Sei sulla pagina 1di 7

Least weight design of steel pile

foundations
A. S. Hoback and K. Z. Truman
Department of Civil Engineering, Washington University, St. Louis, Missouri
63130, USA
(Received 15 July 1992; revised version accepted October 1992)
The layouts of steel piles under slab foundations are optimized to find
the least weigtlt systems. The pile inertias and batters are varied using
an optimality criteria approach. A Gauss-Seidel procedure is used to
evaluate the Lagrange multipliers. The types of design limits controlled
by the optimization are the pile stresses, displacements, sizes and
orientations. Three examples are given, two of them are dam structures,
and the other is a control tower foundation. The first dam structure
demonstrates that a feasible design can be reached even though the
initial design is highly infeasible. The control tower foundation demonstrates that significant weigtlt reductions can be achieved given an
understressed initial design.
Keywords:
criteria

piles, pile foundations, structural optimization, optimality

Pile foundation systems can be very expensive, therefore


finding the optimal pile layout is of great importance.
The objective is to optimize HP-14 steel pile foundations
to produce a least weight design.
Structural optimization methods are numerical processes which improve an initial design by iterating until
an optimal design is reached. Very little previous work
has been performed in the area of pile optimization; one
investigation has been reported. Hill 1 tried to reduce the
weight of steel by using a pile deletion process. His
optimization procedure consisted of first finding the
optimal pile batters or slopes, then finding the optimal
pile spacing within specified zones, and removing piles
until the stresses and displacements were near their
limits. The piles were removed in an iterative process by
eliminating the most and/or the least stressed piles. This
method is very inefficient and does n o t always produce
the optimal design.
An optimality criteria approach was used in this study.
Optimality criteria methods were first used in the 1970s.
Gellatly and Berke z and Venkayya 3 applied the optimality criteria method to the structural optimization of
trusses and frames. The method was developed to solve
optimization problems with a large number of a variables. The method quickly arrives at an optimal design
while using a relatively small amount of computational
effort.

Notation
A
ci
d~

hj

area of a given member


extreme fibre distance in direction of pile's ith
local axis
design variable number i
constraint j

Ixx

Iyy
m
Mi
n
N
Pi
r
v
Vi

Wr
0
2j
p
~b

major axis moment of inertia


minor axis moment of inertia of a given member
number of constraints
global moment load component i
number of variables
number of piles
global force load component i
convergence control parameter
current optimization iteration number
volume of element i
total weight of steel piles
angle of pile flange alignment
Lagrange multiplier for jth constraint
specific weight of steel
angle of pile rotation from x-axis

Pile layouts
The cross-sectional properties and the orientations of the
piles will be variables. All of the pile tips are at the same
depth within a design regardless of batter. The initial coordinates of the head of the piles will be fixed, and the pile
spacing will not change.
The batter of the piles is the ratio of the depth of the
pile to the horizontal distance between the pile head and
tip as shown in Figure l(a). Phi (4) is the angle of
rotation of the pile about the vertical axis from the global
X-axis as shown in Figure 1 (b), measures the pile flange
alignment as shown in Figure 1 (c). The batter is the only
topological parameter that is variable in the examples
shown. The angle ~b is not a variable in the examples
because symmetry exists in each example; it is a variable
when out-of-plane loads are applied.
Initially the cross-sections are not restricted to discrete
HP-14 steel pile sizes which are available from fabri-

0141-0296/93/050379-07
1993 Butterworth-Heinemann ktd

Eng. Struct. 1993, Volume 15, Number 5 379

Pile optimization. A. S. Hoback and K. Z. Truman

Pile Cap

The maximum size of HP-14 section available is 1,., -1220 in 4 (50780 cm4), and the minimum is Ix~ = 729 in ~
(30343 cm4). Similar equations can be developed for
regular and irregular cross-sections 5.

