T* H cP 3jo .

--- ct-

NON-LINEAR VIBRATIONS, STABILITY,

AND DYNAMICS OF
STRUCTURES AND MECHANISMS
IUN

June 1-3, 1988

Abstracts

Sponsored by
Army Research Office

Donaldson Brown Continuing Education Center

Virginia Polytechnic Institute and State University

Chairmen: A. H. Nayfeh and D. T. Mook

Department of Engineering DTIC
Science and Mechanics ELECT_
Virginia Polytechnic Institute SEP 121989
and State University Sr
Blacksburg, VA 24061 B

IIWd1 __pmbUc r.*-mq

 UNCLASSIFIED MASTER COPY -FOP. REPRODUCTION PURPOSES
SECURITY CLASSIFICATION OF TH15 PAGE

REPORT DOCUMENTATION PAGE

I&. REPORT SECURITY CLASSIFICATION lb. RESTRICTIVE MARKINGS
Unr1_aqqtif1 A_________________________
2a. SECURITY CLASSIFICATIO AUTNORITY 3. DISTRIBUTION/ AVJLAAIUTY OF REPORT
ECLSSIFCATON
/OWNRADIG
2b. SHEDUE--Approved for public release;
2b.DOWGRAING
ECLSSIFCATON
CHEULEdistribution unilimi ted.
4. PERFORMING ORGANIZATION REPORT NUMBER(S) 5. MONITORING ORGANIZATION REPORT NUMBER(S)

ARO 25614.1-EG-CF
6t. NAME OF PERFORMING ORGANIZATION 6ib. OFFICE SYMBOL 7a. NAME OF MONITORING ORGANIZATIN

Va Polytechnic Inst & State U 1 (if___________ U. S. Army Research Office

6c. ADDRESS (01y, State, and ZIP Code) 7b. ADDRESS (0ty, State, and ZIP Code)

Blacksburg, VA 24061 P. 0. Box 12211

Research Triangie'Park, NC 27709-2211
S.. NAME OF FUNDING /SPONSORING Sb. OFFICESYMBOL 9. PROCUREMENT INSTRUMENT IDENTIFICATION NUMBER

U. S. Army Research Office DAAL03-88-G-0013

Sc ADDRESS (City, State, anid ZIP Code) 10. SOURCE OF FUNDING NUMBERS
P.0.Bx 221PROGRAM NO.
PROJECT
NO.
TASK
NO0.
IWORK UNIT
P0.Bx121ELEMENT CCESSION NO.
Research Triangle Park, NC 27709-2211
11. TITLE (include Security Clasficktion)

Non-Linear Vibrations, Stability, and Dynamics Of Structures and Mechanisms

12. PERSONAL, AUTHORS
A. F.Nyeh and D. T. Hook
13a. TYPE OF REPORT 13b. TINE COV?)fEE 14TAE FRPOZYFa5/thDy3S1AE O.
Final ~~FROM~'' &3/9 Ags i O '

16.SUPLMETAY 40ATONThe view, opinions and/or findings contained in this report are those
of h aut
shuld or e cosr
r().an as n ficia D rtment of the Army position,

19. ABSTRACT19SBJCTTEM (Continue on reverse if necefary and identify by block niumber)

The conference was held as scheduled. Abstracts of the papers presented at the conference
are contained in a report distributed by the Virginia Polytechnic Institute and State

20. DISTRIBUTION /AVAILABILITY OF ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION

[3UNCLASSIFIEDILUNUMITED 03 SAME AS RPT. O3DTIC USERS Unclassified
22a. NAME OF RESPONSIBLE INDIVIDUAL I22b. TELEPHONE (linclude Area Code) I22c. OFFICE SYMBOL

DD FORM 1473,a&4MAR 53 APR edition may be used until exhausted. SECURITY CLASSIFICATION OF THIS PAGE
All other editions are obsolete. UNCLASS IFIED
I
I
I
NON-LINEAR VIBRATIONS, STABILITY,
AND DYNAMICS OF
I STRUCTURES AND MECHANISMS
/
June 1-3, 1988
I
II Abstracts
!
I Sponsored by
Army Research Office

IDonaldson Brown Continuing Education Center

Virginia Polytechnic Institute and State University

Chairmen: A. H. Nayfeh and D. T. Mook

Department of Engineering
Science and Mechanics
Virginia Polytechnic Institute
and State University
II Blacksburg, VA 24061
I
I
i Reception and Registration at CEC Tuesday, 1900-2100, May 31

Session 1: Chaotic Motions Wednesday, AM, June 1

Chairman: R. Miller, Air Force Office of Scientific Research

0800 OPENING REMARKS, A. H. Nayfeh and 0. T. Mook, Virginia Polytechnic Institute and
State University

WELCOME, P. Torgersen, Interim President and Dean, College of Engineering, Virginia

5 (Polytechnic Institute and State University
0830 K-CHAOTIC OSCILLATIONS IN MECHANICAL SYSTEMS)>
/ E. H. Dowell, Duke University

0900 "INTERPOLATED MAPPING TECHNIQUES", B. H, Tongue, Georgia Institute of Technology

0915-)N UNDERSTANDING CHAOS IN NONLINEAR OSCILLATORS HAVING A SINGLE EQUILIBRIUM

I W 3
POSITION,
arsa W. Szemplinska-Stupnicka and P. Niezgodski, Polish Academy of Sciences,
l and

0930 Lf'IFURCATIONS IN A FORCED SOFTENING DUFFING OSCILLATOR' A. H. Nayfeh and N. E.

3 4
Sanchez, Virginia Polytechnic Institute and State Univer;Ity

0945 "EXTENSIONS AND NEW APPLICATIONS OF MELNIKOV'S METHOD FOR PREDICTING THE ONSET OF
CHAOS, S. W. Shaw, Michigan State University

1000 Rensselaer OF CHAOS BY

!-SUPPRESSIONPolytechnic NONLINEAR
Institute U. S. Army I. Research
and DAMPING", Office and G. L. Anderson,
G. Tadjbakhsh

I 1015 Coffee Break

Session 2: Chaotic Motions Wednesday, AN, June 1

Chairman: G. Anderson, U. S. Army Research Office

1045 "SYMMETRY BREAKING BIFURCATIONS IN MECHANICAL SYSTEMS AND IN SURFACE WAVES-, P. R.

Sethna, University of Minnesota
1115 "ASYMPTOTIC TECHNIQUES AND CHAOS IN WEAKLY NONLINEAR FORCED MECHANICAL SYSTEMS", A.
K. Bajaj, Purdue University

3 1130 "NONLINEAR NONPLANAR PARAMETRIC RESPONSES OF AN INEXTENSIONAL BEAM", A. H. Nayfeh

and P. F. Pal, Virginia Polytechnic Institute and State Univer-';

1145-- $MOOAL INTERACTION IN THE RESPONSE OF SHELLS TO A SUBHARMON .XCITATIONN, A. H.

Nayfeh, R. A. Raouf, and J. F. Nayfeh, Virginia Polytechnic Lnstitute and State
University

1200 "THE EVALUATION OF THE REGION OF CHAOTIC MOTION IN CURVED STRUCTURES-, C. F. Ng, C]
NASA Langley Research Center u1
1215 "PARAMETER IDENTIFICATION IN CHAOTIC DYNAMIC SYSTEMS", 0. J. Mook and P-H. Tong,
State University of New York at Buffalo
1230 Lunch at CEC , '-bUL10D/_
3L --
AvaiabiityCodes

S013t~ j
Special
"
1/
II
Session 3: Nonstatlonary and Random Vibrations Wednesday, PM, June 1
Chairman: M. P. Singh, Virginia Polytechnic Institute and State University

1330 "NONLINEAR NONSTATIONARY PROCESSES-, R. M. Evan-Iwanowski and G. L. Ostiguy,

University of Central Florida and Ecole Polytechnique, Montreal, Canada

1400 "RESPONSE OF A SINGLE-DEGREE-OF-FREEDOM SYSTEM TO A NONSTATIONARY PARAMETRIC

EXCITATION - THEORY AND EXPERIMENT , A. H. Nayfeh and H. L. Neal, Virginia
Polytechnic Institute and State University

1415 "ON THE PROBABILITY DENSITY FUNCTION OF THE RESPONSE OF NARROW-BANZ" EXCITED
NONLINEAR OSCILLATIONS", Q. Liu and H. G. Davies, University of New Brunswick,
Canada

1430 "WIDE BAND RANDOM EXCITATION OF A THREE-DEGREE-OF-FREEDOM SYSTEM WITH PRINCIPAL

INTERNAL RESONANCES", R. A. Ibrahim and W. Li, Wayne State University

1445 NA GENERALIZED METHOD OF AVERAGING FOR DETERMINING THE RESPONSE OF NONLINEAR SYSTEMS
TO RANDOM EXCITATIONSO, A. H. Nayfah and S. J. Serhan, Virginia Polytechnic
Institute and State University

1500- >DYNAMIC SNAP-BUCKLING UNDER STOCHASTIC LOADS, S. T. Ariaratnam and W-C. Xie,
University of Waterloo, Canada

1515 "A NEW APPROACH TO STOCHASTIC FLAP-LAG STABILITY OF A ROTOR BLADE IN HOVER, I N. S.
Namachchivaya and J. Prussing, University of Illinois at Urbana-Champaign

1530 Coffee Break

Session 4: Nonlinear Vibration Control Wednesday, PM, June 1

Chairman: E. H. Dowell, Duke University

1600 "NONLINEAR OSCILLATIONS OF LARGE SYSTEMS WITH LOCALIZED NONLINEARITIES', P.

Hagedorn, Technische Hochschule, Darmstadt, FRG

1630 "ON THE ADAPTIVE CONTROL OF DYNAMIC SYSTEMS WITH FLEXIBLE STRUCTURES"; R. L. Kosut,
Integrated Systems, Inc., Santa Clara, CA

1645 "ON THE STABILIZATION OF TETHERED SATELLITE SYSTEMS", E. H. Abed and 0. -C. Liaw,
University of Maryland

1700 "A MATHEMATICAL FORMULATION AND CONTROL EXPRESSIONS FOR COMPLIANT CONTROL OF
CONSTRAINED ROBOT MANIPULATORS", C. S. Yoon and F. M. A. Salam, Michigan State
University

1715 "FLEXIBLE ROBOT MODELS WITH REVOLUTE AND PRISMATIC JOINTS - HANDLING OF CLOSED
LOOPS, M. Riemer and J. Wauer, Universitat Karlsruhe, FRG

1730 "STABILITY OF FLEXIBLE MULTIBOOY SYSTEMS DYNAMICS IN THE PRESENCE OF CLOSED LOOPS",
F. M. L. Amirouche, University of Illinois at Chicago

1930 Dinner at CEC

i

mirm mmmm mm mmmmm lmm

mmm
nmm I mm
I
I Session 5: Experiments in Nonlinear Dynamics Thursday, AM, June 2
Chairman: R. Evan-Iwanowski, University of Central Florida

0830 OFRACTALS AND CHAOS IN ELASTIC SYSTEMS8, F. C. Moon, Cornell University

0900 -'NONLINEAR OSCILLATION OF A FLEXIBLE CANTILEVER: EXPERIMENTAL RESULTS5 ," A. G.
I Haddow and S. M. Hasan, Michigan State University
0915 "NONLINEAR RESONANCES AND CHAOS IN A PARAMETRICALLY EXCITED FLEXIBLE BEAMS', T. D.
5 Burton, Washington State University
0930 "THE NONLINEAR RESPONSE OF A SLENDER BEAM CARRYING A LUMPED MASS TO A PRINCIPAL
PARAMETRIC EXCITATION', L. 0. Zavodney and A. H. Nayfeh, Ohio State University and
I Virginia Polytechnic Institute and State University
0945 "THEORETICAL AND EXPERIMENTAL INVESTIGATION OF COMPLICATED RESPONSES OF A TWO-
DEGREE-OF-FREEDOM STRUCTURE', A. H. Nayfeh, B. Balachandran, M. A. Colbert, and M.
A. Nayfeh, Virginia Polytechnic Institute and State University

I 1000 Coffee Break

IChairman:
Session 6: Nonlinear Analysis of Beams and Rotor Blades Thursday, AM, June 2
R. A. Wehage, US Army Tank-Automotive Command, Warren, MI

1030 NONLINEAR OSCILLATIONS IN A ROTATING SHAFT SYSTEM', T. Yamamoto and Y. Ishida,

MeiJo University and Nagoya University, Japan

I 1045- 'NONLINEAR EFFECTS IN THE STATIC AND DYNAMIC BEHAVIOR OF BEAMS AND ROTOR BLADES' 0.
H. Hodges, M. R. M. Crespo da Silva, and 0. A. Peters, Georgia Institute of
Technology and Rensselaer Polytechnic Institute

1100 *NONLINEAR FLEXURAL-FLEXURAL-TORSIONAL EXTENSIONAL DYNAMICS OF BEAMS: FORMULATION

AND RESPONSE-, M. R. M. Crespo da Silva, Rensselaer Polytechnic Institute

I 1115 *NONLINEAR MODAL COUPLING IN THE RESPONSE OF INEXTENSIONAL BEAMS', M. R. M. Crespo

da Silva and C. L. Zaretzky, Rensselaer Polytechnic Institute
1130 "MODAL INTERACTIONS IN THE
RESPONSE OF BEAMS TO A HARMONIC EXCITATION-, J. F.
Nayfeh, A. H. Nayfeh and 0. T. Mook, Virginia Polytechnic Institute and State
University

I 1145 *ANALYSIS AND DESIGN SENSITIVITY CALCULATION OF NONLINEAR FORCED VIBRATION OF

STEPPED BEAMS', J. W. Hou, Y. X. Xue, C. Mei and C. Jackson, Old Dominion University
1200 ONON LINEAR OSCILLATIONS AND LIMIT CYCLES - A TIME FINITE ELEMENT APPROACH', M.
Borri and P. Mantegazza, Georgia Institute of Technology

1215 *DYNAMICS OF BEAMS WITH TIP MASSES AND ATTACHED TO A MOVING BASE, S. Hanagud and S.
Sarkar, Georgia Institute of Technology

L 1230 Lunch at CEC

I"
I
 I
I
Session 7: Nonlinear Analysis of Plates and Shells Thursday, AN, June 2
Chairman: 0. Frederick, Virginia Polytechnic Inst" "e and State University m

1330 NON-LINEAR OSCILLATIONS OF AN INEXTENSIBLE, AIR-INFLATED, CYLINDRICAL MEMBRANEN, R.

H. Plaut, Virginia Polytechnic Institute and State University

1345 "IMPACT OF SHELLS, C. R. Steele, Stanford University 1

1400 "DYNAMIC STABILITY OF AN ANNULAR SECTOR PLATE SUBJECTED TO INPLANE DYNAMIC NOMENTS, I
K. Takahashi, M. Hirakawa and Y. Natsuaki, Nagasaki University, Japan 5
1415 NONLINEAR DYNAMIC STABILITY OF LAMINATED COMPOSITE SHEAR BEFORMABLE FLAT PANELS
SUBJECTED TO IN-PLANE PERIODIC EDGE LOADS', L. Librescu and S. Thangjitham, Virginia
Polytechnic Institute and State University
1430 "THE METHOD OF MULTIPLE SCALES FOR NONLINEAR RESONANCES IN THE FORCED RESPONSE OF
ORTHOTROPIC RECTANGULAR PLATES, H. Eslami and 0. A. Kandil, Embry-Riddle University 1
and Old Dominion University

14454'A FINITE ELEMENT METHOD FOR NONLINEAR PANEL FLUTTER OF COMPOSITE LAMINATES', C. Mel
and C. E Gray, Jr., Old Dominion University anm NASA Langley Research Center I
1500 DYNAMIC STABILITY OF LAMINATED COMPOSITE SHELLS USING A SHEAR DEFORMATION THEORY",
E. Yogeswaren and J. N. Reddy, Virginia Polytechnic Institute and State University 1
1515 Coffee Break
I
Session 8: Miscellaneous Topics in Nonlinear Dynamics Thursday, PH, June 2
Chairman: F. C. Moon, Cornell University
1545--'NLANAR RESPONSE OF ELASTIC CABLES.TO A SUBHARMONIC OR SUPERKARMONIC EXCITATION', F.
Benedettini and G. Rega, Universita dell Aquila, Italy

1600 *NONLINEAR OSCILLATIONS UNDER TWO-FREQUENCY PARAMETRIC EXCITATION-, R. H. Plaut, J. I

J. Gentry, and 0. T. Mook, Virginia Polytechnic Institute and State University
1615 "THE RESPONSE OF SEISMICALLY MOUNTED ROTOR SYSTEM WITH CUBIC NON-LINEARITIES-, A. I
Ertas and E. K. Chew, Texas Tech University
1630 "QUENCHING OF SELF-EXCITED VIBRATIONS", K. R. Asfar, Jordan University of Science
and Technology, Irbid, Jordan
1645 'EFFICIENT SOLUTION OF CONSTRAINED EQUATIONS OF MOTION BY RECURSIVE PROJECTION-, R.
A. Wehage, US Army Tank-Automotive Command, Warren, MI I
1700 "GLOBALLY CONVERGENT NUMERICAL ALGORITHMS FOR NONLINEAR SYSTEMS OF EQUATIONS", L. T.
Watson, Virginia Polytechnic Institute and State University -f

1900 Banquet at Marriott Hotel

"CHANGING PERCEPTION OF SCIENTISTS AND ENGINEERS IN SOCIETY", M. Salkind, Air Force
Office of Scientific Research

I
i I-- l II• ll /ll I
I
I Session 9: Multibody Dynmics Friday, AM, June 3
Chairman: E. L. Marsh, National Science Foundation

3 0830 "SURVEY ON THE GENERATION AND SOLUTION OF MULTIBODY SYSTEM EQUATIONS',

Schwertassek and C. Fuhrer, DFLVR and Technische Universitat Munchen, FRG
R.

' 0900 "ON THE USE OF THE FINITE ELEMENT METHOD AND CLASSICAL APPROXIMATION TECHNIQUES IN
THE NONLINEAR DYNAMICS OF MULTIBOOY SYSTEMS', A. A. Shabana, University of Illinois
at Chicago

I0915 COMPARATIVE STUDY OF METHODOLOGIES EMPLOYED IN CONSTRAINED MULTIBOOY DYNAMICS',

M. L. Amirouche, University of Illinois at Chicago
F.

0930 "PRIMAL AND MIXED FORMS OF HAMILTON'S PRINCIPLE FOR CONSTRAINED RIGID AND FLEXIBLE
DYNAMICAL SYSTEMS: NUMERICAL STUDIES', M. Borri, F. Mello, M. lura and S. N. Atluri,
Georgia Institute of Technology

I 0945"'ANALYSIS OF NONLINEARLY CONSTRAINED, NONHOLONOMIC MULTIBODY SYSTEMS', M. Xu, C.

<7 Liu, and R. C. Huston, Chongqing University, China, University of Cincinnati

I 1000 Coffee Break

Session 10: ultitbody Dynamics Friday, AM, June 3

Chairman: R. C. Huston, University of Cincinnati

1030 *NONLINEAR ANALYSIS OF LOSS OF STABILITY OF PERIODIC SOLUTIONS WITH AN APPLICATION

TO ROBOTIC MOTIONS', E. Lindtner, A. Steindl, H. Troger, Technische Universitat,
Vienna, Austria

'SYSTEMATIC REDUCTION OF MULTIBOOY EQUATIONS OF MOTION TO A MINIMAL SET',

3 1100
Nikravesh, University of Arizona
P. E.

5 1115 'USE OF LINEAR AND NONLINEAR STRUCTURAL THEORIES IN FLEXIBLE MULTIBODY DYNAMICS', R.
R. Ryan and H. H. Yoo, University of Michigan
1130 ."NONLINEAR LARGE ROTATIONAL STRUCTURAL DYNAMICS', M. Iura and S. N. Atluri, Georgia
Institute of Technology
1145 "DYNAMICS AND BIFURCATIONS OF ELASTIC SPACE STRUCTURES', M. Levi, Boston University

I 1200 Lunch at CEC

I
I
I'
I
I
Session 11: Friction and Damping Friday, PM, June 3
Chairman: T. L. Herdman, Virginia Polytechnic Institute and State University

1300 "THE EFFECT OF SOFTENING OF THE SUPPORTS ON THE STABILITY AND DYNAMICS OF
STRUCTURES', M. A. Souza, Pontificia Universidade Catolica do Rio de Janeiro, Brasil
1315 OJOINT DAMPING AND JOINT NONLINEARITY EFFECTS IN DYNAMICS OF SPACE STRUCTURES% M.
Bowden and J. Dugundjl, Massachusetts Institute of Technology

1330 'EFFECTS OF VELOCITY-DEPENDENT FRICTION ON PERFORMANCE OF A RESILIENT-FRICTION BASE

ISOLATOR, L. Su, G. Ahmadi and I. G. Tadjbakhsh, Clarkson University, Rensselaer
Polytechnic Institute 3
1345 *NONLINEAR DYNAMIC RESPONSE OF FRACTIONALLY DAMPED STRUCTURAL SYSTEMS', J. Padovan,
University of Akron

1400 'PERIODIC RESPONSE OF A CLASS OR HYSTERETIC OSCILLATORS',

Vestroni, Universita' dell' Aquila, Italy
0. Capecchi and F. m
1415 "DYNAMICS OF NONLINEAR AUTOMOBILE SHOCK-ABSORBERS', J. Wallaschek, Technische I
Hochschule, Darmstadt, FRG
U
I
', I
I

mI I I I II I
 SESO
CHOI3OIN
WENSAI80-11
Jue1,18
 CHAOTIC OSCILLATIONS IN
MECHANICAL SYSTEMS

I Earl H. Dowell

I Duke University

5 ABSTRACT

Chaotic oscillations have now been observed in

nonlinear mechanical systems by analytical,
numerical and experimental methods. Nevertheless
a more fundamental understanding of why and when
such oscillations occur is of great importance.
This goal will be pursued here by considering the
relationship between chaos induced by forced
oscillations vs self-excited oscillations, the
relationship of indeterminancy of the final
equilibrium state in the initial value problem to
chaos in the sustainei oscillation problem, com-
parison of theory to physical experiment,
necessary and sufficient conditions for chaos to
occur, and the question of convergence of systems
of modal ordinary differential equations which
derive from partial differential equations.

Acknowledgement: This work was supported in part

by the Army Research Office, Contract DAAL03-
87K0023, and the National Science Foundation,
Grant MSM-8504105. Drs. Gary Anderson and
Elbert Marsh are the program directors. This
paper is a chapter from a forthcoming book,
"Nonlinear Aeroelasticity," to be published by
Springer-Verlag in 1988. The co-authors are
Earl H. Dowell and Marat Ilgamov.

£
I
I
I
I
 I
I
INTERPOLATED vA.PPING TECHNIQUES*

Benson H. Tongue

The George W. Woodruff

School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, GA 30332

Introduction. In order to completely characterize the response of nonlinear systems, it

is often necessary to calculate time responses for many different initial conditions. This is
due to the well known fact that nonlinear systems often support several distinct steady state
solutions depending upon the particular initial conditions used [1]. In addition, one often wishes
to determine the system response under a variety of system parameter values. Unfortunately,
exploring a system's responses in this way is extremely time consuming due to the extremely
I
large number of possible trajectories that have to be simulated.
In an attempt to alleviate this problem, Hsu and co-workers [2-71 introduced two new
methods of analysis. The central idea, common to both techniques, is to treat phase space not
as a continuum, but as a finite set of discrete states. The way in which this is done is to divide
phase space into a rectangular array of "cells". In the Simple Cell approach, the center of each
cell is used as the initial condition for every state that lies within that particular cell. The cell
that this trajectory terminates in is recorded as the target cell for all points lying within the
original cell. In this way the continuous system is reduced to that of a discrete mapping The
difficulty arises from the fact that continuous trajectories cannot be generated due to the step
I
discontinuities that occur at the beginning of each mapping. This introduces a distortion into
the phase flow that effectively eliminates the possibility of accurately determining long term
trajectories or local stability characteristics.
I
The second approach, Generaiied Cell Mapping, constructs a probabilistic characterization
of the phase flow by utilizing several trajectories in each cell and determining what fraction of i
these end in particular target cells. Based on the distribution of these target cells, a probabilis-
tic mapping is constructed which allows one to determine steady state responses via Markov
chain theory. This technique also is hampered by the impossibility of determining continuous
trajectories for a given system.
The topic of the present work is an approach (Interpoiated Mapping) that allows an accu-
rate global determination of a system's dynamical respunses without the restrictions that have
I
hampered previous methods [8-111. The notable feature of this methodology is that instead of
viewing phase space as a discrete array that maps onto itself under the system's dynamics, the
notion of a continuous phase space is retained. The continuous deformation of state space under
the action of the governing dynamical equations is approximated from a knowledge of the exact
reponse of a finite number of points. This information is then used along with an interpolation
procedure to generate complete trajectories for any given initial conditions within the confines
of the original array.
In the simplest application of the method, the points are arranged within a rectangular 3
* This work was supported by the National Science Foundation, Grant No. MSM-8451186.

i
 grid. Trajectories are computed for all of the points for a fixed time duration, A. If the system
is periodically forced, the natural value for A will be the fundamental period of the excitation.
The terminal position of earh trajectory is then recorded. This information forms the basis of a
I discrete mapping for the system from to to to + A.
Figure 1 illustrates how an initial point, A, maps to the new position, B. To begin
the iteration procedure, the relative position of a trajectory's initial position is determined
with resDect to the four array elements forming the vertices of a rectangle about this point
(C11, C1 2 , C21 and C22 in Figure la). The location of this point after another A increment of
time is found by interpolating between the final locations of the trajectories emanating from the

Cu 's. The relative position of the point A to its surrounding array is shown in Figure la. It is
displaced a distance x1 . to the right from the leftmost array element and a distance x2, up from
the lowermost array element. One mapping later, the relative positions have changed to those
shown in Figure lb. The new position of the discrete trajectory is given by the intersection of
the two line segments (a,b) - (e,f) and (c,d) - (g,h). Extending the trajectory simply involves a
repeated application of the interpolation technique.
jIt is important to observe that the individual segments of a given total trajectory are unique
to that particular trajectory and will not occur in any other trajectory. Thus, even though a finite
number of array elements is used initially, an infinite number of trajectories can be generated, not
just a finite number of cell to cell mappings, as is the case for other existing mapping methods.
Since no overt discretization is imposed on the possible states of the system, arbitrarily long

3technique
trajectories can be computed. The ability to generate a continuum of trajectories allows the
to efficiently be used in bifurcation studies and in determinations of a system's basins
of attraction.
Because previously examined methods of iterative mapping enforced a step discontinuity
in the state of the systems at each mapping interval, errors in the system response were made
unavoidable. The present technique avoids this difficulty. Thus one is free to investigate relative
stability, generate Lyapunov exponents, etc., at a very low cost and with a very high degree of
accuracy. As an example, the basins of attraction and associated attractors for the equations

i2 = -. 02x 2 - .25x, - x' + 8.5cos(t)

are shown in Figures 2 (exact calculation) and 3 (Interpolated Mapping). Note that the two
cases are almost indistinguishable.
The potential impact of these techniques is sizable. Until now, global nonlinear analyses
have been largely confined to single degree of freedom systems because of excessive computer
time requirements. For example, the calculation of the basins of attraction for a simple nonlinear

3 pendulum can require literally hours of time on a supercomputer. Interpolated Mapping allows
the same calculations to be done in just minutes of run time on a minicomputer. The implication
is that one can do all of the currently performed analyses in a greatly accelerated fashion or,
more importantly, attempt complicated, multi-degree of freedom analyses for the first time. The
talk will focus on different types of interpolation strategies, dynamical limitations and future
applications of the method.

I
I
I
 REFERENCES
[1] Nayfeh, A. H. and Mook, D. T., Nonlinear Oscillations, New York: Wiley Interscience
[2] Hsu C. S., 1980, "A Theory of Cell-to-Cell Mapping Dynamical Systems," ASME Journal
of Applied Mechanics, 47, December, 931
[3] Hsu, C.S., and Guttalu, R.S., 1980, -An Unraveling Algorithm for Global Analysis of Dy-
namical Systems: An Application of Cell-to-CeU Mappings," ASME Journal of Applied
Mechanics, 47, December, 940
[4] Hsu C. S., Guttalu, R. S. and Zhu, W. H., 1982, "A Method of Analyzing Generalized Cell
Mappings," ASME Journal of Applied Mechanics, 49, December, 885
[5] Hsu, C.S., 1982, "A Probabilistic Theory of Nonlinear Dynamical Systems Based on the
Cell Space Concept," ASME Journal of Applied Mechanics, 49, December, 895
[6] Hsu, C.S. and Kim, M.C., 1985, 'Statistics of Strange Attractors by Generalized Cell Map-
ping," Journal of Statistical Physics, 38, 735
[7] Hsu, C.S. and Chiu, H.M., 1987, 'Global Analysis of a System with Multiple Responses
Including a Strange Attractor," Journal of Sound and Vibration, 114, No. 2, April, 203
[8] Tongue, B.H., 1987 "On the Global Analysis of Nonlinear Systems Through Interpolated
Cell Mapping," Physica 28D, No. 3, October, 401
[9] Tongue, B. H., "Interpolated Cell Mapping of Dynamical Systems," to appear in the ASME
Journalof Applied Mechanics, 1988
[10] Tongue, B. H. and Gu, K., "A Theoretical Basis for Interpolated Cell Mapping," to appear
in the SIAM Journal on Applied Mathematics, October 1988
[11] Tongue, B. H., "A Higher Order Method of Interpolated Cell Mapping," to appear in the
Journal of Sound and Vibration, 1988

, I (g~h..

(a) (b)

FIGURE 1
3LEGEND *converge to P-3
LEGEND
* attractor I p-3
*converge to strange attractor ottroctor2 strange attractor

X2X2~ 404~:

j.

'1.5 18 * 2.1 2.4 2.7 Y. 16 4.2 1'.& Z.1 ;f.4 17 X3 0 1 - S9

X
(a) ,()

ILEGEND converge to p-3 *

LEGEND
attractor I p-3
Converge to strang. attractor attractor2 strong. attractor

-------
-0- - - - - - - -

- - - - - - - - - -

It

15 4 182.1
2 0 3. 3 42 1.5 1.8 2. 12.4 2. 0P .3 .S.63.

(a) X, (b) X

U FIGURE 3
 On Understanding Chaos in Nonlinear Oscillators
Having a Single Equilibrium Position
W. Szemplinska-Stupnicka* and P. Niezgodzki
Institute of Fundamental Technological Research
PAS, Swietokrzyska 21, 00-049 Warsaw, Poland.

Introduction
This paper is concerned with the chaotic response of a single-
degree-of-freedom system governed by the following equation of motion
x +
2
w0x + hi + a IX
2
+ a 2x
3
P cosvt, 2n
= = - (1)

where h > 0 and the coefficients of the quadratic and cubic nonlinear
terms, a and a , satisfy the condition that x = 0 is the unique stable
rest position ai P = 0. Equations of this type arise in the analysis of
large-amplitude vibrations of flexible structures when a single-mode
solution is assumed. The behavior of the system is studied extensively
by approximate analytical methods.
Since the year 1979, however, when a distinctly new type of steady-
state solution known as chaotic motion was obtained by computer
simulation, it became clear that our knowledge about the system behavior
is far from complete. While chaotic motion in systems having two stable
equlibrium positions (two-well potential system) can be intuitively
explained by some physical arguments, the arguments fail in the system
for which x = 0 is the only stable rest point.

In the present paper some attempts are made to provide a physical

interpretation of chaotic motion in this class of systems by using
arguments and mathematical tools based on the approximate theory. This
became possible because of observations made by numerous authors that
zones of chaotic motions separate two different types of periodic
responses. Usually qT-periodic and T-periodic solutions are involved.

Computer-Simulation Results and the Concepts of the "Filtered" Chaotic

Component of Motion
A typical example is sketched in Fig. 1. The frequency spectrum of
x(t) in Fig. la corresoonds to a subultraharmonic resonance, and the qT
periodic components are denoted by dashed lines. The solution can be
approximately described as
x(t) = x(t + qT) = Z Ancos(nvt + tn +
A cos(R vt + .p)

nnp q p
n=0,1,2... p=1,2,3... (2)

* Currently Visiting Professor, Department of Engineering Science and

Mechanics, Virginia Polytechnic Institute and State University,
Blacksburg, VA 24061.
I
where the first term contains harmonic components with period T, and the
second comprises the harmonic components of period qT. On varying the
frequency v we arrive at the stability limit of this type of solution,
and eventually to the T-periodic solution shown in Fig. ic. If the
forcing parameter P is sufficiently smd!7, the decay of the qT-periodic
component occurs at certain v=v., and hence only the transient motion
and the associated jump phenomena are observed. At higher values of P,
however, the decay of the qT-periodic component is preceded by a zone of
chaotic motion. The most essential observation that comes from the
frequency spectrum of the chaotic solution (Fig. ib) is that continuous
segments of the spectrum spread only around those harmonic components,
which are due to decay. Because only the continous portion of the
frequency spectrum contributes to the chaotic motion, a conclusion was
drawn that the time histories of the individual chaotic components of
the response might give insight into the nature of the response.

To make such an analysis possible, a numerical technique was

developed to obtain the "filtered" response i(t), i.e., the component of
the complete signal x(t), that corresponds to the individual continuous
i segment of the frequency spectrum, denoted by av in Fig. lb.
An analysis of many samples of x(t) clearly indicated that it can
be approximated by a single-harmonic component with the amplitude
fluctuating randomly with time and the frequency being close to the
middle frequency, but depending on the amplitude: [Fig. 2 and 31

i (t) = a(t)cos[R(a)t + .l (3)

This brings an appealing idea that the "chaotic component" of motion
3 here can be interpreted as a sort of "free-vibration" motion.

Secondary Resonances-Theoretical Analysis of the Stability Limits

SIn the approximate study of secondary resonances we start with the

general solution of the linear, undamped system, Eqs. (i) for f(t) h =
x(O).t = a cos(w 0 t + o) + Afcosvt,
Af 2P (4)
0

where a, p are arbitrary constants depending on initial conditions,

and v being far from wo"

If the damping terms is included, the free- vibration comoonent

decays with time. Where the complete nonlinear Eq. (1) is considered,
however, it may happen that the "free-vibration" term is sustained,
giving rise to the first approximate stable solution

x(t) = a cos(wt + t) + Afcos(Vt + u), (5)

Sda do 0 a 0 (6)
dt dt
The necessary condition for such a solution, and hence for the secondary
resonance, to exist is that the frequencies be commensurable:

I
 I I
l
Aft A I

3Y y 3Y" y S
Figure 1. 5
iU)

Figure 2. Example I. Figure 3. Example II.

aI
0.

0.6
0.4 *, Figure 4. Exanple 1. II

.0.2

I
|I
I
W RV, p, q = 1, 2, 3 ... p * q (7)
and hence the steady-state solution (5) be qT-periodic. The loss of
stability of the solution can be interpreted as the loss of the
frequency entrainment (6).

I The approximate methods lead immediately to the conclusion that the

free-vibration component decays at the stability limit, a - 0 as t - =.
The time history of the filtered response, however, suggests another
type of instability, the instability for which the motion exhibits
random-like fluctuation of the amplitude a in a certain zone of v.

3 To verify this idea we have to compare the frequency-amplitude

relationships obtained from the~filtered response with the theoretical
ones. To find the theoretical w = w(a) at the stability limit we
rewrite Eq. (5) in a more general form:
on )
x(t) = z(t) + xf(t), where xf(t) = =On,.. ncos(nvt +
An (9)

and insert it into Eq. (1).

z(a):
3 + zw2 + 2al(A + AnCose ) + 3a( n A
A2n + 1 2nAcS2enn
h o n=112...n2

+ JAnAmCossnCosam)I + z [a, + 3a 2 (A0 + JAnCosen) + a 2z = F(vt)

where 9n = nvt + , n = 1,2... and F(vt) stands for all time-dependent
terms. Eventuallynneglecting all time-dependent terms, we find that the
"free-vibration" components of motion z(t) is governed by equation
2 2 3
z+hiz+z + z (a, + 3a2A0) +a 2 z = 0 (10)

2 o A2 (1)
Q ~ + 2a1 A0 2~ t n
n=1,2...
where A, - An are the amplitudes of the forced term at the stability
j limit.

On the assumption of the simple-harmonic solution for z(t) the

3 theoretical first approximation to the amplitude-frequency relation is
2( = 2 2 + 2:A +
2
2 + 3a
3 -2 (12)
W(a)
0 A0 20 n

I Example i
2
2W0 0.38, CL = 1.07, 22 1, P = 0.16, h = 0.05, v = 1.04

The chaotic zone occurs in the neighborhood of the stability limit of

the 1/2 subharmonic resonance

x(t) = x(t + 2-) = Acos(! t + t + A + A +

2 0
with the continuous frequency spectrum around v/2. At he stability
limit A = - 0.276, A1 = 0.24. Equation ( ) yields a = 0.1056 and
1
w(a) = 0.1056 + 3/4 a

The theoretical w
= 1(a) is represented in Fig. 4 by the solid line.
Results obtained by an analysis of the filtered response R(t) are
represented by the dots.

Example 11
2
W =0,
a,= , 2 =1.0, P = 12.0, h = 0.1, V = 0.99.
The chaotic zone considered here occurs close to the stability limit of
the 7/3 ultraharmonic resonance

x(t) = x(t + 3T) - I Ancos(Vt + 0n) + Z A p/3cos(p/3vt + )

n=1,3,5 p=1,7
At the stability limit: A, = 1.67, A 3 = 1.29, A s = 0.27, yields

a = 3 = 2.60
n=1 2,3 n
The filtered response x(t) in Fig. 3 corresponds to a narrow band
of av in the neighborhood of 7/3v so that the changes of with the
amplitude can hardly be observed. The averaged frequency was obtained
as
W 22,50

Conclusions

The major point in the study of the nature of chaos is an

observation and analysis of the "filtered response", i.e., the component
of the complete chaotic response, which corresponds to an individual
narrow-band continuous segment of the frequency spectrum. Because the
filtered response x(t) shows a pattern similar to a single harmonic
function of time with randomly varying amplitude and amplitude-dependent
frequency, the idea of considering it to be a sort of "free-vibration"
was developed. The theoretical explanation of this phenomenon was found
by considering the concept of the stability limit of the secondary
resonances in the light of the approximate theory.

Comparisons of the properties of the filtered response with those

obtained by the first approximate theoretical analysis shows
surprisingly good agreement.

References

1. Y. Ueda, Randomly transitional phenomena in the systems governed

by Duffing's equation, J. Stat. Phys. 20, 1981-156 (1979).

2. Y. Ueda, Explosions of strange attractors exhibited by Duffing

equation, Ann. N. Y. Acad. Sci. 357, 422-433 (1980.
U
1 3. W. Szemplinska-Stupnicka and J. Bajkowski, The subharmonic
resonance and its transition to chaotic motion in a non-linear
5 oscillator, Int. J. Non-Lin. Mech. 21, 401-418, (1986).

4. W. Szemplinska-Stupnicka, Secondary resonances and approximate

models of routes to chaotic motion in non-linear oscillators, J.
Sound and Vib. 113, 155-172 (1987).

5. W. Szemplinska-Stupnicka, Bifurcations of harmonic solution

leading to chaotic motion in the softening type Duffing's
oscillator, Int.J. Non-Lin. Mech. 23 (1988).

6. G. M. Zaslavskii and B. V. Chirikov, Stochastic instability of

nonlinear oscillations, Sov. Phys. Uspekti 14, 549-672 (1972).

I
I

I
£
I
I
I
U
I
I
I
 I
Bifurcations in a Forced Softening Duffing Oscillator
A. H. Nayfeh and N. E. Sanchez
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
I
Abstract
Blacksburg, VA 24061
The response of a damped Dufllng oscillator of the softening type to a harmonic force Is analyzed in I
a two parameter space including the frequency and amplitude of excitation. An approximate procedure to
generate the bifurcation diagram in the parameter space of interest is developed by obtaining perturbation
solutions in the neighborhood of the nonlinear resonances of the system and, through Floquet analysis,
determining the bifurcation locus of these solutions. The results obtained are evaluated by comparing them
with analog simulations, which show escape from the potential well among other behaviors.

I. Introduction
Interest in the Dufing oscillator with softening nonlinearity lies in the vari-ty of physical phenomena
that it models (e.g., the rolling motion of a ship [I I ) and the fact that it is isomorphic to other systems of
importance in physics and engineering (e.g., Josephson junction oscillators and Foucault's pendulum).
Particularly interesting is the response of the Duffing oscillator in the presence of harmonic forcing and
I
damping, which has been found [2-41 to exhibit, among other features, hysteretic and chaotic behavior.
Thus, we considered this latter system governed by a nondimensional differential equation of the form 3
i + 2Ai + x - x3 =fcos Flt u,a>O feF FleW FW_ R (1)
where F and W represent appropriate domains to analyze the dependence of the solution on the parameters
f and fQ, which can be alternatively changed. Our aim was to identify the regions of the parameter space
P = F x W where bifurcations occur by using approximate analytical techniques and simple numerics.
I
Additionally, we carried out detailed analog simulations on a system governed by (1).
Figure I shows the potential well and the phase diagram of the Hamiltonian system associated with
(I) when ;&= 0.2 and a = 1. It is evident that under some conditions the system can escape from the
I
potential well. We are especially interested in this event due to the catastrophic implications that it could
carry in a physical system, e.g., the capsizing of a vessel. To determine the dependence of x(faf) on the
parameters, in Section 2 we develop an approximate solution using perturbation techniques. The stability
of this solution is then determined using Floquet theory, which signals the location of bifurcation points.
By sequentially applying this procedure, we generate a bifurcation diagram in the parameter space P for the
T-periodic approximate solution. In Section 3 we present the results of analog simulations and compare
these observations with the calculated solutions of Section 2.
2. Perturbation Solution
An approximate analytical solution of (1) may be obtained by assuming that the coefficients u, a, and
g
f are small. This smallness can be characterized using a single coefficient e, where &<< 1, as scaling factor.
Thus, we rewrite (1) as
ji + x =(-21x + ax 3 + fcos fit) ua>O feF fOeW FWCR 1 (2) 3
A second-order straightforward expansion 151 of the type
x(t, = x0(t) + ext(t) + Cx 2(t)... (3)
quickly identifies resonances in (2) occurring when fi = 1,fQ = 1/3, Q = 3, which render (3) nonuniform
and consequently inappropriate. We determine a uniform second-order approximate solution for the
primary resonance (fi 1) by applying the method of multiple scales, as presented by Nayfeh [51. The
I
subharmonic (fa = 3) , and superharmonic (fQ - 1/3) resonances were found to produce considerably
smaller amplitudes of response than the latter, so they were excluded from the frequency interval considered. 3
2.1 Multiple-Scales Solution for Primary Resonance
In this case the proximity of2m
Q to unity car. be expressed as
a2 =i+ea (4)
where a is a detuning parameter. The natural frequency of the linear oscillator in (2) can be written in terms
of f0 using (4), resulting in the following form: 3
I
 W+ei C(OX + 2 Li +fcos
-2x
X3 (5)

x(t, -)-x0 (To, T1, T2) + xI(TO, T1, T2) + Cx2(To, T, T2) + (6)
where To =t, T, = at, and T2 = alt. Substituting (6) into (5) and equating coefficients of like powers of a,
we obtain
2o 2. (7)

ox, + -2x = -2DoD 1 xO + ax0 - 2AsOxV + a4 +fcos fit (8)

D2x 2 + fi x2 = -2DoD 2 x0 - 2DoDtx, - D12xo - 2ADox, - 2MDx o + ax, + 3ax, (9)
The solution of (7) can be written as
xO= a cos(flt + ) (10)
Eliminating secular terms from (8) and (9), we find that
x(t) - a cos(t-It + fl) - aa 3 cos)(3 t + 3#l) +(1)

I
where a and # are given by
3
jff
--4Q fi
2 f
8.3
gxf a2.
3 z-a ) sinG
m-a- Z3ag fa-
i- 4
3+ 2f Cos#-(a-+a + . (12)
n2 4a2B2n
s
2 A2 2 a 3a 2 3aa 3 2 15a 2
2- 16C13 256CI 3

-( If. + 22 of + 2 3 f a 2) cosp - 2 4 sinGP (13)

For steady-state periodic responses i = 0 and = 0so that (12) and (13) become a set of algebraic equations
and can be solved numerically to determine a and P.

2.2 Stability Analysis

To ascertain the stability of the approximate solution (11), we examine the time evolution of the orbit
after the application of an infinitesimal disturbance of arbitrary type (t) in the form
A
x() = x(t + (t) (14)
S tttn A()it
Substituting x(t) into (2) and keeping only linear terms in (t), we obtain

I + + ( I- 2 )t(t (15)
which is a linear ordinary-differential equation with periodic coefficients having the period T = 21r/fl. The
existence of nontrivial solutions can be shown via Floquet's theorem [51, which calls for solutions of the
form
(t + 7) = At(t) (16)
where A is an eigenvalue (also called Floquet multiplier) of the monodromy matrix C associated with a
fundamental matrix solution 0(t) of (15) through the relation
(t + ) = D(t)C (17)

I In the case t = 0, 0(0) = I, we have C(t) = (1J) , which can be computed numerically by integrating (15)
in [0,T] subject to the initial conditions ( (0) = 1.0, (0) = 0 .) and ( (0) = 0., (0) = 1.0) , for each set of

3 parameters (f.fl). The periodic coefficient x(t) of (15) is evaluated using (11). The solution of x(t) is stable
provided that (t) does not grow with t. This requires that I A 1 <5 I . Additionally, -.he manner in which A

I
 1
leaves the unit circle defines the bifurcation that occurs [61. Determination of the stability of the periodic
orbit (11) or its Poincare map, when a parameter is varied, can be accomplished by characterizing its
codimension one bifurcations from the information provided by the Floquet multipliers A. For the
I
dissipative one-degee-of-freedom system described by (2), there are two types of bifurcations to be
considered: period-doubling (or flip) and saddle-node (or tangent), which occur when A goes through -I or
+ Irespectively [61.
2.3 Bifurcation of Symmetric Solutions
The preceding procedure allows us to predict instabilities in the T-periodic solutions of (2) when either
f or f is being changed. Period doubling has been found not to take place when the inversion symmetry
of the system is shared by the solution (2). This implies that the symmetry of the solution must be broken
before undergoing the bifurcation [3]. Since (11) satisfies x(t + T/2) = -x(t) ( i.e., it is symmetric), we can
only observe saddle-node bifurcations of this orbit, which induce either symmetry breaking, or tangent
instability depending on the region of the parameter space where the changes occur. Floquet analysis was
B
performed on the multiple-scales approximation (11) for values of C in the interval W = [0.4,1.01 and F
= [0.2,0.61 for g= 0.2 and a= 1.0 . The results are shown in Figure 2, where inserts (a), (b), and (c)
correspond to the stable orbits found in the regions A, B, and C of the parameter space P. These regions
can be defined as follows: S
A= (f,fl)ePI x(t)-X 1, as t-oo (x(O), i(0))e'Vj, P=FxW (18) 3
where Xi is the invariant set representing the large attractor and 'V is its basin of attraction. A similar
definition holds for B referring to its invariant set X2 and its basin of attraction 'Vi. Consequently, C has
an invariant set X1 UXz and a basin of attraction T, n IF,. The large attractor in insert (a) is then stable in
regions A and C. On the other hand, the small attractor in insert (b) is stable in regions B and C. Therefore,
1
we observe bistability in C, where the two attractors coexist. Domains of stable solutions are, in this case,
separated by lines of saddle-node bifurcations which indicate the loss of the T-periodic solution as either
of the two parameters undergoes a change across the line. Attractor (a) loses stability across the solid line,
while attractor (b) is lost when the parameters are changed across the dashed line. At very low forcing levels
U
no bifurcation occurs and the two attractors merge into one. The direction in which the bifurcation lines
are active is marked by arrows.

2.3 Bifurcation of Asymmetric Solutions

In order to predict period-doubling starting with a symmetric orbit we have to go to the second
instability. This can be achieved by modifying the form of the solution (11), to allow for the appearance
of even harmonics. A solution of the form
I
= A, cos(kflt + kp) (19)
k=O
can be used. The method of harmonic balance is appropriate to determine the values of the constants A&
and ftthat satisfy (I). Substituting (19) into (1)and equating the coefficients of each of the three harmonics
and constant term, we find I
-zA3_ 3
-- -(A°'A 1z +A 0 A oAAr 3 +--AAz
A 2+ +A )Ao =0 (20)I
- (AOA+ + AAA + -L "4 3)

2 A0, 3 A,:A
-fcos(fl)- A 1f 1+A 2=0+LAA+ )! (21)
-3a(AOAI + IjA+ +A0AA 3 +IA + AiAzA3 + 2 3 3
- 4fA 2 + A2 =0 (22)
fsin(ff) 2
+ 2A =0
+ A2 A
!a(A + -LAA'
+
2
A
J&
+AA
2
9A +2
(23)
(24
(
I
3a(A'A3 °A~ 2 + -L A 3" LA A IA ,2 + -L A'A3 3-0

A0,4A A2
3+LA3-9023+A(24
4lm 1Oll4l 1~ 2I l2 l
 This set of equations can be solved numerically to determine the A, and f coefficients for any given
set of parameters (fl)e P. The stability of (19) in the parameter space P can be settled again following the
procedure described in 2.2, but replacing x(t) by i in (15). The results are summarized in Figure 3, which
shows this time the occurrence of period-doubling bifurcations (solid curve) and, as before, the tangent
instabilities (dashed). The circles mark the results obtained in section 2.3 for the symmetric orbit.
3. Analog Simulation
As a means to verify the behavior predicted in the previous sections, we used an analog computer to
model the system described by (1). The results obtained are summarized in Figure 4, where we can observe
the characteristic Y-shaped zones 171 encountered in many phase-locking systems and obtained previously
from the T-periodic solution (Figure 3). However, this time we can get a complete picture of the events
taking place after the T-periodic solution loses stability. The region marked B lies below the arm of the Y
with negative slope, that we denote by yi. Tangent instabilities take place when any of the parameters
crosses yl, and the future fate of the system is determined by the attracting set existing in the subset of the
parameter space where it lands. When the system is initially at B and crosses yi into A, we observe the small
orbit becoming unstable and the system hopping to the large attractor, which is the attracting set in this
region. On the other hand, if y, is crossed into D, where
D=[(f f)*Pl x(t)--*co, as (--Poo (x(0),i'(0))eR1} P=FxW (25)
the solution becomes unbounded, since infinity is the only attracting set with nonzero basin.
The size of the domain A has not changed considerably. Two coalescent curves y2 and Y3 lie on the
side of the Y with positive slope, where yz marks the locus of the period doubling bifurcations observed in
the oscilloscope and confirmed by the appearance of half-frequency components in the frequency content
of a FFT(Fast Fourier Transform) of the solution. The symmetry breaking precursor was observed but the
bifurcation line was not included to avoid obscuring the diagram. Shortly after the first period-doubling
occurs, across y2, what is believed to be a period-doubling sequence takes place. Only a few period
multiplications can be observed before the broad-band spectrum appears in the FFT, indicating the
presence of a chaotic attractor. When either of the parameters crosses the Y3 curve, the existing chaotic
attractor vanishes, causing the system to jump to one of the attractors in the new domain. In this case the
jump could be to the small attractor if the crossing goes into B, or to infinity if it goes into D. The last curve
bordering domain A is y4, which represents the locus of tangent instabilities causing the large attractor to
jump to the small attractor.
Because of the inversion symmetry of (1), between y2 and y3 in Figure 4, x(t) and -x(t) are solutions;
thus, two asymmetric attractors coexist and undergo a period-doubling sequence to chaos. Finally, we note
that below 0 = 0.3 we observe a small Y notch replicating in small scale all the behavior previously
described, but this time located in the superharmonic frequency range where the system (1) has another
resonance. Figure 5 shows data points (stars) obtained from the analog simulations for comparison with the
predictions from section 2. Although the computed diagram only indicates the occurrence of saddle-node
and period-doubling bifurcations on the arms of the Y, the fact that no other attractor is present in D should
hint the possibility of escape.
Acknowledgement
This work was supported by the United States Office of Naval Research under grant #
N00014-83-K-0184 NR 4322753 and the Air Force Office of Scientific Research under grant #
AFOSR-86-0090.
References
I. A. H. Nayfeh and A. A. Khdeir, Nonlinear rolling of ships in regular beam seas. IntL Shipbldg. prog. 33, 40 (1986)
2. B. A. Huberman and J. P. Crutchfield, Chaotic states of anharmonic systems in periodic fields. Phys. Rev. Let. 43, 1743
(1979)
3. R. Raty, J. Von Boehm, and H. M. Isomiki, Absence of inversion symmetric limit cycles of even periods and the chaotic
motion of DufTing's oscillator. Phys. Lett. 103A , 289 (1984)
4. S. N. Rasband, Marginal stability boundaries for some driven, damped, non-linear oscillators. Intl. J. Non-Unear Mech. 22,
477 (1987)
5. A. H. Nayfeh. Introduction to Perturbation Techniques, Wiley, New York (1981)
6. G. loos and D. D. Joseph, Elementary Stability and Bifurcation Theory, Springer-Verlag, New York (1981)
7. N. B. Abraham, J. P. Gollup, and H. L Swinney, Testing nonlinear dynamics. Physica I ld, 252 (1983).
 V(x) (a)
0.3

0.0M

-10 0.0 10

(b)
X

00.

-10 00 0

Figure I (a) Potential well V(x) = X2 x4 , (b) Phase diagram of oscillator

X i-x-x=-O. 2 4

(c) "

0.5

0.4-
(b) / a)

0.3 1
B" A
0.2 .....

0.0 0.2 0.4 0.6 0.8 1.0

Figure 2 Bifurcation diagram of the approximate T-periodic multiple scales solution for
,u = 0.2, a = 1.0. All curves represent represent Faddle-node bifurcations. Inserts show
existing attractors in A,B, and C. Arrows point the direction in which the bifurcations
are active.
 0.6

S0.50

0.41 0

0.5' s • :"-~ .*"0

* A
0.2' , ,

1 0.0 0.2 0.4 0.6 0.8 1 .0

Figure 3 bifurcation diagram of the approximate T-periodic solution for I -0.2, a= 1.0.
Harmonic balance results are shown in solid (period doubling) and dashed
(saddle-node) curves. Circles show the multiple scales results.

1 0.6

1 0.5 '

(b) ' .:' . (a)

D y'
I 0.3 B A

0.2 ....
0.0 0.2 0.4 0.6 0.8 1.0

Figure 4 Bifurcation diagram from analog simulations

3£ 0.3
0.4 ... 'q 0~b

0.2

0.0 0.2 0.4 0.6 0.8 1.0

0.212

I Figure 5 Predicted bifurcations; Period doubling (solid), saddle-node (dashed). Observed

I bifurcations in analog simulation (stars)

 S

EXTENSIONS AND NEW APPLICATIONS OF MELNIKOV'S METHOD

FOR PREDICTING THE ONSET OF CHAOS*
Steven W. Shaw
Department of Mechanical Engineering
Michigan State University
East Lansing, MI 48824

Melnikov's method (1] is one of very few methods which can provide 1
necessary conditions for the existence of chaotic motions explicitly in
terms of the system parameters. It is an asymptotic method which is capable
of detecting the existence of transverse homoclinic points, which in turn
guarantees the existence of horseshoes via the Smale-Birkhoff homoclinic
S
theorem [2]. Horseshoes represent an unstable type of chaos which may be
transient in nature, but they are often a precursor to the onset of
sustained chaotic dynamics [3]. This analytical tool has been widely
I
applied to single degree of freedom systems with periodic excitation (see
conservative systems with more than one adegree
[3-7] for example) and to work of
freedom, [8-10]. Recent has extended the method to include wider
range of applications including:

i)
ii)
systems
systems
with small amplitude multi-frequency inputs
with large amplitude, low frequency inputs
3
iii)
iv)
certain
systems
dissipative multi-degree of freedom systems
of type iii) with inputs of types i) or ii). 3
These extensions are due primarily to Wigging and are outlined in his
forthcoming book (10].
In this presentation we will review the usual, planar Melnikov
method and will then proceed to discuss specific examples of physical
systems which are amenable to the extended methods. These wi include the
forced, damped spherical pendulum, a buckled beam with low amplitude, multi- I
frequency inputs, and the simple planar pendulum with large amplitude, low-
frequency excitation. The results will indicate the nature of the chaotic
motions which are expected to occur in each situation. I
*Supported in part by NSF and DARPA. 3
REFERENCES 3
1. V.K. Melnikov, "On the stability of the center for time periodic
perturbations," Trans, Moscow Math. Soc. 12, 1-57, 1963. 3
2. J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical
Systems and Bifurcations of Vector Fields, Applied Mathematical
Sciences #42, Springer Verlag, 1983.

3
I
3. F. Moon, J. Cusumano and P. Holmes, "Evidence for Homoclinic Orbits as
a Precursor to Chaos in a Magnetic Pendulum," Physica 240, 383-390,
1987.
4. F.M.A. Salam and S. Sastry, "Dynamics of the Forced Josephson Junction
Circuit: The Regions of Chaos," IEEE Transactions on Circuits and
Systems CAS-32, No. 8, 784-796, 1985.

5. S.W. Shaw and R.H. Rand, "The transition to chaos in a simple

mechanical system," submitted.
6. P. Holmes, "A Nonlinear Oscillator with a Strange Attractor,"
Philosophical Transactions of the Royal Society A292, 419-448, 1979.
7. B.P. Koch and R.W. Leven, "Subharmonic and homoclinic bifurcations in a
parametrically forced pendulum," Physica 16D, 1-13, 1985.
8. P. Holmes and J. Marsden, "Horseshoes in perturbations of Hamiltonians
with two degrees of freedom," Communications in Mathematical Physics
82, 523-544, 1982.
9. F.M.A. Salam. Marsden, J. and P.P. Varaiya, "Chaos and Arnold Diffusion
in Dynamical Systems," IEEE Transactions on Circuits and Systems CAS 30
(9), 697-708, 1983.
10. P. Holmes and J. Marsden, "Horseshoes and Arnold diffusion for
Hamiltonian systems on Lie groups," Indiana U. Math. J. 32, 273-310,
1983.
11. S. Wiggins, Global Bifurcations and Chaos-Analytical Methods, Springer
Verlag, to appear.
 I

SUPPRESSION OF CHAOS BY NONLINEAR DAMPING

Iradj G. Tadjbakhsh
Rensselaer Polytechnic Institute, Troy, New York
3
Gary L. Anderson
U.S. Army Research Office, Durham, North Carolina

Introduction 3
This paper describes a new method of achieving vibration reduction of
beams by means of damping of their axial displacements. Axial
displacements are of a smaller order of magnitude than the flexural
deflections. These displacements can be mechanically magnified in order to
S
produce appreciable axial damping that will inhibit vibrations. The result
is a dynamical system whose damping coefficient increases quadratically
with the amplitude of oscillations.
I
The problem of nonlinear vibrations of rods and beams also has a long
history. An account of its development is given in Nayfeh and Mook [11 I
where it is pointed out that the model that forms the basis of most non-
linear studies takes into account the nonlinear effect due to variation of
the axial force. Crespo da Silva and Glynn [2] using variational approach
showed that a consistent third order model has nonlinear contributions from
1
inertia and flexure as well. In this paper a consistent third order model
forms the basis of our study which however, is derived from Newton's laws
of motion and agrees with Crespo da Silva's and Glynn's result.
£
Equations of Motion

The basic system that is considered consists of a simply supported

beam that is axially damped. As shown in Figure 1, axial displacements of
the beam caused by transverse oscillations is resisted by a dashpot (C) or
another means of energy absorption such as friction. A rack-and-pinion
element (M) can suitably magnify axial displacements to appreciable levels
as necessary. Letting s denote the arc length along the beam and u(s,t)
1
and v(s,t) the time dependent planar components of displacements in the x
and y directions we have

(1+u) +v s =1 (v)

where subscripts denote differentiation.

The axial force T acting on the end s = L of the beam also determined I
to second order of nonlinearity becomes

TL = C(u t)sL' (2) 1

I
I
where C denotes the product of the damping coefficient of the dashpot and
the factor magnifying axial motions. Now the axial displacement of the
beam at its end s = L can be shown from (I) to be given by

1 L 2
u(L,t) =_L f0 (v,) ds. (3)

When (3) is inserted into (2), the results is found to be

UL TL f0 v v ds. (4N

From the equation of motion in the x-direction

(T (1+u s ) - N v i = utt , (5)

where T = T(s,t) is the axial force at an arbitrary point s along the center
line of the beam, and from the equation of dynamic angular equilibrium
2
N =-M -El [(I+V /2) vs (6)

one can determine that

T = - fS [S (v )tt ds 0 ]ds - E1 v s vss s + TL (7)

I Here m is the mass per unit length of the beam, M is the bending moment and
the boundary conditions u(O,t) = 0 and M(L,t) = 0 have been assumed. Sub-
5 stitution of (7) into the y-component of equation of motion

[Tv s + (+u s)NJs + f - mvtt (8)

yields
I m s S1 2]d }
mvtt + El {Vssss + [vs(vsvss)s] s } + { s fL
j if (vs)tt d s 0
s

3 - (TL Vs) s - f (9)

5 where f(s,t) is the applied force in the y-direction.

