Dissertation

Modeling and Simulation of MEMS Devices
Xiaopeng Zhao
Dissertation submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulllment of the requirements for the degree of
Doctor of Philosophy in Engineering Mechanics
Ali H. Nayfeh, Chair Harry Dankowicz Scott L. Hendricks Ziyad N. Masoud Craig A. Woolsey
August 2, 2004 Blacksburg, Virginia
Keywords: MEMS, Finite Element Method, Reduced-Order Modeling, Nonsmooth Dynamics, Discontinuity Mappings Copyright 2004, Xiaopeng Zhao
Modeling and Simulation of MEMS Devices

Xiaopeng Zhao (ABSTRACT)
The objective of this dissertation is to present a modeling and simulation methodology for MEMS devices and identify and understand the associated nonlinearities due to large deections, electric actuation, impacts, and friction. In the rst part of the dissertation, we introduce a reduced-order model of exible microplates under electric excitation. The model utilizes the von Karman plate equations to account for geometric nonlinearities due to large plate deections. The Galerkin approach is employed to reduce the partial-dierential equations of motion and associated boundary conditions into a nite dimensional system of nonlinearly coupled ordinary-dierential equations. We use the reduced-order model to analyze the mechanical behavior of a simply supported microplate and a fully clamped microplate. Eect of various design parameters on both the static and dynamic characteristics of microplates is studied. The second part of the dissertation presents comprehensive modeling and simulation tools for impact microactuators. Nonsmooth dynamics due to impacts and friction are studied, combining various approaches, including direct numerical integration, root-nding technique for periodic motions, continuation of grazing periodic orbits, and local analysis of the near grazing dynamics. The transition between nonimpacting and impacting long term motions, referred to as grazing bifurcations, indicates the transition between on and o states of an impact microactuator. Three dierent on-o switching mechanisms are identied for the Mita microactuator. These mechanisms also generalize to arbitrary impacting systems with a similar nonlinearity. A local map based on the concept of discontinuity mapping provides an ecient and accurate tool for the grazing bifurcation analysis. Nonlinear impacting dynamics of the microactuator are studied in detail to identify various bifurcations and parameter ranges corresponding to chaotic motions. We nd that the frequency-response curves of the impacting dynamics are signicantly dierent from those of the nonimpacting dynamics.
Dedication
To my parents, my wife, and my son
iii
Acknowledgments
I would like to wholeheartedly thank Dr. Ali Nayfeh for his invaluable guidance and advice throughout my Ph.D. studies. His encouragement, support, and thoughtfulness are deeply appreciated. His patience, resourcefulness, and passion for research are greatly admired. I would also like to thank my committee members: Dr. Harry Dankowicz, Dr. Scott Hendricks, Dr. Ziyard Masoud, an Dr. Craig Woolsey for their invaluable comments and advice. Specially, I would like to thank Dr. Dankowicz for his leading me into the eld of nonsmooth dynamics and his great help on my research. I would like to thank Dr. Hendricks for giving me an valuable understanding for dynamics. I also thank Dr. Liviu Librescu, Dr. Ramesh Batra, Dr. Dean Mook, and Dr. Zafer Gurdal for enhancing my knowledge with their excellent teaching. I would like to thank all of the people I have met in the Nonlinear Vibration Laboratory. In particular, I thank Dr. Pramod Malatkar, Dr. Haider Arafat, Mr. Chevva Konda Reddy, and Mr. Gregory Vogl for their friendship, encouragement, and inspiring ideas. I thank Dr. Mohammad Younis, Dr. Eihab Abdel-Ralman, Dr. Khalid Alhaza, Dr. Samir Emam, Dr. Waleed Faris, Mr. Mohammad Daqaq, and Mr. Imran Akhtar for their friendship. Thanks are also due to Mrs. Sally Shrader for her help in administrative matters. I would like to thank the friends at Blacksburg, whom I was pleased to meet and know. Thanks go to Dr. David Gao, Dr. Hsu-Kuang Ching, Dr. Xiaolan Song, Dr. Hong Wang, Dr. Johnny Yu, Dr. Jesse Chen-Yu, Dr. Zhanming Qi, Dr. Zhicun Wang, Dr. Zhigang Wei, Ms. Miao Sun, and Mr. Yunkai Lu. I would like to express my sincere gratitude and deep appreciation to my beloved parents for
iv
their love, encouragement, and believing in me over all these years. I also thank my sister Xiaona and brother in law Lifei for their love and support. I would like to deeply thank my wife Li for her support, patience, and sacrice, without which this work would not have been done. Thanks also go to my son Eric for the great happiness he has brought to our family.
Contents
1 Introduction 1.1 1.2 What are MEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Challenges in MEMS Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 1.2.2 1.2.3 1.2.4 1.3 Electric Actuation and Pull-in Instability . . . . . . . . . . . . . . . . . . . . Electrically Actuated Microplates . . . . . . . . . . . . . . . . . . . . . . . . . Impacts and Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fluid-Structure Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1 1 3 3 5 8 9
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 13
2 Literature Review 2.1
Nonlinear Analysis of Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.1.1 2.1.2 2.1.3 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Methods of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2
Dynamic Tools for Nonsmooth Systems . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.2.1 Dynamical systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
vi
2.2.2 2.2.3 2.2.4 2.2.5
Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Periodic solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Continuation of periodic solutions . . . . . . . . . . . . . . . . . . . . . . . . 28 Continuation of grazing periodic solutions. . . . . . . . . . . . . . . . . . . . . 28 31
3 Mechanical Behavior of Electrically Actuated Microplates 3.1 3.2 Motivation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Microplates with Simply Supported Boundary Conditions . . . . . . . . . . . . . . . 33 3.2.1 3.2.2 3.2.3 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Linear Mode Shapes and Corresponding Natural Frequencies . . . . . . . . . 41
3.3
Microplates with General Boundary Conditions . . . . . . . . . . . . . . . . . . . . . 43 3.3.1 3.3.2 3.3.3 3.3.4 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Linear Mode Shapes and Corresponding Natural Frequencies . . . . . . . . . 51
3.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 56
4 Modeling and Simulation of Impact Microactuators 4.1 4.2 4.3 Motivation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Mitas Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 Parametric Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.3.1 4.3.2 Switch on and o mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Voltage responses and frequency responses . . . . . . . . . . . . . . . . . . . . 68 vii
4.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 72
5 Local Analysis of Grazing Bifurcations 5.1 5.2 5.3 Motivation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Transition Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Discontinuity mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 5.3.1 Near-grazing Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.4 5.5
Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 103
6 Bifurcations and Chaotic Motions of Post-Grazing Dynamics 6.1 6.2 6.3 6.4 6.5 Motivation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Bifurcation Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Chatter and Sticking Motions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Frequency-Response Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 114
7 Conclusion and Recommendation for Future Work 7.1 7.2
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 Recommendation for Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 118 128 129
Bibliography A The Implicit Function Theorem B Hierarchical Finite Element Method
viii
List of Figures
1.1 1.2 1.3 2.1 3.1 3.2 3.3 The size of a micro-thruster relative to a penny. . . . . . . . . . . . . . . . . . . . . . A schematic of a parallel-plate capacitor. . . . . . . . . . . . . . . . . . . . . . . . . 2 4 5
A schematic of an air-gap capacitor. . . . . . . . . . . . . . . . . . . . . . . . . . . .
