Professional Documents
Culture Documents
Low Tang 1997 Probabilistic Slope Analysis Janbu PDF
Low Tang 1997 Probabilistic Slope Analysis Janbu PDF
121-142, 1997
Q 1997 Elsevier Science Ltd. All tights reserved
Printed in Great Britain
PII:SO266-352X(97)00019-0 0266-352X/97 $17.00 + 0.00
ELSEVIER
(Received 30 January 1997; revised version received 27 May 1997; accepted 2 June 1997)
ABSTRACT
INTRODUCTION
dQ (Moment equilibrium)
T = -E tan alt + ht g - ZQ dx (3)
B. K. Low and W. H. Tang 123
in which t = dT/dx. The mobilized shear stress t at the base of a slice was
expressed in terms of the Mohr-Coulomb strength criteria and the global
factor of safety Fs:
r=c+(o-u)tan@
(Limit equilibrium)
FS
where c is the cohesion, o the total stress normal to the base of a slice, u the
average pore water pressure at the base of a slice, and @Jthe angle of friction
of soil. A further equation was given for overall horizontal equilibrium:
XAE=&-E, (5)
From Eqns (l-5), after rearrangement, Janbu proposed the set of equations
shown in Fig. 2 as the working formulae for implementing slope stability
analysis by an iterative procedure.
An illustrative slope stability analysis involving four slices was given in
Fig. 22 and Table 1 of Janbu [l]. The slope has a gradient (vertical : hori-
zontal) 1 : 2 4, and a height of about 13 m. The soil is homogeneous with
a unit weight (y) of 2 tons m-3, a cohesion (c’) of 1 ton m-*, an angle of
friction (4’) such that tan 4’ is 0.67. The pore water pressure at the base of
any slice is assumed to be equal to 0.4~2, where z is the vertical thickness
of the slice. The iteration sequence was clearly explained, and a factor of
safety of 1.48 was obtained after iteration. A computer program is normally
Boundary forces:
E,, T,, at x = a; E,, T,, at x = b
AP
p=Iyz+q+x
(a) B=AQ+(p+t)&tana
k-0 A= AIn
/ n
ZA
(e) F=
E, -Eb +XB
dE
_=-- LIE AEi +LW,,
(h)
dr- Ax- Aq+Aq+,
t=AT
OE) / Ax
A
(0 Cm) o’=p+t--Ztana-u
’ = F(l + tan’ ar)Ax
Fig. 2. Janbu’s equations for generalized procedure of slices. (Notations as in Fig. 1).
needed to conduct such iterations. For instance, Table 2 in Janbu [I], which
involved seven slices, was produced by a computer program. It will be shown
in the next section that a powerful spreadsheet-automated iterative method
can be used to obtain the same solution as Janbu’s examples without the
need for programming on the user’s part.
The example in Janbu’s [l] Table 1 will be solved using the widely available
spreadsheet software Microsoft Excel. Other spreadsheet software packages
are likely to be (or will be) adequate for the task in hand. The point to note is
B. K. Low and W. H. Tang 125
(a) In step 2 the formula for F,-Eqn (e) in Fig. 2Lcannot be entered
until CA and CB are defined. An empty cell (tantamount to a value of
zero) is also not permissible because it will lead to division by zero
when the formulas for n,-Eqn (c>-are entered. Hence a temporary
non-zero value is entered.
(b) When the formula for F, is finally entered at step 5, a circular reference
is formed, because F, is a function of A, and hence of n,, which in turn
is a function of F,. By activating the iteration option and setting the
criterion for halting iteration at maximum change of 1 x 10e6, the F,
value and the four columns of B, A’, n,, and A will blink for a few
seconds until convergence. The converged values at this stage are only
transitional-they are for the condition where the columns from “AE”
to “t” are empty (equivalent to zero), and hence are not the same as
those shown in Fig. 3.
(c) In step 5, in the Tools/Options/Calculation dialog box, the default
maximum iterations is 100, which is adequate for the problem in hand.
The number can be set higher if desired. If necessary, one can also
press a key (F9, for Microsoft Excel) to test if convergence has been
reached.
