Professional Documents
Culture Documents
New Developments in
Multibody System Dynamics
and Its Applications
Guest Editors: Xiaoting Rui, Chuanzeng Zhang, and Caishan Liu
New Developments in Multibody System
Dynamics and Its Applications
Advances in Mechanical Engineering
This is a special issue published in “Advances in Mechanical Engineering.” All articles are open access articles distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
Editorial Board
Koshi Adachi, Japan Luı́s Godinho, Portugal David R. Salgado, Spain
Mehdi Ahmadian, USA Tian Han, China Mohammad R. Salimpour, Iran
Rehan Ahmed, UK Francisco J. Huera-Huarte, Spain Sunetra Sarkar, India
Muhammad T. Akhtar, Japan Davood Jalali-Vahid, Iran Pietro Scandura, Italy
Nacim Alilat, France Jiin Y. Jang, Taiwan A. S. Sekhar, India
M. Affan Badar, USA Zhongmin Jin, UK Liyuan Sheng, China
Luis Baeza, Spain Xiaodong Jing, China Xi Shi, China
R. Balachandran, UK S.-W. Kang, Republic of Korea Seiichi Shiga, Japan
Claude Bathias, France Xianwen Kong, UK Chow-Shing Shin, Taiwan
Adib Becker, UK Michal Kuciej, Poland Ray W. Snidle, UK
Leonardo Bertini, Italy Yaguo Lei, China Margaret M. Stack, UK
Liam A. Blunt, UK Zili Li, The Netherlands Neil Stephen, UK
Noël Brunetière, France Yangmin Li, Macau Kumar K. Tamma, USA
Marco Ceccarelli, Italy Jaw-Ren Lin, Taiwan Yaya Tan, China
Fakher Chaari, Tunisia Cheng-Xian Lin, USA Anand Thite, UK
Chin-Lung Chen, Taiwan Seyed N. Mahmoodi, USA Cho W. Solomon To, USA
Hyung H. Cho, Republic of Korea Oronzio Manca, Italy Yoshihiro Tomita, Japan
Seung-Bok Choi, Korea Ramiro Martins, Portugal Shan-Tung Tu, China
Ahmet S. Dalkilic, Turkey Aristide F. Massardo, Italy Sandra Velarde-Suárez, Spain
Kangyao Deng, China Francesco Massi, Italy Moran Wang, China
Francisco D. Denia, Spain T. H. New, Singapore Junwu Wang, China
T. S. Dhanasekaran, USA Kim C. Ng, Singapore Jia-Jang Wu, Taiwan
Nihad Dukhan, USA Cong T. Nguyen, Canada Fengfeng Xi, Canada
Farzad Ebrahimi, Iran Hirosi Noguchi, Japan Gongnan Xie, China
Ali El Wahed, UK Hakan F. Oztop, Turkey Wei Mon Yan, Taiwan
Bogdan I. Epureanu, USA Duc T. Pham, UK Jianqiao Ye, UK
Mohammad R. Eslami, Iran Jurij Prezelj, Slovenia Byeng D. Youn, USA
Ali Fatemi, USA Xiaotun Qiu, USA Bo Yu, China
Mario L. Ferrari, Italy Pascal Ray, France Jianbo Yu, China
Siegfried Fouvry, France Robert L. Reuben, UK Zhongrong Zhou, China
Ian Frigaard, Canada Pedro A. R. Rosa, Portugal
Mergen H. Ghayesh, Canada Elsa de Sá Caetano, Portugal
Contents
New Developments in Multibody System Dynamics and Its Applications, Xiaoting Rui,
Chuanzeng Zhang, and Caishan Liu
Volume 2014, Article ID 671604, 1 page
Automatic Deduction Theorem of Overall Transfer Equation of Multibody System, Xiaoting Rui,
Jianshu Zhang, and Qinbo Zhou
Volume 2014, Article ID 378047, 12 pages
Optimal Design of One-Folded Leaf Spring with High Fatigue Life Applied to Horizontally Vibrating
Linear Actuator in Smart Phone, Ki Bum Lee, Chang Hyun Park, and Jin Ho Kim
Volume 2014, Article ID 545126, 7 pages
Application of Krylov Reduction Technique for a Machine Tool Multibody Modelling, M. Sulitka,
J. Šindler, J. Sušeň, and J. Smolı́k
Volume 2014, Article ID 592628, 9 pages
Study on Launch Dynamics of Self-Propelled Artillery Based on Transfer Matrix Method of Multibody
System, Hailong Yu and Xiaoting Rui
Volume 2014, Article ID 308049, 9 pages
Higher Order Variational Integrators for Multibody System Dynamics with Constraints, Jieyu Ding and
Zhenkuan Pan
Volume 2014, Article ID 383680, 8 pages
Free Vibration Characteristic of Multilevel Beam Based on Transfer Matrix Method of Linear
Multibody Systems, Laith K. Abbas and Xiaoting Rui
Volume 2014, Article ID 792478, 16 pages
Case Study on Human Walking during Wearing a Powered Prosthetic Device: Effectiveness of the System
“Human-Robot”, Svetlana Grosu, Pierre Cherelle, Chris Verheul, Bram Vanderborght, and Dirk Lefeber
Volume 2014, Article ID 365265, 9 pages
Controller Parameters Tuning Based on Transfer Matrix Method for Multibody Systems,
Hossam Hendy, Xiaoting Rui, Qinbo Zhou, and Mostafa Khalil
Volume 2014, Article ID 957684, 8 pages
All Terrain Vehicle Flexible Multibody Dynamic Simulation for Fatigue Prediction, Jia-Shiun Chen and
Hsiu-Ying Hwang
Volume 2013, Article ID 725315, 6 pages
Projectile Impact Point Prediction Based on Self-Propelled Artillery Dynamics and Doppler Radar
Measurements, Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy
Volume 2013, Article ID 153913, 12 pages
Study on the Dynamics of Laser Gyro Strapdown Inertial Measurement Unit System Based on Transfer
Matrix Method for Multibody System, Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang,
and Qinbo Zhou
Volume 2013, Article ID 854583, 9 pages
Recursive Formulation for Dynamic Modeling and Simulation of Multilink Spatial Flexible Robotic
Manipulators, Zhenjie Qian, Dingguo Zhang, and Jun Liu
Volume 2013, Article ID 216014, 15 pages
Dynamic Control and Disturbance Estimation of 3D Path Following for the Observation Class
Underwater Remotely Operated Vehicle, Hai Huang, Qirong Tang, Yueming Li, Lei Wan, and Yongjie Pang
Volume 2013, Article ID 604393, 16 pages
Application of Finite Volume Method to Structural Stochastic Dynamics, Weidong Chen, Yanchun Yu,
Ping Jia, Xiande Wu, and Fengchao Zhang
Volume 2013, Article ID 391704, 10 pages
Numerical Simulation on Dense Packing of Granular Materials by Container Oscillation, Jun Liu and
Dongxu You
Volume 2013, Article ID 284693, 7 pages
Research on Flexible Joint Stiffness Test and Trajectory Planning of Space Manipulator, Yongtai Yang,
Jili Rong, Jian Li, and Ling Tang
Volume 2013, Article ID 280453, 7 pages
Case Study on Influence of Step Blast-Excavation on Support Systems of Existing Service Tunnel with
Small Interval, Shaorui Sun, Ling Yue, Jimin Wu, Jin Liu, and Jihong Wei
Volume 2013, Article ID 257457, 13 pages
The Effect of First-Order Bending Resonance of Wheelset at High Speed on Wheel-Rail Contact
Behavior, Shuoqiao Zhong, Xinbiao Xiao, Zefeng Wen, and Xuesong Jin
Volume 2013, Article ID 296106, 19 pages
Dynamic Modeling and Analysis of Hub-Tapered Beam System, Si-jia Chen and Ding-guo Zhang
Volume 2013, Article ID 313279, 12 pages
Cooperative Search by Combining Simulated and Real Robots in a Swarm under the View of Multibody
System Dynamics, Qirong Tang and Peter Eberhard
Volume 2013, Article ID 284782, 11 pages
Global Impact Dynamic Modeling and Verification of a Flexible Beam with Large Overall Motion,
Yuechen Duan, Dingguo Zhang, and Jiazhen Hong
Volume 2013, Article ID 362317, 8 pages
Evaluating Energy Flux in Vibrofluidized Granular Bed, N. A. Sheikh, S. Manzoor, and U. Ghani
Volume 2013, Article ID 327379, 7 pages
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 671604, 1 page
http://dx.doi.org/10.1155/2014/671604
Editorial
New Developments in Multibody System Dynamics and
Its Applications
Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Multibody system dynamics is nowadays a widely used tool- and precision, the model and modal reduction techniques
box in the development of modern engineering science and for multibody system dynamics, the optimization and control
technology. It provides powerful dynamic design theories, of a swarm of robots, the optimal design of smart systems,
computational and analytical methods, and test technology the dynamics and control of underwater vehicle and space-
for innovative engineering applications. In order to sys- craft formation, the perturbation-finite volume method for
tematically show the important effects of multibody system stochastic multibody system dynamics, the blast-excavation
dynamics on the development of modern science and tech- dynamics, the granular system dynamics, the vehicle-track
nology and to strengthen the academic exchanges in the coupling dynamics, and the human-robot dynamics. Thus,
research field of multibody system dynamics and its applica- the general fundamental theories and several key engineering
tions, we organized this special issue containing twenty-five applications are covered representing the state of the art of
papers selected from the contributions coming from Belgium, recent intensive and extensive research activities in the field
Czech Republic, Egypt, Germany, Iraq, Netherlands, Pak- of multibody system dynamics to a certain extent.
istan, China, Republic of Korea, United Kingdom, Taiwan, The guest editors hope that this special issue will pro-
and so forth. vide some valuable information and guidance for scientists
The twenty-five papers selected for this Special Issue are working on multibody system dynamics and its applications
devoted to the following topics: an automatic deduction the- in modern engineering science and technology.
orem of the overall transfer equations of the transfer matrix
method for multibody systems (MSTMM), the application
of MSTMM in dynamics of a multilevel beam system, the Acknowledgments
application of MSTMM in dynamics of an elastic-plastic Many thanks are due to Professor Edwin Kreuzer from
system subjected to earthquakes, the application of MSTMM Hamburg University of Technology and Professor Dieter
in dynamics of controlled cars, the application of MSTMM Bestle from Brandenburg University of Technology Cottbus-
in dynamics of a laser gyro strapdown inertial measurement Senftenberg for their great unselfish help in organizing the
unit system, the application of MSTMM in launch dynamics special issue, to reviewers and authors, and to the reviewers
of a self-propelled artillery system and flight dynamics of for their voluntary works and constructive comments. All
projectile, the modeling of trivariate isogeometrics for flexible of these promoted greatly the projected publication of this
multibody dynamics, the modeling of hub-tapered beam special issue.
system dynamics, the formulations and modeling approaches
of spatial flexible robotic manipulators, the impact in flexible Xiaoting Rui
systems, the modeling and test methods for flexible joint Chuanzeng Zhang
stiffness of a space manipulator, the computational stability Caishan Liu
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 378047, 12 pages
http://dx.doi.org/10.1155/2014/378047
Research Article
Automatic Deduction Theorem of Overall Transfer
Equation of Multibody System
Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Transfer matrix method for multibody System (MSTMM) is a new multibody dynamics method developed in recent 20 years. It
has been widely used in both science research and engineering for its special features as follows: without global dynamics equations
of the system, high programming, low order of system matrix, and high computational speed. Based on MSTMM and its above
features, a theorem to deduce automatically the overall transfer equations of multibody systems by handwriting or by computer
is proposed in this paper. The theorem is effective for multibody systems with various topological structures, including chain
systems, closed-loop systems, tree systems, general systems composed of one tree subsystem, and some closed-loop subsystems.
This theorem makes it possible to program large scale software of multibody system dynamics with much higher programming,
and much higher computational speed because of the above features of MSTMM. Formulations of the proposed method as well as
two examples are given to verify this method.
transfer equations of various multibody systems mentioned time derivatives of the generalized coordinates describing
above by handwriting and by computer is proposed in this the deformation of flexible bodies using modal method; the
paper. superscript 𝑛 is the highest order of the modal considered, r,̈
𝜃,̈m, and q are the column matrices of accelerations, second
2. General Theorems and Steps of MSTMM time derivatives of the space-three angles, internal torques,
and internal forces in 𝑜𝑥𝑦𝑧, respectively.
2.1. Basic Idea of MSTMM. The basic idea of MSTMM [19] is For body and hinge elements moving in a plane, a similar
to break up a multibody system into the elements contain- definition of the state vector can be introduced, which is a
ing bodies (including rigid bodies, flexible bodies, lumped special example of spatial motion.
masses, etc.) and hinges (including joints, ball-and-sockets, The transfer equations of the 𝑗th element can be obtained
pins, springs, rotary springs, dampers and rotary dampers, easily by rewriting its dynamics equations as follows [19, 21,
etc.) whose dynamics properties can be readily expressed 30]
in matrix forms. These matrices of elements are considered
as building blocks that provide the dynamics properties of z𝑗,𝑗+1 = U𝑗 (𝑡𝑖 ) z𝑗,𝑗−1 , (5)
the entire system when assembling them together according
to the topology of the system. Particularly, the positions where U𝑗 (𝑡𝑖 ) is the transfer matrix of the 𝑗th element, which
of bodies and hinges are considered equivalent in transfer expresses the relationship between the state vectors of its
equations and transfer matrices, which is totally different output end z𝑗,𝑗+1 and input end z𝑗,𝑗−1 and is already known
from ordinary methods for MSD [1–18] and results in the at time instant 𝑡𝑖 .
very low order of system matrix and very high computational It should be pointed out that there are general linear
speed in MSTMM. relations among accelerations, angular accelerations, forces,
and torques of any mechanics system in an inertial coordinate
2.2. State Vector, Transfer Equation, and Transfer Matrix of system 𝑜𝑥𝑦𝑧, according to Newton motion law and Euler
Element. The same coordinate systems and sign conventions theorem of moment of momentum. It is to say that there are
as are described in [19, 21, 30] will be used. The state vector strict linear relations between the state vectors of output end
of the connection point between any rigid body and hinge and input end of any element and among all state vectors
moving in space is defined as of a multibody system. Thus, the transfer equation (5) is a
general equation and effective for any mechanics element in
𝑇 the inertial coordinate system.
z = [𝑥̈𝑦̈𝑧̈𝜃𝑥̈ 𝜃𝑦̈ 𝜃𝑧̈ 𝑚𝑥 𝑚𝑦 𝑚𝑧 𝑞𝑥 𝑞𝑦 𝑞𝑧 1] , (1)
(1) A circle I denotes a body element and the number Figure 1: Topology figure of a dynamics model of tree multibody
inside this circle is the sequence number of the body system.
element.
(2) An arrow → denotes a hinge element and the transfer 4. Automatic Deduction of Overall Transfer
direction of state vectors; the number beside the
arrow is the sequence number of the hinge element. Equations of System
4.1. Automatic Deduction of the Overall Transfer Equation of
(3) Each body element is dealt with single output end and
a Chain System. The topology figure of any chain system is
single input end if the body has two connection ends
shown in Figure 2.
with other elements; otherwise, it is dealt with single
It is clear that we can rewrite the overall transfer equation
output end and multiple input ends if the body has
(6) of the chain system as
more two connection ends.
𝑇
(4) For a nonboundary end, the first and second sub- Uall [z𝑇1,0 z𝑇𝑛,0 ] = 0, Uall = [−I U𝑛−1 ] . (8)
scripts, 𝑖 and 𝑗 (𝑖, 𝑗 ≠0), in a state vector z𝑖,𝑗 of the
end denote the sequence numbers of the adjacent From equations (6) or (7), it can be seen clearly that
body element and hinge element, respectively. For the overall transfer matrix of any chain system Uall can be
a boundary end, the second subscript 𝑗 = 0 in deduced automatically by handwriting and by computer. In
the state vector z𝑖,𝑗 ; that is, the second subscript 0 the overall transfer matrix, the coefficient matrix of the state
means boundary end; then the first subscript 𝑖 in the vector of root is a minus unit matrix I, while the coefficient
state vector z𝑖,0 of the boundary end stands for the matrix of the state vector of tip U𝑛−1 is the successive
sequence number of the element involved. premultiplication of the transfer matrices of all elements in
the transfer path from this tip to the root as shown in equation
(5) In a multibody system, only one boundary end is
(7).
considered as the root; the state vector of root is noted
The highest order of the overall transfer matrix is 13 for
as z𝑖,0 , where 𝑖 is the sequence number of the root
spatial chain multi-rigid-body system or (13 + 𝑛) for chain
element; all of other boundary ends are considered as
multi-rigid-flexible-body system, where 𝑛 is the highest order
the tips; the state vectors of tips are denoted as z𝑗,0 ,
of the modal considered.
where 𝑗 is the sequence number of the tip element.
The transfer directions of a system are always from its
tips to the root. 4.2. Automatic Deduction of the Overall Transfer Equations
of a Closed-Loop System. For any closed-loop system, whose
(6) The subscript 𝑖 in transfer matrix U𝑖 denotes the topology figure is shown in Figure 3, after “cutting” at the
sequence number of element 𝑖. The subscript (𝑖 − 𝑘) junction of any two adjacent elements such as body 1 and
in the transfer matrix U𝑖−𝑘 and the partitioned matrix hinge 𝑛 as shown in Figure 4, consider the couple of “cutting
u𝑖−𝑘 means from element 𝑖 to element 𝑘. U𝑖−𝑘 and u𝑖−𝑘 point” as the “boundary ends” with the same state vectors
mean the successive premultiplication of the transfer noted as z𝑛,0 and z1,0 . Then the original closed-loop system
matrices of all elements in the transfer path from the becomes a chain system with the same “boundary ends” state
element 𝑖 to element 𝑘 of the system. vectors, z𝑛,0 and z1,0 , as shown in Figure 4.
4 Advances in Mechanical Engineering
0 n − 1 ··· 7 5 3 1 0 Root from the state vector when it premultiplies z𝑗,𝐼𝑘 , acted on the
𝒛n,0 n 6 4 2 same point.
𝒛1,0
However, in equation(13), the number of unknown vari-
Figure 2: Topology figure of a chain system. ables is more than that of algebraic equations. Therefore,
geometrical equations of the body, which describes the
geometrical relationship between the first input end and
kth (𝑘 = 2, 3, 4, . . . , 𝐿) input end of the body, should be
7 ··· n−1 n 1 introduced for body elements with single output end and
multiple input ends. It is verified later that the geometrical
6 equation can be written in the form of
2
𝒛n,0 𝒛1,0 By equations (16), (18), and rewriting them in the form of
7 ··· n−1 0 0 1
equation (13), transfer matrices can be obtained as
n
I3 E1 O3×3 O3×3 E2
6
2 [ O3×3 I3 O3×3 O3×3 O3×1 ]
[ ]
U𝑗 = [
[E6 E3 + E7 E6 E4 + E8 I3 E6 E6 E5 + E9 ]
],
5 3 [ E3 E4 O3×3 I3 E5 ]
4
[ O1×3 O1×3 O1×3 O1×3 1 ]
Figure 4: Topology figure of a closed-loop system after “cutting” the (19)
hinge 𝑛.
O3×3 O3×3 O3×3 O3×3 O3×1
[O3×3 O3×3 O3×3 O3×3 O3×1 ]
[ ]
U𝑗,𝐼𝑘 =[
[O3×3 O3×3 I3 ̃r𝐼1 𝐼𝑘 O3×1 ]
]
[O3×3 O3×3 O3×3 I3 O3×1 ] (20)
y2
O [O1×3 O1×3 O1×3 O1×3 1 ]
y
x2 (𝑘 = 2, 3, 4, . . . , 𝐿) ,
C
o2 IL E1 = −Ãl𝐼𝑂H, E2 = Ã
𝜔𝜔̃l𝐼𝑂 − Ãl𝐼𝑂Ḣ𝜃̇
𝐼,
I1 I2
E3 = −𝑚I3 , E4 = 𝑚Ãl𝐼𝐶H,
o z2
x
E5 = f𝐶 − 𝑚Ã
𝜔𝜔̃l𝐼𝐶 + 𝑚Ãl𝐼𝐶Ḣ𝜃̇
𝐼,
z
E6 = ̃r𝐼𝑂, E7 = 𝑚̃r𝐼𝐶, E8 = AJ𝐼 H,
Figure 5: A spatial motion rigid body with more than two ends.
𝜔J𝜔 + AJḢ𝜃̇
E9 = −m𝐶 − ̃r𝐼𝐶f𝐶 + Ã 𝐼,
(21)
̃ = A𝑇 A,̇
𝜔 𝜔 = H𝜃,̇
regarded as a special case of rigid body with more than two Similarly, the geometrical equations corresponding to
ends (multiple input ends and single output end). body elements 10 and 4 can be deduced as
Based on the transfer equations and geometrical equa-
tions of elements, it is then easy to get the overall transfer G17−10 z17,0 + G18−10 z18,0 = 0,
equation of the system automatically.
G15−4 z15,0 + G16−4 z16,0 + G17−4 z17,0 (30)
According to the topology figure of the system shown in
Figure 1, the relations among the state vectors and transfer + G18−4 z18,0 = 0,
equations of elements can been described more intuitively
and directly using topology described by state vectors and where
transfer equations as shown in Figure 6. G17−10 = −H10 U13 U17 ,
Then the main transfer equations of the system in
Figure 6 can be easily deduced, that is, G18−10 = H10,14 U14 U18 ,
𝒛15,11 = 𝑼15 𝒛15,0 𝒛16,12 = 𝑼16 𝒛16,0 𝒛17,13 = 𝑼17 𝒛17,0 𝒛18,14 = 𝑼18 𝒛18,0
𝒛9,11 = 𝑼11 𝒛15,11 𝒛9,12 = 𝑼12 𝒛16,12 𝒛10,13 = 𝑼13 𝒛17,13 𝒛10,14 = 𝑼14 𝒛18,14
𝒛7,8 = 𝑼8 𝒛8,9
𝒛7,5 = 𝑼7 𝒛7,8
𝒛2,3 = 𝑼3 𝒛4,3
𝒛2,1 = 𝑼2 𝒛2,3
𝒛1,0 = 𝑼1 𝒛2,1
𝑛 is equal to 3 for planar motion or 6 for spatial motion. −I T15−1 T16−1 + T19−1 C T18−1
Considering the method proposed in Section 4.3 for a [ O G15−9 G16−9 O ]
[
Uall = [ ],
tree system, and regarding equation (34) concerned with the O O G19−10 C G18−10 ]
relationship between z16,0 and z19,0 , the overall transfer equa- [ O G15−4 G16−4 + G19−4 C G18−4 ] (38)
tions of a general system with a closed-loop subsystem shown
in Figure 8 can be obtained automatically by handwriting and z1,0
[z15,0 ]
by computer: zall = [ ]
[z16,0 ] ,
Uall zall = 0, (36) [z18,0 ]
8 Advances in Mechanical Engineering
0
𝒛15,0
0 𝒛18,0 0 𝒛15,0 0 𝒛 𝒛19,0 0 𝒛18,0
16,0
15 16 17 18 15 16 19 17 18
19
11 12 13 14 11 12 13 14
9 10 9 10
8 8
7 6 6
7
5 5
4 4
3 3
2 2
Root 1 1
0 Root 0
𝒛1,0 𝒛1,0
Figure 7: Topology figure of a nontree system. Figure 8: Topology figure of a nontree system becoming a tree
system after “cutting” the hinge 19.
where
(2) For a tree system, in the first line of the overall
T15−1 = U1 U2 U3 U4 U5 U7 U8 U9 U11 U15 , transfer matrix, the coefficient matrix of the state
vector of root is a minus unit matrix, while each
T16−1 = U1 U2 U3 U4 U5 U7 U8 U9,12 U12 U16 , coefficient matrix of the state vector of a tip is the
successive premultiplication of the transfer matrices
T19−1 = U1 U2 U3 U4,6 U6 U10 U13 U17 U19 ,
of all elements in the transfer path from this tip to
T18−1 = U1 U2 U3 U4,6 U6 U10,14 U14 U18 , the root; besides the first line in the overall transfer
matrix, all coefficient matrices of state vectors in
G15−9 = −H9 U11 U15 , the first column are zero matrices. Except the first
line, in each row, each nonzero partitioned matrix
G16−9 = H9,12 U12 U16 , corresponds to the coefficient matrix of the tip state
(39)
G19−10 = −H10 U13 U17 U19 , vector, from which there is a transfer path to the input
end of the element with multiple input ends. Each
G18−10 = H10,14 U14 U18 , nonzero coefficient matrix of the state vector of a
tip is the successive premultiplication of all transfer
G15−4 = −H4 U5 U7 U8 U9 U11 U15 , matrices of elements in the transfer path from this tip
to the 𝑘th input end 𝐼𝑘 of jth body element, which has
G16−4 = −H4 U5 U7 U8 U9,12 U12 U16 ,
more than two ends, then premultiplied by −H𝑗 for
G19−4 = H4,6 U6 U10 U13 U17 U19 , 𝑘 = 1 or premultiplied by H𝑗,𝐼𝑘 for 𝑘 = 2, 3, 4 ⋅ ⋅ ⋅ 𝐿.
G18−4 = H4,6 U6 U10,14 U14 U18 . (3) For a chain system, its overall transfer matrix is
deduced automatically by successive premultiplica-
tion of the transfer matrices of all elements in the
5. Automatic Deduction Theorem of Overall transfer path from the tip to the root of system. In
Transfer Equation fact, any chain system can be considered as a special
example of the tree system in the case with only two
The following features of the overall transfer equation of a boundary ends.
multibody system can be clearly concluded from equations
(4) For a closed-loop system, its overall transfer equation
(23), (24), (28), and (30). These features make up the theorem
is deduced automatically as the chain system, after
to deduce automatically the overall transfer equation as the
treating the original system as the chain system by
following for tree system (1, 2), for chain system (1, 3), for
“cutting” a junction of any two adjacent elements and
closed-loop system (1, 3, 4), and for general system (1, 2, 5).
letting the couple of “cutting points” as the tip and
(1) The state vectors involved in an overall transfer root with the same state vectors of the chain system.
equation are the column matrix comprising the state (5) For a general system composed of one tree subsystem
vectors of all boundary ends of the system. and some closed-loop subsystems, its overall transfer
Advances in Mechanical Engineering 9
0 𝐾𝑥,7 = 1000 N/m, 𝐾𝑦,6 = 𝐾𝑦,7 = 500 N/m, and 𝐾6 = 𝐾7 =
8 150 N ⋅ m/rad.
1 T6−1 = U1 U2 U3 U4 U5 U6 ,
𝒛1,0
Root 0 T8−1 = U1 U2 U3 U4 U5,7 U7 U8 ,
(43)
Figure 10: Topology figure of the system. G6−5 = −H5 U6 ,
G8−5 = H5,7 U7 U8 ,
equation is deduced automatically as the tree system, where z1,0 is state vector of root, z6,0 and z8,0 are the state
after treating the original system as the tree system by vectors of tips, and U1 , U2 , U3 , U4 , U5 , U5,7 , U6 , U7 , and U8
“cutting” the junctions of any two adjacent elements are the transfer matrices of body 1, fixed hinge 2, uniform
in every closed-loop subsystem and letting every beam 3, fixed hinge 4, body 5, elastic hinge 6, elastic hinge 7,
couple of “cutting points” as new “boundary ends” and body 8. All of the transfer matrices can be found directly
with the same state vectors of the tree system. from the library of transfer matrices in [19, 30].
There are boundary conditions of the system
The theorem above is effective for various multi-rigid-body
𝑇
systems, chain multi-rigid-flexible-body systems, and any z1,0 = [𝑥̈𝑦̈ 𝜃𝑧̈ 0 0 0 1]1,0 ,
closed-loop multi-rigid-flexible-body systems and is effective
for various tree multi-rigid-flexible-body systems and general 𝑇
z6,0 = [0 0 0 𝑚𝑧 𝑞𝑥 𝑞𝑦 1]6,0 , (44)
multi-rigid-flexible-body systems if the bodies with more
than two ends are rigid bodies. For more general systems 𝑇
z8,0 = [𝑥̈𝑦̈𝜃𝑧̈ 0 0 0 1]8,0 .
including the flexible bodies with more than two ends, the
theorem to deduce automatically overall transfer equation is
The system experiences a step upward force at the mass
undergoing study and will be discussed in another paper.
center of body element 1 at time instant zero, while the
initial displacement and velocity of the whole system are zero.
6. Numerical Examples The computational results of the system dynamics obtained
by the proposal method and by Newton-Euler method are
By comparison with Newton-Euler method, the numerical shown in Figure 11. It can be seen from Figure 11 that the
examples here are carried out to validate the proposed computational results obtained by the above two methods
method. have good agreements.
Example 1. A tree multi-rigid-flexible-body system moving Example 2. A multi-rigid-body system containing a close
in plane, as shown in Figure 9, consists of two fixed hinges, 2 loop moving in plane is shown in Figure 12. The planar rigid
and 4, two elastic hinges, 6 and 7, three rigid bodies, 1, 5, and 8, bodies (numbered 1, 3, 5, 7, 9, 11, and 13) are connected by pin
and one uniform beam element, 3, with three boundary ends. hinges (numbered 2, 4, 6, 8, 10, 12, and 14), and rigid body
The simulation parameters are given as follows: 𝑚1 = 𝑚5 = 1 is connected with the ground by a smooth pin. Each rigid
𝑚8 = 7.8 kg, 𝐽𝐶,1 = 𝐽𝐶,5 = 𝐽𝐶,8 = 0.013 kg ⋅ m2 , 𝑙3 = 3 m, body has the identical dynamics parameter as 𝑚 = 1 kg and
𝐸𝐴 3 = 1000 N, 𝑚3 = 0.78 kg/m, 𝐸𝐼3 = 166.67 N ⋅ m2 , 𝐾𝑥,6 = 𝐽𝐶 = (1/6) kg ⋅ m2 .
10 Advances in Mechanical Engineering
0
0.2
1
0.15
Rotation angle (rad)
3
0.1
4 10
5 11
0.05
6 12
7 13
0
0 1 2 3 4 5 8 14
9
t (s)
Figure 12: A general system moving in plane.
MSTMM
Newton-Euler method
0 𝒛9,0
Figure 11: Computational results of the angle of the right end of the 𝒛14,0
beam. 9 0
8 14
Uall zall = 0, 4 10
(45)
I1 3 I2
where the overall transfer matrix takes the form: 2
−I T9−1 T14−1 1
Uall = [ O G9−3 G14−3 ] . (52) 0 𝒛1,0
[O C −I ]
Figure 13: Topology figure of the system.
State vectors of all boundary ends are
z1,0
zall = [ z9,0 ] . (46) The initial angle of rigid body 1 is (𝜋/6) rad and the
relative angles of pin hinges (numbered 2, 4, 6, 8, 10,
[z14,0 ] 12, and 14) are all zero. The system moves from the rest
According to the proposed sign conventions, one can acquire under the effect of gravity. The computational results of the
system dynamics are obtained by the proposal method and
T9−1 = U1 U2 U3 U4 U5 U6 U7 U8 U9 , by Newton-Euler method. The time history of rigid body
1’s angle is exhibited in Figure 14, which shows that the
T14−1 = U1 U2 U3,10 U10 U11 U12 U13 U14 , computational results obtained by the above two methods
(47)
G9−3 = −H3 U4 U5 U6 U7 U8 U9 , have good agreements.
Research Article
Optimal Design of One-Folded Leaf Spring with
High Fatigue Life Applied to Horizontally Vibrating
Linear Actuator in Smart Phone
Received 4 July 2013; Revised 12 November 2013; Accepted 25 November 2013; Published 24 February 2014
Copyright © 2014 Ki Bum Lee et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Horizontally vibrating linear actuator (HVLA) instead of VVLA has been under study in a few past years and recently HVLA
with thickness of 2.5 mm was developed. The one-folded leaf spring to guide the moving part is newly designed and applied in
HVLA, but unfortunately it makes HVLA be wider. Accordingly, this paper presents the optimal design of one-folded leaf spring,
which results in reduction of HLVA width. The commercial design optimization tool “PIAnO” was utilized based on design of
experiments (DOE), approximation techniques, and optimization algorithm. In addition, for the vibration modal analysis and
harmonic response analysis to generate metamodeling, the software “ANSYS” is utilized. The optimal width of leaf spring was
reduced by 46% compared to the initial one, while all the design constraints were satisfied, which clearly showed the validity of the
proposed design approach.
1. Introduction part of the actuator and enable the actuator to vibrate elas-
tically. Accordingly, the various designs of guide spring had
The slimming of a smart phone has become a competitive been presented to achieve the high fatigue life [3, 4] and
issue in the smart phone industry because the display panel is among them the one-folded leaf spring introduced by Lee and
enlarged to show more information to the user. A smart Kim has the highest fatigue life [5]. Figures 2(a) and 2(b) show
phone is comprised of numerous components: the battery, the the one-folded spring and the schematic diagrams of the
camera module, the vibrating actuator, and so forth. One of HVLA which has the one-folded spring, respectively. Table 1
the thickest components is vibrating actuator which imposes shows the performance characteristics of the HVLA.
the restrictions on the slimming of smart phones. In currently This one-folded leaf spring enables the HVLA to improve
manufactured smart phone in the market, a vertical vibrating the vibration force and product life. Instead, the HVLA has
linear actuator (VVLA) is being widely used as a vibration been enlarged in width due to the wide structure of one-
motor, as shown in Figure 1. folded spring. Figure 3 shows the top view of the HVLA with
However, it imposes the thickness problem on a smart one-folded leaf spring.
phone, because it needs vibration space in the vertical direc- This one-folded leaf spring has occupied 45% of total
tion [1]. Therefore, a horizontally vibrating linear actuator width of HVLA. Therefore, the design of one-folded leaf
(HVLA) has been developed in recent years, but it has not spring needs be optimized to reduce its width. Several
been commercialized in the smart phone industry so far [2]. researches about the optimal design of the mechanical springs
The primary reason is that the fatigue fracture of guide spring such as helical, coil, and leaf spring had been reported in
due to the cyclic compression and tension loads has not been past decades. These springs are usually designed to avoid
resolved. The guide spring has the role to support the moving resonance, but one-folded leaf spring of HVLA needs to
2 Advances in Mechanical Engineering
Case
2.1.2. Design Constraint on the Maximum Stress. A one-
Spring folded spring that is affected by cyclic compression and ten-
sion loads should have a yield stress of approximately 215 MPa
Yoke
in order to maintain its fatigue life [7]. The one-folded spring
is represented as follows:
Magnet
𝜎max < 𝜎yield . (2)
Coil
2.1.3. Design Constraint on the Mechanical Natural Frequency.
The mechanical natural frequency of HVLA ranges from
175 Hz to 185.5 Hz:
Figure 1: Schematic diagram of VVLA.
175 ≤ 𝑓 ≤ 185.5. (3)
15 mm
m
m
.5
16
Case
Leaf spring
Housing
Weight
Magnets Yoke
Coil
PCB
Bracket
(a) (b)
Figure 2: (a) The one-folded spring of HVLA and (b) schematic diagram of HVLA.
Table 2: Initial, lower, and upper bound values of the selected design variables.
Width Width W
of spring of spring
Length L
of weight
R
H
Figure 3: Top view of HVLA with the one-folded leaf spring. saturated points is the minimum required number of sam-
pling points to generate a full quadratic polynomial model.
An orthogonal array with the main effect and interaction of
according to each design variable. 𝐿 98 (715 ) was selected each factor was represented by a table, which allows the
by considering the number of saturated points and levels design of experiments to be easily established [8]. An orthog-
according to the number of design variables. The number of onal array, regardless of the lack of theoretical background,
4 Advances in Mechanical Engineering
(1) Design of
Extract experiments
experiments
(2) ANSYS
PIAnO
(modal and harmonic response)
(3) Approximation Generate meta models
(4) Design
optimization Derive the optimal
148
Amplitude (MPa)
117
93.1
58.8
46.7
37.1
23.4
174 175 178 180 183 185 188 190 193
Frequency (Hz)
189 200
188
First frequency (Hz)
150
187
Stress (MPa)
186 100
185
50
184
183 0
1.79 2.2425 2.695 3.1475 3.6 1.79 2.2425 2.695 3.1475 3.6
x1 width x1 width
The stiffness of a cantilever beam is directly proportional response analyses. Metamodels were regenerated with respect
to the cube of the thickness. Therefore, the thickness of the to the width of the one-folded spring, maximum stress, and
leaf spring has a nonlinearity, so we selected the Kriging natural frequency of the first mode. At this point, we added
model, which is one of the metamodels provided in the only 39 types of sampling points to satisfy the 15 multiples
PIAnO software that was selected. The Kriging model, which of the number of design variables corresponding to the
a type of interpolation model, was mathematically estab- sampling points (75). We used augmented Latin hypercube
lished by Metheron in 1963 based on research conducted design (ALHD) that is not overlapped with the existing
by mining researcher Krige in 1951 [13, 14]. And it shows sampling points [16]. Also, it is widely known for excellent
superb predictive performance under many design variables performance of the space filling. Nevertheless, the accuracy of
2
and is in strongly nonlinear systems, provides a statistical 𝑅pred was only about 80%. Therefore, 98 sampling points were
estimation [15]. Also, there are no parameters that depend on reextracted in order to obtain higher accuracy over 90% using
the experience and intuition of customers when choosing the orthogonal array 𝐿 98 (715 ) at a time. Finally, in the evaluation
design parameters because the Kriging model can optimize results of the metamodel, the accuracy of 𝑅pred 2
is over 90%.
design parameters through maximum likelihood estimation
(MLE). Therefore, a recent trend is an increase in the use of
Kriging models in the field of engineering. Our first Kriging 3.5. Optimization Technique. We selected an evolution algo-
model was generated by using an orthogonal array 𝐿 36 (313 ), rithm that is provided in PIAnO. The evolution algorithm
which was based on the analytical results of the width of (EA) was proposed by Holland in 1975 and is a global
the one-folded spring, the maximum stress, and the natural optimization technique [17]. This algorithm describes the
frequency. However, an accuracy evaluation of the Kriging evolution of the biological genetic trait, so it searches for a
2
model using 𝑅pred , which corresponds to the maximum stress, global optimum through a process of selection, recombina-
was underestimated 55% as accuracy. Consequently, we tion, and mutation. On the downside, it requires consider-
added 39 types of sampling points to improve the accuracy of able computation time, depending on the analytical model,
the metamodel and then performed the modal and harmonic because more calculation functions are required compared to
6 Advances in Mechanical Engineering
210
Violation area
200
200
Violation area
Maximum stress (MPa)
170
120 Violation area
100 160
Initial Opt meta Opt exact Initial Opt meta Opt exact
(a) Maximum stress at bending area on leaf spring (b) Natural frequency of first mode
3
Width (mm)
3.6. Results. In our results for the optimal design using the
2.5
metamodel (Opt meta), the optimal design satisfied the con-
straints, including the maximum stress and natural frequency 1.9
2 1.9
of first mode, and the width of one-folded leaf spring
decreased by 47% compared to the initial width of 1.9 mm. 1.5
However, the optimal design results can be changed based on
using the metamodel instead of the actual analytical model in 1
this research. The accuracy of the optimization results should Initial Opt meta Opt exact
be verified by actual analysis using ANSYS. To do this, the Figure 11: Comparison of the objective function.
Kriging model results (Opt meta) of the optimal design vari-
ables and the analysis results from ANSYS (Opt exact) were
compared as shown in Figures 10(a) and 10(b). 15 mm 12.2 mm
The Kriging model results (Opt meta) and the ANSYS
model results (Opt exact) were very similar; therefore, we
confirmed the high accuracy of the Kriging model’s predic-
tion. The initial and optimal values of the design variables
16.5 mm
4. Conclusion
We minimized the existing size of a HVLA (16.5 × 15 × 2.5) by
decreasing the width of the leaf spring. The following conclu- Figure 12: Comparison of HLVA according to the initial design and
sions were drawn from our results. optimal design.
Advances in Mechanical Engineering 7
References
[1] J.-H. Kim, H.-K. Il, J.-W. Lee, and J.-H. Yoo, Horizontally Vibra-
tion Motor for Slimming Mobile Smart Phone, vol. 6, Korean
Society for Precision Engineering, 2011.
[2] J.-H. Kim and J.-H. Kim, “A horizontally vibration linear actua-
tor for slimming of smart phone,” in Proceedings of the Interna-
tional Conference on Computer, Electrical and Systems Confer-
ence (ICCESSE ’11), Phuket, Thailand, December 2011.
[3] JA HWA ELECTRONICS CO., LTD., Linear type vibration
motor of horizontal vibration capable of generating vibration to
horizontal direction by linearly reciprocating mover, KOEAR
Patent, 1007201970000 (20070514).
[4] SAMSUNG ELECTRO-MECHANICS CO., LTD, Linear vibra-
tion motor capable of minimizing a friction area with adja-
cent components when a motor is driven, KOREA Patent,
1010595990000 (20110819).
[5] K.-B. Lee and J.-H. Kim, “Design of leaf spring with high fatigue
life applied to horizontal linear vibrating actuator,” The Korea
Academia-Industrial Cooperation Society, vol. 13, no. 12, pp.
5684–5688, 1975.
[6] PIDOTECH Inc, Process Intergration, Automation, and Opti-
mization, PIAnO User’s Manual, 2010.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 512583, 9 pages
http://dx.doi.org/10.1155/2014/512583
Research Article
Adaptive Terminal Sliding Mode Control of Electromagnetic
Spacecraft Formation Flying in Near-Earth Orbits
Received 25 August 2013; Revised 17 November 2013; Accepted 9 December 2013; Published 11 February 2014
Copyright © 2014 Jingrui Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
An adaptive terminal sliding mode control for six-degree-of-freedom electromagnetic spacecraft formation flying (EMFF) in near-
Earth orbits is presented. By using terminal sliding mode (TSM) technique, the output tracking error can converge to zero in
finite time, and strong robustness with respect to disturbance forces can be guaranteed. Based on a rotated frame 𝐹𝑟 and the
adaptive TSM controller, the special magnetic moment of the steerable magnetic dipole is computed. The angular momentum
management strategy (AMM) is implemented in a periodically switching fashion, by which the angular momentum buildup was
limited. Illustrative simulations of EMFF are conducted to verify the effectiveness of the proposed controller.
1. Introduction forces on the order of 10-9 N/W may be generated using elec-
tromagnetic gradient forces or scattering forces; microwave
Spacecraft formation flying (SFF) represents the concept beam powers of 10 kW can thus produce restoring forces of
of distributing the functionality of large spacecraft among approximately 10 𝜇N, which are sufficient to correct a number
smaller, less-expensive, cooperative spacecraft [1, 2]. Specif- of orbital perturbations. Miller et al. [10, 11] address the
ically, NASA and the U.S. Air Force have identified space- novel concept of electromagnetic formation flying (EMFF) in
craft formation flying as an enabling technology for future which high temperature superconducting (HTS) wire tech-
missions. The practical implementation of the SFF concept nology is used to create magnetic dipoles on each spacecraft
relies on the accurate control of the relative positions and ori- that can be used to maintain and reconfigure the spacecraft
entations between the participating spacecraft for formation formation. However, Since magnetic force on each spacecraft
configuration. The conventional thruster-based schemes may in the formation can be applied in any arbitrary direction
require continuous expenditure of fuel to maintain formation which can be easily created by steerable magnetic dipoles,
geometry that can contaminate the sensitive sensors on board EMFF has advantages in terms of controllability [4].
and mission lifetime also becomes dependent on the fuel A critical component of EMFF is an effective formation
available [3, 4]. flying control. The dynamics and control problem associated
To alleviate these concerns, several propellant-free for- with EMFF become highly challenging, due to the nonlinear
mation flying methods have been proposed in the literature. nature of the magnetic forces. Ahsun and Miller [3] have
The propulsive conducting tethers and spin-stabilized tether presented a hybrid adaptive control scheme in which trans-
systems have been proposed in place of on-board propulsion lation control is implemented in a centralized fashion with
systems to form and maintain satellite formations [5, 6]. a decentralized attitude control. Elias et al. [12] designed a
King et al. [7] have presented Coulomb force approaches linear optimal controller based on the linearized dynamics.
to maintain a formation. The flux pinning technology has Kong et al. [13] addressed the use of electromagnetic dipoles
been applied to achieve passively stable configurations by for relative position and orientation maintenance as needed
HTS electromagnetics [8]. LaPointe [9] has presented the for the terrestrial planet finder (TPF). Reference [14] derived
microwave scattering formation flight method. Radiation the dynamics of an N-spacecraft EMFF (in 2D) for deep
2 Advances in Mechanical Engineering
By approximating the coils on each spacecraft to SMD, the where 𝜏𝑚𝑖𝑗 is the torque exerted on the SMD of 𝑖th spacecraft
magnetic field due to the 𝑗th spacecraft can be written as [3, due to that of 𝑗th spacecraft [3, 10],
10]
𝜇0 𝜇𝑖 3r𝑖𝑗 (𝜇𝑗 ⋅ r𝑖𝑗 ) 𝜇𝑗
𝜇 3𝜇𝑗 ⋅ r𝑖𝑗 𝜇𝑗 𝜏𝑚
𝑖𝑗 = ×( − 3) (13)
B𝑗 (r𝑖𝑗 ) = 0 ( 5
r𝑖𝑗 − 3 ) (6) 4𝜋 𝑟𝑖𝑗5 𝑟𝑖𝑗
4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗
𝜏𝑐𝑤𝑖 and 𝜏𝑑𝑖 are the control and disturbance torques, respec-
and the force between spacecrafts can be written as
tively. 𝜏𝑑𝑖 includes the gravity gradient torques 𝜏𝑔𝑖 and the
F𝑚 Earth magnetic torques 𝜏𝑒𝑖 . Consider the following:
𝑖𝑗 ≜ 𝑓 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 )
𝜏𝑑𝑖 = 𝜏𝑔𝑖 + 𝜏𝑒𝑖 . (14)
3𝜇 𝜇𝑖 ⋅ 𝜇𝑗 𝜇𝑖 ⋅ r𝑖𝑗 𝜇𝑖 ⋅ r𝑖𝑗
= 0 (− 5
r𝑖𝑗 − 5
𝜇𝑗 − 𝜇𝑖 A very important factor presented in the disturbance
4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗 𝑟𝑖𝑗5 (7) torque term is the torque that acts on the spacecraft due to
the Earth’s magnetic field. This torque can be written as [4]
(𝜇𝑖 ⋅ r𝑖𝑗 ) (𝜇𝑗 ⋅ r𝑖𝑗 )
+5 r𝑖𝑗 ) , 𝜏𝑒𝑖 = 𝜇𝑖 × B𝑒 (R𝑖 ) , (15)
𝑟𝑖𝑗7
where B𝑒 is Earth’s magnetic field strength at the location of
where 𝜇0 = 4𝜋 × 10−7 T⋅m/A is the permeability constant. 𝜇𝑖 the spacecraft.
and 𝜇𝑗 are the dipole strength of the 𝑖th satellite and the 𝑗th Substituting (10) into (11), the attitude dynamics can be
satellite, respectively. Note that (7) gives the force on dipole written compactly as
𝑖 (located on 𝑖th spacecraft) due to dipole 𝑗 (located on 𝑗th
spacecraft), which depends on the distance between the two M𝑖𝐴 (J𝑖 , q𝑖 ) 𝜀̈
𝑖 + C𝑖𝐴 (J𝑖 , Ω𝑖 , h) + N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 )
dipoles and the orientation of both dipoles in the inertial = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,
space. It is the dependence on the orientation of the dipoles (16)
that gives rise to the complexity of the expression for the force,
since the orientation of a dipole obviously depends on the where
orientation of the body axes in the inertial space [3].
M𝑖𝐴 (J𝑖 , q𝑖 ) = 2J𝑖 T𝑖 −1 ,
2.2. Attitude Dynamics. To avoid singular points, the Euler 𝑇
C𝑖𝐴 (J𝑖 , Ω𝑖 , h) = Ω×𝑖 (J𝑖 Ω𝑖 + h) , X𝑖 = [𝜌𝑇𝑖 , 𝜀𝑇𝑖 ] ,
parameter is chosen to describe the attitude of the spacecraft.
Let q𝑖 represent the Euler parameters corresponding to the
N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 ) = 2J𝑖 T𝑖 −1 𝑓𝑖 , 𝑓𝑖 = −𝜔×𝑏𝑖 𝜔𝑂𝑖 + 𝜔̇𝑂𝑖 .
attitude of the body frame relative to the orbital frame 𝐹𝑂𝑖 (17)
[19],
𝑇
q𝑖 = [𝜀𝑇𝑖 𝜂𝑖 ] ∈ R4 , 2.3. Combined Attitude and Translational Dynamics. Com-
(8) bining the attitude dynamics in (1) and the translational
𝑇
𝜀𝑖 ≡ [𝜀𝑖1 𝜀𝑖2 𝜀𝑖3 ] ∈ R3×1 , 𝜂𝑖 ∈ R1×1 . dynamics in (16), the following 6-DOF dynamics equation for
formation flying is obtained:
The Euler parameters, which are equivalent to the coeffi-
cients of unit quaternion, have unit norm by definition; hence M𝑖 Ẍ
𝑖 + C𝑖 + N𝑖 = u𝑖 + D𝑖 , (18)
where
𝜀𝑇𝑖 𝜀𝑖 + 𝜂𝑖2 = 1. (9)
I3×3 0 C
Attitude kinematics and dynamics of the spacecraft are M𝑖 = [ ] ∈ R6×6 , C𝑖 = [ 𝑖𝑇 ] ∈ R6×1 ,
0 M𝑖𝐴 C𝑖𝐴
governed by [19]
N𝑖𝑇 F̂
Ω𝑖 = 2T𝑖 −1 𝜀̇𝑖 + 𝜔𝑂𝑖 , (10) N𝑖 = [ ] ∈ R6×1 , u𝑖 = [ 𝑐𝑖 ] ,
(19)
N𝑖𝐴 𝜏𝑐𝑤𝑖
J𝑖 Ω̇𝑖 + Ω×𝑖 (J𝑖 Ω𝑖 + h) = 𝜏𝑚𝑖 + 𝜏𝑐𝑤𝑖 + 𝜏𝑑𝑖 , (11)
𝑇 F̂
D𝑖 ≜ [𝑑1𝑖 𝑑2𝑖 𝑑3𝑖 𝑑4𝑖 𝑑5𝑖 𝑑6𝑖 ] = [ 𝑑𝑖 ].
where T𝑖 ≡ 𝜂𝑖 I + 𝜀×𝑖 3×3
∈ R , the superscript “×” denotes 𝜏𝑚𝑖 + 𝜏𝑑𝑖
skew-symmetric matrix of a vector, and Ω𝑖 = 𝜔𝑂𝑖 + 𝜔𝑏𝑖 , 𝜔𝑏𝑖 is
angular velocity of the spacecraft relative to the orbital frame 3. Adaptive Terminal Sliding
𝐹𝑂𝑖 . 𝜔𝑂𝑖 is the orbital angular velocity of 𝑖th spacecraft. J𝑖 Mode Control Design
is the moment of inertia of the spacecraft. h is the angular
momentum of the flywheel. The magnetic torque term in (11) In this section, an adaptive terminal sliding mode controller
can be written as is designed for the follower spacecraft based on the dynamic
𝑁−1 model in (18) and the TSM technique. With this controller,
𝜏𝑚𝑖 = ∑ 𝜏𝑚
𝑖𝑗 , (12) the follower spacecraft can track the desired attitude and
𝑗=0, 𝑖 ≠
𝑗 relative position trajectories simultaneously.
4 Advances in Mechanical Engineering
3.1. Error Dynamics Equation. The trajectory tracking errors Substituting (24) and (25) into (11), the error attitude
of the follower spacecraft are defined as dynamics are derived as follows [19]:
where 𝜌𝑑𝑖 , 𝜌̇𝑑𝑖 ∈ R3×1 are the relative position/velocity of the = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,
desired trajectory with respect to the leader. Differentiating (27)
(20) and substituting it into (1), one can obtain that −1 −1
where N𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 f𝑖𝑒 , M𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 .
𝑇
𝑒
𝜌̈ 𝑒 𝑑 ̂ ̂ 𝑇 𝑇
𝑖 + C𝑖𝑇 (𝜔𝑂, 𝜌̇
𝑖 ) + N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 , 𝜌̈
𝑖 ) = F𝑑𝑖 + F𝑐𝑖 , Let e𝑖 = [𝜌𝑒𝑖 𝜀𝑒𝑖 ] , and utilizing (18), (21), and (27), the
(21) error dynamics of the whole system can be obtained
𝑇
sign (S𝑖 ) ≜ [sign (𝑠1𝑖 ) sign (𝑠2𝑖 ) sign (𝑠3𝑖 ) sign (𝑠4𝑖 ) sign (𝑠5𝑖 ) sign (𝑠6𝑖 )] . (33)
Substituting (32) into (18) produces the closed-loop dynamics Consider a Lyapunov function as follows:
𝛾−1
𝑖 = −𝛼𝑖 ė
ë 𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 − diag (̂
𝜎𝑖 ) sign (S𝑖 ) + D𝑖 . 1 1 𝑇
𝑉 = S𝑇𝑖 S𝑖 + 𝜎
̃ diag (𝜗𝑖 ) 𝜎
̃𝑖, (35)
(34) 2 2 𝑖
Advances in Mechanical Engineering 5
𝑇
where 𝜗𝑖 ≜ [𝜗1𝑖 𝜗2𝑖 𝜗3𝑖 𝜗4𝑖 𝜗5𝑖 𝜗6𝑖 ] . The estima- It means that the switching planes 𝑠𝑛𝑖 (𝑛 = 1, 2, . . . , 6)
tion error is defined as 𝜎 ̃𝑖 = 𝜎 ̂ 𝑖 − 𝜎∗𝑖 , 𝜎∗𝑖 ≜ converge to zero [20]. On the other hand, in the TSM 𝑠𝑛𝑖 = 0
∗ ∗ ∗ ∗ ∗ ∗ 𝑇
[𝜎1𝑖 , 𝜎2𝑖 , 𝜎3𝑖 , 𝜎4𝑖 , 𝜎5𝑖 , 𝜎6𝑖 ] . (𝑛 = 1, 2, . . . , 6), the system state will reach zero in finite time
[17].
The first derivative of (30) can be expressed as
Remark 1. For the purpose of eliminating chattering, a
Ṡ𝑖 = ë
𝑖 + 𝛼𝑖 ė
𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖
𝛾−1
) ė𝑖 . (36) common practice is to replace the signum function of (32)
with a continuous saturation function
Differentiating (35) and utilizing (36), one can obtain that sat (𝑠𝑛𝑖 , 𝜉𝑛𝑖 )
𝑠𝑛𝑖
{
{ if 𝑠𝑛𝑖 < 𝜉𝑛𝑖 (39)
𝑉̇= S𝑇𝑖 Ṡ𝑖 + ̃̇𝑖
̃ 𝑇𝑖 diag (𝜗𝑖 ) 𝜎
𝜎 𝜉𝑛𝑖
={ 𝑛 = 1, 2, . . . , 6,
(37) {
= S𝑇𝑖 (ë
𝑖 + 𝛼𝑖 ė
𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖
𝛾−1
̃ 𝑇𝑖 S𝑖 ,
) ė𝑖 ) + 𝜎 { sign (𝑠𝑛𝑖 ) if 𝑠
𝑛𝑖 ≥ 𝜉𝑛𝑖 ,
𝑇
sat (S𝑖 , 𝜉𝑖 ) = [sat(𝑠1𝑖 , 𝜉1𝑖 ) sat(𝑠2𝑖 , 𝜉2𝑖 ) sat(𝑠3𝑖 , 𝜉3𝑖 ) sat(𝑠4𝑖 , 𝜉4𝑖 ) sat(𝑠5𝑖 , 𝜉5𝑖 ) sat (𝑠6𝑖 , 𝜉6𝑖 )] . (41)
Remark 2. There exists a possible singularity in sliding mode Substituting (5) and (7) into (44) produces
controller as e𝑖 → 0. Since e𝑖 = 0 only approaches along a
sliding mode, we observe that for a general choice of 𝛾𝑖 𝑁−1 𝑚𝑖 𝑚𝑗
∑ 𝑓𝑖𝑗 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = u𝑖𝑇 . (45)
𝑗=0, 𝑗 ≠
𝑖 𝑚𝑖 + 𝑚𝑗
𝛾
ė𝑖 = −𝛼𝑖 e𝑖 − 𝛽𝑖 diag (e𝑖 ) (42)
Note that (7) gives the force on the 𝑖th SMD due to the 𝑗th
while sliding and that the component in (32) SMD and it depends on the distance between the two SMDs
and the orientation of both SMDs in the inertial space. It rises
𝛾−1 𝛾 2𝛾𝑖 −1 to the complexity of the expression for the magnetic. Here, we
𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 → −𝛾𝑖 𝛽𝑖 (𝛼𝑖 diag (e𝑖 𝑖 ) + 𝛽𝑖 diag (e𝑖 )) .
consider a two-spacecraft electromagnetic formation flying
(43) (EMFF) array. By defining a rotated frame 𝐹𝑟 (see Section 2),
a simplified algebraic form of (7) is obtained
Consequently there will be a singularity in (32) unless 𝛾𝑖 is
chosen so that 2𝛾𝑖 > 1. To satisfy this requirement we set 𝛾𝑖 = 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧
3/5 for the examples to follow. 3𝜇0 [ ]
𝑓𝑖𝑗𝑟 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] . (46)
4𝜋𝑟𝑖𝑗4
[ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]
3.3. Compute the Special Magnetic Moments of SMD. The
adaptive sliding mode controller u𝑖𝑇 gives the desired special Utilizing (45) and (46) produces that
forces that can be used as input for a thruster-based system.
However, for EMFF, the desired special forces are produced 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧 4
by SMD. The SMD of each spacecraft is a complicated func- [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] 4𝜋𝑟𝑖𝑗 𝑚𝑖 𝑚𝑗
[ ]= A A u ,
tion of current position and orientation [3, 10]. For EMFF, 3𝜇0 𝑚𝑖 + 𝑚𝑗 𝑟𝐼𝑖 𝐼𝑏𝑖 𝑖𝑇
the control variables consist of the dipoles of individual [ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]
spacecraft. Utilizing (32) and (4), one can obtain that (47)
Spacecraft Semimajor axis Inclination Right ascension of Argument of Eccentricity Mean anomaly
ascending node perigee
LS 6878 km 1.106538745764 rad 0.523598775598 rad 0 rad 0.003 0 rad
FS 6878 km 1.106538745764 rad 0.523598071512 rad 0.000000315261 rad 0.002999636522 0 rad
the transformation matrix from the body frame 𝐹𝐵𝑖 of 𝑖th on a unique nonlinearity of the magnetic torques [3, 4]. The
spacecraft to ECI frame. force acting between a pair of dipoles depends on the product
𝑟 𝑟 𝑟 𝑟 𝑇 4
Let ŭ𝑖𝑇 ≜ [𝑢̆ 𝑖𝑇𝑥 , 𝑢̆
𝑖𝑇𝑦 , 𝑢̆
𝑖𝑇𝑧 ] = (4𝜋𝑟12 /3𝜇0 )(𝑚1 𝑚2 /(𝑚1 +
of the individual dipole values. By switching the polarity of
𝑚2 ))A𝑟𝐼𝑖 A𝐼𝑏𝑖 u𝑖𝑇 , 𝜇𝑖 = −𝜇𝑗 , and utilizing (47), one can obtain all dipoles in the EMFF, the torque acting on the spacecraft
that due to Earth’s magnetic field changes sign, but the torques
and forces due to the other spacecraft in the system do not.
1 As can be seen from (15), it results in a net cancellation
𝜇𝑖𝑥 = √ √𝑢̆
𝑟2 𝑟2
𝑖𝑇𝑥 + 2𝑢̆
𝑟2
𝑖𝑇𝑦 + 2𝑢̆
𝑟2
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥 ,
2 of the effect of the Earth’s magnetic field on the average
𝑟2
sense.
𝑢̆
𝑖𝑇𝑦
𝜇𝑖𝑦 = ,
𝑟2
√ √𝑢̆ 𝑟2 𝑟2 𝑟2
𝑖𝑇𝑥 + 2𝑢̆
𝑖𝑇𝑦 + 2𝑢̆
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥
(48) 4. Numerical Simulation
𝑟2 The adaptive terminal sliding mode controller equation (40)
𝑢̆
𝑖𝑇𝑧
𝜇𝑖𝑧 = . was simulated for a two-spacecraft formation flying control.
𝑟2
√ √𝑢̆ 𝑟2 𝑟2 𝑟2 Considering the nonlinear dynamics with disturbance D1 ,
𝑖𝑇𝑥 + 2𝑢̆
𝑖𝑇𝑦 + 2𝑢̆
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥
the effectiveness of the proposed controller was verified.
The orbital parameters of leader and follower spacecraft
The special magnetic moments of SMD can be computed in the simulations are listed in Table 1. The parameters of
using (48), by which the follower spacecraft can track the spacecraft are summarized in Table 2. The dynamics includes
desired relative position trajectories. the second harmonic of the gravitational field as well as
the Earth’s magnetic field, of which the vector is calculated
3.4. Angular Momentum Management. For EMFF in near- using the World Magnetic (WMM2005)Model block in the
Earth orbits, a constant disturbance torque may act on each paper.
spacecraft due to the Earth’s magnetic field that causes the The follower was commanded to move around the leader
reaction wheels on each spacecraft to quickly become satu- in an elliptic orbit. The desired trajectory was generated by
rated. In order to avoid this situation, the angular momentum solving nonlinear equation (1) numerically (set F̂ ̂
𝑑1 and F𝑐1
management strategy herein can be utilized, which is based equal to zero) with the following initial condition:
3
The parameters of the adaptive sliding mode controller are 𝛾1 = , 𝜉 = [5, 5, 5, 1, 1, 1] × 10−3 ,
then 5
𝛼1 = diag (2, 2, 3, 250, 80, 80) × 10−4 , 𝜗1 = [15, 22, 22, 80, 125, 125] .
(51)
𝛽1 = diag (10, 15, 15, 1200, 800, 800) × 10−5 ,
Advances in Mechanical Engineering 7
(deg)
0
Ls 100 kg Diag (19, 19, 32) kg⋅m2
FS 100 kg Diag (19, 19, 32) kg⋅m2 −0.2
0 1000 2000 3000 4000 5000 6000
(a)
Relative trajectory rou
Error (𝜃2 )
5 0.2
(deg)
0
z (m)
0 −0.2
∗ 0 1000 2000 3000 4000 5000 6000
(b)
−5
10 Error (𝜃3 )
5 4 0.5
0 2
0
(deg)
y (m −5 −2 ) 0
) −10 −4 x (m
−0.5
Figure 2: Actual trajectory 𝜌1 (𝑡) of the follower spacecraft (∗ the 0 1000 2000 3000 4000 5000 6000
leader spacecraft.) Time (s)
(c)
Error (x-direction)
0.1 Figure 4: Attitude tracking error.
(m)
0.05
0
0 1000 2000 3000 4000 5000
(a) 𝜇x (Am2 )
1000
Error (y-direction) 0
0.8 −1000
0.6 −2000
(m)
hx Conflict of Interests
5
(Nms)
−2 References
0 1000 2000 3000 4000 5000 6000
[1] J. Bristow, D. Folta, and K. Hartman, “A formation flying
(b)
technology vision,” in Proceedings of the AIAA Space Conference
hz and Exposition, Long Beach, Calif, USA, September 2000,
5
AIAA-2000-5194.
(Nms)
Research Article
Application of Krylov Reduction Technique for a Machine Tool
Multibody Modelling
Copyright © 2014 M. Sulitka et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Quick calculation of machine tool dynamic response represents one of the major requirements for machine tool virtual modelling
and virtual machining, aiming at simulating the machining process performance, quality, and precision of a workpiece. Enhanced
time effectiveness in machine tool dynamic simulations may be achieved by employing model order reduction (MOR) techniques
of the full finite element (FE) models. The paper provides a case study aimed at comparison of Krylov subspace base and mode
truncation technique. Application of both of the reduction techniques for creating a machine tool multibody model is evaluated.
The Krylov subspace reduction technique shows high quality in terms of both dynamic properties of the reduced multibody model
and very low time demands at the same time.
developed in [12]. Another approach employing an addition- The Krylov subspace methods [26, 29–31] are very
ally created FE node moving between the end positions of the interesting because of their iterative nature which allows
motion axes stroke has been described in [13]. the reduction of large-scale problems. The computational
The finite element method (FEM) is used as a common efficiency of Krylov subspace based MOR has encouraged
way to simulate the dynamic behavior of a machine tool. wide interest in the method, and therefore wide knowledge
The solution of FE problems usually involves the solution in different fields is available. We will mention the most
of a large set of sparse algebraic equations. The number important observations to date. The passivity and stability
of unknowns in ordinary engineering FE model nowadays preservation has been achieved using the Krylov MOR
is commonly 106 -107 . The computational time needed to methods in [26, 32]. A Krylov algorithm preserving structure
solve static analysis of such problems on current hardware is of second order ordinary differential equations has been pre-
acceptably low (∼hours). The issue arises when the harmonic sented in [33]. Handling of nonlinear convection coefficient
or transient analysis is required. The solution then requires was studied in [34]. Reduction of coupled physics problems
10 s or 100 s of iterations and therefore requires days or even was studied in [35] for the case of a thermomechanical model
weeks to solve. Such long solution time effectively hinders of packages and in [36] for the case of structural-acoustic
this kind of simulations in machine tool virtual model coupled models. Krylov subspace MOR was also successfully
simulations. The model order reduction (MOR) methods are used in optimization of MEMS devices [37] and sensitivity
one of possible ways to speed up the solution and make analysis of structural frequency response [38]. One of the
it feasible in practice. The idea behind MOR is to reduce most important directions in development on the Krylov base
the number of unknowns while producing sufficiently good reductions is a parametric model order reduction (PMOR).
approximation to the input/output behavior. The PMOR allows preservation of parameters which the
The following text gives a basic review of model order system depends on [39, 40]. The dependence of parameters
reduction techniques. The idea of reduction of the number may be either linear or nonlinear.
of unknowns during the solution of the FE model is almost The comparison of different model order reduction meth-
as old as the FE method itself. The first MOR method was ods has been discussed in [17], where the Krylov subspace
MOR method was found to be one of the best methods. The
static reduction proposed by Guyan [14] and Irons [15]. This
comparison of Krylov, CMS, and balanced truncation can be
method was introduced for structural mechanics problems
found in [41].
but it is also valid for thermal analysis and other analyses
Based on the findings presented in previous works the
regardless of the underlying physics. However, this method
Krylov subspace based MOR is very robust and computation-
is of questionable quality when using it for dynamic analyses
ally efficient. The goal of this study is to show a new strategy
as was shown in [16, 17].
of employing the Krylov MOR technique for creating the
To remedy the insufficiencies in the static condensation coupled models of compliant systems. The idea is to reduce
method, the component mode synthesis (CMS, [18]) was each part of the machine tool structure separately and to
proposed by Bampton and Craig. The CMS has become produce the coupled model of the whole assembly. This way
widely used by the engineering community. The CMS was enables coupling of reduced components in any kinematic
used to efficiently conduct not only large-scale structural configuration. To assess the quality of Krylov MOR and to
eigenanalysis [19] but also transient heat conduction analysis show its soundness the following comparison properties will
[20] and heat conduction/convection analysis [21]. Another be evaluated:
field of application of the CMS is coupled physics simulations.
The weakly coupled thermomechanical models were studied (1) low error in approximation,
in [22]. There is still active research regarding improvement (2) fast computation.
of the CMS [23]. A low error in the approximation of a full FE model
Another method enhanced to approximate dynamic sys- is required, as the objective is to replace the full FE model
tems well is the improved reduced system (IRS) proposed by in simulations. Fast computations are required to meet the
O’Callahan in [24]. Later, Friswell et al. developed an iterated requirements in virtual machine tools simulations.
version of IRS in [25]. The static condensation, CMS, and The proposed procedure will be tested on industrial scale
IRS can be viewed as engineering approaches to reduce the (106 -107 DOFs) FE model of machine tool. The harmonic
number of equations. and transient simulations on such large models take a lot of
The global error bounds and the preservation of passivity time (days, weeks), and therefore some kind of acceleration is
and stability are important questions posed on the MOR usually used. Mode truncation [42] is frequently used for fast
methods in a more mathematical point of view. Two of simulations and will be compared with Krylov based MOR.
the MOR techniques proposed in accordance with these Both methods will be employed in reduction of individual
questions are Krylov subspace reduction [26] and Balanced parts of machine tool. The reduced part will be then coupled
truncation [26]. Balanced truncation methods [27] have a in any kinematic position. The purpose is to simulate flexible
great advantage because there exists a priori global error multibody systems accurately and very fast (with possible
bound. But it also has a great disadvantage in that the applications in real-time CNC control simulations). The
Lyapunov equation [28] needs to be solved in order to proposed method of coupling the Krylov based reduced
reduce the system. Thus the usage of balanced truncation in substructures has not yet been reported in the literature, to
reduction of large-scale systems is limited. the best the authors’ knowledge.
