SOILS AND FOUNDATIONS

Vol. 18, No. 4, Dec, 1978

Japanese Society of Soil Mechanics and Foundation Engneering

OBSERVATIONAL PROCEDURE OF
SETTLEMENT PREDICTION
AKIRA AsAOKA *

ABSTRACT

The use of analytical solution of consolidation equation for settlement prediction is not
always effective since such conditions as an initial distribution of excess pore water pressure, drain length, final vertical strain of soils and the coefficient of consolidation are
sometimes quite uncertain in practical engineering problems. For this reason, an observational settlement prediction is freshly presented. First, the linear ordinary differential
equation derived is demonstrated to give a settlement-time relationship.
After that, by
using the difference form of the equation (an autoregressive equation), observational
procedure of settlement prediction is proposed. Two kinds of practical methods are presented. One is a graphical method, the advantage of which is its simplicity. The other
is the method based on the Bayesian inference of the non stationary stochastic process,
which can give a predictive probability distribution of future settlement and then also provide a preliminary theory for reliability-based design of settlement problems.
The proposed methodology is demonstrated to be also applicable for some other special
problems including settlement due to drainage from sand piles.

Key words:

IGC:

cohesive soil, computer application, consolidation, graphical analysis, measurement, settlement, statistical analysis
E2/D5
INTRODUCTION

The one-dimensional consolidation theory is an important contribution of soil mechanics

to practical foundation engineering. Settlement resulting from one-dimensional compression and one-dimensional drainage has been widely recognized to be well explained by this
theory. However, for the theory to be also effective in the case of future settlement prediction, some other engineering judgement has been additionally indispensable.
It is known that a consolidation equation, a partial differential equation of parabolic
type, produces a unique solution when a coefficient of the equation and the initial and
boundary conditions have been prior determined.
This is common knowledge so that the
practical applications of this equation have exhibited a tendency to utilize only this characteristic of the solution. In fact, according to conventional settlement analysis, conditions such as an initial distribution of excess pore water pressure, drain length, final
vertical strain of soils and the coefficient of consolidation were naturally considered to be
given in advance of the analysis. It .is, however, also commonly accepted that the estimation of these conditions usually goes with .a high degree. of uncertainty. Therefore, the

* Research Associate, Department of Civil Engineering, Kyoto University, Sakyo-ku, Kyoto.

Written discussions on this paper should be submitted before October 1, 1979.

NII-Electronic Library Service

ASAOKA

88

engineer working on settlement prediction is usually expected to have had some previous
experience.
In this paper, a new idea of settlement prediction is presented, the philosophy of which
is based on "Observational Procedure." The trend equation of time series data of settlement is derived first from the one-dimensional consolidation equation, after which the
future settlement is predicted by using the past observations. Two kinds of practical methods are proposed. One is a graphical method, the advantage of which is its simplicity.
The other is the method based on the Bayesian inference of a non stationary stochastic
process, which can give a predictive probability distribution of future settlement and then
also provide a preliminary theory for reliability-based design of settlement problems.
The proposed methodology is demonstrated to be also applicable for some special problems including settlement due to drainage from sand piles.

MASTER EQUATION OF SETTLEMENT-TIME RELATION

In this section, the ordinary differential equation derived gives a settlement-time relationship.
As a fundamental equation, Mikasa's equation is adopted, that is
( 1)
m which
e(t, z) : vertical strain (volumetric strain),
t(~O) :time,
z : depth from the top of clay stratum and
cv : coefficient of consolidation.
In Eq. (1) the upper dot, represents time-differentiation and the lower script, z the
differentiation with respect to depth, z.
Even if the coefficients of both permiability and
volume compressibility vary from time to time, the Eq. (1) is still effective when Cv remains constant (Mikasa, 1963). In addition, Eq. (1) makes it easy to express settlement.
For these reasons Eq. (1) is adopted instead of Terzaghi's equation.
The solution of Eq. (1) is expressed here by introducing two unknown functions of
time, T and F, as follows:
e(t, z)

\
= T+_!_(E_t)+J_(
2! Cv
4! Cv

1 ( -zsF, )
+zF+-

3!

