february 15, 2005 benchmarking of nonlinear geotechnical ground response analysis procedures peer...

February 15, 2005

Benchmarking of Nonlinear Geotechnical Ground Response Analysis Procedures

PEER Lifelines Project 2G02http://cee.ea.ucla.edu/faculty/jstewart/groundmotions/PEER2G02/

Meeting Overview

• Review results of code usage exercise

• Discuss verification plan

• Other business

Other Business

• Subcontracts– Request for contract revision: Dec. 3 2004– Current status:

Davinder Gabhi (2-11-05):

The contract between PEER and PEA has been sent to our Sponsored Projects Office for formal paperwork and final signatures. This has been approved by the Lifelines Program Manager.

Other Business

• Turkey Flat project– PI is Charles Real of CGS– Workshop Fall 2005– http://www.quake.ca.gov/turkeyflat.htm

• Web posting of code reports

• Reimbursements

Project Overview

• Two-year project July 2004 – June 2006

• Three general tasks:– Develop parameter selection protocols– Verification studies– Parametric studies

• Effects of parametric variability

• Benefits of NL relative to EL and application in PSHA

Schedule

Tasks1. Model parameter selection protocols2. Model verification- small strain3. Model verification - small to moderate strain4. Model verification - large strain5. a - Parametric unc., soil parameters5. b - Parametric unc., other parameters6. Other parametric studies

PI team

Developers working in consultation with PI team

Jul-04 Jun-05 Jul-05 Jun-06

Today’s Agenda

• Review of code usage exercise (Stewart)– Objective and plan for work– Reporting/response protocols– Common issues for all codes– Code specific issues

• Developer presentations– 10-15 min each– Selection of model parameters and input motions– Analysis results and comparison to UCLA results– Reasons for differences

• Verification plan (Stewart)

Objective of Code Usage Exercise

From September meeting minutes:

“One of the urgent needs is to establish protocols for evaluating input parameters and checking that the results provided are “reasonable.” This gets to the issue of how usable the codes are to users other than the code developers. The establishment of those protocols, and demonstrating that they can be used by novice users, is a key first objective of the project.”

Plan for Code Usage Exercise

– Described in “white paper” dated 9/29/04– Developers provide parameter selection and

code use protocols • Information for all codes forwarded to UCLA team

by mid-October, although some codes unusable in initial form

• Parameter values in sand given Vs, , N, ’• Parameter values in clay given Vs, , PI, Su, ’• Parameter uncertainty• List of common errors and unreasonable results

associated with those errors

Difficult for fitting parameters

Plan for Code Usage Exercise

– Novice user (AK, JS) runs codes for example sites

– Developers run codes in parallel– Based on outcome of above: Refine parameter

selection and use protocols, as needed

Code Usage Exercise

Reporting and response protocols:1. UCLA team completes initial report, sends to

developer2. Developer provides feedback, factual errors in initial

reports are corrected3. Final report prepared and returned to developers with

comments for code and/or user manual improvement4. Developer response:

– Agree with comment and will make change– Agree with comment but insufficient time and resources to

make change– Disagree with comment and change will not be made

Would like to post (3) and (4) to project web page – agree?

Code Usage Exercise

STATUS(1)

UCLA Report

(2)

Developer Feedback

(3)

Revised Report

(4)

Developer Response

DMOD_2

DEEPSOIL, v2.5

OpenSees

SUMDES

TESS

Code Usage Exercise

Common issues for all codes:– Use of reference strain (r) in lieu of shear strength

(mo) to describe G/Gmax and curves

– Input motion specification (outcropping versus within)

– Layer thickness criteria

Code Usage Exercise – Common Issues

Reference strain issue• Typical existing

parameters to describe backbone curve:– Gmax

– mo

– Various fitting parameters• Ref. strain definition

r=mo/Gmax

• Problem: – Parameter mo is unknown,

especially at depth, for most sites

– No guidelines in users manuals

1

G m ax m o


Reference strain issue• Possible solution when data on

mo unavailable:– Estimate r using guidelines

from Darendeli and Stokoe or from material specific G/Gmax curve (r where G/Gmax = 0.5)

