Professional Documents
Culture Documents
Machines
Marcelo A. Silva1; Jasbir S. Arora, F.ASCE2; Colby C. Swan, M.ASCE3; and Reyolando M. L. R. F. Brasil4
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 06/15/14. Copyright ASCE. For personal use only; all rights reserved.
Abstract: The objective of this study is to describe a problem formulation and an optimization procedure for the design of elevated
reinforced concrete foundations for vibrating machines. Special emphasis is placed on structures composed of footings, beams, and
columns. The dimensions of the structure and its reinforcement are the design variables for the optimization problem. The objective
function consists of costs of the concrete, the steel, the form, and the propping form. Constraints related to material and soil failure, as
well as geometrical limits and human comfort are imposed. A new failure surface for columns and beams subjected to biaxial bending and
axial loads is defined and used in the formulation. The main motivation of using the new failure surface is to save a large amount of
computational effort in the solution of this dynamic response optimization problem. The problem of minimizing the structural cost while
satisfying the operating and safety requirements is solved using an augmented Lagrangian method. A large number of constraints 共time
dependent and time independent兲 is treated without any difficulty in the method. The numerical methods used in the solution process are
described. Optimal solutions for an example problem are obtained and discussed.
DOI: 10.1061/共ASCE兲0733-9445共2002兲128:11共1470兲
CE Database keywords: Optimization; Design; Concrete structures; Dynamic loads; Soil-structure interaction.
Structural Analysis
relative to a local coordinate system. The displacement vector for
The finite element procedure is used to develop equations of mo- the node i in the global coordinate system is written as zti
tion for the system consisting of beams, columns, the footing, and ⫽ 关 w i x i y i 兴 . Using the usual shape functions, element stiffness,
the soil. The equations are integrated numerically to determine the and mass matrices, nodal force vector can be generated.
dynamic response of the structure.
Beam and Column Equations
Footing and Soil formulation Consider now the cross section of a beam/column member of a
A typical footing modeled as a plate structure is shown in Fig. 2. three-dimensional framework, as shown in Fig. 4. Such elements
The global reference coordinate system is denoted as 0XY Z. The are referred to as the line elements in this work. L⫽length of the
plate is developed in the 0XY plane. The directions of X, Y, and Z element, while the cross section width is B and height is H. The
are, respectively, the transverse, longitudinal, and vertical direc- cross section geometric properties can be written as A⫽BH
tions of the system. The plate is modeled as an isotropic, homo- 共area兲, I y ⫽BH 3 /12 共moment of inertia related to yy axis兲, I z
geneous, and a linearly elastic material of uniform thickness. The ⫽HB 3 /12 共moment of inertia related to zz axis兲, and J⫽I y ⫹I z
soil system in which the slab foundation is embedded is modeled 共polar moment of inertia兲. The material properties are Young’s
by a system of springs and dashpots. The soil spring constants in modulus E, shear elasticity modulus G, and mass density . The
the X and Y directions are K x and K Y , respectively. K T ⫽soil t
displacement vector for the eth element is ze ⫽ 关 ze t1 ze t2 兴 , where
spring constant related to the rotation about the axis Z. The con- t
stant ⫽soil coefficient of the subgrade reaction; physically, it zei ⫽ 关 u ei v ei w ei exi eyi zi
e
兴 is the displacement at the ith element
represents an elastic spring density per unit area in the Z direc- node. The components u ei , v ei , and w ei are, respectively, the dis-
tion. placements in the x, y, and z directions, while exi , eyi , and zi e
are
A compilation of several procedures to estimate has been the rotations of fibers parallel to the x, y, and z axis, respectively.
performed by Silva et al. 共1999兲. Considering a point in the con- These displacements are given in the local system. The displace-
tact area between the footing and the soil, the spring force F per ment vector of the ith node in the global coordinate system is
unit area is F ⫽ w, where w⫽vertical deflection of the soil– denoted as zti ⫽ 关 u i i w i xi yi zi 兴 . The equations of motion for the
plate interface. The inertial force per unit area is F 1 ⫽hẅ, where axial displacement, bending about two axes, and torsional motion
h⫽plate thickness and ⫽mass density of the plate material.
