Optimal Stiffness Design to Limit

Static and Dynamic Wind Responses

of Tall Steel Buildings
CHUN-MAN CHAN

INTRODUCTION The problem can become quite complex if a large scale three
dimensional asymmetrical building structure exhibiting
With the continuing trend of constructing taller and slender
torsional swaying needs to be considered with multiple
buildings with higher strength materials and lighter structural
stiffness constraints under multiple loading conditions. Out of
systems, modern tall steel buildings are wind sensitive
the many given structural members, one needs to determine
structures that are prone to serviceability problems. Two
which members are critical and to what extent the member
important serviceability limit states for tall building design
sizes should be adjusted. Moreover, any modification of
are lateral deformations and motion perceptions under wind
member sizes requires the structure to be reanalyzed. This
loads. Excessive lateral deformations have been found to
traditional iterative resizing process is often tedious and time
cause windows to rack, non-structural partitions to crack and
consuming.
cladding to collapse. Exorbitant oscillations induced by
With the emergence of structural optimization
turbulent wind have been known to cause occupant
technology, the aforementioned resizing design process can
discomfort and even shatter windows.1,2 The functions of tall
be made in an automatic fashion and thus saving much design
buildings may be disrupted due to inadequate design for
time. Structural optimization is nothing but a numerical tool
serviceability. Indeed, the design of tall slender buildings is
that replaces the conventional trial-and-error design approach
generally governed by serviceability stiffness criteria rather
by a systematic goal-oriented design process. In such an
than by ultimate strength safety requirements.
optimization procedure, the numeric intensive tasks of the
Stiffness design is the most challenging and difficult task
analysisdesign cycle are formalized and the optimal member
in tall building design. When presented with a tall building to
sizes are automatically sought while specified design
design, the structural engineer must select a suitable lateral
constraints are simultaneously satisfied. In recent years,
load resisting system to resist wind and earthquake loads. Of
several design professionals3-6 have developed ad hoc
the two lateral loads, the action of wind loads frequently
optimization software for sizing members of tall steel
determines the design of tall buildings. Common lateral load
building frameworks to satisfy static wind drift. Although
resisting systems for tall steel buildings are rigid frames,
their methods are quite efficient, they are useful only for
frames with shear trusses, outrigger trusses, tubular frames,
building structures with single displacement constraint
and super diagonalized trusses. Often times, several
problems. A number of researchers have developed formal
preliminary structural alternatives are initially devised, and
optimization techniques for large-scale structures.7-9
the choice of preliminary selection is then decided based on
However, their efforts focus mostly on the optimization
the engineer's experience, intuition and some approximate
theory with little practical applications to realistic tall
calculations. Once the topology of the lateral load resisting
building structures.
system is defined, the major effort is to size the structural
Although it has long been recognized that structural
members to satisfy both static and dynamic serviceability
optimization techniques have much to offer in engineering
performance requirements. Since tall building structures
practice, the application of such technology for large scale
usually consist of thousands of members and are very
building frameworks has been quite limited to date. Not until
complex in nature, structural engineers are faced with the
recently, the author has successfully developed an efficient
problem of how to distribute efficiently material throughout
optimization technique for the sizing design of tall practical
the structure to limit the static wind drifts and the dynamic
building frameworks subject to multiple drift constraints and
wind vibrations.
the use of discrete standard steel sections.10,11 In this paper,
the author intends to extend the optimization technology to
include both static wind drift and dynamic natural period
Chun-Man Chan is assistant professor, department of civil
constraints. The design optimization problem is first
and structural engineering, Hong Kong University of Science
explicitly defined and then the details of the optimization
and Technology, Kowloon, Hong Kong.
technique

94 ENGINEERING JOURNAL / THIRD QUARTER / 1998

are developed. Two design examples are presented in the area A for W14, W24, W30 and W36 wide-flange sections
paper. The first example is a simple design problem that is from the AISC-LRFD design manual.12 Using the
used to illustrate the algorithm of the optimization technique. regressional relationships, Equation 1, the six basic design
The second example is a full-scale, 50-story practical variables (i.e., A, AY, AZ, IX, IY, IZ) can be reduced to one
building structure through which the effectiveness and cross sectional area variable (i.e., A) for each member.
practicality of the automated optimal resizing technique are If the topology of a building structural system is
illustrated. predefined, the design objective for a steel framework having
i = 1, 2, ..., N members can be expressed as
OPTIMAL DESIGN FORMULATION
N
Unlike the conventional design method, the formal
optimization approach requires the designer to define
∑ w A → minimum
i i (2)
i =1
explicitly a set of design variables, an objective function to
be optimized and some explicit design constraints. In fact, a where
proper formulation of the optimal design problem is a key to Ai = the cross section area for member i
good solutions. A good problem formulation should not only wi = the corresponding material weight/cost coefficient
represent properly the current design problem but should also per unit cross section area
maintain a high level of accuracy during the evolution
process of the design solution. Formulation details of the Stiffness Design Constraints
stiffness design of tall buildings are discussed as follows.
A constraint is defined as a restriction that must be satisfied
Design Variables and Objective for a design to be acceptable or feasible. There are normally
For a skeletal framework with a prescribed geometric layout, two types of serviceability performance constraints to be
the design variables are the six basic cross sectional considered in tall building design.13-16 The first type of
properties of each member, i.e., the axial area (A), two shear constraints
areas (AY, AZ) and three moments of inertia (IX, IY, IZ). In this
regard, a structure with N structural members should be
theoretically described by 6N design variables. However,
since commercially available standard steel sections are to be
used for the design, section properties such as AY, AZ, IX, IY,
IZ can be accurately related to the cross sectional area A by
certain functions through regressional analysis.10 Linear
relationships between reciprocal section properties are herein
adopted and expressed as
1 C
= AY + CAY
'
(1a)
AY A
1 C
= AZ + CAZ
'
(1b)
AZ A
1 CIX
= + CIX
'
(1c)
IX A
1 CIY
= + CIY
'
(1d)
IY A
1 CIZ
= + C'IZ (1e)
IZ A

where
C and C′ = regressional constants derived under the
assumption that the cross-section maintains
within a constant shape group as it changes
size
Fig. 1. Regressional relationships between
Figure 1 shows graphically the linear reciprocal relationships
strong moment of inertia Iz and cross-section
between the strong moment of inertia IZ and cross sectional area A for selected AISC W-shapes.