Primary Resonance

The boundary conditions are assumed to be

I
I
 I

v Mv -O at s -OL. (10)

In order to evaluate the effect of non-linear damping as represented

by the term (TLVs)s in (9), specifically,

L
(TL Vs )s " (CVs 0 sVst dx),

the response of the beam to a first mode harmonic force of frequency 9 is

considered. Similariy for the response we assume a first mode of variable
amplitude. Then
3/2 s 1/2 s
- 3 Fcos~tsin , v =c LA(t)sin w (12)
L L

where e is a dimensionless small parameter that is introduced for the pur-

pose of the perturbation analysis that will follow. Substitution into (9)
and using Galerkin procedure brings about

2 ""
A + A + 6A + A(A ) + eaA 2 1 - ek cos V T , d( ) (13)

where 6 and B are constants

2
6 - !- 8 - !4 312(
3-2 (14)
212 32

and the remaining quantities are defined by

2
Cit 4 F
w2 EI
v--, n k T - Wt (15)
2mLw W ML4 mLw2

Using the method of multiple scales [11 a solution is sought in the

form

A(t,e) - A0 (T ,T + cA1 (To,T I) + ... (16) 3

where To - T and T1 - cT. For Ao one may assume a harmonic solution of
slowly varying amplitude and phase, i.e., I
A0 - a(T I)Cos* , T0 + 8(T 1 ) (17)

Fig. 3 shows the comparison of results obtained from a Runge-Kutta

based numerical evaluation of (13) and (16) with the approximate result
obtained on the basis of multiple scale analysis. The numerical evalua- I
tions are for the five points depicted with a (*) and which correspond to
I
I
 the values v - 0.8, 0.9, 1.0, 1.1 and 1.2. The agreement is satisfactory
for e - 0.01 and k - a = 10.

To study the characteristics of the undamped motion Poincare' map of

the scaled velocity A/1iAaxI and the scaled displacement A/lAmaxl were
plotted. This map is the two dimensional projection of the three dimen-
sional phase diagram (A, A, r) onto the plane of (A, A) which is sampled at
regular time intervals. At resonance v-i sampling was done at T(2w) with T
3an integer in the interval 50 < T < 1050. Strictly periodic motion of one
period will be represented by a single point on the Poincare' map. Closed
cycles represent recurrent variable motions. More complex motions occupy
regions that can be characterized by means of fractals. All of these types
of motions were observed for the undamped nonlinear motion. Generally more
complex types of motions occurred as the amplitude of the forcing term was
increased. This is observed in the sequence of Poincare maps shown in
3 Figures 4-5.

3 The effect of the increase of nonlinear damping (increasing a) on the

system is
a gradual coalesence or shrinking of points and orbits in the
Poincare' map toward more compact figures and eventually a single point
representing periodic motion completely in phase with the applied force.
For e = 1, k = 2.0, v = 1.0, the progress toward a single periodic solution
was observed as damping parameter a was increased in steps from a - 0.0 to
a - 0.1. A single steady state periodic solution was observed for a - 1.0
with Amax = 0.9200 and Amax ' 0.0504. Fig. 6 is a typical stage in this
suppression process.

3 Acknowledgement

The authors wish to acknowledge useful discussions with M. Shahinpoor

and M.R.M. Crespo da Silva. Contributions of B.C. Lin and Y. Su, graduate
students at RPI, who performed the numerical computations are also
acknowledged. The work of I.G. Tadjbakhsh was supported partially through
a grant by the National Center for Earthquake Engineering Resarch at SUNY,
Buffalo.

3 References

1. A.H. NAYFEH and T.D. MOOK 1979 Non-Linear Oscillations. New York:
3 John Wiley and Sons.

2. M.R.M. CRESPO DA SILVA and C.C. GLYNN 1979 Journal of Structural

Mechanics 6, 437-461. Non-linear Flexural Flexural-torsional dynamics
of inextensional beams I, Equations of motion, II, Forced motions.
 n-V=c ~1 c'Ot-O.Iex
-@e.$Lo
c

rub"m 4, hmossm so ad & vmla Mmans

'..40 *1 0

toil

FREOITOC RATIO502

vmmiQ cMIO g-.ICrum2.O

F Ag

/W OI
(VLCM/8 3A
FEaIA CY ('-fl4,/Ua"
;qRATO 05L Ca 40 /4"so~
as1003"o
uu

ft ,CtAqfMm 01 42406 inj SP10"ME

ftM914L "~a"" dotin eL.t m m jim@~
too go@n ominS*6n Ia tmLm.
 SESSION 2

CHAOTIC MOTIONS

WEDNESDAY - 1045 - 1230

June 1, 1988
U

Symmetry Breaking Bifurcations In

Mechanical Systems and In surface Waves
3 P.R. Sethna

I
This study is concerned with symmetry breaking bifurcations in nonlinear
Idynamical systems with D 4 symmetry when two of the linear natural frequencies are
i nearly equal. The analysis can be shown to apply to a class of dynamical systems
including vibrations of nearly square plates. Detailed results will be given in the case
3 of surface waves in a nearly square container. The analysis shows that the periodic
and quasi-periiodic standing, as well as travelling wave phenomena, occur. For
certain critical values of the parameters, the system also exhibits chaotic phenomena.
3 The theoretical results are verified with the aid of experiments. A video tape of some of
the phenomena will be shown.
U
I
U
I
U
I
I
I
I
 ASYMPTOTIC TECHNIQUES AND CHAOS IN WEAKLY
NONLINEAR FORCED MECHANICAL SYSTEMS

by
A. K. Bajaj
School of Mechanical Engineering
Purdue University
West Lafayette, IN 47907

In many recent investigations into the dynamic behavior of multi-degree-of-freedom

weakly nonlinear and harmonically excited systems it has been observed that the
amplitude equations obtained via an asymptotic analysis can possess complex
responses. These are in addition to the various steady state constant solutions which
give rise to the phenomena of multiple periodic solutions, pump response etc. The
complex dynamic behavior usually arises when a stable constant solution loses stability
by Hopf bifurcation and the amplitude equations have a limit cycle solution. Changes
in system parameters such as the excitation frequency, the damping and the amplitude
of external forcing may lead to the limit cycle solution undergoing a cascade of
period-doubling bifurcations which culminate in chaotic solutions. The theorems for
the method of averaging then imply that for a small enough f the original system has
motiou on an integral manifold. The solution can be physically interpreted as an
amplitude modulated motion. The exact nature of the modulations in the motion is
not predicted.
In the present work, fundamentals of the asymptotic methods are carefully
reviewed and then some particular systems are studied with a view to exploring the
range of applicability of the asymptotic analysis and the correspondence between the
solutions of the averaged equations and those of the original system. The solutions of
the original system equations are investigated for e = 0.1 by using longtime integration
and Poincare section of the steady state motion. For moderate damping when the
averaged equations have only stable limit cycle solutions over the interval in
frequency, the Poincare section of the original equations consists of a set of points
dense in a closed curve. The motion is thus almost periodic with noncommensurate
frequencies. With variation in frequency, this motion undergoes a series of changes
from almost periodic to phase-locked and phase-locked to almost periodic. Poincare
sections clearly reveal that the periodic solutions of the system bifurcate into motion
on a 2-torus. A decrease in damping results in the 2-torus becoming unstable via a
torus-doubling. For small enough damping there is a cascade of torus-doublings
leading ultimately to the destruction of the torus. The predictions of asymptotic
analysis are qualitatively verified for parameter values at which the averaged systems
possess hyperbolic solutions.

mu
rnnmn
nmu•numu
nmnnmn
mm n n
U
NONLINEAR NONPLANAR PARAMETRIC RESPONSES
OF AN INEXTENSIONAL BEAM
All H. Nayfeh and Perngjin F. Pai
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061

U Abstract
The nonlinear integro-differential equations of motion for an inextensional beam are used to
investigate the planar and nonplanar responses of a fixed-free beam to a principal parametric
excitation. The beam is assumed to undergo flexure about two principal axes and torsion. The
equations contain cubic nonlinearities due to curvature and inertia. Two uniform beams with rec-
tangular cross sections are considered: one has an aspect ratio near unity, and the other has an as-
pect ratio near 6.27. In both cases, the beam possesses a one-to-one internal resonance with one
of the natural flexural frequencies in one plane being approximately equal to one of the natural
flexural frequencies in the second plane. A combination of the Galerkin procedure and the method
of multiple scales is used to construct a first-order uniform expansion for the interaction of the two
resonant modes, yielding four first-order nonlinear ordinary-differential equations governing the
amplitudes and phases of the modes of vibration. The results show that the nonlinear inertia terms
produce a softening effect and play a significant role in the planar responses of high-frequency
modes. On the other hand, the nonlinear geometric terms produce a hardening effect and dominate
the planar responses of low-frequency modes and nonplanar responses for all modes. If the non-
linear geometric terms were not included in the governing equations, then nonplanar responses
would not be predlicted. For some range of parameters, Hopf
consists of amplitude- and phase-modulated or chaotic motions.bifurcations exist and the response

3 Introduction
1.
A widely studied phenomenon is the response of axially driven rods. The axial load produces
a parametric excitation, which results in time-dependent coefficients in the governing equations and
boundary conditions. An important property of such systems is that a small excitation can produce
a large response when the excitation frequency is not close to any of the natural frequencies of the
system Ill. And, owing to certain nonlinear terms, the parametrically excited planar motions may
in turn excite a mode which does not lie in the original plane. These phenomena are not disclosed
by a linear approximation to these systems.

3 By taking into account the nonlinear inertia terms and considering linear curvature in the dif-
ferential equations of motion, Haight and King [21 obtained the planar frequency-response curves
of a parametrically excited rod by means of an averaging method. They also identified unstable re-
gions in the planar response curves, which correspond to a plane shift to stable motions in the other
principal plane. They did not find nonplanar motions. In this paper, we extend the analysis of
3 Haight and King by including the nonlinear terms arising from the curvature and determine the
nonplanar motions and their stability.

Most of the studies of the nonlinear dynamics of beams are based either on differential
equations valid for systems in which torsional effects are neglected or on equations obtained by
linearizing the beam's curvature. Only nonlinear inertia and stretching terms are commonly con-
sidered. Crespo da Silva and Glynn 131 showed that the generally neglected nonlinear terms arising
from the curvature are the same order as the nonlinear terms due to inertia. To investigate large
amplitude whirling motions of a simply supported beam constrained to have a fixed length, Ho,
Scott and Eisley 141 neglected the longitudinal inertia and Poisson effects but accounted for large
deformations through the use of Green's strain measure in the longitudinal direction. They showed

3 that the nonlinear terms are cubic. They found both steady whirling motions and whirling motions
of the beating type for some parameters.

I
 I
2. Equations and Method of Solution 3
The equations governing the parametric vibration of the system shown in Fig. 1 are

V+ ci" + yV1'v - (I - fl)[:W"f v"w ' ds - w"' v"w'dsl' S

(I - f), - 9YLV'(V'V1 + w'w")']' 3
lY
f" I
(v'f 3[ 2
W,2)ds]"ds} ' -
+L(VJ - 1)+ v']Bn 2 COS(nt) (1) 3v
1 00

(I -Y)2 [v'fJlv"w"dsds]° - [w'(v'v" + ww")']' 3

- - Ew'J[ (v,2 + w2 )ds]"ds]' - [w"(s - 1) + w']Bfl 2 cos(flt) (2)

and the boundary conditions are

v=w=v'=w'=0 ats=0 (3a) 3
v" =w" =v.' =w.' =0 ats= I (3b)
To investigate the behavior of the beam to a principal parametric excitation in the presence of
a one-to-one autoparametric resonance, we introduce two detuning parameters a and 62 , defined
by
2
Ll = 2whn(l + &2a) and (h/b)2 2
I +6o + e62 (4)
and let wtm = 2,. By using the Galerkin procedure and the method of multiple scales, we obtain
the first-order uniform expansion 5
v(s,t) = cFm(!)al( 2t) cos( -- Ut - Yl) + "- (5a)

w(s,t) = eF,(s)a2 (c2t) cos( I f t - Y2) + ..- (Sb)

where the equations governing the amplitudes and phases are

2U +
2&)ma ' + CRI + R2a2 sin 2(y I - y 2) R 3 sin 2y ,]a, = 0 (6)
2
[2coImyI'- R4 - R 5 a2 + R,a
6 + R 2a2 cos (y, - Y2) + R3 cos 2y 1]a, = 0 (7)
2 2 (8)
w2na 2 ' + [El + E2a sin (y I - Y2) + E 3 sin 2y 2]a 2 = 0

2
1202'- E4 - Esa2- Ea - E 2a2 cos (y I - y2) + E3 cos 2Y2 ]a2 = 0 (9)

Here, the R, and E, are constants obtained by using numerical integration methods.

3. Numerical Results
Case 1: Near Square Cross Section
In this case, 6. = 0.0 and P, = 0.7692. We let A = 0.05 and b = 0.03. Figure 2 shows the re- U
23
sponse curves of the first mode. It is well known that, for planar responses, if only in-plane dis-
turbances are considered then the upper branch of a2 will be stable Ill. But due to the disturbances
in the y-direction , this branch is unstable. It follows from Eqs.(6) and (7) that

-9 and sin 2y2 = b

sin 2y 1

So, if s = 0, then the phase angles y, = = - 900 for the upper branches, and y1 = = 0 for the
lower branches. But the damping changes slightly the phase angles. If we increase j, the upper and
lower branches will move closer to each other. At some critical value of Au,these two will merge.
If we increase b, the separation between these two branches will increase. We note from Fig. 2 that
the planar solution in the stiff direction (i.e., z direction) is always unstable and hence the planar
response is always in the weak direction (i.e., y direction). It follows from Eqs. (6)-(9) that the ef-
fective nonlinear spring coefficients a. and P, in the y and z directions are
2 2 2 2
e= (1 + 60)a4 - 3 ao5wm and e = f4- 3 05C°2n

Since CE,a5, P4, andfls are positive, the nonlinear geometric terms (I + 60)a, and ft. are of the hard-
ening type, whereas the nonlinear inertia terms 2/3a5<0, and 2/3fls W2 are of the softening type.
Hence, the overall type of the effective nonlinearity in the y(z) direction depends on the relative
magnitudes of a,(#,) and a,(#,) and the mode shape. For the lower modes, Fig. 2 shows that the
planar response curves are bent to the right, which implies that the nonlinear geometric terms
dominate the response because they have a hardening effect. Neglecting the nonlinear geometric
terms (i.e., letting a, = fl, = 0 for i = 1, 2, 3, and 4), we obtain the response curves shown in Fig.
3. Comparing Figures 2 and 3 shows that neglecting the geometric nonlinearity yields frequency
curves that are even qualitatively wrong. The nonlinearity changes from a hardening to a softening
type. Moreover, nonplanar responses cannot be predicted without including the geometric non-
linearity.
Figure 4 shows the response curves of the second mode. We note that for the upper branches
of planar motions Y, = Y2=- 0 0 , and for the lower branches y, = Y2= - 90° . These phase angles are
different from those found in the response of the first mode (Figure 2). Owing to disturbances in
the z-direction, the planar motion in the y-direction is unstable along the branch AB. Similarly,
the planar motion in the z-dircction becomes unstable along the branch CD due to disturbances in
the y-direction. Another point is that the planar response curves are bent to the left, which means
that the overall effective nonlinearity is of the softening type and hence the nonlinear inertia terms
dominate the response.
Next, we return to Figures 2 and 4 to discuss the nonplanar response curves of the first two
modes. As a decreases from a value larger than that corresponding to the point B of Fig. 2, the
nonplanar fixed point loses stability with a complex conjugate pair of eigenvalues moving into the
right-half plane. This corresponds to the extensively studied Hopf bifurcation. Based on the Hopf
bifurcation theorem, one expects amplitude- and phase-modulated motions for values of cr near B.
We note that the phase angles of nonplanar responses are not constant, but the difference between
y, and Y2 is always 900. Furthermore, all the nonplanar response curves are bent to the right even
for the second mode. This is not unexpected because as discussed earlier the nonlinear geometric
terms control the nonplanar motion. For the first mode, the nonlinear geometric terms have a
hardening effect, and hence the amplitudes of nonplanar mction are smaller than those of planar
motion. On the other hand, for the second mode, the nonlinear geometric terms overcome the in-
ertia terms and produce nonplanar motions that are larger than the planar motions.
If we increasc the absolute value of 62 (i.e., increase the deviation of the cross section from a
square), the planar response curves of the y-direction motion move away from those of the z-
direction motion. Figures 5 and 6 show the response curves for the first two modes. Because of
out-of-plane disburbances, the planar response is unstable along the branch AB of Fig. 5, which
also shows the nonplanar response of the first mode. Here we have a Hopf bifurcation at point C.
Using a Runge-Kutta routine to integrate Eqs. (6)-(9) for a = - 0.0353 along the branch AC for
a long period of time, we obtain the amplitude-modulation behavior shown in Fig. 7. The phase
difference is not equal to 1800 . Figure 8 shows the projection of the attractor on the a, - at plane.
Since the amplitude and phases are not constant but periodic with a period that is large- than that
corresponding to free oscillations, the resulting motion is nonperiodic having two periods (i.e.,

3
 I
motion on a torus). The spectrum of a, in Fig. 9 shows that the fundamental dimensionless fre-
quency of the attractor is approximately 0.154 and hence its dimensionless period is approximately
6.5. This motion can be better visualized by plotting the motion of the tip-end of the beam, as
shown in Fig. 10. This figure shows that the elliptical route keeps changing the lengths of axes and
direction, and it also shows the twisting motion. Because of the nonlinear terms, the inertia force
in the y-direction is not proportional to v(s,t) and the inertia force in the z-direction is not pro - I
portional to w(s,t), and hence the resultant inertia force is not parallel to the total displacement in
the y-z plane and it induces a twisting moment on the beam. This is a whirling motion of the
beating type. Increasing u further to 0.07627 produces a period-doubling bifurcation of the
attractor as shown in Figs. II and 12. Increasing a further produces a bifurcation of this attractor
to a fixed point, yielding a periodic rather than an aperiodic nonplanar motion. Thus, the response
is a steady whirling motion.
Figure 6 shows also the nonplanar response of the second mode. We note that the Hopf I
bifurcation points E and F are adjacent to stable and unstable branches. It is found that the motion
is chaotic. In Figs. 5 and 6, the nonplanar response curves always bifurcate from the planar re-
sponse curves at frequencies higher than the natural frequency. The reason is that the motion in
the y-direction has the same mode shape as that in the z-direction motion but it has a smaller linear
I
natural freqyency; that is, it belongs to a lower-energy motion.
Case 2: Rectangular Cross Section
In this case, 6. =- 0.9745 (i.e., b/h = 6.2673 ) and 0, = 0.3944 , and we let ;i = 0.05 and
b = 0.03. For this beam, co,, = W0l. We investigate the possibility of nonplanar vibrations comprised
of the first mode in the z-direction and the second mode in the y-direction. We find that there are
two branches of nonplanar response cuives: the left branch of the nonplanar response curve
I
bifurcates from the stable branch of the planar response curve for the motion in the y-direction and,
in that interval, the planar motion in the y-direction is unstable with respect to z-direction dis-
turbances; the right branch of the nonplanar response curve bifurcates from the unstable branch
of the planar response curve for the motion in the z-direction and, in this interval, both the planar
I
motion in the y-direction and the nonplanar motion are stable, but the nonplanar motion belongs
to a higher energy motion because its amplitudes are larger than those of the planar motion.

Acknowledgement
This work was supported by the Air Force Office of Scientific Research under Grant Nos.
AFOSR-86-0090 and F49620-87-C-0088.

References
1. A. H. Nayfeh and D. T. Mook, Nonlinear Oscillations,Wiley-Interscience, New York, 1979. I
2. E. C. Haight and W. W. King, "Stability of Parametrically Excited Vibrations of an Elastic
Rod," Dev. Theor. Appl. Mech., Vol. 5, 1971, pp. 677-714.
3. M. R. M. Crespo da Silva and C. C. Glynn, "Nonlinear Flexural-Flexural-Torsional Dynamics
of Inextensional Beams--- I. Equations of Motion," J. Struct. Mech., 6(4), 1978, pp. 437-448.
4. C.-H. Ho, R. A. Scott, and J. G. Eisley, "Nonplanar, Nonlinear Oscillations of a Beam---I.
Forced Motions,' Int. J. Nonlinear Mechanics, Vol. 10, 1975, pp. 113-127.

I
I

4 I
 A020

II
S 4"
,a.90
89, '
0.04

0. 53 82
-0.4 -0.2 0 0.2 0.4

Y1 (o,0).1co,(M0) Figure 4. Response curves of the second mode

Figure 1. Coordinate systems x-y-z = the for a beam with an aspect ratio b^h= 1.0 mode
inertial reference frame; - n - C= the principal (2,2) , (0 12 = cn,60 = 0.0 ,6 = -0.05, = 0.05,
axes of the beam's cross section at position s. b = 0.03, P, = 0.7692.
'21 5" "- I 2.0

.8i a,12 " 05/ -, / 02"

. / t/4
. A2./
/V ,, .., 0.41 ..-
,.o
, ,'° //, "" ,"0

- - - --. 24 ._.2 . . . 11...

. . /.. , . . .I
/. ,'.- ,_
I :""I-0
-0.4 0 0.04 0.08 0 9 a - 2
o--.136 -0.24 -0.12 0 62
o Qo'24
Figure 2. Response curves of the first mode for
a beam with an aspect ratio b/h= 1.0 mode (1,1) Figure 5. Response curves of the first mode for
I cl,, w,2 , 60=0.0 , 62=-0.05 , M=0.05,
b = 0.03, fl, = 0.7692; a,, a2 = planar response a beam with an aspect ratio ;/h= 1.0: mode (1,1)
amplitudes; a,,, a2, = nonplanar response ampli- (o, = co2 j, 60 = 0.0 , 62 = -0.5 , /.= 0.05,
tudes; (_ ) stable, (---- ) unstable with at least b = 0.03,/3, - 0.7692.
one eigenvalue being positive, (....
) unstable with
the real part of a complex pair of eigenvalues be- 0.20

1 0.161 A /
2 / ,\ //
C CL12

081 00121\

\\\

S0.044
. .~. .. . .4
o 2-- - . - . -6__ .._ 0...-2,2 -- "" " 2-
0.2 '03
a 5 6- 08 1 - C .4 0 C
-C2 -0.08 -0.04 0 0.04 o08 Figure 6. Response curves of the second mode
Figure
the 3. Response
absence curves of
of the nonlinear the firstterms,
geometric mode for
in for
(2,2)a beam
, a,12 =with
con, an aspect
0.0 ,ratio
6 = ;/h= 1.0 :mode
60 = -1.2 0.05,
th asncf h nnieromti tems fo (=2 0.03 , =0.07692.-. ,~s=0.5
a beam with an aspect ratio/h- 1.0: mode (1,1) b =0.03,/3, = 0.7692.
oll 0 21 , 60=0.0 , 62=-0.05 , - 0.05,

I b = 0.03, /l,= 0.7692.

 aa,,

time v(L,t)
Figure 7. The long-time history of the ampli- Figure 10. The path of the tip-end of the beam
tudes for the case of an amplitude- and phase- for the case of an amplitude- and phase-
modulated motion b/h= 1.0 , mode (1,1) ,modulated motion b/h= 1.0 , mode (1,1)
cil0, 6liJo,0.0 6=
2 -0.5 , i0.05, Wi W1 650.0 ,2 =-0.5 IA, .5

b - 0.03, ~,=0.7692, a=-0.0353. b = 0.03, ~,=0.7692 , a=-0.0353 , c-0.5.

a2. a2n'-.

Figure 8. A projection of the trajectory onto the Figure 11. A projection of the trajectory onto theU
a, - a2 plane :b/h= 1.0 , mode (1,1) ,01 -o~=(021, a, - a2 plane : b/hz 1.0 ,mode (1,1) ,Coll = W21,
JOM00162 -0.5 PA 0s.05, b 0.03, 60=o.0 1 62=-0.5 =0.05, b 0.03,
iA

= 0.7692, a=-0.0353. = 0.7692 , a=0.07627.

frequency frequency

Figure 9. The Fourier harmonic analysis of a,: Figure 12. The Fourier harmonic analysis of a,:
;/h= 1.0 ,mode (1,1) ,coil -- 2,, 60 = 0.0 ;if,=z 1.0 ,mode (1, 1) wit = w~i, 6o~ = 0.0
62 - 0 .5 ,A 0s.05, b 0.03, /30.7692 ,62=-o.5 A= ~o.05, b=0.03, a,0.7692
a =-0.0353. a 0.07627.
I
Nonlinear Response of Infinitly Long Circular Cylindrical Shells to
Subharmonlc Radial Loads
by
Ali H. Nayfeh, Raouf A. Raouf, and Jamal F. Nayfeh
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
3 Blacksburg, VA 24061
To second-order, the equations of motion of an infinitely long
circular cylindrical shell in dimensionless form are (Goodier and
McIvor, 1964; Nayfeh and Raouf, 1987; and Raouf, 1985)
• 2 wi .....
v
W+ a (w' +2w +w)- +w =w w) - 10

+ 4 - 2w - 1 + a(1+v2 ) -w)

Eh
and
- +w =ww -2w, + 2 + a(1Eh w P (2)

where the overdot indicates the partial derivative with respect to t,

the prime indicates the partial derivative with respect to e, w and

o are the dimensionless radial and tangential displacements,

respectively, and P is the applied pressure load. Here, 0 2 = h2/12a2,
3 where h and a are the thickness and initial radius of the shell,
respectively.

We use the method of multiple scales (Nayfeh, 1973, 1981) to

determine a second-order uniform expansion of the solution of Eqs. (1)
and (2) for small but finite amplitudes when P is given by
2
a(I-v l p = F cosat (3)
* Eh
when a - 2wm and w_ =2w, where w and wf are the linear natural
frequencies of the breatAing and f~exural modes, respectively. Here,
3 E is a small dimensionless quantity. Thus, we assume that
w(e,t;e) = ew,(e,To,TL) + 2 w2(e,T
Tl) + ... (4)

= ((e,t;e)
el(e,T 0 ,Tl) + E2 2 (9,T 0 Tl) + ... (5)
3 where To = t, a fast scale characterizing motions with the natural and
excitation frequencies, and T, = et, a slow scale characterizing the
modulation of the amplitudes and phases of the modes with damping,
m nonlinearity, and any possible resonances.

The evolution equations are (Nayfeh, Raouf and Nayfeh, 1988)

p'p+ v2q, + 0 p, + 2A,(p 2q2 + P3q 3) + fql = 0 (6)

2 + 2 2 2 =
qI - V2P1 + 4oq, - Aj(P2 P - q2 _ q3) + fP1 0 (7)
 m
P2 + nP2 + A 2 (q P 2 - q2p,) = 0 (8)

- ~11P 2 + Unq2 - A 2 (P1 P 2 + qlq 2) = 0 (9)

3+ vq 3 + UnP3 + A 2 (q1 P 3 - q3PI) = 0 (10)

q; - VP 3 + Unq3 - A 2 (PlP 3 + q1q3 ) = 0 (11)

where u° and pn are modal damping coefficients,

vi = 1
1 + 1
I a V 1
1 (12a)
2~+ 4 29 = 2 021

= 0 + Ca2W = 2wn +ca (12b)

The steady-state solutions of (6)-(11) correspond to p= q, = 0. There
are two possibilities. Either

ao = an = bn = 0 (13) 1
or
a = a* = [ 1 a+ )21 (14)
02 2
b f ao2
an +2 +
=- xi t _ X2) (15)
n nA2 A 2
1
where
whr 2 2 2 1 2 b2 2 2i

ao = p, + q1 , an = P 2 + q 2 bn = P 3 + q3 (16)
X, = [4uoan - 02( 02 + o 1 )]/4A A 2 (17)

=
X2 [P0(a2 + 2o1 ) + 22Un ]/4AIA 2 (18)
We study the flow governed by the autonomous evolution equations
(6)-(11) rather than the original equations of motion. Thus, a fixed
point of the flow corresponds~to a periodic solution of Eqs. (1) and
(2). The detuning parameter a2 = C 2 (f/A2 )- is used as a bifurcation
parameter, all others being held fixed.

In Fig. 1, we show a typical response curve. This figure shows two

phenomena, typical of nonlinear systems, the saturation and jump
phenomena (Nayfeh and Mook, 1979). As the amplitude f of the excitation
increases from zero, only the trivial fixed point attractors exist,
until a threshold is reached at f = f 2. The trivial solution is no
longer stable and a jump phenomenon occurs. As f increases the

mU
U
amplitude of the excited breathing mode saturates at a constant level
and the extra input energy is spilled over into the flexural mode, which
responds with a large amplitude wrinkling of the shell. At f = f3, the
3 fixed point attractors lose stability through a Hopf bifurcation.

Upon decreasing f, the amplitude of the flexural mode decreases and

the amplitude of the breathing mode remains constant, until f = f is
reached. The critical value f = f corresponds to a collision between
a the unstable and stable fixed points causing a jump to the trivial
response. The flow undergoes a fold bifurcation.

Next we study the flow as a2 changes. We take a = 2.0918xi0 - ,

which yields wo z 2w6 and 0, = a1 (f/A2 )- = - 0.73. Moreover, we let

6 = UO,* f/A2) = 0.02.

IFor
2.0353 < a, < -1.9987, the flow asymptotically approaches a
hyperbolic fixed point as t - - and the response of Eqs. (1) and (2) is
either trivial or periodic. The behavior of the flow within the above
interval is summarized in Fig. 2. Here, we make the following
observations:
1. All periodic orbits are born symmetric and lose stability through a
cyclic-fold bifurcation. Figure 3 shows a typical projection of
attractor III and its power spectral density (PSD).
2. Deformation of attractors:
All attractors undergo deformation as a 2 changes. Interesting
behaviors are observed in the following attractors:

a) Attractor IV starts at ^2 = - 1.8200 as a symmetric one but loses

its symmetry just before collision with a repeller.
b) Attractor IX is born at a2 = 1.9953 as a symmetric one, loses
symmetry but regains it before collision.

c) Attractor X is born at a = 1.9953 as a symmetric one. It goes

into a period-three motion, becomes asymmetric, and then regains its
symmetry before collision as shown in Figs. 4(a-d), respectively.

I 3. The evolution equations (6)-(11) exhibit a symmetry apparent in the

projections of the phase trajectories (Fig. 3). The flow is
m invariant under the transformation

P, -1 0 0 0 P1

IT: q, 0 -1 0 0 q(
(19)
P2 0 0 0 1 P 2.

m 0 0 -1 0 q2

I
4. At o = - 1.55 (chaotic region), the Lyapunov exponents are 0.566,
o.006, -0.057, -0.624 and the dimension is df = 3.8.

ACKNOWLEDGEMENT
This work was supported by the National Science Foundation under
Grant # MSM-8521748 and the Air Force Office of Scientific Research
under Grant # AFOSR-86-0090.
REFERENCES

1. Nayfeh, A. H., 1973, Perturbation Methods, Wiley-Interscience New

York.
2. Nayfeh, A. H. and Mook, 0. T., 1979, Nonlinear Oscillations,
Wiley-Intersicence.

3. Nayfeh, A. H., 1981, Introduction to Perturbation Techniques,

Wiley-Interscience, New York.

4. Nayfeh, A. H. and Raouf, R. A., 1987a, "Nonlinear Forced Response

of Infinitely Long Circular Cylindrical Shells," ASME Journal of
Applied Mechanics, Vol. 109, pp. 571-577.

5. Nayfeh, A. H. and Raouf, R. A., 1987b, "Nonlinear Oscillation of

Circular Cylindrical Shells," International Journal of Solids and
Structures, Vol. 23, pp. 1625-1638.
6. Raouf, R. A., 1985 "Nonlinear Forced Response of Circular
Cylindrical Shells," Thesis, Department of Engineering Science and
Mechanics, Virginia Polytechnic Institute and State University,
Blacksburg, VA.

7. Nayfeh, A. H., Raouf, R. A., and Nayfeh, J. F., 1988 "Nonlinear

Response of Long Circular Cylindrical Shells to Subharmonic Radial
Loads," ASME Journal of Applied Mechanics, submitted for
publication.

a
 - 6

I\ 00

f 1f2
f3f

FIG.I TYPCAL RESPONSE AMPLITUDE

- i
-I.9987 - - - hyperbolic
I1.9981- -1.9986 - fixed point

-L820 - r
-1.7893
-1.7861 - 1
-I.690 y
-L6546 y---l.y-,
-1.630
-1.580 -&f-
S4.280 h Os M

I
IA

1.9100 ' Ech-os 3m

1.98127 Q s sym.
1.9816
1.9830- asym.
I 9834- eriod- 3 asym
3 1.9837
1.990 -sym. sym.

1
1.9953-
1.9981
2.0353-
F hyperboli C
fixed point

FIG 2
I
I
 ~~1 3
P,LL I

2 H2

p p U
a)2 Hz 2

A I92OA 1

1. p l 93 14

a) P2) 23

FIG 4 DEFOR MATION OF ATTRACTOR X

U
I
THE EVALUATION OF THE REGION OF CHAOTIC MOTION IN
I CURVED STRUCTURES

3 By
Chung Fai Ng
Resident Research Associate
Structural Acoustics Branch
NASA Langley Research Center
Hampton, VA 23665-5225

For Presentation at
the Second Conference on Non-Linear, Stability,
and Dynamics of Structures and Mechanisms
3 June 1-3, 1988
Blacksburg, Virginia

Curved structures such as postbuckled beams, plates and cylindrical

plates are known to have softening spring behavior. In some cases, as
the effects of softening increase, the effective stiffness is reduced to zero
and instability starts. The dynamic instabilities of a beam or plate was
studied by representing the deformation shapes by one to three terms of
shape function and analyzing the resulting coupled nonlinear equations of
motion using Runge-Kunta Techniques. It is found that the dynamic
instability under sinusoidal excitation takes the form of a non-periodic
chaotic motion. Typically, in a buckled plate, the chaotic motion starts
3 when the r.m.s. displacement reaches 35% of the buckled value and
under sinusoidal excitation at 80% of the linear natural frequency.
However, the chaotic motion stops when the excitation is increased
beyond a certain level and periodic motion resumes. This motion is
shown in figure 1 as an oscillatory motion from C to C' passes through the
region of negative stiffness (between A & B) and its mean position is at
zero, which is an unstable static position. Also, the oscillatory motion
exhibits hardening spring behavior, like that of a flat plate, showing that
3 the effects of softening due to curvature are less significant when the
Jynamic amplitude is large. This can be seen from the fact that a flat and
a buckled plate have similar magnitude of stiffness for large
displacement (from figure 1). In summary, a curved plate starting with a
softening spring system with mean position at the statically stable
position ends eventually with a hardening spring dynamic system with

U
 I
I
mean at a statically unstable position. The transitional region between
the two systems being the region of chaotic motion. In this paper,
emphases are on the evaluation of the lower and upper boundaries of the
chaotic region rather than the characteristics of the chaotic motions itself.
Effects of imperfections and the presence of antisymmetric (2,1) mode
and higher symmetric mode (3,1) on the starting point of chaotic motion
are studied and approximate analytical formulae are derived.
Comparisons are made with experiments on buckled beams and plates
using excitation frequencies from 5 Hz to twice the fundamental
frequency.

The results are very useful for prediction of sinusoidal and random
responses of curved plates.

I
FIG.1

FORCE VS DISPLACEMENT I
FOR FLAT AND BUCKLED PLATES
force p
2 flat buckled - -u

/ H
B / I
-2/
-displacement
2Pp=- q

AI
Ce 4 static position p:O

/ -1

I
I
Parameter Identification In Chaotic Dynamic Systems
by

D. Joseph Mook*
Pao-Hong Tongt
IDepartment of Mechanical and Aerospace Engineering
State University of New York at Buffalo
Buffalo, New York 14260

Abstract
The study of chaotic dynamic systems continues to generate considerable interest
in the mathematics and mechanics research communities. The research work to date
has consisted largely of investigating the properties and solutions of mathematical models
which exhibit chaos, of building simple physical systems which demonstrate chaotic motion,
and of investigating the possibility of the existence of chaos in various physical systems.
There remains a critical issue which must be addressed before many of these results may
be incorporated into engineering analysis and design, which is, how does one obtain an
accurate mathematical model for an actual system known to be chaotic?

Some efforts have recently been made to apply existing well-known parameter identifi-
cation techniques to the problem of identifying unknown parameters in a chaotic dynamic
system model. These efforts have been largely unsuccessful, indicating a need for new
methods to identify chaotic systems. This motivation led to the present study.

In this paper, we present a method for estimating the parameters in a chaotic dynamic
system model from discrete measurements of the system's output. The method works well
on the example problems considered. The effects of measurement noise and frequency are
discussed, and ideas which may be useful in choosing model terms for a chaotic system are
*presented.

I
U
I
* Assistant Professor
*t Graduate Assistant

11
 SESOI
IOSAINR N ADMVBAIN
WENSAI30-13
Jue1,18
I
3 NONLINEAR NONSTATIONARY PROCESSES

R. M. Evan-Iwanowski and G. L. Ostiguy

I ABSTRACT: The objective of this write-up is (i) to present new and refined
results on the nonstationary (NS) manifestations of mechanical systems: the
Duffing oscillator, relaxation vibrations, and dynamic stability which
indicate the departures from the corresponding stationary (ST) responses, or
result in specifically NS behavior; (ii) to identify the factors and the
aspects which affect the NS responses; and (iii) to delineate the NS
unsolved problems related to the systems in (i). This work is viewed as a
contribution for further theoretical developments and for applications in
technology.

I I. INTRODUCTION.
Physical world as prceived by the Systems Dynmics (SD) is nonlinear (NL),
nonstationary (NS) , and random (stochastic) . The mathematical models
(ordinary differential equations (ODEs or simply DEs) in our case), which
represent this world contain nonlinear terms and nonstationary components.

I Inrecent years, the unprecedented achievements were registered in ST SD.

Quantitative and qualitative - geometrical or topological methods - were
used leading to these achievements. The developments in NS systems showed
relatively little activities. The main reason is that, in spite of the NS
manifestations being obvious, they are complex and difficult to track.
Clearly, the ST systems constitute a point of departure for the NS systems,
and at the same time, the NS systems should contain the ST systems as a
special or limiting case.
It has been established that in some cases even small changes of the control

variables or initial conditions may result in considerable changes in the

dynamic responses. It is thus expected that the NS modes may alter substan-
* tially the ST responses.

In the studies in the ST3 SD, primary attention is focused on the three
classes of the most common single degree of freedom (1DF) systems:
U (1) Duffing oscillators, (2) Relaxation oscillations, and (3) Parametric
systems related to dynamic stability. Similar methodology has been followed
in the studies of NS systems.

I*

I *Visiting Professor, University of Central Florida, Orlando, FL 32816

Professor, Ecole Polytechnique, Montreal, Quebec H3C3A7.

1NS systems contain some time dependent components or control

variables.
2
Not treated in this write-up.
3 Many
physical, engineering, and scientific manifestations are modeled
by these equations.

I
 2

This write-up concerns itself with a special NS modes, viz., transitions

through resonances (TtRs)-time varying frequencies and/or amplitudes of the
external excitations (ee).
The studies in TtR fall into two categories (1) when the time varying
parameters are arbitrary, explicit functions of time with extended ranges of
their variations, so-called robust TtR, and (2) when these parameters are
considered to vary very slowly, i.e., they are close to the ST values or
they ever need be defined as explicit functions of time, and they are
referred to as evolving TtR.

2. DEVELOPMENTS AND DELINEATION OF TtR PROBLEMS

The early NS developments were related to the robust NS systems and were
presented in a paper [1] and two monographs [2] and [3], where the asymp-
totic methods of Krylov-Bugolyubov have been extended to include NS pro-
cesses, the KBM method. Numerous illustrations of applications of special
functions and KBM methods to the technological problems were presented [3],
[4].
A formal extension of the KBM method to the multiple degrees of freedom
systems to include combination resonances has been presented in [5] and
applied to a set of problems exhibiting these type of resonances, especially
in the area of dynamic stability [6].
The main factors which affect the TtR responses are the values and direc-
tions of the sweep rates a and . in linear TtR of the frequency
v = V + a t and the amplitude of V B + a t of the ee, a and y in
cycli 2 TtR) v = v + ySinat. Other tim foris of the TtR are in use:
logarithmic, powersof t and exponential.

Other factors affecting the NS responses are: initial conditions (IC)

either ON the stationary curve (SC) or NOT-ON SC, damping parameter, modes
of oscillations, static loads, spatial boundary conditions, static loads.
Duffing Oscillator. x + 2x + 41 + Vx3 = B(t)cose(t). Amplitude-frequency
plot s a, v) in linear TtR, are presented in Fig. 1, originating on SC and
NOT ON SC. It is seen from Fig. 1 that such TtR curves eliminate fold
discontinuity, and they jump or drop in a cascading wavy form, terminating
with dying out high frequency oscillations at the junction with the SC
curve.
The evolving TtR was used, among other means, in [7] to predict the incip-
ient jump. It is remarked there that such prediction has little practical
consequence, because it is too close to the actual ST jump frequency.
'Injecting', however, an NS (TtR) mode ON the NS drop response, a stable
plateau is reached which may delay the drop process Fig. 2. Complete
explanation of the elimination of the discontinuities in robust TtR
responses is lacking.
Another only partly answered practical question is related to the determina-
tion of a corresponding to a given maximum of the TtR amplitude.

4 Deeper
reasons may be identified for the existence of these categories.
 3

The cyclic TtR may be successfully used in dynamic controls by applying

alternately sweeps a and -a to move a point from the position A to the
position B Fig. 3 (schematic).
The effects of the NS damping coefficient ;(t) and the effects of the ee
amplitude B(t) are shown in Fig. 4.
There is an inherent difficulty in defining the NS phase-planes. This is so
because of the inherent existing asynchroneity in the NS responses, due to
time variations of the phase shift * [12].
While for some cases, the NS response amplitude variations exhibit a consid-
erable amount of wavyness, the corresponding TtR 0 plots exhibit a smooth
behavior Fig. 5. It may thus be advantageous in some cases, particularly in
applications, to track the TtR manifestations using 0 rather than a.
Recently obtained partial results by Kitis and Evan-lwanowski on the NS
chaotic responses of a Duffing-Hayashi oscillator offer quite revealing
observations Figs. 6, 7, 8. For a = 10" for the first 90 sec the NS
chaotic response coincides with the It; then they diverge. However, the NS
response remains chaotic till it shows a quasi-periodic character Fig. 7.
The same picture is seen from Fig. 8 for a = 10 the divergence appears at
30 sec, and the structured, quasi-periodiC motion appears at 22 sec. Thus,
the NS modes quench the ST chaotic motion for sufficiently fast TtR.
Further studies of these points may be related to Fig. 9, showing a number
of different ST responses of the Duffing-Hayashi oscillator in ( , B) plane
[15]. Each of these responses could be cast in NS modes. The early studies
on the evolving TtR in Duffing oscillator are found in [81 in undamped
linear Duffing equation and in [9] for the same equation with damping. By
an appropriate selection of the time scales, key equations were obtained
connecting the amplitudes in the first approximation. This method was also
used in [10] for damped, nonlinear Duffing oscillator to establish a sense
ir,which NS response approaches the ST jumps. Lately, in [11] the work in
[10] has been extended for the effects of the initial conditions NOT ON the
3 STC.

Relaxation Systems. x + w2x - C(1 - x B(t) cos e(t). In evolvin TtR

interesting ress have been obtained at Northwestern University 12],
presenting evolving Hopf bifurcation, memory effect, resonance, and nerve
accommodation. Currently, Tran and Evan-Iwanowski are working on these
problems Fig. 10, 11. Needless-to-say that this area demands further
intensive studies.
Dynamic Stability. x + Q21 - 2 p cos 6(t)] x + + ux3 = 0. Dynamic
Sprobems (or parametrically excited systems) find wide and essen-
tial applications in structures of rockets, aircraft, tall antennas,
rotatory machinery, ect. These structures are often subjected to the TtR
modes. For these reasons, they recently received attention of a number of
researches [5], [6], [13, [14] pertaining to linear or nearly linear and
cyclic TtR. These results present some specifically NS manifestations which
indicate a need for further studies. These are: (1) penetration (delayed
NS responses) Fig. 12, whereby the TtR responses initiated outside of the
stable branch penetrate (delay) into the instability region (IR) and then
jump catastrophically on the stable branch; (2) The NS drag-in and
out, initially stable configuration outside of the IR is rendered

I
 4

unstable when subjected to TtR mode and vice versa. This manifestation
depends on damping, IC rate of TtR, the mode, static load, Figs. 13-17;
clearly this TtR behavior presents considerable concern for an analyst-
designer; (3) Three nonoverlapping categories of the TtR C1, C2 and C3 are
shown in Figs. 18, 19. This manifestation may have some effect on the
random vibration methodology when a is random; (4) the most striking
behavior is observed in cyclic TtR, (v(t) = v + Y Sin at in dynamic
stability problems Figs. 20, 21. These are: (a)0 for some TtR parameters Y
and a, the response is centered about the mean value (whether they are
periodic or chaotic is not yet resolved) Fig. 20; (b) for some Y and a, they
drop rapidly to static equilibrium Fig. 21, thus offering an effective means
of stabilization; (c) they wander up and down, which may for suitable
selection of a and/or Y be used for a programmed motion Fig. 22.

Interesting results were obtained on a slender column subjected to two-

harmonic excitations with the frequencies v , and v Firstly, the analysis
indicates that the system is always diveygent. owever, the experiments
showed that for Iv1 - v2J 1 Hz, the column exhibited an "unordered"
motion Fig. 23.

Experimental Work. Experimental work is essential in ascertaining validity

of some predicted phenomena or uncovering some new ones. For instance, the
experimental works established first the existence of chaotic motions, and
explained mathematically later. Experimental works conducted, at Syracuse
University and later carried out at Ecole Polytechnique, Montreal, particu-
larly in parametric stationary and nonstationary modes, show very good
agreement with analytical results. This fact reinforces the effectiveness
of the perturbation methods and computer schemes used.

CONCLUSIONS.
An obvious conclusion is that the NS processes, in our case the TtRs affect
almost all manifestations of the ST SD. A global menu thus for the future
work in this field is: (i) to apply NS processes to the well-established
bifunction theories and the conjectures as well, (ii) to proceed to study NS
attractors including strange attractors (chaos). The NS processes in the
evolving formulations found already their applications in dynamics, nuclear
physics*, chemistry, medicine, biology, etc. The robust formulations of the
NS processes are indispensable in analysis and design in structural engin-
eering, especially in dynamic stability (parametric ee.)

The NS processes constitute an indispensible component of the basic SD

studies. They also constitute an indispensible input to the technological
(engineering) point of view: They may be utilized for gaining mechanical
advantages in design and may serve to establish design criteria, i.e., in
the determination of the parameter ranges within which the predicted
responses are secured. The lack of the detailed knowledge of these
manifestations may result in dire technical consequences.

(2AE14N202TXY)

*Communication from Professor G. Hoher, Inst. Nuclear Physics, Karlsruhe.

 REERNCS
t(1965).
1. Lewis, F.M. (1932). Trans. ASME, 54, 253-261.219.
Davey, New York. vii,137,162,219.
A.P.(1971). - Verlag. Berlin.
2. Mitropoisky, Y.A.
3. Goloskokow, E.G. and Fillipow,
4. Evan- Iwanows kiR.M.( 1969) . Appi. Mech.
Rev.,22,213-219. 5. Evan-Iwanowski, R.M. (1976). Elsevier, New York.
Ivii, 219,260,267,268,402,417,418.
Dissertation, Syracuse University.
6. Ostiguy, G.L. (1976). - Ph.D.
7. Thompson, J.M.T. and Stewart,
H.B.(1986), J. Wiley and Sons, New York, N.Y. p.11, 129-131. 8. Kevorkian,
J.(1971). SIAMJ. Appi. Math.20, pp.364-373. 9. Ablowitz, M.J., Funk B.A.
and Newell, A.C.(1973). Stud.Appl Math., 52, pp. 51-74. 10. Collinge I.R.
3(1986),
and Ockendon,J.R.(1979). SIAMJ.Appl.Math.,37, pp.350-357.
private communication from Dr. Ockendon.
Professor Erneux, Northwestern University.
11. Johnson D.A.
12. Communication from
13. Ostiguy, G.L. and
Eyan-Iwanowski, R.M.(1982), ASME Journal of Mechanical Design, 104,
pp.417-425. 14. Bluteau, C.(1984). Ecole Polytechnique, Montreal. 15. Ueda
Y. (1980), Holmes, P.J.(ed.). SIAMJ. Phila. 311-322.
FIGURES:

2 S.

I.7.
4 8
Fis 1-9 . Dufn Osilao Fg .10,1. Rlxto silto
 a -0 £*0.05 ~,,,,
~.as
6 bi7)

1seat is",

wagaspiq,

20. 90,g 100. AW16

12 Ccltlon Frequency n,alos
Il 170
720 17)0I~ 20

3WI M
J/11 1.0

100. Il IN 1 3 . Rtio. 50

SPTA *IC~ SD

6..

Ilutum Fiil 143

SCOND SPATAL MODE $4* I I'"

AA
:N. -to

U ~ l~ 35 US 1. .*189 ',
lMt~~ ~ 1 Fie.e~y
JA miiw
vs ~ ,,,miro a:
dtoi
peso mnw ViNY
"M * assamam ~ f~~w"R

lid 15

S2
_________________12_23.__Dynamic__Stability
I
RESPONSE OF A SINGLE-DEGREE-OF-FREEDOM SYSTEM TO A NONSTATIONARY
PARAMETRIC EXCITATION - THEORY AND EXPERIMENT
A. H. Nayfeh and H. L. Neal
Department of Engineering Science and Mect inics
Virginia Polytechnic Institute and State Unive, sity
Blacksburg, VA 24061

In this paper, we examine the nonstationary response of a one-degree-of-freedom

nonlinear system to a nonperiodic principal parametric excitation with varying fre-
quency. We use the method of multiple scales to determine two ordinary differential
equations governing the amplitude and phase of the nonstationary response. From
these equations, we obtain the stationary responses and their stability. Next, the
modulation equations are integrated on a digital computer using several different
sweep rates and initial conditions. Then, we simulate the original differential
equation on an analog computer using several sets of values for the parameters in
the governing equation. Finally, we use a mechanical model to qualitatively verify the
resu Its.
3 The governing equations for the system under study are

S+ co2u + 2t/.u + tau 3 E2efu cos9=0 (1)

1 9= D=2w
0 + (2)

Here, w, is the natural frequency, g is the damping coefficient, a is the coefficient of

the cubic nonlinearity, f is a measure of the excitation amplitude, and the dots denote
derivatives with respect to time. The parameter t, which is much less than one,
serves as a bookkeeping device in the method of multiple scales analysis. The
excitation is nonperiodic because its frequency varies linearly with time. The fre-
quency is varied by varying the detuning parameter, o, according to

I = o + r~t (3)

where r is the sweep rate. Because of the parametric excitation, we must include
nonlinear terms in the governing equation. Otherwise, an analysis of the equation
would predict a trivial response up to a critical value of excitation and an infinite re-
sponse thereafter. The nonlinear terms can result from many sources, including
large rotations of structural elements, nonlinear material behavior, or inertia from
distributed or concentrated masses. An example of a nonlinear system with
parametric excitation is a ship excited by a head or follower sea. A simpler example
is an axially-excited beam with a concentrated mass at one end.

3 The response of such systems is nonstationary--it has changing amplitude and

phase-due to the nonperiodic excitation. We examine the response when the
excitation frequency is varied near the principal parametric resonant frequency. The
situation where the excitation frequency is swept through the principal parametric
resonant frequency is usually referred to as passage through resonance. Passage
1 through resonance could occur, for example, when a motor that operates beyond the
principal parametric resonant frequency is started up or shut down.

Using the method of multiple scales, we find that the nonstationary response of
the system is given by
a= a a9ty
(5)
2

I a' = -)ua - fa sin"-5

 3a a 2 f
y'=Z- o
os- (6)

where the primes indicate derivatives with respect to the slow time scale T1 = tt.
We find the stationary responses to periodic excitations by setting a - constant
and a' = y' = 0. There are three stationary solutions. The trivial solution a = 0, which
exists for all values of ar, and it is stable as long as I
<C0 2 +4) (7)
There are two nontrivial stationary solutions; the larger solution is always stable, and n
the smaller solution is always unstable. The three solutions are shown in the a- a
plane in Figure 1. The solid branches represent stable soluitons whereas the broken
lines represent unstable solutions.
Next, we digitally integrate the modulation equations for the nonstationary ampli-
tude and phase. Integrating Eqs. (5) and (6) will give incorrect results when a is
trivial, because we divided by a in deriving them. So we use their equivalent
Cartesian form. Near-trivial initial conditions with magnitudes around 0.01 to 0.001
are used for the digital integration. These initial conditions, which represent the
small disturbances from rest that act on the system, are used since using exactly
trivial initial conditions would require very small integration steps to get equivalent
£
results. The results (Figure 2) show five phenomena in which the nonstationary re-
sponse differs from the stationary response. First, on forward sweeps the trivial re-
sponse penetrates into the region where the stationary trivial response is unstable
I
(Figure 2a). Second, the response amplitude grows and oscillates about the stable
nontrivial stationary solution (Figure 2a). Third, the nonstationary response con-
verges to the stable stationary response (Figure 2a). Fourth, on reverse sweeps, the I
nontrivial response lingers into the region where only the stationary trivial response
exits (Figure 2b). Fifth, on reverse sweeps, the response rebounds to small ampli-
tudes several times after initially becoming trivial (Figure 2b). These nonstationary
phenomena are quantitatively changed by changing the sweep rate or the near-trivial
I
initial conditions, We note that time increases from left to right in Figure 2a and from
right to left in Figure 2b.
We also digitally integrate the original differential equation. With the same sweep
rate and similar initial conditions, this integration agrees well with the integration of
the modulation equations for small values of E except for a small shifting of the sol- I
utions with respect to one another. This shift results from approximations involved
in using the method of multiple scales and from using different measures of ampli-
tude in the two integrations. The close agreement is shown in Figure 3.
Next, the original governing equation is simulated on an analog computer. Eight
different sets of the parameters w0, u, a, f, and i are used in the simulations so that the
effect of each parameter cani be examined. Near-trivial initial conditions are again
used as noise in the analog computer prevents the use of exactly trivial initial condi-
tions. The results agree qualitatively with the digital integration results, and the five
nonstationary phenomena are found to be affected by the values of the parameters =
wo0 , , a, f, and t.. Sample forward and a reverse sweeps are shown in Figures 4a and
4b, respectively.
Finally, a mechanical model governed by a differential equation somewhat more
complex than the one studied here is used to qualitatively verify the previous results.
Sweeps with four positive and five negative sweep rates are conducted, and the five
nonstationary phenomena are again observed. The amplitude traces for a forward
and reverse sweep are shown in Figures 5a and 5b, respectively.

21
I
3 Acknowledgement
This work was supported by the Office of Naval Research under Grant#
N00014-83-K-0184, NR 4322753 and the Office of Scientific Research under Grant #

AFOSR-86-0090.

I
I
I
I
I
I
I
I
I
I
i
I
I
I
I
1
 I2 3
U

01

W I
Figure 2. Sowr
0 - --------
we
sa nr-unstable
10
tationary-rsos
ih crestable
nr In
IP

13
U

UFg - r 3. -
oso
omp0.so
-o. -i.0 dig ta
-e.g-r0at.io.0.8 -
4.8 6.0 7.

214

I I
a M sainr-ntbe - -- osainr

"4 1 1
P
N

-I;
Sze
-SI

*I 4 a £ 0 -t --
S -8 --
4

Figure 4b. Reverse sweep with r -0.2 on analog computer.

I I i I S I 40

Si--4 i(61
16

0 -I- - - Ti .. ."
I
3 ON THE PROBABILITY DENSITY FUNCTION OF THE RESPONSE OF
NARROW-BAND EXCITED NONLINEAR OSCILLATORS

Qiang Liu and Huw G. Davies

Department of Mechanical Engineering
University of New Brunswick
P.O. Box 4400
Fredericton, New Brunswick, Canada E3B 5A3

3 Abstract

The probability density function (pdf) for the envelope of the

response of a nonlinear oscillator driven by random narrow band
noise is discussed. The nonlinearity treated involves general
cubic terms and includes the Duffing and van der Poll oscillators
as special cases. The input excitation envelope is Rayleigh
distributed. The response envelope pdf is obtained for the
limiting case of small excitation bandwidth. It includes cases
of multiple valued response (Duffing) and limit cycles (van der
Poll).

The research was supported by the Natural Sciences and Engineer-

ing Research Council of Canada.

3 Introduction
We consider in this paper the response of a nonlinear oscillator
I described by the equation
0. + 6 + 2
W v + g(y, ) = f (1)
0 -

where f(t) is a narrow band random excitation, and g(y, ) is a

general cubic nonlinearity.
I g(y,') = U1 y2 + U2 2y + 3 k3 + P4 y3 (2)

The excitation is described by

f F s in' ft + F coswft (3)

Fc 2f ) wc - -2 Fc (4)

I F s =() ws - f F s (5)

3 where w
spectru
and w are independent gaussian white noise sources with
level S 0.f has bandwidth Y at excitation frequency wf.
I
I
Now if f in (1) is white noise the stationary two dimensional pdf
p(y, ') can be obtained for certain forms of g(y,k) . In the
present narrow band case, although the Fokker Planck equation for
the four dimensional pdf p(y, ,Fc ,Fs ) is easily obtained, we are
not aware of solutions of this equation for any form of g(y,S).
In what follows, we first obtain the time dependent envelope
equations corresponding to equations (1), (4) and (5) using the
approach of Rajan and Davies [1]. A simplified form of these
envelope equations is then used to find the pdf P(a, a ) where a
and af are the response and excitation envelopes, respectively.
Numerical simulations are used to corroborate the analytic work.

Envelope equations
We start by making the exact transformation (see, for example,
Roberts and Spanos [2])

y = x cos Wft + z sinwft (6)

= -wf x sinw ft + WfZ cos Wft (7)

Equations..(6) and (7) are substituted into (1), (7) being used to
evaluate y , and two separate equations obtained for ± and 2.
These equations are averaged (over one period 2 -/.wf) as in [2]
and then squared and combined with (4) and (5) as in [1]. After
some algebra one finds the results
&2 = _ Sea 2 + P/ 2 wf (8)
P = -(y + ae) P/2 + 6 e Q + af 2 /Wf (9)

+ Be )Q/2 - 6 ep (10)
= -(y

where a2 = (x 2 + z 2 )/2 (11)

af= (Fc 2)/2 (12)

P =Fcz - Fsx (13)

Q = F x + F z (14)
3
u3wf 2 (15)
Be = B + ULa2/2 + a2/2
2
6e = ( f - W - 2wf2 a2/2 - 3u 4 a2/2)/ wf (16)

It is clear that the multiple time scale approach of Nayfeh and

Mook [31 instead of the averaging of [21 will yield the same
results. Be and 6e represent equivalent damping and frequency
difference, respectively. The equations (8) to (16) reduce to
those of [1] for the case Ul = u2 = u3 = 0.

Equations (8) to (10) describe a third order system for output G2

with slowly varying random input af 2 We are still dealing with
I
3 a four dimensional pdf, and again have not found a solution.
Instead we approximate (8) to (10) for the case of very small y.
The response envelop can exhibit multiple values, and a jump
phenomenon from one pseudo stable level to another. Except for
transients associated with the jumps, however, we are dealing
with a slowly varying process that can be described by just the
particular integral component of (8) to (10). In combining (8)
to (10) we are thus justified in assuming, for example, that & 2
= 02 0 (Y), so that second and third order derivatives and prod-
ucts of first and higher derivatives may be neglected. The
resulting first order equation from (8) to (10) is

I A 12 + 8 A 22 = 0
(e 2 + (y + B ) a f 2 /2)/2w.2 (17)
(17)
where

A 1 (G 2 ,af2 ) = 6e2 + 5 8 e 2 /4 + 38e D 1 G2/2

2
S e 2D D2/(26 ) + D 2a f2/(2wf 26

Be A 2(2) = 8 (6 2 + 8
e2 e e e 2/4 + yS e /2)

D 2 = 6 e/G 2 , D 1=
" 6e/02

In order to describe the excitation we return to (4) and (5) and

use the stochastic averaging method of [2]. Average values of
w F and w F are approximated by mean and fluctuating parts. We
t~uS obtaiR s

F2 = y (27S - F 2) + (47ryS oF2 G(t) (18)

where aF2 =2 ;2 and G(t) is unit amplitude white noise. We note

that the pdfl(a F2) obtained from (18) is exponential, so that
p(a F) is Ravleigh as expected. Finally we combine (17) and (18)
to give
1 G F) i(4S y 2) G(t)/( 4wf 2 A1 ) (19)
3 where
-A 2 e2 rS oY (8e- y)a F2

f1 A! +Wf 2A 8wf A1
1

Equations (18) and (19) are two nonlinear state equations for the
excitation and response envelopes with unit amplitude white noise
as the basic input.

I
Probability density function

The stationary form of the Fokker Planck equation for p(a2,o 2)

is easily written. We assume that for small Y p can be written p
= p + Yp + Q(YZ)• To zero order Y the Fokker Planck equation
reduces tJ
a {fl(o2a F2) ,P(a2,F2)) = 0 . (20)

We write p (G2 , F 2 ) = p(aF2) po(a2b0F2). The appropriate

solution of F20) is
6 (fl(72 (21)
pO (a2 a F 2) = (constant) a 2F))

The required pdf can then be obtained in the form

00 a 8
a ) 6 {_e (-A G2 + aF2/8wf da2F
p ( 2) = (constant) f exp (- )
0 o 2lSo A1 2 F f
(22)
For the Duffing case, which is compared with numerical simulation
in the next section, integration of (22) yields (transforming
p (02) to p (a))

pO (0) = (constant) UJ6 e(6e -3 4 92/2wf + 5$2/41

4wf 2

exp (- -- (6 2 +B2/4)2 02) (23)

7TS e
0

This form reduces to the correct Rayleigh distribution in the

linear case Iu
4 = 0. The form (23) however, shows two local
Iitaxima (a bimodal distribution) in cases where the response
envelope is multivalued.