Cv

r)+

25
)
1 ( --F
++ ...
5! c/

( 2)

Eq. (2) shows that

T=e(t, z=O)
and
F=ez(t, z=O)
The vertical strain e in Eq. (1) is originally defined in an Eulerian sense, but, in this
paper, e is approximately regarded as a Lagrangean strain for simplicity.
The formula
of settlement prediction by regrading e as an Eulerian strain has already been established
by the author. The results, however, become a complex nonlinear problem and therefore
they are beyond the scope of practical applicability.
The typical two boundary conditions are considered next.
( 1 ) Drainage from both top and bottom boundaries.
Fig. 1 shows this boundary condition, which is formulated as
e(t, z=O) =e

: constant

( 3)

NII-Electronic Library Service

89

PREDICTION OF SETTLEMENT

"C''~f~ -.,~~,~;'
free drainage

free drainage

:l.-.-. -~~~'~t~,~~~:.

bottom of clay

.... ,. .... ......... ' .... -

fr~~ d~air{;cj;;

bottom of clay

nm

impermeable layer

Fig. 1. Drainage from top and

bottom boundaries

Fig. 2. Upward drainage

e(t, z=H) =f: constant

where H represents the thickness of clay stratum and both
ary values respectively.
Substitution of Eq. (3) into Eq. (2) gives

(4)

e and

_,

the unknown bound-

T=e : constant

( 5)

which shows that upper half of the solution (2) becomes constant.
Then, from Eq. (4),
we have the linear ordinary differential equation of unknown F with constant coefficients,

F+_!_(H2 P)+_!__(H4 P)+ .. = .-e

3!

c/

5!

Cv

( 6)

( 2) Upward drainage (Fig. 2)

When the bottom of the clay layer 1s impermiable, boundary conditions are given as
follows:
e(t, z=O) =e : constant
tz(t, z=H)

( 7)

=0

( 8)

Since Eq. (7) is the same as Eq. (3), it also follows that
. T= e : constant

( 9)

Then, calculating Eq. (8) from the solution (2), we get

1 (H2
\; +1-(H4
.. ) +=0
F+--F
--F
2
2!

4!

Cv

(10)

Cv

which is also a linear ordinary differential equation of F with constant coefficient.

Now, for any boundary condition, the settlement of clay stratum can be expressed as
p (t)

iH

(11)

e (t, z)dz

in which p(t) represents the settlement at time, t. Substituting the solution (2) into Eq.
(11), since T=e: constant, we have
4

2
1
1 (HF )
p(t)=eH+-(H
F)+-

2!

4!

Cv

1 (Hs
+F) +
6! cv2

(12)

By the sequential differentiation of Eq. (12) with respect to t, we get the following set of
,equations:
4

2 .
1
1 (HF
)
p=-(H
F)+-

2!

4!

Cv

1 (HsF
) + ...
+6! cv2
(13)

NII-Electronic Library Service

90

ASAOKA

Comparing the set of equations, Eq. (12) and Eqs. (13), with Eq. (6) pr Eq. (10),
unknown function F can be eliminated. When Eq. (6) is the case, it follows that

P+_l(H2
ti)+_l(H4
rJ)+=H(e+e)
3! Cv
5! Cv 2
2
-

the

(14 a)

Similarly, from Eq. (10), we get

p+_l(H
2! Cv

ti)+_l_(H: rJ)+=He
(14b)
4! Cv
Eqs. (14) give a settlement-time relationship under the condition in which an external
consolidation load is constant, that is, Eqs. (14) are the results derived from the stationary boundary conditions, Eq. (3) and Eq. (7).
It is easily ascertained that Eqs. (14) are absolutely stable for t-H:x), that is, all the
eigenvalues of these equations are negative real numbers which are different from each
other. In succeeding sections, however, we discuss the practical problems in which cv, H
and boundary conditions of drainage and load are entirely uncertain. Therefore, the expressions of the solutions of Eqs. (14) by the use of prominent eigenvalues and boundary values,
this is common in conventional analysis, can not be efficient for settlement prediction.
It should be noted that the higher order differential terms of Eqs. (14) are negligibly
small. Then, the following n-th order approximation equation is adopted as a master
equation of settlement-time relationship:

(n)

p+c1P+C2P++cnp=C
(15)
where c1 , c2 , ... , en and C are regarded as unknown constants. It will be shown later that
Eq. (15) is also applicable for some other settlement prediction problems.
Introducing
discrete time as
t;=4tj, j=O, 1, 2,
4t : constant,

...