ENGINEERING JOURNAL / THIRD QUARTER / 1998 95

concerns with the static lateral deformations under wind M X mX MY mY M Z mZ 
loads. The second type deals with the dynamic wind motion + + +  dx (4)
GI X EIY EI Z  i
perception.
1. Static Wind Drift. where

Excessive magnitudes of lateral wind deflections may cause Li = length of member i

objectionable damage to nonstructural components or E, G = axial and shear elastic material moduli
increase the chance of building instability. The static A, Ay, Az = axial and shear areas for the cross-section
serviceability of a building can be secured by carefully I x, Iy, Iz = torsional and flexural moments of inertia for
controlling lateral deflections within certain allowable limits. the cross-section
Typically, two kinds of lateral deflections need be FX, FY, FZ,
considered. One is the overall building drift, defined as the MX, MY, MZ = member forces and moments due to the ctual
total lateral deflection at the roof top divided by the building wind loads
height, H. Another is the interstory drift, which is the fX, fY, fZ,
differential lateral translation of two adjacent floor levels per mX, mY, mZ = member forces and moments due to a unit
story height, h. While the overall building drift represents the virtual load applied at the location of and in
average lateral translation of a building under wind effects, the sense of δ
nonstructural damage is more dependent on interstory drift. Given a particular section type for each member, one can
The normally accepted range of drift ratio limits for buildings substitute the regressional relationships Equation 1 into
appears to be 1/750 to 1/250, with 1/400 being typical.13 Equation 4 to express instantaneously the displacement δ as
Consider a general 3D steel building framework having i explicit functions of the sizing variable Ai alone as
= 1, 2, ..., N members (or member fabrication groups), j = 1,
N
 ci '
2, ..., M storys, k = 1, 2, ..., S column lines under l = 1, 2, ...,
L lateral loading conditions. The drift constraints can be
δ ( Ai ) = ∑  A + c  i (5)
i =1 i
expressed as
(δ kjl − δ kj −1l ) where the virtual strain energy coefficients, ci and the
d kjl = ≤ d Uj correction factors ci' are respectively given by
hj
Li  FX f X + MY mY CIY + M Z mZ CIZ
(k = 1, 2, ..., S); (j = 1, 2, ..., M); (l = 1, 2, ..., L) (3a) ci = ∫
0

 E
δ kMl
d kMl = ≤ d HU
H FY f Y CAY + FZ f Z CAZ + M X mX CIX 
+  dx (6a)
G i
(k = 1, 2, ..., S); (l = 1, 2, ..., L) (3b)
Equation 3a defines interstory drift ratio dkjl, where δkjl and Li  '
MY mY CIY + M Z mZ CIZ
'

δkj−1l are the lateral translations on a column line k at two

ci' = ∫0

 E
adjacent j and j − 1 floor levels under lateral loading ' 
'
FY f Y CAY + FZ f Z CAY
'
+ M X mX CIX
condition l, hj, is the jth story height and d Uj is the allowable +  dx (6b)
G i
jth story drift limit; Equation 3b defines the overall building
drift ratio dkMl, where δkMl is the lateral translation on column As both the interstory drift constraints Equation 3a and the
line k at the top floor level M under lateral loading condition overall drift constraints Equation 3b are similar to each
l, H is the building height and d HU is the allowable overall other, they can be simplified collectively into one single type
building drift limit. Note that the drift constraints Equation 3 of constraints for convenience of future discussion. The
are expressed in implicit form. In order to facilitate subscript (k, j, l) in the drift constraints Equation 3 can be
numerical solution of the design optimization problem, the changed to a single subscript d to represent the dth constraint
drift constraints Equation 3 must be first expressed explicitly in a collective set of Nd drift constraints, where Nd is the total
in terms of the design variables. number of interstory and overall building drift constraints for
By the principle of virtual work, a displacement of an M-story framework having concern for drift control of S
interest δ can be expressed as column lines under L lateral loading conditions. Using the
explicit displacement expression Equation 5, the drift
N Li 
∑∫
F f F f F f constraints Equation 3 can be expressed in terms of the
δ=  X X + Y Y + Z Z
i =1
0  EA GAY GAZ design variable Ai as

