Professional Documents
Culture Documents
DRauxpajand
DRauxpajand
a r t i c l e i n f o a b s t r a c t
Article history: Twelve new and well known dynamic relaxation methods are considered. Several benchmark frame and
Received 23 May 2011 truss structures, with geometric nonlinear behavior, are solved by these schemes. Based on the total
Accepted 27 August 2012 number of iterations and overall analysis duration, the studied techniques are graded. In this way, the
Available online 18 October 2012
efficiencies of the solvers are found and these algorithms are sorted in terms of their abilities.
Ó 2012 Elsevier Ltd. All rights reserved.
Keywords:
Dynamic relaxation
Efficiency
Convergence rate
Nonlinear analysis
Frame and truss
0045-7949/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruc.2012.08.007
296 M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310
for estimating a better value of the initial displacements. Zhang linear and nonlinear structural behaviors. In nonlinear behavior,
et al. developed the maDR technique and defined the damping fac- the stiffness matrix is a function of displacements and can be
tor for each node of the structure [14]. These researchers revised denoted by SðXÞ. It should be added that the nonlinear systems
the guess of the initial displacement vector as well. It is noted; can be analyzed by an iterative procedure. One of the simple solvers
the critical damping of a single degree of freedom system is pro- suitable for this duty is the dynamic relaxation method. This tech-
portional to the mass and stiffness. Munjiza employed this idea nique transfers Eq. (1) from the static space into the dynamic one
and suggested the Km proportional damping for the DRM [15]. It by adding the fictitious inertial and damping forces, as follows:
should be added, the common DR technique utilizes the first-order
€ þ CX_ þ SX ¼ P
MX ð2Þ
finite difference method. For improving the accuracy, Rezaiee-
Pajand and Taghavian-Hakkak employed the three terms of Taylor It is usual to consider the damping matrix as a factor of the mass
series and formulated a new DR process [16]. In another study, matrix and use the next equality:
Rezaiee-Pajand and Alamatian suggested an addition strategy for
evaluating the fictitious masses and damping factor [17]. C ¼ cM ð3Þ
Some other researchers have focused their efforts on finding the In this equation, c is the damping factor. Employing central finite
optimal time step. Recently, Kadkhodayan et al. minimized the difference method, the following iterative relations can be
residual force to achieve this goal [18]. Minimizing the residual en- achieved:
ergy is another process that was employed by Rezaiee-Pajand and
k k
Alamatian [19]. Lately, Sarafrazi investigated the DR algorithms 1 2 h c _ k1 2h
X_ kþ2 ¼ k
X 2þ k
M1 Rk ð4Þ
and proposed a general formulation [20,21]. In the mentioned 2þh c 2þh c
study, the effect of time step is separated to the value of time step kþ1 1
Xkþ1 ¼ Xk þ h X_ kþ2 ð5Þ
at the current step and the time step ratio. He found the proper
k+1
relations for the optimal time step ratio and critical damping. By In these relations, h is the time step of the finite difference pro-
setting the damping factor to zero, Sarafrazi also proposed an alter- cess. Vector Rk shows the residual force, Rk = P F(Xk). The velocity
native DR algorithm that did not require the damping factor vector at the middle of the interval time is denoted by X_ k1=2 . It is
[20,22]. well known that the damping factor, time step and artificial
The proper value of the time step ratio leads to the near exact mass matrix affect the convergence capability and rate of the
solution. By combining the power iterative process and the DR DRM. In other words, they should be selected in such a way that
technique, Sarafrazi suggested a new method for updating the crit- the iterative procedure converges as soon as possible. To speed
ical damping [20,21]. Recently, Rezaiee-Pajand et al. proposed a and simplify the technique, a diagonal mass matrix is selected for
novel technique for updating the damping factor [23]. In fact, their this process. It should be added that the iterative loop should start
approach improves the convergence rate, but it requires inversing from the initial values of velocity and displacement vectors. By
the stiffness matrix. It should be noted that the Dynamic Relaxa- assigning an acceptable residual error, er, the algorithm is able to
tion process does not assemble the global stiffness matrix. In other converge to the solution of Eq. (1). Usually, analyzers prefer to
words, the displacements are calculated by implementing some utilize zero initial conditions. All aforementioned processes can be
simple vector operations. Having the simplicity, efficiency and outlined as below:
robustness, the DRM has been employed for analyzing the highly
0
nonlinear structures as well [24–26]. To have more information (a) Defining er ; X0 ; X_ 1=2 ¼ 0 and h .
