Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Nonlinear Analysis of Space Trusses Using Modified Normal

Flow Algorithm
H. Saffari1; M. J. Fadaee, M.ASCE2; and R. Tabatabaei3
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

Abstract: In order to evaluate the structures’ behavior accurately, a suitable mathematical algorithm and a set of appropriate assumptions
concerning the structures’ behavior must be adopted. The more accurate the algorithm and the assumptions adopted the more real behavior
of the structures is determined. In the case of nonlinear analysis, the evaluation process is complex and cost effective. Therefore, for this
aim, the researchers have presented simple relationships ignoring the effect of several factors. In the proposed method, in the case of
nonlinear analysis, linear analysis is directly used first, and then by implementing simple or advanced iterative methods convergence is
achieved in the equilibrium path. Using these methods, after passing limit points, the structure behavior is defined and so accurate design
of the structure is possible. In the present study, for passing the equilibrium path of the space trusses, the Newton–Raphson iterative
algorithm is used along the flow path perpendicular to the Davidenko curves with modified convergence rate. Contrary to the previous
methods, this algorithm uses the homotopy approach, is based upon the new mathematical concepts, and has great ability for developing
complex load-displacement paths of the structures with multidegrees of freedom. At the end, through three numerical examples, three
structures have been analyzed using the algorithm presented in this paper and the results are compared with the previous advanced
iterative methods that have been used for nonlinear analysis of those structures. The ability of the proposed method, particularly for
passing the limit points, has been indicated by those numerical examples.
DOI: 10.1061/共ASCE兲0733-9445共2008兲134:6共998兲
CE Database subject headings: Equilibrium; Limit states; Curvature; Trusses; Structural behavior.

Introduction the structures with many degrees of freedom, generally there is no


explicit solution for the nonlinear system of equations and so,
The stability points in nonlinear elasticity may be classified into several numerical solutions have been presented for such equa-
limit points with snapthrough and limit points with snapback tions by the researchers 共Levy and Spillers 2003兲. Papadrakakis
共Ramm 1981兲. For limit points, the path-tracing scheme to suc- 共1983兲 has employed the dynamic relaxation scheme for the
cessively compute the regular equilibrium points on the equilib- second-order and large deflection analysis of trusses. Leu and
rium path and the pinpointing scheme to precisely locate the Yang 共1990兲 have investigated the effects of rigid body motions
singular equilibrium points are sufficient for the computational and member stretching on the geometrically nonlinear response of
stability analysis. For bifurcation points, however, a specific pro- space truss structures. Static analysis of space truss structures
cedure for path switching is also necessary to detect the branching including inelastic material and large deformation geometric non-
paths to be traced in the postbuckling region. Fujji and Ramm linearities has been considered by Blandford 共1996兲. The nonlin-
共1997兲 have described these fundamental strategies, path tracing, ear analysis problem, formulated as an application of minimum
pinpointing, and path switching, in computational bifurcation potential energy principle, is obviously an optimization problem.
theory. There are analytical solutions for simple geometrical non- For optimization of the total potential, there are various tech-
linear trusses 共Pecknold et al. 1985兲. Jagannathan et al. 共1975兲 niques that have been used such as simulated annealing algorithm
have considered a total Lagrangian formulation with elastic ma- 共Toklu 2004兲. The equilibrium stability problem for a simple class
terial properties on the geometric nonlinear analysis of space truss of elastic space trusses in the shape of a regular pyramid has been
using Green–Lagrange representation of the axial strain. But in examined by Ligaro and Valvo 共2006兲.
Some methods are weak for passing through these points and
1
Associate Professor, Civil Engineering Dept., Univ. of Kerman, P.O. are not able to present the real behavior of the structure. Among
Box 76169133, Kerman, Iran. the methods for solving the nonlinear system of equations, the
2
Assistant Professor, Civil Engineering Dept., Univ. of Kerman, P.O. incremental iterative Newton–Raphson method is used much
Box 76169133, Kerman, Iran. more often 共Baltoz and Dhatt 1979兲, but it is time consuming and
3
Ph.D. Candidate, Civil Engineering Dept., Univ. of Kerman, P.O. expensive in the case of nonlinear analysis of huge structures, and
Box 76169133, Kerman, Iran. E-mail: tabatabaei@iauk.ac.ir it diverges during passing the limit points. The reason for these
Note. Associate Editor: Sashi K. Kunnath. Discussion open until disadvantages is that in the structures with complex behavior the
November 1, 2008. Separate discussions must be submitted for individual load-displacement curve is a combination of softening and hard-
papers. To extend the closing date by one month, a written request must
ening states with limit points. Therefore, the analysis of such
be filed with the ASCE Managing Editor. The manuscript for this paper
was submitted for review and possible publication on May 21, 2007; structures is not possible by simple incremental iterative methods.
approved on October 8, 2007. This paper is part of the Journal of Struc- The simple incremental iterative methods are conducted in the
tural Engineering, Vol. 134, No. 6, June 1, 2008. ©ASCE, ISSN 0733- form of load increments or displacements increments. As the load
9445/2008/6-998–1005/$25.00. level is constant in the method of load increments, passing the