Advances in Mechanical Engineering 3
The paper is organized in the following way: Section 1 Here 𝑥𝑢 is a vector of unique DOFs. Because (5) indicates that
contains the introduction and the motivation of work; the substructures DOF are obtained from the unique set 𝑥𝑢 ,
approach to substructuring is considered in Section 2; the compatibility equation (3) is satisfied for any set 𝑥𝑢 :
Section 3 contains the description of Krylov subspace reduc-
tion; Section 4 introduces the used machine tool model; and B𝑥 = BL𝑥𝑢 = 0 ∀𝑥𝑢 . (6)
Section 5 presents the comparison of the results obtained This means that L is null space of B and vice versa. And it is
by the methods. Conclusion suggestions on future work are only necessary to define one of the matrices L and B:
given in Section 6.
L = null (B) ,
2. Substructuring (7)
B𝑇 = null (L𝑇 ) .
One way to couple dynamic systems is to connect their mass
M𝑠 , damping C𝑠 , and stiffness K𝑠 matrices; external force Substituting (5) into (1), one gets equation of motion
vectors 𝑓𝑠 , and internal force vectors 𝑔𝑠 by coupling equations ML𝑥𝑢̈ + CL𝑥𝑢̇ + KL𝑥𝑢 = 𝑓 + 𝑔. (8)
[43]; the subscript 𝑠 denotes the systems being coupled. This
approach is similar to the global matrix assembly in FE Further multiplication of (8) from the left-hand side by L𝑇
software. The equations of motion of 𝑛 coupled subsystems yields the final form of the equation of motion of the coupled
can be written as system
M𝑥̈+ C𝑥̇+ K𝑥 = 𝑓 + 𝑔. (1) M ̃𝑥𝑢̇ + K𝑥
̃𝑥𝑢̈ + C ̃
̃ 𝑢 = 𝑓, (9)
𝑥 = L𝑥𝑢 . (5) 𝑔𝑚 ≜ R𝑇 𝑔.
4 Advances in Mechanical Engineering
Then following, procedure for coupling full matrices (3)– Full model with n physical DOFs
(10), we get coupled reduced system. This procedure allows
reducing substructures independently and then coupling of
reduced systems. The coupling interface is defined using
matrix B (3), and therefore the specification of the coupling
interface is arbitrary.
Mode superposition
𝑥 = T𝑞. (18)
3.2. Krylov MOR. In this section only the basics behind
This equation can be then substituted to (16) and after Krylov reductions will be described. The reader is encouraged
multiplying the result by T𝑇 from the left-hand side, one to read an excellent mathematical description of Krylov based
obtains the equation of motion in the full system in the modal reductions in [26]. An overview of the reduction methods
coordinates of is given in [27]. Although optimal Krylov based reduction
algorithms are available [44], a simpler and a possibly more
E𝑞 ̈+ C𝑞 𝑞 ̇+ Λ𝑞 = T𝑇 𝑓, (19) computationally efficient method will be used in this work—
a block rational Krylov method [45]. The structure of second
where E is identity matrix, C𝑞 is damping ratio matrix, order ODEs in (15) will be preserved using Bai’s algorithm
and Λ is matrix of eigenvalues. The equations in (19) are [33].
Advances in Mechanical Engineering 5
̃̇(𝑡) + 𝑥1 ,
𝑥̇(𝑡) = 𝑥 (20)
q
̃̈(𝑡) .
𝑥̈(𝑡) = 𝑥
V
Substituting (20) into (15),
̃̈(𝑡) + C𝑥
M𝑥 ̃̇(𝑡) + K𝑥̃ (𝑡) = F𝑢 (𝑡) − C𝑥1 − K𝑥0 . (21)
Figure 2: Projection onto K𝑛 .
Right-hand side of (21) is enriched with the constant term
−C𝑥1 − K𝑥0 , and we may consider this term as additional
force. Then we may assume initial conditions (15) without loss where
of generality in form of
M𝑛 = V𝑇𝑛 MV𝑛 ,
𝑥 (0) = 0, 𝑥̇(0) = 0. (22)
This kind of treatment of a nonzero IC was introduced in [46]. K𝑛 = V𝑇𝑛 KV𝑛 ,
(30)
C𝑛 = V𝑇𝑛 CV𝑛 ,
3.2.2. Krylov Subspace Based MOR. The Laplace transform of
(15) has the form of F𝑛 = V𝑇𝑛 F,
−1
H (𝑠) = L𝑇 (𝑠2 M + 𝑠D + K) F. (23) where M𝑛 , Cn , and K𝑛 ∈ 𝑅𝑛×𝑛 , F ∈ 𝑅𝑛 , and L𝑛 ∈ 𝑅𝑛×𝑚 .
The transfer function of the reduced system (30) has the
And the McLaurin series of transfer function (23) has the
form
form
−1
∞ H𝑛 (𝑠) = L𝑇𝑛 (𝑠C𝑛 + K𝑛 ) Q𝑛 . (31)
H (𝑠) = ∑𝑚𝑙 𝑠𝑙 , (24)
𝑖=0 The above procedure assures that the first 𝑛 moments of the
where 𝑚𝑙 are the so-called moments of the transfer function transfer function (23) of the full system equal the first 𝑛
moments of the transfer function (31) of the reduced system
𝑚𝑙 = L𝑇 𝑟𝑙 , [26].
The error induced by the projection (28) in the output
𝑟0 = K−1 F, function 𝑦(𝑡) has the form
(25)
𝑟1 = −K−1 D𝑟0 , 𝜖 = max 𝑦 (𝑡) − 𝑦̃ (𝑡) . (32)
𝑡>0
𝑟𝑙 = −K−1 (D𝑟𝑙−1 − M𝑟𝑙−2 ) . An a priori expression for error norm (32) is not known
although there exist algorithms minimizing the error [44, 47].
The first 𝑛 vectors 𝑟𝑙 span Krylov space
The algorithm used in this paper to produce the reduced
K𝑛 = span (𝑟0 , . . . , 𝑟𝑛−1 ) . (26) order systems is the block Arnoldi algorithm [45].
There exist wide possibilities to improve the computa-
Letting V𝑛 be the orthonormal basis of K𝑛 tional performance of Krylov methods. One of the most
obvious options is parallelization [48]. Another is to use an
K𝑛 = span (V𝑛 ) , V𝑇𝑛 V𝑛 = I, V𝑛 ∈ 𝑅𝑁×𝑛 . (27)
iterative algorithm to solve the system [49]. The presented
The projection of state coordinates 𝑥 onto K𝑛 using V𝑛 is case is of medium size, and it is therefore suitable to use the
called generalized state coordinates 𝑞 ∈ 𝑅𝑛 direct sparse solver [50].
The procedure is easily extended to a multi-input/multi-
𝑥 = V𝑞 + 𝜖. (28) output case where Q, F ∈ 𝑅𝑁×𝑙 and L ∈ 𝑅𝑁×𝑚 . The size of
𝑛
The error 𝜖 ∈ 𝑅 in the projection rises while performing the reduced system is determined by the size of Q and L.
projection of 𝑥 onto K𝑛 (Figure 2). However, it is possible to use the superposition property [51]
We obtain reduced system of (29) by substituting gener- to keep the matrices small.
alized coordinates 𝑞 into (15) and using the Galerkin method.
The reduced equations have the form of 4. Machine Tool Model for the Case Study
M𝑛 𝑞 ̈(𝑡) + C𝑛 𝑞 ̇(𝑡) + K𝑛 𝑞 (𝑡) = F𝑛 𝑢 (𝑡) , The dynamic performance of a machine tool may be evalu-
(29) ated at the TCP using the oriented directional FRFs, which
𝑦̃ (𝑡) = L𝑇𝑛 𝑞 (𝑡) , directly relate according to stability theory to achievable
6 Advances in Mechanical Engineering
Tool tip the X-slide (face coupling)—or by force interaction in the case
of linear guides (symbols of springs in Figure 4).
Z Z
X Y Y
5. Results
The following case studies compare computational efficiency
Figure 3: Machine tool FE model. of full harmonic in ANSYS software package and model
order reduction using mode truncation and Krylov MOR. To
assess the quality of results obtained using MOR the error
Face coupling norm is evaluated. The study is performed in one kinematic
Face coupling
configuration, and it is assumed that properties of MOR
techniques will be similar in other kinematic configurations.
The model used in the study is a multibody system.
Full FE model
Modal reduction Acknowledgment
Krylov MOR
This research has been supported by the Competence Center-
Figure 5: Comparison of FRF, 𝑋-axis. Manufacturing Technology Project TE01020075 funded by
the Technology Agency of the Czech Republic.
×10−5
8 References
7
[1] Y. Altintas, C. Brecher, M. Weck, and S. Witt, “Virtual machine
Compliance (mm/N)
Machine Tools and Manufacture, vol. 51, no. 7-8, pp. 591–604, [29] E. Grimme, Krylov projection methods for model reduction
2011. [Ph.D. thesis], ECE Department, University of Illinois, Urbana-
[12] M. Zaeh and D. Siedl, “A new method for simulation of machin- Champaign, Ill, USA, 1997.
ing performance by integrating finite element and multi-body [30] E. Grimme and K. Gallivan, “Krylov projection methods
simulation for machine tools,” CIRP Annals—Manufacturing for rational interpolation,” 1997 http://citeseerx.ist.psu.edu/
Technology, vol. 56, no. 1, pp. 383–386, 2007. viewdoc/download?doi=10.1.1.54.5011&rep=rep1&type=pdf.
[13] R. Neugebauer, M. Kolouch, M. Richter, V. Wittstock, and S. [31] Z. Bai, “Krylov subspace techniques for reduced-order mod-
N. Saha, “Model updating of machine tools,” in Proceedings of eling of large-scale dynamical systems,” Applied Numerical
the 9th International Conference on Machine Tools, Automation, Mathematics, vol. 43, no. 1-2, pp. 9–44, 2002.
Robotics and Technology (MATAR ’12), Prague, Czech Republic,
[32] A. Odabasioglu, M. Celik, and L. T. Pileggi, “PRIMA: pas-
September 2012, (MM Science Journal 2012, Special Issue).
sive reduced-order interconnect macromodeling algorithm,”
[14] R. J. Guyan, “Reduction of stiffness and mass matrices,” AIAA in Proceedings of the IEEE/ACM International Conference on
Journal, vol. 3, p. 380, 1965. Computer-Aided Design (ICCAD ’97), pp. 58–65, Washington,
[15] B. M. Irons, “Structural eigenvalue problems: elimination of DC, USA, November 1997.
unwanted variables,” AIAA Journal, vol. 3, no. 5, pp. 961–962,
[33] Z. Bai, K. Meerbergen, and Y. Su, “Arnoldi methods for
1965.
structure-preserving dimension reduction of second-order
[16] L. B. Bushard, “On the value of Guyan reduction in dynamic dynamical systems,” in Dimension Reduction of Large-Scale
thermal problems,” Computers & Structures, vol. 13, no. 4, pp. Systems, P. Benner, D. Sorensen, and V. Mehrmann, Eds., vol.
525–531, 1981. 45 of Lecture Notes in Computational Science and Engineering,
[17] P. Koutsovasilis and M. Beitelschmidt, “Comparison of model pp. 173–189, Springer, Berlin, Germany, 2005.
reduction techniques for large mechanical systems,” Multibody
[34] L. H. Feng, E. B. Rudnyi, J. G. Korvink, C. Bohm, and T. Hauck,
System Dynamics, vol. 20, no. 2, pp. 111–128, 2008.
“Compact electro-thermal model of semiconductor device with
[18] M. Bampton and J. R. Craig, “Coupling of substructures for nonlinear convection coefficient,” in Proceedings of the 6th
dynamic analyses,” AIAA Journal, vol. 6, no. 7, pp. 1313–1319, International Conference on Thermal, Mechanical and Multi-
1968. Physics Simulation and Experiments in Micro-Electronics and
[19] Y. Aoyama and G. Yagawa, “Component mode synthesis for Micro-Systems (EuroSimE ’05), pp. 372–375, Berlin, Germany,
large-scale structural eigenanalysis,” Computers & Structures, April 2005.
vol. 79, no. 6, pp. 605–615, 2001.
[35] E. Zukowski, J. Wilde, E.-B. Rudnyi, and J.-G. Korvink, “Model
[20] D. Botto, S. Zucca, M. Gola, E. Troncarelli, and G. Pasquero, reduction for thermo-mechanical simulation of packages,” in
“Component modes synthesis applied to a thermal transient Proceedings of the 11th International Workshop on Thermal
analysis of a turbine disc,” in Proceedings of the 3rd Worldwide Investigations of ICs and Systems (THERMINIC ’05), pp. 134–
Aerospace Conference & Technology Showcase, Toulouse, France, 138, Belgirate, Italy, 2006.
2002.
[36] R. S. Puri, D. Morrey, A. J. Bell, J. F. Durodola, E. B. Rudnyi, and
[21] D. Botto, S. Zucca, and M. M. Gola, “Reduced-order models
J. G. Korvink, “Reduced order fully coupled structural-acoustic
for the calculation of thermal transients of heat conduc-
analysis via implicit moment matching,” Applied Mathematical
tion/convection FE models,” Journal of Thermal Stresses, vol. 30,
Modelling, vol. 33, no. 11, pp. 4097–4119, 2009.
no. 8, pp. 819–839, 2007.
[37] J. S. Han, E. B. Rudnyi, and J. G. Korvink, “Efficient optimization
[22] P. Nachtergaele, D. J. Rixen, and A. M. Steenhoek, “Efficient
of transient dynamic problems in MEMS devices using model
weakly coupled projection basis for the reduction of thermo-
order reduction,” Journal of Micromechanics and Microengineer-
mechanical models,” Journal of Computational and Applied
ing, vol. 15, no. 4, pp. 822–832, 2005.
Mathematics, vol. 234, no. 7, pp. 2272–2278, 2010.
[23] Z. Bai and B.-S. Liao, “Towards an optimal substructuring [38] J. S. Han, “Efficient frequency response and its direct sensitivity
method for model reduction,” in Applied Parallel Computing. analyses for large-size finite element models using Krylov
State of the Art in Scientific Computing, J. Dongarra, K. Madsen, subspace-based model order reduction,” Journal of Mechanical
and J. Waśniewski, Eds., vol. 3732 of Lecture Notes in Computer Science and Technology, vol. 26, no. 4, pp. 1115–1126, 2012.
Science, pp. 276–285, Springer, Berlin, Germany, 2006. [39] U. Baur, C. Beattie, P. Benner, and S. Gugercin, “Interpolatory
[24] J. O’Callahan, “A procedure for an improved reduced system projection methods for parameterized model reduction,” SIAM
(IRS) model,” in Proceedings of the 7th International Modal Journal on Scientific Computing, vol. 33, no. 5, pp. 2489–2518,
Analysis Conference (IMAC ’89), Las Vegas, Nev, USA, 1989. 2011.
[25] M. I. Friswell, S. D. Garvey, and J. E. T. Penny, “Model reduction [40] J. Šindler, A. Suleng, T. Jelstad Olsen, and P. Bárta, “Krylov
using dynamic and iterated IRS techniques,” Journal of Sound model order reduction of a thermal Subsea model,” World
and Vibration, vol. 186, no. 2, pp. 311–323, 1995. Academy of Science, Engineering and Technology, vol. 77, pp. 217–
[26] A. C. Antoulas, Approximation of Large-Scale Dynamical Sys- 224, 2013.
tems, Advances in Design and Control, SIAM, Philadelphia, Pa, [41] W. Witteveen, “Comparison of CMS, Krylov and balanced trun-
USA, 2005. cation based model reduction from a mechanical application
[27] A. C. Antoulas and D. C. Sorensen, “Approximation of large- engineer’s perspective,” in Topics in Experimental Dynamics
scale dynamical systems: an overview,” International Journal of Substructuring and Wind Turbine Dynamics, R. Mayes, D.
Applied Mathematics and Computer Science, vol. 11, no. 5, pp. Rixen, D. Griffith et al., Eds., vol. 2 of Conference Proceedings
1093–1121, 2001. of the Society for Experimental Mechanics Series, pp. 319–331,
[28] T. Stykel, Analysis and numerical solution of generalized Lya- Springer, New York, NY, USA, 2012.
punov equations [Ph.D. thesis], Technical University of Berlin, [42] Z. Qu, Model Order Reduction Techniques: With Applications in
Berlin, Germany, 2002. Finite Element Analysis, Springer, New York, NY, USA, 2004.
Advances in Mechanical Engineering 9
Research Article
Study on Launch Dynamics of Self-Propelled Artillery Based on
Transfer Matrix Method of Multibody System
Copyright © 2014 H. Yu and X. Rui. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Due to the importance of increasing the firing precision of self-propelled artillery system corresponding to the projectile impact
point, the launch dynamics of self-propelled artillery based on the transfer matrix method for multibody system (MSTMM) is
studied in this paper. By using MSTMM to study launch dynamics of self-propelled artillery, there is no need of global dynamical
equation of the system, the order of the system matrix is low, the computational speed is high, and it meets the requirements
of dynamics of self-propelled artillery multibody system. The dynamical model of multibody system and its topological figure,
the overall transfer equation, overall transfer matrix, characteristics equation and generalized coordinates equation, and launch
dynamical equation of self-propelled artillery are developed. The vibration characteristics, initial disturbance of projectile,
dynamical response, and firing dispersion for self-propelled artillery are simulated. The results of eigenfrequency and the time
history of system dynamics got by simulations and experiments have good agreements. The study results can be used to improve
the firing precision and design of self-propelled artillery.
vibration characteristics and is successfully applied to many Taking self-propelled artillery as an example, according
engineering design and test of various types of vehicles, to its system structure, from bottom to top in sequence, it can
airborne, and ship-borne weapons [1, 2, 4]. be divided into road wheels, hull, and revolving part which
In this paper, by using MSTMM and its automatic does not contain elevating part, elevating part which does
deduction method [2, 20], the dynamical model of multibody not contain recoil part, gun breech, and muzzle brake. Each
system for self-propelled artillery is developed; the overall component can be regarded as rigid body and elastic beam
transfer equation, overall transfer matrix, and characteristics according to its natural attribute. Rigid body and elastic beam
equation are deduced. Generalized coordinates equation of are called “body,” the connection between “body” and “body”
self-propelled artillery is obtained by using the orthogonality is called the “hinge,” which can be seen in [2].
of augmented eigenvector and body dynamical equation. The ground that supports the self-propelled artillery is
Combining with launch dynamical equation of projectile, regarded as an infinity rigid body, whose sequence number
launch dynamical equation of self-propelled artillery system is 0. The elastic and damping effect of each road wheel
is developed. The vibration characteristics, initial disturbance and the interaction between ground and each road wheel
of projectile, dynamical response, and firing dispersion are are, respectively, modeled as springs, rotary springs, and
simulated, and the factors of different ground conditions the accompanying dampers connected in parallel, which can
and connection stiffness among various components, which represent relative linear motion and relative angular motion
influence the firing precision and vibration characteristics, in 3 directions at the same time; their sequence numbers are
are analyzed. It provides a theoretical basis and simulation 42, 43, . . . , 53. Each road wheel can be regarded as a rigid
tool for improving the firing precision by adjusting structural body, whose sequence number is 30, 31, . . . , 41. The inter-
parameters of the system to change the vibration characteris- action between each road wheel and hull are, respectively,
tics of the self-propelled artillery. modeled as springs, rotary springs, and the accompanying
dampers connected in parallel, which can represent relative
2. Dynamical Model of Self-Propelled Artillery linear motion and relative angular motion in 3 directions
Multibody System at the same time; their sequence numbers are 18, 19, . . . , 29.
Hull, revolving part, elevating part, gun breech, and muzzle
Main components of self-propelled artillery are muzzle brake, brake can be regarded as a rigid body, respectively, whose
barrel, gun breech, recoil and counter-recoil mechanisms, sequence numbers are 17, 15, 13, 10, and 1. The effect of
cradle, elevating mechanism, equilibrator, turret, travers- traversing mechanism associating the elastic, and damping
ing mechanism, chassis, torsion bar, balance elbow, shock effects of hull, the effect of elevating mechanism and equi-
absorber, track chain, road wheel, and so on. According to the librator associating the elastic and damping effects between
motion state of each component, the firepower system of self- revolving part and elevating part are, respectively, modeled as
propelled artillery can be divided into recoil part, elevating springs and rotary springs accompanying dampers which can
part, revolving part, suspension part, walking part, and so represent relative linear motion and relative angular motion
on. The recoil part contains muzzle brake, barrel, gun breech, in 3 directions at the same time, whose sequence numbers
and recoil and counter-recoil mechanisms. The elevating are 16, 14. The interaction between barrel and elevating
part contains total recoil part, cradle, and the components par, is, respectively, modeled as springs and rotary springs
moving with cradle, which includes elevating mechanism, accompanying dampers which can represent relative linear
equilibrator, and so on. The revolving part contains total motion and relative angular motion in 3 directions at the
elevating part, turret, and the elements moving with turret same time, whose sequence numbers are 11, 12. According
that include traversing mechanism and so on. The walking to its structure characteristics, the barrel is divided into 6
part is used to support the weight of the self-propelled segments, each segment can be regarded as a beam with
artillery and drive self-propelled artillery to run placidly, equal sectional area, whose sequence numbers are 2, 3, 4, 5, 7,
which contains track chain and road wheels. The suspension and 9. The connection points among joints and beam can be
part is used to connect chassis to walking part, which contains regarded as massless rigid body, whose sequence numbers are
torsion bar, balance elbow, and shock absorber. The self- 6, 8. The dynamical model of self-propelled artillery multi-
propelled artillery is shown in Figure 1. rigid-flexible system is composed of 19 rigid bodies, 6 beams,
Advances in Mechanical Engineering 3
and 28 joints, as shown in Figure 2. The topology figure of where, 1 is the sequence number of body and 0 denotes the
dynamical model of the self-propelled system is shown in boundary.
Figure 3. The form of Z10,0 , Z11,13 , Z13,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 ,
Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , and Z53,0 is similar
3. Overall Transfer Equation of to Z1,0 .
Self-Propelled Artillery
3.2. Overall Transfer Equation of Self-Propelled Artillery Sys-
3.1. The State Vector of Self-Propelled Artillery. According to tem. According to MSTMM and the topology figure of
the dynamical model and its topology figure, the state vectors the dynamical model, the overall transfer equation of self-
at boundary points for self-propelled artillery are defined as propelled artillery system is automatically deduced as follows:
follows
𝑇 Uall Zall = 0, (2)
Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]1,0 ,
(1) where overall transfer matrix
−I12 T10−1 T42−1 T43−1 T44−1 T45−1 T46−1 T47−1 T48−1 T49−1 T50−1 T51−1 T52−1 T53−1 T13−1 + T11−1
[006×12 06×12 G42−17 G43−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ 6×12 06×12 G42−17 06×12 G44−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 G45−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 G46−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 G47−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 G48−17 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ ]
Uall = [06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 G49−17 06×12 06×12 06×12 06×12 06×12 ].
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G50−17 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G51−17 06×12 06×12 06×12 ]
[0 ]
[ 6×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G52−17 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G53−17 06×12 ]
[0 06×12 G42−13 G43−13 G44−13 G45−13 G46−13 G47−13 G48−13 G49−13 G50−13 G51−13 G52−13 G53−13 G13,𝐼2 ]
[ 6×12 ]
06×12 G10−8 G42−8 G43−8 G44−8 G45−8 G46−8 G47−8 G48−8 G49−8 G50−8 G51−8 G8−52,𝑏12 G53−8 G13−8
[06×12 G10−6 G42−6 G43−6 G44−6 G45−6 G46−6 G47−6 G48−6 G49−6 G50−6 G51−6 G52−6 G53−6 G13−6 + G11−6 ]
(3)
Elevating part 13
Projectile
Gun breech 10 14 11
12 Barrel 2∼5 Muzzle brake 1
0
16
18 19 20 21 22 23
30 31 32 33 34 35
42 43 44 45 46 47 Road wheel 30∼41
0 0
48 49 50 51 52 53
36 37 38 39 40 41
9 7 5-2
Z10.0 10 I1 8 I1 6 1 Z1.0
24 25 26 27 28 29 I2 I2
I7 I8 I9 I10 I11 I12 12 11
Z11.13
17 15 I1 13 Z13.0
16 14 I2
I1 I2 I3 I4 I5 I6
18 19 20 21 22 23
30 31 32 33 34 35
42 43 44 45 46 47
eigenvalue of the system. Hence, the system characteristics an arbitrary point on beam (barrel) can be obtained. Thus,
equation is obtained by the vibration characteristics of self-propelled artillery are
obtained, which include the eigenfrequency 𝜔𝑘 (𝑘 = 1, 2, . . .)
det (Uall ) = 0. (7) and the mode shape corresponding to each 𝜔𝑘 , and the datum
of mode shape is included in the variables of all state vec-
By solving (7), the eigenfrequency of self-propelled artillery, tors of system, so that the mode shape can be got after the
𝜔𝑘 (𝑘 = 1, 2, . . .), can be obtained. Solving (6), the state vector corresponding variables selected them from the state vec-
𝑘
Zall corresponding to 𝜔𝑘 can be obtained, in other words, Z𝑘all tors.
is obtained. That is to say, the state vectors Z1,0 , Z10,0 , Z13,0 , It can be seen from (6) that the order of the overall
Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , transfer matrix of the self-propelled artillery system is only
Z52,0 , and Z53,0 corresponding to 𝜔𝑘 are obtained. Based on 96 and is much lower than that for other multibody system
these state vectors at the system boundaries, through using dynamics methods so computational speed is high and the
the transfer equation of each element one by one, all the computational ill-condition caused by high matrix order and
state vectors in system including each connection points and large stiffness gradient is avoided.
Advances in Mechanical Engineering 5
5. Launch Dynamical Equation of 5.3. Launch Dynamical Equation of the Projectile. Launch
Self-Propelled Artillery dynamical equation of a projectile is given by [1, 5]
C𝑗 = 𝛼M𝑗 + 𝛽K𝑗 . (12) where the specific meaning of all symbols is no longer given,
which can be seen in [1, 5].
There are
+∞
𝑘
∑ ⟨ ∑ (C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡) , V𝑘𝑗 ⟩ = (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) 𝑑𝑘 . (13) 6. Numerical Simulation and
𝑗 𝑘1 =1 Experimental Validation
Generalized coordinate equation of self-propelled artillery is Launch dynamics of self-propelled artillery can be computed
obtained, by combining (14) and (16). Eigenfrequency, vertical target
dispersion, the time history of barrel recoil displacement,
𝑞𝑘̈(𝑡) + (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) + 𝜔𝑘2 𝑞𝑘 (𝑡) = 𝑝𝑘 (𝑡) , (14)
and the time history of muzzle displacement in the plumb
where direction in cement ground for a self-propelled artillery got
by simulation and test are shown, respectively, in Tables 1 and
𝑘
∑𝑗 ⟨f𝑗 , V𝑘𝑗 ⟩ 2 and Figures 4 and 5. The simulation parameters can be seen
𝑝 (𝑡) =
𝑑𝑘 (15) in [1].
It can be seen from Tables 1 and 2 and Figures 4 and 5
(𝑗 = 1, 2, 3, 4, 5, 7, 9, 10, 13, 15, 17, 30, 31, . . . , 41) . that results got by simulation and test have good agreements,
6 Advances in Mechanical Engineering
Mode order 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Simulation 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8 87.5 94.2 101.5 — 141.0 214.2 225.9 243.7
Test — 15.5 18.7 29.9 42.2 49.8 — — 82.1 — 105.5 128.4 136.3 202.9 229.2 244.4
Error (%) −1.3 −1.0 −2.7 4.3 6.0 6.2 −3.8 3.4 5.6 −1.4 −0.3
0 0.6
0.4
−0.4
0.2
−0.6
0
−0.8
−1 −0.2
0 0.5 1 1.0 2 0 1 2 3
t (s) t (s)
Test Test
Simulation Simulation
Figure 4: The time history of barrel recoil displacement got by Figure 5: The time history of muzzle displacement in plumb
simulation and test. direction got by simulation and test.
× U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , G47−17 = H17,23 U23 U35 U47 ,
× U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , G49−17 = H17,25 U25 U37 U49 ,
8 Advances in Mechanical Engineering
Table 6: Firing dispersion of a self-propelled artillery got by G46−8 = H8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,
simulation on the cement ground.
G47−8 = H8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,
Vertical target dispersion
Different connection stiffness Vertical Direction G48−8 = H8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,
dispersion/m dispersion/m
(2.2 × 107 , 2.2 × 107 , 1.1 × 108 ) G49−8 = H8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,
0.411 0.367
(5.5 × 107 , 1.3 × 107 , 1.1 × 107 )
G50−8 = H8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,
(2.2 × 106 , 2.2 × 106 , 1.1 × 107 )
0.543 0.380
(5.5 × 106 , 1.3 × 106 , 1.1 × 106 ) G51−8 = H8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,
G52−8 = H8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,
G53−8 = H8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,
G50−17 = H17,26 U26 U38 U50 ,
G51−17 = H17,27 U27 U39 U51 , G13−8 = H8,12 U12 U13,0 ,
G52−17 = H17,28 U28 U40 U52 , G42−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 ,
G53−17 = H17,29 U29 U41 U53 , G43−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 ,
G42−13 = H13 U14 U15 U16 U17 U18 U30 U42 , G44−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,
G43−13 = H13 U14 U15 U16 U17,19 U19 U31 U43 , G45−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 ,
G44−13 = H13 U14 U15 U16 U17,20 U20 U32 U44 , G46−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,
G45−13 = H13 U14 U15 U16 U17,21 U21 U33 U45 , G47−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,
G46−13 = H13 U14 U15 U16 U17,22 U22 U34 U46 , G48−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,
G47−13 = H13 U14 U15 U16 U17,23 U23 U35 U47 ,
G49−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,
G48−13 = H13 U14 U15 U16 U17,24 U24 U36 U48 ,
G50−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,
G49−13 = H13 U14 U15 U16 U17,25 U25 U37 U49 ,
G51−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,
G50−13 = H13 U14 U15 U16 U17,26 U26 U38 U50 ,
G52−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,
G51−13 = H13 U14 U15 U16 U17,27 U27 U39 U51 ,
G53−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,
G52−13 = H13 U14 U15 U16 U17,28 U28 U40 U52 ,
G13−6 = H6 U7 U8,12 U12 U13,0 ,
G53−13 = H13 U14 U15 U16 U17,29 U29 U41 U53 ,
G11−6 = −H6,11 U11 C,
G13,I2 = −H13,0 ,
G10−6 = H6 U7 U8 U9 U10 ,
G10−8 = −H8 U9 U10 ,
I O6×6
G42−8 = H8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , C=[ 6 ].
O6×6 −I6
G43−8 = H8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , (A.1)
G44−8 = H8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,
All matrices is no longer given in detail, which can be seen in
G45−8 = H8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 , [1, 2, 20].
Advances in Mechanical Engineering 9
Table 8: Firing dispersion of self-propelled artillery got by simula- [11] X. Rui and W. Schiehlen, “Multibody system dyanamics: pref-
tion on the cement ground. ace,” Multibody System Dynamics, vol. 18, no. 4, pp. 485–492,
2007.
Vertical target dispersion
[12] X. Rui, L. Yun, G. Wang et al., “Augmented eigenvector and its
Different connection stiffness Vertical Direction orthogonality of linear multi-rigid-flexible-body system,” Jour-
dispersion/m dispersion/m nal of China Ordnance, vol. 4, no. 2, pp. 100–105, 2008.
(2.2 × 107 , 2.1 × 107 , 5.3 × 107 ) [13] X. Rui, B. He, B. Rong et al., “Discrete time transfer matrix
0.411 0.367
(5.5 × 107 , 1.1 × 107 , 1.1 × 108 ) method for dynamics of multi-rigid-flexible-body system mov-
(2.2 × 106 , 2.1 × 106 , 5.3 × 106 ) ing in plane,” Journal of Multi-Body Dynamics, vol. 23, no. 1, pp.
0.458 0.371
(5.5 × 106 , 1.1 × 106 , 1.1 × 107 ) 23–42, 2009.
[14] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for
linear multibody system,” Multibody System Dynamics, vol. 19,
Conflict of Interests no. 3, pp. 179–207, 2008.
[15] X. Rui, E. Kreuzer, B. Rong, and B. He, “Discrete time transfer
The authors declare that there is no conflict of interests matrix method for dynamics of multibody system with flexible
regarding the publication of this paper. beams moving in space,” Acta Mechanica Sinica, vol. 28, pp.
490–504, 2012.
Acknowledgments [16] X. Rui, B. Rong et al., “Some new developments in discrete time
transfer matrix method of multibody systems,” in Proceedings
The research was supported by the Research Fund for of the International Conference on Dynamics Vibration and
the Doctoral Program of Higher Education of China Control, Hangzhou, China, 2010.
(20113219110025), the Natural Science Foundation of China [17] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer
Government (11102089), and the Program for New Century matrix method for dynamics analysis of complex weapon sys-
Excellent Talents in University (NCET-10-0075). tems,” Science China Technological Sciences, vol. 54, no. 5, pp.
1061–1071, 2011.
[18] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer
References matrix method for multibody system dynamics,” Multibody
System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.
[1] X. Rui, Y. Liu, and H. Yu, Launch Dynamics of Tank and Self-
Propelled Artillery, Science Press, Beijing, China, 2008. [19] B. He, X. Rui, and G. Wang, “Riccati discrete time transfer
matrix method for elastic beam undergoing large overall mo-
[2] X. Rui, L. Yun, Y. Lu et al., Transfer Matrix Method for Multibody
tion,” Multibody System Dynamics, vol. 18, no. 4, pp. 579–598,
System and its Application, Science Press, Beijing, China, 2008.
2007.
[3] X. Rui, L. Yun, G. Wang et al., Direction to Launch Safety of
[20] X. Rui and J. Zhang, “Automatical transfer matrix method of
Ammunition, National Defense Industry Press, Beijing, China,
multibody system,” in Proceedings of the 2nd Joint International
2009.
Conference on Multibody System Dynamics, Stuttgart, Germany,
[4] X. Rui, Y. Lu, G. Wang et al., Simulation and Test Methods of May-June 2012.
Launch Dynamics of Multiple Launch Rocket System, National
Defense Industry Press, Beijing, China, 2003.
[5] X. Rui, Launch Dynamics of Multibody System, National Defense
Industry Press, Beijing, China, 1995.
[6] X. Rui and Q. Yang, Launch Process of Projectile, South East
University Press, Nanjing, China, 1991.
[7] J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G.
Teubner, Stuttgart, Germany, 1977.
[8] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,
Germany, 1990.
[9] T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft Dynam-
ics, McGraw-Hill, New York, NY, USA, 1983.
[10] X. Rui, D. Bestle, and J. Zhang, “A new form of the trans-
fer matrix method for multibody systems,” in Proceedings of
the ECCOMAS Thematic Conference on Multibody Dynamics,
Zagreb, Croatia, 2013.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 383680, 8 pages
http://dx.doi.org/10.1155/2014/383680
Research Article
Higher Order Variational Integrators for Multibody System
Dynamics with Constraints
Received 25 August 2013; Revised 16 November 2013; Accepted 25 November 2013; Published 23 January 2014
Copyright © 2014 J. Ding and Z. Pan. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The continuous and discrete Euler-Lagrangian equations with holonomic constraints are presented based on continuous and
discrete Hamiltonian Principle. Using Lagrangian polynomial to interpolate state variables and Gauss quadrature formula to
approximate Hamiltonian action integral, the higher order variational Galerkin integrators for multibody system dynamics with
constraints and the computation procedure are given. Numerical examples are provided to show the long-time behavior of the
methods proposed in this paper via comparisons with traditional Runge-Kutta methods.
focus on the cases with holonomic constraints, but the results where 𝐿 𝑑 (q𝑖 , q𝑖+1 , ℎ) and 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) are defined as the
can be easily extended to the cases with nonholonomic discrete approximation of the integration of 𝐿(q, q,̇ 𝑡) and
constraints and the cases with nonconservative forces. 𝜆𝑇 Φ(q, 𝑡) on the interval 𝑡 ∈ [𝑡𝑖 , 𝑡𝑖+1 ],
The paper is organized as follows. In Section 2, the 𝑡𝑖+1
continuous and discrete Euler-Lagrangian equations with 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) = ∫ 𝐿 (q, q,̇ 𝑡) 𝑑𝑡,
𝑡𝑖
holonomic constraints are presented based on continuous (5)
𝑡𝑖+1
and discrete Hamiltonian Principle, respectively; in Section 3, 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) = ∫ 𝜆𝑇 Φ (q, 𝑡) 𝑑𝑡.
we derive the higher order variational Galerkin integra- 𝑡𝑖
tors for multibody system dynamics with constraints using
Lagrangian polynomial to interpolate state variables and Using standard variational method, the discrete Euler-
using Gauss quadrature formula to approximate Hamilto- Lagrangian (DEL) equations are derived as
nian action integral; the computation procedure is given in 𝐷1 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) + 𝐷2 𝐿 𝑑 (q𝑖−1 , q𝑖 , 𝑡𝑖−1 )
Section 4; numerical examples are provided in Section 5 to
show the long-time behavior of the methods proposed in this − ℎΦ𝑇q (q𝑖 , 𝑡𝑖 ) 𝜆𝑖 = 0, (6)
paper via comparisons with traditional methods. The last one Φ (q𝑖+1 , 𝑡𝑖+1 ) = 0,
is concluding remarks including summary and future works.
where 𝐷𝑗 𝐿 𝑑 (𝑗 = 1, 2) is the partial derivative of 𝐿 𝑑 with
2. Discrete Euler-Lagrangian respect to the 𝑗th variable.
Equations with Constraints
3. Higher Order Variational Galerkin
The Hamilton Principle of a multibody system with holo- Integrators of Multibody
nomic constraints and conservative forces can be stated as
Systems with Constraints
𝑡𝑓
𝛿𝑆 = 𝛿 ∫ (𝐿 (q, q,̇ 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡 = 0, (1) The key technique to improve the accuracy of variational
0
integrators is the approximation of the discrete Hamilto-
where 𝑆 is Hamilton action integral, 𝐿 is Lagrangian, q, q̇are nian action integral, which can be realized through general
generalized displacement and velocity, respectively, and 𝜆 is Galerkin methods [24, 25]. Using the same methods in this
Lagrange multiplier vector corresponding to the constraint section, Hamiltonian action integral is approximated in a
subinterval [𝑡𝑖 , 𝑡𝑖+1 ] with an introduced local time parameter
Φ (q, 𝑡) = 0. (2) 𝜏 ∈ [0, 1], where
Via variational method, the following Euler-Lagrange 𝑡 − 𝑡𝑖
𝜏= . (7)
equations of a constrained mechanical system can be got: 𝑡𝑖+1 − 𝑡𝑖
𝑑 𝜕𝐿 (q, q,̇ 𝑡) 𝜕𝐿 (q, q,̇ 𝑡) Given 𝑠 + 1 control points 0 = 𝜏0 < 𝜏1 < ⋅ ⋅ ⋅ < 𝜏𝑠 = 1,
( )− + Φ𝑇 (q, 𝑡) 𝜆 = 0, for 𝑘 = 0, 1, . . . , 𝑠, Lagrange interpolation polynomial 𝑙𝑘 (𝜏) :
𝑑𝑡 𝜕q̇ 𝜕q (3) [0, 1] → 𝑅 is defined as
Φ (q, 𝑡) = 0. 𝜏 − 𝜏𝑗
𝑙𝑘 (𝜏) = ∏ . (8)
It is a typical differential-algebraic equation which can be 𝜏 −𝜏 𝑘 𝑘
0≤𝑗≤𝑠,𝑗 ≠ 𝑗
solved using different traditional numerical methods of DAEs
[2, 3]. The state variables q(𝑡) on [𝑡𝑖 , 𝑡𝑖+1 ] can be interpolated as
Following the variational integrators based on discrete follows:
𝑠
variational principle, the time interval 𝑡 ∈ [0, 𝑡𝑓 ] is divided q𝑑 (𝜏) = ∑ 𝑙𝑘 (𝜏) q𝑖,𝑘 , (9)
into 𝑁 subintervals with time step ℎ = 𝑡𝑓 /𝑁 equally along 𝑘=0
with 𝑁 + 1 grids, 0 = 𝑡0 , 𝑡1 , 𝑡2 , . . . , 𝑡𝑁−1 , 𝑡𝑁 = 𝑡𝑓 , 𝑡𝑖 = 𝑖ℎ,
𝑖 = 0, 1, 2, . . . , 𝑁. The discrete generalized displacement and where q𝑖,𝑘 = q(𝑡𝑖 + 𝑘 ⋅ (ℎ/𝑠)). The derivative of q𝑑 (𝜏) with
velocity and Lagrangian multiplier can be denoted as q𝑖 = ̇ on [𝑡𝑖 , 𝑡𝑖+1 ] as
respect to 𝑡 provides an approximation of q(𝑡)
q(𝑡𝑖 ), q̇𝑖 = q(𝑡
̇ 𝑖 ), 𝜆𝑖 = 𝜆(𝑡𝑖 ), 𝑖 = 0, 1, 2, . . . , 𝑁. Equation (1) can 1 𝑠
be rewritten as the following discrete counterpart: q̇𝑑 (𝜏) = ∑ 𝑙 (𝜏) q𝑖,𝑘 , (10)
ℎ 𝑘=0 𝑘
𝛿𝑆𝑑 (q0 , q1 , . . . , q𝑁−1 , q𝑁, 𝜆0 , . . . , 𝜆𝑁−2 , 𝜆𝑁−1 )
where 𝑙𝑘 (𝜏) = (𝜕/𝜕𝜏)𝑙𝑘 (𝜏). Then, Hamiltonian action integral
𝑁−1 can be approximated on [𝑡𝑖 , 𝑡𝑖+1 ] as follows:
= 𝛿 ∑ 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) 𝑡𝑖+1
𝑖=0
(4) ∫ (𝐿 (q (𝑡) , q̇(𝑡) , 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡
𝑁−1 𝑡𝑖
1
− 𝛿 ∑ 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) (11)
𝑖=0
= ∫ ℎ (𝐿 (q𝑑 (𝜏) , q̇𝑑 (𝜏) , 𝑡𝑖 + 𝜏ℎ)
0
Runge-Kutta P1L-Q2G
3 3
Total energies (N·m)
2 2
1 1
0 0
0 50 100 0 50 100
Time (s) Time (s)
(a) (b)
P2L-Q2G P2L-Q3G
3 3
Total energies (N·m)
2 2
1 1
0 0
0 50 100 0 50 100
Time (s) Time (s)
(c) (d)
y
xj
yj A
xi i
x
B
yi
Oj
Oi
𝜃
O x
P1L-Q2G P1L-Q2G
30 3
20 2
10 1
Total energies (N·m)
Energies (N·m)
0 0
−10 −1
−20 −2
−30 −3
0 20 40 60 80 100 0 20 40 60 80 100
Time (s) Time (s)
Kinetic energy
Potential energy
Total energy
(a) (b)
compared with the Runge-Kutta But using the discrete variational methods presented in this
method in Table 3 for the ter- paper, the time step can be chosen longer as 0.01 s. Table 3 sh-
minal time 100 s. Here, P1L-Q2G means that the order of Lag- ows that, with this longer time step, the discrete variational
range interpolation polynomial is 1 and the number of Gauss methods can keep the total energy and constraints in higher
quadrature points is 2, and the same as P2L-Q2G and P2L- accuracy than Runge-Kutta method, the cost instead is the
Q3G. 𝜀(Η), 𝜀(Φ) are the maximum errors of the total energy longer computation time because of the iterations during the
𝐻 and constraints Φ. nonlinear equations solving process, but it is still shorter than
Using the traditional method such as Runge-Kutta met- the computation time of Runge-Kutta method with smaller
hod, to keep the errors of the total energy 𝐻 and constraints Φ time step to reach the same accuracy of the results.
are all smaller, the time step should be chosen as about 0.001 s It is also shown in Tables 3 and 4 that when the order of
in this example, and the total computation time is 99.7938 s. Lagrange interpolation polynomial is higher, the total energy
6 Advances in Mechanical Engineering
×10−16 ×10−16
4 4
2 2
Constraints 1
Constraints 2
0 0
−2 −2
−4 −4
0 50 100 0 50 100
Time (s) Time (s)
(a) (b)
−16 −16
×10 ×10
4 4
2 2
Constraints 3
Constraints 4
0 0
−2 −2
−4 −4
0 50 100 0 50 100
Time (s) Time (s)
(c) (d)
Figure 5: Constraints of rotary rod slider system by method P1L-Q2G, ℎ = 0.01.
and constraints can be kept better, but there are no big In the case of nonconstant mass matrix, the methods of
differences between 2 or 3 points in Gauss quadrature higher order variational Galerkin integrators are also applica-
formula. ble. For another example, Figure 3 shows a rotary rod slider
Figure 2 shows the energies of the previously mentioned system. 𝑂𝐴 is a rigid rod with uniform mass 𝑚𝐴 and length 𝑙,
methods. With the time step ℎ = 0.01 s, the total energy is which rotates round 𝑂 in the plane 𝑂𝑋𝑌. 𝐵 is a slider with the
up and down around zero with little errors during the long-ti- mass 𝑚𝐵 , and the stiffness of the spring on it is 𝑘; the mass of
me simulation by the methods presented in this paper, wh- the spring is ignored. Suppose only gravity in the plane 𝑂𝑋𝑌
ile increases quickly during the simulation by Runge-Kutta is considered for the system.
method. The results of high order variational Galerkin inte- The state variables are chosen as q = [𝑅𝑥𝑖 𝑅𝑦𝑖 𝑅𝑥𝑗
grators are better than low order integrators. 𝑅𝑦𝑗 𝜃 𝑥𝑖 ]𝑇 . Then, the mass matrix is a nonconstant matrix
The constraint equations are [6] Z. Ge and J. E. Marsden, “Lie-Poisson Hamilton-Jacobi theory
and Lie-Poisson integrators,” Physics Letters A, vol. 133, no. 3, pp.
𝑙 134–139, 1988.
𝑅𝑥𝑗 − cos 𝜃
2 [7] J. M. Wendlandt and J. E. Marsden, “Mechanical integrators der-
( 𝑙 ) ived from a discrete variational principle,” Physica D, vol. 106,
𝑅𝑦𝑗 − sin 𝜃
Φ=(
( 2 ) = 0.
)
no. 3-4, pp. 223–246, 1997.
[8] J. E. Marsden and M. West, “Discrete Mechanics and Variational
(𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) cos 𝜃 + (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) sin 𝜃 − 𝑥𝑖 Integrators,” Acta Numerica, vol. 10, pp. 357–514, 2001.
( (𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) sin 𝜃 − (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) cos 𝜃 ) [9] J. Moser and A. P. Veselov, “Discrete versions of some classical
(18) integrable systems and factorization of matrix polynomials,”
Communications in Mathematical Physics, vol. 139, no. 2, pp.
Given 𝑙1 = 2𝑚, 𝑚𝐴 =2 kg, 𝑚𝐵 = 5 kg, 𝑘 = 1000 N/m. 217–243, 1991.
With the initial step length ℎ = 0.01 s and the terminal [10] C. Kane, J. E. Marsden, and M. Ortiz, “Symplectic-energy-mo-
time 100 s, using the method P1L-Q2G, where the order of mentum preserving variational integrators,” Journal of Mathe-
matical Physics, vol. 40, no. 7, pp. 3353–3371, 1999.
Lagrange interpolation polynomial is 1 and the number of
Gauss quadrature points is 2, the results are also satisfying. [11] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Variational time
integrators,” International Journal for Numerical Methods in
The total computer time is 86.0034 s; the maximum errors of
Engineering, vol. 60, no. 1, pp. 153–212, 2004.
the total energy 𝐻 and constraints Φ are 0.2611 and 3.8858 ×
[12] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “An overview of
10−16 , which is illustrated in Figures 4 and 5. variational integrators,” in Finite Element Methods: 1970s and
Beyond, pp. 85–146, CIMNE, Barcelona, Spain, 2003.
6. Conclusions [13] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Asynchronous
variational integrators,” Archive for Rational Mechanics and
Based on Lagrange interpolation and Gauss quadrature, hig- Analysis, vol. 167, no. 2, pp. 85–146, 2003.
her order variational Galerkin integrators have been devel- [14] A. Pandolfi, C. Kane, J. E. Marsden, and M. Ortiz, “Time-dis-
oped for multibody systems with constraints. Numerical res- cretized variational formulation of non-smooth frictional con-
ults validate the accuracy of the method for longer time step tact,” International Journal for Numerical Methods in Engineer-
and long-time simulation. Nevertheless, as the computation ing, vol. 53, no. 8, pp. 1801–1829, 2002.
is more complex, the computation cost is higher. Further [15] R. C. Fetecau, J. E. Marsden, M. Ortiz, and M. West, “Nonsmo-
study is needed to have more efficient integrators. oth lagrangian mechanics and variational collision integrators,”
SIAM Journal on Applied Dynamical Systems, vol. 2, no. 3, pp.
381–416, 2003.
Conflict of Interests [16] N. Bou-Rabee and H. Owhadi, “Stochastic variational integra-
tors,” IMA Journal of Numerical Analysis, vol. 29, no. 2, pp. 421–
The authors declare that there is no conflict of interests regar-
443, 2009.
ding the publication of this paper.
[17] W. Hu, Z. Deng et al., “Generalized multi-symplectic integrators
for a class of Hamiltonian nonlinear wave PDEs,” Journal of
Acknowledgments Computational Physics, vol. 235, pp. 394–406, 2013.
[18] J. Cortés and S. Martı́nez, “Non-holonomic integrators,” Nonlin-
This research was supported by the National Natural Sci- earity, vol. 14, no. 5, pp. 1365–1392, 2001.
ence Foundation of China (Grants nos. 11272166, 11002075, [19] R. McLachlan and M. Perlmutter, “Integrators for nonholonom-
10972110). ic mechanical systems,” Journal of Nonlinear Science, vol. 16, no.
4, pp. 283–328, 2006.
References [20] S. Leyendecker, J. E. Marsden, and M. Ortiz, “Variational integr-
ators for constrained dynamical systems,” ZAMM Zeitschrift für
[1] A. A. Shabana, Dynamics of Multibody Systems, Cambridge Uni- Angewandte Mathematik und Mechanik, vol. 88, no. 9, pp. 677–
versity Press, 3rd edition, 2005. 708, 2008.
[2] A. Laulusa and O. A. Bauchau, “Review of classical approaches [21] C. Kane, J. E. Marsden, M. Ortiz, and M. West, “Variational
for constraint enforcement in multibody systems,” Journal of integrators and the newmark algorithm for conservative and
Computational and Nonlinear Dynamics, vol. 3, no. 1, Article ID dissipative mechanical systems,” International Journal for Num-
011004, 2008. erical Methods in Engineering, vol. 49, no. 10, pp. 1295–1325,
[3] O. A. Bauchau and A. Laulusa, “Review of contemporary appr- 2000.
oaches for constraint enforcement in multibody systems,” Jour- [22] S. Leyendecker, S. Ober-Blöbaum, J. E. Marsden, and M. Ortiz,
nal of Computational and Nonlinear Dynamics, vol. 3, no. 1, “Discrete mechanics and optimal control for constrained sys-
Article ID 011005, 2008. tems,” Optimal Control Applications and Methods, vol. 31, no. 6,
[4] L. Jay, “Symplectic partitioned Runge-Kutta methods for con- pp. 505–528, 2010.
strained Hamiltonian systems,” SIAM Journal on Numerical [23] S. Ober-Blöbaum, O. Junge, and J. E. Marsden, “Discrete me-
Analysis, vol. 33, no. 1, pp. 368–387, 1996. chanics and optimal control: an analysis,” Control, Optimisation
[5] O. Gonzalez, “Mechanical systems subject to holonomic con- and Calculus of Variations, vol. 17, no. 2, pp. 322–352, 2011.
straints: differential-algebraic formulations and conservative [24] K. Witting, S. Ober-Blöbaum, and M. Dellnitz, “A variational
integration,” Physica D, vol. 132, no. 1-2, pp. 165–174, 1999. approach to define robustness for parametric multi-objective
8 Advances in Mechanical Engineering
Research Article
Free Vibration Characteristic of Multilevel Beam Based on
Transfer Matrix Method of Linear Multibody Systems
Copyright © 2014 L. K. Abbas and X. Rui. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
In this paper, an approach based on transfer matrix method of linear multibody systems (MS-TMM) is developed to analyze the free
vibration of a multilevel beam, coupled by spring/dashpot systems attached to them in-span. The Euler-Bernoulli model is used for
the transverse vibration of the beams, and the spring/dashpot system represents a simplified model of a viscoelastic material. MS-
TMM reduces the dynamic problem to an overall transfer equation which only involves boundary state vectors. The state vectors at
the boundaries are composed of displacements, rotation angles, bending moments, and shear forces, which are partly known and
partly unknown, and end up with reduced overall transfer matrix. Nontrivial solution requires the coefficient matrix to be singular
to yield the required natural frequencies. This paper implements two novel algorithms based on the methodology by reducing the
zero search of the reduced overall transfer matrix’s determinate to a minimization problem and demonstrates a simple and robust
algorithm being much more efficient than direct enumeration. The proposal method is easy to formulate, systematic to apply, and
simple to code and can be extended to complex structures with any boundary conditions. Numerical results are presented to show
the validity of the proposal method against the published literature.
connected complex double-beam system in the case of simply and 𝐼 is the area moment of inertia. Beam mass per unit
supported beams. The classical modal expansion method was length 𝑚 = 𝜌𝐴, where 𝜌 is the material density and 𝐴 is the
applied to determine the dynamic responses of the beams beam cross-section area. Of course, any different materials,
due to arbitrarily distributed continuous loads. Vu et al. [6] different dimensions, or different boundary conditions could
presented an exact method for solving the vibration problem be considered in MS-TMM. The beams have the same length
of a damped double-beam system subjected to harmonic 𝐿 and are joined by the 𝑗-system of spring/dashpot located
excitation. The double-beam system consists of two identical at different positions (e.g., 𝑥1 and 𝑥𝑗 , where 𝑗 is the number
beams with the same boundary conditions on both sides. of spring/dashpot systems between two levels). 𝑘𝑦 and 𝑐 are
The beams are connected by a viscoelastic layer. Natural linear translational spring constant and damping coefficient,
frequencies and mode shapes of vibration of the system respectively, as shown in Figure 1.
are determined and the forced vibrations are investigated.
Gürgöze and Erol [7] determined the natural frequencies
of a clamped-free double-beam system carrying tip masses 3. MS-TMM Strategy in the Context of
to which several spring mass systems are attached across Free Vibration Characteristics
the span. However, there are only few contributions dealing
with the vibration of multibeam systems. That is, probably, 3.1. MS Topologies. According to the natural attribute of
the general vibration analyses of an elastically connected bodies, a complicated MS can be represented by various
multibeam system are complicated and laborious in view bodies (e.g., rigid bodies, elastic bodies, lumped masses,
of a large variety of possible combinations of boundary etc.) interconnected by hinges (e.g., spherical joints, sliding
conditions, and thus, the solution of the governing coupled joints, cylindrical joints, dampers, springs, etc.). In MS-
partial differential equations is difficult [8]. TMM, there are different topologies based on a certain set
Multibody system dynamics (MSD) has become an of modeling variables to formulate the dynamic equations of
important theoretical tool for wide engineering problems MS. Such topologies are chain, tree, closed loop (as illustrated
analysis in the world. Lots of methods of MSD have been in Figure 2 for reference), network, and so forth.
studied by many authors on theory and computational
method [9–15]. Professor Rui Xiaoting and his students have
been enlightened by the method of letting state vectors (SVs) 3.2. State Vector, State Variables, and Transfer Direction.
be transferred into classical transfer matrix method and The state vector (SV) at a connection point of MS is a
built up a new multibody dynamics method called “Transfer column vector denoting the mechanics state of this point.
Matrix Method of Linear Multibody Systems-MS-TMM” It includes the displacements of the point (including angu-
[16]. Using MS-TMM, the eigenvalue of linear multi-rigid- lar displacements) and the corresponding internal forces
flexible-body system is computed easily, the computational (including internal moments). Therefore, the SV is given
ill-condition is overcome, and the computational efficiency is by kinematics (displacements) and kinetics (internal forces)
increased. Over 20 years, MS-TMM has been developed and quantities, called state variables. For convention in this paper,
used widely in engineering applications. z with bold lowercase represents the SV in the physical
Motivated by the interesting study by Kukla [2] which coordinates and Z with bold capital represents the SV in
was published on the problem of the natural longitudinal the modal coordinates. Vibrations in space are described
vibrations of two rods coupled by many translational springs by displacement coordinates 𝑥, 𝑦, 𝑧 along the Cartesian axes
and by the two novel algorithms based on the new recursive and angular rotations 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 about Cartesian axes. Cutting
scanning approach presented by Bestle et al. [17], this paper forces and moments are given by 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 and 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 ,
presents a unique yet simple scenario of obtaining the respectively. Positive directions at input points are shown
exact free vibration characteristics of undamped/damped in Figure 3(a). Positive directions of forces and moments at
multilevel beam coupled elastically. The scenario developed output points (Figure 3(b)) are opposite due to the principle
in this paper is based on MS-TMM and Euler-Bernoulli beam of action equals reaction. In 3D case with 𝑛𝑠 = 12 (𝑛𝑠 is the
theory. number of state variables in the SV), the SVs in physical and
The text is organized as follows. The problem statement modal coordinates at the connection point 𝑝𝑖,𝑘 (where the
is presented in Section 2. In Sections 3 and 4, the general first subscript 𝑖 is the serial number of element for boundary
theorem brief of MS-TMM and problem solution scenario are end and the second subscript 𝑘 is the serial number of the
shown. In Section 5, some results calculated by MS-TMM and hinge element and 𝑘 = 0 for boundary end) are summarized
the other method are given which can validate the proposed in a vector, receptively:
method. The conclusions are presented in Section 6.
z𝑖,𝑘 physical coordinates
2. Problem Statement 𝑇
= [𝑥, 𝑦, 𝑧, 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 , 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 ]𝑖,𝑘 ,
The transversely vibrating system in a plane under considera- (1)
tion consists of multi-level parallel, elastic, and homogeneous Z𝑖,𝑘 modal coordinates
Euler-Bernoulli beam with general boundary conditions. 𝑇
Beam bending stiffness is 𝐸𝐼, where 𝐸 is the elastic modulus = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]𝑖,𝑘 .
Advances in Mechanical Engineering 3
y
General
boundary x Beam 1 m1 , (EI)1
conditions
ky c ··· ky c
Beam 2 m2 , (EI)2
ky c ··· ky c
..
x1 .
xj General
boundary
conditions
ky c ··· ky c
Beam m mm , (EI)m
Hinges
Loop
Bodies
Loop
Base
(a) (b) (c)
Figure 2: Multibody systems (a) chain, (b) tree, and (c) closed loop topology.
In case of 1D or 2D applications, the SV will be reduced direction shown in Figure 3, the transfer equation between the
as shown later. Defining a boundary point of the MS as component input and output is
the transfer end, the direction from all other boundary points
to the transfer end is called transfer direction. Along the z𝑛,0 = U𝑛 z𝑛−1,𝑛
transfer direction, the nodes entering into elements are called
inputs denoted by 𝐼 and the nodes leaving from elements are z𝑛−1,𝑛 = U𝑛−1 z𝑛−2,𝑛−1
called outputs 𝑂. z𝑛−2,𝑛−1 = U𝑛−2 z𝑛−3,𝑛−2 (2)
..
3.3. Transfer Equation, Transfer Matrix, Overall System Trans- .
fer Matrix, and Overall System State Vector. A vibrating MS
comprised of 𝑛-components, see Figure 4, is used as an z1,2 = U1 z1,0 .
example to show how to deduce the overall transfer equation
The constant matrix U𝑘 is the transfer matrix of the
and overall transfer matrix of the system. In order to describe
𝑘th component. Transfer matrices of basic components are
conveniently the idea, the chain topology is considered
considered as building blocks, which can be assembled
in the following. This vibrating system is comprised of 𝑛
together to provide the transfer matrix of the whole system
components and 𝑛 + 1 connection points. The SVs of the
according to the chain MS-TMM topology as follows:
boundary right extremity and other boundary left extremity
of the system are expressed as z𝑛,0 and z1,0 , respectively. 𝑛−1
Transfer direction of the system is always from another z𝑛,0 = Tz1,0 , where T = ∏U𝑛−𝑘 . (3)
boundary end to may call it the root. Following the transfer 𝑘=0
4 Advances in Mechanical Engineering
yI , qyI yO , −qyO
−𝜃xO , mxO
𝜃xI , −mxI
I xI , qxI O xO , −qxO
zI , qzI zO , −qzO
(a) (b)
Transfer direction
2 4 ··· n − 1 n 0 Root
0 1 3
Left end Right end
z1,0 zn,0
I 3 O
{
z2,3 z3,4
Figure 4: TMM-MS in the sense of chain topology. A vibrating system comprised of 𝑛-components with 𝑛 + 1 connections.
Rewrite (2) as from elimination of all columns of Uall (4c) associated zeros
in zall . For harmonic vibrations, solutions maybe written as
Uall 𝑛𝑠 ×(2×𝑛𝑠 ) zall (2×𝑛𝑠 )×1 = 0, (4a) z𝑖,𝑗 = Z𝑖,𝑗 𝑒𝜆𝑡 where 𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 , 𝜆𝑟 , 𝜆𝑖 ∈ R are the
eigenvalues. The real part (−𝜆𝑟 ) is related to the magnitude
where
of damping, where the imaginary part (𝜆𝑖 ) is related to the
𝑇 vibration frequency of the damped system. For undamped
z𝑇all = [z𝑇1,0 z𝑇𝑛,0 ] , (4b)
systems, 𝜆𝑟 = 0 and 𝜆𝑖 = 𝜔. Finally, Uall is only a function of
Uall = [T −I𝑛𝑠 ] . (4c) the unknown 𝜆 𝑖 of the system. For nontrivial solutions, the
Eigenfrequency equation
Herein, Uall is the overall system transfer matrix and zall is the
!
overall system state vector. Δ (𝜆) = det Uall = 0 (5)
3.4. Eigenfrequency Equation of the Whole MS. The overall has to be fulfilled. The natural frequencies of the system can
transfer equation (4a) only involves the boundary SVs, and now be computed.
the SVs at all other connection points do not appear. The
SVs at the boundary are composed of displacements, rotation 3.5. Beam Transfer Matrix. The full derivation of transfer
angles, moments, and shears, which are partly known and matrices for the Timoshenko and Euler-Bernoulli beams
partly unknown. For common boundary conditions, half of vibrating in a plane (with kinematics and kinetics’ SV defined
state variables of zall (4b) are zeros due to known constraints. as z = [𝑦, 𝜃𝑧 , 𝑚𝑧 , 𝑞𝑦 ]𝑇 ) may be found in [18], which is an
Thus, (4a) reduces to Uall zall = 0, where zall is composed of the open access article and the reader may download it from the
unknown state variables and Uall is a square matrix resulting Internet. However, for completeness, only the transfer matrix
Advances in Mechanical Engineering 5
for the Euler-Bernoulli beam will be presented, Figure 5. The where 𝑌 (𝑥) = 𝐴 1 cosh 𝛽𝑥 + 𝐴 2 sinh 𝛽𝑥
differential equation of a Euler-Bernoulli beam is
+ 𝐴 3 cos 𝛽𝑥 + 𝐴 4 sin 𝛽𝑥,
(6)
𝜕4 𝑦 𝜕2 𝑦
𝐸𝐼 4 + 𝑚 2
𝜕𝑥 𝜕𝑡 𝐴 1 , 𝐴 2 , 𝐴 3 , and 𝐴 4 are arbitrary constants, and 𝛽 =
√4 −𝑚𝜆2 /(𝐸𝐼). For the Euler-Bernoulli beam, the linearized
𝜕4 𝑌 (𝑥) 𝑚𝜆2 relations in modal coordinates Θ𝑧 = 𝑌 , 𝑀𝑧 = 𝐸𝐼𝑌 , and
= 0 𝑦 (𝑥, 𝑡) = 𝑌 (𝑥) 𝑒𝜆𝑡 4
+ 𝑌 (𝑥) = 0,
→ 𝜕𝑥 𝐸𝐼 𝑄𝑦 = 𝑀𝑧 maybe added to end up with the transfer relation:
The coefficient vector a = [𝐴 1 , 𝐴 2 , 𝐴 3 , 𝐴 4 ]𝑇 summarizes (1) Break up the complicated MS into components with
the unknown constants to be adopted to boundary condi- simple dynamic properties, which can be expressed in matrix
tions. At input end Z𝐼 (𝑥 = 0), we get Z𝐼 = [B(0)]a. Thus, form and SVs (for each component, it is possible to obtain
the coefficient vector can be expressed as a = [B(0)]−1 Z𝑂 and the close form expression of the transfer matrix giving the
substituting it into (7) for the beam output end at 𝑥 = 𝑙, one displacements and the forces applied to one extremity to
gets the displacements and forces applied to the other extremity).
In other words, on the component level, the governing
Z𝑂 = [B (𝑙)] a = [B (𝑙)] [B (0)]−1 Z𝐼 = UZ𝐼 , (8a) partial and ordinary differential or algebraic equations are
transformed to algebraic transfer equations, where the output
where state results from a product of the input state and an
element specific transfer matrix. These component matrices
U = B (𝑙) B−1 (0) are considered as building blocks. In fact, the transfer matrix
of such components needs not to be rededuced but may be
𝑇 𝑈 𝑉
[ 𝑆
taken directly from a transfer matrix library.
𝛽 𝐸𝐼𝛽 2 𝐸𝐼𝛽3 ]
[ ] (2) Following the transfer direction that has been
[ 𝑇 𝑈 ]
[ ] designed already by the analyst and according to the topology
[ 𝛽𝑉 𝑆 ]
=[
[
𝐸𝐼𝛽 𝐸𝐼𝛽2 ] ,
]
of the MS, these component transfer matrices are then
[ 𝑇 ] assembled and end up with a system of linear algebraic
[𝐸𝐼𝛽2 𝑈 𝐸𝐼𝛽𝑉 𝑆 ]
[ 𝛽 ]
equations called the overall transfer equation.
[ ] (3) Substitute the boundary conditions into the overall
3 2
[ 𝐸𝐼𝛽 𝑇 𝐸𝐼𝛽 𝑈 𝛽𝑉 𝑆 ] transfer equation to construct the eigenfrequency equation.
Consequently, the vibration characteristics such as frequen-
𝑐ℎ + 𝑐 𝑠ℎ + 𝑠
𝑆= , 𝑇= , cies can be deduced as the roots of a transcendental equation.
2 2 Due to narrow couples of natural frequencies, the classical
𝑐ℎ − 𝑐 𝑠ℎ − 𝑠 zero search method is likely to fail. However, a new recursive
𝑈= , 𝑉= , scanning approach for minima of the absolute values of the
2 2
determinant shows much more efficiency and reliability than
𝑐ℎ = cosh (𝛽𝑙) , 𝑠ℎ = sinh (𝛽𝑙) , direct enumeration.
𝑐 = cos (𝛽𝑙) , 𝑠 = sin (𝛽𝑙)
4. Problem Solution Scenario
(8b)
Figure 6 illustrates the suggested scenario for the solution
is the transfer matrix of the Euler-Bernoulli beam compo- of the problem statement and as follows. There are (1 : 𝑚)
nent. multi-level beam. Each beam level is divided into (1 : n)
components, which have (0 : 𝑛 + 1) connection points
3.6. Summary. In the context of free vibration characteristic, and coupled with another beam level by a system (say 1 :
the general strategy of linear MS-TMM, in summary, is as 𝑗) of a viscoelastic material (modeled as a spring/dashpot
follows. system). The massless dummy body as shown in Figure 6
6 Advances in Mechanical Engineering
qyI
𝜃zO mzO
mzI 𝜃zI m, L, EI
qyO
yI yO
i−1 i n−1 n
0 1 2
A G
U1,1 U1,2 ··· U1,i ··· U1,n−1 U1,n 1
1, 1 j, 1
Spring/dashpot component
(e.g., 𝐴, 𝐵, 𝐶, and so on) is a connection point between Figure 7(a) sketches the first step. According to the continuity,
the two beam segment components and spring/dashpot 𝑗 has identical displacements, angles, and moments at input
system(s). The main key of MS-TMM is transferring the and output:
SV from one component to another following the general
transfer equation (2). For the beam segment component, 𝑌𝑂 = 𝑌𝐼 , Θ𝑧𝑂 = Θ𝑧𝐼 , 𝑀𝑧𝑂 = 𝑀𝑧𝐼 . (9a)
the transfer matrix is available (8b), while it is not for
spring/dashpot system at the connection point between two And from the force analysis shown in Figure 7(b), the
or multi-level beam in this paper. However, it is based on the spring/dashpot force 𝐹 changes the shear forces:
kinematics and kinetics of the spring/dashpot to formulate 𝑄𝑦𝑂 = 𝑄𝑦𝐼 + 𝐹. (9b)
the transfer matrix and it needs two steps. First, as an
example, let us consider a connection point 𝑗 between two Equations (9a) and (9b) in the view of the SV as a matrix form
beam segment components and system of spring/dashpot
may be grounded or connected to another connection point. Z𝑂 = Z𝐼 + e4 𝐹. (9c)
Advances in Mechanical Engineering 7
ky c
YI YO
ky c
I j O yI
Z1,I 1 Z1,O
G Level
1
F
Fsys
ky𝑗,1
cj,1
Fsys
Z1,I 1 F
Z1,O Level
Level
A 1 C 2
Z2,I Z2,O
F F
Fsys
Fsys
c1,1 c1,m−1
Fsys Fsys
F F
Level Level
B 2 V 3
Z2,I Z2,O Z3,I Z3,O
(d) (e)
Figure 7: (a) Two beam segments and spring/dashpot system connected at massless dummy body, (b) state vectors (SV) at the massless
dummy body 𝑗-connection point, (c) force analysis of a spring/dashpot system, (d) a system 1,1 connected two levels 1-2 through two
connection points 𝐴 and 𝐵 shown in Figure 6, and (e) two systems 𝑗, 1 and 1, 𝑚−1 connected with multi-level (1-2-3) through three connection
points 𝐺, 𝐶, and 𝑉 shown in Figure 6.
