Professional Documents
Culture Documents
Reliability-Based Optimization of Transmission Line Towers: Pergamon 0045-7949 (94) 00468-4
Reliability-Based Optimization of Transmission Line Towers: Pergamon 0045-7949 (94) 00468-4
1995
Copyright < 1995 Elwier Science Ltd
Pergamon 0045-7949(94)00468-4 Prmted m Great Britam All rights reserved
0045.7949195 $9.50 + 0.00
Abstract-Transmission line towers form the major part of a power grid and contribute a large percentage
to the total cost of the transmission line network. The structural design of a transmission line tower is
most often governed by IS:802-1977, IS:802 (draft)-1989 (Indian standard code for design of
transmission line towers) and Indian electricity rules. There have been failures reported in transmission
line towers, particularly in Andhra Pradesh, a southern state of India, due to cyclonic storms. Against
this background of codal specifications not properly taking into account the uncertainty factors in loading
and material properties, the need for reliability-based design is highlighted in the present paper. Four
independent computer programs for component reliability, reliability analysis, optimization and auto-
mation of failure mode generation were developed. Not only were these programs independently
developed and validated, they were all linked together. This has enabled not only the more economical
design of typical towers but also ensured a particular level of chosen reliability for the towers. It has also
been demonstrated that the program can be used for checking the existing towers for assessing the
reliability levels available in them.
387
388 K. Natarajan and A. R. Santhakumar
/4=4(-b) (3)
---t--l
Call camp. exe.
ministic design of transmission line towers, one uses
a single value (nominal value) to quantify the en-
vironmental phenomena’ producing loads on the
Call rel. exe. system. Nominal values for wind speed, wind direc-
tion, ice thickness and temperature are taken by the
designer, who obtains the design load effect S in each
component. Then, one selects a component with
strength R equal to or larger than S.
In reality, all environmental phenomena are ran-
dom quantities which can only be described proba-
bilistically. The real component strength R is also
random. If the probability density functions (PDF) of
S and R are known, it is possible to calculate the
configuration with system reliability of the component. In general, the PDF of
reliabilitv index
S is not known, since S itself is a complex function
of the various environmental load variables such as:
(1) wind velocity;
(2) wind direction;
Fig. 3. Flow diagram for reliability analysis program.
(3) temperature.
In the present work, only wind speed is treated as
modes of the structure. For the class of structural a probabilistic variable, while the others are treated
system that are built of elements of elasto-plastic (or as deterministic. Statistical data on wind that are
ductile) material such as mild steel, the FMA has needed can be found in the studies of Natarajan [4]
proven to be effective, provided that the dominant and Alam et al. [5].
collapse modes of the system can be easily found. For It is reasonable to assume Gumbel distribution for
the class of structures having brittle components, annual maximum mean-hourly wind speed. Analysis
the number of potential failure modes is increased of wind data from a large number of meteorological
dramatically, and FMA becomes cumbersome and stations shows that, although the annual maximum
ineffective. SCA is suitable for highly indeterminate mean-hourly wind speed may vary from one location
structures with brittle components [3]. to another, the coefficient of variation of this quantity
is sensibly constant for locations of similar exposure.
2.4. Application of reliability theory to transmission
A typical value for the coefficient of variation of
line towers
annual maximum mean-hourly wind speeds is
2.4. I. General remarks. Reliability analysis is car- l2-25%.
ried out by systematically generating the failure Data for the strength of the transmission structure
modes by using the matrix method of structural components are usually scarce. Following are the
analysis. The failure criterion for a statically deter- contributions to the uncertainties in R.
0
Read
Start
1
Read
NDELE, NODE 1, NODE 2
I J
TEMP2 = NODE2
Write
I
LNUM (I)
MNOI(I) MN02(1)
1
(I;
stop
IJ =0
1
Call rel. exe
I
IJ = IJ+l
No
1
I=I+l
c
Choose the least WT.
1
J=O
1
- J=J+l
, I Determine worst
config. and delete
I
Call gload, truss, design
(1) material strength variability; (4) deterioration of strength with time in service.
(2) fabrication and construction variability (geo- A few tests were conducted by Peyrot and
metric); Dagher [6,7] and recommended normal (Gaussian)
(3) professional factor, to account for inability of distribution for the resistance variables with COV
simplified code procedures to predict a true strength, ranging from 0.05 to 0.2, depending upon the mem-
even if all uncertainties are eliminated; ber type. In the present work, normal distribution
392 K. Natarajan and A. R. Santhakumar
I
I Call camp. exe.
I
I Call rel. exe. I
*
I Callopt. exe I
I
(a) Normal condition
program.
IL
B F
average value of the component reliability index
value of the members involved in that mode.