Using the basic moment-curvature relationships and the theory
of elastic plates on elastic spring foundations we obtain equation
of motion for the plate for transverse displacement 共Shigueme
1995; Zienkiewicz and Taylor 1989兲
Dⵜ 4 w⫹w⫹hẅ⫺q 共 X,Y ,t 兲 ⫽0, ᭙t苸 关 t 0 ,t f 兴 (1)
where D⫽Eh / 关 12(1⫺ ) 兴 , and q(X,Y ,t)⫽load per unit area
3 2
l
M zi ⫽flexural moments. In case of columns, generally structural
design codes require the increment of flexural moment due to
l
second order effects; i.e., for M zi and M lyi . A simplified procedure Fig. 5. Typical cross section of line element and plate
described in NB-1 共1978兲, the Brazilian code for design and con-
struction of reinforced concrete structures, was used to calculate
this additional moment.
stress calculations for concrete and steel are described. All these
considerations and specifications are based on NB-1 共1978兲; how-
Global Equation of Motion
ever, any other code can be used.
Assembling the element level mass and stiffness terms, we can
Fig. 5 shows a typical concrete line element and a plate. In
build a global system of equations in the matrix form that in-
Fig. 5共a兲, B⫽width, H⫽height, A s and A ss ⫽respectively, longi-
volves all the structure degrees of freedom 共Zienkiewcz and Tay-
tudinal and transverse steel areas, and c⫽cover of longitudinal
lor, 1989兲
reinforcement, for beams and columns. In Fig. 5共b兲, h
Mz̈⫹Kz⫽p共 t 兲 , ᭙t苸 关 t 0 ,t f 兴 (2) ⫽thickness, A sx and A sy ⫽longitudinal steel area perpendicular to
the y and x axes, respectively, and c p ⫽reinforcement cover, for
To complete the dynamic analysis, it is necessary to compute plates. Elastoplastic behavior of the steel used here is considered
damping effects within the soil and the structure. In this work we with elasticity modulus E s and yield stress f y . This stress must be
consider Rayleigh damping, as divided by a factor ␥ s ⬎1 to get the steel design stress f yd
C⫽a 0 M⫹a 1 K (3) ⫽ f y /␥ s . The strain corresponding to this stress is yd ⫽ f yd /E s .
The distribution of strains on the cross section is linear. The con-
where a 0 and a 1 ⫽proportionality parameters, determined from crete and steel strains are, respectively, s and c . The maximum
the simultaneous solution of the expression tensile strain in steel is smax⫽1% and the maximum compressive
is that we have two shear loads V y and V z . The steel area corre-
sponding to each load is calculated and later added to obtain the Here  w and w ⫽nondimensional coefficients related to the foot-
total area. This procedure is a little conservative, but quite simple ing dimensions and the soil Poisson’s ratio s . Arya et al. 共1978兲
to implement. The minimum volume of the torsion steel in the also presented the following soil elastic constants parallel to tor-
entire element is 0.14% of the ideal prism volume 共Sussekind sional, transverse, and longitudinal directions:
冑
1979兲. To design completely for the torsional moment, it is nec-
essary to adopt another steel area, the longitudinal steel area for 16G s r 30 4 l x l y 共 l 2x ⫹l 2y 兲
K T⫽ ; r 0 ⫽2 (14)
torsion 3 6
u␥ f T k K X ⫽4 共 1⫹ s 兲 G s  x xy 冑l x l y ;
A Ts ⫽ (10) (15)
2A l f yd
K Y ⫽4 共 1⫹ s 兲 G s  y xy 冑l x l y
After calculating the steel area for shear load and torsional
moment it is necessary to compute the concrete stress due to these where r 0 is the footing equivalent radius, and
strengths. A simple expression that can be used for this verifica- lx
tion is given in NB-1 共1978兲 as  x ⫽0.02105 ⫹0.98947
ly
td wd
⫹ ⫺1⭐0 (11) ly
tu wu  y ⫽0.02105 ⫹0.98947 (16)
lx
␥ fTk ␥ fVk
td ⫽ , tu ⫽0.22 f cd , wd ⫽ , wu ⫽0.25f cd 0.55共 2⫺ s 兲 h