8 Advances in Mechanical Engineering
𝑇
e4 = [0 0 0 1] is a unit vector assigning 𝐹 to the points 𝐺, 𝐶, and 𝑉 shown in Figures 6 and 7(e) can be
transfer equation of the massless connection body. The obtained similarly:
spring compression and damping forces are given as (see
Figure 7(c))
U𝑑𝑗,1 U𝑐𝑗,1 0
Z [ ] Z
𝐹spring = 𝑘𝑦 (𝑌𝐼 − 𝑌𝑂) = 𝑘𝑦 Δ𝑌 = 𝑘𝑦 e𝑇1ΔZ { }
{ 1
} [ 𝑐 𝑑 𝑐 ] { 1}
{ }
{ Z2 } = [ 𝑗,1 𝑗,1/1,𝑚−1 1,𝑚−1 ]
[ U U U
] { Z2} ,
{ } [ ] { }
) 𝑦 = 𝑌𝑒𝜆𝑡 = 𝑐𝜆 (𝑌𝐼 − 𝑌𝑂) = 𝑐𝜆e𝑇1 ΔZ [ ]
𝐹dashpot = 𝑐 (𝑦𝐼̇
− 𝑦𝑂̇ {Z3 }𝑂 0 U𝑐1,𝑚−1 U𝑑1,𝑚−1 {Z3 }𝐼
→ [ ]
→ 𝐹sys = (𝑘𝑦 + 𝑐𝜆) e𝑇1ΔZ.
where U𝑑𝑗,1 ≡ I − D𝑗,1 , U𝑐𝑗,1 ≡ D𝑗,1
(10) (12d)
0 0 0 0
[ 0 0 0 0] (11)
where D = (𝑘𝑦 + 𝑐𝜆) e4 e𝑇1 =[
[
]. Now we are in the position to end the solution scenario.
0 0 0 0] From the MS-TMM topology point of view, the system shown
[ 𝑘𝑦 + 𝑐𝜆 0 0 0] in Figure 1 or equivalently Figure 6 can be modeled as a mul-
tichain, following the similar procedures mentioned above
For system 1,1 connected two levels 1-2 through two in Section 3.3 to determine the overall transfer equation and
connection points 𝐴 and 𝐵 shown in Figure 6, the second step ending up with determinant solution of the reduced overall
is how to deduce the transfer equation between the input and transfer matrix.
output related to these two levels. For the connection point 𝐴
(see Figure 7(d)) and using (11) with 𝐹 = 𝐹sys ,
5. Numerical Examples
Z1,𝑂 = Z1,𝐼 + D1,1 (Z2,𝐼 − Z1,𝐼 ) Before the free vibration analysis of a multi-level elastic
beam coupled by a spring/dashpot system(s) is performed,
(I − D1,1 )Z1,𝐼 + D
= ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ ⏟⏟Z2,𝐼
⏟⏟⏟⏟⏟
1,1 the reliability of the suggested scenario and the Matlab
U𝑑1,1 U𝑐1,1
(12a)
computer program developed for this paper are confirmed
by comparing the present results with those obtained from
Z1,𝑂 = U𝑑1,1 Z1,𝐼 + U𝑐1,1 Z2,𝐼 . the existing literature.
y Transfer direction
x m1 , (EI)1
Beam 1 1 3 5
ky1,1 ky2,1 0 0 Level 1
Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0
m2 , (EI)2 Chain 1
Beam 2 ky1,1 ky2,1
2 4
6 7 8
x1 0 Level 2
0 Z Z6,2 Z2,7 Z7,4 Z4,8 Z8,0
x2 = 1 − x1 6,0 Chain 2
x1 L − 2x1 x1
L
Figure 8: (a) Free-free and clamped-clamped beams connected with two linear translational springs, (b) chain TMM-MS: State vectors and
transfer direction of the suggested solution scenario.
Table 1: Common boundary conditions for a beam vibrating in a grids 𝑁𝑥0 = 500, absolute precision tolerance 𝜀 = 10−6 ).
plane. Figure 9 shows log10 |Δ| obtained from the fMin1D algorithm
versus the first 10 dimensionless frequency values (𝜔 =
Support type Zero terms Nonzero terms
Fixed 𝑌, Θ𝑧 𝑄𝑦 , 𝑀𝑧 √4 (𝜆𝑖 ≡ 𝜔)2 𝑚𝐿4 /𝐸𝐼) for 𝑘𝑦 = 𝑘𝑦 = 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ) and
1,1 2,1
ky1,1 ky2,1
1010
x1
x2 = 1 − x1
5
10
log10 |Δ|
100
10−5
10−10
0 2 4 6 8 10 12 14 16
𝜔
106 108
log10 |Δ|
log10 |Δ|
104 106
102
104
10.6 10.7 10.8 10.9 11 11.1 11.2 11.3 11.4 14 14.05 14.1 14.15 14.2
𝜔
𝜔
Figure 9: fMin1D function determinant of a system consisting of free-free and clamped-clamped beams coupled with two springs. 𝑘𝑦 =
100(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.3 m.
16 16
14 14
12 12
10 10
8 8
𝜔
𝜔
6 6
4 4
2 2
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m) x1 (m)
(a) (b)
Figure 10: Dimensionless frequency parameter 𝜔 values as a function of the spring locations (𝑥1 = 0 → 0.5 m) for a system consisting of
free-fee and clamped-clamped beams coupled by two springs. (a) 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ); (b) 𝑘𝑦 = 1000(𝐸𝐼/𝐿3 ).
Transfer direction
1 3 5
0 0 Level 1
Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0 Chain 1
ky1,1 2 ky2,1 4
c1,1 c2,1
6 8 10
0 0 Level 2
Z6,0 Z6,2 Z2,8 Z8,4 Z4,10 Z10,0 Chain 2
or or or or
Z6,7 Z7,8 Z8,9 Z9,10
ky1,2 7 ky2,2 9
c1,2 c2,2
11 12 13
0 0 Level 3
Z11,0 Z11,7 Z7,12 Z12,9 Z9,13 Z13,0 Chain 3
x1 L − 2x1 x1
Figure 11: Chain TMM-MS: state vectors and transfer direction of pinned-pinned multi-level beam connected by spring/dashpot systems.
1010
105
100
10−5
log10 |Δ|
10−10
L 10−15
10−20
10−25
10−30
0 1 2 3 4 5 6 7 8 9 10 11 12
𝜔
Figure 12: fMin1D function determinant to evaluate the lowest three 𝜔 for undamped, uncoupled pinned-pinned three-beam.
Advances in Mechanical Engineering 13
ky1,1 ky2,1
10
ky1,2 ky2,2 9
8
x1 7
x2 = 1 − x1 6
5
𝜔
L
4
3
9.54
2
9.52
1
9.5 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
9.48
𝜔
x1 (m)
9.46 (a) ky = 25(EI/L3 )
9.44
9.42 1010
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m)
105
(a) ky = 25(EI/L3 )
log10 |Δ|
6.6
100
6.55
6.5
6.45 10−5
6.4
𝜔
6.35 10−10
0 2 4 6 8 10 12
6.3
𝜔
6.25
(b) ky = 25(EI/L3 ) and x1 = 0.25 m
6.2
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m)
(a) ky = 25(EI/L3 )
102
100 100
10−2 10−2
log10 |Δ|
log10 |Δ|
10−4 10−4
10−6 10−6
Figure 13: (a) The lowest three 𝜔 values as a function of the spring systems locations (𝑥1 = 0 → 0.5 m) for undamped, coupled pinned-pinned
three-beam; (b) fMin1D function determinant to evaluate the dimensionless frequency parameter 𝜔 for 𝑘𝑦 = 25(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.25 m.
14 Advances in Mechanical Engineering
ky1,1 ky2,1
c1,1 c2,1
ky1,2 ky2,2
c1,2 c2,2
x1
x2 = 1 − x1
2000 1000
1500 800
600
1000
400
500
200
𝜆r = 𝛿
𝜆r = 𝛿
0 0
−500 −200
−400
−1000
−600
−1500
−800
−2000 −1000
−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600
𝜆i = 𝜔 (rad/s) 𝜆i = 𝜔 (rad/s)
(a) (b)
500 600
400
500
300
200 400
100
300
𝜆r = 𝛿
𝜆r = 𝛿
0
200
−100
−200 100
−300
0
−400
−500 −100
−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600
i i
𝜆 = 𝜔 (rad/s) 𝜆 = 𝜔 (rad/s)
(c) (d)
Figure 14: fMin2D first step: damping (𝜆𝑟 = 𝛿) range scanning of a specific frequency band (𝜆𝑖 = 𝜔 ≤ 1500 rad/sec) for damped, coupled
pinned-pinned three-beam, (a) [−2000, 2000], (b) [−1000, 1000], (c) [−500, 500], and (d) [−100, 600].
Advances in Mechanical Engineering 15
Table 2: Chain MS-TMM eigenvalues results of damped, coupled Journal of Sound and Vibration, vol. 269, no. 1-2, pp. 431–438,
pinned-pinned three-beam. 2004.
[8] M. Abu-Hilal and N. Beithou, “Free transverse vibrations of a
𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 Chain MS-TMM results using fMin2D
triple-beam system,” Journal of Mechanical Engineering, vol. 58,
𝜆𝑖 (rad/sec)
no. 1, pp. 30–50, 2007.
𝜆1 −2.463512 × 102 + 4.010226 × 101 𝑖 [9] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,
𝜆2 −2.074369 × 10−9 + 1.462698 × 102 𝑖 Germany, 1990.
𝜆3 −7.409510 × 101 + 1.652322 × 102 𝑖 [10] W. Schiehlen, “Multibody system dynamics: roots and perspec-
𝜆4 −4.577982 × 102 + 4.615502 × 102 𝑖 tives,” Multibody System Dynamics, vol. 1, no. 2, pp. 149–188,
1997.
𝜆5 −2.074369 × 10−9 + 5.850790 × 102 𝑖
[11] J. Wittenburg, Dynamics of Systems of Rigid Bodies, Edited by:
𝜆6 −1.484833 × 102 + 5.861790 × 102 𝑖 B. G. Teubner, Stuttgart, Germany, 1977.
𝜆7 −2.159398 × 102 + 1.243120 × 103 𝑖 [12] J. Wittenburg, Dynamics of Multibody Systems, Springer, Berlin,
𝜆8 −7.448404 × 101 + 1.311750 × 103 𝑖 Germany, 2nd edition, 2008.
𝜆9 −2.074369 × 10−9 + 1.316428 × 103 𝑖 [13] A. A. Shabana, Dynamics of Multibody Systems, Cambridge
University Press, New York, NY, USA, 3rd edition, 2010.
[14] A. A. Shabana, Computational Dynamics, John Wiley & Sons,
investigations of more complex multibody systems of this New York, NY, USA, 3rd edition, 2010.
type with rigid bodies due to simplicity in the formulation [15] A. A. Shabana, “Flexible multibody dynamics: review of past
of the transfer equation, being systematic to apply, and being and recent developments,” Multibody System Dynamics, vol. 1,
easy to program. no. 2, pp. 189–222, 1997.
[16] X. Rui, L. Yun, Y. Lu, B. He, and G. Wang, Transfer Matrix
Method of Multibody System and Its Application, Science Press,
Conflict of Interests Beijing, China, 2008, (Chinese).
The authors declare that there is no conflict of interests [17] D. Bestle, L. K. Abbas, and X. Rui, “Recursive eigenvalue
search algorithm for transfer matrix method of linear flexible
regarding the publication of this paper.
multibody systems,” Multibody System Dynamics, 2013.
[18] L. K. Abbas, M. J. Li, and X. Rui, “Transfer matrix method for the
Acknowledgments determination of the natural vibration characteristics of realistic
thrusting launch vehicle—part I,” Mathematical Problems in
The research was supported by the Research Fund for Engineering, vol. 2013, Article ID 764673, 16 pages, 2013.
the Doctoral Program of Higher Education of China
(20113219110025), the Natural Science Foundation of China
Government (11102089), and the Program for New Century
Excellent Talents in University (NCET-10-0075).
References
[1] S. Inceoğlu and M. Gürgöze, “Bending vibrations of beams
coupled by several double spring-mass systems,” Journal of
Sound and Vibration, vol. 243, no. 2, pp. 370–379, 2001.
[2] S. Kukla, “Free vibration of the system of two beams connected
by many translational springs,” Journal of Sound and Vibration,
vol. 172, no. 1, pp. 130–135, 1994.
[3] S. Kukla and B. Skalmierski, “Free vibration of a system
composed of two beams separated by an elastic layer,” Journal
of Theoretical and Applied Mechanics, vol. 32, no. 3, pp. 581–590,
1994.
[4] Z. Oniszczuk, “Free transverse vibrations of elastically con-
nected simply supported double-beam complex system,” Jour-
nal of Sound and Vibration, vol. 232, no. 2, pp. 387–403, 2000.
[5] Z. Oniszczuk, “Forced transverse vibrations of an elastically
connected complex simply supported double-beam system,”
Journal of Sound and Vibration, vol. 264, no. 2, pp. 273–286,
2003.
[6] H. V. Vu, A. M. Ordóñez, and B. H. Karnopp, “Vibration of a
double-beam system,” Journal of Sound and Vibration, vol. 229,
no. 4, pp. 807–822, 2000.
[7] M. Gürgöze and H. Erol, “On laterally vibrating beams carrying
tip masses, coupled by several double spring-mass systems,”
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 365265, 9 pages
http://dx.doi.org/10.1155/2014/365265
Research Article
Case Study on Human Walking during Wearing a Powered
Prosthetic Device: Effectiveness of the System ‘‘Human-Robot’’
Svetlana Grosu,1 Pierre Cherelle,1 Chris Verheul,2 Bram Vanderborght,1 and Dirk Lefeber1
1
Department of Mechanical Engineering, Vrije Universiteit Brussel, 1050 Brussels, Belgium
2
SayField International, Broeksloot 10, 3474 HP Zegveld, The Netherlands
Copyright © 2014 Svetlana Grosu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
It is known that force exchanges between a robotic assistive device and the end-user have a direct impact on the quality and
performance of a particular movement task. This knowledge finds a special reflective importance in prosthetic industry due to
the close human-robot collaboration. Although lower-extremity prostheses are currently better able to provide assistance as their
upper-extremity counterparts, specific locomotion problems still remain. In a framework of this contribution the authors introduce
the multibody dynamic modelling approach of the transtibial prosthesis wearing on a human body model. The obtained results are
based on multibody dynamic simulations against the real experimental data using AMP-Foot 2.0, an energy efficient powered
transtibial prosthesis for actively assisted walking of amputees.
150
Torque (Nm)
100 HO
50
FF
0 TO HC
−25 −20 −15 −10 −5 0 5 10 15
Angle (deg)
AMP-Foot 2 ankle angle data
AMP-Foot 2 ankle angle average
D. winter data
(a) (b)
Figure 5: The AMP-Foot 2.0 model composition and DoF. 3.2. Modelling and Simulation of Human Walking during
Wearing the AMP-Foot 2.0 Prosthesis. There are specialized
commercially available simulation tools that can be used for
analysis of human walking, such as AnyBody [23], OpenSim
STEP (𝑥, 𝑥0 , ℎ0 , 𝑥1 , ℎ1 ). It has continuous first derivatives. Its
[24], SIMM [25], and LifeMod [26]. However, these tools
second derivatives are discontinuous at 𝑥 = 𝑥0 and 𝑥 = 𝑥1 ,
cannot be used for analyzing human-robot interactions,
where 𝑥 is the independent variable (time, in present model),
which is becoming an essential requirement for modeling and
𝑥0 is a real variable that specifies the 𝑥 value at which the
simulation of robotic systems as their physical interaction
STEP function begins, 𝑥1 is a real variable that specifies the 𝑥
with humans becomes more complex. In general, the model-
value at which the STEP function ends, ℎ0 is the initial value
ing and simulation phase of the development of such systems
of the step, and ℎ1 is the final value of the step.
is becoming demanding. Since the robot and robot-human
The arrows, that can be observed in Figure 6, are repre-
interactions increase in complexity, the simulation with a
senting the actuation and tension forces during the simula-
single simulation tool is not effective anymore. Therefore,
tion.
in this research work the combination of MSC Adams with
Due to cables nonlinear geometric properties and a
LifeMode plug-in was used as the most appropriate.
complex static and dynamic behaviour, the challenge was to
model the cable segments which play an important role in
a force transmission system of the AMP-Foot 2.0. The cable 3.2.1. Methods. The creation of human models begins by
transmission system was modelled using TKC toolbox feature generating a base human segment set, followed by joints,
and the tension forces inside of the system were obtained [22]. soft tissues, and contact elements between the model and the
Advances in Mechanical Engineering 5
(b) (c)
Figure 6: (a, b) Simulation of AMP-Foot 2.0 behaviour. (c) Cable properties GUI.
(a) (b)
environment. In order to study the influences of AMP-Foot lower body (two legs and pelvis) which is rigidly attached
2.0 device on normal overground walking the mechanical to the right extremity AMP-Foot 2.0 prosthesis. Once the
model of the human body was built using the Lifemodeler segments of the lower body are established, joints are created
plug-in in framework of MSC Adams environment. This between the segments. Along an inverse-dynamics simu-
combination of tools supports the analysis of the “human- lation, joints learn angulation patterns while the model is
robot system” effectiveness and mutual interaction. The being driven by the motion capture data. The nominal joints
considered model, see Figure 8, includes a model of human’s stiffness for both legs at the hip, knee, and the ankle is
6 Advances in Mechanical Engineering
20.0
10.0
Angle (deg)
0.0
−10.0
−20.0
−30.0
0.0 0.5 1.0 1.5 2.0
Time (s)
15.0
50000 10.0
Angle (deg)
40000 5.0
30000 0
20000 −5.0
10000 −10.0
0.0
0.0 0.5 1.0 1.5 2.0 −15.0
Time (s) −20.0
−25.0
Torque right Hip 0 10 20 30 40 50 60 70 80 90 100
Torque right Knee Stride (%)
Torque ankle AMP Foot 2.0
Figure 12: Ankle angle comparison: red line: simulation of human
Figure 10: Torque values for right limb. model during wearing the AMP-Foot 2.0; blue line: simulation of
human model without AMP-Foot 2.0; green line: real experiment
data of subject during wearing the AMP-Foot 2.0 device.
60.0
50.0
40.0
Angle (deg)
contact definition (which is not so complex as in case of real [6] H. Bateni and S. J. Olney, “Kinematic and kinetic variations of
foot-ground contact). below-knee amputee gait,” Journal of Prosthetics and Orthotics,
vol. 14, no. 1, pp. 2–10, 2002.
[7] D. A. Winter, “Biomechanical motor patterns in normal walk-
5. Conclusions ing,” Journal of Motor Behavior, vol. 15, no. 4, pp. 302–330, 1983.
The main focus in the framework of the presented study [8] A. H. Hansen, D. S. Childress, S. C. Miff, S. A. Gard, and K.
P. Mesplay, “The human ankle during walking: implications for
was to inspect how the attached trans-tibial prosthesis will
design of biomimetic ankle prostheses,” Journal of Biomechan-
influence normal healthy human walking kinematics and ics, vol. 37, no. 10, pp. 1467–1474, 2004.
how close the AMP-Foot 2.0 working behaviour comes to
[9] S. K. Au, P. Dilworth, and H. Herr, “An ankle-foot emulation
a real human foot subsystem. The effectiveness of human- system for the study of human walking biomechanics,” in
robot interaction was analyzed during the tests with a real Proceedings of the IEEE International Conference on Robotics and
subject and in virtual environment. The authors admit that it Automation (ICRA ’06), pp. 2939–2945, Orlando, Fla, USA, May
is difficult to compare individuals having a different walking 2006.
pattern, even if for simulation a subject with the same physical [10] D. A. Winter and S. E. Sienko, “Biomechanics of below-knee
parameters as the one tested in real experiment was consid- amputee gait,” Journal of Biomechanics, vol. 21, no. 5, pp. 361–
ered. Moreover, the simulation comparison (subject walking 367, 1988.
with and without AMP-Foot 2.0) is almost impossible to [11] G. K. Klute, J. Czerniecki, and B. Hannaford, “Development
arrange in real life. This fact that the simulation results more of powered prosthetic lower limb,” in Proceedings of the 1st
interesting and significant. National Meeting, Veterans Affairs Rehabilitation R&D Service,
Thanks to the built human-prosthesis model further Washington, DC, USA, October 1998.
improvements on prototype design parameters can be [12] S. K. Au, Powered ankle-foot prosthesis for the improvement of
achieved and the prototype can be adapted to certain subject’s amputee walking economy [Ph.D. thesis], MIT Department of
individual properties. The result data demonstrate that AMP- Mechanical Engineering, Cambridge, Mass, USA, 2007.
Foot 2.0 prosthesis can successfully replace the human ankle- [13] J. K. Hitt, R. Bellman, M. Holgate, T. G. Sugar, and K.
foot subsystem, provided some individual adjustments are W. Hollander, “The SPARKy (spring ankle with regenerative
kinetics) project: design and analysis of a robotic transtibial
made.
prosthesis with regenerative kinetics,” in Proceedings of the 2007
ASME International Design Engineering Technical Conferences
Conflict of Interests and Computers and Information in Engineering Conference
(IDETC/CIE ’07), pp. 1587–1596, Las Vegas, Nev, USA, Septem-
The authors, Grosu Svetlana, Pierre Cherelle, Chris Verheul, ber 2007.
Bram Vanderborght, and Dirk Lefeber, declare that there is no [14] R. Versluys, G. Lenaerts, M. Van Damme et al., “Successful
conflict of interests regarding the publication of this paper. preliminary walking experiments on a transtibial amputee
fitted with a powered prosthesis,” Prosthetics and Orthotics
International, vol. 33, no. 4, pp. 368–377, 2009.
Acknowledgments [15] R. Versluys, P. Beyl, M. Van Damme, A. Desomer, R. Van Ham,
and D. Lefeber, “Prosthetic feet: state-of-the-art review and
This work has been partially supported by the European
the importance of mimicking human anklefoot biomechanics,”
Commission 7th Framework Program as part of the CORBYS Disability and Rehabilitation: Assistive Technology, vol. 4, no. 2,
(Cognitive Control Framework for Robotic Systems) project pp. 65–75, 2009.
under Grant Agreement no. FP7 ICT-270219. [16] P. Cherelle, A. Matthys, V. Grosu, B. Vanderborght, and D.
Lefeber, “The AMP-Foot 2.0: Mimicking intact ankle behavior
References with a powered transtibial prosthesis,” in Proceedings of the
IEEE International Conference on Biomedical Robotics and
[1] O. Khatib, K. Yokoi, O. Brock, K. Chang, and A. Casal, “Robots Biomechatronics, pp. 544–549, June 2012.
in human environments: basic autonomous capabilities,” Inter- [17] G. van Oort, R. Carloni, D. J. Borgerink, and S. Stramigioli,
national Journal of Robotics Research, vol. 18, no. 7, pp. 684–696, “An energy efficient knee locking mechanism for a dynamically
1999. walking robot,” in Proceedings of the IEEE International Confer-
[2] A. de Santis, B. Siciliano, A. de Luca, and A. Bicchi, “An atlas ence on Robotics and Automation, pp. 2003–2008, May 2011.
of physical human-robot interaction,” Mechanism and Machine [18] P. Cherelle, V. Grosu, A. Matthys, B. Vanderborght, and D.
Theory, vol. 43, no. 3, pp. 253–270, 2008. Lefeber, “Design and validation of the ankle mimicking pros-
[3] M.-J. Hsu, D. H. Nielsen, S.-J. Lin-Chan, and D. Shurr, “The thetic (AMP-) foot 2.0,” Transaction on Neural Systems and
effects of prosthetic foot design on physiologic measurements, Rehabilitation Engineering, 2013.
self-selected walking velocity, and physical activity in people [19] http://www.youtube.com/watch?v=tXLQq9a5kyI.
with transtibial amputation,” Archives of Physical Medicine and [20] J. G. de Jalón and E. Bayo, Kinematic and Dynamic Simulation
Rehabilitation, vol. 87, no. 1, pp. 123–129, 2006. of Multibody Systems: The Real-Time Challenge, Springer, New
[4] A. Esquenazi and R. DiGiacomo, “Rehabilitation after amputa- York, NY, USA, 1994.
tion,” Journal of the American Podiatric Medical Association, vol. [21] http://www.mscsoftware.com/.
91, no. 1, pp. 13–22, 2001. [22] S. Grosu, P. Cherelle, C. Verheul, B. Vanderborght, and D.
[5] S. Ron, Prosthetics and Orthotics: Lower Limb and Spinal, Lefeber, “AMP-Foot 2.0 prosthesis dynamic behavior, prelim-
Lippincott Williams & Wilkins, Baltimore, Md, USA, 2002. inary computational multibody dynamics simulation results,”
Advances in Mechanical Engineering 9
Research Article
Controller Parameters Tuning Based on Transfer Matrix Method
for Multibody Systems
Copyright © 2014 Hossam Hendy et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Transfer matrix method for multibody systems (MS-TMM) is a rife method to multi-rigid-flexible-body systems dynamics model
deduction due to that there are no needs to establish the global dynamics equations of the system. Its basic idea is transferring a
state vector between the body input(s) and output(s); this idea is close to the linear theories in control analysis and design. In this
paper, three controllers’ parameters tuning techniques for the proposed system model using MS-TMM are utilized; one technique is
applied to get the stability regions via the frequency response of MS-TMM derived model. Another technique considers a classical
PID controller design through the analysis of step input response of the system, and the last technique can be applied in both time
and frequency domains if the model has a known mathematical model. A car suspension system is considered to represent modeling
and tuning problems. In-depth study of MS-TMM with control techniques and defining the controllers’ parameters stability regions
provide an opportunity to formulate a relationship between MS-TMM and control design for novel control applications due to the
powerful strength of MS-TMM dealing with more complex problems of the controlled multibody systems.
1. Introduction based on PID control and [5] indicated that more than 97%
of regulatory controllers utilize the PID algorithm. But the
With the increments of complexity of multibody systems tuning of the controller gains is a problem because many
and the development of their design and control methods, industrial models suffer some burdens such as nonlinearities,
the need for more elegant formulations of the equations higher order, and time delay [6]. The basic idea of MS-TMM
of motion becomes an issue of paramount importance. is transferring the state vector between the body input(s)
Many methods and theories for developing the model of and output(s); this idea is close to linear theories in control
the multibody system dynamics and control are presented analysis and design. The control and feedback variables could
for such reasons. In transfer matrix method for multibody be considered in the transfer equation of MS-TMM. The
systems (MS-TMM) there are two cases to deal with control tuning of the controller parameters is a great issue to get a
element, one is to express the control force with state of stabilized system. However Ziegler-Nichols technique is one
system of previous time such as the delay control, and the of the simplest techniques of PID tuning; there are several
second is that the control force is relative to present state of disadvantages, such as many trials are needed to find the
system, such as real-time control systems. Rui et al. have trials stability limits and also not all tuning values work efficiently
to develop new controlled systems depending on deriving for all situations [7–10]. Shamsuzzoha and Skogestad (2010)
the dynamics equations using MS-TMM [1, 2]. Bestle et al. developed a new procedure for PI/PID tuning method in
reformulated the car suspension system using MS-TMM as closed-loop mode, only one closed-loop step test is required
similar as classical control theory’s structural diagram [3]. to deduce PI controller values. In this method also it is simple
Proportional-Integration-Differentiation (PID) control- to obtain the PID tuning parameters in one step [11, 12]. This
ler is prevalent in industry applications. Reference [4] indi- paper is organized as follows: the proposed system is modeled
cated that more than 90% of feedback control loops are by MS-TMM including the control law in Section 2. The
2 Advances in Mechanical Engineering
0 Disturbance X4,0
𝒁4,0
m1
𝑼g1
K2 fc
𝑼m1
C2
𝑼g
m3
x fc + 𝒁2,1 Output X1,0
r=0 + 𝑼m
C(s) 𝑬c
+ 𝒁1,0
K4 − X1,0
𝑬m
0
Figure 1: Dynamic model of controlled car. Figure 2: The structural diagram of the controlled system [3].
1 0
where U𝑗 = [ 𝑚𝑗 Ω2 1 ] and Ω is the external excitation
design of PI controller values stability region that achieves a
specified gain and phase margins, in addition to tuning the frequency.
controller parameters using different techniques, is presented For the linear spring 4, the transfer equation is
in Section 3. Simulation and results of car active suspension
Z4,𝑂 = U4 Z4,𝐼 (2)
system model with the designed controllers are provided in
Section 4. Conclusions are offered in Section 5.
with transfer matrix U4 = [ 10 −1/𝐾
1
4 ].
Magnitude (dB)
0
Name Value Company −20
𝑚3 50 kg −40
𝑚1 250 kg −60
−80
𝐾2 16 kN/m −100
𝐾4 160 kN/m 0
𝐶2 −45
Phase (deg)
1500 N⋅s/m −90
−135
−180
which finally ends up with −225
−270
−1
Z4,0 = (U1 U2 U3 U4 ) [I2 + 𝐶 (𝑖Ω) U1 E𝑐 E𝑚 ] Z1,0 . (8) 100 101 102 103
Frequency (rad/s)
This can be rewritten considering the boundary condition of
Classical
the system Z1,0 = [ 𝑋0 ]1,0 , Z4,0 = [ 𝑄𝐴𝑥 ]4,0 as MS-TMM
140 30
Stability boundary for 25
120
M = 1 and 𝜑 = 0
20
100 Stability boundary for
15
M = 1 and 𝜑 = 0
KI
80
10
KI
Magnitude (dB)
5
Figure 4: Stability region for frequency range Ω → [0, 56.7]. 0
−5
−10
−15
−20
−25
To determine the missing expressions in (15), (16), and (17), 0
the transfer function in (12) can be divided into odd and
Phase (deg)
−45
even portions for both numerator and denominator, and −90
−135
after the comparison with (14), the following variables can be −180
substituted as −225
10−1 100 101
Frequency (rad/s)
𝑁𝑒 (−Ω2 ) = 𝐾1 𝐾3 , 𝑁𝑜 (−Ω2 ) = 𝐾1 𝐶3 ,
PI controller
𝐷𝑒 (−Ω2 ) = 𝑚2 𝑚4 Ω4 Uncontrolled
(18) Figure 6: Bode plot for the controlled and uncontrolled system.
− (𝐾1 𝑚4 + 𝐾3 𝑚2 + 𝐾3 𝑚4 ) Ω2 + 𝐾1 𝐾3 ,
90 1.6
80 1.4
70
Phase margin (deg)
1.2
60
1
Amplitude
50
40 0.8
30 0.6
20 0.4
10 0.2
0
0 0.5 1 1.5 2 2.5 3 0
0 0.5 1 1.5 2 2.5 3 3.5
Damping ratio
Time (s)
Figure 7: Phase margin versus damping ratio.
KP = 1; KI = 7.5
KP = 1; KI = 25
Passive susp.
the red solid locus shows another design conditions with
gain margin =1 and phase margin 60∘ , and the maximum Figure 8: Step response of the system.
frequency for this region is 31.5 rad/sec.
It is an applicable method to achieve user specified
gain and phase margins to define the values of controller Returning back to (10) and substituting 𝑗Ω = 𝑠, another
parameters with the ability to justify the maximum frequency technique is applied in this section; a developed methodology
without the need of sweeping over the parameters and also by Shamsuzzoha and Skogestad requires only one closed-loop
without using programming for solving inequality problems. step test to obtain PI controller setting. In this method it is
This method can be expanded to estimate the stabilizing also simple to obtain the PID tuning parameters [11, 12].
region of PID controller gains as shown in [17]. Considering an ideal PID controller in the form, as shown
From Figure 6, the plotted line in red solid color indicates in (5)
the output response of the uncontrolled system and also
shows the stability region of frequencies shown in red points 1 𝐾
𝑐 (𝑠) = 𝐾𝑐 (1 + + 𝜏𝐷𝑠) ⇐⇒ 𝑐 (𝑠) = (𝐾𝑃 + 𝐼 + 𝐾𝐷𝑠) .
in the range Ω ∈ [10.2, 15.6] rad/sec, while the blue dashed 𝜏𝐼 𝑠 𝑠
locus indicates the system response after applying the PI (24)
controller which is chosen from the stability region shown in
Figure 6 as 𝐾𝑃 = 1 and 𝐾𝐼 = 2.5; it is obvious that the stability 𝐾𝑐 is the controller gain and 𝜏𝐼 , 𝜏𝐷 are the controller integral
region of frequencies shown in blue squares is expanded in and derivative time, respectively.
the range Ω ∈ [0, 15.5] rad/sec. The defined parameters can be calculated using the
The relationship between the phase margin Φ𝑀, the following basic formulae:
overshoot (OS), and the damping ratio 𝜁 can be expressed as:
2𝜏 + 𝜃 𝜃
𝐾𝑐 = , 𝜏𝐼 = min {(𝑟 + ) , 8𝜃} ,
√−2𝜁2 + √1 + 4𝜁4 3𝑘𝜃 2
Φ𝑀 = 90∘ − tan−1 , (22) (25)
2𝜁 𝑟𝜃
𝜏𝐷 = ,
2𝜏 + 𝜃
where
where 𝑘 is the gain, 𝜏 is the time lag constant, and 𝜃 is
√1−𝜁2 )
OS% = 𝑒−(𝜁𝜋/ × 100%. (23) the time delay. The controller gain is only depending on
the overshoot, while the integral and derivative times are
The relation between Φ𝑀 and 𝜁 is plotted in Figure 7. function of the system peak time. Applying this approach, the
So it is noticeable to get different overshoot values due to tuning results simulation for the designed problem assuming
the changes of the phase margins with the controller design different values for 𝜏/𝜃 is indicated in Table 2.
parameters; Figure 9 shows the effect of changing these values An improved definition of (25) parameters were con-
as follow. cluded in Shamsuzzoha et al. [12], a closed-loop test with
In Figure 8 the red dash-dot line indicates the uncon- P-controller was applied to adjust overshoot value of 0.3021
trolled system step response; the overshoot is about 46%, and for P-only control then assigned the other parameters which
choosing a test point from Figure 3 with values of 𝐾𝑃 = 1 leaded to gain 𝐾𝑐 = 2.5 with 𝜏𝐼 = 0.2040 and peak time 0.153
and 𝐾𝐼 = 25, it is clear that it locates outside the regions for the controller gains.
of stabilities from the used technique, so it is clear from the In spite of Zeigler-Nichols methods’ drawbacks which
figure that the green solid line response overshoot value is are mentioned in previous sections, it is still one of the
about 55%, and unlike the line in blue dot color it is clear that easiest methods for determining the proportional gain 𝐾𝑃 ,
the overshoot value is improved to be 7% when the test point integral time 𝑇𝑖 , and derivative time 𝑇𝑑 based on the transient
is chosen as 𝐾𝑃 = 1 and 𝐾𝐼 = 7.5. response of a given system. Zeigler-Nichols suggested the
6 Advances in Mechanical Engineering
𝐾𝑃 𝐾𝐼 𝐾𝐷
Amplitude: 1.45
1.3
0.1 0.4000 0.6667 0.0333 1.2 System: PI
Amplitude
Time (s): 0.23
1.1 Amplitude: 1.08
0.5 0.6667 0.6667 0.1667 System: shass PID
Time (s): 0.158
1 Amplitude: 1.08
2.5 2.0000 0.6667 0.8333
0.9
10 7.0000 0.8750 3.3333
0.8
20 13.6667 1.7083 6.6667 0.7
50 33.6667 4.2083 16.6667 0.6
100 67.0000 8.3750 33.3333 0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
Table 3: Zeigler-Nichols tuning rules. Passive susp. Shass PID
Z-N PID PI
Type of controller 𝐾𝑃 𝑇𝑖 𝑇𝑑
P 0.5𝐾cr ∞ 0 Figure 9: Step response of the system.
PI 0.45𝐾cr 0.83 𝑃cr 0
PID 0.6𝐾cr 0.5 𝑃cr 0.125 𝑃cr Table 5: System response analysis parameters.
values of the controller parameters according to the shown response for a vehicle coming out of a pothole can be
formulae in Table 3. simulated as a step input with amplitude of 0.08 as [18]:
For the systems which have a mathematical model, root-
locus method can be used to find the critical gain 𝐾cr and the 0 ⇒ 𝑡 ≤ 0,
sustained oscillation frequency 𝜔cr ; then get 𝑃cr = 2𝜋/𝜔cr . Amp = { (26)
0.08 ⇒ 𝑡 ≥ 0.
Also the systems’ open loop system (passive suspension) and
closed-loop system (Active suspension) can be analyzed in The system output displacement is shown in Figure 10.
frequency domain to find the bandwidth, gain, and phase Figure 10 shows that the peak value of the passive suspen-
margins, and then the design parameters can be found easily. sion system in red solid line is nearly 0.116 m although the
Considering the previous controlled car problem, the input step amplitude is 0.08 m; that is, the overshoot is 45%,
results are found as follows: the critical gain is 5.1, and the and after using the estimated values from the described tech-
sustained oscillation frequency is 56.7 rad/sec. It is the same niques, the overshoot and response of the system improved
result shown in Figure 4, where the stability frequency for although the transient time is increased for some of them but
the case of assuming gain margin 1 and phase margin 0 is the displacement action is smoothed; Zeigler-Nichols results
Ω → [0, 56.7]. are plotted in dash-dot line, Shamsuzzoha results are plotted
From the previous values, the P, PI, and PID Zeigler- in dash lines, and PI controller is plotted in x marker line.
Nichols controller values can be tabulated in Table 4. Another simulation model for random roughness road is
Applying the controller parameters obtained from the applied where the road model is expressed as a differential
different tuning techniques, the system response is plotted in equation with the formula [19]:
Figure 9.
From Figure 9 and related to the different tuning meth- 𝑍̈
(𝑡) + 𝑎V𝑍 (𝑡) = 𝑊 (𝑡) , (27)
ods, the systems’ time of the first peak, the peak values, and ̈is the road random incentive and the constant “𝑎”
the overshoot values can be tabulated in Table 5. where 𝑍(𝑡)
depends on road grade; the values of “𝑎” are shown in Table 6.
𝑊(𝑡) is the white noise, with zero mean value and
4. Simulations and Results amplitude equals 1, and V is the vehicle speed.
Focusing for more details can be shown in Figure 12.
After defining the controller parameters, it is necessary to test The desired problem is validated using the results of
the system performance for some situations. The suspension road simulation of grade B and assuming the velocity is
Advances in Mechanical Engineering 7
0.12
0.1
0.1
0.05
Displacement (m)
0.08
Amplitude
0
0.06 −0.05
0.04 −0.1
−0.15
0.02
−0.2
0 0 5 10 15 20 25 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s)
Time (s)
Passive susp. Shass PID
Passive susp. Shass PID PI
Z-N PID
Z-N PID PI
Figure 11: Output displacement of the system.
Figure 10: The output displacement from the system.
0
50 km/hr; the results show the effectiveness of the used
controllers to reduce the displacement motion as shown in −0.05
Figure 11 and zoomed in Figure 12, also results illustrated
that different control techniques can be applied to MS-TMM, −0.1
results of the passive suspension response are plotted in blue 0 1 2 3 4 5 6 7 8 9 10
solid line, Zeigler-Nichols result is plotted in blue dot plot, Time (s)
Shamsuzzohas’ result is plotted in green dash-dot plot, and Passive susp. Shass PID
finally the PI controls’ result is plotted in red dash plot. Z-N PID PI
Research Article
All Terrain Vehicle Flexible Multibody Dynamic
Simulation for Fatigue Prediction
Copyright © 2013 J.-S. Chen and H.-Y. Hwang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
This study presents the flexible-body dynamic analysis and simulated stress recovery of vehicle components to predict their lifetime
when maneuvering on an uneven random road. The subject of this study was an all-terrain vehicle (ATV). The body frame and
suspension system were modeled as flexible elements for multibody dynamic simulation. The simulations in this study revealed the
stress from the flexible elements and predicted the component fatigue life using the retrieved stresses. This approach considers the
interaction between dynamic forces and structure deformation and achieves more accurate structure stress prediction and fatigue
life prediction.
This study uses ADAMS to perform flexible-body where 𝑚𝑖 and 𝐼𝑖 represent the modal mass and modal inertia
dynamic simulation and models all structural components as of moving body, respectively.
flexible elements. The stress history of the structure was then The mass matrix of flexible body, 𝑀, can be expressed as
retrieved from the dynamic simulation for fatigue calcula-
tion. 𝑀𝑡𝑡 𝑀𝑡𝑟 𝑀𝑡𝑚
The remaining sections are as follows: the theoretical [ 𝑇 ]
[ 𝑀𝑡𝑟 𝑀𝑟𝑟 𝑀𝑟𝑚 ]
background of the flexible-body dynamics and fatigue anal- [
𝑀 (𝜉) = [ ], (6)
]
ysis, the simulation model, the simulation results, and the [ 𝑇
]
𝑀𝑡𝑚 𝑀𝑟𝑚 𝑀𝑚𝑚
conclusion. [ ]
where 𝑡, 𝑟, and 𝑚 represent the translational, rotational, and
2. Theory Background modal degrees of freedom, respectively.
2.1. Equations of Flexible-Body Dynamics. In equations of The stiffness matrix of the flexible body, 𝐾, is relatively
motion, the linear deformation of a structure can be repre- simple because there is no rigid body contribution:
sented by the combination of mode shapes and mode coordi-
𝐾𝑡𝑡 𝐾𝑡𝑟 𝐾𝑡𝑚
nates as follows: [ 𝑇 ] 0 0 0
[ ]
𝑀 𝐾 (𝜉) = [ 𝐾𝑡𝑟 𝐾𝑟𝑟 𝐾𝑟𝑚 ] = [0 0 0 ] . (7)
𝑢 = ∑𝜙𝑖 𝑞𝑖 , [ ]
[0 0 𝐾𝑚𝑚 ]
(1) 𝑇 𝑇
𝑖=1 𝐾𝑡𝑚 𝐾𝑟𝑚 𝐾𝑚𝑚
[ ]
where 𝜙 is the mode shape and 𝑞 is the mode coordinate.
The equations of motion with a Lagrange multiplier can
With the Cartesian coordinate (𝑥, 𝑦, 𝑧), Euler’s angle
be written as follows:
(𝜓, 𝜃, 𝜙), mode coordinate 𝑞, and the generalized coordinates
of the flexible element can be expressed as follows: 1 𝜕𝑀 ̇𝑇 ̇ 𝜕𝜓 𝑇
𝑀𝜉 ̈
+ 𝑀̇𝜉 ̇− [ 𝜉] 𝜉 + 𝐾𝜉 + 𝑓𝑔 + 𝐷𝜉 ̇+ [ ] 𝜆 = 𝑄.
𝑥} 2 𝜕𝜉 𝜕𝜉
{
{ }
{
{ 𝑦 }
} (8)
{
{ }
}
𝑥 { }
{𝑧}
{ } The term 𝐷 is the modal damping matrix, and 𝐾𝜉 and 𝐷𝜉,̇
𝜉 = (𝜓) = { 𝜓 } , (𝑖 = 1, 2, 3, . . . , 𝑀) . (2)
𝑞 {𝜃}
{ } respectively, represent the structural internal force caused
{
{ }
}
{𝜙}
{ } by the elastic deformation and velocity. The term 𝜆 is the
{ }
{ }
𝑞 Lagrange multiplier, 𝑓𝑔 is the structure weight, and 𝑄 is the
{ 𝑖} external force.
The position on the deformed body is written as The strain and stress can be retrieved using the mode
𝑟𝑖 = 𝑥 + 𝐴 (𝑆𝑖 + 𝜙𝑖 𝑞) , (3) coordinate [10]. This study calculates the stress in the struc-
ture using the modal stress recovery technique:
where 𝐴 is the transformation matrix between global coordi-
nates and the body local coordinates. The term 𝑆𝑖 represents {𝜀} = [𝐻] {𝑥} ,
the position in the body before deformation, and 𝜙𝑖 is the (9)
mode shape matrix. {𝜎} = [𝐸] {𝜀} .
From the position, the velocity of the deformed body is The term {𝜎} is the stress vector, {𝜀} is the strain vector,
𝑑𝑟𝑖 𝑑𝑥 𝑑𝐴 𝑑 (𝑠𝑖 + 𝜙𝑖 𝑞) and [𝐻] is the finite-element geometric deformation-to-
V𝑖 = = + (𝑠 + 𝜙𝑖 𝑞) + 𝐴 strain matrix. The term [𝐸] is the material property matrix.
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑖 𝑑𝑡
= 𝑥̇− 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝜔 + 𝐴𝜙𝑖 𝑞 ̇ (4)
2.2. Fatigue Prediction. Metal material under cycling stress
can fail even when the stress is under the ultimate stress limit.
= 𝐹 − 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝐵 + 𝐴𝜙𝑖 𝜉.̇ The relationship between fatigue stress and life cycle is usually
The term 𝜔 is the angular velocity vector of the body represented by a stress-cycles curve (𝑆-𝑁 diagram). For steel,
coordinate, and 𝐹 is the transfer matrix between the time a cycle life greater than 106 is treated as a lasting life, and the
derivative of Euler’s angle and the angular velocity. stress corresponding to a 106 cycle is the endurance limit 𝑆𝑒
The kinematic energy of the body is [11].
The 𝑆-𝑁 diagram is plotted in log scale, and the formula
1 of 𝑆-𝑁 curve can be simplified as
𝑇= ∫ 𝜌V𝑇 V𝑑𝑉
2
log 𝑆𝑓 = log 𝑎 + 𝑏 log 𝑁 ⇒ 𝑆𝑓 = 𝑎𝑁𝑏 , (10)
1𝑁
≈ ∑ (𝑚𝑖 V𝑖𝑇 V𝑖 + 𝜔𝑖𝑇 𝐼𝑖 𝜔𝑖 ) (5)
2 𝑖=1 where 𝑏 is the slop of curve and 𝑁 is the life cycle.
During vehicle maneuvering, the structural stress is fluc-
1
= 𝜉𝑇̇𝑀 (𝜉) 𝜉,̇ tuating. The amplitude of stress variation varies with the driv-
2 ing and road conditions. The rain-flow counting technique
Advances in Mechanical Engineering 3
Revolute
joint
Bushing
Ball
joint
y x
Figure 1: Vehicle structure and suspension frames. z
3.1. Components. The structure of vehicle model included 4. Simulation and Result
three primary parts: the body frame, the front suspension,
and the rear suspension (Figure 2). The powertrain and In the simulation, the vehicle maneuvered in straight runs
passenger weights were attached to the structure, and the total over a random uneven surface. The vehicle accelerated from
weight of vehicle was 423 kg. All the structure components standing still, and the data for analysis was collected when
were modeled as a flexible body. The tires were simulated as it reached a constant speed of 50 km/h. The duration of data
force elements in dynamic simulation [14]. Figure 2 shows collecting was 10 s. Several thicknesses of the structure frame
the connecting joint and forces applied between components tube were applied in the simulation: 1 mm, 1.5 mm, 2 mm, and
[14, 15], and Table 1 shows material properties of the frame 2.5 mm.
[16].
In addition to the weight of the vehicle structure and the 4.1. Dynamic Simulation. In flexible-body dynamic simula-
driveline, two dummies weighing 80 kg each were modeled tion, ADAMS can retrieve the structural stress which could be
and seated on the vehicle structure. The dummies were applied to failure or fatigue analysis. Figure 4 shows the high-
modeled as two parts (the upper torso; pelvis and legs) est stress locations during the dynamic simulation. Table 4
4 Advances in Mechanical Engineering
10.0 y
z x
V-axis length (mm)
5.0
Figure 5: Joint locations.
0.0
−5.0
simulation cannot determine the structural stress, the max-
−10.0 imum rear suspension joint forces were used to present the
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 difference between these two types of simulation (Table 5).
H-axis length (m) Figure 5 shows the joint locations. The frame tube thickness
of the model was 2.0 mm in this simulation.
Figure 3: 2D random road profile.
In addition to the suspension and tire elements, the
flexible element in the vehicle structure also absorbed part of
582079 dynamic energy during the uneven road simulation. Thus, the
585076 joint force of the flexible-body model was smaller than that of
von Mises stress (MPa) the rigid-body model during the simulation.
592.32
533.09
470.86
586151
582887 4.2. Fatigue Analysis. The FEM model in this study adopts
414.62 several thicknesses of the frame tube. The flexible elements
355.39 were built based on these models, and flexible-body dynamic
296.16 simulations were run with different tube thicknesses. The
236.93 tube thicknesses were 1 mm, 1.5 mm, 2 mm, and 2.5 mm. The
177.7 maximum stress was retrieved and plotted in a stress-time
118.46 curve. Figure 6 shows the stress history of the structure with
59.23 a 2.5 mm tube thickness in 10 s simulation at a constant speed
5.48E − 014 of 50 km/h.
The rainflow counting method was applied to the stress
history curve, and the stress cycle number of different stress
amplitudes was used for fatigue analysis. Figure 7 shows the
Figure 4: Highest stress locations at random road simulation. stress range and cumulative cycle number of different frame
tube thicknesses.
Based on the stress cycle and material 𝑆-𝑁 diagram, the
structural life can be predicted using the Miner rule. Using
shows the maximum von Misses stress of these locations in a 10 s simulation cycle, Table 6 presents the structural life
uneven random road simulations. of different frame tube thicknesses. This duration time and
To compare the differences in dynamic load running on the maximum stress position listed in Table 4 show that
an uneven road surface, this study created both rigid-body the structure connecting points of two or more structure
and flexible-body models. Because the rigid-body dynamic members are the most vulnerable to fatigue damage.
Advances in Mechanical Engineering 5
Frame tube thickness (mm) Road surface Max. stress grid point Max. stress (MPa)
Flat 582887 215.36
1
Random 585076 498.30
Flat 582887 163.08
1.5
Random 586151 430.28
Flat 582887 137.87
2
Random 586151 361.84
Flat 582887 105.33
2.5
Random 585079 347.21
Frame tube thickness (mm) Right and left joints average force at rear Difference
suspension (N)
Flexible-body model 623.01 —
Rigid-body model 758.09 21.7%
200
5. Conclusions
150
Conventional analysis methods model the vehicle body as a
100
rigid body in dynamic simulations. In this case, the dynamic
50 energy passing from the tire running over an uneven road can
only be absorbed and damped by tires, springs, and dampers
0 in suspension, and bushings. After this rigid body dynamic
0.0 2.0 4.0 6.0 8.0 10.0 simulation, the component load was applied to a finite-
Time (s) element model to determine the structural stress of the
Figure 6: Stress history of the max. stress grid point (585079).
component for further analysis. The approach used in this
paper is different from the conventional one. The dynamic
model in this study uses flexible bodies. The advantage of
using flexible-body dynamic simulation is that it simulates
450
structural stresses in dynamic analysis and can be retrieved
400
through the simulation. The dynamic response considers the
350 effect of structural deformation. This provides more realistic
Stress range (MPa)
References
[1] H. Ohchida, “Analysis of service failures of Hitachi products
1970–1975,” Hitachi Company Report, 1979.
[2] S. Lu, Y. Li, and Y. Dong, “Dynamic modeling and virtual test
of a new all terrain off-road vehicle,” Applied Mechanics and
Materials, vol. 86, pp. 125–128, 2011.
[3] B. R. Miao, W. H. Zhang, L. M. Zhang, T. Zhu, and H. T. Yin,
“Research of modern railway vehicle structure fatigue design
method based on multibody dynamic simulation,” Advanced
Materials Research, vol. 118–120, pp. 327–331, 2010.
[4] J. Cuadrado, R. Gutiérrez, M. A. Naya, and M. González,
“Experimental validation of a flexible MBS dynamic formu-
lation through comparison between measured and calculated
stresses on a prototype car,” Multibody System Dynamics, vol. 11,
no. 2, pp. 147–166, 2004.
[5] I.-D. Moon, H.-S. Yoon, and C.-Y. Oh, “A flexible multi-body
dynamic model for analyzing the hysteretic characteristics and
the dynamic stress of a taper leaf spring,” Journal of Mechanical
Science and Technology, vol. 20, no. 10, pp. 1638–1645, 2006.
[6] A. A. Shabana and J. R. Sany, “A survey of rail vehicle track simu-
lations and flexible multibody dynamics,” Nonlinear Dynamics,
vol. 26, no. 2, pp. 179–210, 2001.
[7] S. Zhu, Z. Xiao, and X. Li, “Vehicle frame fatigue life prediction
based on finite element and multi-body dynamic,” Applied
Mechanics and Materials, vol. 141, no. 1, pp. 578–585, 2012.
[8] J. Yang, Q. Shi, G. Zhang, and J. Zhang, “The fatigue life simula-
tion of the wheel of CHR3 EMU in random loading,” Advanced
Materials Research, vol. 430–432, pp. 1424–1427, 2012.
[9] C. Rathod, R. Chamorro, J. L. Escalona, M. El-Sibaie, and A. A.
Shabana, “Validation of three-dimensional multi-body system
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 153913, 12 pages
http://dx.doi.org/10.1155/2013/153913
Research Article
Projectile Impact Point Prediction Based on Self-Propelled
Artillery Dynamics and Doppler Radar Measurements
Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy
Institute of Launch Dynamics, Nanjing University of Science and Technology, 200 Xiao Ling Wei, Nanjing 210094, China
Copyright © 2013 Mostafa Khalil et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Any trajectory calculation method has three primary sources of errors, which are model error, parameter error, and initial state
error. In this paper, based on initial projectile flight trajectory data measured using Doppler radar system; a new iterative method
is developed to estimate the projectile attitude and the corresponding impact point to improve the second shot hit probability. In
order to estimate the projectile initial state, the launch dynamics model of practical 155 mm self-propelled artillery is defined,
and hence, the vibration characteristics of the self-propelled artillery is obtained using the transfer matrix method of linear
multibody system MSTMM. A discrete time transfer matrix DTTM-4DOF is developed using the modified point mass equations
of motion to compute the projectile trajectory and set a direct algebraic relation between any two successive radar data. During
iterations, adjustments to the repose angle are made until an agreement with acceptable tolerance occurs between the Doppler
radar measurements and the estimated values. Simulated Doppler radar measurements are generated using the nonlinear six-
degree-of-freedom trajectory model using the resulted initial disturbance. Results demonstrate that the data estimated using the
proposed algorithm agrees well with the simulated Doppler radar data obtained numerically using the nonlinear six-degree-of-
freedom model.
actual thrust made during the flight of rocket-assisted projec- equation of self-propelled artillery was obtained to compute
tiles using Doppler radar measurements. In 2008, an iterative the corresponding vibration characteristics. In order to itera-
algorithm had been developed [7] using maximum likelihood tively compute the projectile attitude, a discrete time transfer
method to predict the drag characteristics of projectiles in matrix DTTM-4DOF is developed based on linearization
motion by processing data acquired by Doppler radar system, of the nonlinear modified point mass equations of motion
where the point mass model was used due to the projectile using second-order Taylor series [16]. The system state vector,
small angle of attack along the trajectory. which defines the projectile trajectory parameters during
The movements of artillery and projectile are very com- Doppler radar observations, is defined by 13 state variables,
plex in launch process because of the complex structure of which are the projectile position, velocity, spin rate, repose
self-propelled artillery and the severe environments, such angle, and wind velocity. During iterations, adjustments to
as, high temperature, high pressure, high speed, instanta- the repose angle are made until agreement with a certain tol-
neous state, multibody, and mutation, in launch process erance occurs between the Doppler radar velocity measure-
[8]. Therefore, gun launch dynamics has a primary concern ments and the value predicted using DTTM-4DOF. Finally,
in designing embedded navigation and guidance systems projectile impact point can be predicted precisely depending
in case of guided projectiles and evaluating the projectile on the length of radar data available, where the data estimated
initial disturbance in case of unguided projectiles. In 2000, at the end of radar observations is used as an initial state to
the launch dynamics of self-propelled artillery [9] based the 6-DOF trajectory model.
on transfer matrix method of multibody system MSTMM
was developed for studying the vibration characteristics and 2. Launch Dynamics Model for
dynamics of self-propelled artillery and the projectile’s initial Self-Propelled Artillery System
disturbance. In 2001, a finite element analysis [10] had been
done for the breech closure for the 155 mm Cannon M199, To accurately estimate the projectile initial state, launch pro-
which is normally mounted on the Towed Howitzer M198. cess has to be decided precisely by studying the artillery and
The analysis is for a 9 body problem with 13 contact surfaces projectile dynamics from the moment of firing to the state
and was solved for both static and dynamic load cases. In of muzzle point. Four interactions have to be considered, the
2004, a simplified model [11] had been done to predict joint interaction between projectile and artillery, the interaction
forces and accelerations along the length of a simplified between the artillery transverse and longitudinal motion, the
155 mm projectile using modal superposition. These pre- interaction between rigid and flexible body, and, finally, the
dicted loads used to locate all guidance equipment, sensors, interaction between space coordinate and time coordinate for
joints, and computers that must operate reliably after exiting describing motion parameters.
the gun. In 2005, a sophisticated 3D-finite-element model The launch dynamics model of self-propelled artillery is
[12] had been developed to investigate the survivability a multi-rigid-flexible system which has been studied in [8].
of embedded electrical systems by simulating the launch This system is composed of 51 elements (23 bodies and 28
dynamics of a surrogate Excalibur projectile. This study joints) as shown in Figure 1. It has 3 boundary ends including
determined free mounting locations trouble for sensitive ground, gun breech, and muzzle. The ground (0) is regarded
components and parametric investigation in identifying sen- as an infinity rigid body and the first input point, the gun
sitive factors affecting the muzzle-exit motion of projectile breech (44) is regarded as rigid body and the second input
substructures. In 2008, a new method [13] named as transfer point, and the muzzle (52) is the output point. The road
matrix method of linear multibody system MSTMM for lin- wheels (13–24) are regarded as lumped masses. The hull
ear hybrid multibody system dynamics was developed. This (37), revolving part (39), elevating part (41), and muzzle
method had been applied to a Multiple Rocket System as a brake (51) are regarded as rigid bodies; the barrel is divided
linear multi-rigid-flexible-body system. In 2011, the dynamics into 6 segments (45–50) and each is regarded as a beam
problem of a shipboard gun system had been solved [14] using with equal sectional area. The interactions (1–12) between
the discrete time transfer matrix method of multibody system ground and road wheel and the connections (25–36) between
MSDTTMM [15]. road wheel and the hull are, respectively, modeled with
In this paper, an iterative method is developed to deter- springs and dampers in 3 directions. Connections (38, 40,
mine the projectile kinematics including attitude and angular 42, and 43) are modeled with springs and rotary springs
motion using Doppler radar measurements. Doppler radar accompanying dampers to represent relative linear motion
measurements are only available during the first portion and relative angular motion in 3 directions at the same time.
of projectile trajectory including the range, the range-rate, All connections between barrel and gun breech, barrel and
and the azimuth and elevation angles (position and velocity muzzle brake, and each segment of barrel are regarded as
vectors), in order not to be recognized by the enemy coun- fixed. The masses of traversing mechanism, elevating mecha-
terattack systems. Simulated Doppler radar measurements nism, and equilibrator fall into elevating part, revolving parts,
are generated during flight time using 6-DOF trajectory and hull, respectively. According to the launch dynamics
model including the projectile initial disturbance problem. model of self-propelled artillery and the sequence number of
The projectile initial disturbance is computed by solving each element, there are 49 connection points which are 𝑃𝑖,𝑖−12 ,
the self-propelled artillery and projectile dynamics as a 𝑃𝑖,𝑖+12 (𝑖 = 13–24), 𝑃37,𝑖 (𝑖 = 25–36), 𝑃37,38 , 𝑃39,38 , 𝑃39,40 ,
multi-rigid-flexible system using Transfer Matrix Method of 𝑃41,40 , 𝑃41,42 , 𝑃41,43 , 𝑃44,45 , 𝑃45,46 , 𝑃46,47 , 𝑃47,48 , 𝑃48,49 , 𝑃49,50 ,
Multibody System MSTMM [8], where the overall transfer and 𝑃51,50 and 14 boundary points which are 𝑃0,𝑖 (𝑖 = 1–12),
Advances in Mechanical Engineering 3
Elevating part 41
Projectile
Breech 44 42 40 43 Beams 45–50 Muzzle brake 51
52
38
Revolving part 39
Hull 37
25 26 27 28 29 30
13 14 15 16 17 18
1 2 3 4 5 6 Road wheels 13–24
0
BT
and the forces that acted on the barrel and muzzle brake V1 h3
x1
by propellant gas [8] are regarded as external forces of the O
artillery. The recoil resistance between the barrel and elevat-
ing part produced by recoil and counter-recoil mechanism is Figure 2: Coordinate system and directions of ballistic target.
decomposed into external force and internal force, and the
internal force is proportional to the relative displacement
𝑝𝛼 𝑝
between barrel and elevating part. where 𝐶𝐷, 𝐶𝐿𝛼 , 𝐶𝑁 , and 𝐶𝑙 are the aerodynamic drag
force, lift force, Magnus force, and spin damping moment
3. Projectile Flight Trajectory Model coefficients, respectively; V is the projectile velocity vector in
earth fixed coordinate; Vt is the projectile total velocity vector
Due to the complexity of projectile initial disturbance deter- with respect to air; V is the projectile total aerodynamic speed;
mination, a modified point mass 4-DOF model for an 𝜌 is the air density; 𝑚, 𝐼𝑥 , 𝐼𝑦 , 𝑑, and 𝑆ref are the projectile mass,
unguided rigid projectile was stated in [18, 19]. These equa- axial and transverse moments of inertia, reference diameter,
tions assume that the epicyclic pitching and yawing motion and reference area, respectively; and g and Λ are the earth’s
of the projectile are small everywhere along the trajectory; gravitational and Coriolis accelerations vectors.
therefore, the yaw and pitch moments considered in 6-DOF The projectile repose angle 𝛼𝑅 is defined by [19]
trajectory model [19] are neglected.
The projectile modified point mass equations of motion 𝛼𝑅 = x − (cos 𝛼𝑡 ) i, (4)
with respect to earth fixed coordinate system (𝑋1 , 𝑋2 , 𝑋3 ) as
shown in Figure 2 are given by [19] where x is the projectile’s axis of symmetry unit vector and i
is the unit vector in the direction of Vt .
𝑇
And the total aerodynamic angle of attack is [3, 18, 19]
Ẋ= V = [𝑉1 𝑉2 𝑉3 ] , (1)
Vt ⋅ x
𝜌V𝑆ref 𝐶𝐷 𝜌𝑆 𝐶𝛼 𝛼𝑡 = cos−1 . (5)
V̇= − Vt + ref 𝐿 V2 𝛼𝑅 V
2𝑚 2𝑚
𝑝𝛼
(2) The simplified form of the repose angle is given by [19]
𝜌𝑆ref 𝑑𝐶𝑁
+ 𝑝 [Vt × 𝛼𝑅 ] + g + Λ,
2𝑚 2𝐼𝑥 𝑝
𝛼𝑅 = [V̇× Vt ] , (6)
2 𝜌𝑆ref 𝑑V4 𝐶𝑚
𝛼
𝜌𝑆ref 𝑑 V 𝑝
𝑝̇= 𝑝𝐶𝑙 , (3)
2𝐼𝑥 𝛼
where 𝐶𝑚 is the aerodynamic pitching moment coefficient.
4 Advances in Mechanical Engineering
For discrete systems, a good approximation was done for The equations of motion matrix A, summarizing (1)–(9)
the projectile repose angle [20] as multiplied by the time step Δ𝑡, are given by
2𝐼𝑥 𝑝
𝛼𝑅 = [g × Vt ] . (7) Δ𝑡2
𝜌𝑆ref 𝑑V4 𝐶𝑚
𝛼
[G3×3 I3 ⋅ Δ𝑡 O3×1 O3×3 O3×3 ]
[ 2 ]
The Coriolis acceleration due to earth’s rotation and gravita- [G ⋅ Δ𝑡 VV ⋅ Δ𝑡 V𝑝 ⋅ Δ𝑡 V𝛼 ⋅ Δ𝑡 VW ⋅ Δ𝑡]
̇ ̇ ̇ ̇
[ 3×3 ]
tional acceleration are given by [3, 19] A=[
[ O1×3 O1×3 𝑝𝑝̇ ⋅ Δ𝑡 O1×3 O1×3 ]
],
[ ]
0 − cos 𝜇 sin 𝜆 − sin 𝜇 𝑉1 [𝛼 G 𝛼V̇V̇V 𝛼V̇V̇𝑝 𝛼V̇ V̇𝛼 𝛼V̇ V̇W ]
[ V̇3×3 ]
Λ = 2Ω [cos 𝜇 sin 𝜆 0 cos 𝜇 cos 𝜆] [𝑉2 ] ,
ẆV ⋅ Δ𝑡 O3×3
[ O3×3 O3×3 O3×3 ]
[ sin 𝜇 − cos 𝜇 cos 𝜆 0 ] [𝑉3 ] (14)
𝑇
𝑋1 2𝑋2
g = −𝑔 ⋅ [ 1− 0] , where I3 is the 3 × 3 identity matrix.
𝑅𝑒 𝑅𝑒
All elements of matrix A in (14) are stated as follows:
(8)
𝑋12
𝐻 = 𝑋2 + . (9) V̇V = ΛV − 𝜌𝑎 1 𝐶𝐷 VI3 ,
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
2 ⋅ 𝑅𝑒 (16)
V̇
W
(10)
where 𝑎1 = (𝑆ref /2𝑚), 𝑎2 = ((𝑆ref 𝑑)/2𝑚), 𝑏1 = ((𝑆ref ⋅ 𝑑2 )/(2 ⋅
DTTM-4DOF is the relation between any two successive state 𝐼𝑥 )), 𝑏2 = (𝑑/𝑏1 ), and B1 =Vt × 𝛼𝑅 .
vectors, where
The change in wind velocity vector with the projectile
S𝑖+1 𝑖 position ẆV is given by
10×1 = DTTM-4DOF10×13 ⋅ S13×1 . (11)
The 2nd-order Taylor series is given by [16]
𝜕W/𝜕𝑡 𝜕W/𝜕𝑡 𝜕W
2 ẆV = = = . (18)
Δ𝑡 𝜕V 𝜕X/𝜕𝑡 𝜕X
S𝑖+1
10×1 = S𝑖10×1 + Ṡ
𝑖 ̈
𝑖
10×1 ⋅ Δ𝑡 + S10×1 ⋅ , (12)
2
where The Jacobian matrix J of the system, mentioned in (13b),
which represents discretization errors in matrix A, is given
Ṡ
10×1 ⋅ Δ𝑡 = I10×13 A13×13 ⋅ S13×1 , (13a) by
𝑖
̈ Δ𝑡2 𝜕Ṡ
𝑖
Δ𝑡2 𝜕Ṡ Δ𝑡2
𝑖
S10×1 ⋅ =[ ⋅ ] = [( ) ⋅ Ṡ
𝑖
13×1 ⋅ ] O3×3 I3 O3×1 O3×3 O3×3
2 𝜕𝑡 2 𝜕S 10×13 2 [O
[ j22 V̇𝑝 O3×3 j25 ]]
J = [ 3×3 ]. (19)
= [T10×10 J10×13 A13×13 S13×1 ] . [O1×3 j32 𝑝𝑝̇ O1×3 −j32 ]
(13b) [O3×3 j42 j43 O3×3 j45 ]
Advances in Mechanical Engineering 5
Radar data
Si[10×1] DTTM-4DOF [ X V p 𝛼R ]T Si+1 = DTTM ∗ Si −+
[Xi+1 Vi+1 ]T
No
[Xo Vo 𝜃0 po ]T Calculate ΔVo No
ΔX (i=1) ≤ tolerance
Yes j=0
eq. (29a)
i = 0, j = 0 No
Calculate Δ𝛼Ri No ΔVi ≤ tolerance
Radar data length < i
eq. (30a)
Yes
No
i>0 Calculate 𝛼t , x 𝛼R
i= i+1
eq. (31)–(32)
Yes
Calculate h Calculate 𝛼̇ ̇
R , i, ẋ
eq. (36) eq. (33)–(35) Projectile impact point
No
Yes
(3) The rate of change of the projectile total velocity unit Table 1: Corresponding relation between the input point and the
vector is given by [19] sequence number.