– Calculate mo as r Gmax

– Then use fitting parameters

• Provides excellent fit (in all codes) to G/Gmax curve

• How does r Gmax compare to mo (when known)

0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

max

0.0001 0.001 0.01 0.1 10

10

20

30

40

Da

mp

ing

Rati

o (

%)

0.0001 0.001 0.01 0.1 1Cyclic Shear Strain (%)

0

0.2

0.4

0.6

0.8

1

G/G

max


0

10

20

30

40

Dam

pin

g R

atio

(%

)

G1 G1

G2 G2

a

r

G

G

1

1

max

r = f(PI, OCR, ’), defined uncertaintya = 0.92


Input motion issue• Two general formulations

– Lumped mass (DMOD, DEEPSOIL)

– Continuum (OpenSees, SUMDES, TESS)

• Extensive email discussion on correct form of input motion when recording is from outcropping site: – Modify recorded outcropping motion to within (using

SHAKE)

– Original outcropping motion


Input motion issue

Walt Silva’s thoughts (2-7-05):

“I still feel an essential issue is outcrop verses total (in layer) motions. It is simply not acceptable to have a nonlinear code that does not treat control motions as outcrop, there is no good reason for this restriction. To treat control motion as total motion, a nonlinear code can treat the control point as a rigid half space. This is exact for this case. To treat the control motion as outcrop, the control point can be taken as a flexible half-space. I hope this gets clarified at the meeting.”


Input motion issue (lumped mass)

– Similar to dynamic response of structure

– Requires total motion at base as input

– From Oct. 2004 correspondence, recommendation was to use SHAKE within motion

– Use of outcropping motion may be preferred (following slides…)

Graphic: Y. Hashash

guIMuCuKuM

Test I

• Treasure Island soil profile

• Linear soil properties

• Input motion: outcrop motion

• Frequency domain analysis– Input at bedrock+ elastic base– Input at bedrock+ rigid base– Input at outcrop+ elastic base– Input at outcrop+ rigid base

Y. Hashash

Result of Test I

0

0.5

1

1.5

2

2.5

3

3.5

0.01 0.1 1 10

Period(sec)

Sa(

g)

Outcrop motion input atoutcrop+rigid base(Frequency)

Outcrop motion input atoutcrop+elasticbase(Frequency)

Outcrop motion input atbedrock+rigidbase(Frequency)

Outcrop motion input atbedrock+elasticbase(Frequency)

1

2

3

4

1. Case 2 is correct case2. In SHAKE, there are two options to input motion. If inputting at outcrop, then rock base is treated as elastic. If inputting at bedrock, then rock base is treated as rigid. Therefore, if choosing bedrock as input, no matter using rigid base or elastic base, the result is the same

Y. Hashash

Test II

• Treasure Island profile• Linear soil properties• Input at bedrock• Time domain analysis

– Input motion: outcrop motion• Input at bedrock+ rigid base• Input at bedrock+ elastic base

– Input motion: within motion (I convert it from outcrop motion)

• Input at bedrock+ rigid base • Input at bedrock+ elastic base

• Two red case should have identical result

Y. Hashash

Result of Test II

(outcrop motion + elastic base) is equal to (within motion + rigid base)

0

0.5

1

1.5

2

2.5

3

3.5

0.01 0.1 1 10

Period(sec)

Sa(

g)

outcrop motion input atbedrock+rigid base (Time)

outcrop motion input atbedrock+elasticbase(Time)

within motion input atbedrock+rigid base(Time)

within motion input atbedrock+elasticbase(Time)

Y. Hashash

Compare time domain and frequency domain analysis

1. Case 2 is correct one2. If we follow rules of “outcrop motion + elastic base” and “within motion + rigid base” doing time domain analysis, we can get almost identical result as frequency domain analysis

0

0.5

1

1.5

2

2.5

3

3.5

0.01 0.1 1 10

Period(sec)

Sa(

g)

