Recurdyn Solver - Theoretical Manual

R
R
e
e
c
c
u
u
r
r
D
D
y
y
n
n
/
/
S
S
o
o
l
l
v
v
e
e
r
r

T
T
h
h
e
e
o
o
r
r
e
e
t
t
i
i
c
c
a
a
l
l
M
M
a
a
n
n
u
u
a
a
l
l

Copyright 2005 FunctionBay, Inc. All rights reserved

This document may not be copied, disclosed, or modified without the prior
written consent of FunctionBay, Inc.
Information described in this document is furnished for general information
only, is subject to change without notice, and should not be constructed as a
warranty or inaccuracies that may appear in this manual.
The software described in this document is provided under written license only,
contains valuable trade secrets and proprietary information, and is protected by
the copyright laws of the United States and Other countries. UNAUTHORIZED
USE OF SOFTWARE OR ITS DOCUMENTATION CAN RESULT IN CIVIL DAMAGES AND
CRIMINAL PROSECUTION.

Edition Note

This theoretical manual documents the theoretical background of the
RecurDyn / Solver.

Trademarks of FunctionBay, Inc.

RecurDyn is a registered trademark of FunctionBay, Inc.
RecurDyn/Professional, RecurDyn/SOLVER, RecurDyn/SOLID,
RecurDyn/FLEX, RecurDyn/NodalFlex, RecurDyn/LINEAR,
RecurDyn/CONTROL, RecurDyn/TRACK_HM,
RecurDyn/TRACK_LM, RecurDyn/CHAIN, RecurDyn/MTT2D,
RecurDyn/MTT3D, RecurDyn/BELT, RecurDyn/HAT,
RecurDyn/Hydraulic, RecurDyn/Gear, RecurDyn/Hydraulic,
RecurDyn/Tire are trademarks of FunctionBay, Inc.

Revision History
First printed, April 2001
1
st
Revision, January 2002
2
nd
Revision, July 2002
3
rd
Revision, August 2002
4
th
Revision, September 2003
5
th
Revision, September 2005

RecurDyn
/SOLVER THEORETICAL MANUAL

TABLE OF CONTENTS

1. GENERALIZED RECURSIVE FORMULATION 1- 1

1.1. INTRODUCTION ... 1- 1

1.2. RELATIVE COORDINATE KINEMATICS .. 1- 2

1.2.1 COORDINATE SYSTEMS . 1- 2
1.2.2 RELATIVE KINEMATICS FOR A PAIR OF CONTIGUOUS BODIES 1- 4

1.3. GENERALIZATION OF THE VALOCITY RECURSIVE FORMULA ... 1- 6

1.4. GENERALIZATION OF THE FORCE RECURSIVE FORMULA 1- 8

1.5. GRAPH REPRESENTATIONS OF MECHANICAL SYSTEMS ... 1- 9

1.6. EQUATIONS OF MOTION AND DAE SOLUTION METHOD .. 1-10

1.7. GENERALIZED RECURSIVE FORMULAS . 1-13

1.7.1 RECURSIVE FORMULA FOR . x B X
& &
= 1-14
1.7.2 RECURSIVE FORMULA FOR RM BOLD ...
q q
Bx X ) ( =
1-14
1.7.3 RECURSIVE FORMULA FOR
q
T
q
G) (B g =
1-15
1.7.4 RECURSIVE FORMULA FOR AND ..
q q
x) B ( X
& &
=
v v
x B X ) (
& &
=
1-18

1.8. APPLICATION OF THE GENERALIZED RECURSIVE FORMULAS . 1-18

1.8.1 COMPUTATION OF THE RESIDUAL F 1-18
1.8.2 COMPUTATION OF THE JACOBIAN
q
F
1-19

1.9. NUMERICAL EXAMPLES ..... 1-20

1.9.1 A GOVERNOR MECHANISM .. 1-20
1.9.2 A MULTI-WHEELED VEHICLE .. 1-22

1.10. CONCLUSIONS .. 1- 24

RecrDyn

/ Solver THEORETICAL MANUAL

REFERENCES ... 1- 25

APPENDIX A : RECURSIVE FORMULAS 1- 26

2. DECOUPLING SOLUTION METHOD FOR IMPLICIT

NUMERICAL INTEGRATION .. 2- 1


2.2. IMPLICIT NUMERICAL INTEGRATION FOR DIFFERENTIAL
ALGEBRAIC EQUATIONS .. 2- 2

2.3. A DECOUPLING SOLUTION METHOD FOR IMPLICIT NUMERICAL INTEGRATION
.. 2- 4

2.4. NUMERICAL ALGORITHM .. 2-7

2.5. NUMERICAL EXAMPLES . 2-7

2.5.1 QUICK-RETURN MECHANISM .. 2-7
2.5.2 AIR COMPRESSOR . 2-10

2.6. CONCLUSIONS ..... 2-12

REFERENCES ..... 2-13

3. FLEXIBLE MULTIBODY DYNAMICS USING A VIRTUAL BODY

AND JOINT . 3- 1


3.2. KINEMATICS TWO CONTIGUOUS FLEXIBLE BODIES . 3- 3

3.2.1. COORDINATE SYSTEMS AND VIRTUAL BODIES 3-3
3.2.2. JOINT CONSTRAINTS BETWEEN TWO RIGID BODIES .. 3-4
3.2.3. FLEXIBLE BODY JOINT CONSTRAINT BETWEEN A FLEXIBLE BODY
AND A RIGID VIRTUAL BODY . 3-5

3.3. EQUATIONS OF MOTION . 3- 9
3.3.1. COEFFICIENT MATRIX OF CONVENTIONAL AUGMENTED
FORMULATION 3-10
3.3.2. COEFFICIENT MATRIX OF PROPOSED AUGMENTED FORMULATION 3-11
3.3.3. NON-SINGULARITY OF AUGMENTED MASS MATRIX 3-12

3.4. COMPUTER IMPLEMENTATION AND DISCUSSIONS . 3-13

3.4.1. NUMERICAL ALGORITHM . 3-13
3.4.2. COMPARISON OF DIFFERENT IMPLEMENTATION METHODS 3-14

3.5. NUMERICAL RESULTS .. 3-17

3.5.1. FLEXIBLE SLIDER CRANK MECHANISM ... 3-17
3.5.2. FLEXIBLE PENDULUM MECHANISM . 3-20

3.6. SUMMARY AND CONCLUSIONS 3-22


4. GENERALIZED RECURSIVE FORMULATION FOR FLEXIBLE

MULTIBODY DYNAMICS 4- 1


4.2. RELATIVE COORDINATE KINEMATICS OF TWO CONTIGUOUS
FLEXIBLE BODIES .... 4- 3

4.2.1. COORDINATE SYSTEMS AND VIRTUAL BODIES . 4-3
4.2.2. RELATIVE KINEMATICS FOR A FLEXIBLE BODY JOINT 4-5
4.2.3. RELATIVE KINEMATICS FOR A RIGID BODY JOINT .. 4-7
4.2.4. GRAPH REPRESENTATIONS OF MECHANICAL SYSTEMS .. 4-8

4.3. FORWARD RECURSIVE FORMULAS . 4-10

4.3.1. GENERALIZATION OF THE VELOCITY RECURSIVE FORMULA . 4-10

RecrDyn


4.3.2. RECURSIVE FORMULA FOR
q q
) (Bx X =
4-11

4.4. BACKWARD RECURSIVE FORMULAS ... 4-13

4.4.1. GENERALIZATION OF THE FORCE RECURSIVE FORMULA 4-13
4.4.2. Recursive Formula for
k k
q
T
q
) ( G B g =
4-15

4.5. THE GOVERNING EQUATIONS OF SOLUTION .. 4-17

4.5.1. IMPLICIT INTEGRATION OF THE EQUATIONS OF MOTION 4-17
4.5.2. APPLICATION OF THE GENERALIZED RECURSIVE FORMULAS 4-19

4.6. NUMERICAL RESULTS . 4-20

4.7. CONCLUSIONS . 4-23


5. RELATIVE NODAL METHOD FOR LARGE DEFORMATION PROBLEM
.... 5-1

5.1. INTRODUCTION ... 5-1

5.2. RELATIVE DEFORMATION KINEMATICS . 5-2

5.2.1. GRAPH THEORETIC REPRESENTATION OF A STRUCTURE 5-2
5.2.2. KINEMATIC DEFINITIONS .. 5-3

5.3. GOVERNING EQUATIONS OF EQUILIBRIUM 5-6

5.3.1. STRAIN ENERGY 5-6
5.3.2. EXTERNAL FORCE . 5-7
5.3.3. CONSTRAINT . 5-8
5.3.4. EQUATIONS OF EQUILIBRIUM 5-10

5.4. NUMERICAL ALGORITHM 5-11

5.5. NUMERICAL EXAMPLES .. 5-12


6. DYNAMIC ANALYSIS OF HIGH-MOBILITY TRACKED VEHICLES
.................... 6-1


6.2. HIGH-SPEED, HIGH-MOBILITY TRACKED VEHICLES 6-3

6.3. KINEMATIC RELATIONSHIPS AND EQUATIONS OF MOTION .. 6-6

6.4. A COMPLIANT TRACK MODEL .. 6-7

6.4.1. SINGLE PIN CONNECTION 6-8
6.4.2. DOUBLE PIN CONNECTION .. 6-10

6.5. MEASUREMENT OF TRACK COMPLIANCE CHARACTERISTICS .. 6-12

6.6. CONTACT FORCES .. 6-15

6.6.1. INTERACTION BETWEEN TRACK AND ROAD WHEEL, IDLER, AND
SUPPORT ROLLER . 6-15
6.6.2. TRACK CENTER GUIDE AND ROAD WHEEL INTERACTIONS 6-16
6.6.3. INTERACTION BETWEEN SPROCKET TEETH AND TRACK LINK PINS 6-17
6.6.4. GROUND AND TRACK SHOE INTEGRATION .. 6-18

6.7. METHOD OF NUMERICAL INTEGRATION . 6-19

6.7.1. ACCURACY ANALYSIS 6-19
6.7.2. STABILITY ANALYSIS . 6-20
6.7.3. IMPLEMENTATION OF A VARIABLE STEPPING ALGORITHM . 6-22

6.8. NUMERICAL RESULTS . 6-24

RecrDyn




7. DYNAMIC TRACK TENSION OF HIGH MOBILITY TRACKED VEHICLES

.
7-1


7.2. NUMERICAL MODEL OF A HIGH MOBILITY TRACKED VEHICLE ... 7-3

7.3. INTERACTION GROUNDS . 7-8

7.4. MEASUREMENT OF THE DYNAMIC TRACK . 7-16

7.5. NUMERICAL INVESTIGATION OF DYNAMIC TRACK TENSION . 7-17

7.6. FUTURE WORK AND CONCLUSIONS . 7-22


8. EFFICIENT CONTACT AND NONLINEAR DYNAMIC

MODELING FOR TRACKED VEHICLES .. 8-1


8.2. MULTIBODY TRACKED VEHICLE MODEL AND PARAMETER EXTRACTIONS
. 8-4

8.3. EFFICIENT CONTACT SEARCH ALGORITHM . 8-5

8.3.1. ROAD WHEEL-TRACK LINK CONTACT 8-5
8.3.2. SPROCKET-TRACK LINK CONTACT ... 8-6
8.3.3. GROUND-TRACK LINK CONTACT . 8-7

8.4. EQUATIONS OF MOTION . 8-10
8.5. EXTENDED BEKKERS SOIL MODEL FOR MULTIBODY TRACK SYSTEM 8-11
8.6. SUMMARY AND CONCLUSIONS 8-15


9. AN EFFICIENT CONTACT SEARCH ALGORITHM FOR

GENERAL MULTIBODY SYSTEM DYNAMICS . 9-1


9.2 KINEMATIC NOTATIONS OF A CONTACT PAIR . 9-3

9.3. DIVISION OF THE CONTACT DOMAIN .. 9-4

9.4. PRE-SEARCH 9-6

9.5. POST-SEARCH AND COMPLIANCE CONTACT FORCE .. 9-7

9.6. KINEMATICS AND EQUATION OF MOTION FOR THE RECURSIVE FORMULAS
9-9

9.7. NUMERICAL INTEGRATION STRATEGY 9-12

9.8. NUMERICAL EXAMPLE . 9-13

9.9 CONCLUSIONS ... 9-15


10. LINEARIZED EQUATIONS OF MOTION FOR MULTIBODY

SYSTEMS WITH CLOSE LOOPS .. 10-1

RecrDyn


10.1. INTRODUCTION .. 10-1

10.2. RELATIVE COORDINATE KINEMATICS . 10-4

10.3. EQUATIONS OF MOTION 10-7

10.4. ELIMINATION OF LAGRANGE MULTIPLIERS AND LINEARIZATION OF
THE EQUATIONS OF MOTION . 10-8

10.5. NUMERICAL EXAMPLE. 10-10

10.5.1. FOURBAR MECHANISM WITH A SPRING .. 10-10
10.5.2. CANTILEVER BEAM DRIVEN BY A MOTION 10-13
10.5.3. A SPRING SYSTEM WITH 2 D.O.F 10-14
10.5.4. A CANTILEVER BEAM . 10-15

10.9 CONCLUSIONS ... 10-18


11. NONLINEAR DYNAMIC MODELING OF

SILENT CHAIN DRIVE 11-1


11.2. MULTIBODY MODELING OF SILENT CHAIN DRIVE .. 11-4

11.2.1. SPROCKET 11-4
11.2.2. SILENT CHAIN LINK . 11-4
11.2.3. TENSIONER AND CHAIN GUIDE .. 11-6
11.2.4. EQUATIONS OF MOTION AND INTEGRATION .. 11-6

11.3. CONTACT FORCE ANALYSIS .. 11-8

11.3.1. STRATEGE OF CONTACT SEARCH 11-8
11.3.2. LINE-ARC CONTACT 11-9
11.3.3. ARC-POINT CONTACT . 11-12
11.3.4. ARC-ARC CONTACT . 11-13

11.3.5. LINE-POINT CONTACT . 11-13
11.3.6. CONTACT FORCE MODEL 11-14
11.4. NUMERICAL STUDY OF AN AUTOMOTIVE SILENT CHAIN SYSTEM . 11-14
11.5. FUTURE WORK AND CONCLUSIONS . 11-18


12. DYNAMIC ANALYSIS AND CONTACT MODELING FOR

TWO DIMENSIONAL MEDIA TRANSPORT SYSTEM . 12-1


12.2. TWO DIMENSIONAL FLEXIBLE MULTIBODY SHEET . 12-3

12.3. CONTACT FORCE ANALYSIS .. 12-5

12.3.1. KINEMATICS NOTATIONS . 12-6
12.3.2. SHEET AND ROLLER INTERACTIONS ... 12-7
12.3.3. ROLLERS INTERACTIONS 12-9
12.3.4. SHEET AND ROLLER INTERACTIONS .. 12-9

12.4. EQUATIONS OF MOTION 12-12

12.5. NUMERICAL RESULTS 12-14

12.6. CONCLUSIONS 12-16


13. HYDRAULIC AUTO TENSIONER (HAT) FOR

BELT DRIVE SYSTEM ... 13-1


RecrDyn


13.2. MULTIBODY SIMULATION MODEL ... 13-3

13.3. EQUATIONS OF MOTION .. 13-4

13.4. HYDRAULIC FORCES 13-5

13.4.1. OIL FLOW RATES THROUGH THE CHECK VALVE . 13-6
13.4.2. OIL FLOW RATES THROUGH THE LEAK
BETWEEN PLUNGER AND CYLINDER ... 13-7

13.5. CONTACK OF THE CHECK BALL .. 13-10

13.6. BELT DRIVE SYSTEM.. 13-12




14. DYNAMIC ANALYSIS OF CONTACTING SPUR

GEAR PAIR FOR FAST SYSTEM SIMULATION .. 14-1


14.2. TOOTH PROFILE OF SPUR GEAR .. 14-2

14.3. EFFICIENT CONTACK SEARCH ALGORITHM AND
CONTACK FORCE MODEL . 14-4

14.3.1. ARC-ARC CONTACT . 14-4
14.3.2. ARC-POINT CONTACT .. 14-8

14.4. KINEMATICS AND EQUATION OF MOTION FOR SYSTEM DYNAMICS
USING THE RECURSIVE FORMULAS .. 14-9


RecrDyn




1

GENERALIZED RECURSIVE FORMULATION

1.1. INTRODUCTION

In Ref. 1, the equations of motion for the constrained mechanical systems
were derived with respect to Cartesian coordinates. Then the equations were
transformed into the corresponding ones that employ the relative coordinates by
using the velocity transformation method. Since the virtual displacement and
acceleration of the entire system were simultaneously substituted into the
variational form of the equations of motion, the resulting equations of motion
were compact. In spite of the compactness, they are not computationally efficient
since the recursive nature of the relative kinematics was not exploited.
In Ref. 2, Hooker proposed a recursive formulation for the dynamic analysis
of a satellite which has a tree topology. It was shown that the computational
cost of the formulation increases only linearly with respect to the number of
bodies. In Ref. 3, Featherstone proposed a recursive formulation to calculate the
acceleration of robot arms using screw notation. These ideas were extended by
using the variational vector calculus for constrained mechanical systems in Ref.
4.
Constrained mechanical systems are represented by differential equations of
motion and algebraic constraint equations, which are often called differential
algebraic equations (DAE). Several DAE solution methods using the BDF have
been proposed in Refs. 5-7. In particular, the parameterization method treated
the DAE as an ordinary differential equation (ODE) on the kinematic constraint
manifolds of the system. The stability and convergence of the method were
proved in Ref. 8. The present research employs this method, due to its
mathematical soundness.
In Ref. 9, a recursive formulation was presented to obtain the Jacobian in the
linearization of the equations of motion. Recursive formulas for each term in
the equations of motion were directly derived, using the state vector notation.

RecurDyn

Similar approach was taken in Ref. 8 to implement the implicit BDF integration
with the relative coordinates. Since the recursive formulas were derived term
by term, the resulting equations and algorithm became much complicated.
To avoid the complication involved in Ref. 8, the equations of motion are
derived in a compact matrix form by using the velocity transformation method in
the present study. Computational structure of the equations of motion in the joint
space is carefully examined to classify all computational operations that can be
done in a recursive way into several categories. The generalized recursive
formula for each category of the computational operations is then developed and
applied whenever such a category is encountered. Many common fact ors,
which are not easily observed when they are derived term by term, can be
observed among terms in the equations of motion. Furthermore, the matrix form
of the equations makes it easy to debug and understand the program while
computational efficiency is achieved by the recursive computational operation. A
library of the generalized recursive formulas is developed to implement a
dynamic analysis algorithm using the backward difference formula (BDF) and
the relative generalized coordinate.
Section 2 introduces relative coordinate kinematics. Generalization of velocity
and force recursive formulas is treated in Sections 3 and 4, respectively. Also,
computational equivalence between the recursive method and velocity
transformation method for a mechanical system is shown in Section 3. Section 5
presents a graph representation of mechanical systems. Section 6 presents the
equations of motion and a solution method for the DAE. A library of the
generalized recursive formulas are developed and applied in Sections 7 and 8.
Numerical examples are given in Section 9. Conclusions are drawn in Section 10.

1.2. RELATIVE COORDINATE KINEMATICS

1.2.1 COORDINATE SYSTEMS

Orientation of a body in Fig. 2.1 is given as

1-3

| h g f A =
(
(
(
=
33 32 31
23 22 21
13 12 11
a a a
a a a
a a a
| (2-1)

where , , and h are unit vectors along the f g x , y , and axes,
respectively. The
z
z y x frame is the body reference frame and the
frame is the inertial reference frame. Z Y X

Z
X
Y
rp
r
p
x
y
z
s
o

Fig. 2.1 Coordinate systems and a rigid body

Velocities and virtual displacements of point in the X O Z Y frame are
defined as
(2-2a)
(
w
r&
(
r
(2-2b)

Their corresponding quantities in the z y x frame are defined as

RecurDyn

(2-3a)
(
=
w A
r A
w
r
Y
T
T
& &
(2-3b)
(
T
T
A
r A r

1.2.2 RELATIVE KINEMATICS FOR A PAIR OF CONTIGUOUS BODIES

A pair of contiguous bodies is shown in Fig. 2.2. Body 1) (i is assumed
to be an inboard body of body and the position of point is i
i
O

1) i(i 1)i (i 1)i (i 1) (i i
+ + = s d s r r (2-4)

The angular velocity of body in its local reference frame, using Eq. 2-3a and
defining , is
i
i
T
1) (i 1)i (i
A A A

=

1)i (i 1)i (i
T
1)i (i 1) - (i
T
1)i (i i
+ = q H A w A w & (2-5)

where is determined by the axis of rotation. H

zi
X
Z
Y
r
i-1
r
i
s(i-1)i
si(i-1)
yi-1
x
i-1
zi-1
zi-1
x
i-1
x
i
y
i
x
i
z
i
y
i
yi-1
d(i-1)i
o
i o
i-1

Fig. 2.2 Kinematic relationship between two adjacent rigid bodies

1-5

Differentiation of Eq. 2-4, using Eq. 2-3a, yields

1)i (i
'
1)i (i 1) (i
'
i
'
1) i(i i
'
1) (i
'
1)i (i 1) (i
'
1) (i
'
1)i (i 1) (i
'
1) (i 1) (i
'
i i
1)i (i
) (
~
~
~

+
+
=
q d A
s A d A
s A r A r A
q
&
& &

(2-6)

where symbols with tildes denote skew symmetric matrices comprised of their
vector elements that implement the vector product operation (Ref. 1) and
denotes the relative coordinate vector. Substituting of Eq. 2-5 and
multiplying both sides of Eq. 2-6 by yields
1)i (i
q
'
i
T
i
A

1)i (i 1)i (i
1)i (i
T
1) i(i 1)i (i 1)i (i
1)i (i
T
1) (i
1)i (i
T
1) - i(i 1)i (i 1)i (i 1)i (i
1)i (i
T
1) (i
1)i (i
T
i
)
~
) ( (
)
~
~
~
(
1)i (i

+ +
+
=
q H A s A d A
A s A d s A
r A r
q
&
& &
(2-7)

where
i i i
~
= A A
&
is used. Combining Eqs. 2-5 and 2-7 yields the recursive
velocity equation for a pair of contiguous bodies.

1)i (i 1)i2 (i 1) (i 1)i1 (i i
+ = q B Y B Y
&
(2-8)
where
(
+
(
I 0
A s A d s I
A 0
0 A
B
)
~
~
~
( 1)i (i
T
1) i(i 1)i (i 1)i (i 1)i (i
1)i (i
T
1)i (i
T
1)i1 (i

(
(
+
(

1)i (i
1)i (i
1)i (i
T
1) i(i 1)i (i 1)i (i
1)i (i
T
1)i (i
T
1)i2 (i
~
) (
1)i (i
H
H A s A d
A 0
0 A
B
q
(2-9)

It is important to note that matrices and are functions of only
relative coordinates of the joint between bodies
1)i1 (i
B
1)i2 (i
B
(i 1) and . As a
consequence, further differentiation of the matrices and in Eq.
i
1)i2 1)i1 (i
B
(i
B

RecurDyn

2-9 with respect to other than yields zero. This property plays a key role
in simplifying recursive formulas in Section 7.
1)i (i
q
1)i1 (i

2
Y
1)m1
Y
1) - (m
Similarly, the recursive virtual displacement relationship is obtained as
follows.
(2-10)
1)i (i 1)i2 (i 1) (i i

+ = q B Z B Z

1.3. GENERALIZATION OF THE VELOCITY RECURSIVE FORMULA

1
n-1
n
0

Fig. 3.1 A serial chain mechanism

Before proceeding to generalize the recursive velocity formula, the
computational equivalence between the recursive method and the velocity
transformation method is demonstrated using the mechanical system shown in
Fig. 3.1. The Cartesian velocity is obtained by replacing by in
Eq. 2-8.
m
i m

1)m (m 1)m2 (m 1) (m (m m
+ = q B B Y & (3-1)

Substitutions of Eq. 2.8 for , , . . . , and yield Y
2) - (m
Y
0
Y

1-7

1)m - (m 1)m2 - (m
1 - m
1 j
1)j - (j 1)j2 - (j
j - m
1 k
1)1 - m m)(-k k (
m
1 k
0 1)1 - m m)(-k k ( m
q B
q B B
Y B
&
&
+
)
`
|
|
.
|
\
|
+
=

= =
+ +
=
+ +
Y
(3-2)

Thus, the Cartesian velocity for all bodies is obtained as Y

q B Y & = (3-3)

where is the collection of coefficients of and B
1)i (i
q&

| |
T
1 nc
T T
2
T
1
T
0
=
n
Y , , Y , Y , Y Y K (3-4)
| |
T
1 nr
T
) 1 (
T
12
T
01
T
0
=
n n
q , , q , q , Y q & K & & & (3-5)

where and nc nr denote the numbers of the Cartesian and relative coordinates,
respectively.
The Cartesian velocity , with a given , can be evaluated either
by using Eq. 3-3 or by using Eq. 2-8 with recursive numerical substitution of .
Since both formulas give an identical result, and recursive numerical substitution
is proven to be more efficient in Ref. 4, matrix multiplication with a given
will be evaluated by using Eq. 2-8.
nc
R Y
nr
R q&
i
Y
q B&
q&
Since in Eq. 3-3 is an arbitrary vector in , Eqs. 2-8 and 3-3, which are
computationally equivalent, are actually valid for any vector such that
q&
nr
R
nr
R x

x B X & = (3-6)

and
1)i - (i 1)i2 - (i 1) - (i 1)i1 - (i i
x B X B X + = (3-7)

where is the resulting vector of multiplication of and . As a
nc
R X B x

RecurDyn

result, transformation of into is calculated by recursively
applying Eq. 3.7 to achieve computational efficiency.
nr
R x
W
nc
R Bx
G
nc
R Q
Q Z
* T
Q q =
+
*
) ! i(i
Q
( {
+
+
1 i 1)2
Q
(
T
q =
=
=
1 - n
0 i

=
+
1 - n
0 i
T
i(i
q
T
=0

1.4. GENERALIZATION OF THE FORCE RECURSIVE FORMULA

It is often necessary to transform a vector in into a new vector
in . Such a transformation can be found in generalized force
computation in the joint space with a known force in the Cartesian space. The
virtual work done by a Cartesian force is
nc
R
G B g
T
=
nr
R

= (4-1)

where must be kinematically admissible for all joints in Fig. 3.1.
Substitution of
Z
q B Z = into Eq. 4-1 yields

T
Q B W (4-2)

where . Equation 4-2 can be written in a summation form as Q B Q
T *

+
T
1) i(i
q W (4-3)

On the other hand, the symbolic substitution of the recursive virtual
displacement relationship Eq. 2-10 into Eq. 4-1, along the chain in Fig. 3.1
starting from the body n toward inboard bodies, yields

)}
)
+ +
=
1 i
T
i(i 1)
S B W (4-4)

where
2 i 2 i
2)1 1)(i (i
1 i
0
+ +
+ +
+
+ S Q B S
S
(4-5)

1-9

Equating the right sides of Eqs. 4-3 and 4-4, the following recursive formula for
is obtained:
*

Q

( ) 0 ...., 1, - n i ,
1 i 1 i
)2 1 i(i
T
1) i(i
*
= +
+ +
+ + S Q B Q (4-6)

where is defined in Eq. 4-5.
1 i+
S
Since is an arbitrary vector in , Eqs. 4-5 and 4-6 are valid for any
vector in . As a result, the matrix multiplication of is
evaluated to achieve computational efficiency by
Q
G
nc
R
nc
R G B
T

( )
( ) 0 ...., 1, i
1 i 1 i
)1 1 i(i
T
i
n
1 i 1 i
T
1)2 i(i 1) i(i
= +
=
+ =
+ +
+
+ + + +
S Q B S
0 S
S G B g
(4-7)

where is the result of . g G B
T

1.5. GRAPH REPRESENTATIONS OF MECHANICAL SYSTEMS

In the previous section, a serial chain mechanism is considered to derive
recursive formulas for and B where is a vector in and G
in . In general, a mechanical system may have various topological
structures. To cope with the various topological structures, an automatic
preprocessing is required for a general purposed program, which employs a
relative coordinate formulation. The preprocessing identifies the topological
structure of a constrained mechanical system to achieve computational efficiency.
A graph theory was used to represent bodies and joints for mechanical systems
in Refs. 1 and 4. A node and an edge in a graph represented a body and a joint,
respectively. The preprocessing based on the graph theory yielded the path and
distance matrices that are provided to automatically decide execution sequences
Bx G
T
x
nr
R
nc
R

RecurDyn

for a general purposed program. As an example, a governor mechanism and its
graph representation are shown in Figs. 5.1 and 5.2.

4
3
2
8 7
6
5
1
U1
R2 R3
U2
S1 S2
T2
R1
T1
: Cut joint

Fig. 5.1 Governor mechanism

7
6
4
5
3
8
2
1
R1
T1
R3
U2
R2
U1
T2
Cut Joint
Cut Joint

Fig. 5.2 Graph representation of the governor mechanism

1.6. EQUATIONS OF MOTION AND DAE SOLUTION METHOD

The variational form of the Newton-Euler equations of motion for a
constrained mechanism is

1-11

0 ) Q Y M Z

T
= +
&
( (6-1)

where Z must be kinematically admissible for all joints except cut joints [1].
In the equation, and , respectively, denote the cut joint constraint and the
corresponding Lagrange multiplier. The mass matrix and the force vector
are defined as
M
Q

(
nbd 2 1
, , , diag M M M M L = ) (6-2)
(
=
i
i
i
J 0
0 I m
M (6-3)
| |
T
nbd
T
3
T
2
T
1
, , , , Q Q Q Q Q L = (6-4)
~
(

=
J n
r m f
Q
~
i
&
(6-5)

where denotes the number of bodies, denotes the identity matrix,
denotes the moment of inertia, denotes the external force, and denotes
the external torque. Substituting the virtual displacement relationship into Eq.
6-1 yields
nbd I J
f n

{ } 0 ) Q Y M B q

T
= +
&
(
T
(6-6)

Since q is arbitrary, the following equations of motion are obtained:

0 ) Q Y M B F

T
= + =
&
( (6-7)

The equations of motion, the constraint equations, q v = & , and
constitute the following differential algebraic equations[8]:
a v = &

RecurDyn

0
a v
v q
a v q
v q
q
) a , v q ( F
=
(
(
(
(
(
(
(
(
) (
) (
) (
&
&
& &
&
t , , ,
t , ,
t ,
t , , ,
(6-8)

Application of 'tangent space method' in Ref. 7 to Eq. 6-8 yields the following
nonlinear system that must be solved at each time step:

0
a v U
v q U
a v q
v q
q
) a , v q ( F
p H =
(
(
(
(
(
(
(
(
+ +
+ +
) (
) (
) (
=
) (
) (
t , , ,
t , ,
t ,
t , , ,
) (
2 n 0 n
T
0
1 n 0 n
T
0
n n n
n n n
n n
n n n n n
n
n
& &
&
(6-9)

where | |
T T
n
T
n
T
n
T
n n
, a , v , q p =
0
U
, , , and are determined by the
coefficients of the BDF, and is an
0
nr
2
ncut) - (nr such that the augmented

square matrix is nonsingular.
(
(
U
T
0
Applying the Newton's method to solve the nonlinear system in Eq. 6-9 yields

H p H
p
= (6-10)
1,2,3,... i ,
i
n
1 i
n
= + =
+
p p p (6-11)

where

1-13

(
(
(
(
(
(
(
(
=
0 U U 0
0 0 U U
0
0 0
0 0 0
F F F F
H
T
0
a v q
v q
q
q q q q
p
T
0 0
T
0 0
T
0
& & & & & &

& &
(6-12)

Since and are highly nonlinear functions of , , , and F q v a , care
must be taken in deriving the non-zero expressions in , so that they can be
efficiently evaluated.
p
H

1.7. GENERALIZED RECURSIVE FORMULAS

Inspection of the residual and Jacobian matrix shows that types of
necessary recursive formulas are classified into Bx , , , ,
H
p
H
G B
T
x B
&
( )
q
Bx
( )
q
G B
T
, ( )
q
x B
&
, and ( )
v
x B
&
, where into G are arbitrary constant
vectors, and are relative coordinates. Formulas and were
derived in Sections 3 and 4, and the formulas for the rest will be derived in this
section. All recursive formulas are tabulated in Appendix A. Note that the
recursive formulas are quite simple. This simplicity is achieved by exploiting
the relative kinematics in the local reference frame instead of the global
reference frame.
nr
R x
nc
R
Bx
q G B
T
To derive the formulas systematically, bodies in a graph are divided into four
disjoint sets (associated with a generalized coordinate ) as follows:
k
q
( )
k
q q I having joint the of body outboard body outborad {
k
=
} coordinate d generalize its

as
( ) ( ) { }
k k
of bodies outboard all q I q II =

( )
( )
)
`
=
body inboard and base the including
, of body inboard the and body base e between th bodies all
k
k
q I
q III

RecurDyn

( ) ( ) ( ) ( ) { }
k k k k
of set ary complement the q III q II q I q IV U U =

For example, the body sets associated with q (relative coordinate between
bodies 2 and 4) for the graph shown in Fig. 5.2 are obtained as follows:
24

( ) { 4 Body
24
= q I }
}
}
}

( ) { 7 and 6 Bodies
24
= q II

( ) { 2 and 1 Bodies
24
= q III

( ) { 8 and 5 3, Bodies
24
= q IV

1.7.1 RECURSIVE FORMULA FOR x B X
& &
=

Recursive formula for is easily obtained by differentiating Eq. 3-7.
nc
R x B
&

1)i (i 1)i2 (i 1 i 1)i1 (i 1 i 1)i1 (i i
+ + = x B X B X B X
& & & &
(7-1)

This recursive formula can be applied to compute the Cartesian acceleration
with known relative velocity and acceleration.
Y
&

1.7.2 RECURSIVE FORMULA FOR RM BOLD
q q
Bx X ) ( =

To obtain the recursive formula for , Eq. 3-7 is partially differentiated
with respect to .
q
Bx) (
nr ..., 1, k ,
k
= q

1)i (i 1)2 (i 1 - i 1)i1 - (i 1 i 1)i1 (i i
k k k
) ( ) ( ) ( ) (

+ + = x B X B X B X
q q q q
k
(7-2)

Since matrices and depend only on the relative coordinates for
joint , their partial derivatives with respect to generalized coordinates
other than are zero. In other words, the partial derivatives are zero if
1)i1 (i
B
1)i1
1)i2 (i
B
1)i - (i
q
(i k
q

1-15

does not belong to set . Therefore if body i is an element of set ,
Eq. 7-2 becomes
( )
k
q I II( )
k
q
(q III
( )
k
q
i
k
) = X
q
24
T
G) (B

k k
) ( ) (
1 i 1)i1 (i i q q
X B X

= (7-3)

If body i belongs to set ) ( )
k k
q IV U , is not affected by . As a
result, Eq. 7-3 is further simplified as follows
i
X
k
q

0 X
q
=
k
) (
i
(7-4)

If body i is an element of set , body is naturally its inboard body
and it belongs to set . Using Eq. 7-4, Eq. 7-2 becomes
(
k
q I ) 1) - (i
III

1)i (i 1)2 (i 1 i 1)i1 (i
k k
) ( ) ( (

+ x B X B
q q
(7-5)

This recursive formula can be applied to compute the partial derivative of the
Cartesian velocity with respect to relative coordinates Y . For example, if
in Fig. 5.2, is shown in Fig. 7.1.
q
24 k
q q =
q
Y

q
T
q
G) (B g =

Recursive formula for is obtained by using the recursive formula in
Eq. 4-7. By replacing i by , Eq. 4-7 can be rewritten as
q
- (i 1)

) (
) (
i i
T
1)i1 - (i 1 - i
i i
T
1)i2 - (i 1)i - (i
S G B S
S G B g
+ =
+ =
(7-6)

Taking partial derivative of Eq. 7-6 with respect to yields
k
q

RecurDyn

(Y1)q24
= 0
(B671)q24 = 0,
(B672)q24 = 0
(B461)q24 = 0,
(B462)q24 = 0
(B241)q24 ,
(B242)q24
(B121)q24 = 0,
(B122)q24 = 0
(B231)q24 = 0,
(B232)q24 = 0
(B351)q24 = 0,
(B352)q24 = 0
(B281)q24 = 0,
(B282)q24 = 0
(Y2)q24
= 0
(Y4)q24
=(B241)q24Y2
+(B242)q24q24
(Y3)q24
= 0
(Y8)q24
= 0
(Y5)q24
= 0

.
(Y6)q24
=
B461 (Y4)q24
(Y7)q24
=
B671 (Y6)q24

Fig. 7.1 Computation Sequence for
24
q
Y

k k k
k k k
) ( ) ( ) ( ) ( ) (
) ( ) ( ) ( ) ( ) (
i i
T
1)i1 (i i i
T
1)i1 (i 1 i
i i
T
1)i2 (i i i
T
1)i2 (i 1)i (i
q q q
q q q
S G B S G B S
S G B S G B g
+ + + =
+ + + =

(7-7)

Since is a constant vector, . If .
nc
R G 0 G
q
=
k
( ) ( ) (
k k k
q IV q III q II U U ) i ,
and are not functions of . Therefore their partial derivatives
with respect to are zero. As a result, Eq. 7-7 can be simplified to
1)i1 (i
B
1)i2
k
q
(i
B
k
q

k k
k k
) ( ) ( ) (
) ( ) ( ) (
i
T
1)i1 (i 1 i
i
T
1)i2 (i 1)i (i
q q
q q
S B S
S B g

=
=
(7-8)

Since for the tree end bodies, by the second equation 0 S
q
=
k
) (
i
0 S
q
=
k
) (
1 - i

1-17

of Eq. 7-8 for ( ) ( )
k k
q IV q II U i . Thus, for ( ) ( )
k k
q IV q II U i
)
, Eq. 7-8
becomes
g
q
k
) (
1)i (i
1) (i + 0 S
q
=
k
) (
i
( ) (
( ) (
T
1)i1 (i 1 i
T
(i 1)i (i
k
k
B S
B g
q
q

=
=
24 k
q =
= 0
= 0
4
6
3
5
(g35)q24 = 0
(S3)q24 = 0
(g23
(S2
)q24 = 0
)q24 = 0
(g12)q24 = (B122)( S
(S12)q24 =(B121)(S
2) q24
2) q24
24
24 q
q g G=
0 = (7-9)

If , body (
k
i q I belongs to set ( )
k
q II , and . Thus, Eq. 7-7
becomes
) ( )
) ( )
i
i 1)i2
k
k
S
S
q
q
(7-10)

For example, if in Fig. 5.2, is shown in Fig. 7.2. q G B q
24
T

1
(g67)q24
(S6)q24
2
8

(g28)q24 = 0
(S2)q24 = 0
(g24)q24 = (B242)q24 S4
(S2)q24 =(B241)q24 S4
(g46)q24 = 0
(S4)q24 = 0
7
Fig. 7.2 Computation Sequence for
T
B

RecurDyn

1.7.4 RECURSIVE FORMULAS FOR AND
q q
x) B ( X
& &
=
v v
x B X ) (
& &
=

To obtain the recursive formula for and ( , Eq. 7-1 is partially
differentiated with respect to and for
q
x) B (
&
k
v
x) B
&
nr ...,
k
q v 1, k = .

1)i (i 1)i2 (i 1 i 1)i1 (i 1 i 1)i1 (i
1 i 1)i1 (i 1 i 1)i1 (i i
) ( ) ( ) (
) ( ) ( ) (

+ + +
+ =
x B X B X B
X B X B X
q q q
q q q
k k k
k k k
& & &
& & &
(7-11)
1)i (i 1)i2 (i 1 i 1)i2 (i
1 i 1)i1 (i 1 i 1)i1 (i i
) ( ) (
) ( ) ( ) (

+ +
+ =
x B X B
X B X B X
v v
v v v
k k
k k k
& &
& & &
(7-12)

The recursive formulas for and are obtained as in Appendix A
by following the similar steps taken in the previous sections.
q
x) B (
&
v
x) B (
&

1.8. APPLICATION OF THE GENERALIZED RECURSIVE FORMULAS

A library of the generalized recursive formulas is developed in Section 7. This
section shows how the library can be utilized to compute the terms in and
in Eqs. 6.9 and 6.12. Inspection of and reveals that the residual
and partial derivatives of F , , , , , and need to be
computed. Only and are presented in this section and the rest are
omitted for simplicity of the presentation.
H
p
H
F
H
a
F
p
H
&
q v
F
q
q q
& &
F
q
F

1.8.1 COMPUTATION OF THE RESIDUAL F

The generalized force , and the Cartesian acceleration need to be
computed to obtain shown in Eq. 6-7. The term is obtained by applying
the recursive formula in Eq. 7.1. The recursive formula with
in Eq 4.7 can be applied to evaluate in since
is a vector in
Q
Z
T
Y
&
F
)
nc
R .
Y
&
G B
T
nr
R (
T
Q Y M G
Z
+ =
&
G
F

1-19

1.8.2 COMPUTATION OF THE JACOBIAN
q
F
In Eq. 6-7, differentiation of matrix with respect to vector results in a
three dimensional array. To avoid the complexity, Eq. 6-7 is differentiated with
respect to a typical generalized coordinate . Thus,
B q
k
q

nr ....., 2, 1, k , ) (
) (
k
k k
T T
T T
= + +
+ =
q Z
Z q q
Q Y M B
Q Y M B F
&
&
(8-1)

Since the term can be easily expressed in terms of the Cartesian
coordinates, is obtained by applying the chain rule, as
) (
T
Q
Z

k
)
T
q Z
Q (

k
T T
) ( ) (
k
B Q Q
Z Z q Z
= (8-2)

where B
q
Z
=
is used and denotes the kth column of the matrix . The

resulting equation for becomes
k
B B
k
q
F

nr ....., 2, 1, k ), ) ( (
) (
k
T T
T T
k
k k
= + +
+ =
B Q Y M B
Q Y M B F
Z Z q
Z q q
&
&
(8-3)

The first term in Eq. 8-3 can be obtained by applying the recursive formula for
, with , as explained in section 7.3. Collection of
, for all k, constitutes ( , which is equivalent to
. Matrix consists of columns which are
vectors in . Therefore, the application of , where is each column
of matrix , yields the numerical result of . Finally,
the second term in Eq. 8-3 is also obtained by applying , where
.
G B q
k
T
T
(
Z

T T
((
Z
B
( M G=
) (
T
Q Y M G
Z
+ =
&
k
T
T
) )
Z
T
(
Z

T
)
Z
Q
) )
k
T
B Q
Z Z

) B Q
Z
Q
nc
R
T
(
Z
(
k
Y
q
+
&
B Q
Z Z
)
T
T
)
Z
G B
T
Q nc
T
( B
G
T
Z
T
)
Z
Q
G B
T

RecurDyn

1.9. NUMERICAL EXAMPLES

1.9.1 A GOVERNOR MECHANISM

The mechanism shown in Fig. 9.1 consists of seven bodies, a spring-damper,
five revolute joints, and a translational joint. The material properties and spring
and damping constants of the system are shown in Table 9-1. The mechanism
has redundant constraints that are removed by the Gaussian elimination with full
pivoting. Consequently, it has only 2 degrees of freedom.
Dynamic analysis is carried out for 2 seconds with error tolerance of
for the system. The Z acceleration of body 4 is drawn in Fig. 9.2. The result
obtained by the other commercial program and that obtained by the proposed
method are almost identical. The average step size, the numbers of residual
function evaluations, and CPU time on SGI R3000 are shown in Table 9-2. The
CPU time spent by the other commercial program is about 6 times larger than
that by the proposed method. Note that the number of function evaluations of the
proposed method is smaller than that of the other commercial program.
5
10 3

4
3
2
5
6
1
R4
R2
R3
R5
S1 S2
R1
T1
7
Y
X
Z
0
.
1
6
0.5
0.2
0
.
1
0
9
45
o

Fig. 9.1 A governor mechanism

1-21

Table 9-1 Inertia properties of the governor mechanism and spring and damping
constants
Mass
x
I
y
I
z
I
xy
I
yz
I
zx
I
Body 1 (Ground) not necessary
Body 2 200.0 25.0 50.0 25.0 0.0 0.0 0.0
Body 3 1.0 0.1 0.1 0.1 0.1 0.1 0.1
Body 4 1.0 0.1 0.1 0.1 0.1 0.1 0.1
Body 5 1.0 0.15 0.125 0.15 0.0 0.0 0.0
Body 6 0.1 0.1 0.1 0.1 0.0 0.0 0.0
Body 7 0.1 0.1 0.1 0.1 0.0 0.0 0.0
Spring constant 1000
Damping constant 30

Table 9-2 Integration output information
Program TOL Average step size No. fevals CPU time (sec)
Other
5
10 3

2
10 1 . 1

748 41
Proposed
5
10 3

2
10 2 . 1

441 7

PROPOSED
OTHER
Fig. 9.2 Z acceleration of Body 4

RecurDyn

1.9.2 A MULTI-WHEELED VEHICLE

A vehicle example shown in Fig. 9.3 is chosen to show the practicality of the
proposed method. The vehicle runs on a bump whose radius is 0.3048(m).
The system consists of a chassis and twelve road wheels and arms. The
material properties and spring and damping constants are shown in Table 9-3.
The road wheel and arm are considered as a single body. As a result, the
system has 18 degrees of freedom.