2A l e Ac xy ⫽1⫹
(12) 冑4l x l y /
In Eq. 共12兲 A c ⫽B(H⫺c) is the useful concrete area of the cross are nondimensional coefficients related to the footing dimensions
section and e⫽B/6. The values of tu and wu must be, respec- and the soil Poisson’s ratio. The masses associated with K X , K Y ,
tively, smaller than 3.92 and 4.41 MPa. and K T are, respectively
The verification of the longitudinal steel area in line elements
4
subjected to biaxial bending and axial load will be treated in m X ⫽m Y ⫽4hl x l y ; m T ⫽ hl x l y 共 l 2x ⫹l 2y 兲 (17)
‘‘Biaxial Moment and Axial Load Constraints’’ considering a new 3
failure surface and a new failure criterion developed by the au- Generally and s can be determined as functions of the
thors. standard penetration test value 共Carter 1983兲, usually abbreviated
as N S PT . An expression for in MN/m3, is ⫽2N S PT . Bowles
共1996兲 has given an expression for the allowable bearing capacity
Soil Parameters
of soil under a rectangular footing in terms of N S PT , h, and l x .
A close interaction exists between the structure and its support However, in the present work, s is considered as given data.
共foundation and soil兲. The load on the structure is transmitted to
the foundation and the soil. A part of the work of the load applied Optimal Design: Problem Formulation
to the soil is stored in the form of strain energy, and damping due
to friction dissipates the remaining part. The physical constants
Design Variables and Cost Function
involved in this problem are functions of the soil geotechnical
parameters. Therefore knowledge of the soil representative prop- The design variables for a problem are the ones that define the
erties is required for structural and foundation analysis. Generally, system. Let b⫽ 关 b 1 b 2 ¯b n 兴 t represent a design variable vector,
the following information is required: Poisson’s ratio of soil, s ; where n⫽total number of design variables. In this work, we are
shear modulus of soil, G s , at several levels of strain 共or magni- dealing with structures having specified geometric configuration
tude of bearing pressure兲; coefficient of subgrade reaction of soil, and material. Therefore only structural dimensions and steel area
, if above parameters are not accurately known; and allowable are the design variables. Thus the design variables related to the
bearing capacity of soil, s , which is a function of the soil footing design are btf ⫽ 关 hA sx A sy l x l y 兴 , where h⫽plate thickness,
strength properties 共friction angle s and cohesion c s ), the system A sx and A sy ⫽steel areas in the cross section perpendicular to axis
foundation geometry, and ␥ so . y and x, respectively, and 2l x and 2l y ⫽plate sides length. The
The main parameters used in the analysis and design process design variables vector for the eth line element is btle
are and s . They can be determined by field or laboratory tests. ⫽ 关 B e H e A se A sse 兴 , where B e and H e ⫽respectively, width and
The coefficient of subgrade reaction, , can be defined as the rate height of the cross section, and A se and A sse ⫽respectively, lon-
q s /⌬, where q s ⫽pressure applied on the footing and ⌬ gitudinal reinforcement steel area and transverse reinforcement
steel area. The footing is discretized into n no nodes and n e ele- Design Constraints
ments, and the framework is discretized into n pe line elements
with n pno nodes. Then the entire design variable vector is given as For structures subjected to dynamic loads, the constraints are
bt ⫽ b btf btl1 ¯btl npe c . grouped into two types: static and dynamic constraints. The static
