Sunteți pe pagina 1din 4

GUYED TOWER OPTIMIZATION

LANSFORD C. BELL
Department
of CivilEngineering,
AuburnUniversity,Auburn,AL 36830,U.S.A.

and

DAN&XM. BROWN
Department
of CivilEngineeting,
Vanderbilt
University,Nashville,TN 37235,U.S.A.

(Received10August1975)

Abstract-Auniqueprocedurefor the optimum design of a guyed tower structure is presented. The method depends
on a carefully established design hierarchy which is based on the observed independence of the design variables in
certain portions of the design space. Powell Search and Branch-and-Bound sub-optimization routines are used to
generate locally optimum designs in terms of cable area, initial cabie tension, mast area and anchor to tie locations. A
designexample is presented and computer program capabilities are discussed.

1. AXON The procedure obviously requires a detailed knowledge


The optimum design of a unique, complex or extremely of the various design variable interactions. Feedback
large indeterminate structure is not easily obtained using loops and redesign iterations must be specified when
formal mathematical programming techniques. Penalty variable changes affect the results of a previous optimiza-
functions [l], gradient projection [2] and techniques based tion step. Experience with the interactive problem-
on the Simplex method [3] have been applied to structural oriented design package TOWER[4] has convinced the
problems of various types. These methods are severely authors that a variable independence does in fact exist in
limited, however, by their inability to realistically handle certain portions of the design space. It was further noted
discrete design variables. Moreover, idealized or overly that a judicious ordering of the design process hierarchy
simplified mathematical models are often required when will maximize variable independence and therefore
dealing with full scale structures. minimize the number of feedback loops or redesigns
The purpose of this paper is to demonstrate a heuristic required. Once the design hierarchy has been established
approach to the optimization of a specific type of com- and broken into stages, simplified search type algorithms
plex, nonlinear able-suited structure, namely the can be applied to each indi~dual stage.
guyed tower. Mathematics pro~am~g techniques are This procedure does not necessarily produce an overall
pickily diicuh to apply to the guyed tower structure global optimum solution although the resulting designs
for the following reasons: can seldom be improved by an experienced designer using
1. Mechanical properties (area, weight, strength, etc.) an interactive program.
of neither the cable nor the tower mast can be approxi- The investigation is co&red to a class of guyed towers
mated as continuous variables; lOO-5OOft (30.5 m- 152.4m) in elevation that support
2. Simplified reanalysis techniques are complicated by heavy microwave antermas. Development of a detailed
the non-linear behavior of both cables and mast; optimization algorithm is justified by the large number of
3. Constraints in terms of both stress and displacement such structures in use for the transmission of voice
are involved thus increasing the likelihood of non-fully- commumcations.
stressed optimal designs.
The proposed procedure depends on two proper- 2. PROBLEMFORMULATION
ties which have generally been observed by Tower design is defined as the process by which the
designers: design variables shown in Fig. 1 are selected. Note the
1. The decision making process iogically evolves into a process en~mpasses more than the selection of member
hierarchial progression which proceeds generally from sizes in that decisions must be made with respect to initial
topological properties to geometric properties and thence cable tensions and the physical placement of tie and
to mechanical properties, with increasing attention to anchor locations. The design objective is thus to minimize
detail as the design proceeds; the cost of the overall structure:
2. During each stage of the process, modification of the
variables associated with that stage has relatively little M~z=ki~(nCL,)i+ki~(LA),
effect on the optimum values of variables associated with (1)
earlier stages (although some iteration is usually re-
quired). subject to the stress and rotation constraints:
It seems logical, therefore, that optimization algorithms
could be applied to individual stages of the design pro-
cess. Optimum cable areas, for example could be deter- s/s,iow
-< 1.Oin each cable of all cable sets i (2)
mined before the selection of opium tower mast sec- S/S&0$?L 1.0 at all critical points in each tower span i
tions. Initial cable tensions could likewise be adjusted to (3)
optimum values independent of the design for mechanical
properties. ete puorv
I 1.Oat each antenna location (4)
447
LANSFORD C. ~ELL and DANIEL M. BROWN

compute
quadmtic
bahaviof
mc&Jls
forcabbstramand tcmr rot&n

OQimia cabb ten&m to


min mu (Cable stresslallow, tower b/allow)