A ow passes a discontinuity surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 A schematic of a micropump. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Geometry of a microplate-based capacitor. . . . . . . . . . . . . . . . . . . . . . . . . 35 Variation of the maximum nondimensional deection wmax with the nondimensional electrostatic load 2 Vp2 when = 1 and 1 = 2. . . . . . . . . . . . . . . . . . . . . . 39
3.4
Variation of the nondimensional electrostatic load at pull-in (2 Vp2 )P L with the number of modes when = 1 and 1 = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5
Variation of the maximum deection (wmax )P L at pull-in with the number of modes when = 1 and 1 = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6
Variation of the electrostatic load (2 Vp2 )P L with 1 at pull-in: (a) nonlinear plate model with immovable inplane constraints (+), (b) nonlinear plate model with movable inplane constraints ( ), and (c) linear plate model (solid line) when = 1. . . . 41
ix
3.7
Variation of the maximum deection (wmax )P L at pull-in with 1 : (a) nonlinear plate model with immovable inplane constraints (+), (b) nonlinear plate model with movable inplane constraints ( ), and (c) linear plate model (solid line) when = 1. 42
3.8
Variation of (2 Vp2 )P L with for 1 = 0.5 (+), 1 = 1.0 ( ), 1 = 1.5 ( ), and 1 = 2.0 (x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.9
Variation of (wmax )P L with for 1 = 0.5 (+), 1 = 1.0 ( ), 1 = 1.5 ( ), and 1 = 2.0 (x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.10 Variation of the maximum deection with (2 Vp2 ) when 1 = 0.5. . . . . . . . . . . . 45 3.11 Inuence of 1 on the normalized fundamental natural frequency. . . . . . . . . . . . 45 3.12 The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3.13 The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 2.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.14 Comparison of wmax calculated using the reduced-order model with the experimental results (+) of Francais and Dufour [1999]. . . . . . . . . . . . . . . . . . . . . . . . . 48 3.15 Variation of the electrostatic pull-in load with 1 for a simply supported square microplate: analytical model developed in Section 3.2 (o) and present model (solid line). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.16 Variation of the maximum deection at pull-in with 1 for a simply supported square microplate: analytical model developed in Section 3.2 (o) and present model (solid line). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.17 Variation of the maximum deection wmax with the electrostatic load 2 Vp2 for a fully clamped square plate when 1 = 1. . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.18 Variation of the electrostatic pull-in load with 1 for a fully clamped square plate: nonlinear plate theory (+) and linear plate theory (solid line). . . . . . . . . . . . . . 50
3.19 Variation of the maximum deection at pull-in with 1 for a fully clamped square plate: nonlinear plate theory (+) and linear plate theory (solid line). . . . . . . . . . 51 3.20 Comparison of the electrostatic load at pull-in obtained for a fully clamped square plate (+) with that obtained for a simply supported square plate (solid line) for various value of 1 . The pull-in loads for the simply supported plate have been shifted upward by 120 for convenience of comparison. . . . . . . . . . . . . . . . . . . 52 3.21 Variation of the normalized fundamental natural frequency 1 with the electrostatic load 2 Vp2 for various values of 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.22 The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.23 The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 2.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.1 4.2 4.3 4.4 4.5 The schematic of the impact microactuator. . . . . . . . . . . . . . . . . . . . . . . . 59 Flowchart of the algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Step displacement of the microactuator under very low frequency inputs. . . . . . . . 63 Steady-state time histories of q1 and q2 when = 0.5 and f = 0.24. . . . . . . . . . 64
Type I switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: G indicates a grazing bifurcation and CF indicates a cyclic-fold bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.6
Type I switch on and o mechanism: variations in the maximum displacement of the movable block and the impact velocity under changes in the applied voltage: G indicates a grazing bifurcation and CF a cyclic-fold bifurcation. . . . . . . . . . . . . 66
4.7
Type II switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: G indicates a grazing bifurcation and PD indicates a period-doubling bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . 67
xi
4.8
Type III switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: CF indicates a cyclic fold bifurcation and PD indicates a period-doubling bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.9
Variation of switch on and o voltages with respect to the excitation frequency. . . . 69
4.10 Voltage response when = 0.5: PD indicates a period-doubling bifurcation, CF indicates a cyclic-fold bifurcation, and G indicates a grazing bifurcation. . . . . . . . 70 4.11 Frequency response when f = 0.18: CF indicates a cyclic-fold bifurcation and G indicates a grazing bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 5.1 Intersections of nonimpacting and impacting trajectories with the Poincar e section P . The nonimpacting trajectory reaches its local maximum in q2 at the intersection. The impacting trajectory reaches P virtually by neglecting the existence of the discontinuity surface D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 5.2 A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across I (G=grazing contact, SN=saddle-node bifurcation). 5.3 . . . . . . . . . . . . . . . . . . . . . . . . . 76
A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across II (G=grazing contact, PD=period-doubling bifurcation). Here, the black regions correspond to impacting chaotic attractors. An enlargement near the period-doubling bifurcation in (a) is shown in (b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
xii
5.4
A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across III (G=grazing contact, SN=saddle-node bifurcation, PD=period-doubling bifurcation, C=global crisis). Here, the black regions correspond to impacting chaotic attractors. An enlargement near the period-doubling bifurcation in (a) is shown in (b). . . . . . . . 77
5.5
A collection of bifurcation curves in the parameter space, where represents the

sn1 represents the locus of saddle-node bifurcations when crossing grazing curve; Vamp pd I ; Vamp represents the locus of period-doubling bifurcations when crossing II and sn2 represents the locus of saddle-node bifurcations when crossing cr III ; Vamp III ; Vamp
represents the locus of the global crisis of the chaotic attractor when crossing III . Panels (b) and (c) show enlargements of the neighborhoods of the boundary point between I and II and the boundary point between II and III , respectively. . . 78 5.6 Schematic bifurcation scenarios associated with the switching between impacting motions and nonimpacting motions at the boundary between I and II in Panel (a) and those at the boundary between II and III in Panel (b) (G=grazing contact, PD=period-doubling bifurcation). Here, solid curves correspond to stable periodic motions and dashed curves to unstable periodic motions. The black regions correspond to impacting chaotic attractors. 5.7 . . . . . . . . . . . . . . . . . . . . . . . . . 79