(d) In step 6, when the formulas-Eqns. (f k(j) of Fig. 2-for AE, E,
AE/Ax, T, and AT are entered, the spreadsheet performs normal
calculations without iteration since no new circular reference is
formed. With the entering of Eqn (k) in the column “t”, a second
126 B. K. Low and W. H. Tang
Y E, Eb
2 0 0
0 5 10 m a
1 1 J
ht
slice# tam Ax P ” c tan4 AQ o tanat
2
3
4
1 1.13
0.5
0.18
-0.04
4.4
11
11
6
5.3
10.1
8.6
2.9
2.12
4.04
3.44
1.16
1
1
1
1
0.67
0.67
0.67
0.67
0
0
0
0
1.2
1.0
1.1
0
0.63
0.33
0.16
-
0F*
1.480
slice# B A z 0’
1 21.39 10.83 0.6639 16.315 10.36 10.36 1.78 -4.394 -4.394 -0.999 1.1 0.938
2
3
4
53.27 52.61 0.9811 53.622 17.04 27.4 0.05 -8.954 -4.56 -0.415
17.99 52.61 1.0475 50.224 -15.94 11.46 -1.612 -3.607 5.3465 0.486
-0.84 15.41 0.9803 15.721 -11.46
91.8 I w x
0 -1.91 0 3.6071 0.6012 7
2.63 4.328
2.99 5.108
1.77 2.412
‘Procedure:
1) Enter Input data in columns “tana”, “Ax”,....... “hr”, and “tan&“.
2) Enter a value of 1 in cell F, initially. (To be changed in step (5) below)
3) Enter formulas for B, A’, n, and A in their respective columns. (Drag
formula of slice#l to autofill the entire column.). All columns from AE
to CY’are blank at this time except for column headings.
4) Enter formulas for summation of the A and B columns.
5) Enter formula for F, in cell F,. Activate Too/s\Options\Ca/cu/ation\iterate,
and set maximum change to le-6.
.6) Enter formulas for columns AE, AE/Ax,T, AT, t, z, and 6’.
Fig. 3. Spreadsheet automated iterations for the case in Fig. 22 and Table 1 of Janbu[ 11.
circular reference is introduced (as shown in the next section) and all
the boxed columns in the spreadsheet again blink for a few seconds
until iteration converges. The converged final results are as shown in
Fig. 3.
B. K. Low and W. H. Tang 127
Note that the boundary values E,, T,, Et,, and Tb are all zero for the case
in hand. When autofilling equations in the columns E and T by dragging,
one should take care not to drag equations into the cells of Et, and Tb.
Extending the proposed spreadsheet approach to more slices and nonhom-
ogeneous soil profile is straightforward. Also, computing the average factor
of safety F, [Janbu’s Eqn (1 lo)] with respect to shear failure along each
interface requires only one additional column in the spreadsheet.
Powerful though the circular iteration is, there is a caveat: a seemingly
harmless error (for example, accidentally entering a zero into the F, cell of
the spreadsheet of Fig. 3) can be propagated rapidly in a spreadsheet that
contains circular references; the spreadsheet may even be paralyzed. Fortu-
nately, the damage is more apparent than real. One can easily restore such a
seemingly paralyzed spreadsheet to its original working condition if one
understands the formula path. The need to conduct the conceptually
straightforward formula-path analysis in order to handle such a potential
snare of circular reference is explained next.
That numerical difficulty may result from error in input data or in formulae
need no explanation. What also needs to be borne in mind is that iterative
calculations among spreadsheet cells connected by circular references will
produce a solution only if the iterative calculations lead to convergence.
With methods of slices as used in slope stability analysis, numerical difficul-
ties have been noted and discussed by Whitman and Bailey [20], Ching and
Fredlund [2 13and Duncan [ 171, among others. Janbu’s GPS is no exception,
as some slip surfaces will not lead to convergence by the iterative method.
When that happens, one can select a different trial slip surface. Prior to
this, however, the spreadsheet-paralyzed by the previous slip surface that
produced divergent iterative results-would need to be restored to normal
conditions. To do this requires an understanding of the formula path, which
is fairly simple.