Different failure modes are investigated and the
A Not to Scale corresponding system reliability index is calculated.
The system reliability index of the tower is evaluated
as the least value of the different failure modes
generated.
h 3.51 m r(
6 7
162
IO 10 11
14 I5
19
58
parameters are then randomly generated (initial com- ative procedure is adopted for the optimum tower
plex). Reflection steps were implemented to achieve design, it is obvious that the use of a computer is
improvement of the designs generated in the initial essential.
complex. The algorithm used for optimum tower design is
In method 1, only the member sizes are treated as similar to that given by Lipson and Agarwal[8],
variables, whereas the geometry was assumed to which presumes that an initial feasible configuration
be fixed. Method 2 also treats the geometry as is available for the structure. The tower is divided
a variable and gets the most preferred geometry. into a number of groups and the externally applied
The geometry developed by the computer results in loads are obtained. For the given configuration, the
the minimum weight of tower for any practically upper limits and the lower limits on the design
acceptable configuration. For solution, since an iter- variables, namely the joint coordinates, are fixed. The
Reliability-based optimization of transmission line towers 395
(k - 1) new configurations are generated randomly number of design variables and r,j is the random
as: number for the ith coordinate of the jth configur-
ation, the random numbers having a distribution over
x,, = !, + r,i(Uj - 1,) (4) the interval 0 to 1, and u is the upper limit and I is
the lower limit of the ith independent variable.
i=1,2...n Thus, the complex containing k number of feasible
solutions is generated and all these configurations will
j-1,2,...k, satisfy the explicit constraints, namely, the upper and
lower bounds on the design variables. Next, for all
where k is the total number of configurations in the these k configuratims. analysis atld fully stressed
complex, usually larger than (n + i), where n is the designs are carried out and their corresponding total
396 K. Natarajan and A. R. Santhakumar
weights determined. Since the fully stressed design (1) Guess a value for fl.
concept is an economical and practical design, it is (2) Initialize the design vectors at the mean values,
used for steel area optimization. Every area optimiz- m,.
ation problem is associated with more than one
analysis and design. For the analysis of the truss, the X,* =m,, i = I, 2,. n.
stiffness method has been used. Therefore, all the
generated configurations also satisfy the implicit con- Here, n is the number of design variables.
straints, namely the allowable stress constraints. (3) Start the iteration, i.e.
Reflection steps were implemented to achieve im-
provement of the designs generated in the initial
ITERATION = 1.
complex.
From the value of the objective function (total
weight of the structure) of k configurations, the (4) Compute partial derivatives, g) = Sg/SX,, at
vector which yields the maximum weight is searched design points:
and discarded, and the centroid c of each joint of the
k - 1 configurations is determined from
a, =
si6
i=l,2,3...n,
r i (sg/sx,)2(al’)2 lj2’
L/=1 1
where x,, and x,,, are the ith coordinates of the
centroid c and the discarded configuration W. 1.52 m
Then a new configuration is generated by reflecting
the worst configuration through the centroid, x,, Reference
That is, 90X90X8
15 x 15 x 8
15X75X6
x,,,= Xl,. + a (x,, - x,,,1 (6) 60X60X6
50 x 50 x 6
50 x 50 x 5
i = I, 2,. . n, 45X45X5
(6) Modify the design vector using the sensitivity the structure. Each time the input data has to be
coefficients and the jl value: modified by removing the failed member. Further
analysis thus requires renumbering the members.
x: = (m:‘- aiBa:‘).
383
j_151
(7) Find the convergence, if the difference is within
676:i” ,i”
the tolerance limit (specified by the user), then the
next step is executed, otherwise the steps numbers 4-7
c676
are repeated:
XD = (xl’)“+ ‘) - Xf)/(Xf).
1.14 m
I-;Ji
c 110x 110x10
D I 110x110x 8 I
E
s
::
to Scale
-
4.88 m
Also, the residual strength of the member which determined. The following steps are involved in the
has failed is applied as equivalent external forces at automation procedure:
the nodes connecting the member in X, Y and Z (I) The following input variables are read by the
directions. program:
Due to the above tedious process in the failure
mode approach, there is a tendency to generate fewer (a) NEL total number of el-
modes of failure, which will apparently result in ements
over-estimation of system reliability index of the (b) NLN number of loaded
tower. This is one of the main drawbacks of the nodes
failure mode approach. This drawback is overcome (c) LNUM (I) member numbers
in the present work by developing an automation (d) MN01 (I), MN02(1) node numbers
program. (e) NDELE element number to be
The algorithm adopted for this program is shown deleted
in Fig. 4. The program asks the user to specify the (f) NODEI, NODE2 node numbers con-
member number to fail. Next. the program automati- necting the deleted
cally calculates the forces and also generates the input member
data necessary for the space truss analysis program. (8) R resistance of the
Due to this automation, it was possible to generate deleted member
nearly 100 modes of failure for each loading case (h) Xl, Yl. ZI, X2, ~2, 22 coordinates of the
of the tower. Based on such a large number of nodes connecting the
failure modes, the system reliability index value was deleted member.