Trajectories associated with the discontinuity mapping D. Here S stands for the stick manifold, D is the discontinuity surface, and P is the Poincar e section. We note that S , D, and P are all 4dimensional hyper-surfaces in a 5dimensional space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.8
Fixed points of the composite Poincar e map P corresponding to period-1 nonimpacting orbits (solid), period-1 impacting orbits with a single impact per period (dashed). Here, penetration is dened as sign (q2 )
q2 |q2 |, b = Psmooth , (x ), and q2 c = 1 Psmooth , (x ) c1 + Psmooth, (x ) c2 . . . . . . . . . . . . . . . . . . . . . . . 92
xiii
5.9
Fixed points of the composite Poincar e map P corresponding to period-1 nonimpacting orbits (solid), period-1 impacting orbits with a single impact per period (dashed). Here, penetration is dened as sign (q2 ) |q2 |. . . . . . . . . . . . . . . . . . . . . 93
5.10 Comparison of the bifurcation results predicted from the local map with those obtained through numerical integration when is at the boundary between I and II . Panel (a) shows variation of q2 with the applied voltage, where the impacting motions are in black and the nonimpacting motions are in gray. We note that the predicted results are indistinguishable from the numerical results. Panel (b) shows the dependence of the real and imaginary parts of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (solid) and as predicted by the local mapping (dashed). . . . . . . . . . . . . . . . . . . . . 95 5.11 Comparison of the bifurcation results predicted from the local map with those obtained through numerical simulation when = 0.4888. Panel a) shows variation of q2 with the applied voltage, where the impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, and dashed curves correspond to unstable periodic motions. Panels (b) and (c) show the dependence of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (dots for stable and circles for unstable) and as predicted by the local mapping (solid for stable and dashed for unstable). . . . . . 96 5.12 Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.485. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors. Panels (c) and (d) show the dependence of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (dots for stable and circles for unstable) and as predicted by the local mapping (solid for stable and dashed for unstable). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xiv
5.13 Bifurcation curves obtained numerically (Panels (a) and (b)), compared to those predicted from the local map (a)), and those predicted from the Taylor expansions (b)). In both panels, triangles, circles, and + correspond to the numerically obtained grazing, period-doubling, and saddle-node bifurcation points, respectively; the solid lines, dashed lines, and dotted lines correspond to the predicted grazing, period-doubling, and saddle-node bifurcation points, respectively. . . . . . . . . . . 100
5.14 Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.4803. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors. . . . . . . . . . . . . . 100
5.15 Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.4793. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors. . . . . . . . . . . . . . 101
5.16 Comparison of predicted bifurcation points with those obtained using the simulated data (solid line - numerically obtained grazing bifurcation points, dots - predicted grazing bifurcation points, + signs - numerically obtained crisis points, circles predicted crisis points, * signs - numerically obtained saddle-node bifurcation points, triangles - predicted saddle-node bifurcation points. . . . . . . . . . . . . . . . . . . . 101 5.17 Dependence of the largest Lyapunov exponent of the chaotic attractors on the applied voltage as obtained from the local mapping. . . . . . . . . . . . . . . . . . . . . . . . 102 6.1 Phase portraits in the x2 , x2 space: nonimpacting periodic orbit at (a) V = 0.18, (b) V = 0.2; periodic grazing orbit at (c) V = 0.2272; impacting periodic orbit at (d) V = 0.18, (e) V = 0.21, (f) V = 0.24. We note that the rigid stopper is placed at x2 = 0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
xv
6.2
One example of the transition between nonimpacting and impacting motions. Here, the dashed line stands for nonimpacting periodic motions, the solid line stands for stable impacting periodic motions, and the dotted line stands for the unstable impacting periodic motions: G indicates a grazing bifurcation and SN indicates a saddle-node bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.3
Bifurcation diagram for the impact microactuator when = 0.5: (a) shows the dierence between the maximum and minimum displacements x2 of mass m2 sampled per period of forcing, and (b) shows the velocity x2 of mass m2 at impact. . . . . . . 106
6.4
Numerically computed phase portraits of mass m2 : (a) P (1, 1) motion at V = 0.55, (b) P (2, 2) motion at V = 0.575, (c) near grazing P (2, 2) motion at V = 0.58, and (d) P (3, 2) motion at V = 0.58. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.5 6.6
An enlargement of Figure 6.3 in the range 0.78 V 0.815. . . . . . . . . . . . . . 108 Bifurcation diagram showing two stable branches of P (3, 1) motions connected through an unstable branch of P (3, 1) motions. Here, the solid lines stand for stable motions, the dotted line stands for unstable motions, and SN indicates a saddle-node bifurcation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.7
Time history (a) and phase portrait (b) of m2 representing a P(3, 1) chattering motion at V = 0.82. We note that the three impact velocities are positive in (b). An enlargement of the boxed area is shown in Figure 6.8 (a). . . . . . . . . . . . . . 110
6.8
Time histories of the displacement x2 at excitation voltages showing the transition from chatter to sticking. Here, (a) shows an enlargement of the boxed area in Figure 6.7 (a) and (b) (f) show zooms near the maximum displacement of the corresponding time history to show the increasing number of chatters and sticking: (a) P (3, 1) chattering motion at V = 0.82, (b) P (4, 1) chattering motion at V = 0.85, (c) P (5, 1) chattering motion at V = 0.86, (d) P (6, 1) chattering motion at V = 0.864, (e) P (7, 1) chattering motion at V = 0.868, and (f) sticking motion at V = 0.872. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
xvi
6.9
Bifurcation diagram showing a route of chatter to sticking. Here, cn indicates the voltage at which the P (n, 1) chattering motion originates. . . . . . . . . . . . . . . . 112
6.10 Nonimpacting frequency-response curve showing variation of the maximum minus minimum displacements with forcing frequency, when V = 0.175. . . . . . . . . . . . 112 6.11 Impacting frequency-response curves when V = 0.175. The dashed line corresponds to nonimpacting periodic motions, the dotted lines correspond to unstable impacting P (1, 1) motions, and the solid line corresponds to stable impacting P (1, 1) motions; SN stands for saddle-node bifurcation, and G stands for grazing bifurcation. . . . . . 113
xvii
Chapter 1
Introduction
1.1 What are MEMS