1 0.75
Mode shape
0 0.25
−0.5 0
0 40 80 120 160 200
Input point
−0.25
1st mode, X-direction
2nd mode, Z-direction −0.5
0 0.25 0.5 0.75 1
Figure 4: The 1st and 2nd mode shapes in 𝑋- and 𝑍-direction, xo /L b
respectively.
Y-direction
Z-direction
1
Figure 8: Projectile lateral displacement.
0.5
Y-direction
10
0
8
0
xo (m)
4
−0.0015
0 0.25 0.5 0.75 1
0
0 0.25 0.5 0.75 1 xo /L b
Nondimensional launch time Plumb
Sidewise
Figure 6: Projectile longitudinal displacement.
Figure 9: Projectile angular displacement during launch.
1500 0.6
Roll rate (rad/s)
1000 0.4
𝛼R|1 (rad)
0.2
500
0
0
0 0.25 0.5 0.75 1 −0.2
xo /L b
−0.4
0 0.1 0.2 0.3 0.4 0.5
6 Time (s)
Swing rate (rad/s)
0.3
3
0.15
0
𝛼R|2 (rad)
0
−3
−0.15
−6
0 0.25 0.5 0.75 1
−0.3
xo /L b 0 0.1 0.2 0.3 0.4 0.5
Plumb Time (s)
Sidewise 0.4
Figure 10: Projectile angular velocity during launch. 𝛼R|3 (rad) 0.2
−0.2
438
−0.4
433 0 0.1 0.2 0.3 0.4 0.5
V1 (m/s)
Time (s)
428
Estimated
6DOF-R
423
6DOF-N
418 Figure 12: Projectile total angle of attack vector versus flight time.
0 0.1 0.2 0.3 0.4 0.5
Time (s)
552 Table 2: Simulation results of initial disturbance of projectile.
547 Initial disturbance Plumb Sidewise
V2 (m/s)
1
𝐿 𝑏 is the total barrel length. Figure 8 shows the projectile
0 center of gravity plumb and sidewise displacement with
−1 respect to gun tube axis as function of nondimensionalized
−2
longitudinal displacement.
0 0.1 0.2 0.3 0.4 0.5 The projectile axes orientation with respect to line of sight
Time (s) LOS, which includes the projectile roll, plumb and sidewise
swing components, angular displacement, and velocity, are
Estimated
6DOF-R
illustrated in Figures 9 and 10, respectively. The correspond-
6DOF-N ing projectile angular displacement and velocity components
at muzzle are shown in Table 2, where the deflection angle
Figure 11: Projectile velocity vector versus flight time. is the angle between the projectile velocity vector and LOS
Advances in Mechanical Engineering 9
𝑉1 (m/s) 𝑉2 (m/s) 𝑉3 (m/s) 𝛼 (10−3 rad) 𝛽 (10−3 rad) 𝑝 (rad/s) 𝑞 (rad/s) 𝑟 (rad/s)
429.3847 551.9871 1.021021 1.40134 −1.52758 1279.72 −2.92724 −3.15983
1
30
0.6 0
x1
0.4 −15
−30
0.2 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
20
1
0.6 −10
−20
0.4 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
30
0.4
𝛿𝛼R3 /𝛿t (rad/s)
15
0.2
0
0
x3
−15
−0.2
−30
−0.4 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
Estimated Estimated
6DOF-R 6DOF-R
6DOF-N 6DOF-N
Figure 13: The projectile’s axis of symmetry unit vector. Figure 14: Rate of change of the projectile AOA.
while projectile is just flying off from muzzle, and the swing projectile’s position and velocity vectors, using the initial
angle is the angle between projectile axes and LOS at the same flight conditions listed in Table 3. The Doppler radar output
moment. data frequency is assumed to be 100 Hz. The 6DOF-R data
The projectile initial flight conditions corresponding to is assumed to be the real projectile trajectory data. The
the self-propelled artillery launch dynamics, and hence, the corresponding impact point range, drift, maximum altitude,
projectile initial disturbance calculations proposed before are velocity, and flight time are 19689.2 m, 1027.85 m, 8180.9 m,
listed in Table 3, which include the projectile initial velocity 338.76 m/s, and 81.932 s, respectively.
𝑉1 , 𝑉2 , 𝑉3 , initial roll 𝑝, pitch 𝑞 and yaw 𝑟 rates, initial angle Secondly, a nondisturbed trajectory simulation has been
of attack 𝛼, and side-slip angle 𝛽. performed as nominal trajectory 6DOF-N, where the initial
flight conditions used are 700 m/s, 52∘ , and 1279.7 rad/s for
7. Estimation Results muzzle velocity, elevating angle, and spin rate, respectively.
The corresponding impact point range, drift, maximum
A computer program has been developed by using the non- altitude, velocity, and flight time are 19729 m, 995 m, 8162 m,
linear 6-DOF dynamic equations stated in [19]. All simula- 338.66 m/s, and 81.829 s, respectively.
tions were computed with fixed time step size of 0.001 s and Finally, the iterative algorithm shown in Figure 3 has been
integrated using Runge-Kutta-Gill method. applied to estimate the projectile attitude, using the generated
Firstly, a simulated noiseless Doppler radar velocity Doppler radar data 6DOF-R, where the initial flight condi-
measurements data 6DOF-R are generated, including the tions are as the data used to simulate the projectile nominal
10 Advances in Mechanical Engineering
90
30
80
𝛿x1 /𝛿t (s−1 )
15
h1 (rad/s)
0 70
−15 60
−30
0 0.1 0.2 0.3 0.4 0.5 50
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
20
100
10
𝛿x2 /𝛿t (s−1 )
90
h2 (rad/s)
0
−10 80
−20
0 0.1 0.2 0.3 0.4 0.5 70
Time (s) 0 0.1 0.2 0.3 0.4 0.5
Time (s)
30
20
15
𝛿x2 /𝛿t (s−1 )
10
0
h3 (rad/s)
0
−15
−10
−30
0 0.1 0.2 0.3 0.4 0.5
−20
Time (s) 0 0.1 0.2 0.3 0.4 0.5
Estimated Time (s)
6DOF-R Estimated
6DOF-N 6DOF-R
6DOF-N
Figure 15: Rate of change of the projectile’s axis of symmetry unit
vector. Figure 16: Projectile angular momentum per unit 𝐼𝑦 .
trajectory 6DOF-N. Figures 11, 12, and 13 show that the be −0.0205% and −0.0448%, respectively, as the radar data
estimated projectile velocity, AOA, and attitude, respectively, availability is 15% of total flight time.
are matching very well with the simulated Doppler radar
trajectory 6DOF-R, although the nominal (non-disturbed) 8. Conclusion
trajectory results 6DOF-N are mismatching the 6DOF-R.
Figures 14 and 15 show the rate of change of the projectile The main objective of this paper is determining the effect
AOA and attitude, respectively, during flight time, where the of self-propelled artillery dynamics on the projectile impact
estimated values are matched very well with 6DOF-R except point prediction accuracy, and hence, an iterative projectile
at some peaks/dips data points due to the large difference attitude determination algorithm was developed to accurately
between the data frequency obtained from radar and the compute the projectile ballistic trajectory and the corre-
simulation speed of the real trajectory data (1 : 10). Therefore, sponding impact point parameters by using a Doppler radar
the final estimated value shown in Figure 16, which rep- data. The projectile initial disturbance was computed by
resents the projectile angular momentum per-unit 𝐼𝑦 has solving the interaction between the self-propelled artillery
mismatched with 6DOF-R data during these peaks/dips data and the projectile dynamics during launch process. A discrete
points. time transfer matrix DTTM-4DOF was developed by lin-
For nominal trajectory data 6DOF-N, the impact range 𝜀𝑅 earizing the projectile nonlinear modified point mass equa-
and drift 𝜀𝐷 errors relative to 6DOF-R are 0.202% and −3.2%, tions using second-order Taylor series. The main advantages
respectively. All impact point parameters inaccuracies for the of using DTTM-4DOF are (1) linearly relating any two
proposed algorithm as function of simulated Doppler radar successive radar data, and (2) increasing the computation
data availability are shown in Figure 17, where these errors, speed due to solving a set of algebraic equations instead of
𝜀𝑅 and 𝜀𝐷, can be reduced by approximately 8 and 7 times, using traditional numerical methods. An iterative projectile
respectively, as the radar data is available for the first 10% attitude determination algorithm was developed to accurately
of total flight time and approximately by 10 and 70 times to compute the projectile ballistic trajectory and predict the
Advances in Mechanical Engineering 11
0
0.05
−0.01
0
𝜀R (%)
𝜀R (%)
−0.02
−0.05
−0.03
−0.1 −0.04
0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100
Data availability (%) Data availability (%)
10 1.2
5 0.6
𝜀D (%)
𝜀D (%)
0 0
−5 −0.6
−10 −1.2
0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100
Data availability (%) Data availability (%)
Figure 17: Impact range and drift error versus Doppler Radar data availability.
2
impact point parameters by using a Doppler radar data. In Δ𝑡
[m13 m14 m15 ] = [V̇𝑝 V̇𝛼 V̇W ] ,
order to validate the proposed algorithm, a simulated noise- 2
less trajectory data was generated as Doppler radar velocity Δ𝑡
measurements data 6DOF-R for 155 mm HE spin stabilized m21 = (I3 + j22 ) GΔ𝑡,
2
projectile using the nonlinear six-degree-of-freedom trajec-
2
tory model including projectile initial disturbance problem. Δ𝑡
m22 = I3 + V̇V Δ𝑡 + (j22 V̇V + j25 ẆV ) ,
A nominal ballistic trajectory 6DOF-N was generated with 2
52∘ firing elevating angle and without including the effect of 2
Δ𝑡
launch dynamics to show the importance of the projectile ini- m23 = V̇𝑝 Δ𝑡 + [j22 V̇𝑝 + V̇𝑝 𝑝𝑝̇] ,
tial disturbance. Based on the results obtained, the proposed 2
algorithm estimated the projectile velocity and attitude very Δ𝑡2
well compared to the 6DOF-R trajectory data. The impact m24 = V̇𝛼 Δ𝑡 + j22 V̇𝛼 ,
2
range and drift inaccuracies were reduced from 0.202% and 2
Δ𝑡
−3.2% in case of nominal trajectory 6DOF-N to be −0.0205% m25 = V̇W Δ𝑡 + j22 V̇W ,
and −0.0448% by applying the proposed algorithm as the 2
radar data availability is 15% of total flight time. As the projec- Δ𝑡2
tile dynamic stability increased, the radar data availability can m31 = j32 G ,
2
be decreased. Finally, the proposed algorithm can estimate
2
the projectile impact point accurately without computing the Δ𝑡
m32 = j32 [V̇V − ẆV ] ,
projectile initial disturbance, where different projectiles have 2
different initial disturbances based on many factors, such
Δ𝑡2
as, gun barrel wear and erosion, projectile initial position m33 = 1 + 𝑝𝑝̇Δ𝑡 + [j32 V̇𝑝 + 𝑝𝑝̇
2
] ,
and attitude inside barrel, and the propellant charge amount, 2
composition, and homogeneity. 2
Δ𝑡
[m34 m35 ] = j32 [V̇𝛼 V̇W ] ,
2
Appendix
m41 = [𝛼V̇
+ j42 Δ𝑡] G,
Modified Point Mass Discrete Time
+ j42 Δ𝑡] V̇V + j45 ẆV Δ𝑡,
m42 = [𝛼V̇
Transfer Matrix Elements
All DTTM-4DOF elements mentioned in (26) are expressed + j42 Δ𝑡] V̇𝑝 + j43 𝑝𝑝̇Δ𝑡,
m43 = [𝛼V̇
as follows:
+ j42 Δ𝑡] V̇𝛼 ,
m44 = [𝛼V̇
m11 = I3 + GΔ𝑡2 ,
2 + j42 Δ𝑡] V̇W .
m45 = [𝛼V̇
Δ𝑡
m12 = I3 ⋅ Δ𝑡 + V̇V , (A.1)
2
12 Advances in Mechanical Engineering
Research Article
Study on the Dynamics of Laser Gyro Strapdown Inertial
Measurement Unit System Based on Transfer Matrix Method for
Multibody System
Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang, and Qinbo Zhou
Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China
Copyright © 2013 Gangli Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The dynamic test precision of the strapdown inertial measurement unit (SIMU) is the basis of estimating accurate motion of various
vehicles such as warships, airplanes, spacecrafts, and missiles. So, it is paid great attention in the above fields to increase the dynamic
precision of SIMU by decreasing the vibration of the vehicles acting on the SIMU. In this paper, based on the transfer matrix
method for multibody system (MSTMM), the multibody system dynamics model of laser gyro strapdown inertial measurement
unit (LGSIMU) is developed; the overall transfer equation of the system is deduced automatically. The computational results
show that the frequency response function of the LGSIMU got by the proposed method and Newton-Euler method have good
agreements. Further, the vibration reduction performance and the attitude error responses under harmonic and random excitations
are analyzed. The proposed method provides a powerful technique for studying dynamics of LGSIMU because of using MSTMM
and its following features: without the global dynamics equations of the system, high programming, low order of system matrix,
and high computational speed.
x1,2 3
6 5
f1,2
1 4 2
x3,2
7
10
f3,2 11
16
Figure 3: Free body diagram of spring-and-damper hinge 2.
15
8 12 y
9
where O
T 13 14 z x
Uall = [−I T] , Zall = [ZT1,0 ZT3,2 ] ,
Figure 4: Dynamics model of LGSIMU system.
1
1 − (9)
[ 𝑘 + iΩ𝑑 ]
T = U 1 U2 = [
[
].
2 𝑚Ω2 ] 1∼3, platform 7, and foundation 12 are considered as rigid
𝑚Ω 1 −
[ 𝑘 + iΩ𝑑 ] bodies, respectively; since the accelerometers are fixed on
the platform and their masses are relatively small, they
The boundary conditions of the system are are included in the platform; the connections 4∼6 between
T T platform and gyros are considered as spatial spring-and-
Z1,0 = [𝑋1,0 0] , Z3,2 = [𝑊 𝐹3,2 ] . (10) damper hinges; meanwhile, platform and foundation are
linked by isolators 8∼11, which are regarded as spring-and-
Substituting the boundary conditions into (8) yields damper hinges longitudinally vibrating in space. The motion
of the vehicle is the input of LGSIMU system. The isolators
Uall Zall = −B𝑊, (11) 13∼16 between foundation and platform are considered as
spring-and-damper hinges longitudinally vibrating in space.
where
The inertial coordinate system shown in Figure 4 is used
1 to describe the motion of elements. The state vectors of the
−1 −
[ 𝑘 + iΩ𝑑 ] input and output points of every element and the boundary
Uall = [ ],
T
Zall = [𝑋1,0 𝐹3,2 ] , [ 𝑚Ω2 ]
points take the form
0 1− (12)
[ 𝑘 + iΩ𝑑 ] T
Z = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ] . (15)
1
B=[ ].
𝑚Ω2 The state variables in the state vector are complex
amplitudes of displacements, angular displacements, internal
Solving (11), the unknown elements in the state vectors of forces, and internal torques when the system undergoes
boundary points can be obtained as harmonic excitation, respectively.
−1 𝑊 (𝑘 + iΩ𝑑) 1
Zall = −Uall B𝑊 = [ 2] . (13) 4. Topology Figure of the Dynamics
−𝑚Ω2 + 𝑘 + iΩ𝑑 −𝑚Ω
Model of LGSIMU System
As a result, the frequency response function of the system
is The dynamics model of LGSIMU system is constructed with
dynamics elements including bodies and hinges. In order
𝑋1,0 𝑘 + iΩ𝑑 to describe the transfer relationship among the state vectors
𝐻 (Ω) = = . (14)
𝑊 −𝑚Ω2 + 𝑘 + iΩ𝑑 of elements in the system, the topology figure of the model
will be used for deducing the overall transfer equation of
Equation (14) can easily be verified using classical fre- LGSIMU system. Isolators 8∼11 and 13∼16 can be treated
quency response function analysis (i.e., Newton’s second law as one equivalent hinge element, respectively, which will be
of motion and Fourier transform). shown in Section 5.3 ; thus the dynamics model of LGSIMU
system can be dealt with a tree system. Its topology figure can
3. Dynamics Model of LGSIMU System be got readily as illustrated in Figure 5.
Besides the sign conventions introduced in [10], the sign
According to the elements dynamics properties of LGSIMU conventions used in this paper are introduced as follows.
system, the dynamics model shown in Figure 4 is established
as a multibody system. The system can be divided into (1) A circle I denotes a body element and the number
elements which can be modeled as follows: three gyros inside it is the sequence number of the body element.
4 Advances in Mechanical Engineering
0 0 0 0 where Z𝑖,I and Z𝑖,O denote the state vectors of input and
output ends of element 𝑖, respectively. The transfer matrices
13 14 15 16
are
5 6 (𝑖 = 1, 2, 3, 12) ,
0 2 7 3 0
where C denotes the mass center and 𝑚 is the mass of rigid
4 body, JI is the inertia matrix relative to the input point I, ̃lIO
and ̃lIC are the cross product matrices from the input point I
Root
1 to the output point O and mass center C, respectively, and ̃lCO
0 is the cross product matrix from the mass center C to output
point O.
Figure 5: Topology figure of LGSIMU system.
16 Foundation (𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 ) (𝑖 = 13 ∼ 16). The same goes for their output points
in 𝑂2 𝑥𝑦𝑧. The stiffness and damping coefficients of isolators
y 13∼16 are (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) and (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16),
x
O2 15 respectively. Simplify the system of forces acting on the input
z y
O1
ends of isolators 13∼16 into a system of forces only acting on
x the input point of the new element, and simplify the system
z
of forces acting on the output ends of isolators 13∼16 into a
13
Vehicle system of forces only acting on the output point of the new
14
element. Further, since the principle vectors and principle
Figure 6: Equivalent model of isolators. moments of the two new systems of forces should be equal,
the transfer equation of the equivalent spring-and-damper
hinge can be obtained as
bodies, respectively. Hereby, isolators 13∼16 are taken as an
example to deduce the transfer equation and transfer matrix Z13∼16,O = U13∼16 Z13∼16,I . (21)
of the new element. The input and output points of isolator
13 are selected as those of the new element, respectively. The transfer matrix is
Two coordinate systems 𝑂1 𝑥𝑦𝑧 and 𝑂2 𝑥𝑦𝑧 shown in Figure 6
are introduced for deduction, whose origins are at the −1
equilibrium positions of the input and output points of the U13∼16 = [ I6 (K13∼16 + iΩD13∼16 ) ] , (22)
new element. And their directions are the same with the O6×6 I6
coordinate system shown in Figure 4. The coordinates of
the input points of isolators 13∼16 in 𝑂1 𝑥𝑦𝑧 are noted as where Ω is the frequency of harmonic excitation,
K13∼16 =
16 16 16 16 16
[ 0 − ∑ 𝑘𝑦𝑖 𝑐𝑖 − ∑ 𝑘𝑧𝑖 𝑏𝑖 ∑ (𝑘𝑦𝑖 𝑐𝑖2 + 𝑘𝑧𝑖 𝑏𝑖2 ) − ∑ 𝑘𝑧𝑖 𝑎𝑖 𝑏𝑖 − ∑ 𝑘𝑦𝑖 𝑎𝑖 𝑐𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 16 16 ]
[ 2 2 ]
[ ∑ 𝑘𝑥𝑖 𝑐𝑖 0 − ∑ 𝑘 𝑎 − ∑ 𝑘 𝑏 𝑎 ∑ (𝑘 𝑐 + 𝑘 𝑎 ) − ∑ 𝑘 𝑏 𝑐 ]
[ 𝑧𝑖 𝑖 𝑧𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑧𝑖 𝑖 𝑥𝑖 𝑖 𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ 16 ]
[ 16 16 16 16 ]
[− ∑ 𝑘 𝑏 ∑ 𝑘 𝑎 0 − ∑ 𝑘 𝑐 𝑎 − ∑ 𝑘 𝑐 𝑏 ∑ (𝑘 𝑏 2
+ 𝑘 𝑎2 ]
) (23)
[ 𝑥𝑖 𝑖 𝑦𝑖 𝑖 𝑦𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑦𝑖 𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ ].
[ 16 16 16 ]
[ ]
[ − ∑ 𝑘𝑥𝑖 0 0 0 − ∑ 𝑘𝑥𝑖 𝑐i ∑ 𝑘𝑥𝑖 𝑏i ]
[ 𝑖=13 ]
[ 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 ]
[ 0 − ∑ 𝑘 0 ∑ 𝑘 𝑐 0 − ∑ 𝑘 𝑎 ]
[ 𝑦𝑖 𝑦𝑖 i 𝑦𝑖 i ]
[ 𝑖=13 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 ]
[ ]
0 0 − ∑ 𝑘𝑧𝑖 − ∑ 𝑘𝑧𝑖 𝑏i ∑ 𝑘𝑧𝑖 𝑎i 0
[ 𝑖=13 𝑖=13 𝑖=13 ]
Replacing the stiffness coefficients (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) (𝑖 = 13 ∼ 16) platform and gyros 2∼3, respectively. The output point is
in K13∼16 by damping coefficients (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16), the connection point between the platform and gyro 1. The
one can immediately achieve D13∼16 . The transfer equation transfer equation of platform 7 is
and transfer matrix of the equivalent spring-and-damper
Z7,O = U7 Z7,I1 + U7,I2 Z7,I2 + U7,I3 Z7,I3 . (24)
hinge of elements 8∼11 have the same forms as (21) and (22).
The transfer matrices are
5.4. Transfer Equations and Transfer Matrices of Platform 7. I3 −̃lI1 O O3×3 O3×3
Platform 7 is considered as a rigid body with three input ends [ ]
[ O3×3 I3 O3×3 O3×3 ]
and single output end. The first input end is the connection U7 = [
[𝑚Ω2̃l 2 ̃ ̃ ̃l ]
],
point between platform 7 and isolator 8 and the second [ CO −Ω (𝑚lI1 O lI1 C + JI1 ) I3 I1 O ]
2 2̃
and third input points are the connection points between [ 𝑚Ω I3 −𝑚Ω lI1 C O3×3 I3 ]
6 Advances in Mechanical Engineering
O3×3 O3×3 O3×3 O3×3 theory of vibration [20], the steady state response can be
[O3×3 O3×3 O3×3 O3×3 ] expressed as 𝑦7,C = 𝑌7,C eiΩ𝑡 and the complex amplitude 𝑌7,C
U7,I𝑟 =[ ] (𝑟 = 2, 3) .
[O3×3 O3×3 I3 ̃lI𝑟 O ] is the frequency response of LGSIMU system.
[O3×3 O3×3 O3×3 I3 ] The boundary conditions of LGSIMU system are
(25) T
Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]1,0 ,
The geometrical equations of platform 7 are
T
H7 Z7,I1 = H7,I𝑟 Z7,I𝑟 , 𝑟 = 2, 3 (26) Z13∼16,0 = [0, 1, 0, 0, 0, 0, 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]13∼16,0 ,
T
where Z2,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]2,0 ,
I3 O3×3 O3×3 O3×3
H7 = [ ], T
Z3,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]3,0 .
O3×3 I3 O3×3 O3×3
(27) (32)
I ̃lI1 I𝑟 O3×3 O3×3
H7,I𝑟 =[ 3 ] (𝑟 = 2, 3) .
O3×3 I3 O3×3 O3×3 Substituting these boundary conditions into (28) yields
The notations in these matrices are similar to those in Uall Zall = −B, (33)
Section 5.1.
where Uall is a 24 × 24 square matrix composed of the 1st∼
6. Automatic Deduction of the Overall 6th, 19th∼30th, and 37th∼42nd columns of Uall , Zall is a
Transfer Equation of LGSIMU System column matrix consisting of the unknown elements in Zall ,
and B is the 14th column of Uall . For a damped system, Uall
According to the topology figure and the automatic deduc- is a nonsingular square matrix. Solving (33), the unknown
tion method of MSTMM, one can obtain the overall transfer elements in the state vectors of boundary points can be
equation of LGSIMU system: acquired as
Uall Zall = 0. (28) −1
Zall = −Uall B. (34)
The overall transfer matrix is
Using the transfer equations of elements, one can easily
−I12 T13∼16−1 T2−1 T3−1 obtain the frequency response function:
Uall = [O6×12 G13∼16−7 G2−7 O6×12 ] , (29)
[O6×12 G13∼16−7 O6×12 G3−7 ] 𝐻 (Ω) = 𝑌7,C = E7,I1 C U8∼11 U12 U13∼16 Z13∼16,0 , (35)
where where E7,I1 C = [0, 1, 0, −𝑐C , 0, 𝑎C , 0, 0, 0, 0, 0, 0], and (𝑎C , 𝑏C ,
T13∼16−1 = U1 U4 U7 U8∼11 U12 U13∼16 , 𝑐C ) denotes the coordinate of the platform’s mass center in the
body fixed coordinate system.
T2−1 = U1 U4 U7,5 U5 U2 , Similarly, the frequency response functions of other
points can be got. Using the frequency response function, one
T3−1 = U1 U4 U7,6 U6 U3 , can easily obtain the responses under sinusoidal and random
(30)
G13∼16−7 = −H7 U8∼11 U12 U13∼16 , excitations.
If the input is 𝑤 = 𝑊 sin(Ω𝑡+𝛼), the steady state response
G2−7 = H7,5 U5 U2 , is
G3−7 = H7,6 U6 U3 . 𝑦7,C = 𝑊 |𝐻 (Ω)| sin (Ω𝑡 + 𝛼 + 𝜑 (Ω)) , (36)
Zall is a column matrix consisting of the state vectors of system where |𝐻(Ω)| is the amplitude of 𝐻(Ω) and 𝜑(Ω) is the phase
boundary points, that is angle of 𝐻(Ω).
T If the input is a stationary random excitation 𝑤(𝑡) with
Zall = [ZT1,0 ZT13∼16,0 ZT2,0 ZT3,0 ] . (31) PSD 𝑆𝑤 (Ω), the response is also a stationary random process
and its PSD can be expressed as
7. Dynamic Responses of LGSIMU System
𝑆𝑦7,C (Ω) = |𝐻 (Ω)|2 𝑆𝑤 (Ω) . (37)
In order to introduce the method based on MSTMM to
calculate the frequency response, the input and output are For the case that the input is a stationary acceleration
assumed as the displacements of the vehicle and the mass random excitation 𝑤(𝑡)̈ with PSD 𝑆𝑤̈ (Ω) and the output is
center of platform along y-axis, respectively. Setting the input ̈ (𝑡) whose PSD
the acceleration of platform’s mass center 𝑦7,C
to be a harmonic excitation 𝑤 = eiΩ𝑡 and according to the is denoted as 𝑆𝑦7,C
̈ (Ω), regarding the relationship between
Advances in Mechanical Engineering 7
acceleration PSD and displacement PSD 𝑆𝑤̈ (Ω) = Ω4 𝑆𝑤 (Ω), If the response PSD of LGSIMU system undergoing
4
𝑆𝑦7,C
̈ (Ω) = Ω 𝑆𝑦7,C (Ω), and (37), one can obtain random excitation is 𝑆O (Ω), the mean value of the attitude
error caused by random vibration is
2
̈(Ω) = |𝐻 (Ω)| 𝑆𝑤̈
𝑆𝑦7,C (Ω) . (38)
𝐸 (𝛿Φ̇Algo )
According to the Wiener-Khintchine theorem, the auto- ∞ 1 sin ΩT𝑙
correlation function of the response can be obtained as = ∫ Ω [1 + (1 − cos ΩT𝑙 )] − 1 𝑆O (Ω) dΩ.
0 3 ΩT𝑙
(44)
1 ∞
𝑅𝑦7,C
̈(𝜏) = ∫ 𝑆 ̈(Ω) eiΩ𝜏 dΩ. (39)
2𝜋 −∞ 𝑦7,C
9. Numerical Example
Substituting (38) into (39) and setting 𝜏 = 0, the response
For the LGSIMU system shown in Figure 4, the mass and
mean square value can be written as
inertia matrix of foundation 12 are given as
1 ∞ 𝑚12 = 1.2 kg,
𝜙𝑦27,C
̈= 𝑅𝑦7,C
̈(0) = ∫ |𝐻 (Ω)|2 𝑆𝑤̈
(Ω) dΩ. (40)
2𝜋 −∞
0.0045 0 0 (45)
Similarly, the response PSDs and mean square values of J12,C = [ 0 0.0045 0 ] kg ⋅ m2 .
other points can be readily achieved. Then one can analyze [ 0 0 0.0045]
the dynamics performance and errors of LGSIMU system.
The mass and inertia matrix of platform 7 are
8. Error Analysis of LGSIMU 𝑚7 = 2 kg,
In order to improve the precision of LGSIMU in the dynamic 0.0075 0 0 (46)
environment, the errors must be compensated. According to J7,C = [ 0 0.0075 0 ] kg ⋅ m2 .
the error analysis method presented in [18], the attitude errors 0 0 0.0075]
are taken as examples to discuss the errors under sinusoidal [
and random vibrations in this section. The masses and inertia matrices of gyros 1∼3 are
Assuming that LGSIMU system is excited by a sinusoidal
excitation 𝑊 sin (Ω𝑡 + 𝛼), the angular motion response of the 𝑚𝑖 = 0.2 kg,
platform can be written as
0.0006 0 0
𝜃 (𝑡) = e𝑥 𝜃𝑥 (𝑡) + e𝑦 𝜃𝑦 (𝑡) + e𝑧 𝜃𝑧 (𝑡) , (41) J𝑖,C = [ 0 0.0006 0 ] kg ⋅ m2 (𝑖 = 1 ∼ 3) .
[ 0 0 0.0006]
where (47)
𝜃𝑥 (𝑡) = Θ𝑥 sin (Ω𝑡 + 𝜙𝑥 ) , The amplitude and phase frequency responses of the
LGSIMU system got by the proposed method and Newton-
𝜃𝑦 (𝑡) = Θ𝑦 sin (Ω𝑡 + 𝜙𝑦 ) , (42) Euler method are shown in Figures 7 and 8, respectively. It
can be seen that the results got by the two methods have good
𝜃𝑧 (𝑡) = Θ𝑧 sin (Ω𝑡 + 𝜙𝑧 ) . agreements, which verifies the effectiveness of the proposed
method. The attitude error rate under sinusoidal vibration
e𝑥 , e𝑦 , and e𝑧 are the unit vectors of the body-fixed coordinate is demonstrated in Figure 9, where the excitation frequency
frame of platform. Θ𝑥 , Θ𝑦 , Θ𝑧 and 𝜙𝑥 , 𝜙𝑦 , 𝜙𝑧 are the varies from 10 to 2000 Hz, the time interval is 20 ms, and the
amplitudes and phase angles of the angular motion steady amplitude of excitation is 0.2 mm. From this figure, it can be
state responses, respectively, which can be computed using concluded that when the excitation frequency is close to the
MSTMM as presented above. natural frequencies of the LGSIMU system (25 Hz, 46 Hz, and
The error caused by sinusoidal vibration is 85 Hz), attitude error rate reaches maximum values. So the
natural frequencies cannot be close to excitation frequency.
1 Currently, the vibration reduction system of LGSIMU is
𝛿Φ̇Algo = e𝑧 ΩΘ𝑥 Θ𝑦 sin (𝜙𝑥 − 𝜙𝑦 ) in the development phase. Typically, its vibration reduction
2
(43) index is the root mean square value of acceleration response
1 sin ΩT𝑙 should be less than 2 g, and the PSD curve comes down
× {[1 + (1 − cos ΩT𝑙 )] − 1} ,
3 ΩT𝑙 rapidly after 200 Hz [21]. In the following, the dynamics
response and the attitude error response of the LGSIMU
where 𝛿Φ̇Algo is the average attitude error rate, 𝑙 is high speed system under random excitation is computed and analyzed.
algorithm computation cycle index, and T𝑙 is time interval The PSD of the excitation is shown in Figure 10 and its
between 𝑙 cycles. root mean square value is 6.03 g. The PSD of the centroid
8 Advances in Mechanical Engineering
10 ×10−4
2
0
1
−10
0
−20
−1
−30
−2
−40
−3
−50
101 102 103 104 −4
0 500 1000 1500 2000
f (Hz)
f (Hz)
MSTMM
Newton-Euler method Figure 9: Attitude error rate under sinusoidal vibration.
−50
PSD (g 2 /Hz)
0.02
𝜑(Ω) (∘ )
−100 −3 dB/dec
3 dB/dec
−150
20 80 1300 2000
−200 f (Hz)
101 102 103 104
f (Hz) Figure 10: Input PSD.
MSTMM
Newton-Euler method
Figure 8: Phase frequency response. In this paper, the method for calculating the frequency
response function of LGSIMU system is established by using
MSTMM. Based on this method, the responses of LGSIMU
acceleration of platform along y-axis is shown in Figure 11 system are obtained and the dynamics performance is dis-
and its root mean square value is 1.55 g. Since the root cussed. Furthermore, the attitude errors under sinusoidal
mean square value is less than 2 g and the PSD curve comes and random vibrations are calculated and the effect of
down rapidly after 200 Hz, the vibration reduction index damping coefficient on the errors is analyzed. The proposed
is achieved. The isolators 8∼11 are usually identical. The method can also be used to analyze the velocity and position
mean value of the attitude error rate versus the damping errors caused by vibration. The simulation results verify the
coefficient of isolators 8∼11 is shown in Figure 12. The mean effectiveness of the proposed method. Compared with the
value of the attitude error rate decreases with the increasing ordinary method, the proposed method does not need the
of the damping coefficient. Consequently, within the index global dynamics equation. It is also highly stylized, flexible
of vibration reduction, the damping coefficient of isolators for modeling, and easy to program and provides a powerful
should be as large as possible. technique for studying the dynamics of LGSIMU.
Research Article
Recursive Formulation for Dynamic Modeling and Simulation of
Multilink Spatial Flexible Robotic Manipulators
Copyright © 2013 Zhenjie Qian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The dynamics for spatial manipulator arms consisting of n flexible links and n flexible joints is presented. All the transversal,
longitudinal, and torsional deformation of flexible links are considered. Within the total longitudinal deformation, the nonlinear
coupling term, also known as the longitudinal shortening caused by transversal deformation, also is considered here. Each flexible
joint is modeled as a linearly elastic torsional spring, and the mass of joint is considered. Lagrange’s equations are adopted to
derive the governing equations of motion of the system. The algorithmic procedure is based on recursive formulation using 4 × 4
homogenous transformation matrices where all the kinematical expressions as well as the final equations of motion are suited for
computation. A corresponding general-purpose C++ software package for dynamic simulation is developed. Several examples are
simulated to illustrate the performance of the algorithm.
Here, 𝑥𝑖𝑗 , 𝑦𝑖𝑗 , and 𝑧𝑖𝑗 are the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 components of
the elastic linear displacement mode 𝑗 of link 𝑖 at the origin of
the coordinate (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝜃𝑥𝑖𝑗 , 𝜃𝑦𝑖𝑗 , and 𝜃𝑧𝑖𝑗 are
ith link the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 rotation components of the elastic angular
Kti displacement mode 𝑗 of link 𝑖 at the origin of the coordinate
(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝛿𝑖𝑗 is the time-varying amplitude
q2i of mode 𝑗 of link 𝑖, and 𝑁𝑖 is the number of modes used to
Jri 𝜙i2 describe the deformation of link 𝑖.
Tran
smi
ssio q1i
Define 0 W𝑖 or W𝑖 to be the 4 × 4 homogeneous transfor-
n mation matrix from the base coordinate frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0
𝜏i Roto to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then, we have
r
(i − 0 ̂ 𝑖−1 A𝑖 ,
1)th
lin k
W𝑖 = W𝑖 = W𝑖−1 E𝑖−1 Ai = W (7)
Figure 2: Flexible joint model. 2.4. Velocity of a Point of Link 𝑖. Let 𝑖 h𝑖 (𝜂) be the homo-
geneous coordinates in the system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 of a point of
axis of link 𝑖 in undeformed shape). This will be referred the deformed link 𝑖 at position 𝜂 with the link under an
to as the base reference frame of link 𝑖. Fix the coordinate undeformed condition from the origin of (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then,
system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 at the distal end of link 𝑖. This is the 𝑖
h𝑖 (𝜂) can be approximated as
distal frame of link 𝑖. When link 𝑖 is in its undeformed
state, the distal frame can be located by a pure translation 𝑁𝑖
𝑖 T T
of the base reference frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 along the length 𝐿 𝑖 h𝑖 (𝜂) = [1 𝜂 0 0] + ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]
of link 𝑖. Let (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 and (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 be two Denavit- 𝑗=1
̂̈ ̈ ̇ ̇ ̈ + ∑ ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘
̇𝑇 ,
𝑖𝑗𝑘
W𝑖 = W𝑖 E𝑖 + 2W𝑖 E𝑖 + W𝑖 E𝑖 , 𝑖=1 𝑗=1 𝑘=1
(18)
where
where
𝑁𝑖
Ė ̇ 1 𝐿𝑖
𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 , 𝑇𝑖𝑗𝑘 = ∫ 𝐽 𝜃 𝜃 𝑑𝜂,
𝑗=1 2 0 𝑥𝑖 𝑥𝑖𝑗 𝑥𝑖𝑘
(14) 1 𝐿𝑖 𝑖 ̇𝑖 ̇
𝑁𝑖 B1𝑖 = ∫ 𝜇 h hT 𝑑𝜂,
Ë ̈
𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 .
2 0 𝑖 𝑖 𝑖
(19)
𝑗=1 1 𝐿𝑖 𝑖 𝑖 ̇
B2𝑖 = ∫ 𝜇 h hT 𝑑𝜂,
2 0 𝑖 𝑖 𝑖
3. Dynamics of Flexible Robots 1 𝐿𝑖 𝑖 𝑖 T
B3𝑖 = ∫ 𝜇 h h 𝑑𝜂.
To use Lagrange’s equations, we need the kinetic and potential 2 0 𝑖 𝑖 𝑖
energy of the system. With the consideration of 𝑖 h𝑖 expressed in (8) and its
derivative, consider
3.1. The System Kinetic Energy. The system kinetic energy 𝐾 𝑁𝑖
contains two parts: the links kinetic energy 𝐾𝑏 and the joints 𝑖
ḣ ̇
𝑖 (𝜂) = ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]
T
𝑗=1𝑘=1
Assume that the links are slender beams, so the rotary
inertia and shear effects can be neglected. Therefore, the the matrices B1𝑖 , B2𝑖 , and B3𝑖 can be written as
present analysis is based on the Euler-Bernoulli beam theory. 𝑁𝑖 𝑁𝑖
Also assume that the links can undergo a large overall rigid B1𝑖 = ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘
̇D
𝑖𝑗𝑘
motion, however the elastic displacements are small. The 𝑗=1 𝑘=1
kinetic energy of the 𝑖th link is 𝑁𝑖 𝑁𝑖 𝑁𝑖
+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘
̇𝛿 (−F − FT )
𝑖𝑙 𝑖𝑗𝑘𝑙
1 𝐿𝑖 𝑖𝑗𝑘𝑙
𝐾𝑖𝑏 = ∫ 𝜇 (𝜂) Tr {ḣ ̇
T
𝑖 h𝑖 } 𝑑𝜂
𝑗=1 𝑘=1 𝑙=1
2 0
(16) 𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖
1 𝐿𝑖 𝜕𝜃 (𝜂, 𝑡)
2
+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙̇𝛿𝑖𝑠 E𝑖𝑗𝑘𝑙𝑠 ,
+ ∫ 𝐽𝑥𝑖 (𝜂) ( 𝑥𝑖 ) 𝑑𝜂, 𝑗=1 𝑘=1 𝑙=1 𝑠=1
2 0 𝜕𝑡
(21)
where Tr{⋅} is the trace operator; 𝜇(𝜂) and 𝐽𝑥𝑖 are the mass per
𝑁𝑖 𝑁𝑖 𝑁𝑖
unit length and the polar moment of inertia per unit length
of the link about the neutral axis 𝑥, respectively. For slender B2𝑖 = ∑ 𝛿𝑖𝑗̇D𝑖𝑗 + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 DT𝑖𝑗𝑘
𝑗=1 𝑗=1 𝑘=1
beams with uniform cross section area along 𝑥 axis, 𝜇(𝜂) = 𝜇𝑖 .
The first term in (16) is the kinetic energy of link 𝑖 accounting 𝑁𝑖 𝑁𝑖
for the rigid-body motion and the lateral and longitudinal + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 (−E𝑖𝑗𝑘 )
deformation due to flexibility, whereas the second term is the 𝑗=1 𝑘=1
Advances in Mechanical Engineering 5
𝑁𝑖 𝑁𝑖 𝑁𝑖
1 Case 2. Link 𝑖 consists of a flexible beam with the con-
+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 (− FT𝑖𝑗𝑘𝑙 − F𝑖𝑙𝑗𝑘 ) centrated mass 𝑚𝑖 at its proximal end. To account for the
𝑗=1 𝑘=1 𝑙=1 2
contribution of the concentrated mass to the kinetic energy
𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖 of link 𝑖, the extra term D𝑖 should be added to the matrix D𝑖 ,
1
+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑙𝑘𝑠 ), where
𝑗=1 𝑘=1 𝑙=1 𝑠=1 2
1
D𝑖 = 𝑚𝑖 [1 0 0 0] [1 0 0 0] .
T
(26)
(22) 2
𝑁𝑖 𝑁𝑖 𝑁𝑖
B3𝑖 = D𝑖 + ∑𝛿𝑖𝑗 (D𝑖𝑗 + DT𝑖𝑗 ) + ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 D𝑖𝑗𝑘 Should the concentrated mass 𝑚𝑖 locate at the position
𝑗=1 𝑗=1 𝑘=1
(𝑥0 , 𝑦0 , 𝑧0 ) near the proximal end, the extra matrix D𝑖 is
modified to be
𝑁𝑖 𝑁𝑖
1 1
+ ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 (− E𝑖𝑗𝑘 − ET𝑖𝑗𝑘 ) 1
D𝑖 = 𝑚𝑖 [1 𝑥0 𝑦0 𝑧0 ] [1 𝑥0 𝑦0 𝑧0 ] .
T
2 2 (27)
𝑗=1 𝑘=1 2
𝑁𝑖 𝑁𝑖 𝑁𝑖
(23)
1 1 Case 3. Link 𝑖 consists of a flexible beam with the concen-
+ ∑ ∑ ∑𝛿𝑖𝑗 𝛿𝑖𝑘 𝛿𝑖𝑙 (− F𝑖𝑗𝑘𝑙 − FT𝑖𝑗𝑘𝑙 ) trated mass 𝑚𝑖 at its distal end. Considering the contribution
𝑗=1 𝑘=1 𝑙=1 2 2
of the concentrated mass to the kinetic energy of link 𝑖, the
𝑁 𝑁 𝑁 𝑁 extra terms D𝑖 , D𝑖𝑗 , D𝑖𝑗𝑘 , E𝑖𝑗𝑘 , F𝑖𝑗𝑘𝑙 , and E𝑖𝑗𝑘𝑙𝑠 should be added
𝑖 𝑖 𝑖 𝑖
1
+ ∑ ∑ ∑ ∑𝛿𝑖𝑗 𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( E𝑖𝑗𝑙𝑘𝑠 ), to the corresponding matrix D𝑖 , D𝑖𝑗 , D𝑖𝑗𝑘 , E𝑖𝑗𝑘 , F𝑖𝑗𝑘𝑙 , and E𝑖𝑗𝑘𝑙𝑠 ,
𝑗=1 𝑘=1 𝑙=1 𝑠=1 4 respectively. Here,
where 1
D𝑖 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝐿 𝑖 0 0] ,
T
1 𝐿𝑖 2
D𝑖 = ∫ 𝜇 L LT 𝑑𝜂, 1
2 0 𝑖 𝑖 𝑖 D𝑖𝑗 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )] ,
T
2
1 𝐿𝑖
D𝑖𝑗 = ∫ 𝜇 L UT 𝑑𝜂, 1
D𝑖𝑗𝑘 = 𝑚𝑖 [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )]
T
2 0 𝑖 𝑖 𝑖𝑗
2
1 𝐿𝑖
D𝑖𝑗𝑘 = ∫ 𝜇 U UT 𝑑𝜂, × [0 𝑥𝑖𝑘 (𝐿 𝑖 ) 𝑦𝑖𝑘 (𝐿 𝑖 ) 𝑧𝑖𝑘 (𝐿 𝑖 )] ,
2 0 𝑖 𝑖𝑗 𝑖𝑘
(24) 1
E𝑖𝑗𝑘 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝑥𝑖𝑗𝑘 (𝐿 𝑖 ) 0 0] ,
T
1 𝐿𝑖
E𝑖𝑗𝑘 = ∫ 𝜇𝑖 L𝑖 ST𝑖𝑗𝑘 𝑑𝜂, 2
2 0 1
T
1 𝐿𝑖 F𝑖𝑗𝑘𝑙 = 𝑚𝑖 [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )]
F𝑖𝑗𝑘𝑙 = ∫ 𝜇 U ST 𝑑𝜂, 2
2 0 𝑖 𝑖𝑗 𝑖𝑘𝑙 × [0 𝑥𝑖𝑘𝑙 (𝐿 𝑖 ) 0 0] ,
1 𝐿𝑖 1
E𝑖𝑗𝑘𝑙𝑠 = ∫ 𝜇 S ST 𝑑𝜂, T
E𝑖𝑗𝑘𝑙𝑠 = 𝑚𝑖 [0 𝑥𝑖𝑗𝑘 (𝐿 𝑖 ) 0 0] [0 𝑥𝑖𝑙𝑠 (𝐿 𝑖 ) 0 0] .
2 0 𝑖 𝑖𝑗𝑘 𝑖𝑙𝑠 2
in which (28)
L𝑖 = [1 𝜂 0 0] ,
T
For the calculation of the kinetic energy of the joint 𝑖,
T
we can lump its mass to link 𝑖 − 1 in accordance with the
U𝑖𝑗 = [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)] , (25) assumptions made in [21] for simplicity. Thus, the kinetic
T energy needs to be included only in the part accounting for
S𝑖𝑗𝑘 = [0 𝑥𝑖𝑗𝑘 (𝜂) 0 0] . spinning kinetic energy of the rotor of joint 𝑖 as follows:
In (21)–(23), the terms with underline remained in high- 1
order approximation coupling model. 𝐾𝑖𝑟 = 𝐽𝑟𝑖 𝜑𝑖̇
2
. (29)
2
It should be noted that the link shape mentioned above is
restricted to be the slender beam type. In fact, the link shape Considering the relationship in (2). We can obtain the
can further be extended to the other cases. kinetic energy of the joints as follows:
Case 1. Link 𝑖 is the rigid-body with irregular shape. In this 𝑛 𝑛
1
case, 𝑁𝑖 = 0, B1𝑖 = B2𝑖 = 0, and B3𝑖 = D𝑖 . The term D𝑖 is 𝐾𝑟 = ∑𝐾𝑖𝑟 = ∑ ( 𝐽𝑟𝑖 𝜙𝑖2 𝑞1𝑖
2
̇) . (30)
the equivalent of the inertia moment of the rigid-link. It is 𝑖=1 𝑖=1 2
actually the pseu-matrix of the inertia moment in [25] with
its more complex form compared with that of (23). Thus, for 3.2. The System Potential Energy. The potential energy of the
rigid-link, the link shape can be arbitrary, and one should flexible manipulators 𝑉 mainly includes the elastic potential
only input the corresponding matrix D𝑖 . energy of flexible joints 𝑉𝑟 , the elastic potential energy of the
6 Advances in Mechanical Engineering
flexible links 𝑉𝑏 , and their gravitational potential energy 𝑉𝑔 . Finally, by using the same procedure given by Book [14],
Therefore, the whole system potential energy is the gravity potential of the system is
𝑉 = 𝑉𝑟 + 𝑉𝑏 + 𝑉𝑔 . (31) 𝑛
𝑉𝑔 = −gT ∑W𝑖 r𝑖 , (37)
Firstly, the elastic potential energy of joint 𝑖 is 𝑖=1
where 𝜃𝑦𝑖𝑘 and 𝜃𝑧𝑖𝑘 are mentioned in (6). By integrating 𝑑𝑉𝑖𝑏 For the joint variable 𝑞2𝑗 , the following simply joint
of (33) over the link, and summing over all 𝑛 links, one can variable equation:
obtain 𝑉𝑏 as 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉
𝑛 𝑁𝑖 𝑁𝑖
( )− + =0 (𝑗 = 1, 2, . . . , 𝑛) . (41)
𝑏
𝑛 𝐿𝑖
1 ̇
𝑑𝑡 𝜕𝑞2𝑗 𝜕𝑞2𝑗 𝜕𝑞2𝑗
𝑉 = ∑ (∫ 𝑑𝑉𝑖𝑏 ) = ∑ ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 𝐾𝑖𝑗𝑘 , (35)
𝑖=1 0 2 𝑖=1 𝑗=1 𝑘=1
For the deformation variable 𝛿𝑗𝑓 , the following simply
where deformation variable equation:
𝐾𝑖𝑗𝑘 = 𝐾𝑥𝑖𝑗𝑘 + 𝐾𝑦𝑖𝑗𝑘 + 𝐾𝑧𝑖𝑗𝑘 + 𝐾𝑡𝑖𝑗𝑘 , 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉
( )− + =0
𝐿𝑖 𝜕𝑥𝑖𝑗 𝜕𝑥𝑖𝑘 ̇
𝑑𝑡 𝜕𝛿𝑗𝑓 𝜕𝛿𝑗𝑓 𝜕𝛿𝑗𝑓 (42)
𝐾𝑥𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐴 𝑥𝑖 (𝜂) 𝑑𝜂,
0 𝜕𝜂 𝜕𝜂 (𝑗 = 1, 2, . . . , 𝑛; 𝑓 = 1, 2, . . . , 𝑁𝑗 ) .
𝐿𝑖 𝜕𝜃𝑦𝑖𝑗 𝜕𝜃𝑦𝑖𝑘
𝐾𝑦𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐼𝑦𝑖 (𝜂) 𝑑𝜂, Upon the substitution of the system kinetic energy of (15),
0 𝜕𝜂 𝜕𝜂 (36)
the system elastic potential energy of (31), and the gravity
𝐿𝑖 𝜕𝜃𝑧𝑖𝑗 𝜕𝜃𝑧𝑖𝑘 energy of (37) into (40), (41), and (42), thus becoming as
𝐾𝑧𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐼𝑧𝑖 (𝜂) 𝑑𝜂,
0 𝜕𝜂 𝜕𝜂 follows.
𝐿𝑖 𝜕𝜃𝑥𝑖𝑗 𝜕𝜃𝑥𝑖𝑘 For the joint variable 𝑞1𝑗 ,
𝐾𝑡𝑖𝑗𝑘 = ∫ 𝐺𝑖 𝐼𝑥𝑖 (𝜂) 𝑑𝜂.
0 𝜕𝜂 𝜕𝜂 𝐽𝑟𝑗 𝑛𝑖2 𝑞1𝑗
̈ = −𝐾𝑡𝑗 𝑞1𝑗 + 𝐾𝑡𝑗 𝑞2𝑗 + 𝜏𝑗 (𝑗 = 1, 2, . . . , 𝑛) . (43)
Advances in Mechanical Engineering 7
𝑛 𝑁 𝑁 𝑁
𝜕W𝑖 𝑖
T ̈
𝑖 𝑖
̇𝛿̇
[B3𝑖 Ẅ
T 𝑁𝑖 𝑁𝑖
2∑ Tr { 𝑖 + ∑ H W 𝛿
𝑖𝑘 𝑖 𝑖𝑘 + ∑ ∑ L𝑖𝑘𝑙 WT𝑖 𝛿𝑖𝑘 𝑖𝑙
𝑖=𝑗 𝜕𝑞2𝑗 𝑘=1 𝑘=1 𝑙=1
H𝑖𝑗 = D𝑖𝑗 + ∑ 𝛿𝑖𝑘 DT𝑖𝑗𝑘 + ∑ 𝛿𝑖𝑘 (−E𝑖𝑗𝑘 )
𝑘=1 𝑘=1
𝑁𝑖
̇]}
+ 2 ∑ H𝑖𝑘 ẆT𝑖 𝛿𝑖𝑘 𝑁 𝑁
𝑘=1
𝑖 𝑖
1
+ ∑ ∑𝛿𝑖𝑘 𝛿𝑖𝑙 (− FT𝑖𝑗𝑘𝑙 − F𝑖𝑙𝑗𝑘 )
𝑛 𝑘=1 𝑙=1
2
𝜕W𝑖
= gT ∑ r𝑖 + 𝐾𝑡𝑗 𝑞1𝑗 − 𝐾𝑡𝑗 𝑞2𝑗 .
𝑖=𝑗 𝜕𝑞2𝑗 𝑁 𝑁 𝑁
𝑖 𝑖 𝑖
1
(44) + ∑ ∑ ∑𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑙𝑘𝑠 ) ,
𝑘=1 𝑙=1 𝑠=1
2
For the deformation variables 𝛿𝑗𝑓 (𝑗 =
𝑁𝑖
1, 2, . . . , 𝑛; 𝑓 = 1, 2, . . . , 𝑁𝑗 ), (46)
L𝑖𝑗𝑘 = −E𝑖𝑗𝑘 + ∑𝛿𝑖𝑙 (−F𝑖𝑙𝑗𝑘 )
𝑁 𝑁 𝑁 𝑙=1
𝑛
𝜕W𝑖 𝑖
T ̈
i 𝑖
̇𝛿̇
2 ∑ Tr { [B3𝑖 Ẅ
T
𝑖 + ∑ H 𝑖𝑘 W 𝑖 𝛿𝑖𝑘 + ∑ ∑ L𝑖𝑘𝑙 WT𝑖 𝛿𝑖𝑘 𝑖𝑙
𝑖=𝑗+1 𝜕𝛿𝑗𝑓 𝑘=1 𝑘=1 𝑙=1
𝑁 𝑁
𝑖 𝑖
1
+ ∑ ∑𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑘𝑙𝑠 + ET𝑖𝑗𝑙𝑘𝑠 − E𝑖𝑗𝑙𝑘𝑠 ) ,
𝑙=1 𝑠=1
2
𝑁𝑖
̇]}
+ 2 ∑ H𝑖𝑘 ẆT𝑖 𝛿𝑖𝑘 𝑁𝑖 𝑁𝑖 𝑁𝑖
𝑘=1 N𝑖𝑗𝑘 = DT𝑖𝑗𝑘 + ∑𝛿𝑖𝑙 (−FT𝑖𝑗𝑘𝑙 − F𝑖𝑘𝑗𝑙 ) + ∑ ∑𝛿𝑖𝑙 𝛿𝑖𝑠 ET𝑖𝑗𝑙𝑘𝑠 ,
𝑙=1 𝑙=1 𝑠=1
𝑁 𝑁 𝑁
{ 𝑗 𝑗 𝑗
+ 2 Tr {[ẄH + ∑ W N 𝛿 ̈ + ∑ ∑ ̇ 𝛿̇
W𝑗 P𝑗𝑓𝑘𝑙 𝛿𝑗𝑘 𝑁𝑖
𝑗 𝑗𝑓 𝑗 𝑗𝑓𝑘 𝑗𝑘 𝑗𝑙
𝑘=1 𝑘=1 𝑙=1 P𝑖𝑗𝑘𝑙 = −FT𝑖𝑗𝑘𝑙 + ∑𝛿𝑖𝑠 ET𝑖𝑗𝑠𝑘𝑙 .
{[ 𝑠=1
𝑁𝑗 𝑁𝑗
}
+ 2 ∑ Ẇ𝑗 N𝑗𝑓𝑘 𝛿𝑗𝑘̇ ] WT + 2 ∑ 𝛿 ̈
𝑗} 𝑗𝑘 𝑇𝑗𝑓𝑘 Finally, by complicated derivation and assembling, and
𝑘=1 𝑘=1
] } using the recursive scheme to reduce the number of calcu-
𝑛
𝜕W𝑖 lation, the dynamic equations of the system are obtained in
= gT ∑ ( r𝑖 ) + gT W𝑗 𝜀𝑗𝑓 the following formulation:
𝑖=𝑗+1 𝜕𝛿 𝑗𝑓
𝑁𝑗 𝑁𝑗 = R𝐼 ,
Jz̈ (47)
T
− g W𝑗 ∑ 𝛿𝑗𝑘 𝜅𝑗𝑘𝑙 − ∑ 𝛿𝑗𝑘 𝐾𝑗𝑓𝑘 .
𝑘=1 𝑘=1
(45) where the generalized coordinates column z is defined as
T
z = [𝑞11 𝑞21 𝛿11 𝛿12 ⋅ ⋅ ⋅ 𝛿1𝑁1 𝑞12 𝑞22 𝛿21 ⋅ ⋅ ⋅ 𝛿2𝑁2 ⋅ ⋅ ⋅ 𝑞1𝑛 𝑞2𝑛 𝛿𝑛1 ⋅ ⋅ ⋅ 𝛿𝑛𝑁𝑛 ] , (48)
where for 𝑗 = 1, . . . , 𝑛 − 1; ℎ = 1, . . . , 𝑗,
𝑛
𝑗̃
Fℎ = ∑ 𝑗̃ ̃ T,
W𝑖 B3𝑖 ℎ W 𝑖 (52) 𝐼𝑗𝑓ℎ = 2 Tr {W ̂ ℎ−1 Uℎ [ 𝑗 𝜙̃T MT + ℎ W ̃ 𝑗 H𝑗𝑓 ] WT } ; (59)
ℎ 𝑗𝑓 𝑗
𝑖=max(ℎ,𝑗)
for 𝑗 = 1, . . . , 𝑛 − 1; ℎ = 𝑗 + 1, . . . , 𝑛 − 1,
ℎ̃ 𝑗
W𝑖 = Eℎ Aℎ+1 ⋅ ⋅ ⋅ E𝑖−1 A𝑖 . (53) 𝐼𝑗𝑓ℎ = 2 Tr {(W𝑗 M𝑗𝑓 ) 𝜙̃ℎ UTℎ W
̂T } .
ℎ−1 (60)
for ℎ = 1, . . . , 𝑗 − 1; 𝑗 = 2, . . . , 𝑛, for 𝑗 = ℎ; ℎ = 1, . . . , 𝑛 − 1,
̂ 𝑗−1 U𝑗 ) 𝑗 Fℎ MT WT } .
𝐽𝑗ℎ𝑘 = 2 Tr {(W (56) 𝐼𝑗𝑓ℎ𝑘 = 2 Tr {M𝑗𝑓 𝑗 𝜙ℎ MT𝑗𝑘 + N𝑗𝑓𝑘 } + 2𝑇𝑗𝑓𝑘 ; (64)
ℎ𝑘 ℎ
Here, for 𝑗 = 2, . . . , 𝑛 − 1; ℎ = 1, . . . , 𝑗 − 1,
𝑛
𝑗 𝑗̃
Fℎ = ∑ W𝑖 B3𝑖 ℎ WT𝑖 , 𝐼𝑗𝑓ℎ𝑘 = 2 Tr {W𝑗 [M𝑗𝑓 𝑗 𝜙ℎ + HT𝑗𝑓 WT𝑗 ] MTℎ𝑘 WTℎ } ;
ℎ
(65)
𝑖=max(ℎ+1,𝑗) (57)
ℎ
W𝑖 = Aℎ+1 Eℎ+1 ⋅ ⋅ ⋅ E𝑖−1 A𝑖 , for 𝑗 = 1, . . . , 𝑛 − 2; ℎ = 𝑗 + 1, . . . , 𝑛 − 1,
𝐼 𝐼
The 𝑅𝑟𝑗 is the remains of (43) with the second derivatives The 𝑅𝐽𝑗 is the remains of (44) with the second derivatives
removed as follows: removed as follows: for 𝑗 = 1,
𝐼
𝐼
𝑅𝑟𝑗 = −𝐾𝑡𝑗 𝑞1𝑗 + 𝐾𝑡𝑗 𝑞2𝑗 + 𝜏𝑗 (𝑗 = 1, . . . , 𝑛) . (69) 𝑅𝐽1 = −2 Tr {U1 Q1 } + gT U1 P1 + 𝐾𝑡1 𝑞11 − 𝐾𝑡1 𝑞21 ; (70)
Advances in Mechanical Engineering 9
for 𝑗 = 2, . . . , 𝑛, ̂̈
The value of W𝑖 can be obtained similarly as follows:
𝐼
𝑅𝐽𝑗 ̂ 𝑗−1 U𝑗 Q𝑗 } + gT W
= −2 Tr {W ̂ 𝑗−1 U𝑗 P𝑗 + 𝐾𝑡𝑗 𝑞1𝑗 − 𝐾𝑡𝑗 𝑞2𝑗 . 𝑖 𝑁ℎ
(71) ̂̈
W ̂ ℎ
̈
𝑖 = ∑ (Wℎ−1 Uℎ W𝑖 𝑞2ℎ + ∑ Wℎ Mℎ𝑘 ℎ W ̈
̂ 𝑖 𝛿ℎ𝑘 ̂̈
)+W k𝑖 ,
ℎ=1 𝑘=1
𝐼 (76)
The 𝑅𝑗𝑓 is the remains of (45) with the second derivatives
removed as follows.
For 𝑗 = 1, . . . , 𝑛 − 1, where
ℎ
𝐼
𝑅𝐽𝑗 = −2 Tr {W𝑗 M𝑗𝑓 A𝑗+1 Q𝑗+1 W𝑖 = Eℎ Aℎ+1 ⋅ ⋅ ⋅ A𝑖 E𝑖 ,
ℎ̂
(77)
𝑁𝑗 W𝑖 = Aℎ+1 Eℎ+1 ⋅ ⋅ ⋅ A𝑖 E𝑖 .
+ [Ẅ ̇ ̇
k𝑗 H𝑗𝑓 + 2W𝑗 ( ∑ N𝑗𝑓𝑘 𝛿𝑗𝑘 )
𝑘=1
For accelerating the computation of the generalized
[ ̂̈ 𝑗
𝑁𝑗 𝑁𝑗
inertia matrix of the system, the terms Ẅ ̃
k𝑖 , Wk𝑖 , Q𝑗 , P𝑗 , Fℎ ,
𝑗̃ 𝑗
̇ 𝛿̇)]
+W𝑗 ( ∑ ∑P𝑗𝑓𝑘𝑙 𝛿𝑗𝑘 𝜙ℎ , ℎ F𝑗 , and 𝜙ℎ defined in (52)–(76), respectively, can be
𝑗𝑙
𝑘=1 𝑙=1 calculated recursively as follows:
] (72)
T
×W𝑗 } Ẅ 2
̇,
k1 = U21 𝑞21
̂̈ ̂̇
𝑁𝑗
Ẅ ̂ 𝑖−1 U2𝑖 𝑞2̇,
̇+W
k𝑖 = Wk𝑖−1 A𝑖 + 2W𝑖−1 U𝑖 𝑞2𝑖
− ∑ 𝛿𝑗𝑘 𝐾𝑗𝑓𝑘 + gT W𝑗 M𝑗𝑓 A𝑗+1 P𝑗+1 2𝑖
𝑘=1 ̂̈
W ̈ ̇ ̇
k𝑖 = Wk𝑖 E𝑖 + 2W𝑖 E𝑖 ,
𝑁𝑗
T T 𝑁𝑛
+ g W𝑗 𝜀𝑗𝑓 − g W𝑗 ∑ 𝛿𝑗𝑘 𝜅𝑗𝑓𝑘 ; Q𝑛 = B3𝑛 Ẅ
T ̇ ̇T
k𝑛 + 2 ( ∑ H𝑛𝑘 𝛿𝑛𝑘 ) W𝑛
𝑘=1
𝑘=1
for 𝑗 = 𝑛, 𝑁𝑛 𝑁𝑛 (78)
̇ 𝛿̇) WT ,
+ ( ∑ ∑L𝑛𝑘𝑙 𝛿𝑛𝑘 𝑛𝑙 𝑛
𝑁𝑛
𝑘=1 𝑙=1
I
𝑅𝑛𝑓 = −2 Tr [Ẅ ̇ ̇
k𝑛 H𝑛𝑓 + 2W𝑛 ( ∑ N𝑛𝑓𝑘 𝛿𝑛𝑘 ) 𝑁𝑗
𝑘=1 ̇
Q𝑗 = B3𝑖 Ẅ
T ̇T
k𝑖 + 2 ( ∑ H𝑗𝑘 𝛿𝑗𝑘 ) W𝑗
𝑁𝑛 𝑁𝑛
𝑘=1
̇ 𝛿̇) WT ]
+ W𝑗 ( ∑ ∑P𝑛𝑓𝑘𝑙 𝛿𝑛𝑘 𝑛𝑙 𝑛 𝑁𝑗 𝑁𝑗
𝑘=1 𝑙=1
(73) ̇ 𝛿̇) WT + 𝑗 W
+ ( ∑ ∑L𝑗𝑘𝑙 𝛿𝑗𝑘 ̃ 𝑗+1 Q𝑗+1 ,
𝑁𝑛 𝑗𝑙 𝑗
𝑘=1 𝑙=1
− ∑ 𝛿𝑛𝑘 𝐾𝑛𝑓𝑘 + gT W𝑛 𝜀𝑛𝑓
𝑁𝑛 𝑁 𝑁
𝑘=1 1 𝑛 𝑛
P𝑛 = r𝑛 = M𝑛 r𝑟𝑛 + ∑ 𝛿𝑛𝑘 𝜀𝑛𝑘 − ∑ ∑𝛿 𝛿 𝜅 , (79)
T
𝑁𝑛
𝑘=1
2 𝑘=1 𝑙=1 𝑛𝑘 𝑛𝑙 𝑛𝑘𝑙
− g W𝑛 ∑ 𝛿𝑛𝑘 𝜅𝑛𝑓𝑘 .
𝑗
𝑘=1 P𝑗 = r𝑗 + W ̃ 𝑗+1 P𝑗+1 . (80)
Here,
𝑗
As for F ̃ ℎ , for 𝑗 = ℎ = 𝑛,
𝑛 𝑁𝑖
̃ 𝑖 B3𝑖 Ẅ
Q𝑗 = ∑ (𝑗 W 𝑗̃ ̇T ̇
k𝑖 + 2 ∑ W𝑖 H𝑖𝑘 W𝑖 𝛿𝑖𝑘 𝑛̃
𝑖=𝑗 𝑘=1 F𝑛 = B3𝑛 , (81)
𝑁𝑖 𝑁𝑖
𝑗
̇𝛿̇) ,
̃ 𝑖 L𝑖𝑘𝑙 WT 𝛿𝑖𝑘 for 𝑗 = 𝑛; ℎ = 1, . . . , 𝑛 − 1,
+∑ ∑ W 𝑖 𝑖𝑙
(74)
𝑘=1 𝑙=1 𝑛̃
𝑛
̃ T,
Fℎ = B3𝑖 ℎ W (82)
𝑛
𝑗
̃ 𝑖 r𝑖 ,
P𝑗 = ∑ W
𝑖=𝑗 for 𝑗 = ℎ; ℎ = 1, . . . , 𝑛 − 1,
𝑖 𝑖−1 𝑁ℎ 𝑗̃ ℎ T 𝑗+1
̈ ̃ + (E𝑖 A𝑗+1 ) ⋅
Fℎ = B3𝑗 ⋅ W ̃ℎ,
F
Ẅ ̂ ℎ̃
̈
𝑖 = ∑ Wℎ−1 Uℎ W𝑖 𝑞2ℎ + ∑ ∑ Wℎ Mℎ𝑘 ℎ W𝑖 𝛿ℎ𝑘 + Ẅ
k𝑖 .
𝑗
ℎ=1 ℎ=1 𝑘=1 T (84)
𝑗̃ ℎ
(75) Fℎ = ( F̃𝑗) .
10 Advances in Mechanical Engineering
×10−6 × 10−5
1.0
0.6 high-order
0.0
0.5
0.6 zero-order
2 high-order
uy (m)
uy (m)
0.0
−2.0
2 zero-order
−0.5
−4.0 −1.0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t (s) t (s)
Figure 4: 𝜔0 = 3 rad/s.
