Professional Documents
Culture Documents
Dynamic Relaxation Analysis of The Non-Linear Static Response of Pretensioned Cable Roofs
Dynamic Relaxation Analysis of The Non-Linear Static Response of Pretensioned Cable Roofs
Dynamic Relaxation Analysis of The Non-Linear Static Response of Pretensioned Cable Roofs
m
Printed in GreatBritain. 0 1984 PergamooPress Ltd.
and
K. R. RUSHTON
Department of Civil Engineering, The University of Birmingham, England
Abstract-A summary of various existing methods of cable roof analysis is followed by a detailed
description of a dynamic relaxation technique that can be applied for the solution of such structures.
Theoretical analysis of static response of pretensioned cable networks is presented using a non-linear
energy approach. Solutions are obtained by dynamic relaxation, a numerical finite difference technique.
Detailed results are presented for a pretensioned cable truss. Comparisons are made with alternative
numerical and experimental results. The extension of dynamic relaxation to more complex cable-cladding
problems is outlined.
NOTATION
high-power computers the analysis of these highly
A stiffness matrix redundant structures has become feasible.
U column vector of displacement Cable structures exhibit highly geometrically non-
AU change of displacements
linear behaviour which is often compared to the
P external force vector
i suffix indicating direction (x, y or z)
behaviour of structural mechanisms. The non-
suffix indicating node linearities arise because, for equilibrium to be
RJ residual or out-of-balance force vector satisfied, the deformations of the structure must cause
sufficient strain in the cables to produce the required
or Cartesian co-ordinates changes of tensions. Therefore any theoretical anal-
x,. xz, x3 t ysis must take into account the deformed geometry of
m, mass in ith direction the cables under the external static load.
4 total potential energy of the system
V potential energy of external loads
u internal strain energy CURRENT METHODS OF THEORETICAL ANALYSIS
internal strain energy in member ‘m’
[$
A fundamental assumption made in the analysis of
transposed column vector of external loads
pretensioned cable roofs is that the cables are the
VI column vector of displacements
p: external force in ith direction at node y’ main load-carrying elements, the axial and shear
d,“, 4 displacements of nodes ‘n’ and ‘j’ in the ith stiffness effects of the cladding being ignored. Litera-
direction ture relating to the analysis of such “bare” cable
F, elastic force in member ‘m’ networks is now considerable. The analytical meth-
E Young’s modulus ods discussed later represent varoius ways of ap-
L unstrained length plying the deformation theory of pretensioned cable
L, unstrained length of member ‘m’ networks, which deal with the deflections and cable
AL, increase of length due to elastic strains forces when the structure is subjected to super-
A cross-section area
number of cable links meeting at node ‘j’
imposed loading. Generally, there are three ap-
5 proaches adopted:
[I” tension coefficient (force per unit length)
damping coefficient in the ith direction (1) Variously formulated linear analysis or stiffness
matrix analysis corrected for non-linearity
d,,; velocity and acceleration in the ith direction
n; number of iterations for one cycle of oscillations effects[l-61.
/ frequency of oscillations
(2) Analagous membrane solutions[7-91.
At time increment (3) Non-linear analysis based on an energy app-
L J suffices indicating step increments in space roach[lO, 111.
The principles involved in the linear analysis can be
INTRODUCTION explained following Siev’s approach. In the analysis
Cables, because of their flexibility, offer interesting a set of equilibrium equations is derived, based
possibilities for shaping attractive three-dimensional initially on the assumption of linear behaviour, which
roofing forms. In recent years there have been a represented in a matrix form, gives:
number of cable roofs constructed to cover both large
and medium span areas. With the development of AU= -P (1)
989
990 W. J. LEWISet al.
or. after inversion: modates the problem of slack tensile cables and
buckled compression members. The computational
U,= -A-‘P, (2) algorithm is attractively simple, and the iterative
technique is numerically stable except under one,
where A-’ is the stiffness matrix representing the easily predeterminable, condition of the time in-
influence coefficients for the displacements Ui at any crement.
joint as a function of the load components P, in the The dynamic relaxation technique was devised by
ith direction. J. R. H. Otter in 1964 in the course of analytical
The subsequent correction for non-linearity is im- investigation of prestressed concrete nuclear vessels.
plemented in the iterative process using the residuals Since then the technique has been widely applied to
R which represent an out-of-balance load. Whenever a variety of structural problems, reflected in
the effect of non-linearity is small, the matrix A-’ Refs. [12-151. The first application of dynamic relax-
may be assumed valid, and the change of displace- ation to cable networks was by Day and Bunce[l 11.
ments AU, is:
Differentiation of eqn (5) with respect to the dis- SOLUTION OF EQUATIONS BY DYNAMIC RELAXATION
placements d/ at all nodes gives: If the cable system is not in equilibrium the vector
iT+/ad,will not be zero but equal to some residual
$.=i,$P/ (6)
force Ri.