998 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008

J. Struct. Eng., 2008, 134(6): 998-1005


nate system is used as global coordinate system 关Fig. 1共a兲兴. So,
the relationship between the end forces of the member in global
and local coordinates is obtained

兵F其 = 兵B其Q 共1兲


in which Q⫽axial force of the bar in the local coordinate system
and 兵B其⫽transformation matrix that indicates the relationship be-
tween the nodal forces of the member in global and local coordi-
nates systems in the following form:

兵B其 = 兵 l m n − l − m − n 其 共2兲
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

Here l, m and n⫽cosine directors of the deformed member which,


for an arbitrary large nodal displacement, are

共X2 − X1兲 + 共␦4 − ␦1兲


l= 共3兲
L⬘

共Y 2 − Y 1兲 + 共␦5 − ␦2兲
Fig. 1. The internal forces and displacements in the global and local m= 共4兲
coordinate systems: 共a兲 the force and deformation of the member in L⬘
general coordinate system; 共b兲 the member in the local coordinate
共Z2 − Z1兲 + 共␦6 − ␦3兲
n= 共5兲
L⬘
limit points is not possible. Also, if there are serious changes in
the load-displacement path, the number of iterations for con- where X j, Y j, and Z j⫽global coordinates of node j = 1 , 2 of unde-
vergence will be increased and so the method will be time con- formed member; L⫽length of the member before deformation;
suming and expensive. Similarly, in the method of incremental and L⬘⫽length of the member after deformation which is ob-
displacement passing the limit points is difficult. For resolving tained by the following relationship
such disadvantages, advanced analysis methods have been devel-
L⬘ = 关兵共X2 − X1兲 + 共␦4 − ␦1兲其2 + 兵共Y 2 − Y 1兲 + 共␦5 − ␦2兲其2
oped. In these methods, an auxiliary equation is needed for solv-
ing the equilibrium equation. There are several methods for + 兵共Z2 − Z1兲 + 共␦6 − ␦3兲其2兴1/2 共6兲
setting up such auxiliary equation like the methods by Wempner
The axial deformation of the member, u, in local coordinates sys-
共1971兲, Riks 共1979兲, Crisfield 共1981, 1983兲, Forde and Stiemer
tem can be directly obtained from the global displacement 兵␦其 as
共1987兲, Al-Rasby 共1991兲, and Bashir-Ahmad and Xiao-zu 共2004兲.
follows:
The advanced incremental iterative methods have been devel-
oped based upon the arclength approach. In this approach, pro-
u = L − L⬘ 共7兲
portional to the load factor obtained in any iteration, the load
level will converge to the equilibrium path. This process will be and the member stiffness relationship in local coordinates system
continued until the convergence is achieved with acceptable is
accuracy. In these methods, contrary to the simple incremental
iterative methods, it is possible to pass the limit points. The im- AE
Q= u 共8兲
possibility of passing the limit points in the case of highly non- L
linear behavior state is one of the disadvantages of the advanced
incremental iterative methods. In this work, Q is positive when it is compressive 共Fig. 1兲.
In this paper, using modified normal flow algorithm, the num-
ber of the iterations in the vicinity of the mentioned points de- Equilibrium Equations
creases in addition to having the possibility of passing the limit
Applying external nodal loads 兵Fext其 on the structure causes nodal
points in the case of highly nonlinear behavior state. In the nor-
displacement 兵␦i其 stresses and the resultant of the nodal internal
mal flow algorithm the iterations are conducted along the normal
loads 兵f其, so the system of the equilibrium equations of the truss
flow path instead of conducting the iterations concerning one of
can be stated as
the arclength approaches.
F共␭,␦兲 = Fext − f共␦1,␦2, . . . ,␦N兲 = 0, i = 1,2, . . . ,N 共9兲
Geometrical Nonlinear Analysis of Space Truss in which ␭⫽load factor; N⫽number of degrees of freedom of the
structure; and F共␭ , ␦兲⫽vector of the residual forces.
It must be noted that the components of the internal forces of
Force–Displacement Relationships in Global and Local
the members are nonlinear functions of the nodal displacements,
Coordinates Systems
so the equilibrium equations 共9兲 are nonlinear and cannot be
Figure 1 indicates a bar having cross-sectional area equal to A and solved explicitly. Therefore, in the nonlinear analysis of the struc-
modulus of elasticity equal to E. The vectors 兵F其 and 兵␦其 show the tures, the system of the equations is changed into incremental
components of the end forces and end displacements in global form first, and then is solved through several iterative steps. The
coordinates, respectively 共Kassimali 1983兲. Eulerian local coordi- incremental form of the system of equations is