11.1 m/sec
0.3048m

Fig. 9.3 A multi-wheeled vehicle

Figure 9.4 shows the vertical acceleration of the chassis. It is shown that the
proposed method and the other commercial program yield almost identical
results. The average stepsize, number of residual function evaluations, and
CPU time on SGI R3000 are shown in Table 9-4. It can be shown that the
proposed method performs much smaller number of residual function
evaluations with larger stepsizes, and the CPU time by the proposed method is
much shorter than that by the other commercial program. Since there is no closed
chain in the system, the governing equations of motion are formulated as an
ODE problem by the proposed method. On the other hand, the equations of
motion by the other commercial program are formulated as an DAE problem.
The DAE problem is generally more difficult to solve than the ODE problem.
This general argument is supported by the numbers of function evaluation and
average stepsize.

1-23

Table 9-3 Inertial properties of the vehicle mechanism and spring and damping constants

Mass
x
I
y
I
z
I
xy
I
yz
I
zx
I
Body 1 (Ground) not necessary
Body 2 40773. 231800 60840 251700 - 234. -
Body 3
~ Body 14 340.27 32.86 20.76 26.85 0.0 0.0 0.0
Spring constant
200000
Damping constant
40000

Table 9-4 Integration output information
Program TOL. Average step size No. fevals CPU time (sec)
Other
4
10

3
10 4

1359 330
Proposed
4
10

3
10 6 . 6

1167 69

OTHER
PROPOSED
Fig. 9.4 Vertical acceleration of the chassis

RecurDyn

1.10. CONCLUSIONS

The recursive formulas are generalized in this research. The velocity
transformation method is employed to transform the equations of motion from
the Cartesian to the joint spaces. Computational structure of the equations of
motion is examined to classify all necessary computational operations into
several categories. The generalized recursive formula for each category is then
applied whenever such a category of computation is encountered. Since the
velocity transformation method yields the equations of motion in a compact form
and computational efficiency is achieved by the generalized recursive formulas,
the proposed method is not only easy to implement but also efficient. A dynamic
analysis algorithm using the backward difference formula (BDF) and the relative
generalized coordinate is implemented using the library of generalized recursive
formulas developed in this research. Numerical studies showed that obtained
solutions were numerically stable and computation time was reduced by an order
of magnitude compared to a well-known commercial program.

1-25

REFERENCES

1. J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G. Teubner, Stuttgart, 1977.
2. Hooker, W., and Margulies, G., The Dynamical Attitude Equtation for an n-body
Satellite, Journal of the Astrnautical Science, Vol. 12, pp. 123-128, 1965.
3. R. Featherstone, The Calculation of Robot Dynamics Using Articulated-Body Inertias, Int.
J. Roboics Res., Vol 2 : 13-30, 1983.
4. D. S. Bae and Edward J. Haug, A Recursive Formulation for Constrained Mechanical
System Dynamics: Part II. Closed Loop Systems, Mech. Struct. and Machines, Vol. 15,
No. 4, pp. 481-506
5. Potra, F. A. and Petzold, L. R., ODAE Methods for the Numerical Solution of Euler-
Lagrange Equations. Applied Nume. Math., Vol. 10, pp. 397-413, 1992
6. Potra, F. A. and Rheinboldt, W. C., 1989, On the Numerical Solution of Euler-Lagrange
Equations, NATO Advanced Research Workshop on Real-Time Integration Methods for
Mechanical System Simulation, Snowbird, Utah, U. S. A..
7. Jeng Yen, Edward J. Haug, and Florian A. Potra, 1990, Numerical Method for
Constrained Equations of Motion in Mechanical Systems Dynamics, Technical Report R-
92, Center for Simulation and Design Optimization, Department of Mechanical
Engineering, and Department of Mathematics, The University of Iowa, Iowa City, Iowa.
8. Ming-Gong Lee and Edward J. Haug, 1992, Stability and Convergence for Difference
Approximations of Differential-Algebraic Equations of Mechanical System Dynamics,
Technical Report R-157, Center for Simulation and Design Optimization, Department of
Mechanical Engineering, and Department of Mathematics, The University of Iowa, Iowa
City, Iowa.
9. Lin, T. C. and Yae, K. H., 1990, Recursive Linearization of Multibody Dynamics and
Application to Control Design, Technical Report R-75, Center for Simulation and
Design Optimization, Department of Mechanical Engineering, and Department of
Mathematics, The University of Iowa, Iowa City, Iowa.

RecurDyn

APPENDIX A : RECURSIVE FORMULAS

Recursive
formulas
) ( i
k
q I ) ( i
k
q II
q q
Bx X ) ( =
1)i (i 1)2 (i
1 i 1)i1 (i i
k
k k
) (
) ( ) (

+
=
x B
X B X
q
q q

k k
1 i 1)i1 (i i q q
) (X B ) (X

=
q
T
q
G) (B g =
i
T
i i i
i
T
i i i i
k
k
( S B S
S B g
q q
q q
) ( )
) ( ) (
1 ) 1 ( 1
2 ) 1 ( ) 1 (
k
k

=
=

0 S
0 g
q
q
=
=
k
k
) (
) (
1 i
1)i (i

q q
x) B ( X
& &
=
1)i (i 1)i2 (i
1 i 1)i1 (i
1 i 1)i1 (i i
k
k
k k
) (
) (
) ( ) (

+
+
=
x B
X B
X B X
q
q
q q
&
&
& &

k
k k
) (
) ( ) (
1 i 1)i1 (i
1 i 1)i1 (i i
q
q q
X B
X B X

+
=
&
& &

v v
x B X ) (
& &
=
1)i (i 1)i2 (i
1 i 1)i1 (i i
k
k k
) (
) ( ) (

+
=
x B
X B X
v
v v
&
& &

k
k k
) (
) ( ) (
1 i 1)i2 (i
1 i 1)i1 (i i
v
v v
X B
X B X

+
=
&
& &

Recursive
formulas
) ( i
k
q III ) ( i
k
q IV
q q
Bx X ) ( = 0 X
q
=
k
) (
i
0 X
q
=
k
) (
i

q q
G B g ) (
T
=
k k
k k
) ( ) (
) ( ) (
i
T
1)i1 (i 1 i
i
T
1)i2 (i 1)i (i
q q
q q
S B S
S B g

=
=

0 S
0 g
q
q
=
=
k
k
) (
) (
1 i
1)i (i

q q
x B X ) (
& &
= 0 X
q
=
k
) (
i
&
0 X
q
=
k
) (
i
&

v v
x B X ) (
& &
= 0 X
v
=
k
) (
i
&
0 X
v
=
k
) (
i
&

Recursive
formulas
) ( i
k
q I or i or or ) (
k
q II ) ( i
k
q III ) ( i
k
q IV
Bx X = 1)i (i 1)i2 (i 1 i 1)i1 (i i
+ = x B X B X
G B g
T
=
) (
0
) (
1 i 1 i
T
1 ) 1 i ( i i
n
1 i 1 i
T
2 ) 1 i ( i ) 1 i ( i
+ + +
+ + + +
+ =
=
+ =
S G B S
S
S G B g

x B X
& &
= i ) 1 i ( 2 i ) 1 i ( 1 i 1 i ) 1 i ( 1 i 1 i ) 1 i ( i
+ + = x B X B X B X
& & & &

1-27

RecurDyn


2
DECOUPLING SOLUTION METHOD FOR
IMPLICIT NUMERICAL INTEGRATION

2.1. INTRODUCTION

The dynamic behavior of a constrained mechanical system is often represented
by differential algebraic equations (DAEs)[1]. Solutions of DAEs are generally
more difficult to obtain than those of ordinary differential equations (ODEs)[2].
To solve DAEs, a direct discretization method was proposed by Gear[3]. Since
the solution obtained by Gear does not satisfy the velocity level constraints,
consistent initial conditions cannot be obtained. It was found that the
inconsistency often resulted in a poor local error estimation[4]. A series of
stabilization methods[5-7] which employ either Lagrange multipliers or
constraint violation penalty terms were followed.
Recently several solution methods[8], projecting the differential equations on
the inflated constraint manifolds, have appeared. Two kinds of solution process
are available. In the first solution process, the numerical integration is carried
out first and the integrated variables are corrected so that the position level
constraints, the velocity level constraints, and the acceleration level constraints
are satisfied. Since the correction is made sequentially level-by-level, the size
of system equations to be solved remains small. However, the integration
stepsize can be excessively small for highly nonlinear or stiff problems due to a
narrow stability region of the explicit method. In order to overcome this
difficulty, the second solution process is developed. In the second solution
process, the numerical integration formula, kinematic constraints and their
derivatives, and equations of motion are solved simultaneously. Therefore, the
size of the system equations to be solved becomes larger although the problem of
excessive small step size is resolved. In addition to the problem of large size of
the matrix equation, the condition of the matrix becomes poor as the stepsize
gets smaller for discontinuous systems. The poor condition of the matrix often

RecurDyn

DECOUPLING SOLUTION METHOD FOR IMPLICIT NUMERICAL INTEGRATION
results in large error in the solution of the matrix equation.
In this paper, a decoupling solution method for the implicit numerical
integration method is proposed. This method is free from the problems of the
poor matrix condition and the excessively small step size as well as the large
matrix size.
In section 2, overdetermined DAEs for constrained mechanical systems are
given. A decoupling solution method is given in section 3. In section 4, the
numerical algorithm is provided. The numerical examples are given in section
5 to demonstrate the efficiency of the proposed method. Conclusions are drawn
in section 6.

2.2. IMPLICIT NUMERICAL INTEGRATION FOR DIFFERENTIAL
ALGEBRAIC EQUATIONS

The equations of motion for a constrained mechanical system can be
implicitly described as

0 q v = (2.1.a)
0 ) a, v, F(q, = (2.1.b)
0 (q) = (2.1.c)

where is the generalized coordinate vector in Euclidean space q
n
R , and
is the Lagrange multiplier vector for constraints in
m
R , represents the
position level constraint vector in
m
R , and its Jacobian is expressed
that is assumed to have full row-rank. Successive differentiations
of Eq. 2.1.c yield velocity and acceleration level constraints,
n
q

m
R

0 v v) (q,
q
= = (2.2.a)
0 a a) v, (q, = =
q

&
(2.2.b)

Equations 2.1 and 2.2 comprise a system of overdetermined differential algebraic

2-3

equations (ODAE). An algorithm based on backward differentiation formula
(BDF) to solve the ODAE is given in Ref. 1 as follows:

0
q v U
v a U
(q)
v
a
) a, v, F(q,
R U
R U
F(x)
H(x)
T
2
T
1
q
q
2
T
2
1
T
1
=
(
(
(
(
(
(
(
(
(
(
(
|
|
.
|
\
|

|
|
.
|
\
|

=
(
(
(
(
(
(
(
(
(
(
(
|
|
.
|
\
|
|
|
.
|
\
|
=
2
0
1
0
0
0
b
h
b
h
b
h
b
h
&
& &
(2.3)

where
=

=
k
1 i
1 n i
0
1
b
b
1
v ,
=

=
k
1 i
1 n i
0
2
b
b
1
q
[ x =
i
U
T
q
, k is the order of integration and

are the BDF coefficients. Here, and the columns of
constitute bases for the parameter space of the position
and velocity level constraints. are chosen so that has an inverse.
Therefore, the parameter space spanned by the columns of and the subspace
spanned by the columns of constitute the entire space
i
b
i
U
] q , v , a ,
T T T T
) 2 , 1 i (
) m n ( n
=

R
(
(
(
T
q
U
i
i
U
n
R .
The number of equations and the number of unknowns in Eq. 2.3 are the same,
so Eq. 2.3 can be solved. Newton's numerical method can be applied to obtain
the solution . x

i i i
H x H
x
= (2.4.a)
i i 1 i
x x x + =
+
(2.4.b)

LU-decomposition of the matrix not only increases the computation time
i
x
H

RecurDyn

but also produces an ill-conditioned matrix as h approaches zero [4]. In order to
eliminate these problems, Eq. 2.4.a will be divided into several pieces to obtain
, , and q v a separately in the next section.
h
a
F
a
m
R
1

v

2.3. A DECOUPLING SOLUTION METHOD FOR IMPLICIT NUMERICAL
INTEGRATION

Equation 2.4.a can be rewritten in detail as follows:

0 x F F a F v F q F
a v q
= + + + + ) ( (3.1.a)
0 x q v a
q v q
= + + + ) (
& & & & & &
(3.1.b)
0 x q
q
= + ) ( (3.1.c)
0 x q v
q q
= + + ) (
& &
(3.1.d)
0 x R v a U = + )) ( (
1
T
1
h h (3.1.e)
0 x R q v U = + )) ( (
2 2
h h
T
(3.1.f)

where
0
b
h
1
U
. Equation 3.1.e can be rewritten in an equivalent inflated form
by choosing such that . as follows [7]: 0
T
q
1
a
T
1
=
F U

0 F U x R v = + +

1
1
1
) (
T
q a
T
i
h h h (3.2)

where is a mass matrix and is generally nonsingular. The can be
singular if a parametric formulation is employed. If is singular, Eqs. 3.1
must be solved simultaneously to obtain
a
F
a
F
q , v , a and . The vector
is a new unknown variable. The a is thus obtained from Eq. 3.2 in
terms of as
1
T 1
a 1
) (
h
1
F x R v a
q
= (3.3)

2-5

Substituting Eq. 3.3 into Eq. 3.1.a yields

) ( ) ( ) (
h
1
x R F x F v
F
F q F
a 1
T
q
a
v q
+ = + |
.
|
\
|
+ + (3.4)

Equation 3.1.f can be rewritten in an equivalent inflated form by choosing
such that
2
U
0
F
F U
T
q
a
v
= |
.
|
\
|
+
1
T
2
h
as follows:

0
F
F x R q v
T
q
a
v 2
= |
.
|
\
|
+ + +

2
1
) (
h
h h h (3.5)

where
h
+
a
v
F
F is assumed to be a nonsingular matrix and is a new
unknown variable. The solution process for the case of a singular matrix will be
explained later in this section. Equation 3.5 can be solved for in terms of
as follows:
m
2
R
v
q
2
1
2
h
) (
h
1

F
F x R q v
T
q
a
v
|
.
|
\
|
= (3.6)

Substituting Eq. 3.6 into Eq. 3.4 and multiplying both sides of Eq. 3.4 by
yields
2
h
3
2
h R q K
T
q
*
= + (3.7)
where

a v q
*
F F F K + + h h
2
(3.8.a)
2 1
(3.8.b)
) ( ) h ( h ) ( h ) ( h
2 1
2 2
3
x R F F x R F x F R
a v a
+ + + (3.8.c)

RecurDyn

Equations 3.7 and 3.1.d are combined to obtain

(
=
(
(
(
) ( h 0
3
2
x
R
K
q
T
q
*
(3.9)

Equation 3.9 is then solved for and . Note that is scaled by to
make the coefficient matrix of Eq. 3.9 ill-conditioned, even as approaches to
zero.
q
2
h
h
Multiplying both sides of Eq. 3.5 by
h
+
a
v
F
F yields

) (
h h
1
h
2 2
x R q
F
F v
F
F
a
v
T
q
a
v
|
.
|
\
|
= + |
.
|
\
|
+ (3.10)

Equations 3.10 and 3.1.c are combined to obtain

(
(
(

|
.
|
\
|
+
=
(
(
(
+
q x
x R q
F
F

F
F
q
a
v
q
T
q
a
v
) (
) (
h h
1
0
h
2
2
(3.11)

where has been obtained from Eq. 3.9. Equation 3.11 is solved for the
and . Multiplying both sides of Eq. 3.3 by yields
q
2
v
a
F

|
.
|
\
|

= + v x R F a F
a a
h
1
) (
1 1
T
q
(3.12)

Equations 3.12 and 3.1.b are combined to obtain

2-7

(
(
(

|
.
|
\
|

=
(
(
(
(
q v x
v x R F
F
q v
a
q
T
q a
& & & & & &
) (
h
1
) (
0
1
1
(3.13)

Equation 3.13 is solved for a and . Once , and are obtained, the
is evaluated from Eq. 3.8.b, as follows:
1

1

2 1
+ + = (3.14)

Since is a mass matrix and is a tangent damping matrix, is
generally not ill conditioned. If an ill-conditioned case is encountered, Eqs. 3.1
must be solved simultaneously to obtain
a
F
v
F
v a
F F h +
q , v , a and . However, the
and
a
F
v
F h
a
F + are rarely singular, so q , v , a and are obtained by
using Eqs. 3.9, 3.11, 3.13, and 3.14 for most of practical problems.

2.4. NUMERICAL ALGORITHM

The DASSAL subroutine [4] is employed to integrate the system variables.
Computational flow for the proposed DAE solution method is given in Fig.
1.(Page 2-7)


2.5.1 QUICK-RETURN MECHANISM

The quick-return mechanism as shown in Fig. 5.1 is mounted on a body
translating with respect to the ground. The system consists of 6 bodies, 2
translational joints, and 5 revolute joints. The system has two degrees of freedom
if the redundant constraints are eliminated.
Dynamic analyses were performed for 1 sec with error tolerances of 10
-4
and
10
-6
by using the program developed in this paper and the other commercial
program

RecurDyn

Read initial conditions
Compute initial Accelerations and Lagrange
multipliers from Eqs. 2.1.b and 2.2.b
t = t + h
Predict q v a , , , and
t > tout ?
Y
N
End
Y
F
F
v
a
+
h
'
N
Compute in Eq. 3.9
qand
Compute in Eq. 3.11
v and
2
Compute
in Eqs. 3.1
q, v, a, and
Update q v a , , , and
Convergence?
N
Compute in Eq. 3.13
a and
1
Compute in Eq. 3.14
F
a
or
is singular ?
Y

Fig. 1 Flowchart for the proposed DAE solution method

2-9

Front view Side view
Body1
Body2
Body3
Body5
Body6
Body4

Fig. 5.1 A quick-return mechanism

, which employs the implicit numerical integration with the BDF. The results are
shown in Fig. 5.2. and the integration information is shown in Table 5.1.

OTHER
PROPOSED
Fig. 5.2 Results of the quick-return mechanism

RecurDyn

Table 5.1 Integration information for the quick return mechanism
Method
Error
Tolerance
No.
Steps
No.
Function
Evaluation
No.
Jacobian
Evaluation
No.
Newton
Iteration
Failure
No.
Integration
Failure
CPU
Time
Proposed 1.0d-4 293 342 180 15 0 16 sec
Other
1.0d-4 115 554 NA NA NA 34 sec
Proposed 1.0d-6 315 722 336 22 0 22 sec
Other 1.0d-6 Failed to integrate.
(NA means Not Available)

Note that the other commercial program failed to integrate while the proposed
method did not as the error tolerance became small (10
-6
). The integration failure
was caused by the ill-conditioned Jacobian matrix

2.5.2 AIR COMPRESSOR

This system was modeled as four bodies, two revolute joints, two translational
joints, and 2 ball joints as shown in Fig. 5.3. The system has 1 degree of
freedom if the redundant constraints are eliminated. Dynamic analyses were
carried out for 1.0 sec with initial angular velocity. The proposed method and the
other commercial program yielded identical results, as shown in Fig. 5.4. The
system is conservative and the total energy should be constant. Figure 5.5
shows the total energy change during the integration. It is shown that the total
energy obtained from the present program is numerically more stable than that
obtained from the other commercial program. Thus, the other commercial
program failed to integrate (while the proposed method did not) as the error
tolerance became small. The integration information is also given in Table 5.2.
5 0 r a d / s e c

Fig. 5.3 An air compressor mechanism

2-11

Table 5.2 Integration information for the air compressor mechanism
Method
Error
Tolerance
No.
Steps
No.
Function
Evaluation
No.
Jacobian
Evaluation
No.
Newton
Iteration
Failure
No.
Integration
Failure
CPU
Time
Proposed 1.0d-4 349 707 351 0 0 16 sec
Other 1.0d-4 295 1185 NA NA NA 31 sec
Proposed 1.0d-6 529 1067 531 0 0 20 sec
Other 1.0d-6 Failed to integrate.

OTHER
PROPOSED
Fig. 5.4 Results for the air compressor

RecurDyn


PROPOSED
OTHER
Fig. 5.5 Total energy comparison for the air compressor

2.6. CONCLUSIONS

A decoupling solution method for the implicit numerical integration is
proposed in this paper. The size of the Jacobian matrix is significantly reduced
by decoupling the iteration equations. The ill-conditioning problem of the
implicit numerical integration is resolved in this method. Numerical study
showed that the proposed method yields numerically more stable solution than
the commercial program with smaller number of function evaluation.

2-13

REFERENCES

1. J. Yen, Constrained Equations of Motion in Multibody Dynamics as ODE's on Manifolds,
SIAM J. Numer. Anal., vol. 30 , pp. 553-568, (1993).
2. P. L. stedt and L. R.. Petzold, Numerical Solution of Nonlinear Differential Equations
with Algebraic Constraints I: Convergence Results for Backward Differentiation
Formulas, Math. Comp., vol. 46, pp. 491-516, (1986).
3. C. W. Gear, The Simultaneous Numerical Solution of Differential Algebraic Equations,
IEEE Trans. Circuit Theory, vol. 18, pp. 89-95, (1971).
4. K. E. Brenan, S. L. Campbell and L. R. Petzold, Numerical Solution of Initial-Value
Problems in Differential-Algebraic Equations, SIAM Press, (1995).
5. J. Baumgarte, Stabilization of Constraints and Integrals of Motion in Dynamical
Systems, Comput. Methods Appl. Mech. Engrg., vol. 1, pp. 1-16, (1972).
6. Javier Garcia de Jalon and Eduardo Bayo, Kinematic and Dynamic Simulation of
Multibody Systems, Springer-Verlag, (1993).
7. F. A. Potra, Implementation of Linear Multistep Methods for Solving Constrained
Equations of Motion, SIAM J. Numer. Anal., vol. 30, pp. 74-789, (1993).
8. Ming-Gong Lee and Edward J. Haug, Stability and Convergence for Difference
Approximations of Differential-Algebraic Equations of Mechanical System Dynamics,
Technical Report R-157, August, (1992).

RecurDyn


3
FLEXIBLE MULTIBODY DYNAMICS USING A
VIRTUAL BODY AND JOINT

3.1. INTRODUCTION

A rigid body in space is described by the position and orientation generalized
coordinates with respect to the inertial reference frame. Contrast to
implementation of a rigid body dynamic analysis program, it is generally
complicated to implement a flexible body dynamic formulation and to expand it
for a general purpose program, regardless of whatever formulation has been
chosen. This is because the flexible body dynamic formulations handle
additional generalized coordinates to these of the rigid body dynamics. One of
the most tedious works involved with the implementation of the flexible body
dynamics is to build a set of joint and force modules. Whenever a new force or
joint module is developed for the rigid body dynamics, the corresponding
module for the flexible body dynamics has to be formulated and programmed
again. In order to avoid such a repetitive process, this investigation proposes a
concept of virtual body and joint.
Shabana [1] presented a coordinate reduction method for multibody systems
with flexible components. The local deformation of a flexible component was
expressed in terms of the nodal coordinates and was then spanned by a set of
mode shapes obtained from a mode analysis. Yoo and Haug [2] spanned the
deformation by a set of static correction modes obtained by applying a unit force
or unit displacement at a node where a large magnitude of force is expected
during the dynamic analysis. Mani [3] used Ritz vectors in spanning the local
deformation and the Ritz vectors were generated by spatially distributing the
inertial and joint constraint forces on a flexible body. Gartia de Jalon et al [4]
presented a fully Cartesian coordinate formulation for rigid multibody dynamics.
This formalism was extended to the flexible body dynamics by Vukasovic et al
[5]. Nonlinearity associated with an orientational transformation matrix was

RecurDyn

FLEXIBLE MULTIBODY DYNAMICS USING A VIRTUAL BODY AND JOINT
relieved by defining all necessary vectors for the equations of motion and
constraints as the generalized coordinates.
Several formulations have been recently developed for flexible body systems
that undergo large deformation. Simo [6] had formulated the equations of motion
for a flexible beam, based on the inertial reference frame. Since displacement of
a point on the beam was directly measured from the inertial reference frame, the
inertia terms become linear and uncoupled, while the strain energy related terms
become nonlinear. Yoo and Ryan [7] proposed a mixed formulation of inertial
and floating reference frames for a rotating beam. Axial deformation was
measured from a deformed state of the rotating beam, while other deformations
were measured from an undeformed state. Shabana [8,9] presented a non-
incremental absolute coordinate formulation in which the global location
coordinates and slopes were defined as the generalized coordinates. Since the
finite rotation coordinates were not used as the generalized coordinates, the
difficulties associated with the finite rotation were resolved.
Contrast to implementation of a rigid body dynamic analysis program, it is
generally complicated to implement a flexible body dynamic formulation and to
expand it for a general purpose program, regardless of whatever formulation has
been chosen. This is because the flexible body dynamic formulations handle
concept of virtual body and joint. The kinematics of virtual body and joint is
presented in Section 2. The equations of motion for a flexible body system are
presented in Section 3. Computer implementation and its impact on a sparse
oriented algorithm are explained in Section 4. Two flexible body systems are
dynamically analyzed by using the proposed method to show its validity in
section 5. Conclusions are drawn in Section 6.

3-3

3.2. KINEMATICS OF TWO CONTIGUOUS FLEXIBLE BODIES

3.2.1 COORDINATE SYSTEMS AND VIRTUAL BODIES

Figure 1 Two adjacent flexible bodies

RecurDyn

Two flexible bodies connected by a joint and their reference frames are shown
in Fig. 1. The frame is the body reference frame of flexible body
and the frame is the inertial reference frame. Suppose there exists a
joint between the and frames, and a force applied at the
origin of the frame. Kinematic admissibility conditions among the
reference frames can be divided into two categories. One is the admissibility
conditions between the two joint frames and the other is the admissibility
conditions among the frames within a flexible body. These two types of
conditions have been mixed in formulating the kinematic joint constraints and
generalized forces in the previous works. As a result, every joint and force
modules in a flexible multibody code, such as ADAMS [10] and DAMS [11],
has been developed separately for rigid and flexible bodies. This would take long
time for computer implementation and prone to coding errors. Especially,
flexible body programming requires much more effort than rigid body
programming does due to complexity associated with flexibility generalized
i i i
Z Y X , ,
i i
Y X
1 1
, ,
i i i
Z Y X
2 2 2
, ,
i
Z Y X , ,
i
Z
1
j j j
Z Y X
1 1 1
, ,
coordinates. In order to minimize the programming effort, a concept of the
virtual body is introduced in this section. At every joint and force reference
frames, a virtual rigid body, whose mass and moment of inertia are zero, is
introduced.

Figure 2 Two adjacent flexible bodies and three virtual bodies

As an example, three rigid virtual bodies are introduced for two adjacent
deformable bodies as shown in Fig. 2. This makes the flexible body has no joint
or applied force and is subjected to only the kinematic admissibility conditions
among its body frame and the virtual body frames. Therefore, the joint and force
modules are developed only for rigid bodies and one flexible body joint is to be
added in the joint module. The kinematic admissibility conditions for the flexible
body joint are formulated in the following subsections.

3.2.2 JOINT CONSTRAINTS BETWEEN TWO RIGID BODIES

A joint has been represented by imposing condition of parallelism or
orthogonality on vectors attached to two adjacent rigid bodies. A library of such
condition for rigid bodies has been well developed and becomes the primitives in
building various joints [10, 11]. The conditions are formulated by using

3-5

geometric vectors that are defined within or between two joint reference frames.
A joint reference frame does not generally coincide with the body reference
frame. The body reference frame for a virtual body also serves as a joint
reference frame in the proposed method. Therefore, the kinematic admissibility
conditions for a joint connecting a virtual body is simplified and the number of
non-zero entries of the constraint Jacobian is reduced.

3.2.3 FLEXIBLE BODY JOINT CONSTRAINT BETWEEN A FLEXIBLE BODY AND A
RIGID VIRTUAL BODY

Figure 3 Flexible body joint constraint between a flexible and a virtual body

Origin of the body reference frame for the virtual body in Fig. 3 can be
expressed as follows:
(
i
f
i i i
i i i i
u u A R
u A R r
+ + =
+ =
+
0
1
)
(1)

where
i
0
u and
i
f
u are the undeformed location vector and deformation vector of
a point on the body with respect to a body reference frame and is the
orientation matrix of body reference frame. The deformation vector
i
A
i
f
u at the

RecurDyn

nodal position can be spanned by linear combination of a set of mode shapes
[12] as
i
f
i
R
i
f
p u = (2)

where is a modal matrix whose columns consist of the translational mode
shapes and is a modal coordinate vector.
i
R
i
f
p
Orientation of the virtual body is obtained as follows: 1 + i

1 , 1 + +
=
i i i
f
i i
A A A A (3)

where is the relative orientation matrix induced by the rotational
deformation and is the orientation matrix between the reference frames of
the flexible body and virtual body
i
f
A
1 , + i i
A
i 1 + i
i
f
in an undeformed state. If the Bryant
angle (1-2-3) [13] is employed, the is expressed as follows: A

(
(
(
+
+

=
i
y
i
x
i
z
i
y
i
x
i
z
i
x
i
z
i
y
i
x
i
z
i
x
i
y
i
x
i
z
i
y
i
x
i
z
i
x
i
z
i
y
i
x
i
z
i
x
i
y
i
z
i
y
i
z
i
y
i
f

cos cos sin sin cos sin sin cos sin cos sin sin
cos sin sin sin sin cos cos cos sin sin sin cos
sin sin cos cos cos
A (4)

If is infinitesimal, the matrix can be approximated as |
T
i
z
i
y
i
x
i
= |
i
f
A

(
(
(
1
1
1
i
x
i
y
i
x
i
z
i
y
i
z
i
f

A (5)

The rotational deformation vector can be represented by linear
combination of rotational mode shapes of body as
i
i

i
f
i i
p

= (6)

where is a modal matrix whose columns are composed of rotational mode
i

3-7

shapes and is the vector of modal coordinate.
C
i
q
|
=
0
0
0
i
f
p
Finally, kinematic constraints between two body frames of the flexible and
virtual bodies can be obtained from Eqs. (1) and (3) as follows:

( ) 0 u u A R r C = + =
+ i
f
i i i i i
R 0
1
(7)
0
h A A f h A A f
h A A g h A A g
g A A f g A A f
=
(
(
(
(
=
+ +
+ +
+ +
1 , 1
1 , 1
1 , 1
i i i
f
T i
T
i T
i i i
f
T i
T
i T
i i i
f
T i
T
i T
(8)
where
| |
(
(
(
=
1 0 0
0 1 0
0 0 1
h g f (9)

Orthogonality conditions would have been used in deriving the orientational
constraints. However, the in Eq. (8) is employed in this research for simple
implementation. Eqs. (7) and (8) yields algebraic constraint equations that
describe the flexible joint between flexible body and virtual body .
Taking variation of Eqs. (7) and (8) yields
i
C
i 1 + i

( )
( )
( )
0 q
C
C
q C
q
q
q
=
(
(
i
i
R i
flex
i
(10)
where
| |
T
T
i
T
i
T
i
f
T
i
T
i i 1 1 + +
= r p R (11)

and the constraint Jacobian matrix ( )
flex
i
q
C is obtained as
( ) |
( )
(
(
(
(

=

+
+
+
1
1
1
i
h
T i f
h
T i
h
T
i
h
T i f
h
T i
h
T
i
g
T i f
g
T i
g
T
i
i
R
i i i
R
B f 0 B f B f
B g 0 B g B g
B f 0 B f B f
C
0 I A B I C
q
q
(12)

RecurDyn

where
h g k
skew
skew
skew
skew
i
T
i i
k
i i f
k
i i
T
i i
k
i i i
, ,
) (
) (
) (
) (
1 1
1 ,
1
=
=
=
=
=
+ +
+
+
k A A B
k A B
A k A A B
u A B
(13)

and the vectors ) (
i
skew u , , , and are the
skew symmetric matrices of vectors,
) (
1
k A
+ i
skew ) (
1 ,
k A
+ i i
skew ) (k skew
i
u , , , and , respectively.
In order to obtain the acceleration level constraint, one can differentiate Eqs. (7)
and (8) twice with respect to time to yield
k A
1 + i
A
1 , + i i
k k

( ) ( ) ( ) ( )
( ) ( )
(
(
(
(
(
+ + +
+ + +
+ + +

=
= =
4 3 2 1
4 3 2 1
4 3 2 1
2
2
2
) ( 2 ) (
h
T
h
T
h
T
h
T
h
T
h
T
h
T
h
T
g
T
g
T
g
T
g
T
i
f
i
R
i i i
flex
i
c
i i
flex
i i
flex
i
skew skew
H f H f H f H f
H g H g H g H g
H f H f H f H f
p A u A
Q q q C q C
q
q q
&
& & & &
(14)

where the is the angular velocity with respect to the body reference frame
and the generalized velocity vector is q&

| |
T
T
i
T
i
T
i
f
T
i
T
i i 1 1 + +
= r p R q & &
&
& (15)
and

( ) ( )
h g k
skew
skew skew
skew skew skew
skew skew
i
f
i i i i
f
i
f
i
k
i i i
T
i
k
i i
T
i i
k
i
T
i i i
k
, ,
) (
) ( ) (
) ( ) ( ) (
) ( ) (
1 , 4
1 1 1 3
1 1 2
1 1
=
=
=
=
=
+
+ + +
+ +
+
p k A A p H
k A A H
k A A H
k A A H
& &

(16)

3-9

3.3. EQUATIONS OF MOTION

Even though the proposed method is applicable to a general system consisting
of many flexible bodies, a slider crank mechanism with one flexible body in Fig.
4(a) is used to clearly show the impact of the proposed method on the equations
of motion. An equivalent virtual system, modeled by using the rigid virtual
bodies proposed in this investigation, is shown in Fig. 4(b). The augmented
equations of motion for the system is obtained by using the general form of
equations of motion as [11]

(
+ +
=
(
(
(
c
s v e
T
Q
Q Q Q
q
0 C
C M
q
q
& &
(16)

where is the mass matrix of the system. The vector consists of
translational acceleration for rigid and flexible bodies, angular acceleration, and
modal acceleration for the flexible body.
M q& &

FLEXIBLE BODY FLEXIBLE BODY
RIGID BODY RIGID BODY
C
R
A
N
K
C
R
A
N
K
SLIDER SLIDER
C
O
U
P
L
E
R
C
O
U
P
L
E
R
1
2
3
Y
Z
X
P1 P1

(a) Two rigid bodies and one flexible body

RecurDyn

FLEXIBLE BODY FLEXIBLE BODY
VIRTUAL BODY VIRTUAL BODY
C
R
A
N
K
C
R
A
N
K
C
O
U
P
L
E
R
C
O
U
P
L
E
R
SLIDER SLIDER
VIRTUAL BODY VIRTUAL BODY 1
2
4
5
3

(b) Two rigid bodies, one flexible body and two virtual bodies
Figure 4 Slider crank mechanism with one flexible body

The is the vector of Lagrange multipliers and , , , and are
the strain energy terms, velocity induced forces, externally applied forces, and
the vector absorbs terns that are quadratic in the velocities, defined clearly
by Shabana [11].
s
Q
v
Q
e
Q
c
Q
c
Q

3.3.1 COEFFICIENT MATRIX OF CONVENTIONAL AUGMENTED FORMULATION

The mass matrix for the system in Fig. 4(b) is

(
(
(
=
3
2
1
r
r
f
M 0
M
0 M
M (17)

where the mass matrices for virtual bodies, and are the mass matrix for
flexible body and for a rigid body, as
f
M
r
M

3-11

) 3 , 2 ( ,
6 6
) 6 ( ) 6 (
1
=
(
=
(
(
(
+ +
k
symmetric
k
k
rr k
r
nf nf
ff f fr
r
rr
f

m 0
0 m
m m m
m m
m
M
M
(18)

where is the number of modal coordinates. The constraint Jacobian matrix
of the slider crank mechanism with flexible crank is
nf
C
) (
q
C

( )
( )
( )
( ) (
(
(
(
(
=
int
30
int
23
,
12
,
01
) (
jo
jo
c flex
c flex
c
q
q
q
q
q
C
C
C
C
C (19)

where, ( )
c flex,
q
C is the constraint Jacobian matrix of the flexible joint obtained
by the conventional method[11].

3.3.2 COEFFICIENT MATRIX OF PROPOSED AUGMENTED FORMULATION

The mass matrix for the system in Figure 4b is

(
(
(
(
(
(
=
5
4
3
2
1
r
r
v
f
v
M
M 0
M
0 M
M
M (20)

where the mass matrix for virtual body, , the mass matrix for flexible body,
v
M

RecurDyn

f
M , and the mass matrix for rigid body, are
r
M
symmetric
m
,
6 6
(
(
(
(
(
(
(
(
(
int
int
int
,
,
int
p
p

| | 3 , 1 ,
6 6
= =

k
v
0 M
) 6 ( ) 6 (
2
nf nf
ff f fr
r
rr
f
+ +
(
(
(
=
m m
m m
m
M

(21)
) 5 , 4 ( =
(
= k
k
k
rr k
r
m 0
0 m
M

The proposed constraint Jacobian matrix ( )
p
q
C of the slider crank
mechanism with flexible crank is

( )
( )
( )
( )
( )
( )
( )
=
50
45
34
23
12
01
jo
jo
jo
flex
flex
jo
p
q
q
q
q
q
q
q
C
C
C
C
C
C
C (22)

where ( )
p flex,
q
C is the constraint Jacobian matrix of the flexible body joint
obtained by the proposed method. As shown in Eq. (22), the constraint Jacobian
matrix can be clearly divided into flexible and rigid body joint modules by
introducing rigid virtual bodies.

3.3.3 NON-SINGULARITY OF AUGMENTED MASS MATRIX

If the constraint Jacobian matrix C has a full row rank, the coefficient
matrix of Eq. (16) is non-singular, which can be proved by showing that the
following equations have only trivial solutions under the same assumption.
q

3-13

( ) 0 y C y
q
= +
3 1
N
T
N
M (23)
( ) 0 y C
q
=
3
V
T
(24)
( ) ( ) 0 y C y C
q q
= +
2 1
V N
(25)

where ,
N
M ( )
N
q
C , and ( )
V
q
C are the mass matrix of non-virtual body, the
Jacobian of Eqs. (23) and (24) after pre-multiplying by Eq. (23) and by
Eq. (24) yields
T
1
y
T
2
y
( ) ( ) ( ) 0 y M y y C y C y y M y
q q
= = + +
1 1 2 1 3 1 1 N
T
V N
T
N
T
(26)

where Eq. (25) is used. As a result, . Eqs. (23) and (24) reduces to 0 y =
1

0 y C
q
=
3
T
(27)

Since the has full row rank, must be zero. Substituting into Eq.
(25) yields
q
C
3
y 0 y =
1
( ) 0 y C
q
=
2
V
(28)

Since rank of ( )
V
q
C
y =
3
is the same as the size of , must be zero. Since
, , are only solutions of Eqs. (23), (24), and (25), their
coefficient matrix is non-singular.
2
y
2
y
0 y =
1
0 y =
2
0

3.4. COMPUTER IMPLEMENTATION AND DISCUSSIONS

In previous sections the flexible equations of motion and kinematic constraints
using virtual body techniques are presented. In this section, the computer
implementation methods for the equations developed in section 2 and 3 are
illustrated.

3.4.1 NUMERICAL ALGORITHM

RecurDyn

A general purpose program for the dynamic analysis of mechanical systems
can be implemented in many different ways, depending on the DAE solution
method employed. The generalized coordinate partitioning method [12] is
employed in this investigation and the proposed program structure is shown in
Fig. 5. Note that there exist joint and force modules only for rigid bodies, and
one flexible body joint is added in the joint library. Those modules can handle
any system consisting of rigid bodies as well as flexible bodies.
POSITION ANALYSIS
VELOCITY ANALYSIS
ACCELERATION ANALYSIS
FORCE ANALYSIS JOINT MODULES FOR RIGID BODY
FLEXIBLE BODY JOINT
JOINT MODULES
GENERATE JOINT EQUATIONS
FORCE MODULES FOR RIGID BODY
T = T_END
NO
YES
END
START
T = T + STEPSIZE
MAIN PROGRAM MAIN PROGRAM MODULE LIBRARY MODULE LIBRARY

Figure 5 A program structure for proposed flexible multibody dynamics

3.4.2 COMPARISON OF DIFFERENT IMPLEMENTATION METHODS

The joint and force modules must be expanded whenever a user group of the
flexible body dynamics code demands a special type of joint or force element.
Since the proposed implementation method for a flexible body dynamics code
reuses all joints and force modules for the rigid body, only necessary modules to
be added to a rigid body dynamics code are the flexible body joint and the
equations of motion for a flexible body. As a result, the proposed method is not
only easy to implement but also to maintain, because the proposed method
eliminates the additional programming effort for the flexible body modules when
an expansion of the joint or force library is required.
However, there are some computational overheads, because extra bodies and
joints must be introduced to a flexible body system if the proposed method is
employed. It is very difficult to analyze the computational overheads for general

3-15

rigid and flexible multibody systems, because various models and flexible body
dynamics theories may end up with various situations. In order to simplify the
presentation, the slider crank mechanism in section 3 is reconsidered in this
section.
Numerical experiments with the Cartesian coordinate formulation [12] showed
that more than 70% of the total computation time is consumed in the Gaussian
elimination of matrices arising from various equations. Direct Gaussian
elimination of Eq. (16) would require a number of arithmetic operations
proportional to approximately cube of the matrix size. However, the number of
arithmetic operations for a sparse solver such as the Harwell Library [14] is
increased only linearly to the number of non-zero entries if the structure of the
non-zero entries is exploited. A sparse solver reduces the number of operations
by minimizing the number of fill-ins and performing the Gaussian elimination
only on the non-zero entries and fill-ins. Therefore, it is important to add the new
non-zero entries so that overall non-zero structure of the resulting matrix is not
disturbed and is well suited for minimization of the fill-ins. The structures of the
non-zeros are shown in Eqs. (20) and (21), respectively. No non-zero entry in the
mass matrix of the proposed method is added, because the mass and moment of
inertia of the virtual body are zero. Total numbers of non-zero entries of Eq. (16)
are shown in Table 1. Note that redundant constraints are eliminated and
coincidence of the virtual body and joint reference frames is utilized in reducing
the number of non-zeros. Since the new non-zero entries in Eq. (16) are scattered
around the existing ones, the overall structure of the non-zeros is not disturbed
and a similar reordering sequence in sparse Gaussian elimination to the original
reordering sequence in a sparse linear solver can be used. As a result, expected
computation time increment with the proposed method would be about 50% for
the slider crank mechanism, when a sparse solver is employed.

Table 1 Number of non-zero entries for the slider-crank mechanism
Implementation Methods No. of non-zero entries
Conventional 122+10nmode
Propose 188+12nmode
* nmode: the number of mode shapes

RecurDyn


The number of non-zeros for a most frequently used joints such as, revolute
joint, spherical joint and translational joint, are also given in Table 2. It can be
easily shown that the percent ratio of the computation time would become
smaller if the number of flexible bodies in a system is small, which is true in
many cases. However, the computation time may be increased significantly for a
flexible body system which has many joints and force elements, because the
number of virtual bodies in such a system is large.

Table 2 Number of non-zero entries for the slider-crank mechanism
Joint Increment of non-zero entries
Revolute joint (33 + nomde) nvirtual
r

Spherical joint (33 + 3nomde)nvirtual
s

Translational joint (33 + nomde) nvirtual
t

* nvirtual
r
: the number of virtual bodies which are connected with revolute
joint
* nvirtual
s
: the number of virtual bodies which are connected with spherical
joint
* nvirtual
t
: the number of virtual bodies which are connected with
translational joint

Another way of implementing the virtual body concept is to mix the proposed
implementation method with the conventional one. The conventional method
may be used to implement the frequently used joint and force modules such as
the revolute and translational joints and an applied force at a point. Meanwhile,
the proposed method may be used to implement the less frequently used joint
and force modules such as an universal joint or a planar joint. This
implementation method will improve both the computational overhead as well as
the coding convenience. This mixed formulation can be very effective if a set of
basic joint and force modules have already been developed and more modules
for the flexible bodies need to be added.

3-17

3.5. NUMERICAL RESULTS

Dynamic analysis of a flexible slider crank mechanism and a flexible
pendulum mechanism is presented in order to validate the results from the
proposed method. The examples are solved by using both the proposed method
and the nonlinear approach developed by Simo [6].

3.5.1 FLEXIBLE SLIDER CRANK MECHANISM

The system consists of two rigid bodies and one flexible body, as shown in
Fig. 4. Length, cross sectional area, and area moment of inertia of the elastic
crank are 0.4 m, 0.0018m
2
, and 1.215 10
-4
m
4
, respectively. The crank is
modeled by using 10 two-dimensional elastic beam elements of equal lengths.
The material mass density of the beam is 5540.0 kg/m
3
and its Young's modulus
is 1.010
9
N/m
2
. Vibration analysis of the crank is carried out with fixed-free
boundary condition and the resulting mode shapes are shown in Fig. 6.
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
X(M)
M
A
G
N
I
T
U
D
E
.
1st mode
2nd mode
3rd mode
4th mode

Figure 6 Mode shapes of the crank

Four mode shapes are selected to span the deformation of the crank. As a result,

RecurDyn

the system has 5 degrees of freedom. Dynamic analysis using the generalized
coordinate partitioning method is performed for 5 sec under the constant
acceleration condition of the joint between the ground and the body 1. The
acceleration, displacement, and relative deformation of the pin joint connecting
the crank and the coupler both from the proposed method and the nonlinear
approach [6] are shown in Figs. 7, 8, and 9, respectively. Note that since the
results from both models are almost identical as shown in these figures, the
proposed implementation methods using rigid virtual body can be validated.

