seek LIGHT | The University of Adelaide
Guided Wave based Quantitative Identification of Damage in Beams Using a Bayesian Approach
Dr. (Alex) Ching-Tai Ng
Australian Network of Structural Health Monitoring (ANSHM) 2013 Annual Workshop 19th November 2013, Melbourne, Australia
seek LIGHT | The University of Adelaide Slide 1
Hanning windowed sinusoidal tone burst
pulse Giurgiutiu & Bao (2004) Struct. Health. Monitor. Animation from [www.me.sc.edu/Research/lamss/]
Piezoceramic transducer
• Guided Wave § Sensitive to small and different types of damages § Long travel distance
Introduction and Background
Fundamental symmetric mode (S0) +
+
Fundamental anti-symmetric mode (A0) +
-
seek LIGHT | The University of Adelaide Slide 2
Introduction and Background
Challenges: • Requirement of baseline data
• Temperature variation & effect of external loading conditions • Difficult to achieve quantitative identification of damages
Excitation
Sensor Damage
Incident pulse reflected from beam end
Excitation
seek LIGHT | The University of Adelaide Slide 3
• To quantitatively identify the location and size of the damage
• To improve the computational efficiency of the proposed damage identification method using frequency domain spectral finite element simulation
• To quantify the uncertainties associated with the damage identification results using a Bayesian approach
• To provide an experimental verification of the proposed method
Objectives
seek LIGHT | The University of Adelaide Slide 4
Frequency-domain Spectral Finite Element Method
§ Mindlin-Herrmann theory § Describes the longitudinal wave using two coupled partial differential
equations**
§ Each element has 2 nodes & each node has 2 DoFs § Account the axial displacement & lateral contraction effect
Spectral finite element
** Krawczuk M, Grabowska J and Palacz M, J Sound Vib 2006, 295:461-478
seek LIGHT | The University of Adelaide Slide 5
Frequency-domain Spectral Finite Element Method
§ The governing equations are reduced to two ordinary differential equations and assumes the solutions in the forms
§ Formulate the dynamic stiffness matrix in frequency domain (at frequency ) by considering the boundary conditions
seek LIGHT | The University of Adelaide Slide 6
Excitation signal in time domain, f(t)
Excitation signal in frequency domain, f(ωn) for n = 1,…,N
FFT
Dynamic stiffness matrix K(ωn)
Calculate displacement in frequency domain
n = n+1
Excitation signal in time domain, f(t) iFFT
Frequency-domain Spectral Finite Element Method
n = N
seek LIGHT | The University of Adelaide Slide 7
f(t)
Damage location, L1
L1
Damage depth, d
L2
L3
EL1 EL2 EL3
E, A1
E, A1
E, A2
Node 2 Node 3 Node 4Node 1
f(t)
Damage length, L2
EL0
Free end
Spectral finite element model of a beam with step damage
f(t)
Damage length (dL)
Damage depth (dd)
Damage length
Damage location (Ld)
Free end Infinite beam end Measurement point
Frequency-domain Spectral Finite Element Method
seek LIGHT | The University of Adelaide Slide 8
§ Using the Bayes’ theorem, the probability of the set of uncertain damage parameters (θ) with a given set of dynamic data is **:
Bayesian Approach
( ) ( ) ( ), , |p D M cp D M p M=θ | |θ θ
( ) ( ) ( )22, 2o
oNNNN J
p D M e σπσ− −
=θ
|θ
( ) ( ) ( ) 21
1 ;N
oko
J k kNN =
= −∑θ q S y θ
where c is normalisation constant.
