You are on page 1of 7

Dynamic Relaxation Using

Continuous Kinetic
Samuel Jung
Mechanical Engineering,
Pusan National University,
Damping—Part I: Basic Algorithm
Busan 609-735, South Korea
e-mail: jung40L@hanmail.net Dynamic relaxation (DR) is the most widely used approach for static equilibrium analy-
ses. Specifically, DR compels dynamic systems to converge to a static equilibrium
Tae-Yun Kim through the addition of fictitious damping. DR methods are classified by the method in
Mechanical Engineering, which fictitious damping is applied. Conventional DR methods use a fictitious mass
Pusan National University, matrix to increase the fictitious damping while maintaining numerical stability. There are
Busan 609-735, South Korea many calculation methods for the fictitious mass matrix; however, it is difficult to select
e-mail: tykid76@gmail.com the appropriate method. In addition, these methods require a stiffness matrix of a model,
which makes it difficult to apply nonlinear models. To resolve these problems, a new DR
Wan-Suk Yoo1 method that uses continuous kinetic damping (CKDR) is proposed in this study. The pro-
posed method does not require the fictitious mass matrix and any tuning coefficients, and
Professor
it possesses a second-order convergence rate. The aforementioned advantages are unique
Fellow ASME
and significant when compared to those of conventional methods. The stability and con-
Mechanical Engineering,
vergence rate were analyzed by using an eigenvalue analysis and demonstrated by simu-
Pusan National University,
lating nonlinear models of a pendulum and cable. Simple but representative models were
Busan 609-735, South Korea
used to clearly demonstrate the features of the proposed DR method and to enable the
e-mail: wsyoo@pusan.ac.kr
reproducibility of the verification results. [DOI: 10.1115/1.4039838]

Keywords: dynamic relaxation, static equilibrium, form finding, kinetic damping,


nonlinear model

1 Introduction The first method consists of DR with VDR that constitutes a


classic DR method [4]. In the VDR process, an artificial dynamic
Most mechanical systems are in static equilibrium when they
system is constructed by adding fictitious damping to the original
are not in operation. The shape of a mechanical system in static
system. This is followed by solving the artificial system with a
equilibrium is termed a static configuration. A static configuration
time integrator such that the responses converge to the steady-
provides very important information that is used in static and
state values that satisfy the static equilibrium condition of the
dynamic analyses.
original dynamic system. However, it is necessary to determine
From a static viewpoint, the stable static configuration is the
a fictitious viscous damping coefficient with estimation
preferred geometry in structural design because the potential
methods [2].
energy of a static equilibrium structure corresponds to a local min-
The second method corresponds to DR with KDR, which is a
imum. This process is widely known as form finding and is used
variant of VDR and was proposed by Cundall [5]. The kinetic
to design the equilibrium shapes of self-supporting structures such
damping is applied to an artificial dynamic system by resetting the
as airbags and architectures [1]. From a dynamic point of view,
velocity to zero at each peak of the kinetic energy, as shown in
the static configuration is the initial condition that does not cause
Fig. 1. This process is continued until the artificial dynamic sys-
a natural response in a dynamic simulation.
tem reaches static equilibrium. In contrast to VDR, the KDR
Therefore, determination of the static configuration of a
method does not require a fictitious damping coefficient. How-
mechanical system is an important process in computer-aided
ever, the KDR method uses conditional operations such as energy
engineering. Most practical models are generally very complex;
peak detection; thus, it is difficult to predict the convergence rate
thus, the static configuration can only be achieved by a numerical
and stability in the process. Here, the convergence rate is a term
process termed the static equilibrium analysis.
that indicates how quickly the initial configuration of a model
Dynamic relaxation (DR) is a widely used technique for static
converges to the final solution. A higher convergence rate means
equilibrium analysis. The DR technique is based on the fact that a
damped system ultimately comes to rest in the displaced position
of the static equilibrium [2]. The addition of fictitious damping
compels the dynamic system to converge to a static equilibrium
[3]. The traditional DR methods use the central finite difference
method, which is an explicit method, and fictitious mass matrix is
introduced to maintain numerical stability [2]. The DR technique
is categorized into two groups on the basis of the manner in which
fictitious damping is applied—namely, relaxation with viscous
damping (VDR) and kinetic damping (KDR) [1].