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008 / 999

J. Struct. Eng., 2008, 134(6): 998-1005


关␶兴兵⌬u其 = 兵⌬Q其 共10兲 关Ktg兴兵⌬␦其 = 兵⌬Fext其 共11兲
where 兵⌬Q其 and 兵⌬u其⫽internal forces increment and the dis- where 关K 兴⫽tangential stiffness matrix of the space truss in glo-
tg

placement increment in the local coordinates system, respectively, bal coordinate system which can be determined as follows:
and 关␶兴 is known as tangential stiffness matrix.

Tangential Stiffness Matrix of the Member


关Ktg兴 = 冉 冊
AE
L
兵B其兵B其T + Q关g兴 共12兲

Using the above-mentioned subjects and the tangential stiffness where A, E, and L⫽cross-sectional area, the modulus of elasticity,
matrix concept, the incremental load–displacement relationship in and the length of the member, respectively, and 关g兴⫽geometric
global coordinate system is in the following form: matrix, which can be found as
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

冤 冥
− 共m2 + n2兲 lm ln 共m2 + n2兲 − lm − ln
lm − 共l2 + n2兲 mn − lm 共l2 + n2兲 − mn
1 ln mn − 共l2 + m2兲 − ln − mn 共l2 + m2兲
关g兴 = 共13兲
L⬘ 共m2 + n2兲 − lm − ln − 共m2 + n2兲 lm ln
− lm 共l2 + n2兲 − mn lm − 共l2 + n2兲 mn
− ln − mn 共l + m2兲
2
ln mn − 共l2 + m2兲

Normal Flow Algorithm 兵Fext其ij = ␭toti


j
兵FI其 共16兲
In the present work, the normal flow algorithm has been used for in which, ␭toti
j
⫽factor of total external load and 兵FI其⫽reference
convergence of the equilibrium path. For this aim, sequential it- external load vector.
erations of the Newton–Raphson method have been conducted If Point a in Fig. 3 concerns the converged point in the step
along the path perpendicular to the Davidenko flow curves i − 1 of the load-displacement path, then the following relationship
共Allgower and Georg 1980兲. The nonlinear system of the equa- will exist for the Step i:
tions governing the problem and the structure equilibrium path,
F 苸 RN, is
F共␭,␦兲 = 0 共14兲
␭tot
j
i
= ␭toti−1
conv
+ ␭1i + 兺
j=2
⌬␭ij 共17兲