128 B. K. Low and W. H. Tang
From Janbu’s equations (Fig. 2), the formula path as shown in Fig. 4 is
easily established. Thus F, determines A which in turn determines F,. In like
manner F, determines t (via AE and AT) which in turn determines F,. The
paths from A to F, and from AT to t are marked with a star symbol to
indicate that they are the critical paths, in the sense that if these two paths
were severed there would not be circular references. Temporary severance of
these critical paths is the key to restoring a worksheet that has errors circu-
larly propagated through it. To break a path from AT to t requires only
deleting the formula in column t, thereby rendering t independent of AT.
To illustrate, suppose that after the spreadsheet in Fig. 3 had been set up
correctly as shown, a zero was accidentally entered into the cell of F,. The
columns from n, to (T would be filled with the unintelligible “#” signs due to
the division by zero in ~1, and hence in A and beyond. Simply retyping the
formula in the F, cell would only re-establish the path A -+ F, and thereby
propagate the error back to F,, thereby completing the “cycle of errors”.
(Such predicament of seemingly inextricable error loops may also arise when
the geometry of the slip surface is such that iteration leads to divergence
rather than convergence.) To restore the spreadsheet the two critical paths
A --f F, and AT -+ t (Fig. 4) need to be temporarily broken, as follows:
The spreadsheet will iterate at steps 3 and 4 to produce the original Fig. 3
within seconds. Execution of the above four steps requires only one minute
or less. Alternatively, the four steps can automatically be recorded using any
spreadsheet’s “record-macro” tool, so that whenever restoration of a work-
sheet is necessary it can be done by merely clicking the macro name or the
button assigned to it.
0
Friction angle, 0 (degrees)
in which [RI-’ is the inverse of the correlation matrix. This equation will be
used to set up the ellipsoid in the spreadsheet since the correlation matrix R
displays the correlation structure more explicitly than the covariance matrix C.
The spreadsheet reliability evaluation approach proceeds as follows: (a)
The mean and standard deviations of c, 4 and water pressure coefficient R,
and the correlation matrix are added to Fig. 3. It is assumed that c and 4 are
normally distributed, and negatively correlated with correlation coefficient
pc+ = -0.5. Th e resulting spreadsheet is as shown in Fig, 6; (b) The formulas
within boxes are entered based on the “x values”, which are initially set equal
to the mean values. Subsequent steps are:
(1) Select the entire correlation matrix, and define a name for it, say
“crmat”, using the menu command insert/name/define. Likewise,
define the column vector [(xi - mi)/ai] as “nxv”, say.
132 B. K. Low and W. H. Tang
(2) The formula Eqn (7) of the reliability index, /? = sqrt ([(x- WZ)/O]~
Wb - md)> is typed in a cell: “ = sqrt(mmult(transpose(nxv),
mmult(minverse(crmat), nxv)))“, followed by pressing “Enter” while
holding down the “Ctrl” and “Shift” keys. (“mmult”, “transpose”,
and “minverse” are Microsoft Excel’s built-in spreadsheet functions
for matrix operations.)
(3) Solver is invoked, to “Minimize”/?, “By Changing” the “xvalues”,
“Subject To” the constraints F,(x) < 1, “xvalues” 20, and c# 2 0.
slice# fana Ax
1 1.13 4.4
2 0.5 11
3 0.18 11
4 -0.04 6
slice#
1
2
3
4
Correlation matrix
r
L
x-m
(T
1
J
j” E] E]
Deterministic analysis with search for the critical non-circular slip surface
tan+
R.
