You are on page 1of 4

Sivakugan, N. & Johnson, K. (2004). Géotechnique 54, No.

7, 499–502

TECHNICAL NOTE

Settlement predictions in granular soils: a probabilistic approach


N. S I VAK U G A N * a n d K . J O H N S O N †

KEYWORDS: footings/foundations; design; settlement used 22 different methods: the most popular ones among the
consultants were the Schmertmann et al. (1978) and Burland
& Burbidge (1985) methods, whereas the academics pre-
INTRODUCTION ferred finite element methods.
It is believed that, for typical shallow foundations in gran- Settlements computed by different methods had been com-
ular soils, settlement considerations are more critical than pared against measured settlements by several researchers
bearing capacity considerations, especially when the founda- (Jeyapalan & Boehm, 1986; Tan & Duncan, 1991;
tion width is greater than 1.5 m (Jeyapalan & Boehm, 1986; Papadopoulos, 1992; Berardi & Lancellotta, 1994; Sivakugan
Tan & Duncan, 1991; Berardi & Lancellotta, 1994). The et al., 1998). Substantial scatter was reported for all meth-
settlement prediction exercise in Texas, USA, in 1994 clearly ods. The scatter may be due to the randomness associated
demonstrated the inadequacy of the current state of the art with the soils, to inaccuracies in the prediction model, or to
of settlement prediction (Briaud & Gibbens, 1994). In spite the soil stiffness derived from the penetration resistance
of having abundant soil data through extensive laboratory using empirical correlations. Berardi et al. (1991) attributed
and in-situ tests, the predictions were quite poor. In a typical some of these differences to the fact that in most cases the
design situation, where the engineer has only access to very settlements were so small that they were not measured
limited soil data, the problems can only be compounded. precisely. In addition, some of the settlement records re-
Probabilistic/reliability studies and risk assessments have ported in these analyses used estimated soil pressures
become increasingly popular in geotechnical engineering (D’Appolonia et al., 1968). In view of the significant scatter
over the past few decades. Geotechnical engineers will in the predictions, Sivakugan et al. (1998) proposed artificial
continue to use their preferred settlement prediction meth- neural networks as an effective tool for predicting settle-
ods, but some guidance on the risk associated with the ments of shallow foundations in granular soils. Once trained
predictions will be very valuable. The objective of this note by the data from literature, the method gave predictions
is to propose a simple probabilistic model with design better than most traditional methods.
charts, for four different settlement prediction methods, The three most popular settlement prediction methods are
which enable the designer to quantify the probability that those proposed by Terzaghi & Peck (1967), Schmertmann
the settlement will exceed a specific limiting value. For the et al. (1978), and Burland & Burbidge (1985). In these three
same input data, the settlement predictions can be quite methods the soil stiffness is taken into account through the
different depending on the method employed. At present penetration resistance from the standard penetration test in
there is no rational procedure for comparing these different the form of blow count, or from the cone penetration test in
predictions. The probabilistic design charts proposed herein the form of cone resistance. The same stiffness would be
will provide a baseline from which all settlement prediction used for any strain level. Noting the fact that the soil
methods can be compared against each other. stiffness decreases with strain level, Berardi & Lancellotta
(1994) proposed a new iterative procedure whereby it is
possible to account for the strain level beneath the founda-
CURRENT STATE OF THE ART tion and use appropriate values for the soil stiffness.
Traditional settlement prediction methods recognise that
applied soil pressure, soil stiffness and foundation width are
the most important factors affecting settlements. Two other STATISTICAL ANALYSIS
factors that contribute to a lesser extent are the shape and Sivakugan & Johnson (2002) carried out a statistical
depth of the foundation. The most difficult part in the analysis on settlements predicted by the above four methods.
settlement prediction exercise is assigning appropriate values They showed that the settlement ratio, defined as the ratio of
for the soil stiffness. predicted to actual settlements, follows a beta distribution,
The poor state of the art for prediction of settlements in and presented histograms along with the beta distribution
granular soils is well documented in the literature. Douglas parameters. The probability density function of a random
(1986) reported the existence of more than 40 different variable X following a beta distribution is given by
methods for predicting settlements in granular soils. There is ˆ(Æ þ ) 1
still no unanimous agreement on which method is the best. f (X ) ¼ (x  a)Æ1 (b  x)1 ,
ˆ(Æ)ˆ( ) (b  a)Æþ1
Geotechnical engineers often have their preferred methods
based on their beliefs. For example, in the Settlement ’94 for a < x < b (1)
prediction session held in Texas, the 31 international experts
where ˆ is the gamma function, defined as
ð1
Manuscript received 2 December 2003; revised manuscript accepted
ˆ( ) ¼ x 1 ex dx
25 March 2004. 0
Discussion on this paper closes on 1 April 2005, for further details
see p. ii.
* Department of Civil and Environmental Engineering, James Cook Here a and b are the minimum and maximum possible
University, Townsvillle, Australia. values for x. The advantage of the beta distribution is that it
† Coffey Geosciences Pty Ltd, Townsville, Australia. allows lower and upper limits and skewness to be associated

