Title: Use of FWD data analysis for mechanisticbased pavement analysis and design
1Use of FWD data analysis for mechanistic-based
pavement analysis and design
16th Annual Falli8ng Weight Deflectometer Users
Group Conference Des Moines, Iowa - October 1-2,
2007
- Kasthurirangan Gopalakrishnan,
- Research Scientist
- and
- Halil Ceylan, Assistant Professor
- Civil, Construction and Environmental Engineering
2Acknowledgements
- The support provided by the Iowa DOT for the
Nondestructive Evaluation of Iowa Pavements
research project is greatly appreciated.
3Pavement Design
State-of-Practice
State-of-the-Art
Mechanistic- Empirical
Empirical
Mechanistic
(Schwartz, 2001)
4Empirical vs. Mechanistic-Based Design
Wood Floor Joist Example
P
P
d
d
L
L
(Schwartz, 2001)
5Empirical Design
- Based on results of experiments or experience
- Requires many observations to establish links
between design variables and performance - Not necessary to establish scientific basis of
observed relationships - Examples AASHTO design methodology, CBR
thickness design, R-value based design
6Mechanistic-Based Design
- Combines both mechanistic and empirical aspects
- Mechanistic component involves determining
pavement responses due to loading through
mathematical models - Empirical component relates the pavement
responses to pavement performance - Each key distress type is associated with a
critical pavement response
7Inputs Structure Materials Traffic
Climate Both mean and standard deviations inputs
are required
Selection of Trial Design
Structural Responses (s, e, d)
Performance Prediction Distresses Smoothness
Revise trial design
Design Reliability
Design Requirements Satisfied?
Performance Verification Failure criteria
No
Yes
Mechanistic-based Design Framework and Components
Final Design
8M-E Pavement Design Guide (MEPDG)
General Information
Inputs Traffic Climate Structure
View Results and Outputs
Status and Summary
9How FWD data can best be used with the MEPDG?
- FWD based materials characterization
- Backcalculated pavement layer moduli
- Environmental effects model
- Distress models
- Primary response models
- Stress-dependent characteristics
10Structural Evaluation of Existing Pavements
- Obtain and compile historical data
- First (windshield) survey
- Preliminary data evaluation
- Second (detailed) survey
- Visual distress survey
- Nondestructive testing
- Coring, sampling, and destructive testing
- Laboratory testing
11Nondestructive Testing - FWD
- FWD tests have the following advantages
- Simulate a moving wheel load
- Measure pavement response-deflection basin
- Require no fixed reference
- Relatively fast
12Falling Weight Deflectometer
13FWD Sensor Configuration
12 in.
12 in.
12 in.
12 in.
12 in.
12 in.
12 in.
Loading Wheel Contact Area
Sensor
14FWD Sensor Offsets
Sensor Offsets (in)
Sensor Arrangement
-12
0
8
12
18
24
36
48
60
72
?
?
?
?
?
?
?
Uniform Spacing
?
?
?
?
?
?
?
SHRP Specification
Iowa DOT FWD machine
?
?
?
?
?
?
?
?
?
15Example of a FWD Deflection Basin
P
Load plate
d36
d12
d0
d60
AREA
16FWD-Deflection Testing
- Typical uses of FWD deflection data
- Determine areas with excessive deflections
- Determine variability along project
- Backcalculate pavement material properties
- To compute LTE for JCP
- Void detection under PCC slabs
17Backcalculation Software
- Flexible pavements
- BOUSDEF, EVERCALC, MODULUS, MODCOMP, and others
- Rigid pavements
- Best fit spreadsheet, AREA spreadsheet, and others
18Advanced Approaches to FWD Data Analysis
- ANN based models
- Analyze complex and nonlinear systems
- Nonlinear geomaterials characterization
- State-of-the-art FE-based structural model
- Real-time analysis
- Analyze 100,000 deflection basins in less than 1
second! - Perfect tools for both project specific and
network level FWD testing
19ANN Methodology
INPUTS Pavement Layer Properties(E, h)
Knowledge Database(E, h, Deflections, s, e)
Structural Model Cri. Pavement
Responses(Deflections, s, e)
Neural Network Training
Non-destructive testing (Deflections, h)
Layer Properties Predictions(E, s, e)
Neural Network Models
20Pavement Systems
- Flexible pavements (ILLI-PAVE database)
- Conventional Flexible Pavements (CFP)
- Full Depth Pavements (FD)
- Rigid pavements (ISLAB database)
- Composite pavements (DIPLOMAT database)
21ANN Based Models
D0
D8
D12
D18
D24
D36
D48
D60
4-Deflection ModelsD0,D12,D24,D36 6-Deflection
ModelsD0,D12,D24,D36,D48,D60 7-Deflection
ModelsD0,D8,D12,D18,D24,D36,D60 8-Deflection
ModelsD0,D8,D12,D18,D24,D36,D48,D60
22Training Set Parameters
23ANN Backcalculation Models
Meier and Rix (1994)
24ANN Based Models
Output Vector
Target Vector
Input Vector
Najjar (2000)
25ANN Performance Prediction of EAC
7
ANN Model CFP-E
-(4)-(5-21 kips)
AC
Millions
Testing Set 1,500
6
AAE () 1.03
RMSE 37 ksi
5
(psi)
2
R
0.99
AC
4
3
ANN Predicted E
2
1
Line of Equality
0
0
1
2
3
4
5
6
7
Millions
Given E
AC
26ANN Performance Prediction of EPCC
(x103 psi)
PCC
E
ANN Predictions for
27ANN Performance Prediction of ks
ANN Model 7 Deflection-k
s
1000
Testing Set 1,500
AAE () 0.26
RMSE 2 psi/in
800
(psi/in)
2
R
1.0
s
600
400
ANN Predicted k
200
Inputs D
, D
, D
, D
, D
,
0
8
12
18
24
D
,
D
, h
, h
36
60
AC
PCC
Output k
Line of Equality
s
0
0
200
400
600
800
1,000
Given k
(psi/in)
s
28Evaluation of ANN Models Using Field Data
29EAC Prediction Performance
Iowa-(5-3-2005)Clarke County I-35(CFP - 4DEFL
- 5-21kips)
Millions
6
5
FILTERED DATA
4
EAC (psi)
3
2
1
0
33
34
35
36
37
38
39
Conventional Flexible Pavement
Mileage
30ERi Prediction Performance
16
Carlyle - Staley - Windsor (IL)
Thousands
12
8
ERi (psi) ANN Prediction
4
Carlyle
Staley
Windsor
0
0
4
8
12
16
Thousands
Full Depth Asphalt Pavements
ERi (psi) Statistical Algorithms
31EPCC Prediction Performance
Rigid Pavements
National Airport Pavement Test Facility FWD
Data
32FWD Data Analysis Software
33FWD Data Analysis Software
34FWD Data Analysis Software
35Conclusions
- ANNs can learn complex multi-dimensional mappings
- ANN based models as surrogates for response
analysis models in MEPDG - Backcalculation of stress dependent layer
properties
36Conclusions
- Do not require seed moduli
- Accuracy of ANN depends on quality of FWD data
- FWD-GPR integration
37THANK YOU!