Foundation analysis and design limits

Pile
Batter
a

1.0

Pile
b

X
Y

HP
Sections

x
Y

0 -- O. d e g r e e s

0 - 90. d e g r e e s

Figure 1 Pile co-ordinate systems. (a), elevation view; (b), plan


view; (c), Flange alignment

cators. The branch and bound algorithm is used in the


last phase of the optimization to change the crosssections to usable sizes. The cross-sections found during
the intermediate iterations are called pseudodiscrete
since they are found using a specified set of equations
which bound the HP-14 sections.
The optimization algorithm requires that one variable
will be used to represent the cross-sectional properties.
All of the cross-sectignal properties will be expressed in
terms of one primary variable which is the major axis
moment of inertia (Ixx). The secondary variables are
approximated as functions of the major axis inertia. The
secondary variables are the area (A), the minor axis
moment of inertia (Irr), and the extreme fibre distances
(cx, cy). The following approximations are developed for
HP-14 sections from the AISC Manual of Steel
Construction 4.
Iyr = 0.37067I~x - 9.2200 (in 4, 1 in = 2.54 cm)

(1)

A = 0.049153Ix 921s (in 2)

(2)

cx = 1.2110

(3)

10 -3

lx~ + 12.7192 (in)

cy = 6.110 x 10-4lxx + 14.139 (in)

380

Eng. Struct. 1993, Volume 15, Number 5

(4)

Thick concrete monolith foundations are analysed assuming that the pile cap is rigid in the manner of Basic
Pile Group Behaviour 6. The structural foundation is
assumed to rigidly connect all the piles. Therefore the pile
cap has only six degrees-of-freedom, three translations
and the three rotations.
The pile group stiffness coefficients are calculated in
the manner of Saul 7. A beam on an elastic foundation
theory is used to find the lateral pile stiffness. The pile
and soil interaction is represented at the head of the pile
by a linearly elastic pile stiffness. The global stiffness
matrix is calculated by summing the coefficients from all
the piles. The displacements in the stiffness method are
found by inverting the stiffness matrix and multiplying it
by the applied loads.
Stress interaction equations are used to evaluate the
feasibility of the piles 6. The top portion of the piles are
subject to axial load and bending. The point of maximum
moment is a function of the soil characteristics and the
pile connection fixity. The bending moments dissipate in
the top portions of the piles, therefore, only the axial
forces are carried through the lower regions. An increased factor of safety is used in the lower regions
because the pile tips are subject to damage during
driving. The axial load at the tip must be less than the
damaged cross-section capacity and the soil bearing
capacity.
Pile structures are subject to a variety of design limits
or constraints. Some of the constraints are:
(1) Member stress constraints. The most highly stressed
member in any group is prevented from becoming
over-stressed
(2) Size constraints. The cross-sections are prevented
from advancing beyond the maximum or minimum
section sizes available
(3) Interference constraints. Piles cannot pass through
one another. In two-dimensional problems and the
less complicated three-dimensional problems this
constraint can be established by preventing the pile
tips from meeting
(4) Minimum batter constraints. The piles may not be
battered with a slope beyond a specified angle
(5) Displacement constraints. Each displacement component may be limited to a specified value
The constraints are represented by the value hj wherej
is the constraint number. Constraint j is satisfied when
hj < 0, exactly satisfied (active) when hj = 0, and violated
when hj > 0. A design is feasible when all the constraints
are satisfied. A computer program O P T P I L E was
created for the analysis and optimization 8. All of the
above constraints may be controlled by user specified
values.

Mathematical formulation of the method


The objective function is the function to be minimized.
The goal of the p~le optimization process is to find the

Pile optimization: A. S. Hoback and K. Z. Truman


The expected value of --j
h .+1 for the active constraints
should be zero. Substituting equation (10) into equation
(11) yields:

minimum weight of steel that can be used to satisfy the


given constraints. The objective function is:
N

(5)

wT = T, p, v,

/Ohjdh,\
I c3dl c3di I

i=l

where WT is the total weight, p~ is the specific weight, and


V~is the volume for the structural element i, and N is the
number of piles.
The optimality criteria method is developed using the
Lagrangian of the weight and constraints. Satisfaction of
the Kuhn-Tucker conditions is necessary for a point in
design space to be a local minimum of the given problem.
The first necessary Kuhn-Tucker condition requires the
derivative of the Lagrangian to be zero. This results in
the optimality criteria:

- ~ 2j
j =a

ew,

Ohj
C~dl

-i

i = 1 .... ,n

s=l

i=I

Ohj
i=1 ~//

j = 1..... m

(12)

where j and s sum over all constraints.