Reliability-based optimization of transmission line towers 399
(2) The nodes connecting the deleted member, WRITE LNUM (I - I), MN01 (I), MN02 (I).
namely NODE1 and NODE2, are stored in arrays
TEMPI and TEMP2, respectively.
(IO) Forces in the X, Y and Z directions are
(3) The member number is designated by the
calculated for nodes, NODEI and NODEZ.
variable 1 and initialized to zero.
(11) The nodes, NODE1 and NODE2, and Forces
(4) In this step, I = I + I is done.
on the nodes are written to an output file.
(5) It is checked whether I > fixed NDELE.
(6) If in step number 7, the answer obtained is NO,
4.4. Tower optimization program
then the program proceeds to step number 9, other-
wise step number 11 is executed. The flow diagram for the program is shown in
(7) It is checked whether LNUM (I) = NDELE. If Fig. 5, and the procedure is already explained in
the answer obtained is YES, then step numbers 6 and Section 3.2. An explanation of the notations used in
7 are repeated. If the answer obtained is NO, then the flow diagram is given below:
step number IO is executed.
(8) The member numbers and the node numbers GENERN sub-program to generate tower
connecting them are written: configurations
REL.EXE external program to compute
WRITE LNUM (I), MN01 (I), MN02 (I). system reliability index of the
tower
(9) The member numbers (reduced by one) and the GLOAD, sub-programs to generate Ioad-
nodes connecting them are written: ing data for the tower analysis
-
T Reference
110x llOx12mm
110x IlOx
100x 100x 8
100x 100x 8
100x 100x 6
75 x 75x 6
70x 70x 6
65 x 65x 6
Not to Scale
5.88 m
TRUSS, DESIGN sub-programs to perform program developed, and were compared with other
tower analysis and design, methods of design. A target reliability index range of
respectively 2.0-2.5 was given as the input for the reliability
J, NT, IJ counter for area optimization. analysis, and the optimization of tower weight subject
number of tower configur- to this reliability constraint was achieved.
ations and number of iter-
ations, respectively. 5.2. ReliuhilitJ-bused optinzizatiorl of’ 1 IO kV tower
4.5. Ocerall network of’ the developed computer 5.2. I Towr properties
programs (i) General information:
Figure 6 shows the flow diagram of the overall
tower voltage IlOkV
function of the computer program. Following are the
number of circuits double circuit
steps involved in the reliability-based optimum design
type of tower tangent
procedure: 2
angle of deviation
Step l-details of the structures. properties of
normal span 320 m
angle and coordinates of initial configuration, which
weightjnormal span ratio I .25.
are for the given tower specification. are read as input
by the reliability analysis program.
(ii) Wind: extreme annual wind speed with Gum-
Step 2-the program TRUSS.EXE is called to
bel distribution. mean = 45 m s ’ and coefficient of
perform the three-dimensional analysis of the tower,
variation (COV) = 0.2.
to determine the forces in the members.
(iii) Component strength: allowable strengths
Step 3-the program COMP.EXE is called to
Gaussian with COV = 0.15.
determine the component reliability index of all the
members of the tower. 5.2.2. Design promlure. The 100 kV double circuit
Step 4-reliability analysis is carried out on the design process was started by initially specifying the
initial configuration to check whether the system range of target reliability index. Figure 7 shows the
reliability index range lies between the chosen range, loading tree for the tower. The tower was analysed
i.e. 2.0 and 3.0. If the system reliability index for the following conditions:
lies within the required range, then optimization
procedure is started. Otherwise the configuration is (i) normal condition [Fig. 7(a)];
revised by calling the routine GENERN, until (ii) broken wire condition I [Fig. 7(b)];
this condition is satisfied. The program called for in (iii) broken wire condition II [Fig. 7(c)].
the reliability analysis is REL.EXE.
Step 5-the optimization procedure requires the The system reliability index range specified as a
user to specify target reliability index range and the constraint during the optimization process was
number of iterations, The program called is 2.0~~2.5. The initial configuration obtained based on
OPT.EXE.
Step 6--final configuration of the tower with Table 4. Coefficient of variation (load and resistance) vs
system reliability index is printed. system reliability index
System Coelfcient of System reliability
no. carnation index
5. ILLWiTRATIVE EXAMPLES
I 0.20 2.80
5. I General
2 0.28 1.70
The tangent towers, I IO and 220 kV, were designed 3 0.56 0.80
4 0.64 0 65
based on reliability concepts using the computer
Fig. 14. 220 kV tower-member nun&l -ing for three-dimensional analysis.