R=_Z. (10)
where PJ is the vector of externally applied loads a4
acting at node j in the ith direction.
The static configuration of an elastic system can be
It can be shown that for a member m between
regarded as a consequence of a previous dynamic
nodes n, j, extracted from the cable system, as shown
state. Making use of the d’Alembert equation of
in Fig. 1, the following relation is true:
dynamic equilibrium:
R,“‘“‘=~(jl+‘_~~)+~(li:+‘+li,.) (12)
or
“cl step
Equations (9), (13), (14) can be solved in a cyclic The boundary conditions, such as zero displacements
order as outlined below: at the nodes fixed to the rigid boundary, are intro-
duced between steps (4) and (5). Based on the numer-
(1) Setting the deflections d,, velocities di. cable ical procedure outlined above is the flow chart of
forces F,,, and residuals R, at their initial values at dynamic relaxation analysis of pretensioned cable
time t = 0. networks presented in Fig. 3.
(2) Calculating the residuals at time ;At (eqn 9).
(3) Calculating the new velocities ci, at time 1At DYNAMIC RELAXATION PARAMETERS
(eqn 13). There are three basic factors that enter the equa-
(4) Substituting the velocities into eqn (14) to tions of motions in the dynamic relaxation formu-
obtain the displacements at time IfAt. lation: namely, the time increment At, damping
(5) Obtaining new values of member forces F, at coefficients c,, and masses m,.
time l;At. The value of time increment At is critical to the
(6) Returning to step (2) unless the residuals R, are stability and convergence of the iterative technique.
close enough to zero. To reduce the number of iterations. At must be as
Qstart
IlO
large as possible. However, too large a time in- span of the structure, as shown. The nodes lie on two
crement will cause the divergence of the solution. parabolic curves. The remainder of the data is as
Numerical instability is readily detected as values of fol1ows:
internal forces and deflections start to oscillate be-
tween quite unrealistic values. An upper bound of the Span =9144m(300ft)
time increment At quoted by Day and Bunce[ll] as: Depth = 9.14 m (30 ft)
Cross-section area
of top cables = 1045.2 mm* (1.62 sq in)
Cross-section area
where A denotes the cross-sectional area of cables; of bottom cables = 2090.3 mm* (3.24 sq in)
and E is the Young’s modulus of the cables; gives a Cross-section area
good indication of the value of At that should be used of struts = 1290.3 mm*(2 sq in)
in the analysis for arbitrary chosen masses. Young’s modulus
The damping coefficients ci are important in regu- of the cables = 165.55 kN mm-*
lating the speed of convergence but they do not affect (24 x lo6 lbf per sq in)
the stability of the iterative solution. The critical Young’s modulus
damping coefficients that allow the structure to reach of the struts = 209.92 kN mm-*
equilibrium most rapidly can be found from a short (30 x lo6 lbf per sq in)
undamped run on the computer, as illustrated in Figs. Tension coefficient = 35.016 kN m-t
4 and 5. The critical damping is given as: (200 lbf per in)
Load P = 44.8 kN (1000 lbf)
c, = 4nmf (15)
The values of the dynamic relaxation parameters
where f= l/NAt is the fundamental frequency of used in the analysis are as follows:
oscillations; and N is the number of iterations for one
complete cycle of oscillations. Time increment At = 0.00095 s
It is recommended that only one value of damping Masses m, = 74 kg
coefficients ci is used in the analysis, and thus should Damping coefficients c, = 1956 kg/s (based on
be based on the dominant deflections of the structure 500 iterations for
and the associated fundamental frequency of oscil- 1 cycle)
lations. The process of determining the critical damp-
ing coefficient can be facilitated by the use of auto- The program took 750 iterations to reduce the re-
matic plotting routines, such as in the Gino F library siduals to a minimum.
of computer programs. The mode1 structure has been previously analysed
The value of masses mj have no bearing on the by Buchholdt [ lo] who published some theoretical
stability of the numerical solution, and can be chosen results obtained by utilising the method of steepest
arbitrarily to calculate the damping coefficients cP descent. Some results from the numerical analysis by
the dynamic relaxation technique are compared with
those of the method of steepest descent, in Tables 1
NUMERICAL EXAMPLES and 2. The values of y deflections are zero. Figure 7
A wide range of pretensioned cable networks have shows the deflected shape of the girder due to the
been studied using the dynamic relaxation technique. applied loading. The values of displacements are
The results are compared with solutions from alterna- scaled up three times in comparison with the scale of
tive methods, and with experimental data. the model.