Outcrop motion input atoutcrop+rigid base(Frequency)

Outcrop motion input atoutcrop+elasticbase(Frequency)

outcrop motion input atbedrock+elasticbase(Time)

within motion input atbedrock+rigid base(Time)

1

2

3

4

Y. Hashash


Input motion issue (continuum)– Motion transformed to shear stress time history applied

to base of soil column– Wave equation solution implies:

• Input could be specified as incident (1/2 of outcropping)• Reflected calculated as part of solution• Total motion taken as incident + reflected

– Recommendations from Oct. 2004 correspondence• OpenSees: input is ½ of outcrop (??)• SUMDES: input is full outcrop motion• TESS: user specifies full outcrop, code has ½ modifier built in

(??)


Layer thickness issue

• Soil layers cannot propagate waves with f > fmax = Vs/4H.

• Results sensitive to layer thickness, especially at high frequencies

• User’s manuals need to make note of this issue

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Ps

eu

do

Sp

ec

tra

l Ac

ce

lera

tio

n (

g)

Input W ith in M otion a t 328 ft

S urface R esponse Spectrum from SH AKE

S urface R esponse Spectrum from D -M O D _2 (25 Layers)


5% D am ping

Code Usage Exercise – Code Specific Issues

• DMOD_2• DEEPSOIL, v2.5• SUMDES• TESS• OPENSEES


Treasure Island Site

0 1000 2000 3000 4000 5000 6000 7000 8000 9000Shear W ave Ve locity (ft/s)

500

450

400

350

300

250

200

150

100

50

0

De

pth

(ft

)

0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

ma

x

0.0001 0.001 0.01 0.1 1Cyclic Shear Stra in (%)

0

5

10

15

20

25

Dam

ping

Rat

io (

%)

0-44 ft (T1)

> 44 ft (T2)

P I=15 (V&D , 1991)

P I=30 (V&D , 1991)

P I=50 (V&D , 1991)

Source: Darragh and Idriss, 1997


Treasure Island Site: SHAKE results

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)

Input O utcropp ing M otion

W ith in M otion a t 328 ft from SH AK E


5% D am ping

0 2000 4000 6000 8000Shear Wave Velocity (ft/s)

600

500

400

300

200

100

0

Dep

th (

ft)

0 0.2 0 .4 0 .6 0 .8 1Maxim um Shear Strain (% )

600

500

400

300

200

100

0

S H A K E


Gilroy II Site

Source: Darragh and Idriss, 1997

0 1000 2000 3000 4000Shear W ave Velocity (ft/s)

650

600

550

500

450

400

350

300

250

200

150

100

50

0

De

pth

(ft

)

0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

ma

x

0.0001 0.001 0.01 0.1 1Cyclic Shear Stra in (%)

0

10

20

30

Da

mp

ing

Ra

tio (

%)

0-40 ft (G 1)

40-80 ft (G 2)

80-130 ft (G 3)

> 130 ft (G 4)

P I=15 (V&D , 1991)

P I=30 (V&D , 1991)

P I=50 (V&D , 1991)


Gilroy II Site: SHAKE results

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)


W ith in M otion a t 560 ft from SH AK E


5% D am ping

0 1000 2000 3000 4000Shear Wave Velocity (ft/s)

600

500

400

300

200

100

0

Dep

th (

ft)

0 0.02 0.04 0.06 0.08 0.1Maximum Shear Strain (%)

600

500

400

300

200

100

0

S H AKE


DMOD_2• MKZ model overestimates the damping at large strain

• How to trade off between fitting MR vs. damping curves?

• Clearer guidelines for more advanced parameters (gray literature references)0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

max

0.0001 0.001 0.01 0.1 10

10

20

30

40

Da

mp

ing

Ra

tio

(%

)


0

0.2

0.4

0.6

0.8

1

G/G

max


0

10

20

30

40

Dam

pin

g R

atio

(%

) M easured

KZ M odel

M K Z M ode l

G3 G3

G4 G4


DMOD_2: Results

Underprediction at high frequency: Possibly due to simplified Raleigh damping?