499
500 SIVAKUGAN AND JOHNSON
with the random variable X, and enables a more realistic PROBABILISTIC DESIGN CHARTS
distribution to be fitted to a given set of data. Further, The probability of failure, pf , can be defined as the
previous studies by Berardi & Lancellotta (1994) showed probability that the actual settlement will exceed the speci-
clearly that settlement ratios are better modelled by the beta fied limiting value. When the predicted settlement is S, and
distribution than by the log-normal or other distributions. the measured settlement is Y, the settlement ratio X (¼ S/Y)
Using the parameters given by Sivakugan & Johnson is a random variable modelled by the probability distribu-
(2002), and the procedures described by Harr (1977) for tions given in equations (2)–(5). If the limiting value is
computing gamma functions, the following probability den- 25 mm, the probability of failure, pf , is given by
sity functions were determined for the four settlement pre- pf ¼ p[Y > 25] ¼ p[S=X > 25] ¼ p[X < S=25]
diction methods:
For example, when the settlement predicted using the Terza-
Berardi & Lancellotta: ghi & Peck method is 35 mm,
: :
f (x) ¼ 0:0499x 0 21 (4:33  x)1 65 for 0 < x < 4:33 (2) pf ¼ p[X < 35=25] ¼ p[X < 1:4]
Burland & Burbidge: Using the BETADIST function in Microsoft Excel 2000,
0:09 2:03 with the beta distribution parameters for the Terzaghi &
f (x) ¼ 0:0142(x  0:16) (6  x) Peck method given in Sivakugan & Johnson (2002), pf ¼
for 0:16 < x < 6:00 (3) 0.35. Here, the BETADIST function gives the probability of
X being less than the specified value, which is 1.4.
Terzaghi and Peck:
For a series of different values of S, and different limiting
: : settlement values, pf values were determined, and this was
f (x) ¼ 0:0251(x  0:55)0 51 (14:29  x)0 77
repeated for all four settlement prediction methods. The
for 0:55 < x < 14:29 (4) values of pf were plotted against the predicted settlement, as
Schmertmann et al.: shown in Figs 2–5. From the settlement predicted by any
: :
one of these four methods, one can obtain the probability
f (x) ¼ 0:0046(x  0:39)0 29 (11:3  x)1 72 ( pf ) that the actual settlement will be greater than a limiting
for 0:39 < x < 11:3 (5) value, using this chart. Limiting values in a wide range of
15–40 mm are considered in developing the design charts.
Here x is the value taken by the settlement ratio X, which is As the 25 mm limit is the most common, this curve is
a random variable. The lower and upper bounds, a and b, shown darker. Most of the discussion hereafter, unless stated
are simply the minimum and maximum settlement ratios otherwise, is based solely on the limiting settlement of
observed within the settlement records. These probability 25 mm.
density functions are plotted along with the histograms in The computed settlement value can have quite a different
Fig. 1. All four probability distributions are of reverse-J meaning depending on the settlement prediction method
shape, strongly skewed to the right, taking the maximum used. For the same predicted settlement, the probability of
value of 1 at x ¼ a. exceeding the limiting settlement can vary significantly,

1·4 66·08 1·4 66·08


Probability density function

1·2 56·64
Probability density function

1·2 56·64
1 47·2 1 47·2
Frequency

Frequency
0·8 37·76 0·8 37·76
0·6 28·32
0·6 28·32
0·4 18·88
0·4 18·88
0·2 9·44
0·2 9·44
0 0
0 1 2 4 6 8 10 12 14 0 0
Predicted settlement/Actual settlement 0 1 2 4 6 8 10 12 14