about the progress of the DRM and the related applications, a short (b) k = 0.
review of the events can be found in the recent publications (c) Assembling the internal force vector and applying boundary
[22,23]. conditions.
The aforementioned statements briefly show the existence of a (d) Evaluating the residual forces, artificial mass matrix and
verity of dynamic relaxation algorithms. Most of the creators of damping factor.
these methodologies describe the capabilities of their techniques, (e) Updating the value of time step h.
individually. In other words, and more accurately, there is no com- (f) Calculating X_ kþ1=2 and Xk+1.
prehensive comparison study available to clarify relatively the effi- (g) If kRk+1k 6 er then stop the algorithm.
ciencies and robustness of these schemes. Based on this fact, (h) k = k + 1.
twelve of the mentioned techniques are selected in this article. (i) If k 6 kmax, continue the DR iteration from step (c).
First, a short overview of these formulations will be presented.
Afterwards, the related algorithms will be compared with each Here, kmax is the maximum allowable number of iterations,
other in the nonlinear analysis of the truss and frame structures. which should be defined by the analyst. So far, several methods
This comparative study is based on the total number of iterations have been suggested to evaluate the DR parameters. These tech-
and overall analysis duration. Finally, the performances of men- niques have some virtues and demerits. Some of the famous or re-
tioned strategies will be graded and the best methods will be cent techniques are selected and reviewed in the next sections.
introduced. Afterward, these techniques will be numerically compared with
each other.
It was shown that the error ratios of DRM were related to the fi ðxk1 Þ þ fi ðxk Þ
sl;k
ii ¼ ð17Þ
eigenvalues of B = D1S matrix. Papadrakakis performed an error k1=2
hx_ i
analysis and found the optimum values of h2/q and ch/q ratios as
below: It should be noted, the lowest circular frequency was assumed less
! than two, such as 1.9, when the result of (xk)TSl,kxk was negative.
2
h 4
¼ ð8Þ
q
opt
kBmax þ kBmin 2.3. Qiang DR technique
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ch 4 kBmax kBmin P
¼ ð9Þ Qiang calculated the mass matrix via the mii ¼ j jsij j relation
q opt kBmax þ kBmin
and found the optimum values of ch and h as below [11]:
In these equations, kBmin and kBmax are the lowest and the highest pffiffiffiffiffiffiffiffiffiffiffi
eigenvalues of B, respectively. As a common way, Papadrakakis 4 kBmin 2
ch ¼ and h ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð18Þ
established the highest eigenvalue via the Gerschgörin theorem: 1 þ kBmin 1 þ kBmin
X
n
The value of the lowest eigenvalue, kBmin, was calculated utilizing
jkBmax j < max jbij j ð10Þ
i
j¼1
the next Rayleigh relation:
On the other hand, the lowest eigenvalue was computed using the ðfXk gT SXk Þ
next DRM characteristic polynomial: kBmin ¼
ðfXk gT MXk Þ
k2DR kDR b þ a
kBmin ¼ ð11Þ
kDR c 2.4. The maDR method
The parameter kDR is the rate of the error decay in the sequential
iterations, and a, b and c are related to the DR parameters as below: In 1989, Zhang and Yu suggested a modified adaptive dynamic
relaxation (maDR) process [13]. They calculated DR mass matrix by
2
2 ch=q 2h =q utilizing Eq. (14), and evaluated damping factor via c = 2x0.