The linear equations apply to only the constraints
which are active or assumed to be active. The equations
must be solved subject to the Kuhn-Tucker conditions.
A Gauss-Seidel iteration method (method of successive
displacements) was used to solve for the 2j values.
Equation (12) is equivalent to:

/OhjOh,~
- - - c~dil
--n I -c3dl

(6)

".2. Ohj

Od~
\

where 2j is the Lagrange multiplier for the constraintj, m


is the total number of constraints, and there are n design
variables d~. This criteria must be satisfied for the n
variables.
The other necessary conditions are that the Lagrange
multipliers must be positive and the constraints must not
be violated.
).j>0

j = l ..... m

(7)

hj<_O j = 1..... m

(8)

j = 1..... m

~ - - ~ d,

Initially the 2j values are set to zero. The 2j values are


successively solved for until their values converge. During the successive solutions if a value of 2j becomes
negative then it is set equal to zero to satisfy equation (7).

Examples
The design process consists of estimating an initial design
and then performing an optimization. The first example
is a dam structure, it demonstrates that given a highly
infeasible initial design the optimality criteria algorithm
can find a feasible design and then find the optimal
design. The second example is a control tower foundation. The third example is an existing dam structure.
The piles are placed in groups which maintain the
same sizes and batters in each group. This grouping is
also called linking. The individual piles in a group are not
allowed to independently vary in size or batter. Placing
piles into groups provides the consistency that is desired
when installing the piles.

Linear recurrence equations are used to update the


variables as the design progresses toward a more optimal
design. The ith variable (dl) is altered by the value of the
ith optimality criteria in the following recurrence formula

c~di

(10)

where v is the index of the iteration number and r is the


convergence control parameter.
The values of the Lagrange multipliers must be estimated during every iteration of the optimality criteria
process, a linear constraint equations method is used to
estimate them.
The equation for the change in thejth active constraint
can be written as:

Ohj
Ahj=h~ +1 - h ~ = i=lWJ-Ad
iuui

j = l ..... m

(13)

Od~

(9)

/-

j = 1..... m

i=1 OWr

The final condition requires that ).~ is zero when hj is not


zero, and 2i is not zero when hj is zero.

2jhj=O

c3di /

)lq=

Example 1
This example is a dam with 29 piles. The piles are
arranged in three rows. The piles in each row are linked.
Linking causes the piles to maintain the same orientation
and cross-section as the other piles in the same row. The
loading is shown in Table 1. The initial pile layouts are

(11)

Tab~ 1.

Loading for example 1 (1 kip = 4.448 kN, 1 h = 0.3048 m)

Load
case

Px
(kips)

P
(~ips)

Pz
(kips)

M
(~-k)

-1300

1500

-10000

1580

M
(~V-k)

M
(~-k)

Eng. Struct. 1993, Volume 15, Number 5 381

Pile optimization." A S. Hoback and K. Z. Truman


Table2.
Iteration

Layouts and sizes for example 1 (1 in = 2.54 cm)


Pile
group

/ xx
(in 4)

Batter

~
(degrees)

729

100

90

2
3
1
2
3

729
729
729
729
729

100
100
3.96
2.35
4.14

90
90
270
90
270

0
0
0
0
0

Spacing 5'

LJ~ 2

~p,

Ir

~3

-q-

(degrees)

Ir

~
[~

10

.~Y

A r r o w s indicate
batter direction.

--T

"~3

.J

N*
N"
N'19

Spacing 5'

~'~ 11

Groups:

29 I r

N-~19

~ _ 10

2~
3m

Arrows indicate
batter direction.

N*
N*
N"

Groups:

2 9 "~

2[]
3

Cap
Cap

Spacing
15'
i

Pile group 1 Pile group 2

I Depth" z'
Pile group 3

II

D e p t h = 70'

Figure 2

Initial pile layouts for example 1. (a), plan view; (b),


elevation view. (1 ft = 0.3048 m)

given in Table 2 and Figure 2. The piles have batters of


100 which are essentially vertical. The initial design is not
feasible. Originally the piles are overstressed in load case
2 by 41%. The maximum pile stress for load case 1 is 39%
of the allowable stress.
The final layouts which were found after eight optimization iterations are given in Table 2 and Figure 3. The
final design is feasible. The initial weight of steel was
147 849 lb (657.6 kN) and the final weight is 154 897 lb
(689.0 kN). The weight convergence is shown in Figure 4.
The weight initially increased in size in the first three
iterations as the optimality criteria algorithm created a
feasible design. In iterations four to eight the designs were
all feasible and the weight decreased with each iteration.
The largest final allowable stress ratios were found in
pile group 2 (piles 11-19). The ratios of actual to allowable stresses were 86% and 98% for load case 1 and 2,
respectively. The constraints actively preventing the design from further reductions in weight are the stresses in
piles 11-19 for load case 2; and the minimum crosssection sizes for all piles.
The optimization procedure does not guarantee that

382

Eng. Struct. 1993, Volume 15, Number 5

Figure 3

Final pile layouts for example 1. (a), plan view, (b),


elevation view

the final design is absolutely the best design. The KuhnTucker conditions guarantee that this is the locally
optimum design. Multiple starting points could be used
to search for the absolute best design.
3E5Weight

Obs.)