where ␭ and ␦⫽load factor and the displacement vector, respec-


tively. From the mathematical point of view, Davidenko flow where ␭toti−1
conv
⫽factor of the total external load at the end of the
curves are defined by a perturbation parameter, ␩, in the nonlinear previous step in which the solution has been converged;
system of equations governing the problem as the following form: ␭1i ⫽assumed load increment at the beginning of the calculations;
and ⌬␭ij⫽increment of the load level calculated in each iteration.
F共␭,␦兲 = ␩ 共15兲 In anticipation phase, calculating the tangential stiffness matrix
By changing the perturbation parameter, ␩, and solving the 关Ktg兴ij=0 at Point a, the tangential displacement 兵␦I其ij=1 is deter-
nonlinear system of equations 共15兲, a set of structural response mined first through the following relationship:
paths is obtained. Generally, these response paths are called
Davidenko flow curves. In Fig. 2, the dashed curves indicate
Davidenko flow curves for a system with one degree of freedom.
Also, in Fig. 2 the path perpendicular to the flow is shown. As
mentioned before, in order to obtain the acceptable convergence
rate, the iterations for solving the problem are conducted along
the path perpendicular to the Davidenko flow curves using the
Newton–Raphson technique. For implementing the method, hav-
ing the iteration point j in order to reach j + 1th point, the analysis
is done in jth iterative step. Now, consider the iterative point j on
one of Davidenko curves as shown in Fig. 2. The shortest path for
convergence to the equilibrium path is the normal path.

Iteration Process in the First Step of Normal Flow


Algorithm
Every step of the normal flow algorithm consists of two phases;
anticipation and modifying iterations. If i⫽number of the step and
j⫽number of the modifying iteration, the vector of external load
applying on the nodes, 兵Fext其ij, will be stated as Fig. 2. The modifying iterative steps in the normal flow algorithm

1000 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008

J. Struct. Eng., 2008, 134(6): 998-1005


关兵␦I其ij兴T兵⌬␦R其ij
⌬␭ij = − 共24兲
关兵␦I其ij兴T兵␦I其ij
where 兵␺其ij−1⫽vector of the residual of internal forces 共in other
words, unbalanced forces兲 and 兵⌬␦R其ij⫽vector of unbalanced dis-
placement such that


兵␺其ij−1 = 兵Fint其ij−1 − ␭toti−1
conv
+ ␭ij=1 + 兺
j

j=2
⌬␭ij−1 冊 兵FI其 共25兲

in which Fint⫽vector of resultant internal forces at the nodes. The


vector of unbalanced displacement 兵⌬␦R其 is computed by the fol-
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

lowing system of equations:


关Ktg兴ij−1兵⌬␦R其ij = − 兵␺其ij−1 共26兲
Solving Eq. 共23兲 the particular solution of V is calculated. It must
be noted that in this work, the condition equation presented in the
method of minimum unbalanced displacement has been used
Fig. 3. Comparing normal flow algorithm method with the other 共Chan 1988兲.
advanced iterative methods Step 2: Using Eq. 共27兲, the minimum solution of the norm is
calculated
兵V其T兵␦I其ij
关Ktg兴ij=0兵␦I其ij=1 = 兵FI其 共18兲 兵⌬␦其ij = 兵V其 − 兵␦I其ij 共27兲
储␦I储
Then in the first iteration j = 1, the displacement increment 兵⌬␦其ij=1
in which 兵␦I其ij⫽vector of tangential displacement in the converged
is found using the following relationship:
point in i − 1 step. As mentioned before, this solution is equal to
兵⌬␦其ij=1 = ␭1i 兵␦I其ij=1 共19兲 the step size of Newton–Raphson iterations, Sij. Since in this
paper the method of controlling the displacement has been used,
So, the total displacement or the approximate answer of the equi- the solution is equal to the vector of displacement increment. The
librium path at Point b is updated as follows: vector 兵␦I其ij is the same as the tangential displacement vector
which is converged in step i − 1.
兵␦其ij=1 = 兵␦其i−1
conv
+ 兵⌬␦其ij=1 共20兲
In the next phase, computing the Newton–Raphson step in each Iterations Process for Achieving the Equilibrium Path
iteration j, the solution is obtained and is modified along the real
equilibrium path. After calculating the vector of increment displacement, 兵⌬␦其ij, the
vector of total displacement is updated in jth iteration as