Micro-electromechanical systems (MEMS) are referred to as microsystems in Europe and micromachines in Japan. Ranging in size from a few micrometers to millimeters, MEMS devices combine mechanical and electric components and have the ability to sense, control, and actuate on the micro scale. Electronics in MEMS are fabricated using integrated circuit (IC) technology. On the other hand, the micromechanical components are fabricated by processes, such as bulk micromachining, surface micromachining, high-aspect-ratio micromachining, microelectrodischarge machining, laser etching, 3D printing, wafer bonding, and micromolding. Figure 1.1 shows a micro-thruster prototyped by TRW Inc (http://spaceightnow.com/news/n0105/17trwmems). The micro-thruster is fabricated as a three-layer silicon-glass sandwich and consists of multiple small propellant cells, each of which is a separate thruster. Cell thrusters are ignited in controlled sequences to generate propulsion in discrete increments. Although research in MEMS dates from the 1950s, it is not until the mid-1990s that MEMS appeared in commercial products and applications. So far, only a few applications have been established, such as accelerometers used in automotive airbags, pressure sensors for medical applications, inject printer heads, color projection displays, and micropositioners in data storage systems. Nonetheless, MEMS have shown substantial promise in a broad eld, which includes
Xiaopeng Zhao
Chapter 1. Introduction
Figure 1.1: The size of a micro-thruster relative to a penny. integrated microsensors, microactuators, microinstruments, micro-optics, and microuidics. The greatest potential for MEMS lies in applications with telecommunications, biomedical, and process control areas. MEMS technology is inherently multidisciplinary, which has resulted in synergies between previously unrelated elds, such as biology and microelectronics. MEMS are compatible with integrated circuit processes, which allow the fabrication of thousands of MEMS devices with increased reliability and reduced cost. Moreover MEMS allow the manufacture of products that can not be made by other methods. These advantages make MEMS more attractive than conventional manufacturing technologies. As one of the most promising technologies for the 21st Century, MEMS technology has the potential to dramatically aect our lives and the way we live. On the other hand, computational tools that allow engineers to quickly design and optimize these micromachines have not kept pace. The development and prototype of MEMS devices rely heavily on trial and error and consume considerable time and expense. An interface, which connects design and fabrication, should be made to provide MEMS designers with convenient modeling and simulation tools.
Xiaopeng Zhao
1.2
Challenges in MEMS Modeling
Nonlinear phenomena are prevalent in MEMS devices. It is therefore necessary to identify and understand these nonlinearities to ensure better system performances. Generally, the nonlinearity appears as nonlinear or discontinuous functions of system variables and their derivatives in the equations of motion, boundary conditions, and/or constitutive laws. Examples of various nonlinearities can be found in the books of Nayfeh and Mook [1979], Sathyamoorthy [1997], and Strogatz [1998]. In this section, we identify several nonlinearities associated with MEMS devices, which pose a challenging modeling problem and need to be studied to better understand the behavior of MEMS devices.
1.2.1
Electric Actuation and Pull-in Instability
MEMS devices require low-input power, which can be provided by thermal, magnetic, piezoelectric, hydraulic, and electric actuations. However, the vast majority of devices have relied on electric actuation because of its high-energy density, high-mechanical exibility, and low-current requirement. To predict performance and enable design optimization of such devices, a coupled electromechanical simulation is necessary. In general, this requires solving Laplaces equation for the electrostatic eld with certain boundary conditions. As the deformation of mechanical structures is unknown a prior: this problem usually prohibits an exact solution. The relaxation method is used in commercial softwares, such as CoventorWare and Intellsuite, where the mechanical structure is discretized using the nite-element method and the electric eld is discretized using the boundaryelement method. Based on the original undeected structure conguration, the initial electric load is computed. This electric load then actuates the structure to a new deected position, which in turn induces a change in the capacitance and thereby a new electric load. A new deformation is then achieved from the updated electric load, and so forth. This iteration has to be investigated carefully to guarantee convergence [Chen et al., 2004]. On the other hand, many MEMS devices possess a small aspect ratio; therefore the potential eld can be asymptotically simplied [Pelesko, 2001]. In many microsystems, polysilicon beam- or plate-based actuation components are deformed using a potential dierence between the structure
Xiaopeng Zhao
and a relatively rigid electrode. The distributed electric force can be approximated by F (x, y, V ) = V 2 , 2 (d w (x, y ))2 (1.1)
where w (x, y ) is the out-of-plane deection, is the permittivity of the dielectric material, V is the potential dierence, and d is the zero-voltage gap. The electric load can be further simplied in devices such as optical switches and inchworm actuators. Here the actuation forces can be described using a nonlinear parallel-plate capacitance model, as shown in Figure 1.2. The attractive force between two parallel-plate conductors is F = where A is the area of the plate. AV 2 , 2(d x)2 (1.2)
k d
c x
Figure 1.2: A schematic of a parallel-plate capacitor. The interaction between the electric eld and the structural deformations brings a new instability called pull-in. Nathanson et al. [1967] were the rst to observe the pull-in phenomenon when studying an electrically actuated resonant gate transistor. They abstracted the transistor as a mass-spring model of electrostatic actuation and explained the pull-in instability using this model. At about the same time, Taylor [1968] found the same instability during his study of the
Xiaopeng Zhao
electrostatic deection of soap lms. Ackerberg [1969] presented a mathematical model of Taylors experiments and numerically discovered that innitely many steady-state congurations of the system exist for certain values of the applied voltage. Pull-in instability imposes a design limit on numerous MEMS/NEMS devices, where the achievable capacitance range is limited by this instability.
1.2.2
Electrically Actuated Microplates
An electrically driven microplate forms one side of the variable capacity air-gap capacitor shown in Figure 1.3. As the capacitive force deforms the microplate, it changes with the plate deection, resulting in coupling of the electrical and mechanical forces. Such capacitors are the actuation components in many micropumps, micromirrors, microphones, and microsensors [Zengerle et al., 1992; Zhang et al., 2001; Hsu et al., 1998; Scheeper et al., 1999; Tilmans and Legtenberg, 1994]. Applications of these MEMS devices spread over biotechnology, image processing, automotive, chemical, food, and mining industries.