-30
-20
-10
0
10
20
30
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
TIME (SEC)
Y

(
M
/
S
E
C
^
2
)
.
.
NONLINEAR
PROPOSED

Figure 7 Y Acceleration of P1

3-19

-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
TIME (SEC)
Y

(
M
)
NONLINEAR
PROPOSED

Figure 8 Y Displacement of P1

-0.015
-0.010
-0.005
0.000
0.005
0.010
0.015
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
TIME (SEC)
Y

(
M
)
NONLINEAR
PROPOSED

Figure 9 Deformation of P1

RecurDyn


3.5.2 FLEXIBLE PENDULUM MECHANISM

BEAM BEAM
(FLEXIBLE BODY) (FLEXIBLE BODY)
REVOLUTE REVOLUTE
JOINT JOINT
GRAVITY GRAVITY
X
Y

Figure 10 Simple flexible pendulum model

The pendulum body shown in Fig. 10 is modeled with 10 beam elements
having a length of 0.4m, a cross sectional area of 0.0018m
2
, and a mass of
3.9888kg. Dynamic analysis is performed for 1 sec under the free falling
condition. Mode shapes of the pendulum are obtained by ANSYS[15] with the
simply supported-free(pin-free) boundary condition. Mode Shapes of the
pendulum are shown in Fig. 11. The acceleration and relative transverse
deformation of the tip point both from the proposed method and the nonlinear
approach [6] are shown in Figs. 12, and 13, respectively. It is clear from these
results that the proposed method and nonlinear approach are in good agreement,
accordingly.
-1.2
-0.8
-0.4
0.0
0.4
0.8
1.2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
X(M)
M
A
G
1st mode
2nd mode
3rd mode
4th mode

3-21

Figure 11 Mode Shapes of the pendulum

-30
-20
-10
0
10
20
30
40
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Y

(
M
/
S
E
C
^
2
)
.
.
NONLINEAR
PROPOSED

Figure 12 Y Acceleration of beam tip

-0.0015
-0.0010
-0.0005
0.0000
0.0005
0.0010
0.0015
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Y

(
M
)
NONLINEAR
PROPOSED

Figure 13 Deformation of beam tip

RecurDyn


3.6. SUMMARY AND CONCLUSIONS

An implementation method is proposed for general purpose rigid and flexible
multibody dynamics with the Cartesian coordinate formulation. A concept of the
virtual body and joint is introduced to make a flexible body free from all
kinematic admissibility conditions except these from the virtual-flexible body
joint. This eliminates extra programming efforts for the flexible body whenever a
joint or force module is added to a general purpose dynamic analysis program.
The computational overhead of the proposed method is turned out to be
moderate if a sparse solver is employed, while implementation convenience is
dramatically improved. A flexible slider crank mechanism and a simple
pendulum are analyzed and the results are validated against these from a
nonlinear approach.

3-23

REFERENCES

1. A. A. Shabana, "Substructure Synthesis Methods for Dynamic Analysis of Multibody
Systems", Computers & Structures, Vol. 20. No. 4, pp 737-744, 1985
2. W. S. Yoo, and E. J. Haug, "Dynamics of Flexible Mechanical Systems Using Vibration
and Static Correction Modes", Journal of Mechanisms, and Transmissions, and
Automation in Design, 1985
3. H. T. Wu, and N. K. Mani, "Modeling of Flexible Bodies for Multibody Dynamic Systems
Using Ritz Vectors", Journal of Mechanical Design, Vol. 116, pp. 437-444, 1994.
4. J. Garcia de Jalon, J. Unda, and A. Avello, "Natural Coordinates for the Computer
Analysis of Three-Dimensional Multibody Systems", Computer Methods in Applied
Mechanics and Engineering, Vol. 56, pp. 309-327, 1985.
5. N. Vukasovic, J. T. Celigueta, J. Garcia de Jalon, and E. Bayo, "Flexible Multibody
Dynamics Based on a Fully cartesian System of Support Coordinates", Journal of
Mechanical Design, Vol. 115, pp. 294-299, 1993.
6. J. C. Simo, and L. Vu-Quoc, "On the Dynamics of Flexible Beams Under Large Overall
Motions-The Plane Case: Part I", Journal of Applied Mechanics, Vol. 53, pp. 849-854,
1986.
7. H. H. Yoo, R. R. Rion, and R. A. Scott, "Dynamics of Flexible Beams Undergoing
Overall Motions", Journal of Sound and Vibration, Vol. 181, pp. 261-278, 1994
8. A. A. Shabana, A. P. Christensen, "Three Dimensional Absolute Nodal Coordinate
Formulation : Plate Problem", International Journal for Numerical Methods in
Engineering, Vol. 40, pp. 2775-2790, 1997
9. A. A. Shabana, H. A. Hussien, and J. L. Escalona, "Application of the Absolute Nodal
Coordinate Formulation to Large Rotation and Large Deformation Problems", Journal
of Mechanical Design, Vol. 120, pp. 188-195, 1998
10. ADAMS Reference Manual, Mechanical Dynamics, 2301 Commonwealth Blvd, Ann
Arbor, MI 48105.
11. A. A. Shabana, Dynamics of Multibody Systems, John & Wiley, New York, 1989.

RecurDyn

12. R. A. Wehage and E. J. Haug, "Generalized Coordinate Partitioning for Dimension
Reduction in Analysis of Constrained Dynamic Systems", Journal of Mechanical Design,
Vol. 104, pp. 247-255, 1982.
13. P. E. Nicravesh, Computer-Aided Analysis of Mechanical systems, Prentice-Hall, 1988
14. I. S. Duff, A. M. Erisman, and R. K. Reid, Direct Methods for Sparse Matrices,
Clarendon Press, Oxford, 1986
15. ANSYS Reference Manual, ANSYS, Inc., Southpointe 275 Technology Drive,
Canonsburg, PA 15317.

3-25

RecurDyn


4
FOR FLEXIBLE MULTIBODY DYNAMICS

4.1. INTRODUCTION

The equations of motion for the general constrained mechanical systems were
derived in terms of the relative coordinates by Wittenburg [1]. The velocity
transformation method with the graph theory was employed to transform the
equations of motion in the Cartesian coordinate space to the joint space
systematically. Hooker [2] proposed a recursive formulation for the dynamic
analysis of a satellite which has a tree topology. It was shown that the
computational complexity of the formulation increases only linearly to the
number of bodies. Fetherstone [3] used the recursive formulation to perform the
inverse dynamic analysis of manipulators. Bae and Haug [4] further developed
the formulation for constrained mechanical systems by using the variational
vector calculus. The recursive formulation was applied to linearize the equations
of motion [5]. Recursive formula for each term in the equations of motion was
directly derived, using the state vector notation. Similar approach was taken in
Ref. 6 to implement the implicit BDF integration with the relative coordinates.
Since the recursive formulas were derived term by term, the resulting equations
and algorithm became much complicated. To avoid the complication, the
equations of motion were derived in a compact matrix form by using the velocity
transformation method in Ref [7]. The generalized recursive formula for each
category of the computational operations was developed and applied whenever
such a category was encountered. This research applies the generalized recursive
formulas for the multibody flexible dynamics.
Shabana [8] presented a coordinate reduction method for multibody systems
with flexible components. The local deformation of a flexible component was
expressed in terms of the nodal coordinates and was then spanned by a set of
mode shapes obtained from a mode analysis. A fully Cartesian coordinate

RecurDyn

GENERALIZED RECURSIVE FORMULATION FOR FLEXIBLE MULTIBODY DYNAMICS
formulation for rigid multibody dynamics by Jalon [9] was extended to the
flexible body dynamics by Vukasovic and Celigueta [10]. Nonlinearity
associated with an orientational transformation matrix was relieved by defining
all necessary vectors for the equations of motion and constraints as the
generalized coordinates. Variational equations of motion for flexible multibody
systems were derived in Ref. 11. The variational approach was applied to extend
the rigid body recursive formulation to flexible multibody systems. An extended
kinematic graph concept was employed to develop a new recursive formulation
for the dynamic analysis of flexible multibody systems by Lai and Haug [12].
Cardona and Geradin [13] dealt with substructuring for dynamic analysis of
flexible multibody systems. The joint coordinates and the finite element method
were employed for the flexible body dynamics by Nikravesh [14]. Pereira [15]
presented a systematic method for deriving the minimum number of equations of
motion for spatial flexible multibody systems.
Contrast to implementation of a rigid body dynamic analysis program, it is
generally complicated to implement a flexible body dynamic formulation and to
expand it for a general purpose program, regardless of whatever formulation has
been chosen. This is because the flexible body dynamic formulations handle
concept of virtual body and joint. The relative coordinate kinematics and the
virtual body concept are presented in section 2. A graph representation of
flexible multibody systems is presented in section 3. The forward recursive
formula and backward recursive formula respectively are treated in sections 4
and 5. A solution method of the equations of motion for a flexible body system is
presented in section 6. Flexible slider crank mechanism is dynamically analyzed
by using the proposed method to show its validity in section 7. Conclusions are
drawn in section 8.

4-3

4.2. RELATIVE COORDINATE KINEMATICS OF TWO CONTIGUOUS
FLEXIBLE BODIES

4.2.1 COORDINATE SYSTEMS AND VIRTUAL BODIES

Figure 1 Two adjacent flexible bodies

The Z Y X
O
frame is the inertial reference frame and the

frame
is the body reference frame in Fig. 1. Velocities and virtual displacements of
point in the
z y x
Z Y X frame are respectively defined as

(
r
(1)
and
(
(2)

Their corresponding quantities in the z y x frame are respectively defined
as
(
(
=
(
=
A
r A
r
Y
T
T
& &
(3)

RecurDyn

and
(
(
=
(
=
A
r A
r
Z
T
T
(4)

where is the orientation matrix of the A z y x frame with respect to the
Z Y X frame. Two flexible bodies connected by a joint and their reference
frames are shown in Fig. 1.
Suppose there exists a joint between the
1 1 1 i i i
z y x and
frames, and a force applied at the origin of the
1 1 1 j j j
z y x
2 2 2 j j j
z y x frame. Kinematic
admissibility conditions among the reference frames can be divided into two
categories. One is the admissibility conditions between the two joint frames and
the other is the admissibility conditions among the frames within a flexible body.
These two types of conditions have been mixed in formulating the kinematic
joint constraints and generalized forces in the previous works. As a result, every
joint and force modules in a flexible multibody code, such as ADAMS [16] and
DAMS [17], have been developed separately for rigid and flexible bodies. This
would take long time for computer implementation and prone to coding errors.
Especially, flexible body programming requires much more effort than rigid
body programming does due to complexity associated with flexibility
generalized coordinates and the strain energy.

Figure 2 Two adjacent flexible bodies and three virtual bodies

4-5

In order to minimize the programming effort, a concept of the virtual body is
introduced in this section. At every joint and force reference frames, a virtual
rigid body, whose mass and moment of inertia are zero, is introduced. The
virtual body and the original flexible body are then connected by a virtual joint.
As an example, three virtual rigid bodies are introduced for two adjacent
deformable bodies as shown in Fig. 2. Note that the flexible bodies have no joint
or applied force except the virtual joints which are represented by the kinematic
admissibility conditions among the flexible body frame and the virtual body
frames. Therefore, the joint and force modules are developed only for rigid
bodies and one flexible body joint of the virtual joints to be added in the joint
module. The recursive kinematic relationships representing the admissibility
conditions of the flexible body joint are formulated in the following subsections.

4.2.2 RELATIVE KINEMATICS FOR A FLEXIBLE BODY JOINT

Figure 3 Flexible body joint between a flexible body and a virtual body

A virtual body is always connected to the original flexible body by a flexible
body joint. Origin of the virtual body reference frame in Fig. 4 can be expressed
as follows:

RecurDyn

) (
) 1 ( ) 1 ( 0 1 1 i i i i i i i
+ + = u s A r r (5)

where and
i i ) 1 ( 0
s
i i ) 1 (
u are the undeformed location vector and deformation
vector of the origin of the virtual body with respect to the flexible body reference
frame, is the orientation matrix of the flexible body reference frame. The
deformation vector
1 i
A
i i ) 1 (
u can be spanned by linear combination of a set of
mode shapes [8] as
f
i i
R
i i i ) 1 ( 1 ) 1 (
= q u (6)

where is a modal matrix whose columns consist of the translational mode
shapes and superscript f in denotes the modal coordinate vector. Subscripts
and denote the generalized coordinate between the and i body
reference frames.
R
i 1
1 i
f
i i ) 1 (
q
i i 1
The angular velocity in the local reference frame is obtained as follows

f
i i i
T
i i i
T
i i i ) 1 ( 1 ) 1 ( 1 ) 1 (
+ = q A A &
(7)

where is used. Differentiating Equation (5) and multiplying by
yields
i
T
i
T
i i
A A A
) 1 ( ) 1 (
=
T
i
A

f
i i i
T
i i i i i
T
i i i
T
i i i ) 1 ( 1 ) 1 ( 1 ) 1 ( ) 1 ( 1 ) 1 (
~

+ = q A s A r A r & & &

(8)

where
i i i i i i ) 1 ( ) 1 ( 0 ) 1 (
~

+ = u s s , symbol with tilde denotes skew symmetric matrix
which consists of their vector elements, and wide tilde are used.
Combining Equations (7) and (8) yields the following recursive velocity
equation for a flexible body joint.
i i i
A A =
~
&
i

f
i i
f
i i i
f
i i i ) 1 ( 2 ) 1 ( 1 1 ) 1 (
+ = q B Y B Y & (9)
where

4-7

(
(
=
(
(
1 ) 1 (
1 ) 1 (
2 ) 1 (
) 1 (
) 1 ( ) 1 ( ) 1 (
1 ) 1 (
~
i
T
i i
R
i
T
i i
f
i i
T
i i
i i
T
i i
T
i i f
i i
A
A
A 0
s A A
B
B
(10)

It is important to note that matrices and are function of only
modal coordinates of the flexible body i-1. As a result, further differentiation of
the matrices and B in Equation (9) with respect to other than bf
yields null. This property will play a key role in simplifying recursive
formulas in sections 4 and 5.
f
i i 1 ) 1 (
B
f
i i 2 ) 1 (
B
f
i i 1 ) 1 (
B
f
i i 2 ) 1 (
f
i i ) 1 (
q
Equation (9) defines the kinematic relationships between an inboard flexible
body and an outboard rigid body. The kinematic relationships between an
inboard rigid body and an outboard flexible body can be derived similarly.
Similarly, the recursive virtual displacement relationship between a flexible body
and a virtual body is obtained as follows

f
i i
f
i i i
f
i i i ) 1 ( 2 ) 1 ( 1 1 ) 1 (
+ = q B Y B Y & (11)
where
(
(

=
(
(
1
1 1 ) 1 (
2 ) 1 (
) 1 (
) 1 ( ) 1 ( ) 1 (
1 ) 1 (
~
~
i
R
i i i i
r
i i
T
i i
T
i i i i
T
i i r
i i
s
B
A 0
A s A
B
(12)

4.2.3 RELATIVE KINEMATICS FOR A RIGID BODY JOINT

The recursive velocity relationship for a rigid body joint connecting two rigid
bodies can be derived by following the similar steps as in Equations (5)-(9) as

r
i i
r
i i i
r
i i i ) 1 ( 2 ) 1 ( 1 1 ) 1 (
+ = q B Y B Y & (13)

RecurDyn

where superscript r denotes the generalized co-ordinate from a rigid body joint
and
(
(
+
=
(
(
+
=

i i
T
i i
i i
T
i i i i i i i i
T
i i r
i i
T
i i
T
i i i i i i i i i i
T
i i
T
i i r
i i
i i
) 1 ( ) 1 (
) 1 ( ) 1 ( ) 1 ( ) 1 ( ) 1 ( ) 1 (
2 ) 1 (
) 1 (
) 1 ( ) 1 ( ) 1 ( ) 1 ( ) 1 ( ) 1 ( ) 1 (
1 ) 1 (
)
~
)
~
((
)
~
~
~
(
) 1 (
H A
H A s A d A
B
A 0
A s A d s A A
B
q
(14)

where is determined by the axis of rotation. Node that the matrices are
function of only .
i i ) 1 (
H B
r
i i ) 1 (
q

4.2.4 GRAPH REPRESENTATIONS OF MECHANICAL SYSTEMS

The graph theory was used to automatically preprocess mechanical systems
having various topological structures in References [1, 4]. A node and an edge in
a graph have represented a body and a joint, respectively. The preprocessing,
based on the graph theory, yields the path and distance matrices that are provided
to automatically decide computational sequences. Two computational sequences
are required in a general purpose program. One is the forward path sequence
starting from the base body and moving towards the terminal bodies. The other is
the backward path sequence starting from the terminal bodies and moving
towards the base body.

Figure 4 Flexible slider crank mechanism

4-9

Figure 5 Graph representation and computational sequence

In order to derive systematically the recursive formulas, bodies in a graph are
divided into four disjoint sets (associated with a generalized coordinate ) as
follows :
k
q

) (
k
q I ={adjacent outboard body of the joint having as its generalized co-
ordinate}
k
q
) (
k
q II ={all outboard bodies of , excluding all bodies in } ) (
k
q I ) (
k
q I
) (
k
q III ={all bodies between the base body and the inboard body of ,
including the base and inboard bodies and excluding all bodies in }
) (
k
q I
) (
k
q I
) (
k
q IV ={ the complementary set of } ) ( ) ( ) (
k k k
q q q III II I

RecurDyn


As an example, the graph theoretic representation and computational path
sequences of the system in Fig. 4 are shown in Fig. 5. The four disjoint sets for
the system in Fig. 5, if belongs to the joint between bodies 3 and 4, are
={body 4}, ={bodies 5, 6, and 7}, ={bodies 1, 2, and 3},
={bodies 8, 9, 10, 11, and 12}
k
q
) (
34
q I
(
34
q IV
) (
34
q II ) (
34
q III
)

4.3 FORWARD RECURSIVE FORMULAS

4.3.1 GENERALIZATION OF THE VELOCITY RECURSIVE FORMULA

Generalization of the velocity recursive formula can be achieved by
computational equivalence between the recursive method and the velocity
transformation method. The velocity Y for all bodies in a system can be
obtained by repetitive symbolic substitutions of the recursive formula in
Equations (9), (11) and (13), depending on the type of a joint, along the forward
path sequence of a graph and by appending the trivial equation of as
follows :
f f
q q & & =

q B
q
q
I 0
B B
q
Y
Y &
&
&
&

(
=
(
=
f
r
zf zr
f
(15)

where and q are the relative and modal co-ordinates vectors for a system,
respectively. The dimension of , and are, respectively, assumed to
be ,
r
q
n
f
Y
r
q&
f
q&
nc r , and . The velocity nf
nf nc
R
+
Y with a given can be
evaluated either by using Equation (15) obtained from symbolic substitutions or
by using (9), (11) and (13) with recursive numeric substitution of 's. Since
both formulas give an identical result and recursive numeric substitution is
proven to be more efficient [4], matrix multiplication with a given will
be actually evaluated by using Equations (9), (11) and (13). Since q in
nf nr
R
+
q&
i
Y
q B& q&
&

4-11

Equation (15) is an arbitrary vector in
nf nr
R
+
, Equations (9), (11), (13) and (15)
which are computationally equivalent, are actually valid for any vector
such that
Bx
x
X
=
(
(
BX BX+
q
) (Bx
i ) 1 (
+ B
i
i i
(
1 1
X
nf nr
R
+
x

X

(16)
and

X = (17)

where
nf nc
R
+
X is the resulting vector of multiplication of and and B
matrices depend on a joint type. As a result, transformation of
B x
nf nc
R
+
Y into
nf nc
R
+
Bx is actually calculated by recursively applying Equation (17) to
achieve computational efficiency in this research.

q
X =

Equation (17) is partially differentiated with respect to for
to obtain the recursive formula for as follows.
k
q
) ( ,..., 1 nf nr k + =
q
) (Bx

i i q i i q i i i q i i q i
k k k k
) 1 ( 2 ) 1 ( 1 1 1 1 ) 1 (
) ( ) ( ) ( ) (

+ = X B X X B X (18)

Since matrices depend only on the generalized coordinates for joint
, their partial derivatives with respect to generalized coordinates other than
become null. In other words, the partial derivatives become null if
does not belong to set . If body is an element of set , Equation
(18) becomes
B
i i ) 1 (
i i ) 1 (
q
k
q
) (
k
q I ) (
k
q II

(19)
k k
q i q i
) ) (
) 1 (
= B X

RecurDyn


Figure 6 Computation sequence of
ii
q
Y

If body belongs to set , is not affected by . As a
result, Equation (18) is further simplified as follows.
i ) ( ) (
k k
q q IV III
i
X
k
q

0 X =
k
q i
) ( (20)

4-13

There are two recursive formulas in the case of body . If body is an
element of set , body
) (
k
q I i i
) (
k
q I 1 i is naturally its inboard body and belongs to set
. Equation (18) becomes ) (
k
q III

i i q i i i q i i q i
k k k
) 1 ( 2 ) 1 ( 1 1 ) 1 (
) ( ) ( ) (

+ = X B X B X (21)

If bodies and are elements of set , the recursive formula in
Equation (18) is expressed as follows:
1 i i ) (
k
q I

i i q i i q i i i i q i i q i
k k k k
) 1 ( 2 ) 1 ( ) 1 ( 1 ) 1 ( 1 1 ) 1 (
) ( ) ( ) ( ) (

+ + = X B X B X B X (22)

As an example, the recursive formula in Equation (19)-(22) can be applied to
compute
34
q
Y for the system in Fig. 4, as shown in Fig. 6. Note that since the
recursive formulas for and can be obtained similarly, they are
omitted.
Bx
q
) (Bx

4.4 BACKWARD RECURSIVE FORMULAS

4.4.1 GENERALIZATION OF THE FORCE RECURSIVE FORMULA

A generalized recursive formula for transformation of
nf nr
R
+
x into a new
vector Bx X = in
nf nc
R
+
G
is derived in section 4. Inversely, it is often necessary
to transform a vector in
nf nc
R
+
into a new vector in G B
T
g =
nf nr
R
+
. Such
a transformation can be found in the generalized force computation in the joint
space with a known force in the Cartesian space. The virtual work done by
is obtained as follows.
nf nc+
R Q

| |
(
=
f
c
fT
T T
Q
Q
q Z Q Z W (23)

where Z must be kinematically admissible for all joints in a system and
c
Q

RecurDyn

and are the Cartesian and modal forces, respectively. Substitution of virtual
displacement relationship into Equation (23) yield
f
Q
Q

* *
) (
f fT r rT f c T
zr
fT c T
zr
rT
Q q Q q Q Q B q Q B q W + = + + = (24)

where and Q . Equation (24) can be written in a
summation form as
c T
zr
r
Q B
* f c T
zf
f
Q Q B +
*

+
+ +
+
+ +
+ =
fjts i i
f
i i
fT
i i
rjts i i
r
i i
rT
i i
) 1 (
*
) 1 ( ) 1 (
) 1 (
*
) 1 ( ) 1 (
Q q Q q W (25)

where rjts and fjts respectively denote all rigid body joints and all flexible body
joints.
On the other hand, the symbolic substitution of the recursive virtual
displacement relationship into Equation (23) along the chain (starting from the
terminal bodies toward inboard bodies) and the reorganization of the equation
about the virtual relative displacement and modal displacement yield

( )
( )

+
+ + + + +
+
+ + + +
|
|
.
|
\
|
+ +
|
|
.
|
\
|
+ =
+
+
fjts i i q l
i
c
i
fT
i i
f
i i
fT
i i
rjts i i q l
i
c
i
rT
i i
rT
i i
i i
i i
) 1 ( ) (
1 1 2 ) 1 ( ) 1 ( ) 1 (
) 1 ( ) (
1 1 2 ) 1 ( ) 1 (
) 1 (
) 1 (
I
I
S Q B Q q
S Q B q W

(26)

where
+ +
+ +
+ + +
+
+
+
) (
2 2 1
1
) 2 )( 1 (
) 2 )( 1 (
) (
body terminal a is 1 if
i i
i i
q l
i
c
i
T
i
i
i
I
S Q B S
0 S
(27)

The recursive formula for bf and is obtained by equating Equations
(25) and (26) as follows:
* f
Q
* r
Q

+ + + + +
+ + =
) (
1 1 2 ) 1 ( ) 1 (
*
) 1 (
) 1 ( i i
q l
i
c
i
T
i i i i i i
I
) S (Q B Q Q (28)

4-15

where for a rigid body joint and for a flexible body joint connecting
an inboard flexible body and an outboard virtual body, and for a
flexible body joint connecting an inboard virtual body and an outboard flexible
body, and is defined in Equation (27).
0 =
+ ) 1 (i i
1 + i
S
Q
f
i i i i ) 1 ( ) 1 ( + +
= Q Q

Since Q in Equation (23) is an arbitrary vector in
nf nc
R
+
, Equations (23) and
(28)are valid for any vector in G
nf nc
R
+
. As a result, the matrix multiplication
of is actually evaluated to achieve computational efficiency in this
research by
G B
T
+ + + + +
+ + =
) (
1 1 2 ) 1 ( ) 1 ( ) 1 (
) 1 (i i
q I l
i
c
i
T
i i i i i i
) S (G B G g (29)

where is the result of and is defined as in Equation
(28) and
g G B
T
) 1 ( + i i
G
) 1 ( + i i
Q
body terminal a is 1 if
1
+
+
i
i
0 S (30)
+ +
+ + + + +
+
) (
2 2 1 ) 2 )( 1 ( 1
) 2 )( 1 (
) (
i i
q l
i
c
i
T
i i i
I
S Q B S (31)

Recursive formula in Equation (29) must be applied for all joints in the
backward path sequence to obtain where is a constant vector in G B g
T
= G
nf nc
R
+
.

k k
q
T
q
) ( G B g =

The Recursive formula for is obtained by replacing by in
Equation (29) and
k
q
T
) ( G B i 1 i
1 + i by 1 i in Equation (31) and taking partial derivative
with respect to yield
k
q

+ +
+ + =
) (
2 ) 1 (
) (
2 ) 1 ( ) 1 ( ) 1 (
) 1 (
) 1 (
) ( ) ( ) (
i i
k
i i
k k k
q l
q i
c
i
T
i i
q l
i
c
i q
T
i i q i i q i i
I
I
) S (G B
) S (G B G g
(32)

RecurDyn


+ + + =
) (
1 ) 1 (
) (
1 ) 1 ( ) 1 (
) 1 ( ) 1 (
) ( ) (
i i
k
i i
k k
q l
q i
c
i
T
i i
q l
i
c
i q
T
i i q i
I I
) S (G B ) S (G B S (33)
Since is a constant vector, . If
, matrices are not functions of . Therefore,
their partial derivatives with respect to become null. As a result, Equations
(32) and (33) can be simplified as follows.
nf nc
R
+
G
) (
k
q IV III
0 G =
k
q
k
q ) ( ) (
k k
q q i II B
k
q

=
) (
2 ) 1 ( ) 1 (
) 1 (
) ( ) (
i i
k k
q l
q i
T
i i q i i
I
S B g (34)

=
) (
) 1 ( 1
) 1 (
) ( ) (
i i
k k
q I l
q i
T
i i q i
S B S (35)

Since for the terminal bodies, for .
Thus, for , Equation (34) becomes
0 S =
k
q i
) (
) (
k
q i II
0 S =
k
q i
) ( ) ( ) (
k k
q q i IV II
) (
k
q IV

0 g =
k
q i i
) (
) 1 (
(36)

There are two recursive formulas in the case of body . If body
and body
) (
k
q i I
) (
k
q i I 1 + i belongs to set , and . Thus, Equation
(32) and (33) become
) (
k
q II 0 S =
k
q i
) (

+ =
) (
2 ) 1 ( ) 1 (
) 1 (
) ( ) (
i i
k k
q l
i
c
i q
T
i i q i i
I
) S (G B g (37)

+ =
) (
1 ) 1 ( ) 1 (
) 1 (
) ( ) (
i i
k k
q l
i
c
i q
T
i i q i
I
) S (G B S (38)

where must be saved when is computed. This recursive formula can
be applied to compute . As an example,
i
S G B
T
q
T
) ( G B ( )
34 34
) (
q
T
q
G B g =
34
q q
k
=
for the system
in Fig. 4 is obtained, as shown in Fig. 7 for the case of . Note that the
components of ( )
34
q
g are either zero or simple to compute.

4-17

Figure 7 Computation sequence of .
ii ii
q q
) ( ) ( BG g =

4.5. THE GOVERNING EQUATIONS OF SOLUTION

4.5.1 IMPLICIT INTEGRATION OF THE EQUATIONS OF MOTION

The dynamic equations of motion for a constrained mechanical system in the
joint space have been obtained in Reference [1] by the velocity transformation

RecurDyn

method as follows.
) Q Y (M B F
Z
+ =
T T &
(41)

where and , respectively, denote the cut joint constraint and the
corresponding Lagrange multiplier. The
vector including external forces, strain energy terms, and velocity induced forces.

M
is a mass matrix and is a force
The equations of motion, the constraint equations, v q = & , and
constitute the differential algebraic equations(DAE). Application of 'tangent
space method' in Reference [19] to the DAE yields the following nonlinear
system of equations
a v = &

0
) , z , v , (q
) v , (q
) , (q
) , , a , v , F(q
) a (v U
) v (q U
) H(p =
(
(
(
(
(
(
(
(
+ +
+ +
=
n n n n
n n n
n n
n n n n n
n n
T
n n
T
n
t
t
t
t
& &
&
,
2 0 0
1 0 0
(42)

where , , , and are determined by the coefficients
of the BDF. The must be chosen such that the augmented square matrix
is nonsingular. Applying Newton's method to solve the nonlinear system
in Equation (42) yields
|
T
T
n
T
n
T
n
T
n
T
n
a v q p , , , =
0
U
|
0
(
(
U
T
0

H p ) H(p =
n
(43)
,... 3 , 2 , 1 ,
) ( ) 1 (
= + =
+
i
i
n
i
n
p p p (44)
where

4-19

(
(
(
(
(
(
(
(
=
0
0 0
0 0 0
F F F F
0 U U 0
0 0 U U
) H(p
a v q
v q
q
a v q
& & & & & &
& &
T T
T T
n
0 0 0
0 0 0
(45)

Since and are highly nonlinear functions of q, and
q
F
q
a v,
, some
cautions must be taken in deriving the non-zero expressions in matrix so
that they can be efficiently evaluated.
p
H

4.5.2 APPLICATION OF THE GENERALIZED RECURSIVE FORMULAS

A set of the generalized recursive formulas has been developed in the sections
3 and 4. This section shows how these formulas can be utilized to efficiently
compute the in H in Equation (45). Inspection of reveals that partial
derivatives of F , , , , and are needed to be computed.
Only the is presented in this section and the rest can be derived similarly.
q
F
p
v
F
p
H
q a
F
q
&
q
& &
q
F
In Equation (41), differentiation of matrix with respect to vector q
results in a three dimensional matrix. To avoid the notational complexity for the
three dimensional matrix, Equation (41) is differentiated with respect to each
generalized coordinate one by one. Thus,
B
k
q

nf nr k
k k
k k
q
T
q
T
T T
q q
+ = + +
+ =
,..., 3 , 2 , 1 ), ( ) ( ) Q Y (M B
) Q Y (M B F
Z
Z
&
&
(46)

Since the term ) Q
Z

T
( can be easily expressed in terms of the Cartesian
coordinates,
k
q
T
) Q
Z
( is obtained by applying the chain rule as follows.

RecurDyn

nf nr k
k
T
q
T
k
+ = = ,..., 3 , 2 , 1 , ( ( B ) Q ) Q
Z
Z Z
(47)
where B q / Z =
G B
T
q
k
is used and B denotes the th column of the matrix .
The first term in Equation (46) can be obtained by applying the recursive
formula for with
k
k B
) Q Y (M G
Z
+ =
T &
. Collection of for all
k constitutes ) Q
Z

T
( , which is equivalent to ) Q
Z

T
( . Matrix
T
Z
)
G
T
Z
Q (
consists of nc+nf column vectors in . Therefore, the application of ,
where is each column of matrix
nf nc
R
+
B
T
q
k
G
T
Z
T
Z
) Q ( , yields the numerical result of
) Q
Z

T
( . Finally, the second term in Equation(46) is also obtained by applying
, where G B
T
q
k
) Q Y (M G
Z
+ =
T &
and
k
q
Y
&
is recursively obtained.


Dynamic analysis of a flexible slider crank mechanism is presented in order to
validate the results from the proposed method. The example problem is solved
by using both the proposed method and the nonlinear approach developed by
Simo [19].
The system consists of two rigid bodies and one flexible body, as shown in
Fig. 4. Length, cross-sectional area, and area moment of inertia of the elastic
crank are 0.4 m, 0.0018 m
2
, and 1.3510
-7
m
4
, respectively. The crank is
modeled by using 10 two-dimensional elastic beam elements of equal length.
The material mass density of the beam is 5540.0 kg/m
3
and its Young's modulus
is 1.0 times 10
9
N/m
2
. Vibration analysis of the crank is carried out with fixed-
free boundary condition and the resulting mode shapes are shown in Fig. 8, 9.
Four mode shapes are selected to span the deformation of the crank. As a result,
the system has 5 degrees of freedom.
Dynamic analysis is performed for 5 sec under the constant acceleration
condition of the joint between the ground and the body 1. The acceleration,
displacement, and relative deformation of the pin joint connecting the crank and
the coupler both from the proposed method and the nonlinear approach[19] are
shown in Figs.10,11 and 12, respectively. Note that since the results from both
models are almost identical as shown in these figures, the proposed
implementation methods using rigid virtual body can be validated.

4-21

Figure 8 Mode shapes of the crank

Figure 9 Mode shapes of coupler

RecurDyn


Figure 10 acceleration of Y 1 P

Figure 11 Relative deformation of 1 P

4-23

Figure 12 The strain energy of crank

4.7. CONCLUSION

This research extends the generalized recursive formulas for the rigid
multibody dynamics to the flexible body dynamics using the backward
difference formula(BDF) and the relative generalized coordinate. When a new
force or joint module is added to the general purpose program in the relative
coordinate formulations, the modules for the rigid bodies are not reusable for the
flexible bodies. In order to relieve the implementation burden, a virtual rigid
body is introduced at every joint and force reference frames and a virtual flexible
body joint is introduced between two body reference frames of the virtual and
original bodies. The notationally compact velocity transformation method is used
to derive the equations of motion in the joint space. The terms in the equations of
motion which are related to the transformation matrix are classified into several
categories each of which recursive formula is developed. Whenever one category
is encountered, the corresponding recursive formula is invoked. Since
computation time in a relative coordinate formulation is approximately
proportional to the number of the relative coordinates, computational overhead
due to the additional virtual bodies and joints is minor. Meanwhile,
implementation convenience is dramatically improved.

RecurDyn

REFERENCE

1. J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G. Teubner, Stuttgart, 1977.
2.W. Hooker, and G. Margulies, "The Dynamical Attitude Equtation for an n-body Satellite",
it Journal of the Astrnautical Science, Vol. 12, pp. 123-128, 1965.
3. R. Featherstone, "The Calculation of Robot Dynamics Using Articulated-Body
Inertias",it Int. J. Roboics Res., Vol 2, pp. 13-30, 1983.
4. D. S. Bae and E. J. Haug, "A Recursive Formulation for Constrained Mechanical
System Dynamics: Part II. Closed Loop Systems", it Mech. Struct. and Machines, Vol. 15,
No. 4, pp. 481-506.
5. T. C. Lin, and K. H. Yae, "Recursive Linearization of Multibody Dynamics and
Application to Control Design", it Technical Report R-75, Center for Simulation and
Mathematics, The University of Iowa, Iowa City, Iowa, 1990.
6. Ming-Gong Lee and E. J. Haug, "Stability and Convergence for Difference
Approximations of Differential-Algebraic Equations of Mechanical System Dynamics", it
Technical Report R-157, Center for Simulation and Design Optimization, Department of
Mechanical Engineering, and Department of Mathematics, The University of Iowa, Iowa
City, Iowa, 1992.
7. D. S. Bae, J. M. Han, H. H. Yoo, and E. J. Haug. "A Generalized Recursive Formulation
for Constrained Mechanical Systems", it Mech. Struct. and Machines, To appear.
8. A. A. Shabana, "Substructure Synthesis Methods for Dynamic Analysis of Multibody
Systems", it Computers & Structures, Vol. 20. No. 4, pp 737-744, 1985.
9. J. Garcia de Jalon, J. Unda, and A. Avello, "Natural Coordinates for the Computer
Analysis of Three-Dimensional Multibody Systems", it Computer Methods in Applied
10. N. Vukasovic, J. T. Celigueta, J. Garcia de Jalon, and E. Bayo, "Flexible Multibody
Dynamics Based on a Fully cartesian System of Support Coordinates", it Journal of
Mechanical Design, Vol. 115, pp. 294-299, 1993.
11. S. S. Kim and E. J. Haug, "A Recursive Formulation for Flexible Multibody

4-25

dynamics:Part I, Open loop systems", it Comp. Methods Appl. Mech.Eng, Vol. 71,
pp.293-314, 1988.
12. H. J. Lai, E. J. Haug, S. S. Kim, and D. S. Bae. "A Decoupled Flexible-Relative
Coordinate Recursive Approach for Flexible Multibody Dynamics", it International
Journal for Numerical Methods in Engineering, Vol. 32, pp.1669-1689, 1991.
13. A. Cardona and M. Geradin, "Modelling of Superelements in Mechanism Analysis", it
International Journal for Numerical Methods in Engineering, Vol. 32, pp.1565-1593,
1991.
14. P. E. Nikravesh and A. C. Ambrosio, "Systematic Construction of Equations of Motion
for Rigid-Flexible Multibody Systems Containing Open and Closed Kinematic Loops", it
International Journal for Numerical Methods in Engineering, Vol. 32, pp.1749-1766,
1991.
15. M. S. Pereira and P. L. Proenca, "Dynamic Analysis of Spatial Flexible Multibody
Systems Using Joint Coordinates", it International Journal for Numerical Methods in
Engineering, Vol. 32, pp.1799-1812, 1991.
16. ADAMS Reference Manual, Mechanical Dynamics, 2301 Commonwealth Blvd, Ann
Arbor, MI 48105.
17. A. A. Shabana, Dynamics of Multibody Systems, John & Wiley, New York, 1989.
18. Jeng Yen, E. J. Haug, and F. A. Potra, "Numerical Method for Constrained Equations
of Motion in Mechanical Systems Dynamics", it Technical Report R-92, Center for
Simulation and Design Optimization, Department of Mechanical Engineering, and
Department of Mathematics, The University of Iowa, Iowa City, Iowa2 1990.
19. J. C. Simo, and L. Vu-Quoc, "On the Dynamics of Flexible Beams Under Large Overall
Motions-The Plane Case: Part I", it Journal of Applied Mechanics, Vol. 53, pp. 849-854,
1986.

RecurDyn


5
RELATIVE NODAL METHOD FOR LARGE
DEFORMATION PROBLEM

5.1. INTRODUCTION

Geometrically nonlinear analyses[1-4] have been investigated by many
researchers. Their equations of equilibrium are based on either the total
Lagrangian formulation or the updated Lagrangian formulation. Since all
displacements are referred to the initial configuration in the total Lagrangian
formulation, the resulting equations of equilibrium are relatively simple.
However, if a structure undergoes a large displacement, some difficulties may be
encountered due to the nonlinearity associated with rotation. All displacements
are referred to the last calculated configuration in the updated Lagrangian
formulation and the rotational nonlinearity is relieved if the load increment is
small. The same difficulties as the total Lagrangian formulation can be
encountered in the case of a large load increment.
Avello[5] referred kinematic variables relative to the initial configuration and
he expressed the strains in a moving frame. Therefore, the strains were invariant
for finite rigid body deformations. Shabana[6-8] presented an absolute nodal
coordinate formulation for flexible multibody dynamics. All finite elements were
reformulated. Shimizu[9] considered the rotary inertia effects. This method is
based on the absolute nodal coordinate formulation.
Moving reference frame approaches were proposed by some researchers in
Refs. 10-14. A moving reference frame is introduced to represent a finite rigid
body motion. Deformation at a point of a flexible body was super-imposed on
the rigid body motion.

RecurDyn

RELATIVE NODAL METHOD FOR LARGE DEFORMATION PROBLEM
5.2. RELATIVE DEFORMATION KINEMATICS

5.2.1 GRAPH THEORETIC REPRESENTATION OF A STRUCTURE

This paper proposes to use the relative nodal displacements in formulating the
equations of equilibrium. Since the absolute nodal deformations are obtained by
accumulating the relative deformations along a path, element connectivity
information must be identified prior to generating the equations of equilibrium
for a general system. Therefore, a topology analysis must be carried out for a
structural system discretized into many finite elements.
0 1 2 3 4

Figure 1(a) A cantilever beam with five nodes
Forward path sequence
Backward path sequence
0 1 2 3 4

Figure 1(b) Graphic theoretic representation for the cantilever beam

The discretized systems can be represented by a graph. A node and an element
are represented by a node and an edge in the corresponding graph, respectively.
As an example, the graph theoretic representation for the system in Fig. 1(a) is
shown in Fig. 1(b). If a structure possesses a loop in its graph theoretic
representation, it is called as a closed loop system. Otherwise, it is called as an
open loop system.
A spanning tree denotes a graph which does not have a closed loop. A node
which does not have a child node is called as a terminal node. A node which
does not have a parent node is called as a base node. The terminal node and the
base node for the system in Fig. 1(b) are nodes 4 and 0, respectively. Two

5-3

computational sequences must be defined in the proposed relative displacement
formulation. One is the forward path sequence which traverses a graph from the
base node towards the terminal nodes. The other is the backward path sequence
which is the reverse of the forward path sequence. Two sequences for the graph
in Fig. 1(a) are shown in Fig. 1(b).