The objective function that determines the relative merits of constraints are imposed on dimensions of structural elements, the
various designs 共Arora 1989兲 is the construction cost of the struc- steel areas, static displacement and differential displacement,
ture. We consider the costs of concrete, steel, form, and propping stress in soil, and natural frequencies of vibration. These are ex-
form. Determination of the concrete cost is direct, because it is pressed as
proportional to the structural volume. Computation of the steel minimum footing thickness: h⭓h l (23)
cost demands a larger computational effort because the determi-
nation of the steel is not trivial, especially in the line elements. It minimum and maximum value for A sx :
is function of the stresses within the structure. The costs of form 0.0015h⭐A sx ⭐0.04h (24)
and propping form are proportional to external structure surface
area. minimum and maximum value for A sy :
The problem is to minimize the cost of the structure shown in
0.0015h⭐A sy ⭐0.04h (25)
Fig. 6. This structure is composed of footing, columns, and
beams. The cost function for the problem is f (b), given as minimum value for l x and l y : l x ⭓l x min ;l y ⭓l y min (26)
f 共 b兲 ⫽C c ⫹C s ⫹C f ⫹C p f (18)
main natural vibration frequency: 0.5⭓ 1 /⍀ ⭓1.5 (27)
where C c , C s , C f , and C p f ⫽respectively, costs of concrete,
maximum footing static displacement:
steel, form, and propping form:
再
footing:4l x l y hc c
C c ⫽ line element e:B H L c
e e e c
(19)
兩 w si 兩 ⭐z su , i⫽1,...,n no
maximum footing static differential displacement:
(28)
再
footing: 关 2 共 l x A sx ⫹l y A sy 兲 s 兴 c s
C s ⫽ line element e: A ⫹2A
关 se sse 共 B e ⫺2c⫹H e ⫺2c 兲兴 L e s c s
兩 w si ⫺w s j 兩 ⭐d su ,
maximum static bearing capacity:
i⫽1,...,n no , j⫽i⫹1,...,n no (29)
再
(20)
footing: 关 4l x l y ⫹4h 共 l x ⫹l y 兲兴 c f 兩 w si 兩 ⭐ s /␥ ss , i⫽1,...,n no (30)
C f ⫽ beam e: 共 B e ⫹2H e 兲 L e c f (21) verification of the minimum longitudinal steel area for line ele-
column e: 共 B e ⫹H e 兲 2L e c f ments:
C p f ⫽ 兵 beam e:B e L e c p f (22) beam e:A se ⭓0.0015B e H e , e⫽1,...,n pe
In these expressions c c ⫽concrete cost per unit volume, s (31)
column e:A se ⭓0.005B e H e , e⫽1,...,n pe
⫽steel density, c s ⫽steel cost per unit mass, c f ⫽form cost per
unit area, c p f ⫽propping form cost per unit area, and L e ⫽eth verification of the maximum longitudinal steel area for line ele-
element length. ments
2. Lateral displacement of a column cannot exceed L/150, 共34兲–共39兲 and 共42兲–共44兲 contemplate human comfort, as well as
where L⫽column length. machine conditions and structural damage. The limits on dis-
Dynamic constraints need to be imposed over the entire time placements, velocities, and accelerations were taken from the
interval t苸 关 t 0 ,t f 兴 . The predicted behavior of the structures is graphs in Arya et al. 共1978兲.
checked or compared with a number of design requirements,
many of which come from design codes. The dynamic constraints Biaxial Moment and Axial Load Constraints
imposed are
The verification of the longitudinal steel area in line elements
maximum footing displacement: subjected to biaxial flexural moment and axial load was done
using a procedure developed by the authors 共Silva et al. 2001兲.
兩 w di 兩 ⭐z du , i⫽1,...,n no (34) Consider the strengths obtained from the structural analysis: the
axial load N⬅N i , and the flexural moments M a ⬅M lyi and M b
兩 u dp 兩 ⭐z du ; 兩 dp 兩 ⭐z du (35)
⬅M zil
, that compose the point (N,M a ,M b ) in the three-
maximum footing velocity: dimensional space. The resultant moment is M r and
⫽arctan(M b /M a)⫽angle between the direction of M r and a axis.