1
Corresponding author.
Contributed by the Design Engineering Division of ASME for publication in the
JOURNAL OF COMPUTATIONAL AND NONLINEAR DYNAMICS. Manuscript received July 7,
2017; final manuscript received March 21, 2018; published online July 6, 2018.
Assoc. Editor: Zdravko Terze. Fig. 1 Kinetic energy history of kinetic relaxation

Journal of Computational and Nonlinear Dynamics AUGUST 2018, Vol. 13 / 081006-1


C 2018 by ASME
Copyright V

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


that a smaller number of time steps are required in the DR 1 h
process. vikþ1 ¼ vi þ ri (8)
1 þ cDR h k ð1 þ cDR hÞm
 ii kþ1
The present study proposes a new DR method using continuous
kinetic damping (CKDR) to resolve the aforementioned problems. i
Continuous kinetic damping is applied to an artificial dynamic dkþ1 ¼ dki þ hvikþ1 (9)
system by resetting the velocities to zero at every time-step. This
concept leads to a new DR method that is completely different If Eqs. (8) and (9) are repeatedly calculated, then dkþ1
from the conventional method. Furthermore, it is possible to rigor- converges to a static displacement vector dst that satisfies the fol-
ously prove the second-order convergence rate and absolute sta- lowing expression:
bility of CKDR through an eigenvalue analysis. These advantages
are specific to CKDR as opposed to conventional DR methods. r ¼ 0 ¼ f  pðdst Þ (10)
Section 2 explains the conventional DR method in detail. In
Sec. 3, the CKDR method is derived from the implicit Euler This process allows the VDR method to obtain a static configu-
method, and its convergence rate and stability are analyzed by ration of the dynamic system. However, the convergence rate of
using an eigenvalue analysis. The algorithm and pseudocode of VDR depends on the fictitious damping coefficient cDR ; thus, it is
CKDR are provided in Sec. 4. Simulations for comparative verifi- necessary to estimate the optimal value at each step [2].
cation were performed, and the results are discussed in Sec. 5.
The conclusions are described in Sec. 6. 2.2 Kinetic Damping. The KDR method does not use ficti-
tious viscous damping to attenuate the artificial dynamic system.
Therefore, cDR ¼ 0, and this is followed by simplifying Eqs. (8)
2 Conventional Dynamic Relaxation Method and (9) as follows:
The KDR method is derived from VDR; thus, VDR is described h i
first in this section, and this is followed by a description of KDR. vikþ1 ¼ vik þ r (11)
 ii kþ1
m
This is followed by a description of the advantages and disadvan-
tages of VDR and KDR. i
dkþ1 ¼ dki þ hvikþ1 (12)

For each step, the kinetic energy Tkþ1 is calculated as follows:


2.1 Viscous Damping. The matrix equation of structural
dynamics is given as follows: 1 X ii  i 2
Tkþ1 ¼  vkþ1
m (13)
2 i
Ma þ Cv þ pðdÞ ¼ f (1)

where M and C denote the mass and damping matrices, respec- If Tkþ1 < Tk is detected when Eqs. (11) and (12) are repeatedly
tively; and p and f denote the internal and external force vectors, calculated, then it is necessary to reset vkþ1 to zero. This process
respectively. Additionally, a, v, and d denote the acceleration, is repeated such that rk converges to zero, and the kinetic energy
velocity, and displacement, respectively. The VDR method intro- peaks disappear, as shown in Fig. 1. When the norm of rk
duces the following artificially damped dynamic system for fast decreases below a tolerance, the KDR process is terminated, and
energy dissipation to the original dynamic system of Eq. (1) as the last displacement vector is output; this approximately satisfies
follows: Eq. (10).

