at different locations.
43
Dr. M. ŞAHİN
CONCLUSIONS -Steel Beam
• It can be concluded from the ANN predictions that the better accuracy has been achieved in severity predictions than the location ones in noise-free case.
• Introducing an artificial noise on noise-free data has adversely affected the severity predictions although the results are still accurate for the location predictions obtained from each ANN used in the verification by using an experimental data.
Dr. M. ŞAHİN
STRUCTURE
Cantilever Sandwich Beam• Core: Linear polymer foam (Core-Cell® A500)
are used during the manufacturing of sandwich beam specimens.
44
Dr. M. ŞAHİN
STRUCTURE
Cantilever Sandwich Beam450mm x 40.5mm x 14.6mm beam
Total thickness of 1.3mm on each side of 12mm thick foam core.
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Fabrication of the Sandwich Beam Specimens with Embedded FOSs
45
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Fabrication of the Sandwich Beam Specimens with Embedded FOSs
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Fabrication of the Sandwich Beam Specimens with Embedded FOSs
46
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Fabrication of the Sandwich Beam Specimens with Embedded FOSs
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
• A Teflon tape is inserted between the foam core and the GFRP skin along the length of the sandwich beam.
• The locations are measured from the fixed end to the centre of the damage.
Introducing of Damage
47
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
• The first three resonant frequencies of the intact and four damaged beams are obtained under random excitation in
the range of 10Hz to 710Hz.
Frequency Measurements
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Frequency Measurements
48
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
450mm x 40.5mm x 14.6mm beam
49
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Effect of Boundary Condition
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Effect of Boundary Condition
50
Dr. M. ŞAHİN
STRUCTURE – Experimental Study
Effect of Boundary Condition
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Frequency Measurements
51
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Sensitivity Analysis on Changes in Frequency
• Damage scenarios are extended by interpolating the normalised natural frequencies for the damage extents between 10mm and 50mm by an increment of 1mm which gives 41 different damage extents at 26 different damage locations along the beam.
• The normalised natural frequencies from 1066 damage cases are obtained from the first three vertical bending modes.
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Variation of normalised natural frequencies: Mode 1
52
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Variation of normalised natural frequencies: Mode 2
Dr. M. ŞAHİN
STRUCTURE – Numerical Study
Variation of normalised natural frequencies: Mode 3