(a) Predicted settlement/Actual settlement

(b)
1·4 66·08
1·4 66·08
Probability density function

1·2 56·64
Probability density function

1·2 56·64
1 47·2
1 47·2
Frequency

Frequency

0·8 37·76
0·8 37·76
0·6 28·32
0·6 28·32
0·4 18·88
0·4 18·88
0·2 9·44
0·2 9·44
0 0 0 0
0 1 2 4 6 8 10 12 14 0 1 2 4 6 8 10 12 14
Predicted settlement/Actual settlement Predicted settlement/Actual settlement
(c) (d)

Fig. 1. Histograms and beta distributions for the settlement ratios: (a) Terzaghi & Peck (1967); (b) Schmertmann et al. (1978); (c)
Burland & Burbidge (1985); (d) Berardi & Lancellotta (1994)
SETTLEMENT PREDICTIONS IN GRANULAR SOILS 501
1·0 1·0

0·9 0·9

0·8 0·8
Allowable settlement limit ⫽ 15 mm

p [actual field settlement ⬎ allowable limit]


p [actual field settlement ⬎ allowable limit]

0·7 0·7
20 mm

0·6 0·6
25 mm

0·5 Allowable settlement limit ⫽ 15 mm 0·5


30 mm

20 mm 35 mm
0·4 0·4
40 mm
25 mm
0·3 30 mm 0·3
35 mm
40 mm 0·2
0·2

0·1 0·1

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Predicted settlement: mm
Predicted settlement: mm

Fig. 4. Design chart for Burland & Burbidge (1985) method


Fig. 2. Design chart for Terzaghi & Peck (1967) method

1·0 1·0

0·9 0·9 Allowable settlement limit ⫽ 15 mm

0·8 0·8 20 mm
p [actual field settlement ⬎ allowable limit]
p [actual field settlement ⬎ allowable limit]

25 mm
0·7 0·7
30 mm

0·6 0·6 35 mm
Allowable settlement limit ⫽ 15 mm 40 mm
0·5 0·5

0·4 20 mm 0·4

25 mm
0·3 0·3
30 mm
35 mm
0·2 40 mm 0·2

0·1 0·1

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Predicted settlement: mm Predicted settlement: mm

Fig. 3. Design chart for Schmertmann et al. (1978) method Fig. 5. Design chart for Berardi & Lancellotta (1994) method

depending on the selected method. This is illustrated in probability that the actual settlement in the field will exceed
Table 1, where probability values of the field settlement 25 mm.
exceeding 25 mm are shown for the four methods. For It can be seen from Figs 2–5 that Terzaghi & Peck’s
example, a 20 mm settlement computed by the Terzaghi & method is the most conservative of the four. For example,
Peck or Schmertmann et al. method implies a 0.20 (20%) when the method predicts 25 mm settlement, there is only a
probability that the actual settlement in the field is greater 0.26 (26%) probability that the actual settlement will exceed
than 25 mm. If the same settlement of 20 mm is estimated 25 mm. On the other hand, the more sophisticated Berardi &
using Berardi & Lancellotta method, there is a 0.52 (52%) Lancellotta method gives predictions closer to the actual
502 SIVAKUGAN AND JOHNSON
Table 1. Probability of exceeding 25 mm settlement in the field
Predicted Probability of exceeding 25 mm settlement in field
settlement:
mm Terzaghi Schmertmann Burland & Berardi &
& Peck et al. Burbidge Lancellotta
1 0 0 0 0.06
5 0 0 0.03 0.19
10 0 0.02 0.15 0.32
15 0.09 0.13 0.25 0.43
20 0.20 0.20 0.34 0.52
25 0.26 0.27 0.42 0.60
30 0.31 0.32 0.49 0.66
35 0.35 0.37 0.55 0.72
40 0.387 0.42 0.61 0.77

values. Here, a 25 mm prediction implies that there is a 0.60 ACKNOWLEDGEMENTS


(60%) probability that the actual settlement will exceed This research was funded by the Australian Research
25 mm. For predicted settlements up to about 40 mm the Council and DETYA, and their financial support is gratefully
Terzaghi & Peck and Schmertmann et al. methods have acknowledged.
similar levels of probability that the actual settlement in the
field will exceed 25 mm.