a¼ ; b ¼ a þ 1 and c ¼ ð12Þ
Furthermore, the following equation was employed to calculate
2 þ ch=q 2 þ ch=q
the lowest circular frequency:
Papadrakakis approximated the value of q by utilizing the following
relation: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðXk ÞT f k
x0 ¼ ð19Þ
kxkþ1 xk k ðXk ÞT Mk Xk
kDR ¼ ð13Þ
kxk xk1 k
Zhang and Yu employed non-zero initial displacements, as follows:
It should be noted, the value of kBmin and kBmax are guessed in the
first iteration. xi þ x
x0i ¼ i
ð20Þ
2
2.2. Underwood procedure
In the last equation, xi and x
i are the first minimum and maximum
Underwood proposed the common DR method [10]. The value displacements in the un-damped system.
of time step was taken to be equal to one in this algorithm. How-
ever, the mass matrix was calculated by utilizing the Gerschgörin 2.5. DODR technique
theorem as below:
By introducing the damping factor for each node separately,
1 2X
n
Zhang and his coworkers developed the maDR algorithm [14]. In
mii P h jsij j ð14Þ
4 j¼1 this process, the artificial damping matrix for node i is calculated
from ci = 1imi. Here, 1i and mi are the damping factor and mass
To assure a stable DR process, the analyzer can utilize h = 1.1. It is matrix of the mentioned structural node, respectively. These
obvious; the stiffness matrix changes when the structural behavior researchers employed the next Rayleigh principle for each node
is nonlinear. Consequently, the artificial mass matrix should be and obtained 1i:
updated. Underwood suggested recalculating the mass matrix when
the maximum value of defined error in the system becomes greater vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u k T k
u x f
than one. The value of error in each degree of freedom was 1i ¼ 2t iT i ð21Þ
calculated as follows: xki mki xki
h €xki €xk1
2
ek
¼ i
ð15Þ It is worth emphasizing that Zhang et al. formulation ignores the
i
4 xki xk1
i
effects of other nodes on the total energy of node i. This algorithm
usually converges to the unacceptable solution, when the truss and
Moreover, the fictitious damping factor was obtained by using
frame structures are analyzed utilizing the stiffness method [27]. To
c = 2x0. In this relation, x0 is the lowest circular frequency of the
increase the solver ability, the same damping factors are used for all
artificial dynamic system. The Rayleigh principle was employed to
nodes in the present study. It should be noted; Zhang et al.
calculate x0 in the kth iteration, as below:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi improved Zhang and Yu relation for initial displacements. They sug-
ðXk ÞT Sl;k Xk gested performing the DR process by arbitrary initial displacements
x0 ¼ ð16Þ to find the first extreme point of the response curve. These analyz-
ðXk ÞT Mk Xk
ers averaged the obtained solution and utilized X0 to update X0.
The matrix Sl,k in the last equation is a local stiffness matrix and is Two or three repetitions of the mentioned approach give the
evaluated by utilizing the subsequent internal force vector: acceptable value of the initial response.
298 M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310
2
h €k
Xkþ1 ¼ Xk þ hX_ k þ X ð22Þ
2
These authors suggested calculating X € k from the dynamic equilib-
rium equation, Rk ¼ MX € k þ CX_ k , and X_ k by utilizing X_ k ¼ ðXk
Xk1 Þ=h. Furthermore, they considered the artificial mass matrix
as a factor of the stiffness matrix, M = aS. By assuming the time step
equal to one in this algorithm, the convergence rate increases when
the value of a is less than one. The value a = 0.6 was employed for
all truss structures. The damping matrix is calculated as a factor of
the mass matrix, similar to the Qiang suggestion [11]. The men-
tioned algorithm is called RPTH technique in this paper.
Fig. 1. Truss TR1.
2.7. Kinetic damping method
Table 1
Comparing the DR processes in analysis of TR1.
It is noted that the present article utilizes the kdDR abbreviation for
the aforementioned technique. Fig. 5. Truss TR3.
The optimum value of the time step in the DRM was studied by
Kadkhodayan et al. [18]. By employing Eqs. (14) and (21), they cal-
culated the DR mass matrix and damping factors, respectively.