Newton–Raphson Step in Normal Flow Algorithm 兵␦其ij = 兵␦其ij−1 + 兵⌬␦其ij 共28兲


Continuing the phase of modifying iterations in Step i, the On the other hand, in any iteration the load level changes by the
Newton–Raphson step size must be found for the jth iteration. As amount of ⌬␭ij which is obtained from the condition equation 共24兲
mentioned before, the iterations are continued along the flow path and is updated by Eq. 共17兲. The process is continued until the
perpendicular to the Davidenko flow curves until the convergence solution at Point c is converged to the equilibrium path of the
is achieved. Generally, the nonlinear system of equations 共9兲 can structure with acceptable accuracy.
be set up as
Selecting the Amount of Load Increment for Steps i > 1
J共␭ij−1,␦ij−1兲Sij = − F共␭ij−1,␦ij−1兲 共21兲
For steps i ⬎ 1, after obtaining the converged point at the first step
where J共␭ij−1 , ␦ij−1兲⫽Jacobian matrix of order N ⫻ 共N + 1兲 and
i = 1, the load level must be modified selecting suitable load in-
Sij⫽Newton–Raphson step size which is found through crement and the process is repeated similar to the first step. In this

J共␭ij−1,␦ij−1兲 = 冋 ⳵ F共␭ij−1,␦ij−1兲 ⳵ F共␭ij−1,␦ij−1兲


⳵␭ ⳵␦
册 共22兲
work the direct method of updating has been used. In the direct
method of updating, the load increment is related to the number
of iterations and the sign of the determinant of the tangential
In normal flow algorithm, the Newton–Raphson step size is stiffness matrix of the previous step and can be computed through
the minimum solution of the system of equations 共21兲. It must be
noted that Eq. 共21兲 has an infinite number of solutions, but their
minimum solution is unique 共Watson et al. 1981兲. This solution
␭i+1
j
= ⫾ ␭ij 冉 冊
JD
JM

, ␥ = 0.5 共29兲

can be found through two steps: where JD⫽assumed number of iterations at the beginning of the
Step 1: Select a condition equation in the following form and calculations, J M ⫽number of iterations in the previous step, and
solve it together with Eq. 共21兲, the particular solution V will be ␭i+1
j
will be negative if the determinant of the stiffness matrix of
obtained the previous step is negative. Therefore, in the presented algo-
rithm, if the structure has linear behavior, the number of iterations
关Ktg兴ij−1兵V其 = ⌬␭ij兵FI其 − 兵␺其ij−1 共23兲 of the previous step is low and, so, the load increment will be

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008 / 1001

J. Struct. Eng., 2008, 134(6): 998-1005


Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Comparing different methods for passing the complex


equilibrium path

increased based upon Eq. 共29兲 which affects the speed of the
problem solution significantly. On the other hand, if the determi-
nant of the tangential stiffness matrix is negative or equal to zero,
it implies passing the limit point. Also, the negative sign indicates
the negative slope of the equilibrium path of the structure. Stick-
ing to this process, the presented algorithm affords the possibility
of passing the complex equilibrium paths such as shown in Fig. 4.

