Risk Assessment in Geotechnical Engineering

D.V. Griffiths

Colorado School of Mines, USA

University of Colorado University of California

Boulder Berkeley

ASCE/G-I Orange County Chapter, Education Seminar

Risk Assessment and Mitigation in Geotechnical Practice
February 9th 2018 1
University of California at Santa Cruz

THE BANANA SLUG

2
 “……in earthwork engineering the designer has to
deal with bodies of earth with a complex structure
and the properties of the material may vary from
point to point.”
K. Terzaghi
Prefce to the Inaugural Edition of
Géotechnique (1948)

“Two specimens of soil taken at points a few feet

apart, even if from a soil stratum which would be
described as relatively homogeneous, may have
properties differing many fold.”
Donald W. Taylor
Introduction to Fundamentals of Soil Mechanics
Wiley, (1948)
3
It is only relatively recently however, that methodologies such as the
Random Finite Element Method (RFEM) have been developed to explicitly
model the variability discussed by Terzaghi and Taylor.
Bearing Capacity

Qult

Bearing failure of a silo in

Manitoba, Canada (1913)

4
 Outline

1. Slope Stability Analysis by Finite Elements

• “Seeking out failure”
• Variable soils

2. Risk Assessment in Geotechnical Engineering

• Three levels of probabilistic analysis
 Event Trees
 First Order Methods
 Monte Carlo
• Modeling spatial variability.
The Random Finite Element Method (RFEM)

3. Concluding Remarks
5
 1. Slope Stability Analysis by Finite Elements
• Gravity loads are applied to the mesh. soil is given simple
elastic-perfectly plastic
stress-strain model

σ′ "plastic" (φ ′, c′ )
Coulomb

"elastic" ( E ′,υ ′ )

ε
• Compute elastic stresses and check for elements violating Coulomb

τ
M<0 φ′
(illegal) M=0

=τ σ ′ tan φ ′ + c′
M>0
c′ (elastic)

σ 3′ σ 1′ σ′ 6
 M≈
< 00 • Element with elastic stresses
violating Coulomb (M < 0)

• Stress redistribution while

maintaining global equilibrium

Bishop and Morgenstern (1960)

FS=1.27
SRF
• Strength reduction to failure 0.00 0.25 0.50 0.75 1.00 1.25 1.50
0.00

c′  tan φ ′  FS=1.27
c′f = φ ′f arctan 
0.05

SRF  SRF  0.10

δmax0.15
At failure FS ≈ SRF 0.20

0.25

0.30
7
 “Seeking out failure”

James Bay Dike using Finite Elements

60 18 56.3 18 60
Units in m and kN
12 φ ′ 30,=
= c′ 0,=
γ 20 φu 0,=
= =
cu 41, γ 20
4
8 φu 0,=
= =
cu 34.5, γ 18
6.5 φu 0,=
= =
cu 31.2, γ 20.3

FS=1.27

• Failure mechanism “seeks out” the path of least resistance.

• Slope fails “naturally” through zones where the shear strength is unable to resist the
shear stresses.

8
 Another example with a 2-layer undrained slope.

cu = 60 FS = 1.12

cu = 95

cu = 60 Would a limit equilibrium

FS = 1.14 method find both these
failure mechanisms?
cu = 97.5

cu = 60
FS = 1.16

cu = 100
 2) Risk Assessment in Geotechnical Engineering

FS=1 FS=1

Two slopes with the same factor of safety

WHAT ABOUT THE CONSEQUENCES OF FAILURE?

10
Definition of RISK
Probability of Failure Baecher (1982)
weighted by the
Consequences of Failure
UNACCEPTABLE?

What is ACCEPTABLE?

acceptable
risk?
11

11
 A Risk Assessment study starts with a Probabilistic Analysis

Goal of a probabilistic geotechnical analysis…..?

To estimate the “Probability of failure ( pf )” as an alternative,

or complement to, the traditional “Factor of Safety ( FS )”
Alternatives might be the
“Probability of inadequate performance”
“Probability of design failure”

Some investigators prefer a more optimistic terminology.....e.g.

“reliability”
“reliability (index)”

.....so what, if any, is the relationship between pf and FS ??

 CONSIDER TWO EXAMPLES OF SLOPE STABILITY
Find the factor of safety of a 2H:1V slope shown:

H
1.5H ru = 0

φ=′ 23°
Example 1 c′ FS = 1.5
= 0.048
γH
Solution from charts, e.g. Michalowski (2002),

φ=′ 32°
Example 2 c′ FS = 2.0
= 0.048
γH

….so the slope in Example 2 is “safer”…..?

13
Following a probabilistic analysis we may get more information
on the statistical distribution of the Factor of Safety in these Examples.

Suppose such an analysis reveals that:

for Example 1:
=µ FS 1.5,
= σ FS 0.18

and for Example 2:

=µ FS 2.0,
= σ FS 0.5
14
Consider once more, the two slopes from a probabilistic standpoint

µ FS = 1.5
area under curve = 1 σ FS = 0.18

µ FS = 2
σ FS = 0.5
FS=1
area under curve = 1

15
 µ FS = 1.5
Probability of Failure is given
by the area where FS < 1.0

µ FS = 2

1.0 Factor of Safety

The “safer” slope has a higher “probability of failure”!