1
0.67
0.4
-f
2
Ea
0
Eb
0 rlFS
1.472
.xb,_ Yb YT hii ht
13 0 tana Ax p u C tan+ AQ 0 tanat
13 3.0288 1.415 2.14 3.029 1.2115 1 0.67 0 1.01
11.375 4.7474 0.836 4 7.776 3.1104 1 0.67 0 1.58
9.75 5.4931 0.593 4 10.24 4.0962 1 0.67 0 1.83
8.125 5.6682 0.448 4 11.15 4.4605 1 0.67 0 1.89
6.5 5.458 0.356 4 11.12 4.4465 1 0.67 0 1.82
4.875 4.9065 0.268 4 10.36 4.1458 1 0.67 0 1.64
3.25 3.9268 0.161 4 8.833 3.5333 1 0.67 0 1.31
1.625 2.5319 0.058 4 6.459 2.5835 1 0.67 0 0.84
0 0.312 -0.149 4 2.844 1.1376 1 0.67 0 0.1
0 0 -0.572 0.545 0.312 0.1248 1 0.67 0 l-l
T
s# I3 A’ n, A hElAx 0 t CT'
1 8.824 4.581 0.5475 8.3663 3.141 3.141 2.301 -0.247 -0.247 -0.115
2 22.31 13.55 0.8126 16.689 10.99 14.13 2.514 -4.662 -4.416 -1.104
3 22.33 18.26 0.9396 19.436 9.127 23.26 1.781 -7.952 -3.289 -0.822
4 19.1 20.63 1.0028 20.576 5.119 28.37 0.825 -9.888 -1.936 -0.484
5 15.63 21.48 1.0313 20.829 1.481 29.86 -0.12 -10.48 -0.589 -0.147
6 11.38 21.3 1.0467 20.348 -2.442 27.41 -1.158 -9.532 0.944 0.236
7 6.115 19.93 1.0462 19.046 -6.821 20.59 -2.042 -6.963 2.569 0.642
8 1.701 16.88 1.0228 16.508 -9.512 11.08 -2.453 -3.234 3.729 0.932
9 -2.16 10.68 0.9121 11.711 -10.11 0.966
10 -0.147 0.672 0.5571 1.2056 -0.966
105.1 ] [ 154.7 1
Deterministic analysis, with spreadsheet automated search for the critical slip surface.
xb Yb YT xb Yb YT
( 38 13.00 13 34.13622 13 13
13 32 6 13
11.375 26 4.3 11.375
9.75 24 2.1 9.75
8.125 20 0 8.125
6.5 16 -1.4 6.5
4.075 12 -2.6 4.875
3.25 8 -3.3 3.25
1.625 4 -3.2 1.625
0 0 -0.35127 0
0 -0.54477 0 0
Fig. 8. Coordinates of the initial arbitrary slip surface of Figs 7-10, and an example of a slip
surface that does not converge during iteration.
generated as the slip surface geometry is being varied during the optimiza-
tion process. This is part of the numerical difficulty that could arise with the
method of slices and has been recognised in the literature. An example of a
slip surface that fails to converge using the general procedure of slices is also
given in Fig. 8. When that happens the circular iteration will fail to converge
and the performance of the spreadsheet’s optimization tool will likewise be
affected.
It was noted in connection with Fig. 4 that there are two critical paths
responsible for the circular references: A + F, and AT + t. To avoid a slip
surface that leads to divergence during iteration, a two-stage optimization
strategy has been used. In the first stage the formula in column “t” is deleted;
Solver obtains the slip surface labelled “Temp” in Fig. 7. This was followed
in the second stage by reinstalling the formula in column “t”; Solver then
obtains the slip surface labelled “Final”.
136 B. K. Low and W. H. Tang
The details of implementation are: in the Solver dialog box, the factor
of safety F, was minimized, by changing the shaded cells of the Xb and yb
columns (i.e. Xbo, ybi - Yb9, and xbio), subject to the constraints that:
mfXb&% -iofybl -J+_&% -wXblo G - 0.1, F, >= 0.
Chen and Morgenstern [5] extended the Morgenstern and Price [3] method
of slices by expressing force and moment equilibrium in two equations
involving multiple integrals. These multiple integral equations can be solved
in a spreadsheet (Low et al., to be published), with automatic search for the
critical slip surface using the spreadsheet’s optimization tool. Figure 9 shows
the critical slip surface obtained by Solver, for side force inclination obeying
a half sine function. The critical slip surface obtained using Janbu’s GPS
(from Fig. 7) is also shown for comparison. For the case in hand, the
spreadsheet approach yields practically the same Fs value and the same cri-
tical surface for both methods. Further validation of the critical slip surface
location is provided by trying a different initial slip surface for the case in
Fig. 7; the same final critical slip surface is obtained.