5.2.2 KINEMATIC DEFINITIONS

Consider a system consisting of two beam finite elements as shown in Fig.
2(a) and (b). Nodes i and are assumed to be inboard nodes of nodes
and in a graph, as shown in Fig. 2(b), respectively. is the
inertial reference frame and
1 i
i
1 + i Z Y X
k k k
z y x ) , ( j i k = is the nodal reference frame
attached to a node , and is a position vector of the node k
k
r k .
is the reference frame attached to a node and the first
subscript denotes the inboard node number of the second subscript . The
orientation of coincides with that of in
the undeformed state. The absolute nodal displacements measured in the
frame have been solved for in the conventional finite element analysis
methods(see Refs. 1-4). In contrast to conventional methods, the relative nodal
displacements measured in its inboard nodal reference frame are solved in this
paper.
i i i i i ) 1 ( ) 1 ) 1 (
z x

1 i
i 1 (
x

X
i (
y
Z Y
i
) 1 ( ) 1
z y

i
i
( i i ) i 1 (
z
i i ) 1 (
y
i )

) 1 (
x
i
i
x
i
z
i
y
1
x
i
1
z
i
1
y
i
i-1
i
X
Z
Y
i
r
1 i
r
i i ) 1 (
x

i i ) 1 (
z

i i ) 1 (
y

i+1

RecurDyn

Figure 2(a) Two finite beam elements
Forward path sequence
Backward path sequence
L L
i i-1 i+1

Figure 2(b) Graphic theoretic representation for the beam elements

The generalized coordinates for the relative nodal position and orientation
displacements of a node are denoted by u and , respectively. The
nodal position and orientation of node in the
'
) 1 ( i i
i
'
) 1 ( i i
Y X Z frame can be
expressed in terms of these of node 1 i and the relative nodal displacements as
follows:
( )
'
) 1 (
'
0 ) 1 ( ) 1 ( ) 1 ( i i i i i i i
+ + = u s A r r (1)
i i i i i i i i ) 1 (
'
) 1 ( ) 1 ( 1
) (

= C D A A (2)
where
[
T
i i i i i i i i
'
3 ) 1 (
'
2 ) 1 (
'
1 ) 1 (
'
) 1 (
= ] (3)

In Eqs.(1) and (2),
k
A ) , 1 ( i i k = denotes the transformation matrix for nodal
reference frame , denotes the constant transformation matrix from
to , denotes the location vector of node
measured in in the undeformed state, and denotes the
deformation vector of node relative to the nodal frame
k
i ) 1
)

i i ) 1 (
i i i ) 1 ( ) 1
z

) 1 (
z

i
i
C
i (
y
) 1 (
y
i
i i i
z y x
i (
x
1 ( i
'
0 ) 1 ( i i
s i
x
'
) 1 ( i i
u
1 i
i i ) 1 (
x

. is the
transformation matrix due to a rotational displacement of
relative to the nodal frame
i i ) 1 (
D
i i ) 1 (
y
i ) i 1 (
z

1 i and can be expressed by the 1-2-3 Euler angle

as
) ( ) ( ) (
'
3 ) 1 ( 3
'
2 ) 1 ( 2
'
1 ) 1 ( 1 ) 1 ( i i i i i i i i
= D D D D (4)

Taking a variation of Eq. (1) yields

( )
'
) 1 ( ) 1 (
'
) 1 (
'
) 1 (
'
0 ) 1 ( ) 1 (
'
) 1 ( ) 1 (
' ~ ~
i i
T
i i i i i i i
T
i i i
T
i i i
+ + = u A u s A r A r (5)

5-5

where a symbol with tilde denotes a skew symmetric matrix which consists of its
vector elements, and is defined as
i i ) 1 (
A
i
T
i i i
A A A
) 1 ( ) 1 (
= (6)

The virtual rotation relationship between nodes and i 1 i is given as

'
) 1 ( ) 1 ( ) 1 (
'
) 1 ( ) 1 (
'
i i i i
T
i i i
T
i i i
+ = H A A (7)
where
) cos( ) cos( ) sin( 0

) cos( ) sin( ) cos( 0
) sin( 0 1
'
2 ) 1 (
'
1 ) 1 (
'
1 ) 1 (
'
2 ) 1 (
'
1 ) 1 (
'
1 ) 1 (
'
2 ) 1 (
) 1 (
i i i i i i
i i i i i i
i i
i i

H (8)

Combining Eqs.(5) and (7) yields the following recursive virtual displacement
equation for a pair of contiguous elements.:

i i i i i i i i ) 1 ( 2 ) 1 ( ) 1 ( 1 ) 1 (
+ = q B Z B Z (9)

where
[ ] ) , 1 ( ,
'
i i k
T
T
k
T
k k
= = r Z (10)
[ ]
T
T
i i
T
i i i i
'
) 1 (
'
) 1 ( ) 1 (
= u q (11)
I 0
u s I
A 0
0 A
B
)
~ ~
(
'
) 1 (
'
0 ) 1 (
) 1 (
) 1 (
1 ) 1 (
i i i i
T
i i
T
i i
i i
(12)
i i
T
i i
T
i i
i i
) 1 ( ) 1 (
) 1 (
2 ) 1 (
H
I
A 0
0 A
B (13)

It is important to note that matrices and are only functions of the
relative displacement between nodes
1 ) 1 ( i i
B
2 ) 1 ( i i
B
i i ) 1 (
q 1 i and i .
The virtual displacement relationship between the absolute and relative nodal
coordinates for the whole system can be obtained by repetitive application of Eq.
(9) along a chain in a graph. As an example, the virtual displacement relationship

RecurDyn

for the Cartesian and relative coordinate systems in Fig. 1 is as follows.

q B Z = (14)
where
[
T
T T T T
4 3 2 1
Z Z Z Z Z = ]
]
(15)
[
T
T T T T
34 23 12 01
q q q q q = (16)
=
342 232 341 122 231 341 012 121 231 341
232 122 231 012 121 231
122 012 121
012
B B B B B B B B B B
0 B B B B B B
0 0 B B B
0 0 0 B
B (17)

5.3. GOVERNING EQUATIONS OF EQUILIBRIUM

5.3.1 STRAIN ENERGY

The strain energy in a finite element having multiple nodes is affected only by
the relative displacements of nodes relative to the inboard nodal frame of the
element and is free from its rigid body motion. As a result, the variational form
of the strain energy for a system can be obtained in a summation form as
=

= =
n
k
T
k k k k
T
k k
W
1
) 1 ( ) 1 ( ) 1 (
Kq q q K q (18)

where q must be kinematically admissible for all constraints. Since the
stiffness matrix is generated in the nodal reference frame, the strain energy due
to a rigid body motion of a node does not appear in Eq. (18). The element
stiffness matrix is contributed from linear and nonlinear terms as (see
Ref. 3)
k k ) 1 (
K

nL
k k
L
k k k k ) 1 ( ) 1 ( ) 1 (
+ = K K K (19)
where

=
k k
l
k k
L
k k
T
k k
L
k k
dx
) 1 (
0
*
) 1 ( ) 1 (
*
) 1 ( ) 1 (
K (20)

5-7

+

=
) 1 (
0
*
) 1 ( ) 1 ( ) 1 (
*
) 1 ( ) 1 (
) (
i i
l
k k k k
nL
k k
T
k k
nL
k k
dx q K (21)

In Eqs. (19) - (21), denotes a linear stiffness matrix, denotes a
nonlinear stiffness matrix, and l denotes the undeformed length of the
element between the nodes and . Note that the significance of
depends on the magnitude of . becomes negligible when the
magnitude of is small, which is true when the element size is small. It is
very difficult analytically to prove the significance of . As a consequence,
the significance of has been demonstrated through a numerical example
in 5.
L
k k ) 1 (
K
nL
k k ) 1 (
K
nL
k k ) 1 (
K
k k ) 1 (
k k ) 1 (
q
1 k k
nL
k k ) 1 (
K
nL
k k ) 1 (
K
k k ) 1 (
q
nL
k k ) 1 (
K

5.3.2 EXTERNAL FORCE

The virtual work done by both nodal forces described in the absolute
nodal coordinate system and described in the relative nodal coordinate
system is obtained as follows:
Q
R

R q Q Z
T T
W + = (22)

where Z must be admissible for the kinematic relationship between Z and
q . Substitution of q B Z = into Eq. (22) yields

( )
*
Q q R Q B q
T T T
W = + = (23)
where
R Q B Q + =
T *
(24)

RecurDyn

5.3.3 CONSTRAINT
cut
0
2
3
4
5
1

Figure 3 A closed loop system

A nodal displacement is measured relative to its inboard nodal frame in the
proposed method. The relative nodal displacement can be defined only in
structures having a tree topology. Therefore, if a structural system has a closed
loop, it must be opened to form the tree topology. The cut joint method (see Ref.
12) is employed to treat the closed loops. A node in a closed loop is removed and
the corresponding cut constraint equations are introduced to compensate for the
removed node. As an example, Fig. 3 shows a closed loop system. The graphical
representation of the system is presented in Fig. 4.
0
1 5
2 4
3
cut

Figure 4 Graphic representation of the system of the closed loop system in Fig. 3

5-9

0 2 3 1 4 5

Figure 5 Tree structure corresponding to the system in Fig. 3

A cut has been made at node 5 to form the tree structure shown in Fig. 5. The cut
constraint can be formulated from the geometric compatibility relationships.
From Eqs. (1) and (2), the position and orientation matrix of node 5 is obtained
along the forward path sequence as
( )
*
5
'
) 1 (
'
0 ) 1 (
5
1
) 1 ( 5
r u s A r = + =

=
k k k k
k
k
(25)
=

= =
5
1
*
5 ) 1 ( ) 1 ( 0 5
k
k k k k
A C D A A (26)

where and are given by the boundary conditions at node 5. Since Eq.
(26) comprises of nine dependent equations, only three are independent. The
three independent constraint equations can be extracted by imposing
perpendicularity between the axes of reference frames. As a result, the six
independent constraint equations are given as
*
5
r
*
5
A

=
*
52 51
*
52 53
*
51 53
*
5 5
a a
a a
a a
r r
T
T
T
(27)

where
[ ]
53 52 51 5
a a a A = (28)
[ ]
*
53
*
52
*
51
*
5
a a a A = (29)

In Eqs. (28) and (29), and
i 5
a
*
5i
a ) 3 , 2 , 1 ( = i denote the -th column vector
of and , respectively.
i
5
A
*
5
A

RecurDyn


5.3.4 EQUATIONS OF EQUILIBRIUM

For a closed loop system, relative deformation q is not independent, and q
must satisfy the constraint Eq. (27). Taking variation of Eq. (27) yields

0 q
q
= = (30)

The Lagrange multiplier theorem (see Refs. 12 and 15) can be applied to obtain
the following equations of equilibrium for a constrained system:

( ) 0 Q Kq q
q
= +
T T *
(31)

where the q is arbitrary. Since q is arbitrary, its coefficient must be zero,
which yields
( ) 0 Q Kq q F
q
= + =
*
,
T
(32)

Since the number of equations is less than that of unknown variables in Eq. (32),
the unknown variables cannot be determined. Thus, constraint equations given in
Eq. (27) are supplemented to find the solution of and q . Deformations
can be obtained by solving Eqs. (27) and (32) simultaneously. Since the , ,
and in the equations are the nonlinear function of , can be solved by
using Newton-Raphson method as
q
q

*
Q q q

F q
0
F
q
q q
T
(33)
where
( )
q
q q
Q K F
*
+ =
T
(34)

By solving Eq. (33), the improved solution of for the next iteration can be
obtained as follows:
q
q q q + = (35)

5-11

By using Eqs. (33) and (35), the iteration continues until the solution variance
remains within a specified allowable error tolerance. Before solving Eq. (33), it is
necessary to calculate . However, the calculation of F is numerically
difficult and tedious.
q
F
q
In order to save computing time in solving Eq. (33), some numerical
approximation techniques may be applied. As an example, the coefficient matrix
of Eq. (33) may remain near constant if the variation of is small, which is the
case when the lengths of finite elements are small. In such case, the coefficient
matrix of Eq. (33) can be hold during Newton-Raphson iterations, which
significantly reduces the computation time. However, the approximation
technique may not converge for a system whose is large. To overcome this
numerical difficulty, a combined incremental and iterative method (see Ref. 16)
can be used.
q
q

5.4. NUMERICAL ALGORITHM

Kinematics of the relative nodal displacements and the equations of
equilibrium are presented in the section 3. This section explains how the
equations are implemented to obtain the relative and absolute nodal
displacements of a structure. The numerical algorithm for closed loop systems is
as follows:

1) Perform the graph theoretic preprocessing to determine computational path
sequences.
2) Form a stiffness matrix . K
3) Compute , , and for in the backward path sequence.
q
*
Q
k
q
4) Solve the Eq. (37) to obtain and q .
5) If and F q remains within the specified allowable error tolerance, then
go to step 6. Otherwise, improve the solution using Eq. (35). Go to step 3.
6) Compute the Cartesian deformations in the forward path sequence by using
Eqs. (1) and (2).

RecurDyn



Static analysis of a cantilever beam subjected to end moment M , as shown in
Fig. 6 is carried out.

M
X
Y
] [ 0 . 1
] [ 0 . 1
] [ 0 . 12
0 . 0
] / [ 10 0 . 3
4
2
2 7
m I
m A
m L
m N E
=
=
=
=
=
L

Figure 6 A cantilever beam subjected to end moment

-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
-2
0
2
4
6
8
10
12
14
16
Undeformed
Proposed
ANSYS: nonlinear
ANSYS: linear
Y

[
m
]
X [m]

Figure 7 Deformed shape of the beam

5-13

In the figure, E , , L , , and denotes Young's modulus, Poisson ratio,
the length of the beam, the cross sectional area of the beam, and the second area
moment of the cross section, respectively. M =6.54510
6
[Nm] is applied at
the end node. Fig. 7 shows the deformed shapes of the beam by the proposed
method by the proposed method and a commercial program ANSYS. In the
figure, Proposed, ANSYS: nonlinear, and ANSYS: linear denote numerical
results by the proposed method, a commercial program ANSYS using nonlinear
analysis, and ANSYS using linear analysis, respectively. It shows that the
numerical results obtained by the proposed method and ANSYS(nonlinear
analysis) are almost identical, but the numerical results by ANSYS(linear
analysis) shows large difference with the remaining two numerical results.

A I
0 2 4 6 8 10
-10
-8
-6
-4
-2
0
u
e
: Proposed
u
e
: ANSYS
u
e

[
m
]
The number of elements

Figure 8 Convergence of axial deformation at the end node vs. the number of elements

Fig. 8 shows the convergence of the axial deformation at the end node.
When fewer elements are used for static analysis, the numerical results of
ANSYS are more accurate than those of the proposed method, but obtained
e
u
e
u

RecurDyn

by the proposed method converges rapidly as the number of elements is
increased. In the figure, the numerical results with more than 6 elements by the
two methods are almost identical. From the analysis results, it is known that the
effect of the nonlinear stiffness matrix is diminished rapidly as the number of
elements is increased.

-1 0 1 2 3 4 5 6 7 8
0
1
2
3
4
(F = [3.0E+4, -1.0E+04 ]
T
[N], M= 0.0E+0[Nm])
Undeformed shape
Deformed shape: 20 elements
Deformed shape: ANSYS
Y

[
m
]
X [m]

F
Y
M
X
] [ 002 . 0
] [ 01 . 0
] [ 14 . 14
0 . 0
] / [ 10 0 . 3
4
2
2 7
m I
m A
m L
m N E
=
=
=
=
=
L
P
Figure 9 A closed loop system subjected to concentrated force and Moment
Figure 10 Comparison of deformed shapes of the closed system

5-15

Fig. 9 shows a closed loop system subjected to a concentrated force and
moment
F
M at a point P . When =[310 F
4
-110
4
]
T
[N] and M =0.0
[Nm] are applied at the point P the deformed shapes of the system are shown
in Fig. 10. It shows that the numerical results obtained by the proposed method
with 20 elements and a commercial program ANSYS are almost identical.
-1 0 1 2 3 4 5 6 7 8
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
(F=[3.0E+4, -3.0E+4]
T
[N], M=3.0E+4[Nm])
Y
[
m
]
X[m]
Undeformed shape
Deformed shape: 20 elements

Figure 11 Undeformed and deformed shapes of the closed loop system

0 1 2 3 4 5 6 7 8
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
(F=[3.0E+4, -3.0E+4]
T
[N], M=3.0E+4[Nm])
Y
[
m
]
X[m]
0.33F, 0.33M
0.67F, 0.67M
1.00F, 1.00M

Figure 12 Deformed shapes of the closed loop system at each load step

RecurDyn


When =[310 F
4
-310
4
]
T
and M =3.010
4
[Nm] are applied at the point
P, the deformed shape of the system is shown in Fig. 11. While the numerical
solution by the proposed method converges after the 7-th iteration, that by the
commercial program ANSYS does not converge. Fig. 12 shows the deformed
shape of the system at each load step.

5.6. CONCLUSIONS

A geometric nonlinear formulation for structures undergoing large
deformations is investigated in this research. Nodal displacements in the
proposed method are referred to its adjacent nodal reference frame. Since the
nodal displacements are measured relative to its inboard nodal frame, quantity of
the nodal displacements is still small for a structure undergoing large
deformations if the element sizes are small. Relative coordinate kinematics is
developed to define relative position and orientation of the nodal displacements.
As a consequence, many element formulations developed under small
deformation assumptions are reusable for structures undergoing large
deformations, which makes it easy to develop a computer program. A structural
system is represented by a graph to systematically develop the governing
equations of equilibrium for general systems. Closed loops are opened to form a
tree topology by cutting nodes. Two computational sequences are defined for a
graph. One is the forward path sequence that is used to recover the Cartesian
nodal deformations from relative nodal displacements and traverses a graph from
the base node towards the terminal nodes. The other is the backward path
sequence that is used to recover the nodal forces in the relative coordinate
system from the known nodal forces in the absolute coordinate system and
traverses from the terminal nodes toward the base node. A solution algorithm is
developed to implement the proposed method. Static analyses are performed for
structures undergoing large deformations. The proposed method can solve the
problem which cannot be solved by the commercial program ANSYS.

5-17

REFERENCES

(1) El Damatty, A. A., Korol, R. M and Mirza, F. A., "Large Displacement Extension of
Consistent Shell Element for Static and Dynamic Analysis," Computers & Structures, Vol.
62, No. 6, (1997), p. 943-960.
(2) Mayo, J and Domnquez, J., "A Finite Element Geometrically Nonlinear Dynamic
Formulation of Flexible Multibody Systems using a New Displacements
Representation," J. Vibration and Acoustics, Vol. 119, (1997), p.573-580.
(3) Dhatt, G and Touzot, G., The Finite Element Method Displayed, John Wiley & Sons,
(1984).
(4) Bathe, K. J., Finite Element Procedures, Prentice-Hall, (1996).
(5) Avello, A. J., Joln, G. D. and Bayo, E., "Dynamics of Flexible Multibody Systems
using Cartesian Co-ordinates and Large Displacement Theory," Int. J. Numer. Methods
Eng., Vol. 32, No. 8, (1991), p.1543-1564.
(6) Shabana, A. A, "An Absolute Nodal Co-ordinate Formulation for the Large Rotation
and Deformation Analysis of Flexible Bodies," Technical Report MBS 96-1-UIC,
Department of Mechanical Engineering, University of Illionois at Chicago, (1996).
(7) Shabana, A. A. and Christensen, A., "Three Dimensional Absolute nodal coordinate
formulation: Plate Problem," Int. J. Nuner. Methods Eng., Vol. 40, No. 15, (1997),
p.2275-2790.
(8) Shabana, A. A., Dynamics of Multibody Systems, 2nd edition, Cambridge University
Press, (1998).
(9) Takahashi, Y. and Shimizu, N, Study on Elastic Forces of the Absolute Nodal
Coordinate Formulation for Deformable Beams, Proceedings of the ASME Design
Engineering Technical Conferences, (1999).
(10) Featherstone, R., "The Calculation of Robot Dynamics Using Articulated-Body
Inertias, " Int. J. Roboics Res., Vol. 2, (1983), p. 13-30.
(11) Bae, D. S. and Haug, E. J., "A Recursive Formulation for Constrained Mechanical
System Dynamics: Part I. Open Loop Systems," Mech. Struct. and Machines, Vol. 15, No.

RecurDyn

3, (1987), p. 359-382.
(12) Bae, D. S. and Haug, E. J., "A Recursive Formulation for Constrained Mechanical
System Dynamics: Part II. Closed Loop Systems," Mech. Struct. and Machines, Vol. 15,
No. 4, (1987), p. 481-506.
(13) Lin, T. C. and Yae, K. H., Recursive Linearization of Multibody Dynamics and
Application to Control Design, Technical Report R-75, Center for Simulation and
Mathematics, The University of Iowa, Iowa City, Iowa, (1990).
(14) Yoo, H., Ryan, R. and Scott, R., "Dynamics of Flexible Beams undergoing Overall
Motion," J. Sound and Vibration, Vol. 181, No. 2, (1995), p.261-278.
(15) Haug, E. J., Computer-Aided Kinematics and Dynamics of Mechanical Systems:
Volume I. Basic Methods, Allyn and Bacon, (1989).
(16) Crisfield, M. A., Non-Linear Finite Element Analysis of Solids and Structures, Wiley,
(1997).

5-19

RecurDyn


6

DYNAMIC ANALYSIS OF HIGH-MOBILITY
TRACKED VEHICLES

6.1. INTRODUCTION

High-speed, high-mobility tracked vehicles are subjected to impulsive
dynamic loads resulting from the interaction of the track chains with the vehicle
components and the ground. These dynamic loads can have an adverse effect on
the vehicle performance and can cause high stress levels that limit the
operational life of the vehicle components. For this reason, high speed, high
mobility tracked vehicles have sophisticated suspension systems, a more
elaborate and detailed design of the links of the track chains, and improved
vibration characteristics that allow the vehicle to perform efficiently in hostile
operating environments.
Galaitsis [1] demonstrated that the predicted dynamic track tension and
suspension loads in a high speed tracked vehicle developed by an analytical
method are useful in evaluating the dynamic characteristics of the tracked
vehicle components. The predicted track tension was compared with the
measured data from a military tracked vehicle. Bando et al [2] developed a
planar computer model for rubber tracked bulldozers. Steel and fiber molded
continuous rubber track is discretized into several rigid bodies connected by
compliant force elements. Characteristics of track damage, vibration, and noise
are investigated using the simulation results. Nakanishi and Shabana [3]
developed a two-dimensional contact force model for planar analysis of
multibody tracked vehicle systems. The stiffness and damping coefficients in
this contact force model were determined based on experimental observations of
the overall vibration characteristics of the tracked vehicle. The nonlinear
equations of motion of the vehicle were obtained using the Lagrangian approach
and the algebraic constraint equations that describe the joints and specified

RecurDyn

DYNAMIC ANALYSIS OF HIGH-MOBILITY TRACKED VEHICLES
motion trajectories are adjoined to the system equations of motion using the
technique of Lagrange multipliers. The generalized contact forces associated
with the system generalized co-ordinates were obtained using the virtual work.
Choi [4] presented a large-scaled multibody dynamic model of construction
tracked vehicle in which the track is assumed to consist of track links connected
by single degree of freedom pin joints. In this detailed three-dimensional
dynamic model, each track link, sprocket, roller, and idler is considered as a
rigid body that has a relative rotational degree of freedom. Scholar and Perkins
[5] developed an efficient alternative model of the track chains considering
longitudinal vibrations. The track is assumed to consist of a finite number of
segments and each is modeled as a continuous uniform elastic rod attached to the
vehicle wheels. Each segment consists of several track links which are
collectively lumped as a single body so that overall chain stretching effects are
accounted for.
A detailed three-dimensional tracked vehicle model as the one developed by
Choi [4] may have hundreds or thousands differential and algebraic equations.
These equations are highly non-linear and can only be solved using matrix,
numerical and computer methods. In addition to this dimensionality problem,
tracked vehicles are characterized by impulsive forces due to the contacts
between the track links and the vehicle components as well as the ground. The
impulsive contact forces cause serious numerical problems when the vehicle
equations of motion are integrated numerically. The degree of difficulty may
significantly increase if compliant elements, instead of the ideal pin joints, are
used to model the connection between the links of the track chains, as it is the
case in this investigation. The compliant elements must have very high stiffness
coefficients in order to maintain the link connectivity. These stiffness
coefficients, which are determined experimentally in this investigation, introduce
high frequency oscillatory components to the solution, thereby forcing the
numerical integration routine to take a very small time step size. It is, therefore,
important to adopt a numerical scheme that can be efficiently used in modeling
this type of vehicle. Newmark [6] presented an absolutely stable second-order
numerical integrator in the area of structural dynamics. The Newmark
integrator was modified by Wilson [13] so that highly oscillatory state variables
are numerically damped out. The numerical damping algorithms are extended

6-3

and generalized in implicit and explicit forms with a constant step size by Chung
[7,8]. The algorithm developed by Chung is employed in this investigation due to
its easy implementation and large stability region.
The objective of this investigation is to develop a computational procedure for
the nonlinear dynamics of high-speed, high-mobility tracked vehicles. The
model developed in this investigation differs from the low-speed tracked vehicle
model previously developed by Choi [4] in two important aspects summarized as
follows:

(1) The high speed tracked vehicle considered in this investigation has a
sophisticated suspension system that consists of road arms and wheels instead
of the simple roller type suspension system previously developed by Choi.
(2) In the model previously developed by Choi [4], the links of the track
chains are connected by pin joints that have one degree of freedom. In the model
developed in this investigation, compliant force elements are used to model the
connectivity between the links of the track chains. The characteristics of these
compliant elements are determined experimentally as discussed in Section 5.
The application of the numerical integration scheme developed by Chung
[7,8] to tracked vehicle dynamics is investigated in this paper using different
simulations scenarios that include accelerated motion, high speed motion,
braking and turning motion.

6.2. HIGH-SPEED, HIGH-MOBILITY TRACKED VEHICLES

In this section, the high-speed, high-mobility tracked vehicle model used in
this investigation is described. The three-dimensional model, which is shown in
Fig. 1, represents the third generation of a military vehicle weighing
approximately 50 tons and can be driven at a speed higher than 60 km/h.

RecurDyn


Figure 1. High mobility tracked vehicle.

The vehicle consists of a chassis and two track subsystems. The chassis
subsystem includes a chassis, sprockets, support rollers, idlers, road arms, road
wheels and the suspension units. The sprockets, support rollers, and road arms
are connected to the chassis by revolute joints. The suspension unit includes a
Hydro-pneumatic Suspension Unit (HSU)[17], and torsion bar that are modeled
as force elements whose compliance characteristics are evaluated using
analytical and empirical methods. The HSU systems are mounted on front and
rear stations to damp out pitching motion and to decrease the vehicle speed when
the vehicle is running over large obstacles. The spring torque of the HSU
systems can be written as
1 HSU
PAL T = (1)

where P is the gas pressure, is the area of piston, and is the distance
shown in Fig. 2. The pressure
A
1
L
P in the gas chamber of HSU system with
respect to rotation angle of a road arm is defined as

+
=
) (
2 2
L L l
l
P P
i s
i
i
(2)

where , , and are the initial pressure and distances when the road arm
is in its initial configuration,
i
P
i
l
i
L
2
is a constant which is equal to 1.4, and is
the distance shown in Fig. 2.
2
L

6-5

Figure 2. Schematic diagram of spring-damper suspension units: hydro pneumatic
suspension unit and torsion bar systems.

The distance l can be adjusted by charging or discharging oil into the oil
chamber. The torsion bars are mounted on the middle stations for this vehicle
model. A simple torsional spring model is used in this investigation to represent
the stiffness of the torsional bars. The stiffness coefficient of the torsion bar
spring is approximately Nm/rad. Figure 2 shows the schematic diagram
of the HSU and the torsion bar systems. Figure 3 shows the spring characteristics
which are employed in this investigation.
s
4
10 5

Figure 3. Spring characteristics of suspension unit

Each track subsystem is modeled as a series of bodies connected by rubber
bushings around the link pins which are inserted into a shoe plate with some
radial pressure in order to reduce the non-linear effect of the rubber. When the

RecurDyn

vehicle runs over rough surfaces, the track chains are subjected to extremely
high impulsive contact forces as the result of their interaction with the vehicle
components such as road wheels, idlers, and sprocket teeth, as well as the ground.
The rubber bushings and double pins tend to reduce the high impulsive contact
forces by providing cushion and reducing the relative angle between the track
links. About 10 percent of the vehicle weight is given as the pre-tension for the
track to prevent frequent separations of the track when the vehicle runs at a high
speed. About 14 degrees of a pre-torsion is also provided in order to reduce the
fluctuation of the torque in the rubber bushing when the track links contact the
sprocket and idler.
The vehicle, which presents a high-mobility military tracked vehicle, consists
of one hundred eighty nine bodies; body 1 is the chassis, bodies 2 and 3 are the
right and left driving sprockets, bodies 4 - 17 are the right and left arms, bodies
18 - 31 are the right and left wheels, bodies 32 - 37 are the right and left support
rollers, and bodies 38 - 113 and 114 - 189 are the right and left track links,
respectively. The sprockets, rollers and arms are connected with chassis by 22
revolute joints, and wheels are connected with arms by 14 revolute joints, each
of which has one degree of freedom. This vehicle model has 152 bushing
elements between track links, and 954 degrees of freedom.

6.3. KINEMATIC RELATIONSHIPS AND EQUATIONS OF MOTION

In this investigation, the relative generalized co-ordinates are employed in
order to reduce the number of equations of motion and to avoid the difficulty
associated with the solution of differential and algebraic equations. Since the
track chains interact with the chassis components through contact forces and
since adjacent track links are connected by compliant force elements, each link
in the track chain has six degrees of freedom which are represented by three
translational co-ordinates and three Euler angles [9]. Recursive kinematic
equations of tracked vehicles were presented by Choi [4,16], who showed that
the relationship between the absolute Cartesian velocities of the chassis
components can be expressed in terms of the independent joint velocities as

6-7

r
i
q B& = q& (3)

where , , and q are relative independent co-ordinates, velocity
transformation matrix, and Cartesian velocities of the chassis subsystem,
respectively. The equations of motion of the chassis that employs the velocity
transformation defined in the preceding equations are given as follows:
r
i
q B

) q B M Q B q MB B
r
i
T r
i
T
( &
&
& & = (4)

where is the mass matrix, and is the generalized external force vector
of the chassis subsystem. Since there is no kinematic coupling between the
chassis subsystem and the track subsystems, the equations of motion of the
chassis subsystem can be obtained using the preceding equation as follows:
M Q

C
i
r
i
C
i
Q q M = & & (5)

where , . MB B M
T C
i
= ) (
r
i
T C
i
q B M Q B Q &
&
=
For the track subsystems, the equations of motion can be written as

t t t
Q q M = & &

(5)

where , and denote the mass matrix; and the generalized
coordinate and force vectors for the track subsystem, respectively. Consequently,
the accelerations of the chassis and the track links can obtained by solving
Equations (5) and (6).
t
M
t
q
t
Q

6.4. A COMPLIANT TRACK MODEL

RecurDyn

Two models can be used to connect the track links of the high-mobility
tracked vehicle chains. These two models are shown in Fig. 4. In the first model,
shown in Fig. 4(a), a single pin is used to connect two links of the chain. In the
second model, shown in Fig. 4(b), two pins are used to connect the track links. In
both models, rubber bushings are inserted between the pins and the track links,
and as a consequence, the relative rotations between the pins and the links are
relatively small. In this section, the force models used for the single pin and
double pin connections are described.

(a) Single pin track links (b) Double pin track links

Figure 4. Track links of high mobility tracked vehicle

6.4.1 SINGLE PIN CONNECTION

Figure 5. Single pin connection

6-9

Figure 5 shows the details of the link, pin and bushings connection of a single
pin track link. In this investigation, a continuous force model is used to define the
pin joint connections. This force model is a non-linear function of the co-
ordinates of the two links. In order to define the generalized compliant bushing
forces, several coordinate systems are introduced. Two centroidal body
coordinate systems and for the track links and
i
b
i
b
i
b
Z Y X
j
b
j
b
j
b
Z Y X i j ,
respectively; a joint coordinate system
i i i
Z Y X whose origin is assumed to be
located at the geometric center of the circular groove containing the pin and the
bushing; and a pin coordinate system
j j j
Z Y X whose origin is rigidly attached
to the center of the pin. Note that because of the bushing effect, the origins of the
joint and pin coordinate systems do not always coincide. The displacement of the
pin coordinate system
j j j
Z Y X with respect to the joint coordinate system
i i i
Z Y X is a function of the bushing stiffness. Also note that the location and
orientation of the joint coordinate system
i i i
Z Y X can be determined as a
function of the generalized co-ordinates of link . For simplicity, it is assumed
in this investigation that the location and orientation of the pin coordinate system
can be defined in terms of the co-ordinates of link
i
j . The deviation
shown in Figure 5 can be used to determine the generalized
forces acting on the two links and
T
z y
] , ,
x R
[ =
i j as the result of the bushing effect. The
bushing force and torque applied to the frame j are given as follows:

0
0 C
K
K
Q
Q
&
&
R
R R R
j
j
R
C 0
0

where and are the 3
R R
, , C K K

C 3 diagonal matrices that contain the
stiffness and damping coefficients of the bushing, and is translational force
vector and is the vector of translational deformations of the frame
j
R
Q
R
j relative
to the frame . Similarly, is the rotational force vector and is the
vector of relative rotational deformations of the frame
i
j
Q

j relative to the frame .
The force and torque applied to the frame i are assumed to be equal in
i

RecurDyn

magnitude and opposite in direction to the force and torque acting on frame j .
Once these forces are determined, the generalized bushing forces associated with
the generalized co-ordinates of the track links and i j can be determined.

6.4.2 DOUBLE PIN CONNECTION

Figure6. Double pin connection

In the double pin assembly, shown in Figure 6, two adjacent track links are
connected with a connector element using two pins and rubber bushings. The
mass and mass moment of inertia of the connector element are relatively small as
compared to those of the track links. Therefore, the dynamic effects of connector
element are modeled in this investigation. This approach has the advantage of
reducing the number of degrees of freedom of the system. The double pin
assembly can be modeled by considering one radial, one axial, and three
rotational springs. The radial spring provides the restoring force due to the
combined translational deformation of the two rubber bushings along the radial
direction of the connector as shown in Fig. 6. The axial spring restricts the
translational motion of the two links along the lateral direction as shown in Fig.
6. The rotational springs are used to model the relative rotational deformation

6-11

between the two track links. The length of the radial spring is assumed to be
the distance between the origins of the coordinate systems
l
i i i
Z Y X and
j j j
Z Y X shown in Fig. 6. This distance is defined as
( )
ij ij 2
d d l
T
= (8)

The magnitude of the force produced by the radial spring is

( ) l C l l K F
r 0 r r
&
+ = (9)

where is the spring stiffness coefficient, and is the damping coefficient,
and is obtained by differentiating Equation (8) with respect to time. Similarly,
the restoring force due to the translational spring along the
r
K
r
C
i
i
Z axis is

i , j
z Rz
i , j
z Rz z
C K F =
&
(10)

where is translational deformation of the
i , j
z
j j j
Z Y X frame with respect to
the
i i i
Z Y X frame along the Z axis, and are the stiffness and
damping coefficients.
Rz
K
Rz
C
The first two components of the bushing restoring torque as the result of the
relative rotation of link with respect to link i j are given by

i , j
x x
i , j
x x x
C K T

&
= (11)
i , j
y y
i , j
y y y
C K T

&
= (12)

where and are relative rotational deformations of the
i , j
x
i , j
y
j j j
Z Y X frame
about x-axis and y-axis with respect to the
i i i
Z Y X frame, respectively, ,
, , and are stiffness and damping coefficients. The restoring bushing
torque about the
x
K
y
K
x
C
y
C
j
Z axis due to the rotation of link j with respect to link i

ib , j
z z
ib , j
z z z
C K T

&
= (13)

RecurDyn


where is relative angle between the link and the connector element and
can be obtained by defining the components of the in the coordinate
system of link as
ib , j
z
i
ij
d i
i
ij i
d A d
T
=
=
ij
z
ij
y
ij
x
ij
d
d
d
(14)
It follows that
( )
ij
x
ij
y
ib , i
z
d / d
-1
tan = (15)

where is the transformation matrix that defines the orientation of link
with respect to the global frame. Note that since the inertia of the connector
element is neglected, the resultant force acting on this element must be equal to
zero. Using the spring forces defined in this section, the generalized bushing
forces acting on the track links can be systematically defined.
i
A i

6.5. MEASUREMENT OF TRACK COMPLIANCE CHARACTERISTICS

In order to determine the stiffness and damping coefficients of the contact
force models used in this investigation, an experimental study is conducted to
examine the road wheel and track link contact as well as the interaction between
the track links. Since the experimental results are to be used in the dynamic
simulation of the multibody tracked vehicle, the dynamics of the contact is also
considered in the measurement process.
While a viscous damping force is proportional to the velocity, in many cases,
analytic expressions for the damping forces are not directly available. It is,
however, possible to obtain an equivalent viscous damping coefficient by
equating energy expressions before and after a contact. In this investigation, the
effective stiffness and damping coefficient are obtained by employing the
hysteresis loop method [10]. The effective stiffness and damping coefficient of
single degree of freedom system are given as follows [10]:

6-13

+ = cos
x
F
m
0
0 2
eff eff
K (16)

sin
0
0
x
F
C
eff
= (17)

where , , and are the effective mass, the magnitude of
applied force, the magnitude of displacement, the natural frequency, and the
phase angle of displacement, respectively.
eff
m ,
0
F
0
x ,
In these experimental studies, forces are applied to the center of the road
wheel which is in contact with a track link fixed to a rigid frame. A LVDT
sensor is attached between the center of the wheel and a track link fixed base to
measure the relative displacement. For a static test, the actuator force is
increased gradually up to 10 ton with 2mm/min velocity. For a dynamic test, the
actuator force is excited harmonically up to 35 Hz. Frequencies higher than 35
Hz are not considered in the measurement because of noise and system
resonance. The relationship between the effective stiffness, damping coefficient,
and frequencies is given by Park et al [11]. It can be shown that the effective
stiffness increases up to a frequency of 10 Hz and does not significantly change
after this frequency. On the other hand, the effective damping coefficient
decreases as the frequency increases.
A LVDT sensor is attached between two adjacent track links to measure the
relative displacement. For a static test, the actuator force is increased gradually
up to 10 ton with 2mm/min velocity. Figure 7 shows the resulting load-
displacement relationship. For a dynamic test, a harmonic actuator force with a
frequency up to 50 Hz is used. Figure 8 shows the hysteresis loop when the load
frequency is 10 Hz with 5 ton pre-static applied force. It can be observed that the
effective stiffness increases up to 12 Hz and does not significantly change after
this frequency. The effective damping coefficient, on the other hand, decreases
as the frequency increases.

RecurDyn


Figure 7. Load-displacement relationship for Radial static measurement
(Pre-static load: 5ton, Forcing freq.: 10HZ)

Figure 8. Hysteresis loop for radial dynamic test

A connector end is welded to the fixed track shoe plate and the other end of
the connector is attached to a load cell, which is connected to an actuator
cylinder by revolute joint. Fourteen degrees of the pre-set angle is given in the
pin. A torque of 500 ton-mm is applied along the directions of rotation. The

6-15

static torque versus the rotational angle, the effective torsional stiffness versus
frequency, and the effective damping coefficient versus frequency are plotted by
Park et al [11]. The experimental results showed that the effective torsional
stiffness is less sensitive to the loading frequency and the effective damping
coefficient decreases to a small value when the frequency exceeds 20 Hz.
In this investigation, for the sake of simplicity, the stiffness and damping
coefficients used in the force models are determined using empirical methods
based on the results of the static test only. A spline curve fitting is used to obtain
the compliant characteristics between measurements.

6.6. CONTACT FORCES

In this section, the methods used for developing the contact force models used
in this investigation are briefly discussed. The scenarios of the contacts between
the track links and the road wheels, rollers, sprockets, and the ground are
explained. A more detailed discussion on the formulation of the contact forces is
presented by Choi, et al, [4, 12], and Nakanishi and Shabana [3].

(a) inner surface contact (b) edge contact

Figure 9. Track link and wheel interactions

6.6.1 INTERACTION BETWEEN TRACK AND ROAD WHEEL, IDLER, AND SUPPORT
ROLLER

As shown in Fig. 9, each roller of the vehicle model used in this investigation

RecurDyn

consists of two wheels which are rigidly connected. There are four different
possibilities for the roller and track interaction. The first possibility occurs when
a track link and one wheel of the roller are in contact. In this case, a concentrated
contact force is used at the center of the contact surface of the wheel. The contact
force acting on the link is assumed to be equal in magnitude and opposite in
direction to the force acting on the roller. The second possibility occurs when
both wheels of the roller are in contact with the track link. In this case, two
concentrated contact forces are applied to the roller and the track link. The third
and fourth possibilities occur, respectively, when either one wheel or both
wheels are in contact with the edges of track link. In such a case, one or two
concentrated contact forces are applied to the wheel and the edge of the track
link.

6.6.2 TRACK CENTER GUIDE AND ROAD WHEEL INTERACTIONS

(a) side wall contact (b) top surface contact
Figure 10. Center guide and wheel interactions

Figure 10 shows a schematic diagram for a track center guide and a road
wheel when they are in contact. As previously pointed out a road wheel of the
vehicle model used in this investigation consists of two wheels, which are rigidly
connected, and therefore, there are four possibilities for the track center guide
and wheel interactions, as shown in the Figure 10. The first possibility is the case
in which the right side plate of the wheel is in contact with the left side wall of

6-17

the track center guide. In the second possibility, the left side plate of the wheel is
in contact with the right side wall of the track center guide. The third possibility
occurs when one bottom surface of wheel and the top surface of track center
guide are in contact. In these three contact cases, a concentrated contact force is
introduced at the contact surface of the road wheel, and that contact force is
equal in magnitude and opposite in direction to the force acting on the track link.
The fourth possibility occurs when the two road wheels are not in contact with
the track center guide. In this case, no generalized contact forces will be
introduced.

6.6.3 INTERACTION BETWEEN THE SPROCKET TEETH AND TRACK LINK PINS

In this investigation, five tooth surfaces are used to represent the spatial
contact between the sprocket teeth and the track link pins. During the course of
engagement between the sprocket teeth and the track links, several sprocket teeth
can be in contact with several track link pins, as shown in Fig. 11. The sprocket
used in this investigation has ten teeth, and each tooth has five contact surfaces.
These surfaces are the top, the left, the right, front, and back surfaces. A
Cartesian coordinate system is introduced for each surface. The surface
coordinate system is assumed to have a constant orientation with respect to a
selected tooth coordinate system. The tooth coordinate system has a constant
orientation with respect to the sprocket coordinate system. Therefore, the
orientation of a surface coordinate system can be defined in the global system
using three coordinate transformation matrices; two of them are constant and the
third is the time dependent rotation matrix of the sprocket. Using these
coordinate transformations and the absolute Cartesian co-ordinates of the origin
of the sprocket coordinate system, the location and orientation of each tooth
surface can be defined in the global coordinate system. Using the track link
coordinate system, the global position vector of the center of the track link pin
can be defined.

RecurDyn


(a) sprocket teeth and connector contact (b) teeth side wall and link side wall
contact

(c) teeth top surface and link inner surface contact

Figure 11. Sprocket tooth and track interaction

This vector and the global co-ordinates of the tooth surfaces can be used to
determine the position of the track link pins with respect to the sprocket teeth.
The relative position of the track link pins, with respect to the sprocket teeth can
be used to develop a computer algorithm that determines whether or not the track
link pin is in contact with one of surfaces of the sprocket teeth. The interactions
between the track link pins and the sprocket base circle are also considered in
this investigation. To this end, the distance between the center of the track link
pin and the center of sprocket is monitored. When this distance is less than the
sum of the pin radius and the sprocket base circle radius, contact is assumed and
a concentrated force is applied to the sprocket and the track link pin.

6.6.4 GROUND AND TRACK SHOE INTERACTIONS

The track link used in this investigation has a single or double shoe plate, and
therefore, there are one or two surfaces on each track link that can come into
contact with the ground. The global position vectors that define the location of
points on the shoe plates are expressed in terms of the generalized co-ordinates
of the track links and are used to predict whether or not the track link is in
contact with the ground. In this investigation, contact forces are applied at
selected six points on the track link shoe when it comes into contact with the

6-19

ground. The normal force components are used with the coefficient of friction to
define the tangential friction forces [4, 12].

6.7. METHOD OF NUMERICAL INTEGRATION

The equations of motion of a tracked vehicle are formulated as a set of
differential equations, as described in Section 3. The solution of the differential
equations can be obtained by step-by-step numerical integration. There are two
types of integration methods; one is the implicit method and the other is the
explicit method. The implicit method generally has a larger stability region, but
it requires solving a system of nonlinear equations. The explicit method, on the
other hand, has relatively smaller stability region, but it requires solving only a
system of linear equations. In this investigation, an explicit method is employed.
The dynamics of tracked vehicles is characterized by high impulsive forces
resulting from the contact between the track chains and the vehicle components
as well as the ground. Because of the high frequency impulsive forces, the
numerical integration routine is forced to take a small time step, and as a
consequence, the simulation of a complex tracked vehicle model, as the one
described in this paper, represents a challenging task. Nonetheless the high
frequency oscillations may have little influence on the low frequency motion. In
this case, the high oscillations can be damped out to obtain the gross motion of
the track link. Various dissipation algorithms for time integration of structural
systems have been proposed [7,8,13]. In this investigation, the method proposed
by Chung and Lee [7] is considered because of its easy implementation and
computational efficiency. Accuracy and stability conditions must be considered
in carrying out a numerical integration of the tracked vehicle equations. The
accuracy and stability conditions are obtained by using the truncation error and
the error propagation analyses. A variable step algorithm is proposed in the
following subsections.

6.7.1 ACCURACY ANALYSIS

The following numerical integrator proposed by Chung and Lee [13] is

RecurDyn

employed in this research.

) q , N(q M q
1
n n n
& & &

= (18)
] ) 2 / 3 ( ) 2 / 1 [(
n 1 n n 1 n
q q q q & & & & & & + + =
+
t (19)
] ) 27 / 28 ( ) 54 / 29 [(
n 1 n n n 1 n
q q q q q
2
& & & & & + + + =
+
t t

(20)

The can be expanded by the Taylor series as follows:
1 n+
q

) ( O ) 2 / (
3
n
2
n n 1 n
t t t + + + =
+
q q q q & & &

(21)

where is collection of higher order terms. Subtracting Equation (20)
from Equation (21) yields the truncation error as follows:
) ( O
3
t

) ( O ) )( 54 / 29 ( ) (
3 2
n 1 n n
t t t + =

q q & & & &

(22)
) ( O
dt
d
) 54 / 29 (
3 3
n
t t + q& &

(23)

which shows that the proposed integrator achieves the second-order accuracy for
non-linear dynamic systems.

6.7.2 STABILITY ANALYSIS

Since it is difficult to analyze the stability condition for a general nonlinear
system, the following linear, undamped, and unloaded system is considered:

0
2
= +
n n
q q & &

(24)

where is a natural frequency. Applying Equation (24) with the integration
formula proposed in this section yields the one step form of the numerical
scheme
n n
HX X =
+1

} 1 ,......, 2 , 1 , 0 { N n (25)
where
T
n
2
n n n
] q , q , q [ & & & t t = X (26)
and

6-21

=
0 0
2 / 1 1 ) 2 / 3 (
54 / 29 1 ) 27 / 28 ( 1
2
2
2
H (27)

in which t = . The characteristic equation for

is obtained as follows: H

0 ) 27 / 1 ( } ) 27 / 2 ( 1 { } ) 27 / 28 ( 2 { ) det(
2 2 2 3
= + + = I H

(28)

where is the I 3 3 identity matrix and

denotes the eigenvalue. The
stability characteristics of the method are determined by the condition that the
roots of the characteristic equation remain in or on the unit circle of the complex
plane as follows:
1 , { }
3 2 1
, , max = (29)

where is called the spectral radius. Stability analysis can be assessed by
using the transformation of Eq. 7.9 to map the interior of the unit circle into the
left half-plane and by applying the Routh-Hurwitz criteria to the transformed
characteristic equation. The stability condition for the algorithm is obtained by
applying the Routh-Hurwitz criteria as follows:

0 ) 27 / 31 ( 4
2

(30)

0 ) 27 / 23 ( 4
2

(31)

which are reduced to
) / 8665 . 1 ( t (32)

Equation (32) provides a guideline in choosing a step size that satisfies the
stability condition.