As for 𝑗 Fℎ , only if 𝑁ℎ > 0, calculate
𝑗
𝑗
Fℎ = F ̃ ℎ+1 ⋅AT (𝑗 = 1, . . . , 𝑛; ℎ = 1, . . . , 𝑛 − 1) . (85) [3, 8], and they are given as follows. The length 𝐿 1 = 8 m,
ℎ+1
the cross-section area 𝐴 𝑥1 = 7.2968 × 10−5 m2 , the area
As for 𝑗 𝜙̃ℎ , it can be obtained from ℎ F𝑗 easily as follows: moment of inertia 𝐼 = 8.2189 × 10−9 m4 , the mass density
𝜌 = 2.7667 × 103 kg/m3 , and the elasticity modulus 𝐸1 =
𝑗̃ T 6.8952 × 1010 N/m2 . For the flexible cantilever beam without
𝜙ℎ = ( ℎ F𝑗 ) . (86)
large overall motion, the fundamental frequency is 0.464 Hz.
The large rotating motion law of the system adopted 𝜃̇in [3, 8]
As for 𝑗 𝜙ℎ , only if 𝑁𝑗 > 0 and 𝑁ℎ > 0, calculate
is given by
𝜔 𝜔
{ 0 𝑡 − 0 sin ( 2𝜋 𝑡) , 0 ≤ 𝑡 ≤ 𝑇,
𝑗
𝑗
𝜙ℎ = 𝜙̃ℎ+1 ⋅ATℎ+1 (𝑗 = 1, . . . , 𝑛 − 1; ℎ = 𝑗, . . . , 𝑛 − 1) .
(87) 𝜃̇= { 𝑇 2𝜋 𝑇 (88)
{ 𝜔 0 , 𝑡 > 𝑇,
×10−4
8
4s 0s
0.0
5s
1s
0
Y (m)
−0.5
8 high-order 6s
uy (m)
8 zero
-order 6.2 s 1.7 s
−8 3s
−1.0 6.3 s
2s 1.9 s
10 high-order
−10 0 10 20
−1.5 10 zero-order X (m)
0 5 10 15 20 25 30 0s 2s 6s
t (s) 1s 3s 6.2 s
1.7 s 4s 6.3 s
Figure 5: 𝜔0 > 3 rad/s. 1.9 s 5s
Wrist yaw B 8
Wrist roll ×10−4
B7
B6
2.0
B1
B4
B2
5
10
5th link 15
5th joint
4th link Figure 9: The torsional angle of 1st joint, 𝜀1 .
4th joint
𝜃2
3rd link
YO 2nd link As shown in Figure 6, Canadarm2 consists of four parts:
3rd joint
2nd joint shoulder, upper arm boom, lower arm boom, and wrist, in
which the shoulder and the wrist both have 3 rotational
k
upper arm and the lower arm has 1 rotational DOF. Hence,
1st joint the total system can be regarded as a 7-DOF manipulator
𝜃1
XO with seven links and seven revolute joints. The links making
O up the shoulder and the wrist are short and thick in shape;
ZO therefore, they can be regarded as rigid links, whereas the
upper and lower arms are assumed to be flexible due to
Figure 7: A spatial flexible robot arms. their slender beam type. Therefore, Canadarm2 system is
12 Advances in Mechanical Engineering
×10−3 0.02
8.0
wz (m)
0.00
ux (m)
0.0
−0.02
2.5 3.0 3.5
2.5 3.0 3.5
t (s)
t (s)
5
5 10
10 15
15
Figure 12: 𝑤𝑧 of 1st link tip.
Figure 10: 𝑢𝑥 of 1st link tip.
0.1
0.02
𝜃x (rad)
0.00
y (m)
0.0
−0.02
Link Mass (kg) Centre (m) Transverse parameters (kg × m−2 ) Length (m)
𝐼𝑥𝑥 = 3.225, 𝐼𝑦𝑦 = 998.630, 𝐼𝑧𝑧 = 997.180, 𝐼𝑦𝑥 = −1.928,
1 139.35 (3.1883, 0.0, 0.0282) (6.377, 0, 0.1524)
𝐼𝑧𝑥 = 9.185, 𝐼𝑦𝑧 = −0.0302
𝐼𝑥𝑥 = 1.59, 𝐼𝑦𝑦 = 632.9, 𝐼𝑧𝑧 = 632.51, 𝐼𝑦𝑥 = 0.8021,
2 87.42 (3.53, 0.0, −0.0066) (7.06, 0, 0.1524)
𝐼𝑧𝑥 = −3.786, 𝐼𝑦𝑧 = −0.08
𝐼𝑥𝑥 = 0.0893, 𝐼𝑦𝑦 = 0.402, 𝐼𝑧𝑧 = 0.402, 𝐼𝑦𝑥 = 0.0, 𝐼𝑧𝑥 = 0.0,
3 8.48 (0.2286, 0.0, 0.0) (0.4572, 0, 0)
𝐼𝑦𝑧 = 0.0
𝐼𝑥𝑥 = 0.408, 𝐼𝑦𝑦 = 4.86, 𝐼𝑧𝑧 = 4.86, 𝐼𝑦𝑥 = 0.0, 𝐼𝑧𝑥 = 0.0,
4 45.94 (0.381, 0.0, 0.0) (0.762, 0, 0)
𝐼𝑦𝑧 = 0.0
𝐼𝑥𝑥 = 1.903, 𝐼𝑦𝑦 = 3.816, 𝐼𝑧𝑧 = 2.823, 𝐼𝑦𝑥 = 0.385, 𝐼𝑧𝑥 = 0.966,
5 45.21 (0.332, 0.0, 0.0) (0.6604, 0, 0)
𝐼𝑦𝑧 = 0.353
0
q23 (rad)
q21 (rad)
−30
−2
−60
−4 0 3 6 9
0 3 6 9
t (s) t (s)
0.6 2.66
0.6 2.66
1.33 5
1.33 5
Figure 14: The 1st joint, 𝑞21 . Figure 16: The 3rd joint, 𝑞23 .
References
q24 (rad)
Research Article
Trivariate Isogeometric Analysis for
Flexible Multibody Dynamics
Ting Pi
Wuhan Second Ship Design & Research Institute, Wuhan 430064, China
Copyright © 2013 Ting Pi. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Isogeometric analysis (IGA) has been a fundamental step forward in the computational mechanics for the past few years, which
maintains the accuracy of the description of computational domain geometry throughout the analysis process. However, the
research on IGA in the area of flexible multibody dynamics is little and mainly concentrates on the univariate or bivariate NURBS
geometry. This paper applies the trivariate IGA to the flexible multibody dynamics and proposes a continuum mechanics-based
method to construct the system dynamic equations within the framework of IGA. A significant feature of this method is that it only
employs the position coordinates of the control points as the system variables. To solve the large rotation and deformation coupled
problems without introducing any rotation angles, the Green-Lagrange strain tensor is adopted. The evaluation of the elastic force
and its Jacobian is easy and accurate by exploiting the appropriate mathematical transformation. In addition, the mass matrix and
the generalized gravity force remain constant, and the centrifugal and Coriolis inertia forces equal zero. A numerical experiment
is conducted using a thin-plate pendulum, which proves the feasibility and effectiveness of this method.
Suppose that the position vector r of an arbitrary point 3.1. Inertia Forces. The virtual work of inertia forces can be
on a flexible body can be expressed in the global coordinate written as
system as
𝑇
e = [xe𝑇 ye𝑇 z𝑇e ] , (7) and the mass matrix is
For isotropic elastic materials, the virtual work of elastic of elastic forces with respect to generalized coordinates are
forces, 𝑈𝑠 , can be written as necessary for the calculation of the Jacobian. Reference [16]
gives an elaborately deduced form of the Jacobian:
𝜆 + 2𝐺 2
𝑈𝑠 = ∫ (𝜀11 + 𝜀222 + 𝜀233 ) 3
𝑉 2 𝜕Q𝑠 𝜆 + 2𝐺
=∑ ∫ ((e𝑇 S𝑇,𝛼 S,𝛼 e) S𝑇,𝛼 S,𝛼
𝜕e 𝛼=1 2 𝑉
+ 𝜆 (𝜀11 𝜀22 + 𝜀11 𝜀33 + 𝜀22 𝜀33 )
+2S𝑇,𝛼 S,𝛼 ee𝑇 S𝑇,𝛼 S,𝛼 − S𝑇,𝛼 S,𝛼 ) 𝑑𝑉
+ 2𝐺 (𝜀212 + 𝜀213 + 𝜀223 ) 𝑑𝑉
3 3
𝜕r 𝜆 + 2𝐺 (17) 𝜆
+∑∑ ∫ ((e𝑇 S𝑇,𝛽 S,𝛽 e) S𝑇,𝛼 S,𝛼
= ∫ 𝑜 [ (𝜀211 + 𝜀222 + 𝜀233 ) 𝛼=1 𝛽=1 2 𝑉
Ω 𝜕𝜉 2 𝛽 ≠
𝛼
+ 𝜆 (𝜀11 𝜀22 + 𝜀11 𝜀33 + 𝜀22 𝜀33 ) +2S𝑇,𝛼 S,𝛼 ee𝑇 S𝑇,𝛽 S,𝛽 − S𝑇,𝛼 S,𝛼 ) 𝑑𝑉
𝜕𝜀22 𝜕𝜀11 𝜕𝜀 With (18) and (20), the elastic forces and the Jacobian can
+ 𝜆 (𝜀11 + 𝜀 + 𝜀11 33
𝜕e 𝜕e 22 𝜕e be effectively evaluated. In engineering problems, mechanical
properties of isotropic elastic materials are usually given in
𝜕𝜀11 𝜕𝜀 𝜕𝜀 terms of Young’s modulus, 𝐸𝑌 , and Poisson’s ratio, 𝜇. The
+ 𝜀 + 𝜀22 33 + 22 𝜀33 )
𝜕e 33 𝜕e 𝜕e relationships between these parameters and Lame’s constants
𝜕𝜀 are the following:
𝜕𝜀12
+ 2𝐺 (2𝜀12 + 2𝜀13 13
𝜕e 𝜕e 𝜇𝐸𝑌
𝜆= ,
𝜕𝜀23 (1 + 𝜇) (1 − 2𝜇)
+2𝜀23 )] 𝑑Ω, (21)
𝜕e 𝐸𝑌
(18) 𝐺= .
2 (1 + 𝜇)
where
𝜕𝜀𝑖𝑗 1 𝑇
= (S S + S𝑇,𝑗 S,𝑖 ) e, 𝑖, 𝑗 = 1, 2, 3, 𝑗 ≥ 𝑖. (19) 3.3. Body Forces. Taking gravity forces as an example, the
𝜕e 2 ,𝑖 ,𝑗 virtual work of body forces can be written as
It can be seen from (14), (16), (18), and (19) that S𝑖,𝜉𝑗 is
the key to evaluate the elastic forces. S𝑖,𝜉𝑗 is the collection of ∫ f𝑏 ⋅ 𝛿r𝑑𝑉 = ∫ [0 0 −𝜌𝑔] S𝑑𝑉𝛿e
𝑉 𝑉
all of the derivatives of NURBS bases with respect to their (22)
𝜕r
parameters, which has been fully studied in computer graph-
= [0 0 −𝜌𝑔] 𝑜 ∫ S𝑑Ω𝛿e.
ics [15]. In the numerical integration methods, the derivatives 𝜕𝜉 Ω
Advances in Mechanical Engineering 5
2
(1) q𝑛+1 = q𝑛 + ℎq̇ 𝑛 + ℎ (0.5 − 𝛽)a;
(2) q̇𝑛+1 = q̇𝑛 + ℎ(1 − 𝛾)a;
(3) 𝜆𝑛+1 = 0;
(4) a = 1/(1 − 𝛼𝑚 )(𝛼𝑓 q̈𝑛 − 𝛼𝑚 a);
(5) q𝑛+1 = q𝑛+1 + ℎ2 𝛽a;
(6) q̇𝑛+1 = q̇ 𝑛+1 + ℎ𝛾a;
(7) q̈𝑛+1 = 0;
(8) for 𝑖 = 1 : 𝑖max do
(9) compute system residual error Υ;
(10) if ‖Υ‖ ≤ 𝜖 then
(11) break;
(12) end
Δq̈
(13) [ ] = J−1 Gen-𝛼 Υ;
Δ𝜆
q̈ q̈ Δq̈
(14) [ 𝑛+1 ] = [ 𝑛+1 ] − [ ];
𝜆𝑛+1 𝜆𝑛+1 Δ𝜆
(15) q𝑛+1 = q𝑛+1 − (1 − 𝛼𝑓 )/(1 − 𝛼𝑚 )𝛽ℎ2 Δq;̈
(16) q̇
𝑛+1 = q̇𝑛+1 − (1 − 𝛼𝑓 )/(1 − 𝛼𝑚 )𝛾ℎΔq;̈
(17) end
(18) a = a + (1 − 𝛼𝑓 )/(1 − 𝛼𝑚 )q̈𝑛+1 ;
4. Computational Strategies
0.75 s
Multibody systems usually contain various forms of con-
1.25 s
straints. Different kinds of system equations for the con-
strained multibody systems have been proposed over the Figure 1: The configuration of the thin plate at different moments
years [17, 18]. This paper adopts the frequently used DAE (36 elements).
system of index-3 with holonomic constraints which is
usually written as
Figure 3: The configuration of the plate and related control points at different moments (1 element).
solve the system equation (24). With this method, system Algorithm 1 sums the process of generalized-𝛼 scheme to
equations of time step 𝑛 + 1 are discretized as solve dynamic equilibrium at time 𝑡𝑛+1 .
The Jacobian, JGen-𝛼 of system equations (25) in
M (e𝑛+1 ) ë𝑛+1 + (Φ𝑇e 𝜆)𝑛+1 − Q𝑛+1 = 0, Algorithm 1 is
1 − 𝛼𝑚 (25)
Φ (e𝑛+1 , 𝑡𝑛+1 ) = 0, JGen-𝛼
(1 − 𝛼𝑓 ) 𝛽ℎ2
2
1 − 𝛼𝑓 1 − 𝛼𝑓
̈ 𝑇 𝑇
where ℎ is the time step, 𝛼𝑚 = (2𝜌∞ − 1)/(𝜌∞ + 1), 𝛼𝑓 = = [M + 𝛽ℎ 1 − 𝛼𝑚 (M (e) e + Φe 𝜆 − Q)e − 𝛾ℎ 1 − 𝛼𝑚 QėΦe ] .
(𝜌∞ )/(𝜌∞ + 1), 𝜌∞ ∈ [0, 1], and 𝛽 = (1 + 𝛼𝑓 − 𝛼𝑚 )2 /4. 𝜌∞ [ Φe 0 ]
(28)
is a parameter controlling numerical dissipation (𝜌∞ = 0 for
maximal dissipation). The position and velocity variables of
time step 𝑛 + 1 are discretized as 5. Numerical Tests
In this section, a numerical experiment with a thin-plate
ℎ2 pendulum is performed to verify the proposed continuum
e𝑛+1 = e𝑛 + ℎė
𝑛 + [(1 − 2𝛽) q𝑛 + 2𝛽q𝑛+1 ] ,
2 (26) mechanics-based trivariate isogeometric analysis for the large
rotation and deformation coupled problems.
ė
𝑛+1 = ė
𝑛 + ℎ [(1 − 𝛾) q𝑛 + 𝛾q𝑛+1 ] , As shown in Figure 1, the plate is fixed in one corner
where 𝛾 = 1/2+𝛼𝑓 −𝛼𝑚 ; q𝑛 is constituted by the acceleration- without any deformation at the initial time. Then, it falls down
like auxiliary variables, which require that freely under the action of gravity. Both the length and width
of the plate are set to 3 m, and its thickness is chosen to be
(1 − 𝛼𝑚 ) q𝑛+1 + 𝛼𝑚 q𝑛 = (1 − 𝛼𝑓 ) ë𝑛+1 + 𝛼𝑓 ë𝑛 , 0.1 m. The density, Young’s modulus, and Poisson ratio of the
(27) material are assumed to be 5000 kg/cm3 , 1.0 × 107 N/m2 , and
q0 = ë0 . 0.3, respectively.
Advances in Mechanical Engineering 7
Figure 4: Comparison of plate deformation of different meshing plans (1 element: yellow; 36 elements: cyan; 100 elements: red).
The employed plate element adopts NURBS bases of order 2,2/3, 5/6, 1, 1, 1, 1}, Ω𝑧 = {0, 0, 1, 1}) and 100-element (Ω𝑥 =
3 in the direction of both length and width. In order not to Ω𝑦 = {0, 0, 0, 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1, 1, 1},
make a distraction, these NURBS bases are assumed to be Ω𝑧 = {0, 0, 1, 1}) meshing plans are performed separately.
𝐶2 -continuous at all of the internal knot values, which are A 2 s simulation with 2−12 s step size is performed for
uniformly distributed between 0 and 1. The polynomial order each of the above three meshing plans. Figure 1 shows
in the thickness direction is chosen to be 1. Therefore, the the configurations of the moving thin plate at different
number of control points in a single element is 4 × 4 × 2 = 32. moments for the 36-element meshing plan. Figure 3 shows
To better illustrate the NURBS element, Figure 2 presents the position relationship between the control points and
the relationship between the element and its related control the thin plate configuration for the single-element meshing
points when the whole plate is regarded as a single element, plan at different moments. Figure 4 compares the plate
that is, Ω𝑥 = Ω𝑦 = {0, 0, 0, 0, 1, 1, 1, 1}, and Ω𝑧 = {0, 0, 1, 1}. deformation of single-element, 36-element and 100-element
As shown in the figure, control points are marked with circles meshing plans at different moments. It is obvious that the
and connected with dashed line to form the control net. plate configuration changes significantly when the number
It should be noted that this is a special case that NURBS of elements increases from 1 to 36, while it differs not that
basis functions are interpolatory at all of the corners of the much from 36 to 100. But the ratio of CPU time used in
element. In most cases, control points do not need to lie on the three cases is 1 : 22 : 66. Therefore, it can be concluded
the geometry. that, using this method, a balance between the accuracy
To see the influence of refinement on the plate’s defor- and efficiency can be achieved with a relatively coarse
mation, other 36-elements (Ω𝑥 = Ω𝑦 = {0, 0, 0, 0,1/6, 1/3, 1/ meshing.
8 Advances in Mechanical Engineering
Research Article
Dynamic Control and Disturbance Estimation of
3D Path Following for the Observation
Class Underwater Remotely Operated Vehicle
Hai Huang,1 Qirong Tang,2 Yueming Li,1 Lei Wan,1 and Yongjie Pang1
1
National Key Laboratory of Science and Technology on Underwater Vehicle, Harbin Engineering University, Harbin 150001, China
2
University of Stuttgart, Pfaffenwaldring 9, 70569 Stuttgart, Germany
Copyright © 2013 Hai Huang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This study addresses the question of 3D path following for the observation class underwater remotely operated vehicle. The dynamic
model of the investigated remote operated vehicle is taken as a coupled multibody system composing of a flexible body and a rigid
body. For precise control, the tether cable disturbance has been investigated as well via a dynamic model. Each element of the
tethered cable even has been taken as an elastic body, and the waves and current disturbances have been taken into consideration.
Based on the multibody system model, an adaptive backstepping sliding mode controller has been designed. To improve the
controller’s systematic robustness against disturbances, the sliding mode surface and adaptive control rule have been designed,
too. Experiments have been performed in a tank, including the 3D path following controls of depth, heading, advance, sideway,
polygon line, and spiral line. With current and wave disturbances having been taken into consideration, the tether effect has been
analyzed, the efficacy and superiority of adaptive backstepping sliding mode control have been verified. It is further confirmed from
the comparisons that the investigated method outperforms those S surface based controllers.
Wind V0 a1
H0 Oc d1 Ti
𝜁 Water plane
Mother ship 𝜂 s
Current Fi,wind
s 𝜙i
z and W h1
wave 𝜙
k Ti+1
x
O Cable R
y l1 i
OR Wi,air
j
ROV
Figure 3: The 𝑖th microunit length of tethered cable in the air and
Figure 2: Model of tethered cable of ROV system in the water. its forces analysis.
Wireless Missionary
joystick
Fault
manipulation
Motion
tolerable control
Joystick
control
Hand
communication
Fault
Autocontrol diagnosis
Surface control
Wireless
interface Surface
computer
Surface Strong tracking
platform Kalman filter
Tether Surface
platform
Socket
Socket Socket
Thruster
Embedded
micro- Motion sensors
processor
DVL
Depth gauge
SY-II ROV Magnetic compass
Underwater Environmental sensors
vehicle
Underwater CCD
Image sonar
Ultrasonic thickness
gauge
(a) Water tank with wave generator (b) Local current generation device
Figure 6: Water tank with wave generator and local current generation device.
only considers for the horizontal direction. If k𝑤𝑐 is assumed Considering the boundary condition at 𝑠 = 0, that is, at
to be the resultant velocity of wave and current velocities, point 𝑂𝑐 , the cable tension force is
then the relative velocity between the cable and this resultant
velocity is
T𝑂𝑐 = √𝐻02 + 𝑉02 . (12)
krel = k𝑡 − k𝑤𝑐 . (10)
At the point 𝑅 on ROV, the cable tension force can be
Thus, if the drag coefficient in the water is represented by 𝐶𝑓 ,
calculated by
then
1 𝑅
F𝑓 = 𝑑𝜉𝐶𝑓 𝜌𝑤 𝐴 krel krel .
sur
(11)
2 𝐻𝑅 + 𝐹wind + 𝐹𝑓 − 𝐻0 = ∫ 𝑚a𝑖,𝐻 + ∫ 𝑚a𝑗,𝐻,
𝑂𝑐 sur
In this study we consider that the wave and current forces are (13)
sur 𝑅
in horizontal direction, for example, in Figures 2 and 4, and
𝑉0 − 𝑊 − 𝑉𝑅 = ∫ 𝑚a𝑖,𝑉 + ∫ 𝑚a𝑗,𝑉
they are in the right horizontal direction. 𝑂𝑐 sur
Advances in Mechanical Engineering 5
7 5
6 4.9
5
4.8
4
4.7
3
4.6
2
4.5
1
0 4.4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Time (s) Time (s)
4
Moment (N·m)
0
0 20 40 60 80 100 120 140
Time (s)
Backstepping control
S surface control
assuming that 𝐻𝑅 is in right horizontal direction and 𝑉𝑅 is where 𝐿 𝑠 is the total length of the cable, that is, from 𝑂𝑐 to 𝑅,
in down vertical direction. Here 𝐻𝑅 = −𝐻𝑅(𝑟) , 𝑉𝑅 = −𝑉𝑅(𝑟) and 𝐿 𝑎 is the length of the cable out of water, that is, from 𝑂𝑐
in which 𝐻𝑅(𝑟) and 𝑉𝑅(𝑟) are the reactive forces in horizontal to the water surface.
and vertical directions, respectively, at point 𝑅 on ROV. In the following, we analyse the cable tension force at the
At the right side of (13), a𝑖,𝐻, a𝑗,𝐻, a𝑖,𝑉, and a𝑗,𝑉 are the water surface. This boundary condition can be considered
corresponding microunits accelerations. The cable tension either from the cable part in the air or the part in the water.
Considering the cable in the air, one has
force at point 𝑅 is |T𝑅 | = √𝐻𝑅2 + 𝑉𝑅2 . sur
air
In (13), other corresponding terms can be governed by 𝐹wind − 𝐻0 + 𝐻sur =∫ 𝑚a𝑖,𝐻,
𝑂𝑐
1 2 (15)
𝐹wind = Fwind = 𝑑1 𝐶wind 𝜌air 𝐴kwind , sur
2 air
𝑉0 − 𝑊𝐿 𝑎 − 𝑉sur =∫ 𝑚a𝑖,𝑉,
1 2 𝑂𝑐
𝐹𝑓 = F𝑓 = ℎ1 𝐶𝑓 𝜌𝑤 𝐴krel , (14)
2 where 𝐻sur
air
and 𝑉sur
air
are the cable tension forces at horizontal
𝑊 = 𝐴𝐿 𝑠 𝜌𝑡 𝑔 − 𝐴 (𝐿 𝑠 − 𝐿 𝑎 ) 𝜌𝑤 𝑔, and vertical directions, respectively. The actual directions of
6 Advances in Mechanical Engineering
9 5.2
8 5.1
7 5
5
4.8
4
4.7
3
4.6
2
4.5
1
0 4.4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Time (s) Time (s)
Backstepping control Backstepping control
S surface control S surface control
Desired depth
6
4
Moment (N·m)
0
0 20 40 60 80 100 120 140
Time (s)
Backstepping control
S surface control
these cable tension forces are determined via the positive or where 𝐻sur
water
and 𝑉sur
water
are the cable tension forces at hor-
negative characteristics of the force values. If it is positive, izontal and vertical directions, respectively, on the water
its actual direction is the same as assumed, otherwise in the surface and considered at the water side. In addition, the
opposite of the assumed. Here we assume that 𝐻sur air
is in right following equations should be met which are
horizontal direction and 𝑉sur is in down vertical direction. In
air
air water
𝑅
𝐻sur = −𝐻sur ,
water
𝐹𝑓 + 𝐻𝑅 − 𝐻sur = ∫ 𝑚a𝑗,𝐻, (17)
air water
sur 𝑉sur = −𝑉sur ,
(16)
𝑅
−𝑉𝑅 − 𝑊𝐿 𝑠 −𝐿 𝑎 ,𝜌𝑡 −𝜌𝑤 + water
𝑉sur = ∫ 𝑚a𝑗,𝑉, air )2 + (𝑉air )2 = √(𝐻water )2 + (𝑉water )2 .
sur Tsur = √(𝐻sur sur sur sur
Advances in Mechanical Engineering 7
−10 6.4
6.2
−15
6
5.8
−20
−25 5.4
5.2
−30
5
4.8
−35
4.6
−40 4.4
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time (s) Time (s)
4
Moment (N·m)
−2
−4
−6
0 20 40 60 80 100 120
Time (s)
Backstepping control
S surface control
when it is in the equilibrium state. Here 𝑎1 is the horizontal where x1 = q ∈ 𝑅6×1 is the state vector and Y is the
length of the cable in the air, 𝑙1 is the horizontal length of the observation variable. Set Y𝑑 to be the desired trajectory of
total cable. ROV and it is twice differentiable. The main steps of the
8 Advances in Mechanical Engineering
80 9
8.5
70
8
60 7.5
7
50
6.5
40
6
30 5.5
5
20
4.5
10 4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Time (s) Time (s)
2
Moment (N·m)
−2
−4
−6
0 20 40 60 80 100 120 140
Time (s)
Backstepping control
S surface control
adaptive backstepping sliding mode control can be described The corresponding Lyapunov function is selected as
as follows.
1
𝑉1 = e𝑇1 e1 . (23)
Step 1. According to the path following objective, e1 is 2
selected as the trajectory tracking error and defined as
Thus,
e1 = Y − Y𝑑 ,
(20) 1 𝑇 1
ė ̇ ̇ ̇ 𝑉1̇= ė e + e𝑇 ė
1 = Y − Y𝑑 = x2 − Y𝑑 . 2 1 1 2 1 1
Then define virtual control coefficient 𝑇 𝑇
= ė
1 e1 = e1 ė
1
𝛼1 = 𝑘1 e1 , (21) (24)
= e𝑇1 (e2 − 𝛼1 )
where 𝑘1 > 0. The velocity error of the ROV is defined as
e2 = ė ̇ ̇
1 + 𝛼1 = x2 − Y𝑑 + 𝛼1 = x2 − Y𝑑 + 𝑘1 e1 . (22) = e𝑇1 e2 − e𝑇1 𝑘1 e1 .
Advances in Mechanical Engineering 9
0.6 0.15
0.4 0.1
0 0
East (m)
−0.2 −0.05
−0.4 −0.1
−0.6 −0.15
−0.8 −0.2
−1 0 1 2 3 4 5 6 7 8 9 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
North (m) Transverse velocity (m/s)
20 3.5
18
3
16
2.5
14
Tension force (N)
Moment (N·m)
12 2
10 1.5
8
1
6
0.5
4
2 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Time (s) Time (s)
Figure 11: Advance toward north with two main thrusters, 𝑉curr = 0.
̇ (e2 + Ẏ
⋅ (Ttru + 𝜏𝑡 − (C (q)̇+ D (q)) 𝑑 − 𝛼1 )
S = 𝑐1 e1 + e2 , (25)
− G (q) − Δd) − Ÿ
𝑑 + 𝛼̇
1] .
where S is the switching function of the terminal sliding (27)
mode, 𝑐1 ≥ 0. Then the Lyapunov function is chosen as
To satisfy the Lyapunov stability theory and the reaching
law of the controller as in [25], the derivative Ṡis defined as
1
𝑉2 = 𝑉1 + S𝑇 S. (26) Ṡ= −ℎ𝛽 sgn (S) − ℎS, (28)
2
10 Advances in Mechanical Engineering
5 0.4
4 0.3
2
0
1
−0.1
0
−0.2
−1 −0.3
−1 0 1 2 3 4 5 6 7 8 9 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
North (m) Transverse velocity (m/s)
45 7
40 6
35
5
Tension force (N)
30
Moment (N·m)
4
25
3
20
2
15
10 1
5 0
0 50 100 150 200 250 0 50 100 150 200 250
Time (s) Time (s)
Figure 12: Advance toward north with two main thrusters, 𝑉curr = 0.4 m/s normal eastward.
where ℎ and 𝛽 are positive constants. We design the back- Step 3. In order to estimate further disturbances Δd and
stepping sliding mode controller as avoid the undetermined upper bounds, we define Lyapunov
function:
u = Ttru
1 ̃𝑇 ̃
𝑉3 = 𝑉2 + Δd Δd, (30)
= M (q) [−𝑐1 (−𝑘1 e1 + e2 ) + Ÿ 2𝛾
𝑑 − 𝛼̇
1 − ℎ𝛽 sgn (S) − ℎS]
8 0.25
7
0.2
5 0.15
East (m)
4
0.1
3
2
0.05
1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 −0.04 −0.02 0 0.02 0.04 0.06 0.08
North (m) Transverse velocity (m/s)
19 3
18
2.5
17
16 2
Tension force (N)
Moment (N·m)
15
1.5
14
13 1
12
0.5
11
10 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
Time (s) Time (s)
Figure 13: Lateral motion toward east with two side thrusters, 𝑉curr = 0.
̇ (e2 + Ẏ
⋅ (Ttru + 𝜏𝑡 − (C (q)̇+ D (q)) 𝑑 − 𝛼1 ) ̇ (e2 + Ẏ
− 𝜏𝑡 + (C (q)̇+ D (q)) 𝑑 − 𝛼1 )
9 0.3
8
0.25
7
5 0.15
East (m)
4 0.1
3
0.05
2
0
1
0 −0.05
−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
North (m) Transverse velocity (m/s)
50 6
5
40
Tension force (N)
Moment (N·m)
4
30
3
20
2
10 1
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (s) Time (s)
Figure 14: Lateral motion toward east with two side thrusters, 𝑉curr = 0.4 m/s normal northward.
By selecting proper values of ℎ, 𝑐𝑖 , and 𝑘𝑖 to satisfy ‖Q‖ ≥ 0; cable are a neutrally buoyant tether which is attached to
that is, the transformation matrix Q is regarded as the positive the vehicle at the tail. The control commands are sent
definite matrix; then through network communication between surface computer
and PC104 embedded processor. It is equipped with depth
𝑉3̇= −e𝑇 Qe − ℎ𝛽S𝑇 sgn (S) ≤ 0, (37) gauge, ultrasonic Doppler velocity meter (DVL), magnetic
compass as motion sensors, underwater CCD, image sonar,
where e = [e1 , e2 ]𝑇 . This means the third subsystem is also ultrasonic thickness gauge as environmental sensors, and 6
stable which is sufficient for our ROV system. thrusters including 2 main thrusters, 2 side thrusters, and
2 vertical thrusters. Its tethered cable and hydrodynamic
4. Experiments and inertial parameters are illustrated in Tables 1, 2, and
3, respectively. The experiments have been performed in a
4.1. Experimental Setup. In order to verify and analysis the 50 m × 30 m × 10 m tank at the Key Laboratory of Science
investigated tethered cable dynamic model and the backstep- and Technology on Underwater Vehicle in Harbin Engi-
ping controller, experiments have been performed based on neering University. Comparisons have been made between
an SY-II remote operated vehicle; see Figure 5. The SY-II ROV the S surface controller investigated in [6] and our adaptive
is an open frame underwater vehicle, which is purposed for backstepping sliding mode controller. Current and waves are
ship hull inspection. Its power supply and communication generated from local current generation device (Figure 6(a))
Advances in Mechanical Engineering 13
26
24
22
20
16
14
8 12
10
6
8
Depth (m)
Backstepping control
2
S surface control
0 14
12
20 20
Ea 10
st 10
(m 10
) )
Moment (N·m)
th (m
0 0 Nor 8
6
Backstepping control
S surface control 4
Desired curve
2
0
0 500 1000 1500 2000 2500 3000
Time (s)
Backstepping control
S surface control
Figure 15: 3D path-following experiments, polygon path, 𝑉curr = 0.2 m/s, 𝑉wave = 0.2 m/s normal northward.
and multidirectional waves generator (Figure 6(b)), respec- Table 1: Tethered cable parameters.
tively, in the experiments to simulate real oceanic conditions.
𝜌𝑡 𝐶𝑓 𝐸 Cwind 𝐷 𝐿 total
Item
4.2. Experiments and Results Analysis. In the depth control (kg/m3 ) (m−1 ) (10 MPa)
5
(m−1 ) (m) (m)
experiments of Figures 7 and 8, the tension force from Value 1000 1.2 0.02 0.4 10−2 300
tethered cable upon SY-II ROV changes little mainly because
its neutral buoyancy, and the peak values of tethered cable Table 2: SY-II hydrodynamic parameters.
moment are decided by diving speed. The tethered cable
makes little effect upon depth control when diving is not Dimensionless x y z r q p
very fast. Adaptive backstepping sliding mode controller has coefficient
improved the stability and speed constringency in depth First-order 27.65 47.26 44.86 62.63 57.55 141.33
control in comparing with S surface controller since it terms
provides dynamic compensation during the experiments.
In the heading control experiments of Figures 9 and 10,
the desired heading is 40∘ and −30∘ relative to the normal from tethered cable upon SY-II ROV change a lot at first
north of earth’s magnetic field. The tension force and moment during revolution process. The tethered cable and ROV’s
14 Advances in Mechanical Engineering
35
30
25
15
8 10
5
6 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Depth (m)
Time (s)
4
Backstepping control
S surface control
2
15
0
10
20
10 20
0 10
Moment (N·m)
Ea
st ( 0 5
m) −10 )
−10 th ( m
−20 −20 Nor
0
Backstepping control
S surface control −5
Desired path
−10
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (s)
Backstepping control
S surface control
Figure 16: 3D path-following experiments, spiral path, 𝑉curr = 0.2 m/s, 𝑉wave = 0.2 m/s, normal northward.
Table 3: SY-II inertial parameters. 𝑉curr = 0.4 m/s normal east (𝑉curr is the currents speed),
are compared. Here the current has made great influence
𝐼𝑥 𝐼𝑦 𝐼𝑧 𝐼𝑥𝑦 𝐼𝑥𝑧 𝐼𝑥𝑧
Mass (kg) upon tethered cable and generates tension force and moment.
(N⋅m2 ) (N⋅m2 ) (N⋅m2 ) (N⋅m2 ) (N⋅m2 ) (N⋅m2 )
Nonlinear disturbance from current and tether has produced
111.9 97.3 26.1 56.8 0 0 0 much greater errors for the S surface controller than the
adaptive backstepping sliding mode controller.
rectangle shape have made nonlinear effects to the ROV During the eastward lateral motion process (see Figures
manipulation. As a result, the S surface controller cannot 13 and 14), the SY-II ROV that is laterally moving mainly
realize precise heading control, while adaptive backstepping depends on two lateral thrusters. Two conditions, that is,
sliding mode controller qualifies, since the controller delivers 𝑉curr = 0 and 𝑉curr = 0.4 m/s normal north, are compared.
command according to its dynamic manipulation and char- At first SY-II ROV has been blown a few meters away,
acter. and then the control started. The current has generated
During the northward advancing process (see Figures 11 great tension force and moment effect upon SY-II ROV.
and 12), the SY-II ROV that is advancing mainly depends on Adaptive backstepping sliding mode controller has realized
two main thrusters. Two conditions, that is, 𝑉curr = 0 and more accuracy result compared to the S surface controller.
Advances in Mechanical Engineering 15
During the 3D path-following experiments in the tank 51279221, and no. 61100006, the Doctoral Fund of Ministry of
(see Figures 15 and 16), the disturbances are current and Education for Young Scholar with no. 20122304120003, State
waves. The current speed is 𝑉curr = 0.4 m/s, the wave Key Laboratory of Ocean Engineering of Shanghai Jiao Tong
is a sine wave with speed 𝑉wave = 0.2 m/s, and height University no. 1102, State Key Laboratory of Robotics and
𝐻wave = 0.2 m, respectively. Polygon path and spiral path are Systems of Harbin Institute of Technology no. SKLRS-2012-
followed, respectively. As the experiments we made above, ZD-03, and the Fundamental Research Funds for the Central
environmental disturbances exerted on the tethered cable Universities with no. HEUCFR1101.
have contributed nonlinear effect to the ROV. The S surface
controller manages to follow the polygon path but have to References
adjust continuously and heavily to follow the spiral path.
Comparatively, adaptive backstepping sliding mode con- [1] K. Zhu and L. Gu, “A MIMO nonlinear robust controller for
troller can compensate more disturbances and can follow the work-class ROVs positioning and trajectory tracking control,”
desired path more precisely. Deduced from dynamic model, in Proceedings of the Chinese Control and Decision Conference
adaptive backstepping sliding mode controller is established (CCDC ’11), pp. 2565–2570, Mianyang, China, May 2011.
to calculate and compensate nonlinear disturbances caused [2] A. R. Marzbanrad, M. Eghtesad, and R. Kamali, “A robust
by current and tether. Therefore, it outperforms S surface adaptive fuzzy sliding mode controller for trajectory tracking of
controller in control accuracy and robustness. ROVs,” in Proceedings of the 50th IEEE Conference on Decision
For the experiments to be performed successfully, we and Control and European Control Conference, pp. 2863–2870,
have also run some preliminary simulations to get the initial Orlando, Fla, USA, December 2011.
value of some uncertainties. The simulation and experimental [3] M. Kokegei, F. He, and K. Sammut, “Fully coupled 6 degree-of-
results agree with each other well. Details of simulation freedom control of an over-actuated autonomous underwater
vehicle,” in Autonomous Underwater Vehicles, N. A. Cruz, Ed.,
results can be found in [26, 27].
pp. 147–170, InTech, Rijeka, Croatia, 2011.
[4] X. Bian, Y. Qu, Z. Yan, and W. Zhang, “Nonlinear feedback
5. Conclusions control for trajectory tracking of an unmanned underwater
vehicle,” in Proceedings of the IEEE International Conference on
The main contributions of this paper are summarized as Information and Automation (ICIA ’10), pp. 1387–1392, Harbin,
follows. At first, the dynamic tether model has been estab- China, June 2010.
lished based on the lumped mass cable model. Each element [5] A. Bagheri, T. Karimi, and N. Amanifard, “Tracking perfor-
of the cable is taken as an elastic body and takes into mance control of a cable communicated underwater vehicle
consideration waves and current disturbances. Based on the using adaptive neural network controllers,” Applied Soft Com-
dynamic model, an adaptive backstepping sliding mode con- puting Journal, vol. 10, no. 3, pp. 908–918, 2010.
troller has been designed. Secondly, the controller has taken [6] J. Liu, H. Yu, and Y. Xu, “Improved S plane control algorithm for
dynamic model into consideration at the first and second underwater vehicles,” Journal of Harbin Engineering University,
steps (backstepping steps) based on Lyapunov functions. To vol. 23, no. 1, pp. 33–36, 2002 (Chinese).
improve the controller’s systematic robustness and against [7] J. Wang, G. Wu, L. Wan, Y. Sun, and L. Wang, “Controller design
disturbances, the sliding mode surface has been defined. of underwater robots based on generalized S-plane,” Electric
Machines and Control, vol. 13, supplement 1, pp. 144–148, 2009
An adaptive control rule has been chosen to further resist
(Chinese).
environmental disturbance which is the most difficult part for
[8] P. W. J. van de Ven, C. Flanagan, and D. Toal, “Neural network
ROV 3D path following.
control of underwater vehicles,” Engineering Applications of
Experiments have been performed in the tank, including Artificial Intelligence, vol. 18, no. 5, pp. 533–547, 2005.
depth control, heading control, advance control, sideway
[9] W. Luo and Z. Zou, “Neural network based robust controller
control, polygon line, and spiral line 3D path-following con- for trajectory tracking of underwater vehicles,” China Ocean
trol. With current and waves disturbances being taken into Engineering, vol. 21, no. 2, pp. 281–292, 2007.
consideration, tethered cable effect has been analyzed, and [10] L. Zhang, X. Qi, and Y. Pang, “Adaptive output feedback control
the efficacy and superiority of our designed controller have based on DRFNN for AUV,” Ocean Engineering, vol. 36, no. 9-10,
been verified. The comparisons to the S surface controller pp. 716–722, 2009.
have shown better performances of the investigated adaptive [11] P. Patompak and I. Nilkhamhang, “Adaptive backstepping
backstepping sliding mode controller. sliding-mode controller with bound estimation for underwater
robotics vehicles,” in Proceedings of the 9th International Confer-
ence on Electrical Engineering/Electronics, Computer, Telecom-
Conflict of Interests munications and Information Technology, pp. 1–4, Phetchaburi,
Thailand, May 2012.
The authors declare that there is no conflict of interests
regarding the publication of this paper. [12] L. Lapierre and B. Jouvencel, “Robust nonlinear path-following
control of an AUV,” IEEE Journal of Oceanic Engineering, vol. 33,
no. 2, pp. 89–102, 2012.
Acknowledgments [13] G. Antonelli, F. Caccavale, S. Chiaverini, and G. Fusco, “A
novel adaptive control law for underwater vehicles,” IEEE
This project is supported by National Science Foundation of Transactions on Control Systems Technology, vol. 11, no. 2, pp.
China under the Grants of no. 51209050, no. 51179035, no. 221–232, 2003.
16 Advances in Mechanical Engineering
Research Article
Application of Finite Volume Method to
Structural Stochastic Dynamics
Weidong Chen, Yanchun Yu, Ping Jia, Xiande Wu, and Fengchao Zhang
Harbin Engineering University, 145 Nantong Street, Harbin 150001, China
Copyright © 2013 Weidong Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The stochastic dynamic problems were becoming more difficult after considering the influences of stochastic factors and the
complexity of the dynamic problems. To this background, the finite volume method combined with Perturbation Method was
proposed for the stochastic dynamic analysis. The equations of perturbation-finite volume method were derived; the explicit
expressions between random response and basic random variables were given; the method of stochastic dynamic analysis was
discussed; and the examples were presented to verify the perturbation-finite volume method. The results of perturbation-finite
volume method were compared with the results of Monte Carlo Method, which proved that the proposed method was correct and
accurate. Because the proposed method was simple and clear, the equations were easy to establish and the efficiency was improved.
Meanwhile, the proposed method was successfully applied to the stochastic dynamic analysis of linear multibody system, which
was verified through the example in this paper.
Boundary conditions are as follows: of displacement field {𝑢} on 𝑌𝑖 at point {𝑌}. {𝑢𝑖𝑗(2) } was
the second-order partial derivatives of displacement
𝜎 ⋅ n = X, field {𝑢} on 𝑌𝑖 and 𝑌𝑗 at point {𝑌}.
(8) For the governing equation (6), due to the randomness of
u = u,
load or structure, the displacement and stress were random.
Therefore, the displacement and stress were expressed as
where X was the outside force of boundary and u was the second-order perturbation expressions. The second-order
displacement of boundary. perturbation expressions of displacement were shown in (15).
The first moment and second moment of {𝑢} were as follows:
Initial conditions are as follows:
𝑛
u 𝑡=0 = u0 , 𝐸 [{𝑢}] = {𝑢} + ∑𝐸 [𝛼𝑖 ] {𝑢𝑖(1) }
(9) 𝑖=1
u̇𝑡=0 = u̇
0.
1 𝑛 𝑛
+ ∑ ∑ 𝐸 [𝛼𝑖 𝛼𝑗 ] {𝑢𝑖𝑗(2) }
Explicit central difference scheme was used in time 2 𝑖=1 𝑗=1
domain [22], in the form of
1 𝑛 𝑛
𝑢̇𝑛+1/2 𝑛−1/2
= 𝑢̇ 𝑛 𝑛
+ Δ𝑡 𝑢̈, = {𝑢} + ∑ ∑𝐸 [𝛼𝑖 𝛼𝑗 ] {𝑢𝑖𝑗(2) } ,
2 𝑖=1 𝑗=1
(10)
𝑢𝑛+1 = 𝑢𝑛 + Δ𝑡𝑛+1/2 𝑢̇
𝑛+1/2
, 𝑛 𝑛
𝑇
𝐷 [{𝑢}] = ∑ ∑𝐸 [𝛼𝑖 𝛼𝑗 ] {𝑢𝑖(1) } {𝑢𝑗(1) }
where Δ𝑡𝑛+1/2 = 𝑡𝑛+1 −𝑡𝑛 and Δ𝑡𝑛 = 𝑡𝑛+1/2 −𝑡𝑛−1/2 = (Δ𝑡𝑛−1/2 + 𝑖=1 𝑗=1
(16)
Δ𝑡𝑛+1/2 )/2. In order to ensure the stability of the numerical 𝑛 𝑛 𝑛
results, the value of the time step should satisfy Δ𝑡1 = 𝑙/𝑐; (2) 𝑇
+∑∑∑ {𝑢𝑖(1) } {𝑢𝑗𝑘 }
here, 𝑐 was the speed of sound. 𝑖=1 𝑗=1 𝑘=1
When the stochastic field was not dependent on time, it −𝐸 [𝛼𝑖 𝛼𝑗 ] 𝐸 [𝛼𝑘 𝛼𝑙 ]) ,
could be expressed as
where
{𝑌} = {𝑌} + {𝛼} , (12)
𝐸 [𝛼𝑖 𝛼𝑗 ] = Cov (𝑌𝑖 , 𝑌𝑗 ) . (17)
where {𝑌} was the mean value of the stochastic field, Generally higher moments of the stochastic field were
𝑇 difficult to obtain and the error was large, so the first item
{𝑌} = [𝑌1 𝑌2 ⋅ ⋅ ⋅ 𝑌𝑛 ] . (13) was used as follows:
𝑛 𝑛
𝑇
{𝛼} was the small perturbation on the mean value point {𝑌}, 𝐷 [{𝑢}] = ∑ ∑𝐸 [𝛼𝑖 𝛼𝑗 ] {𝑢𝑖(1) } {𝑢𝑗(1) } . (18)
where 𝑖=1 𝑗=1
𝑇
{𝛼} = [𝛼1 𝛼2 ⋅ ⋅ ⋅ 𝛼𝑛 ] = {0} . (14) The second-order perturbation expression of stress was
𝑛
1 𝑛 𝑛
If the stochastic field was dependent on time, then, for any {𝜎} = {𝜎} + ∑𝛼𝑖 {𝜎𝑖(1) } + ∑ ∑𝛼𝑖 𝛼𝑗 {𝜎𝑖𝑗(2) } , (19)
instant, (12) was correct. 𝑖=1 2 𝑖=1𝑗=1
Since the displacement field {𝑢} depended on stochastic
field, {𝑢} could be expanded by second-order perturbation as where {𝜎} was the mean value of stress field {𝜎} at point
follows: {𝑌}. {𝜎𝑖(1) } was the first-order partial derivatives of stress
𝑛 field {𝜎} on 𝑌𝑖 at point {𝑌}. {𝜎𝑖𝑗(2) } was the second-order par-
1 𝑛 𝑛
{𝑢} = {𝑢} + ∑𝛼𝑖 {𝑢𝑖(1) } + ∑ ∑ 𝛼𝑖 𝛼𝑗 {𝑢𝑖𝑗(2) } (15) tial derivatives of stress field {𝜎} on 𝑌𝑖 and 𝑌𝑗 at point {𝑌}.
𝑖=1 2 𝑖=1 𝑗=1
The relationship between stress matrix and strain matrix
was
where {𝑢} was the mean value of displacement field {𝑢} at
point {𝑌}. {𝑢𝑖(1) } was the first-order partial derivatives {𝜎} = [𝐷] {𝜀} . (20)
4 Advances in Mechanical Engineering
60 mm
(21) z x
{𝜎𝑖𝑗(2) } = [𝐷] {𝜀𝑖𝑗(2) } + [𝐷𝑖(1) ] {𝜀𝑗(1) } 150 mm
30 mm
+ [𝐷𝑗(1) ] {𝜀𝑖(1) } + [𝐷𝑖𝑗(2) ] {𝜀} .
Figure 2: The calculation model of cantilever beam.
According to (21), the mean value and variance of stress
were
1 𝑛 𝑛 ×10−4
𝐸 [{𝜎}] = {𝜎} + ∑ ∑ 𝐸 [𝛼𝑖 𝛼𝑗 ] {𝜎𝑖𝑗(2) }
2 𝑖=1 𝑗=1 1.0
𝑛 𝑛
𝑇
−1.5
𝐷 [{𝜎}] = ∑∑ 𝐸 [𝛼𝑖 𝛼𝑗 ] {𝜎𝑖(1) } {𝜎𝑗(1) }
𝑖=1 𝑗=1 −2.0
𝑛 𝑛
Figure 3: Mean value of displacement on node (0.03, 0.06, 0.1) m on
𝑇 direction 𝑦.
= ∑∑ 𝐸 [𝛼𝑖 𝛼𝑗 ] ([𝑆𝑖(1) ] [𝑅𝑎𝑎 ] [𝑆]
𝑖=1 𝑗=1
𝑇
+ [𝑆] [𝑅𝑎𝑖 𝑎 ] [𝑆𝑖(1) ]
of the finite volume method for solving stochastic dynamic
𝑇 problems. The partial derivative of (6) to random variable was
+ [𝑆] [𝑅𝑎𝑖 𝑎𝑗 ] [𝑆] ) ,
(22) (1) 1 𝑚 (1)
ü
𝑖 = ∑ A (𝜎 ) ,
𝑀 𝑛=1 𝑛 𝑛 𝑖
where D was elastic matrix and B was strain matrix. We have (24)
(2) 1 𝑚 (2)
ü = ∑ A (𝜎 ) .
[𝑆𝑖(1) ] = [𝐷𝑖(1) ] [𝐵] , [𝑆] = [𝐷] [𝐵] , 𝑖𝑗
𝑀 𝑛=1 𝑛 𝑛 𝑖𝑗
[𝑅𝑎𝑎 ] = {𝑢} {𝑢}𝑇 , If the physical quantities at time 𝑡 were known, the partial
𝑇 𝑇
(23) derivatives of displacement at time 𝑡 + Δ𝑡 could be obtained
[𝑅𝑎𝑖 𝑎 ] = [𝑅𝑎𝑎𝑖 ] = {𝑢} {𝑢𝑖(1) } , by explicit central difference as follows:
(1) (1) (1)
[𝑅𝑎𝑖 𝑎𝑗 ] = {𝑢𝑖(1) } {𝑢𝑗(1) } . 𝑡+Δ𝑡/2
(u̇ 𝑡−Δ𝑡/2
)𝑖 = (u̇ 𝑡
)𝑖 + (ü )𝑖 Δ𝑡, (25)
×10−9 ×108
3.5
5
2.5 4
2.0 3
1.5
2
1.0
0.5 1
0.0 0
−5.0
0.0 3.0 6.0 9.0 12.0 15.0
0.0 3.0 6.0 9.0 12.0 15.0
×10−4 ×10−4
t (s) t (s)
P-FVM P-FVM
Monte Carlo Monte Carlo
2.0
Variance of equivalent stress (Pa 2 )
(1) (1)
(𝜎𝑡+Δ𝑡 )𝑖 = 𝐷𝑖(1) Bu𝑡+Δ𝑡 + DB(u𝑡+Δ𝑡 )𝑖 , 1.0
(2) (1)
(𝜎𝑡+Δ𝑡 )𝑖𝑗 = 𝐷𝑖𝑗(2) Bu𝑡+Δ𝑡 + 𝐷𝑖(1) B(u𝑡+Δ𝑡 )𝑗 (29) 5.0
𝜕2 𝜎 𝜕2 X (31)
⋅n= , 𝜕u̇
𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 = u̇ 0,
𝜕𝑥𝑖 𝑡=0
𝜕2 u 𝜕2 u 𝜕2 u̇
= . = u̇ 0.
𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝑡=0
6 Advances in Mechanical Engineering
y ×10−11
y
2.5
z x
1.5
150 mm
30 mm 1.0
Figure 7: The calculation model of clamped beam.
0.5
×10−5 0.0
0.5
0.0 1.0 2.0 3.0
×10−4
Mean value of displacement (m)
0.0 t (s)
Monte Carlo
P-FVM
−0.5
Figure 9: Variance of displacement on node (0.03, 0.06, 0.1) m on
direction 𝑦.
−1.0
−1.5
×107
14.0
−2.0
Mean value of equivalent stress (Pa)
2.0
For the boundary conditions of load, the partial deriva-
tives of load to random variables were added to (24). For 0.0
the boundary conditions of displacement, if a degree of
0.0 1.0 2.0 3.0
freedom of a node was constrained, the partial derivative of ×10−4
t (s)
the acceleration was forced to zero at each time step.
Monte Carlo
P-FVM
4. Examples Analysis
Figure 10: Mean value of equivalent stress on node (0.03, 0.06,
The stochastic dynamic problems for cantilever beam and 0.0) m.
clamped beam subjected stochastic dynamic load were solved
by perturbation-finite volume method. The results were
compared with that produced by monte carlo method, which
proved the proposed method was correct. To ensure the Therefore, 10000 samples were used in Monte Carlo
accuracy of Monte Carlo, according to the law of large num- Method to satisfy the accuracy.
bers, a large number of samples were required to ensure the
credibility of the results. Through the results of perturbation- (A) Cantilever Beam. The calculation model of cantilever
finite volume method, the structural failure probability was beam was shown in Figure 2; the parameters of material were
estimated, and the value was 44%. The literature [23] recom- shown in Table 1.
mended that the number of samples should satisfy certain The expression of load was
conditions:
100 𝑡
𝑁≥ . (32) 𝑃 (𝑡) = 𝑃𝑚 exp (− ) , (33)
𝑃𝑓 𝜃
Advances in Mechanical Engineering 7
×1014 ×10−10
14.0 8.0
Variance of equivalent stress (Pa 2 )
12.0 7.0
4
Mean value of displacement (m)
1.0
3
0.0
2
−1.0
1
−2.0 0
−3.0 −1
0.0 2.0 4.0 6.0 8.0 10.0
−4.0 t (s) ×10−4
0.0 2.0 4.0 6.0 8.0 10.0
t (s) ×10−4 Figure 14: Mean value of equivalent stress on node (0.03, 0.06,
0.0) m.
Figure 12: Mean value of displacement on node (0.03, 0.06, 0.1) m
on direction 𝑦.
Table 1: The parameters of material.
1.13
𝑤1/3 The comparison charts of the two methods were shown
𝑃𝑚 = 52.4( ) (MPa) , in Figures 3, 4, 5, and 6.
𝑅 (34)
(3.4 MPa < 𝑃𝑚 < 138 MPa) , (B) Clamped Beam. The basic parameters of loads and
materials were the same as example A (Figure 7), and then the
−0.23
1/3 𝑤1/3 comparison charts of the two methods were shown in Figures
𝜃 = 0.048𝑤 ( ) (ms) . (35) 8, 9, 10, and 11.
𝑅
From the results of the two examples, it showed that
the results of two methods were basically the same, and
Here, 𝑤 and 𝑅 were the independent random variables. the calculation accuracy was verified. It has be seen from
The statistical characteristics of random variables were shown the comparison of the two methods that the mean value of
in Table 2. the displacement and stress fitted better, while the variance
8 Advances in Mechanical Engineering
Table 2: Statistical characteristics of random variables. Table 3: Statistical characteristics of random variables.
Random Coefficient of Random Coefficient of
Distribution Mean value Distribution Mean value
variables variation variables variation
𝑤 (kg) Normal distribution 200.0 0.20
𝑤 (kg) Normal distribution 200.0 0.20
𝑅 (m) Normal distribution 50.0 0.20
𝐸 (GPa) Normal distribution 200.0 0.002
×1014
16.0 Table 4: The parameters of material.
14.0 Modulus of
Variance of equivalent stress (Pa 2 )
4.0
2.0
0.0
The calculation model of linear multibody system was shown
−2.0 in Figure 16. The sections of copper and steel were the same.
0.0 2.0 4.0 6.0 8.0 10.0 The parameters of materials were shown in Table 4.
t (s) ×10−4 The load and the statistical characteristics of random
Figure 15: Variance of equivalent stress on node (0.03, 0.06, 0.0) m. variables were the same as example A. The mean value
and variance of the responses solved by the perturbation-
finite volume method were shown in Figures 17, 18, 19, and
20.
y y y
Aluminum
Aluminum
20 mm
60 mm
Copper
Copper Steel
x x
50 mm z 20 mm
150 mm 150 mm
30 mm
×10−2 ×107
2.0 10.0
8.0
1.0
6.0
0.5
4.0
0.0
−0.5 2.0
−1.0
0.0
−1.5
0.0 0.5 1.0 1.5
0.5 1.0 1.0
t (s) ×10−2
0.0
t (s) ×10−2
Figure 19: Mean value of equivalent stress on node (0.0, 0.058,
Figure 17: Mean value of displacement on node (0.0, 0.06, 0.35) m 0.213) m.
on direction 𝑦.
×10−5
2.5 ×1014
7
Variance of displacement (m2 )
2.0
Variance of equivalent stress (Pa 2 )
5
1.5
4
1.0 3
2
0.5
1
0
0.0
−1
0.0 0.5 1.0 1.5
0.0 0.5 1.0 1.5 t (s) ×10−2
t (s) ×10−2
Figure 18: Variance of displacement on node (0.0, 0.06, 0.35) m on Figure 20: Variance of equivalent stress on node (0.0, 0.058,
direction 𝑦. 0.213) m.
10 Advances in Mechanical Engineering
Acknowledgment [17] M. Xiao, C. Gao, and L. Ye, “Dynamic reliability analysis for
structure seismic response based on wavelet,” World Earthquake
This work was supported by the Science and Technol- Engineering, vol. 18, no. 1, pp. 27–33, 2002.
ogy on Combustion and Explosion Laboratory Foundation [18] H. Cao, M. Lai, and S. Bai, “Estimations of the reliability of
(9140C350406150C35126). structures based on wavelet transform,” World Information on
Earthquake Engineering, vol. 16, no. 4, pp. 70–77, 2000.
References [19] H. Qiao and Z. Lv, “Dynamic reliability analysis of stochastic
structures under non-stationary random excitation,” Chinese
[1] Q. Chen and X. Liu, Stochastic Finite Element Method and Journal of Solid Mechanics, vol. 29, no. 1, pp. 36–40, 2008.
Engineering Application, Southwest Jiaotong University Press, [20] N. Fallah, “A cell vertex and cell centred finite volume method
Chengdu, China, 1993. for plate bending analysis,” Computer Methods in Applied
[2] W. An, W. Zhu, and X. Yan, “Application of stochastic finite Mechanics and Engineering, vol. 193, no. 33–35, pp. 3457–3470,
element method to analysis of uncertainties,” Journal of Harbin 2004.
Engineering University, vol. 23, no. 1, pp. 132–135, 2002. [21] G. Xia and C.-L. Lin, “An unstructured finite volume approach
[3] H. Cui and W. Wen, “Stochastic FEM and its applications in for structural dynamics in response to fluid motions,” Comput-
engineering,” Journal of Nanjing University of Aeronautics & ers and Structures, vol. 86, no. 7-8, pp. 684–701, 2008.
Astronautics, vol. 32, no. 1, pp. 91–98, 2000. [22] C. Yan, The Method of Computational Fluid Dynamics and Its
[4] X.-C. Yan and Y. Hua, “Structural system reliability analysis Application, Beijing University of Aeronautics and Astronautics
based on stochastic finite element method,” Journal of Wuhan Press, Beijing, China, 2003.
University of Technology, vol. 32, no. 9, pp. 69–75, 2010. [23] W. An, Y. Cai, and W. Chen, Reliability Analysis and Optimiza-
[5] Z. Rong and S. Lin, “Reliability analysis for connecting-rod tion Design of Stochastic Structural System, Harbin Engineering
based on SFEM,” Ship Sciences and Technology, vol. 33, no. 9, University Press, Harbin, China, 2005.
pp. 68–70, 2011.
[6] L. Yang and Z. Li, “Vectorial cooperative response surface
method for stochastic analysis of structures,” Engineering
Mechanics, vol. 29, no. 11, pp. 58–64, 2012.
[7] P. Jing and S. Huang, “Analysis on structure reliability colloca-
tion based stochastic finite element method,” Sichuan Building
Science, vol. 38, no. 2, pp. 39–41, 2012.
[8] X. Xu, Z. Lv, and P. Wei, “Probability and compound uncer-
tainty importance measures of structure stochastic analysis and
their computational method,” Chinese Journal of Computational
Mechanics, no. 2, pp. 64–71, 2013.
[9] T. Xu, T. Xiang, C. Yang, and R. Zhao, “Uncertainty analysis of
long-term deformation for prestress concrete beam,” Journal of
Civil Architectural & Environmental Engineering, vol. 35, no. 1,
pp. 111–116, 2013.
[10] A. Saltelli, “Sensitivity analysis for importance assessment,” Risk
Analysis, vol. 22, no. 3, pp. 579–590, 2002.
[11] H. C. Frey and S. R. Patil, “Identification and review of
sensitivity analysis methods,” Risk Analysis, vol. 22, no. 3, pp.
553–578, 2002.
[12] A. Saltelli, S. Tarantola, and F. Campolongo, “Sensitivity analysis
as an ingredient of modeling,” Statistical Science, vol. 15, no. 4,
pp. 377–395, 2000.
[13] W.-P. Mu, “An extended precise integration method for
response of a structure subjected to evolutionary random
exciation,” Journal of Vibration and Shock, vol. 28, no. 7, pp. 131–
142, 2009.
[14] X. Lv, Y. Zhao, X. Y. Huang, G. H. Xia, and X. H. Su, “A matrix-
free implicit unstructured multigrid finite volume method for
simulating structural dynamics and fluid-structure interaction,”
Journal of Computational Physics, vol. 225, no. 1, pp. 120–144,
2007.
[15] J.-B. Chen and J. Li, “Dynamic response and reliability analysis
of non-linear stochastic structures,” Probabilistic Engineering
Mechanics, vol. 20, no. 1, pp. 33–44, 2005.
[16] C. S. Putcha, R. Patev, and M. A. Leggett, “Time-dependent
reliability analysis of steel miter gates,” International Journal of
Modelling and Simulation, vol. 23, no. 1, pp. 1–12, 2003.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 284693, 7 pages
http://dx.doi.org/10.1155/2013/284693
Research Article
Numerical Simulation on Dense Packing of Granular
Materials by Container Oscillation
Copyright © 2013 J. Liu and D. You. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The packing of granular materials is a basic and important problem in geomechanics. An approach, which generates dense packing
of spheres confined in cylindrical and cuboidal containers in three steps, is introduced in this work. A loose packing structure is
first generated by means of a reference lattice method. Then a dense packing structure is obtained in a container by simulating
dropping of particles under gravitational forces. Furthermore, a scheme that makes the bottom boundary fluctuate up and down
was applied to obtain more denser packing. The discrete element method (DEM) was employed to simulate the interactions between
particle-particle and particle-boundary during the particles’ motions. Finally, two cases were presented to indicate the validity of
the method proposed in this work.
Cylindrical container
o x o x
(a) (b)
Figure 1: The sketch map of reference cuboid for cylindrical (a) and cuboidal (b) container, respectively.
accepted or not by identifying overlap and variety of potential generating coordinates of spheres’ centers. The size of lattice
energy. Finally, a dense packing can be reached by repeating is determined by the sizes of particles and container. Finally,
the above course many times [12]. These above mentioned a random point is selected within each cubic lattice and the
methods mainly focus on reaching the highest packing point is regarded as a particle’s center. It should be noted
density instead of describing real mechanical characteristics that the random point within a cubic lattice must meet the
during the course of forming dense packing. request of no existing intersection between particle’s surface
In this work, the objective is to generate a dense packing and lattice’s faces.
structure of particles to model practical dense packing, for Let 𝑁𝑝 represent the number of particles; 𝑟max and 𝑟min
instance, structure of concrete, packing of grain, soil, and denote the maximal and minimal radius of the particles;
sand, and so forth, instead of only obtaining the highest 𝑅cylinder and 𝐻cylinder are the radius and height of cylinder; the
packing density. Therefore, the mechanical characteristics size of the cuboidal container is 𝐿 cuboid × 𝐿 cuboid × 𝐻cuboid ;
of forming dense packing are considered in our study. The and the size of the reference cuboid is 𝐿 𝑟 × 𝐿 𝑟 × 𝐻𝑟 . For the
Discrete Element Method (DEM) is employed to simulate cylindrical container, the size of reference lattice is limited by
particles’ packing confined in container (cylindrical con- the following equation:
tainer and cuboidal container). A loose structure was first
2𝑅max ≤ 𝐿 lattice ≤ √2𝑅cylinder . (1)
generated. Then the dropping of particles onto the bottom
of the container due to gravity is simulated. Furthermore, The size of reference cuboid for cylindrical container is
the bottom boundary of the container was fluctuated up and
down to reach a more dense packing. √2𝑅cylinder
𝐿 𝑟 = mod ( ) 𝐿 lattice ,
𝐿 lattice
2. Initial Loose Packing
A loose packing structure must be generated in order to { 𝑁𝑝 }
𝐻𝑟 = mod { 2 } 𝐿 lattice ,
simulate the process of particle packing under gravitational [mod ( √2𝑅cylinder /𝐿 lattice )] + 0.5
forces. Loose packing means that there is no overlap existing { }
(2)
between particle and its neighbors or between particle and
boundaries. The objective is to avoid interactions among where mod means remaining integral part of the result.
particles and between particles and boundaries at initial For the cuboidal container, the size of reference lattice is
state. A simple scheme has been applied to generate loose
packing. First, a reference cuboid whose projection in 𝑥𝑜𝑦 2𝑅max ≤ 𝐿 lattice ≤ 𝐿 cuboid . (3)
plane is a square has been constructed within container. The
symmetrical axis of reference cuboid is the same as that of The size of reference cuboid for cuboidal container is
container. For a cylindrical container, the reference cuboid 𝐿 cuboid
has the maximal size when it is inscribed to the container 𝐿 𝑟 = mod ( ) 𝐿 lattice ,
(see Figure 1(a)), but for cuboidal container, the maximal 𝐿 lattice
size of reference cuboid is the same as that of the container (4)
𝑁𝑝
within the range of the container height (see Figure 1(b)). 𝐻𝑟 = mod { 2
} 𝐿 lattice .
The size of the reference cuboid does not have a determinate [mod (𝐿 cuboid /𝐿 lattice )] + 0.5
value for a certain container because the length of its section
must be integral times of the size of lattice. The height of the 3. Mechanical Model
reference cuboid is determined by the number of particles.
Secondly, the reference cuboid is partitioned using cubic In the current study, the Discrete Element Method (DEM)
lattice. The cubic lattices are regarded as reference lattice for [13] has been employed to simulate the interactions among
Advances in Mechanical Engineering 3
particles and between particles and boundaries. DEM is a where 𝑚𝑙 , 𝐼𝑙 represent the mass and moment of inertia of
finite difference scheme. It is used to study assemblies of particle 𝑙, 𝑙 = 1, 2, . . . , 𝑁𝑝 , Ẍ𝑙 , 𝜃̈𝑙 represent acceleration and
individual particles. By monitoring the interaction between angular acceleration of particle 𝑙, F𝑙 , M𝑙 are the resultant force
particles, the behavior of the material is known. and resultant moment of particle 𝑙, and ∑ M𝑙 = ∑ F𝑇 𝑟𝑙 .
The equations of motion (11) are solved by means of Verlet
3.1. Contact Forces. The sketch map of contact model is integrator. See [10] for details.
shown in Figure 2.
The normal contact force 𝐹𝑁 and shear contact force 𝐹𝑇 3.3. Oscillations of Bottom Boundary. To improve the volume
can be expressed by fraction, the oscillations of bottom boundary are applied
after a dense packing formation under gravity forces. A
𝐹𝑁 = 𝑘𝑑 𝛿𝑖𝑗 + 𝜂𝑑 V𝑖𝑗 ⋅ n𝑖𝑗 , (5) sinusoidal function has been used to describe increment
of containers’ bottom boundaries displacements; see the
𝐹𝑇 = 𝑘𝜏 𝜏𝑖𝑗 + 𝜂𝜏 V𝑖𝑗 ⋅ t𝑖𝑗 , (6) following equation:
where 𝑘𝑑 , 𝑘𝜏 represent the normal and shear spring constant; 2𝜋
𝜂𝑑 , 𝜂𝜏 are the normal and shear damping coefficient due to Δ𝑈 (𝑡) = 𝐴 sin ( 𝑡) , (12)
𝑁Δ𝑡
the dissipative force; n𝑖𝑗 and t𝑖𝑗 are the unit vector in normal
direction and tangential direction, respectively, which can be where Δ𝑈(𝑡) is the increment of bottom boundary displace-
specified by ment, 𝐴 is the amplitude of increment, 𝑁Δ𝑡 is the period of
oscillation, 𝑁 is the number of time step within a period, and
X𝑖 − X𝑗 𝑡 is the current time. The increment of displacement must be
n𝑖𝑗 = ,
X𝑖 − X𝑗 small enough in each time step to keep stability of calculation.