After that, these authors introduced the following relation for the
rate reduction in the sum of the squares of the nodal residual
forces, ZR:
X
q
2 X
q
k 2
pikþ1 fikþ1 6 ZR ri ð26Þ
i¼1 i¼1
In the last equation, the rates of changes in the internal forces were
P
denoted by f_ kþ1=2 , which is approximately equal to qj¼1 sij x_ j
kþ1=2
. By
minimizing the parameter ZR, Kadkhodayan et al. found the subse- Utilizing the optimum time step leads to the modified fictitious time
quent optimum time step: process (MFT) [18]. It is worth emphasizing that the second-order
Pq k _ kþ1=2 derivative of ZR with respect to hk+1 is always positive. When satis-
kþ1 r f fying this condition, Eq. (28) guides the highest convergence rate. It
h ¼ Pi¼1 i ð28Þ
_
ðf kþ1=2 Þ2 ought to be added that the obtained value for time step is
Table 2
Comparing the DR processes in analysis of TR2.
Table 3
Comparing the DR processes in analysis of TR3.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h iffi
They also employed c ¼ x20 4 ðhk Þ2 x20 to calculate damping
factor. Furthermore, the lowest circular frequency is calculated via
Eq. (19).
By minimizing Eq. (31), with respect to the time step, one can find a
second-order function, which gives two values for the time step.
One of these time steps makes the related second-order derivative
positive. This value of the time step is optimum one. Taking advan-
tage of this time step will accelerate the DRM, even within a few ini-
tial iterations [20]. If there is no feasible time step, the one
calculated by Eq. (28) is employed.
Table 4
Comparing the DR processes in analysis of TR4.
Table 5
Comparing the DR processes in analysis of TR5.
The twelve aforementioned algorithms are programmed to ana- 3.1. Simple truss
lyze trusses and plane frames. In addition, some benchmark prob-
lems are selected and the geometric nonlinear behaviors of these Fig. 1 shows a single degree of freedom truss that is denoted by
structures are traced by applying the entire loads in ten steps. TR1. In this structure, the value of AE is 68,029,310 kgf and P equals
The acceptable residual errors are the same for all solutions and to 32 105 kgf [17]. The load–displacement curve of TR1 is shown
are equal to 104 (er = 104). Also, the total number of iterations in Fig. 2. The dots in the mentioned figure show the responses of
and the analysis durations are recorded for each case. All the con- the converged methods. Furthermore, Table 1 presents the total
verged algorithms were shown by dots. These schemes had the number of iterations Ii and analysis duration ti for the studied DR
M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310 303
Table 6
Comparing the DR processes in analysis of TR6.
Fig. 14. The load–displacement curve for Truss TR7. 3.3. Chan truss
Fig. 5 shows TR3 structure that was analyzed by Chan and Lau.
procedures. Based on the values of Ii and ti, the scores EiI and EiT are This truss has a strong nonlinear behavior [28]. The value of AE for
respectively calculated and presented in Table 1. all structural members is 9 103. The parameters k and P are set to
According to the results, the maDR, DODR and MFT approaches 5 and 1, respectively. The static path for index displacement is
cannot analyze this structure. The rate of declining residual error in illustrated in Fig. 6. Table 3 demonstrates the analyses outcomes
these methods is very slow. Even by using the maximum allowable and comparison results.
number of iterations in each loading step (kmax), which is defined It should be noted, in step 10 of loading, the Qiang and RPTH
60,000 for the computer program, the acceptable residual error procedures did not converge to acceptable residual error in the
(er) is not attained. Based on the scores, the other techniques are admissible number of iterations (i.e. 60,000 iterations).
graded from one to nine as recorded in Table 1. The column of SI
shows the grades of each technique based on the iteration scores. 3.4. 22-element truss (TR-4)
On the other hand, the marks of analysis duration are shown in
the column of St. In this example, the grades of maDR, DODR and The structure of Fig. 7 has 22 elements . The member’s area
MFT processes are equal to zero. sections are 20 and 40 in.2 for diagonal and other members,
304 M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310
Table 7
Comparing the DR processes in analysis of TR7.
