Do Models Really Work? Verification and Validation in ... Buttlar.pdf · Verification and...
-
Upload
nguyenngoc -
Category
Documents
-
view
232 -
download
6
Transcript of Do Models Really Work? Verification and Validation in ... Buttlar.pdf · Verification and...
Do Models Really Work? Verification and Validation in Pavement Engineering
Road MaterialsandPavementDesign
Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
William G. Buttlar, PhD, PE
Professor of Civil and Environmental Engineering
University of Illinois at Urbana-Champaign
Road MaterialsandPavementDesign
Pavement Complexity: Closed-Form Solutions do Not Consider:
>Damage
>Propagating Cracks
>Complex Interface Behavior
>Viscoelasticity, Visco-elasto-
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
>Viscoelasticity, Visco-elasto-plasticity
>Material Gradients
>Effects of Multiple Wheel Loads for Nonlinear Problems
>Complex Load Spectra
Road MaterialsandPavementDesign
Simulation in Pavement Design and AnalysisSelected Historical Developments
>Layered Elastic Analysis: BISAR, CHEVRON, JULEA, LEAF, ILLIPAVE, MICHPAVE, KENLAYER (1970-1990)
>TCMODEL (SHRP Thermal Cracking Model ~ 1993)
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
>TCMODEL (SHRP Thermal Cracking Model ~ 1993)
>PCASE, LEDFAA, FAARFIELD (Airfield Pavement Design ~ 1990’s - Present)
>MEPDG (Mechanistic-Empirical Pavement Design Guide) Software (2002 - Present)
Road MaterialsandPavementDesign
Numerical Modeling/Simulation
>Allows for Complexity in Material Properties, Loading Configurations, and Material Response/Distress
>The Time is Now!
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
>The Time is Now!
� Moore’s Law (double in computing power every 1.5 years)
� IBM Blue Waters at University of Illinois (Petascale computing)
Road MaterialsandPavementDesign
As Engineers, How Can We Handle the Black Box???
Out
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
???In Out
Phases of Modeling and Simulation
From Oberkampf et al. (Appl. Mech. Rev., vol. 57, no. 5, Sept. 2004)
Phases of modeling and simulations and the role of V&V
Model Qualification
‘‘Determination
of adequacy of the
conceptual model to
provide an
acceptable level of
agreement for the
Road MaterialsandPavementDesign
Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
From Oberkampf et al. (Appl. Mech. Rev., vol. 57, no. 5, Sept. 2004)
agreement for the
domain of intended
application.’’
Model Verification
The process of
determining that a
model implementation
accurately represents
the developer’s
conceptual description
of the model and the
From Oberkampf et al. (Appl. Mech. Rev., vol.
57, no. 5, Sept. 2004)
of the model and the
solution to the model.
Road MaterialsandPavementDesign
LEA Example for Model Verification
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
Khazanovich and Wang, Trans. Res. Rec., Vol.
2037, pp. 63-75, 2007.
Model ValidationThe process of
determining the
degree to which a
model is an accurate
representation of the
real world from the
perspective of the
Road MaterialsandPavementDesign
Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
From Oberkampf et al. (Appl. Mech. Rev., vol.
57, no. 5, Sept. 2004)
perspective of the
intended uses of the
model.
Validation ProcessHomogeneous Tests
(Direct Tension,
Annular Shear, etc.)
Bench scale, Possibly
Standard Tests
(SE(B), DC(T), SCB, etc.)
Road MaterialsandPavementDesign
Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
From Oberkampf et al. (Appl. Mech. Rev., vol.
57, no. 5, Sept. 2004)
Intermediate Scale/
Structural/ Simulation Tests
(Overlay Tester, Layered
Beam, MMLS, etc.)
APT and Field Sections
(ATLaS, FABAC, HVS, MLS,
Mn/ROAD, LTPP, etc.)
Road MaterialsandPavementDesign
Verification and Validation of Code is Important!!!
Computerized Controls for Braking, Accelerating, Crash Avoidance
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201012
Accelerating, Crash Avoidance
Simulation for Pavement Design; What Will be Our Legacy?
Road MaterialsandPavementDesign
Calibration vs. Validation
In Mechanistic – Empirical modeling, which is commonly needed in pavement engineering, we have empirical constants
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201013
engineering, we have empirical constants which require calibration. Thus, model calibration is generally needed between verification and validation.
!!! Calibration and Validation are often confused in pavement engineering !!!
Road MaterialsandPavementDesign
MEPDG Example for Model Calibration
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010From: MEPDG User’s Manual, Chapter 3 (2006).
Road MaterialsandPavementDesign
SHRP Example for Model Validation - TCMODEL
OBSERVED CRACKING
ZERO MED HIGHLOW
ZERO
LOW
PR
ED
ICT
ED
CR
AC
KIN
G16
2
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010From: SHRP A-357 Final Report (1993).
LOW
MED
HIGH
PR
ED
ICT
ED
CR
AC
KIN
G
4
11
2
2
The statistical ‘PRESS’ procedure was used to independently validate model results when only limited field sections was available.