Convergence Criteria
Fig. 5. Geodesy dome of Example 1
In nonlinear analysis methods suitable criteria must be chosen for
ending the iterations process. In the present paper, the modified
Euclidian criteria for displacement has been used which is in the and going along the hardening branch. The characteristics of
following form: this dome are: the cross-sectional area of the members
A = 645.2 mm2, and the modulus of elasticity E = 68.95⫻ 106 kPa.
关兵⌬␦其ij兴T兵⌬␦其ij
ⱕ ␧c 共30兲 Also, the following parameters are assumed for this dome:
关兵␦其ij兴T兵␦其ij P = 44.48 kN, ⌬␭11 = 0.01, ␭max = 0.5, JD = 2, Jmax = 15, and
where ␧c⫽assumed error at the beginning of the calculations. For ␧c = 10−5.
ending the analysis process two other criteria are introduced as This structure has been analyzed using the method developed
well: the maximum number of the iterations, Jmax, and the maxi- in this paper and the equilibrium path has been drawn as shown in
mum factor of total external load, ␭max. In the numerical examples Fig. 6. Also, the result obtained using the advanced iterative
it is indicated how to use these criteria. methods of Crisfield 共1981兲 and Chan 共1988兲 is indicated in Fig.
6. Further, the total number of increments, the cumulative sum of
total iterations, and the CPU time used by each method is shown
Numerical Examples in Table 1. As it can be seen, in advanced iterative methods which
use the arclength factor, the number of loads steps and conse-
In this section three space trusses are analyzed using the pre- quently the number of iterations is more than the number of it-
sented algorithm. Every node of the truss element has three tran- erations of the method of this paper. Therefore, it can be
sitional degrees of freedom. In Example 1 a 24-member geodesy
truss which has hardening behavior with uniform slope, in Ex-
ample 2 a 168-member truss which has uniform hardening behav-
ior with complex limit points and snapthrough branch and, in
Example 3 a truss having 33 members and 32 degrees of freedom
are considered. For comparison, all examples have been solved
using methods with a 32 bit Pentium 1800 MHz processor.

Example 1
Fig. 5 shows a 24-member geodesy dome under external concen-
trated loading. This structure has 21 degrees of freedom. The
simple hardening behavior of this dome has been paid attention to
by many researchers and is discussed in different references using
different methods 共Kassimali and Bidhendi 1988; Blandford
1996; Papadrakakis 1981兲. The equilibrium path of this structure
consists of a branch with high slope. It must be noted that the load Fig. 6. Comparing the results of different methods for the geodesy
increment iterative methods are weak in passing the limit point dome of Example 1

1002 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008

J. Struct. Eng., 2008, 134(6): 998-1005


Table 1. Performance and Iteration for Example 1
Number of Total Time
Method increments iterations 共s兲
Crisfield 共1983兲 19 69 40.58
Chan 共1988兲 18 68 40.05
Present study 16 66 38.72

concluded that in the case of uniform slope and for hardening


Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

branch with high slope, the normal flow algorithm can analyze the
system more quickly in addition to having the ability of passing Fig. 8. Comparing the results of different methods for the
the equilibrium path of the structure. 168-member dome of Example 2