respectively [16]. The load–displacement curve is drawn in Fig. 8. the bridge versus the load variation is plotted in Fig. 10. Table 5
The analysis results for this truss are recorded in Table 4. shows the total number of iterations and overall analysis durations
All discussed formulations converge to the true nonlinear solu- for different DR algorithms. All the twelve DR processes are graded
tion of this truss. Besides, the Underwood and zdDR techniques in the mentioned table, as well.
perform a fewer iterations and require a smaller amount of the In this example, the Papadrakakis method diverges in loading
times for solving this problem. steps 3–10. Furthermore, it is interesting to notice that the RPTH
and kdDR techniques converge to wrong solutions after step 2 of
3.5. Bridge truss loading. Based on these unacceptable results, the grades of these
solvers are equal to zero.
TR5 is a bridge truss as shown in the Fig. 9. The values of section
area for the bottom and top of the longitude members are respec- 3.6. Truss TR-6
tively 0.003 and 0.005. The other cross section areas of the truss
members are equal to 0.001. The modulus of elasticity for the This three dimensional truss with 24 elements is drawn in
structure is 2 1010 [20]. The curve of maximum deflection of Fig. 11. The value of section area and modulus of elasticity of its
M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310 305
t/m
q=100
60 , u
q=100
88
q=100
90
q=100
Fig. 16. The load–displacement curve for Truss TR8. 60
members are 3.17 and 303,000, respectively [16]. Also, P is equal to q=100 ton
650. Fig. 12 shows the load–displacement curve for top point of the 30
truss. Moreover, the outcomes of analyses are arranged in Table 6.
3.9. Five-story frame and loading of the mentioned structure that is denoted by FR1.
The columns and beams of the frame are W2150 and W1835,
In this example, a five-story frame structure with rigid beam to respectively. The frame modulus of elasticity is equal to
column connections is considered. Fig. 17 presents the geometry 2 107 t/m2 [20]. The curve of index displacement via the load is
Table 8
Comparing the DR processes in analysis of TR8.
Table 9
Comparing the DR processes in analysis of FR1.
q=250 kgf/cm
90000 kgf
q=25
75000
q=250
60000
q=250
45000
6@350 cm
q=250
30000
q=2
15000
550 cm
Table 10
The section properties of FR2.
Table 11
Comparing the DR processes in analysis of FR2.
Table 12
Comparing the DR processes in analysis of FR3.
Table 13
Comparing the DR processes in analysis of FR4.
Table 14
Comparing the DR processes in the analysis of Arch structure.
Table 15
Grading the DR algorithms based on the analysis duration.
i Method Cij Gi
j=0 j=1 j=2 j=3 j=4 j=5 j=6 j=7 j=8 j=9 j = 10 j = 11 j = 12
1 Papadrakakis [9] 5 0 0 0 0 0 2 2 0 1 2 0 1 24
2 Underwood [10] 1 2 6 1 1 0 0 0 0 1 0 1 0 74
3 Qiang [11] 4 1 0 0 0 1 0 0 1 1 2 1 2 25
4 maDR [13] 2 0 0 1 0 0 0 1 6 1 1 1 0 35
5 DODR [14] 2 0 0 0 1 0 3 3 2 1 1 0 0 42
6 RPTH [16] 4 2 0 0 0 1 0 1 2 0 0 3 0 35
7 kdDR [7] 1 6 1 0 1 3 0 0 0 1 0 0 0 77
8 MFT [18] 2 1 0 1 1 2 2 2 1 1 0 0 0 53
9 mdDR [17,19] 0 0 0 1 2 2 0 2 0 4 1 1 0 49
10 MRE [19] 0 1 1 3 2 3 2 0 0 0 0 0 1 71
11 RPS [20,21] 0 0 2 2 3 1 2 2 0 0 1 0 0 68
12 zdDR [20,22] 1 0 3 4 2 0 2 0 1 0 0 0 0 71
Table 16
Grading the DR algorithms based on the total number of iterations.
