Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and...
Transcript of Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and...
![Page 1: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/1.jpg)
Sinkhole Detection and Characterization with
2-D and 3-D Full Waveform Tomography
by
Khiem Tran, Ph.D.
Department of Civil and
Environmental Engineering
Clarkson University
3-D Sinkhole Imaging Workshop
Gainesville, FL
10/2017
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
200
400
600
800
1000
200
400
600
![Page 2: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/2.jpg)
2
Outline of presentationNeed for sinkhole detection
FWI motivation
FWI challenges at geotechnical scales
Overview of FWI methods
2-D waveform tomography method
• Methodology
• Synthetic data application
• Florida sinkholes, Ohio abandoned mine voids
3-D waveform tomography method
• 3-D FWI using Adjoint gradient
• 3-D FWI using Gauss-Newton
• Synthetic data application
• First field data application
Conclusion
![Page 3: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/3.jpg)
3
Need of sinkhole detection
Sinkhole collapses
Sinkhole problem Structural collapses that lead to
significant property damage and even fatalities
Site investigation Typical invasive testing SPT, CPT
– tests < .1% of material
Seismic methods can test over large volume of materials
Soil/rock property and stratigraphy, and embedded voids/anomalies
![Page 4: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/4.jpg)
4
FWI Motivation
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
Observed data
time
(s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
Estimated data
time
(s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
Residual
time
(s)
Receiver position (m)
measured
synthetic
Vp, Vs
Most conventional seismic methods analyse travel times of certain wave types
• inversion of P-wave first arrival travel time
• inversion of surface wave dispersion
• migration
• use only phase, not magnitude
FWI is wave-equation based and has the potential to
• use full information content (waveforms), both phase and magnitude
• consider all measured wave types (P-, S-, Rayleigh waves)
• characterize both Vp and Vs at high resolution (meter pixel)
![Page 5: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/5.jpg)
5
FWI challenges at geotechnical scales
inconsistent wave excitation, unknown source
signatures (inversion artifacts near source
locations)
strong variability of near surface soil/rock, poor
priori information (shallow inversion artifacts,
local minimum)
dominant Rayleigh waves, small body waves
with strong attenuation (large model updates at
shallow depths, poorly resolved deeper
structures)
![Page 6: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/6.jpg)
Overview of full waveform inversion
6
Inversion method:
Forward modeling d = f(m) • 2-D and 3-D elastic wave equations
dest = f(mest)
Model updating to match dest ͌ d• Global optimization: simulated annealing,
genetic algorithm
• Deterministic optimization: Gradient, Newton,
Gauss-Newton methods
Vs, Vp (model m)
?
Inverse problem
Seismic
testing
Measured
wave field
d
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25
0
5
10
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25
0
5
10
400
600
800
1000
1200
200
400
600
800
![Page 7: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/7.jpg)
7
2-D FWI
zxt
v
zxt
v
zzxzz
xzxxx
1
1
• Eq. governing particle velocity:
Forward modeling
x
v
z
v
t
x
v
z
v
t
z
v
x
v
t
zxxz
xzzz
zxxx
2
2
• Eq. governing stress tensor:
Tran K.T. and Hiltunen D.R. (2012), “Two-Dimensional Inversion of Full Waveform Using Simulated Annealing”,
Journal of Geotechnical and Geoenvironmental Engineering.
PML
No PML
• Perfectly Matched Layer (PML) at
bottom and 2 vertical boundaries
![Page 8: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/8.jpg)
8
2-D FWI
Model updating by Gauss-Newton
Tran K.T., McVay M., Horhota D., and Faraone M. (2013), “Sinkhole Detection Using 2D Full Seismic Waveform
Tomography”, Geophysics.
ddm t
2
1)(E
Residual wave field:
Misfit function:
m)FddmFd ()( ,,,,, kijikijiji
di,j and Fi,j (m): measured and estimated
data
di,k and Fi,k (m): reference traces from
measured and estimated data
Source-independence inversion
0 0.2 0.4 0.6 0.8-0.5
0
0.5
1
Time (s)
Mag
nitu
de
0 10 20 30 40 500
50
100
Frequency (Hz)
Mag
nitu
de
![Page 9: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/9.jpg)
9
2-D FWI
Model updating
Tran K.T., McVay M., Horhota D., and Faraone M. (2013), “Sinkhole Detection Using 2D Full Seismic Waveform
Tomography”, Geophysics.