(16)

Eq. (15) can be reduced to a difference form,

n

P;= So+ LJ SsPJ-s

(17)

S=l

in which P; denotes p(t;), the settlement at the time t=t; and the coefficients So and S.
(s=l, 2, ... , n) are unkown plant parameters. Eq. (17) will give us an idea of observational settlement prediction.
For the convenience sake of succeeding discussions, the first order approximation equation,

p+cif;=C
is examined here.

(18)

Let initial condition be

(19)
p(t=O) =Po
in which the time, t=O, should be taken at the time after loading works since Eqs. (14)
have been derived from the stationary boundary conditions which do not vary from time
to time. When this is the case, Eq. (18) can be easily solved as

p(t) ={)1-(P1-Po)exp( - :

(20)

where

P1=C,
the final settlement, sometimes referred to as the stable state of p.
the first order difference equation is expressed as
P;=So+S1P;-1

On the other hand,

(21)

NII-Electronic Library Service

91

PREDICTION OF SETTLEMENT

The coefficients,

Bo

B1

and

have the following values.

Substituting the stable state,

(22)

PJ=PJ-1=P1,

into Eq. (21), we obtain

P1=

l~~l'

(23)

Furthermore, by the recursive operation with respect to j, Eq. (21) yields

-- 1-B
Bo

PJ-

which can be compared with Eq. (20).

_
Llt {
lnB1--C1

which suggests that

ent of load.

B1

{1-B
Bo.

-Po

}cD)J.

From this, it follows that

=-

t~

=-

H 2 L1 t : the case of upward dramage

2cv

(24 )

P1

Llt : the case of both top and bottom drainage

(25)

does not depend on the boundary values,

(26)

e and

f;

B1

1s independ-

GRAPHICAL SETTLEMENT PREDICTION

We first examine the effectiveness of the first order approximation equation.

Assume
that we have had n+l settlement observations, (p 0 , p 1 , ... , Pn) generated by a constant
external load. Using these observations we can plot n points, (Pk, Pk- 1 ) for k= l, 2, ... , n,
on the (PJ, PJ- 1 ) co-ordinate. From this, we can visually see whether all these points lie
on a straight line as Eq. (21) suggests or not. Figs. 3 and Fig. 4 show the actual results
obtained from the settlement observations of extensive reclaimed lands in Japan. The data
for Figs. 3 are from the reclaimed land at Kobe Port.
Fig. 4 is the rearranged result
from the data published by Aboshi (1969).
In every case of these figures, settlement was
considered to be resulting from one-dimensional consolidation. Through these figures,

--------yl --- . -.----------------/ 1

1 71
1968
'69
70
'72
'73
'74
'75
'76
'77
'78
6 12 6 12 6 12 6 12 6 12 6 12 6 12 6 12 6 12 6 12 6 12

s
C)

... 10

..

s::Q)

sQ)
r-l

..
..

Q)

C/l

200

Fig. 3. (a)

Settlement observations of the reclaimed land at Kobe Port

NII-Electronic Library Service

92

ASAOKA

300

300

,.r/

Kobe Port No.l

I<obe Port No.2

:?

..-;'/~--'/

,,

....

.II'

..;/
,'/

S 200

//
./

CJ

.'/

/
:/

"/

: /

~-~

., /'/

100

./
.
../ / /

.. '/..:/

100

./ /

. ./..//

../7

,'/

';,

200

,Ji'

//

100

200

pk-1'

300

100

200

pk-1'

cm

300

300

cm

/'

Kobe Port No.3

.-;-'/
.//

.'/

200

,:/

.'/

'/

,'/

,"/

.'/
//

.. -J

100

..

/
/

100

200

pk-1'
Fig. 3. (b)

300

cm

PJ-PH relationship

a time interval, LI t was taken to be 3 months ( 91-92 days).

As shown in the figures, the first order difference equation, Eq. (21), gives a good fitting to the observations, which enables us to predict future settlement by graphical method. This procedure is illustrated in Fig. 5. Since the rough estimates of /3 0 and /3 1
are given by the intercept and the slope of a fitted straight line, it is possible to predict
settlement of any future time, j, by using Eq. (24). Moreover, the intercepting point of
this line with the 45-line represents the final settlement, because {) 1 is given by {)J = PJ-l
If necessary, the unknown coefficient of time factor, cv' / H2, can be easily calculated backward by Eq. (25) or (26).
In the case of multi-staged loading, the straight line, {)J=f3 0 +{3 1 pJ-i, will be moved up
as is also shown in Fig. 5. When the settlement is relatively small compared to the thickness of clay layer, the shifted line becomes almost parallel to the initial line because /3 1
is determined not from the external load but from the thickness, H and the coefficient of
consolidation, Cv,
This graphical method is quite useful in situations in which simplicity is more valuable
than strictness, but followings are noteworthy.