RecurDyn

6.7.3 IMPLEMENTATION OF A VARIABLE STEPPING ALGORITHM

Since the governing equations of motion for the tracked vehicle system are
highly nonlinear, the integration step size must be varied so that both the
accuracy and stability conditions are satisfied. For the accuracy condition,
ignoring the higher order terms in Equation (22) yields the local truncation error
formula as follows:
2
n 1 n n
| ) ( | ) 54 / 29 ( ) ( t t =

q q & & & & (33)

The allowable stepsize with a given error tolerance is obtained by solving
Equation (33) for t

as follows:

2 / 1
n 1 n
| } | ) ( | 29 /{ 54 | q q & & & & =

t (34)

For the stability condition, the apparent frequency method proposed by Park
and Underwood [14] is employed in this research. An apparent frequency is
estimated by substituting q and into the following equation: q& &

0
1
2
1
= +
+ +
i
n app
i
n
q q & & i

(35) } ,......, 2 , 1 , 0 {
q
n

where is the apparent frequency and is the number of generalized co-
ordinates. The highest apparent frequency is selected as the reference frequency
in determining the step size.
app
q
n
The step size determination algorithm is shown in Figure 12. Note that the
stability condition of instead of in Equation
(32) is used for conservative numerical integration. The integration step size
employed by the variable step integration algorithm used in this investigation,
when the vehicle maximum acceleration, steady state velocity at 50 Km/h and
stiff deceleration of braking, is shown in Figure 13. This figure shows that the
integration step size is relatively depended on the vehicle speed. The increment
of vehicle speed will enlarge impulsive contact forces and oscillation of track
links, and integration step size should be decreased, accordingly.
app
/ 5 . 0 < t
app
/ 8665 . 1 < t

6-23

Figure 12. Variable stepsize algorithm

Figure 13. Stepsize of variable step integration algorithm

RecurDyn


The high mobility tracked vehicle shown in Figure1 is used as a simulation
model in order to demonstrate the use of the methods proposed in this paper.
Several simulation scenarios, including acceleration, high speed motion, braking
and turning motion, are presented in this investigation. In the simulation of
acceleration, high speed motion, and braking of the vehicle, the same angular
velocity is used for both left and right sprockets in order to obtain straight line
motion. The angular velocities of the sprockets are increased linearly up to -45
rad/s in 10 s, kept constant for 3 s, and then decreased linearly to 0 rad/s in 4 s.
The coefficient of friction between the track links and ground is assumed to be
0.7 in the case of rubber and concrete contact. The double pin track link is used
in the numerical study presented in this section. Figures 14-18 show the
numerical results of simulation of the acceleration, steady-state velocity and
deceleration.

Figure 14. Vertical motion of a track link

6-25

Figure 15. Radial tension of track link

The vertical displacement of a track link with respect to the global coordinate
system during the constant velocity motion is shown in Figure 14. This figure
clearly shows the effect of three support rollers, idler, six road wheels and
sprocket on the vertical displacement of the track link. The track tension can
have a significant effect on the dynamic behavior of tracked vehicles, such as
preventing the separation of the track chains [18,19] from chassis, distribution of
mean maximum pressure(MMP), power efficiency, and the life of the track chain.
As previously pointed out in Section 2, about 10 percent of the vehicle weight
[15] is used as track pre-tension. Simulation results showed that the track tension
significantly decreases after the start of the motion.

(a) (b)

RecurDyn


(c) (d)

(e)
Figure 16. Sprocket teeth loading contour: (a) acceleration; (b) cruise at high speed; (c)
braking; (d) turning (right sprocket); (e) turning ( left sprocket)

Figure 15 shows the longitudinal track tension in the bushing between track links,
while Figure 16 shows sprocket teeth loading contour. Heavy duty high mobility
tracked vehicles as one used in this study have, in general, double pin track links.
One of the main advantages of using double pin track is that the shear stress on
the rubber bushings can be significantly reduced as compared to the single pin
track link. In order to compare the loadings on the track bushings in the case of
single or double pins, new driving conditions are examined. The rotational speed
of both sprockets is decreased linearly up to - 9 rad/sec in 2 sec, and then kept
constant velocities.
Figure 17 illustrates the moment on the rubber bushings in the case of the
single and double pin track. The results presented in this figure demonstrate the
significant reduction of the load on the rubber bushings when a double pin track
is used. Figure 18 shows the norm of the contact forces exerted on one of the
links of the right track chain as the result of its interaction with the road wheels,

6-27

support rollers, idler, sprocket, and ground. Figure 19 shows a road arm angle
and HSU gas pressure of the second road wheel.

Figure 17. Torsional moments of track rubber bushing.

Figure 18. Contact forces of track link

RecurDyn


Figure 19. Gas pressure of HSU system

The second simulation scenario used in this study is a turning motion. The
turning motion is obtained by providing two different values for the angular
velocities of the sprockets. The angular velocity of the right sprocket is
decreased linearly to -9 rad/s and the angular velocity of the left sprocket is
increased linearly to 9 rad/s in 2 s. The angular velocities are then kept constant
velocities. Using these values for the sprocket angular velocities, the vehicle
rotates counter clock wise as result of opposite rotation directions of right and
left sprockets, the upper part of the right side of the track chain is loose, and the
upper part of the left side of the track chain is tight as shown in Figure 20. Figure
21 shows the forces of contact between side wall of the wheels and center guide
of a track link.

6-29

Figure 20. Tension adjuster force

Figure 21. Track center guide and wheel contact forces

RecurDyn


6.9. SUMMARY AND CONCLUSIONS

The dynamics of a high speed, high mobility multibody tracked vehicle is
investigated in this paper. Compliant forces are used to define the connectivity
between the links of the track chains instead of an ideal pin joint. Two track link
models are considered in this study. These are the single pin and double pin track
models. In the single pin track model, only one pin is used to connect two track
links in the chain. In the double pin track model, two pins are used with a
connector element to connect two links of the track chain. Rubber bushings are
used between the track links and the pins. The stiffness and damping
characteristics of the contact forces are obtained using experimental testing. By
using experimental data, the generalized contact and bushing forces associated
with the generalized co-ordinates of the tracked vehicle are developed. The
tracked vehicle model used in this investigation includes significant details that
include modeling the chassis, sprockets, idlers, road wheels, road arms, and the
multi-degree of freedom track chains. The vehicle model is assumed to consist of
189 bodies, 36 pin joints, and 152 bushing elements. The model has 954 degrees
of freedom. Because of the high frequency contact forces, numerical difficulties
are often encountered in the simulation of multibody tracked vehicles. An
explicit numerical integration method that has a large stability region is
employed in this study. The method employs a variable time step size in order to
achieve better computational efficiency. It was observed that the time step size
significantly decreases as the vehicle speed increases. Several simulation
scenarios are examined in this investigation. These include accelerated motion,
high speed motion with a constant velocity, braking, and turning motion. The
simulation results demonstrate the significant effect of the bushing stiffness on
the dynamic response of the multibody tracked vehicle. It was also shown that
the use of the double pin track model leads to a significant reduction in the
bushing forces as compared to the single pin track model.

6-31

REFERENCES

(1) Galaitsis A.G., 1984, A Model for Predicting Dynamic Track Loads in Military
Vehicles, ASME, Journal of Vibration,Acoustics, Stress, and Reliability in Design, Vol.
106/289
(2) Bando, K., Yoshida, K., and Hori, K., 1991, The Development of the Rubber Track
for small Size Bulldozers, International off-Highway Powerplants Congress and
Exposition, Milwaukee, WI, Sept. 9-12
(3) Nakanishi, T., and Shabana,(1994)"Contact Forces in The Nonlinear Dynamic Analysis
of Tracked Vehicle," International Journal For Numerical Methods in Engineering, 1994
1251-1275.
(4) Choi, J. H., 1996 "Use of Recursive and Approximation Methods in The Dynamic
Analysis of Spatial Tracked Vehicle," Ph. D. Thesis, The University of Illinois at Chicago
Scholar C. and Perkins N., 1997, Longitudinal Vibration of Elastic Vehicle Track
System
(5) Newmark NM. A method of computation for structural dynamics. Journal of the
Engineering Mechanics Division, ASCE 1959; 85 (EM3):67-94
(6) J. Chung, J. M. Lee,(1994) A New Family of Explicit Time Integration Methods for
Linear and Non-linear Structural Dynamics, International Journal for Numerical
Methods in Engineering, Vol.37, 3961-3976
(7) J. Chung,(1992) Numerically Dissipative Time Integration Algorithms for Structural
Dynamics, Ph.D. dissertation, University of Michigan, Ann Arbor
(8) Shabana A,(1989) Dynamics of Multibody Systems, John Wiley & Sons, New York
(9) Shabana, A,(1996) Theory of Vibration, An Introduction, Second Edition, Springer-
Verlag, New York
(10) Park DC, Seo IS, Choi JH. Experimental study on the contact stiffness and damping
coefficients of the high mobility multibody tracked vehicle. Journal of Korea Society of
Automotive Engineers 1999; 7:348-357
(11) Lee, H. C., Choi., J. H., Shabana, A. A., Feb. 1998 "Spatial Dynamics of Multibody

RecurDyn

Tracked Vehicles: Contact Forces and Simulation Results," Vehicle System Dynamics,
Vol. 29, pp. 113-137
(12) E. L. Wilson,(1968) A Computer Program for the Dynamic Stress Analysis of
Underground Structures, SESM Report No. 68-1, Division of Structural Engineering
and Structural Mechanics, University of California, Berkeley
(13) K. C. Park and P. G. Underwood,(1980) A Varialbe-step Centeral Difference Method
For Structural Dynamics Analysis Part 1. Theoretical Aspects, Computer Methods in
Applied Mechanics and Engineering 22, 241-258
(14) Owen J. Guidelines for the Design of Combat Vehicle Tracks. Dew Engineering and
Development Ltd., Ottawa, Canada.
(15) Choi, J. H., Lee., H. C., Shabana, A. A., Jan. 1998 "Spatial Dynamics of Multibody
Tracked Vehicles: Spatial Equations of Motion," International Journal of Vehicle
Mechanics and Mobility, Vol. 29, pp. 27-49
(16) Bruce Maclaurin (1983) Progress in British Tracked Vehicle Suspension Systems,
830442 Society of Automotive Engineers(SAE)
(17) Ketting Michael,(1997) Structural Design of Tension Units for Tracked Vehicles,
especially Construction Machines Under The aspect of Safety Requirements, Journal of
Terramechanics, Vol. 34, No. 3, pp. 155-163.
(18) Trusty RM, Wilt MD, Carter GW, Lesuer DR. Field measurement of tension in a T-142
tank track. Experimental techniques, 1988.

6-33

RecurDyn


7

DYNAMIC TRACK TENSION OF HIGH
MOBILITY TRACKED VEHICLES

7.1. INTRODUCTION

The track tension of tracked vehicles plays significant roles for the dynamic
behaviors, such as separation of the track system from the chassis system,
distribution of wheel supporting pressure, power efficiency, vibration and noise,
and the life of the track system. Due to the importance of the track tension in
designing tracked vehicles, study of the dynamic track tension has long been a
subject for many researchers in manufacturers and academia. However, it is very
difficult to clearly understand the nonlinear behaviors of the dynamic track
tension while a vehicle runs, even though both experimental and numerical
works have been attempted [1-7].
Both numerical and experimental investigations are carried out in this paper.
For the experimental investigation, strain gages are attached on track pin-
bushing locations of track shoe body, and signal processing and recording
modules are installed on the inside of track shoe body. Only limited results can
be collected through the experiment due to small installation space inside of a
track shoe body, high impulsive shock and vibration, and high temperature over
150 cent degrees. In order to make up the limitation of experimental results, a
tracked vehicle model developed in [9] is used to obtain various numerical
results. Each track link is modeled as a body which has six degrees of freedom
and is connected by a bushing force element. The numerical results are validated
against the experimental results before they are used for investigations.
Doyle and Workman [1] presented a static prediction of track tension when
the suspensioned-tracked vehicle traverses obstacles using two dimensional
finite element methods. An elastic beam element subjected to tension,
compression and bending loads was utilized to model track links. Galaitsis [2]

RecurDyn

DYNAMIC TRACK TENSION OF HIGH MOBILITY TRACKED VEHICLES
demonstrated that the analytically predicted dynamic track tension and
suspension loads of a high speed tracked vehicle are useful in evaluating the
dynamic analysis of the vehicle. The predicted track tensions were compared
with the empirically measured track tensions. A detailed track tension
measurement methodology and results are presented by Trusty et al [3]. Strain
gages connected to a portable data acquisition system were installed in the track
link. The flat ground, quick acceleration, traversal of obstacle courses, pivot
turns, moving uphill, and pre and post tension, were used for the tension
investigation scenarios. McCullough and Haug [4] designed a super element that
represents spatial dynamics of high mobility tracked vehicle suspension systems.
The track was modeled as an internal force element that acts between ground,
wheels and the chassis of the vehicle. Track tension was computed from a
relaxed catenary relationship. Empirical normal and shear force formulas based
on constitutive relations from soil mechanics were used to model the soil-track
interface. Choi [5, 6] presented a large scale multibody dynamic model of a
construction tracked vehicle in which the track is assumed to consist of track
links connected by single degree of freedom pin joints. In this detailed three
dimensional dynamic model, each track link, sprocket, roller, and idler is
considered as a rigid body that has a relative rotational degree of freedom.
Scholar and Perkins [7] developed an efficient alternative model of the track
chains considering longitudinal vibrations. The track is assumed to consist of a
finite number of segments, each of which is modeled as a continuous uniform
elastic rod attached to the vehicle wheels. Overall chain stretching effects are
accounted for.
The purpose of this paper is to investigate the dynamic track tensions of a high
mobility tracked vehicle maneuvering under various driving conditions. Both
numerical and empirical methods are employed and the effects of pretensions,
friction forces, interacting proving grounds, vehicle speeds, and driving torque
are explored for the sake of understanding dynamic behaviors of the track system.

7-3

7.2. NUMERICAL MODEL OF A HIGH MOBILITY TRACKED VEHICLE

Turret
Chassis system
Track system

Figure 1. High mobility multibody tracked vehicle model

A three-dimensional multibody tracked vehicle model shown in Fig. 1 consists
of a chassis subsystem and two track subsystems. The chassis subsystem
includes a chassis, sprockets, support rollers, idlers, road arms, road wheels and
the suspension units. The sprockets, support rollers, and road arms are connected
to the chassis by revolute joints. The track link subsystem includes a shoe body,
a pin, rubber bushings, and a rubber pad. Rubber bushings and pin are inserted
into the hole of a shoe body with a radial pre-pressure and a rubber pad is
mounted on the ground interaction side of the shoe body. The vehicle model
used in this investigation consists of 189 bodies; 37 bodies for the chassis
subsystem, 76 bodies for each track subsystem, 36 revolute joints and 152
bushing elements and has 954 degrees of freedom.

RecurDyn


Figure 2. Hydraulic track tension adjustor system

Suspension systems and tension adjustor: The suspension units of the
vehicle include a Hydro-pneumatic Suspension Unit(HSU), and torsion bar that
are modeled as force elements whose compliance characteristics are obtained
from analytical and empirical methods. The HSU systems are mounted on first,
second, and sixth stations to damp out pitching motion and to decrease an impact
when the vehicle is running over large obstacles. The torsion bars are mounted
on the middle stations for this vehicle model. A simple torsional spring model is
used in this investigation to represent the stiffness of the torsional bars. The
hydraulic passive tension adjustor is installed on the idler to maintain a proper
track tension of the tracked vehicle model. Figure 2 shows the schematic
diagram of the tensioner system of the vehicle. The hydraulic ram of the tension
adjustor is modeled as an equivalent linear spring-damper force element.
Track link connection: Each track subsystem is modeled as a series of
bodies connected by rubber bushings around the link pins which are inserted into
a shoe plate with a radial pressure to reduce rattling of the pin. When the vehicle
runs over rough surfaces, the track systems are subjected to extremely high
impulsive contact forces as the result of their interaction with the vehicle
components such as road wheels, idlers, and sprocket teeth, as well as the ground.
The rubber bushings tend to reduce the high impulsive contact forces by
providing cushion and reducing the relative angle between the track links. In this

7-5

investigation, a continuous force model is used to represent the pin connections.
This force model is a non-linear function of the coordinates of the two links. Note
that because of the bushing effect, the origins of the joint and pin coordinate
systems do not always coincide.
Contact detection and forces: In this section, the contact force model and the
contact detection algorithms between the track links and the road wheels, rollers,
and sprockets are briefly discussed. A more detailed discussions on the
formulation of the contact force model is presented by Choi, et al, [5, 6, 9]. As
shown in Fig. 1, when a track link travels around vehicle components, its
trajectory is determined by the contact forces. These forces are created by
detecting on contact conditions. The contact detection algorithms monitor the
contacts of, wheel and track link contact, center guide and wheel contact,
sprocket tooth and track link pin contact, and side wall of track link and sprocket
contact. Once a contact condition is satisfied, contact forces are applied at the
contacted position to restitute each other.

Figure 3. Interaction between track shoe body and triangular patch element

RecurDyn


Interacting ground representations: The ground interacting surface of a track
link can be single or multiple, and therefore, there are one surface or multiple
surfaces on each track link that can come into contact with the ground. The
interacting surface of ground is discretized and each contact node points were
defined. The global position vectors that define the locations of points on the
shoe plates surface of track link are expressed in terms of the generalized
coordinates of the track links and are used to predict whether or not the track link
is in contact with the ground. Since the contact surface of track link consists of
rubber pad and steel shoe plate, the contact forces at each node point are
evaluated by using their own stiffness and damping coefficients. In order to
construct various geometries of tracked vehicle paved proving ground [10], such
as bumping courses, trench course, inclined course, standard cross country
courses, descritized terrain representation methods using triangular patch
element are used in this investigation. The plane equation of interacting ground
profiles for a triangular patch element which has three nodes and a unit normal
vector is employed as illustrated in Fig. 3.
Equations of motion: Since the track system interacts with the chassis
components through the contact forces and adjacent track links are connected by
compliant force elements, each track link in the track system has six degrees of
freedom which are represented by three translational coordinates and three Euler
angles [11]. The equations of motion of the chassis that employs the velocity
transformation defined by Choi [5, 9] are given as follows:

) (
r
i i
q B M Q B q MB B
T T
&
&
& & = (1)

where , B and q are relative independent coordinates, velocity
transformation matrix, and Cartesian velocities of the chassis subsystem, and M
is the mass matrix, and Q is the generalized external and internal force vector of
the chassis subsystem, respectively. Since there is no kinematic coupling
between the chassis subsystem and track subsystem the equations of motion of
the track subsystem can be written simply as
r
i
q &

7-7

t t t
Q q = & & M (2)

where , and denote the mass matrix; and the generalized
coordinate and force vectors for the track subsystem. Consequently, the
accelerations of the chassis and the track links can obtained by solving Eqs. 1
and 2.
t
M
t
q
t
Q
G-Alpha integrator : Many different types of integration methods can be
employed for solving the equations of motion for mechanical systems. Explicit
methods have small stability region and are often suitable for smooth systems
whose magnitude of eigenvalues is relatively small. Contrast to the explicit
methods, implicit methods have large stability region and are suitable for stiff
systems whose magnitude of eigenvalues is large. One of the important features
of the implicit methods is the numerical dissipation. Responses of mechanical
systems beyond a certain frequency may not be real, but be artificially
introduced during modeling process. In the model used in this investigation, a
contact between two bodies is modeled by compliance elements. Lumped
characteristics of the spring and damper must represent elastic and plastic
deformations, and hysterisis of a material. Such characteristics may include
artificial high frequencies which are not concern of a design engineer. Unless
such artificial high frequency is filtered, an integration stepsize must be reduced
so small that integration can not be completed in a practical design cycle of a
mechanical system. To achieve this goal, generalized-alpha method [8, 9] has
developed to filter frequencies beyond a certain level and to dissipate an
undesirable excitation of a response. One of the nice advantages of the
generalized-alpha method is that the filtering frequency and dissipation amount
can be freely controlled by varying a parameter in the integration formula. As a
result, the generalized-alpha method is the most suitable integration method for
integrating the equations of motion for stiff mechanical systems. Figure 4 shows
the animation of high mobility tracked vehicle when the vehicle runs over a
trench profile.

RecurDyn


(a) time = 0.0 sec (b) time = 3.0 sec

(c) time = 6.0 sec (d) time = 9.0 sec
Figure 4. Computer animation of multibody tracked vehicle running
over trench ground profile

7.3. INTERACTION GROUNDS

The ground interacting surface of a track chain link can be single or multiple,
and therefore, there are one surface or multiple surfaces on each track link that
can come into contact with the ground. The interacting surface of chain link is
discretized and each contact node points were defined. The global position
vectors that define the locations of points on the shoe plates surface of chain link
are expressed in terms of the generalized coordinates of the track chain links and
are used to predict whether or not the track chain link is in contact with the
ground. Since the contact surface of track chain link consists of rubber pad and
steel shoe plate, the contact forces at each node point are evaluated by using their
own stiffness and damping coefficients. In order to construct various geometries
of tracked vehicle paved proving ground, such as bumping courses, trench
course, inclined course, standard cross country courses, discretized terrain
representation methods using triangular patch element are used in this
investigation. A triangular patch element has three nodes and a unit normal
vector to describe plane equations of interaction grounds [13].

7-9

(a) Series of triangular patch for generalized virtual body

(b) Triangular patch surface
Figure 5. Discretized terrain representation

Discretized terrain representation: The virtual terrain model used in this
investigation is a general three dimensional surface defined as a series of
triangular patch elements. Figure 5 (a) shows an example of virtual ground using
8 points and 6 elements. Most geometries of various paved proving ground for
tracked vehicle can be represented by using triangular patch elements. The
equation for the plane defined from three nodes can be written as

z x a x a x a = + +
3 2 1
(3)

The three coefficients, a , , and of the equations of plane can be
obtained by given three locations of triangular patch shown in Fig. 5 (b), and by
using Cramer's rule [14], these coefficients are
1 2
a
3
a

RecurDyn


A
A
det
det
k
k
a =
, = 1, 2, 3 (4) k
where
=
=
=
=
3 3 3
3 3 2
3 3 1
3 3
] [
] [
] [
] [
k k k
k k
k k
k k
z y x A
z x A
y z A
y x A
I
I
I
and (5)
T
] 1 1 1 [ = I

Then the unit normal vector of the plane is defined as n

T
a a
a a
] 1 [
1
1
2 1
2
2
2
1
+ +
= n
(6)

Virtual proving ground : Until recent development of computer simulation model
[6, 7, 9], the development process of tracked vehicle have been depended on
inefficient technologies of repeated procedures ; construction prototype vehicle
based on basic calculations and simple computer simulation, test on proving
ground, then modification. This expensive design procedure can be diminished
by recent developments of computer simulation.
In this investigation, only paved ground models are developed for the virtual
test of dynamic analysis of three dimensional tracked vehicle. The developed
computer models of grounds are stored into the created ground library.
As shown in Figs. 6 and 7 the variety of virtual proving grounds, symmetric
and unsymmetric bump courses, trench and ditch courses, longitudinal and
laternal inclined courses, and standard cross-country courses of RRC9 and
Profile IV, are constructed by using triangular patch elements. When a vehicle
runs over these virtually created proving grounds, the nonlinear behaviors of
track chains resulting from the interacting with the test grounds are obtained in
this numerical investigation.

7-11

(a) Single bump course (d) Obstacle course

(b) Trench course (e) Grade ability slope

(c) Ditch course (f) Side slope
Figure 6. Various paved virtual proving ground using triangular patch

RecurDyn


(a) Series of triangular bump

(b) Series of trapezoidal bump
Figure 7. Simulated cross-country course (APG Profile IV)

Methods of finite track chain-ground contact point: Unlike wheel and surface
contact, the interactions between track chain link and ground are very
complicated problems. This is because the track chain link has irregular contact
geometry and different material properties. According to large number of track

7-13

chain links of each track subsystem, commonly used contact theory of surface to
surface interactions in finite element community can not be employed for this
work. Choi [15] suggested that element free finite contact nodes were distributed
on the contact surface of track chain link, which have their own stiffness and
damping characteristics. The relative indentations of each nodes were monitored
and positions are restored. The use of element free finite contact node methods
demonstrated clearly the computational efficiency for dynamic analysis of track
chain system. Based on the method developed by Choi [15], the interactions
between track chain link surface and triangular patch surface are developed in
this paper. Figure 3 shows the interaction between finite contact nodes of track
link and triangular patch surface. The perpendicular deformation scalar of
contact node
ij
k
d
j of link i on patch plane can be defined as k

ij
k
P ij
k
d n r
1
= (7)

where is shown in Fig. 3 and unit vector is defined in Eq. 6.. The
criterion of necessary condition for the contact to occur of node
P 1
r
ij
k
n
j , which is
not sufficient, is
>
separate d
contact d
ij
k
ij
k
0
0
(8)

If this conditions is satisfied, the position vector shown in Fig. 3 is used
to compute the node location whether contact point
jk
B
r
B of node j is on the
patch plane . The position vector can be written as k
jk
B
r

ij
k
ij
k
ij
p
i i jk
B
d u A R r n + =
(9)

where is the transformation matrix associated with the orientation
coordinates of link and
i
A
i
ij
p
u is the local position vector of node j in the
track chain link coordinate system. On the other hand, using scalar triple product

RecurDyn

if one of the following conditions is satisfied

0
0
0
3 31
2 23
1 12
n
B
n
B
n
B
u r r
u r r
u r r
or (10)

0
0
0
3 31
2 23
1 12
n
B
n
B
n
B
u r r
u r r
u r r

then the node j of link is in contact with patch element . i k
If the node j is in contact with patch plane , the contact force at the
contact node can be computed using the equation as
k

ij
k
ij
k
ij
k
ij
k
ij
k
d C d K F
&
+ =
(11)

where and C are, respectively, the stiffness and damping coefficients of
the contact force model at node
ij
k
K
ij
k
j of body on patch plane . Using the
expression for the contact force as defined by the preceding equation, the contact
force vector can be defined as
i k

ij
k
ij
k
ij
k
F n F = (12)

where is a unit normal vector shown in Fig. 3. The virtual work of the
contact force at the nodes is given by
ij
k
n

=
=
n
j
ij
k
i
k
W W
1

=
=
n
j
ij
k
ij
k
ij
k
d F
1
n | |
(
=
i
i
T
i
T
i
R
R
Q Q
(13)
where
=
=
n
j
ij
k
T
i
R
1
F Q
=
=
n
j
ij
k
T ij
p
i
T
i
1
)
~
( F u A Q
(14)

are the generalized contact forces associated with the Cartesian and orientation

7-15

coordinates of link , and i
ij
p
is the skew symmetric matrix associated with the
vector
ij
p
u . In order to evaluate the tangential component of these contact forces
for friction effect at each contact nodes, the smooth Coulomb friction model [6]
is employed in this investigation. Figure 8 shows the computer animation of
multibody tracked vehicle running over APG Profile IV test ground.
u
~
Note that the proposed element free finite contact node method have several
advantages such as, simple computer implementation, easy contact detecting
algorithm for irregular surface, independent contact coefficients, and distribution
of concentrated contact forces, however, in the penalty function approach used in
this contact force model the determinations of spring and damping coefficients
may be a black art. These coefficients may not correspond to familiar physical
properties that can be measured experimentally. Careful numerical calibration
process is necessary to obtain reliable model, accordingly.

Time = 0.0 sec Time = 9.0sec

Time = 3.0 sec Time = 12.0 sec

RecurDyn


Time = 6.0 sec Time = 15.0 sec

Figure 8. Computer animation of multibody tracked vehicle running over Aberdeen
profile IV proving ground

7.4. MEASUREMENT OF THE DYNAMIC TRACK

The measurement system is composed of strain gages, signal processor, data
storage, and power unit. The system is installed inside of a track shoe body.
When the switch is on, the system will start to measure and store the tensional
forces from the strain gages into data storage processor. The measurement results
are then downloaded into a laptop computer through communication port.
The basic platform of dynamic track instrumentation system is developed by
Kweenaw Research Center at Michigan Technology University [12]. The tension
measurement system records 2 channels, which are track tensions at both ends of
a track link, at the rate of 800 samples per second for 160 seconds. The tension
data in the system memory is offloaded to a computer for storage after the test
vehicle is stopped. A track shoe body was carved to attach full bridge of strain
gages on the outside and inside edges of the body. A track link, as a sensor
system to measure the dynamic track tension of the high mobility tracked vehicle,
was carefully calibrated at the center. A known load, Shunted Engineering Unit
Value, can be simulated by shunting one leg of the strain gage bridge using a
58,900 ohm resistor inside the measurement system. The known load is about
20,000 lb [12]. If any load is applied to the measurement system, the load as an
engineering unit can be determined by a linear interpolation or extrapolation
using the engineering unit value.
Figure 9 shows the comparison of simulation and experimental results when
the vehicle runs on flat ground with the velocity of 10 km/h. The figure shows

7-17

that there are four disagreement areas between experimental and numerical
results. These disagreements are due to the extra deformation of the strain gage
when the track link moves around sprocket, idler, and first and last road wheels.
The extra deformation makes the track tension look much higher than it actually
is
12 13 14 15 16 17
0
10k
20k
30k
40k
50k
60k
70k
80k
Time(sec)
T
e
n
s
i
o
n
(
N
)
Experiment

Simulation

Figure 9. Dynamic tension of a track link

7.5. NUMERICAL INVESTIGATION OF DYNAMIC TRACK TENSION

Extended numerical simulations are carried out to compensate for the
experimental limitations due to space and environment. The track tension is
monitored in two different views of track link following view and chassis fixed
view. In order to acquire the track tension for the chassis fixed view, the track
tensions are recorded until all links pass through one point of the hull. For the
track link following view, the track tension of one selected track link is recorded
when it is moving around vehicle components of idler, road wheels, sprocket and
support rollers.
Key physical quantities influencing the track tension are pre-tension, vehicle
speed, ground profile, traction force, driving torque, and turning resistance,
respectively. The pre-tensions of 25 kN, 50 kN and 100 kN are given to observe
their influences on the dynamics of the vehicle. Three different speeds of 5 km/h,
20 km/h, and 40 km/h are given on both driving sprockets using velocity

RecurDyn

constraint equations. Various ground profiles as defined in real proving ground
[10] are developed by using the triangular patch elements. Three friction
coefficients of 0.1, 0.4, and 0.7, between the track shoe body and the ground, are
used for different traction force modeling. The track tensions are observed for a
pivot turning, right and left turning, backward motion, acceleration and braking
motions.
Effect of pre-tension: One of the most critical variables for the dynamic
track tension is the pre-tension. Although an optimal pre-tension has long been a
major subject for academia and industry, researchers only relied on experimental
and field experiences. Most of high mobility suspenioned tracked vehicles,
approximately 10 % of the vehicle total weight is loaded as a track pre-tension.
Figure 10 shows the track tensions of a selected track link in the link following
view with three different pre-tensions. These pre-tensions are 25 kN, 50 kN, and
100 kN, respectively. Both sprockets have constant angular velocity of -17.8
rad/sec which can produce 20 km/h vehicle speed. As illustrated in this figure,
increment of the pre-tension linearly increases the dynamic track tension.

0 1 2 3 4
0
20k
40k
60k
80k
100k
120k
5
50kN pre-tension
100kN pre-tension
25kN pre-tension
T
e
n
s
i
o
n
(
N
)
Time(sec)

Figure 10. Track tensions of pre-tension effect

7-19

Effect of vehicle speed: Like a tire of wheeled vehicles, revolution of a track
system can cause the movement of tracked vehicles. The vehicle speed varies
time to time due to random and irregular vehicle operations. Several numerical
and empirical studies showed that the amplitude of track tension does not change
much as the speed changes. However, the frequency of the track tension changes
significantly. In the case of a bump run, the track tension around contacted
region increases significantly at a higher speed when the vehicle hits a bump. It
is mainly because of large increment of impact force between a track link and the
ground.
Effect of ground profile: In the previous section, a generalized method for
building the proving ground profiles are introduced. Many profiles representing
real testing grounds are developed by using the triangular patch. Since the
ground contact forces are directly transferred to the track links, the track tension
is strongly related to the surface geometry of a given ground profile.
Effect of traction forces: Force transmission of a tracked vehicle can be
understood as three force conversions. When a track system rotates, traction
forces are generated between the track system and the ground in opposite
direction to the velocity of the track system. Contact forces between track link
pins and sprocket teeth are converted to the sprocket moment in the tangential
direction of the pitch circle of the sprocket. The sprocket torque can be converted
again to a translational force acting on the axis of the sprocket center. Finally,
this translational force on the axis of the sprocket can cause movement of a
tracked vehicle. During the force conversion process the traction forces can be
replaced directly by tensional forces of track system. The amount of the traction
forces is determined by a friction model between the track system and the
ground. The track tensions between middle road wheels with different friction
coefficients are shown in Fig. 11. Three different friction coefficients, 0.1, 0.4,
and 0.7 are used in this numerical investigation. In order to show the effect of the
friction, the vehicle is accelerated from zero to 40 km/h in ten seconds. As
shown in this figure, increment of the friction coefficient causes an increment of
the track tension.

RecurDyn

0.0 0.5 1.0 1.5 2.0 2.5 3.0
5x10
4
6x10
4
7x10
4
8x10
4
=0.4
=0.7
=0.1
T
e
n
s
i
o
n
(
N
)
Time(sec)

Figure 11. Track tensions of traction force effect

Effect of sprocket torque : In this investigation power-pack, engine and
transmission, are modeled by using velocity constraints or the sprocket torque. In
the real world there are two major disturbances to keep steady sprocket torque.
These are irregular driver inputs and impacts of transmission shifts. The sprocket
torque is converted to the contact force between the sprocket teeth and track link
pins. The sprocket contact force repeats to pull and push, which makes the track
tension vary. To observe the effects of sprocket torque, step, sinusoidal, and
linear-steady torques are applied on the sprocket. Figure 12 shows the track
tension changes near the sprocket when the step torque is applied.

7-21

1 2 3 4 5 6 7 8 9 10 11
0
5k
10k
15k
20k
T
o
r
q
u
e
(
N
m
)
Time(sec)
20k
30k
40k
50k
60k
tension
torque

T
e
n
s
i
o
n
(
N
)

Figure 12. Track tensions of sprocket torque effect
0 2 4 6 8 10 12
0
50k
100k
150k
200k
250k
road wheels
upper
right track tension
left track tension
sprocket
upper
T
e
n
s
i
o
n
(
N
)
Time (sec)

Figure 13. Track tensions of turning resistance effect

Effect of turning motion: Heading direction of a tracked vehicle is turned by
a speed difference of left and right sprockets, which causes different traction
forces. The traction forces of track systems are converted to the forces at the
center of both sprocket axes. Then the chassis system is rotated with respect to
the vertical axis by a force difference of both sprocket axes. Suppose a track
vehicle is stuck to the ground. If angular velocities of both sprockets are constant

RecurDyn

with different speeds, significant differences of track tensions may be observed
due to the revolution of the chassis.
Figure 13 shows the dynamic changes of right and left track tensions when the
vehicle makes right turning with angular velocities of = -4.5 rad/sec and
= -5.4 rad/sec. It can be shown that the track tension of the upper part of
the left sprocket goes up significantly, while the track tension of the lower part
goes down.
right
left

7.6. FUTURE WORK AND CONCLUSIONS

The dynamic track tension for a high mobility tracked vehicle is investigated
in this paper. The three dimensional multibody tracked vehicle consists of the
hull, sprockets, road arms, road wheels, support rollers, and sophisticated
suspension systems of hydro-pneumatic and torsion bars. A compliant force
model is used to connect the rigid body track links. The tracked vehicle model
has 189 bodies, 36 pin joints and 152 compliant bushing elements and has 954
degrees of freedom. Various ground profiles are developed by using triangular
patch elements. Numerical results are validated against experimental results.
Numerical simulations have been carried out under various maneuvering
conditions and effects of several conditions are discussed . Numerical results
showed that the optimal track tension may not be necessarily 10 % of the total
vehicle weight as many track vehicle researchers have believed. Further studies
must be carried out to find the optimal track tension.

7-23

REFERENCES

[1] G. R. Doyle and G. H. Workman, 1979, ''Prediction of Track Tension when Traversing
an stacle'', Society of Automotive Engineers, 790416
[2] A.G. Galaitsis, 1984, "A Model for Predicting Dynamic Track Loads in Military
Vehicles", ASME, Journal of Vibration, Acoustics, Stress, and Reliability in Design, Vol.
106/289
[3] R.M. Trusty, M.D. Wilt, G.W. Carter, D.R. Lesuer, 1988, ''Field Measurement of Tension
in a T-142 Tank Track'', Experimental Techniques
[4] M. K. McCullough, and E. J. Haug, 1986, ''Dynamics of High Mobility Tracked
Vehicles'', ASME, Journal of Mechanisms Transmissions, and Automation in Design,
Vol.108, pp. 189-196
[5] Choi, J. H., Lee., H. C., Shabana, A. A., Jan. 1998, ''Spatial Dynamics of Multibody
Tracked Vehicles: Spatial Equations of Motion'', International Journal of Vehicle
Mechanics and Mobility, Vehicle System Dynamics, Vol. 29, pp. 27-49
[6] Lee, H. C., Choi., J. H., Shabana, A. A., Feb. 1998, ''Spatial Dynamics of Multibody
Tracked Vehicles: Contact Forces and Simulation Results'', International Journal of
Vehicle Mechanics and Mobility, Vehicle System Dynamics, Vol. 29, pp. 113-137
[7] C. Scholar and N. Perkins, 1997, "Longitudinal Vibration of Elastic Vehicle Track
System", SAE, 971090, International Congress and Exposition, Detroit, MI, Feb. 24-27
[8] J. Chung, J. M. Lee, 1994 ''A New Family of Explicit Time Integration Methods for
Linear and Non-linear Structural Dynamics'', International Journal for Numerical
Methods in Engineering, Vol.37, 3961-3976
[9] H. S. Ryu, D. S. Bae, J. H. Choi and A. Shabana, 2000 ''A Compliant Track Model For
High Speed, High Mobility Tracked Vehicle'', International Journal For Numerical
Methods in Engineering, Vol. 48, 1481-1502
[10] Changwon Proving Ground Construction Manual, 1996, Agency for Defense
Development, GWSD-809-960634
[11] Shabana A, 1989 ''Dynamics of Multibody Systems'', John Wiley & Sons, New York

RecurDyn

[12] Glen Simula, Nils Ruonavaara, and Jim Pakkals, 1999 "DTIS operation manual",
KRC, Michigan Technological University.
[13] ADAMS Reference Manual, Mechanical Dynamics, 2301 Commonwealth Blvd, Ann
Arbor, MI 48105.
[14] E. Kreyszig, 1983 Advanced Engineering Mathematics, 5th edition John Wiley &
Sons, New York
[15] Choi, J. H., 1996 Use of Recursive and Approximation Method in the Dynamic
Analysis of Spatial Tracked Vehicle, Ph. D. Thesis, The University of Illinois at
Chicago

7-25

RecurDyn


8

EFFICIENT CONTACT AND NONLINEAR
DYNAMIC MODELING FOR TRACKED VEHICLES

8.1. INTRODUCTION

In the dynamic analysis of vehicle system the mathematical modeling for the
system can be very different according to the objective of analysis. Sometimes
the mathematical modeling methods of vehicle systems are pursuing more
simplified effective models such as for real-time analysis or for the system
design which doesnt require highly nonlinear effects. Oppositely, due to rapid
developments of computer hardware and numerical technologies, researchers
and engineers can construct super detail nonlinear dynamic models which have
several hundreds, even several thousands of degrees of freedom systems, which
has the same phenomena as physical system. The objective of this research is to
built a reliable tracked vehicle dynamic analysis model so as to design a dynamic
track tensioning control system for high speed tracked vehicle based on
multibody dynamic modeling techniques. One of the key points for the dynamic
analysis of tracked vehicle is to predict the dynamic track tension when the
vehicle operates on various ground. In order to satisfy such objective of the
research, the track links of the track system should be modeled as a rigid body
which has six degrees of freedom connected by bushing force elements.
In early 80s several dynamic modeling techniques for track systems have
been developed in universities, research institutes and companies. McCullough
and Haug[1] designed a super element that represents spatial dynamics of high
mobility tracked vehicle suspension systems. The track was modeled as an
internal force element that acts between ground, wheels and the chassis of the
vehicle. Track tension was computed from a relaxed catenary relationship.
Empirical normal and shear force formulas based on constitutive relations from
soil mechanics were used to model the soil-track interface. Frank Huck[2]

RecurDyn

EFFICIENT CONTACT AND NONLINEAR DYNAMIC MODELING FOR TRACKED VEHICLES
introduced a planar multibody dynamic model of track type tractor by using
DRAM software. In this investigation each track link was modeled as a rigid
body and contact force analyses of sprocket, rollers and soil ground were
represented as a pioneer works. Similar modeling technique was also developed
by Tajima[3] at the similar period. The Komatsu Ltd. in-house multibody
program developed by Tajima is used to simulate planar multibody tracked
vehicles. The contact search mechanics and dynamic analysis of planar rigid
body track system are clearly introduced by Nakanishi and Shabana[4].
Nakanishis work was extended for three dimensional analysis by Choi and
Shabana[5, 6], and simultaneously Wehage[7] also developed the full three
dimensional tracked vehicle model, which considers the track link as a rigid
body, under research project in Caterpillar Inc. Whereas Choi used to connect
each rigid tracked link by one degree of freedom pin joint, bushing force
elements were used to connect for rigid track link by Wehage. Chois work
shows an possibility of very difficult numerical solution however it fails to give
more freedoms in real world than Wehages approach. Ryu et al.[8] extended
previously developed track system modeling techniques for the high mobility
military tracked vehicle which adopts sophisticated suspension and tensioning
systems. In this investigation a new variable step algorithm is implemented into
G-Alpha integrator which gives high numerical damping to integrate smoothly
high frequency and impulsive contact and bushing forces.
There are several reasons why many researchers has tried to develop rigid
multibody track systems even though such modeling techniques burden heavily
for numerical solutions. Unlike tires of wheeled vehicles track system causes
many problems such as separations or failures of connections, etc., furthermore it
is very expensive to maintain and has relatively weak durability. Because of
superiority of track system on very hostile terrain it cannot be replaced by
wheeled system, thus researchers should have solved these difficulties of the
tracked vehicle system. In the beginning of the research several simple modeling
techniques had been introduced however those gave a conclusion that each track
link should be considered as a rigid body to satisfy requirements. For instant, one
of the key issue for tracked vehicle is track tension since track tension has
significant roles for the vehicle maneuverings as focused in this investigation.
Very few works have been performed for the analysis of track tension based on

8-3

empirical or simple numerical analysis. Doyle and Workman[9] presented a
static prediction of track tension when the suspensioned tracked vehicle traverses
obstacles using two dimensional finite element methods. An elastic beam
element subjected to tension, compression and bending loads was utilized to
model track links. Galaitsis[10] demonstrated that the analytically predicted
dynamic track tension and suspension loads of a high-speed tracked vehicle are
useful in evaluating the dynamic analysis of the vehicle. The predicted track
tensions were compared with the empirically measured track tensions. A detailed
track tension measurement methodology and results are presented by Trusty et
al.[11]. Strain gages connected to a portable data acquisition system were
installed in the track link. The flat ground, quick acceleration, traversal of
obstacle courses, pivot turns, moving uphill, and pre and post tension, were used
for the tension investigation scenarios. Choi et al.[12] predicted and showed the
effect of dynamic track tension for the vehicle by using multibody techniques.
This research focuses on a heavy military tracked vehicle which has
sophisticated suspension and rubber bushed track systems. Various virtual
proving ground models are developed to observe dynamic changes of the track
tension. The predicted dynamic track tensions are validated against the
experimental measurements.
In this investigation for the sake of efficient development of dynamic track
tensioning system for suspensioned high speed military tracked vehicle, detail
nonlinear dynamic modeling methods which can partially replace physical
prototype models are presented. For the multibody dynamic modeling techniques
of the tracked vehicle used in this research several new methods are developed
and suggested. Those are efficient contact detecting kinematics for sprockets,
wheels and track links, parameter extraction techniques from component
experimental test, and a method how to apply Bekkers[13] soil theory for
multibody track and soil interactions. The simulation results are correlated by
newly developed experimental measurement techniques in this investigation.