Step length:
Model updating: ,[ 21
1dmm ttttnnn J IIPPJJ -1]
Jacobian matrix: ,)()(
,
,,
,
p
ki
jiki
p
ji
i,jmm
mFdd
mF J
].)([[
,][][
])([][
21 dmF
dmF
nttttn
nttnt
ntntn
g
gg
g
J IIPPJJ
JJ
J
1-]
Filter, focus, balance gradient vector
![Page 10: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/10.jpg)
10
Gauss-Newton vs Adjoint Gradient Method
True model
Distance (m)
De
pth
(m
)
Vs, m/s
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Vp, m/s
0 5 10 15 20 25 30 35
0
5
10
15
300
400
500
600
0
100
200
300
Initial model
Distance (m)
De
pth
(m
)
Vs, m/s
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Vp, m/s
0 5 10 15 20 25 30 35
0
5
10
15
0
200
400
600
0
100
200
300
Distance (m)
De
pth
(m
)
Vs, m/s
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Vp, m/s
0 5 10 15 20 25 30 35
0
5
10
15 520
540
560
580
600
620
150
200
250
300
GN inverted at
first iteration
Distance (m)
De
pth
(m
)
Gradient Vs
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Gradient Vp
0 5 10 15 20 25 30 35
0
5
10
150
5
10
15
x 10-17
0
5
10
x 10-16
dtJ
Distance (m)
De
pth
(m
)
Gauss-Newton Vs
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Gauss-Newton Vp
0 5 10 15 20 25 30 35
0
5
10
15 -20
0
20
40
60
80
0
50
100
150
d tttt J IIPPJJ -1]21[
Gradient inverted
at first iteration
Distance (m)
De
pth
(m
)
Vs, m/s
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
De
pth
(m
)
Vp, m/s
0 5 10 15 20 25 30 35
0
5
10
15
500
550
600
200
250
300
![Page 11: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/11.jpg)
Data Acquisition
on top of void
sources & geophones at
1 to 3 m spacing
10-20 lb. sledgehammer
or Propelled energy
generator (5-50 Hz
signals)
P-, S-, and Rayleigh
waves are all recorded
11
S-waveP-wave
![Page 12: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/12.jpg)
Data Analysis
Start analysis at lowest
frequencies and move up
Low frequencies (large
wavelengths) require less
detailed information of initial
model
Adding high frequency data
gradually helps to resolve
variable near surface
structures
12
Misfit function
![Page 13: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/13.jpg)
13
Synthetic Test on Embedded Void
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
Observed data
Tim
e (
s)
Receiver position (m)
Shot 1
Shot 13
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
Observed data
Tim
e (
s)
Receiver position (m)
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
Test configuration
• 24 receivers at 1.5 m spacing
• 25 shots at 1.5 m spacing
![Page 14: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/14.jpg)
14
Synthetic Test on Embedded Void
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
Initial model
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
True model
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
5 Hz
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
10 Hz
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
15 Hz
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
15
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
15
0
500
1000
1500
0
200
400
600
800
20 Hz
Tran K.T., McVay M., Horhota D., and Faraone M. (2013), “Sinkhole Detection Using 2D Full
Seismic Waveform Tomography”, Geophysics.