Example 2 Example 3

The 168-member dome indicated in Fig. 7 is another example that In this example a truss having 33 members and 32 degrees of
has been considered in the previous studies 共Yang et al. 1997兲. freedom is considered. This structure has serious nonlinear behav-
The number of the degrees of freedom of this structure is more ior because of unsymmetric geometry and loading. The character-
than the number of the degrees of freedom of the structure of istics of the structures are: the cross-sectional area of the
Example 1, and also the load characteristics of this structure is members A = 3 cm2, the modulus of elasticity E = 3 ⫻ 104 kN/ cm2,
changed compared to Example 1. The characteristics of this struc- the length L = 10 cm, the high of truss H = 11 cm, and W = 7 cm.
ture are: the cross-sectional area of the members A = 645.2 mm2, The following initial values are assumed at the beginning of
and the modulus of elasticity E = 68.95⫻ 106 kPa. Also, the fol- the analysis: P = 2,000 kN, ⌬␭11 = 0.10, ␭max = 2, JD = 5, Jmax = 20,
lowing parameters are assumed for this dome: P = 444.8 kN, and ␧c = 10−4.
⌬␭11 = 0.25, ␭max = 2, JD = 2, Jmax = 15, and ␧c = 10−4. This structure has complex behavior as shown in Fig. 9. There
As in Example 1, the iterative methods with load increment are snapthrough and snapback branches in the load-displacement
factor are weak in passing the limit point and the snapthrough and curve of this structure. Therefore, this example is very suitable for
the snapback branches. The load-displacement path for this struc- comparing analytical methods. The analyses results are indicated
ture is shown in Fig. 8 and the results are compared to the results in Fig. 10. Based upon Fig. 10 the load increment method does
of the methods of Crisfield 共1983兲 and Chan 共1988兲. As can be not have the ability of analysis and going along the stability path.
seen, the method of the present paper has passed the equilibrium The methods of the other references have long steps when they
path having returning load and returning displacement branches. pass the limit points. But the method of the present work travels
Table 2 summarizes the number of increments, total iterations, along the stability path with fewer iterations compared to the
and the CPU time. In all the advanced iterative methods with other methods because of transmitting the condition equation to
arclength factor, the number of iterations is almost the same, but the normal flow algorithm path as shown in Fig. 10. Table 3
in this example, the normal flow algorithm has the minimum compares number of the total iterations and CPU time used by
number of iterations. each method.

Fig. 7. The 168-member dome of Example 2

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008 / 1003

J. Struct. Eng., 2008, 134(6): 998-1005


Table 2. Performance and Iteration for Example 2 Table 3. Performance and Iteration for Example 3
Number of Total Time Number of Total Time
Method increments iterations 共s兲 Method increments iterations 共s兲
Crisfield 共1983兲 32 113 66.48 Crisfield 共1983兲 62 182 38.22
Chan 共1988兲 31 110 64.70 Chan 共1988兲 54 162 37.31
Present study 27 104 61.17 Present study 48 158 37.11
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