For the van der Poll case, with U2= = u 0 and U1 = -8, we
have 6 = 6(1 - 02/2). The pdf (22) in this case exhibits a
delta Ainction at the limit cycle 02 = 2.

Numerical simulation

Numerical results are shown in the figures for the Duffing

oscillator. Figures 1A and lB show the time histories of the
response envelope; 1A is obtained from a direct integration of
(1) to (5), lB shows that the approximate equations (18) and (19)
are adequate for this very low Y. Figures 2A,B,C are the pdf
p(c) ; 2A obtained from the time history A., 2B from 1B, and 2C
from the analytic result (23) . The multivalued nature of the
response shows up clearly in the time histories and the pdf.
 References

1. S. Rajan and H.G. Davies, J. Sound Vibn. accepted for

publication (1987).

I 2. J.B.
(1986).
Roberts and P.D. Spanos, Int. J. Nonlin. Mech. 21

3. A.H. Nayfeh and D.T. Mock, Nonlinear Oscillations, Wiley

(1979).

I.

I........
, ..

t 49
 I
U
WIDE BAND RANDOI EXCITATION OF A THREE-DEGREE-OF-FREEDON SYSTEM

WITH PRINCIPAL INTERNAL RESONANCES 3

R. A. Ibrahim and W. Li I
Wayne State University
Department of Mechanical Engineering
Detroit, MI 48098

Introduction I
This investigation is a continuation of previous work [1,21 on the random
excitation of a three-degree-of-freedom structural model whose normal coordin-
ates are nonlinearly coupled. The modal coupling is due to inertia quadratic
I
nonlinearity and results in four different internal resonance conditions. One
of these conditions involves the nonlinear interaction of three modes, while
the other three give rise to autoparametric interaction between two modes.
The t ree internal resonance conditions are of principal type, (w. 2w.,
where w. and w. are two normal mode frequencies of the system), and ?ave dii-
ferent Aegrees of response dynamic characteristics. The random response cor-
I
responding to the three cases is determined by using the Fokker-Planck
equation approach to generate a general differential equation for the response
joint moments. This equation is found to constitute an infinite coupled set
I
of differential equations which are closed via a non-Gaussian closure scheme.
The closure scheme is based on the properties of higher order response cumu-
lants. The anaiytical derivations of the equations of motion in the Markov
vector form and the response moment equations are performed by using the com-
puter manipulation software MACSYMA. 5
The linear
conditions.
modal analysis
These are:
of the system reveals four possible internal
3
ii.
.W3 =
2= 2
+w2
£I
iii. = 2
iv. w2 = 2 w,

The system response corresponding to combination internal resonance (i) is

determined in references (1,21 by using Gaussian and non-Gaussian closure
schemes. The Gaussian closure solution predicts nonlinear interaction
I
between second and third modes at the internal detuning parameter r=w 3 /(w1+w2 )

1
= 1.18. This unexpected result is scrutinized and it is found that at r =
1.18 the second and third modes are in exact internal tuning, i.e. w =2w 2.
The non-Gaussian closure, on the other hand, successfully predicts nonlinear
three-mode interaction in the neighborhood of r=1.0. The autoparametric
interaction occurs among the three modes in such a way that the mean square
response of the first two modes are always greater than the linear solution,
while it is less for the third mode. A new feature of considerable interest
is the contrast in the form of the mean square response curves above the exact
detuning ratio r>l.0 for a certain combination of system parameters and
excitation level. This is indicated by multiple solutions over a finite
portion of internal detuning parameter. The well known saturation phenomenon
[3] which usually occurs in deterministic systems with quadratic nonlinearity
is not predicted because the excitation is random and includes a wide range of
frequencies which always excite the system modes.

The present analysis is extended to analyze the system response in the

vicinity the three principal internal resonance conditions (ii) through (iv).
This means that the analysis is restricted to autoparametric interaction bet-
ween two normal modes in a three-degree-of-freedom system. If Gaussian
closure is applied one should generate 14 differential equations in the first
and second order response moments. On the other hand, if non-Gaussian closure
is used one should generate 69 moment equations in the first through fourth
order moments. The solutions of these two sets of moment equations are ob-
tained for each principal internal resonance condition.

First and Second Modes Interaction: w2 = 2w,

The Gaussian closure solution predicts both stationary and nonstationary

responses depending on the value of the internal detuning parameter r = w2/wl.
The response is stationary when r is well remote from the exact internal
tuning r=2. The non-Gaussian closure solution yields a stationary response
for all value of r. The time history records display the interaction in a
form of energy exchange between the two modes during the transient period. The
basic characteristics of the mean square responses occur in such way that the
motion of the first mode acts as a vibration absorber to the second mode. It
is observed that the nonlinear interaction takes place over a narrow band of
internal detuning parameter r. Contrary to the first case of combination
internal resonance the nonlinear interaction is less sensitive to the external
excitation level. In other words, the interaction is only manifested under
relatively higher excitation spectral density level, DI2C 2

First and Third Modes Interaction: w3 = 2w,

Unlike the previous case, the Gaussian closure fails to predict any nonlinear
interaction and all solutions are identical to the linear response for all
possible values of system parameters, excitation spectral density levels and
initial conditions. The Gaussian closure scheme, on the other hand, gives
results which are different from the linear response and involve nonlinear
interaction only for excitation spectral density D/2C 3 level greater than 40.
Below that level the response is completely linear.

Second and Third Modes Interaction: w 3 = 2w2

In the neighborhood of the exact internal resonance the Gaussian closure

scheme gives a quasi-stationary response. The nonlinear interaction is found

2
to take place over a small range of r = 2 + O(e). The non-Gaussian closure
solution shows fluctuations in a form of energy exchange between the two modes
during the transient period. These fluctuations are completely vanished
during the steady state response. The steady state is also determined by
setting the right-hand sides of the closed 69 equations. The resulting
nonlinear algebraic equations are solved numerically by using the IMSL
subroutine ZSPOW. The results are determined as function of the internal
detuning parameter r and it is found that the region of autoparametric inter-
action becomes more wider as the nonlinear coupling parameter increases.

Conclusions

The random response of a three-degree-of-freedom structural model is deter-

mined in the neighborhood of three different principal internal resonance
conditions. The response statistics are sensitive to small levels of the
excitation spectral density when the third normal mode frequency is twice the
second normal mode frequency. However, the autoparametric interaction is only
sensitive to a relatively high excitation level when the first and second or
the first and third modes are internally tuned. The stochastic interaction of
the three cases is characterized by irregular energy exchange between the
interacted modes.

Acknowledgment

This research is supported by a grant from the Air Force Office of Scientific
Research under grant no. AFOSR 85-0008. Dr. Anthony K. Amos is the program
manager.

References

1. Ibrahim, R. A. and Hedayati, Z., "Stochastic Modal Interaction in Linear

and Nonlinear Aeroelastic Structures," Probabilistic Engineering Mechanics
1(4), pp. 182-191, 1986.

2. Ibrahim, R. A. and Li, W., "Structural Modal Interaction with Combination

Internal Resonance under Wide Band Random Excitation," Journal of Sound
and Vibration 123(2), 1988.

3. Nayfeh, A. H. and Mook, D. T., Nonlinear Oscillations, Wiley-Interscience,

New York, 1979.

3
I
IA GENERALIZED METHOD OF AVERAGING FOR DETERMINING THE
RESPONSE OF NONLINEAR SYSTEMS TO RANDOM EXCITATIONS
by
Ali H. Nayfeh and Samir J. Serhan
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, VA 24061

In this paper a generalized method of equivalent linearization is

developed for the determination of the response of systems with cubic
nonlinearities to random excitations. The method explains some observed
nonlinear phenomena such as non-Gaussian responses, broadening effects,
and shift of the resonant frequency. The method is illustrated by
studying the response of the Duffing oscillator to a Gaussian white
noise.

Problem Formulation
We consider a second-order nonlinear system with stochastic
3 excitation described by the equation
2
u + w0 u + 2 Eu6 + u3 = F(t) (1)
I where wo is the linear natural frequency, u and a are constants, e is a
small but finite constant, and F is a zero-mean Gaussian random
3 excitation with a white power spectral density So of 0(e).
The method of equivalent linearization is based on replacing the
nonlinear equation (1) by an equivalent linear equation whose solution
3furnishes an approximate solution to Eq. (1); that is,
2
U + weu + 1e 6 = F(t) (2)
The equivalent stiffness we and damping ue parameters can be determined
by minimizing the mean-square value of the difference between Eqs. (1)
and (2). Equation (2) shows a linear relation between the statistics of
the excitation and those of the response. The equivalent linear model
given by Eq. (2) fails to account for many of the phenomena peculiar to
nonlinear r items. These include non-Gaussian responses, multi-valued
responses, jumps, modal frequency shifts, broadening effects,
superharmonic, subharmonic, ultrasubharmonic and combination resonances,
period-multiplying bifurcations and chaos. The proposed method accounts
for some of these phenomena.

In investigating the response of panels to acoustic loading, test

results (1,5,6) demonstrated the broadening of the response curves and
the increase of the resonant frequency of the fundamental mode at high
noise levels. To explain these phenomena, Mei and Prasad (2) had to
include nonlinear damping terms in Eq. (1). Using the proposed method,
we can explain the experimental observations without the resort to
nonlinear damping.
I
I
Proposed Method
The random excitation in Eq. (1) is divided into two uncorrelated
parts as F = fl + f2- The first part is f, whose spectral density is
around wo. The second part is f2 whose spectral density is away
from wo" Moreover, we express the solution of Eq. (1) as

u = x +v (3)
where
2
x + w0 x + 2eui = f2 (t) (4)
Substituting Eqs. (3) into Eq. (1) and using Eq. (4), we obtain the
following equation governing v:
2 2 2 3 3
v + wv+ 2 +3ex v+3xv
+ + aV f(t)
3 =xx (5)

Next, we replace Eq. (5) with the equivalent linear equation

2
V + Wev + 2euv = g(t) (6)
where 2 2 2232

We = Wo + <3ecx v + 3eaxv + eav > / <v > (7)

g fI - ex (8)
3

The presence of Ex in the excitation g provides 2t~e non-Gaussian

feature of the nonlinear response u. The term <x v > accounts for both
an increase in the linear frequency and ultrasubharmonics, the term 3
<xv > accounts for the subharmonic of order one-third, and the term x
in g accounts for the superharmonic resonance and all combination
resonances. If we neglect the subharmonic and ultrasubharmonic
resonances, Eq. (7) becomes
2 2 2 2
We = W0 + 3Ea<x > + C<v4>/<v > (9)
If we assume that v is psuedo-sinusoidal, then <v4> : <v2> and Eq.
(9) becomes 2
2 2 + 2 3 2 (
We = o 3e<x> + 7a<v
2 > (10)
Using Eq. (4), the mean-square value of x cin be written as

I
U
2
S(W+Wd)2+E 2 C 2 2 (Wc+Wd)2 2 2
So + 1. (w-w)+ (+)+
> =
<x [[in - in + 1n
(2l2d)2 2 2 2+ 2 2 2 2 2]
4ow d d+E (W2-W) (W2+Wd) +EU
S [tan0 w-d + tan- W +d c-Id i Wc+d
+ [ - + tan - I w + tan-
3 2u
2uw 0 E4 l E

tan'
- 2+wd (1
tan - I W +WI
-u-

where wc is the cutoff frequency and

W =W - S2 ,
2 o +12 , and = (W - 2 u ) (12)
Here, s is the bandwidth of the narrow-band process f,.

I As a result, the nonlinear equation (1) is replaced by two

equivalent linear equations (4) and (5) and the mean-square response has
form
2the

2 2 2 2 2 2 2 32

2EU 1loWe( o-_e) +(2 o+6)(2 1 + 2;Je

L1+BTSe)] (
where w is given by Eq. (7) and

Sx2(we) = 9<x2>Sx() + 6 Sx(r)S(r 2 )S(W - r r 2 )drdr 2

. -O e (14)

Here, S (w) is the output spectral density of the linear response x in

Eq. (4 )x
Sf 2(W)
x()S = 2 2 222 2(15)

I Conclusions

A method is presented for determining the response of nonlinear

systems to random excitations. The original nonlinear governing
equation is replaced by two equivalent linear equations. The method
provides some means to describe the non-Gaussian feature of the
nonlinear response and preserves its wide-band character. The
interaction and feedback between the nonlinearity and excitation creates
nonlinear resonances. These lead to a linear shift in the frequency and
an adjustment of the strength of the primary excitation. The former is
responsible for the increase in the modal frequency and the latter
contributes to the experimentally observed broadening of the response
3 curves.

I
I
Acknowledgement
This work was supported by the Air Force Office of Scientific
Research under Grant # AFOSR-86-0090 and Grant # F49620-87-C-0088.
References

1. C. MEI and K. R. WENTZ May 1982 23rd Structures, Structural

Dynamics and Materials Conference, New Orleans, LA 514-520.
Analytical and experimental nonlinear response of rectangular
panels to acoustic excitation.

2. C. MEI and C. B. PRASAD 1986 27th Structures, Structural Dynamics

and Materials Conference 644-653. Effects of nonlinear damping
on random response of beams to acoustic loading.
3. A. H. NAYFEH 1973 Perturbation Methods. New York: Wiley.

4. A. H. NAYFEH 1981 Introduction to Perturbation Techniques. New

York: Wiley.
5. J. SOOVERE 1981 J. Aircraft 19, 304-310. Sonic fatigue testing of
an advanced composite aileron.

6. R. G. WHITE Oct. 1978 Composites 251-258. Comparison of the

statistical properties of the aluminum alloy and CFRP plates to
acoustic excitation.

• II11 I
g
I
I
I
I
I DYNAMIC SNAP-BUCKLING
UNDER
3 STOCHASTIC LOADS

S. T. Artj'atuxaz and We-Chan Xie,

Solid Moeoduics Divimon,
Facnity nf u5IpCecring,
Universtity or WaLerloo,
A Waterloo, Ont.%no, Canad, N21. 3(;.

In shallow CurVwd .tructurts such w *reies and shells under Ayn.iietrically distributed loads,
tuiiiaUy only the symmotric node of deformatiol; i exiited. 1owpier, tue to the inhere.n4 wa-
linear coupling, the antiJynmietric modc can albo be excited under cerrain cuudiuons and become
unstable leading to .nap-bckLiug of the structure.
We consider the p1e motion of -, simply supported shallow vch of uniform err -ecLion
whoec axis in the inloaded confignritinn im,a form w4(z), on which thc dynamic delection
w(r.t) is superpoed. There is a lateral lu'sdiijS p(zJ) isshown in Figure 1. The problm te; be
considered here is that of a low, half-.ine, pined arch lo.aded stocLastically by iP, f.buc spar
tially distributed load. The initi-l ishape i3 taken in the form

where q. Lsthe initial rMe paramekr. The cxpreseiin fir the loading, directei upward, is givcn by

£ whe~re F11) is
I-
random procese with meau value E1Ij)I=F.) O. The dyunnuc defletion may be
a
Prprreieuted approximately by r

I where qj. 42 ar ei.amplitudes of the symmeitric: and Lhe aaitiaymmetric tnind...

tione of motion are gien by
Then, the equa-

I
I
 2I
4
pq+ ( •, ' 4 E . (q 1 )q+4?.q ,)

wiere tAis the ma per unit lsr.th. it i., tc viscous damping crowrricirtiL per unit length, EA %nd
Kt'l drizvie. respuctvcly. the aia[ And the nxeaurid 3LtjfncW of the arch Pib.
If EP(I)I-Po 0 #o, we take A(t)- o- 0 ( ), where 4t) it a zero ,,,iua, *ide-band procees
which c.t be approximated by a ('xatsmian white name, i.e.

bifj- 0. and E'4t)"(1+r)I =jef(r),

where x is the intensity of thr kP'd fluctuation, and 6(r) te the )irac delta ftuction.

It is known (Paing and Kaplan *iS2')that under a p)rply .atdtic load Pa, it the initJ ri.o 7
is such that qO/2jo>V'5.S, 3nap-buckling into a n1oa-ytiunetric atodc can oacur ;f P>,P.,-
2l'pfiL)4 lq,,a+3(q /4, 3 -4) 1 ', where ( The effct of a %Utwhxtirlood on the
stability of tfic itfucture when qr,,2p>'%.!j5 Andl Pu'<P, is considered in the follnwing.
It amntA.d that only the symmetric mode is directiy xcited and the ntifytruetric moud is
at rest. Then the first of equations (1) becomes

-- T'0 +€t ). (2)

.m Uijug q - qj,4-X 1(t),where el, to the deflection due to the 3t&tic load P,I
Wrr,.pe.),iu..
and Xj(t) to the addtional deflection due to the random load Ot). the static deflection q. is
'iven by
kA 'q 4 2q qi. f =(a
4 L
Then, subtrxetie equation (3) from (2) leads to the e vatioo for the deflection component
corresponding to the fluctutiwual part of the load:

The AtaionAty mnarigiul probability density of the displacement proequcM X,(L), obtained by
.owving the associated lokker-Planck equation is

E4 W 1
)YEA) '2q '+q I+J^

where C, in the normalisation constnt

To mvestigawe the stability of both the ecited mode and the r.t mode de.eribed by equ&-
tions (I), %jubstitltingthe pertihed sohttion
I
I
I 3

q: - qi.+X,+2t, M0 0 (a)
equations (1), neglecting all nonlinoar trm aud uuikiug use of equations (3), and (4) recullt.
Wnto
in

3 .. . .+ .(7)

or
, i - 1,2,

where

fi I=
. 1.,

'- "l (-! ' ( 4p 2+ 2q/ O q q ' ,+3 I ), 21 = - [ J l p . + , F ( , ,) O

M d ," .. .. A r
RS , .. .=

3 quAtions (8)are of the forn0)

= 0,
where g(i) is an ergodic random proc..a. Since the prubability density of the random proem X,
p(=z). is known, it is posible to find th prImility deu.ity of the random precs A. According
to the mltiplicative ergodl theorem of Oao.l dle 1M8], equition (10) possescse two real
Li~puaov exponents X1,12 (X: > Xj) definod by

I 1.o) = mrIPI 1(t ,zo)II, i-,,

±IOj- (I)

ror appropriate initial random variable sa.

If both *,, X, are negative, the solutions are uymptotically stable with pmh:Lhiliky 1 lw.p,l).
The solutions art unstable a soon as ).2=0. Tn general, the Liarunov expolaeau are diffi-
cult to evaluate oven ntmenrieally. Howevor, for eefitation (10., it v possible Lo obtain upper
bowids to X. asig a metlhod due to laute '19081 ad its extension by Kozm and Wu [19731,
from wjch sufficient condirion% for .ymj)'Ati .sblwleb w p.1 wuay be ioingdo
m As au exajulie, We Cutivider a 3ieel arch, of rectangular crosseecio having the following
parameters: density of material -7850 kg m-, Young's modulus E-U.'Xl0' 2 N in4 . recrangu-
lar erom-setion with height - 0.03 m, width = 0.02 m, length of span I - I m, initial rise

3 q,- 0.07 in. Two v3lu for the mean load Pj are considered, namely PI-- IX104 N,,,XI04 IV,
corresponding to static deflection q 1-- )-.3"2lxO ?,n, -C.I7!RIxtO a in, rtpectively.

I
!
I
 4

S'mmcieon .s.. aymptotic st&ability i, nit for both the vibrating mnde and the rest aLode are
obtained by using the Schwan iequAlJity ;nd an opt~isaiun mcthod the result are Tpl ,tedin
Figures 2 and 3.
Fnr further details, reference may he made to Araratuam and Xi I19881

ACKNOWLEDGEMENT
This research was supported by thu National .ctenecs and Engineering Re efaret Council of
Cauada through Grant NoA-IRJ5.
ALEFERZNCES
Ari2kriam . S.T arid Xe, W.-C., 1488, "Dynamic Snap-Buckling of Structures under S'cbastic
Loads*, to appear in R.'cru Progress in Stochastic Strurfurat Dtmantaie, S.T. Ariaratnam, G
Schup]ler and I. Eishako( (Eds.). Elsevier, I.K.

Fung, Y C and Kaplan, A., 19.5, "Buckling of Low Arches or Curved Beam of S-udl Curva-
turc', NACA Techa. Note 2840.
TnfAntp, E F., 1968, I)ai the Stability nf -Some Linear Nouautonomoun Random Systems', J.
.4ppi. .forh , Vol.35, No.1, pjp. 7 -12 .

Kojn, F. and Wu, C., 1973, "On the Stability of .inpar Stochastic )tffereuti], Equatione", J.
Appt. Mech., Vol. 40, pp.7.92.

Onael ee, Y1.. 1962, "A Multiplieauv Ergodie Theorem, Liapunov Characteristic Ni 'nber for
Dynamireal Systems", Trudst Mooko. Aat. Obec., Vol. 11, pp.17I-210, (Enolish Translation).

Fig.I Two-Pinned Arch and Loading

I
I
I
I
I
I
I

Z7777 sabe regioas u /1 ~p77*77~stable relions

L P 0 -5x10'N

P~LXIOjV 4
I 4PO-IX10 N

IIx
U4
1 2

I .0 0.2 0.4 0.0 0.9 0.0 0.2 0.4 0.6 0.5

I Regiong of Almost-Sure Asymptotic Stability Fig.3 Regiors ot'Almost-Sure Asymptotie Stabi

(First Mode) (Second Mode)

3- Stability lonndary via Schwars Inequality Stability Boundary via Schwam [nequw

-
I
I
I
I
 A NEW APPROACH TO

STOCHASTIC FLAP-LAG STABILITY

OF A ROTOR BLADE IN HOVER

by

N. Sri Namachchivaya
and
John Prussing
Department of Aeronautical and Astronautical Engineering
University of Illinois at Urbana-Champaign
104 South Mathews Avenue
Urbana, IL 61801

ABSTRACT

In recent years, considerable progress has been made towards

understanding the dynamics of helicopter rotor blade systems in both hover and
the forward flight conditions [1-4]. From an operational, as well as safety
point oT view, the most important problem is certainly the stability of the
system. However, theoretica' investigations on rotor blade stability
published to date, with a few notable exceptions, haie been restricted to
deterministic analyses.

In the service life of a helicopter, numerous encounters with

thunderstorm and clear-air turbulence can be expected. Furthermore, because
of the very nature that lift is generated by blade rotation, some level of
self-created turbulence is also unavoidable. Therefore, randor turbulence in
the atmosphere should be included in a realistic analysis. This has been done
partially in, for example, References [5-7], where the vertical component of
the turbulence velocity has been taken into consideration. The inclusion of
random vertical inflow results in additional non-parametric excitation terms
appearing on the right hand side of the equations. The objective was to
calculate certain statistical properties of the structural response, including
the spectral density, level crossing averages, and peak magnitude
distribution.

It is well known, however, that stability of a dynamic system depends

only on parametric excitations. The inclusion of vertical inflow in Refs. [5-
7] does not change the stability analysis; that is, the stability analysis
remains deterministic.

A preliminary analysis of the stochastic stability of flap-lag rotor

blade motion was made in Ref. [8], In which the turbulence was modeled as
white noise and the method of stochastic averaging was employed to dutermine
first and second moment stability. Reference 8 is a continuation of the work
 2

in Refs. [9-10]. Related analyses of rotor blade stochastic stability and

response are given in Refs. [11-13).

In the present analysis the problem of Ref. 8 is approached from a

different point of view, namely that in Refs. [14-153. The flap-lag system is
analyzed as having a single critical mode (lead-lag) and a highly dampeu mode
(flap). Using the methods of Refs. [14-15] the nth mode stability can be
analzed with significantly less algebra than the mean-square (second moment)
3 stabiiity of Ref. 8.

3 The linearized equations of flap-lag motion in hover are given

terms of the flap angle B and the lead-lag angle 4 as [Ref. 8]
in

68 6 6 0
() [+
[E c1( )+ [i + K] 64 )1 (0

£ Iwhere

2- h(2 -) h(-2 + 2Cd /a)

-- , ( 0)

P - he B . Z - he

oe
+ Re2( -2) W hej
B 2

£ represent the deterministic damping and stiffness, and

C 4h/3 [ 1
VJ
V 'V h
I -8h/3 4h-13 -2"

!0 -8ho/3, 0 |-2i 0

3 K -v = vh
06 L /3
 3

represent the stochastic contributions, where v(t) represents the vertical

turbulence component, and the other variables represent various system
parameters (see Ref. 8).

Equation (1) can be written in first-order form as 3

x A x + v(t) B x (2)

The analysis of Eq. (2) becomes easier if it is brought to the

simplest partially diagonal form that still retains real variables. To this
end, consider a transformation x = Ty, i.e., X =

x 2(c u1 + d u2 ) + 2(c 2 + d )2 I(
where superscripts 1 and 2 correspond to the eigenvectors of the critical and
Substituting Eq. (3) in Eq. (2) and
3
stable eigenvalues, respectively.
premultiplying by the adjoint-eigenvectgor b yields

SC 0 u + (Ku + My) v(t) (4a)

v D v + &Nu + Lv) v(t) (4b)

where

c= w oj c= L'
[ 1
DO-
F
L2 2 Wl
2

2 I
prime denotes the differentiation with respect to w o ,
I
For the stochastic problem, i.e., v(t ) 0 it has been shown by Sri
0,
Namachchivaya and Lin C15] that the stochastic terms of Eq. (4b) contribute to
the drift and diffusion coefficients of Eq. (4a). Thus, utilizing the ideas
from the determinJstic and stochastic averaging theorems, as indicated in
[15J, the amplitude and phase converge weakly to a Markov diffusion process
°
with infinitesimal generator L .
I
L f(a,$) = m (a) L f(a,) a03a
+ m (a L f(a, ) #
 4

£_ [C°a i2 (a,O) + [oo] i 2f (a,o)

a_ + [°°y]a a (a,*)
a2r, o (5)
(5

2 [ aa --2 42 0 aa

IConsider the Fokker-Planck equation associated with L ° .

this equation over * (a3suming periodicity in *) yields
Integrating

I
=
2
at -a-a '
)aa[{I p--
p] + -L 3aa 2 [a2 p] (6)

3 Iwhere
C ' 8- 2 sL(2"
(- 1 )
) + , Y L [2K ( (2w)]

3 SJ(WI + 2) + 'J4S "1 w2 ) - . 2 )

I
3P6 - 'i - W2)
4

S (M) - 2 f0 R ,(T)cos WT d-r, SE(W) - 2 f 0 e 2 RE{, (r)cos wT d,

4,(Wi) = 20 R (T) sin wT dT, T4 (W) = 2 fO e R,,(T)sin wT dT

I 00

p(a,t) -L
210
f p(a,¢,t) do

and (, I - 1,2,...,6 are defined in terms of coefficients of matrices K, M, N

in [1 ]. The above equation is similar to that obtained in [14]. From Eq.
(6), the stability condition for the nth moment of the linear system can be
written as
16(6'n + + (n + 2) K 2 S(2w I) 2n 1
K15&(0) < 0 (7)

I Equation (7) represents a significant simplification over the mean

square stochastic stability analysis of Ref. 8. In addition, higher moment
I
 I
5
5
stability can be assessed. The simplification occurs because there is a
single critical mode. This allows a problem of smaller dimension to be
analyzed. 1

REFERENCES 1
1. Sissingh, G. J., "Dynamics of Rotors Operating at High Advance Ratios,"
Journal American Helicopter Society, Vol. 13, No. 3, July 1968,
pp. 56-63.
2. Sissingh, G. J. and Kuczynski, W. A., "Investigations on the Effect of
Blade Torsion on the Dynamics of Flapping Motion, "Journal American
Helicopter Society, Vol. 15, No. 2, April 1970, pp. 2-9.

3. Peters, D. A. and Hohenemser, K. H., "Application of Floquet Transition

Matrix to Problems of Lifting Rotor Stability," 26th National Forum of
the American Helicopter Society, Paper No. 412, Washington, DC, June
1970. 3
4. Peters, D. A., "Flap-Lag Stability of Helicopter Rotor Blades in Forward
Flight," Journal of the American Helicopter Society, Vol. 20, No. 4,
October 1975, pp. 2-13.

5. Gaonkar, G. H. and Hohenemser, K. H., "Stochastic Properties of

Turbulence Excited Rotor Blade Vibrations," AIAA Journal, Vol. 9, No. 3,
March 1971, pp. 419-424. e
6. Wan, F.Y.M., "Nonstationary Response of Linear Time-Varying Dynamical
Systems to Random Excitations," ASME Paper No. 73-APM-6, June 1973.

7. Gaonkar, G. H., "Peak Statistics and Narrow-Band Features of Coupled

Torsion-Flapping Rotor Blade Vibrations to Turbulence," Journal of Sound
and Vibration, Vol. 34, No. 1, 1974, pp. 35-52.

8. Prussing, J. E. and Lin, Y. K., "Rotor Blade Flap-Lag Stability in

Turbulent Flows," Journal American Helicopter Society, Vol. 27, No. 2,
3
April 1982, pp. 51-57.
9. Lin, Y. K., Fujimori, Y. and Ariaratnam, S. T., "Rotor Blade Stability in I
Turbulent Flows - Part I," AIAA Journal, Vol. 17, No. 6, June 1979,
pp. 545-552. 3
10. Fujimori, Y., Lin, Y. K. and Ariaratnam, S. T., "Rotor Blade Stability in
Turbulent Flows - Part II," AIAA Journal, Vol. 17, No. 7, July 1979, pp.
673-678. 3
11. Fuh, J. S., Hong, C-Y.R., Lin, Y. K. and Prussing, J. E., "Coupled Flap-
Torsional Response of a Rotor Blade in Forward Flight Due to Atmospheric
Turbulence Excitations," Journal of the American Helicopter Society, Vol.
1
28, No. 3, July 1983, pp. 3-12.

12. Prussing, J. E. and Lin, Y. K., "A Closed-Form Analysis of Rotor Blade 1
Flap-Lag Stability in Hover and Low-Speed Forward Flight in Turbulent
I
 6

Flow," Journal of the American Helicopter Society, Vol. 28, No. 3, July
1983, pp. 42-46.
13. Prussing, J. E., Lin, Y. K. and Shlau, T-N., "Rotor Blade Flap-Lag
Stability and Response in Forward Flight in Turbulent Flow," Journal of
the American Helicopter Society, Vol. 29, No. 4, October 1984, pp. 81-87.

14 Namachchivaya, N. Sri, "Hopf Bifurcations in the Presence of Both

Parametric and External Stochastic Excitations," Journal of Applied
Mechanics, to appear.

15. Namachchivaya, N. Sri and Lin, Y. K., "Application of Stochastic

Averaging for Systems with High Damping," Journal of Probabilistic
Engineering Mechanics, to appear.

|4
 SESOI
NOLNAIIRTO OTO
WENSAI60-13
Jue1U18
U
I
3 NONLINEAR OSCILLATIONS OF LARGE SYSTEMS
WITH LOCALIZED NONLINEARITIES
3 P. HAGEDORN
Institut fur Mechanik, TH Darmstadt, HochschulstraBe 1, 6100 Darmstadt
W.Germany

U Abstract
Numerical simulations of the dynamics of large mechanical and electromecha-
nical systems are becoming more and more common in the area of aerospace
structures and in other fields. In many of these systems the nonlinearities
are concentrated in a few elements only. the larger part of the system
being linear. This class of systems forms the subject of the present paper.

!x
Fig. l: Complex system with linear

I
I

*t and nonlinear subsystems

We consider a system according

subsystem with input
to Fig. I composed of a large lizear

x(t) = (Xl (t), x2 (t), Xn(0))T

x.. (1)

and output

I Iy(t) = (yl(t), Y2(t) ... Ym(t)) T (2)

lcontaining also several nonlinear elements. The dynamic behavior of the

linear part can be described in the time domain as

y(t) %[H(T) x~t-T) dT. (3)

I the m x n matrix H(t) being the impulse response matrix. i.e. the element
h ij(t) is the response of the output y1 (t) to an input of the type
x (t) = 8(t). For practical purposes the upper boundary in this integral
can be substituted by a finite "decay time", depending on the damping pre-
sent in the system.

In addition to the large linear subsystem, the system of Fig.1 con-

tains also nonlinear elements or subsystems, through which some of the
outputs, say YI+I, yI+2
1 . .... , ym are fed back to some of the inputs, for
example Xk+l , xk+2 ' -. Xn . Only the inputs x I , x 2 ..... xk are then
accessible and only these variables are inputs for the complex system (i.e.
the composed system). The input-output relations between the x 1, x .....
xk and the yl. Y2. ..., y1 in the complex system of Fig.1 are sought, that
is, the outputs yl(t), Y 2 (t) ..... yl(t) are to be obtained via numerical
simulations from given inputs x 1 (t), x 2 (t), ... . xk(t). In the present
paper we solve this problem using directly the transfer properties of the
linear subsystem.

Suppose for a moment that the nonlinearities are such that the Yl+,"
Yl+2 ..... Ym are simply functions of the values of the input variables
Xk+l, xk+ 2 ...- xn taken at the same time, as would be the case for
example for nonlinear springs with displacements and forces as inputs and
outputs. Since the x 1 . x2 , ....xk are given as time functions and the xk+ ,
Xk+2. ....xn as functions of some of the Y. Y2 .... Ym' we can write

x(t) = xCy(t),t]. (4)

This notation does not reflect the fact that x does not depend on yl, Y2
.... Y, but has the advantage of being very simple. With (4) in (3) we now
obtain the vector integral equation

y(t) = H(T) x y(t-T), t-T] dT. (5)

in which the time function y(t) is the unknown, while x(yt) is given. This
integral equation has to be solved numerically for y(t) in order to obtain
the system response.

If the nonlinear elements are such that the xk+1. k+2 . x n are
functions not only of the values of yI+I' Y 2 ...
. Ym' but for example
also of their derivatives, then (4) has to be substituted by
3
I
x(t) = xCy(t), ;(t). t]. (6)

I Similarly, higher order derivatives could also be included. The problem can
3 ables. always be recast into the form
however
abl
s.driver
(5). by introducing additional vari-

U right track t r
left trock

Fig.2: Simple model for the

3 dynamics of a car

The integral equation (5) has been used successfully for numerical
simulations in several cases by the author. A simple example is the simula-
tion of passenger cars, mainly in studies of riding comfort. In this aspect
of vehicle dynamics usually almost the whole system is to a large extent
linear, the only essential nonlinearities being the dampers and the pneuma-
tic tyres. The corresponding mechanical model of Fig.2 is obviously a dyna-
mic system of the type depicted" in Fig.l. Since this system is still rela-
tively complicated, we discuss instead the simpler system of Fig.3, which
3 contains most of the essential features of the original system. It is
formed by two masses and three linear springs and dampers, all of which are
supposed to be identical for greater simplicity. In addition, there is a
nonlinear element connecting the two point masses, which for simplicity we
assume to be a nonlinear spring with a piecewise constant stiffness, as
shown in Fig.4. The external input into the system is by means of the dis-
placement s(t) of the point A. The linear subsystem has therefore two input
variableb, namely v(t) and f(t) (the force at the nonlinear damper) and two
output variables zI(t), z2 t), i.e.

. (7)
I x = Cf. v)T

3 y= (Z, z2)T" (8)

I
I
 I
I

d c- 3
d -Z m

di I

A
v(*)
wt
1 3Z
Fig.3: Nonlinear system with Fig.4: The characteristics ofI
two degrees of freedom the nonlinear spring

The dynamics of the linear subsystem is obviously described by the equa-3

tions of motion

1 + c(2z1
mzi - z2 ) + d(2 1 - 2) = - f(t) + ofOv(i) dit + dv(t). (9

mz2+c(2z2
.0 - zI) + d(2;2 - I) = f(t). (10)
I
The solution can be written in the form (5). withI
F .(t) 4 3 e-e
3 Non osrt-!ls n n g -Te c oa
cos-2tsin ,

12IC 1 (11 2 1
etc., where the abbreviations 6 : d/2m: th lc/ine 2 -
I
1(t)= -(2,
The : 3w2- 962 were used. As an example the system response is calculated
0 - 62( 1
3
r t e a b e i t o s d/
=<h m0. > = .o4ir:
for etc
.,the function

ms1 + {(2z I Cos f2t), -0 t . 3

I
U
m Fig,5 shows some numerical results obtained in this manner for m = 1.
d = 0.04, c = 1. Q = 1. h 0 = 1. The integrations in (5) were performed with
a simple first order numerical scheme for different values of At. These
results are compared with a "numerically exact" solution of the differen-
tial equations, obtained via a fourth order Runge-Kutta method (with
3 At = 0.5).

* 2

-2 I 29
* ai
4

S2

-2 •8l 28
-4

1 4
I
I -2

-4 "
' I 18
l 20

Fig.5: Numerical results

In the paper, additional numerical results are given and it is also

shown how the idea of splitting up the complex system into linear and non-
linear subsystems can be used in the measurement of transfer properties.
U
U
I
U
 On the Adaptive Control I
of
Dynamic Systems with Flexible Structures

I
Robert L. Kosut 3
Integrated Systems,Inc., 2500 Mission College Blvd., Santa Clara, CA 95054
and
3
Information Systems Lab,Stanford University, Stanford,CA 94305 i

Abstract

The purpose of this talk is to present a practical method for analyzing the stability prop-
erties of adaptive control systems in general, and in particular, dynamic systems with
fle.xible structures. The basis for the stability analysis is the application of the classical
method of averaging for analyzing th,? behavior of ordinary differential equations with a
small parameter. The theoretical analysis fcveals comman properties of many adaptive
3
algorithms, including causes of instability and the means to counteract them*. The lim-
itations and practical use of the theory is discussed. The theory will be applied to both i
MRAC (M vodel Reference Adaptive Control) systems and to STR (Self Tuning Regulators).
Specific applications will include: iarge flexible space structures, robotic systems, and disc
drives.

I
]
]
[
*Theoretical background material may be found in Stability of Adaptive Systems: Pas- 3
sivity and Averaging Analysis (MIT Press, 19S6) by B.D.C. Anderson, R.R. Bitmead,
C.R. Johnson, Jr., P.V. Kokotovic, R.L. Kosut, I.M.Y. Mareels, L. Praly, and B.D. Riedle.

I
I
5 ON THE STABILIZATION OF TETHERED SATELLITE SYSTEMSt

E.H. Abed and D.-C. Liaw

3 Department of Electrical Engineering
and the Systems Research Center
University of Maryland, College Park, MD 20742 USA

Abstract
U The nonlinear dynamic model of the shuttle/tethered satellite system (TSS) stud-
ied by Liaw and Abed [1] and the associated results are reviewed and extended in this
talk. Recall [1]-'3] that the TSS consists of a shuttle and a satellite connected by a
5 tether, in orbit around the Earth. In this work, issues of stability and stabilization in
the station keeping and the deployment and retrieval processes are considered. Two
3 approaches are used in the investigation. The first is based on a Liapunov function for
the nonlinear model. The second method relies on Hopf bifurcation theory, as in [1].
3 Two other issues are briefly considered: the existence of an invariant manifold for the
dynamics, and controllability of the TSS via tension control alone. In particular, it is
observed that purely in-plane motion of the ISS corresponds to an invariant manifold
of the dynamics. Moreover, the full nonlinear system is shown to be uncontrollable if
I the only available control is the tension in the tether. This fact is interesting in the
light of the favorable stabilizability properties of the system.
The model of the TSS developed in [1] is a sixth order lumped parameter model,
with state variables o (out of plane angle), 0 (in plane angle), and C(tether length),
as well as their time derivatives -6, 'o and C. respectively. The model is obtained
5 using the svstem Lagrangian in [1] under the assumptions of a massless, rigid tether,
a satellite of mass in very small compared to the shuttle mas6 rn,, no aerodynamic
3 drag forces. and a circular orbit of the shuttle. The model is. in the notation of [1], as
follows:

t Supported in part by the Air Force Office of Scientific Research under URI Grant
AFOSR-S7-0073, and by the NSF under Grants ECS-86-575G1 and CDR-SK-00108.

I
I
 S

= 2v 1 (p)(+Q) 2 f 2 ro ro
T cos 0sin0(l r- (2)
3

fl 2 r~sin9( r3
= 2V(
Loe + Q) + 2 tan $(we + Q) (1- _) (4)
2 ~
=~~ + -2_ ~ ~~ + ~ o r3

2
+Q ro cos cos 0(1 - r -) + T
T(6)
In (1)-(6), r0 is the constant radius of the shuttle orbit, Q is the constant angular

velocity of the shuttle in its orbit, and rm denotes the varying radius of the satellite
orbit, given by
rn = r + 2 + 2roecoscos 3
and T denotes the tension in the tether.
It is shown in [1] that, at an equilibrium point, the system (1)-(G) nominally pos-
sesses two pairs of purely imaginary eigenva.ues. One of these pairs may be stabilized 3
by linear output feedback, while the other pair is uncontrollable. This motivates the
use in [1] of results on stabilization of Hopf bifurcations [4] to yield a family of stabi-
lizing tension control laws for station keeping. Liapunov functions for nonautonomous
linearized systems associated with (1)-(G) are used in [1] to prove stability of deploy-
ment and instability of retrieval for constant angle deployment and retrieval strategies
for which the system retains an equilibrium point.
The Liapunov function candidate for the system (1)-(6) is taken as the total system
energy, the sum of the kinetic and potential energies (KE and PE, respectively), whereI
2
KE = in,2,.2 + m[C2 +e 2 '2 + e2 cos 2 0(9 + Q)

+!p2r2 + 2Mroecos sinO - 2!QroesinosinO¢

+ 2QroCcos o cos0(6 + ()],

 PE=- G-Im, G-lm
ro rm
This Liapunov function candidate is useful in deriving stabilizing tension control laws
T in feedback form. This is most easily seen for the station keeping application studied
in [1], wherein the TSS is to be stabilized about a fixed tether length.
Work is proceeding on the question of the quality of the total system energy as
a Liapunov function for the system, in terms of the size of the predicted region of
attraction. In addition, generalization of the results summarized above to allow for
flexibility of the tether, nonzero tether mass, noncircular orbit, and appreciable satellite
mass is in progress.
The existence of an invariant manifold for Eqs. (1)-(6) is easy to check. Note
that if, in Eqs. (1), (2), at some instant of time 6 = 0, w = 0, then this holds for
all subsequent times. Therefore the set = 0, wo = 0 is an invariant manifold of
(1)-(6), regardless of the form of the tension control law T. The attractivity of this
invariant manifold can only be determined from the nonlinear terms, since the Jacobian
matrix of Eqs. (1), (2) with respect to 6.w has a pair of purely imaginary eigenvalues
(a critical case). Moreover, the existence of this invariant manifold, regardless of the
tension control law T, implies that the system (1)-(6) is uncontrollable.

References
(1] D.-C. Liaw and E.H. Abed. "Stability Analysis and Control of Tethered
Satellites," Proc. USAF/NASA Workshop on Model Determination
for Large Space Systems, Pasadena. CA, March 10S. in press.
[2] C.C. Rupp and J.H. Laue, "'Shuttle/Tethered Satellite System,*" The Journal
of the Astronautical Scitnces. Vol. 26. No. 1. pp. 1-17. 107$.
[3] D.A. Arnold, "The Behavior of Long Tethers in Space," The Journal of the
Astronautical Sciences, Vol. 35. No. 1. pp. 3-1S, 1087.
[4] E.H. Abed and J.-H. Fu, -'Local Feedback Stabilization and Bifurcation
Control, I. Hopf Bifurcation." Systems and Control Letters. Vol. 7.
pp. 11-17, 1986.
 U
I
A MATHEMATICAL FORMULATION AND CONTROL EXPRESSIONS FOR
COMPLIANT CONTROL OF CONSTRAINED ROBOT MANIPULATORS
I
Choong Sup Yoon and Fathi. M. A. Salam 3
Department of Electrical Engineering and Systems Science 3
Michigan State University
East Lansing, MI 488241

ABSTRACT I
We formulate the compliant control problem mathematically employing the frame- U
work of constrained Hamiltonian systems. We then derive nonlinear control expressions for
the force and the motion on the constraint surface. The derivations reveal conditions that
define the class of constraint surfaces allowable in the formulations. Two examples are
then given to illustrate the formulatin and the methodology.
Introduction

Most robots are currently used for very limited tasks usually characterized by
position-to-position movements, e.g., pick-and-place, spot welding and spray painting.
Other essential but complicated tasks involve contact with the manipulator's environment,
e.g., inserting a pin into a hole, assembling, plasma welding, contour following, deburring,
I
grinding, etc., see [1,2). Such contact usually results in the generation of external forces
acting on the end effector of the manipulator. External contact forces such as the ones
introduced by constraint surfaces always modify the dynamical behavior of a manipulator.
3
Consequently, issues of appropriate modeling and of effective new control strategies arise.
Compliant control is concerned with the control of a robot manipulator in contact with
its environment, see [3-4]. The end effector of the manipulator first converges to the con-
straint surface at a specified position generating a specified force upon contact. Then, the
end effector moves along a desired path on the surface while maintaining a desired contact
force profile (along this path). Thus, compliant motion calls for the input torque to achieve
tracking for a specified path on the constraint surface, and with a specified contact force.
In principle, such tracking is possible because the constraint surface limits movement
to a submanifold (on surface) and consequently frees some components of the input torque
to control the contact force with the surface. However, the nonlinearity of the governing
dynamics as well as the constraint equations potentially make the control process difficult
if not impossible. The difficulty may translate mathematically to the presence of singulari-
ties at some points on the constraint surface or to the lack of well-posedness of the govern-
ing system of equations.
We choose to formulation the problem in joint space. An advantage of this choice is
that the constraint now applies to the joint angles directly; consequently the constraint
applies to the links of the manipulator and not merely to its end effector as it is the case in
 -2-

the task space formulation. Another advantage is that once the class of allowable con-
straints is specified in the joint space, the simpler and direct use of the forward kinematic
would provide the corresponding class of the allowable constraints surfaces in the joint
space. We remark, however, that determining useful, in terms of applications, class of sur-
faces is a nontrivial research problem.
The control process we envision may take the following steps. The end effector is first
steered to a point on the constraint surface using, e.g., the linear feedback control strategies
reported in [5-7]. In addition, one must also guarantee that at the final (desired) position on
the surface, a specified (normal) force is generated. Once the end effector is located at a
specific position and with a specified force, one may then apply compliant control stra-
tegies to generate or to track a desired path with a desired contact force profile. Some
results on compliant control have been reported in [3-4].
In this work, we propose a control strategy which consists of the sum of two non-
linear controls. One control restricts (the end effector of) the manipulator to the constraint
surface; this control represents the force control part. The other control stetrs (the end
effector of) the manipulator along a specified path on the constraint surface; this control
represents the position control part. Then we show that these nonlinear controls can be sup-
plied by the input torque vector at the joints. Specifically, we give an expression for the
(physical) torque which would generate the desired nonlinear controls. (It is possible to
include the dynamics of the actuators and consider the actuator voltages as the physical
inputs) Further work need to exploit force and velocity feedback to achieve attractivity of
the constraint surface in order for the formulation and the control to be robust.
We employ the geometric tools of symplectic Hamiltonian systems in setting up our
framework. Although these tools have been used in [3], our emphasis is quite different:
we assume that the amplitude (modulo a multiplicative constant) of the desired force is
given as a function defined on the constraint surface; then we derive the control required to
maintain that desired force. We also derive the second component of the control strategy
which generates desired paths or trajectories on the constraint surface. The derivations
require that the constraint surfaces satisfy conditions in terms of a matrix of Poisson brack-
ets being nonsingular. These conditions in fact specify the class of constraint surfaces
allowable in our formulation.
The formulation does not yet take advantage of feedback of error signals, with respect
to a desired position, velocity or force. Moreover, while our analytical results are valuable
on their own merits, we recognize that in application one has to take into account the
effect of disturbances, unmodeled dynamics, -,the dynamics of the material of the con-
straint surface itself. We hope to pursue these + ues in future works to blend our theoreti-
cal derivations with practical applications. . , ould be recognized however that the
theoretical framework provides guidance and deep insights into how to properly devise and
apply the control strategies.
Summary of Results

The Hamiltonian of the overall constrained robot system (HT) can be shown to be
I ~HT =H+F,(Xj +i2)4 +
2m 2n-2m j,()

where H is the free Hamiltonian, Xj is a Lagrangian multiplier, Oi represents one of the set
of equations that model constraint, and 'J represents one of the corresponding set of
orthogonal complement to the set of constraint equations. firepresents the 2m-dimensional
force control input vector and ii represnts the (2n-2m)-dimensional compliant motion con-
trol input vector. These controls can be derived to equal ([8])
a = C H.b (2.i)
 -3- 3
co = [( i)] = [tI, i)

1i= SiTT SHIP (2.ii)

sTvp = wpi) ] =[{I',ilq
( H* - H T, p2--2,,} ]T
SW.= [{H* - HT, IF' ) .......
I
The dynamic equations of the Hamiltonian HT can be expressed as
D)H 2m V
a.o 2,,-2,, oaFi
- = + • + ai)-- +
apt--Iap, ,=I
Yls- p i=1 (3.i) U
_H 2m aV 2n--2mn api
+) - IU ...... (i)
'aq'--I aq-
The vector form of (3) can be written as
q= -aaH+ AI(a + X.) + Bla (4.i) I
P1= - - A 2 (a + X) - B 2 4 (4.ii)
aq
where I

ap I3pIaq q
A= A2 (5.i)

0 DI a~b~ a~* a(D

apn ap" D, an
2
a~p D n-2 m DAPl _______

B,= b2 = (5.ii)

apI' api'-l aq" azI-f

The required torque input at the joints of the manipulators is given by

= (-A 2 + AM(q)A 1 + M(q),)2 + M(q)AIa
+ (-B 2 + Mi(q)A1 + M(q)Bl)i + M(q)Bli (6)

REFERENCES

[1] M.Brady et.al.(editor), "Robot Motion: Planning and Control," MIT Press, 1982.
U
1 -4-

[2] R.P.C.Paul, "Robot Manipulators: Mathematics, Programming, and Control," MIT

Press, 1981.
[3] N. Harris McClanmroch and Anthony M.Bloch, "Control of Constrained Hamiltonian
Systems and Applications to Control of Constrained Robots," Dynamical Systems
Approaches to Nonlinear Problems in Systems and Circuits edited by FMA Salam
and ML. Levi, SIAM, 1988, pp. 394-403.
[4] M.H.Raibert and J.J. Craig, "Hybrid Position/Force Control of Manipulators," ASME
J. of Dynamic, Systems, Measurement and Control, 102, 1981, 126-133.
[5] M.Takegaki and S.Arimoto, "A New Feedback Method for Dynamic Control of Mani-
pulators," J. of Dynamic Systems, Measurement, and Control Vol.102 June 1981, pp.
119-125.
[6] M.Takegaki and S.Arimoto, "Stability and Robustness of PID Feedback Control for
Robot Manipulators of Sensory Capability," Robotics Research, M.Brady and R.Paul
(editors), MIT Press, T984, 783-799.
[7] F.M.A.Salam, "Velocity feedback Control and Dynamic Properties of Robot Manipu-
lators, " The 26th IEEE Conference on Decision and Control, Los Angeles, December
9-11, 1987.
[8] C. S. Yoon and F. M. A. Salam, "CompliantControl of ConstrainedRobot Manipula-
tors: A Mathematical Formulation and Control Expressions," 27th IEEE Conference
3 on Decision and Control, 1988, submitted.

U
I
I
I

I
I
I
I
 FLEXIBLE ROBOT MODELS WITH REVOLUTE AND PRISMATIC JOINTS-
- HANDLING OF CLOSED LOOPS

Michael Rlemer and .org Wauer

Institut fUr Technische Mechanlik
Universltat Karlsruhe
KalserstraBe 12, D-7500 Karlsruhe 1, FRG

Svnopsis. Supplementing recent investigations generating the equations of

motion for flexible industrial robot models with revolute and prismatic
Joints here the treatment of closed loops Is presented.

1. Introduction
In the dynamics of muitibody systems with deformable components,
structures with material boundary conditions, e.g., revolute Joints, and
tree structure, e.g., open chains, have been considered. Until quite
recently, only a few contributions have taken Into consideration the
effects of non-material constraints, e.g., by means of prismatic Joints
/1/ and the handling of closed loops /21, respectively. But these effects
have been considered neither completely nor simultaneously. The first
detailed Investigations deriving the equations of motion for such
non-material systems, but without closed loops, have been carried out In
the recent past /3,4/. As a supplement, here the extension to closed
loop systems will be discussed, namely, In the first part with an example
of a simple one-body distributed parameter system and in the second part
for a planar two-body system with beam-shaped structural members as a
typical sub-class of general flexible non-material multibody systems
involving closed loops. The results of /3,4/ and this short communication
will be combined In a forthcoming paper /5/.

2. Axially moving string as a simple model

The simplest prototype for a non-material flexible system involving
closed loops Is represented In Figure 1. The string pre-stressed by an
axial constant force H., performs an overall motion s(t) In the axial
direction and superimposed small vibrations described by the vector field
of displacements u. The motion Is constrained by the material, moving
support A and the non-material, locally fixed support C. For the
co-ordinate system we choose the Inertial frame e (k-1,2) and the
locally attached tangential unit base vectors ek (k-1,2). They define the
non-deformed reference configuration and the Lagranglan co-ordinate (LC)
tl of the
string.

el=

_ _ _ _ _ Figure I

m~m~nmu
-
I
I Here the transformation

between the reference frames Is very simple, but an essential property

holds, in generalt Locally attached tangential base vectors (here ek) are
independent of the deformations of the body they belong to. Applying
KIrchhoff-Hamilton's principle (KHP) /6/
t2 t2
If (T-V)dt + f Wvir dt - 0 (2)
ti ti
m the kinetic and potential energies T and V as well as the virtual work
W,, have to be established In advance. For a string their non-linear
formulation Is given in /7, eq.(9)/. In order to handle the problem in
the sense of multibody systems, first of all we consider the underlying
tree structure, which Is obtained by removing support C. Then (2) yields
by means of a classical mathematical procedure, the non-linear field
3 equations /6/
-Iau.,,t +N, 1 - 0, N-EA(ut,+. 1u 1 +.u2 ' 1), EA-flexurai stiffness,
5 - utt +0 ,i- 0, Q-EAu,. 2,
1 , IL-mass per length (3)
for the one-fleld problem with geometrical boundary conditions at t1-0
(support A) and dynamical ones at the free end t -1. If s(t) is not
prescribed, the equation for the overall motion is
IS-rF(t)-j 1 (s+u,,)dt 1-0 f (3a)

with~the given force F(t). Now the above discussed tree structure (Fig. 1
without support C) is closed by adding the non-material support C at the
position
3 r'r e,, (rk) " [b] (4)

In the Inertial frame. Hence, the corresponding constraint equation for

the stiing Is
Ck
C
xr,(ti_4 c0t)A
x ) ,t) -- r 0O ork-1,2
(5)
- -~(t1-tc(t),t)
wherein x-(Es(t)+k1+u1]a.+ua*2
1 2 ))eA1; 6
(6)
is the position vector of an arbitrary material point 0 in the deformed
state u-ukek measured In the inertial frame e^. Therewith (5) reads

3 k-i: s(t)+-C(t)+u(tc,t)-b, k-2: u2 (Cft)-0. (7)

I
I
Undoubtedly M 2 is obvious looking at Figure 1, but only a formalism
like (4), (5) and (6) can be used to establish algorithms for general
multibody systems. (7), determines the LC 4C(t) of the actual material
point (MP) at support C. The additive supplement to KHP (3) then is

Bx - zJt2(t) Ck dt, (8)

ti
with a minimal set of Lagrangian multipliers (LM's) kt depending on time
t only. In C , (5), It Is essential that only the actual MP with the LC
kl- C(t) Is considered. This requires the variational operator 6z in (8)
Instead of the classical one, St. The relationship between the locally
measured, non-material variation (z-const. In Fig. 1) 6z at support C and
the material one (tl-const.) 6t Is given by the chain rule /7/
5z 1] = RI. MC + St 1] (9)

Indicating the change of MP's at C by the convective part [], 2 z. (8)

can be evaluated In the form

5X - fs(t)6zCk dt - t2 ( +Bull x-u,2''dt (10)

ti ti

wherein all variations are Independent and also (5). Is taken Into
account. Because of support C, the Integration In (2) over the Interval
[0,1] must be split Into the two non-material Intervals [0, c(t)) and
(QC(t),], which lead also to variations sz defined in (9). Following the
arguments In /6/, together with (10), the new equations of motion for the
closed-iop case can be derived: The two field equations (3) remain valid
for the two fields left and right of C, respectively. The four material
boundary conditions remain unchanged, too. The "open-chain" equation
(3a) for the overall motion s(t) suffers a correction due to
r(1+u, )21 tc +0
S+x.+ Eu];-u, [U]* - -A (1)

Additionally the four non-materlal boundary conditions at C occur in the

form
X+ [U+,].-0, x- + [W+013-0,
(uY-,u 2 -0 ~-- (j+7')2t] 0C+ (12)

together with three equations for the constraints

ck- - [(W+Q)u 2 .+ (U+N)u. ]+-O (13)
 IA
I accordance with the number of additional unknowns
beeing In
and It(t). In order to discuss the method of "body-doubling" /8/,
(t) (k-1,2)
the
system in Figure 1 must be divided Into an open chain (system I: string
U1 plus support A only) and the doubled body (system II: string with support
C only) having tree structure, too. Each of the bodies have half the
original mass. To fit the two subsystems I,II it Is necessary to match
the node co-ordinates by means of their position vectors. Using the
reference frames In Figure 1 for both I and II, we obtain
AA

Ak - Iuk(l-ot)-O, C1 . s(t)+ c+ u (C, t)-b-O, C2 -1u 2 ( ,t)-O, (14)

wherein Ck-o Is identically the constraint equation (5), here for body I.
Matching the MP's in the fields results in
fk. Tuk_ IUk. 0 , k-1,2. (15)

In order to generate the supplements for the existing equations of motion

(3),(3a), six LM's have to be introduced due to
S2 " X _l
A
[X (t) Ck + AX k(t) Ak+ I tkl1t)
(
fk dt' dt. (6

ti 0

Compared with (10), the method of "body-doubling" need more LM's and,
hence, four additional equations. But the most important difference Is
the dependence of two of the position co-ordinate t', leading to
considerable complications during the discretization procedure, which is
usually used to obtain approximate solutions.

1 3. Two-body robot model (see /5/)

The considerations carried out in the last chapter will now be extended
to a more practical system. It consists of two elastic bars I and J (see
Figure 2). They perform a general overall motion x(t) and T(t) together
with a rotational and translational rigid body motion a(t) and s(t) of
the second component, J, relative to the other one, I. The relative
motion is realized by the non-material revolute-prismatic joint G. Small
plane elastic vibrations denoted by the vector fielas of displacements
lu,Ju and the scalar fields of bending angles k@J. are superimposed. In
Timoshenko's beam theory u Is independent of I due to the shear
deformation. The transformation between the reference frames can be
performed analogously to (1) as well as the generation of the equations
of motion for the underlying tree structure (after removing joint C).
Closing the open chain by prescribing the path of C in the Inertial frame
along a smooth curve r result in constraint equations (similar to
(4),(5)) with a minimal set of LM's. The associated supplement (similar
to (8)) together with the operators of the tree structure lead to five
Integro-differential equations for the overall motion, nine partial
differential equations for the deformations, three non-material boundary
conditions at G, and three additional constraint equations. All equations
are coupled to each other and contain non-classical terms. For details

I see reference /5/.

I__ _____
 I
I

reerenc courlpration

]oil.'oof bar J
/ UU
4. // /Cnuo I
ferefrce conigatin

//rewi"e c.,,flg,,ritl.,, I

EtsoFigure 2: Two-body system

(I~~Isse)w ith revolIute-pr istoat ic•
joint G

4. ConclusionsI

Extensions of flexible muitibody systems with tree structure to such

cases involving closed loops denote an important topic in the dynamics of
robotics, for example. If the connecting joints belong to the group of
so-called non-material constraints, e.g., a prismatic joint, not only the
analytical description of the kinematics and the non-classical handling
of variational and differential operators, but also the treatment of
closed loops lead to considerable complications. The formalism presented
I
here results in a minimal set of equations of motion, which can be
reduced by direct variational approaches to ordinary differential
equations. The method remains fundamentally unchanged when applied to
I
spatial motions or three-dimensional components.

References
/i/ LILOV, L., WITTENBURG, J., Dynamics of Chains of Rigid Bodies and
I
Elastic Rods with Revolute and Prismatic Joints, In Biancl/Schiehlen
(eds.), Dynamics of Multibody Systems, Springer, 1986, p.141-152.

I
m mm"" I
/2/ LILOV, L. K., Dynamics of Elastic Muitibody Systems Involving Closed
Loops, In Bianci/Schlehlen (eds.)s Dynamics of Multibody Systems,
Springer, 1986, p.129-140.

/3/ RIElIER, M., WAUER, 3., Zur Behandlung von Schubgelenken In

Mehrkorpersystemen mit verformbaren Teiistrukturen, ZAMI (1988), to
appear.
/4/ RIEMER M., WAUER, J., Equations of Motion for Hybrid Industrial
Robot Models with Revolute and Prismatic Joints, ICIAM-Conference
Paris, 1987, unpublished.
/5/ RIEMER, M., WAUER, 3., Equations of Motion for Flexible Multibody
Systems with Revolute and Prismatic Joints, CISM-IFTolIM-Conference,
Udine, 1988, to appear.