HZ HI (Pcwell starch)
Fig. 1. Design variables. C Cable area; T Initial cabIe tension; H
AnchorIocation;V Tie iocatioa; A Mast area.

where k, and kl are relative costs of cable and tower steel;


L and I.,, are lengths of mast and cable; S and 0 are stress
and rotation; r and s are the number of cable sets and Fwelbbdmlgnwithmepea N
tower mast sections respectively and n is the number of f0wbbmeeendtoworrotstion?
cables in the set. The rotation constraint (tilt) is critical in
that the antenna beams must maintain a prescribed align-
ment under load.
For simplicity, the objective function is stated as a
quasi weight function in that the total amount of tower
steel is the quantity to be minimized. If sufficient data are
available, additional terms may be added to the objective
function without affectiug the optimization algorithm. For
example, site cost could be specified as a function of
anchor location and cable connection costs could be
specified as a function of cable tension.
Such for olimr fmibb combindons
3. DiLSKN OFTIMIZAT~ON ofcabbar@nthatreduwtotaltoworcort
Design experience and a careful study of variable
~te~ctions has produced a design stage hierarchy which (Branch and Baund)
is shown somewhat simpli6ed in Fig. 2. An initial infeasi-
ble design is selected and initial cable tensions optimized
using the relatively efficient Direct Search Method of
Powell [S].
This method minimizes a given function one or more
times in each coordinate direction and then in an as-
sociated pattern direction where a pattern direction vector
is defined by:

sq=xq- xq-, (5)


where n is the number of steps preceding the move. One
coordinate direction is then discarded in favor of the
pattern direction for inclusion in the next m minimiza-
0 End

Fig. 2. Heuristic design Rowchart.


tions. After the next cycle of minimizations, a new pattern
direction again replaces one of the coordinate directions.
A combination of initial cable tensions that minimizes infeasible member that results in a minimum cost in-
stress and rotation values at critical locations is computed crease;
and cabfe areas are increased until a feasible design with 2. Use a branch and bound algorithm to explore the
respect to cable stress and tower rotation is located. Mast portion of the feasible region between the given solution
sections are then selected ~de~ndent of the cable optim- and the problem cons~aints;
ization design stages. Other feasible combinations of cable 3. Expiore solutions adjacent to the local minimum
area may possibly exist in which case a branch and bound generated above by “bouncing away” from the con-
procedure is used to explore alternate integer combina- straints.
tions of cable area. The procedure employed is analogous Adjustments to cable tie and anchor locations are
to the three phase method of Cella and Logcher [6] which, performed in the &ml design stage although cable tie
for discrete structural components, proceeds as fohows: Iocations are somewhat predetermined by antenna loca-
1. Input an initial infeasible solution and modify the tion and bracing details. Optimum values previously de-
Guyedtoweroptimization 449

termined are relatively unaffected by minor adjustments


in this stage.
It is very important that the number of analysis itera-
tions be held to a minimum. Quadratic behavior models [ 11
are therefore used to predict changes in cable stress and
tower rotation caused by variations in initial cable ten-
sion. For a tower with r cable sets, a total of 2n + 1
analysis iterations are required to compute the desired
relationships. Using a step length of t, the quadratic
equation is: 300’ 150’

F = a + bt + ct2 (6) Cableset Cable Tower section Leg angle


(in order of size (in order of size
where: increasing (in.) increasing (in.)
elevation) elevation)
a =f@)
1 518 1 4~4x7116
b = [4!(t) - 3f(O)-f(2?)]/2t 2 5/8 2 4x4~318
3 518 3 4x4~318
\C =U(21)+f(0)=2f(t)]/2t2 4 518 4 4x4~5116
5 518 5 4x4~5116
Each cable tension and tower rotation is approximated
under load by expressions of the form of eqn (6). Exact Fig. 3. Initial infeasible design.
values of these parameters, denoted by f, are used to
compute the constants a, b, and c.