Ma  ¼r
 þ cDR Mv (2) 2.3 Drawbacks of Conventional Dynamic Relaxation
Methods. The VDR and KDR methods are widely used for static
r ¼ f  pðdÞ (3) equilibrium analysis and are applied to certain commercial soft-
ware packages for structural analysis, such as RADIOSS [7]. How-
where cDR is the fictitious damping coefficient, r is a residual ever, the conventional DR method requires the resolution of the
force vector, and M  is the fictitious mass matrix and assumed to following two problems.
be diagonal. There are many methods for calculating the fictitious First, conventional DR methods require the calculation of the
mass matrix, which are introduced in detail in Ref. [2]. Assuming fictitious mass matrix for the stability of the algorithm. The ficti-
that the fictitious mass matrix is given, Eq. (2) is expressed as tious mass matrix can be calculated from the actual stiffness
follows: matrix, and there are many calculation methods according to
Ref. [2]. However, finding a suitable method is one of the difficul-
 ii ai þ cDR m
m  ii vi ¼ r i (4) ties of applying conventional DR methods.
Second, the KDR method performs a conditional process due to
To solve Eq. (4), various time integrators are used according to kinetic energy peak detection, and this yields discontinuous
the specific purpose [6,7]. However, in the present study, the fol- results, as shown in Fig. 1. This discontinuity makes it difficult to
lowing implicit Euler method is used to enable a fair comparison predict the convergence rate and stability of the method.
with the CKDR method as follows: To resolve these problems, the present study proposes the
CKDR method that does not require the fictitious mass matrix and
qkþ1 ¼ qk þ hq_ kþ1 (5) any tuning coefficients, and it is performed in a continuous pro-
cess. In Sec. 3, the CKDR method is derived, and its characteris-
where h and k denote the step size and step index, respectively. tics are analyzed in detail.
The second-order form of Eq. (5) is expressed as follows:

vkþ1 ¼ vk þ hakþ1 (6) 3 Continuous Kinetic Damping Process


As mentioned earlier, DR uses a time integration method to
dkþ1 ¼ dk þ hvkþ1 (7) solve an artificially damped dynamic system. However, there are
several types of time integration methods (e.g., Euler, central
Substituting Eq. (6) into Eq. (4), the VDR method is expressed finite difference, trapezoidal, Runge–Kutta, and generalized-a
as follows: [8]). It is obvious that the integrator type influences the stability

081006-2 / Vol. 13, AUGUST 2018 Transactions of the ASME

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


and convergence rate of DR. The CKDR method is based on an dkþ1 ¼ dk þ h2 akþ1 (22)
implicit Euler method, which is a first-order method and an abso-
lute stable (A-stable) method for a stiff system. The reason as to vkþ1 ¼ 0 (23)
why the implicit Euler method is best suited for CKDR is
explained by performing a comparison with the other integrators When compared to the KDR method in Eqs. (11)–(13), the
in Sec. 3.3.
CKDR method does not require the fictitious mass matrix and any
tuning coefficients. Only the acceleration response is used during
3.1 Derivation of Continuous Kinetic Damping. The equa- the iterative calculation process. This feature solves the model
tion of motion for a damped system with an external force is given dependencies of conventional DR methods. The convergence rate
as follows: and stability of the CKDR method are described in Sec. 3.2.

a þ 2fxn v þ x2n d ¼ f (14)


3.2 Convergence Rate and Stability of Continuous Kinetic
Damping. The characteristics of an iterative numerical method
where a, v, and d denote the acceleration, velocity, and displace-
can be analyzed with the eigenvalues of its amplification matrix
ment, respectively. Additionally, f, xn , and f denote the damping
[10]. To obtain the amplification matrix of CKDR, the difference
ratio, natural frequency, and external force, respectively. Note
matrix equation of CKDR is obtained from Eq. (19) as follows:
that the single degree-of-freedom linear model of Eq. (14) is
assumed for eigenvalue analysis of CKDR, but it can also be ^y ^y
applied to multi-degree-of-freedom nonlinear models. With L^ kþ1 ¼ R^ k (24)
respect to the time integration, Eqs. (6) and (7) can be expressed
as follows: where
     
dkþ1 ¼ dk þ hvk þ h2 akþ1 (15) dk ^¼ 1 1 ^ ¼ 1 0
y^k ¼ ; L ; R
h2 ak h2 xn 2 1 h2 xn 2 1
vkþ1 ¼ vk þ hakþ1 (16)
The recursive form of Eq. (24) is expressed as follows:
If the external force f of Eq. (14) is assumed to be constant at
steps k and k þ 1, the responses at steps k and k þ 1 must satisfy ^y
y^kþ1 ¼ P^ (25)
k
the following equation: 2 3
1
akþ1 þ 2fxn vkþ1 þ x2n dkþ1 ¼ ak þ 2fxn vk þ x2n dk (17) 6 1 7
^ ¼6
^ 1 R
^¼L h2 xn 2 þ 1 7
P 6 7 (26)
4 1 5
For the eigenvalue analysis, the difference matrix equations of 0 2 2
Eqs. (15)–(17) can be expressed as follows: h xn þ 1