![Page 15: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/15.jpg)
15
Sinkhole Detection in Florida
Search for Sinkholes
dry retention pond in
Newberry, FL
fine sand and silt,
underlain by highly
variable limestone
top of limestone varies
from 2 m to 10 m in
depth
no indication of voids on
the ground surface
25 lines (A to Y) at 3 m
spacing
![Page 16: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/16.jpg)
16
Search for Sinkholes 10 testing lines at 3 m apart
(line K, L, M, N, O, P, Q, R,
S, and T)
each line 36 m long
24 geophones at 1.5 m
spacing
25 shots at 1.5 m spacing
20 lb. sledgehammer for
source
Newberry, FL
![Page 17: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/17.jpg)
17
Data Analysis
Power spectrum
Frequency (Hz)
Ra
yle
igh
Wa
ve
Ve
locity (
m/s
)
5 10 15 20 25 30
100
200
300
400
500
600
700
800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Initial model
Distance (m)
De
pth
(m
)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
Distance (m)
De
pth
(m
)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
400
500
600
700
800
200
250
300
350
400
4 inversion runs at 6, 10, 15, and 20 Hz central
frequencies
![Page 18: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/18.jpg)
18
Results of Line P
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
200
400
600
800
1000
200
400
600
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Observed data
Tim
e (
s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Final estimated data
Tim
e (
s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Final residual
Tim
e (
s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Observed data
Tim
e (
s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Final estimated data
Tim
e (
s)
Receiver position (m)
0 10 20 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Final residual
Tim
e (
s)
Receiver position (m)
![Page 19: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/19.jpg)
19
Results of Line Q
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25 30 35
0
5
10
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25 30 35
0
5
10
200
400
600
800
1000
200
400
600
0
1
2
3
4
5
6
7
8
0 10 20 30 40
Dep
th (
m)
SPT N
Tran K.T., McVay M., Horhota D., and Faraone M. (2013), “Sinkhole Detection Using 2D Full Seismic
Waveform Tomography”, Geophysics.
Void
![Page 20: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/20.jpg)
20
Abandoned mines in Ohio
Problem
8,000 abandoned mines, 1,200
lane miles of Ohio’s highway
system underlain by mine voids
Significant risk to the health and
safety of the traveling public
Refraction tomography, GPR,
Resistivity, and Micro gravity often
fail, because mine voids are deep
(40-60 ft in depth)
Subsidence pit on I-70 (Crowell, 2010)
Subsidence stabilization
![Page 21: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/21.jpg)
21
US33, Athens, OH
Search for abandoned
mine voids
located at the edge of a
large abandoned mine
complex (no mine map)
overburden is
interbedded clay shales
and sandstones,
variable bedrock
![Page 22: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/22.jpg)
22
US33, Athens, OH
Search for
abandoned mine
voids
• Land-streamer of 120 ft.
length
• 24 geophones at 5 ft.
spacing
• Propelled energy
generator (PEG 40 kg)
• 2 lines of about 1000 ft.
each
Land-streamer
Propelled
Energy
Source
Operator
Controlled
![Page 23: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/23.jpg)
23
Results: US33, Athens, OH
Distance (ft)
De
pth
(ft)
S-Wave Velocity (m/s)
400 450 500 550 600 650 700
0
20
40
60
Distance (ft)
De
pth
(ft)
P-Wave Velocity (m/s)
400 450 500 550 600 650 700
0
20
40
60 500
1000
1500
2000
200
400
600
800
1000
1200
1400
Sullivan B., Tran K.T, and Logston B. (2016), “Characterization of Abandoned Mine Voids Under Roadway Using
Land-streamer Seismic Waves”, Journal of Transportation Research Board
![Page 24: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/24.jpg)
24
Results: US33, Athens, OH
![Page 25: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/25.jpg)
25
3-D FWI
3,2,1,
jiwheref
xt
vi
j
iji
jix
v
x
v
t j
i
k
kij
if2
jix
v
x
v
t i
j
j
iij
if
Forward modeling
by 3-D wave equations
PML is used at bottom and 4 vertical boundaries.
Nguyen D.T. and Tran K.T. (201x), “Site Characterization with 3-D Elastic Full Waveform Tomography”, Geophysics,
under review.