Mic
ropl ate
Stat
i o na
ry E lect rode
Figure 1.3: A schematic of an air-gap capacitor. A common approach in the literature is to assume a linear relationship between the excitation force and the plate deection. The linear plate equation is then solved by numerical methods,
Xiaopeng Zhao
such as the Galerkin method, the Rayleigh-Ritz method, and the nite-element method [Fletcher, 1984; Zienkiewicz, 2000]. However, the linear plate theory is only appropriate for small deections. When the plate deection is comparable to the thickness of the plate, which is the case in most microplates in typical MEMS devices, it produces erroneous results and one has to resort to a nonlinear plate theory [Timoshenko and Woinowsky-Krieger, 1959; Nayfeh and Pai, 2004]. A group of researchers neglect the bending stiness and treat the microplate as a membrane element. Bourouina and Grandchamp [1996] studied a square-shape membrane under an electrostatic load using electrical analog and equivalent network. Cozma and Puers [1997] studied the static deection of the exible diagram in a pressure sensor using the nite-element method. Francais and Dufour [1998] investigated the static instability of a square membrane. Saif et al. [1999] treated an annular actuating element in a micropump as a membrane, therefore neglecting the bending stiness of the element. Yang [2002] examined the static deformation of a microscale membrane under an electrostatic loading. Pull-in voltage was determined analytically and the contact problem after the pull-in collapse was also studied. Pelesko and Chen [2003] presented a mathematical model for an electrostatically actuated device, which consists of a circular elastic membrane suspended above a rigid plate. Using symmetry methods, they investigated solutions of the system under a prescribed voltage. They also studied the capacitive-control scheme by examining the control eect on the bifurcation diagram. Malhaire [1999] found experimentally that, in the case of clamped actuating elements in MEMS devices, the bending stiness can not be neglected. Francais and Dufour [1999] studied the static deection of a diaphragm under an electrostatic excitation. They attacked the linear plate equation using the Rayleigh-Ritz method and polynomials as basis functions. They developed an iterative procedure to solve for the deection at a given voltage. Their numerical results are in good agreement with their experimental measurements. Their analysis and results are limited to small deections only. Using a linear circular plate theory, Rajalingham and Bhat [1998] calculated the static deection of an electrically actuated circular diagram and they also investigated the vibration characteristics of the diagram around the deected positions. The pull-in instability associated with MEMS devices was studied by Konig and Wachutka [1999]. Numerical results were presented for micromirrors and membranes. Ng et al. [2004] studied the dynamic characteristics of an electrically actuated microplate. The Laplace
Xiaopeng Zhao
equation governing the potential dierence between the electrodes was solved using the boundaryelement method (BEM). Geometric nonlinearities due to mid-plane stretching were included in the plate model, which was discretized using the nite-element method (FEM). The coupled electromechanical equations were then solved by iteration. They found qualitative dierences between the results of nonlinear analysis and those of linear analysis when the plate undergoes large deections. Because numerous variables are generated by FEM models, this approach makes it dicult to map the design space and identify the relationships among the design parameters [Younis et al., 2003]. Nayfeh and coworkers have investigated the nonlinear modeling of annular plates [Faris et al., 2002], clamped circular plates [Vogl and Nayfeh, 2003], and simply supported rectangular plates [Zhao et al., 2003, 2004a]. They determined the static deection using either a numerical shooting technique [Faris et al., 2002] or a reduced-order model [Vogl and Nayfeh, 2003; Zhao et al., 2003, 2004a]. Faris et al. [2002] studied the mechanical behavior of an electrostatic micropump employing an annular actuation member, accounting for bending stiness. They modeled the annular plate as a distributed-mass structure deecting in bending under electrostatic actuation. They adopted the von K arm an plate model under axisymmetric loading conditions to account for the bending stiness and large deections. They considered two cases of boundary conditions to simulate practical boundary conditions in MEMS devices. Vogl and Nayfeh [2003] presented an analytical reduced-order model (macromodel) for an electrically actuated clamped circular plate by using the Galerkin approach, thereby reducing the distributed-parameter equations into a nite system of ordinary-dierential equations in time. The reduced-order model accounts for residual stresses and geometric and electric nonlinearities. They solved the reduced equations for the equilibrium states due to a general electric potential and determined the natural frequencies of the axisymmetric modes. They validated the model by using data from experiments performed on silicon-based MEMS systems. Zhao et al. [2003; 2004a] studied the mechanical behavior of electrically driven rectangular plates. The reduced-order modeling of rectangular plates is computationally more complicated than that of annular or circular plates because the former problem is two-dimensional in space and the latter ones are one-dimensional. Moreover, the linear undamped mode shapes, which serve as
Xiaopeng Zhao
the basis functions in the reduced-order modeling, are not analytically available for rectangular plates with general boundary conditions. Static deections and associated vibration characteristics for simply supported and fully clamped rectangular plates were solved by Zhao et al. [2003] and Zhao et al. [2004a], respectively.
1.2.3
Impacts and Friction
Micro or nano distance manipulations are of prime importance in the MEMS industry. Microdevices are ideal for micropositioning systems due to their small sizes. Microactuators used to produce small displacements would need large actuation forces and a long driving distance. This would require large voltages to produce the desired forces. Actuators based on impulsive forces provide a solution to this problem. The proper function of these actuators relies on actuation due to impacts between their components and the stick-slip phenomenon due to friction between the device and the support medium. Probably the simplest way of modeling systems with impact and friction is to use the rigid multibody approach because it neglects wave eects during impacts between bodies and stick-slip transitions of bodies in contact. Instead, the impact and stick-slip phenomena are described by instantaneous contact laws. The advantage of the rigid multibody approach is that sti dierential problems are avoided. Nonetheless, the rigid-body approach leads to nonsmooth mathematical formulations with discontinuities in the generalized velocities due to impacts and in the vector eld due to friction. The dynamics of impacting systems and the bifurcations therein have been addressed by several authors [Shaw and Holmes, 1983; Foale and Bishop, 1994; Ivanov, 1996; Nordmark, 1997; Brogliato, 1999; Dankowicz and Nordmark, 1999; Leine et al., 2000; Wagg and Bishop, 2001; Adolfsson et al., 2001; Dankowicz et al., 2002]. The central interest in studying these systems is the stability and persistence properties of impacting orbits, transitions between nonimpacting and impacting orbits, and the associated appearance of chaotic motions. Whereas bifurcations in smooth systems are well understood [Nayfeh and Balachandran, 1995; Strogatz, 1998], the study of bifurcations in nonsmooth systems is more challenging [Leine et al., 2000]. Impacting systems can exhibit a plethora of typical nonlinear phenomena, such as saddle-node bifurcations, ip bifurcations,
Xiaopeng Zhao
multiple coexisting attracting solutions, Feigenbaum cascades, as well as nonlinear phenomena not seen in smooth dynamical systems, such as grazing bifurcations, chatter, and sticking. Therefore, a careful study of the dynamics is essential in order to ensure a consistent performance of the device. A starting point for the analysis of oscillatory dynamical systems is the study of periodic solutions. A suitable formulation of the condition for the existence of a periodic orbit leads to the use of the Newton-Raphson scheme for precisely locating a periodic orbit (stable or unstable) given a good initial guess. Combined with an estimate of the change in the position of the periodic orbit under changes in system parameters, the Newton-Raphson scheme is the method of choice for so-called parameter continuation. Here, an initially found periodic solution is followed in state space under changes in system parameters. The parameter continuation approach is able to follow periodic solutions even through bifurcation points associated with a change of stability and the possible birth of additional periodic or nonperiodic oscillatory motions. The methodology is thus signicantly more versatile than an approach based entirely on direct numerical simulation, as such a scheme would not be able to detect unstable periodic solutions.