The central coordinates of the bottom of containers vary with
(7)
V𝑖𝑗 − (V𝑖𝑗 ⋅ n𝑖𝑗 ) n𝑖𝑗 calculation time; that is,
t𝑖𝑗 = ,
V𝑖𝑗 − (V𝑖𝑗 ⋅ n𝑖𝑗 ) n𝑖𝑗 P (𝑡) = P0 + Δ𝑈 (𝑡) n𝑑 , (13)
where X𝑖 and X𝑗 are the positions of particle 𝑖 and 𝑗, where P(𝑡) is the central position vector of the bottom
respectively; V𝑖𝑗 is the relative velocity of particle 𝑖 and 𝑗 at boundary of container, n𝑑 is a normal vector of the bottom
contact points when forming contact, which can be expressed boundary, and P0 is the initial value of P(𝑡) (when 𝑡 = 0).
by
N Shear
damper
N k
d
𝜂d Normal
Normal 𝜂𝜏
z T spring
damper k𝜏
i Slider Shear spring
y j T j
x
(a) (b)
Figure 2: The sketch map of calculating contact force ((a) two particles in contact; (b) the mechanical model of contact. 𝑁-normal direction,
𝑇-tangential direction).
z
y z
y
x x
(a) (b)
Figure 3: Initial loose packing configuration for cylindrical (a) and cuboidal (b) container, respectively.
where 𝑉𝑝 is the sum of particles volume, 𝑎 is the side length structure can reach a higher volume fraction. However, there
of the bottom boundary, 𝑧𝑝 max is the maximal 𝑧 coordinate of is no definite relationship between oscillation frequency and
particles during particles motions, and 𝑟 is the corresponding the resulted volume fraction, as shown in Figure 9.
radius of the particle. For cylindrical container, replace 𝑎2 by
𝜋𝑅cylinder 2 . 5. Conclusions and Discussions
Figures 6(a), 6(b), 6(c), and 6(d) are the simulation
results before and after container oscillations for cylindrical The simulating results implicate that oscillation of container
and cuboidal container, respectively. In addition, the dense can significantly affect the packing density. However, different
packings of particles with four kinds of grain size distribution oscillation amplitudes and frequencies have different effects
were also simulated for cylindrical and cuboidal container, on packing density. For a given particles system, there
respectively, as shown in Figure 7. Figure 7 shows that the exists an optimal amplitude of oscillation corresponding
particles with smaller sizes can reach a larger volume fraction to the maximum volume fraction. But there is no definite
both for cylindrical and cuboidal containers. Furthermore, relationship between the frequency of oscillation and the
the effects of amplitude of oscillation on volume in the volume fraction.
case of keeping other factors the same were analyzed. An The packing density from our simulation does not exceed
optimal oscillation amplitude exists as shown in Figure 8. that from the method in the view of pure geometry. The
The frequency of oscillation was also analyzed in this work. main reason lies that the mechanical behaviors of interactions
In some values of oscillation frequency, the dense packing between particles and between particles and boundaries
Advances in Mechanical Engineering 5
Figure 4: The dropping process of particles under gravity forces within a cylindrical container ((a), (b), (c) and (d) are the configuration at
0.2 s, 0.35 s, 0.65 s, and 1 s, resp.).
Figure 5: The dropping process of particles under gravity forces within a cuboidal container ((a), (b), (c), and (d) are the configuration at
0.2 s, 0.35 s, 0.65 s, and 1 s, resp.).
Figure 6: The dense packing of particles in cylindrical and cuboidal container before and after container oscillation ((a), (c) before container
oscillation; (b), (d) after container oscillation).
6 Advances in Mechanical Engineering
Density (Kg/m3 ) Elastic modulus (N/m) Time length (s) Time step (s) Contact damping
2700 1.5 × 108 1 10−7 0.27
0.61 0.620
0.615
0.60
0.610
Volume fraction
Volume fraction
0.59 0.605
0.600
0.58
0.595
0.57 0.590
0.585
2 3 4 5 6 7 8 9 10 12 14 16 18 20
Average size of particles (mm) Frequency (Hz)
Figure 7: The relationship between volume fraction and average size Figure 9: The relationship between volume fraction and oscillation
of particles. frequency.
0.570
0 2 4 6 8 10 References
Oscillation amplitude (mm)
[1] D. J. Daley, “Packings and approximate packings of spheres,”
Cylindrical container Tech. Rep. 104, National Institute of Statistical Sciences,
Cuboidal container Research Triangle Park, NC, USA, http://www.niss.org.
Figure 8: The relationship between volume fraction and oscillation [2] J. A. Elliott, A. Kelly, and A. H. Windle, “Recursive packing of
amplitude. dense particle mixtures,” Journal of Materials Science Letters,
vol. 21, no. 16, pp. 1249–1251, 2002.
Advances in Mechanical Engineering 7
Research Article
Research on Flexible Joint Stiffness Test and
Trajectory Planning of Space Manipulator
Copyright © 2013 Yongtai Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Both Natural Coordinate Formulation describing rigid bodies and Absolute Nodal Coordinate Formulation describing flexible
bodies are used to model a flexible manipulator with flexible joint and flexible link. The torsional stiffness of flexible joint is tested
using a specialized stiffness test equipment, and then the nonlinear torsional stiffness is determined by fitting the experimental data.
A new trajectory planning function called the cosine-based function is proposed to design the joint trajectory, which is smoother
than the fifth-polynomial and cycloidal motion functions. Finally, a one-link manipulator with flexible joint and flexible link is
used to compare the performance of the three trajectory planning functions. Results show that residual vibration can be remarkably
reduced by the proposed cosine-based function, which exhibits a significantly better performance than the fifth-polynomial and
cycloidal motion functions.
Flexible joint 𝜂
u
X
Figure 1: One-link manipulator with flexile joint and flexible link.
Z
Y ri rj Normal (before)
𝛼
Normal (after)
2𝛼
X
The reflected laser beam (before) A0
Z Mirror (after)
The reflected laser beam (after)
Figure 3: Three-dimensional beam element of two nodes described ΔA
by ANCF.
A1
The torque-angle curve has two segments: nonlinear and ̇ (𝜃𝑓 − 𝜃0 ) 2𝜋𝑡
𝜃̃ (𝑡) = [1 − cos ( )] ,
linear. The quadratic polynomial is adopted to fit the nonlin- 𝑡𝑓 𝑡𝑓
ear segment, whereas the least square method is adopted to fit (15)
the linear segment. Finally, the relationship between torque ̈ (𝜃𝑓 − 𝜃0 ) 2𝜋𝑡
and angle can be expressed by the following formula: 𝜃̃ (𝑡) = [2𝜋 sin ( )] .
𝑡𝑓2 𝑡𝑓
2
{𝜏1 ( 𝛼 ) ,
{
|𝛼| < 𝛼1 ,
𝜏={ 𝛼1 (11) Cycloidal motion can evidently satisfy all boundary
{ conditions described by (12).
{𝜏1 + 𝐾1 (𝛼 − 𝛼1 ) , |𝛼| ≥ 𝛼1 , (3) In this paper, a cosine-based function is proposed to
where (𝛼1 , 𝜏1 ) is the turning point between nonlinear segment plan the joint trajectory, and it can be written as
and linear segment and 𝐾1 is the stiffness of linear segment.
The nonlinear stiffness of the joint, as shown in Figure 7, 𝜋𝑡 3𝜋𝑡
can be obtained by differentiating the above equation. 𝜃̃ (𝑡) = (𝜃𝑓 − 𝜃0 ) [𝑎0 + 𝑎1 cos ( ) +𝑎3 cos ( )] + 𝜃0 .
𝑡𝑓 𝑡𝑓
(16)
4. Trajectory Planning of the Joint
According to the six boundary conditions described by
In engineering, smooth and continuous functions are typi-
(12), the three unknown coefficients 𝑎0 , 𝑎1 , and 𝑎3 can be
cally adopted to plan the joint trajectory, the first and second
determined. Finally, the angle, angular velocity, and angular
derivatives of which are also smooth and continuous, such as
acceleration of the joint are as follows:
the fifth-polynomial and cycloidal motion functions. The six
boundary conditions of angle, angular velocity, and angular
1 9 𝜋𝑡 1 3𝜋𝑡
acceleration of the joint are as follows: 𝜃̃ (𝑡) = (𝜃𝑓 − 𝜃0 ) [ − cos ( )+ cos ( )] + 𝜃0 ,
2 16 𝑡𝑓 16 𝑡𝑓
̇ ̈
𝜃̃ (0) = 𝜃0 , 𝜃̃ (0) = 0, 𝜃̃ (0) = 0,
(12) ̇ (𝜃𝑓 − 𝜃0 ) 9 𝜋𝑡 3 3𝜋𝑡
𝜃̃ (𝑡𝑓 ) = 𝜃𝑓 ,
̇
𝜃̃ (𝑡𝑓 ) = 0,
̈
𝜃̃ (𝑡𝑓 ) = 0, 𝜃̃ (𝑡) = [ 𝜋 sin ( ) − 𝜋 sin ( )] ,
𝑡𝑓 16 𝑡𝑓 16 𝑡𝑓
where 𝑡𝑓 is the planning time, 𝜃0 is the initial angle of joint, (𝜃𝑓 − 𝜃0 ) 9 2
̈ 𝜋𝑡 9 3𝜋𝑡
and 𝜃𝑓 is the final angle of joint. 𝜃̃ (𝑡) = [ 𝜋 cos ( ) − 𝜋2 cos ( )] .
(1) The fifth-polynomial is adopted to plan the joint 𝑡𝑓2 16 𝑡𝑓 16 𝑡𝑓
trajectory. The polynomial coefficients are obtained by the six (17)
Advances in Mechanical Engineering 5
Angle (∘ )
𝜃̃max 8 𝑡𝑓 𝑡𝑓 4 𝑡𝑓
̈ 10 𝜃𝑓 − 𝜃0 𝜃𝑓 − 𝜃0 √3 2 𝜃𝑓 − 𝜃0 2
𝜃̃max ( 2 ) 2𝜋 ( ) 𝜋 ( 2 )
√3 𝑡𝑓 𝑡𝑓2 2 𝑡𝑓
1
final angle of the joint is set to be 5∘ . Figure 9: Joint angular velocity of the three planning functions.
The geometrical parameters and mechanical parameters
of the flexible link are shown in Table 2. Also, six ANCF
three-dimensional beam elements are used to discretized the
flexible link. motion (1 s ≤ 𝑡 ≤ 2 s), the residual angle errors are,
The above three planning functions are used to plan the respectively, 0.028∘ , 0.019∘ , and 0.002∘ . The tracking errors
joint trajectory. The motor is driven by the speed mode; caused by the three functions are evidently similar in the
that is, the motor rotor is assumed to be capable of abso- course of motion, but the residual vibration caused by the
lutely tracking the given trajectory using the three planning proposed cosine-based function is far less than that caused
functions, respectively. To determine the effects of the three by the fifth-polynomial and cycloidal motion functions after
planning functions on manipulator vibration, two cases are motion.
studied: (1) with joint flexibility considered and (2) with joint Figure 12 shows the angle errors of the joint of the three
flexibility and link flexibility considered. planning functions with joint flexibility and link flexibility
Figure 11 shows the angle errors of the three planning considered. Figure 13 shows the vibration of the endpoint of
functions with joint flexibility considered. In the course of the three planning functions with joint flexibility and link
motion (0 s ≤ 𝑡 ≤ 1 s), the maximal angle errors of the fifth- flexibility considered. In the course of motion (0 s ≤ 𝑡 ≤ 1 s),
polynomial, cycloidal motion, and cosine-based functions the maximal angle errors of the fifth-polynomial, cycloidal
are, respectively, 0.056∘ , 0.052∘ , and 0.056∘ , whereas after motion, and cosine-based functions are, respectively, 0.069∘ ,
6 Advances in Mechanical Engineering
0.10
40
Angular-acceleration (∘ /s2 )
20 0.05
Angle errors (∘ )
0
0.00
−20
−0.05
−40
0.10
3
0.05 2
Vibration of endpoint (cm)
Angle errors (∘ )
1
0.00
0
−0.05 −1
−2
−0.10
0.0 0.4 0.8 1.2 1.6 2.0 −3
Time (s) 0.0 0.5 1.0 1.5 2.0
Time (s)
Fifth-polynomial function
Cycloidal motion function Fifth-polynomial function
Cosine-based function (proposed) Cycloidal motion function
Cosine-based function (proposed)
Figure 11: Angle errors of the joint (with joint flexibility considered).
Figure 13: Vibration of the endpoint (with joint flexibility and link
flexibility considered).
Acknowledgments
This work is supported by Grant 10972033 of the Natural
Science Foundation of China. This work is also supported by
Grant 20100141107 of the China Academy of Space Technol-
ogy.
References
[1] S. Dubowsky, “Dealing with vibrations in the deployment
structures of space robotic systems,” in Proceedings of the 5th
International Conference on Adaptive Structures, pp. 5–7, Sendai
International Center, Sendai, Japan, 1994.
[2] M. W. Spong, “Modeling and control of elastic joint robots,”
Journal of Dynamic Systems, Measurement and Control, Trans-
actions of the ASME, vol. 109, no. 1, pp. 310–319, 1987.
[3] S. K. Ider and M. K. Özgören, “Trajectory tracking control of
flexible-joint robots,” Computers and Structures, vol. 76, no. 6,
pp. 757–763, 2000.
[4] G. Magnani, P. Rocco, and A. Rusconi, “Modeling and position
control of a joint prototype of DEXARM,” in Proceedings of
the 10th International Workshop on Advanced Motion Control
(AMC’ 08), pp. 562–567, March 2008.
[5] J. Martins, M. Ayala Botto, and J. S. Da Costa, “Modeling
of flexible beams for robotic manipulators,” Multibody System
Dynamics, vol. 7, no. 1, pp. 79–100, 2002.
[6] S. K. Tso, T. W. Yang, W. L. Xu, and Z. Q. Sun, “Vibration
control for a flexible-link robot arm with deflection feedback,”
International Journal of Non-Linear Mechanics, vol. 38, no. 1, pp.
51–62, 2003.
[7] A. Abe, “Trajectory planning for residual vibration suppression
of a two-link rigid-flexible manipulator considering large defor-
mation,” Mechanism and Machine Theory, vol. 44, no. 9, pp.
1627–1639, 2009.
[8] D. J. Garcı́a and E. Bayo, Kinematic and Dynamic Simulation
of Multibody Systems: the Real-Time Challenge, Springer, New
York, NY, USA, 1994.
[9] A. A. Shabana and R. Y. Yakoub, “Three dimensional absolute
nodal coordinate formulation for beam elements: theory,”
Journal of Mechanical Design, vol. 123, no. 4, pp. 606–613, 2001.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 257457, 13 pages
http://dx.doi.org/10.1155/2013/257457
Research Article
Case Study on Influence of Step Blast-Excavation on Support
Systems of Existing Service Tunnel with Small Interval
Shaorui Sun, Ling Yue, Jimin Wu, Jin Liu, and Jihong Wei
College of Earth Science and Engineering, Hohai University, Nanjing 210098, China
Copyright © 2013 Shaorui Sun et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
During the construction of newly built tunnel (NBT) adjacent to the existing service tunnel (EST), stability of the EST with small
interval is affected by vibration waves which are caused by blasting load. The support structures of the EST will be cracked and
damaged, while the unreasonable blast-excavation methods are adopted. Presently, the studies on behavior of support structure
in the EST under blasting load are not totally clear, especially for the bolts system. Besides, the responses of support structure on
blasting load are lacking comprehensive research. In this paper, New Zuofang tunnel is taken as a study case to study the influence
of step blast-excavation in NBT on support structures of the EST through field experiment and numerical simulation. Some data,
such as blasting vibration velocity (BVV) and frequency of support structures, are obtained through field measurement. Based on
these data, the formula of BVVs is obtained. Research on stability of tunnel support structures affected by step blast-excavation
is conducted using numerical simulation method. The dynamic-plastic constitutive model is adopted in the software ABAQUS to
assess safety of support structures. The range and degree of damage for the support structures are obtained. In addition, change
laws of axial force and stress with time for the bolts are analyzed.
1. Introduction masses. Kuszmaul [3] has obtained the “KUS damage” con-
stitutive model for fragmentation of rock under the dynamic
The structural types of tunnel with small interval (less than 1.5 loading. Oriard [4] and Law et al. [5] have researched the
times of tunnel width) exist widely in the tunnelling projects. accumulative damage of surrounding rock due to multitimes
The major risks resulted from vibration failure induced by repetitive blasting loads by monitoring changes of peak
blasting load in NBT when they are excavated by the drill BVVs and velocity of acoustic wave. Preece and Thlone [6]
and blast method. The blasting load in NBT will be easy to have studied the detonation time and fragmentation using
cause the adverse influence of blast-induced vibrations in the PRONTO-3D dynamic finite-element program and obtained
EST lining. How to protect the support structures of the EST a modified “KUS damage” constitutive model. Doucet et al.
during blast-excavation in NBT has become to a significant [7], Villaescusa et al. [8], and Ramulu et al. [9] have studied
and valuable research work. It is necessary to carry out both the damage degree and damage range of surrounding rock
experiments and numerical analysis to research the impact of under multi-times blasting by use of sliding micrometer
BVVs on the support structures of the EST. and imaging well log technique. Singh [10] has introduced
The damage of concrete lining in the EST under blasting the blasting vibration damage to underground coal mines
load has been studied by lots of researchers. For example, from adjacent to open-pit blasting. Toraño et al. [11] have
Hisatake et al. [1] have researched a dynamic method with researched FEM models including randomness and its appli-
respect to effects of adjacent blast operation on vibration cation to the blasting vibrations prediction. Nateghi et al.
behavior of the EST. Krajeinovie [2] has introduced the [12] have researched a negative effects method of blasting
damage acoustic wave of surrounding rock and obtained an waves on concrete by analyzing parameters of underground
important result that damage cracking of surrounding rock vibration. The control method of blasting waves has been
can be expressed by change rate of acoustic wave in rock proposed based on the peak BVVs of underground structure.
2 Advances in Mechanical Engineering
(a) Monitoring points in the EST (b) Monitoring points in the field
4.0
9.0
Peak vibration velocity (cm/s)
3.0
7.0
2.0 5.0
1.0 3.0
0.0 1.0
21.0 23.0 25.0 27.0 29.0 31.0 33.0 35.0 21.0 23.0 25.0 27.0 29.0 31.0 33.0 35.0
Distance (m) Distance (m)
(a) Vertical BVVs (b) Horizontal BVVs
Figure 4: Relationship between horizontal and vertical peak BVVs and distance to the blasting center.
are 1.185 and 3.544 for the vertical BVV and 2.443 and 2.427 The dominant vibration frequency is less than 50 Hz when
for the horizontal BVV, respectively. the BVV is more than 6 cm/s. The vibration frequency at
the position of the greater BVV is mostly low frequency.
3.2. Application of Sadov’s Formula. The BVVs of concrete In general, the frequency decreases with distance increasing
lining at different positions and excavation steps are predicted in the inhomogeneous rock masses. The existing joints and
by using Sadov’s formula and shown in Figure 5. As seen, the fractures in intermediate strata have a great effect on the stress
BVV values of the 4th and 7th steps are bigger at point 1, wave spread. High BVV and low frequency intensify the dam-
and these values are close to the limited safety velocity con- age of concrete lining, because the low vibration frequency
trolled by “Blasting Safety Code” published by Chinese State is close to the natural frequency of the EST lining (about 10–
Administration of Work Safety. In addition, the cracks occur 15 Hz) and causes resonance of concrete lining. Therefore, the
at the hance of the EST at the 4th step and the 7th step; this concrete lining is not safe at the position with the maximum
phenomenon is consistent with the observation in the field. BVV. The high BVV occurs mostly at the 4th step, which leads
to the greater possibility of damage at the position of hance
3.3. Analysis on Dominant Vibration Frequency. The dom- and spandrel facing the blasting side.
inant vibration frequency distribution of BVVs measured
at the different monitoring points is shown in Figure 6. As 3.4. Time-Domain Analysis of BVV. The attenuation laws of
seen, BVVs less than 5 cm/s have a high frequency and their particle BVV are obtained by the time-domain analysis of
dominant vibration frequencies are distributed at the range BVV. The curve of BVV of point 2 at the 4th step is shown in
of 50–150 Hz. This indicates that the vibration waves caused Figure 7. As seen, the time-dominant curve has lots of wave
by blasting load are low-frequency waves, and its frequency peaks and troughs with continuous intervals. This is caused
is much larger than the natural frequency of concrete lining. by the millisecond blasting in each borehole and vibration
Advances in Mechanical Engineering 5
6.0
12.0
5.0 10.0
Vibration velocity (cm/s)
3.0 6.0
2.0 4.0
1.0 2.0
0.0 0.0
1 2 3 4 7 1 2 3 4 7
Excavation steps Excavation steps
1 4 1 4
2 5 2 5
3 6 3 6
(a) Vertical BVVs (b) Horizontal BVVs
500.0 BVV, which lasts for 0.55 s. The massive stress waves generate
Frequency of main vibration (Hz)
6.92 3.56
0 0
−6.92 −3.56
−0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 −0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95
Time (s) Time (s)
(a) Vertical BVVs (b) Horizontal BVVs
Materials Weight/kN/m3 Elastic modulus/GPa Poisson’s ratio Internal friction angle/∘ Dilation angle/∘ Cohesion/kPa
Surrounding rock 25.50 2.06 0.3 30.6 30.6 20
Bolts 78.50 210 0.2 / / /
under the different blast-excavation steps. Finally, the plastic reliable to analyze the law of stress and strain of support
damage in the concrete lining and the laws of response on structures under different blasting excavation steps.
bolts are studied by introducing dynamic-plastic constitutive
model of concrete. 4.3. Analysis on Damage Degree of the EST Lining
4.3.1. Damage Mechanism of Concrete Lining under Cyclic
4.1. Physical-Mechanical Parameters of Rock Masses Load. The representation of damage variable proposed by
and Concrete Kachanov is widely used to describe the damage variable. The
formula is listed as follows:
(1) Physical-Mechanical Parameters of Surrounding Rock. The
plastic state under blasting load might appear in the sur- 𝐴∗ 𝐴 − 𝐴∗ Δ𝐴
𝐷=1− = = , (2)
rounding rock, so the Drucker-Prager model is used to sim- 𝐴 𝐴 𝐴
ulate damage of concrete lining, and the elastic model is used
where 𝐷 is the damage variable value, 0 ≤ 𝐷 ≤ 1. It indicates
to simulate bolts. According to the reports of engineering
no damage when 𝐷 is equal to zero. It indicates complete
geological investigation and the results of laboratorial exper-
damage when 𝐷 is equal to 1.0; 𝐴 is initial sectional area of
iments, the physical-mechanical parameters are listed as
damage zone; 𝐴∗ is the effective bearing area after the struc-
Table 2.
ture is damaged.
(2) Physical-Mechanical Parameters of Concrete Lining. Dam- The process of damage in the concrete lining can be
age-plastic constitutive model is defined as the constitutive regarded as a cyclic loading process with high loading rate,
model of shotcrete and secondary lining. The physical- because of the influence of repeated vibration caused by
mechanical parameters of concrete lining are shown in multi-times blast-excavation and a high loading rate being
Table 3. 5 m/s with a longer time of blasting vibration. This process
is also regarded as an accumulative process of multi-times
dynamic damages. The distribution of damage in the concrete
4.2. Rational Verification of Numerical Calculation. The aver- lining and degree of damage in different points can be
age values of vertical BVVs at point 1 by the numerical simula- obtained by the damage-plastic constitutive model of con-
tion based on the software ABAQUS and the field experiment crete in the process of numerical analysis.
are shown in Table 4. The BVVs obtained by numerical simu-
lation are larger than those of field experiment. This is because 4.3.2. Comparison Analysis on Damage under Different Blast-
there are a large number of discontinuities in rock masses Excavation Steps. The damage process of the EST lining is
in field experiment to make the stress waves decay faster irreversible because of the sublevel millisecond blasting. The
than those in the rock masses regarded as a homogeneous fatigue damages of concrete lining occur on the support
material in numerical simulation. In addition, the calculation structures of the EST under a longtime dynamic load. The
formula of blasting-borehole pressure is set up in an ideal damage process of concrete lining caused by blasting load at
condition. The blasting energy is totally translated into the different excavation steps is shown in Figure 8.
force to throw the rock into the air. The allowed error As seen in Figure 8, there is no damage appearing in
range of BVVs larger than actual values is about 10%, so concrete lining at the 1st step and the 2nd step, which means
the calculation results obtained by numerical calculation are that the stress wave is not strong enough to produce damage
Advances in Mechanical Engineering 7
Table 4: Comparison between vertical BVVs of field experiment and that of numerical calculation.
Excavation steps The 1st step The 2nd step The 3rd step The 4th step The 7th step
Field experiment/cm/s 2.642 3.622 4.740 8.616 8.668
Numerical simulation/cm/s 3.123 4.342 6.346 10.125 10.084
Error 18.2% 19.9% 33.9% 17.5% 16.3%
(a) The 1st step and the 2nd step (C) (b) The 3rd step (C) (c) The 4th step (C)
Damage (C)
Damage (T) Damage (T)
(average: 75%) (average: 75%)
+1.825e − 01 (average: 75%)
+1.684e − 01 +0.000e + 00 +8.213e − 01
+1.554e − 01 +0.000e + 00 +7.529e − 01
+1.404e − 01 +0.000e + 00 +6.844e − 01
+1.263e − 01 +0.000e + 00 +6.160e − 01
+1.123e − 01 +0.000e + 00 +5.475e − 01
+9.826e − 02 +0.000e + 00 +4.791e − 01
+8.422e − 02 +0.000e + 00 +4.106e − 01
+7.019e − 02 +0.000e + 00 +3.422e − 01
+5.615e − 02 +0.000e + 00 +2.738e − 01
+4.211e − 02 +0.000e + 00 +2.053e − 01
+2.807e − 02 +0.000e + 00 +1.369e − 01
+1.404e − 02 +0.000e + 00 +6.844e − 02
+0.000e + 00 +0.000e + 00 +0.000e + 00
(d) The 7th step (C) (e) The 1st step and the 2nd step (T) (f) The 3rd step (T)
Damage (T)
Damage (T) (average: 75%)
(average: 75%) +9.640e − 01
+8.882e − 01
+8.898e − 01
+8.142e − 01
+8.157e − 01
+7.401e − 01
+7.415e − 01
+6.661e − 01
+6.674e − 01
+5.921e − 01 +5.932e − 01
+5.181e − 01 +5.191e − 01
+4.441e − 01 +4.449e − 01
+3.701e − 01 +3.708e − 01
+2.961e − 01 +2.966e − 01
+2.220e − 01 +2.225e − 01
+1.480e − 01 +1.483e − 01
+7.401e − 02 +7.415e − 02
+0.000e + 00 +0.000e + 00
(g) The 4th step (T) (h) The 7th step (T)
Figure 8: Damage of concrete lining under different excavation steps (C: compression; T: tensile).
8 Advances in Mechanical Engineering
because the wave intensity attenuates while spreading to the on the range and degree of damage. Damage appears at the
concrete lining. Damage of concrete lining begins at the 3rd 3rd step, the 4th step and the 7th step, and the largest one
step. The range and area of damaged zone of secondary lining is at the 4th step. Therefore, the blasting parameters should
are almost the same to those of shotcrete. The compressive be controlled mainly at the 4th step and the fifth step. The
damage zone focuses on secondary lining and arch foot facing maximum damage value is between 0.8 and 0.9 with some
the blasting side. The cyclic-dynamic stress concentration microcracks appearing at the damage zone. The total damage
takes place in the aforementioned positions due to their in the concrete lining after every round blast-excavation is
acute angle types. This position can be damaged easily under shown in Figure 9. As seen, the macrocracks appear at the
compressive stress. The maximum damage value less than 1.0 hance facing the blasting side of concrete lining, arch foot,
under the compressive stress is smaller than the limited value, and arch spandrels, and at spandrel back facing the blasting
so only a few of microcracks appear on the subsurface of side of concrete lining.
concrete lining. The buckling failure of concrete lining and It can be seen from Figure 9(b) that the damage degree
shotcrete is not produced at the 3rd step. The macrocracks of concrete lining at back facing the blasting side is higher
begin to emerge when the maximum damage value caused by than that of facing the blasting side. This phenomenon is
tensile stress in the concrete lining is 0.8213. However, these consistent with the distribution of peak principle stress.
cracks without connection fail to result in the failure of con- The simulation results can be proved by field observation.
crete lining. The maximum damage value in the shotcrete is Some reinforcement measures of concrete lining are taken by
0.6974. The damage degree of shotcrete is smaller than that of railway administrative department. Two main reasons lead to
secondary lining, but the damage range of shotcrete is greater the cracks in the concrete lining. Firstly, because of the close-
than that of secondary lining. The minor damage at the back ness of EST to NBT, the charge is too excessive to produce
facing the blasting side of concrete lining can be seen from a large blasting load. Secondly, lots of initial defects exist in
the damage graph. This phenomenon shows that the intensity the concrete lining because of concrete aging. Comparing
diffracted here is still very strong when the stress waves Figure 9(a) with Figure 9(b), it can be found that the range
spread to this position. At the 4th step, the minor compressed and degree of damage due to tensile stress are larger than
damage in shotcrete and small range of compressive damage those of damage due to compressive stress. This is because the
in secondary lining appear at the same position with the 3rd cyclic principle tensile stress in the concrete lining is big and
step. The maximum damage value is 0.6123, and the degree it is close to tensile strength of concrete. Besides, the duration
of damage is larger than that at the 3rd step. The maximum of blasting load with 65 ms promotes the development of
damage value in tensile stress zone is 0.8720, and the macro- damage zone.
cracks begin to emerge but not posing failure of concrete
lining. Larger damage zone exists at arch foot, which is similar 4.3.3. Development Process of Damage in the Concrete Lining.
to the 3rd step, and the maximum damage value is 0.8882. The expansion of damage range and deepening of damage
Comparing with the 3rd step, there is no damage emergeing degree at the 4th step are taken as example to analyze the
at the back facing the blasting side of concrete lining. Damage development of damage in the concrete lining under the
generation concentrates mainly on the inside of invert at the successive blasting load.
7th step. The range and degree of damage are small under
the compressive stress and become bigger under the tensile (1) Expansion Process of Damage Zone. The range of damage
stress. The punctuate distribution of damage zones appears zone in the concrete lining becomes larger with the excava-
in the concrete lining at the 7th step, and the damage degree tion steps firstly and then tends to be stable. The distribution
in shotcrete is far less than that in secondary lining under the range of damage at different time is shown in Figure 10. As
tensile stress. seen, the damage caused by tensile stress occurs firstly near
According to the aforementioned analysis, the distance subsurface facing the blasting side at 0.012 s, and the damage
from blasting center to monitoring point has great impact value is 0.01463. The damage zone enlarges to two sides with
Advances in Mechanical Engineering 9
the lasting function of stress wave, and the degree of damage 1.0
deepens continually. The damage zone of hance facing the
blasting side enlarges obviously at 0.057 s, and the maximum 0.8
value is up to 0.1480. The damage zone emerges at arch foot
facing the blasting side, and two aforementioned sections are
0.6
Damage value
(2) Deepening Process of Damage Degree. The damage emer- Figure 11: Development process of damage value at the 4th step.
gence of concrete lining in the EST is caused by the blast-
excavation. Under the cyclic loading, the mechanical param-
eters of concrete lining began to decrease, and damage degree
enlarged gradually. Meanwhile, the damage value 𝐷 began to The increase of strain and displacement does not coincide
increase from zero to a stable value. with stress under high strain rate load. This means that it
The damage process of concrete lining with time at the 4th does not have enough time to make deformation increase
step is shown in Figure 11. As seen, the damage of concrete with increase of stress under the inertia function. This is the
lining does not appear immediately but later after blast- theory of damage effect lag that the cracks lag to appear in
excavation. This may be caused by two main reasons. Firstly, the concrete lining. The theory based on the deformation of
it requires enough time to make stress wave spread from concrete lining is the direct reason of damage generation.
blasting center to the concrete lining. Secondly, the concrete It needs some time to make the damage of concrete lining
lining has inertia effect under the dynamic cyclic loading. reach the maximum value once the concrete lining emerges
10 Advances in Mechanical Engineering
5 6
4 7
2234 3 8
2 9
1 10
16.0
150.0
0.0
12.0 0.0 3.0 6.0 9.0 12.0 15.0 18.0
−150.0
Displacement (mm)
Stress (kPa)
8.0
−300.0
4.0 −450.0
−600.0
0.0
0.0 3.0 6.0 9.0 12.0 15.0 18.0
−750.0
−4.0 −900.0
Distance (m) Distance (m)
Figure 14: Stress and displacement at the arch section (H: horizontal direction; V: vertical direction; NA: no bolt; A: bolt).
Table 5: Peak axial forces on the middle of bolts under different excavation steps (units: kN).
that of without bolts. The arrangement of bolts enhances influence appears at the 2nd step. This is because the small
equivalently the stiffness of surrounding rock at the arch distance between monitoring point and blasting center has
section. Comparing with surrounding rock without bolts, the great effect on axial forces of bolts and the change of blasting
capacity of bearing dynamic load of surrounding rock with center and angles of bolts have significant impact on axial
bolts is enhanced obviously. forces of bolts.
4.4.2. Axial Force Analysis of Bolts at the Arch Section under 4.4.3. Time-History Analysis on Axial Stress of Bolts. The
Different Steps. It can be seen from Figure 13 the bolts under strength of bolts is enhanced obviously after lasting stress
different blasting loads. They also suffer different axial forces waves from surrounding rock. The axial stress of bolts
with different angles even at the same excavation step. The vibrates intensely with time. The time-history analysis on
peak axial force on the middle of No. 1, No. 3, No. 5, No. 8, axial stress on the middle of No. 3, No. 5, and No. 8 bolts is
and No. 10 bolt is shown in Table 5. shown in Figure 15.
As seen in Table 5, the peak axial forces of bolts at the As seen in Figure 15, bolts with different angles have dif-
arch section facing the blasting side are generally larger than ferent change trends of axial stress. It indicates that the stress
those of bolts at other section. This is because the bolts in this waves with different angles have the different spread laws.
position are influenced by stress waves firstly, and the stress Just as the time-history of axial stress in the concrete lining,
waves attenuate when they spread to the other side. About the the stresses in bolts vibrate intensely and last for 0.1 s, which
axial forces of bolts at the same step, the sequence from big to is almost consistent with the function time of blasting load,
small is No. 3, No. 1, No. 5, No. 8, and No. 10 bolts, respectively. and then concuss slightly because of inertia function. They
It indicates that the angle of bolts has great effect on the do not decrease to zero during the calculation time while
axial force of bolts. Dynamic load has the greatest impact on vibrating appreciably around a certain value. The permanent
inclined bolts, horizontal bolts secondly, and vertical bolts at compressive stress and tensile stress in bolts caused by the
last. About the axial force of the same bolt, the sequence from vibration function of blasting load, rotation, or extrusion of
big to small is the 4th step, the 7th step, the 3rd step, the 1st rocks in the arch section. The tensile stress or compressive
step, and the 2nd step, respectively. Stress waves generating at stress of bolts prevents the rotation or extrusion of rock
the 4th step have greatest impact on bolts, and the smallest masses. These stresses are put together with static stress of
12 Advances in Mechanical Engineering
100.0 4.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
80.0
−4.0
60.0 −8.0
Axial stress (MPa)
0.0 −24.0
0.0 0.2 0.4 0.6 0.8 1.0
−28.0
Time (s)
−20.0 Time (s)
(a) No. 3 bolt (b) No. 5 bolt
15.0
10.0
5.0
Axial stress (MPa)
0.0
0.0 0.2 0.4 0.6 0.8 1.0
−5.0
−10.0
−15.0
Time (s)
(c) No. 8 bolt
bolts caused by ground stress. Bolts force will increase if two the largest peak principle stress appears at the 4th step and the
kinds of stress are in the same direction. On the contrary, 7th step, the damage zones of the concrete lining are mainly
bolts force will decrease. The concussion process of stress concentrated in hance and arch foot facing the blasting side.
in bolts with different angles exists significant differences, Stress in support structures concusses with many peak values.
for example, the superposition phenomenon of axial stress The concussion lasts for very long time which is far longer
appears in bolts (as No. 3 bolt) facing the blasting side under than that of stress in borehole. It is shown on the stress wave
blasting load. However, it does not appear in bolts back facing curve that the first peak of stress is the largest one, and then
the blasting side (as No. 5 and No. 8 bolts), which are vibrated peaks attenuate gradually. Stress waves which spread in the
by tensile stress and compressive stress circularly. As a result, concrete lining present superposition phenomenon at the 1st
the yield strength of bolts decreases obviously under cyclic step and the 2nd step.
load. (3) Damage of support structures happens at the 3rd step,
the 4th step, and the 7th step. It is mainly concentrated on
5. Conclusions the concrete lining facing the blasting side. The most serious
damage happens at the 4th step. This shows that the damage
(1) At the same monitoring point, BVVs caused by different degree is related to the distance from blasting center in terms
blast-excavation steps decrease with the distance to blasting of the same charge. The cracks at hance and spandrel facing
center increasing. After analysis on particle BVVs from the blasting side are caused by the cumulative damage of these
different positions to blasting center, Sodev’s formula which is three steps. The area of damage zone, as well as distribution
suitable for similar projects is obtained. The particle BVVs in law of peak damage value, begins to extend from one point on
the concrete lining subsurface are forecasted by this formula. the subsurface of concrete lining to any direction. At the same
(2) Under different blast-excavation steps, the peak prin- monitoring point, damage does not happen immediately after
ciple stress appears at spandrel at the 1st step and the 3rd step blast-excavation occurs, but it experiences a delay time before
and appears at hance at the 2nd step and the 4th step. Because enlarging rapidly.
Advances in Mechanical Engineering 13
(4) It can be seen from the distribution of damage zone a railway tunnelling project—a case study,” Tunnelling and
that the range and degree of damage in the surrounding rock Underground Space Technology, vol. 24, no. 2, pp. 208–221, 2009.
at the arch section are very small. This proves that bolts [10] P. K. Singh, “Blast vibration damage to underground coal mines
improve the capacity of bearing dynamic load for the sur- from adjacent open-pit blasting,” International Journal of Rock
rounding rock efficiently. In addition, bolts with different Mechanics and Mining Sciences, vol. 39, no. 8, pp. 959–973, 2002.
angles suffer different stresses. The inclined bolts suffer larger [11] J. Toraño, R. Rodrı́guez, I. Diego, J. M. Rivas, and M. D. Casal,
stress than that of horizontal bolts and vertical bolts. This “FEM models including randomness and its application to the
shows that stress wave with oblique incident is stronger than blasting vibrations prediction,” Computers and Geotechnics, vol.
that with forward incident. After blast-excavation is finished, 33, no. 1, pp. 15–28, 2006.
the fact that the residual stress is left in bolts proves that the [12] R. Nateghi, M. Kiany, and O. Gholipouri, “Control negative
interaction dislocation happens in the surrounding rocks at effects of blasting waves on concrete of the structures by
the arch section. analyzing of parameters of ground vibration,” Tunnelling and
Underground Space Technology, vol. 24, no. 6, pp. 608–616, 2009.
[13] M.-N. Wang, X.-M. Pan, C.-M. Zhang, X.-D. Wen, and K.-K.
Acknowledgments Wang, “Study of blasting vibration influence on close-spaced
This paper is financially supported by the National Natural tunnel,” Rock and Soil Mechanics, vol. 25, no. 3, pp. 412–414,
2004.
Science Foundation of China (Grants nos. 41002089 and
41102162) and Jiangsu Overseas Research & Training Program [14] J.-R. Li, X. Xia, H.-B. Li, X.-W. Wang, and Q.-C. Zhou, “Study
on blast-induced bedrock damage extension for a nuclear
for University Prominent Young and Middle-Aged Teachers
power station project,” Chinese Journal of Rock Mechanics and
and Presidents. The authors gratefully acknowledge M. S. Engineering, vol. 23, no. A01, pp. 4674–4678, 2005.
Chuanlei Zhang at the School of Earth Sciences and Engi-
[15] X. Xia, J. Li, H. Li et al., “Study on damage characteristics of
neering, Hohai University, China, for his contribution to the rock mass under blasting load in Ling’ao Nuclear Power Station,
field experiment. The authors would also like to acknowledge Guangdong Province,” Chinese Journal of Rock Mechanics and
the editors and reviewers of this paper for their very helpful Engineering, vol. 26, no. 12, pp. 2510–2516, 2007.
comments and valuable remarks. [16] Z. R. Feng and X. Wen, “Numerical analysis on existing service
tunnel under blasting load in new tunnel,” Blasting, vol. 25, no.
References 4, pp. 20–23, 2008.
[17] Q.-S. Zhang, L.-P. Li, S.-C. Li, W.-T. Ding, and W.-L. Hong,
[1] M. Hisatake, S. Sakurai, and T. Ito, “Effects of adjacent blast
“Experimental study of blasting dynamic vibration of closely
operation on vibration behavior of existing tunnel,” Proceedings
adjacent tunnels,” Rock and Soil Mechanics, vol. 29, no. 10, pp.
of the Japan Society of Civil Engineers, vol. 1983, no. 332, pp. 67–
2655–2666, 2008.
74, 1983.
[18] Y. S. Shen, B. Gao, and Z. J. Wang, “Effect of blasting in double
[2] D. Krajeinovie, “Continuum damage mechanics,” Applied
line tunnel on existing tunnel,” Chinese Journal of Underground
Mechanics Reviews, vol. 37, no. 1, pp. 1–32, 1984.
Space and Engineering, vol. 5, no. 5, pp. 980–984, 2009.
[3] J. S. Kuszmaul, “New constitution model for fragmentation
of rock under dynamic loading,” in Proceedings of the 2nd [19] L. J. Chen, “Experimental study on blasting vibration influence
International Symposium on Rock Fragmentation by Blasting, pp. on nearby tunnels with small interval,” Blasting, vol. 27, no. 2,
412–423, 1987. pp. 36–39, 2010.
[4] L. L. Oriad, “The scale of effects in evaluating vibration damage [20] Y.-B. Yang, M.-G. Liu, G.-H. Zhang, and Q. Li, “Analysis of con-
potential,” in Proceedings of the 15th Annual Conference on struction parameter optimization for new large cross-section
Explode Sives and Blasting Research, pp. 161–176, International tunnel next to existing tunnels,” Rock and Soil Mechanics, vol.
Society of Explosives Engineers, New Orleans, La, USA, 1989. 31, no. 4, pp. 1217–1226, 2010.
[5] T. M. Law, J. May, A. T. Spathis, A. T. du Plessis, and A. M.
Palmer, “Blast damage and blast dilution control: the application
of bulk emulsion systems at the WMC St Ives junction mine,”
Fragblast, vol. 5, no. 1-2, pp. 1–20, 2001.
[6] D. S. Preece and B. J. Thlone, “A study of detonation time and
fragmentation using 3-D finite element techniques and a dam-
age constitutive model,” in Proceedings of the 5nd International
Symposium on Rock Fragmentation by Blasting, pp. 147–156,
1996.
[7] C. Doucet, A. Cameron, and Y. Lizotte, “Effects of rock mass
characteristics on fragmentation in controlled blasting exper-
iments in small development headings,” in Proceedings of the
22nd ISEE’s Annual Conference on Explosives and Blasting Tech-
nique, pp. 1–10, Orlando, Fla, USA, February 1996.
[8] E. Villaescusa, I. Onederra, and C. Scott, “Blast induced damage
and dynamic behaviour of hangingwalls in bench stoping,”
Fragblast, vol. 8, no. 1, pp. 23–40, 2004.
[9] M. Ramulu, A. K. Chakraborty, and T. G. Sitharam, “Damage
assessment of basaltic rock mass due to repeated blasting in
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 296106, 19 pages
http://dx.doi.org/10.1155/2013/296106
Research Article
The Effect of First-Order Bending Resonance of
Wheelset at High Speed on Wheel-Rail Contact Behavior
Copyright © 2013 Shuoqiao Zhong et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
The first-order bending deformation of wheelset is considered in the modeling vehicle/track coupling dynamic system to investigate
its effect on wheel/rail contact behavior. In considering the effect of the first-order bending resonance on the rolling contact of
wheel/rail, a new wheel/rail contact model is derived in detail in the modeling vehicle/track coupling dynamic system, in which
the many intermediate coordinate systems and complex coordinate system transformations are used. The bending mode shape and
its corresponding frequency of the wheelset are obtained through the modal analysis by using commercial software ANSYS. The
modal superposition method is used to solve the differential equations of wheelset motion considering its flexible deformation
due to the first-order bending resonance. In order to verify the present model and clarify the influence of the first-order bending
deformation of wheelset on wheel/track contact behavior, a harmonic track irregularity with a fixed wavelength and a white-noise
roughness are, respectively used as the excitations in the two models of vehicle-rail coupling dynamic system, one considers the
effect of wheelset bending deformation, and the other does not. The numerical results indicate that the wheelset first-order bending
deformation has an influence on wheel/rail rolling contact behavior and is easily excited under wheel/rail roughness excitation.
Carriage
The second
The first bending
suspension
Bogie frame of wheelset
causing
polygonal wear
Z
The primary
suspension
Track
central line
X
Rail Clearance
Y
Fastening O Slab
Rubber pad
Rail cant 𝛼 Rail cant 𝛼
as a lumped and discrete structure by Popp et al. In [4, 6– deformation on the wheel/rail rolling contact behavior. The
8], the existing models of the flexible wheelset were contin- main object is to analyze the effect of wheelset flexibility
uous models. A vehicle/track dynamic model developed by on the wheel/rail contact behavior mainly in mid-frequency
Meinders considering wheelset flexibility was based on the range. The mid-frequency components of the excitation from
cosimulation of finite element analysis software ANSYS and the wheel/rail interaction system are filtered by the suspen-
multibody dynamics simulation software NEWEUL [9, 10]. sion system, and the wheel/rail mid-frequency excitation
Baeza et al. [11–13] proposed a numerical dynamic system cannot easily excite the flexibility of the car-body and bogies.
model originating from a rotating elastic cylinder model, Hence, the influence of the structural flexibility of car-body
taking inertial effect and effect of Coriolis force into account. and bogies is ignorable in this present paper. In Section 2,
In terms of wheel/rail rolling contact modeling, the the model of the flexible wheelset and the wheel/rail contact
effect of wheelset deformation in mid- and high-frequency model are introduced. In Section 3, a harmonic track irregu-
ranges should not be neglected [14]. Though many literatures larity with a fixed wavelength and white-noise roughness of
presented different methods to model flexible wheelsets, the the rail are adopted as the excitations to analyze the effect of
detailed derivations of rolling contact geometry of a flexible the wheelset flexibility.
wheelset and a pair of flexible rails and the detailed calcu-
lations and discussions on the effect of the high-frequency 2. Modeling
bending of wheelset on the rolling contact behavior were not
given. This is a very difficult research. The influence exerted 2.1. Vehicle/Track Model. Conventional vehicle system mod-
by the wheelset bending formation is much concerned with els consist of one carriage, two bogies, and four wheelsets,
wheelset polygonal wear. In [15], the effect of wheelset flexible which are considered as rigid bodies or lumped masses
deformation on the creepages and the creep forces in static connected by spring-damping systems. A ballasted track is
state was indicated to be significant. The flexible deformation usually modeled as a triple-layer model of discrete elastic
of a wheelset in service always contains mid- and high- support. The rail is modeled as a Timoshenko beam on elastic
frequency components. The rolling contact behavior of the point supporting foundation. The two ends of the calculation
wheel/rail caused by wheelset deformation would be in mid- rail are simply supported. The vertical and lateral bending
and high-frequency ranges, which brings a great difficulty deformations and twisting of the rails are taken into account.
into the on-line calculation of wheel/rail contact geometry. The sleepers are modeled as rigid bodies, and the ballasts
And the existing models for the wheel/rail contact geometry are regarded as discrete equivalent bodies. The equivalent
are not suitable for the model with flexible wheelsets and need spring-damping systems are used as the connections between
to be further improved. the rails and the sleepers, the sleepers and the equivalent
In this paper, a new wheel/rail contact model is developed ballast bodies, and the ballast bodies and the subgrade.
to consider the effect of the wheelset first-order bending Figure 2 illustrates such a model of the vehicle and the track.
Advances in Mechanical Engineering 3
FzfL FzfR
FxfR
FyfL X
OL O OR OfR FyfR
FxfL OfL
Y
MRx
NRx + FRx
MLy NLy + FLy Z
FyfL xL4 −(NLx + FLx )yL3 X −(NRx + FRx )yR3 FyfR xR4
−FyfL yL4 +(NLy + FLy)xL3 + MLz +(NRy + FRy)xR3 + MRz −FxfR yR4
O Y
2l
(a)
Y
O
2l
(b)
Figure 4: Force analysis diagrams (a) in the 𝑋-𝑌 plane parallel to the track level plane and (b) in the 𝑍-𝑌 plane parallel to the track level
plane.
+ (−𝐹𝑦𝑓𝐿 𝑧𝐿4 + 𝐹𝑧𝑓𝐿 𝑦𝐿4 ) 𝛿 (𝑦 − 𝑦𝐿4 ) The boundary conditions of (1) and (2) are given as
Table 1
Physical parameter Value Unit Notation
𝑀𝑐 33.766 × 103 kg Car-body mass
𝑀𝑏𝑖 2.4 × 103 kg The 𝑖th bogie mass
𝑀𝑤𝑖 1.85 × 103 kg The 𝑖th wheelset mass
𝐶𝑡𝑦 20 × 103 Ns/m The equivalent lateral damping of secondary suspension (considering
damping of lateral shock absorber joint)
The equivalent lateral stiffness of secondary suspension (considering
𝐾𝑡𝑦 18.13 × 106 N/m stiffness of lateral shock absorber joint and lateral stiffness of air
spring)
𝐶𝑡𝑧 40 × 103 Ns/m The equivalent vertical damping of secondary suspension
(considering vertical damping of air spring)
𝐾𝑡𝑧 299 × 103 N/m The equivalent vertical stiffness of secondary suspension (considering
vertical stiffness of air spring)
𝐶𝑓𝑦 0 Ns/m The equivalent lateral damping of primary suspension
𝐾𝑓𝑦 6.47 × 10 6
N/m The equivalent lateral stiffness of primary suspension (considering
the lateral stiffness locating node of the axle-box rotary arm)
𝐶𝑓𝑧 15 × 103 Ns/m The equivalent vertical damping of primary suspension (considering
damping of vertical shock absorber joint)
𝐾𝑓𝑧 6.076 × 106 N/m The equivalent vertical stiffness of primary suspension (considering
stiffness of vertical shock absorber joint and steel spring)
𝑀𝑟 60.64 kg/m Rail mass per unit length
𝑀𝑠 349 kg Mass of sleeper
𝑀𝑏 466 kg Mass of ballast element
𝐿𝑠 0.6 m Sleeper bay
𝐸 2.06 × 1011 N/m2 Young’s modulus
𝐾𝑝𝐿𝑖 20 × 106 N/m Lateral stiffness of the 𝑖th pad
𝐶𝑝𝐿𝑖 5 × 104 N/m Lateral damping of the 𝑖th pad
𝐾𝑝𝑉𝑖 40 × 106 N/m Vertical stiffness of the 𝑖th pad
𝐶𝑝𝑉𝑖 5 × 104 N/m Vertical damping of the 𝑖th pad
𝐾𝑏V(𝐿,𝑅)𝑖 80 × 106 N/m Vertical stiffness between sleeper and the ith ballast element
𝐶𝑏V(𝐿,𝑅)𝑖 1 × 105 Ns/m Vertical damping between sleeper and the ith ballast element
𝐾𝑤 78 × 106 N/m Vertical stiffness between sleeper and the ith ballast element
𝐶𝑤 8 × 104 Ns/m Vertical damping between sleeper and the ith ballast element
𝐾𝑓V(𝐿,𝑅)𝑖 65 × 106 N/m Vertical stiffness between road bed and the ith ballast element
𝐶𝑓V(𝐿,𝑅)𝑖 31000 N/m Vertical damping between road bed and the ith ballast element
where “ ” means one-order partial derivative with respect to 𝑃𝑥𝑘 and 𝑃𝑧𝑘 are, respectively, given as
𝑦 coordinate.
Based on the modal superposition principle, the solutions
to (1) and (2) are, respectively, given as 𝑃𝑥𝑘 = (𝑁𝐿𝑥 + 𝐹𝐿𝑥 ) 𝑋𝑘 (𝑦𝐿3 ) + 𝐹𝑥𝑓𝐿 𝑋𝑘 (𝑦𝐿4 )
Using the normalized modal functions which are mutu- × 𝑋𝑘 (𝑦𝑅3 )
ally orthogonal, we have that
+ (−𝐹𝑥𝑓𝐿 𝑦𝐿4 + 𝐹𝑦𝑓𝐿 𝑥𝐿4 ) 𝑋𝑘 (𝑦𝐿4 )
𝑃𝑧𝑘 = − (𝑁𝐿𝑧 + 𝐹𝐿𝑧 ) 𝑍𝑘 (𝑦𝐿3 ) + 𝐹𝑧𝑓𝐿 𝑍𝑘 (𝑦𝐿4 ) The lateral primary suspension forces on the four wheel-
sets read as
− (𝑁𝑅𝑧 + 𝐹𝑅𝑧 ) 𝑍𝑘 (𝑦𝑅3 ) + 𝐹𝑧𝑓𝑅 𝑍𝑘 (𝑦𝑅4 ) + 𝐺𝑍𝑘 (0)
𝐹𝑦𝑓(𝐿,𝑅)𝑖
× (− (𝑁𝐿𝑦 + 𝐹𝐿𝑦 ) 𝑧𝐿3 + (𝑁𝐿𝑧 + 𝐹𝐿𝑧 ) 𝑦𝐿3 + 𝑀𝐿𝑥 )
= 𝐾𝑓𝑦 (𝑢𝑦𝑖 (∓𝑙𝑤 , 𝑡) − 𝑌𝑏 (𝑡) + 𝐻𝑏𝑤 𝜙𝑏 (𝑡) + (−1)𝑖 𝑙𝑏 𝜓𝑏 (𝑡))
× 𝑍𝑘 (𝑦𝐿3 )
̇ ̇ 𝑖
𝑦𝑖 (∓𝑙𝑤 , 𝑡) − 𝑌𝑏 (𝑡) + 𝐻𝑏𝑤 𝜙𝑏 (𝑡) + (−1) 𝑙𝑏 𝜓̇
+ 𝐶𝑓𝑦 (𝑢̇ 𝑏 (𝑡)) .
+ (− (𝑁𝑅𝑦 + 𝐹𝑅𝑦 ) 𝑧𝑅3 + (𝑁𝑅𝑧 + 𝐹𝑅𝑧 ) 𝑦𝑅3 + 𝑀𝑅𝑥 ) (14)
𝐹𝑧𝑓(𝐿,𝑅)𝑖 NMR
̇(𝑡))
+ ∑ 𝑍𝑅𝑘 (∓𝑙𝑤 ) 𝑞𝑅𝑧𝑘𝑖
= 𝐾𝑓𝑧 (𝑍𝑏 (𝑡) − 𝑧𝑖 (∓𝑙𝑤 , 𝑡) + (−1)𝑖 𝑙𝑏 𝛽𝑏 (𝑡) ∓ 𝑙𝑏 𝜙𝑏 (𝑡)) 𝑘=1
+ 𝐶𝑓𝑧 (𝑍̇ 𝑖 ̇ ̇
𝑏 (𝑡) − 𝑧̇
𝑖 (∓𝑙𝑤 , 𝑡) + (−1) 𝑙𝑏 𝛽𝑏 (𝑡) ∓ 𝑙𝑏 𝜙𝑏 (𝑡)) . + (−1)𝑖 𝑙𝑏 𝛽𝑏̇
(𝑡) ∓ 𝑙𝑏 𝜙𝑏̇
(𝑡) ) .
(13)
Advances in Mechanical Engineering 7
The flexible wheelset model is successfully coupled with indicated in Figure 5. It translates with the wheelset
the bogie through (14) and (15). The calculation formulas of mass center. Its base vector is that e1 = [i1 , j1 , k1 ].
the wheel/rail contact forces are to be discussed below.
(3) 𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) : they are the local coordinate
2.3. Wheel/Rail Rolling Contact Model systems attached to the two rigid wheels, as indicated
in Figure 5. Their origins 𝑂𝐿 and 𝑂𝑅 are at the centers
2.3.1. Description of Coordinate Systems for Wheel/Rail Rolling of the nominal rolling circles of the left and right
Contact Calculation. It is complex and tedious to carry out wheels, respectively, which means that their poses
the rolling contact behavior calculation of a flexible wheelset are determined by the wheel poses and their 𝑌(𝐿,𝑅)
running on a pair of rails in service. The calculation is carried axes are always perpendicular to the corresponding
out with the aid of the transformations of multicoordinate nominal rolling circle planes. The nominal rolling
systems. A wheel/rail rolling contact model considering circles of the left and right wheels coincide with the
flexible wheelset effect is much different from that without planes 𝑂𝐿 -𝑋𝐿 𝑍𝐿 and 𝑂𝑅 -𝑋𝑅 𝑍𝑅 , respectively. But the
considering flexible wheelset effect. rolling motions of the left and right wheels are around
Figure 5 indicates a detailed contact geometry description the axes 𝑋𝐿 and 𝑋𝑅 , respectively, and their yawing
of a flexible wheelset in rolling contact with a pair of rails, the motions are around the axes 𝑍𝐿 and 𝑍𝑅 , respectively.
global coordinate system, and the local coordinate systems So the axes 𝑋𝐿 and 𝑋𝑅 are always parallel to the
used in the rolling contact calculation. In Figure 4, the rolling horizontal plane, and the axes 𝑍𝐿 and 𝑍𝑅 are always
angles, 𝜙𝑤(𝐿,𝑅) , are, actually, the projections of the rotation parallel to the vertical plane. Their base vectors are
angles of the normal line of the axle cross-section at the wheel that e(𝐿,𝑅) = [i(𝐿,𝑅) , j(𝐿,𝑅) , k(𝐿,𝑅) ]. The unit vectors i(⋅) ,
nominal rolling circle centers in the 𝑂-𝑍𝑌 plane, but the j(⋅) , and k(⋅) indicate the directions of the axes 𝑋(⋅) , 𝑌(⋅) ,
yawing angles 𝜓𝑤(𝐿,𝑅) of the wheels, which are the projections and 𝑍(⋅) , respectively.
in the 𝑂-𝑋𝑌 plane, are not shown in the figure and can be
described in the top view of Figure 5 that is omitted here. (4) 𝑂(𝐿,𝑅)1 -𝑋(𝐿,𝑅)1 𝑌(𝐿,𝑅)1 𝑍(𝐿,𝑅)1 : their axes 𝑌𝐿1 and 𝑌𝑅1 ,
The left and right contact points are regarded as, respec- respectively, coincide with the lines 𝐴 𝐿 𝑂𝐿1 and
tively, the origins 𝑂𝐿1 and 𝑂𝑅1 of the local coordinates 𝑂(𝐿,𝑅)1 - 𝐴 𝑅 𝑂𝑅1 , as indicated in Figure 5. The planes 𝑂𝐿1 -
𝑋(𝐿,𝑅)1 𝑌(𝐿,𝑅)1 𝑍(𝐿,𝑅)1 . 𝑜𝐿 and 𝑜𝑅 are the centers of the left and 𝑋𝐿1 𝑌𝐿1 and 𝑂𝑅1 -𝑋𝑅1 𝑌𝑅1 coincide with the common
right instant rolling circles, respectively. The two auxiliary tangent planes of the contact surfaces of the left
lines are created, and they are, respectively, perpendicular to and right wheels/rails, respectively. The axes 𝑍𝐿1 and
the left and right instant circles and pass through the centers 𝑍𝑅1 coincide with the common normal lines of the
𝑜𝐿 and 𝑜𝑅 . These two auxiliary lines, respectively, intersect contact surfaces of the left and right wheels/rails,
with the common planes of the wheel/rail contact surfaces, respectively. Their base vectors are that e(𝐿,𝑅)1 =
and points 𝐴 𝐿 and 𝐴 𝑅 are, respectively, their intersection [i(𝐿,𝑅)1 , j(𝐿,𝑅)1 , k(𝐿,𝑅)1 ].
points, as indicated in Figure 5. Making a line connects point
𝐴 𝐿 to the left contact point 𝑂𝐿1 and the other line connects
(5) 𝑂𝑟(𝐿,𝑅) -𝑋𝑟(𝐿,𝑅) 𝑌𝑟(𝐿,𝑅) 𝑍𝑟(𝐿,𝑅) : these two coordinate sys-
point 𝐴 𝑅 to the left contact point 𝑂𝑅1 . The included angle
tems are used to describe the configurations and
between line 𝐴 𝐿 𝑂𝐿1 and line 𝐴 𝐿 𝑜𝐿 and the one between line
motions of the left and right rail cross-sections which
𝐴 𝑅 𝑂𝑅1 , and line 𝐴 𝑅 𝑜𝑅 are exactly the contact angles 𝛿𝐿 and are assumed to be rigid. Their origins are fixed at the
𝛿𝑅 , respectively. Each coordinate system in Figures 2 and 4 left and right rail cross-session centers, 𝑂𝑟𝐿 and 𝑂𝑟𝑅 ,
has the following detailed description and requirements. and their rotation and translation are determined by
(1) 𝑂 − 𝑋𝑌𝑍: this coordinate system is illustrated in those of the rail cross-sections. Their base vectors are,
Section 2.2. Its base vector is that e = [i, j, k]. Its unit that, e𝑟(𝐿,𝑅) = [i𝑟(𝐿,𝑅) , j𝑟(𝐿,𝑅) , k𝑟(𝐿,𝑅) ].
vectors, i, j, and k are, respectively, in the longitudinal,
vertical and lateral directions of the tangent track. The transformations from the base vector of coordinate
(2) 𝑂1 -𝑋1 𝑌1 𝑍1 : it is a translating coordinate system that system 𝑂1 -𝑋1 𝑌1 𝑍1 to those of the coordinate systems 𝑂(𝐿,𝑅) -
is always parallel to the coordinate system 𝑂-𝑋𝑌𝑍, as 𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) read as
XL X1 XR
Y1 𝜙wR AR
𝜙wL
O1 YR
OL OR O
OL YL R
AL X XR1 𝛿rR
ZL Z1 ZR
𝛿L XL1 Y YR1
RL RR
O
OL1 OR1
Z
ZL1 ZR1 XrR
YL1
XrL
OrR
OrL YrR
YrL
ZrL ZrR
𝜙0
𝑇 𝑇
𝑂1 of 𝑂1 -𝑋1 𝑌1 𝑍1 to the point 𝑂. [𝑥𝑂(𝐿,𝑅) , 𝑦𝑂(𝐿,𝑅) , 𝑧𝑂(𝐿,𝑅) ] are 𝑥𝑂(𝐿,𝑅) { i } 𝑥(𝐿,𝑅) {i}
the coordinates of the nominal rolling circle centers 𝑂(𝐿,𝑅) = −[𝑦𝑂(𝐿,𝑅) ] { j } + [𝑦(𝐿,𝑅) ] A𝑤(𝐿,𝑅) { j }
in the coordinate system 𝑂-𝑋𝑌𝑍. It is noted that in (17) [ 𝑧𝑂(𝐿,𝑅) ] {k} [ 𝑧(𝐿,𝑅) ] {k}
the bases [i, j, k] are the same as [i1 , j1 , k1 ], as indicated
𝑇
in Figure 4. The transformation in (17) includes two times 𝑥 {i}
of translations and one rotation of the coordinate systems = [ 𝑦] { j } .
𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) . Consider the following: [ 𝑧 ] {k}
𝑇 (17)
𝑥(𝐿,𝑅) { i(𝐿,𝑅) }
[𝑦(𝐿,𝑅) ]
{ j(𝐿,𝑅) }
[ (𝐿,𝑅) ] {k(𝐿,𝑅) }
𝑧 The transformations from the base vectors of the local
coordinate systems 𝑂(𝐿,𝑅)1 -𝑋(𝐿,𝑅)1 𝑌(𝐿,𝑅)1 𝑍(𝐿,𝑅)1 at the contact
𝑇
𝑥(𝐿,𝑅) { i1 } → points to the absolute coordinate system read as
→
= 𝑂(𝐿,𝑅) 𝑂1 + [𝑦(𝐿,𝑅) ] A𝑤(𝐿,𝑅) { j1 } + 𝑂1 𝑂
[ 𝑧(𝐿,𝑅) ] {k 1 }
{ i(𝐿,𝑅)1 } {i}
𝑥(𝐿,𝑅)
𝑇 j
{ (𝐿,𝑅)1 } = A (𝐿,𝑅)1 { j } , (18)
→ { i1 }
= 𝑂(𝐿,𝑅) 𝑂 + [𝑦(𝐿,𝑅) ] A𝑤(𝐿,𝑅) { j1 } k
{ (𝐿,𝑅)1 } {k}
[ 𝑧(𝐿,𝑅) ] {k 1 }
where
The transformations from the bases of the local coordi- cross-section centers to those of the coordinate system 𝑂1 -
nate systems 𝑂𝑟𝐿 -𝑋𝑟𝐿 𝑌𝑟𝐿 𝑍𝑟𝐿 and 𝑂𝑟𝑅 -𝑋𝑟𝑅 𝑌𝑟𝑅 𝑍𝑟𝑅 at the rail 𝑋1 𝑌1 𝑍1 (or 𝑂-𝑋𝑌𝑍) are given as
Advances in Mechanical Engineering 9
{ i𝑟(𝐿,𝑅) }
{ j𝑟(𝐿,𝑅) } The instant rolling circle
Y3
{k𝑟(𝐿,𝑅) } dB
X3 O3R
1 0 0 B
= [0 cos (𝜙𝑟(𝐿,𝑅) + 𝜙0 ) sin (𝜙𝑟(𝐿,𝑅) + 𝜙0 ) ] (20) A
𝛿R
[0 − sin (𝜙𝑟(𝐿,𝑅) + 𝜙0 ) cos (𝜙𝑟(𝐿,𝑅) + 𝜙0 )] O3 𝜙
{ i1 } { i1 } {i} Z3
× { j1 } = A𝑟(𝐿,𝑅) { j1 } = A𝑟(𝐿,𝑅) { j } . X
CR
{k 1 } {k 1 } {k} Y
𝜓
The wheel/rail rolling contact calculation consists of three O
parts, wheel/rail contact points, wheel/rail normal forces, and O3
wheel/rail tangential forces. They are elaborated as follows. rigid
Yow CR
A
2.3.2. Wheel/Rail Contact Point Calculation. Before the cal-
culation of the wheel/rail contact points, the given discrete B
data points of the profiles of the wheel treads are usually Z
described in the local coordinate system associated with
the wheel tread, and the rail top profile described in the Figure 6: Schematic diagram of wheel/rail contact spatial geometry.
local coordinate system associated with the rail [19]. In the
calculation, they have to be transformed into the expression
in the absolute coordinate system 𝑂-𝑋𝑌𝑍 through a series
𝑂3𝑅 is the intersection point of the wheelset axle centerline
of the coordinate system transformations which include
and the common tangent line of the contact surfaces, 𝑑𝐵 is
rotating and translating the coordinate system.
the distance between the point 𝑂3 and the rolling circle, 𝐵
The prescribed profile of the wheel is usually described in
and 𝐵𝐶𝑅 are, respectively, the center and radius of the rolling
the local coordinate system associated with the wheel tread.
circle, 𝐴 is the intersection point of the common normal line
The local coordinate system associated with the wheel tread
of the contact surfaces and the wheelset axle centerline, 𝑂3 -
is, actually, parallel to the rigid wheel coordinate system,
𝑋3 𝑌3 𝑍3 is the coordinate system of the rigid wheelset, and
𝑂𝐿 -𝑋𝐿 𝑌𝐿 𝑍𝐿 , or 𝑂𝑅 -𝑋𝑅 𝑌𝑅 𝑍𝑅 , and their origins are at the
𝑂3 𝐴 𝐵 𝐶𝑅 are, respectively, the projections of 𝑂3 𝐴𝐵𝐶𝑅 on
lowest points of the nominal rolling circles. In order to obtain
the coordinates of the wheel tread profiles described in the the 𝑂-𝑋𝑌 plane, so the line 𝑂3 𝐵 is the projection of the axle
absolute coordinate system 𝑂-𝑋𝑌𝑍, firstly, the coordinates centerline on the 𝑂-𝑋𝑌 plane.
of the profiles of the wheels described in the local coordinate The wheel contact point trace line, on which the contact
systems associated with the wheel treads have a translation point 𝐶𝑅 is, in the absolute coordinate system 𝑂-𝑋𝑌𝑍 can be
that equals the radius of the nominal rolling circles, and calculated with (21) according to [18] as follows:
can be described in 𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) . Then, they are
transformed into the description in the absolute coordinate 𝑥 = 𝑥𝐵 + 𝑙𝑥 𝑅𝑅 tan 𝛿𝑅 ,
system 𝑂-𝑋𝑌𝑍 by using formula (17) and with the help of the
wheelset coordinate system 𝑂1 -𝑋1 𝑌1 𝑍1 . 𝑅𝑅 (𝑙𝑥2 𝑙𝑦 tan 𝛿𝑅 + 𝑙𝑧 𝑚)
The transformation from 𝑂𝑟(𝐿,𝑅) -𝑋𝑟(𝐿,𝑅) 𝑌𝑟(𝐿,𝑅) 𝑍𝑟(𝐿,𝑅) to 𝑦 = 𝑦𝐵 − ,
(1 − 𝑙𝑥2 ) (21)
𝑂-𝑋𝑌𝑍 is through translating and rotating the coordinates of
the rail top profiles described in 𝑂𝑟(𝐿,𝑅) -𝑋𝑟(𝐿,𝑅) 𝑌𝑟(𝐿,𝑅) 𝑍𝑟(𝐿,𝑅) . 𝑅𝑅 (𝑙𝑥2 𝑙𝑧 tan 𝛿𝑅 − 𝑙𝑦 𝑚)
→ 𝑧 = 𝑧𝐵 − ,
The translation is just done along the vector 𝑂(𝐿,𝑅)1 𝑂 or R(𝐿,𝑅) , (1 − 𝑙𝑥2 )
and (20) is the transformation from the base vector of the
coordinate systems 𝑂𝑟(𝐿,𝑅) -𝑋𝑟(𝐿,𝑅) 𝑌𝑟(𝐿,𝑅) 𝑍𝑟(𝐿,𝑅) to that of the
coordinate system 𝑂-𝑋𝑌𝑍. where
The existing models for the calculation of rigid wheel/rail
contact geometrical relationship need to be further improved 𝑙𝑥 = − cos 𝜙 sin 𝜓,
to be applicable to the contact geometry calculation of a
𝑙𝑦 = cos 𝜙 cos 𝜓,
flexible wheelset in contact with a pair of rails. Figure 6
illustrates the spatial contact geometrical relationship of a (22)
𝑙𝑧 = sin 𝜙,
wheel of a rigid wheelset in contact with a rail [19], in which
the contact state on the right side is shown. In Figure 6, 𝐶𝑅
𝑚 = √1 − 𝑙𝑥2 (1 + tan2 𝛿𝑅 ).
is the contact point of the right wheel/rail, 𝑂3 is the center
of the rigid wheelset, the left and right wheels have the same
rolling angle and yawing angle, the wheelset axle centerline (𝑥𝐵 , 𝑦𝐵 , 𝑧𝐵 ) are the coordinates of the instant rolling circle
𝑂3 𝑂3𝑅 is perpendicular to the instant rolling circle plane, center 𝐵 in the absolute coordinate system, and they are given
10 Advances in Mechanical Engineering
as follows: Once the contact circles and the rail top profiles are
rigid described in the absolute coordinate system 𝑂-𝑋𝑌𝑍, it will
𝑥𝐵 = 𝑋𝑜𝑤 + 𝑑𝐵 𝑙𝑥 , be easy to calculate the minimum distances Δ𝑍min 𝐿 (𝑡) and
rigid Δ𝑍min 𝑅 (𝑡) between the wheel treads and rail top surfaces, as
𝑦𝐵 = 𝑌𝑜𝑤 + 𝑑𝐵 𝑙𝑦 , (23)
well as the coordinates of the wheel/rail contact points. The
rigid wheel/rail contact angles can be subsequently calculated with
𝑧𝐵 = 𝑍𝑜𝑤 + 𝑑𝐵 𝑙𝑧 , the known coordinates of the wheel/rail contact points and
the given wheel/rail profiles.
where (𝑋𝑜𝑤 rigid
, 𝑌𝑜𝑤
rigid
, 𝑍𝑜𝑤
rigid
) are described in the coordinate
system 𝑂3 -𝑋3 𝑌3 𝑍3 defined as a rigid wheelset coordinate sys-
tem. So the series of the wheel rolling circles can be deter- 2.3.3. Wheel/Rail Normal Forces. The vertical instant com-
mined with 𝑑𝐵 . pressions of wheel/rail at the contact points, 𝛿𝑍𝐿 (𝑡) and
But, for the contact between a flexible wheelset and a pair 𝛿𝑍𝑅 (𝑡), are written as
of rails, due to the axle deformation, as indicated in Figure 5,
strictly speaking, 𝑑𝐵 is not constant, and the directioncosine
components, (𝑙𝑥 , 𝑙𝑦 , 𝑙𝑧 ), of the central lines of the left and 𝛿𝑍𝐿 (𝑡) = 𝑍𝑜𝑤𝐿 (𝑡) − (Δ𝑍min 𝐿 (𝑡) − Δ𝑍𝐿 (0)) − 𝑍irr𝐿 (𝑡) ,
right wheels should be modified. (𝑙𝑥 , 𝑙𝑦 , 𝑙𝑧 ) in (22) and (23)
should be replaced by the direction cosine components, 𝛿𝑍𝑅 (𝑡) = 𝑍𝑜𝑤𝐿 (𝑡) − (Δ𝑍min 𝑅 (𝑡) − Δ𝑍𝑅 (0)) − 𝑍irr𝑅 (𝑡) .
(𝑙(𝐿,𝑅)𝑥 , 𝑙(𝐿,𝑅)𝑦 , 𝑙(𝐿,𝑅)𝑧 ), of the normal lines at the instant rolling (25)
circle centers in 𝑂-𝑋𝑌𝑍; namely, in the calculation of
(𝑙(𝐿,𝑅)𝑥 , 𝑙(𝐿,𝑅)𝑦 , 𝑙(𝐿,𝑅)𝑧 ), the rolling and yawing angles (𝜙, 𝜓) of
The instant normal compressions, 𝛿𝑍𝐿𝑐 (𝑡) and 𝛿𝑍𝑅𝑐 (𝑡),
the wheelset should be replaced by (𝜙𝑤(𝐿,𝑅) , 𝜓𝑤(𝐿,𝑅) ) of the
are calculated according to (26) and the contact geometrical
left and right wheels of the flexible wheelset. It should be
relationship shown in Figure 6, and read they as
noted that the normal lines through the instant rolling circle
centers are, actually, those through the centers of the nominal
rolling circles since the wheels are assumed to be rigid. For 𝛿𝑍𝐿 (𝑡)
the flexible wheelset, the coordinates of the centers of the left 𝛿𝑍𝐿𝑐 (𝑡) = ,
cos (𝛿𝐿 − 𝜙𝑤𝐿 )
and right instant rolling circles should be written as, based on (26)
formula (23) and Figure 6, 𝛿𝑍𝑅 (𝑡)
𝛿𝑍𝑅𝑐 (𝑡) = .
rigid cos (𝛿𝑅 + 𝜙𝑤𝑅 )
𝑥𝑜(𝐿,𝑅)
= 𝑋𝑜𝑤 + Δ𝑥 + 𝑑𝑂1 𝑜(𝐿,𝑅)
𝑙(𝐿,𝑅)𝑥 ,
𝑦𝑜(𝐿,𝑅)
rigid
= 𝑌𝑜𝑤 + Δ𝑦 + 𝑑𝑂1 𝑜(𝐿,𝑅)
𝑙(𝐿,𝑅)𝑦 , (24) The left and right wheel/rail normal forces 𝐹𝑤𝑟𝑛(𝐿,𝑅) are
calculated by
rigid
𝑧𝑜(𝐿,𝑅)
= 𝑍𝑜𝑤 + Δ𝑧 + 𝑑𝑂1 𝑜(𝐿,𝑅)
𝑙(𝐿,𝑅)𝑧 ,
𝐹𝑤𝑟𝑛(𝐿,𝑅) (𝑡)
where 𝑑𝑂1 𝑜(𝐿,𝑅)
are the distances between the center 𝑂1 of
3/2
the flexible wheelset and the centers 𝑜(𝐿,𝑅) of the left and { 1
right instant rolling circles, respectively. And Δ𝑥, Δ𝑦, and {
{ { [𝛿𝑍(𝐿,𝑅)𝑐 (𝑡) − 𝛿𝑍(𝐿,𝑅)𝑐 (0)]} , 𝛿𝑍(𝐿,𝑅)𝑐 (𝑡)
{
Δ𝑧 are the increments of the wheelset center translation in { 𝐺
{
={ −𝛿𝑍(𝐿,𝑅)𝑐 (0) > 0,
the 𝑋-, 𝑌-, and 𝑍-directions, respectively. They are caused by {
{
{
{ 0, 𝛿𝑍(𝐿,𝑅)𝑐 (𝑡)
the bending deformation of the flexible wheelset. Comparing {
{ −𝛿𝑍(𝐿,𝑅)𝑐 (0) ≤ 0.
(23) with (24), for the flexible wheelset, the calculation of the
coordinates of the wheelset center 𝑂3 and the instant rolling (27)
circle center (𝑥𝐵 , 𝑦𝐵 , 𝑧𝐵 ) considers the combined effect of the
rigid motion and flexible deformation of the wheelset. But Then, the normal force should be decomposed into three
its stretching and compressing deformations caused by the components in the 𝑋-, 𝑌-, and 𝑍-directions in establishing
axle bending deformation in the axle direction are very small, the motion equations of the wheelset and the rails.
compared with the axle bending deformation amplitudes.
So Δ𝑦 is neglected. Since the analyzed wheelset is not a
powered wheelset, the component along the track of the 2.3.4. Wheel/Rail Creep Forces and Spin Moments. The wheel/
wheel/rail force is quite small, compared with the vertical rail creep forces are calculated by using the model developed
component, the bending deformation of the wheelset is very by Shen-Hedrick-Elkins, as described in [18]. In the calcu-
small in a horizontal plane of the track. So Δ𝑥 caused by lation of each time step, the wheel/rail creepages 𝜉𝑥 , 𝜉𝑦 , and
the wheelset bending deformation can also be neglected. 𝜉𝑧 should be calculated in advance. The existing creepage
It should be noted that (𝑋𝑜𝑤 rigid
, 𝑌𝑜𝑤
rigid
, 𝑍𝑜𝑤
rigid
) are obtained calculation formulas need to be improved and can be used to
through the dynamic analysis using the traditional multibody calculate the creepages of a flexible wheelset in rolling contact
dynamic model of the vehicle/track, which is introduced in with a pair of rails. The creepages represent the ratios of the
the present paper. relative velocities of the contact surfaces at the contact points
Advances in Mechanical Engineering 11
⇀
to the forward velocities, V(𝐿,𝑅) , of the instant rolling circles of to time. R𝑂𝑂(𝐿,𝑅) are the vectors 𝑂𝑂(𝐿,𝑅) . Ṙ 𝑂𝑂(𝐿,𝑅) represent
the wheels, as shown in (28). Consider the following: the translational velocities of the centers of the nominal
ΔV(𝐿,𝑅)𝑥 rolling circles of the left and right wheels, described in the
𝜉𝑥(𝐿,𝑅) = , absolute coordinate system. 𝜔(𝐿,𝑅) are determined by the
V(𝐿,𝑅) angular velocity vectors of the left and right wheels, 𝜓̇ 𝑤(𝐿,𝑅)
ΔV(𝐿,𝑅)𝑦 ̇ , as indicated in (32). It is noted that 𝜔(𝐿,𝑅) is
and 𝜙𝑤(𝐿,𝑅)
𝜉𝑦(𝐿,𝑅) = , (28) just the rotation of the transport coordinate systems, 𝑂(𝐿,𝑅) -
V(𝐿,𝑅)
𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) , relative to the absolute coordinate system
Δ𝜔(𝐿,𝑅)𝑧 𝑂-𝑋𝑌𝑍, as shown in Figure 5. Consider the following:
𝜉𝑧(𝐿,𝑅) = ,
V(𝐿,𝑅) ̇ 𝑇
𝜙𝑤(𝐿,𝑅) {i}
where ΔV(𝐿,𝑅)𝑥 , ΔV(𝐿,𝑅)𝑦 , and Δ𝜔(𝐿,𝑅)𝑧 are, respectively, the 𝜔(𝐿,𝑅) = [ 0 ] {j} ,
relative longitudinal velocity, the relative lateral velocity, and [𝜓̇
𝑤(𝐿,𝑅) ] {k }
angular velocities. The relative velocities are firstly calculated 𝑇
(32)
in the absolute coordinate system and then transformed into 𝑅̇
i
] { j(𝐿,𝑅) }
(𝐿,𝑅)𝑥
[𝑅̇
the expressions in the local coordinate systems associated R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 = [ (𝐿,𝑅)𝑦 ] { (𝐿,𝑅) } .
with the contact surfaces, which will be discussed in the ̇
k
following. [𝑅(𝐿,𝑅)𝑧 ] { (𝐿,𝑅) }
V(𝐿,𝑅) can be calculated in two ways. The first way is It should also be noted that 𝜔(𝐿,𝑅) do not include the
to make a summation of the rigid motion and the flexible rolling (𝜔0 ) of the rigid wheelset over the track and neglect
deformation of the wheelset, and the other way is to take the the rolling angular velocities of the wheels due to the twisting
derivatives of the coordinates of the centers of the instant deformation of the wheelset, which produce the longitudinal
rolling circles with respect to time directly. But, actually, creepages. The relative velocities k𝑤𝑟(𝐿,𝑅) are the velocities
the difference between V(𝐿,𝑅) and V is quite small due to the of the left and right contact points described in the local
very small effect of the rigid yaw, motion and the bending coordinate systems 𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) , due to the rolling
deformation of the wheelset on the forward speeds, V(𝐿,𝑅) , of the wheelset over the two rails. They are written as
of the instant rolling circles. So, by using formulas (28), for
simplicity, V(𝐿,𝑅) is replaced with V. V is the speed of the vehicle k𝑤𝑟(𝐿,𝑅) = 𝜔(𝐿,𝑅) × R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 , (33)
or the nominal forward speed of the wheelset center.
Based on the composite motion theorem of rigid body where
motion, the absolute velocity k𝑎 of a particle is the resultant 𝑇
vector of the relative velocity k𝑟 and the transport velocity k𝑒 , 0 { i(𝐿,𝑅) }
given in (29). In other words, k𝑟 is the velocity of the moving 𝜔(𝐿,𝑅) = [−𝜔0 ] { j(𝐿,𝑅) } . (34)
point, described in the moving coordinate system, and k𝑒 is [ 0 ] {k(𝐿,𝑅) }
the velocity of the moving coordinate system relative to the
absolute coordinate system. Consider the following: Equation (34) is expressed in the local coordinate systems
𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) .
k𝑎 = k 𝑒 + k 𝑟 . (29) Substituting (31) and (33) into (30), the equation could be
deduced as follows:
The wheel coordinate systems 𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) , as
shown in Figure 5, can be regarded as the moving coordinate k𝑤𝑎(𝐿,𝑅) = k𝑤𝑒(𝐿,𝑅) + k𝑤𝑟(𝐿,𝑅)
systems. So the absolute velocities of the left and right contact
points 𝑂(𝐿,𝑅)1 , respectively, on the left and right wheels, have = Ṙ
𝑂𝑂(𝐿,𝑅) + 𝜔(𝐿,𝑅) × R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 (35)
the following expressions:
+ 𝜔(𝐿,𝑅) × R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 .
k𝑤𝑎(𝐿,𝑅) = k𝑤𝑒(𝐿,𝑅) + k𝑤𝑟(𝐿,𝑅) , (30)
It should be noted that k𝑤𝑟(𝐿,𝑅) should be transformed into
where k𝑤𝑒(𝐿,𝑅) and k𝑤𝑟(𝐿,𝑅) are, respectively, the transport the expression in absolute coordinate system 𝑂-𝑋𝑌𝑍 through
velocities and the relative velocities. The relative velocities using transformation matrices A𝑤(𝐿,𝑅) , as indicated in (16),
are the velocities of the wheel contact points 𝑂(𝐿,𝑅)1 relative before being added to the calculation.
to the origins 𝑂(𝐿,𝑅) of the moving coordinate systems. The The absolute velocities of the contact points on the left and
transport velocities k𝑤𝑒(𝐿,𝑅) are the velocities of two points in right rails are written as
𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) , and these two points coincide with
𝑇 𝑇
the contact points 𝑂(𝐿,𝑅)1 , respectively. So k𝑤𝑒(𝐿,𝑅) read as 𝑋̇𝑟(𝐿,𝑅) {i} 𝛿𝑋̇ 𝑟(𝐿,𝑅) {i}
krail(𝐿,𝑅) = [ 𝑌𝑟(𝐿,𝑅)
̇] j
{ } + ̇]
[ 𝛿𝑌𝑟(𝐿,𝑅)
{j} , (36)
k𝑤𝑒(𝐿,𝑅) = 𝜔(𝐿,𝑅) × R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 + Ṙ
𝑂𝑂(𝐿,𝑅) , (31) ̇ ̇
[ 𝑟(𝐿,𝑅) ] { } [ 𝑟(𝐿,𝑅) ] {k}
𝑍 k 𝛿 𝑍
⇀
where R𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 indicate the vectors 𝑂(𝐿,𝑅) 𝑂(𝐿,𝑅)1 . Ṙ
𝑂𝑂(𝐿,𝑅) ̇ and 𝑍̇
where 𝑌𝑟(𝐿,𝑅) 𝑟(𝐿,𝑅) are, respectively, the lateral and
indicate the derivatives of the vectors R𝑂𝑂(𝐿,𝑅) with respect vertical velocities of the left and right rail cross-sections and
12 Advances in Mechanical Engineering
̇ and 𝛿𝑍̇
𝛿𝑌𝑟(𝐿,𝑅) 𝑟(𝐿,𝑅) are, respectively, the rates of changes 7(b), and 7(c). This is caused by the excited resonance of first-
of the lateral and vertical irregularities on the left and right order bending mode. An increase in the wheel/rail normal
railheads. force caused by the first-order bending resonance is quite
Using (35) and (36), the velocity differences Δk(𝐿,𝑅) small, compared with the static wheel/rail forces. But such
between the contact surfaces of the wheels and the rails can a small difference of the force applied on wheels at a fixed
be written as frequency could cause a polygonal wear on high-speed wheels
𝑇 in a long-time service. For instance, now the high-speed
ΔV𝑥(𝐿,𝑅) { i } vehicles in service widely use the wheelsets with diameters in
Δk(𝐿,𝑅) = k𝑤𝑎(𝐿,𝑅) − krail(𝐿,𝑅) = [ΔV𝑦(𝐿,𝑅) ] { j } . (37) the range of 840 mm–920 mm, and the operating speed of the
[ΔV𝑧(𝐿,𝑅) ] {k} high-speed trains is 300 km/h. Large amounts of wheel tread
roughness measured from the sites show that the trilateral
The velocity differences Δk(𝐿,𝑅) , are expressed in the absolute wear on the wheels is very common due to the wheelset first-
coordinate system, shown in (37), and then need to be trans- order bending resonance excited probably.
formed into the expressions in the local coordinate systems Figure 8 indicates the PSDs (power spectral densities)
𝑂(𝐿,𝑅)1 -𝑋(𝐿,𝑅)1 𝑌(𝐿,𝑅)1 𝑍(𝐿,𝑅)1 through (18). of the vertical forces, corresponding to those analyzed in
Then, plugging (18) into (37), the velocity differences are time domain in Figure 7. The frequencies of the dominant
𝑇 vibration components (at peaks) in Figure 8 are the excitation
ΔV𝑥(𝐿,𝑅) { i(𝐿,𝑅)1 } frequencies of the harmonic track irregularity, which are,
Δk(𝐿,𝑅) = [ΔV𝑦(𝐿,𝑅) ] A(𝐿,𝑅)1 −1 { j(𝐿,𝑅)1 } respectively, 46 Hz, 69 Hz, 92 Hz and 117 Hz at the speeds of
[ΔV𝑧(𝐿,𝑅) ] {k(𝐿,𝑅)1 } 200 km/h, 300 km/h, 400 km/h, and 500 km/h. This indicates
𝑇
(38) that track irregularity plays a dominant role in the influence
ΔV(𝐿,𝑅)𝑥 i
[ ] { j(𝐿,𝑅)1 } on wheel/rail vertical force. In other words, the influence
= [ΔV(𝐿,𝑅)𝑦 ] { (𝐿,𝑅)1 } . of the wheelset flexibility is not significant if the resonance
k
[ΔV(𝐿,𝑅)𝑧 ] { (𝐿,𝑅)1 } frequency of the first-order bending mode is far away from
the passing irregularity frequency. The lines of DIFF illustrate
Equation (38) is plugged into (28) to calculate the lateral the PSDs of the differences between the vertical forces applied
and longitudinal creepages. The calculation of the spin on the rigid and flexible wheelsets. In Figures 7(a), 7(b), and
creepages by using the third formula in (28) neglects the 7(c), three frequency components dominate the PSDs of the
effect of the wheelset axle bending since its effect on the spin force differences. The first is that 𝑓1 = 31 Hz, which is the nat-
creepages is quite small. ural frequency of the wheel-rail contact system independent
of the operating speed. The second is the passing harmonic
3. Results and Discussions track irregularity frequency, 𝑓2 that is equal to, respectively,
46 Hz, 69 Hz, 92 Hz and 117 Hz, corresponding to the speeds
In order to verify the present model considering the effect of 200 km/h, 300 km/h, 400 km/h and 500 km/h. The third
of the first-order bending modes of the wheelset in the two is the first-order bending resonance frequency, 𝑓3 . At the
planes mentioned above on the rolling contact behavior, a speed of 500 km/h, the peak value, as seen in Figure 8(d), is
harmonic track irregularity with fixed wavelength of 1.18 m is about two orders of magnitude larger than the values at the
used. The excitation frequency of the irregularity is 117 Hz at frequency of 117 Hz at the other three speeds. So the difference
the speed of 500 km/h, which is exactly equal to the resonance in the dynamic behavior of the rigid and flexible wheelsets has
frequency of the wheelset first-order bending mode. The been clearly understood.
fluctuating amplitude of the irregularity is 0.05 mm. The The differences of the lateral wheel/rail forces and their
calculation cases include the speeds of 200 km/h, 300 km/h, PSDs have almost the same features. So the discussion on
400 km/h, and 500 km/h. The parameters used in the calcu- them is omitted in this paper.
lation are listed in Table 1 [20]. The excited wheelset bending mode makes a contribution
In Figure 7, the right vertical axis denotes the difference to the wheel/rail interaction to some extent, as known from
(DIFF) between the normal forces calculated by the models of the above analysis on the wheel/rail forces. The first-order
the rigid (RW) and flexible (EW) wheelsets. The lines of DIFF bending resonance causes the oscillations of the wheel/rail
in Figure 7 are obtained by subtracting the lines of RW from contact points and the relative velocity of the wheel/rail
the corresponding lines of EW. The numerical computation contact surfaces. Figure 9 shows the changes of contact points
benefits from fast double precision. Figure 7 shows that the on the rigid and flexible wheelsets with the travelling distance.
amplitudes of the total vertical wheel/rail forces increase with In Figure 8, the vertical axis represents the coordinates of the
the speed increase, as indicated by the red and blue broken wheel contact points in the direction of 𝑌(𝐿,𝑅) axis of the local
lines. The amplitude of the normal force difference denoted coordinate systems 𝑂(𝐿,𝑅) -𝑋(𝐿,𝑅) 𝑌(𝐿,𝑅) 𝑍(𝐿,𝑅) . The black heavy
by DIFF, that is, the normal force variation caused by the first- lines indicate the traces of the contact points on the rigid
order wheelset bending deformation, is about 0.2 N when the wheelset and the fine red lines denote those on the flexile
speed is less than 500 km/h, as shown in Figures 7(a), 7(b), wheelset, which are, respectively, indicated by RW and EW.
and 7(c). In the case of 500 km/h, as shown in Figure 7(d), The contact points on the rigid wheelset are closer to the
the maximum amplitude of the difference reaches about 1.2 N, nominal rolling circles, and the contact points on the flexible
which is 6 times larger than those of the cases of Figures 7(a), wheelset move toward their corresponding field sides due to
Advances in Mechanical Engineering 13
×10−4 ×10−4
65 8 65 8
45 0 45 0
×10−4 ×10−4
65 8 65 20
1.18 m (92 Hz) 1.18 m (117 Hz)
Vertical force difference (kN)
55 4 55 10
Dominant vibrations caused by a Dominant vibrations caused by a
50 sine wave irregularity 0.95 m(117 Hz) 2 50 sine wave irregularity 5
45 0 45 0
RW RW
EW EW
DIFF DIFF
(c) 400 km/h (d) 500 km/h
Figure 7: Wheel/rail vertical forces at the speeds of (a) 200 km/h, (b) 300 km/h, (c) 400 km/h, and (d) 500 km/h. RW and EW denote the
cases of the rigid wheelset and the elastic wheelset, respectively, and DIFF illustrates the difference between the analysis results of the rigid
and the elastic wheelsets.
the bending deformation. The first-order bending resonance It is very interesting that the contact point traces of
of the wheelset in the plane perpendicular to the track level the rigid wheelset do not present the obvious oscillation at
causes the opposite rolling angles of the left and right wheels, the passing frequency of the harmonic regularity excitation,
which are nearly symmetrical about the wheelset centre. This as indicated by the black lines of RW. However, the RW
mode shape is similar to the downward bending deformation lines show the position oscillations of the contact points
of the wheelset axis, as shown in Figure 3. With the increase on the rigid wheelset at a low-frequency about 22 Hz. The
of the speed the passing irregularity frequency gradually frequency is probably the natural frequency of the bogie in
approaches the wheelset resonance frequency of the first- the lateral direction. The resonance of the bogie is caused by
order bending mode, so that the trace of the contact points the harmonic track irregularity. When the speed is 400 km/h,
oscillates more and more seriously. When the vehicle speeds the low frequency oscillation reaches the largest, as indicated
up to 500 km/h, as displayed in Figure 9(d), the first-order in Figure 9(c), and when the speed is 500 km/h, the oscil-
bending resonance of the wheelset occurs and the left and lation disappears, as indicated by the coarse black lines in
right contact points on the elastic wheelset fiercely fluctuate Figure 9(d). The further discussion on dynamic behavior of
around the traces of the contact points on the rigid wheelset. the bogie does not belong to the task of this paper.
The fluctuating amplitude reaches 0.1 𝜇m, and is five times The above analysis shows the evident impact of the
larger than those at the speeds of 200 km/h 300 km/h and wheelset bending deformation on the contact points. The
400 km/h. wheel/rail contact forces rely on not only the contact point
14 Advances in Mechanical Engineering
10000 10000
PSD of vertical force (kN2 /Hz) 100 100
Figure 8: The PSDs of the vertical wheel/rail forces at the speeds of (a) 200 km/h, (b) 300 km/h, (c) 400 km/h, and (d) 500 km/h: 𝑓1 = 31 Hz,
𝑓2 = 46 Hz in (a), 𝑓2 = 69 Hz in (b), 𝑓2 = 92 Hz in (c), 𝑓2 = 117 Hz in (d), and 𝑓3 = 117 Hz.
positions but also the relative speeds of the contact surfaces independent of the operation speed. In Figures 11(a) and 11(b),
represented by the creepages. There is a small difference 𝑓𝑠 is the passing frequency of the sleepers and 2𝑓𝑠 is the
between the creepages calculated with the models of the twice passing frequency of the sleepers or called harmonic
elastic and rigid wheelsets, as indicated in Figure 10. frequency due to the excitation of the multisleepers with an
As well known, rail corrugation and wheel polygonization equal pitch between the neighboring sleepers. In Figures 11(c)
are mainly related to the sensitive resonances of the wheel/rail and 11(d), 𝑓𝑚1 denotes the bending resonance frequency of
system [1, 21, 22]. In order to verify that the first-order the wheelset. 𝑓𝑚1 equals 117 Hz, and it is evident that the
bending resonance of the wheelset is active and find the resonance frequency is independent of the vehicle speed. In
sensitive resonance frequencies of the wheel/rail systems, all the figures of Figure 11, 𝑓𝑡 always appears at about 380 Hz.
the present model of the high-speed vehicle coupled with It is one of the resonance frequencies of the system consisting
the track is used to analyze the dynamic behaviour of the of the track and the bogie of the vehicle. It will be further
wheel/rail systems under a white-noise excitation and at analyzed in the following.
different speeds. The analyzed frequency range is from 0 Carefully making a comparison of Figures 11(a), 11(b),
to 500 Hz and the operation speed is from 200 km/h to 11(c), and 11(d) shows the interesting results, in which one
400 km/h. Figure 11 shows the wheel/rail force responses ana- should carefully pay attention to the vertical coordinate scales
lyzed in frequency-domain. Figures 11(a) and 11(b) illustrate of the figures. Under the present excitation of the white-noise,
the linear spectra of the lateral and vertical wheel/rail forces the wheel/rail force response has high peaks at the passing
of the rigid wheelset, respectively, and Figures 11(c) and 11(d) frequencies of the sleepers, or the wheel/rail force response
indicate the corresponding results of the flexible wheelset. to the excitation of the periodically discrete sleepers is clearly
In Figure 11(b), 𝑓0 indicates the resonance frequency of the characterized by using the existing model of the coupling
wheel/rail system, as mentioned above. It is about 31 Hz, and vehicle/track without considering the effect of the first-order
Advances in Mechanical Engineering 15
EW EW
RW RW
(c) 400 km/h (d) 500 km/h
Figure 9: Lateral coordinates of contact points on wheel treads in associated coordinate systems at the speeds and of (a) 200 km/h, (b)
300 km/h, (c) 400 km/h, and (d) 500 km/h.
)
250
/h
)
300
/h
m
300
m
(k
350
(k
350
ed
ed
400
e
400
Sp
e
Sp
0 100 200 300 400 500 0 100 200 300 400 500
Frequency (Hz) Frequency (Hz)
(a) Rigid wheelset (b) Rigid wheelset
250
)
/h
/h
m
300 300
m
(k
(k
350 350
ed
ed
400 400
e
e
Sp
Sp
0 100 200 300 400 500 0 100 200 300 400 500
Frequency (Hz) Frequency (Hz)
(c) Flexible wheelset (d) Flexible wheelset
Figure 11: Wheel/rail force amplitudes in frequency domain at different speeds and under a white-noise roughness excitation: (a) the lateral
forces on the rigid wheelset; (b) the vertical forces on the rigid wheelset; (c) the lateral forces on the flexible wheelset; and (d) the vertical
forces on the flexible wheelset.
374 Hz 390 Hz
Mx
Sleeper Sleeper
Figure 12: Track mode shapes with resonance frequencies near 380 Hz.
for the track with a bogie is developed and used to calculate 4. Conclusions
the resonance frequencies and the corresponding mode
shapes of them. In Figure 12, the two figures indicate the This paper presents a new vehicle/track coupling dynam-
mode shapes of the track with a bogie, the correspond- ics model considering the effect of the first-order bend-
ing frequencies of which are, respectively, 374 Hz and 390 Hz. ing resonance of the flexible wheelset. A new wheel/rail
They are close to 380 Hz. contact geometry model is derived in detail and considers
Advances in Mechanical Engineering 17
the interaction of a flexible wheelset and a pair of rails. So the 𝑁𝑅𝑥 , 𝑁𝑅𝑦 , 𝑁𝑅𝑧 : The components in 𝑋-, 𝑌-, and
existing coupling dynamics model of railway vehicle/track 𝑍-directions of wheel/rail normal
is further improved. In order to verify the new model and force on the right wheel
clarify the influence of the first-order bending resonance of 𝐹𝐿𝑥 , 𝐹𝐿𝑦 , 𝐹𝐿𝑧 : The components in 𝑋-, 𝑌-, and
the flexible wheelset on the wheel/rail contact behavior, the 𝑍-directions of wheel/rail tangent
new and existing models are used to analyze the difference of force on the left wheel
the wheel/rail contact behavior under the excitation of a har- 𝐹𝑅𝑥 , 𝐹𝑅𝑦 , 𝐹𝑅𝑧 : The components in 𝑋-, 𝑌-, and
monic track irregularity with a fixed wavelength and a white- 𝑍-directions of wheel/rail tangent
noise at different operating speeds. The results calculated by force on the right wheel
the two models are compared to understand the effect of the 𝐹𝑥𝑓𝐿 , 𝐹𝑦𝑓𝐿 , 𝐹𝑧𝑓𝐿 : The components in 𝑋-, 𝑌-, and 𝑍-
wheelset first-order bending on the rolling contact behavior directions of primary suspension
of the wheel/rail system. The conclusions are given as follows. forces on the left sides of a wheelset
𝐹𝑥𝑓𝑅 , 𝐹𝑦𝑓𝑅 , 𝐹𝑧𝑓𝑅 : The components in 𝑋-, 𝑌-, and
(1) In the analysis cases that indicate the different oper- 𝑍-directions of primary suspen-
ating speeds and a harmonic irregularity and white- sion forces on the right sides of a
noise excitations, the first-order bending mode of the wheelset
wheelset in service can be excited easily. When the 𝑡: Time
passing irregularity frequency equals the natural fre- 𝑥: The displacement of the bending
quency of the first-order bending mode of wheelset, deformation in the directions of 𝑋-
the resonance of the wheel/rail systems occurs. In axis
such a situation, the wheelset bending has a great 𝑧: The displacement of the bending
influence on the wheel/rail rolling contact behavior. deformation in the directions of 𝑍-
(2) When the first-order bending mode in the verti- axis
cal plane is excited at the speed of 500 km/h, the 𝐸: Young’s modulus of the wheelset
wheel/rail forces and contact points oscillate more material
seriously than those at the other speeds at the first- 𝜌: The wheelset density
order bending resonance frequency. Besides, the res- 𝐴: The equivalent cross-sectional area
onance easily occurs between the flexibility wheelset of the axle
and the rails when the passing frequency of the 𝐺: The gravity of the wheelset
track irregularity excitation is equal or close to the 𝐼𝑧 , 𝐼𝑥 : The inertia moments of the cross-
wheelset bending resonance frequency at any speed. section with respect to the 𝑍-axis
Though the numeric differences between the dynamic and the 𝑋-axis, respectively
responses of the rigid and flexible wheelsets are small, 𝑄𝑥 , 𝑄𝑧 : The concentrated forces on the
they could contribute to the development of the wheelset axle in the 𝑋- and 𝑍-
polygonal wear on the wheel treads after a long-time directions
service of the wheelset. 𝑀𝑧 , 𝑀𝑥 : The concentrated moments on the
wheelset axle, respectively, in the
(3) Developing the coupling dynamics model of railway negative 𝑍-axis direction and the
vehicle and track considering the effect of structural 𝑋-axis direction
flexibility of the key parts is a complex subject. The (𝑥(𝐿,𝑅)3 , 𝑦(𝐿,𝑅)3 , 𝑧(𝐿,𝑅)3 ): The coordinates describing the dis-
present work is just at a preliminary stage. The next tances from the wheelset axle to the
step will be to consider modeling the high-order contact points of the left and right
bending resonances of wheelset, wheel flexibility and wheels/rails, respectively
powered wheelset flexibility in the wheel/rail rolling (𝑥(𝐿,𝑅)4 , 𝑦(𝐿,𝑅)4 , 𝑧(𝐿,𝑅)4 ): The coordinates describing the dis-
contact model. tances from the axle to the left and
right primary suspensions, respec-
tively
Nomenclature 𝑋𝑘 , 𝑍𝑘 : The 𝑘th normalized modal func-
𝑖 (=1–4): The number of wheelsets tions describing the bending defor-
𝑗 (=1-2): The number of bogies mations of the axle in the planes
𝑀𝐿𝑥 , 𝑀𝐿𝑧 : The components in 𝑋- and parallel and perpendicular to the
𝑍-directions of the spin moment track level plane, respectively
on the left wheel 𝑞𝑥𝑘 , 𝑞𝑧𝑘 : The 𝑘th generalized coordinates of
𝑀𝑅𝑥 , 𝑀𝑅𝑧 : The components in 𝑋- and the bending deformations in the
𝑍-directions of the spin moment directions of the 𝑋-axis and the 𝑍-
on the right wheel axis, respectively
𝑁𝐿𝑥 , 𝑁𝐿𝑦 , 𝑁𝐿𝑧 : Components in 𝑋-, 𝑌-, and ̈, 𝑞𝑧𝑘
𝑞𝑥𝑘 ̈: The second-order derivatives of
𝑍-directions of wheel/rail normal 𝑞𝑥𝑘 (𝑡) and 𝑞𝑧𝑘 (𝑡) with respect to
force on left wheel time 𝑡
18 Advances in Mechanical Engineering
𝑞𝐸𝑥𝑘 (𝑡), 𝑞𝐸𝑧𝑘 (𝑡): The generalized coordinates corre- (𝑋𝑜𝑤 rigid
, 𝑌𝑜𝑤
rigid
, 𝑍𝑜𝑤
rigid
): The coordinates of a rigid wheelset
sponding to the flexible modes center 𝑂3
𝑞𝑅𝑥𝑘 (𝑡), 𝑞𝑅𝑧𝑘 (𝑡): The generalized coordinates corre- 𝛿𝑍(𝐿,𝑅) : The relative vertical displacements
sponding to the rigid modes of the left and right wheels of the
𝜆 𝑥𝑘 , 𝜆 𝑧𝑘 : The 𝑘th circular frequencies flexible wheelset, respectively
𝑋𝐸𝑘 (𝑦), 𝑍𝐸𝑘 (𝑦): The flexible modal functions nor- 𝑍(𝐿,𝑅) (0): The initial vertical compressions of
malized in the directions of the 𝑋- the left and right wheels at 𝑡 = 0
axis and the 𝑍-axis, respectively 𝑍𝑜𝑤(𝐿,𝑅) : The vertical displacements of the
𝑋𝑅𝑘 (𝑦), 𝑍𝑅𝑘 (𝑦): The rigid modal functions normal- left and right wheels caused by
ized in the directions of the 𝑋-axis the wheelset bending deformation
and the 𝑍-axis, respectively occurred in the transverse section
NM: The total number of the modes in of the track
the calculation 𝑍irr(𝐿,𝑅) : The vertical irregularities of the left
NMR: The total number of the rigid modal and right rails
functions 𝛿𝑍(𝐿,𝑅)𝑐 : The left and right wheel/rail normal
NME: The total number of the flexible compressions
modal functions 𝐹wrn(𝐿,𝑅) : The left and right wheel/rail normal
𝑙𝑤 : The half distance between the left forces
and right action points of the 𝜉𝑥 , 𝜉𝑦 , 𝜉𝑧 : The longitudinal, lateral, and spin
primary suspension forces on the creepages
wheelset V(𝐿,𝑅) : Forward velocities for left and right
𝑙𝑏 : The half distance between the left wheels
and right action points of the ΔV(𝐿,𝑅)𝑥 , ΔV(𝐿,𝑅)𝑦 : The relative longitudinal and lateral
primary suspension forces on the velocities
bogie Δ𝜔(𝐿,𝑅)𝑧 : The relative angular velocities
𝑙axle : The half of the length of the wheel- k𝑎 : The absolute velocity
set axle k𝑟 : A relative velocity
𝜓𝑏 : The yawing angle of the bogie k𝑒 : A transport velocity
𝜙𝑏 : The pitching angles of the bogie k𝑤𝑒(𝐿,𝑅) : The transport velocity used in
𝛽𝑏 : The rolling angles of the bogie creepage calculation of the flexible
𝐾𝑓𝑥 , 𝐾𝑓𝑦 , 𝐾𝑓𝑧 : The primary suspension stiffness in wheelset
𝑋-, 𝑌-, and 𝑍-directions, respec- k𝑤𝑎(𝐿,𝑅) : The relative velocities used in
tively creepage calculation of the flexible
𝐶𝑓𝑥 , 𝐶𝑓𝑦 , 𝐶𝑓𝑧 : The primary suspension damping wheelset
in 𝑋-, 𝑌-, and 𝑍-directions, respec- 𝜔(𝐿,𝑅) : The angular velocity vectors of the
tively left and right wheels
𝐻𝑏𝑤 : The vertical distance between the Δk(𝐿,𝑅) : The velocity differences of the con-
mass centers of the bogie and the tact points on left and right rail and
wheelset wheel, respectively.
𝛿(𝐿,𝑅) : The left and right wheel/rail contact
angles
𝜙𝑤(𝐿,𝑅) : The rolling angles of the left and Acknowledgments
right wheels
𝜓𝑤(𝐿,𝑅) : The yawing angles of the left and The present work is supported by the National Natural
right wheels Science Foundation of China (U1134202), the National Basic
𝜙0 : The rail cant Research Program of China (2011CB711103), and the Program
𝜙𝑟(𝐿,𝑅) : The left and right rail rolling angles for Changjiang Scholars and Innovative Research Team in
𝑑𝐵 : The abscissa ordinate of the instant university (IRT1178 and SWJTU12ZT01).
rolling circle center
𝑙𝑥 , 𝑙𝑦 , 𝑙𝑧 : The cosine components of the axle References
centerline in the 𝑋-, 𝑌-, and 𝑍-
directions, respectively, of a rigid [1] X. S. Jin, L. Wu, J. Y. Fang, S. Q. Zhong, and L. Ling, “An
wheelset investigation into the mechanism of the polygonal wear of
(𝑙(𝐿,𝑅)𝑥 , 𝑙(𝐿,𝑅)𝑦 , 𝑙(𝐿,𝑅)𝑧 ): The cosine components of the axle metro train wheels and its effect on the dynamic behavior of
centerline in the 𝑋-, 𝑌-, and 𝑍- a wheel/ rail system,” Vehicle System Dynamics, vol. 50, no. 12,
directions, respectively, of left and pp. 1817–1834, 2012.
right wheels of a flexible wheelset [2] K. L. Knothe and S. L. Grassie, “Modelling of railway track and
𝜙: The rolling angle of a rigid wheelset vehicle/track interaction at high frequencies,” Vehicle System
𝜓: The yawing angle of a rigid wheelset Dynamics, vol. 22, no. 3-4, pp. 209–262, 1993.
Advances in Mechanical Engineering 19
[3] T. B. Meinders, P. Meinke, and W. O. Schiehlen, “Wear estima- in Proceedings of the 22th Symposium of the International
tion in flexible multibody systems with application to railroads,” Association of Vehicle System Dynamics (IAVSD ’11), Manchester,
in Proceedings of the ECCOMAS Thematic Conference on Multi- UK, 2011.
Body Dynamics, pp. 21–24, Madrid, Spain, June 2005. [22] X. S. Jin, B. Q. Wu, L. Wu, J. Y. Fang, and S. Q. Zhong, “An
[4] T. Szolc, “Medium frequency dynamic investigation of the rail- investigation into the mechanism of the polygonal wear of
way wheelset-track system using a discrete-continuous model,” metro train wheels and its effect on the dynamical behavior
Archive of Applied Mechanics, vol. 68, no. 1, pp. 30–45, 1998. of wheel/rail system,” in Proceedings of the 22th Symposium
[5] K. Popp, H. Kruse, and I. Kaiser, “Vehicle-track dynamics in the of the International Association of Vehicle System Dynamics
mid-frequency range,” Vehicle System Dynamics, vol. 31, no. 5-6, (IAVSD ’11), Manchester, UK, 2011.
pp. 423–464, 1999.
[6] N. Chaar, Wheelset structural flexibility and track flexibility in
vehicle/track dynamic interaction [Ph.D. thesis], Royal Institute
of Technology, Stockholm, Sweden, 2007.
[7] M. Meywerk, “Polygonalization of railway wheels,” Archive of
Applied Mechanics, vol. 69, no. 2, pp. 105–120, 1999.
[8] T. Szolc, “Medium frequency dynamic analysis of various
designs of the high-speed railway bogies,” in Proceedings of the
7th Mini Conference on Vehicle System Dynamics, Identification
and Anomalies, pp. 255–262, November 2000.
[9] T. Meinders, “Modeling of a railway wheelset as a rotating elastic
multibody system,” Machine Dynamics Problems, vol. 20, pp.
209–219, 1998.
[10] T. Meinders and P. Meinker, “Rotor dynamics and irregular
wear of elastic wheelsets,” in System Dynamics and Long-Term
Behaviour of Railway Vehicles, Track and Subgrade, vol. 6 of
Lecture Notes in Applied Mechanics, pp. 133–152, 2002.
[11] J. Fayos, L. Baeza, F. D. Denia, and J. E. Tarancón, “An
Eulerian coordinate-based method for analysing the structural
vibrations of a solid of revolution rotating about its main axis,”
Journal of Sound and Vibration, vol. 306, no. 3–5, pp. 618–635,
2007.
[12] L. Baeza, P. Vila, A. Roda, and J. Fayos, “Prediction of corruga-
tion in rails using a non-stationary wheel-rail contact model,”
Wear, vol. 265, no. 9-10, pp. 1156–1162, 2008.
[13] L. Baeza, P. Vila, G. Xie, and S. D. Iwnicki, “Prediction of rail
corrugation using a rotating flexible wheelset coupled with a
flexible track model and a non-Hertzian/non-steady contact
model,” Journal of Sound and Vibration, vol. 330, no. 18-19, pp.
4493–4507, 2011.
[14] A. A. Shabana, “Flexible multibody dynamics: review of past
and recent developments,” Multibody System Dynamics, vol. 1,
no. 2, pp. 189–222, 1997.
[15] X. Jin, P. Wu, and Z. Wen, “Effects of structure elastic defor-
mations of wheelset and track on creep forces of wheel/rail in
rolling contact,” Wear, vol. 253, no. 1-2, pp. 247–256, 2002.
[16] X. Xiao, X. Jin, Z. Wen, M. Zhu, and W. Zhang, “Effect of
tangent track buckle on vehicle derailment,” Multibody System
Dynamics, vol. 25, no. 1, pp. 1–41, 2011.
[17] X. Xiao, X. Jin, and Z. Wen, “Effect of disabled fastening systems
and ballast on vehicle derailment,” Journal of Vibration and
Acoustics, vol. 129, no. 2, pp. 217–229, 2007.
[18] X. Xiao, X. Jin, Y. Deng, and Z. Zhou, “Effect of curved track
support failure on vehicle derailment,” Vehicle System Dynamics,
vol. 46, no. 11, pp. 1029–1059, 2008.
[19] W. M. Zhai, Vehicle/Track Coupling Dynamics, Chinese Science
Press, 3rd edition, 2007.
[20] S. G. Zhang, Gaosu Lieche Sheji Fangfa Yanjiu, Study on
Design Method of High-Speed Trains, China Railway Publish-
ing House, Beijing, China, 2009.
[21] W. Li, X. Li, L. Wu et al., “Investigation into mechanism of metro
rail corrugation using experimental and theoretical methods,”
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 313279, 12 pages
http://dx.doi.org/10.1155/2013/313279
Research Article
Dynamic Modeling and Analysis of Hub-Tapered Beam System
Copyright © 2013 S.-j. Chen and D.-g. Zhang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
The dynamic modeling of a hub-tapered flexible beam system which is rotating in a plane is investigated. In the modeling, the high-
order terms related with the nonlinear coupling term are retained, which are ignored in the first-order approximation coupling
modeling. So, this model can also be called a high-order approximation coupling model, which cannot only be used in the case
of small deformation but also in the case of large deformation. Solving for this mathematical model, we can obtain the dynamics
of the system. Examples are given to indicate the differences between the high-order approximation coupling model and the first-
order approximation coupling model. Also, the natural frequencies are studied through the transversal bending vibration analy-
sis. The width ratio and the height ratio of the tapered beam are shown to have great influences on the transversal bending natural
frequencies.
𝑥
𝑇
H (𝑥) = ∫ Φ𝑦 (𝜉) Φ𝑦 (𝜉) 𝑑𝜉. (9)
Figure 3: The deformation of the beam. 0
𝑇
The velocity vector of the point 𝑃(𝑥) can be obtained by Take the generalized coordinates q = (𝜃, A𝑇 , B𝑇 ) and
differentiating Equation (2) as follows: substitute the kinetic and potential energy expressions (Equa-
tions (5) and (6)) into the second kind of Lagrange’s equation
ṙ= (𝑤̇1 (𝑥, 𝑡) + 𝑤̇c (𝑥, 𝑡) − 𝑤2 (𝑥, 𝑡) 𝜃)̇ i
(4) 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉
+ [(𝑎 + 𝑥 + 𝑤1 (𝑥, 𝑡) + 𝑤c (𝑥, 𝑡)) 𝜃̇+ 𝑤̇2 (𝑥, 𝑡)] j ; ( )− =− + 𝐹𝑞 . (10)
𝑑𝑡 𝜕𝑞 ̇ 𝜕𝑞 𝜕𝑞
then, the kinetic energy of the system can be expressed as
Then, we will get the dynamic equation of the system
1 1 𝐿
𝐾 = 𝐽oh 𝜃2̇ + ∫ 𝜌𝑆 (𝑥) r𝑇̇ r𝑑𝑥,
̇ (5) written in compact form as
2 2 0
where the first part is the kinetic energy of the hub and the Mq̈= Q, (11)
second part is the kinetic energy of the beam.
2.2. Potential Energy. As the beam is constrained in the hori- where M is called generalized mass matrix and Q is called
zontal plane, the gravitational potential energy can be ignored generalized force matrix. M and Q can be formed as
here. So, the potential energy of the system can be expressed
as
𝑀11 M12 M13 𝑄𝜃 𝜏
2
1 𝐿 𝜕𝑤 (𝑥, 𝑡) M = [M21 M22 M23 ] , Q = [Q𝐴] + [0] , (12)
𝑉 = ∫ 𝐸𝑆 (𝑥) ( 1 ) 𝑑𝑥
2 0 𝜕𝑥 [M31 M32 M33 ] [ Q𝐵 ] [0]
(6)
2
1 𝐿 𝜕2 𝑤2 (𝑥, 𝑡)
+ ∫ 𝐸𝐼 (𝑥) ( ) 𝑑𝑥, where
2 0 𝜕𝑥2
where the first part is the stretching energy of the beam and 𝑀11 = 𝐽oh + 𝐽ob + 2S𝑥 A + A𝑇M𝑥 A + B𝑇 M𝑦 B − B𝑇 CB
the second part is the bending energy of the beam.
1 𝐿
2.3. Dynamic Equations. The assumed mode method (AMM) + ∫ 𝜌𝑆 (𝑥) (B𝑇 HBB𝑇 HB) 𝑑𝑥
4 0
is used to describe the deformation of the flexible beam. The
axial deformation 𝑤1 (𝑥, 𝑡) and the transversal deformation 𝐿
𝑤2 (𝑥, 𝑡) can be expressed as − ∫ 𝜌𝑆 (𝑥) (A𝑇 Φ𝑥 𝑇 B𝑇 HB) 𝑑𝑥
0
𝐿
M31 = M13 𝑇 = S𝑦 𝑇 + M𝑥𝑦 𝑇 A
M𝑥 = ∫ 𝜌𝑆 (𝑥) Φ𝑥 𝑇 (𝑥) Φ𝑥 (𝑥) 𝑑𝑥
𝐿 0
1
+ ∫ 𝜌S (𝑥) (HBΦ𝑦 B − Φ𝑦 𝑇 B𝑇 HB) 𝑑𝑥 𝐿
0 2
M𝑦 = ∫ 𝜌𝑆 (𝑥) Φ𝑦 𝑇 (𝑥) Φ𝑦 (𝑥) 𝑑𝑥
𝐿 0
M32 = M23 𝑇 = −∫ 𝜌S (𝑥) (HBΦ𝑥 ) 𝑑𝑥 𝐿
0
M𝑥𝑦 = ∫ 𝜌𝑆 (𝑥) Φ𝑇𝑥 (𝑥) Φ𝑦 (𝑥) 𝑑𝑥
0
𝑄𝜃 = − 2𝜃̇ [S𝑥 Ȧ+ A𝑇 M𝑥 Ȧ+ B𝑇 M𝑦 Ḃ− B𝑇 CB]
̇ 𝐿
𝐿 C = ∫ 𝜌𝑆 (𝑥) (𝑎 + 𝑥) H (𝑥) 𝑑𝑥
̇ 𝑑𝑥
− ∫ 𝜌S (𝑥) (B𝑇 Φ𝑦 𝑇 Ḃ𝑇 HB) 0
0 𝐿
𝐿
K1 = ∫ 𝐸𝑆 (𝑥) Φ𝑇
𝑥 (𝑥) Φ𝑥 (𝑥) 𝑑𝑥
0
+ 𝜃̇∫ 𝜌S (𝑥) (ȦT Φ𝑥 𝑇 B𝑇 HB + 2A𝑇 Φ𝑥 𝑇 B𝑇 HḂ
0 𝐿
K2 = ∫ 𝐸𝐼 (𝑥) Φ𝑇
𝑦 (𝑥) Φ𝑦 (𝑥) 𝑑𝑥
0
̇ 𝑑𝑥
−B𝑇 HBB𝑇 HB)
𝐿
𝐽ob = ∫ 𝜌𝑆 (𝑥) (𝑎 + 𝑥)2 𝑑𝑥.
Q𝐴 = 𝜃2̇ (S𝑇𝑥 +M𝑥 A) +2𝜃M
̇ B−K
𝑥𝑦
̇ 1A 0
(14)
𝐿
1
− 𝜃2̇ ∫ 𝜌𝑆 (𝑥) (Φ𝑥 𝑇 B𝑇 HB) 𝑑𝑥
2 0
3. First-Order Approximation Coupling
𝐿 (FOAC) Model
̇ 𝑑𝑥
+∫ 𝜌𝑆 (𝑥) (Φ𝑥 𝑇 Ḃ𝑇 HB)
0
When evolving the kinetic energy expression, since the cou-
pling deformation 𝑤𝑐 (𝑥, 𝑡) is the second-order amount of the
Q𝐵 = 𝜃2̇ (M𝑦 −C) B−2𝜃M
̇ 𝑇 A−K
𝑥𝑦
̇ 2B
transversal deformation 𝑤2 (𝑥, 𝑡), some high-order amount
𝐿
respected to 𝑤𝑐 (𝑥, 𝑡) can be ignored. Thus, the parts with
1
+𝜃2̇ ∫ 𝜌𝑆 (𝑥) ( HBB𝑇 HB−HBΦ𝑥 A) 𝑑𝑥 double underlines in (13) can be ignored. So, (11) can be
0 2 transferred as another form
𝐿
−2𝜃̇∫ 𝜌𝑆 (𝑥) (HBΦ𝑦 B−Φ ̇ 𝑑𝑥
̇ 𝑦 𝑇 B𝑇 HB)
𝑀11 M12 M13 𝜃̈ 0 0 0 𝜃̇
0
[M21 M22 0 ] [Ä] + [0 0 G23 ] [Ȧ]
𝐿 [M31 0 M33 ] [ B̈
] [0 G32 0 ] [ Ḃ]
−∫ 𝜌𝑆 (𝑥) (HBḂ𝑇 HB)
̇ 𝑑𝑥. (15)
0 0 0 0 𝜃 𝑄𝜃 𝜏
(13) + [0 K22 0 ] [A] = [Q1 ] + [0] ,
[0 0 K33 ] [ B ] [ 0 ] [0]
Equation (11) is the rigid-flexible coupling dynamic equa-
tion of the system. The parts with single underline in (13) are where
called the one-order coupling terms. The parts with double
underlines in (13) are called the high-order coupling terms. If
without the large overall motion, (11) can be degenerated as 𝑀11 = 𝐽oh + 𝐽ob + 2S𝑥 A + A𝑇 M𝑥 A + B𝑇 M𝑦 B − B𝑇 CB (16)
the structure dynamic equation. Actually, (11) contains four
different models: the high-order approximation coupling M22 = M𝑥 (17)
(HOAC) model with large overall motion, the first-order
approximation coupling (FOAC) model with large overall M33 = M𝑦 (18)
motion, the zero-order approximation coupling (ZOAC)
model with large overall motion, and the structure dynamic M21 = M12 𝑇 = −M𝑥𝑦 B (19)
model without large overall motion. The constant matrices in
(13) can be expressed as M31 = M13 𝑇 = S𝑦 𝑇 + M𝑥𝑦 𝑇 A (20)
̇
G23 = −G𝑇32 = −2𝜃M (21)
𝐿 𝑥𝑦
S𝑥 = ∫ 𝜌𝑆 (𝑥) (𝑎 + 𝑥) Φ𝑥 (𝑥) 𝑑𝑥
0 K22 = K1 − 𝜃2̇M𝑥 (22)
𝐿
S𝑦 = ∫ 𝜌𝑆 (𝑥) (𝑎 + 𝑥) Φ𝑦 (𝑥) 𝑑𝑥 K33 = K2 − 𝜃2̇ (M𝑦 −C) (23)
0
Advances in Mechanical Engineering 5
Table 1: The dimensionless natural frequencies of the uniform beam in transversal bending vibration with different 𝛾 (𝛿 = 0, 𝛼 = 𝛽 = 1).
ysis, the angular speed of the large overall motion is assumed + 𝛾2 ∫ [(1 − 𝜉) + 𝜉𝛼] [(1 − 𝜉) + 𝜉𝛽] (𝛿 + 𝜉) H (𝜉) 𝑑𝜉
0
to be constant as 𝜃̈= 0. Compared to the transversal defor-
𝑇
mation, the longitudinal deformation has little influence on 𝜉 𝜕Φ𝑦 (𝜂) 𝜕Φ𝑦 (𝜂)
the transversal bending vibration. So, the transversal bending H (𝜉) = ∫ ( ) ( ) 𝑑𝜂.
0 𝜕𝜂 𝜕𝜂
vibration equation of the system with constant angular speed
(29)
can be simplified as
Let 𝜅2 = e𝜆𝜔𝜍 Θ; we can get the eigenvalue equation
M33 B̈+ K33 B = 0. (26)
𝜔2 M33 Θ = K33 Θ, (30)
For convenience of discussion, dimensionless parameters where 𝜆 is imaginary, 𝜔 is the dimensionless natural fre-
are used in the present analysis and they are defined as quency, and Θ is a constant column matrix.
Table 1 shows the dimensionless natural frequencies of
𝑡 𝑥 B 𝑎 the uniform beam without hub in the transversal bending
𝜍= , 𝜉= , 𝜅2 = , 𝛿= ,
𝑇 𝐿 𝐿 𝐿 vibration under different 𝛾, where 𝛿 = 0, and 𝛼 = 𝛽 = 1,
(27) truncate the first four modes when calculating. From Table 1,
𝑏 ℎ
𝛾 = 𝑇𝜃,̇ 𝛼 = 1, 𝛽 = 1, we can see that the dimensionless natural frequencies here are
𝑏2 ℎ2 the same as the results of paper [13], using another modeling
method. That proves the correctness of this model. So we
where 𝑇 = (𝜌S0 L4 /𝐸𝐼0 )1/2 and 𝑆0 = 𝑏2 ℎ2 , 𝐼0 = (1/12)𝑏2 3 ℎ2 . will truncate the first four modes when calculating in the
And 𝛿, 𝛾, 𝛼, and 𝛽 will denote the hub radius ratio, the angular following analysis. Additionally, we can see that the dimen-
speed ratio, the width ratio, and the height ratio, respectively. sionless natural frequencies increase as the angular speed
The dimensionless form of the free vibration equation ratio 𝛾 increases.
is finally obtained by using the dimensionless variables and In Table 2, two models are compared, one is the FOAC
parameters defined in (27) model in this paper, and the other is the traditional ZOAC
model (without the single underline parts). Also, the beam
is considered to be a uniform one for simplifying analysis,
M33 𝜅̈
2 + K33 𝜅2 = 0, (28) where 𝛿 = 𝛼 = 𝛽 = 1. From (28), we can find that the two
6 Advances in Mechanical Engineering
Table 2: The dimensionless natural frequencies of the uniform beam in the transversal bending vibration with different models (𝛿 = 𝛼 = 𝛽 =
1).
Table 3: The dimensionless natural frequencies of the tapered beam in the transversal bending vibration with different 𝛿 (𝛼 = 𝛽 = 0.5).
Table 4: The first dimensionless natural frequencies in the transversal bending vibration with different 𝛼 (𝛿 = 𝛽 = 1).
𝛾=2 𝛾 = 10
𝛼
First natural frequency Change (%) First natural frequency Change (%)
0 6.20400 15.8566
19 10
0.3 4.99493 14.2455
5.4 3.0
0.5 4.72715 13.8222
3.5 1.9
0.7 4.56147 13.5539
3.5 1.9
1 4.40063 13.2934
models are same when 𝛾 = 0. So the natural frequencies are Also, the first natural frequency is in inverse proportion to the
equal when 𝛾 = 0. When 𝛾 increasing, the natural frequencies height ratio 𝛽. As we can see, the changes of the first natural
of the two models become discriminative. When the angular frequency under different 𝛽 have the same variation as the
speed of the large overall motion is low (𝛾 = 2), the natural changes of the first natural frequency under different 𝛼.
frequencies of the two models have a small distinction. But
when the large overall motion is at high speed, the natural
3.2. Transversal Bending Vibration Analysis Including Longitu-
frequencies of the ZOAC model will be imaginary, which is
dinal Deformation Effect. The transversal bending vibration
impractical, while the natural frequencies of the FOAC model
analysis including longitudinal deformation effect will be
are still real. So we can say that the traditional ZOAC model
studied in this section. Also, the angular speed of the large
without the coupling deformation is inapplicable when the
large overall motion of the beam is at high speed. overall motion is assumed to be constant as 𝜃̈= 0, and Q1
Table 3 shows the dimensionless natural frequencies of is neglected here for the free vibration analysis. So, the free
the tapered beam in the transversal bending vibration under vibration equation of the system with constant angular speed
different 𝛿, where 𝛼 = 𝛽 = 0.5. It is obvious that the can be formed as
dimensionless natural frequency is proportional to the hub
radius ratio 𝛿, especially when the large overall motion is at M 0 Ä 0 G23 Ȧ
( 22 )( ) + ( ) ( ̇)
high speed. 0 M33 B̈ G32 0 B
Table 4 shows the first dimensionless natural frequencies (31)
of the tapered beam with different 𝛼 when 𝛿 = 𝛽 = 1. As we K 0 A
+ ( 22 ) ( ) = 0.
can see, the first natural frequency is in inverse proportion 0 K33 B
to the width ratio 𝛼. When 𝛼 changes from 0 to 0.3, the first
natural frequency will have a great change. Compared to the Other dimensionless parameters are defined as
change of the first natural frequency when 𝛾 = 2, the change
of the first natural frequency is much smaller when 𝛾 = 10. 1/2
A S0 L2
Table 5 shows the first dimensionless natural frequencies 𝜅1 = , 𝜀=( ) , (32)
of the tapered beam with different 𝛽 when 𝛿 = 𝛼 = 1. 𝐿 I0
Advances in Mechanical Engineering 7
Table 5: The first dimensionless natural frequencies in the transversal bending vibration with different 𝛽 (𝛿 = 𝛼 = 1).
𝛾=2 𝛾 = 10
𝛽
First natural frequency Change (%) First natural frequency Change (%)
0 7.74540 16.2564
27 12
0.3 5.65629 14.3832
9.8 3.4
0.5 5.09972 13.8975
6.9 2.2
0.7 4.74692 13.5916
7.3 2.2
1 4.40063 13.2934
𝜅̇1
𝜅̇
z = ( 2) .
where 𝜅1
𝜅2
Table 6: Comparison of the first transversal natural frequencies with and without longitudinal effect (𝛼 = 𝛽 = 0.5, 𝜀 = 70).
0.6 4
3
0.4
2
Tip deformation (m)
0.0 0
−1
−0.2
−2
−0.4
−3
−0.6 −4
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s) Time (s)
HOAC HOAC
FOAC FOAC
(a) 𝜏0 = 36 N⋅m (b) 𝜏0 = 136 N⋅m
Table 7: Data of the hub-tapered beam system. conditions of a cantilever beam are used in the simulations.
And the torque applied at the hub 𝜏 can be expressed as
Property Symbol Value
Effective length of the beam 𝐿 8m {𝜏 sin ( 2𝜋 𝑡) 0 ≤ 𝑡 ≤ 𝑇
Tip width of the tapered beam 𝑏1 1.5 × 10−2 m 𝜏 (𝑡) = { 0 𝑇 (40)
Bottom width of the tapered beam 𝑏2 2.5 × 10−2 m {0 𝑡 > 𝑇,
Tip height of the tapered beam ℎ1 1.5 × 10−2 m where 𝑇 = 2𝑠.
Bottom height of the tapered beam ℎ2 2.5 × 10−2 m Figure 4(a) gives the tip deformations of the tapered
Mass density of the beam 𝜌 2.7667 × 103 kg ⋅m−3 beam with HOAC model and FOAC model when 𝜏0 =
Elastic modulus of the beam 𝐸 6.8952 × 1010 N ⋅m−2 36 N⋅m. When the torque is applied on the hub, the max tip
Radius of the hub 𝑎 1m deformation is 0.56 m, which is only a small deformation to
Rotary inertia of the hub 𝐽oh 5 kg ⋅m2
the effective length of the beam. In this case, the tip deforma-
tions of the HOAC model and the FOAC model are exactly
coincidental, while Figure 4(b) gives the tip deformations
of the two models when 𝜏0 = 136 N⋅m. When the torque
equations for convenience. But they will have a great influ- applying on the hub, the max tip deformation of the HOAC
ence on the dynamics of the system with large deformation. model reached 2.32 m, this is obviously a large deformation
Here, two examples will be given to approve it. to the effective length of the beam. So, in this case, the result of
the FOAC model becomes divergent as shown in Figure 4(b).
4.1. Beam Undergoing Prescribed Translation and Rotation. A On the other hand, the result of the HOAC model is still con-
tapered beam made of aluminum with the dimensions and vergent. Then, we can say that the FOAC model is inappli-
material properties given in Table 7 is used. The boundary cable in the situation of large deformation, while the HOAC
Advances in Mechanical Engineering 9
0.0 0s
0 2 4 6 8 1s
4.2. Single Pendulum with Gravity. To illustrate the effect of
−0.5 large deformation, a flexible tapered pendulum with free
i (m) 2s
falling is investigated as shown in Figure 8. The pendulum is
−1.0 assumed to fall under the effect of gravity.
−1.5 0.5 s To a flexible tapered pendulum with free falling, the grav-
itational potential energy should be included. So, the gravita-
Figure 5: The deformation in rotating frame. tional potential energy will be added to the system potential
energy
𝐿
4 1.5 s
2.5 s
𝑉𝑔 = −g𝑇W ∫ 𝜌𝑆 (𝑥) r𝑑𝑥 = −g𝑇WR, (43)
0
1s
cos 𝜃 − sin 𝜃
g𝑇 = [𝑔𝑥 , 𝑔𝑦 ] W=[ ]
1 sin 𝜃 cos 𝜃
(44)
𝐿
0s
0 R = ∫ 𝜌𝑆r𝑑𝑥.
0 2 4 6 8 0.5 s 10 0
i (m)
−1
Since the FOAC model could not be used in the large
Figure 6: The motion in inertial reference frame. deformation problems, paper [16] investigated the applica-
tion of an absolute nodal coordinate formulation in the cou-
pling dynamics of flexible beams with large deformation. The
model is applicable. That is because the high-order terms with simulation of a flexible pendulum indicated that the absolute
double underlines respected to 𝑤𝑐 (𝑥, 𝑡) were ignored in the nodal coordinate formulation is suited for the beams with
FOAC modeling. large deformation, while the FOAC model cannot be used
The deformation of the tapered beam in the rotating in that case. Here, we will simulate the example of Figure 4
frame 𝑜 − i j , with HOAC model when 𝜏0 = 136 N⋅m, is in paper [16] by using the HOAC model. In reference to paper
shown in Figure 5. And the motion in the inertial reference [16], the flexible pendulum is a uniform one, 𝜌 = 2.76667 ×
frame 𝑜 − ij is shown in Figure 6. 103 kg.m−3 , 𝐿 = 1.8 m, 𝑆(𝑥) = 2.5 × 10−4 m2 , 𝐼(𝑥) = 1.3 ×
In order to research the dynamic response of the tapered 10−9 m4 , and 𝐸 = 6.8952 × 109 N⋅m−2 .
beam, two ratios are defined here, the width ratio 𝑅𝑏 = Figure 9 shows the tip deformation of the flexible pen-
𝑏1 /𝑏2 and the height ratio 𝑅ℎ = ℎ1 /ℎ2 . And two situations dulum with large deformation. As we can see, the result of
are assumed here, on the assumption that the weight of the the FOAC model is divergent, while the result of the HOAC
tapered beam stays invariable. model is nearly the same as the result of paper [16]. This
proves the validity of the HOAC model and the applicability
(1) Suppose the height of the beam keeps uniform, as in the large deformation situation. As we know, the absolute
ℎ1 = ℎ2 = 0.02 m. Then, we will get nodal coordinate method cannot distinguish the rigid body
motion from the elastic deformation of the flexible body.
