Upload
ngokiet
View
223
Download
2
Embed Size (px)
Citation preview
3D Crack Detection Using an XFEM Variant and GlobalOptimizaion Algorithms
K. Agathos1 E. Chatzi2 S. P. A. Bordas1,3,4
1Research Unit in Engineering ScienceLuxembourg University
2Institute of Structural EngineeringETH Zurich
3Institute of Theoretical, Applied and Computational MechanicsCardiff University
4Adjunct Professor, Intelligent Systems for Medicine Laboratory, The University of WesternAustralia
June 1, 2016
K. Agathos et al. XFEM based crack detection 1/6/2016 1 / 21
Outline
1 Inverse problem formulation
2 Global enrichment XFEM
3 Parametrization and constraints
4 Numerical examples
5 Conclusions
K. Agathos et al. XFEM based crack detection 1/6/2016 2 / 21
Background
Nondestructive evaluation (NDE)Methods used to examine an object, material or system without impairingits future usefulness
Available Techniques:impedance tomography,radiography, ultrasounds,acoustic emission
SHM - Damage Detection:Monitor changes in the dynamicproperties of a structure
K. Agathos et al. XFEM based crack detection 1/6/2016 3 / 21
Inverse problem
→ Detection of cracks in existing structures
→ Measurements are available
→ A computational model is employed
→ The difference between the two is minimized
→ Information regarding the cracks is obtained
K. Agathos et al. XFEM based crack detection 1/6/2016 4 / 21
Inverse problemMathematical formulation:
Find βi such thatF (r (βi ))→ min
where
βi Parameters describing the crack geometry
r (·) Norm of the difference between measurements and computedvalues
F Some function of the residual
The CMA-ES algorithm is employed to solve the problem.K. Agathos et al. XFEM based crack detection 1/6/2016 5 / 21
Inverse problem
Solution process:
→ Generation of initial population (βi ) with CMA-ES
→ Fitness function (F (r (βi ))) evaluation using XFEM andmeasurements
→ Population is updated with CMA-ES
→ The procedure is repeated until convergence
K. Agathos et al. XFEM based crack detection 1/6/2016 6 / 21
Inverse problem
During the optimization proccess:
A large number of crack geometries is tested
The computational model is solved several times
An efficient and robust method is required
K. Agathos et al. XFEM based crack detection 1/6/2016 7 / 21
XFEM
FEM vs XFEM for fracture:
FEM⇒
XFEM⇒
No crack
Crack 1 Crack 2
K. Agathos et al. XFEM based crack detection 1/6/2016 8 / 21
XFEM
FEM vs XFEM for fracture:
FEM⇒
XFEM⇒
No crack Crack 1
Crack 2
K. Agathos et al. XFEM based crack detection 1/6/2016 8 / 21
XFEM
FEM vs XFEM for fracture:
FEM⇒
XFEM⇒
No crack Crack 1 Crack 2
K. Agathos et al. XFEM based crack detection 1/6/2016 8 / 21
XFEM approximation
XFEM approximation:
u (x) =∑∀I
NI (x) uI︸ ︷︷ ︸FE approximation
+∑∀I
N∗I (x) Ψ (x) bI︸ ︷︷ ︸enriched part
where:
NI (x) are the FE shape functions
uI are the nodal displacements
N∗I (x) are functions forming a PU
Ψ (x) are the enrichment functions
bI are the enriched dofs
K. Agathos et al. XFEM based crack detection 1/6/2016 9 / 21
Enrichment functions
Jump enrichment functions:
H(φ) ={
1 for φ > 0− 1 for φ < 0
Tip enrichment functions:
Fj (r , θ) =[√
r sin θ2 ,√
r cos θ2 ,√
r sin θ2 sin θ,√
r cos θ2 sin θ]
K. Agathos et al. XFEM based crack detection 1/6/2016 10 / 21
XFEM
Some drawbacks of XFEM:
The use of tip enrichment in a fixed area around the crack front(geometrical enrichment) is required for optimal convergence
The use of geometrical enrichment causes conditioning problems
Blending problems the enriched and the standard part of theapproximation
K. Agathos et al. XFEM based crack detection 1/6/2016 11 / 21
Global enrichment XFEM
An XFEM variant is employed which:
Enables the application of geometrical enrichment to 3D
Employs weight function blending
Employs enrichment function shifting
K. Agathos et al. XFEM based crack detection 1/6/2016 12 / 21
Global enrichment XFEM
Special front elements are introduced:
crack surface
crack front
front elementboundary
front node front element
K. Agathos et al. XFEM based crack detection 1/6/2016 13 / 21
Global enrichment XFEM
Front element shape functions:
boundaryfront elementnode
front element
= 2η
= 3η
1−=ξ
= 0ξ
= 1ξ21=ξ
21−=ξ
Ng (ξ) =[1− ξ
21 + ξ
2
]
K. Agathos et al. XFEM based crack detection 1/6/2016 14 / 21
Global enrichment XFEMDisplacement approximation:
u (x) =∑I∈N
NI (x) uI + ϕ (x)∑
J∈N j
NJ (x) (H (x)− HJ)bJ+
+ ϕ (x)
∑K∈N s
NgK (x)
∑j
Fj (x)−∑
T∈N tNT (x)
∑K∈N s
NgK (xT )
∑j
Fj (xT )
cKj
where:
ϕ, ϕ are weight functions
NgK are front element shape functions
HJ ,Fj are nodal values of the enrichment functionsK. Agathos et al. XFEM based crack detection 1/6/2016 15 / 21
Problem parametrization
Elliptical cracks are considered:
x
y
z
0x
2tn
1t
a
b
Parameters:
Coordinates of centerpoint x0 ({x0, y0, z0})
Rotation about the threeaxes θx , θy and θz
Lengths a and b
K. Agathos et al. XFEM based crack detection 1/6/2016 16 / 21
Problem parametrization
Scaling of parameters:
pi = pi1 + pi22 + pi2 − pi1
2 sin(βi10 ·
π
2
)where:
βi are design variables
pi are geometrical parameters of the crack
pi1 , pi2 are lower and upper values for the parameters
K. Agathos et al. XFEM based crack detection 1/6/2016 17 / 21
Penny crack in a cube
Geometry and sensors:
a
xL
yLzL
Sensor locations
4xL
4xL
4xL
4xL 4
yL4
yL4
yL4
yL
4zL
4zL
4zL
4zL
K. Agathos et al. XFEM based crack detection 1/6/2016 18 / 21
Penny crack in a cubeOptimization problem convergence:
evaluations500 1000 1500 2000
fitn
ess
func
tion
10-2
10-1
100
K. Agathos et al. XFEM based crack detection 1/6/2016 19 / 21
Penny crack in a cubeBest solution after different numbers of iterations
Actual crack
Detected crack
Initial guess 500 evaluations 1000 evaluations
1500 evaluations 2000 evaluations
K. Agathos et al. XFEM based crack detection 1/6/2016 20 / 21