![Page 26: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/26.jpg)
26
3-D FWI
Model updating by Adjoint Gradient
Displacement residual:
𝐸 𝒎 =1
2Δ𝒖𝑡Δ𝒖, where Δ𝒖 = Δ𝑢𝑖,𝑗 , 𝑖 = 1, . . , 𝑁𝑆, 𝑗 = 1, . . , 𝑁𝑅 1
Misfit function:
Δ𝑢𝑖 ,𝑗 𝑡 = 𝐹𝑖 ,𝑗 𝒎, 𝜏 𝑑𝜏
𝑡
0
− 𝑑𝑖 ,𝑗 𝜏 𝑑𝜏
𝑡
0
Gradients for Lame parameters:
δ𝜆 = − 𝑑𝑡 𝜕𝑢𝑥
𝜕𝑥+
𝜕𝑢𝑦
𝜕𝑦
𝜕𝜓𝑥
𝜕𝑥+
𝜕𝜓𝑦
𝜕𝑦 +
𝜕𝑢𝑥
𝜕𝑥+
𝜕𝑢𝑧
𝜕𝑧
𝜕𝜓𝑥
𝜕𝑥+
𝜕𝜓𝑧
𝜕𝑧 +
𝜕𝑢𝑦
𝜕𝑦+
𝜕𝑢𝑧
𝜕𝑧
𝜕𝜓𝑦
𝜕𝑦+
𝜕𝜓𝑧
𝜕𝑧
𝑇
0
𝑁𝑆
𝑖=1
δμ = − 𝑑𝑡
𝜕𝑢𝑥
𝜕𝑦+
𝜕𝑢𝑦
𝜕𝑥
𝜕𝜓𝑥
𝜕𝑦+
𝜕𝜓𝑦
𝜕𝑥 +
𝜕𝑢𝑥
𝜕𝑧+
𝜕𝑢𝑧
𝜕𝑥
𝜕𝜓𝑥
𝜕𝑧+
𝜕𝜓𝑧
𝜕𝑥 +
𝜕𝑢𝑦
𝜕𝑧+
𝜕𝑢𝑧
𝜕𝑦
𝜕𝜓𝑦
𝜕𝑧+
𝜕𝜓𝑥
𝜕𝑧
+2 𝜕𝑢𝑥
𝜕𝑥
𝜕𝜓𝑥
𝜕𝑥+
𝜕𝑢𝑦
𝜕𝑦
𝜕𝜓𝑦
𝜕𝑦+
𝜕𝑢𝑧
𝜕𝑧
𝜕𝜓𝑧
𝜕𝑧
𝑇
0
𝑁𝑆
𝑖=1
Nguyen D.T. and Tran K.T. (201x), “Site Characterization with 3-D Elastic Full Waveform Tomography”, Geophysics,
under review.
![Page 27: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/27.jpg)
27
3-D FWI
Model updating by Adjoint Gradient
Gradients for Vs, Vp: 𝛿𝑉𝑃 = 2𝜌𝑉𝑃𝛿𝜆
δ𝑉𝑆 = −4𝜌𝑉𝑆δ𝜆 + 2𝜌𝑉𝑆δμ
Regularization: δ∗𝑉𝑃 = 𝑅𝑉𝑃𝐿𝑉𝑃 + δ𝑉𝑃
δ∗𝑉𝑆 = 𝑅𝑉𝑆𝐿𝑉𝑆 + δ𝑉𝑆
Model update:𝑉𝑃
𝑛+1 = 𝑉𝑃𝑛 − 𝛼𝑃δ
∗𝑉𝑃
𝑉𝑆𝑛+1 = 𝑉𝑆
𝑛 − 𝛼𝑆δ∗𝑉𝑆
Conditioning Gradients:
tampering to suppress large gradient values near source and receiver
locations
tapering to linearly increase the gradient scales with depth to better resolve
deeper structures
![Page 28: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/28.jpg)
28
3-D FWI
Model updating by Gauss-Newton
ddm t
2
1)(E
Velocity residual:
Misfit function:
jijiji ,,, )( dmFd
Model updating: ,[ 21
1dmm ttttnnn J IIPPJJ -1]
Jacobian matrix:p
ji
i,jm
)(, mF J
Gauss-Newton inversion is done in frequency domain to reduce
RAM
nt
l
ttlωlΔt)u(u1
),(exp),(~ 1 xx
![Page 29: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/29.jpg)
29
3-D FWI: Synthetic test
24 x 36 x 18 m model,
4.5x4.5x4.5 m at 9 m depth
Test configuration
• 8x12 (96) receivers at 3 m
spacing
• 9x13 (117) shots at 3 m
spacing
0
10
200
10
20
30
0
10
x-axis [m]
Vs [m/s]
y-axis [m]
z-a
xis
[m
]
0
10
200
10
20
30
0
10
x-axis [m]
Vp [m/s]
y-axis [m]
z-a
xis
[m
]
100
200
300
400
500
600
700
200
400
600
800
1000
1200
![Page 30: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/30.jpg)
30
3-D FWI: Synthetic test
2 inversion runs at 15 and 25 Hz central frequencies
about 40 hours for both Adjoint gradient and Gauss-
Newton inversions on a desktop computer (32 cores
of 3.46 GHz each and 256 GB of memory)
Initial model used for both
Adjoint and GN inversion
![Page 31: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/31.jpg)
31
3-D FWI: Synthetic test results
Adjoint gradient
Gauss-Newton
![Page 32: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/32.jpg)
32
3-D FWI: plane comparison at void center
Gauss-
Newton
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
Vp [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
200
400
600
200
400
600
800
1000
1200
True
model
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
Vp [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
200
400
600
200
400
600
800
1000
1200
Initial
model