96 ENGINEERING JOURNAL / THIRD QUARTER / 1998

 N
 eid  N Li  F2 F2 F2 M2 M2 M2 
dd = ∑  + eid'  ≤ ddU (d=1, 2, …, Nd) (7) W= ∑∫  X + Y + Z + X + Y + Z  dx (10)
i =1  i   EA GAY GAZ GI X EIY EI Z  i
A i =1
0

where for interstory drifts, where

cikjl − cikj −1l '
cikjl − cikj
'
− 1l
FX, FY, FZ,
eid = ; eid' = (8a, b) MX, MY, MZ = internal member forces and moments due to
hj hj
the external inertia loading condition, F
and for overall building drifts, Using again the linear regression section relationships
cikMl ' '
cikMl Equation 1, the total internal work Equation 10 can be
eid = ; eid = (8c, d) expressed as explicit functions of the sizing variables Ai
H H
alone as
2. Wind-Induced Vibrations
N e 
It is widely accepted that wind-induced acceleration has W ( Ai ) = ∑  iτ + ei'τ  (11)
i =1  Ai 
become the standard for evaluation of motion perception in
buildings.15,16 Semi-empirical formulas have been derived where
from numerous wind tunnel studies to predict the acceleration
responses of buildings in an urban environment. These τ = the fundamental mode of vibration
formulas are expressed in terms of the wind velocity, the eiτ and e iτ' = are respectively given by
building's shape, damping ratio, mass and fundamental
Li  FX2 + MY2 CIY + M Z2 CIZ
natural periods.2,17 Wind-induced accelerations can be
reduced by changing the building shape to maintain better
eiτ = ∫0

 E
aerodynamic stability, but unfortunately this is often beyond
the control of the engineer. In practice, the mass of a building FY2 C AY + FZ2 C AZ + M 2X CIX 
is hardly variant and it has not been common to design +  dx (12a)
G i
damping into a structural system. A common approach to
suppress wind-induced vibrations remains to limit the natural Li  MY2 CIY
'
+ M Z2 CIZ
'
periods.18
Using the Rayleigh method, the fundamental circular
ei'τ = ∫
0

 E
frequency of vibration ω for an undamped structure can be
FY2 C 'AY + FZ2 C 'AY + M 2X CIX
'

found by equating the maximum kinetic energy of the system +  dx (12b)
at zero displacement to the work done by the inertia forces as G i
the system moves from zero to maximum displacement as
follow: By definition, the natural period T is inversely related to the
1 2 T circular frequency ω as
ω φ Mφ (kinetic energy)
2 2π
T= (13)
1
= φT F (work done by inertia forces) (9a) ω
2
To limit the natural period T, one can increase ω by
φT F φ T Kφ K*
⇒ ω2 = T = T = (9b) increasing the structure stiffness according to Equations 9
φ Mφ φ Mφ M * and 13. If one assumes temporarily constant initial inertia
where force F such that the internal member forces and moments are
invariant, an increase in the structure stiffness can be
M, M* = structure mass matrix and generalized mass
achieved by increasing member sizes and thus reducing the
K, K* = structure stiffness matrix and generalized
internal work W. Assuming that the initial mode shape φo,
stiffness
inertia force F and structure mass M are invariant, one can
φ = the computed mode shape under the inertia
limit the natural period T by the following explicit equivalent
force matrix F
design constraint on internal work as
Initially, if the structure has circular frequency ωo and mode
N e 
shape φ0, the inertia force F can be obtained as ω 02 Mφ0. For W ( Ai ) = ∑  iτ + ei'τ  ≤ W U (14)
i =1  Ai 
an undamped skeletal framework, the external work done by
inertia force F is equal to the total internal work W
represented by the total sum of internal strain energy for all where the allowable work limit WU can be shown equal to
members as (TU / To)2(Wo) in which

ENGINEERING JOURNAL / THIRD QUARTER / 1998 97

 TU = targeted natural period method, which has been shown computationally very efficient
To = initial natural period for large-scale structures is employed.10,11 The OC method
Wo = initial strain energy for the framework having the first involves the derivation of a set of necessary optimality
original member sizes criteria for the design. Then, a recursive algorithm is applied
to resize the structure to satisfy the optimality conditions and
Under the influence of dynamic wind loads, tall buildings
thus indirectly optimize the structure. The basic essence of
vibrate in the alongwind, acrosswind and torsional directions.
the OC technique is herein presented in this paper. Further
Normally, there is one natural period to be limited
details of the technique can be found in references 10 and 11.
respectively in each of these three wind sensitive directions,
i.e. two translational and one torsional mode of vibrations. Optimality Criteria
While slight modifications on the Rayleigh method are
needed to obtain a higher mode shape and frequency,19 the In classical optimization theory, the necessary optimality
same form of explicit strain energy constraints as given in criteria for the constrained optimal design problem (Equation
Equation 14 can be expressed for (τ = 1, 2, ..., Nτ ) multiple 16) can be obtained indirectly by first converting the
natural period constraints as constrained problem to an unconstrained Lagrangian function
and then solving for the stationary condition of the
N e 
Wτ ( Ai ) = ∑  iτ + ei'τ  ≤ Wτu (τ=1, 2, …, Nτ) (15) Lagrangian function. Temporarily omitting the sizing
i = 1  Ai  constraints Equation 16c, the unconstrained Lagrangian
where function can be formulated as,21-24
N N,
 N e  
∑ ∑ ∑
Nτ = total number of natural period constraints
L( Ai , λ s ) = wi Ai + λ s   is + eis'  − gUs  (17)
Recognizing the fact that the explicit lateral drift Equation 7 i =1 s =1  i =1  Ai  
and equivalent period constraint Equation 15 are very much
alike, one can express collectively the stiffness design The first part of the Lagrangian function describes the
optimization problem in terms of sizing variables Ai as structure weight, whereas the second part involves the design
follows: constraints multiplied by their corresponding Lagrange
multipliers, λs. For minimization problems, the Lagrange
Minimize:
multipliers must be positive for active constraints (i.e. gs =
N
g Us ) or equal to zero for inactive constraints (i.e. gs < g Us ).
∑ wi Ai (16a)
i =1 Differentiating the Lagrangian function Equation 17 with
subject to: respect to the design variables Ai and setting the derivative to
zero, one can obtain the following necessary conditions at the
N
 eis ' 
gs ( Ai ) = ∑  A + e  ≤ g is
U
s (s=1, 2, …, Ns) (16b) optimum as
i =1 i
Ns
 −eis 
AiL ≤ Ai ≤ AiU (i = 1, 2, ..., N) (16c) wi + ∑ λ  A s
i 
2
 = 0 (i = 1, 2, …, N) (18)
s=1
Equation 16b denotes the stiffness constraints, where the
constraint gs with subscript s = 1, 2, ..., Nd represents the which can be rearranged to
static drift constraint in which gUs = ddU as given in Equation Ns