RecurDyn


8.2. MULTIBODY TRACKED VEHICLE MODEL AND PARAMETER
EXTRACTIONS

The tracked vehicle model used in this investigation is a military purpose high
speed tank system which has sophisticated suspension system to damp out
impacts from hostile ground. In general this type of vehicle can be divided four
subsystems for overall motion analysis of vehicle dynamics. These subsystems
are two track subsystem with suspension units, main body subsystem with power
pack, and turret subsystem with main gun. The each right and left track
subsystems is composed of rubber bushed track link, double sprockets with
single retainer, seven road wheels and arms, and three upper rollers. The
sprockets, road arms, road wheels, upper rollers and turrets are mounted on main
body by revolute joints which allow single degrees of freedom. Total 38 revolute
joints are used for the vehicle modeling and generate 190 nonlinear algebraic
constraint equations. Two busing force elements to connect each track links and
total 304 bushing forces elements for both track systems are used in this
investigation. The modeled vehicle has 191 rigid bodies and 956 degrees of
freedom. Figure 1 shows a computer graphics model for tracked vehicle used in
this investigation.

Turret system
Chassis system
Track system

Figure 1. Computer graphics of high speed tracked vehicle model

8-5

8.3. EFFICIENT CONTACT SEARCH ALGORITHM

The interactions between the track links and the road wheels, rollers, and
sprockets are explained in this section. When a track link travels around vehicle
components, its trajectory is controlled by contact forces. The contact forces can
be generated computationally by detecting of contact conditions. The contact
collision algorithms are composed of five main routines such as search routines
for, wheel and link contact, center guide and wheel contact, sprocket tooth and
link pin contact, side wall of link and sprocket contact, and ground and link shoe.
The contact points and penetration values are defined from the searching
routines. Then a concentrated contact force is used at the contacted position of
the contact surface of the bodies. A detailed discussion on the formulation of the
contact collision is represented by Choi et al[5,6] and Nakanishi and Shabana[4].
However, it is not efficient for each chassis component such as road wheels and
sprockets to search all track links in detail. Efficient search algorithms and
discretized terrain representation method are investigated, respectively.

8.3.1 ROAD WHEEL-TRACK LINK CONTACT

Each road wheel is usually composed of two wheels. The interactions between
road wheel and track link can be divided into two types contact, as shown in Fig.
4. One is road wheel-track link body contact and the other is wheel side-track
link center guide contact. Each track subsystem has 6 road wheels and 76 track
links. In order to search wheel-track link contact efficiently, the pre-search and
post-search algorithm is applied. In the pre-search, bounding circle relative to
road wheel center is defined. All of track links are considered to detect a starting
link and ending link which has a possibility of wheel contact. Post-search means
a detailed contact inspection for track links in a bounding circle. Once a starting
and an ending link are found at one time through pre-search prior to analysis,
only detailed search is carried out by using the information of starting link and
ending link from the next time step.

RecurDyn


Search direction
Bounding circle
ending link
starting link
Search direction

Figure 4. Wheel and track link interactions

8.3.2 SPROCKET-TRACK LINK CONTACT

The interactions between sprocket and track link can be divided into two types
contact, as shown in Fig. 5. One is sprocket-track link pin(end connector) contact
and the other is sprocket side-track link side. Each track subsystem has 1
sprocket and 76 track links, moreover a sprocket has many teeth. For the
efficient search of the sprocket-track link contact, contact search algorithm is
composed of the pre-search and post-search. In the pre-search, bounding circle
relative to sprocket center is defined. All of track links are employed to detect a
starting link and ending link which has a possibility of sprocket contact. Then,
track links from starting link are investigated the engagement with sprocket
valley. Post-search means a detailed contact inspection for track links in a
bounding circle. Once a starting and ending link is found at one time through
pre-search prior to analysis, only detailed search is carried out by using the
information of starting link and ending link from the next time step.

8-7

ending engagement
Search direction
Bounding circle
s
Search direction
tarting engagement
ending link
starting link
Figure 5. Sprocket and track link interactions

8.3.3 GROUND-TRACK LINK CONTACT

The ground interacting surface of a track link can be single or multiple, and
therefore, there are one surface or multiple surfaces on each track link that can
come into contact with the ground. The interacting surface of track link is
discretized and each contact node points were defined. The global position
vectors that define the locations of points on the shoe plates surface of track link
are expressed in terms of the generalized coordinates of the track links and are
used to predict whether or not the track chain link is in contact with the ground.
In order to construct various geometries of tracked vehicle paved proving ground,
such as bumping courses, trench course, inclined course, standard cross country
courses, discretized terrain representation methods using triangular patch
element are used in this investigation. A triangular patch element has three nodes
and a unit normal vector to describe plane equations of interaction grounds [16].

RecurDyn


DISCRETIZED TERRAIN REPRESENTATION
The virtual terrain model used in this investigation is a general three dimensional surface
defined as a series of triangular patch elements. Figure 6 shows an example of
obstacle course created by triangular patch surfaces. Most geometries of various
paved proving grounds for tracked vehicle can be easily represented by using
triangular patch.

Figure 6. Terrain representation (obstacle course)

The equation for the plane defined from three nodes can be written as

z a y a x a = + +
3 2 1
(3)

The three coefficients, , , and of the equations of plane can be
obtained by given three locations of triangular patch, and by using Cramer's rule,
these coefficients can be obtained[16].
1
a
2
a
3
a
Then the unit normal vector of the plane is defined as n

[
T
a a
a a
1
1
1
2 1
2
2
2
1
+ +
= n ]
(4)

METHODS OF FINITE CONTACT NODES FOR GROUND INTERACTIONS
Unlike wheel and surface contact, the interactions between track link and
ground are very complicated problems. This is because the track link has

8-9

irregular contact geometry and different material properties. Due to large number
of track links of each track subsystem, commonly used contact theory of
surface to surface interactions in finite element community can not be employed
for this work. Choi[17] suggested that element free finite contact nodes were
distributed on the contact surface of track link, which have their own stiffness
and damping characteristics. The relative indentations of each node were
monitored and positions are restored. The use of element free finite contact node
methods demonstrated clearly the computational efficiency for dynamic analysis
of track system. Based on the method developed by Choi[17], the interactions
between track link surface and triangular patch surface are developed in this
investigation.

Figure 7. Interaction between track shoe body and triangular patch element

Figure 7 shows the interaction between finite contact nodes of track link and
triangular patch surface. The perpendicular deformation scalar of contact
node of link on patch plane can be defined as
k
d
j i k

RecurDyn

k
P ij
k
d n r
1
= (5)

where is shown in Fig. 7 and unit vector is defined in Eq. (4). The
criterion of necessary condition for the contact to occur of node , which is not
sufficient, is
P 1
r
k
n
j

>
seperate d
contact d
ij
k
ij
k
: 0
: 0
(6)

If this conditions is satisfied, the position vector shown in Fig. 7 is used
to compute the node location whether contact point
jk
B
r
B of node is on the
patch plane . The position vector can be written as
j
k
jk
B
r

k k
ij
p
i i jk
B
d n u A R r
+ =
(7)

where is the transformation matrix associated with the orientation
coordinates of link and
i
A
i
ij
p
u is the local position vector of node in the
track link coordinate system. On the other hand, using scalar triple product if one
of the following conditions is satisfied
j

(8)
3 31
2 23
1 12
3 31
2 23
1 12
k
B
k
B
k
B
k
B
k
B
k
B
or
n r r
n r r
n r r
n r r
n r r
n r r

, then the node of link is in contact with patch element . j i k


In this investigation, the relative generalized coordinates are employed in
order to reduce the number of equations of motion and to avoid the difficulty

8-11

associated with the solution of differential and algebraic equations. Since the
track chains interact with the chassis components through contact forces and
adjacent track links are connected by compliant force elements, each track chain
link in the track chain has six degrees of freedom which are represented by three
translational coordinates and three Euler angles. Recursive kinematic equations
of tracked vehicles were presented by [8] and the equations of motion of the
chassis are given as follows :

) (
r
i
T r
i
T
q B M Q B q MB B &
&
& & = (9)

where and are relative independent coordinates, velocity
transformation matrix, and is the mass matrix, and is the generalized
external and internal force vector of the chassis subsystem, respectively. Since
there is no kinematic coupling between the chassis subsystem and track
subsystem, the equations of motion of the track subsystem can be written simply
as
r
i
q B
M Q
t t t
Q q M =
& &
(10)

where , and Q denote the mass matrix, the generalized coordinate
and force vectors for the track subsystem, respectively. Consequently, the
accelerations of the chassis and the track links can be obtained by solving Eqs.
(9) and (10)..
t
M
t
q
t

8.5. EXTENDED BEKKERS SOIL MODEL FOR MULTIBODY TRACK
SYSTEM

The interactions between track link and soil used in this investigation consist
of the normal pressure-sinkage and shear stress-shear displacement relationships.
Bekker[13] developed the bevameter technique to measure terrain characteristics
by the plate penetration and shear tests. He also proposed the equation for
pressure-sinkage relationship, given by

RecurDyn

n c
z k
b
k
z p ) ( ) (

+ =
(11)
where p is pressure, is the width of a rectangular contact area, b z is
sinkage, is the soil cohesive modulus, is the soil frictional modulus and
is the exponent of soil deformation. The value of , , and can be
obtained from empirical test. From the experimental observations[13], the range
between unloading and reloading can be approximated by a linear function in the
pressure-sinkage relationship.
c
k

k
n
c
k

k n

) ( ) ( z z k p z p
u u u
=
(12)

where p and z are the pressure and sinkage, respectively, during unloading
or reloading; and are the pressure and sinkage, respectively, when
unloading begins; and is the average slope of the unloading-reloading line.
The slope of the unloading-reloading represents the degree of elastic rebound. If
the slope is vertical, there is no elastic rebound. That means the terrain
deformation is entirely plastic.
u
p
u
z
u
k
The shear stress-shear displacement relationship proposed by Janosi and
Hanamoto[13] is used for tangential shear forces, given by

) 1 )( tan ( ) , (
/ K j
e p c z j

+ =
(13)

where is the shear stress, p is the normal pressure, is the shear
displacement, and
j
c are the cohesion and the angle of internal shearing
resistance of the terrain, respectively, and K is the shear deformation modulus.
In summary, the proposed equations are applied for track system and soil
interactions as;
Loading condition ( ) :
p
z z >

n c
z k
b
k
z p ) ( ) (

+ =
(14)
) 1 )( tan ( ) , (
/ K j
e p c z j

+ =
(15)

8-13

Unloading, Reloading condition ( ) :
u
z < z
u r
z z z if < <

) ( ) ( z z k p z p
u u u
=
(16)
) 1 )( tan ( ) , (
/ K j
e p c z j

+ = (17)
r
z z if <

0 ) ( = z p
(18)
0 ) , ( = z j
(19)

Loading condition after reloading ( )
u
z z >

n c
z k
b
k
z p ) ( ) (

+ =
(20)
) 1 )( tan ( ) , (
/ K j
e p c z j

+ =
(21)

where is sinkage at the previous time step and is sinkage when the
plastic effect of terrain is started during unloading. Figure 8 shows the
simulation response to normal load of a track link on dry sand terrain when the
vehicle is accelerated from the rest. The soil conditions for simulation
are , , = ,
p
z
95kN
r
z
1
/ . 0
+
=
n
c
m k
2
/ 43 . 1528
+
=
n
m kN k
c kPa 04 . 1 = , and =1.1. The

pattern of result agrees to the experimental result shown in reference[13].
o
28 n

u
z
r
z
0.00 0.03 0.06 0.09 0.12 0.15 0.18
0.0
5.0x10
4
1.0x10
5
1.5x10
5
2.0x10
5
2.5x10
5
reloading
unloading
loading
P
r
e
s
s
u
r
e

(
N
/
m
2
)
sinkage (meter)

Figure 8. Simulation response to normal load of a dry sand terrain

RecurDyn


Contact detect node
Contact segment area
Track link i
Figure 9. Mesh areas and detect nodes of a track link

As shown in the Fig. 9, if the node of link i is in contact with triangular
patch ground, the contact force at the contact segment area can be computed
using the equation as
j

) ( segment th j of area p F
ij ij
p
=
(22)
) ( segment th j of area F
ij ij
s
=
(23)

where and are the normal pressure and shear stress, respectively.
Using the expression for the contact force as defined by the preceding equation,
the contact force vector can be defined as
ij
p
ij

k
ij
s k
ij
p
ij
F F t n F
+ =
(24)

where and t
are a unit normal vector and tangential vector of ground

patch . The virtual work of the ground force at a track link, which has the
number of rectangle surface, is given by
k
n
n
k
k

= =
= =
n
j
ij
T
ij
n
j
ij i
W W
1 1
F r
(25)
where is a th node position vector of link i defined by inertia
reference frame.
ij
r j

8-15

8.6. SUMMARY AND CONCLUSION

For the sake of efficient component development of tracked vehicle at early
design stage, it is clearly proved that the multibody dynamic simulation methods
can be very useful tool. The presented three dimensional multibody tracked
vehicle consists of the hull, sprockets, road arms, road wheels, support rollers,
and sophisticated suspension systems of hydro-pneumatic and torsion bars. A
compliant force model is used to connect the rigid body track links. The tracked
vehicle model has 191 bodies, 38 pin joints and 304 compliant bushing elements
and has 956 degrees of freedom. The suspension, contact and bushing
characteristics are extracted by empirical measurements and implemented into
the simulation model. The efficient kinematic contact search algorisms between
track system and chassis components are suggested and implemented. Two
methods are developed for the interactions between track shoe body and ground.
When the distributed node points on shoe body surface detect contact condition,
direct forces are calculated based on the contact deformation on node points, or
pressure and shear forces on each segment areas of the contact surface are
calculated based on pressure-sinkage relationship and shear stress-shear
displacement relationship. In order to validate and construct the simulation
database, positions, velocities, accelerations and forces of the tracked vehicle are
measured empirically. The simulation results show very good agreements with
experimental measurements. Therefore, the suggested methods by using the
multibody dynamic technologies can be used efficiently for tracked vehicle
developments.

RecurDyn

REFERENCES

[1] M. K. McCullough, and E. J. Haug, 1986, ''Dynamics of High Mobility Tracked
Vehicles'', ASME, Journal of Mechanisms Transmissions, and Automation in Design,
Vol.108, pp. 189-196.
[2] F.B. Huck, ''A Case for Improved Soil Models in Tracked Machine Simulation'',
Caterpillar, Inc.
[3] Tajima, and T. Nakanishi Technical discussions Komatsu Ltd.
[4] Nakanishi, T., and Shabana, 1994 "Contact Forces in the Nonlinear Dynamic analysis
of Tracked Vehicles", International Journal for Numerical Methods in Engineering,
Vol.37, pp. 1251-1275.
[5] Choi, J. H., Lee., H. C., Shabana, A. A., Jan. 1998, ''Spatial Dynamics of Multibody
Tracked Vehicles: Spatial Equations of Motion'', International Journal of Vehicle
Mechanics and Mobility, Vehicle System Dynamics, Vol. 29, pp. 27-49.
[6] Lee, H. C., Choi., J. H., Shabana, A. A., Feb. 1998, ''Spatial Dynamics of Multibody
Tracked Vehicles: Contact Forces and Simulation Results'', International Journal of
Vehicle Mechanics and Mobility, Vehicle System Dynamics, Vol. 29, pp. 113-137.
[7] R. Wehage, F. Huck Technical discussions Caterpillar Inc.
[8] H. S. Ryu, D. S. Bae, J. H. Choi and A. Shabana, 2000 ''A Compliant Track Model For
High Speed, High Mobility Tracked Vehicle'', International Journal For Numerical
Methods in Engineering, Vol. 48, 1481-1502.
[9] G. R. Doyle and G. H. Workman, 1979, ''Prediction of Track Tension when Traversing
an stacle'', Society of Automotive Engineers, 790416.
[10] A.G. Galaitsis, 1984, "A Model for Predicting Dynamic Track Loads in Military
Vehicles", ASME, Journal of Vibration, Acoustics, Stress, and Reliability in Design, Vol.
106/289.
[11] R.M. Trusty, M.D. Wilt, G.W. Carter, D.R. Lesuer, 1988, ''Field Measurement of
Tension in a T-142 Tank Track'', Experimental Techniques.
[12] J. Choi, D. Park, H. Ryu, D. Bae, K. Huh, 2001 Dynamic Track Tension of High
Mobility Tracked Vehicles Proceedings of DETC01, ASME Third Symposium on

8-17
[13] J. Wong, 2001, Theory of Ground Vehicles 3
rd
Ed. John Wiley & Sons.
Multibody Dynamics and Vibration, Pittsburgh, PA, USA.
[14] Shabana A. 1996 Theory of Vibration, An Introduction, 2
nd
Ed. Springer: New York.
[15] Berg, M., 1998 ''A Non-Linear Rubber Spring Model for Rail Vehicle Dynamics
Analysis'', International Journal of Vehicle Mechanics and Mobility, Vehicle System
Dynamics, Vol. 30, pp. 197-212.
[16] ADAMS Reference Manual, Mechanical Dynamics, 2301 Commonwealth Blvd, Ann
Arbor, MI 48105.
[17] Choi, J. H., 1996 Use of Recursive and Approximation Methods in The Dynamic
Analysis of Spatial Tracked Vehicle, Ph. D. Thesis, The University of Illinois at
Chicago.

RecurDyn


9

AN EFFICIENT CONTACT SEARCH ALGORITHM
FOR GENERAL MULTIBODY SYSTEM DYNAMICS

9.1. INTRODUCTION

This paper presents a contact analysis algorithm employing the relative
coordinate system for the multibody system dynamics. Multiple-contact higher
pairs are widely used in mechanical systems such as walking machines, feeding
systems, driving chains, and tracks of off-road vehicles. Common design
problems due to the multiple contacts among bodies are undercutting, jamming,
backlash, and body interference.
The configuration space representation of a higher pair was proposed by
Lozano-Perez [1] for robot motion planning. Sacks extended the configuration
space concept in [2] for efficient detection of contact pairs. The relative position
and orientation of a pair were mapped into the configuration space. The degrees
of freedom of a pair became the dimension of the configuration space, which is
divided into free space and contact space in the preprocessing stage of a dynamic
analysis and is tabulated into a database. Run time query is made to decide
whether a pair is currently in contact or not. When a higher pair has many
degrees of freedom, formation of the configuration space and processing effort
for a run time query may become extensive.
Wang presented an interference analysis method in [3]. Relative coordinates
were defined for a contact pair and a kinematic closed loop including the contact
pair was formed. Constraint equations arising from closed loops are solved for
the relative coordinates including the ones for the contact pair. The canonical
Hamiltonian formulation is used to derive a minimal set of dynamic equations of
motion.
Mirtich proposed a contact detection algorithm consisting of narrow and broad
phases in [4]. Candidate features are selected in the broad phase and contact

RecurDyn

AN EFFICIENT CONTACT SEARCH ALGORITHM FOR GENERAL MULTIBODY SYSTEM DYNAMICS
inspection is carried out in the narrow phase only among the candidate features.
Haug presented a formulation for domains of mobility that characterizes
kinematic boundaries of multiple contact pairs in [5]. A surface-surface contact
joint was developed by Nelson in [6]. Piecewise dynamic analysis method for a
contact problem was employed in [7, 8]. Dynamic analysis is halted when a
contact pair is detected to be in contact and is resumed with new velocities that
are calculated from the momentum balance equations. One of drawbacks of this
method is that too frequent halting and resuming of the numerical integration
may occur when a contact pair toggles between contact and not contact status.
Zhong summarized many contact search algorithms in the area of the finite
element analysis in [9]. All geometric variables necessary to detect a contact
were expressed in the absolute Cartesian coordinate system. The penalty and
Lagrange multiplier methods were proposed. The compliant contact model that
is based on the Herzian law was used in [10]. Since the contact force is large and
varied significantly, the differential equations of motion for this method are
generally stiff.
A recursive formulation using the relative coordinates was proposed by Bae in
Ref. [11]. The equations of motion were derived in a compact matrix form by
using the velocity transformation method. The actual computation was carried
out by using the recursive formulas developed for each joints. Realtime
simulation of a vehicle system is carried out by the recursive method in Ref. [12].
The Jacobian matrix was updated once in while during time marching of the
numerical integration. The recursive method was extended to the flexible body
dynamics of constrained mechanical systems in Ref. [13]. A virtual body
concept was employed to relieve the implementation burden of the flexible body
dynamics coding. A compliant track link model was developed for tracked
vehicles in Ref. [14]. A minimum set of the equations of motion was obtained by
the recursive method. Concept of the configuration design variable with the
recursive formulation was introduced in Ref. [15]. The recursive method is
applied to efficiently detect a contact in this research.
This paper presents a hybrid contact detection algorithm of the configuration
space method and bounding box method in conjunction with the compliant
contact model. Two bodies of a contact pair are logically considered as a defense
body on which the contact reference frame is defined and as a hitting body that

9-3

moves relative to the defense body, respectively. Contour of the defense body is
approximated by many triangular patches which are projected on axes of the
contact reference frame. Bounding box inside which contains base surface is
divided into several blocks each of which is indexed on axis of the contact
reference frame. Contact inspection for a contact pair is processed in the
sequence of broad and narrow phases. Relative position vector of the hitting
body to the defense body is projected on the axes of the contact reference frame
and select candidate features that may come in contact shortly in the broad
inspection phase, which greatly reduces the searching effort. It is not needed any
database to be built prior to an analysis. Since the searching algorithm is coupled
with stepping algorithm of the numerical integration, a strategy for deciding an
integration stepsize is proposed. A numerical example is presented to
demonstrate the validity of the proposed method.

9.2. KINEMATIC NOTATIONS OF A CONTACT PAIR

Consider a contact pair shown in Fig. 1. Two bodies of the contact pair will be
referred as a hitting body and a defense body for convenience in the following
discussions, respectively. The contours of the hitting and defense bodies will be
referred as the hitting and target boundaries, respectively.

Figure 1 Kinematic notations of a contact pair

RecurDyn


Figure 2 Contact reference frame and generalized coordinate

The coordinate system is the inertial reference frame and the
primed coordinate systems are the body reference frames. The
orientation and position of the body reference frame is denoted by and ,
respectively.
Z Y X
z y x
A r
Double primed coordinate systems are the node reference frame of the hitting
body and the surface and contact reference frames of the defense body,
respectively. All geometric variables of the defense body are measured on the
surface reference frame. The contact reference frame for the contact pair is
defined on the left corner of the bounding box of the defense body, as shown in
Fig. 2. The relative position and orientation of the hitting body to the defense
body are defined as the generalized coordinates, which are denoted by and
as shown in Fig. 2. Therefore the generalized coordinates are directly used
to detect a contact for the pair.
ch
d
ch
A

9.3. DIVISION OF THE CONTACT DOMAIN

A surface-to-surface contact problem can be replaced by multiple sphere-to-
surface contact problems. Therefore, the sphere-to-surface contact problem will
be discussed in this research.

9-5

Contour of a smoothly shaped body has been represented by the 3D
NURBS(Non-Uniform Rational B-Spline)[16] in many commercial CAD
programs. Since it is computationally extensive to find intersection lines or
points between two surfaces, the defense surface is approximated by triangular
patches and the boundary of the hitting body is represented by a set of spheres,
as shown in Fig. 3. The numbers of patches and spheres must be decided by the
degree of accuracy required.

Figure 3 Approximated defense and hitting surfaces

The bounding box of the defense surface in space can be divided into many
blocks each of which has a list of patches lying inside or on the block to
efficiently process a contact detection, as shown in Fig. 4. Since the block
locations are tabulated with respect to the contact reference frame attached to the
defense body, they are constant. As a result, the locations do not needed to be
calculated at every time steps, which significantly reduces computation time
associated with the contact search.

RecurDyn


Figure 4 Relationship patch and block: The patch, p belongs in the block, b

9.4. PRE-SEARCH

Every pairs of the boundary nodes of the hitting body and the patches on the
defense body must be examined to detect a contact between two bodies, which is
computationally extensive. In order to save the extensive computation, each node
of the hitting body searches to find blocks of the contact domain to which it
belongs in the pre-search stage, as shown in Fig. 5.
The relative position and orientation of the hitting body reference frame with
respect to the contact reference frame shown in Fig. 2 can be directly available
from the generalized coordinate d and . Therefore, the relative nodal
position of the hitting body with respect to the contact reference frame is
obtained as
ch

ch
A

n ch ch cn
s A d d + =
(1)

where is the nodal position with respect to the hitting body reference
frame. Direct comparison of the
n
s
cn
d with this of the block locations of the
defense body yields the state of a contact.

9-7

Figure 5 Node and blocks in pre-search stage

If a pair of a node and a block is in contact, post-search step will be proceeded.
The bounding box of the defense body is divided into many blocks. Each block
has a list of patches lying within or on the block boundary. Therefore, the post-
search step will be carried out only for the patches belonging to the blocks that
have found to be in contact in pre-search step, as shown in Fig. 5.

9.5. POST-SEARCH AND COMPLIANCE CONTACT FORCE

The candidate patches on the defense surface have been selected for the post
search step in the pre-search step. For the candidate patches, it is necessary to
compute the amount of penetration to generate the contact forces, as shown in
Fig. 6.

Figure 6 Node and patch in post-search stage

RecurDyn

The relative position of a node with respect to the patch reference frame
is obtained as follows.
pn
d
1
p cn pn
s d d =
(2)
where
The vector
pn
d is projected into the patch reference frame as

pn
T
p pn
d C d =
(3)

where is the orientation matrix of the patch reference frame with respect
to the contact reference frame.
p
C
The first step in the post search is to check whether the node is in contact with
the patch or not by inspecting
pn
d . In case of non-contact, the rest of procedures
must be skipped. Otherwise, the penetration of the node into the patch is
calculated by

pn
T
p
- r d n =
(4)

where is always positive. The
p
n is a normal vector of a patch and a
constant vector with respect to the patch reference frame.
Thus, the contact normal force is obtained by

3
2
1
m
m
m
n

c k f
&
&
&
+ =
(5)

where k and are the spring and damping coefficients which are determined
by an experimental method, respectively and the is time differentiation of .
The exponents m and generates a non-linear contact force and the
exponent yields an indentation damping effect. When the penetration is
very small, the contact force may be negative due to a negative damping force,
which is not realistic. This situation can be overcome by using the indentation
damping exponent greater than one.
c
&

1 2
m
3
m

9-9

The friction force is obtained by

n f
f f =
(6)

where is the friction coefficient and its sign and magnitude can be
determined from the relative velocity of the pair on contact position.

9.6. KINEMATICS AND EQUATION OF MOTION FOR THE RECURSIVE
FORMULAS
A contact search algorithm is proposed in the previous sections. The proposed
method makes use of the relative position and orientation matrix for a contact
pair. This section presents the relative coordinate kinematics for a contact pair as
well as for joints connecting two bodies.
Translational and angular velocities of the z y x frame in the
frame are respectively defined as
Z Y X

w
r
&
(7)

Their corresponding quantities in the z y x frame are defined as

=
w A
r A
w
r
Y
T
T
& &
(8)

where is the combined velocity of the translation and rotation. The
recursive velocity and virtual relationship for a pair of contiguous bodies are
obtained in [17] as
Y
1)i (i 1)i2 (i 1) (i 1)i1 (i i
+ = q B Y B Y
&
(9)

where denotes the relative coordinate vector. It is important to note that
matrices and are only functions of the . Similarly, the
1)i (i
q
1)i1 (i
B
1)i2 (i
B
1)i (i
q

RecurDyn

recursive virtual displacement relationship is obtained as follows

1)i (i 1)i2 (i 1) (i 1)i1 (i i

+ = q B Z B Z
(10)

If the recursive formula in Eq. (9) is respectively applied to all joints, the
following relationship between the Cartesian and relative generalized velocities
can be obtained:
q B Y
&
=
(11)

where is the collection of coefficients of the and B
1)i (i
q&

[ ]
T
1 nc
T T
2
T
1
T
0
=
n
Y , , Y , Y , Y Y K (12)
[ ]
T
1 nr
T
) 1 (
T
12
T
01
T
0

=
n n
q , , q , q , Y q
&
K
& & &
(13)

where nc and nr denote the number of the Cartesian and relative coordinates,
respectively. Since in Eq. (11) is an arbitrary vector in q&
nr
R , Eqs. (9) and (11),
which are computationally equivalent, are actually valid for any vector
such that
nr
R x &
x B X & =
(14)
and
1)i - (i 1)i2 - (i 1) - (i 1)i1 - (i i
x B X B X + =
(15)

result, transformation of into is actually calculated by
recursively applying Eq. (15) to achieve computational efficiency in this
research.
nc
R X B x
nr
R x
nc
R Bx
Inversely, it is often necessary to transform a vector in G
nc
R into a new
vector in G B g
T
=
nr
R . Such a transformation can be found in the generalized
force computation in the joint space with a known force in the Cartesian space.
The virtual work done by a Cartesian force is obtained as follows.
nc
R Q

9-11

Q
Z W = (16)

where must be kinematically admissible for all joints in a system.
Substitution of
Z
q B Z = into Eq. (16) yields

* T T T
Q q Q B q W = = (17)
where . Q B Q
T *
The equations of motion for constrained systems have been obtained as

follows.
0 ) Q Y M B F

T
= + =
&
( (18)

where the is the Lagrange multiplier vector for cut joints [18] in
m
R
and represents the position level constraint vector in
m
R . The and Q
are the mass matrix and force vector in the Cartesian space including the contact
forces, respectively.
M
The equations of motion and the position level constraint can be implicitly
rewritten by introducing v q= & as

0 a , v q F = ) , , (
(19)
0 q = ) (
(20)

Successive differentiations of the position level constraint yield

(21)
0 v v q
q
= = ) , (
&
0 v v v q
q
= = & &
& &
) , , (
(22)

RecurDyn

Equation (19) and all levels of constraints comprise the over determined
differential algebraic system (ODAS). An algorithm for the backward
differentiation formula (BDF) to solve the ODAS is given in [19] as follows.
0
v v U
v q U
v v q
v q
q
) v , v q ( F
x H =
+ +
+ +
) (
) (
) (
=
) (
) (
, ,
,
, ,
) (
2 0
T
0
1 0
T
0
&
&
& &
&
&
(23)

where [ ]
T T T T T
, v , v , q x & = , , and are determined by the
coefficients of the implicit integrators and is an nr matrix such
that the augmented square matrix is nonsingular.
0
U
T
0
1
0
U m) (nr
The number of equations and the number of unknowns in Eq. (23) are the
same, and so Eq. (23) can be solved for . Newton Raphson method can be
applied to obtain the solution .
n
x
n
x

H x H
x
=
(24)
1,2,3,... i ,
i 1 i
= + =
+
x x x
(25)
0
0 U U 0
0 0 U U
0
0 0
0 0 0
F F F F
H
T
0
a v q
v q
q
q q q q
x
=
=
T
0 0
T
0 0
T
0
& & & & & &

& &
(26)

Recursive formulas for and in Eq. (24) are derived to evaluate them
efficiently.
x
H H

9.7. NUMERICAL INTEGRATION STRATEGY

The sufficient condition for a successful numerical integration step is to
satisfy both accuracy and stability of the state variables for a system without

9-13

contact. Satisfaction of the accuracy and stability is not sufficient for a system
with a contact. Suppose a bullet collides with an object. If the object is thin, the
bullet passes through the object without noticing it. If the object is thick and a
moderately large step size satisfies both the accuracy and stability, the bullet
penetrates too deep
at the first step of a contact. Large and sudden contact force due to the large
penetration generally introduces a large numerical error in the state variables.
The large numerical error often causes the integration step to fail. Therefore, the
contact condition must be considered in deciding an integration step. In order to
make a system transition from a non-contact status to a contact status smooth as
much as possible, time of contact must be predicted accurately. However, the
computationally extensive search algorithm must be triggered to predict the
exact time of a contact even though two bodies of a contact pair are located in a
distance. Easy and practical solution to this problem is to use the method of
backtracking.

Figure 7 Buffer radius of a node

This paper adopted the concept of buffer radius shown in Fig. 7. In post-search
stage, if no nodes with radius in the hitting body is contacted with the candidate
lines in the defense body and some nodes with buffer radius are contacted, the
integrating step will be decreased.

9.8. NUMERICAL EXAMPLE

The proposed algorithm is implemented in the commercial program RecurDyn.

RecurDyn

A paper-feeding problem of a copying machine is solved to demonstrate the
efficiency and validity of the proposed method.

Figure 8 Copying machine

The system has 255 degree of freedom and consists of five roller pairs and one
paper shown in Fig. 8. Each roller pair is modeled by using two driving rollers,
two idlers, two driving bars, two idler bars, six joints and one nip spring. The
paper is modeled by using 40-segmented bodies and 28 plate force elements. The
segmented paper bodies and the roller pairs are contacted and it is modeled by
using 160 sphere to surface contacts.
The paper goes through a path while contacting the roller pairs. The angular
velocity of each driving roller reaches 10 rad/sec during one second. The
tangential velocities of a driving roller and a leading segment body of the paper
are shown in Fig. 9. The static and dynamic friction coefficient is 0.5 and 0.3,
respectively.
The analysis was performed on an IBM compatible computer (PIII-933Mhz)
and took about 260 sec. per 1 sec. for simulation. A copying machine is solved to
demonstrate the effectiveness of the proposed algorithm

9-15

Figure 9 Tangential velocities of a driving roller and a leading segment body of paper
on a contact point

9.9. CONCLUSIONS

This research proposes an efficient implementation algorithm for contact
mechanisms. The contact domain is divided into many blocks each of which
contains the list of patches inside it. The search process consists of pre-search
and post search steps. In the pre-search step, the bounding box technique is
employed to find approximate contact state. Once the contact is detected in the
pre-search step, the detailed contact condition is further examined in the post-
search step. The compliance contact model is used to generate the contact force
which is applied to the hitting and defense bodies. The relative coordinate
formulation is used to generate the equations of motion. The local
parameterization method is used to solve the differential algebraic equations.
The integration stepsize is automatically reduced when a contact is expected
soon. The proposed algorithm is implemented in the commercial program
RecurDyn and a copying machine example is successfully solved.

RecurDyn

REFERENCES

1. Lozano-Perez, T., "Spatial Planning: A Configuration Space Approach", IEEE
Transactions on Computers, Vol. C-32, IEEE Press, 1983.
2. Sacks, E. and Joskowicz, L., "Dynamical Simulation of Planar Systems with Changing
Contacts Using Configuration Spaces", "Journal of Mechanical Design", Vol. 120, pp.
181~187, 1998.
3. Wang, D., Conti, C. and Beale, D., "Interference Impact Analysis of Multibody Systems",
"Journal of Mechanical Design", Vol. 121, pp. 121-135, 1999.
4. Mirtich, B. V., "Impulse-based Dynamic Simulation of Rigid Body Systems", Ph. D thesis,
University of California, Berkeley, 1996.
5. Haug, E. J., Wu, S. C. and Yang, S. M., "Dynamic mechanical systems with Coulomb
friction, stiction, impact and constraint addition-deletion, I: Theory", "Mech. Mach.
Theory", Vol. 21(5), pp. 407-416, 1986.
6. Nelson, D. D. and Cohen, E., "User Interaction with CAD Models with Nonholonomic
Parametric Surface Constraints", Proceedings of the ASME Dynamic Systems and
Control Division, DSC-Vol. 64, pp. 235-242, 1998.
7. Wang, D., Conti, C., Dehombreux, P. and Verlinden, O., "A Computer-aided
Simulation Approach for Mechanisms with Time-Varying Topology", "Computers and
Structures", Vol. 64, pp. 519-530, 1997.
8. Wang, D., "A Computer-aided Kinematics and Dynamics of Multibody Systems with
Contact Joints", Ph. D Thesis, Mons Polytechnic University Belgium, 1996.
9. Zhong, Z. Z., "Finite Element Procedures for Contact-Impact Problems", Oxford
University Press, 1993
10. Lankarani, H. M., "Canonical Impulse-Momentum Equations for Impact Analysis of
Multibody System", ASME, "Journal of Mechanical Design", Vol. 180, pp. 180-186,
1992
11. Bae, D. S., Han, J. M., and Yoo., H. H., A Generalized Recursive Formulation for
Constrained Mechanical System Dynamics, Mech. Struct. & Mach., Vol. 27(3), pp.

9-17
12. Bae, D. S., Lee, J. K., Cho, H. J., and Yae, H., An Explicit Integration Method for
Realtime Simulation of Multibody Vehicle Models, Computer Methods in Applied
293-315, 1999.
13. Bae, D. S., Han, J. M., Choi, J. H., and Yang, S. M., A Generalized Recursive
Formulation for Constrained Flexible Multibody Dynamics, International Journal for
Numerical Methods in Engineering, Vol. 50, pp. 1841-1859, 2001.
14. Ryu, H. S., Bae, D. S., Choi, J. H., and Shabana, A. A., A Compliant Track Link Model
for High-speed, High-mobility Tracked Vehicles, International Journal for Numerical
Methods in Engineering, Vol. 48, pp. 1481-1502, 2000.
15. Kim, H. W., Bae, D. S., and Choi, K. K., Configuration Design Sensitivity Analysis of
Dynamics for Constrained Mechanical Systems, Computer Methods in Applied
16. Farin, G., "Curves and Surfaces for Computer-aided Geometic Design", Academic
Press, 1997.
17. Angeles, J., "Fundamentals of Robotic Mechanical Systems", Springer, 1997.
18. Wittenburg, J., "Dynamics of Systems of Rigid Bodies", B. G. Teubner, Stuttgart, 1977.
19. Yen, J., Haug, E. J. and Potra, F. A., "Numerical Method for Constrained Equations
of Motion in Mechanical Systems Dynamics", Technical Report R-92, Center for
Simulation and Design Optimization, Department of Mechanical Engineering, and
Department of Mathematics, University of Iowa, Iowa City, Iowa, 1990.

RecurDyn


10

LINEARIZED EQUATIONS OF MOTION FOR
MULTIBODY SYSTEMS WITH CLOSED LOOPS

10.1. INTRODUCTION

Linearization is an important tool in understanding the system behavior of a
nonlinear system at a certain state. As an example, the eigenvalues of the
linearized equations of motion are very useful information in developing control
logics. Linearization of an unconstrained system is relatively easier than that of
the constrained systems due to the algebraic constraint equations and
corresponding Lagrange multipliers. This research proposes a linearization
method for the constrained mechanical systems and compares the results with
those obtained from other methods.
Sohoni [1] presented an approach for automatically generating a linearized
dynamical model, which is derived from the nonlinear equations of motion. The
Lagrange multiplier term was kept constant in the linearized equations of motion.
The velocity and acceleration level constraints have not been considered in the
resulting linearized equations of motion. Neuman symbolically generated the
dynamic robot model by Lagrange-Euler formulation and linearized the dynamic
model about a nominal trajectory [2]. Balafoutis presented a computational
method for recursive evaluation of linearized dynamic robot model about a
nominal trajectory [3]. The formulation was applied to the robot systems, which
are unconstrained systems. This formulation was generalized by Gontier [4] for
general unconstrained mechanical systems. Similar formulations have been
developed by the variational approach in Refs. [5,6]. A recursive formulation
using the relative coordinates was proposed by Bae in Ref. [7]. The equations of
motion were derived in a compact matrix form by using the velocity
transformation method. The actual computation was carried out by using the
recursive formulas developed for each joints. Realtime simulation of a vehicle

RecurDyn

LINEARIZED EQUATIONS OF MOTION FOR MULTIBODY SYSTEMS WITH CLOSED LOOPS
system has been carried out by the recursive method in Ref. [8]. The Jacobian
matrix was updated once in a while during time marching of the numerical
integration. The recursive method was extended to the flexible body dynamics of
constrained mechanical systems in Ref. [9]. A virtual body concept was
employed to relieve the implementation burden of the flexible body dynamics
coding. A compliant track link model was developed for tracked vehicles in Ref.
[10]. A minimum set of the equations of motion was obtained by the recursive
method. Concept of the configuration design variable with the recursive
formulation was introduced in Ref. [11].
The equations of motion for multibody systems are highly nonlinear with
respect to the relative positions, velocities, and accelerations. The equations of
motion are perturbed to obtain the linearized equations of motion. Since the
equations of motion are highly nonlinear, their perturbation involves with many
arithmetic operations for a multibody system consisting of many bodies and
joints. In case of open loop systems which do not have any constraints, the
equations of motion result in the ordinary differential equations whose partial
derivatives with respect to the relative coordinates, velocities, and accelerations
has been obtained by several different methods in Refs. [2,3,4]. In case of closed
loop systems which have constraints, these method cannot be used directly any
more due to the constraints and corresponding Lagrange multipliers.
One of the intuitive methods to handle the constraints is to directly express the
equations of motion only in terms of the independent relative positions,
velocities, and accelerations. In order to achieve this goal, the relative
coordinates must be divided into the independent and dependent coordinates and
the dependent coordinates, velocities, and accelerations must be directly
expressed in terms of independent ones. However, the independent and
dependent coordinates, velocities, and accelerations are tightly and nonlinearly
coupled by the position, velocity, and acceleration level constraints and the
equations of motion are implicit function of the coordinates, velocities, and
accelerations. As a result, it is very difficult to directly express the dependent
coordinates, velocities, and accelerations in terms of independent ones and
consequently to express the equations of motion only in terms of the independent
coordinates, velocities, and accelerations.
The null space of the constraint Jacobian is first pre-multiplied to the

10-3

equations of motion to eliminate the Lagrange multiplier and the equations of
motion are reduced down to a minimum set of ordinary differential equations.
The resulting differential equations are still functions of all relative coordinates,
velocities, and accelerations. Since the coordinates, velocities, and accelerations
are tightly coupled by the position, velocity, and acceleration level constraints,
direct substitution of the relationships among these variables yields very
complicated equations to be implemented. As a consequence, the reduced
equations of motion are perturbed with respect to the variations of all coordinates,
velocities, and accelerations, which are coupled by the constraints. The position,
velocity and acceleration level constraints are also perturbed to obtain the
relationships between the variations of all relative coordinates, velocities, and
accelerations and variations of the independent ones. The perturbed constraint
equations are then simultaneously solved for variations of all coordinates,
velocities, and accelerations only in terms of the variations of the independent
coordinates, velocities, and accelerations. Finally, the relationships between the
variations of all coordinates, velocities, accelerations and these of the
independent ones are substituted into the variational equations of motion to
obtain the linearized equations of motion only in terms of the independent
coordinate, velocity, and acceleration variations.
The proposed method is implemented in the commercial program RecurDyn.
Vibration analyses of a four bar mechanism and a vehicle system are carried out
to demonstrate the validity of the proposed method.

RecurDyn


10.2. RELATIVE COORDINATE KINEMATICS

X
Z
Y
i
r
i
y
i
x
i
z
i
O
O
i
f
i
g
i
h
Figure 1 Coordinate systems and a rigid body

Figure 1 shows the coordinate system fixed on a body . In the figure, the
frame is the body reference frame and the
i
i i i
z y x Z Y X frame is the
inertial reference frame. Point O is the origin of Z Y X , point is the
origin of
i
O
i i i
z y x , and is the position vector of from O . The , ,
and are unit vectors along the
i
r
i
O
i
f
i
g
i
h x ,
i
y , and
i
z axes, respectively.
Orientation matrix of the body is given as

|
i i i i
h g f A = |
(1)
Velocities and virtual displacements of point in the X
i
O Z Y frame are
defined as (see Refs. [4-5])
(
=
i
i
i
r
Y
&
(2)
(
=
i
i
r
Z
(3)

Their corresponding quantities in the
i i i
z y x frame are defined as

10-5

(4) (
i
T
i
i
T
i
A
r A
&
=
i
i
i
r
Y
&
(5) (
=
i
T
i
i
T
i
i
i
A
r A

r
Z
Figure 2 Kinematic relationships between two adjacent rigid bodies

X
Z
Y
i
y
i
x
i
z
) 1 ( i i
s
i i ) 1 (
d
) 1 (
y

i
) 1 (
x

i
) 1 (
z

i
i
O
) 1 ( i
O
i
r
) 1 ( i
r
i i ) 1 (
s

A pair of contiguous bodies is shown in Figure 2. Body 1 i is assumed to
be an inboard body of body and the position of point is i
i
O

1) i(i 1)i (i 1)i (i 1) (i i
+ + = s d s r r
(6)

By using Eq. (5), the angular virtual displacement of body in its local
reference frame is
i
(7)
1)i (i 1)i (i
T
1)i (i 1) - (i
T
1)i (i i

+ = q H A A

where is determined by the axis of rotation and is defined as
i i ) 1 (
H
1)i (i
A

i
T
1) (i 1)i (i
A A A

=
(8)

Taking variation of Eq. (6) yields

RecurDyn

1)i (i 1)i (i
1)i (i
T
1) i(i 1)i (i 1)i (i
1)i (i
T
1) (i
1)i (i
T
1) - i(i 1)i (i 1)i (i 1)i (i
1)i (i
T
1) (i
1)i (i
T
i
)
~
) ( (
)
~
~
~
(

1)i (i

+ +
+
=
q H A s A d A
A s A d s A
r A r
q
(9)

where symbols with tildes denote skew symmetric matrices comprised of their
vector elements that implement the vector product operation and denotes
the relative coordinate vector.
1)i (i
q
Combining Eqs. (7) and (9) yields the recursive virtual displacement equation
for a pair of contiguous bodies

1)i (i 1)i2 (i 1) (i 1)i1 (i i

+ = q B Z B Z
(10)

where

(
+
(
(
I 0
A s A d s I
A 0
0 A
B
)
~
~
~
( 1)i (i
T
1) i(i 1)i (i 1)i (i 1)i (i
1)i1 (i
1)i (i
1)i (i
T
T
(11)
(
(
+
(
(

1)i (i
1)i (i
1)i (i
T
1) i(i 1)i (i 1)i (i
1)i (i
1)i (i
1)i2 (i
~
) (
1)i (i
H
H A s A d
A 0
0 A
B
q
T
T
(12)

It is important to note that matrices and are functions of only
relative coordinates of the joint between bodies
1)i1 (i
B
1)i2 (i
B
(i 1) and . As a
consequence, further differentiation of the matrices and in Eqs.
(11) and (12) with respect to other than yields zero. The virtual
displacement relationship between the absolute and relative coordinates for the
whole system can be obtained by repetitive application of Eq. (10) as
i
(i 1)i1 (i
B
1)i
1)i2
B
(i
q

q B Z =
(13)

where is the velocity transformation matrix with relationship between
Cartesian and relative coordinates. The relationship between Cartesian velocity
B

10-7

Y q& and relative velocity can be derived in the same manner.
+
=
M
=

q B Y
&
=
(14)


The variational form of the Newton-Euler equations of motion for a
constrained multibody system is

0 Q Y M Z

=

) (
T T
&
(15)

where and Q are the mass matrix and general force vector in Cartesian
space, respectively.
M
Z must be kinematically admissible for all joints except
cut joints [12]. In the equation, and , respectively, denote the constraint
equations and the corresponding Lagrange multiplier in in which m is the
number of the constraint equations. Substituting the virtual displacement
relationship and acceleration relationships into Eq. (15) yields (see
Ref. [5])
m
R
q B q B &
&
& & + = Y
&
n * T *
R = + F 0 Q q M F
q
& &
(16)

where n is the number of generalized coordinates and the mass matrix
and force vector are defined as
*
M
*
Q
B M B
T
=
*
(17)
) (
*
q B M Q B Q
T
&
&
(18)

A recursive method has been proposed to compute Eqs. (17) and (18) in Ref.
[7].