where J(θ) is:
Likelihood Prior distribution
(Allow the inclusion of engineering judgment about the possible damage)
Simulated signal Measured signal
** Beck JL and Katafygiotis LS, J. Eng. Mech. ASCE. 1998, 124(4), 455-461
The minimisation problem is solved by Hybrid Particle Swarm Algorithm
seek LIGHT | The University of Adelaide Slide 9
§ The updated (posterior) PDF of damage parameters for given data and model class can be approximated as a weighted sum of Gaussian distributions:
§ The weightings are given by:
( ) ( ) ( )( )( )1
1
ˆ ˆ| ,I i i
ii
p D w N −
=≈ ∑θ θ A θ
( )( ) ( )( )12ˆ ˆi i
i Nw π−
= θ A θ
Bayesian Approach
seek LIGHT | The University of Adelaide Slide 10
Experimental Verification
Backing mass
Piezoceramic transducer
Longitudinal guided waveStep damage
Laser head
Aluminum beam Beam cross-section: 12x6 mm2
Beam length: 2 m Excitation frequency: 80 kHz
seek LIGHT | The University of Adelaide Slide 11
Experimental Verification
Preliminary study of measurement location using 3D finite element method (LS-DYNA)
Scaled contour plot of displacements
seek LIGHT | The University of Adelaide Slide 12
Experimental Verification
Preliminary study of measurement location using 3D finite element method (LS-DYNA)
Reflected longitudinal wave propagates toward the beam end
Flexural waves propagate toward the beam end
seek LIGHT | The University of Adelaide Slide 13
-1 -0.5 0 0.5 1-1
-0.83-0.67
-0.5-0.33-0.17
00.170.33
0.50.670.83
1
Normalized AmplitudeNor
mal
ized
Thr
ough
-thic
knes
s Lo
catio
n
S0 guided wave
• Modeshapes
-1 -0.5 0 0.5 1-1
-0.83-0.67
-0.5-0.33-0.17
00.170.33
0.50.670.83
1
Normalized AmplitudeNor
mal
ized
Thr
ough
-thic
knes
s Lo
catio
n A0 guided wave
Experimental Verification
Longitudinal wave
Flexural wave
seek LIGHT | The University of Adelaide Slide 14
Experimental Verification
0 100 200 300 400 500 600
-1.5-1
-0.50
0.51
1.5
Time (µs)
Norm
alize
d Am
plitu
de
Flexural wave reflected from damage
0 100 200 300 400 500 600
-1.5-1
-0.50
0.5
11.5
Time (µs)
Norm
alize
d Am
plitu
de
Flexural wave reflected from damage
0 100 200 300 400 500 600 -1.5
-1 -0.5
0 0.5
1 1.5
Time ( m s)
Nor
mal
ized
Am
plitu
de Incident longitudinal wave
Longtitudinal wave rebounded from left beam end
Longtitudinal wave reflected from damage
seek LIGHT | The University of Adelaide Slide 15
Experimental Verification
Amplifier Function generator
Oscilloscope
Computer
Positioning system
Laser head
Piezoceramictransducer
Beam
Backing mass
Step damage
Ld
LLdd
(Left end)
(Right end)
seek LIGHT | The University of Adelaide Slide 16
0 1 2 3 4 5 6 7x 10-4
-1
-0.5
0
0.5
1
Time (sec)
Nor
mal
ized
am
plitu
de
b)
0 1 2 3 4 5 6 7x 10-4
-1
-0.5
0
0.5
1
Time (sec)
Nor
mal
ized
am
plitu
de
a)
Incident wave Waves reflected from damage
Incident wave
Reflected wave rebounded from beam end
Waves reflected from damage
Reflected wave rebounded from beam end
Experimental Verification
Case C1
Case C2
0 1 2 3 4 5 6 7x 10-4
-1
-0.5
0
0.5
1
Time (sec)
Nor
mal
ized
am
plitu
de
b)
0 1 2 3 4 5 6 7x 10-4
-1
-0.5
0
0.5
1
Time (sec)
Nor
mal
ized
am
plitu
de
a)
Incident wave Waves reflected from damage
Incident wave
Reflected wave rebounded from beam end
Waves reflected from damage
Reflected wave rebounded from beam end
Blue line: Experimental data Red line: Spectral finite element simulation with identified damage parameters
seek LIGHT | The University of Adelaide Slide 17
Normalized marginal PDF of the identified damage length and depth for a) Case C1 and b) C2
Experimental Verification
seek LIGHT | The University of Adelaide Slide 18
• A method has been proposed to provide quantitative identification of damage in beams using longitudinal guided wave
• The method is able to identify damage location and size • Frequency-domain spectral finite element has been
employed to improve the computational efficiency • The proposed method is also able to quantify the
uncertainties associated with the damage identification results
• The proposed method has been experimentally verified • The proposed method is currently extending to address
the multiple damages situation and structures with complicated configurations
Conclusions
seek LIGHT | The University of Adelaide Slide 19
Thank You!