Treasure Island Gilroy II

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)




5% D am ping

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from D -M O D _2(48 Layers)

5% D am ping


DEEPSOIL

• Utilizes modified MKZ model – so similar issues with fit of MR and damping curves as with DMOD: – Damping at large strain is overestimated – How to trade off between good fits of MR and damping curves?

• Modified MKZ model includes pressure-dependent coefficients – When use coefficients vs. specifying depth-dependent curves?– Need recommendations for selecting coefficients


DEEPSOIL• Viscous damping

formulation: – 3 possible formulations– Select matching frequencies

that provide good match of the linear time domain and frequency domain solutions

– Examples of good and poor matches needed to assist users

• Issues with equivalent linear model

Extended Rayleigh damping (ERF)

0

1

2

Frequency (Hz)

Eff

ecti

ve d

ampi

ng r

atio

, (

%)

fn

fm

fo

fp

Simplified RF Full RF

Figure from Hashash


DEEPSOIL: results

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from D EEPSO IL

5% D am ping

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pse

ud

o S

pectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from D EEPSO IL

5% D am ping



SUMDES• Used Model 6 for simplified total stress analysis

• Problems matching large strain damping

• Hr fixed at 0.7726 due to r definition

• Viscous damping contribution not included in code-generated damping plot0.0001 0.001 0.01 0.1 1

0

0.2

0.4

0.6

0.8

1

G/G

max

0.0001 0.001 0.01 0.1 10

10

20

30

40

50

Dam

pin

g R

atio

(%

)


0

0.2

0.4

0.6

0.8

1

G/G

max


0

10

20

30

40

Dam

pin

g R

atio

(%

)

M easured

W ang M odel

T1 T1

T2 T2


SUMDES: results

Same viscous damping formulation as DMOD (except match frequency specified as 1 Hz): why results so different?

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)

Input O utcropp ing M otion (w ith e lastic base)


S urface R esponse Spectrum from SU M D ES

5% D am ping

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)

Input O utcropp ing M otion(w ith e lastic base)


S urface R esponse Spectrum from SU M D ES

5% D am ping



TESS• Need to synthesize and update code documentation

• Five possible levels of analysis: we use Level 1

• Guidelines needed for selection of higher-level parameters (which are also required for Level 1 analysis)

• Good match of MR and damping curves0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

max

0.0001 0.001 0.01 0.1 10

5

10

15

20

25

Dam

pin

g R

atio

(%

)


0

0.2

0.4

0.6

0.8

1

G/G

max


0

10

20

30

40D

am

pin

g R

ati

o (

%)

M easured

TES S

T1 T1

T2 T2


TESS: resultsTreasure Island Gilroy II

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from TESS

5% D am ping

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from TESS

5% D am ping


OpenSees• Nonlinear soil curves:

– Can specify MR, damping calculated automatically per Masing– Can adjust MR iteratively to reduce damping error– Issues of trade off between fitting MR vs. damping curves– Pressure-dependent coefficients option – see DEEPSOIL

comments• Viscous damping contribution not included in code-

generated damping plot 0.0001 0.001 0.01 0.1 10

0.2

0.4

0.6

0.8

1

G/G

max

0.0001 0.001 0.01 0.1 10

10

20

30

40

Dam

pin

g R

atio

(%

)


0

0.2

0.4

0.6

0.8

1

G/G

max


0

10

20

30

40D

amp

ing

Rat

io (

%)

M easured

O PEN SEES M odel(use m easured M R curve)

O PEN SEES M O D EL(fit M R and dam ping itera tive ly)

T1 T1

T2 T2


OpenSees• Viscous damping

formulations:– 2 options for Raleigh

damping– Simplified + Full– Guidelines needed

regarding frequencies where damping specified

• Clearer guidelines for parameters of more advanced models

• Documentation needed for new GUI version of code

Full Rayleigh damping (CRF & RF)

0

1

2

Frequency (Hz)