RecurDyn

10.4. ELIMINATION OF LAGRANGE MULTIPLIERS AND LINEARIZATION
OF THE EQUATIONS OF MOTION

The relative coordinates can be partitioned into dependent coordinates
and independent coordinates q such that the sub-Jacobian is well-
conditioned. Variational form of the cut constraint equations can be written as
q
D
q
I
D
q

0 q q
q q
= + =
I D

I D
(19)

The can be obtained from Eq. (19) as
D
q

I
1
D

I D
q q
q q
= (20)

By using the relationship in Eq. (20), is represented as
I
1
D

I D
q q
q q
=

I
q N q = (21)
where
(
=

I

N
q q
I D
1
(22)

Direct calculation of shows that is the null space of as
T T
q
N N
q

| | 0
I N
q
q
q q q
=
(
(
=
T
T
T 1 - T T T
I
D
D I
) (
(23)

As a result, pre-multiplication of Eq. (16) by gives
T
N

0 Q N q M N F = =
* T * T *
& &
(24)

where Lagrange multiplier term was eliminated since is the null space of . N
q

10-9

However, the equations of motion are dependent on not only the dependent variables ,
and but also independent variables , and q& . Taking variation of Eq. (24)
yields
*
F
q
*
| |
&
*
q
q
q
+
+
=
q
q
q
& &
&
F q
q& q& & q q& &
D
D D I
q+
&
0
+
=
I
0
0
0
I
F
q& &
*
q
& &
(
(
(
(
(
0
0
0
I
0
T
q& q
q F q F
q
= + =
& &
&
* *
(25)

Equation (25) can be rewritten in a matrix form as

0
q
q
q
F F F
q q q
=
& &
&
& &
* *
(26)

Variations of position, velocity and acceleration level constraints are

0 q
q
0
q q q
q q
q
=
&
& & &
&
&
(27)

Appending the trivial identity relationships for the variations of independent
coordinates, velocities and accelerations to Eq. (19) yields

(
(
(
(
(
I
I
I
q
q
q
0
0
0
I

0
0 0
q q q
q q
q
& &
&
& & &
&
(28)

Equation (28) is solved for the { } q& & and substituted into the
linearized equations of motion in Eq. (26) to yield the following linearized
equations of motion only in terms of the variations of independent coordinates,
velocities and accelerations:

RecurDyn

(29)
| | 0
q
q
q
I
0 0 0
0 0 0
0 0 0
I

0
0 0
F F F
q q q
q q
q
q q q
=
(
(
(
(
(
(
(
(
(

I
I
I
1
* * *
& &
&
& & &
&
& & &

Direct comparison of Eq. (29) and the following linearized equations of
motion yields the M
)
, C
)
and K
)
matrices:

0 q K q C q M F
q
= + + =
I I I
*
*
& & &
(30)

10.5.1. FOURBAR MECHANISM WITH A SPRING
Figure 3 shows a four bar mechanism with a spring. The system consists of
four revolute joints and one spring and their material properties are defined in
Table 1. As a result, three generalized coordinates, , and are defined
for the first three revolute joints and the remaining one revolute joint is defined
as a cut joint. The constraint equations are introduced from the cut joint.
1
Link
Link
Link
500
400
Cut joint
Figure 3 A four-bar mechanism with a spring

10-11

Table 1 Material property of bodies and a spring
Mass (kg)
Inertia Moment
(kg*mm^2)
Link A 7.707 161760.83
Link B 3.946 53005.79 Body
Link C 7.707 161760.83
Stiffness (N/mm) Damping (N*sec/mm)
Spring
10.0 0.0

Dynamic analysis of the mechanism is performed to obtain the time domain
response. FFT of the time response is performed to extract dominant frequency
domain response. Figures 4 and 5 show the time and frequency responses,
respectively.
The proposed linearization method is applied for the system. The dominant
frequency and corresponding mode shape are shown in Figure 6 and Table 2.
The frequency obtained from the proposed method and that obtained from FFT
analysis of the time domain responses are shown to be very close, which
validates the proposed method.

Figure 4 Angle of link C in time domain

RecurDyn


Figure 5 Response in frequency domain
Figure 6 Mode shape of fourbar mechanism
Table 2 Undamped natural frequency and mode shape from the proposed method
Undamped Natural
Frequency (Hz)
Mode
1

2

3

5.040164E+00
5.773503E-01 -5.773503E-01 5.773503E-01

10-13

10.5.2. CANTILEVER BEAM DRIVEN BY A MOTION
The system characteristics of a rotating cantilever beam differ from those of
beam in a static state, because the stiffness of the beam is changed by a
centrifugal force due to the rotational motion. (see Ref. [13]). A cantilever beam
rotating with the angular velocity is shown in Figure 7.
Figure 7 A rotating cantilever beam

Length of the beam is 6.8 m, density of the material is 14705.88 kg/m
3
,
Young's modulus of the material is 7.010
8
N/m
2
. Area of the cross section is
0.002 m
2
, the moment of inertia 4.010
-7
m
4
. The beam is divided into 21
lumped mass and 20 beam elements. Figure 8 shows the lowest three natural
frequencies of the rotating beam. As the angular speed increases, the bending
natural frequencies are shown to be increased.

Figure 8 The relationship between angular velocity and natural frequencies

RecurDyn

10.5.3 A SPRING SYSTEM WITH 2 D.O.F.

A spring model shown in Fig. 9 is a system with two D.O.F, and the system
has two masses, joints and spring elements. Their material properties, spring and
damping coefficients are shown in Table 3.

Figure 9 A spring model

Table 3 Material properties, spring and damping coefficients
Mass1 5 Kg
Mass2 3 Kg
Length of m1 300 mm
Spring coefficient (k1) 10 N/mm
Spring coefficient (k2) 20 N/mm

If the rotational angle is small, sin and the equation of motion of this
system can be derived as:
0 =
(
(
(
(
+
+
(
y
k k
l
k
l
k
l
k l
y
m
I
2 2
2 2
2
1
2
2
2
2 4
0
0
& &
& &
(31)
From Table 3, Eq. (31) can be replaced as:

(32)
0
20000 3000
3000 1350
3 0
0 15 . 0
=
(
+
(
y y

& &
& &

10-15

The characteristic equation of this spring system is derived from Eq (32).

0
3 20000 3000
3000 15 . 0 1350
=
(33)

Also, the analytic natural frequencies can be computed as:

) Hz ( 76 . 17 f sec) / rad ( 6 . 111 12455
) Hz ( 019 . 9 f sec) / rad ( 66 . 56 3211
2 2
1 1
= = =
= = =
(34)

Finally, the eigenvalues of this spring system is validated shown in Table 4.

Table 4 Eigenvalues of spring model
Undamped Natural Frequency (Hz)
Mode
number
RecurDyn/Eigenvalue Analytic solution
1 9.01862E+00 9.019
2 1.77621E+01 17.76

10.5.4 A CANTILEVER BEAM

Two cantilever beam models shown in Figs. 10 and 11 have a fixed-free end
condition and ten lumped masses. One is modeled by using ten beam force
elements and the other is modeled by using one flexible body of RecurDyn. The
flexible beam model is originally generated in ANSYS. The material properties
and geometry conditions of the beam are shown in Table 4.

Figure 10 Beam model using RecurDyn/Beam element

RecurDyn


Figure 11 Beam model using RecurDyn/Flexible body element

Table 5 The material properties and geometry conditions of beam
Length 0.4 m
Mass 3.9888 Kg
Youngs
modulus
9
10 1 N/m
2

Inertia of area
8
10 215 . 1

m
4

Area 0.0018 m
2

In Ref. [14], the analytic natural frequencies of these beams are computed as:

4
2
1
875 . 1
AL
EI
=
,
4
2
2
694 . 4
AL
EI
=
,
4
2
3
855 . 7
AL
EI
=
(35)

By replacing Eq. (35) with Table 5, the natural frequencies can be computed as:

8601 . 3 2537 . 24 875 . 1
4
2
1
= = =
n
f
AL
EI

1929 . 24 0085 . 152 694 . 4
4
2
2
= = =
n
f
AL
EI

7477 . 67 6714 . 425 855 . 7
4
2
3
= = =
n
f
AL
EI

10-17

Finally, the eigenvalues of this beam model is validated shown in Table 6.

Table 6 Eigenvalues of cantilever beam model
Undamped Natural Frequency (Hz)
Mode
number
Beam element Flexible Body Analytic solution
1 3.84002E+00 3.84259E+00 3.8426
2 2.37455E+01 2.38154E+01 23.8154
3 6.55744E+01 6.60152E+01 66.0152
4 1.26483E+02 1.28016E+02 128.016
5 2.05481E+02
6 2.65264E+02

In addition, RecurDyn can show the mode shapes of the beam model through 3D
animation, as shown in Figs. 12 and 13.

(a) 1
st
mode shape (b) 2
nd
mode shape

(c) 3
rd
mode shape
Figure 12 The mode shapes of model using RecurDyn /Beam element

RecurDyn


(1) 1
st
mode shape (2) 2
nd
mode shape

(3) 3
rd
mode shape

Figure 13 The mode shapes of model using RecurDyn/Flexible body

10.6. CONCLUSIONS

In this paper, a linearization method for constrained multibody systems is
proposed for the non-linear equations of motion employing the relative
coordinates. Null space of the constraint Jacobian is pre-multiplied to the
equations of motion to eliminate the Lagrange multipliers and to reduce the
number of equations. The set of differential equations are perturbed in terms of
all relative positions, velocities and accelerations. The position, velocity and
acceleration level constraints are perturbed to express the variations of all
relative positions, velocities and accelerations in terms of the variations of
independent positions, velocities and accelerations, which are substituted into the
perturbed equations of motion. The equations of motion perturbed with respect
to the , and q& finally become the corresponding equations perturbed with
respect to the , and . Eigenvalues and eigenvectors are then computed
from the equations of motion perturbed with respect to the , and . The
proposed method is implemented in a commercial program RecurDyn.
Numerical results obtained from the proposed method are in good agreement
with the results reported in the literature and obtained by other methods.
q q& &
q&
I
q
I I
q& &
I
q
I
q&
I
q& &

10-19

REFERENCES

1. Sohoni VN, Whitesel J. Automatic Linearization of Constrained Dynamical
Models. ASME. Journal of Mechanism, Transmission, and Automation in Design,
Vol. 108, pp 300-304, 1986.
2. Neuman CP, Murray JJ. Linearization and Sensitivity Functions of Dynamic
Robot Models. IEEE Transactions on Systems, Man, and Cybernetics, Vol. SMC-
14, No.6, pp.805-818, 1984.
3. Balafoutis CA, Misra P, Patel RV. ecursive Evaluation of Linearized Dynamic
Robot Models. IEEE Journal of Robotics and Automation, Vol.RA-2, No.3,
pp.146-155, 1986.
4. Gontier C, Li Y. Lagrangian Formulation and Linearization of Multibody System
Equations. Computers & Structures, Vol.57. No.2, pp.317~331, 1995.
5. Bae DS, Haug EJ. A Recursive Formulation for Constrained Mechanical System
Dynamics: Part I, Open Loop Systems. Mech. Struct. and Machines, Vol. 15, No.
3, pp.359-382, 1987.
6. Bae DS, Haug EJ. A Recursive Formulation for Constrained Mechanical System
Dynamics: Part II, Closed Loop Systems. Mech. Struct. and Machines, Vol. 15,
No. 4, pp. 481-506, 1987.
7. Bae DS, Han JM, Yoo HH. A Generalized Recursive Formulation for
Constrained Mechanical System Dynamics. Mech. Struct. & Mach., Vol. 27(3),
pp. 293-315, 1999.
8. Bae DS, Lee JK, Cho HJ, Yae H. An Explicit Integration Method for Realtime
Simulation of Multibody Vehicle Models. Computer Methods in Applied
9. Bae DS, Han JM, Choi JH, Yang SM. A Generalized Recursive Formulation for
Constrained Flexible Multibody Dynamics. International Journal for Numerical

RecurDyn

Methods in Engineering, Vol. 50, pp. 1841-1859, 2001.
10. Ryu HS, Bae DS, Choi JH, Shabana AA. A Compliant Track Link Model for
High-speed, High-mobility Tracked Vehicles. International Journal for
Numerical Methods in Engineering, Vol. 48, pp. 1481-1502, 2000.
11. Kim HW, Bae DS, Choi KK. Configuration Design Sensitivity Analysis of
Dynamics for Constrained Mechanical Systems. Computer Methods in Applied
12. Wittenburg J. Dynamics of Systems of Rigid Bodies. B. G. Teubner Stuttgart,
1977.
13. Southwell R, Gough F. The Free Transverse Vibration of Airscrew Blades.
British A.R.C. Reports and Memoranda No. 766, 1921.
14. L. Meirovitch, Analytical Methods in Vibrations, MACMILLAN, 1967.

10-21

RecurDyn


11

NONLINEAR DYNAMIC MODELING OF
SILENT CHAIN DRIVE

11.1. INTRODUCTION

Chain drives are widely used in the power transmission applications in the
automotive field for a long time because they are capable of transmitting large
power at high efficiency and low maintenance cost. However, the noise and
vibrations created by chain drives have always been major problems, especially
for higher speed, lighter weight, and higher quality. Noise and vibrations in
chain systems are largely caused by chordal(polygonal) action and impacts
between chain and sprocket. The links of the chain form a set of chords when
wrapped around the circumference of the sprocket. As these links enter and leave
the sprocket, they impart a jerky motion to the driven shaft by chordal action.
The chordal action causes chain span longitudinal and transverse vibrations.
Whereas, impact between sprocket and link excites high frequency vibration and
is a major source of noise in chain drives at high speeds. In order to minimize
such problems, silent chains are introduced in many camshaft drives of
motorcycle/automobile engines and the primary drive between the engine and
transmission, as well as in other high-speed applications. It is also used with the
object of increasing chain life. However, in spite of the widespread use of silent
chain drives, surprisingly little works have been published about their dynamic
analysis. This may be due to three major difficulties; the first is the complexity
of the contact algorithms among components, the second is small integration step
size resulting from the impulsive contact forces and the use of stiff compliant
elements to represent the joints between the chain links, and the third is the large
number of the system equations of motion to solve.

RecurDyn

EFFICIENT CONTACT AND NONLINEAR DYNAMIC MODELING OF AUTOMOTIVE SILENT CHAIN DRIVE

Camshaft Sprocket
(Idler Sprocket)
Crankshaft
Sprocket
Pivot Guide
Tensioner
Fixed Guide
Figure 1. Silent Chain Drive Model of Automotive Engine

Chen and Freudenstein [1] presented a kinematic analysis of chain drive
mechanism with the aim of obtaining insight into the phenomena of chordal
action, with the associated impact and chain motion fluctuation. Veikos and
Freudenstein [1] developed a lumped mass dynamic model based on Lagranges
equations of motion and showed chain drive dynamics and vibrations. Wang [3,
4] investigated the stability of a chain drive mechanism under periodic sprocket
excitations and studied the effect of impact intensity in their axially moving
roller chains. Kim and Johnson [5, 6] developed a detailed model of the roller-
sprocket contact mechanics that allowed the first determination of actual
pressure angles and a multi-body dynamic simulation. This investigation is based
on Kanes dynamic equations. Choi and Johnson [7, 8] investigated the effects of
impact, polygonal action, and chain tensioners into the axially moving chain
system and showed the transverse vibration of chain spans. Quite recently Ryu et

11-3

al [9] developed very detailed chain models including contact forces for links,
sprockets and idlers with special application to large-scaled civilian and military
tracked vehicles. There has been some design analysis in the view of dynamic
behaviors of silent chain in powertrain industry and commercial software [10].
However they showed some primitive dynamic analysis and design of silent
chain system because it has high frequency contact forces, speedy revolution and
large number of bodies.
The purpose of this work is to investigate and suggest the dynamic modeling
and analysis of silent chain drive mechanism with high speed revolution using
multibody dynamic techniques. In this investigation, numerical skills of
multibody chain dynamic analysis are employed and showed very good
agreement of physical phenomenon of silent chain system. Dynamic tension,
impact forces, and vibration of chain links are explored for the sake of
understanding dynamic behaviors of the chain system.

RecurDyn


11.2. MULTIBODY MODELING OF SILENT CHAIN DRIVE

As shown in Fig. 1, in general a chain drive mechanism has four main
components, which are sprockets, chain links, guides and tensioner element. The
sprockets can be recognized as drive sprockets and idler sprockets. The chain
link can includes link plates, guide plates, and pins. The tensioner element
maintains stable tension during operation by adjusting pressure force to the chain
link system. While roller chain mechanism has engagements between pins and
sprocket, since silent chain mechanism engages between chain link teeth and
sprocket teeth, there is much less chodal vibrations and can transmit the power
more quietly. In this investigation the dynamic analysis and numerical modeling
techniques are presented by using multibody methods.

11.2.1 SPROCKET
The sprockets of the chain system are interacted by the introduced contact and
friction forces acting on between the chain and the sprocket teeth. The crank
sprocket of the system is driven by motion constraint. This motion constraint can
be constant or time dependent. In this investigation the sprocket is modeled as a
rigid body and attached on ground by revolute joint. The geometry of sprocket
teeth profiles consists of a series lines and arcs with different length and radii as
shown in Figure 2. The sprocket of silent chain is shaped more like a gear than
one of roller chain.

Figure 2. Geometry of Silent Chain Sprocket

11-5

11.2.2 SILENT CHAIN LINK

Roller chains, although having excellent wear and strength capability, are
inherently noisy and oscillatory. As a result, inverted-tooth chain mechanisms
were developed in order to reduce the forcing function of the noise-producing
mechanism. The difference in noise performance between silent and roller chains
can be attributed to the manner in which they engage and disengage the sprocket
teeth. After the sprocket tooth initially contacts the chain link, and as the
engagement proceeds, a combination of rolling and sliding motion occurs
between the tooth and link contacting surfaces. Such an engagement mechanism
effectively spreads the engagement time over a significant interval, thereby
minimizing tooth/link impact and its inherent noise generation.
A silent chain consists of several layers of links connected with pins. Since
there is no advantage for the modeling of pins and multi layer links as separate
components, in this investigation these multi layer links are treated as a rigid
body with mass and inertia property which takes into account the effects of the
pins. An individual silent chain link looks much different comparing to a roller
chain link. The geometry of link profile, which resembles a tooth, consists of
several lines and arcs in a complex arrangement as shown in Fig. 3. As used in
the roller chain from previous work, the connections between links are modeled
with bushings to account for the flexibility in this investigation. Though the
sprockets of the silent chain serve in the same function of the rolling chain
system, however, they are designed to engage specifically with the links of the
silent chain with different tooth contour as illustrated.

RecurDyn

Figure 3. Components of Silent Chain Link System

11.2.3 TENSIONER AND CHAIN GUIDE

In a chain drive system, the chain guide ensure that the chain remains on the
path, while tensioner try to keep constant tension of chain system. Usually the
chain guide directs the tight chain portion which runs from the driven sprocket to
the driving sprocket. Conversely, the chain arm directs the slack portion of the
chain which runs opposite (from the driving to the driven). The pivot guide also
serves to distribute the force on the chain from the hydraulic tensioner to
maintain certain level of chain tension. In this investigation hydraulic tensioning
force model is used which is offered from hydraulic tensioner manufacturer.
The chain guide and the chain arm are both modeled as separate rigid body
parts. The geometric profiles of the guides consist of a series arcs with different
radii. If desired, the chain guides can be modified so that they are constructed as
flexible bodies for the calculation of vibrations, stresses and bending moments,
etc.

11.2.4 EQUATIONS OF MOTION AND INTEGRATION

Since the chain system interacts with the frame component through the contact
forces and adjacent chain links are connected by compliant force elements, each
chain link in the chain system has six degrees of freedom which are represented
by three translational coordinates and three Euler angles. The equations of
motion of the frame structure such as sprockets that employs the velocity
transformation defined by Choi [9] are given as follows :
) (
r
i
r
i
q B M Q B q MB B
T T
&
&
& &
=
(1)
where and B are relative independent coordinates and velocity
transformation matrix of the engine chassis subsystem, and M is the mass matrix,
and Q is the generalized external and internal force vector of the frame structure
subsystem, respectively. Since there is no kinematic coupling between the frame
structure subsystem and chain subsystem, the equations of motion of the chain
subsystem can be written simply as
r
i
q

11-7

t t t
Q q M =
& &
(2)
where , and Q denote the mass matrix, the generalized coordinate
and force vectors for the chain subsystem, respectively. Consequently, the
accelerations of the frame structure components and the chain links can be
obtained by solving Eqs. (1) and (2).
t
q M
t t
Many different types of integration methods can be employed for solving the
equations of motion for mechanical systems. Explicit methods have small
stability region and are often suitable for smooth systems whose magnitude of
eigenvalues is relatively small. Contrast to the explicit methods, implicit
methods have large stability region and are suitable for stiff systems whose
magnitude of eigenvalues is large. In the model used in this investigation, a
contact between two bodies is modeled by compliance elements. Lumped
characteristics of the spring and damper must represent elastic and plastic
deformations, and hysterisis of a material. Such characteristics may include
artificial high frequencies which are not concern of a design engineer. Unless
such artificial high frequency is filtered, an integration stepsize must be reduced
so small that integration cant be completed in a practical design cycle of a
mechanical system. To achieve this goal, the implicit generalized-alpha method
[9, 11] has been employed to filter frequencies beyond a certain level and to
dissipate an undesirable excitation of a response in this investigation. One of the
nice advantages of the generalized-alpha method is that the filtering frequency
and dissipation amount can be freely controlled by varying a parameter in the
integration formula. As a result, the generalized-alpha method is the most
suitable integration method for integrating the equations of motion for stiff
mechanical systems.

RecurDyn


11.3. CONTACT FORCE ANALYSIS

The contact collision algorithms for a silent chain drive used in this
investigation are composed of three main routines such as search routines for,
sprocket teeth and chain link contact, chain guide and chain link contact, and
side guide of chain link and sprocket contact. The contact positions and
penetration values are defined from the kinematics of components in searching
routines. Thereafter a concentrated contact force is used at the contacted position
of the contact surface of the bodies. A detailed discussion on the formulation of
the contact collision is represented in this section, respectively. Efficient search
algorithms should be considered seriously because there are large number of
chain link bodies and sprocket which take long time to search all the bodies
whether they are in contact or not.

11.3.1 STRATEGE OF CONTACT SEARCH
For the efficient search of the sprocket-chain link contact kinematics, the
contact search algorithm is divided by pre-search and post-search. In the pre-
search, bounding circle relative to sprocket center is defined. All of chain links
are employed to detect a starting link and ending link which has a possibility of
sprocket contact. Then, chain links from starting link are investigated the
engagement with sprocket valley. Post-search means a detailed contact
inspection for chain links in a bounding circle. Once a starting and ending link is
found at one time through pre-search prior to analysis, only detailed search is
carried out by using the information of starting link and ending link from the
next time step. There are four contact possibilities such as, arc-line, arc-point,
arc-arc and line-point contact for interaction between the sprocket teeth and
chain link.

11-9

11.3.2 LINE-ARC CONTACT
X
Y
Z
i
X
i
Y
i
Z
i
t
X
i
t
Y
i
t
Z
i
R
i
t
u
j
p
X
j
Y
j
Z
j
R
j
p
u
p
t
ij
k
u
a
j
p
Y
j
X
Link arc
Tooth line
Chain link
coordinate system
Sprocket
coordinate system
Global coordinate system
Tooth
coordinate system

Figure 4. Line-Arc Contact Kinematics
The contact conditions between the sprocket teeth line segment and the chain
link arc segment can be determined. A coordinate system is attached
to each of the sprocket surfaces shown in Fig. 4. The surfaces of the tooth line
are approximated by plane surfaces and the axis of each surface coordinate
system is assumed to be parallel to the tooth surface. The surfaces of the chain
link arc segment are approximated by plane surfaces and the axis of each arc
origin coordinate system is assumed to be directed to the starting arc point from
arc origin. The orientation of the tooth surface coordinate system with
respect to the global system is defined by
i
t
i
t
i
t
Z Y X
j
p
X
i
t
X
k
i
k
i i
t
A A A =
(3)

RecurDyn

where is the transformation matrix that defines the orientation of the
coordinate system of the sprocket and is the transformation matrix that
defines the orientation of the tooth line surface coordinate system
with respect to the sprocket coordinate system. The orientation of the link arc
coordinate system with respect to the global system is defined by
i
A
i
i
k
A
k
i
t
i
t
i
t
Z Y X

j
l
j j
a
A A A = (4)
where is the transformation matrix that defines the orientation of the chain
arc surface coordinate system with respect to the chain link
coordinate system.
j
l
A
l
j
p
j
p
j
p
Z Y X
The global position vector of the coordinate system of the tooth surface is
defined as
k
i
t
i i i
t
u A R r + = (5)
where
i
R is the global position vector of the coordinate system of the
sprocket and i
i
t
u is the position vector of point with respect to the origin
of the sprocket coordinate system .
t
i i i
Z Y X
The global position vector of the center of the chain link arc segment, denoted
as point p , can be defined as
j
p
j j j
p
u A R r + = (6)
where
j
R is the global position vector of the origin of chain link j , is
the transformation matrix of chain link
j
A
j and
j
p
u is the position vector of
point p defined in the chain link coordinate system .
j j j
Z Y X
The position vector of the center of the arc of chain link j with respect to the
origin of the tooth line surface coordinate system can be defined in the global
coordinate system as
i
t
j
p
ij
k
r r u = (7)
The components of the vector along the axes of the tooth line surface
coordinate system are determined as
ij
k
u
[ ]
ij
k
T
i
t
T
ij
z
ij
y
ij
x
ij
u u u u A u = = (8)
Necessary but not sufficient conditions for the contact to occur between the
chain link arc and the sprocket tooth line surface are k
k
ij
x
l u 0
(9)

11-11

p t
ij
z p t
w w u w w +
(10)
r
ij
y
u
(11)
where is the length of the tooth line surface , is half width of the
tooth and is half width of the chain link outer plate and
k
l
w
k
t
w
p
r is the radius of
the chain link arc. If the above conditions are satisfied, it has to be checked if
contact point is existed in the arc range for the next step.

g d =
ji
k
, where (12) ] [ h g f A =
i
k
[ ]
ji
k
T
j
a
T
ji
z
ji
y
ji
x
ji
k
d d d d A d = = (13)
) , ( atan2
ji
x
ji
y k
d d = (14)
a k
0 (15)
where is the opposite signed normal vector of the tooth line surface ,
is the angle of with respect to the link arc segment coordinate system
and is the angle of arc segment.
ji
k
d k
k
ji
k
d
a
If the above conditions are satisfied, the penetration is evaluated as

ij
ij
y
ij
u r = (16)

RecurDyn


11.3.3. ARC-POINT CONTACT
X
Y
Z
i
X
i
Y
i
Z
i
t
X
i
t
Y
i
t
Z
i
R
i
t
u
j
Y
j
Z
j
R
j
p
u
p
t
ij
k
u
a
j
X
Link point
Tooth arc
Chain link
coordinate system
Global
coordinate system
Tooth
coordinate system
Sprocket
coordinate system

Figure 5. Arc-Point Contact Kinematics
There are two arc-point contact possibilities such as convex arc vs. point and
concave arc vs. point contact for arc-point interaction between the sprocket teeth
and chain link. Figure 5 shows a convex arc-point contact kinematics. The arc-
point contact conditions between the sprocket teeth and the chain link can be
determined. A coordinate system is located at the center point of the
sprocket arc surfaces.
i
t
i
t
i
t
Z Y X
The position vector of the point p of chain link j with respect to the center
point of the tooth surface coordinate system can be defined in the global
coordinate system such as in Eqs. (7) and (8)
chain link point and the sprocket tooth surface are k
r u u
ij
y
ij
x
+
2 2
) ( ) ( (17)
p t
ij
z p t
w w u w w + (18)
where r is the radius of the sprocket arc segment, is half width of the
t
w

11-13

tooth and is half width of the chain link outer plate.
p
w
If the above conditions are satisfied, it has to be checked if contact point is
existed in the arc range for the next step.
) , ( atan2
ij
x
ij
y k
u u = (19)
a k
0 (20)
where is the angle of with respect to the sprocket arc segment
coordinate system and is the angle of arc segment.
k
ij
k
u
a

ij
2 2
) ( ) (
ij
y
ij
x
ij
u u r + = (21)

11.3.4. ARC-ARC CONTACT

There are four arc-arc contact possibilities such as convex vs. convex, convex
vs. concave, concave vs. convex, concave vs. concave arc contact for arc-arc
interaction between the sprocket teeth and chain link. Since the radius and angle
of each arc are given at geometry, the contact kinematics between arcs can be
calculated by expanding arc-point contact logic. At the center of the arc a marker
is attached and X axis is fixed to the starting point of arc. The monitoring vector
between arc centers can be easily detected whether they are in contact boundary
or not using the arc angles with respect to the X axis of the marker. If the vector
is in contact boundary and the length between the centers of arcs is less than the
sum of the radii of arcs, they are considered in contact situation.

11.3.5 LINE-POINT CONTACT

The search kinematics of line-point contact is one of the most simple search
algorithms in contact analysis. An axis of marker can be attached on the line and
the vertical vector from the point to line can be evaluated whether the point is in
contact with line, respectively.

RecurDyn


11.3.6 CONTACT FORCE MODEL

In the field of multi-body dynamics, one of the most popular approximation of
the dynamic behavior of a contact pair has been that one body penetrates into the
other body with a velocity on a contact point, thereafter the compliant normal
and friction forces are generated between a contact pair. In this compliant
contact force model, a contact normal force can be defined as an equation of the
penetration, which yields
(22)
&
n m
n
c k f =
where and are an amount of penetration and its velocity, respectively.
The spring and damping coefficients of and c can be determined from
analytical and experimental methods. The order of the indentation can
compensate the spring force of restitution for non-linear characteristics, and the
order n can prevent a damping force from being excessively generated when the
relative indentation is very small. As it happens, the contact force may be
negative due to a large negative damping force, which is not realistic. This
unnatural situation can be resolved by using the indentation exponent greater
than one. A friction force can be determined as follows.
&
k
m
n f
f v f ) ( = (23)

11.4. NUMERICAL STUDY OF AN AUTOMOTIVE SILENT CHAIN SYSTEM

Four cylinder DOHC (double overhead cam) engine valve drive mechanism is
employed for the sake of numerical verification of proposed methods as shown
in Figure 1. A silent chain drive system has 1 crankshaft sprocket, 2 camshaft
sprockets, 1 fixed guide, 1 pivot guide, tensioner element, and 135 chain links.
The crank sprocket of the system is rotated by motion constraint. Resistance
torque is applied at each camshaft sprockets. Hydraulic tensioning force model is
used which is offered from manufacturer.
Figure 6 shows the computer simulation model of automotive silent chain

11-15

system in computer graphic environment. The system consists of 143 rigid
bodies, 270 bushing force elements to connect chain link bodies, 4 revolute joints,
2 resistance torque and a hydraulic force element of tensioner. It has 815 degrees
of freedom.
Figure 7 and 8 demonstrate the trajectory and velocity of the chain link during
the cycle around the system when the engine runs 4000 rpm, respectively. The
X-Y trajectory of the links agrees the defined path of the chain motion and the
magnitude of link velocity with respect to system inertia reference frame reflect
the linear velocity of 4000rpm as clearly shown in Fig. 8. Figure 9 shows the
contact force between a chain link and the sprockets or the chain guides and
figure 10 shows the dynamic chain tension measured between chain links during
simulation. Since the hydraulic auto tensioner is attached on guide arm, the
dynamic tension of the chain is controlled not to have excessive or be loosened.
Dynamic analysis of the silent chain system is performed for 200 milli-sec. It is
found that the CPU simulation times is 4039 sec on a Pentium 1.8 GHz platform
personal computer. Note that since the numerical results from the proposed
methods are almost showing the real physical behaviors and dynamic
characteristics of the chain mechanism, the proposed methods using multibody
dynamic techniques can be valid and suitable for the design of the silent chain
system, accordingly.

Figure 6. Simulation Model of Automotive Silent Chain System

RecurDyn


Figure 7. Trajectory of the Chain Link

Figure 8. Velocity of the Chain Link

11-17

Cam Sp.
Cam Sp. Crank Sp. Cam Sp.
Pivot
Guide
Pivot
Guide
Fixed
Guide
Figure 9. Contact Forces of the Chain Link at 4000 rpm

Figure 10. Dynamic Tension of the Chain Link at 4000 rpm

RecurDyn


11.5. FUTURE WORK AND CONCLUSIONS

It is clearly proved in this investigation that using the multibody dynamic
simulation methods the dynamic analysis of silent chain mechanisms can be
achieved clearly. While previous works showed rough estimations of the silent
chain system, the proposed methods in this paper show the possibility of the
replacement of real prototype at early design stage. The presented three
dimensional silent chain consists of the driving sprocket, idle sprockets, pivot
guide, fixed guide, tensioner, and chain links. Pre and post contact search
algorithms are employed in order to increase the simulation speed significantly.
For the sprocket teeth and link teeth, guide and link contacts, line-arc, arc-point,
arc-arc, and line-point kinematic interactions are presented in this investigation.
A compliant force model is used to connect the rigid body chain links. The silent
chain model has 143 bodies, 4 pin joints, tensioner element and 270 compliant
bushing elements and has 815 degrees of freedom. The numerical study of
automotive silent chain system shows that the tendency of the chain motion and
tensions are close as real system and it shows the characteristics of silent chain
comparing to roller chain with less oscillation.

11-19

REFERENCES
1. C. K. Chen and F. Freudenstein, ''Towards a More Exact Kinematics of Roller
Chain Drives, ASME Journal of Mechanisms, Transmission, and Automation in
Design, Vol.110, No.3, 123-130 (1988)
2. N. M. Veikos and F., Freudenstein, "On the Dynamic Analysis of Roller Chain
Drives: Part1 and 2", Mechanism Design and Synthesis, DE-vol 46, ASME, NY,
431-450 (1992)
3. K. W. Wang, ''On the Stability of Chain Drive Systems Under Periodic Sprocket
Oscillations'', ASME Journal of Vibration and Acoustics, Vol. 114, 119-126
(1992)
4. K. W. Wang, et al, ''On the Impact Intensity of Vibrating Axially Moving Roller
Chains'', ASME Journal of Vibration and Acoustics, Vol. 114, 397-403 (1992)
5. M. S. Kim and G. E. Johnson, Advancing Power Transmission into the 21
st

Centrury, DE-vol. 43-2, ASME, NY, 689-696 (1992)
6. M. S. Kim and G. E. Johnson, Advances in Design Automation, DE-vol. 65-1
(B. J. Gilmore et al., eds), ASME, NY, 257-268 (1993)
7. W. Choi and G. E. Johnson, Vibration of Mechanical Systems and the History of
Mechanical Design, DE-vol. 63 (R. Echenpodi et al., eds), ASME, NY, 29-40
(1993)
8. W. Choi and G. E. Johnson, Vibration of Mechanical Systems and the History of
Mechanical Design, DE-vol. 63 (R. Echenpodi et al., eds), ASME, NY, 19-28
(1993)
9. H. S. Ryu, D. S. Bae, J. H. Choi and A. Shabana, ''A Compliant Track Model For
High Speed, High Mobility Tracked Vehicle'', International Journal For
Numerical Methods in Engineering, Vol. 48, 1481-1502 (2000)
10. Phased Chain System Quietly Transmits Power, Automotive Engineering,

RecurDyn


Dec. (1995)
11. J. Chung, J. M. Lee, ''A New Family of Explicit Time Integration Methods for
Linear and Non-linear Structural Dynamics'', International Journal for
Numerical Methods in Engineering, Vol.37, 3961-3976 (1994)
12

DYNAMIC ANALYSIS AND CONTACT
MODELING FOR TWO DIMENSIONAL MEDIA
TRANSPORT SYSTEM

12.1. INTRODUCTION

Recently the media transport systems, such as printers, copiers, fax, ATMs,
cameras, film develop machines, etc., have been widely used and being
developed rapidly. Especially, in the development of those systems, the media
feeding mechanism for paper, film, money, cloth etc., is an important key
technology for the design and development of the media transport systems.
Tedious and iterative experimental trial and errors methods have been essential
way to determine kinematic mechanisms of parts dimensions, and materials, etc
for the media machine developers. Since the iterative trial & error methods are
truly inefficient, in order to shorten the time, reduce the cost, and improve the
machine performance, it has been absolutely required to develop the computer
simulation tool, which analyses the paper feeding and separation process.
Cho and Choi [1] developed a computational modeling techniques for two
dimensional film feeding mechanisms. The flexible film is divided by several
thin rigid bodies which are connected by revolute joints and rotational spring
dampers. The primitive computer implementation methods for contact search
algorithms are presented. Diehl [2, 5] presented the local static mechanics of
electrometric nip system for media transport system. The nonlinear finite
element method and experimental measurement techniques are used to
investigate the large deformable rollers. Several unique phenomena, such as
skewing sheet, etc., of nip feeding system are well described in this research.
Ashida [3] suggested the computer modeling techniques for the design and
analysis of film feeding mechanisms. The primitive dynamic analysis of two
dimensional film feeding models are presented by using commercial computer

RecurDyn

DYNAMIC ANALYSIS AND CONTACT MODELING FOR TWO DIMENSIONAL MEDIA TRANSPORT SYSTEM

program. The paper feed mechanism with friction pad system is investigated by
Yanabe [4] by using commercial nonlinear FEA program. It show the local
separation phenomenon between papers and roller, and proved very good
agreement with experimental measurements. Shin [6, 7] developed web
simulation and design tools using roll tensions. They show that the control of
tensions of each segment is the key design factors for web system.
In this investigation, a numerical modeling method and dynamic analysis of
the two dimensional flexible sheet for thin flexible media materials such as paper,
film, etc., and their roller and guide contacts are suggested by using multibody
dynamic techniques. Since the flexible sheet undergoes large deformation with
assumed linear material properties, the flexible sheet has been modeled as a
series of thin rigid bars connected by revolute joints with rotational spring
dampers force elements. It shows good visual appearance of the sheet under
severe bending conditions. An efficient contact search and force analysis
between sheet and rollers, and guides are developed and implemented
numerically. The sheet is fed by contact and friction forces when it contacts with
rotating rollers or guides. In order to detect a contact phase efficiently, the
bounding box method is used in this contact search algorithm. The method has
an advantage that the number of contact search can be smaller than conventional
methods for a system in which the position of rollers and guides are fixed on a
point of a base body. The proposed numerical models for media transport
systems will make it possible to confirm the potential problems of jamming by
given different sheet size, weight, stiffness, temperature, humidity extremes,
sheet velocity due to misalignment of drive-driven roller sets, and roller
velocities due to gap, wear or etc.

12-3

12.2. TWO DIMENSIONAL FLEXIBLE MULIBODY SHEET

In general, there are two methods to build a thin 2-D flexible sheet for
dynamic analysis. One is to employ beam element at discretized sheet body, and
the other is small rigid bar interconnected by revolute joint with rotational
spring-damper forces. In this investigation, the second method is used and
proposed the modeling techniques.

Figure 1 Modeling definition of a two dimensional flexible sheet

Several research works show that the most efficient way to model two-
dimensional approximation of the proper behavior of a sheet can be a series of
rigid bars connected by revolute joints and rotational spring-dampers as shown
in Figure 1 [1, 3]. The sheet is divided into a number of rigid bars with mass.
The mass and inertia moment of each rigid bar can be defined as follows

s s
t L m =
(1)
12
) (
2 2
s s
zz
L t
m I
+
=
(2)

where, is a sheet density per unit depth, is thickness, and is length
of each rigid bar. The leading body is connected to a ground by a planar joint to
guarantee an in-plane motion. The planar joint has one rotational and two
translational degrees of freedom. The body is connected to the body
by a revolute joint and rotational spring damper. The revolute joint has one
rotational degree of freedom between two rigid bars. The relative angle of
s
t
s
L
(i i ) 1 +
) 1 ( + i i

RecurDyn


is directly integrated. The torque of the rotational spring-damper is computed as
following

) 1 ( ) 1 ( + +
=
i i i i
c k
&
(3)
s
s
L
t E
k
12
3
=
(4)
k c = (5)

where,
) 1 ( + i i
and are relative angles and angular velocities of the

revolute joints, and
) 1 ( + i i
&
E and are the young s modulus and the structural
damping ratio of a sheet.
Figure 2 Contact geometry of two-dimensional sheet

The contact geometry of a sheet is described as a box and two circles as
shown Figure 2. The x-axis of the body reference frame of each rigid bar is
defined along longitudinal length direction and the y-axis is defined by right
hand rule. The mass center of each rigid bar is located at the center point of box.
In order to generate a continuous contact force, two circles are located on both
sides of the box. Even thought the proposed assumed method for flexible sheet
has an excellent visual appearance of the sheet under severe bending conditions,
this approach shows the lack of continuity between rigid bodies, which can cause
noise problems when the sheet is contact with rollers. It has also rigid leading
and trailing effect of the sheet. Problems can be overcome with introducing a
circular edge at leading and trailing points of each rigid bar.
There can be another approach to assume flexible sheet in dynamic analysis,

12-5

which employs a series of beam forces, and for the contact definitions, a rigid
bar can be attached simply. One of the advantages of this approach is a natural
definition of the flexible properties using the beam elements. However this
approach can cause problems with the contact definitions since it has possible
gaps and the lack of continuity between rigid contact bodies. The contact forces
on the edges of the rigid bodies are amplified as torques applied where the rigid
body is connected to the junction of two beams, and the rigid leading and trailing
edges of the sheet cause unnatural behaviors.

12.3. CONTACT FORCE ANALYSIS

In the field of multi-body dynamics, one of the most popular approximation of
the dynamic behavior of a contact pair has been that one body penetrates into the
other body with a velocity on a contact point, thereafter the compliant normal
and friction forces are generated between a contact pair. Figure 3 shows the
schematic diagram of contact force analysis used in this investigation.
Figure 3 Contact forces between a contact pair

In this compliant contact force model, a contact normal force can be defined
as an equation of the penetration [9], which yields

RecurDyn


(6)
&
n m
n
c k f =

where and are an amount of penetration and its velocity, respectively.
The spring and damping coefficients of k and can be determined from
analytical and experimental methods. The order of the indentation can
compensate the spring force of restitution for non-linear characteristics, and the
order can prevent a damping force from being excessively generated when
the relative indentation is very small. As it happens, the contact force may be
negative due to a large negative damping force, which is not realistic. This
unnatural situation can be resolved by using the indentation exponent greater
than one. The phenomenon is very important for the case of sheet contact
interaction since it is very thin and light. A friction force can be determined as
follows.
&
c
m
n

n f
f v f ) ( =
(7)

where, and
n
f ) (v
are a contact normal force and a friction coefficient,
respectively.