CONCLUSION REFERENCES
The design of shallow foundations in granular soils is Berardi, R. & Lancellotta, R. (1994). Prediction of settlements of
generally governed by the settlement criteria rather than by footings on sand: accuracy and reliability. Proceedings of Settle-
the bearing capacity. Typically, the foundations are designed ment ’94, 1, 640–651.
Berardi, R., Jamiolkowski, M. & Lancellotta, R. (1991). Settlement
such that the settlements are limited to 25 mm. There is of shallow foundations in sands: selection of stiffness on the
substantial uncertainty associated with the different settle- basis of penetration resistance. Geotechnical Engineering Con-
ment prediction methods; most of the time the predicted gress 1991, Geotechnical Special Publication 27, 185–200.
settlement greatly exceeds the actual settlement in the field. Briaud, J.-L. & Gibbens, R. M. (1994). Predicted and measured
Unless the uncertainty is quantified, there is a tendency for behaviour of five spread footings on sand, Geotechnical Special
designers to design the foundation conservatively. Therefore Publication 41. Texas: ASCE.
any attempt to quantify the uncertainty in the different Burland, J. B. & Burbidge, M. C. (1985). Settlement of foundations
settlement prediction methods will be well received by on sand and gravel. Proc. Instn Civ. Engrs 78, No. 1, 1325–
geotechnical engineers and researchers. 1381.
D’Appolonia, D. J., D’Appolonia, E. & Brissette, R. F. (1968).
Using an extensive database of settlement records, it is
Settlement of spread footings on sand. J. Soil Mech. Found.
shown in this paper that, for the four settlement prediction Div., ASCE 94, No. SM3, 735–760.
methods reviewed, the ratio of predicted to measured settle- Douglas, D. J. (1986). State-of-the-art. Ground Engng 19, No. 2,
ment, defined as the settlement ratio, follows a reverse 2–6.
J-shaped beta distribution with a strong skew to the right. Harr, M. E. (1977). Mechanics of particulate media. New York:
Using the beta distribution parameters, separate design charts McGraw-Hill.
were developed for the four methods, which enable the Jeyapalan, J. K. & Boehm, R. (1986). Procedures for predicting
designer to estimate the probability that the actual settlement settlements in sands. In Settlements of shallow foundations on
in the field will exceed a certain limiting value. In addition cohesionless soils: design and performance, (ed. W. O. Martin)
to the traditional 25 mm limiting value of settlement in pp. 1–22. Seattle: ASCE.
Papadopoulos, B. P. (1992). Settlements of shallow foundations
granular soils, other values from 15 mm to 40 mm limits are on cohesionless soils. J. Geotech. Engng, ASCE 118, No. 3,
considered, which gives the user an option, especially in the 377–393.
case of unconventional structures where different settlement Schmertmann, J. H., Hartmann, J. P. & Brown, P. R. (1978).
limits may be recommended. Improved strain influence factor diagrams. J. Geotech. Engng
The proposed design charts are based on settlement Div., ASCE 104, No. 8, 1131–1135.
records representing a wide range of footings, well docu- Sivakugan, N. & Johnson, K. (2002). Probabilistic design chart for
mented in the literature. In using the design charts, it should settlements of shallow foundations in granular soils. Aust. Civ.
be assumed that the reliability and uncertainty associated Engng Trans. CE 43, 19–24.
with the current input parameters, including the field data, Sivakugan, N., Eckersley, J. D. & Li, H. (1998). Settlement predic-
tions using neural networks. Aust. Civ. Engng Trans. CE40,
are comparable to the settlement records. The uncertainties
49–52.
associated with the different settlement methods have been Tan, C. K. & Duncan, J. M. (1991). Settlement of footings on
built into the proposed design chart, which would become a sands: accuracy and reliability. Proc. Geotech. Engng Cong.
very valuable tool in the design of shallow foundations. The 1991, Colorado 1, 446–455.
probabilistic approach proposed herein can be extended to Terzaghi, K. & Peck, R. B. (1967). Soil mechanics in engineering
include other settlement prediction methods as well. practice, 2nd edn. New York: John Wiley & Sons.

You might also like