NII-Electronic Library Service

93

PREDICTION OF SETTLEMENT

ro
ro
0

i-l

l:l

100

(I)

(I)

r-i

.
.

(I)
CJ)

'Z.

8.5 t/m
3

n}i

Duration, year

C)

...

1. 6 t/

50

-~

150

200

"

.. - -

sC)

/
/
120_/~~~-"-~~~__,_~~~-'-~~~2~00
140
160
180
Pk-1' cm
Fig. 4. Settlement observations at Shin-Ube thermoelectric
power station (after Aboshi (1969))

( 1) To obtain /3 0 , we are generally compelled to extrapolate the fitted line to the Pk-i.
=0 axis. So, if the banking period is so long and then Po becomes so large, the extrapolation should be performed very carefully.
( 2) The accuracy of the graphical method depends mainly on a time interval, .Llt.
The longer .Llt becomes, the higher the accuracy is. For this purpose, the rearrangement
of observations is sometimes recommended.
If, as an example, we observe settlement
every week and .Llt is taken to be 10 weeks, then, letting p[k] denote the settlement of
the k-th week, we have the following 10 time-series of observations, that is;

NII-Electronic Library Service

94

ASAOKA

.... :7(\

,7...
. - - ~ -/ , /

d ...

placement of...

//

additional fill

f3o __..

--i''"'-...'

...... / /

I'
;
,,"
1
Q

/3,

final
settlement

final settlement
resulting from

initial ill

//
/

45
/
o~/~~---~~~~~~~~~~~-Pk-1

Fig. 5. Graphical method of settlement prediction

( 1)

p[OJ, p[JOJ, p[20],

( 2)

p[l], p[ll], p[21],

(10)

p[9], p[l9], p[29],

Since {3 0 and {3 1 are independent of an absolute time, all the time-series data will be plotted on the same straight line.
RELIABILITY ANALYSIS

The solution of Eq. (18), the :first order approximation equation, is predicted here by
using Bayesian techniques under the conditions in which c 1 and C are unknown at the
time t=O. One definite advantage that the probabilistic approach has over the graphical
method is that the predicted value is given with its reliability.
As is implied by the previous discussions, the first order antoregressive model,
P;=f3o+f31P;-1+ar;;
(27)
is chosen here as being a suitable one for settlement prediction, in which 1/; is the random
variable with zero mean and unity deviation.
In order to make successive calculations
simple, the followings must be assumed: one is that 1/; is a Gaussian random variable and
the other is

(28)
where E [ J denotes an expectation operation. It may be unnecessary to state that Eq.
(28) does not suggest the statistical independency between P; and Pk Three unknown
parameters, {3 0 , {3 1 and a are sometimes written as (} for simplicity, that 1s,

(} = ({3

0,

{3 i, a)' : unknown parameter vector.

Now suppose that the set of observations,

pt= (Po, Pi, ... , Pt)

(29)

NII-Electronic Library Service

95

PREDICTION OF SETTLEMENT

has been already obtained. We, then, have the posterior Probability Distribution Function
(pdf) of unknown (} as follows:
(30)

where

g ((}) : a-priori pdf of (}

Gaussian pd with mean, ({3 0 +(3 1pJ-i), and variance 0 2
When we have no information at the time, t=O, g(O) can be given by a diffuse prior pdf,
P(PJI(}, PJ- 1)

g((}) o:1(J
If we have some information at that time, g ((})
tive one than Eq. (31). When we take g ((}) as
pdf, g(Olpi), also becomes a Gauss-gamma joint
Next derived is the pdf of PJ conditioned by (}
P(PJI(}, Po)=

J . J

(31)

can be expressed by the more informaa Gauss-gamma joint pdf, the posterior
pdf.
and p0 This pdf is calculated from

k~/(Pkl(}, Pk-1) dp1dP2 dPJ-1

(32)

00

<J=l)