/6/ RIEMER, M., Zur Theorie eiastischer Kontinua mit nichtmaterieiien

Ubergangsbedingungen, Dr.-Ing. Thesis, Univ. Karlsruhe, 1985.
/7/ RIEMER, M., Anwendung des Prinzips von Hamilton auf die mit
konstanter Geschwindigkeit bewegte Saite, ZAMM (1986), p.T78-T80.

/8/ WOLZ, U., Dynamik von Mehrkorpersystemen - Theorie und symbolische

Programmierung, Dr.-Ing. Thesis, Univ. Karlsruhe, Fortschr.-Ber.
VDI-Z., R. 11, Nr. 75, 1985.
 I
STABILITY OF FLEXIBLE MULTIBODY SYSTEMS
DYNAMICS IN THE PRESENCE OF CLOSED LOOPS

Amirouche F.M.L 1
Assistant Professor
Opt of Mechanical Engineering
University of Illinois at Chicago 1
Chicago Ill 60680
U
ABSTRACT

A common problem in the analysis of rigid and flexible constrained

multibody dynamics arises when the Jacobian matrix becomes singular.This I

in fact result in some numerical instability in the integration of the

governing equations of motion.

Usually when a multibody system is subjected to simple nonholomic and

holonomic constraints of the form BY-G(where B denotes the Jacobian matrix

,Y the generalized speeds and G is a function of time t),it is customary

to assume that B has full rank and its orthogonal complement array C in

the constrained space exists.However in certain configuration the loop can

go from 3D to 2D type of formation causing the Jacobian matrix to be

singular.A common approach to insure stability of the system is to avoid 1

the singularity position.This require more efforts than needed and doesn't

guaranty the continuity of the system motion in the neighborhood of I

singular positions. 3
An a3gorithm based on the Pseudo-Uptriangular-Oecomposition -Method(PUTO)

introdu J by Amirouche et al is used to detect when a particular

constr- , equation vanishes in the process of the uptriangulazation of

the Jac,- matrix.This identification is essential as it permits one to

compute the orthogonal complement array even when B changes rank.Two

I
p
i methods are then introduced to insure the numerical stability of complex

dynamical systems with variable closed-loops.

I
a
I
I
I
I
t
II
I,
I
I
I
I
I
I
I
I
I
I
I
!
I
I
SESSION 5
I
IEXPERIMENTS IN NONLINEAR DYNAMICS

I THURSDAY - 0830 - 1000

1 June 2, 1988
I
I

I
I
a
I
 FRACTALS AND CHAOS IN ELASTIC SYSTEMS

Lecture by

Francis C. Moon
Professor and Director
Sibley School of Mechanical
and Aerospace Engineering
Cornell University

ABSTRACT

In recent years new phenomena have been discovered in

nonlinear dynamics, namely chaotic oscillations in deterministic
systems. In this lecture physical examples will be discussed from
mechanical, control, space structures and fluid-elastic systems. To
describe this chaotic phenomena, new mathematical ideas have
entered dynamics such as fractals. In this lecture I will illustrate
how fractal concepts and Poincare maps are used to describe strange
attractors and other properties of chaotic systems such as basin
boundaries. in particular, we show how to calculate the fractal
dimension of a chaotic attractor from experimental data for nonlinear
vibrations of an elastic beam and vibrations of fluid flow through a
flexible tube. A discussion of new experimental techniques in
nonlinear dynamics will be given. A demonstration of chaotic
vibration will be performed.
 Ngnlinear Oscillation of a Flexible
Cantilever: Experimental Results
Alan G. Haddow & Syed M. Hasan
Dept. of Mechanical Engineering
Michigan State University
East Lansing, Michigan

1. Introduction.

A collection of experimental results which describe the nonlinear

oscillatory behavior of a flexible cantilever beam are presented. The
approximate dimensions of the beam are 30" x 0.5" x 0.02" and it is forced
through a sinusoidal base excitation in the axial direction. This is the
"input" term. The response, or the "output" term, is measured by a strain-
gage attached ntar the root of the cantilever. The following sections
describe a number of the phenomena observed. The periodic or almost periodic
responses occur for planar motion, whereas if non-planar motions arise,
various forms of chaos result.

2. Linear Stability and Nontrivial Steady-States.

It is well documented that the dynamic stability of a beam excited by a

sinusoidal displacement in the axial direction is given by a Strutt diagram.
In the displacement-frequency parameter space there are wedge shaped
boundaries which seperate the stable and the unstable regions. The nose of
these regions occur at forcing frequencies that are equal to twice the linear
natural frequencies of the system. They also occur at multiples and
combinations of these natural frequencies. Figure 1 presents experimental
data which define one of these regions. It is associated with the 4th natural
frequency of the system. The excitation frequency is - 92 Hz and for a set of
parameters inside the unstable region, the model responds in the 4th mode.
Linear theory predicts that this unstable response would grow exponentially.
However, as the response grows the effects of non-linearities can not be
ignored. This results in the beam attaining a steady-state. Keeping the
magnitude of the base acceleration constant, it is possible to investigate the
variation of the amplitude of this steady-state response as a function of the
forcing frequency. Figure 2 presents such results. Note the large multi-
valued region in which both trivial and nontrivial steady-state solutions are
found.

3. Transient Chaos.

If the frequency is such that the response of the system is quite large,
e.g. point A on Figure 2, then it is possible to observe transient chaos if
the system is perturbed from its steady-state. For this value of force the
response never remains chaotic, but always returns to either the trivial or
the non-trivial, single mode, steady-state value. The basic form of the
response while in this chaotic regime is similar to that given in Figure 4.
 5 -2-

4. Steady-State and Intermittent Chaos.

Increasing the level of the force slightly (see Figure 1) results in the
frequency response curve presented in Figure 3. If the frequency is decreased
enough an out-of-plane mode is excited and a chaotic response results. The
term steady-state chaos refers to the fact that the response remains chaotic
for all time. A time trace of such a response is presented in Figure 4 along
with its FFT. By altering only the initial conditions the stable, trivial
solution can of course be attained. In addition it is possible to find multi-
mode responses. Figure 5 shows such a case.

5. Extremely Low Subharmonic Response.

A time trace of the input and the output are presented in Figure 6 for a
forcing frequency of - 190 Hz. (Note the change in the time scales.)
Initially the beam responds at the forcing frequency but as time progresses,
the energy seems to cascade down through the modes. The authors believe this
is a consequence of internal resonances. Eventually a very low frequency,
steady-state response is attained.

6. Concluding Remarks
A selection of some experimental observations of the nonlinear behaviour
of a cantilever beam have been presented. Work is in progress exploring a
number of the phenomena in more detail, particularly the extremely low
subharmonic response. This type of energy transfer to remote modes has
received little attention in the past.

I 1400.0- o(PERMAMAL DTA

1200.0- 0

S1000.0,
Lsee Fig. 3
I 800.0-
3 STABLE UNSTABLE STABLE

600.0-

400.0-

200.0 - 0
000... see Fig. 2

80.0
9 90 91 92 93 94

3 FORCING FREQUENCY (Hz)

FIGURE 1. Stability diagram for the fourth mode.

I
 -3-

1000.0- . ZON-y , .
, ". "

500.0- 7C S.f

600.0-
U, S,

' 400.0-
U-U

200.0-
fi--_ ___N'__
:

-200.0 -
850 86.0 87.0 88.0 89.0 90.0 91.0 92.0 93.0 94.0 95.0
FORCING FREQUENCY (H)
FIGURE 2. Steady-state parametric response. Level
of force constant (see Fig. 1).

ZONE ~ S-- .

u 800.0 ,.

600.0 "N

'" 400.0-
I- 4,.,

c 200.0-

0.0-

-200.0 , ' , 94. 95.0

85.0 86.0 87.0 88.0 89.0 90.0 91.0 92.0 93.0
FORCING FREQUENCY (Hz)
FIGURE 3. Steady-state parametric response. Level
of force constant (see Fig. 1).
I
I
I

()(a)
A

I
I .LUjlI - a.in

I D(b) ,

LSI I

LIe LI

3 FIGURE 4. haotic response for ,irorcing rreuencV FIGURE 5 Combination of resonses a¢ a forcin frequmncy

r 85.5 Hz. (a) Time trace of the output, of 85.5 Hz. (a) Tim. trace of the output.
.(b) FFT of (a).

I
I
I
I
I
 -5- £!
I

I
|I

CD0 10
06.

o.-

.0 !
~1
iIi
 NONLINEAR RESONANCES AND CHAOTIC MOTION IN A FLEXIBLE PARAMETRICALLY
EXCITED BEAM

T.D. Burton and M. Kolowith

Department of Mechanical and Materials Engineering
Washington State University
Pullman, WA 99164-2920 USA

ABSTRACT

We present experimental and analytical results for a flexible, parametrically excited, vertically mounted beam.
The exciting frequency Q is near twice the fourth mode natural frequency W 4 . We have observed steady, periodic
motions associated with the fourth mode principal parametric resonance and chaotic motions, both occurring in
a narrow band near Q - 2W4 .

INTRODUCTION

A schematic of the experimental setup is shown in Figure 1. The test specimen, mounted as a vertical
cantilever, is SAE 1094 steel of dimensions 35.625 in x 1 in x 1/32 in. The beam is very flexible in the "in-plane"
direction and very stiff in the "out-of-plane" direction. The in-plane flexural stiffness El = 76.29 lb-in 2 and the
mass per unit length p = 0.2388 x 10 - lb sec 2 in 2
_4
. The vibrational motion was monitored with a strain gage,
which was mounted at X/L 0.2, near the first peak of the fourth mode shape. The harmonic base motion was
monitored with an accelerometer. The five lowest natural frequencies of the beam, determined experimentally,
are approximately 0.5 hz, 4.8 hz, 14.0 hz, 27.8 hz, and 46.0 hz, respectively.

- be am

lateral vibration(v(x,t))

I shaker motion(U(t) =

shaker
bcosivt)

// / / //

Figure 1. Schematic of experimental setup.

The objectives of the experiment were 1) to study the steady state, periodic motions of the nonlinear
resonant response associated with the principal parametric resonance of the fourth mode (Q ; 2W 4 ), 2) to
study the chaotic motions which were observed to occur in this same frequency range, and 3) to study the
connection between the two types of motion, particularly the transitions form periodic to chaotic motion and vice
versa.
Moon [11 has studied experimentally the chaotic motions of a harmonically driven, magnetically buckled
beam. We note three basic differences in our experimental conditions: 1) the beam is driven parametrically rather
than directly, 2) the vertical configuration is statically stable; thus, multiple static equilibria are not present, and
3) the nonlinearities are a result of the geometry cf large amplitude motion, rather than being due to an external
 2

agent. Presented in the following sections are analytical and experimental results for the steady, periodic forth
mode motion and experimental results which characterize the chaotic motion.

ANALYSIS AND EXPERIMENTAL RESULTS: PERIODIC MOTIONS

If the base excitation parameter bQ 2 is relatively small and if Q 2U4, the system exhibits planar,
unimodal, periodic motions characteristic of a single degree of freedom, parametrically excited oscillator having
a softening nonlinearity. The analysis of these motions is based on the following equation of undamped, large
amplitude, planar motion, which was derived by Krishnamurthy [21, based on the previous analysis of Crespo da
Silva and Glynn [3]:

/' [ I I- ~I'(v+ 1 [v' f S '2 dS 2 dS, = (uoQ2 cos QoT - go)[(1 - S)v" - v'].
4(1)

Here dots are derivatives with respect to the dimensionless time T = (EI/pL) 2 t, = I'/L, and 110 = b/L.

In equation (1) the first of the nonlinear terms is a static, hardening nonlinarity arising from the potential
energy stored in bending, while the second nonlinear term is srftening and is a result of the kinetic energy of axial
motion. As noted by Haight and King [4], the latter, inertial nonlinearity is the dominant nonlinear effect.

If one assumes the motion to be dominated by a single mode, v(s, T) q$(S)=(Twhere

F O(S) is the
linear mode four free vibration mode shape, equation (1) may be converted to following modal equation fur the
fourth mode:

+ y[l + 2 q cos 207] + aiy(y2 '+ a2y 3 0. (2)

Here Q = Q/2W4 , dots are derivatives with respect to a new time r = 6 4T, 5 = 10.99554... is the fourth
mode eigenvalue, y = bz, q = .9512Uo6202, al = 14.05 and C2 = 1.09. The maximum value of the
dimensionless displacement y is of the order of 0.10.

The measured mode four frequency response for the case q 0.014 is shown in Figure 2. In the range
55.12hz < Q < 55.88hz, the null solution v(s, T) = 0 is unstable. Stable and unstable nonzero solutions
arise via pitchfork bifurcations at Q 2 = 55.88hz and Q = 55.12hz, respectively. These solutions coalesce
in a saddle node bifurcation at Q3 54.12hz. Thus, one observes the jump phenomenon as Q is decreased
through Q3 and as Q is increased through Q 1.
 ,
.rr, rr= ir,- r [ c .o=_pr - rrr. .-
S:, . : - rt; "C'-t:- -.
.:=:,v . ... <.C :. :. ::: ° - - :V. - . . . - .

075- experimental
theory, equation (4)

base acceleration
600 in/sec

LU

' 0.25.

54 5'5 UNSTABLE-1j 56

54.12 55.16 55.88

EXCITATION FREQUENCY- HZ

Figure2. Frequency response, steady mode 4 motion.

In relating the experimental and analytical results, it is necessary to add some modal damping in equation
(2). A nonlinear damping model was employed, so that the modal equation actually analyzed was

2 (y
j+ 2J(1+I ) +[1 + 2qcos 2(27] + )"+ 2 _0 (3)

The linear damping factor ( = 0.002 was measured form free vibration tests at small amplitude, while /I = 3500
was determined so as to match the observed jump frequency Q3 . (free vibration tests at large amplitude revealed
an effective damping factor 2' 0.005 at an amplitude of 0.75 cm, and this corresponds to U 2 2000).
A harmonic balance (or first order multi-scale) analysis of equation (3), y(Tr) = a cos(OT- + 0) yields the
following relation for the nonzero mode four steady state vibration amplitude a:

- 2 2 (10 +I 3+ (1 ) 0)
2 {q

The comparison of experimental results and those given by equation (4) is shown in Figure 2. While both linear
and nonlinear damping models provide good agreement for Q > 54.Ahz, some nonlinear damping is clearly
necessary if the bifurcation frequency Q 3 is to be determined analytically.
It should be noted that numerical integration of equation (3) yields results which are virtually indistinguish-
able from those obtained from equation (4). This is because the steady state solutions to (3) are dominated by
the fundamental harmonic (this is also true of the experimental results, as shown in Figure 3). This appears to
occur because, for the case Q = 1, a2 = = 0,the exact periodic solution to equation (3) is simple harmonic,
an unusual occurrence in nonlinear oscillators. We conclude that the steady state, periodic mode four motion is
well predicted by the theory. Note, however, that the particular nonlinear damping model used is not necessarily
physically motivated.
 ,1

.... 2t.6 hz

a) top: strain gage vs. time b) FFT of strain gage signal

bottom: base acceleration (log scale)
vs. time

Figure 3 Steady mode 4 response.

EXPERIMENTAL RESULTS: CHAOTIC MOTION

Chaotic motions were observed in certain regions of the parameter (b, Q) space near Q = 2 ,.. The
strain gagc output and frequency spectrum for a typical chaotic motion are shown in Figures 4(a) and 4(b) For
reference, the frequency spectrum for a typical free oscillation, generating by striking the beam, is shown in 4(c).
The chaotic response exhibits the characteristic irregularity in the time series and broadening in the spectrum.
It is apparent that the lowest six or seven in-plane vibration modes are involved in the chaotic motion. Direct
observation of the chaotic motion also indicates the lowest torsional mode to be present, accompanied by an out
of plane displacement due to combined torsion and in-plane curvature.

b) FFT of strain
gage output,
chaotic motion.

0 4 8 12
-.. c) FFT of strain
gage output,
TIME (=eeond,) free oscillation.
a) strain gage output
vs. time.

Figure4 Chaotic response.

Shown in Figure 5 are regions of the parameter plane (b, Q) in which chaotic motions were observed. A
stable chaotic motion was defined as one persisting for three minutes or longer (approximately 10,000 cycles
of excitation). As Figure 5 shows, the minimum base excitation required to sustain chaotic motion occurs at
(Q/2W4 ) ; .985, with a fairly sharp increase in required excitation level on either side of this frequency. In
the "transition" regions the response was observed to cycle back and forth between chaotic and mode 4 motions,
with more time spent in chaos as the base excitation was.increased at fixed driving frequency. This may indicate
a transition to chaotic motion via the intermittency route, in which, as the excitation level q is increased beyond
 5V

'/
a critical value qc, the fraction of time spent in chaos is proportional to (q - q ) . We were unable to
verify this conjecture quantitatively, however, as our exciter exhibited sufficient drift in excitation level over the
long measurement times required (1-3 hours) that reliable determination of the fraction of time spent in chaos
was not possible.

2.5

2.0/ 2.0 periodic response for

all initial conditions

"intermittent" chaotic
. L response

.0 i
17ZZZ71 chaotic response

0.5

.0
52 53 54 5 56 57

EXCITATION FREQUENCY -HZ

Figure 5. Regions of parameter plane for which chaotic motion occurs.

Numerical solutions of the singlemode model, equation (3), did not exhibit chaotic behavior for any parameter
values comparable to those of the experiments. Thus, an analysis of the loss of stability of the node 4 motion
as a parameter is varied will require a multi-mode expansion of equation (1) and possibly the inclusion of torsion
and out of plane motions as well. This is not surprising in view of the obvious participation of multiplemodes in
the chaotic response.

CONCLUDING REMARKS

The experimental and analytical results show the steady node 4 motions to be well described by a single
oiode model in which the kinetic energy of axial motion provides the dominant nonlinear effect. The chaotic
notions observed at higher excitation levels will require a multi-mode model for their prediction. Furthermore,
more precise measurements need to be made in order to characterize experimentally the transitions into/out of
chaotic motion.

REFERENCES

1. Moon, F.C., Experiments on chaotic motiol of a forced nonlinear oscillator: strange attractors, J. Appl. Mech,
47, 639-644 (1980).
2. Krishnamurthy, K., "Dynamics and control of flexible robotic manipulators," PhD thesis, Department of
Mechanical Engineering, Washington State University (1986).
3. Crespo da Silva, M.R.M. and CC. Glynn, "Nonlinear flexural-flexural-torsional dynamics of inextensional beams.
I. Equations of motion," J. Struct. Mech 6 (4), 437-448 (1978).
4. Haight, E.C. and W.W. King, "Stability of parametrically excited vibrations of an elastic rod," Proc. South-
eastern Conf. Theor. Appl. Mech., 5th, Raleigh, NC (1969).
I
The Nonlinear Response of a Slender Beam Carrying a Lumped Mass
to a Principal Parametric Excitation
by
Lawrence D. Zavodney
Department of Engineering Mechanics
The Ohio State University
Columbus, OH 43210
and
All H. Nayfeh
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, VA 24061

Many structural elements can be modelled as a slender continuous

beam with concentrated masses located between the ends. When the
support of a cantilevered beam undergoes motion, the beam is subject to
vibration--either external or parametric, or both. Many researchers
have investigated parametric resonances in structures; the text of
Nayfeh and Mook [11 provides an extensive review of the literature.
Schmidt and Tondl [21 also consider parametric resonances in their
text. Haxton and Barr [31 studied a cantilever beam carrying a tip mass
mounted onto another oscillating large mass; the system was considered
an autoparametric vibration absorber. Sato, et al [41 analyzed the
parametric response of a horizontal beam carrying a concentrated mass in
a gravity field.
1. Derivation of the Nonlinear Differential Equation of Motion

The governing equation of motion of the beam shown in Figure I is

derived using the Euler-Bernoulli theory. We assume that the effects of
shearing deformation and rotatory inertia of the beam can be
neglected. If the beam is kept relatively short (< 30 beam widths), the
transverse vibration is purely in plane (if the lumped mass is
symmetrical with the centerline), and if the excitation frequency is far
below the first torsional mode, then we can safely neglect the torsional
modes of the beam in the analysis. These assumptions are consistent
with observations in the laboratory. Also, we do not observe any
combination or internal resonances.

Keeping up to third-order terms, we obtain L-t rning equation:

EI(v
E(ssss + 2 V s
I Vssss 2 + 3vsvsv
ss ss + Vss) + (I
v 2 Vs
2
'

L
P + m6(s - d)lv - - as s(Nv ) + s [p + mns( - d)lvd&

{J6(s - d)[vs(l + vs + "") + vs s 1}

2 v_ - ...)c' + v V cd =0
where s

I
I .. . .
 NL 1 L 2

2
L f0
S
vn2 ])ttdnld
m S
6(4 - d)[f 0 (V)ttdnld&
n m

+ m(z - g) I 6( - d)d + PL(1 - c)(z - g) . (2)

S

2. Solution of the Linear Problem

The governing equation is nonlinear and does not admit a closed-
form solution. Therefore, an approximate solution will be sought that
satisfies both the equation and the boundary conditions. Since the
boundary conditions are spatial and independent of time, we represent
the solution of the nonlinear problem in the form

v(s,t) = I ro n(s)Gn(t) (3)

n
where r is a scaling factor, * (s) is the shape function of the nth
linear mode, and Gn(t) is the 2ime modulation of the nth mode and
assumed to be harmonic. The undamped linear free vibration problem
(without the rotatory effects of m) is governed by
EIviv + [p + ma(s - d)]v = 0, (4)

Without loss of generality, we will solve explicitly for the first mode,
with the understanding that the eigenfunction of the nth mode and its
associated eigenvalue correspond to the nth characteristic.
The solution of (4) that satisfies the boundary conditicns is

i(s) = C,[(sin
L s - sinh
sin L S-
s) - A(COS L
s)-L~ cosh r s)]

+ C U(s - d){(h1 - ah2)[sin k - d) - sinh (s - d)]

+ (h3 - Ah)[cos - (S - d) - cosh L (s- d)I} (5)

where the characteristic k is the nth root of the characteristic

equation. The hi in (5) are constants associated with k. Details of
I
the derivation and solution are in reference [5].

3. Solution of the Nonlinear Differential Equation of Motion

Since we are analyzing the first mode, this continuous system can
be discretized by Galerkin's method. When this procedure is applied to
(1) for the case of a single mode, we obtain
I
I
I
 G + 2eG + [1 - f cos(oT)IG + LG + C GGT

+ C 2G2G - 2cvG2G = 0 (6)

where c, 3, K1 , K;, v are constants, f is the amplitude of excitation,

and e is a dimensionless parameter. Equation (6) contains cubic
nonlinearities and nonlinear damping, and hence it does not lend itself
to a closed-form solution. It can be analyzed by a perturbation or a
3 numerical technique.
A first-order uniform solution using the method of multiple scales
3 [6,71 is found to be
2

I G(T)=aCrT+0) + at-
4 + (-- K, c)cos[3( .T + 8)]

+ a2sln{3rOT + )]I}+ .... (7)

The frequency-response curve for a typical system is shown in

Figure 2. The dashed curves represent unstable solutions. Figure 3,
which corresponds to region It of Figure 2, shows that parametric
vibrations exist only when the excitation amplitude exceeds a threshold
value. When the frequency of the excitation is increased to region III,
the frequency-response curve is multivalued, resulting in a subcritical
instability. This is shown in Figure 4.
4. Results and Discussion of the Composite Beam Experiments

A symetrical 0-90-90-0 ° 4-ply graphite-epoxy composite plate 0.022

inches thick was fabricated and cut into strips one-half inch wide. The
frequency-response curves of the composite beam for three levels of
excitation amplitude are shown in Figure 5. We observe that the general
behavior is as predicted by the theory. There is, however, a maximum
frequency at which point further increases in the frequency cause a jump
down to the lower branch. We also note the appearance of chaotic
behavior for the largest amplitude response; it was preceded by a
modulati"n in the amplitude. We observe a penetration of a stable
trivial solution into what is typically the unstable region of
parametric resonance. Inside this "unstable" region small disturbances
decayed and large disturbances grew, but as the region was penetrated,
the disturbances that decayed became smaller and smaller until the
trivial solution became unstable to all disturbances. We also see the
lower branch lifting off the frequency axis as the frequency is
decreased from above. This behavior was not predicted and appears to be
intensified due to the higher level and nonlinear nature of the damping
present in the composite beam.
i The nature of the parametric resonance for t = 2.000 and for a
table acceleration level of 1.00 g is shown in Figure 6(a). When the
excitation frequency is increased to o = 2.013 and the model is released
from rest, the lower branch attracts the response, as shown in Figure

I
 |I
6(b). After the system achieved steady state, it was disturbed, and we
note that the disturbance caused the system to jump up to the large
amplitude response. Here the system modulates and does not achieve a
constant steady-state amplitude. I
5. Results and Discussion of the Metallic Beam Experiments

A flexible steel beam was fabricated, instrumented with a strainI

gage and fitted with a small mass. The frequency response curve is
shown in Figure 7. We again note a penetration of a stable trivial
solution into the region predicted to be unstable by the theory, and we
note a modulation in the large steady-state amplitude just before it
I
jumps down. The amplitude response for o = 2.000 is shown in Figure
8. These results also show remarkable agreement with the theory (see
Figure 3).
REFERENCES
I
1. Nayfeh, A. H. and Mook, D. T. (1979). Nonlinear Oscillations,
Wiley-Interscience, New York.
I
2. Schmidt, G. and Tondl, Ales (1986). Non-Linear Vibrations,
Akademie-Verlag, Berlin.
3. Haxton, R. S. and Barr, A. 0. S. (1972). The autoparametric
I
vibration absorber. Transactions of the ASME, Journal of Engineerin g a
for Industry, pp. 119-125.
4. Sato, K., Saito, H. and Otomi, K. (1978). The parametric response
1
of i horizontal beam carrying a concentrated mass under gravity.
Journal of Applied Mechanics, 45, 643-648.
5. Zavodney, L. 0. (1987). A theoretical and experimental
investigation of parametrically excited nonlinear mechanical
I
systems. Ph.D. Dissertation, Virginia Polytechnic Institute and
State University, Blacksburg, VA.
6. Nayfeh, A. H. (1973). Perturbation Methods, Wiley-Interscience, New
York.
7. Nayfeh, A. H. (1981). Introduction to Perturbation Techniques,
Wiley Interscience, New York.
mUdj,) Xl

J ' (j mvlt)I
d~t)

\ mg
Ii''
I
dC.
C)'__ % -N

LL P9 10t
# i ##

I0Ci1 0
200202t042'

figr . , CLnt1'eV-d " with a concetraed 14%as Subjecte to FIu* Z. Theo ticaIlly dt M fdequerymid. C e.
mt'l.I
verticIl be4at
 3. ............ 05 -

0 005 010 015 f 020 025 0 000 01to 00 1 020 025

IW~uoZihtO~td fectto nrgo t4"lu 1htesit"o xiainfi einH

Figure1. 1.tort.--11y determined4
variationof the steady-state Figure4. Theoretically vaition of the steady-state
determined

0 (a
20 * (m

cus 0t e lag s
Wetlop o m
p i ue0f eci a i n 0 2 3 4

2 .0 2 .. ee the triia solu ion When 0

.3?h
tECIOOI~~r~r possible.*

03 03 #01
Lr6.Seteti itroftetithetOttiit

03 )3 *l0l

1 .900 2000 Soto 2040

"0
2060 0 01 02 03 040o.5 06 0o a 01 CIto

I Figure 7.
~aceletlevelsItti.
Vrec~tpiecurves for the mtallIlc ban for tw Figure 8. variation Of the mitwoe t
f Of the *etallic hem, for
with the excittatinlltwIi
*200
 THEORETICAL AND EXPERIMENTAL INVESTIGATION OF COMPLICATED RESPONSES OF
A TWO-DEGREE-OF-FREEDOM STRUCTURE
A. H. Nayfeh, B. Balachandran, M. A. Colbert, and M. A. Nayfeh
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, VA 24061

An experimental study of the response of a two-degree-of-freedom

structure with quadratic nonlinearities and a two-to-one internal
resonance to a primary resonant excitation was conducted. The responses
were analyzed using hardware and software developed for performing time-
dependent modal decomposition. When the driving frequency is close to
the higher frequency periodic and aperiodic responses were observed as
predicted by theory. The experimentally observed frequency response of
the system shows good qualitative agreement with the theoretical
predictions.

We consider the response of internally coupled oscillators to

harmonic excitations. If u, and u2 are the generalized coordinates of
the motion, then the governing equations are of the form
u 1 +w 2lU1 + 2iulu 1 + 6
2
1 + 62ulu 2 + 63u22 + 6 u.21 + 65uzu 2

.2
S6 6U2 + 67U 1 I+ a8U2U1 + 6 9 UlU 2 + 6 10 U 2U 2

+ (hllu, + hl 2u 2 )cosat = F cosat , (1)

2 2 2 .2
U2 + W 2 U 2 + 2u 2 U 2 + a 1 u + G2U1U 2 + a 3U 2 + a u1 + Q 5 U 1 U 2

.2
+ 6 U 2 + a7 uiul +oU 8 U 2 U1 + a 9 Ulu 2 + aoU 2U2

* (h2 1 + h2 2u 2)cosot = G cosat , (2)

where the highest order of terms retained is quadratic. In these

equations, w, and w2 are the linear natural frequencies of the
system, Q is the excitation frequency, u, and U are the modal dampings, m
and F, G, n, wn, 6 n' an and hmn are constants. 2

Considering the case w2 = 2 + 0 and a = W 2 +0 F, where I

al and .02 are small detuning parameters, one finds that, to the first
approximation, the response is [11
+
u = alcos 1t - Y 1 12) (3)

u2 = a 2cOs(at - Y2) + "'" (4) 1

where the amplitudes an and an are governed by the following equations:

I
i
5l a 41a - A~ala 2 sinyl (5)

a2 = - u2 a2 + A2 a siny 1 + g siny 2 (6)

3 au 1 = Alala 2cosy1 (7)

2

where 2 2 = a 1 cos - g COSy(8)

5 Yl = at + B2 - 2si and 12 = 0 2 t - 82 (9)

and
4wA +6 6W2 6W2 (0
S11 =2 + s 2 - 68 1 - 6 9w 2 , (10)

2 2
4w2A2 = al - 4 wi - •(11)

Periodic solutions of equations (1) and (2) correspond to the fixed

points of equations (5)-(9). They are obtained by setting a, = 82
=Y = 2 = 0. There are two possibilities. First,

ia =0 and a,= q-, g = G/2w 2 (12)

i and, to the first approximation the response is

u, = 0 and u 2 = a 2cos(at - Y 2 ) + ... (13)

3 which is
essentially the linear solution. Second,

81 = (A A 2 1X1 ±gA 2 x , (14)

(15)
[1 (al + 2 )2
I where a 2 = a2 s = IA 1

I X = C
02(l + 2) - "0'2 (16)

X2 = a 2i u1 +
4u2 (al 2) , (17)
and to the first approximation the response is given by equations (3)
and (4), where al and a2 are defined in equations (14) and (15).

I We note that not all fixed points are stable. They can lose

I stability in one of two ways

I
 1
1) An elgenvalue crosses the imaginary axis into the right-half plane
along the real axis
or
2) A pair of complex conjugate values crosses transversally into the
right-half plane.

The former is associated with the jump phenomenon while the latter is
associated with the Hopf bifurcation. Figure 1 shows frequency-response
curves for the case A, = 1.0, A2 = 0 = 0.02, a 2 = 0.12 and g
I
= 0.1. The solid branches correspond to stable periodic motions, the
dashed portions correspond to unstable fixed points with at least one
elgenvalue being positive, and the dotted portion (-0.047 s a < -
I
0.0127) corresponds to unstable fixed points with the real part of a
complex conjugate pair of eigenvalues being positive. In the latter
case, the amplitudes and phases are not constant but vary with time.
The corresponding response is an amplitude- and phase-modulated or
chaotic motion. 3
An experiment with a structure composed of two light beams and two
concentrated masses, as in Figure 2, was conducted to observe the
amplitude- and phase-modulate motions and to analyze them. The same
model had been used by Nayfeh and Zavodney [21 to observe amplitude- and
I
phase-modulated motions when the driving frequency is close to the lower
natural frequency. The linear resonant frequencies of the structure 1
were determined using a random excitation. The linear resonant natural
frequencies were found to be fl = 8.130 Hz and f 2 = 16.44 Hz. Figure 2
shows also the associated mode shapes. Strain gages were used to
measure the displacements of the beam ends. The frequency of excitation
ranged from 15.5 Hz to 17.5 Hz and the signals from the strain gage were
analyzed using an FFT analyzer, from which the amplitudes a* and a* were
they are proportional to the modal amplitudes al and a2 , 2
found;
respectively. The frequency response obtained from sweeping up and down
the frequency range at a constant level of excitation is shown in Figure
3. The points where Hopf bifurcation was seen is also shown in the
figure. It should be noted that this occurs only when a > 0 and a2 <
I
0. Comparing Figures 3 and I , one can see the qualitative agreement
between theory and experiment. m

To perform the time-dependent modal decomposition, we used the

signal from the strain gage on the vertical beam. This signal was low-
pass filtered (f = 40 Hz) and dmplified and sent to an IBM PC which
acquired data thFough an 8-bit analog-to-digital converter. A quartz
I
clock was used to set the desired sampling rate. Using an FFT
algorithm, we transformed the data from the time domain into the m
frequency domain, enabling us to do the subsequent digital filtering.
The digital filtering separated the two modes. The separated frequency
components were transformed back into the time domain using an inverse m
FFT (IFFT) algorithm. Using quadrature demodulation on the separated
signals, we extracted the at(t) and si(t). The process was performed as
follows. t a

To determine the amplitude a(t) and phase a(t) of the signal

a(t)cos[wt + s(t)I, we multiply it by sinwt and coswt and express the
resulting expressions as
 a(t)cos[wt + a(t)lcoswt = a(t)[cos(2wt + a(t)) + cos(s(t))] (18)
1
a(t)cos[wt + s(t)Isinwt = a(t)[sin(2wt + s(t)) - sin(s(t))] (19)
Using a low-pass filter to eliminate the 2w component, we obtain
p(t) = a(t)coss(t) and q(t) = - a(t)slnB(t), (20)
Then, we calculate a(t) and B(t) according to
a = 2/p2 + q 2, B = tan - - _ (21)
p
Using the aforementioned procedure, we analyzed the strain gage
signal. In the region of stable periodic motions, we found a point
attractor, which showed up as a point in the a* - a* phase plane. In
the region of amplitude- and phase-modulated motions, a limit cycle was
seen in the projection of the attractor onto the a* - a* phase plane.
Figure 4 shows the strain gage signal, the separated modal components,
the amplitudes at(t) and a*(t), and the limit cycle in the a* - a*
2
plane.
References
1. A. H. Nayfeh, "Application of the method of multiple scales to
nonlinearly coupled oscillators" Chapter for the Lasers, Molecules
and Methods, a Volume in the Advances in Chemical Physics Series,
(1987).
2. A. H. Nayfeh and L. 0. Zavodney, "Experimental observation of
amplitude- and phase-modulated responses of two internally coupled
oscillators to a harmonic excitation," accepted for publication,
Journal of Applied Mechanics, 1988.
 ii I
.41g
t

/ as \

Figure 1. Theoretically predicted frequency-response Curves for the

case At " . A, 0.5. g
, - U2 * 0.02, 0 * 0.12. g - 0.1. I
i

• I

Ftg' 2.Two-deroee-of-freedm model tuned for a 2:1 iter1a

ae 2
Fi eo a c n ac coquey in i ner mod sa ps. a erD 1:
12.40 -o "
Bern?: 0.55 x 12. 02d - x
33.1go d - 90.52S -.
0.O4W1/me tu - 4.0g;

II
f
5
II.0 16, IZO17.5

curves ei en e
allyobta
grment, mn dfod ency-reS Onse
Figure . x naturam
isi lo se to me higher
excitato n frequency
frequency x " 16.44 Hz.
 -I-l 16 r

£•@1 2 4 1 a2

F:;
0 !

~
2
2
3

3
16r

6 I
we S

S ia a 7 a a

Figure 4. (a) Long-tive history of the displacement of mass

12

£!A
(b) Long-tim history of the moda components of the
displacement of ass
Mg .
0(t),
(c) Long-time
&-It). history of the Mocll Upllitudes
() mpftudeS phase plane.

F|ue4 a ogtmhsoyo h lpan fms 2

 SESO
NOLNA£NLSSO BASADRTRBAE
THRDY-100-13
Jue218

...
..
..
 I/

Nonlinear Oscillations in a Rotating Shaft System

Toshio Yamamoto* and Yukio Ishida**
* Professor,Dept.of Sci. and Tech., Meijo Univ.,
Shiogamaguchi, Tenpaku-ku, Nagoya,468, Japan.
** Associate Professor,School of Eng.,Nagoya Univ.
Furo-cho, Chikusa-ku, Nagoya, 464, Japan.

1. Introduction
When a rotating shaft is supported by ball bearings, there appear non-
linear spring characteristics in restoring forces due to clearance in bearngs.
Consequently, many kinds of nonlinear forced oscillations may occur. But,
based on the results of the experiments we have conducted,some of these often
appear with large amplitude and others are not observe)(4). Lateral vibrations
of a shaft become a whirling motion due to a gyroscopic moment, and nonlinear
oscillations have unique characteristics which are not observed in rectilinear
systems. This paper will show that the adoption of polar coordinates helps to
clarify such properties and predict the occurrence of oscillations.

3 2. Equations of Motion and Nonlinc r Spring Characteristics

2.1 Equations of motion We consider . four-degree-of-freedom (FDOF) system
where a rotor is mounted on an elastic shaft. A rectangular coordinate system
O-xyz (z-axis coincides with a bearing center line) is considered, and the
deflection and inclination of the rotor are expressed by r(x, ) and 6(6xej).
Let the rotor mass be m, the polar and diametral moments of inertia be Ipand
I, the static and dynamic unbalance of the rotor be e and T, the angle between
these unbalances be 6, the spring constants be a, y, 5, the angular velocity
of the shaft be w, and the damping coefficients be cj(i,j=1,2). By adopting
the quantity eo=mg/a as a representative value, we define the following
dimensionless quantities:
W'=-/,e,=e/(e0ov-T),
!/, e='ee/(e,--),
x'=x/eo, Y'=z/eo t'=t/m,
ip = 1P/I, I' =, ). ,, =j ,"
Nc-=:/( '' l
C -2,
=C12 /"/CL, C22' = cz vrm-/1, e' = e,le0, 7'=T/ (e0 V )
The equations of motion are expressed as follows:
+c,,y +c,,6 +x + ye +.'v, = ew2 cosnt

i
+jW6.3 + C2 X + Y
+yx+ 6e. + Vo
= (p-l)T2 Cos(Wr+E) (2)
+c
+ce +se +
+'( - 2
we the
where us h ,
e primes
foloin are two degree-of-freedomwt3,
omitted, and nonlinear terms are expressed by X,. gx'V.
1,
In order to explain the physical meanings and how to analyze phenomena,
we use the. following two-degree-of-freedom (TDOF) system.
& + ipW 6 + c@x + 6 +Nex = (l-ip)TW 2COS (3
p J6 + C8I + ej + 3p. = (l-ip)Tw 2
-+l8
iC s ytm (3)
For the dimensionless quantities in Eq. (3), some modifications are necessary
in their definition.
2.2 Nonlinear terms The nonlinear terms are derived from the potential
energy V. In the TDOF system, the energy 7 is expressed as follows when up
to the third order terms are considered in restoring forces.
V
= ( +e )2 + 30oX + E 1P 2+ 12 e e 4E0 1)
3

'. + 3 1 + M e + .1
+a 13
3B &3
X-
+ 1 4) (4)

.
The restoring forces are obtained from this by 6x+N.x=;V/aeB and ey+AV/;6
As the shaft moves in a whirling mode, it is considered that the polar
coordinates is more suitable to explain the phenomena. By the transformation
6x=e cosO, 6j=e sinO, Eq(4) is expressed by the polar coordinates as follows:
V (1/2) e2 +(EE cos%+ ej sino + F'_"cos3o + E3) sin30)e 3
+ cos2o+ep) sin2 +ft_cos4 + )sin4o)eI
3
= (1/2)62 +{(5) cos _'-01) +E3 cos3(0- 3 )}e
+{+c &) Cos2(0-02) + cos4 (0-% )}64 )
(5)
The following relations holds between these coefficients
(1) (I = E3)(3)
i
c (3E 30 +EU )/4, E<s (E21 +3E03 )/4, Jc = (E30 -E 12 )/4, ES
(E21- EO3 )/4, 3 (3B40+
= S2+ 3804 )/8, (40-
(6= 8 4 )/2, (6)
es)= (631+ 813 )/4, ec) ( L0- 8 22+ B0 )/8, 54)= (B31- B13 ) /8. . I
1
The coefficients elI)... land P1 ,...,are obtained by the relation E,
( C+V ,
=tan'(E 'I /ES ). The coefficients itcE and ES (i=1,3) belong to the un-
sy netrical nonlinear spring characteristics and e' and S' (i=0,2,4) belong
to symmetrical ones.
Figure l(a) shows a distribution of potential energy V. The shape of V for
a nonlinear system deviates irregularly from that for a linear system whose
potential energy is V0 = (6x'+e')/2. But this nonlinear spring character-
istic can be classified into regular components if we represent it by polar
coordinates. Figures l(b)-(f) shows cross sections of surface V with a
plane parallel to the e8u -plane, in the cases that one of E(l), EM) ? "
and & exists, respectively. It is seen that E") ,
,
, etc. are coefficients
g
of terms which vary their magnitude n times while the angle 0 changes its
value from 0 to 27r. We designate these nonlinear components by the
notation N(n) in this paper.
In the case of FDOF system, the potential energy is expressed by

V 17 8I
2
V, V

*- 0I .
=

V 'V/\

(a) The distribution of V and te otential (cy istributon 3

/rV0 v _i-' (Ia
<k -<1)e)3u
(d) ,.V+8()@
8(O~o) (b) V-VO+C{a)cos (C) V"T/O+ C( 3)c ° 8% 4
( 8
3()03

=0 -- ~-0 A

(d) VV 0 +8(
0
)94 (6(0)>0) (e) V..V0 +t3(2 co2($..42 )6j4 (f) V.VO+0( 4)cos4(0$44z)Oe4

Fig.l Nonlinear components and the potential energy distribution3

3 3
V (X z +Y 2 ) +y(xex +yej) +-I(e 2 + e ); , '
-4 I
,,,,,
J..
e e (7)7),
Restoring forces are derived by differentiating this energy by x, y. 6x, and
eu, respectively. By the transformation x =rcoso,, y = rsinor, ecose, ex,=
and e= esin~a, this potential energy is transformed into the polar coordi-
nate system (r, 0,, 0,),). We show the obtained equation partially as follows.
V= {r 2 /2 + yrecos (,.-4.) + 6e2/2}
+ (63)c COS P,.+E3Os s inO,)r 3 + ... }
40 r 431c cos(r - ) 631s sin(Pr-Pe)}r 30 + ... ] (8)
In this equation, the angles which are not contained in Eq(5) appear. For
example, the angle (O, -W) is constant during the whirling motion, and
therefore the terms containing (Or - ,) is a component expressed by N(O).
Similar to Eqo(6), the relations s30 = (3 0 +s 12 )/4, ... holds among these
coefficients.
2.3 Frequency equation and resonance points Let the natural frequency
be p. The frequency equation of FDOF system is obtained from Eq. (2) as follows.
(l-p2)(d+iPWp-p) _y 2 =O (9)
This equation has four roots pi'p4. These frequencies are shown in Fig.2 as
a function of w. In a FDOF system with square and cubic nonlinearity,
subharmonic (Sub-H) oscillations and summed-and-differential harmonic
(S-and-D-H) oscillations listed in Table 1 have the possibility to occur
theoretically. We designate the former by the symbol [w =mpl] and the latter
by the symbol [w =nTp +np$ ]. These symbols express relations which hold
at the resonance points. These resonance points arp given by the cross points
in Fig.2.
The frequency equation of TDOF 5000.
system of Eq(3) is given by
1 + iPWp - p2 = 0 (10) ". ., , - .... ./:
G -

which has two roots P.,) and 0

Pb(< ) 4000 .

3. Resonance Curves of the ,, --- /_ .,-,..,

--, - -, .. . .
Subharmonic Oscillation [+(l/2)-w] 300 ' --
--------
As a representative example, we 20 , -- ,
select the Sub-H oscillation of order 2000 *',.,/1
- ..
1/2 with a mode of forward precession. H.--,
First, we discuss a TDOF system in r
order to present an outline of analysis. 1000 -
In the neighborhood of the rotating
speed w = wo where the relation w =2-f
holds, the Sub-H oscillation with 0 WC
frequency (1/2)w appears in addition to wc
the harmonic component with frequency w.
When these two components exist, higher -1000
order components with magnitude of 0()
appear due to the nonlinearity. [The -2000D ,
symbol O(E) means that the quantity has
the same magnitude as the small param-
eter C.] These higher order components -3000
represent small deviation of the orbit. 0 1000 2000 3000 4000 5000
Therefore, we suppose an approximate Angular velocity w rp
solution as follows: Fig.2 p -w diagram
 I

%- Rcose,+Pcos(wt+B) +E(acoseF+b sinj)r

e - R sinf + P sin (wt+B) + E ( a'sinef + bcose,) 5 (ii)
where ef=(l/2-wt + ). By inserting this solution into Eq(3), and using
a method of harmonic balance, we get the following equations.
(U -ip) w R = -c (1/2)wR - 2 (C sin26f- es) cos26Sf)RP
(1 -ir) wR6f- GjR - 2 (E' cos2 6 f+ e,) sin26RP + 4e)(R2 +2P 2 )R (12)
Where, G1 =l+jpw(/2)-(w/2) 2 . The harmonic solution are given in the
accuracy of O(E) as P= (l-ip)TW 2 /{l-(l-jp)W 2 }, S=7. From Eq.(12), we get
the following equations which give the stationary solutions Ro, So.
R0 = 0 (a)
{Gi +4e'(R 2 + 2P 2 )} 2 +c 2 (w/2)2 =411I 2 p2 (b) (13)
We can investigate the stability of these solutions by the same method
used in the literature. Resonance curves are shown in Fig.3, where full
lines and dotted lines represent stable and unstable solutions, respectively.
From Eq.(13), we see that only the components N(O) and N(1) have influence
on this oscillation. Figure 3(a) shows that N(0) (that is, the coefficient
00) determines the inclination of the resonance curve, and Fig.3(b) shows
that N(l) (that is, the coefficients 61and O ) determine the intensity of
the oscillation.
In FDOF system, the same kind of subharmonic oscillation appears at
the point C in Fig.2, where the relation w=2p2or p 2=(i/2)w holds. The
way of the theoretical analysis is almost the same as that mentioned above,
althogh its calculation is more complex and abundant. The equation
corresponding to Eq. (13b) is given by the following equation.
(/2 -P2 - {A(N(0))R'+B(N(0),F)}+C 2w 2 = D(N(l),F) (14)
In this equation, A(N(0)) means a constant containing the coefficients A41,
6oC, ... , B(N(0),F) is a constant containing 4' Sc, .- and the amplitude
Fi of harmonic solutions, C is a constant relating to damping, and D(N(l),F)
is a constant containing E30C, ... and F . By comparing Eq. (13) and Eq. (14),
it is concluded that the vibration characteristics, such as, the shape of
the resonance curve, relating nonlinear components, and so on, in the FDOF
system are qualitatively the same as those in the TDOF system.

5. Summary of Nonlinear Forced Oscillations in the FDOF system

The analytical and experimental results are summarized in Table 1. The
component N(O) has influence on every kinds of oscillation and it determines

0l1 0. 01 C=0. 009S

D0=-0. 0 tI 01 0"O. 33!=. v,0 3 8Wue'=o. 050
/S=.
0.4" j 0. 4C=0. wC0. 1 05
0. 4 -'O. 1
r=o. 1
o --- 0. 2 I 0. 05 o e1'uO.
0. i 02 0.4 048

02
I

e We &O. 046

3.0 3.1 3.2 3.3 3.4 3.1 3.2 3.3 3.4 3.5

(a) Effects of N(O). (b) Effects of N(1).

Fig.3 The effects of the nonlinear components N(O)and
N(1) on the subharmonic oscillation [w =2pi]

mmm mmm
mR• ms~toI
 'p5

Table 1. S arv of theoretical and e esults

Nonlinear Experiments 4
Precessional motion Kinds of Oscllation conent A.11"' B -I i

Sub-H Fonrard(F) w3P

4
w3P2( <1/3) N(O).N(2) x x 0
,.,.-3
pD0.NO x
Oscil . Backward(8) *w43P4, N(O).N(4) X
+ F , 2PFPa 2 (ip<1/2) N()N(2) o resonnce
- W'*Pt+2 p2(ip 1) Points
-- (FB W4 2Pt-P3(1#<ll2).*wJV2pa-p= ()
- F + B (2F+2)
w+whPi-2 p 2 ( ,< 1),* ,p2_p
N(O).N(2)
_
X on
.wI Pi-2 P3VDSkx- a3
$,,,P,-2 p 4(Ui<l).$W'p%-2 P _

( '., et
8 S+w4-pa-2 P4
-2pr-P" N(O).N(4) X X X
XW F+F+B 'FP 1+P2 p4 (i,<i) N(O)

F +B+B W t'P1Ps-p4(P<I) N(O).N(2) X X 0

F (Sub-H
w42P,0<I/2) N(O)N(1) 0 0 0
b-H F w2p3
0ol1 ____ 6 1'-2P4
$ w=
w'-2P 4 N(O).N(3) A x X

S-and F+F W P+P 2 (i,< 1) N(O).N(I)

."0-H 8
F +._______ P,-p,(i,<l).*w~=-p
w* P1P.(i,<l),,W, P.-P N(O),N(I) 0 00 0
0i____

Oscil. B 8 'J4-P-P_ N(0)N(3) A X X

the inclination of resonance curves. In addition, one of the components
N(O) ",N(5) has influence on each oscillation and it determines the intensity
of the occurrence. The experimental results reported previously )'< )are
obtained in the apparatus consisting of an elastic shaft and a disc (ip=2).
In these apparatus, the oscillations with * have their resonance points.
Apparatus A are systems with strong unsymmetrical nonlinearity, and they
are used on the literatures( -) . Apparatus B-I and B-II are systems treated
in the literature . The former apparata B-I is a system with strong
symmetrical nonlinearity with no directional difference, and the latter
apparata B-Il is a system with strong symmetrical nonlinearity with
directional difference. The symbol O means that the oscillation appeared
often with large amplitude in experiments. The symbolnA means that the
oscillation appeared sometimes with small amplitude. The X means that the
oscillation did not appear through many experimemts. By comparing the
theoretical and experimental results, it is known that (a) the oscillations to
which simple-shaped nonlinear componentssuch as N(O) and N(), have influ-
ence in their intensity often appear and (b) those to which complex-shaped
nonlinear components, such as N(3) and N(4), have influence are difficult to
appear. Therefore, it is concluded that, if we check the influencing non-
linear components, we can predict the occurrence of the oscillation.

6. A Simple Rule Predicting the Relating Nonlinear Components

From the above-mensioned results, the importance of nonlinear component
is established. Concerning these nonlinear components which has influence
on the intensity of occurrence, simple rule can be stated: namely, for the
oscillation which appear when the relation aw= bo +cpholds, the nonlinear
component N(k) (k=lc-b-c) have influence on the intensity.
(literatures] (1)Yamaoto, T.,Trans.Japan Soc.Mmch.Enqrs,Vol.21,NO.1ll(1955),p.853. (2)Yama Oto,T., Trans.
72 7
Japan Soc.Mech.Enqrs.Vol.22,so.115(1956),P.1 . (3)Yamamoto, T.,sUl ..S ,Vol3.NO.12(19601,p.39 . (4) 6 4
96
Yamamoto, T. eat al.,Bul.JSI.ME,Vol.18,NO-123(1975)p. 5. (5)yamamoto,T.t al. ,BulI.JSME,Vol.22,No.164,P.1
 I
Nonlinear Effects in the Static and Dynamic
Behavior of Beams and Rotor Blades I
Dewey H. Hodges*
Georgia Institute of Technology
Atlanta, Georgia, U. S. A.
Marcelo R. M. Crespo da Silvat
Rensselaer Polytechnic Institute
Troy, New York, U. S. A.
I
and I
David A. Peters*
Georgia Institute of Technology
Atlanta, Georgia, U. S. A.

Sponsored by the U: S. Army Research Office

Introduction
In work published recently in Vertica, Nagaraj and Sahu [1] analyze the static
and dynamic behavior of an end-loaded cantilever beam. They develop solutions for
the static behavior based on perturbation methods and present these results along
I
with numerical results for the free-vibration frequencies. Contrary to known laws of
physics, they allege that their various analyses exhibit non-negligible differences in tip
deflection, tip rotation and free-vibration frequencies, in spite of the fact that these
analyses are based on only one analytical model with differences arising only from
the treatment of the finite rotation of the beam cross section frame. They conclude
that, in modeling the finite deflections of an Euler-Bernoulli beam, the sequence of
rotational transformations that is used to construct the matrix of direction cosines
(of the deformed-beam cross-sectional frame) "affects the nonlinear corrections to the
bending deflection." The authors also state that "the closed-form solutions derived in
the present paper show that systematic differences exist in these quantities in solutions
based on modified Euler angles ... " (See "Conclusions" on pages 661 - 662).
We show in this paper that these and other conclusions of [11 are patently false
and result from numerous errors in the analysis. Most of the errors in [1] are not new;
previous works with mistakes of a similar nature are discussed in (2]. The present paper
is intended to be a critical analysis and discusion of the work presented in [1]. It seems
* Professor, School of Aerospace Engineering
t Professor, Department of Mechanical Engineering, Aeronautical Engineering, and
Mechanics
I

apparent that only a brief look at the title is sufficient to know that [1] is seriously in
error since the mathematical description of a system cannot affect its physical attributes.
In the paper we discuss the nature of the errors of [1] in terms of the fundamentals of
mechanics, nonlinear analysis, interpretation of experimental results, and rigid-blade
modeling.

Uniqueness of Solution
3 For a given analytical model of any system, we expect to obtain the same answers
for its behavior, regardless of the variables we use to describe the system. In the paper
we discuss the concept of impenetrability and what it implies concerning the use of
orientation angles in dynamics analysis.
One of the most fundamental of all the laws of physics is the law of impenetrability.
n This law states that, at any instant in time, a particle of matter can occupy only one
position in space. On the basis of this law, it is not difficult to show that a rigid body can

£ assume only one orientation at any particular instant in time. In continuum mechanics
textbooks, this law is often stated in terms of the continuity of deformation resulting in
a one-to-one mapping between deformed and undeformed structural configurations. As

Irigid" consequence, when attempting to model a system which consists of particles and/or
bodies, the analyst develops appropriate mathematical expressions which convey
the position of each particle and orientation of each rigid body at a particular time,
generally grouped under the heading of kinematics. If these expressions are valid for
i all time within some range of interest, then the application of laws of motion will allow
for analytical or numerical simulation of the motion of the system in question over the

IIn time interval of interest.

the paper by Nagaraj and Sahu, this fundamental concept is severely violated.

3in Nagaraj and Sahu, as have some other investigators such as some of those discussed
[2], developed more than one expression for the orthonormal transformation matrix
between two reference frames. By unwisely using the same symbol for the third angle
of two distinct sets of orientatic i angles, they were apparently then unable to see that
these quantities are different angles. Since the expressions for the direction cosines are
very different for different treatments of finite rotation, the development in [1] leads to
a fundamental fallacy: that a frame (which is kinematically equivalent to a rigid body)
can have more than one orientation at one time.
The consequences of this error appear in the numerical results of [1]. Recalling the
one-to-one mapping between deformed and undeformed structures, a particular physical
model of a beam can exhibit only one deformation for a given load. On the other hand,
Nagaraj's and Sahu's results show a displacement of the elastic line and an orientation
3 of the cross section at the tip of the beam that depend upon their choice of rotational
variable - which is patently nonsensical!
It is well known to dynamicists that it is the values of the rotational variables
(their being, in a sense, intermediate quantities) that depend on the choice of rotational
!2
I C
 I
variables. That is, at some instant in time the direction cosines of some particular frame
are unique, but the values of the variables used to evaluate the direction cosines will
depend upon what those variables are, whether orientation angles of any of the 24 or
more types [3], of Rodrigues parameters, or of any other measures [4]. However, the final
answer for a component of displacement, a component of angular velocity or velocity, a
strain component, a direction cosine, or other measurable quantities, cannot depend on
the choice of rotational variables. We will see firm evidence of this when we correct the
nonlinear analysis of [1] and properly compare their results with experiment.
When orientation angles (referred to by the authors of [1] as "modified Euler an-
gles") are used, the third orientation angle is not the beam's total rotation at some
particular value of the axial coordinate. It is simply an orientation angle of the rotation
sequence and, thus, its value may, of course, depend on the particular sequence used. In
the paper we show that if one were to assume that all the various torsional kinematical
variables used in [1] were equal, then the distance from a particle to a plane is not
unique at an arbitrary time. Thus, these quantities are not equal and, thus, cannot be
all the same quantity!

Errors in the Nonlinear Analysis of [1]

To analyze nonlinear systems, special attention should be given to the formulation
of the differential equations of motion for the system. Also, if an approximate solution
to the governing differential equations is generated by a perturbation analysis, care
should be taken so that the original nonlinear equations are expanded in terms of a
small parameter, E, to the same order that is desired of the solution. Here we let, the
deflections and torsional kinematical variable each be O(E), where e is a bookkeeping
parameter to identify the order to which each of these variables is retained in the
equations of motion. If the nonlinearities in a set of equations are expanded to 0(0'),
then the solution to the expanded equations is, of course, only valid to 0(e0). If a
solution valid to 0(Em+n) is desired, with m > 0, then all terms of O(em +n) must be
retained in the expanded equations. With this in mind, we address some other erroneous
conclusions and results presented in [1].
There are two basic sources of error in the analysis of the end-loaded cantilever
presented in [1]. The first and more important is in the approach in which the authors
fail to fully develop Eqs. (16) to 0(e0) in order to obtain a set of consistent O(e3 )
approximations for the deflections v and w. The main reason that Nagaraj and Sahu
failed in their attempt to correct their Eqs. (16) to be valid to O(e 3 ) is that they failed
to include O(e 3 ) terms in their expressions for bending curvature (their Eqs. lib and
12b for ky and k.). It turns out that this, alone, is responsible for the authors' erroneous
conclusion that the deflections v and w depend on the rotation sequence. The other
basic source of error happens to be in Eqs. (8). We will prove that Eqs. (8) are in error
by examining a special planar case of those equations. It should be noted that much of
the erroneous analysis in [1] is developed properly in [5,6].

3U

.
I
 In the paper, it is shown that the entire analysis in [1], including static deformation,
dynamics, and lateral buckling, is in error. By far the most serious of the errors are
the failure to distinguish the rotational kinematical variables and the failure to properly
develop Eqs. (16) to O(e 3 ). Therefore, the conclusions of [1] regarding the influence of
transformation sequence, many of which were derived from the faulty analysis therein,
are blatantly in error and physically groundless.

Other Errors in [1]

There are two types of experiments described in [7,8] with which numerical results in
[1] are compared. One deals exclusively with static behavior of an end-loaded cantilever
beam for large deflections, and the other deals with the dynamic behavior. In the former,
the tip displacements and one angle for determining the tip rotation are reported. In the
latter, the fundamental flatwise and edgewise bending frequencies are reported. Nagaraj
and Sahu compare with both and erroneously report a dependence of tip rotation,
tip displacement, lateral buckling load, and fundamental frequency on transformation
sequence. With the tip displacement, buckling load, and fundamental frequency, the
errors noted above are responsible for this false conclusion. Indeed, it is easily shown
that the correct form of the governing equations that were solved in [1] does not depend
on the transformation sequence. For the tip rotation, Nagaraj and Sahu do not properly
identify the quantity that was measured in the experiment. We properly identify it and
show that it is, indeed, independent of transformation sequence.
We also show in the paper that partial differential equations of motion with nonlin-
ear terms retained only through second degree are not sufficiently accurate to determine
the large deflection behavior of end-loaded cantilever beams, a conclusion that can also
be inferred from the work of [9 - 13]. The reason now is clear: recall that in the static
problem, the torsional variable is O(e'2 ) and always appears in the flexural equations
multiplied by terms that are 0(e). Thus, the static equations which stem from dynam-
ical equations with only second degree nonlinear terms, used in finding the nonlinear
equilibrium solution about which to linearize, are not consistent since they contain some
0(6 3 ) terms, but not all. Thus, we should not be surprised that a formulation based on
consistent partial differential equations of motion, with only second degree nonlineari-
ties in v, w, and the torsional variable, would perform in a mediocre manner for this
5- problem.
We also comment on the section of [1] which discusses the flap-lag stability of rigid,
articulated blades. The paper presents stability boundaries for the so-called "sequence-
Ifree" transformation and shows that they are between the boundaries for blades with
flap-lag and lag-flap hinge sequences. This observation is followed by the extraordi-
narily false conclusion, "Since the flap and lag hinges are assumed coincident, there is
no preferred sequence which can be specified for this problem and the sequence-free
formulation appears to be a natural choice."