SiDeD ~

2E5.

~D
1 E5

Figure 4

D~D~D
I
1

Iteration

Convergence of weight for example 1. (1 Ib = 4.448 N)

Pile optimization." A. S. H o b a c k a n d K. Z. Truman

other piles in the same ring. A displacement limit of


0.35 in (0.89 cm) was applied.
The final layouts, which are found after seven iterations, are given in Table 4. During the optimization the
weight changes from 95718 lb (425.8 kN) to 87728 lb
(390.2 kN) as shown in Figure 6. The weight improvement is partially caused by a decrease in the section size
of the piles in group I. Also, the pile slopes are more
vertical in the final design than in the initial design. This
causes a reduced weight since the piles became shorter.
The active constraints are the minimum sizes of both
groups; and the displacement in the x-direction. The
displacement converged to the maximum value of 0.35 in
as shown in Figure 7. The pile allowable stresses did not

Example 2
This example is a control tower with 18 piles arranged in
two rings. The initial design is a design example which is
analysed in the CPGA User's Guide 9. This design will
now be optimized with the optimality criteria method.
The only alterations to the design are that the depth of
piles 1-12 is held at 65 ft (19.81 m), and the depth of piles
13-18 is held at 74 ft (22.56 m). The loading is given in
Table 3, and the initial layouts are given in Figure 5 and
Table 4. Piles 1-12 were originally vertical. The piles in
each ring will maintain the same size and slope as the

1E5r
Group
Weight ? \

L 6 piles

@ 60 de 9,
<k

2,

9E4f

1S piles

\
0.____~0 ....

.-

0 ~ 0

3
4
Iteration

30 de 9
8E4

I
1

-X

I
2

Figure 6 Convergence of weight for example 2


0.500

..._0/0_.____0_~0~0Limit 0,35"

Displ.
Camp. 1

(inches)

"=0- - - (

I "'"

+
0.000

Figure 5 Initial pile layouts for example 2. (a), plan view; (b),
elevation view

3
4
Iteration

Figure 7 Displacement limit for example 2. (1 in = 2.54 cm)

Table 3. Loading for example 2 (1 kip = 4.448 kN, 1 ft = 0.3048 m)


case

Load

P
(kips)

Pv
(kips)

Pz
(kips)

Mx
(ft-k)

My
(ft-k)

M,
(ft-k)

- 142.8

1272

5600

Table 4. Layouts and sizes for example 2


Iteration

0
7

Pile

/xx

group

(in 4)

1
2
1
2

904
729
729
729

Batter

4.
100.
8.80
14.81

(degrees)

(degrees)

See Figure 5
See Figure 5
Unchanged
Unchanged

0
0
0
0

Eng. Struct. 1993, V o l u m e 15, N u m b e r 5

383

Pile optimization: A. S. Hoback and K. Z. Truman


control the design. Initially the largest stress was 53~ of
the allowable stress for pile 15. The final maximum was
58~ for pile 5.

of 90 ft (27.43 m). The loading given in Table 5, and the


initial layouts given in Figure 8 and Table 6. Note that
the pile cap is sloped.
The final layouts which are found after seven iterations, are given in Table 6. The active constraints are the
minimum sizes for all groups; and the stress in pile 229
which is fully stressed. The weight improves from
1 477 348 lb (6571.6 kN) to ! 435 6841b (6386.2 kN) as
shown in Figure 9. The weight was decreased slightly
from the initial design. The initial design is an existing

Example 3
This example is a dam with 240 piles arranged in three
groups. This example is a design which is analysed in the
CPGA User's Guide 9, it will now be optimized. The only
alteration to the design is that all pile tips reach a depth

i 57-:26-i
EL -13 --

-20

Group

Group

Group

_~- 3,5'

61

95'

_I

--

00000

[]

O00OO

[]

[]

[]

[]

[]

00000

eeeeeeeeeee

00000

eeeeeeeeeee

00000

eeeeeeeeeee

[]
[]

O00OO

[]
[]
[]

[]
[]

[]
[]
[]

[]
[]

Group
[]

[]

[]

i0@9':90'

eeeeeeeeee
[]

00000

[]

eeeeeeeeee
[]

OOOOO

[]

eeeeeeeeee

61~
0

11@9'=99'

8'

llseeeeeeeeeee

[]

eeeeeeeeeee
[]

[]

[]

00000

eeeeeeeeeee

O00OO

eeeeeeeeeee

O00OO

eeeeeeeeee

0000~0

eeeeeeeeeee

[]
[]

0
O
[]

[]
[]
0

O
[]
0

....