1.2.4
Fluid-Structure Interaction
Emerging from the MEMS technology, microuidics has become an important research eld and a promising market. The microuidics market was estimated to be 3 to 4.5 billion US Dollars with an annual growth rate of 25%35%, based on a recent report of the System Planning Corporation [Nguyen et al., 2002]. The most promising microuidics products are reported to be devices for DNA, protein analysis, and drug discovery. As one of the most important microuidic components, various micropumps have been developed in the last decade. Micropumps are required in chemical, medical, biomedical and other microuidic systems. A micropump is used to manipulate very small and precise volumes of uids. The uid ow is driven by the vibration of the thin plate and the ow also imposes a resistance to this vibration. Thus, vibrations of the plate and the uid ow are inherently coupled. The vibration characteristics of a micropump highly depend on this uid-structure coupling. Therefore, uid-structure interaction is one of the primary concerns when studying micropumps.
Xiaopeng Zhao
10
The inuence of uids in MEMS devices also appears as viscous damping. As gas, such as air, lls the space between MEMS components, viscous air damping strongly aects the dynamic behavior of these devices. Viscous air damping can be divided into two categories: squeeze-lm damping and slide-lm damping. Squeeze-lm damping occurs when a micromechanical structure is pushed towards a rigid surface with a thin uid lm in between, while slide-lm damping mainly occurs in laterally moving devices.
1.3
Overview
In this Dissertation, we model and simulate several MEMS devices. The primary goal is to illustrate the importance of understanding nonlinear behaviors in MEMS to ensure better device performance. The presented models and simulation results can be used for prototype design, optimization, and control of micromachined systems. In Chapter 2, we present a literature review, which consists of two parts. In part I, we rst briey review nonlinear analysis of plates. Then the equations of motion governing the nonlinear dynamics of isotropic thin plates are derived using the extended Hamilton principle. Several methods of solution are presented to close this part. In part II, we introduce dynamical tools for nonsmooth systems. Traditional dynamical analysis methods for smooth systems have to be modied to take into account the nonsmoothness associated with states and vector elds. This is done by introducing a discontinuity mapping. Then repetitive impact motions are detected using a root nding technique based on the Newton-Raphson algorithm. Stability of the motions is analyzed as a byproduct of the root-nding procedure. A continuation method is developed to nd the system parameters corresponding to grazing orbits. In Chapter 3, we present a reduced-order model for an electrically actuated MEMS-based microplate. The model accounts for the electric-force nonlinearity and the mid-plane stretching of the plate. The linear undamped vibration modes are calculated numerically using the hierarchical nite-element method. These mode shapes are used in a Galerkin approximation to reduce the partial-dierential equations of motion into a nite-dimensional system of nonlinearly coupled second-order ordinary-dierential equations. The developed model is validated by comparing its
Xiaopeng Zhao
11
results with those obtained experimentally and those obtained by solving the distributed-parameter system. The reduced-order model provides an eective and accurate design tool, which can be used for design optimization and determination of the stable operation range of MEMS devices. The model was used to calculate the deection of the microplate under DC voltages and study the pull-in phenomenon. The natural frequencies and mode shapes around the deected position of the microplate are then calculated by solving the linear eigenvalue problem. The eects of various design parameters on both the static and dynamic characteristics of microplates are studied. In Chapter 4, we numerically study the dynamics of an electrically driven impact actuator, using the techniques described in Chapter 2. As the name suggests, the impact actuator uses impacts between its moving elements to produce nano-displacements. While on one hand, impact actuators provide a way to produce small displacements with moderate actuation voltages, on the other hand, impacts make the underlying dynamics nonsmooth. Impacts are a source of nonlinearity and a careful study of the dynamics is essential in order to ensure a consistent performance of the device. The impact microactuator reported by Mita and associates is modelled using a twodegree-of-freedom system. Impacts in the system are assumed to occur simultaneously and are modeled using a coecient of restitution. Friction between the actuator and the ground is modeled using Coulomb friction during slip and Amontons law during stick. We examine the inuence of the system parameters, such as the excitation frequency and amplitude, on the dynamics of the actuator. Transitions between nonimpacting and impacting asymptotic dynamics are identied as grazing bifurcations, which indicate the on -o switching mechanism for an impact actuator. Three characteristically dierent grazing bifurcations are found numerically. In Chapter 5, we study the three grazing bifurcations found in the previous chapter, using a local map analysis. The local description in the vicinity of a grazing trajectory is non-dierentiable. Grazing bifurcations introduce dramatic changes in the stability and persistence of a previously stable nonimpacting periodic motion under further change in a system parameter. Using the concept of discontinuity mappings, we present a theoretical analysis that predicts the character of each transition (identied as a co-dimension-one bifurcation) from a set of quantities that are computable in terms of the system properties at grazing. Specically, we show how this analysis can be applied to predict the bifurcation behavior in the neighborhoods of two co-dimension-two
Xiaopeng Zhao
12
bifurcation points, which separate the co-dimension-one bifurcation scenarios. The predictions are validated with results from numerical simulations of a model impact microactuator. Both of this and the previous chapters show that the post-grazing dynamics settle down to a stable impacting periodic motion for certain parameter values near those corresponding to the grazing trajectory. In Chapter 6, we consider the post-grazing dynamics for parameters farther away from the grazing bifurcation. The previously found impacting periodic motion persists for certain parameter range and its periodicity then changes due to a plethora of bifurcations, such as period-doubling bifurcations, discontinuous grazing bifurcations, and saddle-node bifurcations. We discuss the accompanying series of low-velocity impacts with a high-velocity impact; that is, chattering phenomenon. We numerically nd an interesting pattern of the evolution of chatter to sticking and a constant associated with the chatter sequence. It is also found that impacts dramatically change the characteristics of the frequency-response curves.