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
Vp [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
200
400
600
200
400
600
800
1000
1200
Adjoint
gradient
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15 200
400
600
Vp [m/s]
x-axis [m]
z-a
xis
[m
]
0 10 20 30
0
5
10
15
200
400
600
800
1000
1200
![Page 33: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/33.jpg)
33
dry retention pond in
Gainesville
test area of 36 x 9 m
96 receivers located
in 24 x 4 grid
52 shots located in
13 x 4 grid
48 geophones twice
PEG active source
3-D FWI:
Field data
0 5 10 15 20 25 30 35
0
2
4
6
8
10
x-axis [m]
y-a
xis
[m
]Test site configuration
Stage 1
Stage 2
![Page 34: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/34.jpg)
34
Sample field data
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Tim
e [s]
Receiver Number
0 5 10 15 20 25 30 35 400
10
20
30
40
50
60
70
80
90
Mag
nitu
de
Frequency [Hz]
measured data
combined from
the two stages
for 96-channel
shot gather
consistent wave
magnitudes and
propagation
pattern
![Page 35: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/35.jpg)
35
3D FWI: Field
data analysis
Power spectrum
Frequency (Hz)
Ra
yle
igh
Wa
ve V
elo
city
(m
/s)
5 10 15 20 25 30 35 40 45 50
100
200
300
400
500
600
700
800
900
1000
2 inversion runs
at 12 and 22 Hz
central
frequencies
About 30 hours
for both Adjoint
gradient and
Gauss-Newton
methods
Initial model
0
5
0
10
20
30
0
10
x-axis [m]
Vs [m/s]
y-axis [m]
z-a
xis
[m
]
0
5
0
10
20
30
0
10
x-axis [m]
Vp [m/s]
y-axis [m]
z-a
xis
[m
]100
150
200
250
300
350
400
450
500
200
300
400
500
600
700
800
900
![Page 36: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/36.jpg)
36
3D FWI: Field data analysis
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
Receiver number
Tim
e [s
]
Estimated data
Observed data
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
Receiver number
Tim
e [s
]
Estimated data
Observed data
Waveform comparison for 2 sample shots
![Page 37: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/37.jpg)
37
3D FWI: Field data results
0
5
0
10
20
30
0
10
x-axis [m]
Vs [m/s]
y-axis [m]
z-ax
is [
m]
0
5
0
10
20
30
0
10
x-axis [m]
Vp [m/s]
y-axis [m]
z-ax
is [
m]
100
150
200
250
300
350
400
450
500
200
300
400
500
600
700
800
900
Adjoint gradient
Gauss-Newton
0
5
10 0
10
20
30
0
10
x-axis [m]
Vs [m/s]
y-axis [m]
z-a
xis
[m
]
0
5
10 0
10
20
30
0
10
x-axis [m]
Vp [m/s]
y-axis [m]
z-a
xis
[m
]
100
200
300
400
500
200
400
600
800
![Page 38: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/38.jpg)
38
3D FWI: Field data results at planes
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500SPT-1
y = 0 m
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
y = 3 m
SPT-2
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
y = 6 m
SPT-3
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
y = 9 m
SPT-4
Adjoint GradientGauss-NewtonVs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
Vs [m/s]