Fundamental node
0 1 2 cycles
1 2 cycles
I
( bottom cable
r
kLLi!c]
Displacement in Cm) of top cable
“Et’ diffe:..c:]
x-displacements z-displacements
2 0.4946 0.4980 0.6
3 0.7804 0.7762 0.5
4 0.8338 0.8326 0.1
5 0.6700 0.6668 0.4
6 0.2791 0.2795 0.1
I
7 -0.1270 -0.1294 1.1
a -0.3542 -0.3588 1.2
9 -0.4144 -0.4126 0.4
Average
Method of
CCSUpreSSiVe
numerical
i analysis
Link Force Link ?orce Link Fbrce Link Force Link FCO-ce
w LEWISI 21-24!-48.89
Dynamic l-2 275.6 7-0 266.0 1-12 776.0 14-15 761.7 25 -32.37
relaxation kN kN !a kN 26-29 -15.74
Dynamic relaxation analysis 995
Fig. 7. Displaced shape of the cable girder (the deflections scaled up 3 times in comparison with the scale
of the structure).
The structure shown in Fig. 8 is analysed numer- that good agreement between the theoretical and
ically by dynamic relaxation, and the results are experimental results is achieved.
verified experimentally. The model consists of 3 x 4
cables attached to a rigid boundary, and forming a Multiple cable truss
part of a single hyperbolic paraboloid. The data used A three-dimensional counterstressed dual cable
in the analysis are as follows: structure is commonly used to support the cladding
in circular roofs. The structure, shown in Fig. 9, is
Average pretension in all solved by dynamic relaxation, and the results are
members = 0.20 kN compared with alternative solution by stiffness
Young’s modulus of cables = 128.3 kN mmm2 method proposed by Jonatowski[6].
Cross-section area of cables = 0.785 mm2
Load P = 0.0157 kN
Corner rise H = 0.45 m
Tabie 3. Comparison of theoretical and experimental test
The dynamic relaxation parameters used in the results for a ‘hypar’ net
analysis are as follows: r
Deflections z hm)
Load
Node
(kN) Numerical (DR) Experimental
Time increment dt =80x 10-6s
Masses m, = 1.8 x 10-3kg
5 0.0157 19.3 19.5
Damping coefficients ci =2.0kgs-’
6 0.0157 25.3 25.3
Table 3. The very small deflections which took place 12 0.0157 29.4 28.8
in X, y directions are not quoted. 15 0.0157 26.4 25.2
The changes in tension of the cables due to the
applied unsymmetrical load are given in Table 4. The
load is shared adequately between the two sets of
cables with no member going slack. The largest
increase in tension is observed in cables that are most
loaded, i.e. 9-13, 14-18, 2-25 (Fig. 8). It can be seen
The following data are used: The dynamic relaxation parameters used in the
analysis are as follows:
Span = 91.44 m (300 ft)
Upper sagf, = 6.706 m (22 ft) Time increment At = 1.3 x 10..3s
Lower sagf, = 3.048 m (10 ft) Masses m, = 1.54 kg
Length of cables Damping coefficients c, = 445.9 kg ss’
2-1, 49, = 2.134m (7 ft)
Cross-section area The program completed the solution within 630
of cables = 1290.3 mm’ (2 sq in) iterations. The results from the dynamic relaxation
Cross-section area and stiffness analyses, described in terms of x, y, z
of hangers = 322.6 mm* (0.5 sq in) components, of displacements, are given in Table 5.
Horizontal components The results prove to be the same to within 0.2mm
of prestress I& = H, = 449.98 kN (100 ki ps) accuracy. The changes of cable tensions are not
Concentrated load P = 222.49 kN (30 ki ps) quoted but both numerical methods confirm that the
Table 5. Comparison of published and test results for a multiple cable truss
-
3 f 0.1176 0.0565
8 0.1132 -0.0153
2 -0.1365 0.0399
7 -0.1371 -0.0181