∑λ
eis
7; the constraint gs with subscript s = Nd + 1, ..., Nd + Nτ s = 1 (i = 1, 2, …, N) (19)
s=1 wi Ai2
represents the period constraint in which gUs = WτU as given
in Equation 15. In the case of having a set of multiple drift The optimality criteria for the optimal design problem
and period constraints, the value Ns represents the total Equation 16 are shown in Equation 19, which have a
number of stiffness design constraints, i.e. equal to Nd + Nτ significant physical meaning for design. Each Lagrangian
Equation 16c specifies each design variable Ai to be selected multiplier, λs, can be interpreted as a sensitivity weighting
from its lower bound size AiL and upper bound size AiU . factor which measures the importance of the corresponding
sth constraint to the optimal design. The larger the value of
OPTIMIZATION TECHNIQUE λs, the more influential is the constraint to the optimum
design. When a constraint does not affect the design, the
Upon formulating the design optimization problem for corresponding λs diminishes to zero. The expression eis/WiA2i
serviceability requirements of tall steel building framework, in Equation 19 represents the strain energy per unit
the next task is to apply a suitable method to solve the weight/cost, which is so called the strain energy density, for
problem. A rigorously derived Optimality Criteria (OC) member i with respect to

98 ENGINEERING JOURNAL / THIRD QUARTER / 1998

the sth constraint. Therefore, the optimality conditions Once the continuous optimal solution is obtained, one
Equation 19 specify that each member must contribute the needs to finalize the design with discrete standard sections. A
same weighted amount of strain energy densities at the pseudo-discrete OC technique20 is herein adopted to achieve
optimum. For the case of only one stiffness constraint, λs a smooth transition from the continuous variable design to the
becomes a constant and the strain energy density, eis / wi Ai2 , optimal final design using discrete standard sections. The
for each member is uniform or, in fact, equal to 1/λs. essence of the pseudo-discrete OC technique is to maintain
the least changes in the structure cost while members are
Recursive Resizing Algorithm progressively assigned standard section sizes. Details of the
pseudo-discrete OC technique can be found in references 11
Equations 19 are stationary conditions, equal to unity at the and 20.
optimum, that can be used to derive a linear recursive
relation for the active sizing variables Ai as follows, Design Procedure
 1  i λ seis
N   The procedure to implement the optimal design method for
Aiv +1 = Aiv 1 +  ∑
 η  s=1 wi Ai2
− 1  (i = 1, 2, …, N −ξ)
 
v
(20) limiting lateral drifts and natural periods of tall steel building
 frameworks is listed as follows:
where 1) Analyze the structure under service wind loads and
virtual loads.
η = step-size parameter that controls convergence of
2) Design members to satisfy strength requirements in
the recursive process
accordance with a steel design standard and adopt the
v and
strength-based member size as the minimum size
v + 1 = successive iterations
bound for each member.
ξ = number of inactive members which are assigned
3) Establish the explicit drift constraint Equation 7.
to either their limiting sizes or fixed discrete sizes
4) Perform an eigenvalue analysis to find the initial
In order to apply Equation 20 to find the new sizing variables natural periods (To) and corresponding mode shapes
Aiv +! , the current values of the Lagrange multipliers λ vs must (φo) for the framework.
first be determined. Considering the change of ( gtv +! − gtv ) in 5) Compute an inertia force F = ω o2 Mφo for each
vibration mode and analyze statically the framework
the tth constraint due to the change of ( Aiv +! − giv ) in the (N –
under inertia force, i.e. F = Kφ.
ξ) active sizing variables, the Ns Lagrange multipliers for the 6) Compute the explicit equivalent period constraints
corresponding constraints can be expressed as a set of Ns Equation 15.
simultaneous equations 7) Combine both explicit drift and period constraints,
N −ξ N −ξ and establish size bounds for members to form the
Ns
 eit eis 
∑ ∑ ∑
eit
λ vs   = − η( gtU − gtv ) (t=1, 2, …, Ns) explicit design optimization problem.
i =1  i i  v
3 v
s=1
wA A
i =1 i 8) Apply the recursive optimization algorithm Equations
20 and 21 until optimal member sizes are obtained.
Equations 20 and 21 together form an iterative algorithm to
9) If the structure is statically indeterminate, return to
solve for the continuous design optimization problem posed
step (1) to repeat the design process until the structure
in Equation 16. Given a set of sizing variables Aiv , the set of
weight converges; otherwise, go to step 10.
unknown Lagrange multipliers λ vs is then determined by 10) Apply a pseudo-discrete section selection procedure to
solving the linear simultaneous Equation 21. With the current finalize the optimal design with discrete standard
values of λ vs , the new set of sizing variables Α1v + 1 can then section sizes and terminate with the optimal building
structure.
be sized using Equation 20. Such an iterative algorithm can
be programmed to repeatedly solve for the sizing variables
ILLUSTRATIVE EXAMPLES
and the associated Lagrange multipliers until their
convergence occurs. At convergence, the optimality criteria Example 1: A 3-Bar Truss
(Equation 19) are satisfied such that the optimal member A simple truss involving two diagonal members in two
sizes are obtained and the importance of each constraint to orthogonal directions and a common vertical column is shown
the optimal design are also determined by the final values of in Figure 2. The purpose of this simple truss example is to
λs. Note that the optimization technique developed herein illustrate the application of the optimization technique for
allows for simultaneous consideration of multiple sets of minimum weight design involving multiple stiffness
serviceability constraints involving the top building drift constraints. The truss is loaded at its top node by two lateral
together with interstory drifts and the translational periods as point loads taken as two separate loading conditions in the X
well as the rotational periods. and