y kþ1 ¼ R
L yk (18) where P^ denotes the amplification matrix of CKDR, and its eigen-
values are as follows:
where  
1
2 3 2 3 z1;2 ¼ ; 1 (27)
dk 1 0 1 h2 xn 2 þ 1
6 7 6 7
yk ¼ 4 hvk 5; L ¼ 4 0 1 1 5;
From the eigenvalues, the differential displacement response
h2 ak h 2 xn 2 2fhxn 1 can be expressed as [11] follows:
2 3
1 1 0
6 7 dn ¼ c1 zn1 þ c2 zn2 (28)
R¼4 0 1 05
h2 xn 2 2fhxn 1 where n denotes the step number, and c1 and c2 denote undeter-
mined coefficients. Because z1 < 1 and z2 ¼ 1 for hxn > 0, the
The continuous kinetic damping is applied to a dynamic system initial displacement d0 and static displacement dst are expressed
by resetting the velocity to zero at every step. This is implemented as follows:
by deleting the second row and column of L and R, which is a
technique used to handle a fixed constraint [9] d 0 ¼ c1 þ c2 (29)
2 32 3 2 32 3
L11 6 L12 L13 dkþ1 R11 6 R12 R13 dk dst ð¼ d1 Þ ¼ c2 (30)
6 76 7 6 76 7
4 6 L21 6 L22 6 L23 54 hvkþ1 5 ¼ 4 6 R21 6 R22 6 R23 54 hvk 5
Substituting Eqs. (29) and (30) into Eq. (28) results in the
L31 6 L32 L33 h2 akþ1 R31 6 R32 R33 h2 ak discrete displacement response by the CKDR method that is
(19) expressed as follows:

This technique allows vkþ1 to be fixed at zero at every step. The dn ¼ ðd0  dst Þgn þ dst (31)
displacement and velocity equations of Eq. (19) are expressed for
the k þ 1 step as follows: where g is defined as the amplification factor of CKDR, and it
replaces the principal eigenvalue z1 of Eq. (27) as follows:
L11 dkþ1 þ L13 h2 akþ1 ¼ R11 dk þ R13 h2 ak (20)
1
g¼ (32)
hvkþ1 ¼ 0 (21) h2 xn 2 þ1

From Eqs. (20) and (21), the equations of the CKDR method According to Eq. (31), an increase in the step number n leads to
are derived as follows: the convergence of dn from d0 to dst with the convergence rate

Journal of Computational and Nonlinear Dynamics AUGUST 2018, Vol. 13 / 081006-3

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 1 Amplification factors of continuous kinetic damping
with various integrators

Integrator type Amplification factor (g)

Implicit Euler 1
h2 xn 2 þ 1
Central finite difference 1
Trapezoidal h2 xn 2  4

h2 xn 2 þ 4
Runge–Kutta fourth 1 4 4 1 2 2
h xn  h xn þ 1
24 2
Generalized-a 2
h2 xn 2 þ 2
Fig. 2 Amplification factor of CKDR

based on g. By rearranging Eq. (31) for steps k and k þ 1, the dis-


placement convergence is expressed as follows:

dkþ1  dst ¼ gðdk  dst Þ (33)

Further, from the second row of Eq. (25), the acceleration con-
vergence is expressed as follows:

akþ1 ¼ gak (34)