4.COMPUTERPROGRAM AND DESIGNEXAMPLE


The computer program AGTD (Automated Guyed
Tower Design) was written to generate locally optimum
designs using the hierarchical approach described above.
Load magnitudes and initial values for all design variables
(Fig. 1) are entered as input. The initial design is normally
infeasible with respect to stress and/or antenna rotation.
Feasible design variables which minimize tower cost are
printed as program output. The capabilities and limita-
tions of AGTD may be summarized as follows:
1. The structure is designed elastically to withstand a Cable set Cable Initial Tower section Leg angle
basic wind load with appropriate gust factor. (in order of size tension (in order of size
2. Tower cross section is either square or triangular increasing (in.) (% breaking increasing (in.)
with 4 or 3 vertical leg angles respectively. Leg angle sizes elevation) strength elevation)
remain constant between cable tie locations.
: 5/8
3/4 0.46
0.50 21 4x4x l/2
4x4x7/16
3. Cable sets consist of 8 or 4 cables if cross section is
square; 6 or 3 cables if cross section is triangular. 3 314 0.50 3 4x4~7116
4. A minimum initial cable tension is maintained to 4 518 0.50 4 4x4~5116
prevent dynamic instabilities. 5 5/8 0.48 5 4x4x5/16
5. The horizontal distance separating the vertical leg
angles is given and remains constant for all elevations. Fig. 4. Optimum design.
6. Cable nonlinearities and the effect of increased
flexure due to axial load have been considered[7,8]. above required less than 2.Omin Xerox Sigma 7 CPU
7. The effects of wind load on the cables have been time. An extremely poor choice of initial design variables
considered. will produce convergence in less than 5.0 min CPU time.
8. Cable areas and mast leg angle size are selected from
an auxiliary data file of available sections. 5. CONCLUSION
9. Allowable stresses and factors of safety in the tower Formal nonlinear programming methods are not easily
mast are based on the AISC Specification. applied to complex cable-supported structures with in-
Applying concentrated and uniformly-distributed loads teger design variables. An intuitive approach to guyed
corresponding to a 90mph wind to the initial design tower optimization was therefore developed whereby
shown in Fig. 3 results in the optimum design shown in search routines are applied to the individual stages of the
Fig. 4. The design was based on a relative objective overall design process.
function that assumes: A computer program was written to optimize the major
design variables shown in Fig. 1 with respect to stress and
Total cost = 3.O(cost of cable/lb) displacement constraints. Locally optimum designs which
cannot be readily improved by alternate methods have
+ l.O(cost of mast steel/lb) (7) been generated at reasonable expense.

Allthough rarely excessive, computer execution time REFFmNclB


depends on the user’s ability to specify reasonable values 1. D. Kavlie and J. Moe, Automated design of statically indeter-
for the initial design variables. The example presented minate frames. ASCE 5th Conf Elecrron. Camp. (1970).
450 LANSFORD
C. BELLand DANIELM. BROH~N

2. D. M. Brown and A. H. S. Ang, Structural optimization by function of several variables without calculating derivatives.
nonlinear programming.J. Strucf. Div. ASCE 92 (ST6), 319-340 Computer J. 7(3), 155-162(1964).
(1966). 6. A. Cella and R. Logcher, Automated optimum design
3. F. Moses, Optimum structural design using linear program- from discrete components. ASCE 5th Conf. Electron. Camp.
ming. I. Struct. Diu. ASCE 90 (ST6), 89-104 (1964). (1970).
4. K. L. Reichelt, D. M. Brown and J. W. Melin, TOWER, design 7. D. L. Dean, Static and dynamic analysis of guy cables. J.
system for guyed towers. J. Struct. Diu. ASCE 97 (STl), Stnrct. Diu. ASCE 87 (STl) (l%l).
237-251 (1971). 8. S. P. Timoshenko and J. M. Gere, Theory of Elastic Stability.
5. M.J. D. Powell, An efficient method for findingthe minimumof a McGraw-Hill, New York (1961).

S-ar putea să vă placă și