I
 I

References
1. Nagaraj, V. T., and Sahu, N., "Influence of Transformation Sequence on Nonlinear
Bending and Torsion of Rotor Blades," Vertica, Vol. 11, No. 4, 1987, pp. 649 - 664.
3
2. Hodges, Dewey H., Ormiston, Robert A., and Peters, David A., "On the Nonlinear
Deformation Geometry of Euler-Bernoulli Beams," NASA TP-1566, 1980.
3. Kane, Thomas R., Likins, Peter W., and Levinson, David A., Spacecraft Dynamics,
McGraw-Hill, New York, 1983, chapter 1. I
4. Hodges, Dewey H., "Finite Rotation and Nonlinear Beam Kinematics," Vertica,
Vol. 11, No. 1/2, 1987, pp. 297 - 307. 3
5. Crespo da Silva, M. R. M., and Glynn, C. C., "Nonlinear Flexural-Flexural-
Torsional Dynamics of Inextensional Beams: I. Equations of Motion," Journal of
Structural Mechanics, Vol. 6, No. 4, 1978, pp. 437 - 448. I
6. Crespo da Silva, M. R. M., and Glynn, C. C., "Nonlinear Flexural-Flexural-
Torsional Dynamics of Inextensional Beams: II. Forced Motions," Journal of Struc-
tural Mechanics, Vol. 6, No. 4, 1978, pp. 449 - 461.
7. Dowell, E. H., and Traybar, J., "An Experimental Study of the Non-linear Stiffness
of a Rotor Blade Undergoing Flap, Lag and Twist Deformations," AMS Report
No. 1194, Princeton University, Jan. 1975.
I
8. Dowell, E. H., and Traybar, J., "An Experimental Study of the Non-linear Stiffness
of a Rotor Blade Undergoing Flap, Lag and Twist Deformations," AMS Report
3
No. 1257, Princeton University, Dec. 1975.
9. Hinnant, Howard E., and Hodges, Dewey H., "Application of GRASP to Nonlin-
ear Analysis of a Cantilever Beam," AIAA Paper 87-0953, Proceedings of the 28th
I
Structures, Structural Dynamics and Materials Conference and Dynamics Special-
ists Conference (Part II), Monterey, California, Apr. 6 - 10, 1987, pp. 966 - 975.
10. Dowell, E. H., Hodges, D. H., and Traybar, J., "An Experimental-Theoretical Study
of Nonlinear Bending and Torsion Deformation of a Cantilever Beam," J. Sound
and Vib., Vol. 50, No. 4, Feb. 22, 1977, pp. 533 - 544.

11. Rosen, A., and Friedmann, P., "The Nonlinear Behavior of Elastic Slender, Straight,
Beams Undergoing Small Strains and Moderate Rotations," Journal of Applied
Mechanics, Vol. 46, 1979, pp. 161 - 168.
12. Rosen, A., Loewy, R. G., and Mathew, M. B., "Nonlinear Analysis of Pretwisted
Rods Using 'Principal Curvature Transformation' Part I: Theoretical Derivation,"
AIAA Journal,Vol. 25, No. 3, 1987, pp. 470 - 478.
13. Rosen, A., Loewy, R. G., and Mathew, M. B., "Nonlinear Analysis of Pretwisted
Rods Using 'Principal Curvature Transformation' Part II: Numerical Results,"
AIAA Journal,Vol. 25, No. 4, 1987, pp. 598 - 604.

....
....
I
I

3w NONLINEAR FLEXURAL-FLEXURAL-TORSIONAL-EXTENSIONAL DYNAMICS OF BEAMS:

FORMULATION AND RESPONSE

I M.R.M. Crespo da Silva

Professor, Department of Mechanical Engineering,
Aeronautical Engineering and Mechanics
Rensselaer Polytechnic Institute
Troy, New York 12180-3590 U.S.A.

I
ABSTRACT

3 The nonlinear dynamics of beams capable of undergoing flexure along

two principal directions (and, thus, flexure in any direction in space),
torsion and extension, is discussed. The analysis presented here is based
on a set of consistent nonlinear differential equations of motion that
are valid for extensional and for inextensional beams. The beam's
material is assumed to be Hookean, and the beam's properties may vary
along its span. The nonlinearities in the differential equations of
motion include contributions from the beam's curvature and torsion expres-
sions, from inertia terms, and from midsurface extension if the beam is
extensional. A number of studies are being conducted at R.P.I. dealing
with the nonlinear flexural-flexural and flexural-flexural-torsional
dynamics of inextensional beams and beam-like structures. In this paper

5 the influence of several types of nonlinearities in the dynamics of inex-

tensional and of extensional beams is discussed.

3 FORMULATION AND ANALYSIS

The nonlinear differential equations governing the flexural-flexural-

torsional motions of Euler-Bernoulli inextensional beams were formulated
by Crespo da Silva and Glynn [1]. The equations developed in [1] are
valid for arbitrary property variations along the beam's span. They are
also valid for the general case where the bending and torsional motions
are of the same order. A number of cases were investigated by the same
authors involving the nonlinear non-planar free and forced response of
inextensional beams and for the case where the torsional natural frequen-
cies were much higher than the bending natural frequencies [e.g. 2-81.
Non-planar motions of extensional beams were considered by Ho, Scott and
Eisley [9,10] by making use of a set of differential equations where
torsional effects and nonlinear contributions to the curvature were
3 neglected a priori.

From a fundamentally rigorous point of view, the inextensional

assumption and the differential equations of motion for beams with fixed-
sliding or with fixed-fixed boundaries should be a by-product of a unified
approach that treats both extensional and inextensional systems. One then
could assess the validity of neglecting nonlinear terms, such as higher-

I
 U
I
order contributions to bending curvature and the torsion terms, when I
analyzing the nonlinear response of such systems.

The details of thb fmr't,latiin of the nonlinear differential 5

equations of motion, and their boundary conditions, for initially straight
Euler-Bernoulli beams able to undergo flexure along two principal direc-
tions, torsion and extension, are presented in [11]. The nonlinearities
present in the equations include contributions from the curvature and
U
torsion expressions, and from inertia terms. The equations developed in
[111 are also valid for the general case beam's stiffness and distributed
mass may vary along its span, and when the flexural and torsional motions
are of the same order. A particular form of these equations, valid for
the simpler case when the beam's properties are constant along its span
and when the distributed mass moments of inertia are neglected (and thus,
when the torsional natural frequencies are much higher than the beding
I
natural frequencies) are given below. Here, x denotes distance, normal-
ized by the undeformed beam's length L, measured along the line joining
the beam's supports at x -0 and at x l, and t is normalized time as
defined in [1].

2
+c + + Y+ SA KK"u (v' 2 +w' )dx

" 2U
w(-) f v"w"dx - w f v"w'dx
0 -" f f vIwfdxd
L 0 1 x

Y
w ","
f"dx
0 1
: -
+K.1
r
'K
0 1
""dx d ' ..

2
+ V ' 2 +W12 )dx + +u f (vP2 + )dx rv2 )djdx

(la) 3
i,+ c, + W"",. Q (X,t) + f (v'2 +w'2 )dx

{(- BY) i v"w"dx + v' f v"w'dx - v"'v'w' - f 1f vdxdXj

+
U1- a )2
_ f
r X
f f v"w"dx dx -
K1
K (xv") f f v"w"dx d
X VS+WWV
4. .I'(v'v"+w'w")'
+ L 0 1 y 0 1

- f ]
- - f (v12+ ,)dx He T [ARcw
+ 0 0 0

(lb) 5
U
I
3 In equations (la,b), v(x,t) and w(x,t) are the components, along
orthogonal inertial directions, of the beam's elastic deformations due to
bending; 8A'EEAL2 /(2EIn), 8YEI /(EI) , $y - GJ/(EI.), where A and L are,
raspecti-;e.,, the beam's "cross-sectional area" and length; EIl and EI;
are the beam's bending stiffnesses and GJ is its torsional stiffness.
The parameters Ku and K. are constants; Ku -0 if the end at x- 1 is free
to move, and Ku M if it is fixed; Ky-if the end at x-1 is free to
rotate, and Ky -- if it is restrained against rotation. The terms multi-
plied by 8A in equations (la,b) are due to midplane stretching of the
beam's mid-surface. For inextensional beams (Ku -0) those terms are
absent from the equations, of course.

wD,/(EAL2 ) is the square of the radius of gyration

The quantity V,
(normalized by the length L of the undeformed beam) of the beam's cross
section and, thus, is very small. For extensional beams, where Ku# O,
the terms multiplied by 8A in equations (la,b) are the "dominant" nonline-
arities in those equations [12]. For such beams, the bending deflections
are O(p) and, thus, very small. When Ku - 0 the boams behaves as inexten-
sional [12]. In this case, all nonlinearities for an initially straight
beam are cubic. For Ku =0, and for K.- 0, equations (la,b) reduce to
equations (5a,b) in [21.

3 NONLINEAR RESPONSE: A BRIEF OVERVIEW

The nonlinear coupling terms in equations (la,b) car cause a resonant

energy exchange between the v and w bending components ot the response,
and between different "modes" associated with the planar or with the non-
planar response of the beams. Figure I shows the single harmonic response
of a homogeneous clamped-free beam (Ku KyO ) with B- (1.01)2, driven by
a periodic force Qv(s,t) -K(s) cos t with Q 1.01x 1.875)2, c 0.05 and
q - 0 K(s)F(s)ds, where F(s) is the beam's first modal eigenfunction
associated with the linearized counterpart to equations (la,b)(s is arc
length along the deformed inextensional beam). For these parameter
values, the planar response is unstable.

5 .25

-.25 op .25

5 Figure 1. Nonlinear response of a beam subjected to a planar excitation.

 I
I
The response shown in Figure 1 is non-planar and stable, and the I
beam's tip describes, to a first approximation, an ellipse in space. The
spatial nrientation of the ellipse depends on By, c, a and q. I

Figure 2 shows the influence of EAL 2 /D in a typical amplitude-

frequency response characteristics of an extensional beam undergoing
planar motion. The response shown is for a clamped-clamped beam subjected
to a harmonic excitation with frequency flnear the beam's first natural
frequency w, and with c -0.002 and q -0.0002.
"'" I
I' 2.Mn

a
a
0.00...2

0/ ~ - I-

Figure 2. Typical amplitude-frequency response for an extensional beam

The same type of response shown in Figure 2 is exhibited by clamped-pinned

and by pinned-pinned beams. An upper bound for the maximum amplitude,
a of the planar resonant response of extensional beams has been
determined in (121. With a - 1 F'2 (x)dx]2 , it was shown in [121 that

a
2
_ . DnI(EAL )j
<< 4w[3 (2)

A number of studies are being conducted at R.P.I. concerning the 3

nonlinear non-planar response of inextensional beams and of beam-like
structures. These include modal interactins both in the flexural-flexural
and in the flexural-flexural-torsional dynamics of such beams. Beams with
constant and with variable properties along the span are being conducted.
3
A number of well-controlled experiments are also being performed at R.F.I.
in order to generate data that can be used to verify the results of the
analytical investigations being conducted.
I
I
Im
n
I

3 REFERENCES

1. Crespo da Silva, M.R.M., and Glynn, C.C., Nonlinear flexural-flexural-

torsional dynamics of inextensional beams. I. Equations of motion,
J. Struct. Mech., Vol. 6, pp. 437-448 (1978).

2. Crespo da Silva, M.R.M., and Glynn, C.C., Nonlinear flexural-flexural-

torsional dynamics of inextensional beams. II. Forced motions,
J. Struct. Mech., Vol. 6, pp. 449-461 (1978).
3. Crespo da Silva, M.R.M., and Glynn, C.C., Nonlinear non-planar resonant
oscillations in fixed-free beams with support asymmetry, Int. J. Solids
and Struct., Vol. 15, pp. 209-219 (1979).

1 4. Crespo da Silva, M.R.M., and Glynn, C.C., Out-of-plane vibrations of a

beam including non-linear inertia and non-linear curvature effects,
3] Int. J. Non-Linear Mech., Vol. 13, pp. 261-271 (1979).

5. Crespo da Silva, M.R.M., Flexural-flexural oscillations of Beck's

column subjected to a planar harmonic excitation, J. Sound and Vibr.,
3s Vol. 60, pp. 133-144 (1978).

6. Crespo da Silva,M.R.M., Harmonic non-linear response of Beck's column

to a lateral excitation, Int. J. Solids and Struct., Vol. 14,
pp. 987-997 (1978).

7. Crespo da Silva, M.R.M., Nonlinear resonances in a column subjected to

J a constant end force, J. Appl. Mech., ASME Trans., Vol. 47, pp. 409-
414 (1980).

8. Crespo da Silva, M.R.M., On the whirling of a base-excited cantilever

beam, J. Acoust. Soc. Am., Vol. 67, pp. 740-707 (1980).

9. Ho, C.H., Scott, R.A., and Eisley, J.G., "Non-planar, non-linear

oscillations of a beam. I. Forced motions, Int. J. Non-linear Mech.,
Vol. 10, pp. 113-127 (1975).

£ 10. Ho, C.H., Scott, R.A., and Eisley, J.G., "Non-planar, non-linear oscil-
lations of a beam. II. Free motions, J. Sound and Vibr., Vol. 47,
pp. 333-339 (1976).

S11. Crespo da Silva, M.R.M., Nonlinear flexural-flexural-torsional-

extensional dynamics of beams. I: Formulation. To appear in the
3 Int. J. Solids and Struct.

12. Crespo da Silva, M.R.M., Nonlinear flexural-flexural-torsional-

extensional dynamics of beams. II: Response analysis. To appear in
the Int. J. Solids and Struct.

I
I
 NONLINEAR MODAL COUPLING IN THE RLSPONSE OF INEXTENSIONAL BEAMS

M.R.M. Crespo da Silva and C. L. Zaretzky

Department of Mechanical Engineering,
Aeronautical Engineering & Mechanics
Rensselaer Polytechnic Institute
Troy, New York 12180-3590 U.S.A.

ABSTRACT

Nonlinear inertia and curvature terms, as well as nonlinear terms

that couple tlexural and torsional motions, can have a dominant role in
the response of inextensional beams or beam-like structural members. In
this paper the effect of modal coupling in the nonlinear response of such
members to a periodic excitation is discussed.

INTRODUCTION

Few authors have addressed problems associated with either nonlinear

modal coupling in beams or nonlinear non-planar (with torsion) dynamic
behavior of beams, or beam-like structures, or both. Such problems can be
of primordial importance when, for example, one is dealing with large beam-
like space-structure members. In this paper, nonlinear modal coupling
phenomena in inextensional beams, due to geometric nonlinearities, are
addressed. Due to space limitations, only a brief overview is presented.
A detailed analysis of such motions is presented in [1].

An analysis of the flexural-flexural single-mode response of a

cantilever subjected to a base excitation was first presented in [2].
The analysis presented in [2] was based on a set of differential equations
of motion where nonlinear contributions from curvature and torsion were
neglected a priori. A stability analysis of the motion was also performed
in [21 based on a set of linearized differential equations with periodic
coefficients. The problem considered in [2] was also addressed in [3] by
making use of a consistent set of differential equations of motion that
take into account all the geometric nonlinearities in the system [4,51.
In this paper, modal interactions in the response of inextensional
structural beam-like elements are addressed. The differential equations
of motion developed in [4] are applied to a beam with one end clamped and
subjected to a periodic excitation that is either distributed or applied
at one or at different points along its span. As in [51, we let Lv(s,T)
and Lw(s,T) denote the bending deflections along two orthogonal inertial
directions y and z, and y(s,T) denote the angle of twist of the beam.
Before deformation, the principal axes along the beam's cross section at
s-s are aligned with the x and y directions. Here s denotes arc-lenfgth
along the inextensional beam, normalized by the beam's length L, and T
denotes time. As shown in [1], the differential equations governing the
flexural-flexural-torsional dynamics of an inextensional beam clamped at
I
I
s -0 and with principal bending stiffnesses EI,(s) and EI(s), torsional
stiffness GJ(s), torsional distributed mass moment of inertia jE(s) and
distributed mass m*(s), can be written as

1.V+WII
m; + C,+
~ II'I
(ByV")
mv~cv.-kyv 1
,W"

+
I (B

-
- 8y)v"w"ds - [(Sn-By)W"]

By)W" [ - (B
II fr

- By)VW
f v w'ds -v'[By(V'V+'"]

]ds
ds

f- (v'2 + w'2)d ds-w" f vids-u g +QV(S)COS(aVt Tv)

mi + C +(Brwol
t
S
V1 f(an,-8y)vllwlds - ((B,,-8S)V"]'
o]
f v'W"ds + l8nIV +W@)

3L(a -

4-JIw
8n -

'
)v f [ik- (8,n- B)v"w"]ds d
0 8~1
-wdsIsv
S

(vit+w'2)d ds- v" I uds+u "g +Qw(s)cos(awC+ Tw)

2 1 Lo ' J]
I (lb)
Li- (3YY') - (Bn -B)V"W" - Qycos(syt +TY) (Ic)

where g = + '"*- v 1w11d T

U
In equations (la-c), ( )' )/s,
/3( ( )'a( )/at, Bn=EIn/Dno,
3 8y =
EII /D O,
By -GJ/DTO, 2
ij- jE/(moL ) and tTVDIn 0/m0L4), where

Dno=f E1 (s)ds, m0 f m*(s)ds and m(s) m*(s)/m 0 ; the parameter c is

0 0
a normalized viscous damping coefficient. Also, Qa(s)cos(i1t+Ta), for

3 = v and w, are the components of the external excitation along the

inertial y and z directions, respectively, and Qy(S)COS(ayt+Ty) is an
externally applied moment. Here we will consider the case where Qw = Q -0.
The small effects of shear [6] and of the distributed mass moments of
inertia on the bending deflections are neglected in these equations.

NONLINEAR MODAL INTERACTIONS

To analyze the response of the system, we first introduce an arbitrary

perturbation parameter E, which is used for "bookkeeping" purposes only,
I
3
 n
three time scales ti - cit(i = 0,1 ,2), and let a - [ Fai(s)ati(t) for
i-I.
- v,w,y. Here the functions Fai are chosen to be the eigenfunctions
associated with the linearized counterpart of equations (la-c). The
temporal part of the solution of the linearized, 0(c), differential equa-
tions is obtained as

ati " Aa (tit 2) coswai t 0 +Bai(tlst 2)] ; cz-v,w,Y (2)

When the solution to the linearized equations is substituted into the

(ei), i> 1, differential equations, a number of resonances between the
bending-bending and bending torsion motions are identified. The steady-
state response (and the stability of the perturbed motion about that
steady state) associated with each resonant motion is then obtained from
the solvability conditions extracted from those differential equations.
The amplitude-frequency response characteristics for the structure are
readily obtained from the solvability conditions.

The bimodal amplitude-frequency response curve for a fixed-free

homogeneous beam with torsional natural frequencies much higher than its
bending natural frequencies is shown in Figure I.. The response shown is
1
for a planar motion with c-0.002 and q-f Fvi(s)Qv(s)ds-0.08, for i3.
0

Av 2 :...: -. 062

v" [-:yAv 3

..........
unstable Av2

'--....008

9 ","'-.. .004

-.03 -.02 -.01 0 .01

Slv/wv3 1

Figure 1. Amplitude (equilibrium) - frequency response curve for a

clamped-free beam [(1): single mode response (Av2) -01. 1
I
i
I

U Figure 2 shows the amplitude response curves for a clamped-pinned/sliding

beam with the same value of c and q (with i2) given above. For the
cantilever beam, the bimodal response exhibits modal interactions between
the second and third linear modes; for the clamped-pinned/sliding beam the
modal interaction is between the first and second modes. Nonlinear modal
interactions in extensional beams were investigated by Nayfeh, Mook and
their co-investigators [7-91. For extensional beams, the main nonlinearity
in the equations of motion is due to mid-surface stretching of the beam,
and this nonlinearity is always of the hardening type [10].

.10

I AvI
-- stable
I unstable
Av2
or

.05 Avl

Av2 ...

-.08 -.04
Qlwv2- 1

1Figure 2. Amplitude (equilibrium) - frequency response for a clamped-

pinned/sliding beam [(l): single mode response (Av -0)].

A number of experiments are being conducted by the authors to generate

laboratory data that can be used to corroborate results of their various
analytical investigations. These experiments involve beams with constant
and variable cross sections, and long frames of the type to be employed in
space structures. The results of these experiments will be reported in
3 the near future.

ACKNOWLEDGMENTS

3 The work presented here was sponsored, in part, by the U.S. Air Force
Contract No. F33615-87-C-3255 with the Flight Dynamics Laboratory, Air
Force Wright Aeronautical Laboratories (AFSC).

I
 REFERENCES I
1. Crespo da Silva, M.R.M., and Zaretzky, C., Nonlinear Modal Coupling in
the flexural-flexural and flexural-flexural-torsional motions of
inextensional beams. In preparation.
I
2. Haight, E.C., King, W.W., Stability of nonlinear oscillations of an
elastic rod. J. Acoust. Soc. Am., Vol. 52, pp. 899-911 (1971).

3. Crespo da Silva, M.R.M., On the whirling of a base-excited cantilever

beam. J. Acoust. Soc. Am., Vol. 67, pp. 704-707 (1980).

4. Crespo da Silva, M.R.M., and Glynn, C.C., Nonlinear flexural-flexural-

torsional dynamics of inextensional beams. I. Equations of motion,
J. Struct. Mech., Vol. 6, pp. 437-448 (1978).

5. Crespo da Silva, M.R.M., and Glynn, C.C., Nonli-ear £lexural-flexural-

torsional dynamics of inextensional beams. II. Forced Motions, J.
Struct. Mech., Vol. 6, pp. 449-461 (1978).

6. Timoshenko, S., Young, D.H., and Weaver, W., Jr., Vibration Problems
in Engineering, Wiley, New York (1974).

7. Nayfeh, A.H., Mook, D.T., and Sridhar, S., Nonlinear Analysis of the
Forced Response of Structural Elements, J. Acoust. Soc. Am., Vol. 55,
3
pp. 281-291 (1974).

8. Nayfeh, A.H., Mook, D.T., and Lobitz, D.W., Numerical perturbation I

method for the nonlinear analysis of structural vibrations, ALAA J.,
Vol. 12, pp. 1222-1228 (1974). 3
9. Nayfeh, A.H., Mook, D.T., and Nayfeh, J.F., Some aspects of the modal
interactions in the response of beams, Paper No. 87-07777, AIAA Struc-
tures, Structural Dynamics and Materials Conference, Monterey, CA,
April 6-8, 1987.

10. Crespo da Silva, M.R.M., Nonlinear flexural-flexural-torsional-

extensional dynamics of beams. II. Response analysis. To appear in
lnt. J. Solids and Struct.
I
I
I
I
I
I
UMODAL INTERACTIONS IN THE RESPONSE OF BEAMS TO A HARMONIC EXCITATION
J. F. Nayfeh, A. H. Nayfeh, and D. T. Mook
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061

I In this paper, we investigate some aspects of modal interactions in

the response of beams. Specifically, we present a nonlinear analysis of
the response of a hinged-clamped beam to a simple-harmonic excitation.
For such a beam, the second natural frequency is approximately three
times the first, leading to a state of internal (autoparametric)
resonance which produces a strong coupling of these modes. This
coupling is responsible for unusual responses that cannot be predicted
by single-mode nonlinear analyses or by multi-mode linear analyses.

For a comprehensive review of the nonlinear response of beams to

single- and multi-frequency excitations, we refer the reader to the
textbook by Nayfeh and Mookl.

A comInation of geometric and material nonlinearitiers is

considered . The ends of the beam are immovable and a large lateral
deformation or large amplitude vibration prduces stretching of the
median line of the beam. The strain-displacement relation becomes
nonlinear and is of the form
0o ~ 2
e = ux
2
Wx (1)
In addition, the beam is assumed to be made of a nonlinear material.
The Ramberg-Osgood material with cubic nonlinearity is used and the
5 stress-strain relation is given by

ax = Ae - Be3 (2)
3 where A and B are the material constants. Neglecting the longitudinal
and rotary inertia and the Zransverse shear, we find that the transverse
deflection w(x,t) is governed by
Sw,tt=+
A
W't - W'XXXX = px
w,t + a lw, (w,x) 2dx + 2a2w, (w,xxx2
I XX 0 'XX
0

+ a ) W,xxx + Q(x~t) (3)

where
1= Abh
2p , -2 3Bbh
400(4 (4)

Here P is the mass density of the beam, b, h and I are the width, height
and moment of inertia of the cross section, c is the damping
coefficient, and Q is the applied transverse load.
i The linear free-vibration solution is

w(x,t) = O(x) cos(Wt + 8) (5)

where

I
 X
b[sln(slx) - i sL nh(s2X ) (6)
2 in
D2 2 si n S2L)sn~~)
s= (0(fl)) (7)

s2tan(s 1 L) - sltanh(s 2 L) = 0
and b is chosen so that
(8)
I
S*2dx =L (9)
Next, we expand the solution of Eq. (3) in terms of the free
undamped linear modes as
w(x,t) = ! vm(t)*m(x) (10)
n=1
Using the Galerkin procedure and assuming modal damping, we obtain the
nondimensional temporal equations of motion
d vn n-
(11)
dt +nVn= 2eund- m, , rnmkzVmVkVz + eFn(t )
where

CCnn p 0 Cn(X)n
no (x)dx (12)

(t) 11QXt*(=x(3

n n0 0

+
L
=2[f;o.
L
dx][f oj- -dx (14) 3
2 ep ccn
nn' F n (15)

Ti
and e is a small parameter, which is introduced as a bookkeeping device
I
and will be set equal to unity in the final analysis.

Method of Analysis m
We use the method of multiple scales 3 to determine a uniform first-
order expansion of the solutions of Eq. (11) when Fn(t) is harmonic;
that is,
Fn (t) = fnCOSat (16) 5
and a is near a2- To express quantitatively the nearness of the primary
and internal resonances, we introduce the detuning parameters al and
a2 defined according to

I
 3a1 + Cal and a - a2 + e02
0= a (17)
Applying the method of multiple scales, we obtain the modulation
equations
- _ -2 1a1T1
21(A + .IA )
1 - 8A 1 YJmAmAm -8 A 2Ale =0 (18)
n
- 3 -iaoT, ia2T
21 2(A,+ i2A2) - 8 2A2 Y2 A A - 8Q2 Aae - f2 e = 0 (19)
n
where
8o,6I = 3r,, 2, 8a262 = r 11 2 (20a)

8 mymj z2(rmmij + 2r mjmj), m * J, 80mYmm =3rm10m (20b)

To analyze the solutions of Eqs. (18) and (19), we let
A = 1 (p, - iql)ei ' and A2 = 1 (P2 - iq2 )e (21)

where
V, = 4 (a2 + a,) and v 2 = G2 (22)
Substituting Eqs. (21) and (22) into Eqs. (16) and (17) and separating
real and imaginary parts, we obtain the following autonomous equations
describing the modulation in amplitude and phase:
p+ ulp + vliql+ y~q( + q2) + yj2q 2p + q2
2 2
+ 61q 2 (P, - q1 ) - 2ipiqP2 = 0 (23)
qlq + - P Y zP ( 2 2 (2 2)
p2I + 4i1q, - v~2q2+ yl 2(pi + qI) - Y12P( 2 + q2)
2 2 =
6P 2 (Pq - q) - 2lplqlq2 0 (24)
P + U2p2 + + y2jq2(p1 +q + 22q2(p2 + 2
p ql - qj)
2( =0(25)

q+ u 2q 2 -V
2p 2- Y 2 1 p 2 (p2 + q 2) - y 22 p2 (p' + q 2)

- 62 (p1 - 3pqI) - f 2 /202 = 0 (26)

Next, we present numerical results for the case = 15.418, W? =

49.965, a = 3.241, a, = 3.710, VI = U2 = .01, 3 02y,,/61 = 3 a2 Y 2 2 /61 -

=1, 316 = 02 and f2 = 1.

Periodic solutions of w correspond to the fixed points of Eqs.
(23)-(26). There are two possibilities. First, p, = q, = a,= 0, and
 IF 2
0 2 "- Y2 2 a2 ± [ 2 - U2 (27)
42a 2

where4a - 1
2 2 2 2 2 2
a, = p, + q1, a 2 = P2 + q2 , = sin-q/a
2/ 2)

and 81 - 3sin-l(ql/a,) - 82 (28)

The single-mode nonlinear solution is shown in Figures 1 and 2. Second,
p1 * 0 and q, * 0.
The stability of the fixed points is determined by the eigenvalues
of the Jacobian matrix of the flow governed by (23)-(26). If all the
etgenvalues have negative real parts, the fixed point is stable,
otherwise it is not stable. Hopf bifurcation occurs when the real part
of a complex conjugate pair of eigenvalues changes sign from negative to
positive. In these ranges, steady state periodic solutions do not exist
contrary to results predicted by linear multi-mode analyses or nonlinear
single-mode analyses. Instead, the energy is continuously exchanged
between the modes involved in the autoparametric resonance (Figure 3).
Moreover, for small damping, the response experiences period-multiplying
bifurcations and chaos. The stability of the periodic solutions is
determined using Floquet theory. One of the Floquet multipliers is
always unity because the system is autonomous. The unit circle is =
traversed at -1 giving a supercritical subharmonlc bifurcation (Figures
4 and 5). An infinite sequence of period doubling takes place
culmigating in chaos (Figure 6). The mechanism is reported by Tonsi and
Bajaj4 in their studies of mechanical systems with cubic nonlinearities.
Acknowledgement 3
This work was supported by the Air Force Office of Scientific
Research under Grant # AFOSR-86-0090.

References
1. Nayfeh, A. H. and Mook, D. T., Nonlinear Oscillations, Wiley-
Interscience, New York, 1979. -
2. Besseling, J. F., Ernst, L. J., et al., "Geometrical and Physical
Nonlinearities: Some Developments in the Netherlands," Computer
Methods Appl. Mech. Engrg., 17/18 (1), pp. 131-157, 1979.
I
3. Nayfeh, A. H., Perturbation Methods, Wiley-Interscience, New York,
1973.
4. Tonsi, S. and Bajaj, A. K., "Period-Doubling Bifurcations and
Modulated Motions in Forced Mechanical Systems," ASME Journal of
Applied Mechanics, Vol. 52, pp. 446-452, 1985.
 02 a2

I
Figure 1. Single-mode nonlinear analysis. Left, effect of changing
amplitude of excitation; right effect of changing damping.

I 002 f-f n02

Figure 2. Single-mode nonlinear analysis. Left, effect of changing

I2 ; right effect of changing damping.

I
* 0~2
I Figure 3. Multi-mode nonlinear analysis.

I
Figure 4. A 3-0 projection of the limit cycle attractor for a2 =
.3805. The needles show the 2-0 projection of the attractor
on the plane a2 -el"

I
I ! I

Figure 5. A 3-0 projection of the period-doubled attractor at a =

.3908. The needles show the 2-0 projection of the attractor
on the a2 - .
 TIM
02
02 A

U 01 TIME

(02 02

* 01 TIME

(32 (32

01 TIME

UFigure 6. Phase diagrams of period-doubling bifurcations leading to

3 chaos on the right and their corresponding wave forms on the
left. From top to bottom a2 = (.3805, .3908, .3952, .4036).
 ANALYSIS AND DESIGN SENSITIVITY CALCULATION OF
NONLINEAR FORCED VIBRATIONS OF STEPPED BEAMS*

J. W. Hou, Y. X. Xue, C. 4ei and C. Jackson

Department of Mechanical Engineering and Mechanics
Old Dominion University
Norfolk, VA 23529-0247

I. INTRODUCTION

In many engineering design textbooks, the failure criteria of fatigue are

usually expressea in terms of stresses obtained by using linear vibration
theory. However, nowadays many light-weight structures often experience large
amplitude vibration. The effects of such large amplitude can not be ignored
in designing structures against fatigue. Figure 1, in which the amplitude of
the vibrating beam is more than half of its thickness, displays the relations
between the number of cycles and the stresses calculated by linear and
nonlinear vibration theories. It indicates that if the frequency of the
harmonically excited force is lower than that of the natural frequency of the
system, the stress obtained by using a linear vibration theory is higher than
one obtained by using a nonlinear theory. Likewise, if the frequency of the
excited force is greater than the natural frequency, the stress obtained by
using a linear vibration theory is lower than one obtained by using a
nonlinear theory. If the frequency of the excited force is greater than the
natural frequency, the stress obtained by using a linear vibration theory is
lower than one obtained by using a nonlinear theory. In other words, based
upon the linear vibration theory, the structural or mechanical system can
either be overdesigned or underdesigned depending upon the relation between
the input force frequency and the natural frequency.

In order to consider this large amplitude in an automatic design

environment, research efforts have been focused upon the enhancement of
analysis capabilities and the development of design sensitivity analysis
techniques. Some progress in this regard is reported in the following
sections.

II. FINITE ELEMENT ANALYSIS OF NONLINEAR FORCED VIBRATION

There are many methods available to analyze responses of beams under

nonlinear forced vibration. Because of its versatility and practical
applications, the study will concentrate on the finite element approach
proposed in reference [1]. It is initially used to find the response of a
uniform beam, and is then modified to solve the responses of a stepped beam.

This was partly supported by NSF Grant No. DIC-8657917 and NASA-Langley
Grant No. NGT-70020.
i
I

am e' ~ ~~~~~~~24000
-iJ
J l J.li l l l lj j ll ff~

Ii ; '3"20 0 . .l-
.- ---

I 2*)l
111
1*; 1l0ll0l8
0l '0 4 6

CYCL::X 1 00 000) CYCL l x .000.000

Fig. la Displacements of Beams Under Fig. lb Stresses of Beams Under

Force,, Vibrations Forced Vibrations

UNIFORM BEAM1
3 The Lacirahgian of a uniform beam under harmonic excitation f(x,t) can be
expressed as

L f 0 7 [E w, 2 + EA
EA (u, + 1 w x] dx- I L
0I m2~l dxm6
l ll

L
I Ii
- f f(x,t) wdx
0

3 After linearizing the strain energy, it can be discretized and approximated by

finitacle1 ment matrix equations:
[K] (x) + [G(lx})] {x} - EH(tx})] [x} - x [M] (xl = 0(1

where [G([xI)J corresponds to the nonlinear part of the strain energy and
[H({xl)J is the harmonic force matrix which is proportional to a constant
3B 0
. The term [H((x})J (x} is a linear spring force which is an approximation
of the harmonic excitation. The mathematical justification of the above
statement can be found in Hsu's work [2]. The spring constant B can be
derived as
B CF 0 Z2 2
B 2 c = ft dx/ f dx (2)

where y is the maximum amplitude, is the normalized mode shape, WL is the

linear fundamental eigenvalue, and F0 cos Vxt is the harmonic force.
 Equation I represents n individual equations. The possible unknowns in
that equation, however, can be the mode shape (x} with n components, the
frequency, and the force density of the harmonic excitation. With different
combinations of given and unknown variables, equation 1 can be used to
simulate many practical applications in which either the mode shape, the force
density, or the frequency of an excitation can be considered as the primary
unknown.

As one practical example, it is assumed that the maximum amplitude of the

mode shape and the force density of the harmonic excitation are given. In
this case, with the specified maximum amplitude and the assumed mode shape,
equation 1 is solved iteratively as a linear eigenvalue problem for the mode
shape and the frequency of the harmonic excitation [1]. The same solution
procedure can also be used to solve a nonlinear vibration problem in which the
frequency of the excitation and the maximum amplitude are determined as a
priori with the harmonic excitation force density and the mode shape
considered as unknowns.

Another important application, from a designer's point of view, is one in

which the harmonic excitation force is completely specified, that is both the
frequency and the force density are known; and the mode shape is the only
unknown. In this case, equation 1 becomes a set of n nonlinear equations
which may be solved by using a root-finding algorithm. However, proving the
existence and uniqueness of the solution is a very subtle issue and has not
yet been solved.

STEPPED BEAM

Not much work has been done in dealing with nonlinear forced vibration of
stepped beams. A method based upon Eq. I has been proposed in reference [3]
in which the spring constant, Bi , is evaluated for each individual beam
element, instead of being evaluated over the entire length of the bem. The
purpose of this modification is to allow the spring constant, B., to be
different for different elements according to the geometric properties of each
element and the actual load applied onto each element. Figure 2o shows the
difference betwten the new and the original definitions of B . The new
formulation of B i defined over the ith beam element is given as

o f -F * dx F
Foi
B. = "_ Ti
fz 1 o dx mp

where Foi is the force density applied 0to the ith beam element. The matrix
equation 1 with the new definition of B can be applied to beams with stepped
cross sections and to beams subjected to nonuniformly distributed harmonic
loads.

For nonlinear vibrations of a uniform beam, a maximum difference of 1.2%

is observed between 0the eigensolutions calculated by the proposed fnite
element method with B i and those calculated by the original method with B . As
I
I
for a stepoed beam, there is no finite element solution available currently.
However, numerical comparison indicates that the nonlinear eigenvalue solution
obtained by the proposed finite element method is at most 1.5% higher than
that obtained by the classical solution.

In short, numerical experience seems to confirm that the proposed finite

element formulation for a nonuniform beam is valid.

II1. NONLINEAR EIGENVALUE-EIGENVECTOR DESIGN DERIVATIVES

The field of design optimization has gone through a period of extensive

development over the last two decades. One fruitful area of such research
efforts is design sensitivity analysis. It has been recognized that design
sensitivity analysis is capable of approximation analysis, analytical model
improvement, and assessment of design trends. Thus, design sensitivity
analysis has become more than a utility as an optimization design tool in its
own right.

There are many valid approaches for determining the design derivatives of
eigensolutions. Nevertheless, the computational procedure presented hereafter
is an extension of the one reported in reference [4].
For simplicity, the coefficient matrices of finite element solutions of
the nonlinear forced vibration can be written as

[A] {x} = 0 (3)

where [A] = [K] + [GI - [H] -x [M]. Premultiplying eigenvector {x} T to the
above equation, one immediately has the following equality

3 {x}T [A] {x} =0. (4)

There are n+1 unknowns in the last two matrix equations that can be used to
find n+1 design derivatives of {x} and x. Let the subscript b denote the
design derivative and [E] denote the derivative of the vector [G-H] {x}, in
which {x} is held fixed, with respect to the eigenvector. With some
manipulations, it is straightforward to obtain the design derivatives of Eqs.
3 and 4 as

1 b x}T [AT]X} + {x}T[E]{X (5)

x [TIM] {x} x b

i and
(] [ M]{x}{x} T[E] bm]X
([A] + [E] L T {Xb [A b]{X} + T {x}T[A ](x}
(xI [M]{x} (x} [11] {x

I
 I
or symbolically, U
[B]{xb} = {c' (6)

where [B] and {c} are defined according to the last equation. In contrast to
the linear eigenvalue problem, as indicated in Eq. 5, the design derivative of
eigenvalue, Ab' can not be calculated without knowing the desin derivative
of eigenvector {xbI in advance. Because {x} [B]{Xb}=O and {x} {c}=O, it can
be also concluded that [B] in Eq. 6 is singular based on the alternative -
theorem.

In order to avoid the singularity of [B] one can express the eigenvector
design derivative {x } as {xb } = (x} + c{xl, where the unknown variables
I
are {R} and the coeff~cient 6. The (R} can be attained by the following
matrix equations ([A] + [E = [Ab]Xl - a [E]{xl (7)

(RIT [?]{xl 0 (8) m

where the constraint Eq. 8 restricts {i} to be orthogonal to {x}. Applying
the theorem of Lagrange multipliers to the above equations yields n+1
simultaneous equations for {x} and u

E[A]
{xT
T+ E]
[M]
[MIX}
0
tI}
u
~ [A b] {x?
0
I[E] (x1
0
(9)I

The leadina coefficient matrix on the left side of the above equation is
nonsymmetric, and there is an undetermined coefficient, a, appearing on the
riaht side. Since the above equation is linear, one may superpose the
solutions as

u u a I
where the first and the second terms are the solutions of Eq. 9 with
T T T T
(([Ab]{x})T, 0}0 and {([EJ{x}) , 0}, as the forcing terms, respectively. The
coefficient, a, is usually determined by considering the normalization of the
corresponding eigenvector. For example, the maximum amplitude is often fixed
in the formation of nonlinear forced vibration. In this case,

il (Xo)
x2 (x) + X(Xo
0 )

where xo is the position of the maximum amplitude.

I
U The proposed computational scheme for design sensitivity analysis has
been validated by several numerical examples. Table 1 tabulates the
comparisons between the predicted and actual changes of the lateral deflection
I measured at a location 40 inches away from the center of the beam. It shows
that the change predicted by the first order design sensitivity wb.Ab can be
used to approximate the actual change Aw. It is also important to note that,
as indicated in the last row of Table 1, che design sensitivity analysis takes
only one-seventh of the CPU time required for the direct analysis.

Reel Caese =,* B"e-,.,._ ,,,_

F, '00-diA

I Fig. 2 Differences between two definitions of B

IV. CONCLUSIONS AND REMARKS

This research has achieved the following goals:

1) Modification of the harmonic force matrix of an existing finite element
method in order to treat steppe beams under nonlinear forced vibration.
2) Design sensitivity analysis of nonlinear vibration of stepped beams where
the effects of both longitudinal displacement and inertia are included.

As mentioned before, many physical behaviors pertaining to nonlinear

vibrations can not be realized by using the linear theory. The impact of
these new physical behaviors on design considerations should be investigated
carefully. Thus, it is important not only to enhance the nonlinear vibration
formulations and analysis capability, but also to emphasize the research in
the design sensitivity analysis of nonlinear eigensolutions. However, further
effGrts are needed in order to achieve the research goal.

I REFERENCE

1. tei, C. and Decha-llmphai, K., "A Finite Element Method for Nonlinear
Forced Vibrations of Rectangular Plates," AIAA Journal, Vol. 23, 1985, pp.
1104-1110.
2. Hsu, C. S., "On the application of Elliptic Functions in Nonlinear Forced
Oscillations," nuarterly of Applied Mathematics, Vol. 17, 1960, pp. 393-407.
3. Xue, Y. X., Analysis and Design Sensitivity Computations of Nonlinear Free
and Forced Vibrations of Stepped Beams, M.S. thesis, Old Dominion University,
NorFolk, VA, December 1987.
4. Hou, J. W., and Yuan, J. Z., "Calculation of Eicenvalue and Eigenvecter
Derivatives for Nonlinear Beam Vibrations," AIAA paper No. 86-0963, presented
at the 27th SDM conference, San Antonio, TX, May 19-21, 1986.

I
 4-) 4.J 4m M . -n m 'O m
4L M M P~
'0 n n-4 PC w
SL -4 Y-.- = M M ON m.

Q C-.

>0
0
Q3- C 4 V-40 0 00C Rr .
- CD C) n
L.- in W4 #s CD
4= U-) 0 00 (n CN M m

CL .C 0 0 0 0 0

L. a) I I0

'00)C

goo
Cu

0)- 00 ("1
in i ) -''0
m In 0
-o 0 n (% Oi -4 a~ C) CO
041 0M~ C00 0 0CC
4-~ 0

0 > (0

4.V CC '

> 00

040( -4 -4 -% 4)1

Ic0 I Id (

C (a- L. 0CC) LC
'0~L cn'

LA0 0)C
4w a ai
C~~~C ' . CC m

.6-)In~0 C) cM i n 0 n
Mni i _
Co 0') - l- l - -l (NJ 4-
'n CIr N -KI T m .4. -- 1.) 4.3* I
r- 0)0 M0
0)4- 4-'C' 0
I
I
Non Linear Oscillations and Limit Cycles: A Time Finite
* Element Approach
M. Borri
Center for the Advancement of Computational Mechanics
Georgia Institute of Technology, Atlanta, GA 30332
and
P. Mantegazza
Dipartimento di Ingegneria Aerospaziale
Politecnico di Milano
Via Golgi 40 - 20133 Milano - ITALY

3 1 Introduction
The importance of non linear oscillations in the study of vibration problems is well known to
engineers and scientists. Several alternative analytical and numerical methods have been developed
to solve these kind of problems [1-2].
These methods often require a lot of manual work even when they are applied to systems of a
few degrees of freedom and are therefore too cumbersome for analyzing systems with many degrees
of freedom.
The aim of the present paper is to show how the finite element approach in the time domain
when applied to Hamilton's principle can provide a fully automated numerical scheme for solving
problems of this type.
During the last decade much attention has been given to the numerical applications of Hamil-
ton's principle, often called Hamilton's law of varying action, promoting also vigorous discussions
about this subject [3-11]. The use of Hamilton's principle to solve periodic non-autonomous prob-
lImes has been proposed in Ref. [12], [13].
The present paper deals with more general periodic non linear problems in which the period is
not known in advance. Finite elements in the time domain applied to Hamilton's principle provide
a suitable numerical approach.
[t is believed that the approach proposed here can be used for a wider class of problem than
that shown. It is sufficient to remark that the method can be applied to initial and final value
problems and that the linear stability can be analyzed by the eigenvalue analysis of the same
matrices computed during the search for the solution [10, 12].
Moreover it is important to note that the proposed method is nothing but a suitable application
of the virtual work principle and of finite element method to dynamics, so that the well known
numerical algorithms and techniques of the finite element method can be profitably used.
In order to simplify the presentation, the method is formulated only for holonomic systems.

I Non-holonomic constraints could be accounted for in a natural way [16].

I
!1
 !
I
2 Hamilton's Principle and Finite Element Approximation
Let us consider a holonomic system with n degrees of freedom. At any given time t, the position I
of a representative point P in the configuration space is described by the set of n coordinates q,
i.e. q = (qlq2, *'" q.) and let r be any trajectory of the system with equation r = F(u) where u is
an aribitrary independent parameter that assumes the role of independent variable.
Let L(q, 4, t) be the ordinary Lagrangian function of the system, and S the Lagrangian action
along any oriented curve r in the configuration space, drawn from the point P (where u = ul and
t = t1) to the point P2 (where u = u 2 > ul and t = t2 > t 1 ) i.e. I
s(r) = ftL(q, 4,t) dt

In addition to the conservative forces, whose potential is accounted for in the Lagrangian of the
system, we will consider also a non-conservative force vector Q for which the virtual work is 6,q"Q
and the virtual action is given by I: S
6A =f 6q" Qdt

Hamilton's principle can be written as:

6S + 6A - 6q"-pl = 0 (1)
In view of the numerical applications, the boundary terms are of vital importance and cannot be
dropped by constraining the virtual displacement to zero at the end points P1 , A. Moreover the
generalized momenta p cannnot be overspecified by constraining them to be equal to since,
their definition isnaturally included in the principle.
This principle is very suitable for numerical approximation using the finite element approach.
To this end we subdivide the interval U2 - u1 into consecutive non overlapping subintervals. We
define with u = (ul, ,-.. , uN+l) the values of the parameter at the nodal points. Without loe
of generality we assume ul = 0 and uNV+1 = 1.
In the numerical formulation of the problem we assume the Lagrangian coordinate vector q
ranging in some limited class of admissible functions f(u) and we set:

q(u) = Efk(u)qk (2)

where Sqk = q(uk) k = 1,2,... N + 1

Moreover in order to preserve the orientation of time t with the parameter u we assume:

t = t, + Tu (3)
where T > 0 is the time interval t 2 - fI.
For notational reasons and for sake of conciseness we define the following vectors X and Y.

X -"(qt,q2,, •qN+) Y = (q1 , q 2,'" q. +, t, t 2 ) (4)

So Eq. 2 can be shortly rewritten as:

q =f.X (5)I
'here and in the following 6.denotes the corotational virtual change

2
I
I
and the collection of Eq.. 2 and 3 is briefly identified by:

r=g-Y (6)

With the present approximation the functional S(F) turns out to be a function of the vector Y
and the virtual action related to external forces Q can be expressed in the following form:

&A = 6,X . f. Qdt =6X . A (7)

where A represents the generalized impulses corresponding to the forces Q. The trailing terms of
the variational principle, can be also changed into the following:

beq"PI =6X.B (8)

~where
w
B = (-p1, 0,O,.0,p N+)
and the variational principle leads to the following:

+ A -B =0 (9)

These equations constitute an approximate functional relation onto which the integral curve r
joining the points P 1 and P 2 must fit. They further constitute the approximate parallel of Hamil-

3 ton's partial differential equations. Obviously to solve a particular dynamic problem the initial or
boundary conditions must also be specified. When applied to autonomous periodic problems, we
enforce qv+l = q, and BN+I = BI.
Morover, we can set to zero the value of the time instant ti so that t 2 = T, and we can also
prescribe the value of one component of the vector qI, then Eq. (6) can be solved for the remaining
components of q and t2 .
Since Eq. (6) are nonlinear, their practical implementation requires the use of a nonlinear
algebric equations solver. We select the use of Newton-Raphson method that, in addition to its
property of quadratic convergence within the attraction domain, implies the use of a consistent
linearization of the equations that can be profitably used for Floquet's stability analysis of the
solution [11-12].
The preceding developments have been verified with a few simple examples. These examples
were related to systems with one or two degrees of freedom chosen with the aim of verifying the basic
concepts and the feasibility of the numerical approximations but without taking care of numerical
efficiency.
The first example refers to a self excited system governed by the equation:

I j = .2(1 - 242)4 - q+ q3 =0 (10)

showing, a stable limit cycle enclosing all the critical points q = -1, q = 0, q = +1.
The harmonic balance method gives the approximation, q= 1.35sin(.605t). The present ap-
proach, with the period subdivided into eight four node elements, gives a limit cycle period
T = 2x/.636 and a peak value qp = 1.46; the contribution of higher harmonics is significant.
The second example refers to a single degree of freedom system possessing two limit cycles and
governed by the equation:
I 4+(.3- .02q2 + .0001q 4)4q 0 (11)

3
 0.'I"

0.0

0.0;

0.0;

0.06

4 237 3, 0 44

Vw..-.11 (as/',

Figure 1: Bending-torsion flutter boundary 0 - FEM Ref. [171

The first harmonic balance approximation gives two limit cycles both possessing the same
circular frequency f] = 1. The first is unstable and has an approximate amplitude equal to 8.514
while the second is stable with a larger amplitude equal to 17.368. The resulting limit cycle periods
obtained with the same finite element mesh as the previous example are respectively 6.322 and
6.392.
The third example refers to the bending torsion flutter of a two dimensional airfoil in incom-
pressible flow and is taken from Ref. (171. The system restoring force is linear in translation and
has the non linear torsional stiffness, with backlash.
Figure 1 shows the results obtained with the present method compared with those of Ref. [17).
Despite the different aerodynamic approximation used the comparision shows an acceptable
matching with a more pronounced separatation in the large and small amplitude limit cycles.

3 Concluding Remarks
The paper has shown the extension of Hamilton's Principle to problems showing periodic solutions I
with unknown period. The present and companion works [101, [121 clearly show thepower of
Hamilton's Principle coupled with finite elements in time domain as a unifying tool in the numerical
solution of dynamic and response problems of mechanical systems.
In fact their coupling allows an easy set up of the approximating equations in linearized form
I
which is the basis for an iterative solution of the nonlinear problem and for the analysis of the
linear stability of periodic solutions of any type.
Nonetheless practical, efficient and robust numerical methods capable of dealing with multiple
I
solutions of nonlinear systems of equations remain to be developed in order to allow the present
method to be used for practical problems.

4I I
I
I 4 References
1. Schmidt, G. and Tondl, A., "Non-Linear Vibrations", Cambridge University Press, 1986.

2. Hagedorn, P., 'Non-Linear Oscillations", Clarendon Press, Oxford, 1981.

3. Baily, C.D., "A New Look at Hamilton's Principle", Foundations of Physics, Vol. 5, No. 3,
1975, pp. 433-451.
4. Simkins, T.E. "Unconstrained Variational Statements for Initial and Boundary-Value Prob-
lems" AIAA Journal, Vol. 16, No. 6, June 1978, pp. 559-563.

5 5. Smith, C.V., Jr., "Comment on 'Unconstrained Variational Statements for Initial and Boundary-
Value Problems'", AIAA Journal, Vol. 17, No. 1, Feb. 1979, pp. 126-127.

6. Simkins, T.E., "Finite Elements for Initial Value Problmes in Dynamics, AIAA Journal,
Vol. 19, No. 10, October 1981, pp. 1357-1362.

7. Baruch, M. and Riff, R., "Hamilton's Principle, Hamilton's Law - 6n Correct Formulations',
AIAA Journal, Vol. 20, No. 5, May 1982, pp. 687-692.
1 8. Smith, C.V., Jr., "Comment on 'Hamilton's Principle, Hamilton's Law - 6n Correct Formu-
lations'", AIAA Journal, Vol. 22, No. 8, Aug. 1984, pp. 1181-1182.

3 9. Borri, M., Lanz, M., and Mantegazza, P., "Helicopter Rotor Dynamics by Finite Element
Time Discretization", L'Aerotecnica Missili e Spazio 60, pp. 193-200, 1981.
10. Borri, M., et al, "Dynamic Response of Mechanical Sy teris by a Weak Hamiltonian Formu-
lation", Computers and Structures, Vol. 20, No. 1-3, 1985, pp. 495-508.
11. Borri, M., Lanz, M., and Mantegazza, P., "Comment on 'Time Finite Element Discretization
of Hamilton's Law of Varying Action'", AIAA Journal, Vol. 23, No. 9, September 1985, pp.
1457-1458.
12. Borri, M., "Helicopter Rotor Dynamics by Finite Element Time Approximation", Special
Issue of Computers and Mathematics with Applications, Vol. 12A, No. 1, 1986.

13. Peters, David A. and Izadpanah, P., "hp-version Fiinite Elements for the Space-Time Do-
main", Computational Mechanics, Vol. 3, No. 2, 1988, pp. 73-88.
14. Agrawal, 0. P., "An Application of Modified Hamilton's Law Varying Action", The Pennsyl-
vania State University, October 7-9, 1985.
15. Synge, J. L., "Principles of Classical Mechanics and Field Theory", Encyclopedia of Physics,
Vol. III/1, Springer-Verlag Berlin Gottingen Heidelberg, 1960.
16. Borri, M., and Mantegazza, P., "Finite Time Elements Approximation of Dynamics of Non-
holonomic Systems", Presented at ASME Congress, Williamsburg, Virginia, August 1986.
17. Shen, S. F., "An Approximate Analysis of Non-Linear Flutter Problems", Journal of Aero/Space
Science, January 1959.

1 5
 I
Dynamics of Beams with Tip Masses and
Attached to a Moving Base
S. Hanagud'and S. Sarkart
I
Georgia Institute of Technology
Atlanta, Georgia-30332

Introduction
Studies involving the transient responses of beams attached to bases that are re-
stricted to translational and angular acceleration have many practical applications.
The results of such an analysis can be applied to spinning helicopter blades starting
from rest, rapid maneuverings of space based a i as which carry payloads from
one place to another, deployment of space stru, .-s and numerous other fields.
Even when such systems are physically stable and controllable, an inaccurate anal-
ysis could lead to a wrong conclusion of unstable motion. In a rotating beam,
contribution towards stability is provided by a nonlinear phenomenon known as
the centrifugal stiffening. The purpose of this paper is to account for some of the
significant nonlinearities in the analysis. The dynamic analysis has been performed
by using Kane's method because it leads to the final equations more directly as
compared to the Hamilton's principle or other virtual work methods.
An elementary analysis of the centrifugal stiffening has been given in Ref.(1).
The beam has been assumed to be inextensible and the virtual work done by the
centrifugal force field due to axial foreshortening has been considered in the potential
energy expression which leads to the centrifugal stiffening terms. This analysis has
been refined by Likins et.al2 by considering a steady state axial stretch term due
to centrifugal force. Vigneron s has a assumed foreshortening of the beam and has
shown that the centrifugal stiffening terms arise from the kinetic energy terms.
Simo 4- has taken all displacements to be measured from an inertial frame instead
of a rotating frame and has used finite strain measures to obtain the centrifugal
stiffening effect. In ad analyses mentioned so far, Hamilton's principle has been
used to derive the equations of motion. In Ref.(6), Kane.et.al have used Kane's
method to obtain the equations by considering stretch and two bending deflections
as the independent variables to describe the neutral axis but the stretch term has
not been applied in a consistent manner. In this paper, the same Kane's approach
has been used to derive the equations of motion but the stretch has been replaced by
the axial variable as the independent variable. Large deflections have been included
by taking into considerations all the significant nonlinear terms in the strain energy
terms. Numerical studies have been performed to study the effect of a tip mass on
the motion of the beam. The result has been compared with the case where no tip
mass is present. The response of the beam has been also studied for the case where
the base is subjected to a linear acceleration along the plane of the beam motion.
Finally, all the second and third order stiffening terms have been droppedI from the
equations and the resulting loss of stability has been observed.

Formulation i
The dynamical equations for the axial, bending, translational and rotational equa-
tions has been derived in this section. The deformation of the neutral axis has been
*Professor I
tGraduate Research Assistant

I
sketched in Fig.(1). The neutral axis lies parallel to the a, axis prior to deforma-
tion. The axial displacements ul and the bending displacements u 2 and u3 can be
written as functions of x. In Ref.(6), the basic kinematical variables to determine
the neutral axis have been the stretch and the bending displacements. They are
written in terms of chosen modal functions. The axial and transverse speeds and
the corresponding partial velocities have been obtained in terms of the independent
variables by using an expression for stretch as given below.

X,+ s(x 1 ,t) = [1 + f. + f12"/d, (1)

The stretch given by Eq.(1) will be correct if the bending displacements iL2 and u3
are expressed in terms of a variable which is the projection of the deformed neutral
line on the al axis, denoted by o, or deformed coordinate. In Ref.(7), Kane has
clarified that it is indeed so. However in all subsequent derivations the bending
displacements have been written as function of the undeformed x coordinate, in
which case, the stretch should be written as below

s~l
X, +zl f) (1 +Ul)2 +2 U2 .++ U23.1/2dX
S(Xi) (2)

Both the stretch expressions given in Eq.(1) and Eq.(2) are correct, they appear
different because the effect of axial displacement has been taken into account in
Eq.(1) in an implicit manner. However, care must be exercized in choosing the
proper independent variable.
In this section dynamical equations have been derived for the independent trans-
verse and axial displacements which are expressed as functions of undeformed co-
ordinate x. The mass and moments of inertia of the base have been assumed to be
zero. The cross-sections have been assumed to remain normal after deformation.
The effects of warping and rotary inertia have not been included in this formulation.
Since the displacements could be large, nonlinear strain-displacement relationships
have been used to determine the strain energy function of the beam and terms
upto fourth order are retained. The generalized active forces due to elastic defor-
mation have been determined by differentiation of the strain energy function while
the generalized inertial forces have been found by taking the dot products between
accelerations and partial velocities.
In fig.(l.a), the quantities a1,a2 and a3 are orthogonal unit vectors which are
attached to the base and they undergo the same inertial motion as that of the base.
The quantities v 1 ,v 2 ,v 3 and W1, W, s are the translation and rotational velocity
components respectively of the base along a,, a 2 and a 3 . The beam displacements
axe given in the rotating frame and they are expressed as functions of x. Then, the
assumed displacements take the following form
N,
uj(x, t) = Oji(x, t)qji(t) j = 1,2,3 (3)

It is to be noted that different sets of genaralized coordinates are chosen for dis-
placement u 1 ,u 2 and u3 . Velocities and accelerations are derived from Eq.(3) and
subsequently generalized inertial forces are obtained. The generalized active forces
due to elastic effects are determined by differentiating the strain energy function.
 II
U = 1/2 [EI2 U,. + E13 ,,..+ EAU2 + EA UI,(u2 + U2,)
+1/4 EA(u, + u. + 2u2 u 2 )]dX (4)

The dynamical equations have been obtained by using the Kane's method 8. EA is
the axial rigidity and EI2 and EI 3 are the bending rigidities. The axial equations
corresponding to the generalized speed 41i axe written as follows
N, N, N2
W 1 1 ,ki.i+ 5 Kjjkjqji + 1/2 5 5 G1 2kmnq mq n +
2
2 2 2
i= m=1n=i=11
N3 N3 N3 N2
1/2 5 5 G13,,,,q mq n + ( 3 3 ;2 + W2W3) 'W 1 s,,qs, + (-w3 + wlw2) E W1 kiq 2 , -
m1 n=1 i=1
N, N3 N2
(WI + ) Wiqli + 2W2 , W13 k,4, - 2W3 5 E W1 2,,,4,.
i=1 i=1 i=1

- -( I + W2V3 - W3 V2 )Wlk + (W2 + w)xk (5)

Similiarly the transverse dynamical equations corresponding to 42i and qi can be
obtained and are given in Ref.(9).
The translational and rotational dynamic equations correspond to vi's and wi's
* For vj, The dynamical equations can be written as
N, N2
(til + w2 v3 - w 3 v 2)M + E W1 ,01 i - 2w3 NV 2
i=1i=
N, N,
+2w2 5W
i=1
3 43, - + Uw)(E Wlql, + J)
i=1
N2 N,
+U1W2 - ( 3 ) E W 2 iq 2 i + (W lU 2 + U;2 ) E W 3 iq 3 i = P1 (6)
i=1 i=1

Similiarly the other two translational equations corresponding to v 2 and v3 can

be obtained. The rotational dynamical equations for w ,is I
N N2 N N3
(W 2 W 3 + Lj1) 5 5 Wkq2kq2i - (W,W 3 - "4i) ,1 W3kiq3kq
k=1 i=1 k1 1
N2 NI N2
-(WlW3 - w )(5 5 W 1,kq 2 2 kqli + 5 Xakq3k)
k= i=1 k=
-(WIW 2 + U;3 )(5,E W3 1kiq3kqli + E X~qk
NN, N N k

+2w( W33kiq3,,43, 4-5 W2 1 qq~k,) -

k= i=1 k=1 =1

U
I
I
5 S13 N
2W3 1: 1 W31 Ikjq3k41
k=l i=
-
NNI
2w)2 1: 1
k=l i=l
W21kiq~kq1 1 +

N3 N2 N 2 N,
U - W2 EE
k=l i=1
W 32kiq 3kq2i +
k=l i=i
W 43- q3i42k)

N3 N2
-, Wakqsk(i2 + w3 vI - wVs) + E W2kq~k(t3 + wv 2 - W2 v1)= M (7)
k=1 k=1

Similiarly, the two rotational equations corresponding to w2 and w3 can be obtained.