ENGINEERING JOURNAL / THIRD QUARTER / 1998 99

 Table 1.
Analysis Results of the 3-Bar Truss
Actual Load Virtual Load Weight Energy Coeff.
Coeff.
Member Fix Fiy fix fiy wi= ρiLi eix eiy
1 0 14.142 0 1.4142 14.142 0 2.8284
2 5.774 –10.000 0.5774 –1.000 10.000 0.3333 1.0000
3 –11.547 0 –1.1547 0 20.000 2.6667 0

Y orthogonal directions. The design objective is to minimize The design process converges to the optimal structure
the structure weight subject to the two top drift constraints weight of 1497.1 after ten iterations, where both the X and Y
corresponding to the two respective X and Y direction loads. drift ratios are found reaching their limit of 0.01. By
The structural geometry, material properties, and the two inspection, the 45° diagonal in the Y direction is structurally
drift ratio limits at the top node are shown in Figure 2. less efficient to resist lateral load than the 30° diagonal in the
Member sizes are assumed to be unbounded and X-direction. Therefore, the truss is more vulnerable under the
nondimensional units of measurement are used for ease of Y-direction load and the design is influenced more by the Y-
illustration. direction drift. Such an intuition is evidently shown in Table
To facilitate the presentation, a subscript i denotes the 2, where the corresponding Lagrange multiplier has a larger
member number, subscript x represents information value of λy = 83481 than the value of λx = 66225 for the X-
pertaining to the X-direction load case and subscript y direction drift. At the optimum of the design problem with
corresponds to the Y-direction load case. Results of the multiple constraints, Table 3 shows that the weighted sum of
analysis of the statically determinate truss is shown in Table the virtual energy densities for each member is equal to unity
1. so as to satisfy the optimality criteria Equations 19 for the
To commence the optimization process, an initial set of multiple drift constraint problem as shown in Table 3.
member sizes ( A10 = A20 = A30 = 1 ) is arbitrarily selected. Example 2: A 50-story Building
Employing a step-size parameter η = 2, the set of
simultaneous equation Equations 21 can be established to A 50-story 7-bay by 10-bay practical building framework
solve for the two Lagrange multipliers λx and λy, associated with 5400 members which is shown in Figure 3 and studied
with the X and Y drift constraints. Once the current values of
λx and λy are determined, the member sizes Ai can then be
resized using the recursive relations Equations 20. Table 2
shows the results of the iterative design optimization process.
Further details of the optimization technique for the solution
of this example are given in Appendix A.

Fig. 2. Three-bar truss example. Fig. 3. 3D model of a 50-story framework.

100 ENGINEERING JOURNAL / THIRD QUARTER / 1998

 Table 2.
Iterative Optimization History for the 3-Bar Truss
Iteration Simultaneous Equation 21 λ vx Structure
v A1v A2v A3v g vx g vy L.H.S. R.H.S. λ vx Weight

0 1.000 1.000 1.000 0.30000 0.38284 3.6667E-03 3.3333E-04 0.8800 225.6 44.14
3.3333E-04 6.6568E-03 1.1285 158.2
1 2.082 1.667 2.004 0.15305 0.19581 4.6570E-04 7.1933E-05 0.4392 850.2 86.20
7.1933E-05 8.4232E-04 0.5674 601.1
2 3.928 3.486 3.830 0.07918 0.10070 6.5900E-05 7.8678E-06 0.2176 3037.7 167.01
7.8678E-06 1.1697E-04 0.2821 2207.4
3 7.584 6.361 7.202 0.04226 0.05302 9.9481E-06 1.2949E-06 0.1068 9758.8 314.92
1.2949E-06 1.6853E-05 0.1390 7500.7
4 13.682 11.633 12.634 0.02397 0.02927 1.8337E-06 2.1174E-07 0.0519 25781.6 562.51
2.1174E-07 2.8438E-06 0.0678 21923.7
5 22.865 18.933 19.921 0.01515 0.01765 4.6611E-07 4.9113E-08 0.0254 49395.6 911.11
4.9113E-08 6.2058E-07 0.0330 49196.0
6 32.949 26.807 26.491 0.01131 0.01232 1.9703E-07 1.7304E-08 0.0139 64078.1 1263.85
1.7304E-08 2.1006E-07 0.0169 75385.4
7 39.354 31.448 29.371 0.01014 0.01037 1.4390E-07 1.0717E-08 0.0104 66200.2 1458.46
1.0717E-08 1.2497E-07 0.0111 83154.5
8 40.807 32.453 29.712 0.01000 0.01001 1.3881E-07 9.7521E-09 0.0100 66225.4 1495.86
9.7521E-09 1.1251E-07 0.0100 83480.6
9 40.861 32.489 29.715 0.01000 0.01000 1.3875E-07 9.7198E-09 0.0100 66225.4 1497.06
9.7198E-09 1.1208E-07 0.0100 83481.0
10 40.861 32.489 29.715 0.01000 0.01000 1.3875E-07 9.7197E-09 0.0100 66225.4 1497.06
9.7197E-09 1.1208E-07 0.0100 83481.0