By using the mathematical inductive method, the P(PJI(}, p0 ) is demonstrated to be a Gaussian pdf, the mean, MJ, and the variance, Z/, of which are given by
MJ= 1~a1 -(

and

ZJ=a/

1~~

p,)(13,)' }
(33)

t=~::J

respectively. Although both MJ and ZJ are monotone increasing functions of j,

lnf3 1 < 0, they are also bounded from above:

1.

f3o
j~n.: M j= 1-(31

lim
j-+oo

smce

(34)

zj = aI J 1- f3 i2

Now, what we should examine here is the meaning of a. If we take a time series observation pi at only one spot on a clay stratum, ZJ represents the random fluctuation of PJ
from the trend MJ. However, we can generally observe settlement at various spots located within a certain area. Let Zpi be a time series observation of settlement at spot "l",
and let
pi= (1pi, 2pi, ... , mpi)
(35)
in which 'in is the number of observation spots. In this case, if a settlement process can
be assumed to be spatially stationary, the a estimated by the posterior pdf,
m

g(Olpi)o:g((}) II II p(Lpjl(}, ZPJ-1)

(36)

l=lJ=l

can be considered to include the property of unequal settlement characterized by the spatial interval between observation spots, l=l, 2, , m. When this is the case, introducing
the concept of an initial distribution of Po, p (p 0 ) , we obtain
P(PJI(})

JJ
.___,.__,

P(Po) J/CPkJ(}, Pk-1) dpodP1"dPJ-1

(37)

(j)

which is different from Eq. (32). An initial distribution of p, p(p 0 ) will represent the
distribution of unequal settlement just after loading. Provided that p(p 0 ) can be expressed

NII-Electronic Library Service

ASAOKA

96

as a Gaussian pdf, P(PJ/()) becomes also a Gaussian pdf just as P(PJ/(), p0 ).

Now back to our prediction problem. We have already been able to derive the predictive pdf of PJ conditioned only by the set of observations, pi, as follows:
200

1. 0

160

0.8

:>--t
.

:>--t
.

rl

rl
Ul

Ul
~

QJ
Q

:::

120

QJ
Q

:>--t
.

:>--t
.

rl

rl
rl
rl

,-j

80

rl
..Q

0.6

0.4

..Q

rj

rj

..Q

..Q

H
Pol

H
iJ..!

0.2

40

0.9791

0.9747

0.9335

/30'

/31
Fig. 6. Posterior marginal pdf's of {1 0 and
December, 1972 at Kobe Port, No. 3

7.88

6.94

6.00

cm

/1 1 conditioned by the data observed untill

0.1

7.
:>--t 6
.
rl
Ul
~

QJ

:>--t
.
rl
r-1
rl

..Q

4
3

1-------11--1-----1------+----------

20 2.1-----1----1-------H

Pol

1. l - - - - l t - - - - 1 - - - - - -

0.4

0.6

0.8

(j, cm
Fig. 7. Posterior marginal of a by
the same data of Fig. 6

270

273

276

279

282

285

288

Settlement, cm

Fig. 8. Predictive pdf of the settlement of December, 1976 from the data of December, 1972

NII-Electronic Library Service

97

PREDICTION OF SETTLEMENT

(38)
where

d(} = d{3 0 d{3 1dd.

Reliability of settlement prediction will be defined from Eq. (38).
ability of taking PJ between Pa and Pb at time j can be given by

For example, the prob-

(39)

The calculation of Eq. (38) or Eq. (39) can not be performed without an electronic computer because of the multiple inegrations in these equations.
Figs. 6, 7 and 8 are the numerical results of the applications of Eq. (30) and Eq. (38)
to the observations of Kobe Port No. 3, which are illustrated in Figs. 3. To obtain the
results, both a diffuse prior pd and the data observed until December, 1972 were employed. The prediction was conducted towards the settlement of December, 1976, the realized
value of which was 281 cm. Fig. 8 is the predictive pd of the settlement. Figs. 6 and
7 are the marginal pdf's of [3 0 , {3 1 and d of Eq. (30), respectively. For these calculations
..d t was taken to be 30,.._,31 days.

SOME SPECIAL PROBLEMS

In this section we derive a general prediction formula of future settlement through investigations of three special problems.
Secondary Compression due to Creep
If in the simplest case, the secondary compression of clay stratum can be approximately
described by a single Voigt model subject to a constant external load. Then in this case
the secondary settlement, Pn, is considered to satisfy
(40)

Pn+cfJn=P,

the first order linear differential equation with constant coefficients.