兩 ẇ di 兩 ⭐ u , i-1,...,n no (36)
Let N ut and N uc ⫽the maximum values of tension and com-
兩 u̇ dp 兩 ⭐ u ; 兩 ˙ dp 兩 ⭐ u (37) pression that can be assumed by N when M a ⫽M b ⫽0, with N
⬎0 for tension and N⬍0 for compression. Let N c ⫽1/2(N ut
maximum footing acceleration: ⫹N uc ) and N r ⫽1/2(N ut ⫺N uc ); let M ua ⫽the maximum value
兩 ẅ di 兩 ⭐a u , i⫽1,...,n no (38) assumed by M a when M b ⫽0 and N⫽N c ; let M ub ⫽the maxi-
mum value assumed by M b when M a ⫽0 and N⫽N c . Authors
兩 ü dp 兩 ⭐a u ; 兩 ¨ dp 兩 ⭐a u (39) have developed the following failure surface for members sub-
jected to biaxial bending and axial loads:
verification of Eq. 共7兲:
d⭓ 1 冑␥ f 兩 M xi 兩 ; d⭓ 1 冑␥ f 兩 M yi 兩 , i⫽1,...,n no (40) 冉 冊 冉 冊 冉
Ma
M ua
␣1
⫹
Mb
M ub
␣2
⫹
N⫺N c
Nr 冊 ␣3
⫺1.0⫽0 if N⬎N c
verification of Eq. 共8兲: (49a)
A sx ⭓
␥ f 兩 M xi 兩
 2d
; A sy ⭓
␥ f 兩 M yi 兩
 2d
, i⫽1,...,n no (41) 冉 冊 冉 冊 冉
Ma
M ua
␣1
⫹
Mb
M ub
␣2
⫹
兩 N⫺N c 兩
Nr 冊 ␣4
⫺1.0⫽0 if N⭐N c
(49b)
maximum framework node displacement:
where the exponents ␣ 1 , ␣ 2 , ␣ 3 , and ␣ 4 were determined using
兩 w di 兩 ⭐z du ; 兩 u di 兩 ⭐z du an optimization procedure. They depend on the cross section di-
(42) mensions, amount and distribution of steel reinforcement, stress-
兩 di 兩 ⭐z du , i⫽n no ⫹1,...,n no ⫹n pno strain characteristics of steel and concrete, concrete cover, and
maximum framework node velocity: size of lateral ties or spiral. When Eq. 共49a兲 or 共49b兲 is equal to
zero, the point (N,M a ,M b ) is on the failure surface; when it is
兩 ẇ di 兩 ⭐ u ; 兩 u̇ di 兩 ⭐ u smaller than zero, the point is below the failure surface; and when
(43) it is larger than zero, the point is outside the failure surface and
兩 ˙ di 兩 ⭐ u , i⫽n no ⫹1,...,n no ⫹n pno
the reinforced concrete element violates the failure constraint.
maximum framework node acceleration: Note that in Eqs. 共49a兲 and 共49b兲, the values of N, M a , and M b
are available from structural analysis and the values of M ua ,
兩 ẅ di 兩 ⭐a u ; 兩 ü di 兩 ⭐a u M ub , N ut , and N uc are calculated from geometry of the section.
(44)
兩 ¨ di 兩 ⭐a u ,i⫽n no ⫹1,...,n no ⫹n pno In the first formulation, the exponents ␣ 1 , ␣ 2 , ␣ 3 , and ␣ 4 are
calculated such that Eq. 共49a兲 or 共49b兲 is always conservative
verification of transverse steel for every node of every line ele- 共i.e., the approximate surface is inside the real surface兲. This is
ment: accomplished by imposing additional constraints on the error be-
tween the real surface and the approximate surface. In the second
A ss e ⭓A ss
V
⫹A ss
T
, e⫽1,...,n pe (45)
formulation, they are calculated to just minimize the sum of
verification of Eq. 共11兲 for every line element: squares of the errors between the real surface and the approximate
surface. This may not give conservative yield surface approxima-
td wd tion everywhere. Optimal solutions with both of these approaches
⫹ ⭐1 (46)
tu wu are obtained and compared later in the paper.
冉 冊 冉 冊 冉
Ma ␣1
⫹
Mb ␣2
⫹
N⫺N c
冊 ␣ 3 ,␣ 4
⫺1⭐0
134 60.96 91.44 8.36 2.00
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 06/15/14. Copyright ASCE. For personal use only; all rights reserved.