As tempting as it is....direct comparison between

the Factor of Safety and the Probability of Failure
should be done with great care.
16
 Geotechnical Analysis: The Traditional Approach
Bearing capacity P
Strip footing

φ′ c′

γB
qult = c′N c + qN q + Nγ e.g. Terzaghi’s
bearing capacity
2 equation

qult

qult
qall =
FS 17
Geotechnical Analysis: The Probabilistic Approach
Bearing capacity P
Strip footing

spatial µtan φ ′ σ tan φ ′ σ c′ µ c′ cross

correlation correlation

γB
qult = c′N c + qN q + Nγ Terzaghi’s
2 bearing capacity
equation

µqult σ qult Fundamental Question

How does variability of
probability tables input affect variability of
output?

p f P [ qult < qall ]

= 18
THREE LEVEL OF PROBABILISTIC ANALYSIS

1. Expert Panel
• Event Trees

2. First Order Methods

• First Order Reliability Methods (FORM)

3. Monte-Carlo
• Single random variable approach (SRV)
• Random Finite Element Method (RFEM)

19
Level 1: Event Trees (e.g. USBR)

Probability of embankment breach

due to foundation liquefaction
p f = 0.7 × 0.3 × 0.0003 × 0.1 × 0.05
= 3.15 × 10−7
Level 1: Event Trees (e.g. USBR)

Probability of embankment breach

due to foundation liquefaction
p f = 0.7 × 0.3 × 0.0003 × 0.1 × 0.05
= 3.15 × 10−7
 Level 2: First Order Reliability Method (FORM)
Probability of
bearing capacity failure Qall = 1200 Square footing
D =1
Units
B=2 in
kN and m
γ = 18
φ ′, c′, γ D Qall
Random q=
all = 2
300
and
2
B
p f P [ qult < 300]
correlated
=
tan φ ′ and c′
Variable Mean St Dev Dist type
c′ 
 ρ = −0.3
4 1 Normal
tan φ ′ 0.577 (300)

0.086 Normal
qult 1071
= =
FS =3.6 (based on mean values) 22
qall 300
 Level 2: First Order Reliability Method (FORM)
Consider a joint probabilty density function of c′ and tan φ ′ that might be used
in a geotechnical stabilty problems of bearing capacity or slope stability.

There is an infinite number of combinations of (c′, tan φ ′)

that might result in failure ( FS = 1) .

A vertical wall cutting across the hill

represents the locus of FS = 1 .

FS = 1
FORM will find
the most likely
values of c′ and tan φ ′ NOT FAILURE
to cause failure. “Most Likely
FS>1
i.e. the values closest Failure Point”
to the top of the hill. FAILURE
FS<1

The probability of failure

is the volume of the hill
on the failure side of the
FS = 1 line 23
First Contours of the
Order Reliability Index
Reliability
Method

tan φ ′
NO FAILURE
FS>1

“Most Likely
Failure Point”

FAILURE
FS<1

c′
24
FORM computes pf as the volume under the hill on the failure side of the straight line
DataSolverSolve

25
26
Level 3: Monte-Carlo (Single Random Variable)

c′ ′= µµc′ ++(Xµσ
tan φ= tan φ ′ X (
c′ c−
ρ c′)( µ 1 φ−′ −ρ tan
′ + Y tan 2
σ φ ′)
)
tan φ ′

etc.

Check
Check Check
ρc′,tan φ ′
µc′ andµσtanc′φ ′ and σ tan φ ′ 27
Compute bearing capacity of each Monte-Carlo simulation

if ( qult < 300,1,0 )

nf

p f = n f ntot
ntot
28
tan φ ′ tan φ ′

ρ c′ tan φ ′ = −0.3 ρ c′ tan φ ′ = 0.3

c′ c′

tan φ ′ tan φ ′

ρ c′ tan φ ′ = −0.8 ρ c′ tan φ ′ = 0.8

c′ c′
The more positive the correlation between c′ and tanφ ′, the higher the p f
29
Level 3: The Random Finite Element Method (RFEM)

• Developed in the 1990s for advanced

probabilistic geotechnical analysis.
• Combines finite element and random field
methodologies in a Monte-Carlo framework.
• Properly accounts for (anisotropic) spatial
correlation structures in soil deposits.
• Properly accounts for element size through
local averaging.
• All programs are open-source.
• Frequent short courses given for ASCE
and internationally
• Now a considerable bibliography on the method
and included in proprietary codes. 30
 Geotechnical Applications
Settlement Mine pillar Stability

spatial
correlation
length

Seepage

Bearing Capacity

Qult

31
Earth Pressures Slope Stability

Laterally
Loaded
Piles

32
33
 3) Concluding Remarks

• The natural variability of geomaterials makes them naturally suited to

analysis using statistical methods

• Numerical discretization methods remain the most powerful methods for

modeling variable soils. In stability analysis, FE “seeks out” the critical
failure mechanism which is essential when dealing with random soils.

• Direct comparison between FS and pf should be done with great care.

• For probabilistic geotechnical analysis, engineers have a toolbox of

methods. Three levels of complexity have been identified, but only
RFEM properly accounts for spatial variability.

All the programs described in this seminar can be downloaded from

www.mines.edu/~vgriffit 34
THANK YOU.

35