Reliability analysis with search for the critical non-circular slip surface
The spreadsheet set-up for reliability analysis is largely the same as in Fig. 7,
except for the following: (a) the three columns of “xvalues”, “Mean”, and
“StDev” (for standard deviations) are set up for the random variables c, 4,
and R,, as shown in Fig. 10. Initially the “xvalues” are equal to the “Mean”
values; (b) equations (as in Fig. 6) are entered in the columns “u”, “c”, and
“tan #“, based on “xvalues”; (c) the correlation matrix R and the column
vector (x - m)/o are set up (as in Fig. 6, not displayed in Fig. lo), for
Y Ea
2 0
xb Yb YT hi
13 AX P u o ten+
13 2.888 3.61 2.1427 0.9325 0.831
11.375 4 8.969 5.3244 0.9325 0.631
9.75 4 11.36 6.7449 0.9325 0.631
8.125 4 12.06 7.1573 0.9325 0.631
6.5 4 11.75 6.9769 0.9325 0.631
4.875 4 10.7 6.3608 0.9326 0.631
3.25 4 8.889 5.2769 0.9325 0.631
1.625 4 6.295 3.7372 0.9325 0.631
0 4 2.579 1.5309 0.9325 0.831
0 0.323 0.173 0.1026 0.9325 0.631
=T,
E T
s# B A’ n, A AE 0 AE/Ax 0P AT t ‘5 d
9 -2.103 8.592 0.8836 9.7235 -11.83 0.73 -2.905 -0.05 3.5148 0.8787
10 -0.056
109.4] 0.347 0.515 I0.6733
169.4 1 -0.73H&2.259'
0 0&0496'0.'='
Fig. 10. Probabilistic analysis, with spreadsheet automated search for the reliability-based
critical slip surface.
138 B. K. Low and W. H. Tang
Fig. 11. Comparison of deterministic critical slip surface with reliability-based critical slip
surface.
computing the reliability index /I by Eqn (7). The results as shown in Fig. 10
were obtained after optimization using Microsoft Excel’s Solver tool: the
reliability index ,!J was minimized, by changing the “xvalues” column of c, 4,
R,, and the shaded cells of the xb and yb columns, subject to the constraints
as in the previous section, except that the constraint for F, is F, = 1.0, and
with the additional constraints that “xvalues” 2 0, ~‘20. To avoid a slip
surface that leads to divergence during iteration, the same two-stage optimi-
zation strategy as described in the previous section has been used. Different
Solver options were also tried after Solver’s default options stabilized.
Figure 11 shows both the deterministic critical surface and the reliability-
based critical surface.
The examples in Figs 3, 6, 7 and 10 are for slopes with homogeneous
strength parameters c,c$, and pore pressure parameter R,. The proposed
spreadsheet implementation of the deterministic Janbu’s GPS and the
probabilistic first-order reliability method (FORM) is applicable also to
nonhomogeneous slopes and slopes with spatially correlated properties. In the
B. K. Low and W. H. Tang 139
deterministic analysis (Figs 3 and 7) this merely means that the R, (or pressure
head hr), c and tan4 could differ from slice to slice; in the probabilistic
analysis (Figs 6 and 10) this means that the columns of “xvalues”, “Mean”,
“StDev”, and the corresponding correlation matrix are larger, and that
autocorrelation formulas are entered in the correlation matrix, as discussed
earlier. (Solver can handle up to 500 constraints-one upper limit and one
lower limit for each changing cell, plus 100 additional constraints-accord-
ing to Microsoft Excel 7’s on-line help. Each changing cell can be con-
strained by a number, another cell or range reference, or formula.)
ACKNOWLEDGEMENTS
This work was performed while the first author was on sabbatical leave at
the Hong Kong University of Science and Technology, from September to
December 1996. The second author would like to acknowledge the support
of a research grant No. HKUST 722/96E sponsored by the Research Grant
Council of Hong Kong.