134 56.21 87.45 48.06 3.28 134 42.62 78.92 63.16 2.16
135 113.29 124.86 83.66 12.80 135 105.95 138.56 196.83 4.63
136 113.62 119.24 72.48 5.49 136 122.12 141.51 162.20 7.19
137 88.37 86.96 48.65 5.92 137 76.73 73.64 79.61 3.06
138 88.44 87.07 48.63 5.93 138 77.15 74.05 79.61 2.85
139 88.18 87.19 48.44 3.48 139 75.57 73.38 78.74 3.77
140 88.04 87.05 48.45 18.86 140 76.22 73.93 78.75 3.28
141 110.49 130.95 84.8 11.38 141 99.04 146.37 175.98 4.70
142 113.71 129.17 74.34 7.86 142 111.41 148.07 151.48 6.05
143 82.20 81.50 35.72 9.79 143 89.30 81.84 50.68 3.44
144 83.43 82.14 37.26 20.78 144 95.13 86.46 45.47 5.59
145 91.13 37.77 23.56 2.97 145 102.87 48.54 25.73 5.17
146 93.31 40.58 27.15 4.06 146 110.26 54.92 31.11 10.00
minimizing the approximation errors, and so the approximate sur- solve the optimization problem. In that method, a composite func-
face could be unconservative at some points. tional was constructed by combining the cost and the constraint
These results shown in Tables 3 and 4 were obtained in 13 functions. Thus a large number of static and dynamic constraints
iterations and spent 341,112 s of total CPU time. The cost of the in the present problem were treated without much difficulty. A
final design was US$ 25,900.01. forward finite difference method was used to calculate the deriva-
The results shown in Tables 5 and 6 were obtained in 13 itera- tives of the composite functional that were needed in the minimi-
tions and spent 301,907 s of total CPU time. The cost of the final zation algorithm.
design was US$ 24,613.07. To overcome the problem of large computational time needed
As expected, the design with conservative failure surface con- for verification of the bending capacity of line elements over the
straint is slightly more expensive compared to the one with less time domain, an empirical failure surface proposed by the authors
conservative approximation. Also note that the final designs ob- 共Silva et al. 2001兲 that approximates the real failure surface was
tained with the two formulations are quite different from each used. The approximation was done in two ways: one required the
other. For example: A sx for footing final design is 46% different; empirical failure surface to be always conservative, and the other
B e for element 132 design is 39% different. allowed the surface to be unconservative while minimizing the
error between the real surface and the empirical surface.
From the results, it is observed that the adoption of different
Discussion and Conclusions failure surfaces does not give much different final costs, even
though it gives different final designs. It is noted here that the
In this paper, optimal design of reinforced concrete 共RC兲 struc- bending capacity of every line element could have been checked
tures subjected to dynamic loads was considered. Specifically, the using the traditional process of integration of the stress. However,
problem of design of elevated foundations for support of vibra- that process would have taken more than 40 days to solve the
tory machines was formulated as an optimization problem. The example problem on a 500 MHz PC compared to less than 4 days
design system consisted of a foundation modeled as an RC plate, used with the proposed approximate surface. It is concluded that
and columns and beams modeled as line elements. Interaction of it is possible to optimize in an integrated way structures and foun-
the foundation with the soil under it was included in the formu- dation for dynamic loading environment in a reasonable compu-
lation. Static as well as dynamic loads were treated. Constraints tational time.
related to human comfort, strength, and stiffness were checked in For further work, the following studies and analyses are sug-
the solution process. Many of the constraints were time- gested:
dependent. Time history of such constraints was treated in the 1. Development of a more realistic cost function for the prob-
solution process. The augmented Lagrangian method was used to lem.
2. Utilization of a more efficient adjoint variable method to
calculate derivatives of the augmented Lagrangian func-
Table 5. Final Design for Footing, Failure Surface Approximation tional. This would reduce the computational time roughly by
may or may not be Conservative a factor of the number of design variables compared to the
h 共cm兲 A sx 共cm2兲 A sy 共cm2兲 l x 共cm兲 l y 共cm兲 finite difference method.
3. Utilization of the genetic algorithm to solve the same prob-
73.94 20.03 13.82 474.9 457.5
lem and to compare the results obtained.