in reference 11 for optimal static drift design is herein are W14 shapes; and columns are also W14 shapes except
considered. The purpose of this example is to illustrate the that the cruciform columns in the core (see Figure 5) use
effectiveness and practical application of the design pairs of two W14 shapes oriented perpendicular to each
automatic optimal sizing technique for large-scale 3D tall other. To satisfy practical construction requirements, beams
building frameworks subject to both static drift and dynamic that are grouped together on each floor as shown in Figure 5
natural period constraints. For a bay width of 4.57 m (15 ft) are specified to have the same section over two adjacent
and a story height of 3.66 m (12 ft), the framework has a storys, while columns in each line are grouped together as
height-to-width aspect ratio of 5.7 in the X-direction and 4.0 having a common section over two adjacent storys, as are
in the Y-direction. Details of the framework are shown in its diagonals in each span. To establish the minimum size
elevation and plan views in Figures 4 and 5. As illustrated in boundary for each member, member strength design is
these figures, one corner of the building is cut off at a 45° carried out after each response analysis process in
angle and the core is eccentrically located in the Y-direction accordance with the AISC LRFD design standard.12 To
(i.e., shifted towards the south exterior face). These features account for serviceability lateral swaying and twisting of
create a built-in asymmetry that causes natural twisting of 1/400 is applied on all the columns at the most distant
the framework under lateral loadings. The framework column lines A, B and C. For the control of dynamic wind-
consists of exterior moment frames and a braced core. All induced vibration, two lateral sway periods about the two
beams and columns are rigidly connected while the diagonal orthogonal X and Y directions of the buildings are limited to
braces are simply connected. As shown in Figure 4, two-story 7.5 seconds respectively.
K-bracing modules are used on both the south and north faces To study the practical application of the optimal design
of the core while single-story knee-bracing is used in the technique, three separate runs are conducted for this
west and east faces of the core to ensure accessibility to the framework. The first run is to determine only member-by-
elevators in the core. member strength design for the structure. The second run
American AISC standard sections are used to size the considers the stiffness optimization subject to lateral drift
members as follows: beams are W24 shapes; diagonals constraints alone

ENGINEERING JOURNAL/THIRD QUARTER/1998 101

 Table 3.
Optimality Criteria for the 3-Bar Truss
λx=66225.4 λy=83481.0
Optimality
Virtual Strain Energy Virtual Strain Energy Density
Condition
eix eiy eix eiy e eiy
λ x ix 2 + λ y
Member i Optimal Al Ai Ai wi Ai2 wi Ai2 wi ai wi Ai2
1 40.861 0 6.922×10–3 0 1.198×105 0.000 + 1.000 = 1.000
2 32.489 1.026×10–3 3.078×10–3 3.158×10–6 9.474×10–6 0.209 + 0.791 = 1.000
3 29.715 8.974×10–3 0 1.510×10–5 0 1.000 + 0.000 = 1.000
Sum = 0.01000 0.01000

while using strength-based sizes as the minimum size bounds. of this framework is governed by lateral stiffness criteria
The last run is similar to the second run; but it includes both rather than member strength requirements. An increase in
the static drift and the period constraints. For the last two structure weight to 5133.0 tons is found when only lateral
stiffness design runs, a pseudo-discrete OC resizing drift constraints are considered. When both lateral drift and
technique20 is applied upon convergence of the continuous period constraints are involved in the design, an additional
optimization solution to finalize the design using discrete slight increase in structure weight to 5414.0 tons is found for
standard sections. the framework. Such a result indicates that the period
Rapid and steady convergence to the optimal design is constraints are more active to the design, or in other words,
found for all three design runs on this large-scale framework they control the design somewhat more than the drift
example. The history of the design process for the three runs constraints. As shown in Figure 6, all three runs exhibit
of the framework is shown in Figure 6. For the first case stable and rapid
where the structure is designed for strength alone, the
structure results in a weight of 3848.8 tons. The optimal
results of the other two stiffness design runs clearly indicate
that the design

Fig. 5. Front and side elevations of

Fig. 4. Plan view of 50-story framework example. 50-story framework example.

102 ENGINEERING JOURNAL/THIRD QUARTER/1998

convergence. Such a quick convergence can be explained by given set of design constraints such that the optimal objective
the fact that the internal member force distribution is quite for the design is achieved. As a result, the optimal design
insensitive to changes in member sizes for building achieved may provide insights to engineers for further design
frameworks. Such a peculiar behavior of building improvements. Moreover, the design results obtained are
frameworks results in good quality approximations of the often found to match closely our engineering intuition
explicit energy based stiffness constraints with consequent concerning the structural behaviour of building frameworks.
rapid convergence to the optimal design.
CONCLUDING REMARKS
Figure 7 shows the deflected profiles of the framework in
the two respective X and Y directions after completion of the An automatic resizing technique for the optimal stiffness
design optimization process involving both lateral drift and design of tall steel building frameworks is presented in this
period constraints. While the X-direction lateral response is paper. Based on the results presented in this paper, the
controlled by drift constraints, the design in the Y-direction following conclusions can be made:
is governed by the corresponding sway period constraints.
1) Economy: Not only the most economical design is
For the X-direction drift, the interstory drift ratios of the
achieved by the optimal resizing technique while
south-east and north-east corner column lines B and C above
satisfying all design constraint requirements, but
approximately the 10th floor are found active and are shown
there is also a considerable saving in designer time
having a slope parallel to the drift ratio limit of 1/400. A
and cost.
slight difference in lateral deflection of about 2.5 cm is found
2) Effectiveness: Rapid and steady convergence is
between the column line B and the column line C in the upper
generally found since the formulation of the design
portion of the building, indicating a twisting rotation of 2.5
problem using the energy approach has exploited to
cm in 45.7 m or about 1/1800 radian. Such a result seems to
advantage the peculiar behavior of building
indicate that for the optimal design of an asymmetric
framework, that the member force distributions for
framework, the lateral load resisting system will be sized by
such structures are somewhat insensitive to changes
the OC procedure such as to distribute its stiffness so that
in member sizes.
little or no building torsion occurs. For the Y-direction
3) Practicality: The results of the three dimensional 50-
lateral response, while lateral drifts are not active and no
story building example have demonstrated the
building twisting is shown, the sway period is found active
practical application of the optimal design technique.
with a value close to its limit of 7.5 sec. This result indicates
The technique developed holds much promise for a
that the optimal design technique may not necessarily end
powerful tool for the design of large-scale tall steel
with a fully constrained design having all the constraints to
building frameworks encountered in professional
reach their limits. Indeed, the automatic resizing technique
practice.
will seek for the best response of the structure within the