[]
[]
[]
0112
240

35' ]

b
Figure 8

384

Inital pile layouts for example 3. (a) elevation view; (b), plan view. (1 ft = 0.3048 m)

Eng. Struct. 1993, Volume 15, Number 5

106'

[]

1~ 58 pites

60 piles

128 piles

Pile o p t i m i z a t i o n : A. S. H o b a c k a n d K. Z. Truman
Tab~ 5.

Loading forexample 3

Load
case

P
(dips)

P
(kips)

Pz
(kips)

M
(~-k)

M
(~-k)

Mz
(~-k)

-12000

20000

-1.E6

Tab~ ~
Iteration
0
7

Layouts and sizes for example 3

Pile
group

~x
(in 4)

Batter

1
2
3
1
2
3

729
729
729
729
729
729

100
2
2
6.58
3.96
3.92

2.00E6Weight

Obs.)
1.50E6

P-'--~-O~o------'O- - - - - 0 ~ 0

1.00E6
0

I
1

I
2

0
I
3
4
Iteration

I
5

I
6

Figure 9 Convergence of weight for example 3

structure that was designed using a significant number of


engineering hours. A large weight improvement was not
expected. The optimization algorithms can be used to
provide a quick design process which reduces the engineering hours required for designing.

Conclusions
Concrete monolith foundations supported by steel H P 14 piles were optimized. An optimality criteria method
was selected because it is suitable for large optimization
problems.
The pile cross-sections and pile batters were allowed to
vary. Initially the variables were not held to discrete
values but were allowed to change in a continuous
manner.
The examples included dams and a control tower. The
first example demonstrated that a feasible design can be
obtained even though an initial design is highly infeasible. The second example demonstrated that constraints

~
(degrees)

90
90
270
270
90
270

0
0
0
0
0
0

(degrees)

such as the displacement constraint are accurately controlled by the optimization.


The third example demonstrated that the optimization
algorithms may not always produce a final design with
dramatically improved weight. The level of improvement
attainable is dependant upon the level of effort used in
finding an initial design. The optimization algorithms
reduce the effort required by designers by automating the
design updating process.

Acknowledgment
The authors wish to thank the US Army Corps of
Engineers for partially funding this work under contract
D A C W 3989M3044.

References
1 Hill,J. L. 'User's guide: computer program for optimal design and
analysis of pile foundations (PILEOPT), Instruction Rep. K-81-5
US Army Crops of Engineers, 1981
2 Gellatly, R. A. and Berke, L. Optimal structural design 1971, USAF
AFFDL-TR-70-165
3 Venkayya, V. B. 'Design of optimum structures', Comput. Struct.
1971, 1, 265-309
4 Manual of Steel Construction (8th edn), American Institute of Steel
Construction, Chicago, IL, 1980
5 Cheng, F. Y. and Truman, K. Z. 'Optimization algorithm of 3-D
building systemsfor static and seismicloading, In Ames, W. F. and
Vichnevetsky, R. (Eds) Modeling and Simulation in Engineering Vol
3. North-Holland, Amsterdam, 1983, pp 315-326
6 'Basic pile group behavior', Tech. Rep. K-83-1 CASE Task Group
on Pile Foundations, US Army Corps of Engineers, 1983
7 Saul, W. E. 'Static and dynamic analysis of pile foundations', J.
Struct. Div., ASCE 1968, 94, (ST5), Prec. Paper 5936
8 Hoback, A. S. and Truman, K. Z. 'Optimization of steel pile
foundations using optimality criteria', Contract Rep. ITL-92-1, US
Army Corps of Engineers 1992
9 'User's Guide: Pile Group Analysis (CPGA) Computer Program',
Tech. Rep. 1TL-88 CASE Task Group on Pile Structures and
Substructures, US Army Corps of Engineers, 1988

Eng. Struct. 1 9 9 3 , V o l u m e 15, N u m b e r 5

385

Potrebbero piacerti anche