II In these equations the following quantities have been defined.

Wmnij =,, p,(z)),0(z)dz 4 Mtk,,,i(I)4nj(l); m,n = 1,2,3; i = 1..N,; j = 1.Nn

I (8.a)
Wpi = 0pkp*(z)dx + Mt Opi(/); p = 1,2,3 i = L..N (8.b)

Xp = jpx(0pjz)dx + Mt I ,pi(l); p = 1,2,3 i = 1..Np (8.c)

The definitions of other quantities can be found in Ref.(9).

I Results and Discussion

The spin-up problem in Ref.(6) is also considered here to study the effect of nonlinear
structural terms. The beam is initially at rest and an angular velocity is given
along a axis at the base. Only the inplane motions are excited in line with the
assumptions. The beam parameters are as follows, E=68950000000 N/mt2 , p= 1.2

3 Kg/mt,Mt(tip mass)=3 Kg, A=0.0004601 nt', 13= 0.0000002031 Mt 4 , 1=10 mt.

The angular velocity history is taken to be identical with the one in Ref.(6).

ws=2/r ~
3= 6/15 t - i 151]
sin rad/s 0 < t < 15 s
W3 = 6 rad/s t > 15 s (9)
The transverse mode shapes are taken as the fixed-free nonrotating eigenfunctions
of an uniform beam under transverse vibration while the longitudinal modes are
taken as the eigen-functions of a fixed-free uniform rod under longitudinal vibra-
tions. The axial and transverse motions are represented by one and three modes
respectively. The axial and bending responses of the tip of the cantilever beam
resulting from the formulation presented in this paper have been shown in Fig.(2.a)
and Fig.(2.b) respectively. The solid and dashed curved are for nonlinear analysis
and they correspond to the cases without tip mass and with tip mass respectively.
In the nonlinear analysis, 'he transverse deflection initially grows in a direction
opposite that of the base motion. After reaching a maximum displacement, the
tip goes back towards the equillibrium position and settles down to a steady os-
cilation. The nonlinear stiffening action in the beam prevents it from instability.
As one might expect, the maximum displacement and I-- amplitude of the final

I steady state oscillations are more when the tip mass is prtsent. The dotted curves

I
 I
I
in Fig.(2.a) and Fig.(2.b) correspond to the cases where all the second and third I
order terms in the final equations are dropped . The beam rapidly goes unstable
which clearly illustrates the necessity of retaining the higher order terms in order
to obtain stable motion. In the analysis of in Ref.(3), foreshortening of the beam
has been assumed a-priori to derive the centrifugal stiffness terms. In the present
analysis, the solid and dashed curves in Fig.(2.a) indicate that the initial foreshort-
ening of the beam is a pure consequence of the imposed base motion. The physical
interpretation is that in the first few seconds, the neutral axis does not stretch and U
as a result an axial shortening results for all bending displacements. After the beam
has reached the maximum bending displacement and starts coming back towards
the equilibrium configuration, the axial displacement grows and finally achieves a
more or less steady state displacement under the centrifugal force field. Finally, the
effects of base translational motion on the axial and bending motion are studied by
imposing a constant base acceleration along a fixed inertial axis and along the plane
of the beam motion for the initial five seconds. The axial and bending responses
are plotted in Fig.(3.a) and Fig.(3.b) respectively.
I
Conclusions I
In this paper, we have formulated the problem of a cantilever beam with a tip mass
and attached to a moving support by using Kane's method. The formulation is
valid for large displacements and all significant geometric nonlinearities have been
I
considered in the strain-displacement relations. The method has been validated by
studying the stability characteristics of a beam under the spin-up maneuver. It has
been demonstrated that structural nonlinearities play a major role in the transient
response characteristics and they cannot be ignored.
I
References i
1.Meirovitch, Leonard.,Analytical Methods in Vibrations.
2.Likins,P.W. et.al, AIAA Journal, 11, 1973, 1251-1258.
3. Vigneron, F.R., AIAA Journal,13, 1975, 126-128.
4.Simo,J,C. and Vu-Quoc,L., J. Appl. Mech, 1986, 53, 849-854
5.Simo,J,C. and Vu-Quoc,L., J. Appl. Mech, 1986, 53, 855-863
6.Kane, T.R. et.al, J. Guid. Cont. Dyn.,10, 2, 1987, 139-151
7.Kane, T,R., " Private communication with S.Hanagud," August,1987.
8.Kane, T.R. et.al," Spacecraft Dynamics," McGraw-Hill Book Company, 1983.
9.Hanagud, S. et.al, J. Guid. Cont. Dyn., To be published. I
I
I
I
I
 Ia £ U3
A
aU,

Figure I Displacements &I-IeUtr&A

axis

0.0 ncinl.Lnear.wLthaut tLo moss nanLi~nsar~wLtha~ut U~ mass

1.0 .oLLnsaR,, LthttLo mss s

5.00.a

a 00

15.0-

0.
C

-25.0 -0.

0.0 S.0 10.0 15.0 20.0 23.0 0.0 5.0 10.0 15.0 20.0 3z.0
t,.ne (seconds] t.Lw. (seconds)

1Ficre. 2.a) Effect of tip mass on axal response Figutre.(2.b) E-ffect of tip mass on bending response

I .0 0.9

L .- ~ .cj Lnear aceLectLon, 2 0 'iout L~rneor accLerotLon

Ftn L-Lnr-AC,
oceerat-L n ---- w.1h narAc2rt~n---

CLC

-3---
C

00

-0.

3 time (seconds] tLme Iseconds I

 SESSION 7

NONLINEAR ANALYSIS OF PLATES AND SHELLS

THURSDAY - 1330 - 1515

June 2, 1988
I

NON-LINEAR OSCILLATIONS OF AN INEXTENSIBLE,

m AIR-INFLATED, CYLINDRICAL MEMBRANE

I Raymond H. Plaut
Department of Civil Engineering
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061

3 ABSTRACT

A long cylindrical membrane, attached to a horizontal base along two generators

and inflated with air, is considered. The material is assumed to be inextensible and its
weight is neglected, so that the equilibrium shape of the cross section is circular, as
5 shown in the figure. Two-dimensional, undamped, nonlinear oscillations about this
equilibrium configuration are investigated.
The radial displacement is a function of the coordinate 9 and time t. Weakly
nonlinear motions are considered, and terms of fourth order and higher are neglected
in the equation of motion. Galerkin's method is applied, first with one term and then
with two terms, to yield ordinary differential equations for the displacement amplitudes
as functions of time. In the one-term case, the equation has the form
ClU + c2 &U + c3 J 2 + c4 U + c5 U2 = 0.
I For anti-symmetric modes, c2 and c5 vanish.
The method of multiple scales is used to obtain approximate solutions. It is
found that the response frequencies tend to decrease as the response amplitude
increases. For the case of two terms, the response may grow large if the internal
resonance w2 = 2w, exists. However, the internal resonance w2 = 3w, has little influence
on the response.

I Numerical integration isalso utilized to determine frequencies and amplitudes.

The results show good agreement with the asymptotic solutions when the motion is
sufficiently small. SO

I II

II . I

I b
I
I
 I

IMPACT OF SHELLS
by
Charles R. Steele
Division of Applied Mechanics
Stanford University
Stanford, CA 94305
Introduction

The analysis of large displacement of thin shells is generally a difficult problem.

Guidance can be found from Love's "principle of applicable surfaces", which is based on
the fact that higher levels of strain energy are required for deformations which involve
extensional distortion of the reference surface, than are required for inextensional bending.
The consequence is that even elaborate post-buckling patterns are usually characterized by
regions of inextensional deformation joined by lines of localized bending. Ashwell (1960)
utilizes this concept for the problem of a spherical shell with a static radial point load, for
which the geometry is quite simple, and showed that a good approximation can be obtained
for the large displacement behavior. The procedure consists of assuming that a portion of
the sphere, a "cap", undergoes a reversal in curvature and becomes a "dimple". Continuity
of stress and displacement is obtained by solutions of the linear equations for the inverted
cap and the remaining portion of shell. Ashwell uses the Bessel function solutions of the
shallow shell equations. Ranjan and Steele (1977) show that a much simpler solution can
be obtained by using the exponential, "steep shell" approximate edge stiffness coefficients.
Furthermore, the accuracy is increased by including the geometric nonlinearity, obtained
from a perturbation expansion of the moderate rotation equations by Ranjan and Steele
(1980). Parnell (1984) gives further considerations to this problem.
The purpose for the present paper is to show that the inverted cap solution permits
insight into the response of a shell in various transient loadings. The results are interesting
because the specific problems are of practical concern and are useful as nontrivial examples
for verification of general purpose finite element programs.
Displacement under static point load

First the basic point load solution is reviewed. A spherical shell under a point load
P has the displacement x according to the linear solution by Reissner (1946) given by I
x PR
c =-8-D=" (1) I
where R .h and D are the spherical radius, thickness and plate bending stiffness,
respectively. The reduced thickness is c, which gives the ratio of bending to stretching
stiffness of the shell wall. The simplest form is obtained with the assumption that the
edge angle a of the inverted cap, or dimple, shown in Fig. 1 is both "shallow":

a < 0.3 (2)

and "steep": I

I
I
n

I
V 4(3)
a >> 1(3

The two conditions are easily satisfied for a range of edge angles when the shell is very

Ut thin. Adding the inverted dimple modifies the potential energy. The dominant terms are the
change in potential of the external force because of the increase in displacement and the
of the remaining
strain energy
obtain of bending
continuity of the
of the slope of edge of the dimpie
the meridian. and the edge
The approximate potential change is shell to

U(ca)=27D ':a -PR a

c (4)

For equilibrium, the potential must be stationary with respect to a change in the edge angle.
This provides the relation between the edge angle and the load magnitude P. Adding the
'I linear result to the additional displacement due to the dimpling gives the total displacement

X= P +K(P*)
h(5)

where the constant is

I2 3 z(6)

The geometric nonlinearity in the edge bending, as obtained by Ranjan and Steele (1980),
provides a 20 percent reduction in the strain energy and an increase in the constant:
2
£N=K(T 1.86 for v=0.3(7

The result (5) agrees well with the experiments of Penning and Thurston (1965) and the
numerical results of Fitch (1968), all on rather thin spherical shells for displacement
magnitude up to about fifteen shell thicknesses. More remarkable are the results of Taber
(1982) showing that the equivalent of (5) provides good agreement with experiments on a
thick shell. In addition, Taber considered the problem of a shell filled with an
incompressible fluid, for which the strain energy of the wall extension due to the internal
pressure must be added. The comparison with experiment is reasonable for displacements
in magnitude up to about half the radius. For displacements much larger than the
thickness, the linear contribution in (5) may be ignored and the result written as a nonlinear
spring:

I, P= YX/2 (8)

I where the constant is

U
I
 46xD
R 'VT(9)

The conclusion is that the key feature is the dimple with the strain energy of edge
bending approximated by the one-term, steep shell solution. In the experiments and
numerical computatoit, bifurcations to nonsymmetric patterns occur. However, a
substantial load loss with such bifurcations apparently does not take place, so the
symmetric solution remains a good approximation.
Impact of a rigid mass

A rigid obstacle of mass M striking the spherical shell with an initial velocity of
sufficient magnitude will cause a large displacement of the shell This displacement can be
approximated by the dimple just as in the static case, as indicated in Fig. 1. Adding the
inertia terms to the previous static formulation leads to the following equation of motion:
dt M+Rphx) 1 +Yx/2=0
TF t (Md
1(10)

in which p is the density of the shell wall. Note that the moving mass of the wall is the
dimple region which increases with the displacement x. When either the obstacle mass or
the mass of the shell wall is dominant, the equation may be integrated exactly. The
numerical solution for the general case is straightforward.
Impact of shell against a rigid wall
In the case of the shell impacting a rigid wall, as indicated in Fig. 2, the contact
force acts against the edge of the dimple region, so the displacement x is half that of the
concentrated load case for a given dimple size. The result for the static effective spring is
~/
P=2/2y (11)

The inertia is easy in this situation, since the dimple region merely changes direction but
must have the same magnitude of velocity as the remaining part of the shell. Therefore the
approximate equation of motion is

2
d t2 3
(12) 1
in which M. is the total mass of the shell and whatever may be attached to the shell, away
from the dimple region of contact. The integrals of this equation are readily obtained. For
instance the maximum displacement is

I
U
 Xm = 2 2y (13)

in which v. is the velocity of the shell at the instant of initial contact.

Extensions
Various additional effects can be readily included. These include the generalization
to a composite shell. The asymptotic edge stiffness coefficients for a composite, as
computed from classical shell theory, are already available from Fertahlioglu and Steele
(1974). The inclusion of transverse shear deformation and composite wall construction in
the analysis of Ranjan and Steele (1980) should be possible. Bennett (1979) shows that
the dimple can be extended to a general elliptic surface. In that case the boundary of the
dimple is an ellipse. The asymptotic solution for the general surface with edge bending is
integrated around the boundary, with the interesting result that the behavior is exactly the
same as that for the sphere but with the spherical radius of curvature replaced by the mean
radius of curvature. Impulsive pressure loading can also be treated in a similar manner.
The effective nonlinear spring is, however, unstable which substantially changes the
response.
It seems that much can be done with the shell of positive gaussian curvature. The
challenging problem is to find an equivalent simple method for the shells of zero and
negative gaussian curvature.

Mass M

I Dipl Veloct

iFigure I - Spherical shell impacted by mass M. For large amplitude displacement an

effective approximation is that an inverted dimple forms. The significant strain energy is in
the localized bending at the edge of the dimple which provides the effe,' of a nonlinear
Uspring.

I
I
I
3
 Rigid Surface

Dimple

Shell
Velocity V

Figure 2 - Impact of shell against a rigid surface. For this problem as well, the inverted
dimple can be utilized. Since the dimple has the same velocity magnitude as the center of
the shell, the kinetic energy is the same with or without the rigid surface.
References
Ashwell, D.G., (1960). "On the large deflection of a spherical shell with an inward point
load", Theory of Thin Elastic Shells, W. Koiter, ed. (North-Holland, Amsterdam),
44-63.
Bennett, B.E. (1979). "Studies in dynamics and the application of asymptotic methods in
solid mechanics", Ph.D. Thesis, Stanford University.
Fettahlioglu, O.A., and Steele, C.R. (1974). "Asymptotic solutions for inhomogeneous,
orthotropic shells of revolution", J. Appl. Mech., 41, 753-758.
Fitch, J.R., (1968). "The buckling and post-buckling behavior of spherical caps under
concentrated load", Int.J. Solids Struct., 4, 421-446.
Parnell, T.K. (1984). "Numerical improvement of asymptotic soludions and nonlinear
shell analysis", Ph.D. Thesis, Stanford University.
Penning, F.A., and Thurston, G.A., (1965). "The stability of shallow spherical shells
under concentrated load", NASA CR-265.
Reissner, E. (1946). "Stresses and displacements in shallow spherical shells", Part I, J.
Math. and Phys., 25, 80-85, Part II, J. Math. and Phys., 25, 279-300.
Ranjan, G.V., and Steele, C.R. (1977). "Large deflection of deep spherical shells under I
concentrated load", Proc. 18th Structures, Structural Dynamics, and Materials
Conference, San Diego, (March 1977).
Ranjan, G.V., and Steele, C.R. (1980). "Nonlinear corrections for edge bending of
shells", J.Appl. Mech., 47, 861-864.
I
Taber, L.A., (1982). "Large deflection of a fluid-filled spherical shell under a point load",
J. Appl. Mech., 49, 121-128.
3
I
I
I
l Nonlinear Dynamic Stability of
Laminated Zomposite Shear Deformable Flat Panels
_ Subjected to In-Plane Periodic Edge Loads"

L. Librescu and S. Thangjitham

Department of Engineering Science & Mechanics
_ Virginia Polytechnic Institute & State University
Blacksburg, Virginia 24061

3 Abstract

3 Laminated composite structures are being increasingly used in the

construction of aeronautical and space vehicles. Employment in their

11 construction of the new material systems that exhibit exotic properties

such as high degrees of anisotropy and high flexibility In transverse

1 shear requires a better knowledge of their instability behavior.

The present paper is devoted to an analysis of the dynamic

stability of laminated composite shear-deformable flat panels subjected

3 to in-plane bi/uniaxial periodic edge forces. The analysis is based on

a refined geometrically nonlinear theory of anisotropic symmetrically-

3 laminated composite panels. The improved theory is developed I) by

incorporating transverse shear deformation and transverse normal stress

m effects, ii) by fulfilling the static conditions on the bounding planes

3 of the panel and iii) by incorporating the higher-order effects. The

theory ;s formulated within the Lagrangian description considered in

5conjunction with the von Kirmin concept allowing one a reduction of the
complexity of nonlinear expressions. The linear counterpart of the

Iobtained governing equation (which is of a Mathieu-type) is used to

3 determine the boundaries of the instability regions while the full

nonlinear equation is used to obtain the behavior of the plate within

5the instability region.

!
I
 I
I
Comparisons of the obtained results with their counterparts

obtained within the first order transverse shear deformation theory and I
the classical one (based upon the Kirchhoff constraints) are made.

These allow one to infer about the influence of transverse shear U

deformation as well as of higher order effects on the parametric

instability boundaries.

In addition, a thorough numerical analysis intended to put into 3

evidence the various effects played by the physical and geometrical

parameters of shear deformable composite panels on the instability I

characteristics is performed and pertinent conclusions are formulated. g
I
I
I
I
I
I
I
I
I
I
I
I The Method of Multiple Scales for Nonlinear Resonances
in the Forced Response of Orthotropic Rectangular Plates

5 Habib Eslami*
Embry-Riddle University
Osama A Kandil**
Old Dominion University
Daytona Beach, FL 32014 Norfolk, VA 23529-0247

Introduction

3 Amplitudes of harmonic transverse deflections of thin laminated composite

plates could be as large as the order of the plate thickness. For these large
deflections, the nonlinear governing equations of motion of laminated
composite plates and nonlinear methods of analysis must be used to study the
nonlinear responses. The literature survey shows that there are significant
contributions in the area of nonlinear vibrations of isotropic plates and
laminated composite plates. For isotropic plates, the reader is referred to
the book by Nayfeh and Mook I , and for composite plates, the reader is referred
to the book by Chia 2 and the survey article by Bert 3 .

I For the nonlinear forced vibrations of rectangular and circular isotropic

plates subjected to harmonic excitations, several investigations have been
reported by Yamaki, et. al. 4, ?lei and Decha-Umphai 5 and Sridhar, et. al. 6,
among others. For the nonlinear forced vibration of laminated composite
plates subjected to harmonic excitations, a few investigations has been
reported, e.g.; Bennett, and ?lei and Chiang. Unfortunately, the literature
lacks research work on the nonlinear forced vibrations of laminated composite
plates including the effect of damping for all possible resonances.

In the present paper, we use the method of multiple scales (MMS) to study
the nonlinear forced oscillations of orthotropic rectangular plates subjected
to harmonic excitations. The Galerkin method is used to reduce the nonlinear
partial differential equations to a set of nonlinear ordinary differential
equations with cubic nonlinearity. The resulting nonlinear equations are
solved by using the method of multiple scales which makes it possible to study
all possible nonlinear resonances; primary as well as secondary resonances
(subharmonics and superharmonics).

Formulation and dethod of Solution

I The nondimensional governing equations, which are based on the von Karman
large-deflection analysis, for a rectangular orthotropic panel subjected to
harmonic loading including the effect of damping and in-plane loadings are
given by

5 *Assistant Professor, Department of Aeronautical Engineering.

m*Professor, Department of Mechanical Engineering and flechanics.

I
 1
+ C w' + + 2D r2 + r4
L(w,F) = w
'TT T 0 nn r nnnn

-1 2o C(F, nn + N + (F, + Na) 2 FS ;] I

n nn En Wn

- Po COSQ = 0 (1) 1
a F, + r2 F nn yr 4 F,nnnn - we ;'nn (2)

where the dimensionless quantities is the deflection, F the Airy stress

function, r = a/b the plate aspect ratio, C the damping parameter, Na and
Na the in-plane loadings, D* = (D1 2 + 2 D6 6 )/D11 D22, r o = ab o =
(2/11//, a A* -- D22/h2 ' ( *
+A* 2
-22/h
(2 A12 + A66 ) =
D11 2 ,
a = A22 h B =
A11 /D1 2/h , Aij and Dij the laminate extensional and bending
stiffness and quantities,
dimensionless h the platethe thickness. For additional
reader is referred definitions
to the paper of the
by Eslami and
Kandil 9 . Assuming a solution of the Galerkin form 1

;(E, n, T) = m
zznl
qmn (T) Xm (d Yn(n) (3)

where Xm and Yn are admissible function- _cisfying the boundary conditions,

substitu'ing the solution into the linear equation, Eq. (2), solving the 1
resultins equation for F as a combination of a homogeneous part and a
particular part, substituting Eq. (3) and the solution of F into
setting the weighted residual equal to zero, we obtain
Eq. (1) and
g
1 1
f0 f o L(;, F) XI (E) Yj(n) d& dn = 0 (4)

Equation (4) yields a set of ordinary differential

nonlinearities which can be reduced to the form
equations with cubic
5
+ + qk + Z mn L k t mn q t = Pok COSMT (5)
qk Ck qk W 4m in
t
nK(5 n

where the time functions qmn have been redefined as

{qk = {q 11 , q 21 ' ' qN1' ".' qN}T (6)

I
I
Using the transformation = 1/2
1 Uk' Eq. (5) reduces to

Uk+ 2c¢k
e + Uk + eE LkUmn U U = Fok cossr (7)
1_l
mmn

1 where Ck 1
k = and Fok = Pok/ (8)

Equation (7) is solved nby using the method of multiple scales by introducing
the time scales T = C T and assuming the expansion

I Uk (T; C) r Uko (TO , T1 ) + e Ukl (To , T1 ) + ..... (9)

The time derivative-CT is transformed by using

-= Do + e D + ....;D i = (10)

I Using the MMS, primary and secondary resonances can be studied. Primary
resonances occur when n - wk (kth mode frequency), and one has to order the
forcing amplitude Fok as Fok = e fk and introduce the detuning parameter
a k such that a = wk + e ak" Secular terms develop in the Ukl-equation and
one has to eliminate them, a process which yields governing equations for the
amplitudes and phases of Uko. For simply supported panel, the deflection
function, Eq. (3), is given by

3 w = E qmn (T) sin(mwE) sin (nwn)

mn
(11)

and it can be shown that only symmetric-symmetric modes are excited.

Considering the case of m = 1, 3 and n = 1, two cases of primary resonance
_ might occur; a = w, or a w3 .

i For secondary resonances, Fok need not be ordered, and for simply
supported panels superharmonic resonances might occur when 3 a - w, or
3 3 - 3 , while sjbharmonic resonances might occur when n - 3 w, or a - 3 3"

Numerical Applications

3 Here, we show a few samples of numerical applications for primary and

secondary resonances. The material considered is a single-layered boron-
9
epoxy and c is kept as e = 0.0001. For the primary resonances, we show the
frequency response curves at different forcing amplitudes with a damping ratio
- = 0.05 for a - w 1 (Fig. 1) and for n - w (Fig. 2). Comparisons of the MMS
3 solution and the 6ethod of harmonic balane show excellent agreement. The
frequency response equations (not given here) show that there is no coupling

I
 m
m
between the modes. For the first mode, the jump phenomen becomes visible as
the forcing function is increased while for the second mode, the jump
phenomenon is not visible for the considered range of Fo . Figure 3, shows the
results for the second mode for the same range of F0 but with a damping ratio
= 0.005, which is one order of magnitude less than that of Fig. 2. Figure
a
4 shows the contribution
1 of the second mode solution for superharmonic 1m
response when n -y l.1 The solution is compared with the superharmonic
response obtained by the first mode only. Additional results will be shown in
the presentation for superharmonic and subharmonic resonances. B
References

1. Nayfeh, A. H., Mook, D. T., Nonlinear Oscillations, Wiley & Sons, Inc. I
2. Chia, C. Y., Nonlinear Analysis of Plates, McGraw-Hill, 1980.
3. Bert, C. W., "Research on Dynamics of Composite and Sandwich Plates,"
Shock and Vibration Digest, Vol. 14, 1982, pp. 17-34.

4. Yamaki, N., Otomo, K. and Chiba, M., "Nonlinear Vibrations of a Clamped

Circular Plate with Initial Deflection and Initial Edge Displacemnt, Pa
I
I: Theory, "J. Sound and Vibration, Vol. 79, 1981, pp. 23-42.
m
5. ?lei, C. and Decha-Umphai, K., "A Finite Element flethod for Nonlinear
Forced Vibrations of Rectangular Plates," Proceedings of 25th Structures,
Structural Dynamics and Materals Conference, Part 2, Palm Springs, CA,
1984, pp. 81-89. Also in AIAA Journal, May 1985.

6. Sridhar, S. D., Mook, D. T. and Nayfeh, A. H., "Nonlinear Resonances in

the Forced Response of Plates," Part II: Asymmetric Responses of Circular
Plates, Journal of Sound an Vibrations, 1978, Vol. 59, pp. 15-170.

7. Bennett, J. A., "Nonlinear Vibration mf Simply Supported Angle-Ply

Laminated Plates," AIAA Journal, Vol. 9, Oct. 1971, pp. 1977-2003. I
8. Mei, C. and Chiang, C. K., "Finite Element Nonlinear Forced Vibration
Analysis of Symmetrically Laminated Composite Rectangular Plates,"
Proceedings of 26th Structures, Structural Dynamics and Materials
Conference, Paper No. 85-0654, Orlando, FL, 1985.

9. Eslami, H. and Kandil, 0. A., "A Peturbation Method for Nonlinear Forced I
Vibration of Orthotropic Rectangular Plates with In-Plane Loads,"
Proceedings of the 28th Structures, Structural Dynamics and Material
Conference, Monterey, CA, 1987, pp. 522-532, accepted for publication in
the AIAA Journal.

I
I
I
 SIMPLY SUPPORTED SIMPLY SUP°ORTED
FIRST MODE SECOND MODE
(30.0S (20.0S
r0l.0
.0 1

IM.M.S.
K1xKya0
H A RM O NIC B ALA N CE Fa i 0
-20.8
K0
K_.
- HM .'K0O NICF a zi
2

M D0 BA0.ANCEN

.02 .00

//Fo25

I 0.6 HAR ONIC

SoPER.03, REPOS

3h -1 O. S....,
- -2 "'-...-....O -2 0 12 3

I 0.?l X3

Fig. I •Frequency response of primry resonance for different forcing Fig- 2, Frequency S.G01O~
response of pres"r resonance of 4 slmlIy-sm0owted
~t
-0. -amolitudit for the ortnotropic simoly-supported panel and the Donal for the second mode and €omprison with the harmonic
compa~rison with the nartmnic balance met.hod. balance method.

SIMPLY SUPPORTED SIMPY SUPPOTED "

I o F 0 :,1-10.. RESPONSEx03
~SUPERHARM4OMIC " .
0.46 SECOND MODE 0.5 ~ -,

.... NM ~OE
SOLU.
o3 K ,=K y =0 i T W0M DE" U

IF 4o10o

02,

0.1 0

Foa5

2 3
I 0 I 2

-1 0 1
w1
3

Fig. 3. The effect of 1oer daning ratio on frequency response of Fig. 14. Contribution of the second mode solution over the single mde
Primary resance for the second made. solution for the super orlonic response whem 30 - wi"

I
!
I
 3
A FINITE ELEMENT METHOD FOR NONLINEAR PANEL FLUTTER
OF COMPOSITE LAMINATES

Chuh Mei
Department of Mechanical Engineering and Mechanics
Old Dominion University, Norfolk, VA 23529-0247

Carl E. Gray, Jr.

Systems Engineering Division
NASA Langley Research Center, Hampton, VA 23665-5225

ABSTRACT

Panel flutter is the self-excited oscillations of a plate in supersonic

flow. Panel flutter differs from wing flutter in that the aerodynamic force
acts only on one side of the panel. In the framework of small deflection
I
linear structural theory, there is a critical value of the dynamic pressure,
X, (or air velocity); above which the panel motion becomes unstable and grows
exponentially with time, and below which any disturbance to the panel
decays. Linear theory can determine the critical dynamic pressure, mode shape
and frequency at the instability but gives no information about the panel's
deflections and stresses. Thus, the service life of the panel can not be
predicted. A great quantity of literature exists on linear panel flutter
using different aerodynamic theories, for example [1-3] and many others. The
aerodynamic theory employed for panel flutter at high supersonic Mach numbers
(f1>1.7) is a quasi-steady first-order aerodynamic piston theory. This theory
calculates the aerodynamic loads on the panel from local pressures generated
by the body's motion as related to the local normal component of the fluid
velocity; thus, a point-function relationship between the normal component of
the fluid velocity (no penetration fluid-panel boundary) and the local panel
I
pressure is known. Assuming exponential dependence on time, the aerodynamic
pressure can be separated into two forces. The primary aerodynamic force is
proportional to the local panel slope, while the other force, which is the
m
aerodynamic damping force (ga), is proportional to the transverse panel
displacement. Hence, the damping force is combined with the usual inertial
force to yield an effective inertial force.
In reality, the panel not only bends but also stretches due to vibrations
with large amplitude. Such membrane tensile forces in the panel due to
stretching, provides a limited stabilizing effect that restrains the panel
I
motion to bounded amplitude limit-cycle oscillations with increasing amplitude
as the dynamic pressure increases. The external skin of a flight vehicle can,
thus, withstand velocities beyond the linear critical value. Nonlinear
structural theory determines the limit-cycle oscillating frequency, and also
panel deflections and stresses. Panel fatigue life, therefore, can be
predicted. For a more thorough understanding of panel flutter behavior, the
geometrical nonlirearity effects due to large deflections should be considered
I
in the formulation. An outstanding survey on the subject for both linear and
nonlinear panel flutter was given by Dowell [41, and most recently by Reed,
Hanson and Alford [5].

!
I ll* ,, l lm ilm
llI llll ll !
12

1 A number of classic analytical methods exists tor the investigation of

limit-cycle oscillations of panels in supersonic flow. In general, the
Galerkin's method is used in the spatial domain, and the panel deflection is
expressed in terms of two to six linear normal modes; and various techniques
in the temporal domain such as the time numerical integration [6-8], harmonic
balance [9,10) and perturbation method [10,11] are also employed. Nonlinear
flutter of orthotropic panels was recently studied [122 using the harmonic
balance method. All of the analytical investigations have been limited to
two-dimensional or three-dimensional rectangular plates with all four edges
simply supported or clamped. The classic approaches also indicated that at
least six linear normal modes are required to achieve a converged solution.

Extension of the finite-element method to study the linear panel flutter

problems was due to Olson [13,14]. Because of its versatile applicability,
effects of aerodynamic damping, complex panel configuration (delta and rhombic
planforms, etc.), flow angularity, inplane pre-stresses, and laminated
anisotropic panel properties can be conveniently included in the
formulation. A review of the linear panel flutter using finite-element
methods was given by Yang and Sung [15).

Application of the finite element method to study the limit-cycle

oscillations of two-dimensional panels was given by Mei and Rogers [16,17].
Rao and Rao [18] investigated the large amplitude supersonic flutter of two-
dimensional panels with ends elastically restrained against rotation. Mei and
Weidman [19], Han and Yang [20], and tlei and Wang [21] further extended the
finite element methods to treat limit-cycle oscillations of three-dimensional
rectangular and triangular isotropic plates, respectively. In deriving the
nonlinear stiffness matrices due to large deflections in refs. 18, 19 and 21,
the strain components (au/ax) and (av/ay) have been neglected from the von
Karman strain-displacement relations. A high-order triangular plate element
with 54 degrees-of-freedom was used in ref. 20 and excellent agreement on
limit-cycle oscillations has been obtained between the finite element single-
mode and the classic analytical six-mode solutions.
The eigenvalues, K, versus dynamic-pressure parameter, X(= 2qa 3/sD), for
a simply-supported square panel at two different amplitude ratios, c/h=O.O and
0.6, for aerodynamic damping ga= 0 are shown in Fig. 1. A 3x8 gridwork in a
half plate was used. The curves for c/h=O.O correspond to the linear small-
deflection theory. When x=O (in-vacuo), the nonlinear flutter problem
degenerates to large amplitude vibrations of plates and the eigenvalues are
real and positive. As x is increased monotonically from zero, the symmetric
stiffness matrix is then perturbed by the nonsymmetric aerodynamic influence
matrix, two of the eigenvalues approach each other, and after coalescence they
become complex conjugate pairs. The critical dynamic pressure xcr corresponds
to the lowest value of x for which first coalescence occurs among all limit-
cycle amplitudes c/h, and it usually corresponds to the linear case c/h=O.O.

In the case of negligible aerodynamic damping 9a+ 0, the flutter boundary

simply corresponds to Xcr . When x<Acr, any disturbance to the panel decays
and c/h + 0. For X>x cr , a limit-cycle oscillation exists with increasing
amplitude as x increases. This situation can be seen more clearly by plotting
the panel damping-rate a and frequency w against dynamic pressure x as shown

I
I
 31
U
in Fig. 2. For the case of very small damping, ga + 0, instability that
corresponds to a positive a does not set in until after the two undamped
natural frequencies w, and w2 have merged. If some damping is present,
ga > 0, the instability sets in a somewhat higher value of x than for the
a
case of zero-damping as indicated by the arrows in Fig. 2. This instability
is, however, not catastrophic. The panel response does not grow indefinitely,
but rather a limit-cycle oscillation is developed with increasing amplitude as
I
dynamic pressure x increases.
The limit-cycle panel deflections for a simply supported square panel at I
several dynamic pressures x are shown in Fig. 3. It should be emphasized that
only an updated single-mode is needed in the finite element solution. 3
In Fig. 4, the panel amplitude of the limit-cycle oscillations is given
as a function of dynamic pressure ratio for simply supported square and
isosceles triangular plates. The critical dynamic pressures xcr obtained from
the analysis are 511.8 and 2022 for square and triangular, respectively. It I
clearly indicates that the triangular panel is much stiffer than the square
one. 3
The increasing use of advanced composite materials
on high-performance
aircraft and missile structures necessitates the determination of limit-cycle
motions of laminated composite panels. A finite element formulation for
analyzing nonlinear flutter of arbitrarily laminated composite rectangular
panels will be presented. The linear stiffness matrix is coupled between the
bending and the inplane nodal degrees-of-freedom due to unsymmetric
lamination, in addition the nonlinear stiffness matrix is also coupled due to
large deflections. The quasi-steady aerodynamic theory is used. Finite
I
element limit-cycle oscillation results of composite laminates will be
presented.
REFERENCES
3
1. Laurenson, R. M. and McPherson, J. I., "Design Procedures for Flutter-
Free Surface Panels," NASA CR-2801, 1977.
3
2. Cunningham, H. J., "Flutter Analysis of Flat Rectangular Panels Based on
Three-Dimensional Supersonic Unsteady Potential Flow," NASA TR R-256,
1967.
I
3. Dugundji, J., "Theoretical Considerations of I ',i Flutter at High
Supersonic Mach Numbers," AIAA J., Vol. 4, July 1905-- op. 1257-1266.
4. Dowell, E. H., "Panel Flutter: A Review of the Aeroelasic Stability of
Plates and Shells," AIAA J., Vol. 8, March 1970, pp. 385-399.

5. Reed, W. H., Hanson, P. W. and Alford, W. J., "Assessment of Flutter

Model Testing Relating to the National Aero-Space Plane," NASP Contractor
Report 1002, 1987.

6. Ventres, C. S. and Dowell, E. H., "Comparison of Theory and Experiment

for Nonlinear Flutter of Loaded Plates," AIAA J., Vol. 8, Nov. 1970,
pp.2022-2030.
I
I
 3 4
7. Dowell, E. H., "Nonlinear Oscillations of a Fluttering Plate II," AIAA
J., Vol. 5, Oct. 1967, pp. 1856-1862.

8. Dowell, E. H., "Nonlinear Oscillations of a Fluttering Plate," AIAA J.,

Vol. 4, July 1966, 1267-1275.

9. Eastep, F. E. and McIntosh, S. C., "Analysis of Nonlinear Panel Flutter

and Response Under Random Excitation or Nonlinear Aerodynamic Loading,"
AIAA J., Vol. 9, March 1971, pp. 411-418.

10. Kuo, C. C., Morino, L. and Dugundji, J., "Perturbation and Harmonic
Balance Methods for Nonlinear Panel Flutter," AIAA J., Vol. 10, Nov.
1972, pp. 1479-1484.
11. Morino, L., "A Perturbation Method for Treating Nonlinear Panel Flutter
Problems," AIAA J., Vol. 7, March 1969, pp. 405-410.

12. Eslami, H., "Nonlinear Flutter and Forced Oscillations of Rectangular

Symmetric Cross-Ply and Orthotropic Panels Using Harmonic Balance and
Perturbation Method," Ph.D. Dissertation, Old Dominion University,
Norfolk, VA 1987.

13. Olson, M. S., "Finite Elements Applied to Panel Flutter," AIAA J., Vol.
5, Dec. 1967, pp. 2267-2270.

14. Olson, M. D., "Some Flutter Solutions Using Finite Elements," AIAA J.,
Vol. 8, April 1970, pp. 747-752.

15. Yang, T. Y. and Sung, S. H., "Finite Element Panel Flutter in Three-
Dimensional Supersonic Unsteady Potential Flow,". AIAA J., Vol. 15, Dec.
1977, pp. 1677-1683.
16. Mei, C. and Rogers, J. L., Jr., "Application of NASTRAN to Large
Deflection Supersonic Flutter of Panels," NASA TM X-3428, Oct. 1976, pp.
67-97.

S17. Mel, C., "A Finite Element Approach for Nonlinear Panel Flutter," AIAA
J., Vol. 15, Aug. 1977, pp. 1107-1110.
18. Rao, K. S. and Rao, G. V., "Large Amplitude Supersonic Flutter of Panels
with Ends Elastically Restrained Against Rotation," Computers and
Structures, Vol. 11, 1980, pp. 197-201.

1 19. Mei, C. and Weidman, D. J., "Nonlinear Panel Flutter-A Finite-Element

Approach," in Computational Methods for Fluid-Structure Interaction
Problems, Ed. by Belytscnke, T. and Geers, T. L., AMD-Vol. 26, ASME,
TT77, pp. 139-165.

20. Han, A. D. and Yang, T. Y., "Nonlinear Panel Flutter Using High-Order
Triangular Finite Elements," AIAA J., Vol. 21, Oct. 1983, pp. 1453-1461.

21. Mei, C. and Wang, H. C., "Finite Element Analysis of Large Amplitude
Supersonic Flutter of Panels," Proceedings International Conference on
Finite Element Methods, Shanghai, China, Gordon and Breach Science
Publishers, Inc. 1982, pp. 944-951.

I
I
 5

4000-

zC

D P S KR

03 1 0 0 0 0 0 0 0

Fig. 1 Variation of eigenvalues with dynamic pressure for a

simply supported square panel at small (c/h=O) and large
(c/h=0.6) deflections, ga=O.

-10 10

20

0 -202

o-

1=" 30- m

20

0 200 400 600 Soo

DYNAMIC PRESSURE PARAMETER,A
Fig. 2 Damping rate and frequency versus dynamic pressure for
a simply supported square panel with several values of
ga at c/h=0.6.
 !600
*6

1.0- SECTION 1-1 -650 y

550 A.-A

0 0.5 9

3a V

1.0-
5 SECTION A-A

U-
w lC 0.5 1.0
-o.2 x
a

Fig. 3 Large-amplitude deflections at various dynamic pressure

for a simply supported square panel, ga=O*
I
.0

0.8- 1 l

Ac-
0- I.

I 0.4-

0.2

F i.0 I .. I I : 1.3
DYNAMIC PIISSURE RATIO, Ac 1.4 1.5

Fig. 4 Limit-cycle amplitude versus dynamic pressure ratio for

simply supported square and triangular panels, ga=O.

I
3
 DYNAMIC STABILITY OF LAMINATED COMPOSITE
SHELLS USING A SHEAR DEFORMATION THEORY

by

E. Yogeswaren and J. N. Reddy

Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, Virginia USA

EXTENDED ABSTRACT

1. Introduction

Dynamic instability of plates and shells includes such phenomena as

buckling under impulsive mechanical loads, vibrations induced by sudden
thermal loads, and instability due to pulsating mechanical loads. This
type of instability has been observed in the first walls of inertial
confinement fusion reactors and in the skin shells of aerospace vehicles
in transonic or low supersonic flights. In these cases failures may
occur at loads far below the static critical load and dynamic stability
analysis is an essential part of the design. For example, a flat plate
subjected to an in-plane load varying at a multiple of plate natural
frequency will exhibit parametric instability and may buckle at load
amplitudes far below the static critical buckling load. The load is
considered parametric since it perturbs the coefficients of the govern-
ing equations of motion and thus appears as a parameter in the equation.

Studies of dynamic instability of laminated composite structures

[1-31 are often based purely on analytical methods thus restricting the
scope of problems to simple geometries, loadings and boundary condi-
tions. The present study deals with the use of a shear deformation
shell theory and the finite element method to investigate dynamic sta-
bility of laminated composite cylindrical shells.

2. Governing Equations and Finite Element Model

The Sanders shell theory [41 used in the present study has the
following equations of motion:
aN1 & 6 -Ttu + T2
aN
ax1 ax2 1 2$1

aN6 aN2 T
ax 1 + ax22 i T ,2
aQ_ aQ2 N, 2N
2 +
,xm_~ - __-T q =1w
ax 1 ax2 R1 21

MI + aM6 Q T2 u + )
ax I x
I
aM6 aM2
axI ax2 Q Tv+T-'

where superposed dot denotes differentiation with respect to time, q is

the distributed transverse load, and Ni, Mi and Qi are the stress resul-
tants

= (k)'1'd Ii = 1,2,6)

(M) = I T
N Ik (4 (k)
5
k=1 k-l
The inertias, Ti (i = 1,2) are defined by the equations,
TT +
1
1 R1 2 1T2 12 +RI 3'9
(11,12,13) = Nk P(k)(1, ,2)dr, (3)

k=1 Ck-1

3 and T; are the same as Ti except that R, is replaced by R2.

The resulting finite element model is of the form
3 [M{u} + [ClIu} + [Klfu} = {0} (4)

where
U [CI = x[MI

[KI = [KL I + [KNL + [Ss ] + [SDICOSwt (5)

and
3 [KL ] - linear stiffness matrix
[KNLI - nonlinear stiffness matrix
I [S s - stability matrix due to static loads

is] stability matrix due to dynamic loads.

The equation (4) with a time-dependent parameter is
a Mathieu-Hill equa-
tion. The analysis presented here establishes the principal and second
5 order unstable regions of the system with and without damping.

I
U
3. Numerical Results
Numerical results for sample problems are discussed here. The
following material properties of a lamina are used:

E1 - 25E 2 ; G2 3 = 0.2E 2 ; G 13 = G = 0.5E 2 ; v12 ' 0.25

A cylindrical shell made of four layer (o°/90*/90*/0*) cross-ply
lamination is studied. An axial sinusoidally oscillating load N, is
used. Advantage of the symmetry is taken to moael a quadrant of the
cylinder (see Fig. la). The instability regions are shown in Fig. lb.
A cylindrical panel made of four-ply (0*/900/900/0*) lamination is
investigated (see Fig. 2a) next. An axial sinusoidally oscillating load
N1 is used, and a quadrant of the cylindrical panel is modeled. The
boundaries of instability are plotted in the Ince-Strutt diagram of Fig.
2b.

In order to study the effect of thickness and aspect ratio on the

instability regions, three cases of plates were considered. The boundary
conditions used are the same as those for the cylindrical panel. The
Ince-Strutt instability charts are shown in Fig. 3 for comparison of
thickness effects and by Fig. 4 for comparison of aspect ratio effect.
Increasing thickness shifts the principal region towards the origin while
narrowing the instability region itself. Larger aspect ratios shift the
principal region away from the origin and narrows the instability region.

References
1. Ambartsumyar, S. A., Theory of Anisotropic Plates, Technomic,
Lancaster, PA, 1970.

2. Ray, H. and Bert, C. W., "Dynamic Instability of Suddenly Heated

Thick Composite Shells," International Joiurnal of Enqineering
Science, Vol. 22, No. 11/12, pp. 1259-1268, 1984.

3. Bert, C. W. and Birman, V., "Dynamic Instability of Shear Deformable

Antisymmetric Angle-Ply Plates," International Journal of Solids and
Structures, Vol. 23, No. 7, pp. 1053-1061, 1987.

4. Reddy, J. N., Energy and Variational Methods in Applied Mechanics,

John Wiley, New York, 1984.
 41 -

4 36-tt JSO~g qqn

4P4~

Ic 4 a

liJ( 00 N" UP

aJO.
 III

3.01
a/h-SO /h 100

rz2.5

S2.0-
N1- N,

0. a-b =10 in,h-0.1, 0.2

0.5"

30.0 32.5 35.0 37.5 40.0

Normalzod frequency,tI

Figure 3. Instability regions for simply-supported four-layer

(0/90/90/0) square laminates under axial compressive load.

I
3.0 3
,z 2.5.

2.0
/b- I a/b-2 3
S1.5

i~ -10 in- h-0.I in.

05

0___ _ .-
35.0 37.5 40.0 215.0 217.5 220.0
Normalized frequency. Q

I
Figure 4. Instability regions for simply-supported cross-ply (0/90/90/0)
laminates under axial compressive load.

I
 SESOU
MICLAEUUOISINNNIERDNMC
THRDY-155-10
Jue2,18
 PLANAR RESPONSE OF ELASTIC CABLES TO A SUBHARMONIC OR
SUPERHARMONIC EXCITATION

Francesco Benedettni and Giuseppe Rega

Dipartimento di Ingegneria delle Strutture, delle Acque e del Terreno

Universitd dell'Aquila
Monteluco di Roio, 67040 L'Aquila, ITALY

ABSTRACT

Planar non-linear oscillations of a onedegree-of-freedom model of elastic cables under

subharmonic or superharmonic resonance conditions are studied. Second-order perturbation
analyses and numerical integrations of the equation of motion are developed. Parametric
investigations are performed for technical cables with various sag-to-span ratios to enlight the
main features of the dynamic phenomena.

INTRODUCTION
The planar motion of a parabolic elastic cable subjected to a vertical distributed load p(x,

t) is described by the equation

iv' +EA/.(y +v') (yv' +v' 2 /2)dx] +p- i = my (1)

in the vertical displacement unknown v(x, t). This equation, which is valid in the case of
moderately large rotations and negligible longitudinal inertia forces, is accurate for studying
nonlinear dynamics of cables used in overhead transmission lines. Referring to assumed spatial
distribution of the excitation and deflected shape of the cable, and applying the Galerkin
method, equation (1) reduces to one dimensionless ordinary differential equation
+ *qi 2q 3q= p* cos Qt (2)
where quadratic and cubic nonlinearities occur, associated with initial curvature and stretching
of cable axis respectively.
Though primary and secondary resonances of systems described by equation (2) have
been analysed in several recent papers (1-41, few works have been concerned with elastic cables
[5-7]. In this paper, subharmonic resonances are studied through second order perturbation
analyses and numerical integration of equation (2), and the main features of these dynamic
phenomena are dis.i.ssed for technical cables with various sag-to span ratios.

ANALYSIS
Secondary resonances due to quadratic nonlinearity

Weakly nonlinear oscillations in the neighbourhood of these resonances are investigated

by letting qe q, w"- L,p* - e p, where q, 4 p are 0(0) variables and P_
is a small but finite
parameter. Equation (2) reads
2 +eC 3 3 =pcosgt (3)
and a second-order approximation to its solution is sought through the multiple time scales
method. Secondary resonances occur if a a 1/2 or Q a 2
To study the former case [21, superharmonic resonance of order two, the frequency of the
excitation in written as 2fl = 1 + e &,where a detuning parameter e &/2,describing quantitatively
the nearness of Q to one half the cable natural frequency, appears. In terms of dimensionless
actual quantities, the frequency response equation for steady-state oscillation is obtained
22 2 2

where 2A = p/(1 - M 2 is the amplitude of the forced term occurring in the time law for q(t) and
a is the amplitude of the superharmonic component of order two. This latter exists for all initial
conditions.
To study the case i -=2, subharmonic resonance of orderone halfwe let Q = 2 + e . With the
ordering assumed for damping and excitation, so called second-order theory [11, the frequency-
response equation for steady-state oscillation reads
22 2 22
a + )2 A2 2 +2
4 28 C2 2 _s
and shows the possible occurrence of both the trivial and a non trivial subharmonic of order one
half in the response. Stable finite amplitude subharmonic occurs only in some regions in the
parameter space of the excitation. Where both the trivial and the non-trivial solution can occur,
which one actually settles down depends on the initial conditions.

Secondary resonances due to cubic nonlinearity

The new ordering q = e q, g* = e2g , p* = e p is let, leading to the equation

+ q+eg + Ec 2 q2 + 2
Ec 3
3q =pcosflt (6)
Secondary resonances are seen to occur to the e-order if Q - 1/3 or a - 3.
To study the former case, superharmonic resonance of order three, we let 30 = 1 + et, and
obtain the frequency-response equation
a =-(c 4 a2 + c5 A2) + (c A 6 /a 2 - 2 /4) 1 2 (7)

where a is the amplitude of the superharmonic component of order three occurring again in the
response for all initial conditions.
To study he case Q _=3, subharmonicresonanceof orderone third,we let 0 = 3 +ee and obtain
the frequency-response equation for the non-trivial subharmonic component of order one third
- -3(c 4 a + cA 2 ) ± 3 Ca2 2/4) 1 (8)

Since the trivial solution is a possible stable steady-state solution, regions of existence and
stability of the finite amplitude subharmonic are determined.
m
I
RESULTS

3Some results are presented for prestressed cables vibrating with the first (symmetric)
mode, with sag-to-span ratio ranging from zero up to about 1/40 and technical values of the
axial rigidity-to-initial tension parameter (EA/H - 500). Equation (4) is plotted in figure 1. The
order-two superharmonic resonance phenomenon is as stronger as the cable is slacker. Apart
from the nearly taut cables showing hardening behaviour, the response is more and more
softening as d/l increases due to the increased importance of the quadratic nonlinearity. Two
curves obtained to the first order approximation are also reported in figure 1 (thick line) for
comparison. Besides the absence of bending, the main difference is concerned with the peak
amplitude value, which is lower for a sagged cable and slightly higher for a nearly taut cable.
The curves obtained by plotting equation (7) for the order-three superharmonic resonance
are very similar to the former ones: the more strongly nonlinear response still occurs for the
slacker cables, notably influenced by the quadratic nonlinearity, but the peak amplitude values
are quite lower than in the order-two resonance.
In the two time laws of steady-state response obtained perturbatively in the
neighbourhood of Q - 1/2 and Ql - 1/3, both the superharmonic component of order two and
that of order three are present, to a different order which is exchanged in the two cases. This
interaction is confirmed by the results of numerical integrations of equation (2) which are in
good agreement with the perturbation results (fig. 2) and show how the 2O-harmonic becomes
the more important one even in the neighbourhood of Q = 1/3 as the excitation amplitude
increases (fig. 3). Moreover, in this case, the numerical results become more and more hardening
and show the occurrence of an upper branch of superharmonic response at amplitude and
frequency values of technical interest, contrary to the perturbation results which predict such
branch to occur at unreliable response parameter values.
Equation (5) for the order one half subharmonic resonance is plotted in the lower part of
figure 4 for a nearly taut cable (thick line) and an actually sagged cable (thin line). Intermediate
cables exhibit higher response amplitudes dose to perfect tuning. In the upper part, the regions
of existence (ID, of possible existence (II) and of non-existence (IDof the stable non-trivial
subharmonic are shown in the parameter space of the excitation. Region I and ll are exchanged
for the two cables since they exhibit different spring behaviour. Results of numerical
integrations validate the predictions of this second-order perturbation solution, except for the
occurrence of the subharmonic component at large negative a-values with the sagged cable, a
physically unrealistic finding which is likely to be due to the order considered in the asymptotic
expansion.
For the same two cables, figure 5 shows in the lower part the frequency response equation
(8) for the order one third subharmonic resonance and in the upper part the regions of possible
existence (Il) and of non-existence (I) of the stable non-trivial subharmonic. These regions differ
notably from each other for the two cables. In particular, as p increases, this subharmonic
phenomenon becomes meaningful mostly for the sagged cable, whose region of existence
extends over perfect tuning. However the response is found to be more weakly nonlinear in the
one third than in the one half subharmonic oscillations, confirming the behaviour obtained in
the corresponding superharmonic.
I
I
I
 CONCLUSIONS

The main conclusions are summarized as follows.

1) The quadratic nonlinearity associated with the initial curvature plays an important role in the
cable forced dynamics. This gives rise to some interesting features of the response. i) The
oscillation amplitude is notably higher for the sagged cables than for the nearly taut cables,
even in the neighbourhood of the secondary resonances due to the cubic nonlinearity. ii) In
this neighbourhood, the response of the system is more weakly non-linear than close to the
resonances due to the quadratic nonlinearity, where the amplitude of the secondary
component can reach important values, some times higher than those of the forced
component. iii) In the superharmonic cases, in which significant interaction between the two
main (2a and 3fQ) superharmonic components of the motion occurs in the neighbourhood of
the two resonant frequencies, the order-two superharmonic becomes the more important
one also near 12 = 1/3 as the excitation amplitude increases. iv) As larger is the sag, as lower
is the excitation amplitude to which the one half subharmonic can occur, and as more
important is the one third subharmonic with increasing excitation amplitude.
2) Use at least of a second-order approximation in the perturbation solutions is essential to
accurately describe the cable nonlinear dynamics. In particular, the so called second-order
theory must be referred to when treating the order one-half subharmonic resonance.
3) The findings of the approximate solutions obtained are validated by the results of numerical
integrations in a rather large range of values of amplitude and frequency around perfect
tunings. As the excitation amplitude increases, the perturbation solutions fail to the
approximation considered.

ACKNOWLEDGMENT

This research was partially supported by the Italian Ministry of Education (M.P.L 60%)

REFERENCES

[1] A.H. NAYFEH 1983 Journalof Sound and Vibration89,457-470. The response of single degree
of freedom systems with quadratic and cubic nonlinearities to a subharmonic excitation.
[2] A.H. NAYFEH 1984 Journal of Sound and Vibration 92, 363-377. Quenching of primary
resonance by a subharmonic resonance.
[31 D.T. MOOK, R.H. PLAUT and N. HAQUANG 1985 Journalof Sound and Vibration 102,473-
492. The influence of an internal resonance on non-linear structural vibrations under
subharmonic resonance conditions.
[41 R.H. PLAUT, N. HAQUANG and D.T. MOOK 1986 Journalof Sound and Vibration 106, 361-
376. Simultaneous resonances in non-linear structural vibrations under two-frequency
excitation.
[5] S.I. AL-NOURY and S.A. ALI 1985 Journal of Sound and Vibration 101, 451-462. Large-
amplitude vibrations of parabolic cables.
[61 F. BENEDETTINI and G. REGA 1987 InternationalJournalof Non-linearMechanics 22,497-509.
Nonlinear dynamics of an elastic cable under planar excitation.
[71 K TAKAHASHI and Y. KONISH1 1987 Journal of Sound and Vibration 118, 85-97. Non-linear
vibrations of cables in three dimensions, Part IL Out-of-plane vibrations under in-plane
sinusoidally time-varying load.
 --
£0.03
a 65 p- 045
01
003 I--O,-4
0.04-
-. ,

0.0 . 1 0.02 ,
f45 4 20

001
0.02i 0
0 .0 1
20

-04 C 0.4 0 0.33 0.5

FREQUENCY
Fig. 1 - Order-two superharmonic Fig. 3 - Amplitudes of superharmonic components with
two excitation amplitudes, numerical results

o 0.02 "
2u .005

S0.004 0.0

1 I 0.01

0.002

0. a) 0.b)
0.25 0.33 0.40 0.4 0.5 0.6
FREQUENCY FREQUENCY
Fig. 2 - Order-three (a) and order-two () superharmonic, perturbation (curves) and numerical (dots) results

004 P
31 X 0.2

002 - j Fg.

-0: 2 0a! 0 -0 0

Fig. 4 - One-half subharmonic Fig. 5 - One-third subharmonic

I
I
 THE RESPONSE OF SEISMICALLY MOUNTED
ROTOR SYSTEM WITH CUBIC
NON-LINEARITIES

A. ERTAS AND E. K. CHEw

Department of Mechanical Engineering
Tezxa Tech Univeroity
Lu6bock, Texaj

The dynamic response of a two-degree-of-freedom seismically mounted rotor system

with cubic non-linearities is investigated. The rotating machine is subjected to internal
forces caused by the eccentricity of the center of mass of the rotor. The equations
of motion of the system are determined using Lagrange's equation. The method of
multiple scales in then used to determine the response of the system. The response of
the system is determined when the excitation frequency is near the first and second
modal frequency under noninternal and internal resonance conditions.

INTRODUCTION
Rotor machinery such as turbines and compressors are designed to operate in a vibra-
tion free manner. However, bearing misalignment, material heterogeneity, geometric
variations, and rotor shaft deflections, collectively or selectively, cause the rotor axial
mass center distribution to be non-coincident with the bearing axis. In general, the lo-
cus of the mass centers of the discretized rotor cross sections defines a space curve with
respect to the bearing axis. Such a deviation is termed rotor mass center eccentricity
and causes time dependent bearing forces to occur in the rotor housing. If the housing
is mounted on a foundation isolation support system, the time dependent bearing forces
will cause a motion of the housing and also gives rise to internal stresses.
The housing support system is supposed to decrease the foundation forces from the
values appearing at the bearing support. However, in order to determine the extent
of the decrease, the motion of the housing, which dictates the transmitted foundation
forces, must be known. In turn, the housing motion is affected not only by the bearing
forces but also by the foundation support system forces. The net result is a complex
interdependent set of relationships between the housing motion and the foundation
forces [1].
Vibration, due to motion of the housing-rotor system, may be enough to cause
malfunctions in sensitive equipment located nearby. The quality of the work from
fine machining may deteriorate as a result of this vibration. Hence, the source of
the vibration should be isolated from sensitive equipment. If the vibration cannot be
reduced by dynamic balancing, or by relocating the rotating machine, it is often possible
to isolate the rotating machine on a seismic mounting [2]. A seismic mounting is formed
by interposing resilient materials between the machine and its supports. The resilient
material is typically in the form of steel springs, rubber isolators, or air springs. One
of the most common materials that is used to isolate machines is rubber [3]. Under
compression, a rubber spring will have the characteristics of hardening spring. Often,
designers ignore the non-linearity of actual springs. However, non-linear systems exhibit
phenomena that are not predicted using a linearized analysis. Many excellent analytical

1!

,,,, ! Ia
I
£
and numerical studies have been done on the oscillations of non-linear systems with
quadratic and cubic non-linearities by Abu-Aris and Nayfeh [4], Burton and Rahman
[5], Mook et al. [6, 7], Nayfeh [8-121, Nayfeh and Zavodney [13], Nayfeh and Jerbil [14],
Plaut et a. [15, 16], Tezak et al. [17], and Ibrahim and Barr [181. To investigate the
effects of the seismic mounting on the dynamic response of a rotating machine the two
mass-hardening spring-damper model is used in this paper. The principal resonances
in the rotor system tuned to an internal resonance due to cubic non-linearities (w2 = 3wj)
are investigated.
a
EQUATIONS OF
MOTION
The equations of motion of the rotating machine are derived by using Lagrange's
equation. The generalized coordinates used for the two-degree-of-freedom rotating
machine, as shown in Figure 1, are x, and X2 which are the vertical displacements of the
center of mass of the housing and base from the static equilibrium positions, respectively.
The springs and dashpots are assumed to be identical for all the supports. All the springs
have a non-linear
described by cubic type hardening characteristics. The load-displacement curve is

F(x) = kz + - (1)

CI

Fig. 1-Schematic of rotating machine

SThe equations of motion of the rotating machine cnbe written a

Z2
'. 3F, * 2

i i,,,. i2 +4kL [(1 + r1,)Z2 - Z,] =4c, [i, - (1+ rc)i2]

L-:r2 r3
- -1 Zl - 1-+ rA,/rd)z] + (z:2 _ X~2) (3)
where M, MD, Mi, c, d, k, rM ai
r r t,o,n, are the mass of the base, mass of the
rotor, mass of the housfth ating coeicient, asymptote of the load-deection curve

!2 mm+ k mm 2 -mzz2)ml-mMm(mMmCOlmlmeimmt)) (2)

I
of spring, stiffness, mass ratio of the rotor to housingj-tass ratio of the base to housing,
ratio of the initial stiffness of springs, excitation fre4ncy, and time, respectively. Dots
indicate differentiation with respect to the time t. The equation of motion of the system
in a non-dimensional normalized form can be written as

2
Y" + YII = c [C.,.Y. + NLi.Y3 + NL 2 .. YIY 2 Y] +nE3 co.(z) , = 1,2 (4)

where the A2, C,., NLIam, NL2,ii, E, and &, are constants,e isa small non-dimensional
parameter, and the prime denotes derivatives with respect to the non-dimensional time
r. In equation (4) the damping matrix C.m can be decoupled by using the modal
orthogonality condition.
METHOD OF SOLUTION AND RESULTS
Following the multiple scale method [191, an approximation to the solution of the
non-dimensional normalized equations (3) and (4) can be expressed as
YI = Y 10(To, T2 ) + C2 Y1 2 (To, T 2 ) + O(s) (5)
= Y (Too,)
0 , T2) + o(#) (6)
Through the solution of equations (5) and (6) the influence of stiffness and damping
of the supports and the overall eccentricity of the rotor housing system on the response
amplitude has been investigated. Some of the numerical results are presented in Figures
2-4 for the following cases:
1. Noninternal resonance conditions with 0 w,1 and n WW2
2. Internal resonance conditions when
a. nl ftwand w2 m3w1
b. flt W2 andW2 ft3W1

o 0

Wo (b)

( 1=.010,( -. 032 M .010,6=.03 -

-,=.012,fs-. 037 ' .012,f -. 0 7 ,

a w .216-2

-o -aooo o o-. o . . .