Fig. 6. Design history of 50-story framework example. Fig. 7. Deflected profile for 50-story framework example.

ENGINEERING JOURNAL / THIRD QUARTER / 1998 103

 ACKNOWLEDGEMENTS Serviceability: A Critical Appraisal and Research Needs,"
Journal of Structural Engineering, Vol. 112, No. 12, 1986, pp.
The authors wish to thank the Hong Kong University of 2646–2664.
Science and Technology and the Research Grants Council of 14. ASCE Task Committee on Drift Control of Steel Building
Hong Kong for providing financial support for this research. Structures, "Wind Drift Design of Steel-Framed Buildings:
State-of-the-Art Report," Journal of Structural Engineering,
Vol. 114, No. 9, 1988, pp. 2085–2108.
REFERENCES
15. Ellingwood, B., "Serviceability Guidelines for Steel Structures,"
1. Council on Tall Buildings & Urban Habitat,Structural Design Engineering Structures, AISC, 1st Qtr., 1989, pp. 1–8.
of Tall Steel Buildings, Monograph on Planning and Design of 16. Griffis, L., "Serviceability Limit States Under Wind Load,"
Tall Buildings, Vol. SB, Chapter SB-5, 1979. Engineering Structures, AISC, 1st Qtr., 1993, pp. 1–16.
2. Tallin, A. and Ellingwood, B., "Serviceability Limit States: 17. Vickery, B. J., Isyumov, N., and Davenport, A. G., "The Role of
Wind Induced Vibrations," ASCE, Journal of Structural Damping, Mass and Stiffness in the Reduction of Wind Effects
Engineering, Vol. 110, No. 10, 1984, pp. 2424–2437. on Structures," Journal of Wind Engineering and Industrial
3. Thorton, C. H., Joseph, L., and Scarangello, T. Z., Aerodynamics, Vol. 11, 1983, pp. 285–294.
"Optimization of Tall Structures for Wind Loading,"Journal of 18. Henige, R. A., "Structural Optimization to Limit Natural
Wind Engineering and Industrial Aerodynamics, Vol. 36, 1990, Periods," Proc. of Tenth Conf. on Electronic Computation,
pp. 235–244. Indianapolis, IN, April 29-May 1, 1991, pp. 253–260.
4. Baker, W., "Sizing Techniques for Lateral Systems in Multi- 19. Briggs, J. M., Introduction to Structural Dynamics, McGraw-
Story Steel Buildings," Proc. of the 4th World Congress on Tall Hill Book Co., 1964.
Buildings, Council on Tall Buildings and Urban Habitat, Hong 20. Chan, C. -M., "A Pseudo-Discrete Optimization Technique for
Kong, 1990, pp. 875–868. Tall Steel Building Design Using Standard Steel Sections,"
5. Charney, F. A., "The Use of Displacement Participation Factors Proc. of Int. Conf. on Advances in Steel Structures, Hong Kong,
in the Optimization of Wind Sensitive Buildings,"Proc. of the Dec. 11-14, 1996, pp. 983–988.
Ninth Structures Congress, ASCE, Indianapolis, IN, 1991, pp. 21. Kirsch, U., Structural Optimization—Fundamentals and
545–547. Applications, Springer-Verlag, Germany, 1993.
6. Gilsanz, R. and Carlson A., "Optimization in Building Design," 22. Templeman, A. B., Civil Engineering Systems, The Macmillan
Proc. of the 2nd Int. Conf. on Computer Aided Optimum Design Press Ltd., 1982.
of Structures 91, Hernandez S. and Brebbia C. A., eds., Boston, 23. Morris, A. J., Foundations of Structural Optimization: A
MA, 1991, pp. 481–492. Unified Approach, John Wiley & Sons, 1982.
7. Khot, N. S., Berke, L., and Venkayya, V.B., "Comparison of 24. Pike, R. W., Optimization for Engineering Systems, Van
Optimality Criteria Algorithms for Minimum Weight Design of Nostrand Reinhold Co., 1986.
Structures," AIAA Journal, Vol. 17, No. 2, 1979, pp. 182–190.
8. Khan, M. R., Willmert, K. D., and Thorton, W. A., "A New APPENDIX A
Optimality Criteria Method for Large Scale Structures,"Proc.
AIAA/ASME/SAE 19th Struct. Dynamics and Mat. Conf., AIAA, This appendix describes further details of the OC
Washington, D.C., 1978, pp. 47–58. optimization technique for the solution of the optimal drift
9. Zhou, M. and Rozvany, G.I.N., "DCOC: An Optimality Criteria design of the simple 3-bar truss example as shown in Figure
Method for Large Systems: Part I: Theory," Struct. 2. Given that the truss structure is statically determinate,
Optimization, Vol. 5, No. 1-2, 1992, pp. 12–25.
member forces can be calculated independently of the
10. Chan, C.-M. and Grierson, G. E., "An Efficient Resizing
Technique for the Design of Tall Steel Buildings Subject to member sizes and are listed in Table 1. Using the assumed
Multiple Drift Constraints," The Structural Design of Tall values of the material density, the modulus of elasticity and
Buildings, Vol. 2, 1993, pp. 17–32. member lengths shown in Figure 2, the strain energy
11. Chan, C. M., Grierson, D. E., and Sherbourne, A. N., coefficients in Equation 7 are calculated as
"Automatic Optimal Design of Tall Steel Building
Frameworks," Journal of Structural Engineering, ASCE, Vol.  FfL  1 0 • 0 • 10 2 1
121, No. 5, 1995, pp. 838–847. e1x =   • = • =0
 E  1x h 100 10
12. American Institute of Steel Construction, Manual of Steel
Construction—Load and Resistance Factor Design, 1st Ed.,
 FfL  1 5774
. • 0.5774 • 10 1
Chicago, IL, 1986. e2 x =   • = • = 0.03333
13. Ad Hoc Committee on Serviceability Research, "Structural  E  2x h 100 10