Eff

ecti

ve d

ampi

ng r

atio

, (

%)

fn

fm

Simplified RF

Figure from Hashash


OpenSees: results

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from O PEN S EES

5% D am ping

Treasure Island

0.01 0.1 1 10Period (s)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pseu

do

Sp

ectr

al A

ccele

rati

on

(g

)



S urface R esponse Spectrum from O PEN S E ES

5% D am ping

Gilroy II

Code Usage Exercise

Synthesis of results• Consistently lower PGA

• Amplification at site period relative to SHAKE:– Less for TI

– Similar for Gilroy 2

• Mixed results at mid-periods (between site period and PGA)

0.01 0.1 1 10Period (s)

0

0.2

0.4

0.6

0.8

Sp

ectr

al A

ccele

ratio

n (g

) S H A K E

D -M O D _2

D E E P S O IL V 2.5

S U M D E S

TE S S

O P E N S E E S

Treasure Island

0.01 0.1 1 10Period (s)

0

0.2

0.4

0.6

0.8

Sp

ectr

al A

ccele

ratio

n (g

) S H A KE

D -M O D _2

D E E PS O IL V 2.5

S U M D E S

TES S

O P EN SE E S

Gilroy 2

Tdegraded = 1.04s

Tdegraded = 1.40s

Verification Plan

• Verification of element behavior

• Verification at different strain conditions– Very small strain (visco-elastic)– Small to medium strain– Large strain

• Goodness of fit

Verification of Element Behavior

• Suggested by Kramer

• Apply cyclic load to single element at various rates

• Plot vs. • Look for spurious features

at zero crossing, upon unloading, etc.

• Is this possible with the codes?

Gsec1

Gsec2

Backbone Curve

Initial LoadingCurve

SubsequentLoading & Unloading Curves

Graphic: Hashash

Verification at Very Small Strain

• Why? – Verify wave propagation

part of the codes– Check effects of viscous

damping formulations– Check input specification

procedure• Take linear frequency domain

elastic solution as exact• Compare to time domain

elastic solution• Specified: Vs, Dmin, layer

thicknesses• Vary:

– Profile depth– Layering of Vs

– Depth variation of Dmin

• Pulse and broadband inputs

V s

Dep

th

V s

Dep

th

F ixed D m in

D epth- variable D m in

Verification at Small to Medium Strains

• Site selection criteria: – Should be vertical arrays or nearby

rock/soil pairs– Deep characterization– Range of input motions– Soft and stiff sites– Reasonably well known dynamic

properties• Silva recommended sites:

– Lotung– Port Island (liq.)– Gilroy I, II– Kik Net (inquery made regarding

data resolution)• Others:

– Frasier River, BC – Garner Valley– La Cienega– Turkey Flat

Vs

Verification at Large Strain

• Vertical array data - ?• Centrifuge data

– UC Davis (http://cgm.engineering.ucdavis.edu)

– RPI ?

Verification at Large Strain

Available data• UCD Experiment series

DKS02, DKS03– dense unsaturated sand– Input motions: sine sweeps

and scaled Santa Cruz LP eqk.

• UCD clay experiments– Performed in early 1990s– Refs: Idriss et al. (1994);

Fiegel et al. (1998)– Data available?

Ref: Stevens et al. 1999

4

3

2

1

0

De

pth

in P

roto

typ

e S

cale

(m

)

120 160 200 240S h e a r W a ve Ve lo city (m /s)

Goodness of Fit

Anderson (2004) criterion

• Based on quality of fit for 10 ground motion parameters

• Scores range from 0 to 10 (perfect agreement) for each parameter

• Overall score = average of 10 scores from each parameter

Arias duration

Energy duration

Arias intensity

Energy integral

Peak acceleration

Peak Velocity

Peak Displacement

Response Spectra

Fourier Spectra

Cross Correlation

february 15, 2005 benchmarking of nonlinear geotechnical ground response analysis procedures peer...

Documents

code developers

business slide

code usage exercise

code usage exercise

schedule slide

needed slide

psha slide

code use protocols information