i Method Cij Gi
j=0 j=1 j=2 j=3 j=4 j=5 j=6 j=7 j=8 j=9 j = 10 j = 11 j = 12
1 Papadrakakis [9] 5 0 0 0 1 0 1 2 0 1 2 0 1 25
2 Underwood [10] 1 2 6 1 1 0 0 1 0 1 0 0 0 76
3 Qiang [11] 4 1 0 0 0 0 0 1 1 1 1 2 2 23
4 maDR [13] 2 0 0 0 1 1 1 2 3 1 2 0 0 39
5 DODR [14] 2 0 0 0 1 1 1 0 3 3 1 1 0 36
6 RPTH [16] 4 1 0 0 1 0 0 1 1 1 1 3 0 29
7 kdDR [7] 1 6 1 0 0 1 3 0 0 0 1 0 0 74
8 MFT [18] 2 1 1 0 1 4 2 1 1 0 0 0 0 57
9 mdDR [17,19] 0 0 1 0 1 1 2 4 0 3 1 0 0 52
10 MRE [19] 0 1 1 2 1 3 3 0 1 0 0 1 0 67
11 RPS [20,21] 0 1 1 4 3 1 0 1 2 0 0 0 0 73
12 zdDR [20,22] 1 0 2 6 2 1 0 0 1 0 0 0 0 72
Table 17
Five top DR methods.
Qiang and RPTH methods are so slow that in the allowable number 4. Conclusions
of iterations for one loading step (i.e. 60,000 iterations), they can-
not attain to the acceptable residual error. Furthermore, in loading In this comparative study, several geometrical nonlinear struc-
steps 1–3, the zdDR technique does not achieve the acceptable tures have been analyzed by 12 famous and recently proposed DR
residual error in 60,000 iterations. algorithms. Based on the total number of iterations and overall
310 M. Rezaiee-Pajand et al. / Computers and Structures 112-113 (2012) 295–310
analysis duration, a number between 1 and 12 is assigned for each [7] Topping BHV, Ivanyi P. Computer aided design of cable-membrane
structures. Saxe-Coburg Publications; 2007.
DR technique in solving each benchmark problem. The mark one
[8] Bunce JW. A note on estimation of critical damping in dynamic relaxation. Int J
stands for the best approach, and the number 12 indicates the Numer Methods Eng 1972;4:301–4.
worst one. Consequently, one can enumerate rank j for the scheme [9] Papadrakakis M. A method for automatic evaluation of the dynamic relaxation
i and defines Cij. For example, the kdDR method has been the fifth parameters. Comput Methods Appl Mech Eng 1981;25:35–48.
[10] Underwood P. Dynamic relaxation. In: Belytschko T, Hughes TJR, editors.
process three times. Therefore, Ci5 is three for the mentioned algo- Computational method for transient analysis. Amsterdam: Elsevier; 1983. p.
rithm. The value of Ci0 demonstrates the number of structure 245–65.
which technique i could not analyze it. Based on the values of Cij, [11] Qiang S. An adaptive dynamic relaxation method for nonlinear problems.
Comput Struct 1988;30:855–9.
the grade of method i is calculated by the following formulas: [12] Al-Shawi FAN, Mardirosian AH. An improved dynamic relaxation method for
the analysis of plate bending problems. Comput Struct 1987;27:237–40.
X
12
[13] Zhang LC, Yu TX. Modified adaptive dynamic relaxation method and its
Gi ¼ 100 C ij ð13 jÞ=156 application to elastic–plastic bending and wrinkling of circular plates. Comput
j¼1 Struct 1989;34:609–14.
[14] Zhang LC, Kadkhodayan M, Mai YW. Development of the maDR method.
For instance, the number 156 is obtained when a technique obtains Comput Struct 1994;52:1–8.
score 12 in all 13 solved problems. Consequently, the Grade Gi for [15] Munjiza A. A Km proportional damping for dynamic relaxation. Int J Eng Model
1996;9:1–9.
this strategy is 100. Based on the analyses duration and total num- [16] Rezaiee-Pajand M, Taghavian-Hakkak M. Nonlinear analysis of truss structures
ber of iterations, the values of countered ranks, Cij, and grades, Gi, using dynamic relaxation. Int J Eng 2006;19:11–22.
are arranged in Tables 15 and 16, respectively. [17] Rezaiee-Pajand M, Alamatian J. The dynamic relaxation method using new
formulation for fictitious mass and damping. Struct Eng Mech
Based on the obtained grades, the five top procedures are ar- 2010;34:109–33.
ranged in Table 17. The kdDR algorithm requires lesser time to per- [18] Kadkhodayan M, Alamatian J, Turvey GJ. A new fictitious time for the dynamic
form structural analysis. On the other hand, the Underwood relaxation (DXDR) method. Int J Numer Methods Eng 2008;74:996–1018.