x-axis [m]
z-a
xis
[m
]
0 5 10 15 20 25 30 35
0
5
10
15
100
200
300
400
500
![Page 39: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/39.jpg)
39
3D FWI vs. SPT results
![Page 40: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/40.jpg)
40
Conclusion
Both Vs and Vp can be characterized at high resolution (meter pixel) to 20 m in depth by 2-D and 3-D FWI methods
Buried void can be identified to a depth of about 3 void diameters with surface measurement
Gauss-Newton provides better results than Adjoint gradient inversion method, particularly for sinkhole/void imaging
![Page 41: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/41.jpg)
Future work
41
3-D viscoelastic waveform tomography
• Account for material damping
• Extract more material properties: seismic
attenuation Qp, Qs
3-D adaptive (non-uniform) mesh waveform
tomography
• Begin with uniform mesh to identify low-velocity
anomalies
• Use refine mesh only at the anomalies to extract
more detailed information
![Page 42: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/42.jpg)
Acknowledgments
42
Presented research is funded by FDOT, ODOT, NSF, FHWA
Research team:
Michael McVay, Dennis Hiltunen, Scott Wasman (UF), David Horhota (FDOT), Khiem Tran (Clarkson)
Graduate students at Clarkson: Trung Nguyen, Brian Sullivan, Duminidu Siriwardane, Justin Sperry, Majid Mirzanejad, Amila Ambegedara
![Page 43: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/43.jpg)
43
References Nguyen D.T. and Tran K.T. (201x), “Site Characterization with 3-D Elastic Full
Waveform Tomography”, Geophysics, under review.
Tran K.T. and Luke B. (2017), “Full Waveform Tomography to Resolve Desert
Alluvium”, Soil Dynamics and Earthquake Engineering, Vol. 9, pp. 1-8.
Sullivan B., Tran K.T, and Logston B. (2016), “Characterization of Abandoned
Mine Voids Under Roadway Using Land-streamer Seismic Waves”, Journal
of Transportation Research Board, Vol. 2580, pp. 71-79.
Tran K.T., McVay M., Horhota D., and Faraone M. (2013), “Sinkhole Detection
Using 2D Full Seismic Waveform Tomography”, Geophysics, Vol. 78 (5), pp.
R175–R183.
Tran K.T. and McVay M. (2012), “Site Characterization Using Gauss-Newton
Inversion of 2-D Full Seismic Waveform in Time Domain”, Soil Dynamics and
Earthquake Engineering, Vol. 43, pp. 16-24.
Tran K.T. and Hiltunen D.R. (2012), “Two-Dimensional Inversion of Full
Waveform Using Simulated Annealing”, Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 138(9), pp. 1075-1090.
![Page 44: Sinkhole Detection and Characterization with 2-D and 3-D ... · Sinkhole Detection and Characterization with 2-D and 3-D Full Waveform Tomography by Khiem Tran, Ph.D. Department of](https://reader034.fdocuments.in/reader034/viewer/2022043012/5fa8d02b7f8d9c71c242b346/html5/thumbnails/44.jpg)
44
Thank You!
Distance (m)
Dep
th (
m)
S-Wave
0 5 10 15 20 25
0
5
10
Distance (m)
Dep
th (
m)
P-Wave
0 5 10 15 20 25
0
5
10
400
600
800
1000
1200
200
400
600
800