References

Allgower, E. L., and Georg, K. 共1980兲. “Homotopy methods for approxi-


mating several solutions to nonlinear systems of equations.” Numeri-
Fig. 9. The 33-member truss of Example 3 cal solution of highly nonlinear problems, W. Forster, ed., North-
Holland, Amsterdam, The Netherlands, 253–270.
Al-Rasby, S. N. 共1991兲. “Solution techniques in nonlinear structural
analysis.” Comput. Struct., 40共4兲, 985–993.
Baltoz, J. L., and Dhatt, G. 共1979兲. “Incremental displacement algorithms
for nonlinear problems.” Int. J. Numer. Methods Eng., 14共1兲, 1262–
Conclusion 1267.
Bashir-Ahmad, M., and Xiao-zu, S. 共2004兲. “Arc-length technique for
The space trusses have highly nonlinear behavior regarding the nonlinear finite element analysis.” J. Zhejiang Univ., Sci., 5共5兲, 618–
number of degrees of freedom and the level of load applied. 628.
Among the nonlinear analysis methods simple iterative and incre- Blandford, G. E. 共1996兲. “Large deformation analysis of inelastic space
mental methods are weak in passing the limit points of truss structures.” J. Struct. Eng., 122共4兲, 407–415.
snapthrough or snapback of the trusses. In fact, they may fail in Chan, S. L. 共1988兲. “Geometric and material nonlinear analysis of beam
columns and frames using the minimum residual displacement
passing the limit points of the load and displacement. On the
method.” Int. J. Numer. Methods Eng., 26共12兲, 2657–2669.
other hand, in all the advanced iterative methods with arclength
Crisfield, M. A. 共1981兲. “A fast incremental/iterative solution procedure
factor, the number of iterations is high and almost the same. In the that ‘Handles snap-through.’ ” Comput. Struct., 13共1–3兲, 55–62.
present method, based upon the mathematical concepts, the con- Crisfield, M. A. 共1983兲. “An arc-length method including line searches
dition equation is transmitted to the normal flow path first, and and accelerations.” Int. J. Numer. Methods Eng., 19共9兲, 1269–1289.
then by incremental iterative method, the stability path of the Forde, B. W. R., and Stiemer, S. F. 共1987兲. “Improved arc-length orthogo-
structure travels in fewer steps. nality methods for nonlinear finite element analysis.” Comput. Struct.,
It has been shown by the numerical examples that the method 27共5兲, 625–630.
developed in this paper, which uses the normal flow algorithm, Fujii, F., and Ramm, E. 共1997兲. “Computational bifurcation theory: Path-
decreases the time and the cost for nonlinear analysis of huge tracing, pinpointing and path-switching.” Eng. Struct., 19共5兲, 385–
trusses in addition to having the ability of passing the complex 392.
equilibrium paths. Jagannathan, D. S., Christiano, P., and Epstein, H. I. 共1975兲. “Nonlinear
analysis of reticulated space trusses.” J. Struct. Div., 101共12兲, 2641–
2658.
Kassimali, A. 共1983兲. “Large deformation analysis of elastic-plastic
frame.” J. Struct. Eng., 109共8兲, 1869–1885.
Kassimali, A., and Bidhendi, E. 共1988兲. “Stability of trusses under dy-
namic loads.” Comput. Struct., 29共3兲, 381–392.
Leu, L. J., and Yang, Y. B. 共1990兲. “Effects of rigid body and stretching
on nonlinear analysis of trusses.” J. Struct. Eng., 116共10兲, 2582–
2598.
Levy, R., and Spillers, W. R. 共2003兲. Analysis of geometrically nonlinear
structures, 2nd Ed., Kluwer, Dordrecht, The Netherlands.
Ligaro, S. S., and Valvo, P. S. 共2006兲. “Large displacement analysis of
elastic pyramidal trusses.” Int. J. Solids Struct., 43共16兲, 4867–4887.
Papadrakakis, M. 共1981兲. “Post-buckling analysis of spatial structures by
vector iteration methods.” Comput. Struct., 14共5兲, 393–402.
Papadrakakis, M. 共1983兲. “Inelastic post-buckling analysis of trusses” J.
Struct. Eng., 109共9兲, 2129–2147.
Pecknold, D. A., Ghaboussi, J., and Healey, T. J. 共1985兲. “Snap-through
and bifurcation in a simple structure.” J. Eng. Mech., 111共7兲, 909–
922.
Ramm, E. 共1981兲. “Strategies for tracing the nonlinear response near limit
points.” Nonlinear finite element analysis in structural mechanics, W.
Fig. 10. Comparing the results of different methods for the
Wunderlich, E. Stein, and K. J. Bathe, eds., Springer, Berlin, 63–89.
33-member truss of Example 3 Riks, E. 共1979兲. “An incremental approach to the solution of snapping

1004 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008

J. Struct. Eng., 2008, 134(6): 998-1005


and buckling problems.” Int. J. Solids Struct., 15共7兲, 529–551. 1–21.
Toklu, Y. C. 共2004兲. “Nonlinear analysis of trusses through energy mini- Wempner, G. A. 共1971兲. “Discrete approximation related to nonlinear
mization.” Comput. Struct., 82共20-21兲, 1581–1589. theories of solids.” Int. J. Solids Struct., 7共11兲, 1581–1599.
Watson, L. T., Holzer, S. M., and Hansen, M. C. 共1981兲. “Tracking non- Yang, Y. B., Yang, C. T., Chang, T. P., and Chang, P. K. 共1997兲. “Effects
linear equilibrium paths by a homotopy method.” Technical Rep. No. of member buckling and yielding on ultimate strength of space
CS81001-R., Virginia Polytechnic Institute and Blacksburg Univ., trusses.” Eng. Struct., 19共2兲, 179–191.
Downloaded from ascelibrary.org by Sri Siddartha Institute of Technology on 09/30/19. Copyright ASCE. For personal use only; all rights reserved.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2008 / 1005

J. Struct. Eng., 2008, 134(6): 998-1005

You might also like