Equations (33) and (34) indicate that the dynamic system of


Eq. (14) converges to the static equilibrium state with the second-
order convergence rate of Eq. (32). Therefore, the amplification
factor g determines the characteristics of CKDR. The amplifica-
tion factor g is a function of hxn , as obtained from Eq. (32), and it
is shown in Fig. 2. The stability and convergence characteristics
of the CKDR method are listed in detail as follows.
(1) It is an absolutely stable (A-stable) method. For all positive
values of hxn , the amplification factor g is less than or
equal to one; thus, the responses do not diverge at any step
size. Fig. 3 Amplification factors of CKDR with various integrators
(2) It possesses a second-order convergence rate. This is
because the convergence rate is proportional to h2 on the hxn
basis of Eq. (32). Therefore, the convergence rate increases h=T ¼ (35)
2p
exponentially as the step size increases.
(3) Oscillations do not occur in the iterative process. This is From Fig. 3, the characteristics of CKDR utilizing the other
because g is a real positive number for all positive integrators include the following:
values of hxn ; thus, the acceleration response does not
change signs according to Eq. (34). Thus, the responses (1) Central finite difference method: Specifically, g ¼ 1; thus,
converge without oscillation, similar to those of a first- the responses do not converge even if the CKDR process is
order system. repeated infinitely.
(4) The convergence rate is not affected by the damping of the (2) Trapezoidal method: If h=T > 1=p, then the convergence
model. This is because the amplification factor g is not a rate is negative; thus, the responses exhibit chattering at
function of f. each step.
(3) Runge-Kutta fourth method: If h=T > 0:55, then g > 1;
thus, the responses exhibit divergence.
3.3 Characterization of Continuous Kinetic Damping by (4) Generalized-a method: Specifically, jgj is lower when com-
the Integrator Type. As previously described, the characteristics pared with that of the implicit Euler method for all h=T;
of DR are influenced by the type of integrator. The implicit Euler thus, the convergence rate is relatively low. According to
method is the most suitable for the CKDR method among the Eq. (32), the convergence rate of CKDR with the implicit
most representative integrators (implicit Euler method, central Euler method is twice that of the generalized-a method.
finite difference, trapezoidal, Runge–Kutta, and generalized-a).
This is because the implicit Euler method has the highest numeri- The aforementioned reasons indicate that the implicit Euler
cal damping of the existing integrators [12]. Excessively high method is the best integrator for the CKDR method. As shown in
numerical damping is a factor that is responsible for lowering the Sec. 3.2, CKDR using implicit Euler method is A-stable and pos-
accuracy in a dynamic analysis. However, it achieves a large con- sesses a second-order convergence rate. The CKDR algorithm is
vergence rate in a static equilibrium analysis. In order to demon- detailed in Sec. 4.
strate the suitability of the implicit method, the amplification
factors with respect to the other integrators were also derived in a 4 Continuous Kinetic Damping Algorithm
manner similar to that in the case of the implicit Euler’s method.
These amplification factors are listed in Table 1 and shown in A second-order dynamic system can be expressed by nonlinear
Fig. 3. The X-axis of Fig. 3 corresponds to the ratio of the time- differential equations as follows:
step to the natural period that is given by the following
expression: aðtÞ ¼ gðt; dðtÞ; vðtÞÞ (36)

081006-4 / Vol. 13, AUGUST 2018 Transactions of the ASME

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 2 Pseudocode of the continuous kinetic damping
algorithm

Input: d0 , t0 , h, ac
v0 0, a0 0, k 0
While Do
Find akþ1 satisfying wðakþ1 Þ ¼ 0
dkþ1 dk þ h2 akþ1
If kakþ1 k1 < ap , Then break while, End
k kþ1
End
Return: dkþ1

Fig. 4 Simple pendulum model


The CKDR is an implicit method; thus, it can be expressed as a
problem of solving the following implicit equations:

wðakþ1 Þ ¼ gðtkþ1 ; dkþ1 ; vkþ1 Þ  akþ1 ¼ 0 (37)

where

dkþ1 ¼ dk þ h2 akþ1 (38)

vkþ1 ¼ 0 (39)

tkþ1 ¼ t0 (40)

Here, Eqs. (38) and (39) correspond to the matrix forms of


Eqs. (22) and (23). It should be noted that Eq. (40) is added to pre-
Fig. 5 Convergence of the pendulum with h 5 0.45
vent time-dependent changes in the model. To start the CKDR
algorithm, the initial displacement d0 is obtained from the initial
configuration of a model. The key part of the CKDR algorithm
involves determining akþ1 that satisfies Eq. (37). A nonlinear
equation solver, such as Newton–Raphson, is used to solve this
implicit equation in which ak is selected as the initial value for the
solver. Subsequently, dkþ1 is calculated from Eq. (38). This proce-
dure is repeated until the following termination conditions are
satisfied:

kakþ1 k1 < ap (41)

where ap ð> 0Þ denotes the acceleration tolerance. On the basis of


the proposed framework, the pseudocode of the CKDR algorithm
is provided in Table 2.