Eq. (40) is given as

Since the solution of

(41)

Pn=Pin- (Pin-Pon)exp( - : )

in which Pon is the initial c::mdition of Eq. (40) and Pin=P, then, the total settlement can
be expressed as follows:

() = (p !I+ P1n) -

(p !I-Po1)exp( - :

)
1

(42)

-(Pin-Pon)exp( - : )

where the settlement resulting from Terzaghi's consolidation theory is denoted by the
lower subscript, 1 , and is expressed also by the solu~ion of the first order approximation
equation, Eq. (20). It may be easily noticed that p of Eq. (42) is the formal solution of
a stable 2 nd order linear ordinary differential equation with constant coefficients.
Then,
by introducing a difference form, the next 2 nd order autoregressive equation,
2

PJ=f3o+ ~ f3sPJ-s

(43)

S=l

is suggested as being suitable for settlement prediction including creep factor.

NII-Electronic Library Service

ASAOKA

98

Settlement of Clay Stratum with Sand Seams

In an ordinary soil exploration, a thin sand seam sometimes goes unnoticed. However,
if the s3.nd seam allows free drainage, this overlooking causes a great error in settlementtime prediction. Fig. 9 shows the simple case of such a situation, in which the total settlement of both stratum I and II is given by

(44)

p(t)=Pr(t)+Prr(t)

where p1 (t) and Prr.(t) represent settlements of stratum I and stratum II respectively. Sup
pose that the sand seam of Fig. 9 allows free drainage.
Furthermore, for simplicity, let
p1 (t) and Prr(t) be expressed by the form of Eq. (20). When this is the case, Eq. (44)
suggests that the total settlement p (t) can also be predicted in the same way as is given
in the discussion of the secondary compression due to creep, that is, by the 2 nd order
autoregressive equation.
Sand Drains
In order to reduce the time for consolidation, vertical sand drains are sometimes used.
As is well known, if the equal vertical strain hypothesis is approximately satisfied, the
degree of consolidation can be expressed by a single mode, that is,

U = l-exp(J.t)

(45)
where an eigenvalue ). is determined from both a horizontal consolidation coefficient and
geometrical dimensions of sand piles. According to the tradition in soil engineering field,
.settlement can be treated as being proportional to the degree of consolidation.
Eq. (45)
shows, therefore, that the settlement by sand drains can be predicted by the first order
autoregressive equation.
General Prediction Formula
The above discussions imply that a general settlement prediction model 1s given by
n

PJ=/3o+ ~ f3sPJ-s

(46)

S=l

An actual settlement may be governed by many different factors. However, if those factors can be mathematically formulated as an eigenvalue problem, prediction formula of Eq.
(46) is quite general.
It should be noted that Eq. (46) is the same form as that of Eq. (17), although the
physical meanings of the coefficients are quite different.
t
I
I

I
I

... ---: ..

.. . _. -'": ..... :- .... ... : -~ ...

Top

':._ ;:-

Clay

fa

-4

Thin sand seam,

...__~~:::...::.::.r;.i);.l-~~:~.:: ::--...~t."".:..:~t;,:,.:,i:.(.: i:::.:-!"._!;: {.

--.

free drainage

.s
i

-6

.j.)

-8

-16
-18

Fig. 9. Drainage from thin

sand seam

-~

/sand blanket
.

20m

:c

10m

=
=
l
1

'i;J -10
-14

Bottom

30m

.. ,.,.

; .

:-.

20m
10m
clay

'\

'

:''I.

30m

peat
clayey peat

Q)

-12

Clay ]f

.
I

-2

C:

peat
clayey peat

0n,,,,,,
bottom of sand drains
/loose sand

silty sand

..

...:-:

~f.:

sand

Fig. 10. Soil profile and embankment cross

section of Iwamizawa test fill

NII-Electronic Library Service

PREDICTION OF SETTLEMENT

99

The statistical parameter identification of Eq. (46) for n22 from past observations is,
however, hopeless at present, since the settlement process is not stationary, that is, E[P,1]
~ const.
An analytical expression for a predictive pd of P.1 by Bayesian statistics is formally possible in the same way as is presented in the case of n=l in the previous section.
The results, however, include super multiple integrations. Therefore, the author recommends the least square estimates of /3 0 and f3s (s=l, 2, ) if the long term observation is
available and the considerably large time interval, Llt=t,1-t,1-i, can be adopted.
Practically speaking, the first order prediction model, Eq. (21) or Eq. (27) should be
taken first. When there exists only one prominent eigenvalue, the first order appoxima8

.
SC:

~:
SC:

l'J
~

6
4

;j

1976

9 10 11 12

1977
1 2 3

9 10 11 1

50
100

su

...