[19] Alamatian J. Numerical integration for structural analysis. Dissertation
procedure requires a fewer iterations to find the answers. How-
submitted in partial fulfillment of the requirement for the degree of doctor
ever, the later methodology increases the analysis duration, and of philosophy, Ferdowsi University, Mashhad, Iran; 2007 [in Persian].
based on this aspect, the mentioned formulation occupies the sec- [20] Sarafrazi SR. Numerical integration for structural dynamic analysis.
ond position. As it is demonstrated in the table, the kdDR strategy Dissertation submitted in partial fulfillment of the requirement for the
degree of doctor of philosophy, Ferdowsi University, Mashhad, Iran; 2010 [in
has second rank in the iteration requirements. It is worth empha- Persian].
sizing that the Underwood, kdDR and zdDR techniques are among [21] Rezaiee-Pajand M, Sarafrazi SR. Nonlinear structural analysis using dynamic
the top techniques, which cannot analyze the structure in one relaxation method with improved convergence rate. Int J Comput Methods
2010;7:1–28.
example. Furthermore, the MFT method cannot solve two exam- [22] Rezaiee-Pajand M, Sarafrazi SR. Nonlinear dynamic structural analysis using
ples. This fact explains that these approaches are not reliable in zero damping dynamic relaxation. Comput Struct 2011. http://dx.doi.org/
the analysis of a general truss or frame structures. 10.1016/j.compstruc.2011.04.005.
[23] Rezaiee-Pajand M, Kadkhodayan M, Alamatian J, Zhang LC. A new method of
fictitious viscous damping determination for the dynamic relaxation. Comput
References Struct 2011;89:783–94.
[24] Rezaiee-Pajand M, Alamatian J. Automatic DR structural analysis of snap-
[1] Frankel SP. Convergence rates of iterative treatments of partial differential through and snap-back using optimized load increments. J Struct Eng ASCE
equations. Math Tables Aids Comput 1950;4:65–75. 2011;137:109–16.
[2] Day AS. An introduction to dynamic relaxation. Engr 1965;219:218–21. [25] Lee K, Han S, Park T. A simple explicit arc-length method using the dynamic
[3] Otter JRH. Computations of prestressed concrete pressure vessels using relaxation method with kinetic damping. Comput Struct 2011;89:216–33.
dynamic relaxation. Nucl Struct Eng 1965;1:61–75. [26] Golmakani ME, Kadkhodayan M. Nonlinear bending analysis of annular FGM
[4] Lynch RD, Kelsey S, Saxe HC. The application of dynamic relaxation to the finite plates using higherorder shear deformation plate theories. Compos Struct
element method of structural analysis. Technical report No. Themis-Und-68-1, 2010. http://dx.doi.org/10.1016/j.compstruct.2010.06.024.
University of Notre Dame, Indiana; 1968. [27] Alamatian J. Structural analysis using dynamic relaxation method. Thesis
[5] Rushton KR. Dynamic relaxation solutions of elastic plate problems. J Strain presented in partial fulfillment of the requirement for the degree of master of
Anal 1968;3:23–32. science, Ferdowsi University, Mashhad, Iran; 2002 [in Persian].
[6] Cundall PA. Explicit finite-difference methods in geomechanics. In: [28] Chan ASL, Lau TB. Further development of the reduced basis method for
Proceedings of the EF conference on numerical methods in geomechanics, geometric nonlinear analysis. Comput Methods Appl Mech Eng
Blacksburg, VA; 1976. p. 132–50. 1987;62:127–44.