5 Verification Simulations
Static equilibrium simulations of simple pendulum and cable
models were performed to verify the CKDR method. The acceler-
ation tolerance ap corresponded to 0.001. The built-in “fsolve.m” Fig. 6 Number of time steps of KDR and CKDR for the pendu-
function of MATLAB was used to solve the implicit equations. The lum model
same time-step, termination condition, and integrator were used to
facilitate a fair comparison of KDR and CKDR. shows the convergence of angles by KDR and CKDR. Although
physical damping was absent, the angle of the pendulum con-
5.1 Simple Pendulum Model. The simple pendulum is one verged quickly from the initial angle to zero in both methods. The
of the simplest nonlinear models, and it is used in the verification KDR method exhibited an underdamped response due to the
of commercial software [13]. The simple pendulum consists of a discontinuous process of KDR. This transient response makes it
mass m that hangs from a string of length L, as shown in Fig. 4. difficult to predict the convergence results. In contrast, CKDR dis-
The equation of motion for the model is given as follows: played a very stable convergence process that is similar to that of
a first-order system, as predicted in Sec. 3.2.
€h ¼  g sin h (42) The number of time steps of each method differed on the basis
L of the step size, as shown in Fig. 6. It should be noted that the step
size of DR is only a parameter that controls the convergence
where g ¼ 9:81 m=s2 , L ¼ 1 m, and h0 ¼ p=4. The natural fre- speed. Therefore, a comparison of the number of time steps at the
quency of the simple pendulum at h ¼ 0 is as follows: same step size does not reveal the efficiency difference between
pffiffiffiffiffiffiffiffi the two methods. However, the change in the number of time
xn ¼ g=L  3:13 rad=s (43) steps according to the step size change indicates the convergence
characteristic of each method. In case of KDR, the number of
The results obtained by KDR and CKDR were first compared, time steps converged to a specific value after exhibiting an incon-
and the convergence rate of CKDR is considered in detail. Figure 5 sistent change when the step size increased. This irregularity

Journal of Computational and Nonlinear Dynamics AUGUST 2018, Vol. 13 / 081006-5

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 7 Amplification ratios of the simple pendulum model Fig. 9 Deflection of node 10 with h 5 0.45

Table 3 Amplification factors of the simple pendulum model

h ¼ 0.2 h ¼ 0.4 h ¼ 0.6

g 0.72 0.39 0.22

Fig. 10 Number of time steps of KDR and CKDR for the pendu-
lum model

Fig. 8 Static shape of the cable model

makes it difficult to predict the convergence rate of the KDR


method. In contrast, the number of time steps of CKDR decreased
monotonically when the step size increased. This is because the
convergence rate of CKDR is proportional to h2 according to
Eq. (32).
To verify the convergence rate of CKDR in further detail, the
amplification ratios of the angular acceleration were calculated
and are shown in Fig. 7, and they correspond to the amplification
factor based on Eq. (34). The amplification ratios converge to spe-
cific values with increases in the step size. For comparison pur-
poses, the amplification factors were calculated by using the
natural frequency of Eq. (43), as summarized in Table 3. It should Fig. 11 Amplification ratio of the cable model at node 10
be noted that the amplification factors in Table 3 are equal to the
amplification ratios in Fig. 7. This indicates that the amplification
factor of the CKDR method was accurately analyzed.
CKDR decreased monotonically with increases in the step size.
However, the number of time steps of KDR converged after
5.2 Cable Model. As an example of the form-finding prob- exhibiting inconsistent changes. As shown in Fig. 11, the amplifi-
lem, the simple cable model with both ends fixed is shown in cation ratio of the acceleration converged to 0.49. This value is
Fig. 8. This cable model is modeled with a spring–mass system equal to the amplification factor that is calculated by using the
and consists of 20 nodes, each with a mass of 0.05 kg, and lowest natural frequency that corresponds to 3.35 rad/s of the
each node is connected with a spring with a spring constant of cable model.
500 N/m. Both the KDR and CKDR methods yielded the same
static shape as that shown in Fig. 8.
Figure 9 shows the deflection of node 10, which is similar to 6 Conclusion
that of the pendulum. The deflection by KDR changed suddenly, Dynamic relaxation is widely used for static equilibrium analy-
although the deflection by CKDR gradually converged. The num- sis. However, conventional DR methods require the calculation of
ber of time steps is shown in Fig. 10, and the tendency was similar a fictitious mass matrix to maintain the stability of the algorithm,
to that of the pendulum model. The number of time steps of and it is necessary to select a suitable calculation method for