Chapter 2
Literature Review
2.1
2.1.1
Nonlinear Analysis of Plates

Review
Plates are important construction elements for structures and are the most studied elements in structural engineering. Classical plate theories deal with a plate whose thickness is small compared to its cross-section dimensions and therefore the transverse shear deformation, the rotatory inertia, and the transverse normal stress are neglected. However, transverse shear strains may become signicant in composite materials because of their relatively low shear strength or in structures with high-frequency vibrations. To account for the eect of transverse shear, numerous theories have been developed, such as the rst-order, Mindlin-type, and higher-order shear-deformation theories. Among many works, we quote those of Hildebrand et al. [1949], Mindlin [1951], Brunelle and Robertson [1976], Roufaeil and Dawe [1982], Librescu [1966; 1969], Murthy [1965], Schmidt [1977], and Librescu and Reddy [1989]. In general, the transverse shear can be neglected in homogeneous isotropic plates when the span-to-thickness ratio is greater than 20, and it may have signicant eects on the deection, buckling load, and fundamental frequency of composite plates [Chia, 1980]. Since homogeneous isotropic plates with high span-to-thickness ratio are our major concern, we focus on classical plate theories in this Dissertation.
13
Xiaopeng Zhao
Chapter 2. Literature Review
14
A large number of publications have been devoted to the linear vibrations of plates. Analytical methods as well as various numerical approaches have been widely used in practice. A comprehensive treatment of the linear problem is given in the text of Leissa [1993]. For an up to date review of linear plate analysis, we refer the reader to the monograph of Nayfeh and Pai [2004]. The problem of nonlinear vibration of plates is of considerable interest as the large-amplitude vibrations of plates are gaining more and more attention in the aerospace industries. A signicant geometric nonlinearity is induced when a plate is deected more than approximately one-half of its thickness, especially when in-plane edge constraints are present [White, 1971]. Due to the geometric nonlinearity, the resonance frequencies and mode shapes are amplitude dependent. The geometric nonlinearity also causes jumps as well as internal resonances. The dynamic analogue of von K arm an equations, which take into account stretching of the mid-surface, were widely used for free and forced nonlinear vibrations of plates. Approximate solutions were obtained by Yamaki [1961] for rectangular and circular plates with various boundary conditions. Kung and Pao [1972] used a combination of the Galerkin method and the method of harmonic balance to analyze axisymmetric vibrations of circular plates. Eisley [1964] used the Galerkin method in combination with the Lindstedt-Poincar e technique to investigate the mainresonant vibrations of buckled rectangular plates. Huang and Sandman [1971] used the Kantorovich method to study axisymmetric vibrations of circular plates. None of these studies considered multi-mode solutions. Sridhar et al. [1975; 1978] and Lobitz et al. [1977] investigated primary resonant responses of a circular isotropic plate. They considered the interaction of modes and the possibility of multi-mode solutions. Using the method of multiple scales, Hadian and Nayfeh [1990] studied the responses of circular plates subjected to a harmonic external excitation. They found that chaotically modulated motions were generated through a Hopf bifurcation resulting from the multi-mode response losing stability. Multi-mode responses of plates were also studied by Zhu et al. [1995] and Oh and Nayfeh [1998]. Shi and Mei [1996] studied large-amplitude free vibrations of plates using a reduced-order model obtained by introducing a modal expansion into the nonlinear equations of motion. Several authors have combined the hierarchical nite-element method and the method of harmonic balance to study geometrically nonlinear vibrations of thin isotropic plates. Based on this method and continuation
Xiaopeng Zhao
15
procedures, Ribeiro and Petyt [1999; 2000] investigated nonlinear vibrations of plates and the eect of internal resonance. The dynamic behavior of plates for large-vibration amplitudes was examined both theoretically and experimentally by Benamar [1990] and Benamar et al. [1993]. Using the nite-element method, Mei [1973], Chiang et al. [1991], and Zhou et al. [1994] investigated the nonlinear forced vibrations of plates. Teng et al. [1999] used Fourier series to study the nonlinear forced vibration of rectangular plates. Azrar [2002] studied the nonlinear forced vibrations of plates by coupling the nite-element method and perturbation methods. A starting point is computed using the nite-element method. Then the nonlinear governing equation is transformed into a sequence of linear problems by applying perturbation techniques around the starting point. They considered various types of harmonic excitation forces, such as distributed and concentrated ones. For more comprehensive details, we refer the reader to the books of Timoshenko [1959], Nayfeh and Mook [1979], Chia [1980], Yu [1996], Sathyamoorthy [1997], Nayfeh [2000], and Nayfeh and Pai [2004].
2.1.2
Equations of Motion
In this section, we derive the equations of motion and boundary conditions governing the nonlinear dynamics of isotropic thin plates, using a combination of the classical plate theory and the von K arm an-type nonlinearity. We restrict our attention to rectangular plates. The linear classical plate theory is based on the Kirchho hypothesis, which assumes that [Nayfeh and Pai, 2004]
(1) the deection of the midplane is smaller than the plate thickness;
(2) the
midplane remains unstrained and neutral during bending; (3) plane sections initially normal to the midplane remain plane and normal to the midplane after bending, which implies that transverse shear strains are neglected; and (4) the transverse normal stress is negligible compared to the other stress components. When the transverse deection of a plate is large, assumptions (1) and (2) are no longer applicable because the midplane is strained.
Xiaopeng Zhao
16
We consider a rectangular plate in the domain 0 x a and 0 y b. In the absence of body forces, the extended Hamilton principle gives
t2 t1
(T + Wnc )dt = 0
(2.1)
where Wnc denotes variation of the nonconservative energy Wnc , and variations of the kinetic and elastic energies T and are given by T =
z A
DdAdz D
(2.2)
and =
z A
(11
11
+ 22
22
+ 33
33
+ 23
23
+ 13
13
+ 12
12 ) dAdz.
(2.3)
Here, is the mass density, A denotes the undeformed area of the reference plane, D denotes the displacement vector of an arbitrary point of the dierential plate element under observation, the ij are the Jaumann stresses, and the
ij
are the Jaumann strains.