REFERENCES
15. Greco, V. R., Efficient Monte Carlo technique for locating critical slip surface.
Journal of Geotechnical Engineering, ASCE, 1996, 122, 5 17-525.
16. Ugai, K. and Leshchinsky, D., Three-dimensional limit equilibrium and finite
element analyses: a comparison of results. Soils and Foundations, 1995, 35, l-7.
17. Duncan, J. M., State of the art: limit equilibrium and finite-element analysis of
slopes. Journal of Geotechnical Engineering, ASCE, 1996, 122, 577-596.
18. Bromhead, E. N., The Stability of Slopes, 2nd edn. Blackie Academic & Pro-
fessional, London, 1992.
19. Abramson, L. W., Lee, T. S., Sharma, S. and Boyce, G. M., Slope Stability and
Stabilization Methods. John Wiley & Sons, New York, 1995.
20. Whitman, R. V. and Bailey, W. A., Use of computers for slope stability analysis.
Journal of Soil Mechanics and Foundations Division, ASCE, 1967, 93, 475498.
21. Ching, R. K. H. and Fredlund, D. G., Some difficulties associated with the limit
equilibrium method of slices. Canadian Geotechnical Journal, 1983,20,661-672.
22. Hasofer, A. M. and Lind, N. C., Exact and invariant second-moment code for-
mat. Journal of Engineering Mechanics, ASCE, 1974, 100, 111-121.
23. Veneziano, D., Contributions to second moment reliability. Research Report
No. R74-33. Department of Civil Engineering, MIT, MA, 1974.
24. Ditlevsen, O., Uncertainty Modeling: With Applications to Multidimensional
Civil Engineering Systems. McGraw-Hill, New York, 198 1.
25. Madsen, H. O., Krenk, S. and Lind, N. C., Methods of Structural Safety. Pre-
ntice Hall, Englewood Cliffs, NJ, 1986.
26. Ang, H. S. and Tang, W. H., Probability Concepts in Engineering Planning and
Design, Vol. II-Decision, Risk, and Reliability. John Wiley & Sons, New York,
1984.
27. Melchers, R. E., Structural Reliability: Analysis and Prediction. Ellis Horwood
Ltd., Chichester, U. K., 1987.
28. Dai, S. H. and Wang, M. O., Reliability Analysis in Engineering Applications.
Van Nostrand Reinhold, New York, 1992.
29. Tichy, M., Applied Methods of Structural Reliability. Kluwer Academic, Dor-
drecht, Boston, 1993.
30. Haldar, A. and Mahadevan, S., First-order and second-order reliability meth-
ods. In Probabilistic Structural Mechanics Handbook, ed. C. (Raj) Sundararajan,
Chapman & Hall, London, 1995, pp. 27-52.
31. Wu, T. H. and Kraft, L. M., Safety analysis of slopes. Journal of Soil Mechanics
and Foundations Engineering, ASCE, 1970, 96, 609-630.
32. Vanmarcke, E. H., Reliability of earth slopes. Journal of Geotechnical Engi-
neering, ASCE, 1976, 130, 1247-1265.
33. Whitman, R. V., Evaluating calculated risk in geotechnical engineering. Journal
of Geotechnical Engineering, ASCE, 1984, 110, 143-188.
34. Li, K. S. and Lumb, P., Probabilistic design of slopes. Canadian Geotechnical
Journal, 1987, 24, 520-535.
35. Mostyn, G. R. and Li, K. S., Probabilistic slope analysis-state of play. Pro-
ceedings of the Conference on Probabilistic Methods in Geotechnical Engi-
neering, Balkema, 1993, pp. 89-109.
36. Tang, W. H., Recent developments in geotechnical reliability. In Probabilistic
Methods in Geotechnical Engineering, ed. K. S. Li and S.-C. R. Lo. A. A. Balk-
ema, Rotterdam, 1993, pp. 3-28.
37. Chowdhury, R. N. and Xu, D. W., Geotechnical system reliability of slopes.
Reliability Engineering and System Safety, 1995, 47, 141-151.
142 B. K. Low and W. H. Tang