150

'\

.
SC:

(l)

200

..

(l)

rl

.
.

250

(l)

Ul

300

.,

350

......... ...

. .. .. .

400

Fig. 11. Settlement observations at sand-drain test section of Iwamizawa test fill

Placement of
additional fill
from Apr. to
1977

250

/
/
/

/
200

250

350

300

pk-1'

400

cm

Fig. 12. Application of proposed graphical method to observations

of Iwamizawa test fill

NII-Electronic Library Service

100

ASAOKA

tion method will give satisfactory results. If the model does not give a good approximation to observations, the second order prediction model, Eq. (46) with n=2, is to be tried
next.
A Practical Example of A Complex Boundary Value Problem
The applicability of the proposed settlement prediction method to a complex multilayered
clay deposit was examined by using actual data provided from the test fill which was constructed during 1975-1978 at Iwamizawa in Hokkaido as a portion of the Do-o expressway
by Japan Highway Public Corporation. The soil profile and the embankment cross section
at the test section is shown in Fig. 10. As shown in the figure, the clay deposit consists.
of 7 subsoil layers including clay, peat and silty clay stratum. Sand piles were driven till
the elevation of -13 m for the purpose of vertical drain through the piles.
Measured
settlements under center of embankment are given in Fig. 11. The embankment was left
alone two times with no additional fill, the first time is from the later part of November,
1976 to the midst of April, 1977 and the other time began in the last of May, 1977. The
proposed graphical settlement prediction method was applied to the observations obtained
during these two periods. The results is illustrated in Fig. 12 which shows the remarkable
applicability of the first order autoregressive prediction model. For the illustration of Fig.
12, the time interval, Lit, is taken to be 15 days.

CONCLUSIONS

The observational method for settlement prediction is freshly proposed in this paper.
For one-dimensional consolidation problems and sand-drain problems, the first order autoregressive equation, Eq. (21) or Eq. (27) is demonstrated to be a suitable one for settlement
prediction. For this prediction formula, both the graphical and reliabilistic approaches were
presented.
The proposed method is considered to have a great advantage to predict settlement of
an arbitrary future time under situations in which a coefficient of consolidation and both
initial and boundary conditions are quite uncertain. Such situations are considered to be
so common in engineering practice that the proposed method will have a certain availability.
As for the unknown conditions, they can be guessed backward, if necessary, from settlement
observations by using the estimated coefficients of the first order autoregressive equation.
The higher order autoregressive equation, Eq. ( 46), suggests to provide more precise
settlement prediction and to be applicable to some special problems, that is, one is concerned with secondary compression and the other is the settlement of multilayered stratum
with sand seams.

ACKNOWLEDGMENT

The author is extremely greatful to Professor Yoshimi Nagao of the Kyoto University
for his continuous encouragement. The author also wishes to express his thanks to Pro-
fessor Minoru Matsuo of the Nagoya University for his valuable comments on the "Obser-
vational Procedure."
REFERENCES
1)
2)

Aboshi, H. (1969): Soil Mechanics, Chap. 4 (edited by Mogami, T.), Tokyo, Giho-do, pp.
464-465 (in Japanese).
Japan Highway Public Corporation (1977): Technical Report of Iwamizawa Test Fill (in Japa-
nese)~

NII-Electronic Library Service

PREDICTION OF SETTLEMENT

3)
4)
5)
6)
7)

101

Matsuo, M. and Asaoka, A. (1978): "Dynamic design philosophy of soils based on the Bayesian
reliability prediction," Soils and Foundations, Vol.18, No. 4, pp. 1-17.
Mikasa, M. (1963): Consolidation of Soft Clay, Tokyo, Kajima-shuppan-kai (in Japanese).
Terzaghi, K. and Frohlich, 0. K. (1936): Theorie der Setzung von Tonschichten, Leipzig.
Terzaghi, K. (1948): Theoretical Soil Mechanics, New York, John Wiley and Sons.
Terzaghi, K. and Peck, R. B. (1967) : Soil Mechanics in Engineering Practice, New York, John
Wiley and Sons.
(Received May 22, 1978)

NII-Electronic Library Service