Fig. 2 The variation of response amplitude a, as a function of detuning parameter a2, with
damping ratio f. (MH = 37.27 Slugs, k = 4,658 lbf/Ft, r B = 10, rMD = 0.3,
eccentricity=e = 0.008 Ft, d = 0.05 Ft, r,= e2 /d 2 ). (a) Noninternal resonance condition
with 0 s wI; (b) noninternal resonance condition with fl W2 .

3
 -
i a I

-a

- = .23
- ~~.002.0032

,:
,/I, / /I
..
V =- / ,",I"-"

aI" rs = .160 e1

- oo .00 I=

N 0 2

02

, with
, as a function of detuning paprmeter
Fig. 3 The variation of response amplitude = 8.44, MD -- 0.3,
symtote d (MN -- 37.27, d Slug, k -- 4,658 lb/Ft, rMB (a), (b) Internal resonance
d 0.03 Ft).
I ~ ~eccentricity--e ---0.008 Ft, 1 --0.02 F, 2 ""

condition with fl --w1 when W2--3w.

/A- 9,,
II 0
a
. 011
K-I 1
12

__ _
a. __ .
_= _ __ _ __ _ __ _

04

0 0.0 ,.
.0
- 1'.o.0
0 -1.
20,,.

IFig. 4 The variation of response amplitude a,

= 37.27 Slugs, k
function
as a4,658
0"2 of detuning
lbf/Ft,
8.44, rMD
rMB = parameter
=
0.3,
=2,with

asymptote d (MH resonance condi-

Ft, dl = 0.02 Ft, d 2= 0.03 Ft). Internal nd r . = 0 .0 71,
eccentricity=e = 0.008 = 3 w2
wj. - a l - , #0 w ith r = 0. 6 a
=2 0w h
w e n
tio n with 0 .1 6 .
0 2 0 w ith r .1 =-
if - - 02 0 w ith
10,
- 0 r, -=0 .07 1 , ... a i # 0O,
2

S
i
 REFERENCES
1. A. Ertas and T. J. Kozik 1987 ASME Journal of Energy Resources Technology 109, 174-179.
Fatigue loads on the foundation due to turbine rotor eccentricity.
2. J. A. Macinanth 1984 Seimic Mounting for Vibration Isolation. New York: Wiley-
Interscience.
3. C. M. Harris and C. E. Crede 1976 Shock and Vibration Handbook. MaGraw-Hill Book
Co.
4. A. M. Abu-Arish and A. H. Nayfeh 1985 Journal of Sound and Vibration 103, 253-
272. The response of one-degree of-freedom systems with cubic and quadratic
non-linearities to a harmonic excitation.
I
5. T. D. Burton and Z. Rahman 1986 Internl J. Non-linear Mechanics 21, 135-146. On the
multiple-scales analysis of strongly non-linear forced oscillators.
6. D. T. Mook, R. H. Plaut and N. HaQuang 1985 Journal of Sound and Vibration 102,
473-492. The influence of an internal resonance on non-linear structural vibrations
U
under subharmonic resonance conditions.
7. D. T. Mook, N. HaQuang and R. H. Plaut 1986 Journal of Sound and Vibration 104,
229-241. The influence of an internal resonance on non-linear structural vibrations
1
under combination resonance conditions.
8. A. H. Nayfeh 1983 Journal of Sound and Vibration 89, 457-470. The response of single
degree of freedom systems with quadratic and cubic non-linearities to a subharmonic
excitation.
I
9. A. H. Nayfeh 1983 Journal of Sound and Vibration 90, 237-244. The response of
multidegree-of-freedom systems with quadratic non-linearities to a harmonic para-
metric resonance.
I
10. A. H. Nayfeh 1983 Journal of Sound and Vibration 88, 1-10. Response of two-degree-
of-freedom systems to multifrequency parametric excitation.
11. A. H. Nayfeh 1983 Journal of Sound and Vibration 88, 547-557. The response of two--
defree-of-freedom systems with quadratic non-linearities to a parametric excitation.
I
12. A. H. Nayfeh 1985 Journal of Sound and Vibration 102, 403-414. The response of non-
linear single-degree-of freedom systems to multifrequency excitations.
13. A. H. Nayfeh and L. D. Zavodney 1986 JournaL of Sound and Vibration 107, 329-350.
The response of two-degree-of freedom systems with quadratic non-linearities to a
combination parametric resonance.
14. A. H. Nayfeh and A. E. S. Jerbil 1987 Journal of Sound and Vibration 115, 83-101. The
response of two-degree-of-freedom systems with quadratic and cubic non-linearities
1
to multifrequency parametric excitations.
15. R. H. Plaut, N. Haquang and D. T. Mook 1986 Journal of Sound and Vibration 106,
361-376. Simultaneous resonances in non-linear structural vibrations under two-
frequency excitation.
16. R. H. Plaut, N. Haquang and D. T. Mook 1986 Journal of Sound and Vibration 107,
309-319. The influence of an internal resonance on non-linear structural vibrations
under two-frequency excitation.
I
17. E. G. Tezak, A. H. Nayfeh and Mook 1982 Journal of Sound and Vibration 85, 459-472.
Parametrically excited non-linear multidegree-of-freedom systems with repeated
natural frequencies.
18. R. A. Ibrahim and A. D. S. Barr 1978 Shock and Vibration Digest 10, 9-24. Parametric
vibration Part II: Mechanics of non-linear problem.
19. A. H. Nayfeh and D. T. Mook 1979 Non-linear Oscillations. Newyork: Wiley-
Interscience.

51 I
 II
II
I

U NONLINEAR OSCILLATIONS UNDER TWO-FREQUENCY PARAMETRIC EXCITATION

I
Raymond H. Plaut
Department of Civil Engineering
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061, USA

5 and

Jeanette J. Gentry and Dean T. Mook

Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061, USA

I ABSTRACT

Forces acting on structural and mechanical systems often appear in

the coefficients of the equations of motion and are called parametric
excitations. A parametric excitation containing two harmonic components is
considered here. In the resonance case which is treated, the sum of the two
parametric frequencies is approximately equal to twice a natural frequency
of the system. Multidegree-of-freedom systems with weak quadratic and cubic
nonlinearities are investigated, and results are obtained by means of the
3 method of multiple scales.

3 ANALYSIS

The following system of equations in un(t) is considered:

2 2 o
+ 2C 2 L + 2u + 2
e E cos ( t + T ) Z OmnU

n nn nn m1 m mj=l I)n

+ E E EiA jkn + E rkuuu= ; n - 1,2 ....

I j~lk=1 knj j=l k-l E.-1 nk

3The w are the natural frequencies of the linearized system; c is a small,

constant parameter: the Ajkn and are constant coefficients of the

I
nonlinear terms; the w are constant damping coefficients; and the Qn , A
and T are constant excitation amplitudes, frequencies, and phases, iespec-
tive[ .

Assume that

X1 2 2 (2)

where a is a detuning parameter. The method of multiple scales leads to the

asymptotic solution

un(t) -a n(T2 ) Cos [*nt + 8n(T2)) + 0(c) (3)

where T 2 2t. In the steady state, a -0 for n~q (assuming in >0), and

solutions a depend on the parameters of tAe system. n

q

RESULTS

Various plots of the steady-state amplitude a are presented in Figures

1-4. Solid lines denote stable solutions and dashed lines denote unstable
solutions. There is always a trivial solution, a =0. When no nontrivial
solutions exist, the trivial solution is stable. When there is one non-
trivial solution, it is stable and the trivial solution is unstable. When
there are two nontrivial solutions, the larger one is stable, the smaller
one is unstable, and the trivial solution is stable.

In Figure 1, a is plotted as a function of a for two sets of system

parameters. The beflayor ma be hardening, as in (A), or softening, as in
(B). For Figure 2, Q - Q - Q and the other excitation amplitudes are
assumed to be zero. qn (A 1i and a are both zero, and the nontrivial
solution a is proportional to Q5 In (B) and (C), u -0 while a has opposite
signs. Inq(D) and (E), P >0 and different sets of parameters are chosen.

The response amplitude depends on the relative values of X. and 1 .

Let Y - X 1 /2 wq . The variation of a with y is shown in Figure 3 'for three
typical cases, with W-0 in (A) an% 1 >0 in (B) and (C). The relative
values of the fxcita.4on amplitudes also affect the response amplitude.
For Figure 4, Q -Q, Q is fixed, and the other excitation amplitudes are
assured to be zMo. q (A) and (D), w -0, while w >0 in (B) anI (C). It
is seen that various types of behavior %an occur, dpendlng on Qqq and the
system parameters.

ACKNOWLEDGEMENT

This research was supported by the U.S. Army Research Office.

I
1 _3.4__ __ 3.0

* LC
2 2.2

- 0_ 1.4
<U .0,/. 0.6

0.6

-0. -0.2
-°- -12 -s 0 6 0 3 6 9 12 15 18
SIGMA SIGMA

Figure 1. Response amplitude aq as a function of detuning parameter a.

I17 7.0 5.4

4.9
14 4.4

I t 2.9
8 -1
a- V2.4
< 5 2.0 < 1.9 N,
1.4

2 0.9
0.4
-1 -0.5 -0.1
0246810 Q 01 Q
2 3 4 0 1 Q2 3

I (A) (B) (C)

17.25 0.55

a 4.75 0.35

2.25 <0.15

_0.25 j -0.05
0 1 2 3 0.0 0.2 0.4 0.6
Q 0
(o) 'E)

Figure 2. Response amplitude a as a function of Q, where Q=Q =Qqq

q

I
 30 3 3

25

Z)20
Cl- E-

:215 2:
<1 <1

5-0- 0-
0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1 .0
GAMMA GAMMA GAMMA
(A) (8) (C)

Figure 3. Response ampl1itude aq as a function of y; y=Xi/ 2 wq-

19 6.5
5.5

02.5

-1.- .
0 5 10 15 20 0 2 4 6 8 1
Q Q
(A) (B)

4.9 1.5

3.9

% 0 .
2.9_

0.9
-0.1 - -0. 1
0 1 2 3 0 2 3 4 5

(C) (D)
Figure 4. Response amplitude a as a function of Q; Q=Q I Q2fixed.
q qq qq
 QUENCHING OF SELF-EXCITED VIBRATIONS
Khaled R. Asfar
Department of Mechanical Engineering
Jordan University of Science and Technology
Irbid, Jordan

3 ABSTRACT
A passive method for suppressing self-excited vibrations is
proposed. The system consists of a mass and damper attached to the main
self-excited system. Free and forced vibrations are considered. A
first-order approximate solution of the governing nonlinear equations is
obtained using the method of multiple scales. The analysis shows that
complete quenching of self-excited free vibrations is possible for some
values of the system parameters. The results are verified by numerical
integration of the governing equations. For the forced self-excited
system, the case of primary resonance is investigated. The analysis
shows that the proposed system is quite effective in suppressing the
amplitude of vibrations. Furthermore, it causes a frequency shift and
l eliminates the unstable region in the response.

1. INTRODUCTION

Quenching of self-excited vibrations has been attempted using

active and passive methods. Asfar e6 al [II analyzed the response of
self-excited oscillators to multifrequency excitations. They concluded
that under certain conditions, the free oscillations may be suppressed
by the presence of non-resonant excitations. The dynamic vibration
absorber for self-excited systems was first investigated by Mansour
(2]. He concluded that an absorber may not be quite effective in
suppressing self-excitation effects due to negative damping. Tondl [31
extended further the analysis of Mansour. In conclusion, the absorber
can be effective in suppressing self-excited vibrations only when the
ratio between its mass and that of the main system is adequately high.
In the present paper, a vibration suppression system known as the
Lanchester damper is proposed for the quenching of free and forced self-
excited vibrations.

1 2. PROBLEM FORMULATION
The differential equations of motion are

d2u l= du 1 1 du 3 1 du2 du(1

= lin - -
udt2 t R t dt-

d u2 du2 du, (2)

I dt 2 Y d2 dt)O
In what follows, we determine a first-order approximation to the
I solutions of equations (1) and (2) for E « 1.

I
l
 3. ANALYSIS

Using the method of multiple scales 141, we seek an approximate

solution in the form

ui(t;e) Ui0 (To,T 1 ) + cuil(To,Tl) + ..., i - 1,2 (3)

The solutions of the equations governing u10 and u2o are written as
iT0 - -iT0
u = A(TL)e +A e (4)
10 eTo0
A() -iT(4
u2 2 20( 1 + Y~iY-i
Y
a,(T,
A+ + I A e 0()
The solution of the equation governing u.. contains secular-producing
terms. Therefore, to render the expansion (3) uniform, these terms must
be eliminated. This condition leads to the following equations
governing the amplitude and phase

a.
a 12 R(6)
= -
e" Ry (7)
2(I+y )
The condition for which a * 0 (written in terms of the system
parameters c,u,k,M) is 2
2
kM _2(km [kin2
Ccr =u k M 1 (8)
Figure 1 shows the relation between the critical damping coefficient and
the mass m. Any point on the curve (c,m) would produce complete
quenching of the self-excited motion amplitude. There is a threshold of
the damper mass m below which no quenching is possible unless a very
high value of damping coefficient is used. Beyond this threshold, a
sharp drop in the value of Ccr occurs.

Equations (1) and (2) are numerically integrated to verify the

result obtained by the perturbation solution. A point with the values c
= 10 and m = 20 is picked on the curve in Figure 1, for K = 100 and M =
100, for the numerical integration. Starting with arbitrary initial
conditions, we find that the amplitude of the self-excited motion decays
with a small rate and continues to decay and goes to zero after a long
time, Figure 2a. If the damping coefficient is decreased below the
critical value (keeping m constant), the amplitude of the limit cycle
remains unchanged and no quenching is produced, Figure 2b. When the
damping coefficient is taken higher than the critical value, the decay
rate increases and a trivial response Is obtained after few time cycles,
Figure 2c.
4. FORCED RESPONSE

Here, the response of the self-excited system to a monofrequency

excitation subjected to the action of the mass-dashpot system is
investigated. The governing equations are
I
I
u C(6- 16) + eR(6 2 - 6) + f cosat (9)
u2 + Y(u2 - u =0 (10)
The case of primary resonance is.considered here. Therefore, the
excitation amplitude f is ordered as f = ef. An approximate solution
using the method of multiple scales is sought in the forms (3)and
(4). The equations governing the amplitude a and phase o are given by

a, =aa -1 Y a + 1f cos (11)

2 j+2 + 2fcs

I2 where 4 -
7l+y
-,)a- -a +-jf si nip (12)
aTj - B. The steady-state frequency-response equation is
j 2+ '
a~ (1 1+y2 4 (13)
where a is defined by a 1 + ea.

Figure 3 shows the effect of varying the amplitude of the

excitation f on the response. For small values of f, there exists two
branches; a lower single-valued branch close to the a-axis,
and a higher closed curve branch in the form of an oval. As f
increases, the oval expands and the lower branch moves closer to the
oval. At some value of f (which depends on R and y), the two branches
coalesce and form one continuous multivalued curve. As f increases
further, the multivalued region decreases and eventually becomes a
single-valued response curve. The curve also moves away from the
a-axis as f increases. In all figures, the solid curve represents the
stable solution, while the dashed line represents the unstable solution.

Figure 4 shows the effect of the parameter R on the response. When

R is equal to zero, the response curves are those discussed in [5]. As
R increases, the two branches approach each other and eventually form a
single continuous curve. The maximum amplitude decreases, and the axis
of symmetry shifts to the right. As R increases further, the curve
becomes single valued and moves to the right and closer to the a-axis.
3y The effective damping in the system becomes positive for R > 5 (for
= 2). All of the response curve becomes stable.

The effect of changing y on the response is shown in Figure 5. For

small values of y, the response curve is single valued and its axis of
symmetry is away from the a = 0 axis. As y increases, the axis of
symmetry shifts to the left, the curve becomes multivalued, and the peak
amplitude increases. As y increases further, the curve separates into
two branches; an oval and a lower branch. The axis of symmetry shifts
further to the left, the oval moves away from the a-axis, while the
lower branch moves closer to the a-axis. For higher values of y, the
axis of symmetry becomes very close to the a = 0 axis, and there is
little change in the response curves. The response of the system
approaches the case with no damper.
I
I
 5. CONCLUSIONS
The results presented in this paper show that the mass-damper
system is quite effective in quenching both free and forced self-excited
vibrations. When compared to the dynamic vibration absorber, it is
found that the present system is superior since it has less components
and does not require any tuning to the excitation frequency.
Furthermore, it can be used in those systems where resilient foundations
are impractical to use.

REFERENCES

1. Asfar, K. R., Nayfeh, A. H., and Mook, 0. T., "Response of Self-

Excited Oscillators to Multifrequency Excitations". Journal of
Sound anC Vibration 79(4), 589-604, 1981.

2. Mansour, W. M., "Quenching of Limit Cycles of a van der Pol

Oscillator". Journal of Sound and Vibration 25, 395-405, 1972.

3. Tondl, A., "Quenching of Self-Excited Vibrations; Equilibrium

Aspects". Journal of Sound and Vibration 42, 251-260, 1975.

4. Nayfeh, A. H., Introduction to Perturbation Techniques. Wiley

Interscience, New York, 1981.m

5. Nayfeh, A. H. and Mook, 0. T., Nonlinear Oscillations. Wiley

Interscience, New York, 1979.
£ U1

II

U. . . ~ . *0

HIM
Figure . Thmerl tgationthe
betee gcaldaing ceffaioint:

(a b Cr c c>Cr (a)co
 -075

/-I I

/ 'I
- -,. . .-- 'N ., ',,..

Figure 3.
- 4 --of f- on the response.
The effect -~~ i-
IiI

.0,
" ,.i 25

SI - - jrIt0UII

Figure 4. The effect of R on the response.

3I

V meffect
-Z. 4. TO.1The
Figure , ofa R on theL response.
T.1's
 Efficient Solution of Constrained Equations of Motion by Recursive Projection

3 by
Roger A. Wehage

Mechanical Engineer
US Army Tank-Automotive Command
i Warren, MI 48397-5000
This paper illustrates a method for representing joints, equations of motion and topology of
articulated mechanical systems in factored matrix form, which results in large systems of loosely
coupled equations amenable to sparse matrix manipulation. Optimal matrix permutation,
partitioning and recursive projection techniques are then applied to symbolically lay out an 0(n)
solution strategy which follows the system's natural topological profile and generates the necessary
uncoupled equations. When combined with symbolic equation generation and optimization
techniques, it offers the potential for highly efficient and general purpose computer programs.
An n degree-of-freedom (dof) mechanical system is composed of joints and generalized
coordinates defining joint and absolute displacements and motion relative to a global reference
frame. Each joint is assigned a reference and referenced side, a center and a positive orientation or
direction as indicated by an arrow. The referenced side of a joint holds other entities such as
inertias, profiles, force elements, joints, etc. and it is the only entry point into an element. The
reference side of a joint can be held by only a single joint (element) or the global frame (ground).
Define an inertial global Cartesian reference frame with 1 5 d 5 6 orthogonal spatial
directions (for example, d = 3 implies planer and d = 6 implies three dimensional or spatial) and
joints a, b, c, ..., p (p = number of joints) with resp. d., db, dc, ..., dp internal dof. Each joint has
d reference and referenced absolute coordinates relative to global. Mechanical systems are
composed of trees formed from roots, branches and leaves of interconnected joints and elements.
The tree represents a path for traversal of the system of joints. Moving from root toward leaves
represents upward traversal of the tree and its joints. All joints are oriented upward in the tree.
Joint connectivity and reachability is defined by sparse, lower triangular, inversely related
(dp x dp) C and Rmatrices with unit determinant. To construct C and R, refer to Figs. 1 and 2 and
think of them as a p x p arrays of d x d cells filled with zeros into which I's are inserted. Label
the rows and columns as a, b, c, ..., p. Trace upward through the tree of joints labeling them in
sequentially increasing order. For each joint *,place I in cell (row *, column *) of C, resp. -I in
cell (*,'-1), where *-1 means holder of *, not necessarily the next lower letter in the sequence. To
construct R, start at each joint * in the tree, trace outward traversing each reachable joint @ in the
path and enter I in the corresponding cell (@, "), or simply invert C.
To simplify the following development, all joint and element displacement, velocity,
acceleration and force components are referenced directly to a global frame. Let joints a and b be
adjacent where joint b allows db < d dof (specific joint characteristics are not important here), then
a d x db velocity influence coefficient matrix Hb, db generalized coordiantes qb and product Hbq b
define the displacement and velocity of b with respect to a. If v a and v b define element a and b
absolute velocities (linear and angular components), then
Vb"V a+Hb b (1)
Differentiating Eq. 1 yields absolute accelerations
8bm a+Hbqb+Yb (2)

I
I
 2

where Yb absorbs components of acceleration which are quadratic in first derivatives. The inertia
matrix is given by d x d matrix Mb. Combined absolute force and moment components are
denoted by d x 1 vectors fb, g b, etc. The unconstrained equation of motion for element b is
+
Mbab" gb fb (3)
where g b absorbs inertial forces due to those acceleration components quadratic in first derivatives
and any external effects such as weight, springs, dampers, etc, and f b represents the reaction and
internal forces at joint b.
The matrix Hb defines tangents to the joint constraint surface along which relative motion
takes place and upon which the internal constraint reaction force Ib acts normal to, so that
b b a0 . The remaining forces in the joint, fb, acting tangent or parallel to the surface (i.e. in the
direction of qb) are called generalized forces 0 b and thus HbTrb,
f 1e. Ob. Let fb,, f-b+ 1ba
and

Hbfb= Gb (4)
Keeping in mind the sign conventions on a joint, the force f b acts positively on element b as
indicated in Eq. 3 and negatively on element a.
Using connectivity matrix C,the equations of motion for an arbitrary mechanical system are
C v -H q (5)
c. mH y+Y (6)
Ma -g +CTf (7)
HTf . (8)
T T iT
where q[qIq,....
a. q, j
qp[Ha.b.... , a. .[.. .Q .... , .,
...v, .. ....4.... ... 1a,
b...,I[a.a,.b
and M - diag Ma, Mb, .... MpJ. Equations 6-8 are combined in matrix form as

C 0 -H -(9)
T
0 H 0 9

Finding an efficient solution of Eq. 9 is relatively straight forward when the matrix
structure is carefully analyzed. To illustrate the procedure, the 4-element serial mechanism shown
in Fig. 1 is analyzed. For any system represented by a tree, the optimum pivotal strategy for
minimum matrix fill and computational overhead requires forward elimination starting from the
leaves, progressing toward the root and the reverse process for back substitution. Matrix Cdefines
the optimum permutation order for forward elimination (up and to the left in C ) and back
substitution (down and to the right in C). Observe that this is exactly the reverse (only to keep the
variables in natural order in the permuted matrices) of what one generally encounters in the
literature, so think of UL factorization instead. For convenience in the previous examples, the
elements were labeled a, b, ..., p in a natural ascending order so Eq. 9 can be permuted into
 £3
1 3
U- -1 Mb 00
Nai ib
Xb
ba
(10)
0 0 Mr ---Wc lid] II id
00 10 g.J

whereWa- l 0 -H 1 001 Xa faI. ba" Ya ,etc. One could solve this problem by
0 HT 0 000 [na] [Qa]
UL factorizaion but the structure of the diagonal matrices makes that impractical. It is more easily
handled by matrix partitioning. Consider the linear system of equations A x - b partitioned as
A11A1 2 [Xl ][bl] whereA22 isnonsingular. ThenAll
x+A x -bl, A 12 2 21 x+A22x 2 -b 2

yields
Sx- A22[b 2 -A
2 2 1x 1 (11)
and
S2A2A
2A 2 2A 2 1]X-bI.A12 A1b (12)
Rename this reduced system of equations in Eq. 12 as B y -c and partition again as above to obtain
11 Y I + 8 12 Y2 -C 1,8 21 Y 1 + 3 22Y 2 a € 2 where B 22 is nonsingular. These new equations yield
y2-B22 [2-1321 Y1] (13)
ard 11 B12 B 22 13 21] Y I -¢ process continues until the last matrix is small
CI" B 12 B 22 c 2"- The
enough so that it is easily invertible which completes the forward elimination step. Say
| "1
B11- 812B22B21 is obtained in the last step, then back substitution starts with

c2 ] and follows with Eqs. 13 and 11 wherex Y-yy 2

Yl =
B1-B12 B 8 B1 1c-1 2B
Eventually the entire vector x is evaluated.

I
Following the above pattern, the first partitioning of Eq. 10 starts as
FI
ao -1 OMb[
0b d
ab
1
-rTd~ -'

-T-ili ]c (14)

3 L0 0 -1 Wd Xd d
where the submatrix Wd must be inverted. This inversion can be simplified by first partitionini Md

and noting that [d - 1 0 "d"

1L" . After some effort and using the basic identity

I
I
 4 1
A
T
-8 PFAIE
"lI
where D"-13TA 'IB]I E -,A'IBD,FT.,D BTA "1 , p-I -EBT and Pis a

projection matrix such that P 2 _ p, it can be verified that the desired inverse is
GdP Ed]
p
d Pd Md Fddj(5

where Md= Md, Dd -[H T MH , Ed'HdDd, Fd'MdEd, Gd -EdHTpd'Fd d,P d, 3

Md PM. - Matrices Md/Mp are the effective/projected (across joint d) inertias of element d.
Now using Eq. 11 as a guide, the partitioned Eq. 14 and Eq. 15, it follows that

Xd.'/"o"
'Ed " d
TI
Fl'bd4°[ ° T'..
L1)
(16) VI
L d d
Only the last equation T
of Eq. T16 is necessary for solving the equations (see Eqs. 6 and 7), thus
(17
+
qd -DdOd EOgd"F[C+d] (17)
where g -gd and go is the effective force on element d. Note the introduction of redundant
effective inertias Md - Md and forces gd- g dat leaf elements for consistency.
Using Eq. 12 as a guide, the matrix corresponding to A 11 "A12 A A21 is now evaluated.
Setting up the matrix products from the partitioned submatrices of Eq. 14 and Eq. 15 reveals that to
updac the entire expression requires only the single operaon - cM d where, asbove, M

[T
is the effective inertia of element c. Updating the right hand side, corresponding to Eq. 12 yields

ia 0 b1 GdPd Ed
b -/b -0 "d d d (18)
LEd
- Fd Dd
which also requires only a single computation
acgc+gd- MYd- Fd0d (19)
d gd The quantity gc is the effective force acting on element c and g is the

projected force across joint d. If element c should happen to hold more than one joint, then Me*and
gc would receive projected components across each such joint. At this point the first step of
forward elimination is complete and the reduced system along with its next partitioning becomes
I
I
I
3 5

_T 0 -

I where -1 I o -Hcj' = c The remaining elimination steps follow by decrementing

I subscripts, i.e. (c b and d -- c), etc, and using the effective mass M( and force g. With the

additional equations, Eqs. 6 and 7 and matrix 0,the sparse matrix algorithm is now complete. A
more detailed development of 0(n) solution algorithms for both open and closed kinematic-loop
I systems is given in Ref. 1.
a1(b)() a a
C,

I
sytm
isgvn
Ref.
1I
in mor dtale
deelpmntof ~nasluio aoihfrbotoe
Figure 1Example 4-Bar Mechanisms
adcsd knmtclo

abcd abcd ab c d

ca IbI 0 0 , b-I I0 ~ II 0
-10ci0-11 cI-010
d 0 -- l d 0-i 0 I d f0
ab-I1100 b b -11 00 c b 11 00

a100 a 1 000 a I 000

o0 01 111
00C RC
d 0 1d Rlii0101I ]0 d i
10

i~ ~R Fre2Cetvt n ecaiiymticsfrtemcaim nFg

References
I I Wehage, R.A., "Application of Matrix Partitioning and Recursive Projection to O(n)
Solution of Constrained Equations of Motion," to appear in the ASME Journal o

U Mechanisms. Transmissions and Automation in Desien.

I
 GLOBALLY CONVERGENT NUMERICAL ALGORITHMS FOR
NONLINEAR SYSTEMS OF EQUATIONS

Layne T. Watson
Department of Computer Science
Virginia Polytechnic Institute & State University
Blacksburg, VA 24061

Abstract. There are algorithms for finding zeros or fixed points of nonlinear systems
of equations that are globally convergent for almost all starting points, i.e., with probability
one. The essence of all such algorithms is the construction of an appropriate homotopy map
and then tracking some smooth curve in the zero set of this homotopy map. There are three
distinct, but interrelated, aspects of homotopy methods: 1) construction of the right homotopy
map, 2) theoretical proof of global convergence for this homotopy map, and 3) tracking the
zero curve of this homotopy map. The first aspect is currently still an art, although this is
much better understood now due to the accumulation of computational experience. Although
much remains to be done, significant progress has been made on the second aspect. Global
convergence has been proved for Brouwer fixed point problems, certain classes of zero finding
and nonlinear programming (both unconstrained and constrained) problems, and two-point
boundary value approximations based on shooting, finite differences, spline collocation, and
finite elements. Recently A. P. Morgan obtained some elegant results for polynomial systems,
for which homotopy methods are guaranteed to find all the solutions.
Homotopies are a traditional part of topology, and only recently have begun to be used
for practical numerical computation. The algorithms described here are known as probability
one globally convergent homotopy algorithms, which are related to, but distinct from, contin-
uation, parameter continuation, incremental loading, displacement incrementation, invariant
imbedding, and continuous Newton methods. These algorithms are also referred to as "contin-
uous" methods, to distinguish them from the simplicial homotopy methods, whose theoretical
foundations date back to the very origins of topology.
The frameworks for fixed point and zero finding problems are slightly different, so they
will be discussed separately. The fixed point problem will be considered first. Let B be the
closed unit ball in n-dimensional real Euclidean space E ' ,and let f : B - B be a C 2 map.
Define p,, : [0, 1) x B - E" by

p.(X, x) = \(z - f(z)) + (1 - X)(x - a). (1)

The fundamental result is that for almost all a (in the sense of Lebesgue measure) in the interior
of B, there is a zero curve 7 C [0, 1] x B of p., along which the Jacobian matrix Dp=(A, x) has
rank n, emanating from (0,a) and reaching a point (1,±), where i is a fixed point of f. Thus
with probability one, picking a starting point a E int B and following -Yleads to a fixed point
of f. This justifies the phrase "globally convergent with probability one".
The zero finding problem
F(x) = 0, (2)

where F: En - En is a C' map, is more complicated. Suppose there exists a C 2 map

p: E' x [0,1) x E" - En

1I
3 such that
1) the n x (m + 1
+ n) Jacobian matrix Dp(a,A,x) has rank n on the set

p-'(0) ={(a,A,x) Ia E En,0 < A < 1,x E E",p(a,A,x) = 0},

and for any fixed a E E',

2) p.(O,z) = p(a,0,x) = 0 has a unique solution zo,
3) p.(1, z) = F(x),
4) p-'(O) is bounded.
Then the supporting theory says that for almost all a E E ' there exists a zero curve -f of p ,
along which the Jacobian matrix Dp , has rank n, emanating from (0,zo) and reaching a zero
2 of F at A = 1. - does not intersect itself and is disjoint from any other zeros of p. . The
globally convergent algorithm is to pick a E Em (which uniquely determines x0), and then
track the homotopy zero curve 7. A simple choice for p, is

p.(A,) = AF() + (1-A)(z-a). (3)

This satisfies properties 1)-3), but not necessarily 4). There are fairly general sufficient condi-
tions on F(z) so that (3) will satisfy property 4), but for some practical problems of interest
the homotopy map (3) will not suffice.
The Transversality Theorem from differential topology provides general conditions un-
der which most homotopies (in a precise sense) will have smooth non-bifurcating curves. In
practice an admissible homotopy is constructed by defining artificial parameters (a) so that a
partial derivative condition is satisfied and then choosing these parameters independently of
the structure and coefficients of the original system (F). For example, random choices of a will
generally work.

Thus the artificial homotopy p(a, A,z) might be chosen so that the jth component includes
aj and not any ak for k 4 j, and so that the partial derivatives

3 Op(a, A,z)
Oai

for j = I to n are nonzero for 0 <_ A < 1,for all z. Here a = (a,,...,a,n) and A are artificial;
that is, they have nothing to do with x or any other parameter of the given problem. Then
a is chosen at random. The Transversality Theorem guarantees that the resulting homotopy
curves will be smooth, without bifurcations or singularities. In fact, in practice they tend to
be very well conditioned. This mysterious usefulness of randomly chosen a is a feature of the
-probability-one" approach to constructing homotopies.
Probability-one globally convergent homotopy curves have no bifurcations (with proba-
bility one) for 0 < A < 1. However, at the end of a curve (when A = 1), singularities may
be encountered. This happens precisely when the original problem is singular at the solution,
because as A -- 1 the homotopy becomes the original system. For some mild singularities in
F, the homotopy can remain nonsingular at A = 1,but in general this is not so.
The zero curve 7 of the homotopy map p. can be tracked by many different techniques.
HOMPACK, currently under development at Sandia National Laboratories, General Motors
Research Laboratories, Virginia Polytechnic Institute and State University, and the University
of Michigan, is a suite of codes for tracking zero curves of probability one homotopy maps, and
provides both high-level and low-level subroutines for three different approaches to tracking 7.

12
I
 I
I
The three algorithmic approaches provided by HOMPACK are: 1) an ODE-based algorit'.m; 2)
a predictor-corrector algcrithm whose corrector follows the flow normal to the Davidenko flow (a
"normal flow" algorithm); 3) a version of Rheinboldt's linear predictor, quasi-Newton corrector
algorithm (an "augmented Jacobian" method). There are qualitatively different algorithms for
dense and sparse Jacobian matrices; HOMPACK provides for both capabilities.
This talk surveys globally convergent homotopy methods, the HOMPACK software, and
some 2.. 'ications to nonlinear dynamics.

References.
2
3
[1] L. T. WATSON, A globally convergent algorithm for computing fixed points of C maps,
Appl. Math. Comput., 5 (1979), pp. 297-311.
(2] L. T. WATSON AND D. FENNER, Chow- Yorke algorithm for fixed points or zeros of C
2 1
maps, ACM Trans. Math. Software, 6 (1980), pp. 252-260.
[3] L. T. WATSON, Numerical linear algebra aspects of globally convergent homotopy methods,
SIAM Rev., 28 (1986), pp. 529-545. I
[4] L. T. WATSON, S. C. BILLUPS, AND A. P. MORGAN, HOMPACK: A suite of codes for
globally convergent homotopy algorithms, ACM Trans. Math. Software, 13 (1987), pp.
281-310.

I
I
I
I
3I I.

II
I
3 3
I
an i I I I IU
 SESSION 9

MULTIBODY DYNAMICS

FRIDAY - 0830 - 1000

June 3, 1988
SURVEY ON THE GENERATION AND SOLUTION OF MULTIBODY SYSTEm EQUATIONS

by

R. Schwertassek
Institut fUr Dynamik der Flugsysteme, DFVLR, Oberpfaffenhofen,
W. Germany

and

C. FUhrer
Mathematisches Institut, Technische Universitdt MUnchen, W. Germany

To be presented at the Nonlinear Mechanics Conference, Session Multi-

body Dynamics, June 1-3, 1988, VPI&SU, Blackbury, Virginia.

Abstract

:ultibody system equations can be generated in various forms. All of

these can be interpreted as results of two basic approaches, the
augmentation- and the eliminatio.n-method. The former method yields
the descriotor form of the system motion, a set of differential/
algebraic equations, and the latter the state space representation,
a set of pure differential equations. Both of these methods are
surveyed.

For simulation purposes one would like to select that set of system
ecuations ;hich can be generated mcst efficiently and for which the
:ost efficient and reliable solution techniques are available. Numeri-
cal solution techniques for pure differential e'uations have been
studied in great detail and they are well-developed. By contrast,
differential/algebraic equations have not been investicated for such
a long t.:ne. The status of in the !attar field is sur-
-evelo*zn
veved and recent results on improving relability and efficiency of
the corresponding solution techniques are discussed. A new method,
avoidina the shortcomings of previous techniques for solving differ-
en:ial/aLcebraic equations, is presented.
 ON THE USE OF THE FINITE ELEM ENT METHOD
AND CLASSICAL APPROXIMATION TECHNIQUES IN
THE NONLINEAR DYNAMICS OF MULTIBODY SYSTEMS

A.A. Shabana
Dept. of Mechanical Engineering
University of Illinois at Chicago
P.O. Box 4348
Chicago, IL 60680

ABSTRACr

In this paper the finite element method and classical approximation

techniques such as Rayleigh Ritz methods are used to develop a set of
generalized Newton-Euler equations for deformable bodies that undergo larce
translational and rotational displacements. In the finite element formula-
tion, a stationary (total) Lagrangian approach is used to formulate the
generalized Newton-Euler equations for each finite element in term of a
set of invariants that depend on the assumed displacement field. The
deformable body invariants are obtained by assembling the invariants of the
finite elements using a standard finite element Boolean matrix approach.
This leads to the nonlinear generalized Newton-Euler equations for the
deformable bodies. These equations are presented in a simple closed form
which is useful in developing recursive formulations for multibody system
consisting of interconnected deformable bodies. Both lumped and consistent
mass formulations are discussed.
 COMPARATIVE STUDY OF METHODOLOGIES EMPLOYED
IN CONSTRAINED MULTIBODY DYNAMICS

F.M.L Amirouche
Assistant Professor
Mechanical engineering Opt
University of Illinois at Chicago
Chicago il 60680
ABSTRACT

Popular methods used in the coordinate reduction of dynamical systems

subjected to simple nomholonomic constraint include the zero eigenvalue

theorem,the singular value decomposition,coordinates partioning based on

LU factorization,and the QR decomposition method.Most recently Amirou:he

et al have introduced a new method called the Pseudo-uptriangular-

3 decomposition which is beleived to be computational more efficient and

provides a more stable algorithm to compute the orthogonal complement

3array to the Jacobian matrix.

3l Unlike the previous method which rely on the solution of eigenvalues and

invertion of matrices,the PUTD is based strictly on multiplication type of

3 operation suited for computer simulation.In this paper we intend to pre

sent a comparative study between methods and draw a conclusion on their

advantages and draw backs as it pertain to the solution of complex

g constrained dynamical systems.

I
I
I
I
 Primal and Mixed Forms of Hamilton's Principle for
Constrained Rigid and Flexible Dynamical Systems:
Numerical Studies
M. Borri, F. Mello, M. lura, and S. N. Atluri
Center for the Advancement of Computational Mechanics
Georgia Institute of Technology
Atlanta, GA 30332

Abstract

Constraint equations arise in the dynamics of mechanical systems whenever there is the need
to restrict kinematically possible motions of the system. In practical applications constraint equa-
tions can be used to simulate complex, connected systems. If the simulation must be carried
out numerically it is useful to look for a formulation that leads straightforwardly to a numerical
approximation.
This can be done if the dynamics of the holonomic or nonholonomic system is cast in a variational
form to which a finite element discretization can be easily applied.
This paper suggests the adoption of a new variational principle and shows how two different
formulations can be used to achieve good numerical results.
Introduction

Very often the vectorial and variational theories of mechanics are considered completely equiv-
alent and the differences between them are considered only a matter of style. Many times, the
variational principles are used only as an alternative approach to obtain the differential equations
of motion.
Here we assert that the variational formulations are superior, not only because they afford a
generalized and unified treatment of complex mechanical systems, but also because they are more
easily implementable in a numerical form. They are extremely well suited for obtaining, in a general
way, an automatic appproximation for the treatment of the stability and response equations of very
complicated nonlinear systems. The numerical approximation can be built on a few basic and easily
controllable hand developed formulae.
It is worth remarking that, in the treatment of practical problems, many existing variational
principles must be revised in order to make them suitable for numerical implementation.
During the last decade the variational formulations for complex dynamic systems and their
numerical approximations have known a renewed interest [1-41. An example of this is the direct use
of Hamilton's Weak Principle for the time finite element approximation of the dynamics of holo-
nomic systems [5]. In the case of nonholonomic systems, however, a general and sound variational
formulation suitable for a direct numerical approximation is not yet available.

1!
mmen nm m
n nnlunlu UU UIml~ anmnN
 To contribute to a possible solution of this problem this paper suggests the adoption of a new
variational principle for holonomically and nonholonomically constrained dynamic systems and
shows how two different finite time element approximations can be derived.
IDifferent Forms of Hamilton's Principle

g Hamilton's Principle for unconstrained dynamics can be written as:

= 6q . p(1)
fttj (64 + bq "Q)dt
where tj, t2 are the ends of the time interval of interest; q and p are respectively the generalized
coordinates and the momenta of the system. Z (q, q, t) denotes the Lagrangian function and Q the
external forces not included in C.
By the use of the Hamiltonian transformation, Eq. (1) can be rewritten in the fowing mixed
form: ffmi(6i, p - 6p -q - 6H + 6q. Q)dt
= (Sq p - 6p" q)i
(2)

where H(p, q, t) = p . 1 - Z denotes the Hamiltonian of the system.

The Eq. (1) and Eq. (2) can be denoted respectively as primal and mixed forms of Hamilton's
Principle and they are very suitable for numerical approximations in the context of finite elements
in time domain. [5]
Let us consider now a constrained system in which the velocity q must satisfy the following
equations:
0,(4, q, t) = A(q, t) - q + a(q, t) = 0 (3)
These equations entail the following constraint on the virtual displacements:

I A. Sq = 0 (4)

In order to enforce Eq. (3) and Eq. (4) in a weak form we use the Lagrangian multiplier technique.
Let i be these multipiers. We then weight Eq. (3) and Eq. (4) with the variation bp and the time
derivative A respectively. Obtaining:

6p - Sq = 0 (5)

The benefit of this form is that it allows another integration by parts that reduces the continu-
ity requirements for the Lagrangian multipliers. So, substituting Eq. (5) into Eq. (1), after this
integration by parts [9], we obtain:

ft (bC + bq - Q)dt = bq. -4 (6)~

where:
-£+p.@ =p+p'- Q=Q+ (~ a) (7)

The Eq. (6) constitutes the modified Hamilton's Principle for constrained systems and , , Q are
respectively the modified Lagrangian function, the modified generalized momenta and the external
forces modified by the reactions due to the nonholonomic constraints [91. It is interesting to note

S * 2
 0.4-
0.4

- 6 0. -. - 0 Su.

-USA

-0.4- .U

seentha
-0.6 IF , O.~ I - 6 - -pA
-0.5 -0.4 40.2 0.0 0.2 0.4 GA . S Ne.S MAo * eu.o
X DISPL.AC2MENI"m

Figure 1: X-Y Displacement Figure 2: Precesion anglrei

.0.2

that are actually a generalized momenta of the augmented Lagrangian, in fact it can easily be
seen that i -=~q
Taking this property into account we can define the modified Hamiltonian function as 1 = m
l "q - and rewrite the Eq. (6) in the following mixed form:

It is worth emphasizing that the modified momenta f, from which the compatible momenta p and
the Lagrangian multipliers p can be recovered by a simple projection, are no longer constrained
and can be viewed as independent variables.
For the sake of simplicity the formulations presented here are for finite degree of freedom
systems, but they can be easily extended to continuous systems as in the ref. [11]. There, wave
propagation in a rod is analyzed with the use of a mixed formulation, similar in concept to the
two feild approach of Eq. (2). It is interesting to note that for linear problems the step by step
marching scheme of that formulation produces very accurate results.

Numerical Results

The preceding developments have been verified with a few simple but significant numerical
examples. The formulation corresponding to Eq. (6) has been used in order to solve the well known
Caplygin's nonholonomic problem [81, [9].
The formulation corresponding to Eq. (8) has been used to solve the spinning top problem [101.
Since we intend to check the holonomic and nonholonomic constraints, the reference point is taken
to coincide with the center of gravity and the velocity of the suspension point is enforced to be
zero by appropriate constraint equations. Moreover, as a nonholonomic constraint, the constancy
of the spinning angular velocity is considered. The results obtained are very encouraging and the
method is very promising.
Figure 1 shows the plot of X and Y displacement of the mass center. This represents approx-
imately 7500 calculation steps corresponding to 500 proper rotations. The numerical approach
shows very stable behavior, even for this dynamically stiff problem. Figure 2 is the plot of pre-
I
cession angle, in which the discontinuity in the representation of finite rotation is clearly shown.

3!
IIlliiilm ill inilnnnnlmmn• ]m
I
This does not present a problem in the numerical approach because the incremental finite rotation
is adopted as a generalized coordinate. The physical data used in this problem are the same as in
Ref. [121.

3Aknowledgements
This work was supported by AFOSR and the SDIO/IST. The assistance of Ms. Deanna Winkler

in the preparation of this manuscript is thankfully acknowledged.

References
1 1.
C.D. Bailey, "Application of Hamilton's Law to Forced, Damped, Stationary System?, Jour.
nal of Sound and Vibration, Vol. 75, pp. 117-126, 1981.

2. M.E. Simkins, "Finite Element for Initial Value Problems in Dynamics, AIAA Journal, Vol.
13, pp. 1154-1157, 1981.

3. M. Baruch and R. Riff, "Hamilton's Principle, Hamilton's Law, 6n correct formulations?,

AIAA Journal, Vol. 21, pp. 687-692, 1982.

4. M. Borri, M. Lanz and P. Mantegazza, "Helicopter Rotor Dynamics by Finite Element Time
Discretization", L'Aeretecnica Miuili e Spazio, Vol. 60, pp. 193-200, 1981.
5. M. Borri, G.L. Ghiringhelli, M. Lanz, P. Mantegazza and T. Merlini, "Dynamic Response of
Mechanical Systems by a Weak Hamilton's Formulation", Computers 8 Structures, Vol. 20,
pp. 495-508, 1985.

6. T. Levi Civita and V. Amaldi, Lezioni di Meccanica Razionale, Zanichelli, Bologna 1979.

7. C. Lanczos, "The Variational Principles of Mechanics", University of Toronto, Toronto 1964.

8. Iu. I. Neimark and N.A. Fufaev, "Dynamics of Nonholonornic System", American Mathemat-
ical Society, Providence, Rhode Island 1972.
9. M. Borri and P. Mantegazza, "Finite Time Elements Approximation of Dynamics of Non-
holonomic Systems", paper presented at ASME Congress, Williamsburg, Virginia, August
1986.
10. M. Borri and S.N. Atluri, "Time Finite Element Method for the Constrained Dynamics of a
Rigid Body', Proceedings of ICES-88, Atlanta, Georgia, April 1988.

5 11. M. lura, M. Borri, and S.N. Atluri, "Analysis of Traveling Wave Responses of Structures",
Proceedings of ICES-88, Atlanta, Georgia, April 1988.

12. P.E. Nikravesh and E.J. Haug, "Generalized Coordinate Partitioning for Analysis of Mechan-
ical Systems with Nonholonomic Constraints", Journal of Mechanisms, Transmissions, and
Automation in Design, September 1983,Vol.105, pp. 379-384

I
*t
 I

I
ABSTRACT

ANALYSIS OF NONLINEARLY CONSTRAINED

NONHOLONOMIC MULTIBODY SYSTEMS
B
I
I
Mingtao Xu
Chengqun Liu*
Department of Engineering Mechanics
Chongqing University
Chongqing, Sichuan
PEOPLE'S REPUBLIC OF CHINA

and

Ronald L. Huston
Department of Mechanical and
Industrial Engineering
University of Cincinnati
Cincinnati, Ohio 45221-0072
U.S.A.

I
I
I
*Currently, Ph.D. Candidate, University of Cincinnati I

I
I
 This paper presents a new method for studying multibody systems subjected
to nonlinear nonholonomic constraints. The method is based upon Kane's
equations as exposited by Kane, Levinson, Huston, et. al. Specifically, the
method employs partial velocity and partial angular velocity vectors in
developing generalized active, generalized constraint, and generalized inertia
force arrays. Setting the sum of these force arrays to zero leads to the
general dynamical equations of the system. Then by appending the constraint
equations the governing equations of the constrained system are obtained.
These equations are coupled nonlinear algebraic/differential equations for the
constraint force components and the dependent variables. The solution to the
equations is obtained by multiplying the matrix of dynamical equations by the
orthogonal complement of the matrix of the constraint equations. This
eliminates the constraint force components leaving a consistent system of
differential equations for the dependent variables (generalized coordinates)
of the system.
The principal analytical features of the method depend upon several
intermediate results and observations. The first of these is the fact that
even though the constraint equations are nonlinear, they may be differentiated
into a linear form in terms of higher derivatives of the generalized
coordinates. Next, it is observed that the velocity and angular velocity
vectors are linear functions of the first derivatives of the generalized
coordinates. The coefficients of these derivatives are the "partial velocity"
and "partial angular velocity" vectors used by Kane, et. al. When the velocity
and angular velocity vectors are differentiated these coefficients become the
coefficients of the higher order derivatives of the generalized coordinates as
occur in the differentiated constraint equations. Finally, it is observed
that the generalized constraint force array may be represented as the
transpose of the matrix of constraint equations multiplied by a constraint
force array.

N The method is illustrated with the classical problem of Appel and Hamel.

I
 SESON1
IUTBD DNMC
FRDY300-10
Jue3,18
I
U
I Nonlinear Analysis of Loss of Stability of Periodic Solutions
with an Application to Robotic Motions
i E. Lindtner, A. Steindl. H. Troger; Technische Universitit Wien

I
1 Introduction
I For many physical or technical systems the fundamental state is a periodic motion. Naturally this
periodic motion is supposed to be stable for the standard operating conditions and the correspond-
ing parameter values. However, it can lose its stability under variation of parameters at a critical
parameter value. Sometimes only this critical parameter value is of interest. Then a linear stability
analysis by means of Floquet's theory ([1]) is sufficient. However, there are many problems, for

3 example in the dynamics of robots ([2]), where one also wants to know how the system behaves
after a loss of stability of the fundamental periodic state. This, of course, requires a nonlinear
analysis.
We want to show how, in a systematic way, such an analysis can be given by making use of the
methods of bifurcation theory ([3,4,5]). The crucial step in doing a nonlinear analysis consists in
replacing the equations of motion given by differential equations by a system of difference equations.
This latter system is a point mapping called the Poincari map ([3,4]). It can be given by a power
series expansion in the neighborhood of the periodic solution, which is a ftxpoint for the map. The
coefficients of the power series expansion can be calculated numerically. If the Poincare map is
known then, in general, by means of Center Manifold theory ([6,7]) a further strong reduction of
the dimension of the problem is possible.

3 In this paper these two steps, not well known to engineers, will be explained with emphasis to the
practical calculations. Finally as an example the periodic motion of a robot will be studied.

1 2 Calculation of the Poincar6 Map

We assume the stability problem of the periodic solution yo(t) = y0(t + T) to be given in the form

ii = A(t.A)z + f 2 (z,t,A) - f 3 (z,t,A) (1)

where A(t,A),f 2 (z.t,A),f 3 (z, t,A) are periodic in t with period T and the N-vecors 12 and f3
contain the nonlinear functions in the variable z of second and third order respectively. (1) is
obtained by introducing y(t) = yo(t) - r(t) into the original system of equations of motion. Hence
the motion to be analyzed for stability is zo = 0. For a linear stability analysis of z0 Floquet theory
([21) could be used. But as it will be made clear below the calculation of the Poincari map also
includes the linear stability analysis supplied by Floquet's theory. The Poincari map is defined in
the neighborhood of the periodic solution by the map obtained from the transversal intersection of
the trajectories with a section surface. For example consider the trajectory in Fig. 1 leaving z, and
intersecting after one revolution in z 2 . Hence

Y
Mt
2
/
1 m 'g mn
1 2g

AM, x

Fig. 1: Definition of the Poincar map in the Fig. 2:DD-robot with 2 degrees of freedom
neighborhood of the periodic solution z0 0, (2, and prescribed circular motion of the
endpoint G

X2 = X(T',XI) = P(z1 ) (2)

where z(t,zl) isthe solution of (1)starting at z1. In order to calculate the mapping P in the
neighborhood of the periodic solution which is given by the fixpoint zo inthe section surface the
following power series expansion is introduced.

P(zo + )= P(zQ) + P'(zo) + !P"(oM)(,) +- P"(o)( ) +... (3)

2 6
where P(zo) = zo and P'(zo),P"(xo).P'(zo).... are obtained from the solutions of a series of
initial value problems ("8,9J). This follows immediately from (2) because

8z 10z 1 z T
x(T,o + ) + -o(T,o)+
((To)* 2 o),+ -- o(,O ) + . (4)

Hence comparing (3) and (4)

P'(zO) = ro),
z-(T, z
P"(zo) = . (5)

is obtained. For the calculation of - the differential equation (1) is written in the form i = F(z, t).
Taking the derivative of both sides with respect to zo a differential equation for of the form
 S(a °aFaz(

is obtained. To obtain the first term in (5) (6) must be integrated from t = 0 until t = T with
the initial conditions -o (0)= I, where I is the unit matrix. To calculate the next term in (5) the
derivative of (6) with respect to zo is taken, yielding

_)*(a2 F ( az 2 F
gated = 2 ;o+ =Z (T)

Again (7)must be integrated from 0 to T with the initial conditions °?--(0) = 0. Proceeding inthis
way also the coefficients of the third and higher order terms can be calculated.
The result of these calculations is a discrete dynamical system

I = L(A), + Q2(V,, A)+ Q3 (, , ,,A) +... (8)

where &, E RNV and A is a parameter vector. Q2,Q3,... are quadratic, cubic and higher order
terms in the variables. For example Q2 is a vector in RNv the components of which have the form
(N = 2) : ai,21 + 01,1C, + 7Y1C2
Returning now to the stability problem of z0. It is determined by the eigenvalues of the linear
part of (8). The periodic solution yo(t) or the corresponding fixpoint of the map is stable if all
eigenvalues of L(A) have absolute value smaller than 1. If only one eigenvalue has absolute value
larger than I then the fixpoint is unstable. If all eigenvalues have absolute value smaller than 1
except some with absolute value equal to I the system is just at the stability boundary.
We assume now A E R'. Then it is shown ([31) that generically at loss of stability only one of
the following three critical eigenvalues can occur: (i) III = 1, (ii) A2 = -1, (iii) A = V ± i17 with
!i3i = 1. If the calculation gives a more complicated case it can be made to disappear -y a small
change of other parameters in the system.

£ 3 Calculation of the point map on the Center Manifold

Following [6,71 it is possible to reduce (8) to a system of dimension k, which locally in the neigh-
borhood of the bifurcation point (A = Ac), describes the stability problem completely. k is equal to
the number of eigenvalues with absolute value 1. Hence for the one parameter bifurcation problem
either k = I for cases (i) and (ii) or k = 2 for case (iii) of the preceding section. As first step in the
reduction process L in (8) is transformed into Jordan form by setting y," = T-z, ([3,4,5,8,91).

3Y+1 = JYn + R 2 (yn, yn) + R3(Y, ,yn,

Y.) +... (9)

where J = TLT - ' is in Jordan form. The equations in (9) are ordered in such a way that the
first k equations correspond to eigenvalues of absolute value equal to 1 and the remaining N - k
to eigenvalues of absolute value smaller than 1. ".ence (9) can be written

I
 y,+1. J,~ + F(#.,n 5
y.+1,. = J.y.,. + (y..,,) (10)

where all the eigenvalues of J, are located at the unit circle and those of J. are inside the unit
circle. J, and J, are k x k and (N - k) x (N - k) matrices respectively. y,, are the active variables
or the amplitudes of the eigenvectors corresponding to the eigenvalues with absolute value equal to
1. y,,, are the passive variables which still show up in the equation for the active variables (10),
and have to be eliminated from them. This can be done by making an ansatz of the form ([6,71):
y,,,, = h(y,,,c), where h has the following two properties: h(O) = 0,h'(0) = 0. Hence if a series
expansion for h is made it starts at least with second order terms.
As an example for the practical calculation we pick the case (i) of the preceding section. Then (10)
takes the form

Y,+1,1 = yn,i + F(y, 1 ,y,,)

Yn~l,i = AYn,i + Fi(Yr,,in,i) i = 2,...,N

where for simplicity it has been assumed that the remaining eigenvalues Ai form a diagonal matrix.
From above follows

Y.,i = h,(yn,l) = Cai, 2 yn,1 + ai,y, 1 4... (12)

The coefficients aij. follow from introducing (12) into (11)2 and making use of (11)1 yielding

a,2 [ynj + F(y ,1 ,h,(y ,)] 2

+ 40,3[Y. + F1 (y.,,,hi(yn,l))]3 +... (13)
= Ai(ai,2 y, + ai, 2 yg.1 +-..) + F(y,,l,h(y,. 1 )).

If, for example, (11)i is only calculated up to third order terms then it is sufficient to calculate
only quadratic terms in h, which will be done now. Equating the quadratic terms in (13) results
in aij2 = (i = 2 ,.., N) where f:,2 are the coefficients of the quadratic terms in the Taylor
expansion of F,. One easily convinces himself that no other terms make a contribution.

4 Application to the motion of a simple robot

In '21 a DD-robot consisting of a planar double pendulum (Fig. 2) with moments acting at the hinges I
is studied. The motion of the endpoint G is supposed to be a circle with constant speed w = j.
Further introducing a control-loop which serves to compensate deviations from the prescribed path,
the equations of motion can be given in the form of (1) with N = 4. Calculating the Poincari map
according to section 2 results in a point map of the form of (8). In this problem depending on
the control mechanism by increasing the speed of the endpoint of the double pendulum all three
types of loss of stability can be found. Hence the following equations on the Center Manifold are
obtained (y,,,i = u,)

(i) U'+ 1 = (I+ )u + oau2+ 3 )

Ou,
(ii) ut,+ = -(1 + e)u, + au3 + O(Iunl s ) (14)
(iii) zn+l = (v - it7)(1 + f + az,,)zn
I
e is the unfolding parameter proportional to w and a and a must be calculated from the system data.
(i) is called transcritical bifurcation, (i) Flip-bifurcation and (iii) Hopf bifurcation ([3] p. 285). In
Fig. 3 the motion of the endpoint G of the double pendulum is shown for the different three cases
(14). For the transcritical bifurcation after loss of stability a shifted motion is obtained. In case of
the Flip-bifurcation a double periodic motion sets in. For the Hopf bifurcation a motion on a torus
is obtained.

I \.
" x

I Figure 3: Motion of the endpoint G after a generic one parameter loss of stability due to: (a) Tran-

I scritical bifurcation (p = 1), (b) Flip bifurcation (L = -1), (c) Hopf-bifurcation (14 = v ± i'7 ).

Acknowledgement
This research has been supported by the "Fonds zur F5rderung der wissenschaftlichen Forschung"
in Austria under project P 5519.

References
1. A. H. Nayfeh, D. T. Mook, Nonlinear Oscillations,J.Wiley and Sons, New York 1979.
2. E. Lindtner, A. Steind, H.Troger. Generic One-parameter Bifurcations in the Motion of a DD-Robot,
submitted for publication t,J. of Computational and Applied Mathematics.
3. V. I. Arnold, Geometrical Methods in the Theory of OrdinaryDifferential Equations, Grundlehren der
Math. Wissenschaften 250, Springer Verlag, Heidelberg, New York 1983.

4. J.Guckenheimer, Ph. Holmes, Nonlinear Oscillations, Dynamical Systems and Bifurcations of Vector
Fields, Appl.Math. Sciences 42, Springer Verlag, Heidelberg, New York 1983.
5. M. Golubitsky, D. Schaeffer, Singularitiesand Groups in Bifurcation Theory, Appl. Math. Sciences 51
Springer Verlag, Heidelberg, New York 1985.
6. J. Carr, Center manifold theory and its applications, Appi. Math. Sciences 35, Springer Verlag, Hei-
delberg, New York 1981.

7. G.looss, Bifurcation of Maps and Applications, Math. Studies 36, North Holland, Amsterdam 1979.
I 8.H. Amann, Gew6hnihche Differentialgleichungen.deGruyter Verlag, Berlin, New York 1983.
9. E. A. Coddington, N. Levinson, Theory of Ordinary Differential Equations, McGraw-Hill, New York

I 1955.
 I
I
I
U
I
SYSTEMATIC REDUCTION OF MULTIBODY EQUATIONS

OF MOTION TO A MINIMAL SET 5

Parviz E. Nikravesh

Department of Aerospace and Mechanical Engineering U

University of Arizona, Tucson, AZ 85721
I
This paper presents a two-step process to convert the equations of motion for

closed-loop systems from Cartesian coordinates to a minimal set of relative joint 3

coordinates. Initially, Cartesian coordinates are used to define the position of each

body, the kinematic joints, and the forces acting on the bodies. Prior to numerical 3
integration of the equations of motion, the equations are converted to a minimal set in

order to gain computational efficiency. It is also shown that the equations of motion 1
can be expressed in terms of the time derivative of the system momenta, instead of the
error and, in turn, to gain
accelerations, in order to reduce numerical integration

computational stability. 3
l
I

I'
I
a
5 Use of Linear and Nonlinear Structural Theories
in
Flexible Multibody Dynamics *

R. R. Ryant and H. H. Yoo*

Department of Mechanical Engineering and Applied Mechanics
University of Michigan, Ann Arbor, Michigan, 48109-2125
I Abstract

The present research involves an investigation of the roles of nonlinearand linear

elastic structural theories in accurately predicting transient dynamic behavior
of flexible multibody systems comprised by structuralelements undergoing arbi-
trarilylarge and arbitrarilyfast overall rotations and translationsas well as small
deformations. Coupling effects between deformation and overall motion are care-
fully scrutinized in the context of assumed-mode discretization techniques. Con-
sistently Linearized beam, plate, and shell formulations involving in-plane stretch
variablesare proposed and shown to yield very accuratesimulation resultsand ex-
tremely fast modal convergence for most motions involving small strains, In some
particularcases, however, in which membrane stiffness dominates bending stiff-
ness, a nonlinearstrain formulation is requiredin order to capturepropercoupling
between deformation and overall motion. Unfortunately, with standard com-
ponent modes, algorithmic formalisms involving nonlinear strain-displacement
expressions show very slow modal convergence. A procedure involving use of
constraint modes is proposed to alleviate this problem.