Road MaterialsandPavementDesign
UIUC Fracture Modeling
Verification, Calibration, and Validation of Cohesive Zone Model
RILEM CAP TG-03
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201016
Example #1 – Fracture Modeling
1. Use DCB for Verification
2. DC(T) for Calibration, and
a
H
2δ
P
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
2. DC(T) for Calibration, and
3. SC(B) for Validation
17
Disk-Shaped Compact Tension, DC(T) Test
>ASTM D7313-07
>Loading Rate:
�CMOD = 1.0-mm/min
>Measurements:
�CMOD
�Load
18
Semi-circular Bend, SC(B) Test
>Loading Rate:
�CMOD = 0.7-mm/min
>Measurements:
�CMOD
�Load
19
Fracture Energy Results (Illinois RILEM Mix)
Peak Loads (CoV shown in parenthesis)
DCT: 2.68 kN (7.5%)
SCB: 6.03 kN (18%)
20
Verification of CZ Modeling Approach using
Double Cantilever Beam Problem
21
a
H
2δ
P
Road MaterialsandPavementDesign
FE Meshes for Fracture Energy Tests
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
CZ Elements
(~ 0.8 mm)
22
DCT Load-CMOD Curves: Measured vs. Calibrated
Calibration Factors:
DCT CMOD Energy: 0.84
IDT Strength: 1.00
23
Predicted SCB Load-CMOD Curves (Validation)
24
� Friction model
Discrete Element Method (DEM)
nr FF µ=
A
B
sF
sFnr FF µ=
C
D
C
D
sF
nS
SS
B
A
P
nU∆
� Central difference scheme for solving the dynamic problem
Example #2: Discrete Element Modeling
25
SS
* PFC2D 3.0, ITASCA, 2002
fG
σ
maxσ
sepδ δ
Softening
pU
Micro-cracking Macro-cracking
Unloading and Reloading
m1
m2Friction Model
HardeningCohesive Softening Model
E
∑=
−−+=1
/
0)1()(
i
t
i
ieDDtDτ
Where, D(t) = Creep compliance at time t
Di = Material constants
τi = Retardation times
Viscoelastic Contact Model
26
τi = Retardation times
−+−±= ++ tt
k
tttDfu
A
Buu
Cf m)1(
1 11
k
K
C
tKA
21
∆+=
k
K
C
tKB
21
∆−=
mmkC
t
KAC
tC
2
1
2
∆++
∆=
mmkC
t
KAC
tD
2
1
2
∆+−
∆=
Where,
* Contact force f t+1 can be calculated from known values for ut+1, ut, , and f tt
ku
−++=
−
)1(11
)( 1
110
0
τ
ησε
t
KMMe
G
t
Gt
−+=
−−−
)(1
)( 11
1 )(
11
10
ττ
ησε
ttt
KMee
G
tt
10 tt <<
tt <1
: Creep
: Recovery
Verification of Creep Response
27
0.0E+00
2.0E-08
4.0E-08
6.0E-08
8.0E-08
1.0E-07
0 4 8 12 16 20
Time (sec)
Ve
rtic
al S
tra
in (
t)
Analytical Solution
DEM
+=
−−10
1000)(ττε
tK
tM
eGeGAtF
Relaxation
Verification of Relaxation Response
28
0.0E+00
2.0E+05
4.0E+05
6.0E+05
8.0E+05
0 4 8 12 16 20
Time (sec)
Co
nta
ct F
orc
e (
t) (
N)
Analytical Solution
DEM
Verification of Mesh-Independence
The cohesive zone model introduces
a length scale in terms of material
properties such as Young’s modulus,
fracture energy, and material strength
(Rice, 1968).
A fracture process zone where damage
occurs can be estimated as:
29
(a) Comparison of Different Cohesive Particle Sizes
: 0.25mm, 0.5mm, 1.0mm
2218ave
c
c
GEl
συ
π
−=
lc >> Particle (or element) size
The cohesive model results will be
independent of the particle size if it is
much smaller than lc.
Road MaterialsandPavementDesign
Example #3: Size Effect Study
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201030
Kim, H., Partl, M.N., Wagoner, M.P., and W.G. Buttlar, “Size Effect Investigation on Fracturing of Asphalt Concrete Using the Cohesive Softening Discrete Element Model,” 7th International RILEM Symposium on Advanced Testing and Characterization of Bituminous Materials, Rhodes, Greece, 2009.
Road MaterialsandPavementDesign
DEM Model Details
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201031
Road MaterialsandPavementDesign
Calibration – Intermediate Size
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201032
Road MaterialsandPavementDesign
Validation of Size Effect Prediction
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201033
Small Specimen
(D=100mm)
Large Specimen
(D=300mm)
Dark Symbols = Model Solid Lines = Experiments
Road MaterialsandPavementDesign
Making Simulations Robust> Recognizing that is often very difficulty to fully verify or validate
a highly complex simulation model, models can be improved by:
� Being thorough in verification…leave no submodel unverified, and
conducting systems-level verification against other codes. Re-
verifying code when changes are made.
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
verifying code when changes are made.
� Not confusing calibration and validation; Using independent
validation data sets following calibration.
� Validating across entire range of anticipated model use;
documenting un-validated model ‘regions’
� Providing ‘rules of thumb’ and ‘reality checks’ (example h/L ratio for
beams). Providing sample problems and verified solutions.
34
Road MaterialsandPavementDesign
Summary>When Developing or Adopting Numerical Models for
Pavement Analysis and Design, the Models should
be: Qualified, Verified, Calibrated, and Validated
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 2010
>Consistent Use of These Terms and Properly
Following These Techniques During Model
Development and Adoption are Essential Steps
Towards the Reliable use of Simulation Models in
Pavement Analysis and Design
35
Road MaterialsandPavementDesign
Thanks for your attention!!!
Road Pavements: Materials, design and performance
Lisboa, LNEC, 25 March 201036