,,
402 K. Natarajan and A. R. Santhakumar
wind speed = 0.2 and COV of component is so high that further increase in weight only reduces
strength = 0.15. The weight of the tower increased, the reliability index.
when the tower was designed for a larger system (3) A study was done to investigate the sensitivity
reliability index. Figure 15 shows the weight vs of the reliability index with respect to the uncertain-
system reliability index relationship graph. It can be ties in the loading and resistant variables on a I 10 kV
seen from the graph that the variation of system tower designed earlier, based on reliability concepts.
reliability with weight of tower is approximately COV of component strength was fixed at 0.15 and
linear in the useful range. Also, it could be inferred COV of wind speed was varied from 0.15 to 0.72. It
from the graph that the reliability index of the tower can be inferred from the study that a change in the
increases with increase in weight up to a value of COV value of wind speed by 1% results in 1.5%
3.325 tonnes. Beyond this transition point, increase in reduction in system reliability index of the tower. This
weight only reduces the system reliability index. This indicates the significance of uncertainty factors on the
is due to enormous increases in self-weight, which design results.
induces large compressive forces in the leg members,
leading to premature failure by unfavourable modes. Acknowledgements-The support of Tamil Nadu Electricity
5.4.2. Coqficient qf variation qfload and resistance Board for the optimization phase of the vvork is gratefully
variables 1’s system reliability index. This study was acknowledged by the authors.
done to investigate the sensitivity of the system
reliability index with respect to the uncertainties in
the loading and resistant variables. The tower chosen
REFERENCES
for this purpose was the 1 IO kV tower, designed
earlier based on reliability concepts. The component I. C. A. Cornell, A probability based structural code. AC1
strength COV was fixed at 0.15 and COV for wind J. 66, 974-985 (1969).
speed was varied from 0.15 to 0.72. Figure 16 shows 2. A. M. Hasofer and N. C. Lind, An exact and invariant
first-order reliability format. J. Eqng Me& Div. ASCE
the variation of system reliability index with respect lOO(EMl), II Ill21 (1974).
to COV of wind speed. Table 4 shows some of the 3. S. T. Quek, Structural system reliability by the
typical values obtained from the graph (shown as method of stable configuration. Ph.D. Thesis, Graduate
dotted lines in Fig. 16). College of the University of Illinois, Urbana-Cham-
paign (I 987).
It can be inferred from the graph in Fig. 16 and
4. K. Natarajan. Wind load analysis-probabilistic ap-
from Table 4 that a change in COV value of wind proach. National Seminar on Analysis and Des&r .fi)r
speed by 1% results in nearly 1.5% reduction in Wind Lauds, Indian Concrete Institute (ICI). Madras.
system reliability index of the tower. This shows the pp. 35-46 (1991).
significance of uncertainty factors on the system 5. M. J. Alam, K. Natarajan and A. R. Santhakumar.
Design wind speeds for transmission line structures. In/.
reliability index value of the tower. J. Struct. 12, l-19 (1992).
6. A. H. Peyrot and H. J. Dagher, Strength of trans-
mission line components. A paper presented at
6. CONCLUSIONS BASED ON EXAMPLES STUDIED
the IEEE-PES Summcv Mwtitzg, Los Angeles, CA
(1983).
(1) The weight of the optimal tower accounting for 7. A. H. Peyrot and H. J. Dagher, Reliability based design
reliability as a constraint for both 110 and 220 kV of transmission lines. J. Strut. Enpzg ASCE llO( I I ),
tangent towers is only 3-4% heavier than the tower 2758-2777 (1984).
designed using the conventional method. 8. S. L. Lipson and K. M. Agarwal, Weight optimization
of plane trusses. J. S/rue/. Dir. ASCE lOO(ST5).
(2) From the relationship studied between weight
865879 (1974).
and reliability index for fixed COV values of 0.2 and 9. IS:802 (l977)-Indian standard code of practice for
0.15 for loading and resistance variables, respectively, use of structural steel in overhead transmission line
on a 1IO kV tangent tower, it could be inferred that towers.
it is approximately linear in the useful range. It could IO. IS:802 (DRAFT) CED7 (4724) (I 989)-Indian stan-
dard code of practice for use of structural steel in
also be understood that reliability index of the tower
overhead transmission line towers,
increases with increase in weight, up to a value of II. S. S. Murthy and A. R. Santhakumar, Trtrtr.vnri.v.\iotr
3.325 tonnes. beyond that the self weight of the tower Line Structures. McGraw-Hill, Singapore (1990).