1. Introduction
Fig. I shows a proposed Earth-orbiting satellite consisting of a number of hinge-
connected rigid and deformable bodies which serve as rotors, antennae, solar panels, and
other structural components designed to accomplish various mission objectives. This satel-
lite is representative of a large class of systems known as flexible multibody systems, which
are characterized by interconnected structural elements undergoing large overall motion and
concomitant small deformation. Since these systems are extremely costly to deploy and are
often difficult to test realistically on Earth, more and more emphasis is being placed on the
prediction of dynamical behavior and the evaluation of active control systems via numerical
simulations.
Development of multibody dy: !rmic analysis formalisms aimed at facilitating such
simulations was begun in the mid-1960's[1,2] and has proceeded continuously to the present
time. Originally, these formalisms were restricted to joint-connected rigid bodies arranged
in specific topologies, but were later extended to treat flexible bodies, arbitrary topologies,
and arbitrary forcing functions. In fact, there now exists numerous nonlinear computational
algorithms (e.g., [3-7]) which were designed to treat quite general systems of rigid bodies,
deformable solids, and flexible structures in open- or closed-loop topologies.
Many of the present multibody formalisms and associated computational algorithms
are based on a flexible body model involving a three-dimensional continuum with mas
3 and stiffness distribution described in terms of modal data derived from a linear finite
element eigensolution. If the flexible body characterized in this way is not undergoing large
* This research work was partially supported by Mechanical Dynamics Inc.
t Assistant Professor
Graduate Student
i (1)

U
 1
overall motion relative to other bodies in the multibody system, and if the overall system
rotational motion remains small, then the formalisms can quite accurately predict system
dynamical behavior. Solving for component displacements in this case is equivalent to a
straightforward expansion in terms of the component modes. However, as shown in detail
in Refs.[8-10, the algorithms are very limited in their ability to accurately treat coupled
small deformation and large overall rotational motions of structural elements, regardless
of speed of overall motion. For example, consider a simple uniform, cantilever channel-
section beam undergoing a smooth slewing (repositional) maneuver from one orientation
to another orientation 1800 away. Suppose the axis of rotation is parallel to the cross-
section symmetry axis of the beam, and assume that the properties of the beam are as
I
specified in Fig. 2. Due to the offset between the centroidal axis and the elastic axis of
the beam, one would expect noticeable torsion of the beam as well as bending of the beam
in both principal planes of the cross-section during the maneuver. The correct solution is
shown by the solid curves of in-plane displacement u 2 (t), out-of-plane displacement u3 (t),
I
and torsional rotation 91 (t) in Fig. 2. Unfortunately, the flexible body model used in the
conventional multibody formalisms does not provide the capability to automatically model
the interaction of the mass offsets and the overall inertial forces necessary to provide an
I
accurate description of the dynamic behavior. The result of applying the formalisms is as
shown by the dotted curves in Fig. 2, where it is clear that both the out-of-plane response
and the torsional response are poorly predicted.
This example, as well as many others involving both slow and fast translational and
rotational overall motions of structural components of multibody systems, has pointed out
the necessity of treating each structural element type distinctly in both the multibody
model and the linear finite element model used to obtain descriptive modal data. It is not 1
sufficient to model a structure as a plate or a beam or an assemblage of such structural
elements in a linear finite element model and then use the resulting modal data in a flexible
multibody model wherein the details of the structural component are ignored in favor of a
simple continuum model.
I
These observations have led to vigorous new efforts to develop element-specific struc-
tural models within the framework of jointed multibody systems. These models must be
capable of treating the intracacies of the structure, representing the proper membrane and 1
bending stiffness, and providing for accurate coupling between overall motion and small
deformation. Recent research efforts in this area have generally followed one of three
paths, namely, (a)some form of physical discretization or substructuring combined with
finite element stiffness descriptionstl 1], (b)straightforward fully nonlinear finite element
I
procedures[12 ,13], and (c)modified assumed-mode representations. Our approach has been
to attempt to retain all of the attractive features of the modal methods, such a ease of
control design, model reduction, and solution; while building element specific models which
I
can be assembled in a multibody formalism.
To do this, we have investigated two possibile avenues. First, we have taken advantage
of the extensive literature and methods formulated for rotational analyses of rotorcraft and
turbomachinery and have developed a set of consistently-linearizedmodels for specific beam,
plate, and shell elements which allow for abritrarily general overall translation as well as
rotation, and permit general boundary conditions. These methods have been shown to yield
very accurate predictions of flexible multibody system behavior for most motions involving
I
small strain (see Fig. 3). These methods also provide rates of modal convergence which are
far superior to fully nonlinear strain-displacement based theories when a standard set of
assumed mode trial functions are employed. This is illustrated in Fig. 3, wherein a consis-
tently linearized beam formulation is used with four standard assumed modal functions to
model a spin-up maneuver of a uniform cantilever beam. This produces a solution which is
indistiguishable from the known correct solution. However, using the same set of assumed
modal functions with a fully nonlinear strain-based model yields the results labelled non-
l
(2) 3
U
I
linear in the same figure. Obviously, covergence has not yet been acheived. Studies with
many more modal functions in the nonlinear model have shown eventual convergence to
the known solution.
In attempting to treat large rotational motions of plate structures which are simply-
supported by rotating rigid frames, we found that the consistently-linearized theories yield
results which exhibit much too little membrane stiffness (see Fig. 4). To deal with problems
where membrane effects are expected to be of predominant importance, we have developed
an appropriate second-order nonlinear theory for various element types in the context of
multibody systems.
In order to overcome the convergence problems with the nonlinear methods, we have
developed a method involving nonlinear constraint modes which more closely match the
structural boundary conditions of the nonlinear models. For more details on these issues,
refer to [14].

REFERENCES
1. Hooker, W. W., and Margulies, G., "The Dynamical Attitude Equations for an n-Body Satel-
lite," The Journal of AstronauticalSciences, Vol. XII, No. 4, 1965, pp. 123-128.
2. Roberson, R. E., and Wittenburg, J., "A Dynamical Formalism for an Arbitrary Number
of Interconnected Rigid Bodies, with Reference to the Problem of Satellite Attitude Con-
trol," Proceedings '" International Congress on Automatic Control (London, 1966) London,
1967, pp. 46D.1-46D.8.
3. Bodley, C. S., Devers, A. D., Park, A. C., and Frisch, H. P., "A Digital Computer Program
for the Dynamic Interaction Simulation of Controls and Structure (DISCOS)," Vols. 1 & 2,
NASA Technical Paper 1219, May 1978.
4. Frisch, H. P., "A Vector-Dyadic Development of the Equations of Motion for N-Coupled Flex-
ible Bodies and Point Masses." NASA Technical Note TN D-8047, Aug. 1975.
5. Ho, J. Y. L., and Herber, D. R., "Development of Dynamics and Control Simulation of Large
Flexible
1985, pp.Space Systems," AIAA Journal of Guidance, Control, and Dynamics, Vol. 8, No. 3,
374-383.
6. Singh, R. P., VanderVoort, R. J., and Likins, P. W., "Dynamics of Flexible Bodies in Tree
Topology - A Computer Oriented Approach," Paper No. AIAA-84-1024, AIAA/ASME/ASCE
2511 Structures, Structural Dynamics and Materials Conf., Palm Springs, Ca., May 14-18, 1984.
7. Song, J. 0., and Haug, E. J., "Dynamic Analysis of Flexible Mechanisms," Technical Report
No. 55, Division of Materials Engineering, University of Iowa, May 1979.
8. Ryan, R. R.,"Flexibility Modeling Methods in Multibody Dynamics," Stanford University
Ph.D Dissertation, University Microfilms, Ann Arbor, Mi. 1986.
9. Kane, T. R., Ryan, R. R., and Banerjee, A. K., " Dynamics of a Cantilever Beam Attached
to a Moving Base," Journal of Guidance, Control, and Dynamics, Vol. 10, No. 2, March-April,
1987, pp. 139-151.
10. Ryan, R. R.,"Deployment, Pointing, and Spin of Active-Controlled Spacecraft Containing Elas-
tic Beam-like Appendages," Presented at the AAS/AIAA Astrodynamics Specialist Confer-
ence, Paper No. AAS 87-478, Kalispell, Montana, Aug. 10-14, 1987.
11. Haug, E. J., Discussion of Modal Substructuring, NASA/SDIO Workshop on Multibody Dy-
namics, Pasadena, California, Sept. 1-3, 1987.
12. Simo, J. C., and Vu-Quoc, L.,"On the Dynamics of Flexible Beams under Large Overall Motions
- the Plane Case: Part I and Part II," Journal of Applied Mechanics, Vol. 53, Dec., 1986,
pp. 849-863.
13. Christensen, E. R., and Lee, S. W., " Nonlinear Finite Element Modeling of the Dynamics of
Unrestrained Flexible Structures," Computers and Structures, Vol. 23, No. 6, 1986, pp. 8 19-829.

3 14. Yoo, H. H., Ryan, R. R., and Scott, R. A., "The Use of Assumed Modes in Large-Displacement
Elastodynamic Plate Behavior," ACC Paper, Atlanta, Ga., June 1988.

I (3)

U
 Fig. 1 - Representative Flexible Multibody System

SMOOTH SPIN-UP MOTION (0 - 6 rad/sec in 15 seconds)

2 liner
4 Oder N rj

Transverse
Tip
__________
Displacement -;.E
(meters) -

Consistently-linearized

Linear

B18 Time (seconds) -g

Fig. 3 - Uniform Beam Spin-up Maneuver

SMOOTH SPIN-UP MOTION (0 - 6 rev/mmi in 30 seconds)

Linear "

Midpoint LI (o)
Trasvrs Consistently- linearized
Displacement
(inches)Cort

6 2nd Order NonlinearU

a 1- 29 Time (seconds) 49

Fig. 4 - Simply-Supported Plate Spin-up Ma,1neuver

U SLOW REPOSITIONAL IANEUVER OF CHANNEL BEAM

Length:
Shear area ratios:
L=8 m
Young's Modulus: 02 =3.174
E=6.895EI0 N/m 2 3 =1.520
Torsional constant:
Shear Modulus: P = 2.4330E-II m 4
G=2.6519EI0 N/M 2
Mass per unit volume: Warping constant:
3
r = 5.0156E-13 m6
p=27766.67 kg/m
Cross-sectional area: Eccentricities:
C2 = 0
A=7.3E-5 M2
e 3 = 0.01875 m
Area moments:
12 = 4.8746E-9 M4 Modes Employed: 5
13 = 8.2181E-9 m 4 - "

I L
L

10 T~(t)= ~(t
1- - L.sin fft) rad 0 <t < 15 seconds
* 0 rad t > 15 seconds

Conventional Multibody Formalism Solution Conventional Multibody Formalism Solutic

I 1au0esr) I \ jSO I

II I In-plane DisplacementU 2 (meters) vs. Time (seconds)

Out-of-plane Displacement u3 (meters) vs. Time (seco:
,I
I
-- 20f
/ / \\ Correct Solution
Crtoi
-25

t Correct Solution
• /I

--
40

Torsional Rotation 01 (radians) vs. Time (seconds)

-20
I .3 Fig. 2 - Channel Beam Reorientation
e5
 ,I
U
U
Nonlinear Large Rotational Structural Dynamics 3
M. Iura and S. N. Atluri 3
Center for the Advancement of Computational Mechanics
Georgia Institute of Technology

SummlaI
The objective of this paper is to present a novel theory, and its computational imple-
mentation, for the analysis of strongly nonlinear dynamic response of highly-flexible beam
structures, and also to extend this theory to the case of highly-flexible shell structures.
The beam model used is based on Timoshenko's hypothesis; the effects of stretching,
bending, torsion and transverse shear are taken into account. This kinematic hypothesis
has been employed by many investigators (Antman and Jordan 1975, Reissner 1973, 1981,
and Simo and Vu-Quoc 1986). In these papers, the existence of configuration-independent
external moments has been postulated a priori. Argyris, Dunne and Scharpf (1978) and
Iura and Atluri (1987), however, have made the point that the external moments generated
by the conservative forces are generally configuration dependent. Therefore, the external
virtual work associated with the moments does not, on first sight, appear to correspond
to the first variation of an external energy functional. Argyris, Dunne, and Scharpf (1978)
have derived a nonsymmetric tangent stiffness matrix at the element level using the rota-
tional degrees of freedom referred to fixed axes of a global Cartesian system. Simo and
Vu-Quoc (1986) have concluded that using the variation of a rotational variable introduced
by Atluri (1984), the tangent stiffness matrix become symmetric only at an equilibrium
configuration, provided that no distributed external moments are assumed to exist. This
lack of symmetry (Argyris, Dunne, and Scharpf 1978) and the recovery of symmetry at
only an equilibrium configuration (Simo and Vu-Quoc 1986) have been attributed to the
fact that the finite rotation field is noncommutative. Iura and Atluri (1987), on the other
hand, have shown that the use of any three independent components of the finite rotation
tensor, as rotational variables, leads to a symmetric tangent stiffness matrix, not only at
the equilibrium but also the nonequilibrium configuration, even if the distributed external
moments exist in the problem. It should be emphasized that the rotation field remains
noncommutative.
The shell model used is based on Reissner's hypothesis; the membrane, bending and
transverse shear effects are taken into account. It is well known that the independent
parameters in this shell model are three translational and two rotational ones, while the
number of independent rotational parameters in the beam model is three. Based on this
fact, Basar (1987) has introduced the finite rotation vector with two ir- ependent param-
eters. The momentum equations of Basar are based on the unacceptabie assumption that
U
I
U
I
the Green strain tensor is a linear function of the shell thickness. Furthermore there are
physically uninterpretable terms which appear in the angular momentum balance condi-
tions. It is shown in this paper that, for certain choices of the rotation parameters, the
momentum equations derived from the energy method take on the same form as those de-
rived from the static method. For other choices of parameters the form of the momentum
balance equations is different but completely equivalent to that derived from the static
method. The resulting momentum equations are physically interpretable. As with the
case of the present beam theory, the external moment vectors caused by the conservative
forces are deformation dependent. In spite of this fact, we can derive the symmetric tan-
gent stiffness matrix for the shell element in a manner similar to that employed in the
beam theory.
The large deformation dynamics of a continuum body have been formulated with the
use of the total Lagrangian, updated Lagrangian, Eulerian, Euler-Lagrangian and the
moving coordinate formulations. The inertia effects are readily taken into account in the
total Lagrangian formulation. Therefore, we employ the total Lagrangian formulation for
both beam and shell theory. It should be noted that no simplification is made in the
present formulation; not only the rotatory inertia but also the Coriolis and the centrifugal
effects are accounted for.
Several numerical examples for transient dynamic responses are considered to demon-
strate the validity and applicability of the theoretical methodology developed in this paper.
A flexible in-plane beam in free flight is simulated in Fig. 1.
Aknowledgements

This work was supported by AFOSR and the SDIO/IST. The assistance of Ms. Deanna
Winkler in the preparation of this manuscript is thankfully acknowledged.

1. S.S. Antman and K.B. Jordan, Proc. R. Soc. Edinb. 73A (5), 85-105 (1975).
2. J.H. Argyris, P.C. Dunne and D.W. Scharpf, Comp. Meth. Appl. Mech. Engng., 14,
401-451 (1978).
3 S.N. Atluri, Comput. and Struct. 18, 93-116 (1984).
3.
4. Y. Basar, Int. J. Solids Structures, 23 (10), 1401-1415 (1987).
1 5. M. lura and S.N. Atluri, Computational Mechanics (In Press).
6. E. Reissner, Studies Appl. Math., Vol. 52, 87-95 (1973).
1 E. Reissner, J. Appl. Math. Phy (ZAMP) 32, 734-744 (1981).
7.
3 8. J.C. Simo and L. Vu-Quoc, Comp. Meth. Appl. Mech. Engng., 58, 79-116 (1986).

12
I
 T3

zZ0200. &APioooo. how. JOIO. H050. a-1o.3

tuG. 75

- - - - - - - t1 * 00U

- - t-.50U

t-1.61

t-1. 90

Fig. 1 Flexible in-plane beam in free flight. Time step AtO.O5 1

3
I
I

DYNAMICS AND BIFURCATIONS OF ELASTIC SPACE STRUCTURES

by Mark Levi,
Mathematics department
Boston University, Boston, MA 02215.

Abstract
Coupling of two well-understood classical systems - an oscillator and a free rigid
body in space produces a system which, on the one hand, is amenable to mathematical
analysis using the ideas and methods of geometric theory of dynamical systems, and on
the other hand, exhibits interesting and perhaps unexpected dynamical phenomena which
give a good indication of what is likely to occur in more complex systems such as beams,
platforms, etc. One such phenomenon is the bifurcation in which stable rotations become
destabilized and vice versa, and more than three pure rotations are possible with the same
parameters.

. ........
. .. ......
. . ... . . - - ,, nmm umumun
nn mu
 SESSION 11

FRICTION AND DAMPING

FRIDAY - 1300 - 1415

June 3, 1988
 THE EFFECT OF SOFTENING OF THE SUPPORTS ON
THE STABILITY AND DYNAMICS OF STRUCTURES

Marco Antonio Souza

Associate Professor
Departamento de Engenharia Civil
PUC-Rio Rio de Janeiro BRAZIL

ABSTRACT
A simplified model is used in order to illustrate how the
softening of the supports change the stability and the
vibration characteristics of structural elements liable to
buckling. Damping is not included in the analysis.

I NTRODUCTI ON
The accuate design of the support conditions represents an
essential part in any experimental set up. It is not unusual
to find out that experimental and theoretical results do not
coincide. There are a few reasons for this to happen, the
inacuracy of the theoretical assumptions being maybe the
most common. But any inadequacy in the design of the supports
to be used in the experimental set-up will also lead to the
very same conclusion that the theoretically predicted results
are different from the ones measured during the tests.

In order to highlight this fact, a very simple model is used

to show how the softening of the supports can affect both the
stability and the vibration characteristics of structural
elements liable to buckling. The stability and the
vibration characteristics will be illustrated by the
equilibrium paths and the characteristic curves, respectively.

THE RIGID-BARS MODEL

The simplified rigid-bars model of figure 1 has been
previously used by the author 11,2] to illustrate different
effects. This model reproduces the behavior of
beam-columrns, plates and shells.

L I L
Figure 1- The Rigid-Bars Model [l
The softening of the supports is modelled by the non-linear
springs S, figure 1, which have the characteristics shown in
figure 2, represented by siS./S versus the rotation at the
supports, e.
 I

Figure 2 - The Mon-linear Springs

The incremental stiffness function s=Sl§ to be used in the

present analysis is defined as followss
13 e e :5
S-1.0 C1)
s =1.0 + A + A
A[VK-8 e -34&
BA
Ki.(- +4 -8A -3 z2 (1-D)C-10 S A --
15A Sft ] Cl2

wheres

111) e E U
s=D C4)

Two cases of the incremental stiffness function a will be used

in the analysis and are shown in figure 3, corresponding to
p - -1.2Rad., DO.5 and D0.0.

1.2
-I
0.9 LEGEND
0=0.0
-
C

0.6
I

AnquaW RotaI.# (Rod.)

Figure 3 - The Incremental Stiffness Function a CVp-1.2Rad.) 1

I
I
U THEORETI CAL ANALYSIS
The change of th total potential energy, V, can be written

I as:
V 2CCe - Ci
32 * 0 .2)KLCsinr - sin99

[.Ce) ICe-e* ) de
IN C53

where 9 can be , 6, or 9 depending on the range

of the angular deflection e at the supports, i.e. e O e 5 U;

The equilibrium paths are obtained from the equilibrium

equation
V .0 CS)

and can be summarized ass

i) sinee

P s -C R 4 a 1 - i Cosa C73

m where:
where ( KLcC2
1a . 4C0; p=2PL/( KLa+4C;
2
R.a/C KL2 +4C0
C83

m ii 5- e - e
r sinU )o9 . Ce - f(-a3
p o i in J Cos 8 sin G 1 - ca +

4 C1 D2A'[RC-,10 15A - 6A) + *<C-8 + SA -3Az)]} C9)

where W, A , a, R are defined by equations C33 and

C8) respectively;

iii) & 2 9

si n e - e)
p M a S cose + Cl - a + DR C103
(, sinesine

Equilibrium paths for different values of the parameters R

and W for a a 0.0 are shown in figure 4 corresponding to
the stiffness function s of figure 3.

I
 j l 0OO~d.A.00
0.6I
t I 4m--=oaoa=05
3-0=0O604d.A=0.O0
.0 , -0 O O~
O RO.R
z0
S

g.0 0.? 0.. g-e 4.0 t.0 I- .* 4 i .@

AnuW b~ (Moa)

Figure 4 - Equilibrium Paths Ca a O.C)

The natural frequency of vibration corresponding to a load
level p is obtained by means of a small perturbation 6
from the equilibrium configuration represented by G. The
equation of motion is obtained from the Euler-Lagrange
equation 131:
T, + V, -0 (11)
6

where T is the kinetic energy Tu(1/2)I I * being the

generalized rotatory inertia, 6 is the velocity associated to
2
6 and C 3dC )'dt and neglecting terms of 0(6 ) can be written
as:
CKL IC. 4C3 f2 6 - 0 C123

where f is the square of the natural frequency given bys

f 2 (1 - cO * R* Ccos28e sngsine) - pcos8 (13)

0

113 :5 e 1
f2 f2 R[ 1 - A 3 CI-D)C40 - 75A' + 36A&)] (14)
li) e > I

f2 f 20 - RC1 - D) (15)

Characteristic curves relating the applied load and the square

of the natural frequency are shown in figure 5, and correspond
I
U
I
to the equilibrium paths shown in figure 4.
L£G m0
a:001td ::z0

S0
&---a:0 00
001 :XO0
3.00

t.2

I1- 'a
.0Wa4aX05.D-

I .
~.e
I.I I .~ 4. ~~ ~Load Farm .. w

Figure 5 - Characteristic Curves (a - 0.0)

CONCLUSIONS
The results presented illustrate how the softening of the
supports affects the stability and the vibration
characteristics of structural elements liable to buckling.
This is clear from the comparison between the curves, figures
4 and 5, corresponding to R0.0 and R-0.0 in the range of
F e <U.
The implications of the results on experiments are clear:
unless the actual support conditions are properly designed the
results obtained can exhibit a different pattern from that
expected theoretically. It will not necessarily mean that the
theory is not accurate only that the theoretical and the
experimental supports do not have the same characteristics.

IACKNOWLEDGEMENT
This work was done during the two months visit to the
Department of Civil Engineering of the University of Alberta,
Canada, to which the author is deeply thankful.

REFERENCES
1. Souza, X. A., Post-buckling Vibration Characteristics of
Structural Elements, Engineering Structures,

Vol. 9, pp. 134-137, 1987.

2. Souza, M. A., The Effects of Initial Imperfection and
Changing Support Conditions on the Vibration
of Structuralof Structural Elements Liable to
Buckling, International Journal of
THIN-WALLED STRUCTURES, to appear, 1988.
3. Clough, R. W. & Penzien, J., Dynamics of Structures,
McGraw-Hill Book Co., 1975.

I
 I
JOINT DAMPING AND JOINT NONLINEARITY EFFECTS IN DYNAMICS O
SPACE STRUCTURES

Mary Bowden
John Dugundji

Space Systems Laboratory

Department of Aeronautics and Astronautics
Massachusetts Institute of Technology
Cambridge, MA 02139

The presence of joints can strongly affect the dynamics of

space structures in weightlessness, especially if the joints are I
numerous, of low stiffness, have damping, or are nonlinear. In
order to obtain an idea of these joint effects, a simple free-
free system of four beams connected by three joints is
investigated, as shown in Figs. 1 and 2.
Linear analyses of these beams were performed by using a
standard finite element formulation and including various linear
joint stiffness values kL and linear joint damping values c .
L
These analyses were formulated in the standard form,

M q + c + K q - F (1)

and then reduced to state space form,

x - A x + P (2)

for ease in obtaining solutions. The corresponding lowest six

symmetric modes and frequencies for no damping, cL - 0, are
shown in Fig.3 for a finite and an infinite joint stiffness.
The effect of including joint damping cL on these modes is shown
in E ;.4. Increasing the joint damping increases resonant
frequencies and modal damping, but only to the point where the
joint gets "locked up" by the damping and approaches a
continuous beam. This behavior is different from that predicted
by proportional damping, as shown in Fig.5. The maximum amount
of passive modal damping obtainable from the joints is greater
for low stiffness joints and for modal vibrations in which many
joints are participating. A joint participation function JPF,
based on geometrical arguments of joint location, was used to
quantify this phenomenon.
Nonlinear effects of the joints were introduced by using
describing functions to represent the first harmonic of the

1 I
I
 nonlinear joint forces. Assuming sinusoidal motion q = A sin
where *- (at, the nonlinear symmetrical joint force FNL(q,4) was
expressed as,

F L a sin 0 + b cos - - a q + be
-q
NLA ACI)q
-cPq + cq (3)

where c and cq are frequency and amplitude dependent stiffness

(phase) and damping (quadrature) coefficients defined as

21

CI
P =A FNL(A sin 0, AC cos 0) sin 0 do
ICA 0(4)

cq = f FNL (A sin 0, AO cos 0) cos 0 dO

A joint with a simple cubic spring nonlinearity FNL- kLq + kcsq 3

was used to illustrate nonlinear behavior, although other

nonlinearities (free play, coulomb friction, etc.) were also
considered. Figure 6 shows the typical forced response of a
single degree of freedom system with a cubic spring. The
response plots nondimensional amplitude - A rX' rather than A
in order to collapse all results onto one universal curve. The
forced response of the four beam model with cubic spring joints,
to a vertical harmonic excitation at its center was then
studied. Figure 7 shows typical log-log responses for the beam
with linear spring joints, while Fig.8 shows the corresponding
responses for the nonlinear cubic spring joints. These were
computed using Newton-Raphson method with appropriate initial
displacements. The nonlinear analyses show the classical single
degree of freedom nonlinear behavior at each resonance:
multiple solutions, jump behavior, resonant frequency shifts,
and non-doubling of response for doubling of forcing amplitudes.
These properties are illuminated by characteristic backbone
curves, which show the locus of resonant peaks for increasing
forcing amplitudes. These peaks shift from the small amplitude
(linear) frequencies to the large amplitude)locked, continuous
beam) frequencies and depend on the amount of joint
participation. A modal coupling due to Joint nonlinearity is
also exhibited, as was the case for linear joint damping.

The present studies help to illustrate how multiple

discrete nonlinearities interact with the global dynamics of
continuous systems. The jointed beam model studied here can be

!2
I
interpreted as truss bays with linear characteristics, and the
joints as bay interfaces with nonlinear characteristics.
Acknowledgement

This work was supported in part by NASA and in part by

McDonnell Douglas Astronautics Company.
References

1. Bowden, M.L., and Dugundji, J., "Effects of Joint Damping

and Joint Nonlinearity on the Dynamics of Space
Structures," AIAA SDM Issues of the International Space
Station Conference, Williamsburg, VA, April 21-28, 1988,
AIAA paper 88-2480.

2. Bowden, M.L., "Dynamics of Space Structures with Nonlinear

Joints," M.I.T. Department of Aeronautics and Astronautics,
ScD. Thesis, May 1988.

3. Crawley, E.F., and O'Donnell, K.J., "A Procedure for

Calculating the Damping in Multi-Element Space Structures,"
27th Congress of the International Astronautical
Federation, Innsbruck, Austria, October 1986.

4. Den Hartog, J.P., Mechanical Vibrations, McGraw-Hill, 4th

Edition, New York, 1956.

5. Woolston, D.S., Runyan, H.L., and Andrews, R.E., "An

Investigation of Effects of Certain Types of Structural
Nonlinearities on W.Lng and Control Surface Flutter," J_ of
Aero. Sciences, Vol. 24, No. 1, January 1957. See also
Vol. 26, No. 1, pp. 51-53, January 1959.

6. Breitbach, E., "Effects of Structural Nonlinearities on

Aircraft Vibration and Flutter," NATO AGARD Report R-665,
January 1978.

3
I
I
S3 $ e 0,

it 6f
6q 7 JI q7

f q6

*j q7 - q6

i Figure I Three Joint Mlode Fgue 2 Joint Force

i
I d. L~ tS-I

9. I:::
?. 5115.

"_ ,_ _ _ _' _ 5
.O.83
• , 1 . .n

V ,t\ S -

S' Figure 4 Root Locus for First 6 Symmetric Modes

9.4 v
I -I

I, *
74.311

~*l
.' i4&P1
*
0B0
,
h34 ,efl
deI~lS90
_

.
_

.
_

.
_ _ -

O,I ,. U/L ,-. l~dl - -0l4 - -I* -*3 - * ,.,

-4

I Figre ,, Jonted and Continuous Beam Modahopa Figre S Nonprpoiia Dampin

I
I ln AI I -
 Ct~bic Spring squiaou of M==o

0./
1.2~~~M wcbi tUlow
.~ indreu=
+;+ ~ nvvlfr
' 0~

Figuredo
Cubicin6h
Spin Nolieait (I do()

0 oce ! dm O ;',:" 0. o ffed # 3

Figure7Resonse fLiguare 6 on ytmFgr Cubic d

NoSiystem( epneo Spring
I
I
IEFFECTS OF VELOCITY-DEPENDENT FRICTION ON
PERFORMANCE OF A RESILIENT-FRICTION BASE ISOLATOR

Lin Su, Goodarz Ahmadi

Department of Mechanical and Industrial Engineering
Clarkson University, Potsdam, NY 13676
and
3 Iradj G. Tadjbakhsh
Department of Civil Engineering
I Rensselaer Polytechnic Institute, Troy, NY 12181

* ABSTRACT
Performances of a resilient-friction base isolator (R-FBI) as an aseismic bearing under
a variety of conditions are analyzed. In particular, the effects of velocity-dependence of
friction coefficient on the overall performance of the R-FBI system are studied. Based
on experimental data, two expressions for velocity-dependence of friction coefficient are
assumed. A nonuniform shear beam structural model is considered and the accelero-
gram of the NOOW component of El Centro 1940 earthquake are used as the earthquake
excitation. The presented results show that for a finite static friction coefficient the
velocity-dependence of dynamic friction coefficient has no noticeable effects on the re-
sponse spectra of the base-isolated structure. On the other hand, a zero static friction
coefficie-nt leads to significant differences and improves the effectiveness of the R-FBI
3 system to a considerable extend.

INTRODUCTION
I Using base isolation systems for aseismic design of relatively stiff structures has at-
tracted considerable interest in the recent years. The main concept is to isolate the
structure from ground during earthquake strong motions. Excellent reviews on the sub-
ject were provided by Kelly [1,2]. Recently, an interesting frictional base isolation system
(R-FBI) was introduced by Mostagel and Khodaverdian (3]. This isolator consists of
concentric layers of teflon coated plates that are in friction contact with each other and
contains a central core of rubber. The system provides isolation through the parallel
action of friction, damping and restoring springs. This design essentially uses a rubber
I bearing and a pure-friction isolator in parallel. Figure 1 shows a schematic diagram of
mechanical behavior of the R-FBI system.
For the friction-type base isolators, the friction coefficient is an important parameter.
In most earlier studies, a constant coefficient of friction in according to Coulumb's law

31
I
was used for response analyses. However, recent experimental data [4,5] suggest that the
friction coefficient is not a constant and varies with velocity, normal pressure, and other
parameters.

Figure 1. Schematic diagram of the R-FBI systems

In this work, a shear beam model for structure is considered and the acceleration record
of El Centro 1940 earthquake is used. The performances of the R-FBI system for different
structural systems are analyzed. Particular attention is given to the effect of velocity-
dependence of friction coefficient. It is shown that for a finite static friction coefficient
the velocity-dependence of dynamic friction coefficient has no noticeable effects on the
response spectra of the base-isolated structure. On the other hand, should a negligible
static friction coefficient for teflon-teflon or teflon-steel interfaces be substantiated, it
could significantly alter the behavior of the frictional base isolators.

TECHNIQUE OF ANALYSIS
The equations of motion of a nonuniform shear beam structure with a base isolation
system subject to an earthquake excitation are described at length in [61 and hence is not
repeated here. It suffices to point out that the first ten modes of vibration are used in
the response analysis. The computer program developed in [6] for numerical integration
of equations of motion is modefied and is used in this study.
A modal damping coefficient of 0.02 for the structure, a nonuniformity coefficient of
0.1 and a mass ratio of 0.75 are used. The recommended values of parameters for the
R-FBI system (p, = 0.04, (, = 0.1 and natural period of 4.0 sec) are employed. The
accelerogram of the NOOW component of El Centro 1940 earthquake is used as seismic
excitation. The peak relative displacements and the maximum absolute accelerations of
the base-isolated shear beam structure at its base raft and its roof under a variety of
conditions are evaluated.

VELOCITY- DEPENDENT FRICTION

Recently Constantinou et al. [5] presented a series of experimental data for frictional
characteristics of teflon-steel interfaces. The following two expressions for the velocity-
dependence of friction coefficient given as

I
I
U /J 0.02 + cl(u/n),(i

= c2 (v/n) + C3 (v/n)2 + C4(v/n) (2)

I are fitted to the data of Constantinou et al. in [5]. Here, v is the slip velocity, n is the
number of friction plates used in the R-FBI system (n = 8 is used in the analysis), and
the values of c's are given as

c, = 1.0762 x 10- 3 sec/cm, c2 = 2.5086 x 10- 3 sec/cm,

C3 = -3.1316 x 10- 3 sec 2 /cm 2 , c4 = 2.2545 x 10- 7 sec 3 /cm 3 . (3)

U Figure 2 compares the predictions of equations (1) and (2) with the experimantal data.
Equation (1) assumes a linear relationship between 1t and the slip velocity with a static
friction coefficient of 0.02. Constantinou et al. [51 have also reported that continuous non-
stick sliding occured in their experiments, which implies that 14 = 0 for v = 0. Equation
(2' satisfies this latter condition and leads to a zero static friction coefficient.

I ~Z40.08
r~x.]DATA
o0.06

- 0.0°4
0.04 -,'- EQ.() o
UU
ir,.w 0.02 - / E-
Q.(2)
0.02

0 10 20 30 40 50 80
VELOCITY (cm/sec)
3 Figure 2. Velocity-dependence of friction coefficient

3 RESULTS
For friction coefficient as gven by equations (1) and (2), as well as a constant value
of 0.02, the response spectra of the base-isolated shear beam structure versus its natural
period T, are evaluated. The results are shown in figure 3. It is observed that the response

* 3
 I
I
spectra for it given by equation (1) are almost the same as those for a constant friction
coefficient. That is, the velocity-dependence of friction coefficient as given by equation
3
(1) does not affect the peak responses. This figure also shows that the response spectra
obtained by using equation (2) defer significantly from those obtained for a constant
friction coefficient. The peak deflection and the peak acceleration are lower by a factor
of 2 to 4 and the peak base displacement is higher by about 10 to 30 percent.

L
z4.0 -0..

S.()c / ,, ...

~2.0
10
: 0.2 -3640

1.0-
TI (sec)

<1
I

0.1I
Esec)
Z! TI

0 0.0 0. 4 0. o.s 1. 0 04 0.16 0. 1.0i

Ti (se) Ti (see)

0.1

I
EQ.((2)

0.0 0.2 0. 6 0.1

6
0.0
s 1.0o 0.0 0.2 0.4 0 .6 0. s 1.0o
TI (sew) TI (see)
Figure 3. Variations of the peak responses of the structure with its natural period for
the R-FBI systems with velocity-dependent friction coefficient

CONCLUSIONS 3
From the presented results, it may be concluded that the response spectra of the
structure with the R-FBI system are not sensitive to the velocity-dependence of friction
coefficient for a nonzero static friction coefficient. However, should it be proved that
continuous nonstick sliding with zero static friction coefficient occurs for certain inter-
I
4
I
I
faces, then the velocity-dependence of /j can significantly affect the peak responses of
the friction-type base isolators. The effects, in this case, are generally favorable and re-
sults in significant reductions in the peak deflection and the maximum acceleration of the
structure. At the present time, however, the available meager data are inconclusive and
additional experiments are needed.

ACKNOWLEDGMENT
This work is supported by the National Center for Earthquake Engineering Research,
State University of New York at Buffalo under the Grants No. NCEER 872007.

I REFERENCES

1 1. Kelly, J.M., Aseismic Base Isolation, Shock Vib. Dig., 14 pp. 17-25 (1982).

3quake2. Kelly, J.M., Aseismic Base Isolation: Review and Bibliography, Soil Dyn. Earth-
Engng., 5, pp. 202-216 (1986).
3. Mostaghel, N. and Khodaverdian, M., Dynamics of Resilient-Friction Base Isolator
(R-FBI), Earthquake Engng. Struct. Dyn. 1j, pp. 379-390 (1987).
4. Kragelskii, I.V., Friction and Wear, Butter'worths, (1965).
5. Constantinou, M.C., Caccese, J. and Harris, H.G., Frictional Characteristicsof
Teifon-Steel Interf-e-s under Dynamic Conditions, EarthquakeEngng. Struct. Dyn.,
15, pp. 751-759 (1987).
6. Su, L., Ahmadi, G. and Tadjbakhsh, I.G., A Comparative Study of Performances
of Various Base Isolation Systems - Part I: Shear Bcam Structures, Report No.

1 MIE-153, Clarkson University, October 1987.

1
I
I

I
 I
I
NONLINEAR DYNAMIC RESPO:SE OF FRACTIONALLY 3
DAMPED STRUCTURAL SYSTEMS

Joseph Padovan i
Dept. Mechanical and Polymer Engineering
The University of Akron
Akron, Ohio 44325 I
Introduction 3
The modelling of structural damping has been a long standing prob-
lem. Most typically, differential type simulations have been employed
to represent such behavior, i.e. Kelvin-Voigt, Maxwell, as well as com-
bined models. Interestingly, this is in spite of the fact that such
I
simulations do not generally define the proper frequency sensitivities
over wide ranges of spectral input. As a result of this, alternative
formulations have been sought. Generally, this has meant the develop-
ment of nonlinear type representations. To extend the range of validity
I
of differential type formulations, the integer derivatives have been
recently replaced by fractional integro differential operators, i.e. of
the Liouville-Riemann form. Such a formulation enables the powers of
I
the various derivatives to be cast in terms of experimentally derived
fractional numbers. Such expressions yield better results over wider
spectral ranges than the usual integer version. Note while fractional
operators yield significant modelling advantages, they are awkward to
I
handle both analytically and numerically [2].

In the context of the foregoing, the presentation will develop effi-

cient and stable numerical schemes enabling the solution of the nonlinear
dynamic response of structure with viscoelastic components, as well as
discretely attached dampers. The damping will be modelled by fractional
integrodifferential operators of the Grunwald type (3]. To generalize
the results, the numerical analysis will be generalized to FE type sim-
ulations.
I
Overall the dampers treated fall into three categories, namely
i) Those attached to external support structures;
ii) Internal dampers which may be linked between various components
of the given structure; and,
iii) Viscoelastic type material behavior.

For the fractional formulation of such problems, the sclution development

consists of two levels, i.e. the establishment of the numerical approxi-
mation of the fractional operator and the glob-l level transient algorithm.
I
Work partially supported by NASA Langley under grant NAG--144. 3
I
I
I
I To generalize the results, the algorithm will be developed for nonlinear
structural simulations involving large deformation kinematics.

For the presentation, detailed discussions will be given on:

i) The numerical approximation of fractional operators;
ii) The FE simulation of nonlinear structure with fractionally
defined dampers; and,
iii) Outlining the benchmarking experiments defining the stability
and efficiency of the overall development.

Generalized Fractional Integrodifferential Dampers

As noted earlier, traditional dissipative characteristics are de-

fined by the following differential type expression, namely
d d2 (
1D Clt
l (Y) + C2 dt (i)

where FD, Ci and Y respectively represent the damper force, damping co-
efficient, and deflection. It is well known that due to the proportional
I nature of (1), C i apply only for a small range of exciting frequencies.
To extend the range of application, dn/dtn ( ); ne[l,..] can be replaced
by fractional operators, i.e. [21
=
F- D iC. Dq( (2)

where Dqi( ) is defined by the Riemann-Liouville relation, namely

- t (-!_.qi+ l (
i( r(qi) Y (t-'r

such that r is the gamma function. The main draw back to employing frac-
tional simulations lies in the fact that from a purely analytical point
of view, such operators are somewhat cumbersome to handle. To bypass
such a shortcoming, we shall employ an alternative but equivalent defi-
nition developed by Grunwald. In particular

-1 N-1
54I Dqi(Y) ^ lim (-) Z (-l)J(9)T(t-jt/N)
N j=0
(4)

The foregoing definition can be extended to represent either an

integral or differential type operator. This is achievea by latting q
range over both positive and negative numbers. As will be seen from the
presentation, (4) can be approximated by employing a finite series where
At is discrete.

I
U
 iI
2
To generalize the results, the algorithm will be developed for nonlinear
structural simulations involving large deformation kinematics.

For the presentation, detailed discussions will be given on:

i) The numerical approximation of fractional operators;
ii) The FE simulation of nonlinear structure with fractionally
defined dampers; and,
iii) Outlining the benchmarking experiments defining the stability
and efficiency of the overall development.

Generalized Fractional Integrodifferential Dampers 3

As noted earlier, traditional dissipative characteristics are de-
fined by the following differential
2
type expression, namely
d
F C ( Y) +
C _ ( Y) + ()
!D ' dt- 2 2 •
where FD, Ci and Y respectively represent the damper force, damping co-

efficient, and deflection. It is well known that due to the proportional

nature of (1), Ci apply only for a small range of exciting frequencies.
To extend the range of application, dn/dtn ( ); nc[l,..] can be replaced
by fractional operators, i.e. (2]
I
FD D= iC qi- (Y)
Dq. (2)

where Dqi( ) is defined by the Riemann-Liouville relation, namely

Dqi(Y)- qi)f (-) Y(t)dr (3)

such that r is the gamma function. The main draw back to employing frac-
I
tional simulations lies in the fact that from a purely analytical point
of view, such operators are somewhat cumbersome to handle. To bypass
such a shortcoming, we shall employ an alternative but equivalent defi-
nition developed by Grunwald. In particular

j (4)
Dqi(Y) % lim (L)-1 (-I )T(t-jt/N)
N-- j=0

The foregoing definition can be extended to represent either an

integral or differential type operator. This is achieved by letting q
3
range over both positive and negative numbers. As .ill be seen from the
presentation, (4) can be approximated by employing a finite series where
At is discrete.
I
I
I
 3
U
3 Finite Element Formulation

The main emphasis of the presentation will be to consider the analy-

sis of nonlinear structure containing varying types of dampers. These
shall be modelled by fractional integrodifferential type operators. The
system nonlinearity treated is a direct result of kinematic, material and
boundary condition induced nonlinearity. In this context, the FE model
will employ the 2nd Piola Kirchhoff stress and Lagrangian strain tensor
combination of measures. The overall "fractional" viscoelastic FE form-
ulation will be established via the virtual work principle.

The solution of the resulting FE model will be developed via an

implicit type formulation. As noted earlier, this consists of two oper-
ational levels, i.e. the fractional operator representation of the damp-
ing characteristics and secondly, the remaining formulation involving
the structural stiffness, boundary conditions as well as inertial effects.
The fractional operator is approximated by an appropriately truncated
Grunwaldean representation. Note truncation is controlled via self-
adaptively updated remainder expressions. These streamline the use of
the Grunwald representation. The results of this level are employed
in conjunction with the implicit transient formulation of the inertial
terms. Overall this yields an incremental time stepping solution. For
demonstration purposes a least square type [3] Newmark Beta type expres-
sion is used to represent the appropriate inertial fields. Due to the
generality of the fractional formulation, both stress and strain rate
dependencies can be handled, i.e. Maxwell-Kelvin-Voigt type representa-
tions.

3 Benchmarking

Based on the Grunwaldean Newmark Beta type implicit formulation, an

incremental algorithm will be developed to enable the nonlinear dynamic
solution of structure containing dampers modelled by more comprehensive
fractional type operators. To benchmark the stability, efficiency and

3 capabilities of such algorithms, the results of several numerical experi-

ments will be presented, namely:

i) The simulation of a full scale rolling vehicle traversing

rough terrain; and,
ii) The response of various structure with damped viscoelastic
coatings.

The simulations will be used to establish the unique modelling features

of fractional type operators.

5 References

1. Ross, B., Lecture notes in Mathematics, Vol. 455, Springer. Berlin,

1975.

I
I
 1987.
J., Computational Mechanics, Vol. 2, pp 27l-2S.
2. adoval,
(in press'.
FE in Analysis and Design
3. Padovan, J., Jr. of
1
I
I
5 PERIODIC RESPONSE OF A CLASS OF HYSTERETIC OSCILLATORS

Danilo Capecchi and Fabrizio Vestroni

I Dipartimento di Ingegneria delle Strutture, Acque a Terreno

Universita' dell'Aquila, 67040 Monteluco di Roio (AQ), Italy

Introduction

A great number of papers, some very sophisticated, have

been devoted to the response of nonlinear systems under
sinusoidal excitation. Most of them concern the large-
amplitude oscillations of structural systems; they can be
described by simplified models which exhibit nonlinear force-
deflection relationship of holonomic type (1].
Only a few studies, however, dealth with the dynamic
analysis of hysteretic oscillators (e.g. Refs.2-6); the
earlier ones concerned bilinear constitutive relationships,
and subsequent ones more general yielding cases. Several
models, whose stationary dynamic behaviour does not seem
altogether enlightened, have been recently introduced to
represent the hysteretic characteristics of different physical
systems under severe excitation.
The aim of this paper is to explore this matter more
throughly by studying the steady-state oscillations of a class
of simple but general hysteretic systems.

I Solution of steady-state response

Let us consider a hysteretic oscillator with a restoring

force f, viscous damping and a frequency wo of small
* amplitude under a sinusoidal excitation F cos wt. If fy is the
maximum value of f and x the displacement at nominal yielding
such that 7 = fy/mxy, ty introducing the non-dimensionalized
quantities:

i = X/Xy,. f = f/fy, F = F/fy, t = o t, a /oo (1)

the equation of motion reads:

kx + 2 " x + f(x) = F cos at (2)

where the tilde has been omitted.

Along a stable cycle the force-deflection relation f(x) has
to be described by two single-valued funtions, fl(x) and f2 (x)
U
for the upper (x>o) and lower (x<O) side of the cycle
respectively. Typical examples are the bilinear model and the
Ramberg-Osgood model, studied in [2,3].
Due to the non-holonomic character of the restoring force,
the steady-state response is obtained by means of an integral
procedure. The solution is sought for in the form:

x(t) = EnAncos nat + ZnBnsin nat (3)

Substitute eq. (3) into eq. (2) and expand the periodic function
f[x(t)] in a Fourier series; An, Bn are determined by equating
the coefficients of the same harmonics.
It is assumed here that n=l and damping is neglected; the
amplitude X and the phase 0 of the solution can be obtained
from:

C(X) - a'X ] + S(X)' F' (4)

tg 0 = (5)
C(X)-aX
where:
C(X) = - f(X cos e) cos e d9 (6)
I J0

S(X) = -J f(X cos 9) sin 8 de (7)

Eq.(4) completely describes the relation h(X,a,F) among the

oscillation amplitude, the intensity of the force and its
frequency.
For the bilinear (EPL) and Ramberg-Osgood (ROM) models the
solution has been obtained in [2,3] and the frequency-response
curves (f.r.c.) are plotted in Fig.l. The main features of the
response are:
a) the 'soft' resonance exibited by the systems
b) the single valued and stable nature of the response curves
c) the occurrence of unbounded resonance above a certain value
of force intensity.
In comparison with nonlinear elastic oscillators the
existence of a maximum value in the restoring force is
responsible for the unbounded amplitude resonance, while the
hysteresis seems to make branches of unstable solutions
disappear, although multi-valued f.r.c. have been obtained in
[4] for a particular hysteretic model.
U
I
I
The stability analysis, performed in many works on periodic
oscillations of nonlinear systems, have shown that when the
response amplitude is multi-valued the part of curve between
two loci of vertical tangency is a region of unstable
solution.

Frequency-rssponse curves

To gain insigth into the behaviour of hysteretic models and

to better ascertain the role of the hysteresis, it is
convenient to develop a qualitative study of the eq.(4). Since
it is clear that the solution is governed by the pattern of
C(X) and S(X), to make reference in the analysis tc a range
that is wide but at the same time meaningful, a class of
hysteretic models, illustrated in Fig.2, has been considered.
Due to linearity of the single branch and a very low number of
parameters, the model is simple and sufficiently general so as
to represent some characteristics of stiffness-strength
degrading force-deflection laws of real engineering systems.
The parameter Q is the post-elastic stiffness which can be
positive or negative; # governs the magnitude of hysteresis,
that is minimum for f =0, while the model becomes bilinear
for #=1. When @ and # change, C(X) and S(X) change as well;
they are plotted in fig. 3 for three different cases. It is
worthing to notice that the functions are regular and S(X) may
or may not decrease.
In Fig. 4 the frequency-response curves for the model are
illustrated with various values of g and o (the latter is
always positive). If p is even a little smaller than unity the
curve becomes multi-valued, showing that the curve of bilinear
model is marginally stable. The frequency range of unstable
solution, for a given force level, becomes wider as
J decreases and tends to disappear for high g values.
Notwithstanding the simplicity of hysteretic model, a rich
variety of the periodic response, depending on the pattern of
C(X), S(X, and on the value of F is evidenced. Study of eq.(4)
leads to the conclusion that the number of the roots of the
I equations:
C(X)' + S(X)' - F , S(X)' -2 F (8)

3 governs the intersections of the curve with the a =0 axis,

(eq.81 ) and with the backbone curve the second (eq.82 ).
For the model considered four different circumstances,
illustrated in fig. 5, are possible according to the number of
the roots of eqs.8 reported in Tab.l. In case (a) S(X) is
I
monotonic, while in the others it is not monotonic; the curves
(b),(c) and(d) refer to the same oscillator for increasing F
values. In cases (b) and (c), there is always a frequency
range where the curve is multi-valued and unstable solutions
occur. In case (a) this can occur depending on the parameters
of the model, in any case it happens for p < 1; curve (d)
seems to be always stable.
Fig.6 describes the relation X-F for different frequency
values in the range a<l for an assigned model. On this plane
for each frequency the value of F above which the response is
unbounded is evident as well as the range where the curve is
multivalued.
A detailed study of the stability of the steady-state
solutions of the hysteretic systems presented is a natural
development of the present work.

References

1. Nayfeh, A.H., and Mook, D.T., Nonlinear Oscillations, John

Wiley & Sons, Inc.,New York,NY,1979.

2. Caughey, T.K., "Sinusoidal excitation of a system with

bilinear hysteresis", Journal of Applied Mechanics, 27, Dec.,
1960,pp.640-643.

3. Jennings, P.C., "Response of a general yielding structure",

Journal of Engineering Mechanics Division, ASCE, 90, No.EM2,
Apr., 1964, pp. 131-166.

4. Iwan, W.D., "The steady-state response of the double

bilinear hysteretic model", Journal of Applied Mechanics, 32,
1965, pp.921-925.

5. Masri, S.F., "Forced vibration of the damped bilinear

hysteretic oscillator", Journal of the Acoustical Society of
America, 57, Jan.,1975, pp.105-112.

6. Capecchi, D., and Vestroni, F., "Steady-state dynamic

analy3is of hysteretic systems", Journal of Engineering
Mechanics, ASCE, 111, Dec.,1985, pp.1515-1531.

I
 3 - X0.00

3 - -

F=0.75.0.85.1.0
EPL 4.0
---
- 0.00.0.25,1.00
0.25
6.0. 10Q 0.10.0.20A040

*~00 _ _ _ 0.0.__
_ _ _

0.00 0.25 0.50 0.75 10 .51.50 0.00 0.25 0.50 0.75 1.00 1.25 1.50

Fig. 1 Fig. 2

5( 1.25* Sx) -

I x
0.50

X 0.25 '

30.00. 0.0 2.0 4.0 6.0 6.0

x~
10.0

xFig. 4
100Fig. 3

0K X*ofrotx 0.-02a . 0.54e

io0o 0.2
j 8.0 - I', 1 2 0 08001 00

3 .0tab.
ttF
1
1 20

6.0.

llotroot
4.0- b cd 4.0- .

2.0 2.0.

0.0 a 0.0 -
0.00 0.25 0.50 0.75 1.00 1.25 1.50 0.00 0.20 0.40 0.60 0.80 1.00

3Fig. 5 Pig. 6
 I
I
DYNAMICS OF NONLINEAR AUTOMOBILE SHOCK-ABSORBERS

Jdrg Wallaschek 3
Institut fUr Mechanik II
Technische Hochschule Darmstadt
Hochschulstr. 1
D-6100 Darmstadt
Wes t-Germany
I
1. Introduction

Active and semi-active control of vehicle dynamics has become a subject of

major interest during the past years, and today simple "damping control
systems- are already incorporated in serial constructions of passenger
cars. The fast progress in the analysis, design and technology of the con-
trol systems leads to the need of an accurate description of the dynamics
I
of all components involved, like e.g. the tire or the shock-absorber.

The models used to analyze vehicle dynamics are mainly linear, with some i
isolated nonlinear elements. Usually, for ride-comfort investigations only
the nonlinear dynamics of the shock-absorbers and the tires have to be
taken into account. If only the system's overall dynamic response is of
interest, these isolated elements can be taken care of by equivalent linea-
I
rization techniques. The parameters of the linearized model can then be
determined analytically, if the governing nonlinear equations are known.
Most often, however, this is not the case and one must resort to experimen-
tal data. The parameters of the linearized model do of course depend on the
type of test-signal. By using various ter
determine the coefficients of an ad-hoL
equatinn for the shock-absorber.
3ignals it
.tulated
is thus possible to
nonlinear differential I
In what follows we first describe a simple mathematical model for shock-ab-
sorbers, discuss the linearization technique and estimate the parameters of
the mathematical model from experimental data.

2. A mathematical model of the shock-absorber U

Here only a very basic description of a shock-absorber is given. Its dyna-
mic behavior does of course strongly depend on constructive details.
Although there are many different types of shock-absorbers, they share one
common feature: in principle they all consist of a piston moving in a
closed cylinder as shown in Fig. 1.

Assuming that the two chambers are interconnected by a channel with

constant circular cross-section a and length t and that the liquid is

I
 incompressible one can derive the piston's equation of motion

mw(th + d
d)I;(t) + dg (t)] + d 3 sgnx (t)] = F(t), (1)

I wth

3 =
1 2
d2 =a P-
2
d 3 =J '
(2)

where A is the effective piston area, v and p are the dynamic viscosity and
the mass density of the fluid and a is a constant. Coulomb friction between
piston and cylinder has been taken into account, wY. being the friction
force. The three damping terms in (1) describe energy dissipation due to
laminar flow, throttle losses and friction, respectively. In some types of
dampers the liquid is suspended by pneumatic springs. Then one has to take
into account this elasticity by introducing a spring in series with the
piston. This model can also be used, if the foaming of oil and air becomes
significant. We will however concentrate on the simplest case of Fig.1.

I 3. Equivalent linearization

From the various linearization methods only harmonic linearization is con-

sidered here. This technique corresponds to the first order approximation
in the method of Krylov-Bogoljubow-Mitropolski for quasiperiodic differen-
tial equations /1/. It is therefore limited to a description of the damper
dynamics for harmonic motion or nearly harmonic motion only.

I The basic idea of harmonic linearization is to replace the nonlinear equa-

tion

Im + g(x(t),x(t)) = F(t), (3)

where g(x,x) usually is an odd function and F(t) is harmonic by a linear

one

mI , dl + cx = F(t) (4)

and to chose the coefficients c and d so as to minimize the difference

between x(t) and y(t) in a certain sense.

For shock-absorbers g(x,;) is not an odd function and the method can not be.
applied directly. Taking into account that the operation point of the
system will not be fixed in the case of a purely harmonic F(t). we allow
F(t) to have a constant component

I
I
 FUt) F+ F stn(flt +# (5)1

so that

T= g(x(t),;(t)) at , T -21 (6)

for

x(t) =x stnff2t + + p9). (7)U

In this way the operating point of the system is locked at x =0, x=0 and
equation (4) now holds for the fluctuating part of F(t) only.

If we then replace (3) by (4), the coefficients c and d can be found from

c = -1 f' g(xstnG,x2cos@) sine dG (8)

d = - 00 g(xsine,4cos8) cos8 d. (9)

Of course also stochastic linearization can be applied /2/. This has been
done in /3/. The resuts obtained so far are very similar, the only diffe-
rence being that more test-signals can be used in the stochastic case.

4. Parameter identification

Applying harmonic linearization to the analytical model (1), we obtain

8d2 4d3

c = 0, d = dI + - +d 3 (10)
3ir rxi2

for the parameters of the linearized equation. These coefficients can easi-
ly be measured experimentally. Since d depends on the amplitude and on the
3
frequency of the test signal one can identify dj. d2 ' d3 by applying
various test signals. Each measurement yields an equation of the form

d1 +#ad 2 +b d%= ct .
(11)1
If at least three linear independent equations are formulated the coeffi-
cients d d22 d3 can be determined. Formulating more than three equations,
the method of least squares can be applied.

I
 5.Experimental results
Tests were performed in the laboratory of the Institut fUr Mechanik for a
shock-absorber of the rear-axle suspension of a passenger car (Ford Sier-
ra). Fig.2 shows the test facility with a damper mounted in the hydro-pulse
machine. The damper end-point motion was controlled by the machine with the
force being measured simultaneously. The parameter d was estimated experi-
mentally using sinusoidal test-signals of different amplitudes and frequen-
cies. Fig.3 shows the results obtained for a low temperature of the shock-
absorber and test-signals of 2 Hz and 6 Hz respectively. Following the
method described in the last chapter, the coefficients d 1 , d2 and d3 were

* found to be

d I = 1.3866 Ns/mm, d2 = 0 , d3 = 1.5767 N

I when the 6 Hz curve is considered.

1 6. Conclusions
In a first step towards an accurate description of the shock-absorber dyna-
Ia mics a simple mathematical model was presented. The unknown coefficients
were estimated using experimental data. The parameter identification tech-
nique used in the analysis is based on equivalent linearization and seems
to be adequate for a large class of applications.

7. References

/l/ Bogoljubow. N. N.. Mitropolski. J. A.: Asymptotische Methoden in der

Theorie der nichtlinearen Schwingungen. Berlin: Akademie Verlag 1965.

/2/ Kozin, F. : Linearization techniques for non-linear stochastic vibra-

tions. In Ziegler, F.. Schueller. C. I. (eds.) Nonlinear stochastic

U dynamic engineering systems, ILTAM Symposium Innsbruck/Igls. Austria.

June 21-26. 1987, Springer-Verlag. Berlin. Heidelberg. 1988.

/3/ Hagedorn. P.. Wallaschek. J.: On equivalent harmonic and stochastic

linearization for nonlinear shock-absorbers. In Ziegler. F..
Schueller. G. I. (eds.) Nonlinear stochastic dynamic engineering sys-
tems. IUTAM Symposium Innsbruck/Igls. Austria. June 21-26, 1967.
Springer-Verlag, Berlin. Heidelberg, 1988.

U
I
I
 I
FU
xU is I

F = F -i

Fig.1 Basic model of a shock-

absorber
Fig.3 Equivalent
harmonic
damping rates for
signazls (f=2 Hz (x)
I
I
I

I
and
fI !

Fig.2 The test facility. 3

I
 Prediction of Seismic Response Spectra for Bilinear Hysteretic Oscillators

S. R. Malushte and M. P. Singh

Department of Engineering Science & Mechanics, Virginia Tech, Blacksburg, VA 24061

A new approach for presenting the seismic response spectra of bilinear

hysteretic oscillators in terms of "b", the ratio of yield displacement to the maximum

displacement of the corresponding elastic oscillator is proposed. Such spectra,

which are defined in 'erms of the above mentioned reduction factor on spring forces

provide a convinient and direct basis for comparison of response spectra for

oscillators with different nonliuear characteristics. It is observed that the bilinear

hysteretic oscillators with incorporation of low to moderate ductility levels cause a

considerable reduction in acceleration in the medium and low frequency ranges.

Also, for the same level of 6 , it is seen that the bilinear hysteretic oscillators are

more effective in controlling the permanent residual displacement than the

corresponding elasto-plastic oscillators.

Estimation methods for predicting the required design ductility and maximum

absolute acceleration are presented. It is seen that the proposed methods produce

good a estimation of maximum acceleration in all frequency ranges. However, the

prediction of ductility requirement is good, mainly for the case bilinear hysteretic

oscillators with non-zero secondary stiffness and not so good for the case of

elasto-platc oscillators.