(1 + 𝑅𝑏 ) 𝑏2 = 0.04. (41) Small deformation problems also should be treated as large
deformation problems, while the FOAC model can only deal
(2) Suppose the width of the beam keeps uniform, as with the small deformation problems. However, the HOAC
𝑏1 = 𝑏2 = 0.02 m. Then, we will get model proposed in this paper can make up for the deficiency
of those two methods.
(1 + 𝑅ℎ ) ℎ2 = 0.04. (42) To further prove the applicability of the HOAC model
in the large deformation situation, a quiet flexible tapered
Figure 7 gives the max tip deformations and the response pendulum will be studied here. The dimensions and material
frequency of the tapered beam with different 𝑅𝑏 and 𝑅ℎ . It properties are still as shown in Table 7, except 𝐸 = 6.8952 ×
is apparent that the max tip deformation increases as 𝑅𝑏 and 105 N⋅m−2 , a = 0 m.
10 Advances in Mechanical Engineering
1.2 1.2
1.0
1.1
0.8
1.0
0.6
0.9
0.4
0.8
0.2
0.0 0.7
0.0 0.5 1.0 0.0 0.5 1.0
Rb , Rh Rb , Rh
Rb Rb
Rh Rh
(a) (b)
j 1
0 0s
−2 0 2 4 6 8
−1 0.4 s
i (m)
−2
i
−3
j (m)
0.8 s
Figure 8: The flexible tapered pendulum. −4
−5
0.25
−6 1.2 s
0.20
−7
0.15
1.6 s
−8
Tip deformation (m)
0.10
0.05 Figure 10: The deformed shape at different times when 𝐸 = 6.8952×
105 N⋅m−2 .
0.00
−0.05
−0.10
large at any time step. If we use the FOAC or ZOAC models,
−0.15 the simulation will be divergent.
−0.20 Figure 11 shows the energy change of the tapered pen-
−0.25
dulum. Curve 1 is the kinetic energy, curve 2 is the elastic
0.0 0.5 1.0 1.5 2.0 2.5 potential energy, curve 3 is the gravitational potential energy,
and curve 4 is the total energy. Obviously, as the pendulum
Time (s)
falls, the kinetic energy increases at the beginning stage. But
HOAC model when the pendulum passes someplace, the kinetic energy will
Absolute nodal coordination method in paper [16] decrease. Also, the elastic potential energy changes the same
FOAC model
law as the kinetic energy, while the gravitational potential
Figure 9: The tip deformation of the uniform pendulum. energy decreases before the pendulum falls to the lowest
point. When the pendulum passes the lowest point, the
gravitational potential energy will increase. Anyway, the total
energy keeps invariant at any time.
Figure 10 shows the deformed shape of the tapered pen- Figure 12 shows the deformed shape of the tapered pen-
dulum with HOAC model at different time steps when 𝐸 = dulum with HOAC model at different time steps when 𝐸 =
6.8952 × 105 N ⋅m−2 . As we can see, the deformation is very 6.8952 × 1010 N⋅m−2 . As we can see, the deformation is very
Advances in Mechanical Engineering 11
300 small at any time step. So, the elastic potential energy is almost
(1)
0 (Figure 13). However, the elastic potential energy is very
200
(2) significant when 𝐸 = 6.8952×105 N⋅m−2 . Also, there are some
100 differences between the kinetic energies when 𝐸 = 6.8952 ×
0 105 N⋅m−2 and 𝐸 = 6.8952 × 1010 N⋅m−2 . The kinetic energy
change is more complex when large deformation occurs.
Energy (J)
−100 (4)
−200 5. Conclusion
(3)
−300 A new rigid-flexible coupling model of a hub-tapered beam
system, which is also called the high-order approximation
−400
coupling (HOAC) model, was established in this paper. The
−500 HOAC model could not only be used in the case of small
deformation, but also in the case of large deformation, while
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 the previous FOAC model could only be used in the case of
Time (s) small deformation. Based on the HOAC model, the motion
(1) Kinetic energy (3) Gravitational potential energy
and the dynamic response of the tapered beam were studied.
(2) Elastic potential energy (4) Total energy Also, the natural frequencies of the FOAC model were
studied through the transversal bending vibration analy-
Figure 11: The energy change of the tapered pendulum when 𝐸 = sis. The dimensionless natural frequencies were shown to
6.8952 × 105 N⋅m−2 . increase as the angular speed ratio 𝛾 and the hub radius
ratio 𝛿 increased. However, they decreased as the width ratio
𝛼 and the height ratio 𝛽 increased. When the large overall
0 0s
motion was at very high speed, the longitudinal deformation
−4 0 4 8 would also have a significant influence on the transversal
i (m)
0.4 s bending vibration.
Acknowledgments
j (m)
0
for dynamic analysis of rigid-flexible multibody systems,” Com-
−100 puter Methods in Applied Mechanics and Engineering, vol. 114,
no. 3-4, pp. 379–396, 1994.
−200 (3)
[4] I. Sharf, “Geometric stiffening in multibody dynamics formula-
−300 tions,” Journal of Guidance, Control, and Dynamics, vol. 18, no.
4, pp. 882–890, 1995.
−400
[5] D. J. Zhang and R. L. Huston, “On dynamic stiffening of flexible
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 bodies having high angular velocity,” Mechanics of Structures
Time (s) and Machines, vol. 24, no. 3, pp. 313–329, 1996.
E = 6.8952e5 [6] H. H. Yoo and S. H. Shin, “Vibration analysis of rotating can-
(1) Kinetic energy
E = 6.8952e10 tilever beams,” Journal of Sound and Vibration, vol. 212, no. 5,
(2) Elastic potential energy
pp. 807–808, 1998.
(3) Gravitational potential energy
[7] J. Y. Liu and J. Z. Hong, “Dynamic modeling and modal trun-
Figure 13: The energy change when 𝐸 = 6.8952 × 105 N ⋅ m−2 and cation approach for a high-speed rotating elastic beam,” Archive
𝐸 = 6.8952 × 1010 N⋅m−2 . of Applied Mechanics, vol. 72, no. 8, pp. 554–563, 2002.
12 Advances in Mechanical Engineering
Research Article
Analysis of Elastic-Plastic Responses of a New
Single-Story Frame-Bent Structure during an Earthquake Based
on the Transfer Matrix Method of Multibody System
Copyright © 2013 Jianguo Ding et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Finite element methods, such as the ANSYS, are widely used as traditional structural dynamic analysis methods. Due to the
low calculation efficiency of the ANSYS, there is a hindrance factor in using the finite element method to compute the elastic-
plastic responses of special engineering structures. In this paper, we detail a pioneer study in which the transfer matrix method
of multibody system (MS-TMM) was introduced to analyze the dynamic responses of civil engineering structures. Although the
responses obtained by the MS-TMM are similar to the responses obtained by the ANSYS, the ANSYS requires a greater amount
of computation time. The results indicate that the MS-TMM can satisfy the requirements of calculation efficiency and calculation
accuracy.
where 𝑥, 𝑦, 𝜃𝑧 , 𝑚𝑧 , 𝑞𝑥 , and 𝑞𝑦 denote the displacements in the In the same coordinate, we obtain
𝑥 and 𝑦 directions, angular displacement around 𝑧, internal
moment, internal force in the 𝑥 direction, and internal force Z𝑗,1 = HU𝑗 Z𝑗,𝑗−1 (𝜃 = −90∘ ) ,
in the 𝑦 direction, respectively. (5)
The transfer process is derived as follows. Z𝑗,𝑗−1 = HU𝑗 Z𝑗,3 (𝜃 = −90∘ ) ,
The side column on the left is as
where
Z2,1 = U1 Z0,1 ,
0 −1 0
Z2,3 = U2 Z2,1 , H1,1 O3×4
H=[ ], H1,1 = [1 0 0] ,
O4×3 H2,2
.. [0 0 1]
.
1 0 0 0 (6)
.. [0 0 −1 0]
. H2,2 =[
[0
].
1 0 0]
Z𝑗,𝑗−1 = U𝑗−1 Z𝑗−2,𝑗−1 , [0 0 0 1]
Advances in Mechanical Engineering 3
x
j 1 2 3 j 1 2 3 j
j−1
j−1 j−1
j−2
j−2 j−2
y
x
4 4 4
z
3 3 3
x1 2 x2 2 x3 2
1 1 1
y1 0 y2 0 y3 0
The frame-bent structure is decomposed into numerous Z0 ,1 = UΙΙΙ Z𝑗 ,1 = UΙΙΙ (U𝑗 E1 UΙ Z0,1
chain systems; thus, the transfer equation is expressed as
+U𝑗 E2 UΙΙ Z0 ,1 ) (14)
Z𝑗 ,3 = UΙ Z0,1 ,
= UΙΙΙ U𝑗 E1 UΙ Z0,1 + UΙΙΙ U𝑗 E2 UΙΙ Z0 ,1 .
Z𝑗 ,𝑗−1 = UΙΙ Z0 ,1 , (7)
UΙΙΙ U𝑗 E1 UΙ Z0,1 + UΙΙΙ U𝑗 E2 UΙΙ Z0 ,1 − Z0 ,1 = 0. (15)
Z0 ,1 = UΙΙΙ Z𝑗 ,1 ,
For 𝑗 , all displacements are equivalent; we obtain
where
E3 Z𝑗 ,3 = E3 Z𝑗 ,𝑗−1 = E3 Z𝑗 ,1 , (16)
UΙ = U3 U2 U1 HU𝑗 U𝑗−1 ⋅ ⋅ ⋅ U2 U1 ,
where E3 = [I3 O3×4 ].
UΙΙ = U𝑗−1 U𝑗−2 ⋅ ⋅ ⋅ U2 U1 , (8) By combining (7), (10), (12), and (16), we obtain the
following equations:
UΙΙΙ = U1 U2 U𝑗−1 HU𝑗 U3 U2 U1 .
E3 UΙ Z0,1 − E3 UΙΙ Z0 ,1 = 0,
The element numbered as 𝑗 is a particle with two inputs (17)
and one output. The state vectors of these two inputs consist (E3 UΙ − E3 U𝑗 E1 UΙ ) Z0,1 − E3 U𝑗 E2 UΙΙ Z0 ,1 = 0.
of Z𝑗 ,3 and Z𝑗 ,𝑗−1 . Referring to (1), we define the state vector
of input point as By combining (15), and (17), we obtain
Therefore, the transfer equation of the element 𝑗 is expressed The boundary conditions are listed as follows:
as follows: 𝑇
Z0,1 = [0, 0, 0, 𝑀𝑧0 , 𝑄𝑥0 , 𝑄𝑦0 , 1] ,
Z𝑗 ,1 = U𝑗 Z𝐼,𝑗 . (12)
𝑇
Z0 ,1 = [0, 0, 0, 𝑀𝑧0 , 𝑄𝑥0 , 𝑄𝑦0 , 1] , (20)
Z𝑗 ,1 = U𝑗 E1 Z𝑗 ,3 + U𝑗 E2 Z𝑗 ,𝑗−1 .
(13) 𝑇
= U𝑗 E1 UΙ Z0,1 + U𝑗 E2 UΙΙ Z0 ,1 . Z0 ,1 = [0, 0, 0, 𝑀𝑧0 , 𝑄𝑥0 , 𝑄𝑦0 , 1] .
4 Advances in Mechanical Engineering
The detailed steps for solving the elastic-plastic responses All transfer matrixes of the elements are illustrated in (21)
of a new single-story frame-bent structure during rarely through (26). The transfer matrix of a rigid body in plane
occurring earthquakes by the MS-TMM are shown in motion is expressed as
Figure 2.
1 0 −𝑦𝐼𝑂 (𝑡𝑖−1 ) 0 0 0 𝑏1 𝐺1 − 𝑏2 𝐺2
[ 0 1 𝑥𝐼𝑂 (𝑡𝑖−1 ) 0 0 0 𝑏1 𝐺2 + 𝑏2 𝐺1 ]
[ ]
[ 0 0 1 0 0 0 0 ]
[ ]
[
U=[ 𝑢 41 𝑢 42 𝑢43 1 𝑢45 𝑢46 𝑢47 ], (21)
]
[− (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) 0 (𝑚 𝐴 + 𝑐 𝐶 + 𝑘) 𝑦𝐼𝐶 (𝑡𝑖−1 ) 0 1 0 𝑢57 ]
[ 𝑟 𝜁 ]
[ 0 − (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) − (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) 𝑥𝐼𝐶 (𝑡𝑖−1 ) 0 0 1 𝑢67 ]
[ 0 0 0 0 0 0 1 ]
1 1
𝑢42 = − (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) 𝑥𝐼𝑂 + 𝑚𝑟 𝐴𝑥𝐼𝐶, (𝛼 = , 𝛽 = ) .
2 4
(23)
𝑢43 = (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) 𝑥𝐼𝐶 (𝑡𝑖−1 ) 𝑥𝐼𝑂
− (𝑚𝑟 𝐴 + 𝑐𝜁 𝐶 + 𝑘) 𝑦𝐼𝐶 (𝑡𝑖−1 ) 𝑦𝐼𝑂 + 𝐽𝐴, When the earthquake acceleration is 𝑥𝑔̈ (𝑡𝑖−1 ) at 𝑡 = 𝑡𝑖−1 , 𝑓𝑥,𝑐 =
−𝑚𝑟 𝑥𝑔̈ (𝑡𝑖−1 ). The transfer matrix of a massless elastic beam in
𝑢45 = −𝑦𝐼𝑂, 𝑢46 = 𝑥𝐼𝑂, lateral vibration is as follows:
(𝑐𝑐1 𝑆𝐼 + 𝑐𝑐2 𝐶𝐼 )
𝑦𝐼𝐶 (𝑡𝑖−1 ) = , The transfer matrix of the particle is expressed as
𝑡𝑖−1
(𝑏1 𝐶𝐼 − 𝑏2 𝑆𝐼 )
𝑥𝐼𝑂 (𝑡𝑖−1 ) = , 1 0 0 0 0 0 0
𝑡𝑖−1 [0 ]
[ 1 0 0 0 0 0 ]
[0 0 1 0 0 0 0 ]
(𝑏1 𝑆𝐼 + 𝑏2 𝐶𝐼 ) [ ]
𝑦𝐼𝑂 (𝑡𝑖−1 ) = , [ ]
𝑡𝑖−1 U = [0 0 0 1 0 0 0 ],
[ ]
[0 0 0 0 1 0 0 ]
1 2 [ ]
𝐺1 = cos 𝜃 (𝑡𝑖−1 ) {1 − [𝜃̇(𝑡𝑖−1 ) Δ𝑡] } [0 − (𝑚𝐴 + 𝑐𝜁 𝐶 + 𝑘) 0 0 0 1 −𝑚𝐵𝑦𝐼 + 𝑐𝜁 𝐷𝑦𝐼 + 𝑓𝑦,𝑐 ]
2
[0 0 0 0 0 0 1 ]
+ 𝜃 (𝑡𝑖−1 ) sin 𝜃 (𝑡𝑖−1 ) , (25)
1 2
G2 = sin 𝜃 (𝑡𝑖−1 ) {1 − [𝜃̇(𝑡𝑖−1 ) Δ𝑡] }
2 where 𝑚 is the mass of the particle and c𝜁 is the damping of
+ 𝜃 (𝑡𝑖−1 ) cos 𝜃 (𝑡𝑖−1 ) , the structure.
Advances in Mechanical Engineering 5
Start
i=1
ti ≥ td ti < td
End
Figure 2: The process of solving the elastic-plastic responses of structures by the MS-TMM.
When the earthquake acceleration is 𝑥𝑔̈ (𝑡𝑖−1 ) at 𝑡 = 4. The Elastic-Plastic Restoring Force
𝑡𝑖−1 , 𝑓𝑦,𝑐 = −𝑚𝑥𝑔̈ (𝑡𝑖−1 ). The transfer matrix of the planar Model of the Structure
hinge is expressed as
Because the structure exhibits elastic plasticity during rarely
occurring earthquakes, the stiffness varies with time; thus,
1
1 0 0 0 − 0 0 it is important to establish relations between the restoring
[ 𝑘𝑥 ] force and the displacement. Currently, the commonly used
[ 1 ]
[0 1 0 0 0 − 0] models are the double restoring force model and the three
[ ]
[ 𝑘𝑦 ] linear restoring force models. The double model is the simpler
[ ]
[ 1 ] model. To reflect the properties of the concrete structure in
U=[
[
0 0 1
𝑘
0 0 0] ,
] (26)
[0 the elastic-plastic region, the model of the trilinear restoring
[ 0 0 1 0 0 0]] force is selected for this paper, as shown in Figure 3 [8].
[ ]
[0 0 0 0 1 0 0] In Figure 3, point 1 is the cracking point, 𝐹𝑐 and 𝑥𝑐 are
[ ]
[0 0 0 0 0 1 0] the corresponding force and displacement, stage 01 is elastic,
[0 0 0 0 0 0 1] and 𝑘1 is the elastic stiffness; point 2 is the yielding point,
𝐹𝑦 and 𝑥𝑦 are the corresponding force and displacement,
stage 12 represents cracking to yielding, and 𝑘2 is the cracking
where 𝑘𝑥 , and 𝑘𝑦 are the spring stiffness in the x, y direc- stiffness; after yielding, 𝑘3 is the yielding stiffness; the secant
tions, respectively, and 𝑘 is the stiffness of the rotational stiffness 𝑘4 is the unloading stiffness; in stage 12, when
spring around z. unloading to zero, the straight line extends to the cracking
The process of solving the elastic-plastic responses of a point for the first reverse loading; in stage 23, when unloading
new single-story frame-bent structure during rarely occur- to zero, the straight line extends to the yielding point for
ring earthquake is as follows (Figure 2). the first reverse loading; for subsequent reverse loadings, the
6 Advances in Mechanical Engineering
F F
2 k3 3(8) 9 3 5
6
Fy 4
k2
2
1 1
Fc 7
k1
k4
7 0 4 10 8
x x
xc xy
10 9
x (mm)
0
the cracking stiffness 𝑘2 , the yielding stiffness 𝑘3 , and the −10
secant stiffness 𝑘4 are 3.16 × 108 N/m, 2.3 × 108 N/m, and −20
3.46 × 108 N/m, respectively; the cracking displacement 𝑥𝑐 −30
−40
is 6.3 mm; the yielding displacement 𝑥𝑦 is 21.8 mm; the mass −50
of saddle plate 𝑚2 is 25.14 × 103 kg; the masses of particles 0 1 2 3 4 5 6 7 8 9 10 11
are 𝑚2 = 𝑚4 = 𝑚6 = 5.93 × 103 kg. The seismic fortification t (s)
intensity is 7, the site type is II, and the classification of design El Centro Wave
earthquake is the 1st classification. Taft Wave
To study the responses of a new single-story bent-frame Nanjing Wave
structure during rarely occurring earthquakes, two natural
Figure 5: The displacement time history curves for the top of the
earthquake waves (the El Centro earthquake wave and the structure during rarely occurring earthquakes (based on the MS-
Taft earthquake wave) and one artificial earthquake wave (the TMM).
Nanjing earthquake wave) are chosen. The duration time is
10 s. According to the seismic codes of the PROC, we adjust
the peak value of the earthquake waves to 220 cm/s2 for the 45
rarely occurring earthquake. 30
15
x (mm)
0
−15
5.1. Analysis of the Elastic-Plastic Displacement of a New −30
Single-Story Bent-Frame Structure during Rarely Occurring −45
Earthquakes Is Based on the MS-TMM. The responses of a 0 1 2 3 4 5 6 7 8 9 10 11
new single-story bent-frame structure during rarely occur- t (s)
ring earthquakes are computed using the MS-TMM (as El Centro Wave
shown in Figure 5) and the finite element software ANSYS (as Taft Wave
shown in Figure 6). When the responses of a new single-story Nanjing Wave
bent-frame structure during rarely occurring earthquakes
Figure 6: The displacement time history curves for the top of
are computed by using the finite element software ANSYS,
the structure during rarely occurring earthquakes (based on the
the BEAM198 and SHELL63 are chosen as the models of a ANSYS).
beam or column and a saddle plate, respectively. Because the
foundation is assumed to be a rigid body, the link model at
the bottom of a column corresponds to the fixed constraint.
As illustrated in Figures 5 and 6, based on the MS- Evidently, the elastic-plastic displacement trajectories on
TMM, the maximum elastic-plastic displacement on the the top of a new single-story bent-frame structure obtained
top of the new single-story bent-frame structure under the by the MS-TMM and the ANSYS under the actions of the
activity of El Centro Wave, Taft Wave, and Nanjing Wave three earthquake waves are very close, as shown in Figures
is 47.93 mm at 𝑡 = 2.78 s, 44.41 mm at 𝑡 = 4.48 s, 5 and 6.
and 46.65 mm at 𝑡 = 8.00 s, respectively. And based on The maximum displacement on the top of a new
ANSYS, those are 43.13 mm at 𝑡 = 2.76 s, 46.12 mm at single-story bent-frame structure obtained by MS-TMM and
𝑡 = 4.46 s, and 42.3 mm at 𝑡 = 7.96 s, respectively. Based ANSYS under the actions of the three earthquake waves is
on the MS-TMM, the maximum elastic-plastic story drift shown in Table 1; their errors are also listed.
angles under the activity of three earthquake waves are As listed in Table 1, the error for the average maximum
1/188, 1/203, and 1/193, respectively. And based on ANSYS, elastic-plastic displacement on the top of a new single-story
those are 1/195, 1/213, and 1/205, respectively. The average bent-frame structure under the actions of three earthquake
maximum elastic-plastic displacement on the top of the waves using the MS-TMM and ANSYS is 5.66%, which
structure got by both methods are 46.33 mm and 43.58 mm is similar to the results of engineering computations. The
under the activity of three earthquake waves, respectively; MS-TMM required only 100 s of post-processing time, but
the average maximum elastic-plastic story drift angles got the ANSYS required more than 1800 s of postprocessing
by both methods are 1/194 and 1/205 under the activity of time to solve the maximum elastic-plastic displacement on
three earthquake waves, respectively. Because determinate the top of the structure. Because the transition matrixes
regulations regarding elastic-plastic story drift displacement and the transmission equations do not require a second
angle of the frame-bent structure are nonexistent, we refer to derivation, they are selected directly from the library of
the concrete frame structure, in which the limit of an elastic- transition matrixes; the transfer matrix method of multibody
plastic story drift angle is 1/50. Thus, the frame-bent structure system is convenient. Computations of the elastic-plastic
satisfies the requirement of the seismic codes of the PROC. displacement during rarely occurring earthquakes with the
8 Advances in Mechanical Engineering
Table 1: The maximum elastic-plastic displacement for the top of a new single-story bent-frame structure during rarely occurring
earthquakes.
120 120
80 80
40 40
V (kN)
V (kN)
0 0
−40 −40
−80 −80
−120
−120
0 1 2 3 4 5 6 7 8 9 10 11
0 1 2 3 4 5 6 7 8 9 10 11
t (s)
t (s)
El Centro Wave El Centro Wave
Taft Wave Taft Wave
Nanjing Wave Nanjing Wave
Figure 7: The shear force time history curves for the bottom of the Figure 9: The shear force time history curves for the bottom of the
side column during rarely occurring earthquakes (based on the MS- middle column during rarely occurring earthquakes (based on the
TMM). MS-TMM).
80 120
40 80
V (kN)
0 40
V (kN)
−40 0
−80 −40
−120 −80
0 1 2 3 4 5 6 7 8 9 10 11 −120
t (s) 0 1 2 3 4 5 6 7 8 9 10 11
t (s)
El Centro Wave
Taft Wave El Centro Wave
Nanjing Wave Taft Wave
Nanjing Wave
Figure 8: The shear force time history curves for the bottom of
the side column during rarely occurring earthquakes (based on Figure 10: The shear force time history curves for the bottom of
ANSYS). the middle column during rarely occurring earthquakes (Based on
ANSYS).
Table 2: The maximum shear forces at the bottom of the side column in a new single-story bent-frame structure during rarely occurring
earthquakes.
El Centro wave Taft Wave Nanjing Wave Average value
MS-TMM (kN) 119.82 97.78 111.14 109.58
ANSYS (kN) 104.42 93.83 101.85 100.10
Error 14.75% 4.21% 9.12% 9.50%
Table 3: The maximum shear forces at the bottom of the middle column in a new single-story bent-frame structure during rarely occurring
earthquakes.
El Centro Wave Taft Wave Nanjing Wave Average value
MS-TMM (kN) 125.81 104.67 116.70 115.73
ANSYS (kN) 112.80 101.30 110.30 108.10
Error 6.22% 3.33% 5.80% 7.06%
Research Article
Cooperative Search by Combining Simulated and Real Robots in
a Swarm under the View of Multibody System Dynamics
Copyright © 2013 Q. Tang and P. Eberhard. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
This paper presents a new approach for cooperative search of a robot swarm. After modeling the robot, the mechanical Particle
Swarm Optimization method is conducted based on physical robot properties. Benefiting from the effective localization and
navigation by sensor data fusion, a mixed robot swarm which contains both simulated and real robots is then successfully used
for searching a target cooperatively. With the promising results from experiments based on different scenarios, the feasibility, the
interaction of real and simulated robots, the fault tolerance, and also the scalability of the proposed method are investigated.
1. Introduction D∗ , nor new methods like genetic algorithms and ant colony
algorithms can produce satisfactory results. In recent years,
Nowadays robots are used more and more since they can the comparatively new stochastic Particle Swarm Optimiza-
do various tasks instead of humans. Among the variety of tion (PSO) algorithms have attracted researchers’ attention.
existing robots, swarms of mobile robots receive growing Many research groups use PSO and its variants for swarm
attention by researchers. A number of large international mobile robots search; see, for example, [9–11]. Unfortunately,
projects have been conducted, for example, SWARM-Bots, these publications have shown shortcomings for cooperative
SWARMMANOID, and SYMBARION [1]. Swarms of mobile search. For example, [9] used PSO for multirobot search;
robots have many applications, for example, for efficient however, its focus is on optimizing PSO parameters and it
collection of oil spill [2, 3], for light or odorant detection does not consider the scalability for a large number of robots.
[4–6], for searching survivors after an earthquake [7], and The work in [10] gives an ad hoc application using a PSO
for assisting humans to handle nuclear leakage [8]. Readers based method for group robots search. Its solutions are valid
may notice from these applications that cooperative search is for a specific robot which has only limited motion ability. At
quite representative for the swarm robotics research since it the moment similar researches mostly stop at a theoretical
covers most of the technical points. Many researches focus stage and are verified only by simulation. Some researchers
on cooperative search; however, this is still at an early stage have used PSO on real robots reluctantly, but they have limi-
in spite of its fast development. Many critical issues have tations such as poor scalability and inadequate search ability.
not been solved yet, including, for example, the challenge in Thus, this study develops a systematic cooperative search
controlling and guiding robots systematically and the transfer method which contains sufficient search ability, scalability,
to workable strategies for real robots. From a technical and fault tolerance, considers robot physical properties, and
perspective, for cooperative search of robot swarms, key qualifies for real-simulated robots interaction.
points are performing collaboration, showing fault tolerance, Section 2 builds kinematic and dynamic models of the
sharing information, distributing the work, and realizing used robot. Section 3 designs the cooperative search algo-
scalability. These are also the advantages of a mobile robot rithm. Experiments performed on a mixed robot swarm for
swarm in contrast to a single mobile robot. Neither the different cases are shown in Section 4, while conclusions are
traditional methods such as artificial potential fields, A∗ or given in Section 5.
2 Advances in Mechanical Engineering
Colour webcam Controller robot and the driven translational velocities V𝑖 of three wheels
2 batteries I/O sockets is governed by
3 motors
Expansion
V1 − sin (𝜙𝑔 + 𝛼1 ) cos (𝜙𝑔 + 𝛼1 ) 𝑅 𝑥𝑔̇
[V 2 ] = [ ]
board
[− sin (𝜙𝑔 + 𝛼2 ) cos (𝜙𝑔 + 𝛼2 ) 𝑅] ⋅ [𝑦𝑔̇ ].
3 encoders
(1)
Steel chassis Protective [V 3 ] [
⏟⏟⏟⏟⏟⏟⏟⏟⏟ − sin (𝜙𝑔 + 𝛼3 ) cos (𝜙𝑔 + 𝛼3 ) 𝑅] [ 𝜙𝑔̇ ]
⏟⏟⏟⏟⏟⏟⏟⏟⏟
guard v
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
S𝑔 u̇
𝑔
3 omni- 9 infrared
directional distance
wheels sensors
Here 𝑅 is the distance from the robot center to the
Figure 1: Festo Robotino and its main components (photo from center of wheel; the matrix S𝑔 is the geometrical relationship
Robotino manual [12]). between the global velocity u̇𝑔 and the translational velocity v
which can also be expressed by the wheel’s angular velocities
𝜔𝑖 multiplied by the wheel radius 𝑟; that is,
Motor driver board If S𝑔 is split into a constant part S𝑙 with the known parameters
CF card 𝛼1 , 𝛼2 , and 𝛼3 and a variable part T−1 rot (𝜙𝑔 ), the kinematic
interface board model is
0 1 𝑅
[ √3 1 ]
Flexible connection board 𝜔1 [− − 𝑅]
[𝜔2 ] = 1 [ [ 2 2 ]
]
𝜔3 ] 𝑟 [ ]
Figure 2: Robotino controller head and its components. ⏟⏟⏟⏟⏟⏟⏟⏟⏟
[ [ √3 1 ]
𝜔 − 𝑅
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
[ 2 2 ]
S𝑙 (4)
xl Incremental encoder
yg
DC motor
1 1,y
𝜙g w1
l,1
yl 1,x
f1
𝛼2 = 120∘ 𝛼1 = 0∘ Omniwheel
f2
𝜙g
w2 l,2 Front side
r xg
Back side
𝛼3 = 240∘ 3
l,i Gear belt
Gear box
R fi
2 f3
l,3
i
w3
Figure 5: Robotino drivetrain.
Figure 4: Geometry and kinematic relations of Robotino base.
moment of inertia of the robot, and ü𝑔 is the generalized Now, substituting (6) and (7) into (5) yields
acceleration. For f𝑔 and the robot local force f𝑙 one has −1
fdrag = (S𝑇𝑙 ) ⋅ T−1
rot ⋅ M ⋅ ü𝑔 .
(8)
𝑓𝑔,𝑥 cos (𝜙𝑔 ) − sin (𝜙𝑔 ) 0 𝑓𝑙,𝑥
[𝑓𝑔,𝑦 ] = [ [ sin (𝜙𝑔 ) cos (𝜙𝑔 ) 0] ⋅ [𝑓𝑙,𝑦 ],
] In the following, the Robotino’s motor dynamic is inves-
(6)
[ 𝑙𝑔,𝜙 ] [
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ 0 0 [ 𝑙𝑙,𝜙 ]
1] ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
tigated as well as its relation to (8). Figure 6 shows the
f𝑔 f𝑙
simplified circuit diagram of the used DC motor. Based
Trot
on this and Kirchhoff ’s voltage law, we get for the current
and the connection of f𝑙 to the wheel drag force fdrag is transient changes 𝑐𝑖̇ = 𝑑𝑐𝑖 /𝑑𝑡 (𝑖 = 1, 2, 3)
√ 3 √3 𝑈1 𝑐1 𝑐1̇ 𝑈𝑚,1
[ 0 − [𝑈2 ] = 𝑅Ω [𝑐2 ] + 𝐿[𝑐2̇] + [𝑈𝑚,2 ],
𝑓𝑙,𝑥 [ 2 2 ] ] 𝑓1 (9)
[𝑓𝑙,𝑦 ] = [ [ 1] [⏟⏟𝑈 𝑐3 ] 𝑐3̇] [ 𝑈𝑚,3 ]
1 ] ⋅ [𝑓2 ]. ⏟⏟ ⏟⏟3⏟]
⏟⏟ ⏟⏟⏟⏟⏟⏟⏟⏟⏟
[ ⏟⏟⏟⏟⏟⏟⏟⏟⏟
[ ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
𝑙 [1 − − ] (7) U c ċ U𝑚
[ 𝑙,𝜙 ]
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ [ 2 [𝑓3 ]
2 ] ⏟⏟⏟⏟⏟⏟⏟⏟⏟
f𝑙
[ 𝑅 𝑅 𝑅 ] fdrag
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ where U and U𝑚 are the input voltage acting on the
S𝑇𝑙 motors and the induced voltage from the motor winding,
4 Advances in Mechanical Engineering
7 70 7
6 60 6
MN
Efficiency 𝜂 (%)
Current c (A)
4 40 4 J = f(M)
3 30 3 N = f(m)
2 20 2
1 10 1
0 0 0
−2 0 2 4 6 8 10 12 14 16 18 20
Torque (Ncm)
𝜗R = 20∘ C
Δ𝜗w = 100 K
Figure 7: Characteristic diagram of the DC motor-GR 42 × 25 DC motor used on Robotino (data from information of Dunkermotoren [13]).
𝑚𝑖 0 0 𝑎𝑖,𝑥 𝑓𝑖,𝑥
End
[ 0 𝑚𝑖 0 ] ⋅ [𝑎𝑖,𝑦 ] = [𝑓𝑖,𝑦 ] , (22) End
Turn
[ 0 0 𝐽𝑖 ] [ 𝛼𝑖 ] [ 𝑙𝑖 ] Start Start
(c) Turn, drive, turn (d) Turn and drive, turn
where 𝐽𝑖 is the rotational moment of inertia, 𝑎𝑖,𝑥 and 𝑎𝑖,𝑦 are
the resultant accelerations in 𝑥 and 𝑦 directions, respectively, Start pose
while 𝑓𝑖,𝑥 and 𝑓𝑖,𝑦 are the corresponding forces. The 𝛼𝑖 End pose
, , Orientation marker
represents the planar rotational acceleration, and 𝑙𝑖 is the cor-
responding torque. Thus, the generalized planar acceleration Figure 9: Robot omnidirectional feature.
for robot 𝑖 can be written as
𝑇
ẍ𝑖 = [𝑎𝑖,𝑥 𝑎𝑖,𝑦 𝛼𝑖 ] . (23)
If one defines the generalized acceleration consumption of energy. In Figure 9 one can see four basic
motion modes for the robot which starts from its initial
𝑇
ẍ= [ẍ1 ẍ2 ⋅ ⋅ ⋅ ẍ𝑖 ⋅ ⋅ ⋅ ẍ𝑁𝑝 ] ∈ R3𝑁𝑝 ×1 (24) pose to the desired final pose not strictly along a preferred
trajectory.
for all 𝑁𝑝 robots and k as the term resulting from Euler equa- The motions in Figures 9(a) and 9(b) are preferred
tions, as well as a definition that q contains the information since our researched robot is omnidirectional, and these
of forces and moments, then the motion of the entire swarm two modes can save energy. Nonetheless, it is still difficult
is described by to run mode (b) directly due to the challenges met in
robot position control since the nonlinearity of this robot
M ⋅ ẍ+ k = q or ẍ= M−1 ⋅ (q − k) = M−1 ⋅ F, (25) motion becomes significant when the robot orientation 𝜙𝑔
is changing simultaneously. The reason behind this is that
where F contains the generalized forces. Here, the multibody the orientation 𝜙𝑔 is contained in many entries of the system
system is a free system since no links or joints are involved. matrix when describing this robot model in the state space for
Thus, designing proper controllers. The changing orientation will
play an important role in contributing the nonlinearity. Thus,
M = diag (𝑚1 , 𝑚1 , 𝐽1 , 𝑚2 , 𝑚2 , 𝐽2 , . . . , 𝑚𝑁𝑝 , 𝑚𝑁𝑝 , 𝐽𝑁𝑝 ) (26) we choose mode (a) as a trade-off using the omnidirectional
feature of the robot. The whole trajectory is formed by
possesses both the information of masses and rotational many small segments. Although the robot does not have the
inertias of all the robots in the swarm. With (21) and (25), torque affection from other robots, some additional rotation
the motion of all 𝑁𝑝 robots over time can be computed by adjustments from the robot itself still can be performed if
necessary. In analogy to (18), the force F𝑠 is assumed to consist
ẋ𝑠 of three parts,
x𝑠+1 x𝑠
[ 𝑠+1 ] = [ 𝑠 ] + Δ𝑡 [ −1 ]. (27)
ẋ ẋ M ⋅ F𝑠
The used simple Euler forward algorithm is neither very f1𝑠 = −h𝑠𝑓1 ⋅ (x𝑠 − xself
best,𝑠
),
stable nor very efficient. However, these aspects are not (28)
critical here, because for signals provided from sensors of real f2𝑠 = −h𝑠𝑓2 ⋅ (x𝑠 − x̂swarm
best,𝑠
), f3𝑠 = −h𝑠𝑓3 ⋅ ẋ𝑠 ,
robots, the introduced inaccuracies are even worse.
Next, the essential connection between the mechanical
motion of robots and the mechanical PSO will be made.
where h𝑠𝑓1 , h𝑠𝑓2 , and h𝑠𝑓3 are force factor matrices. Here, the
At the moment, the robots are only affected by forces such
that no torques appear from other robots; that is, 𝑙𝑖 = 0. swarm mode x̂swarm
best,𝑠
can be adjusted to a neighborhood mode
Of course the robot itself can provide its own torques to xnhood when the swarm size is big which brings difficulties
best,𝑠
perform rotations. In other words, the trajectories generated for the robots communication. Force f1𝑠 has the physical
from mechanical PSO only include the two translational meaning of an attraction force by the last self best position,
motions. There are mainly two reasons to design it like this. f2𝑠 represents the attraction force by neighborhood or swarm
One important reason is that we want to benefit from the best position, and f3𝑠 is a kind of inertia force which is
omnidirectional feature of the robot so as to reduce the proportional to its last velocity and counteracts a change in
Advances in Mechanical Engineering 7
direction. Thus, the motion (27) of the entire swarm robotic Animation
1.5
system is governed by
x𝑠+1 x𝑠
[ ] = [ ]
(I3𝑁𝑝 − Δ𝑡M−1 ⋅ h𝑠𝑓3 ) ⋅ ẋ𝑠
1
ẋ𝑠+1
ẋ𝑠 (29)
0.5
[ −1 𝑠 𝑠 ]
+ Δ𝑡 [ M ⋅ h𝑓1 ⋅ (xself − x ) ] .
best,𝑠
−1 𝑠 𝑠
[+M ⋅ h𝑓2 ⋅ (̂xswarm − x )]
best,𝑠
0
Comparing the developed mechanical PSO model (29) for
generating trajectories for swarm mobile robots cooperative
search to the basic PSO model (18), one can see correspond- −0.5
ing relationships as follows:
Δ𝑡M−1 ⋅ h𝑠𝑓1 ←→ 𝑐1 r𝑠1 , Δ𝑡M−1 ⋅ h𝑠𝑓2 ←→ 𝑐2 r𝑠2 , −1
(30)
I3𝑁𝑝 − Δ𝑡M−1 ⋅ h𝑠𝑓3 ←→ 𝜔.
−1.5
The mass matrix M can be different and can even be changed −1.5 −1 −0.5 0 0.5 1 1.5
during motion for each robot which enhances the flexibility. (a) Simulation environment
Random effects are included in h𝑠𝑓1 , h𝑠𝑓2 , and h𝑠𝑓3 , and all forces
must be created equivalently by local actuators in the robots.
Real robot 2
Target
4. Experiments with a Mixed Robot Swarm
NS projector
4.1. General Experiment Specifications. The experimental
environment is a 3 m × 3 m square, and the origin of the
coordinate system is set at the center. Thus, both for 𝑥-
and 𝑦-directions the experimental range is [−1.5 m 1.5 m];
see Figure 10. Obstacles and robots in the real environment Real robot 1
have a one-to-one counterpart in the simulation. We use
a red circle for representing a simulated robot and a gray
circle for the one which uses position and velocity infor-
mation from a real physical robot. The searched target is
(b) Real experimental environment
set at (−0.5 m, 0.45 m) and represented by a red pentagon.
However, the robots do not know it. Five real obstacles are Figure 10: Environment for robot swarm search.
also shown in Figure 10. The searching robots are guided
by mechanical PSO step by step; that is, the robots are
tracking the mechanical PSO generated trajectories online
based on the derived robot kinematic and dynamic models Initial robot poses can be changed and can be arbitrary.
under the help of robot position control [17]. An obstacle However, the pose is required by mechanical PSO for its
avoidance module which is able to avoid both robot-obstacle update. Therefore, the evaluation center should be informed
and robot-robot collisions is activated if there is the danger about the initial pose although the robots themselves do
of collisions. The obstacle avoidance module has a higher not need to know. If the robot has the ability to self-
priority than mechanical PSO; that is, the latter should be localize its initial position with sufficient accuracy, the user
collision free. Worth to be mentioned is that the simulation can randomly place the robot to the searched environment
and real experiment are synchronized which means we can without the necessity of manually measuring its pose. In
realize simulated robots and real robots in a mixed search this investigation we directly measure the global initial pose
simultaneously. of the “randomly placed robot” since the used robot lacks
Since the robots are searching for a target cooperatively a sufficiently accurate self-localization of its initial pose.
in the environment and the target is described in a general However, our mechanical PSO is designed for robots with
way, the performance of the fitness function in mechanical arbitrary initial poses.
PSO is represented by the distances between current robots
and target. Nevertheless, the target is unknown to the robots. 4.2. Verify Real-Simulated Interactions and Fault Tolerance.
The only change required for instantiation is to change the This experiment focuses on the real-simulated robots interac-
performance type in the fitness function. For example, in the tion and the fault tolerance ability of the developed mechan-
case of light source search, the distances should be replaced ical PSO for robots cooperative search. Thus, during the
by the strength of the detected light from each robot. search process, one simulated robot is artificially set to lose
8 Advances in Mechanical Engineering
2 3
0.5
1 4
y (m)
(a) Phase 1: initial status 0
2 3
−0.5
4
1 −1
8
1 4
0
r1 r4 (a) Phase 1: initial status
r2 r3
−10 6 3
2
5 7
−20
8
−30 1 4
0 25 50 75 100 125 150 175 200 225 250 275
Time (s)
(b) Phase 2: handling obstacles
ẋ
ẏ
6 3
2
Figure 13: Robot 1 global velocity change. 7
5
8
30 4
1
10
Velocity (mm/s)
2 6 3 7
s3 5 8
0 4
s1 1
s2
−10
2 6 3 7
−30
0 25 50 75 100 125 150 175 200 225 250 275 5 8
4
Time (s) 1
ẋ
ẏ
Figure 14: Robot 4 global velocity change. (e) Phase 5: swarm cooperation
3
2 6
along the vertical boundary of the nearby obstacle (left box). 7
5 8
Thus, the time of r3 is also a little bit delayed compared to 1
4
the time of s3. In the area of r4, robot 1 is trying to pass the
challenging narrow channel; thus it has several fluctuations.
Accordingly, robot 4 has the s1–s3 feature points. Robot 4 is
artificially set to fail after about 100 s running. From Figures
(f) Phase 6: final status
13 and 14 we also can see that the velocity of the simulated
robot is much smoother than that of the real robot. This is Figure 15: Critical phases of one of the runs with mixed simulated
mainly due to the slippage of the real robot. and real robots in a swarm.
10 Advances in Mechanical Engineering
by the German Research Foundation (DFG) within the fund- [17] P. Eberhard and Q. Tang, “Sensor data fusion for the localization
ing programme Open Access Publishing. All these supports and position control of one kind of omnidirectional mobile
are highly appreciated. robots,” in Multibody System Dynamics, Robotics and Control, J.
Gerstmayr and H. Gattringer, Eds., pp. 45–74, Springer, Wien,
Austria, 2013.
References
[1] R. Bogue, “Swarm intelligence and robotics,” Industrial Robot,
vol. 35, no. 6, pp. 488–495, 2008.
[2] N. M. P. Kakalis and Y. Ventikos, “Robotic swarm concept for
efficient oil spill confrontation,” Journal of Hazardous Materials,
vol. 154, no. 1–3, pp. 880–887, 2008.
[3] D. Fritsch, Steuerung selbstorganisierender Multi-Roboter-Sys-
teme für dynamische Sammelaufgaben am Beispiel der Bekämp-
fung maritimer Ölverschmutzungen [Ph.D. thesis], University of
Stuttgart, Stuttgart, Germany, 2009.
[4] J. M. Hereford, M. Siebold, and S. Nichols, “Using the particle
swarm optimization algorithm for robotic search applications,”
in Proceedings of the IEEE Swarm Intelligence Symposium (SIS
’07), pp. 53–59, Honolulu, Hawaii, USA, April 2007.
[5] A. T. Hayes, A. Martinoli, and R. M. Goodman, “Swarm robotic
odor localization: off-line optimization and validation with real
robots,” Robotica, vol. 21, no. 4, pp. 427–441, 2003.
[6] W. Jatmiko, K. Sekiyama, and T. Fukuda, “A PSO-based mobile
robot for odor source localization in dynamic advection-
diffusion with obstacles environment: theory, simulation and
measurement,” IEEE Computational Intelligence Magazine, vol.
2, no. 2, pp. 37–51, 2007.
[7] E. Rohmer, T. Yoshida, K. Ohno, K. Nagatani, S. Tadokoro, and
E. Konayagi, “Quince: a collaborative mobile robotic platform
for rescue robots research and development,” in Proceedings of
the 5th International Conference on Advanced Mechatronics, pp.
225–230, Osaka, Japan, October 2010.
[8] R. Quinn, “US Sending Robots to Fukushima,” 2011, http://www
.newser.com.
[9] S. Doctor, G. K. Venayagamoorthy, and V. G. Gudise, “Optimal
PSO for collective robotic search applications,” in Proceedings of
the Congress on Evolutionary Computation (CEC ’04), pp. 1390–
1395, June 2004.
[10] J. M. Hereford and M. Siebold, “Multi-robot search using a
physicallyembedded Particle Swarm Optimization,” Interna-
tional Journal of Computational Intelligence Research, vol. 4, no.
2, pp. 197–209, 2008.
[11] J. Pugh, L. Segapelli, and A. Martinoli, “Applying aspects
of multirobot search to Particle Swarm Optimization,” in
Proceedings of the 5th International Workshop on Ant Colony
Optimization and Swarm Intelligence, pp. 506–507, Brussels,
Belgium, September 2006.
[12] M. Bliesener, C. Weber, K. Kling, U. Karras, and D. Zitzmann,
Festo Robotino Manual, Festo Didactic GmbH & Co. KG,
Denkendorf, Germany, 2007.
[13] Dunkermotoren, Permanent Magnet DC-Motors, 2009.
[14] J. Kennedy and R. Eberhart, “Particle swarm optimization,”
in Proceedings of the IEEE International Conference on Neural
Networks, pp. 1942–1948, December 1995.
[15] Q. Tang and P. Eberhard, “A PSO-based algorithm designed
for a swarm of mobile robots,” Structural and Multidisciplinary
Optimization, vol. 44, no. 4, pp. 483–498, 2011.
[16] S. Xue and J. Zeng, “Controlling swarm robots for target
search in parallel and asynchronously,” International Journal of
Modelling, Identification and Control, vol. 8, no. 4, pp. 353–360,
2009.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 362317, 8 pages
http://dx.doi.org/10.1155/2013/362317
Research Article
Global Impact Dynamic Modeling and Verification of
a Flexible Beam with Large Overall Motion
Copyright © 2013 Yuechen Duan et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The theoretical modeling, numerical simulation, and experimental verification for the impact dynamics of a flexible beam with
large overall motion are researched. Based on the rigid-flexible coupling dynamic theory of flexible multibody system, the rigid-
flexible coupling dynamic equations of the beam are established. On the use of the continuous contact force method (CCFM)
and the contact constraint method (CCM), the respective global impact dynamic equations of the system are derived to achieve
dynamic transformation and solution in different stages including impact and unimpact status. The experimental study on the
impact dynamics of the flexible beam is investigated, and the dynamic response in the impact process is obtained. The experimental
results are compared with the results of the impact dynamic simulation and the finite element method (FEM) simulation. The
system’s dynamic behaviors in the impact process are analyzed, and the accuracy of the two impact dynamic theories is verified.
2 P Q
1 𝐿 𝜕2 𝑢𝑦 1 𝐿
𝑉𝐸 = ∫ 𝐸𝐼( 2 ) d𝑥 = ∫ 𝐸𝐼 (B𝑇 Φ𝑇
𝑦 Φ𝑦 B) d𝑥, C
2 0 𝜕𝑥 2 0
Z
(8a) Y
𝐿 𝐿
𝑉𝐺 = ∫ d𝑉𝐺 = ∫ − (𝜌𝑆g𝑇 W (r0 + u)) d𝑥 = −g𝑇WR. X
0 0 O
(8b)
Figure 2: Nonlinear spring damper model.
In which g is the column vector of the gravity accelera-
tion. 𝑇
The system is holonomic and q = [𝜃, B𝑇 ] are the gen- process to impact process is continuous and smooth. Thus,
eralized coordinates of the system. Let F𝑞 be the generalized the method is widely used in engineering fields, especially in
force corresponding with the nonconservative forces, which elastic impact at low speed. However, the contact parameters
is only related to the driving torque 𝜏 at the joint. Substitute of CCFM do not have a unified form and are difficult to
the system’s kinetic energy and potential energy into the obtain. The method is mostly based on the quasistatic contact
Lagrange equations as follows: theories, and the influence of the large overall motion and the
d 𝜕𝑇 𝜕𝑇 𝜕𝑉 stress wave is not considered. Meanwhile, the assumption that
( )− =− + Fq . (9) the impacting bodies penetrate into each other does not meet
d𝑡 𝜕q̇ 𝜕q 𝜕q the physical reality.
After complex derivation, as discussed by [18], the unim- In this paper, the nonlinear spring damper model is
pact rigid-flexible coupling dynamic equations of the system selected as the CCFM model. The elastic contact force is used
can be obtained as to reflect the restoring force, and the nonlinear damping force
is used to reflect the energy loss in the impact process, as
Mq̈= QΙ , (10) shown in Figure 2. The elastic contact force is along with the
normal direction of impact and is always a compression force.
where M is the generalized mass matrix, and QI is the The direction of the nonlinear damping force is opposite to
unimpact generalized force array of the system. the direction of the relative velocity of the impacting bodies.
The nonlinear spring damper model describes the
contact-impact process based on the Hertz contact theory,
3. Impact Dynamic Modeling
and the nonlinear damping force is added artificially. The
For the impact dynamics of a flexible beam with large overall expression of the total normal impact force is
motion, the large overall motion, the small deformation
motion, the impact process, and the coupling effect of 𝐹 = (𝐹𝑘 + 𝐹𝑑 ) ⋅ 𝐿 (𝛿) ,
them need to be fully considered. The continuous contact (11)
force method and the contact constraint method are used, 0, 𝛿 ≥ 0,
𝐿 (𝛿) = {
respectively, in the impact dynamic modeling of the system. 1, 𝛿 < 0,
For both methods, the corresponding impact solving part
can be added to the unimpact dynamic equations of the where 𝛿 is the penetrating capacity between the impacting
system programmatically and conveniently. The following bodies which is negative when the impacting bodies pene-
assumptions for the impact models are introduced: the trate into each other, 𝐿(𝛿) is a logic function judging contact
impact is point-to-point impact, and the size and inertia of due to 𝛿, 𝐹𝑘 is the normal elastic contact force, and 𝐹𝑑 is the
the local impacting area do not change during the impact nonlinear damping force. As discussed by [10], the concrete
process. expressions are selected as follows:
3.2. Contact Constraint Method (CCM). For CCM, contact- Then, the impact dynamic equations of the system are
impact related constraint equations are added to the system’s
dynamic equations in the impact process to solve the impact M C𝑇q q̈ Q
[ ][ ] = [ ], (20)
dynamic problem, which means that the two impacting Cq 0 𝜆 𝛾
points 𝑃 and 𝑄 on the two impacting bodies keep the same
position in the impact process, as shown in Figure 3. Then the where Cq ∈ 𝑅𝑠×𝑛 is the Jacobian matrix of contact constraints
impact dynamics are solved by adding the unilateral contact and 𝜆 is the Lagrange multiplier array which represents the
constraint equations to the original unimpact dynamic equa- contact constraint forces.
tions, and the impact dynamic equations of the system turn The numerical violation and calculation difficulty may
from ordinary differential equations (ODEs) into differential occur at the initial impact moment due to the uncoordination
algebraic equations (DAEs). The assumptions that the impact of motion by adding contact constraints directly. Therefore,
finishes instantaneously and the impacting bodies penetrate a method is needed to achieve kinematic conversion from
into each other are not needed for CCM. The essence of the unimpact status to impact status at the initial impact moment,
method is that the contact constraints are imposed on the to realize motion coordination and to get the initial impact
system according to the kinematic conditions of the impact conditions by adding contact constraints [14]. It is difficult to
process, and the impact force and the system’s dynamic obtain the initial impact conditions, and scholars do not have
response are obtained by calculating the global dynamic a unified opinion on them by far. Some methods for obtaining
equations rather than given artificially before impact. With the initial impact conditions have been reported, such as
Advances in Mechanical Engineering 5
Laser vibrometer
initial impact moment, and the coefficient of restitution 𝑒 is
set to be 0. Then, the initial impact conditions are obtained, Measuring point 2
and the kinematic conversion from unimpact status to impact
status is achieved. After IMM is used, the impacting point
𝑃 at the beam bonds to the impact surface naturally after Force sensor Measuring point 1
the normal velocity mutates, and the conditions for adding
contact constraints are met.
The impact dynamic equations of IMM are shown in [8]
as follows:
MΔq̇= D𝐼, Figure 4: Experimental schematic diagram.
(21)
𝑇
D Δq̇= − (1 + 𝑒) V0 ,
4. Experimental Verification
The experiment that a flexible beam with large rotary motion
impacts with a rigid hemisphere is carried out to verify
Figure 5: Test bench.
the accuracy of the above two impact dynamic modeling
theories.
The impact of flexible bodies is completed in a very short
time; therefore, the response measurement on the impact As shown in Figure 5, the test bench consists of the aluminum
process raises high requirements for the performance of the frame, flexible beam, cast iron base, sensors fixed standoffs,
experimental platform and the accuracy of the measuring clamping devices, position-limit beam, auxiliary wires, and
instruments. The experimental testing system of the impact fixing equipment. The initial angle and the impact velocity of
of flexible multibody system consists of the test bench, the beam are controlled by the position of the position-limit
laser vibrometer, quartz piezoelectric force sensor, condi- beam.
tioning amplifiers, calibrator of vibration sensor, electric The experimental subject is a flexible beam with large
power, Polytec Scanning Vibrometer (PSV) software system, rotary motion, and the free-fall flexible beam impacts with
and LABIVIEW processing software system. As shown in the aluminum hemisphere of the force sensor. The impact
Figure 4, the laser vibrometer is used to acquire the velocity force response at the impacting position and the normal
signal and the piezoelectric force sensor is used to acquire velocity response of three points on the beam are measured.
the force signal during impact. The beginning and ending of The aluminum beam with cylinder cross-section has the
the impact is determined by the sign of the impact force. The length 𝐿 = 0.7 m, cross-sectional radius 𝑅 = 0.0145 m,
experimental system does not need additional equipment, area moment of inertia 𝐼 = 3.47𝑒 − 8 m4 , mass density 𝜌 =
which reduces the interference in the experimental system. 2767 kg/m3 , Young’s modulus 𝐸 = 68.95 GPa, and Poisson’s
6 Advances in Mechanical Engineering
0 Acknowledgments
Research Article
Evaluating Energy Flux in Vibrofluidized Granular Bed
Copyright © 2013 N. A. Sheikh et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Granular flows require sustained input of energy for fluidization. A level of fluidization depends on the amount of heat flux provided
to the flow. In general, the dissipation of the grains upon interaction balances the heat inputs and the resultant flow patterns can
be described using hydrodynamic models. However, with the increase in packing fraction, the heat fluxes prediction of the cell
increases. Here, a comparison is made for the proposed theoretical models against the MD simulations data. It is observed that
the variation of packing fraction in the granular cell influences the heat flux at the base. For the elastic grain-base interaction, the
predictions vary appreciably compared to MD simulations, suggesting the need to accurately model the velocity distribution of
grains for averaging.
where 𝑈 is the velocity of the incoming particle, 𝑉 is the RMS energy transfer from a vibrating boundary to a granular gas.
base velocity, 𝑁 is the number of particles, 𝑔 is acceleration From the mean change in velocity squared of the particles
due to gravity, and 𝑚 is the mass of single particle. Depending following a collision, they numerically evaluated the energy
on the unknown function 𝑓(𝑈/𝑉), the form of the energy transfer integral, 𝐼𝑒 for a given base velocity magnitude.
transfer can be modelled for a vibrating surface. Reference Keeping in view (3)–(5) and by incorporating momentum
[3] showed that the total energy input through vibrating balance in the heat balance, 𝐼𝑒 can be related to the heat flux
surface can be estimated by considering that each particle through [11] in nondimensional form:
possesses mean velocity. In terms of granular temperature 𝑇,
the proposed heat flux at the vibrating base is given by 𝜂𝑜 𝐺 1 1/2 ∗
𝐽𝑏∗ = ( ) 𝐼𝑒 , (6)
(2𝜋)3/2 𝑇∗
1 𝑚 1/2
𝐸𝑏 = 𝑁𝑚𝑔𝑉2 ( ) . (4)
where 𝐽𝑏∗ is the nondimensional heat flux at the base, 𝜂𝑜 is
2 𝑇
the packing fraction, and 𝑇∗ is the non-dimensional granular
While for Maxwellian particle velocity distribution, [6] a temperature at the base (see Table 1 for details). 𝐺 is the
slightly different relationship for symmetric base profile such pair distribution function at the contact encountering the
as sinusoidal is proposed: extended volume effect.
Calculations shown by Richman [4] quantified the heat
𝑚 1/2
𝐸𝑏 = 2𝑁𝑚𝑔𝑉2 ( ) . (5) and momentum flux for the rapid granular flows of identical
𝜋𝑇 smooth spheres that interact with bumpy boundaries through
Both (4) and (5) coincide with (3) with different scale inelastic collisions. The boundaries were considered to be
factors depending on the choice of 𝑓(𝑈/𝑉). All the formu- nonstationary as well and could translate randomly with
lations rely on the continuum level description of granular given specific mean velocities along with deviations about
temperature and base densities. However, in some scenarios, the mean with specified fluctuation velocities. Richman pro-
the predictions based on continuum models are not neces- duced expressions for mean energy transfer in all the three
sarily correct. At times, the lack of separation of length and/ orthogonal directions moving with a mean square velocity
or time scales can lead to break down of continuum order of 3𝑉2 (𝑉2 for each of the direction), with 𝑉 being the
description. Specifically for granular flow in the Knudsen mean velocity of the base. The relationship includes addition
regime or rarefied flow, such description of continuum level and removal terms scaling with the square of base velocity
state variables is not necessary accurate [8, 9]. For instance, in and granular temperature. The bases of these calculations
a vibrofluidized bed, a continual input of energy is required were upon the Maxwellian velocity distribution functions
to sustain granular flow usually through the vibrating surface. that relate and describe the velocities of both the flow particles
If the order of the vibration time period matches the mean and the boundaries. The rate of exchange was also calculated
free time of the particles between collisions, the separation for the case of linear momentum and kinetic energy between
of time scales at the surface can become indistinct. In such a the granular particle and wall particle for a nearly flat surface.
case, assumption of instantaneous collisions may also become Later, Warr and Huntley showed detailed calculation for a
susceptible. triangular waveform inputting energy [3]. A detailed analysis
It can be concluded that the continuum level description considering all possible combinations of impacts with possi-
of dissipative granular flows has led to the formation of ble speeds and positions of a particle upon a vibrating base
granular flow models. However, with the increase of number is provided by Warr and Huntley. The energy was calculated
density and/or dissipation, the models tend to fail. Thus, it is using the mean change in velocity squared of the particles
necessary to evaluate continuum order description for vari- due to collision, and it was evaluated numerically through
ous flow regimes in order to establish their validity. an energy integral 𝐼𝑒 for a given base velocity magnitude.
For these calculations, the requirement is on the vibration
2. Vibrating Base-Heat period which should be much shorter than the mean time
Flux Boundary Condition between successive collisions made by a given particle and
the base. This analysis showed how the energy integral varies
The analytical foundation of the heat flux boundary con- with the speed of the incident particles and the velocity of the
ditions proposed and used in modelling granular beds has base while taking into account the coefficient of restitution
largely been developed on the argument of instantaneous between the particle and the base. Remarkably, it is shown
binary collisions. Conditions for a steady state of heat flux and that high incoming speeds of particles do not necessarily
momentum balance at a vibrating boundary have successfully guarantee high energy inputs for small base velocities. In fact,
been developed [3–6] and implemented in hydrodynamic it is also possible for incident particles to lose energy on the
models [10, 11]. One particularly effective model used in a average during an impact. This allowed direct comparison
number of simulations was developed by Richman [4] for to the work of Richman for the mean square velocity of
bumpy walls and simplifies rather neatly the limiting case of base (3𝑉2 ) and allowed a calculation of heat flux in a one-
a flat smooth vibrating wall. The results from the Richman dimensional system with flat boundaries.
method should be directly comparable with Warr’s findings Kumaran also proposed energy input expressions for
[3] in the limits of high frequency excitation. In this latter vibrated beds [6]. It gives an energy input proportional to
work, Warr and Huntley showed detailed calculations of the the mean square speed of the bed. Although it is easy to
Advances in Mechanical Engineering 3
Ie∗
𝑉∗ 𝑉/√(𝑔𝑑)
𝐼𝑒∗ 𝐼𝑒 /(𝑔𝑑)2
3. Steady State Hydrodynamic Model For high base velocity, low density, and nearly elastic
granular material, the density can be approximated as
For a vibrated granular cell, one-dimensional steady state
variation in packing fraction or density and granular temper- 𝑁𝑔 𝑔𝑧
𝜌= exp (− ) . (12)
ature can be described using 𝑇𝑜 𝑇𝑜
0.25 18
16
0.2 14
12
0.15
10
T∗
𝜂
8
0.1
6
4
0.05
0 0
0 10 20 30 0 10 20 30
z∗ z∗
0.35
0.3
0.25
0.2
𝜂
0.15
0.1
(a) (b)
0
0 20 40 60 80 100 120
z∗
(c)
Interestingly, the two regions on either sides of the peak
Figure 4: The possible collision scenarios for grain-base interaction. packing fraction are fairly similar, that is, the initial steep
increase and the following decrease of packing fraction (see
Figures 3 and 5). However, the difference in corresponding
variation of granular temperature in both regions is phenom-
a shift in its location (see hydrodynamic predictions in enal. The top layer lying above the “block” region of grains
Figure 5), only to increase the chances of sudden extended has nearly constant granular temperature, though MD sim-
recollision. At the same time, one may notice that the packing ulations suggest slight increase in temperature. The increase
fraction near the base also increases slightly. of granular temperature away from the vibrating base has
6 Advances in Mechanical Engineering
1.2
20
1
0.8 15
Jb∗
Jb∗
0.6
10
0.4
0.2 5
0
0 2 4 6 8 10 12 14 16 18 20 0
∗ 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
z
𝜂
N = 100
MD simulations
N = 300
Richman BC
N = 600
Kumaran BC
Figure 6: Variation of heat flux in the cell with the increase of gran-
ular cell packing fraction. Figure 7: Variation and comparison of heat flux at the base with the
increase of granular cell peak packing fraction at fixed base velocity
of 𝑉𝑏∗ = 1.74.
been previously observed [10]. While the bottom layer has a 5. Conclusion
sharply decreasing granular temperature as the grains move
away from the base, the nearly constant granular temperature Granular flow in one-dimensional vibrated cell is modelled
indicates that the grains in top layer hardly change their using steady state hydrodynamic model. The variation of
energies; that is, there are less collisions between grains at packing fraction and granular temperature in the cell has
that height, while the rapidly decreasing temperature suggests dependence on the heat flux imparted to the dissipative
higher rates of dissipation in the region. media. In the limit of low inelasticity and packing fraction,
The variation of heat flux throughout the granular bed is estimation of heat flux at the vibrating base is predictable.
also shown in Figure 6. As the packing fraction is increased, However, with the increase in the granular cell packing frac-
the heat flux at the base increases. At the same time, the tion, a region of high density granular material is observed
increased dissipation reduces the fluidization height, result- away from the base. The region acts like a top shield for the
ing in shifting the “maximum density point” towards the grains moving upward after receiving momentum from the
base. In such a scenario, one may notice that the calculation base. With increase in dissipation in the region, the likelihood
of averaged heat flux in vibrated cell relies on the correct of grains escaping the block is very low. In such a scenario,
description of velocity distribution in the cell. In Figure 7, a the models for the velocity distribution at the base may
comparison of heat flux predicted using MD simulations is no longer remain valid and shall be accommodated in the
drawn against the Kumaran and Richman models for elastic theoretical models. A comparison with MD simulations also
grain-base collision. One may notice that the predictions of shows appreciable deviation from the theoretical predictions.
the Kumaran as well as the Richman models match reason- Experimental validation of the presented results can help in
ably well with the results of MD simulations for low peak revealing the complex flow physics.
packing fractions. However, as the peak packing fraction
increases, the deviation between the MD simulations and the Nomenclature
Kumaran model becomes apparent. On the other hand, the
predictions of heat flux using the Richman model are slightly 𝐸𝑏 : Energy supplied at base
lower even at low densities. With the increase in the packing 𝐷pp : Dissipation (particle-particle)
fraction, results from MD simulation and the Richman 𝐷pw : Dissipation particle-side wall
model match somewhat better than Kumaran. However, the 𝑉(𝑡𝛿 ): Base velocity at time 𝑡𝛿
discrepancy is still present between the predictions of MD Δ𝑃: Particle change in momentum
simulations and theoretical models. The results suggest that 𝑈: Velocity of particle
the theoretical models need improved predictions of velocity 𝑚: Mass of particle
distributions of the grains near the base. That may provide 𝑔: Acceleration due to gravity
better estimation of heat flux at different packing fraction 𝑉: RMS base velocity
conditions. 𝑁: Number of particles
Advances in Mechanical Engineering 7
𝑇: Granular temperature
𝐽𝑏 : Heat flux at base
𝜌: Density
𝜂: Packing fraction
𝜂𝑜 : Packing fraction at base
𝐺: Pair function
𝐼𝑒 : Energy transfer integral
𝑒: Coefficient of restitution
𝑃: Pressure
𝐾: Thermal conductivity
𝐷pp : Dissipation of the grains
𝑑: Particle diameter
𝑇𝑜 : Granular temperature at the base
𝑧: Vertical height.
References
[1] S. B. Savage and M. Sayed, “Stresses developed by dry cohesion-
less granular materials sheared in an annular shear cell,” Journal
of Fluid Mechanics, vol. 142, pp. 391–430, 1984.
[2] M. W. Richman and R. E. Martin, “Confined granular flows
induced by identical, parallel, vibrating boundaries,” in Proceed-
ings of the ASME Winter Annual Meeting, pp. 49–55, December
1993.
[3] S. Warr and J. M. Huntley, “Energy input and scaling laws for a
single particle vibrating in one dimension,” Physical Review E,
vol. 52, no. 5, pp. 5596–5601, 1995.
[4] M. W. Richman, “Boundary conditions for granular flows at
randomly fluctuating bumpy boundaries,” Mechanics of Mate-
rials, vol. 16, no. 1-2, pp. 211–218, 1993.
[5] J. T. Jenkins and M. Y. Louge, “On the flux of fluctuation energy
in a collisional grain flow at a flat, frictional wall,” Physics of
Fluids, vol. 9, no. 10, pp. 2835–2840, 1997.
[6] V. Kumaran, “Temperature of a granular material “fluidized” by
external vibrations,” Physical Review E, vol. 57, no. 5, pp. 5660–
5664, 1998.
[7] J. T. Jenkins and M. W. Richman, “Boundary conditions for
plane flows of smooth, nearly elastic, circular disks,” Journal of
Fluid Mechanics, vol. 171, pp. 53–69, 1986.
[8] S. McNamara and S. Luding, “Energy flows in vibrated granular
media,” Physical Review E, vol. 58, no. 1, pp. 813–822, 1998.
[9] J. E. Galvin, C. M. Hrenya, and R. D. Wildman, “On the role
of the Knudsen layer in rapid granular flows,” Journal of Fluid
Mechanics, vol. 585, pp. 73–92, 2007.
[10] H. Viswanathan, R. D. Wildman, J. M. Huntley, and T. W.
Martin, “Comparison of kinetic theory predictions with exper-
imental results for a vibrated three-dimensional granular bed,”
Physics of Fluids, vol. 18, no. 11, Article ID 113302, 2006.
[11] T. W. Martin, J. M. Huntley, and R. D. Wildman, “Hydrody-
namic model for a vibrofluidized granular bed,” Journal of Fluid
Mechanics, vol. 535, pp. 325–345, 2005.