081006-6 / Vol. 13, AUGUST 2018 Transactions of the ASME

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


stable computation. In the case of the VDR method, the tuning References
parameters must also be selected appropriately to increase conver- [1] Garcia, J. R., 2011, “Numerical Study of Dynamic Relaxation Methods and
gence rate. The present study proposed a new DR method termed Contribution to the Modelling of Inflatable Lifejackets,” Ph.D. dissertation,
CKDR to resolve this problem. The basic concept of this method Universite de Bretagne Sud, Lorient, France.
[2] Rezaiee-Pajand, M., and Taghavian-Hakkak, M., 2006, “Nonlinear Analysis of
involves resetting the velocity to zero at every step in the time Truss Structures Using Dynamic Relaxation,” Int. J. Eng., 19(1), pp. 11–22.
integration. This concept was valid for a static equilibrium analy- [3] Rezaiee-Pajand, M., and Rezaee, H., 2012, “Fictitious Time Step for the Kinetic
sis, and the implicit Euler method corresponded to the best time Dynamic Relaxation Method,” Mech. Adv. Mater. Struct., 21(8), pp. 631–644.
integration method for CKDR. The proposed method does not [4] Day, A. S., 1965, “An Introduction to Dynamic Relaxation(Dynamic Relaxa-
tion Method for Structural Analysis, Using Computer to Calculate Internal
require calculation of the fictitious mass matrix and tuning of the Forces Following Development From Initially Unloaded State),” Eng.,
damping coefficients, and it exhibits a second-order convergence 219(5688), pp. 218–221.
rate. These characteristics represent significant and unique advan- [5] Cundall, P. A., 1976, “Explicit Finite Difference Methods in Geomechanics,”
tages when compared to conventional methods. Comparison simu- International Conference on Numerical Methods in Geomechanics, pp.
132–150.
lations with conventional KDR verified the convergence rate of [6] Namadchi, A. H., and Alamatian, J., 2016, “Explicit Dynamic Analysis Using
CKDR. An increase in the step size exponentially increased the Dynamic Relaxation Method,” Comput. Struct., 175, pp. 91–99.
convergence rate of CKDR when compared to that of KDR. The [7] AltairVR , 2011, “RADIOSS User’s Guide: Example 16—Dummy Positioning,”

convergence rate of CKDR is closely related to the step size; thus, Altair, Troy, MI.
[8] Chung, J., and Hulbert, G. M., 1993, “A Time Integration Algorithm for Struc-
proper step size selection is very important. Hence, an adaptive tural Dynamics With Improved Numerical Dissipation: The Generalized-a
step size method will be proposed in Part 2 to facilitate the practi- Method,” ASME J. Appl. Mech., 60(2), pp. 371–375.
cal use of CKDR. [9] Petyt, M., 2010, Introduction to Finite Element Vibration Analysis, 2nd ed.,
Cambridge University Press, Cambridge, UK.
[10] Hilber, H. M., Hughes, T. J. R., and Taylor, R. L., 1977, “Improved Numerical
Dissipation for Time Integration Algorithms in Structural Dynamics,” Earth-
Funding Data quake Eng. Struct. Dyn., 5(3), pp. 283–292.
[11] Hulbert, G. M., and Chung, J., 1994, “The Unimportance of the Spurious Root
 This research was supported by a grant for a project managed by of Time Integration Algorithms for Structural Dynamics,” Commun. Numer.
the Agency for Defense Development, “Technology develop- Methods Eng., 10(8), pp. 591–597.
[12] Cellier, F. E., and Kofman, E., 2006, Continuous System Simulation, Springer
ment for rescue robots capable of lifting over 120 kgf,” which Science & Business Media, Berlin.
was funded by the Civil-Military Technology Cooperation [13] FunctionBay, 2014, “Theoretical Manual: RecuryDyn/Solver,” FunctionBay,
Program. Gyeonggi-do, South Korea.

Journal of Computational and Nonlinear Dynamics AUGUST 2018, Vol. 13 / 081006-7

Downloaded From: https://computationalnonlinear.asmedigitalcollection.asme.org on 12/17/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like