104 ENGINEERING JOURNAL / THIRD QUARTER / 1998

  FfL  1 − 11547
. • −11547
. • 20 1 3

∑A
e3x =  eix
 • = • − η(d xU − d x0 ) = 0.3 − 2(0.01 − 0.3) = 0880
.
 E  3x h 100 10 0
i =1 i
= 0.26667
3
eiy
 FfL 
e1 y = 
1 14142
 • =
. • 14142
. • 10 2 1
• = 0.28284 ∑A 0
− η(d Uy − d y0 ) = 0.382843 − 2(0.01 − 0.382843)
 E  1y h 100 10 i =1 i
= 11285
.
 FfL  1 − 10 • −1 • 10 1
e2 y =  • = • = 01
. Therefore, the simultaneous equations Equation 21 in terms
 E  2y h 100 10 of λ 0x and λ 0y can be expressed as
 FfL  1 0 • 0 • 20 1 λ 0x  08800
e3 y =   • = • =0 36667
. × 10−3 33333
. × 10−4  . 
 E  3y h 100 10    0 =  
33333
. × 10− 4 6.6568 × 10− 3  λ y  11285
. 
Note that since the structure is a truss, the correction Solving the above simultaneous equation, the Lagrange
coefficient terms e′ in Equation 7 are equal to zero. multipliers are found such that
Therefore, the X and Y drift constraints can be written as
λ 0x = 225.62 and λ 0y =158.23.
e e e 0.03333 0.26667
g x = 1x + 2 x + 3x = + ≤ 0.01
A1 A2 A3 A2 A3 Having obtained λ 0x and λ 0y , a new set of member sizes can
e1y e2 y e3 y 0.028284 01 .
gy = + + = + ≤ 0.01 be found from the recursive relationships of Equation 20 as
A1 A2 A3 A1 A2
 1
2
 
∑
Using an initial set of member sizes A10 = A 20 = A 30 = 1 , the X e
A11 = A10 1 +  λ s 1s 2 − 1 
and Y drift ratios can be obtained respectively as  η  s=1 w1 A1  0 
g x0 = 0.3000 and g 0y = 0.38284 , which are found to violate  1 0
= 11 +  22562. •
the drift ratio limit of 0.01, indicating that the current design  2 10 2 • 12
is too flexible. 0.28285 
With the current values of member sizes, the +158.23 • − 1 
10 2 • 12

simultaneous linear equations for the two Lagrange
multipliers λx and λy associated with the X and Y drift = 2.082
constraints can then be established. For the initial set of   
1
2

∑
e
member sizes ( A10 = A 20 = A 30 = 1 ), the summation terms on A21 = A20 1 +  λ s 2 s 2 − 1 
the left hand side of the Equation 21 can be written as  η  s=1 w2 A2  0 
 1 0.03333 01
. 
 3 eix eix  0.03333 • 0.03333 0.26667 • 0.26667 = 11 +  22562. • + 158.23 • − 1 
∑
  = 0+
3
 i =1 wi Ai  v 10 • 13
+
20 • 13

= 1667
.
2  10 • 12
10 • 12 

= 36667
. × 10−3
 1
2
 
∑
e
 3
eix eiy  0.03333 • 01 A31 = A30 1 +  λ s 3s 2 − 1 
∑
.
 3
 = 0+ + 0 = 33333
. × 10−4  η  s=1 w3 A3  0 
 i =1 wi Ai  v 10 • 13
 1 0.26667 0 
= 11 +  22562
. • + 158.23 • − 1 
 3 eiy eix  . • 0.03333  2 20 • 1 20 • 1 
2 2

∑
01
  = 0+
3
+ 0 = 33333
. × 10−4 = 2.004
 i =1 wi Ai  v 10 • 13

 3 eiy eiy  0.28284 • 0.28284 01. • 01 After obtaining the new set of member sizes as in the
∑
.
  =
3
+ +0 foregoing, the OC process is repeated in an iterative fashion
 i =1 wi Ai  v 10 2 • 1 3
10 • 13
until convergence of both member sizes and Lagrange
= 6.6568 × 10−3 multipliers occurs. The iteration history of the OC
Employing a step-size parameter η=2, the right hand side of optimization process for the truss design is tabulated in Table
Equation 21 is obtained as 2.

ENGINEERING JOURNAL / THIRD QUARTER / 1998 105