12.3.1. KINEMATICS NOTATIONS

The coordinate system is the inertial reference frame and the single
primed coordinate systems are the body reference frames, and the double primed
coordinate system is the contact reference frame in order to define contact
conditions as shown in Figure 4. The orientation and position of the body
reference frame are denoted by and , respectively.
Y X
A r

12-7

Figure 4 Kinematic notations of a contact pair

12.3.2. SHEET AND ROLLER INTERACTIONS

In this investigation, two kinds of rollers are defined for the system. One is a
fixed roller with one rotational degree of freedom. The fixed roller is linked to
the ground with a revolute joint. The other is a movable roller, which has two
degrees of freedom for a translational and a rotational motion. The movable
roller is linked to rotational axis retainer (RAR) with a revolute joint and the
retainer is linked to the ground with a translational joint. The contact geometry
of rollers is described as a circle as shown in Figure 5
Figure 5 Definition of rollers

RecurDyn


Two different interactions between roller and sheet are introduced in this
investigation. Since the proposed flexible sheet is constructed by linear part and
circular part, these are interactions between linear part and rollers, and circular
part and rollers, as clearly illustrated in Figure 6

Figure 6 Sheet and roller interaction

In the case of linear part contact with rollers, the contacted penetration is
determined as follows:
r
R =
y sr,
d
, (8)
) r (r A d
s r
T
s sr
=

where, is the orientation matrix of a rigid bar, and is the radius of a
contacted roller, respectively. The location of contact between rigid bar and
roller can be defined as follows:
s
A
r
R

=
0
2 / ) (
s y s
t sign
sr,
x sr,
c
d
d
s
, (9)
and
) d s ( A A s
sr sc s
T
r rc
= (10)

where, and t are the orientation matrix of a roller and the thickness of
the sheet. The relative velocity at the contact point can be determined as
r
A
s

12-9

( )
( )
sc r s rc r r
r s rc r r
s w A r s w A
s A r s A r
+
|
.
|
+
s
dt
~ ~
&
r
T
n
T
n
r u
u
=
\
|
=
d
&
&
(11)
c
T
n
d u
&
=
(12)

and tangential relative velocity is

c
T
d u
&
t t
v =
(13)

where, and
r
w
s
w are the angular velocities of a roller and a rigid bar with
respect to each body reference frame, and u and are the normal and
tangent vectors of relative position between rigid bar and roller, respectively.
n t
u

12.3.3. ROLLERS INTERACTIONS

A circle to circle contact is used to describe the interactions between circular
rollers. In this circle to circle contact, the positive normal direction is same in the
direction of the relative position vector between two roller center points. The
tangent direction vector is determined by the right hand rule. The relative
velocity and the contact forces at the contact point can be computed similarly as
the sheet and roller interactions.

12.3.4. SEEET AND GUIDE INTERACTIONS

Guide has three types. Commonly used sheet guides for media transport
machines can be divided into three different types, which are an arc guide with
radius and angle, a linear guide with two points, and a circle guide similar to a
roller. In order to avoid the complex contact detect algorithms. It is assumed that
the arc and line guide are interacted with the circular part of rigid bars of the
sheet. However, in the case of circle guide, both linear and circular parts of the

RecurDyn


sheet are interacted with.

Figure 7 Sheet and arc guide interactions

As shown in Figure 7, the relative displacement between a circular edge of
rigid bar and arc guide can be determined as

g g g s s s gs
s A r s A r d + =
(14)

where, and are the center position and the orientation of the guide,
and the vectors of
g
r
g
A
g
s and s
s
are positions of the arc reference frame and the
circular edge center position with respect to each body reference frame,
respectively. If the vector is projected into the arc reference frame, the
resultant vector can be represented as follows
gs
d

gs gs
d C A d
g g
T
) ( = (15)

where, is the orientation matrix of the arc reference frame. The relative
angle between x-axis of the arc reference frame and the resultant vector of Eq.
15 is within an arc angle, which can be written as
g
C

g

) ( cos 0
1
gs
g
T
gs
d
f d
(16)

12-11

where, is the arc angle and is a constant unit vector of . If
the condition of Eq. 16 is satisfied, the penetration between circular part of sheet
and arc can be defined as follows
g
f 1 | |
T
0 0
g

g s gs
R t + = d
(17)

where, is a radius of the arc guide. The contact positions can be computed
as follows.
g
R

n g g
R u s
c
= (18)
c g g
T
c
c c
s C A A s
d s s
s
g
s s
gs g s
g
=
=
(19)

where, is the normal direction vector and determined
n
u

gs
gs
d
d
u
n

=
(20)

The tangent direction vector is determined by the right hand rule, and the
relative velocity at the contact point is defined as follows.

) (
~
) (
~
) ( ) ( (
gc g g g g s s s s
gc g g g s s s
dt
d
s C s w A r s s w A r
s C s A r s s A r d
g sc
g sc c
+ + + =
+ + + =
& &
&
(21)

where, and
g
w
s
w is the angular velocities of guide and a bar with respect
to each body reference frame, respectively. The contact forces at the contact
point can be computed similarly as described in the sheet and roller interactions.

RecurDyn


Figure 8. Sheet and line guide interactions

The sheet and line guide interactions are clearly illustrated in Figure 8. If the
component of the vector
d
defined in the double primed line guide
reference frame is the range of guide length, simple circle and line contact
algorithm is used in this investigation. After definitions of penetration and its
derivative, the contact force is created to restitute each body as similar as
previous interactions between sheet and guides.
x
gs

12. 4. EQUATIONS OF MOTION

Figure 9 Kinematic relationships between rigid bars of the sheet

12-13

Since the multibody sheet system interacts with the roller and guide
components through the contact forces and adjacent rigid bars are connected by
revolute joint and rotational spring damper forces as shown in Figure 9, each
sub-rigid bar in the sheet system has one degree of freedom which is represented
by one rotational coordinates and the leading body has three free coordinates.
The equations of motion of the sheet system that employs the velocity
transformation defined by Bae [8] are given as follows:

) (
r
i i
q B M Q B q MB B
T T
&
&
& & = (22)

where , and are relative independent coordinates, velocity
transformation matrix, and Cartesian velocities of the media feeding system, and
is the mass matrix, and is the generalized external and internal force
vector of the media feeding system, respectively. The velocity transformation
matrix of the sheet is more explicitly as
r
i
q B q&
M Q
B

(
(
(
(
(
(
=
1)n2 (n 232 1)n1 (n 122 1)n1 (n 012 1)n1 (n
232 122 231 012 121 231
122 012 121
012
B B B B B B B
0 B B B B B B
0 0 B B B
0 0 0 B
B
L L L
M M M M M
L
L
L

where the recursive velocity and virtual relationship for a pair of rigid bars are
obtained [8] as

1)i (i 1)i2 (i 1) (i 1)i1 (i i
+ = q B Y B Y & (23)

and denotes the relative coordinate vector. It is important to note that
matrices and are only functions of the .
1)i (i
q
(i
B
1)i1 1)i2 (i
B
1)i (i
q

RecurDyn


12. 5. NUMERICAL RESULTS

The proposed algorithm is implemented and a film-feeding problem is solved
to demonstrate the efficiency and validity of the proposed method.
Figure 10 Film feeding machine

The system has 29 degrees of freedom and consists of four fixed rollers and
three movable rollers, five line guides, one arc and circle guide and one sheet of
film shown in Figure 10. The sheet is modeled by using 20 rigid bars. The
density and Young s modulus of sheet are 2.2e-6( kg ) and 2250( ),
respectively. And the thickness and length of sheet are 0.5( mm) and 200( ),
respectively.
3
/ mm
2
/ mm N
mm
Figure 11 Slip between rollers and sheet

12-15

The film goes through a path while contacting the roller pairs. The
circumferential speed of each driving roller is 300( ). The slip velocities
between driving rollers and the sheet are shown in Figure 11. The path of first,
second and third segment bodies of the thin film are plotted as shown in Figure
12. The x and y axes of the plot are displacements measured in the directions of
x and y axes in the global reference frame, respectively.
sec / mm

Figure 12 Path of segmented bodies of film

The analysis was performed on an IBM compatible computer (PIII-933Mhz)
and took about 60 sec. per 1 sec. for simulation.

RecurDyn


12. 6. CONCLUSIONS

The dynamics and modeling techniques of two-dimensional media transport
system is investigated in this paper. The flexible sheet is divided by finite
number of rigid bars. Linear motions are constrained in order to allow rotations
between the rigid bars of the sheet. Rotational spring damper force is applied for
the reflection flexible stiffness of the sheet. From previous empirical
measurements in manufacturing process effective stiffness and damping
coefficients are substituted in this investigation. Compliant contact force model
is used for the interactions between sheet rollers, and guides. Kinematics
notations of the contact search algorithms for the media transport system are
clearly represented. A simple film feeding example is represented in this
investigation and manufacture [3] confirms that simulation results have very
good agreement with experimental measurements. The media transport system
manufactures have rely on trial error techniques for the design of their core
mechanisms, however the proposed method by employing multibody dynamics
in this paper can reduce many difficulties at the early design stage.

12-17

REFERENCES
1. H. J. Cho, and J. H. Choi, 2001, 2DMTT development specification Technical
report, FunctionBay Inc.
2. Ted Diehl, 1995, Two dimensional and three dimensional analysis of nonlinear
nip mechanics with hyper elastic material formulation Ph. D. Thesis,
University of Rochester, Rochester, New York
3. Tsuyoshi Ashida, 2000, The meeting material of The Japan Society for
Precision Engineering Japan
4. http://www.yanabelab.nagaokaut.ac.jp
5. http://www.me.psu.edu/research/bension.html
6. http://www.engext.okstate.edu/info/WWW-WHRC.htm
7. Shin, K. H., 1991, Distributed Control of Tension in Multi-Span Web Transport
Systems , Ph. D. Thesis Oklahoma State Univ.
8. D. S. Bae, J. M. Han, and H. H. Yoo, 1999, A Generalized Recursive
Formulation for Constrained Mechanical System Dynamics, Mech. Struct. And
Machines, Vol. 27, No 3, pp 293-315
9. Lankarani H. M. and Nikravesh P. E., 1994, Continuous Contact Force Models
for Impact Analysis in Multibody Systems, Journal of Nonlinear Dynamics,
Kluwer Academic Publishers, Vol. 5, pp 193-207

RecurDyn


13

HYDRAULIC AUTO TENSIONER (HAT) FOR
BELT DRIVE SYSTEM

13.1. INTRODUCTION

The hydraulic auto tensioner is a device that automatically adjusts the tension
for engine belt drive system. By reducing the noise due to play that occurs if the
tension on the belt drive system is insufficient and by holding the tension
constant, an auto tensioner extends the product life of the belt drive system and
is an indispensable part for improving engine reliability [1]. It is important to
analyze and to predict the dynamic behavior and the characteristics of the
hydraulic auto tensioner for design of the system. At this, numerical simulation
models can provide significant advantages in early design stage referred in [2]
and [3]. A simple simulation technique of HAT is applied for the initial design of
belts and chains using commercial multibody software [7]. Figure 1 shows the
hydraulic auto tensioner system. The plunger is connected to the belt drive
system. The spring force and the hydraulic force of the pressure chamber create
the damping force and are balanced with the load that is from belt drive system.
The check ball has the function of the check valve for control the oil flow
through orifice between the plunger and the cylinder.

RecurDyn

HYDRAULIC AUTO TENSIONER (HAT) FOR BELT DRIVE SYSTEM

Plunger
Cylinder
Check Ball
Pressure Chamber
Plunger
Spring
Check Ball
Spring

Figure 1. Hydraulic auto tensioner system

Figure 2 shows the schematic diagram of the operating principle of HAT. As
the tension of the belt drive system decreases, and the pressure of chamber
decreases, the check ball moves down and the check valve opens. Afterward due
to the plunger moves up by the spring force and the plunger pushes the belt, and
then the tension of belt system is increased. As the tension of the belt drive
system increases, the plunger moves down by the load and the plunger pushes
the pressure chamber, and it leads that the pressure of chamber increases, finally
the check ball moves up and the check valve closes. As a result, the oil flows
through the leakdown and the plunger moves down slowly.

Tension Decreasing Tension Increasing
Oil
Flow

Figure 2. Operating principle

13-3

Since the tension of the belt drive system is oscillated over 200~300Hz, the
hydraulic auto tensioner must be a reciprocating hydraulic device that can
respond to frequencies up to 300Hz.
The multibody simulation model of the hydraulic auto tensioner is presented
in the following sections. The differential equations are used to describe the
function and damping characteristics of the hydraulic auto tensioner, and the
circle to curve contact model is used for the movement of the check ball. In this
investigation, the developed HAT model is tested numerically for multibody belt
drive system.

13.2. MULTIBODY SIMULATION MODEL

The hydraulic auto tensioner consists of cylinder, plunger and check ball. The
spring force and the damping force of the plunger relative to the cylinder balance
these bodies. The spring force is built up by the spring preload and the spring
rate multiplied by the spring stiffness. The damping force is a friction force and a
hydraulic force that is proportional to the relative velocity of plunger and
cylinder [3].
The schematic diagram of analysis model is shown in Figure 3. When the
plunger is loaded from belt drive system, the spring force and the hydraulic force
react against the motion of the plunger. The hydraulic force from the check ball
is ignored in this investigation since it is relatively small amount. The motion of
the plunger is assumed to have the parallel direction to the motion of the cylinder.
The check ball has the spring force and the hydraulic force from the plunger. The
motion of the check ball is also assumed to have the parallel direction to the
motion of the plunger. The check ball is contacted between plunger and retainer.

RecurDyn


T
r
a
n
s
l
a
t
i
o
n

J
o
i
n
t

S
p
r
i
n
g

F
o
r
c
e

H
y
d
r
a
u
l
i
c

F
o
r
c
e

T
r
a
n
s
l
a
t
i
o
n

J
o
i
n
t

Check
ball
S
p
r
i
n
g

F
o
r
c
e

H
y
d
r
a
u
l
i
c

F
o
r
c
e

Plunger
C
o
n
t
a
c
t

Cylinder
Belt Drive System

Joint

Figure 3. Schematic diagram of analysis model

11.3. THE EQUATION OF MOTION

When the external load is forced to the plunger, the equation of the plunger
motion is following [3].

load ric c p
c p p hydraulic p p p
F f x x
x x K F x m

=
) sgn(
) (
.
& &
& &
(1)

where, , , and are the displacement of the plunger, and its first and
p
x
p
x&
p
x& &

13-5

second time derivatives, and , , , and are the mass of plunger, the
displacement of the cylinder, its first time derivative, and the stiffness coefficient
of the plunger spring, and , , and are the hydraulic force, the
friction force and the load form belt drive system, respectively.
p c
x
hydraulic
B
K
F
m K
c
ric
B

x&
f
B
hydraulic
x
p
load
F
BO
B
F
x
p
F
.
B B
x = & &
)
(
(
.
&
oil
&
c
x& )
p

p
x & ( V +
&
S =
In the case of check ball, since its motion is forced by the hydraulic force, the
spring force, and contact force, the equation of motion of the check ball can be
written as [3]

contact p
c B
F x
x m
+
) &
(2)

13.4. HYDRAULIC FORCES

The hydraulic forces that interact with the check ball and the plunger are
obtained from the pressure of the pressure chamber. The pressure is caused by
the volume variation of the pressure chamber and the oil flow rate. The volume
variation of the pressure chamber can be described by relative velocity between
plunger and cylinder. The rate change of the chamber volume is given by the
following equation [5].

p B air chamber
Q Q V V + + =
&
(3)

where is the compressed volume rate of pure oil and is the compressed
volume rate of air component in the oil. Q
oil
V
&
air
V
&
p
is leak oil flow rate out of the high
compression chamber at high pressure phase, Q
B
is the oil flow rate through
check valve, and S
p
is the effective area of hydraulic force, respectively.

RecurDyn


13.4.1. OIL FLOW RATES THROUGH THE CHECK VALVE

Accordingly to the check ball moves between the plunger and the retainer, the
check valve opens or closes and the oil flows. When the check valve opens, the
oil flow through the check valve is shown in Figure 4.

Po
Pi
d
r

QB

Figure 4. Oil flow rate through check valve

As the resistance of the oil through the check valve depends on the orifice area,
in this investigation, the dynamic resistance is considered for the turbulent flow
of the oil flow through the check valve [4], which yields,

i o i o d B
P P
g
P P A C Q =
2
) sgn(
(4)

where, is discharge coefficient of check valve, A is the orifice area, g is
gravity acceleration, and is weight density of oil, respectively. Q
d
C
B
represents
the oil stream flowing rate through the opened check valve into or out of the
pressure chamber. The area of orifice is obtained such as

cos sin 2 d r A =
(5)

13-7

13.4.2. OIL FLOW RATES THROUGH THE LEAK
BETWEEN PLUNGER AND CYLINDER

As the pressure of chamber is different comparing to the air pressure, the oil
flow through the gap between plunger and cylinder is shown in Figure 5. The oil
flowing between the plunger and the cylinder is laminar flow. The oil speed is
faster than the plunger speed. As shown in Figure 5, variation of the oil speed is
fully depended on the pressure difference between the pressure chamber and
reserver, and it is not affected by the plunger speed. The oil flow rate between
the gap of plunger and cylinder, Q
p
, can be written as [4]

) (
12
2
3
i o
P
p
P P
l
h r
Q =
(6)

where is the coefficient of viscosity of oil.

P
o
P
i
Q
p
l

r
p
h

Figure 5. Oil flow rate through leak

As shown in the Figure 4 and 5, we can consider about the relationship
between the plunger speed and the flow rate. It is assumed that inflow does not
induce any outflow from the pressure chamber by considering compressibility,

RecurDyn


and expansion and compression processes are isentropic. It is also assumed that
there is no cavitation caused by negative pressure. The volume of air in chamber
is obtained as.

0 .
1
air
i
o
air
V
P
P
V

= (7)

where is the ratio of specific heat and V
air.0
is the initial air volume. Since air
can be compressed, the volume rate is achieved by using the following equation,
such as

i
i
air
air
P
P
V
V
& &
= (8)

, and the volume of oil in chamber can be approximated by

p c p oil
S x x V ) (
(9)

In the case of oil, since it can also be compressed with high pressure, the
volume rate of oil is written as.

i
oil
oil
P
K
V
V
& &

=
(10)

where K is the bulk modulus.

The equations (4), (6), (8) and (10) are substituted into the equation (3), and
the differential equation for the pressure of the camber can be obtained,
accordingly;

13-9

+ +
=
K
V
P
V
P P
g
P P A C P P
l
h r
x x S
oil
i
air
i o i o i o
P
C P P
i
2
) sgn( ) (
12
2
) (
3
& &
p&
)
(11)

The hydraulic force to the plunger and to the check ball yield as

) (
. i o p hydraulic p
P P S F =
(12)

) (
. i o B hydraulic B
P P S F =
(13)

where S
B
can be obtained from Figure 4 as following.

(
2
cos = r S
B
(14)

RecurDyn


13.5. CONTACK OF THE CHECK BALL

The contact analysis of the check ball employs the circle to curve contact
method [6] in this investigation. This method is very efficient algorithm in
contact detection and force generation of the check ball contact.

Figure 6. Concept of circle to curve contack

The candidate lines on the plunger body have been selected for the contact of
the check ball. For the candidate lines, it is necessary to compute the amount of
penetration to generate the contact forces, as shown in Figure 6.

The relative position of a check ball with respect to the contact reference
frame is obtained as follows.
pn
d

1
p cn pn
s d d =
(15)

where the vector
pn
d
is projected into the contact reference frame as

13-11

pn
d
T
p pn
C d =
(16)

where C
p
is the orientation matrix of the contact reference frame. The
penetration of the node into the patch is calculated by

pn
T
p
- d n = r
(17)

where is always positive. The
p
n is a normal vector of a line and a constant
vector with respect to the contact reference frame. Thus, the contact normal force
is obtained by

3
2
1 m
m
m
contact
c k F
&
&
&
+ =
(18)

where k and c are the spring and damping coefficients which are determined by
assumed numerical experiences, or experimental methods, respectively and the
is time differentiation of . The exponents and generates a non-
linear contact force and the exponent yields an indentation damping effect.
When the penetration is very small, the contact force may be negative due to a
large negative damping force, which is not realistic. This situation can be
avoided by using the indentation damping exponent greater than one.
&

1
m
2
m
3
m

RecurDyn


13.6. BELT DRIVE SYSTEM

An automotive belt drive system is used for the simulation of HAT in order to
test numerically. This system is consisted of 5 pulleys and a belt system. A
continuous belt system can be modeled using series of a single body that has six
degrees of freedom and has a matrix (6x6) force element to connect the belt
bodies. Contact forces between the belt and pulleys are defined clearly.

Disturbance roller
Drive
Pulley
HAT
Sensing belt tension
Belt

Figure 7. Belt drive system

As shown in Figure 7, there are a drive pulley, a disturbance roller, four idle
pulleys, and an idle roller equipped with HAT.

13-13

13.7. NUMBERICAL RESULTS
The hydraulic auto tensioner must be a reciprocating hydraulic device that can
respond to frequencies up to 300Hz. When the reciprocating load is applied to
plunger with 300Hz, Figure 8 shows the result of the pressure in chamber and
Figure 9 shows the result the displacement of the check ball. The numerical
results show that the proposed modeling of HAT is acting to the reciprocating
load with 300Hz. As the load increases, the check valve closes and the oil flows
out only through the leak. As the load decrease, the check valve opens and the
oil flows in through the check valve.

Figure 8. Pressure in chamber [300Hz]

RecurDyn


Figure 9. Displacement of check ball [300Hz]

The proposed modeling method of hydraulic auto tensioner is applied for the
belt drive system as shown in Figure 7. The drive pulley rotates with 100 rpm.
As the disturbance roller increases the belt length, the belt tension around HAT
increases such as shown in Figure 10. Due to the belt tension increases, the
pressure in chamber arises as shown in Figure 11 and the oil flows out through
the leak as shown in Figure 12. As a result, the plunger is pushed back and the
belt tension decreases. Figure 10 shows less increase of tension of the belt with
HAT comparing to without it.

13-15

Figure 10. Tension

Figure 11. Chamber pressure

RecurDyn


Figure 12. Oil flow rate through leak

As the disturbance roller decreases the belt length, the tension around HAT
decreases as shown in Figure 13. Due to the belt tension decreases, the pressure
in chamber decreases as illustrated in Figure 14. Figure 15 shows the oil flow
rate through the check valve. As a result, the plunger is pushed to the direction
for increasing the tension by the plunger spring, and therefore the tension
increases. The tension drop can be quickly recovered with proposed HAT
element as shown in Figure 13.

13-17

Figure 13. Tension

Figure 14. Chamber Pressure

RecurDyn


Figure 15. Oil flow rate through check valve

13-19

13. 8. CONCLUSIONS
In this investigation, in order to design automotive power transmitting system
at early design stage, modeling and simulation methods of HAT, which is
necessary component for the tension adjusting system, are presented. The
multibody simulation model is proposed using three rigid bodies, which are
plunger, check ball and cylinder. The plunger and the cylinder bodies can be
connected by constraints and mechanical force elements. The plunger and the
cylinder are interacted by hydraulic force and spring force. The forces between
plunger and check ball are modeled by contact, hydraulic, and spring forces. The
circle to curve contact analysis is employed for the plunger and the check ball
contact efficiently. The differential equations of motion of the components and
the hydraulic force equations are developed in this investigation. It can be
assured that the proposed HAT model is able to respond to frequencies up to
300Hz. The proposed methods of HAT are simulated in different ways,
component level simulation with reciprocating forces, and with automotive belt
system. Both numerical results show reasonable responses as expected. Though
it is necessary to be correlated by experimental results. Therefore the proposed
numerical method of HAT shows the possibility of simulation for automotive
power transmitting system, which has been challenging works for long period.

RecurDyn


REFERENCES

1. http://www.ntn.co.jp/english/corp/news/news/20011001_2.html
2. NTN TECHNICAL REVIEW No. 61
3. NTN TECHNICAL REVIEW No. 67
4. Frank M. White, "Fluid Mechanics", 5th edition, McGraw-Hill International
Editions, 1999.
5. E. Sonntag , Richard, Claus Borgna, kke, and Gordon J. Van Wylen,
"Fundamentals of Thermodynamics", 5th Edition, John Wiley & Sons, Inc., 1998.
6. B. O. Roh, H. S. Anm, D. S. Bae, H. J. Cho, H. K. Sung, "A Relative Contact
Formulation for Multibody System Dynamics", KSME International Journal, Vol.
14, No. 12, pp. 1328-1336, 2000.
7. www.fev.com

14

DYNAMIC ANALYSIS OF CONTACTING SPUR
GEAR PAIR FOR FAST SYSTEM SIMULATION

14.1. INTRODUCTION

All Geared systems are commonly used in many mechanical power
transmitting systems, such as robot manipulator, automotive transmissions, etc.,
so as to transmit motion and power from one shaft to another. One of important
factors in the gear design is the dynamic transmission error, which gear vibration,
noise and other performance can be predicted by. When two mating gear is
operated, the dynamic transmission error is generated by gear dynamic forces.
These forces are caused by contact between meshing teeth. In other words,
contact mechanics between meshing teeth, considering backlash and tooth
geometric profile, is very important in the dynamic analysis of geared systems. A
lot of numerical and experimental works have been published about their
dynamic analysis. One of main topics in these studies is the conventional finite
element analysis. Traditional finite element methods are effective for calculating
quantities such as mesh stiffness, tooth deformations, and stress distributions
under static conditions. But it requires refined meshes to represent the tooth
contact and precise tooth surface shape for gear mechanics. Also, it takes
amazingly long time to analyze the dynamics effects of contacting gears.
Moreover, it is not suitable for analysis of entire system with the sets of gear
pairs as well as other components [1, 2]. Another topic is that concerning the
single degree of freedom(sdof) models of a pair of gears. It is because sdof
model can give relatively accurate results and computational efficiency despite
its simplicity. The sdof model approach in terms of entire system dynamic
analysis with gear pairs is desirable from research and design perspectives. In
sdof model, primitive approach is to model gear pairs with simple constraint or
force element using speed ratio, pressure angle and rotational angles. Gear

RecurDyn

DYNAMIC ANALYSIS OF CONTACTING SPUR GEAR PAIR FOR FAST SYSTEM SIMULATION

systems can be analyzed with fast computational time, but detailed inputs such
as tooth profiles and distance between gears are not considered directly because
it is not real gear teeth contact. More advanced approach is considered by contact
between teeth profile of gears. It enables designer not only to obtain gear contact
position and force exactly but also to simulate with entire system in various
operating conditions [3, 4].
A review of the mathematical models used in gear dynamics was given by
Ozguven and Houser [5], and T. Shing et al. presented an improved model for
the dynamics of spur gear systems with backlash consideration [6]. The torsional
vibration behavior was investigated experimentally by Kahraman and
Blankenship [7, 8, 9]. In the recent studies, a sdof model was proposed, which
considers a time-varying stiffness and backlash of the meshing tooth pairs with
similar formulations. However, most gear models in these numerical
investigations have been used the kinematic relations between the rotational
angles of each gear. It is not real contact model between bodies and needs some
limitation that gear shafts have no translational displacement.
The main purpose of this paper is to develop efficient contact algorithm
between meshing teeth in geared system for better understanding of the dynamic
behavior of entire system. Externally specified dynamic forces, or assumptions
about modeling the mesh forces by time-varying stiffness and static transmission
error are not required since dynamic mesh forces are obtained by contact
analysis at each time step. A simple spur gear pair modeled by using proposed
methods is compared and verified with the measurement results represented by
reference [7]. The dynamic modeling techniques are suggested and efficient &
fast dynamic analysis of a set of complex geared mechanical system is presented
in this investigation.

14.2. TOOTH PROFILE OF SPUR GEAR

The gear teeth profile is usually defined a special profile called an involute
curve for constant speed ratio. However, it is not efficient to use the exact
involute profile in the contact search algorithm because of its complexity of
contact search kinematics. In order to approximate the involute profiles, biarc

14-3
curve fitting method which is proposed by Bolton[11] is employed in this
investigation. The optimum biarc curve passing through a given set of points
along involute curve can be determined by this approximation technique. The
more arcs are used to describe the involute profiles, the less numerical error is
occurred in approximation, but the more calculation time will be required for
contact search of tooth profiles. Consequently, the real geometry of involute
tooth profiles in this investigation is represented by 5 arcs with different radii as
shown in Figure 1, since the error is acceptably small.

Fig. 1 Involute curves by 5 arcs

Arc segment Absolute error (mm) Relative error (%)
1 0.000229 0.00147
2 0.000349 0.00152
3 0.000388 0.00165
4 0.000409 0.00168
5 0.000461 0.00182

Table 1. Absolute and relative error

RecurDyn


Table 1 shows the difference between exact involute curve and approximated
arc segment in spur gear with 24 teeth, 2 mm module, and 20 pressure angle.
Absolute error is an average distance between points on exact involute curve and
points on arc segment from gear center. Relative error is an average difference
percentage that absolute error is divided by average distance of points on
involute curve from gear center. Since the main purpose of the research is to
understand the dynamic behaviors of system with the gear pairs, these kinematic
errors might affect very small for the highly oscillating nonlinear dynamics of
gear system, accordingly.

14.3. EFFICIENT CONTACT SEARCH ALGORITHM
AND CONTACT FORCE MODEL

The contact algorithms for a gear pair are investigated in this section. The
contact positions and penetrated values are defined from the kinematics of
components in searching routines. Thereafter, a concentrated contact force is
generated at the contacted position of the contact surface of the bodies. A
detailed discussion on the formulation of the contact collision is represented in
this section.

14.3.1. ARC-ARC CONTACT

Since the radius and angle of each arc are given at geometry, the contact
kinematics between arcs can be calculated by contact logic. A marker is attached
at the center of the arc and X axis is fixed to the starting point of arc. The
monitoring vector between arc centers can be easily detected whether they are in
contact boundary or not using the arc angles with respect to the X axis of the
marker. If the vector is in contact boundary and the length between the centers of
arcs is less than the sum of the radii of arcs, they are considered as contact
candidate.

14-5

X
Y
Z
i
X
i
Y
i
Z
i
t
X
i
t
Y
i
t
Z
i
R
i
u
j
p
X
j
Y
j
Z
j
R
j
p
u
p
t
ij
k
u
j
j
p
Y
j
X
i
Pinion tooth
coordinate system
Pinion
coordinate system
Gear
coordinate system
t
Gear tooth
coordinate system
Global coordinate
system
Fig. 2 Arc-arc contact kinematics

The contact conditions between the gear tooth convex arc segment and the
pinion tooth convex arc segment can be determined as follows. A coordinate
system and is attached to each arc origin coordinate system
shown in Fig. 2. The surface of the gear tooth arc segment is approximated by
plane surfaces and the axis of each surface coordinate system is assumed to be
directed to the starting arc point from arc origin. The surface of pinion tooth arc
segment is approximated by plane surfaces and the axis of each arc origin
coordinate system is assumed to be directed to the starting arc point from arc
origin. The orientation of the gear tooth arc coordinate system with respect to
the global system is defined by
i
t
i
t
i
t
Z Y X
j
p
j
p
j
p
Z Y X
i
t
X
j
p
X
k

i
k
i i
t
A A A =
(1)

i
A

RecurDyn


coordinate system of the gear and is the transformation matrix that
defines the orientation of the gear tooth arc coordinate system with
respect to the gear coordinate system. The orientation of the pinion tooth arc l
coordinate system with respect to the global system is defined by
i
l
j
A
i
k
A
k
i
t
i
t
i
t
Z Y X
j
j
l
A
j
p
X
A +
i
+
r
j
Z

j j
p
A A =
(2)

coordinate system of the pinion
j
A
and is the transformation matrix that
defines the orientation of the pinion tooth arc coordinate system
with respect to the pinion coordinate system.
l
j
p
j
p
Z Y
The global position vector of the center of the gear arc segment, denoted as
point , is defined as t

i
t
i i i
t
u R r =
(3)
where
i
R is the global position vector of the origin of the gear and
i
t
u is
the position vector of arc center point with respect to the origin of the gear
coordinate system .
t
i i i
Z Y X
The global position vector of the center of the pinion arc segment, denoted as
point p , can be defined as

j
p
j j j
p
u A R r =
(4)

arc center point defined in the pinion coordinate system . p
j j
Y X
The position vector of the center of the arc of pinion with respect to the origin
of the gear tooth arc can be defined in the global coordinate system as

i
t
j
p
ij
k
r u =
(5)

The components of the vector with respect to the gear and pinion tooth
ij
k
u

14-7
coordinate system are determined, respectively, as

[ ]
ij
k
T
i
t
T
i ij
z
i ij
y
i ij
x
i ij
u u u u A u = =
, , , ,
(6)

[ ]
ij
k
T
j
p
T
j ji
z
j ji
y
j ji
x
j ji
u u u u A u = =
, , , ,
(7)

Necessary but not sufficient conditions for the contact to be occurred between
the gear and pinion arc segment are

p t
i ij
y
i ij
x
r r u u + +
2 , 2 ,
) ( ) (
(8)

p t
ij
z p t
w w u w w +
(9)

where and r are the radius of the gear and pinion arc segment respectively,
is half width of the gear tooth and is half width of the pinion tooth.
t
r
p
t
w
p
w

) , ( atan2
, , i ij
x
i ij
y m
u u =
,
) , ( atan2
, , j ji
x
j ji
y n
u u =
(10)

k m
0
, (11)
l n
0

where and are the angle of with respect to the gear and pinion
tooth arc segment coordinate system and and are the angle of gear and
pinion arc segment, respectively.
m
ij
k
u
k l

ij

2 2
) ( ) (
ij
y
ij
x p t
ij
u u r r + + =
(12)

RecurDyn


14.3.2. ARC-POINT CONTACT

The arc-point contact conditions between the gear and the pinion can be
determined. A coordinate system is located at the center point of the
gear arc surfaces.
i
t
i
t
i
t
Z Y X
The position vector of the point p of pinion j with respect to the center point
of the gear tooth arc is defined in the global coordinate system such as in Eqs.
(5) and (6).
pinion point and the gear tooth are k

r u u
ij
y
ij
x
+
2 2
) ( ) (
(13)

p t
ij
z p t
w w u w w +
(14)

where r is the radius of the gear arc segment, is half width of the gear
tooth and is half width of the pinion tooth.
t
w
p
w

) , ( atan2
ij
x
ij
y m
u u =
(15)

k m
0
(16)

where is the angle of with respect to the gear arc segment coordinate
system and is the angle of arc segment.
m
ij
k
u
k

ij

2 2
) ( ) (
ij
y
ij
x
ij
u u r + =
(17)

14-9
14.3.3. CONTACT FORCE MODEL

In the field of multi-body dynamics, one of the most popular approximations
of the dynamic behavior of a contact pair has been that one body penetrates into
the other body with a velocity on a contact point, thereafter the compliant normal
and friction forces are generated between a contact pair. In this compliant
contact force model, a contact normal force can be defined as an equation of the
penetration, which yields

3
2
1
m
m
m
n

c k f
&
&
&
+ =
(18)

where k and c are the spring and damping coefficients which are determined,
respectively and the is time differentiation of penetrated value
&
. The
exponents and generates a non-linear contact force and the exponent
yields an indentation damping effect. When the penetration is very small,
the contact force may be negative due to a negative damping force, which is not
realistic. This situation can be overcome by using the indentation damping
exponent greater than one. The friction force is obtained by
1
m
2
m
3
m

n f
f f =
(19)

where is the friction coefficient and its sign and magnitude can be
determined from the relative velocity of the pair on contact position.

14.4. KINEMATICS AND EQUATION OF MOTION FOR
SYSTEM DYNAMICS USING THE RECURSIVE FORMULAS

Recursive formulas using relative coordinates are very useful for gear system
dynamic analysis since gears in geared systems are usually rotated to one axis

RecurDyn


direction. This section presents the relative coordinate kinematics for a contact
pair as well as for joints connecting two bodies.
Translational and angular velocities of the body coordinate system with
respect to the global coordinate system are respectively defined as

w
r&
(20)

Their corresponding quantities with respect to the body coordinate system
are defined as
=
w A
r A
Y
T
T
&
(21)

where is the combined velocity of the translation and rotation. The recursive
velocity and virtual relationship for a pair of contiguous bodies are obtained in
[16] as
Y

1)i (i 1)i2 (i 1) (i 1)i1 (i i
+ = q B Y B Y &
(22)

where denotes the relative coordinate vector. It is important to note that
matrices and are only functions of the . Similarly, the
recursive virtual displacement relationship is obtained as follows
1)i (i
q
(i
B
1)i1 1)i2 (i
B
1)i (i
q

1)i (i 1)i2 (i 1) (i 1)i1 (i i

+ = q B Z B Z
(23)

If the recursive formula in Eq. (22) is respectively applied to all joints, the
following relationship between the Cartesian and relative generalized velocities
can be obtained:

q B Y & =
(24)

where is the collection of coefficients of the and B
1)i (i
q&

14-11
[ ]
T
1 nc
T T
2
T
1
T
0
=
n
Y , , Y , Y , Y Y K
(25)

[
T
1 nr
T
) 1 (
T
12
T
01
T
0

=
n n
q , , q , q , Y q & K & & & ] (26)

where nc and nr denote the number of the Cartesian and relative coordinates,
respectively. Since in Eq. (24) is an arbitrary vector in q&
nr
R , Eqs. (22) and
(24), which are computationally equivalent, are actually valid for any vector
such that
nr
R x &

x B X & =
(27)

and

1)i - (i 1)i2 - (i 1) - (i 1)i1 - (i i
x B X B X + = (28)

result, transformation of into is actually calculated by
recursively applying Eq. (28) to achieve computational efficiency in this
research. Inversely, it is often necessary to transform a vector in
nc
R X B
G
x
nc
nr
R x
nc
R Bx
R into
a new vector in G B
T
= g
nr
R . Such a transformation can be found in the
generalized force computation in the joint space with a known force in the
Cartesian space. The virtual work done by a Cartesian force Q is
obtained as follows.
nc
R

Q Z W

=
(29)

where must be kinematically admissible for all joints in a system.
Substitution of
Z
q B Z = into Eq. (29) yields

* T T T
Q q Q B q W = =
(30)

RecurDyn


where . Q B Q
T *
The equations of motion for constrained systems have been obtained as

follows.

0 ) Q Y M B F

T
= + =
&
(
(31)

where the is the Lagrange multiplier vector for cut joints [17] in and
represents the position level constraint vector in
m
R
m
R . The and Q are
the mass matrix and force vector in the Cartesian space including the contact
forces, respectively.
M


A spur gear pair system is analyzed for the sake of numerical verification of
proposed methods as shown in Fig. 3. The shafts of the two gears are assumed to
be rigid and the only the compliance of contact force between meshing teeth is
considered in this model. The gear pare model is composed of 2 spur gears, 2
revolute joints, and a gear contact element. Rotational dampers are used for
resistance torque at revolute joints. A gear is driven by steady torque of 10 Nm.

1 1
, r
2 2
, r
Revolute joint & Rotational spring
damper
Applied torque
Contact element

Fig. 3 Gear pair model

14-13

Gear/Pinion
Module 3 mm
Number of teeth 50
Pressure angle 20
Radius of pitch circle 75 mm
Radius of outside circle 78 mm
Radius of base circle 70.477 mm
Radius of root circle 71.25 mm
Tooth width 20 mm
Elasticity modulus
2 9
/ 10 200 m N
Density
3 3
/ 10 85 . 7 m kg
Center distance 150 mm

Table 2 Design parameters of gear and pinion

Table 2 shows design parameters of the spur gear sets which are the inputs of
numerical simulation. Dynamic analysis of a spur gear pair is simulated during
0.08 sec. Gear speed is increased up to 500 rad/sec (4800 rpm) almost linearly as
shown in Fig. 4(a). It is found that the CPU simulation time is just 15 sec on a
Pentium IV 3.0 GHz platform personal computer. Figure 4(b) demonstrates the
dynamic transmission error (DTE= ) with respect to time domain when
a gear is driven at the constant torque of 10 Nm. As rotating speed of gear is
increased, dynamic transmission error (DTE) is changed by gear teeth contact.
Figure 5(a) and 5(b) show the time-domain DTE around mesh frequency of 1900
Hz and 3000 Hz. Magnitude and waveform of DTE are different in each mesh
frequency. Magnitude of DTE is around 30 and 3 micro meter, respectively.
These results show similar magnitude and exact dynamic pattern as compared to
experimental measurement results (in the reference Fig. 6 and 7) introduced by
Blankenship and Kahraman [7]. The minor differences between the proposed
method and referenced [7] might be expected from the dimensions, measurement
settings and noises.
2 2 1 1
r r +

RecurDyn


(a) Rotational velocity of driven gear

(b) Oscillating DTE with respect to time

Fig. 4 Rotational velocity and DTE

14-15

(a) DTE at the mesh frequency of 1900Hz

(b) DTE at the mesh frequency of 3000Hz

Fig. 5 Oscillating DTE time history

RecurDyn


The key advantage of the proposed method is the fast & efficient system
simulation of geared multibody dynamic system without losing the system
dynamic characteristics caused by gear pair contacts and their flexibility. An
Engine system with multi gear sets is illustrated as another geared system
example model. The system has 4 degrees of freedom, which has 13 bodies, 6
revolute joints, one translational joint, 14 fixed joints, and 2 sets of contacting
spur gear pairs. Crankshaft in this model is rotated by gas force and gear sets are
driven by rotation of crankshaft as shown in Fig. 6. In order to examine the
effect of gear contact dynamics, the proposed gear contact force model is
compared by constraint coupler model which should be ideal solution but not
realistic. Figure 7 shows well the difference of output velocity from the final
gear between proposed method and conventional dynamic anaysis using
constraint only. Dynamic analyses of both models are performed for 0.01 sec. It
is found that the CPU simulation time is just 85 sec for the proposed method on
a Pentium IV 3.0 GHz platform personal computer.

Fig. 6 Engine model with multi gear set

14-17

Fig. 7 Rotational velocity in output gear

14.6. CONCLUSION

This research proposes an efficient implementation algorithm of spur gear
contact mechanisms for the fast system dynamic analysis. Externally specified
dynamic forces, or assumptions about modeling the mesh forces by time-varying
stiffness and static transmission error are not required since dynamic mesh forces
are obtained by contact analysis directly at each time step. Arc-Arc and arc-point
kinematic interactions are presented and a compliant force model is used in this
investigation. The relative coordinate formulation is employed to generate the
equations of motion. Two numerical examples, a simple spur gear pair and an
engine transmission system, are illustrated and simulated numerically in this
investigation. A simple spur gear pair model shows the validation of the
proposed method with measurement results illustrated by reference, and engine
transmission system shows the advantages of the proposed method, respectively.
Consequently it is possible to simulate the entire geared system dynamic analysis
without losing its important dynamic characteristics, such as vibration and noise,
etc., with reasonable CPU time as represented in this investigation.

RecurDyn


REFERENCES

1. ANSYS User Manual, ANSYS Inc., PA., USA
2. ABAQUS User Manual, ABAQUS Inc., RI., USA
3. Kahraman, A. and Singh, R., ''Non-Linear Dynamics of a Spur Gear Pair'',
Journal of Sound and Vibration, Vol.142, No.1, pp.47-75, 1990.
4. Amabili, M. and Rivola, A., ''Dynamic Analysis of Spur Gear Pairs: Steady-State
Response and Stability of the SDOF Model with Time-Varying Meshing
Damping'', Mechanical Systems and Signal Processing, Vol.11, No.3, pp.375-390,
1997.
5. Ozguven, H. N. and Houser, D. R., ''Mathematical Models used in Gear
Dynamics a Review'', Journal of Sound and Vibration, Vol.121, pp. 383-411,
1988.
6. Shing, T., Tsai, L., and Krishnaprasad, P., "An Improved Model for the Dynamics
of Spur Gear Systems with Backlash Consideration ", ASME-PUNLICATION-
DE, Vol.65-1, pp. 235-244, 1993.
7. Blankenship, G. W. and Kahraman, A., Gear dynamics experiments, Part-I:
Characterization of forced response, ASME, Power Transmission and Gearing
Conference, San Diego, 1996.
8. Kahraman, A. and Blankenship, G. W., Gear dynamics experiments, Part-II:
Effect of involute contact ratio, ASME, Power Transmission and Gearing
9. Kahraman, A. and Blankenship, G. W., Gear dynamics experiments, Part-III :
Effect of involute tip relief, ASME, Power Transmission and Gearing
10. Parker, R. G. and Vijayakar, S. M. and Imajo, T., ''Non-linear Dynamic
Response of a Spur Gear Pair : Modeling and Experimental Comparisons'',
Journal of Sound and Vibration, Vol.237, pp. 435-455, 2000.

14-19

RecurDyn

11. Bolton, K. M., Biarc curves, Computer Aided Design, Vol.7, No.2, pp.89-92,
1975.
12. Parkinson, D. B. and Moreton, D. N., ''Optimal biarc curve fitting, Computer
Aided Design, Vol. 23, No.6, pp.411-419, 1991.
13. Ryu, H. S., Huh. K. S., Bae, D. S. and Choi, J. H., ''Development of a Multibody
Dynamics Simulation Tool for Tracked Vehicles, Part I : Efficient Contact and
Nonlinear Dynamic Modeling'', JSME International Journal, Series C, Vol.46,
No.2, pp.540-549, 2003.
14. Lankarani, H. M., "Canonical Impulse-Momentum Equations for Impact
Analysis of Multibody System", ASME, Journal of Mechanical Design, Vol. 180,
pp. 180-186, 1992.
15. Bae, D. S., Han, J.

Recurdyn Solver - Theoretical Manual

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Recurdyn Solver - Theoretical Manual

Transféré par

Droits d'auteur :

Formats disponibles

R

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

ncut) - (nr such that the augmented

& & & & & &

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

is used and denotes the kth column of the matrix . The

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

, k is the order of integration and

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

1 i and can be expressed by the 1-2-3 Euler angle

) cos( ) cos( ) sin( 0

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL

/SOLVER THEORETICAL MANUAL