For convenience, we assume that the x- and y -axes lie in the middle plane. Further, we denote the displacement of an arbitrary point (x, y, z ) in the plate by ( u, v , w ) and denote the displacement of the corresponding point (x, y, 0) by (u, v, w). Based on the Kirchho assumptions, we have u = u zw , v = v zw , w = w. The von K arm an nonlinear strains [Nayfeh and Pai, 2004] are given by
11 22 12 33
(2.4)
= e1 zwxx , = e2 zwyy , = 6 2zwxy , =

13
23
= 0,
(2.5)
where the midplane strain components e1 , e2 , and 6 are 1 2 e1 = ux + wx , 2 1 2 e2 = vy + wy , 2 6 = uy + vx + wx wy . (2.6)
Xiaopeng Zhao
17
For thin plates, we adopt the plane stress assumption. The stress-strain relations for an isotropic material are 11 22 = E 1 2 ux 1 0 vy 0 0 (1 ) /2 uy + vx 1 0 z w xx (2.7)
12
. wyy 2w xy
We denote the unit vectors along the axes x, y , and z by i, j, and k , respectively, then the displacement vector D is given by D=u i + v j + w k. (2.8)
Substituting (2.4) into (2.8) and taking the time derivatives and variation of the displacement vector D, we obtain = ( D u zw x )i + ( v zw y )j + w k and D = (u zwx )i + (v zwy )j + wk. (2.9) (2.10)
Substituting (2.9) and (2.10) into (2.2) yields T =

A
(I0 u I1 w x ) u + (I0 v I1 w y ) v + I0 ww
+ (I2 w x I1 u ) wx + (I2 w y I1 v ) wy dA =
A
(I0 u I1 w x ) u + (I0 v I1 w y ) v
+ I0 w (I2 w x I1 u )x (I2 w y I1 v )y w dA
y =a [I2 w x I1 u ]x x=0 wdy x =b [I2 w y I1 v ]y y =0 wdx
(2.11)
where {I0 , I1 , I2 } {1, z, z 2 }dz.

z
(2.12)
We note that I1 = 0 due to the homogeneous assumption.
Xiaopeng Zhao
18
Substituting (2.5) and (2.6) into (2.3) yields =

A z
11 (ux + wx wx zwxx ) + 22 (vy + wy wy zwyy )
+ 12 (uy + vx + wx wy + wy wx 2zwxy ) dAdz =

A
N1 (ux + wx wx ) + N6 (uy + vx + wx wy + wy wx ) + N2 (vy + wy wy ) M1 wxx M2 wyy 2M6 wxy dA =

A
(N1x + N6y ) u + (N2y + N6x ) v
+ M1xx + M2yy + 2M6xy + (N1 wx )x + (N2 wy )y + (N6 wx )y + (N6 wy )x w dA +

y
N1 u + N6 v + M1x + 2M6y
x=a dy x=0
+ N1 wx + N6 wy w M1 wx +
x
N6 u + N2 v + M2y + 2M6x
y =b dx y =0
+ N2 wy + N6 wx w M2 wy 2M6 w
(x,y )=(0,0),(a,b) (x,y )=(a,0),(0,b)
(2.13)
where the internal forces Ni and moments Mi are dened as {N1 , N2 , N6 } = {M1 , M2 , M6 } = {11 , 22 , 12 }dz, z {11 , 22 , 12 }dz. (2.14)
Substituting (2.11) and (2.13) into (2.1) and setting each of the coecients of u, v, and w equal to zero, we obtain the equations of motion as N1x + N6y = I0 u , N6x + N2y = I0 v , M1xx + 2M6xy + M2yy + (N1 wx + N6 wy )x + (N6 wx + N2 wy )y = I0 w I2 (w xx + w yy ) . (2.15) (2.16) (2.17) (2.18)
Xiaopeng Zhao
19
The corresponding boundary conditions are Along x = 0, a : u = 0 v = 0 w = 0 wx = 0 Along y = 0, b : u = 0 v = 0 w = 0 wy = 0 or or or or N6 = 0 N2 = 0 M2y + 2M6x + N6 wx + N2 wy + I2 w y = 0 M2 = 0 or or or or N1 = 0 N6 = 0 M1x + 2M6y + N1 wx + N6 wy + I2 w x = 0 M1 = 0
At (x, y ) = (0, 0), (a, b), (a, 0), (0, b) : w = 0 or M6 = 0 (2.19)
Substituting (2.7) into (2.14) yields N 1 N 2 N6 [Aij ] = 0 M1 M2 M6 where Eh [Aij ] = 1 2 1 0 0 0 (1 )/2 1 0 e1 e2 6 0 1 2 wxx 12 h [Aij ] wyy 2wxy . (2.21) (2.20)
Xiaopeng Zhao
20
Substituting (2.20) and (2.21) into (2.15)(2.17), we obtain the following equations of motion expressed in terms of displacements: 2u 1 2v 1 2 u w 2 w + (1 + ) + (1 ) 2 + 2 x 2 xy 2 y x x2 1 w 2 w 1 w 2 w 1 2 2u + (1 + ) = + (1 ) I 0 2 y xy 2 x y 2 Eh t2 2v 1 w 2 w 2u 1 2v + + (1 + ) + (1 ) y 2 2 xy 2 x2 y y 2 1 w 2 w 1 w 2 w 1 2 2v + (1 + ) + (1 ) = I 0 2 x xy 2 y x2 Eh t2 D4 w + I0 where D
Eh3 . 12(1 2 )
2w 2 2w 2w 2w 2 = I w + N + 2 N + N 2 1 6 2 t2 t2 x2 xy y 2
(2.22)
In general, the inplane natural frequencies are very large compared to the natural frequencies of the transverse motion; therefore eects due to the inplane inertia terms Dropping the inplane inertia terms in (2.15)(2.17) yields N1x + N6y = 0, N6x + N2y = 0, M1xx + 2M6xy + M2yy + N1 wxx + 2N6 wxy + N2 wyy = I0 w I2 (w xx + w yy ) , (2.23) (2.24) (2.25)
2u t2
and
2v t2
are negligible.
Equations (2.23) and (2.24) can be automatically satised by introducing a stress function (x, y, t) as N1 = yy , N2 = xx , N6 = xy . Then, equation (2.25) becomes M1xx + 2M6xy + M2yy + wxx yy 2wxy xy + wyy xx = I0 w I2 (w xx + w yy ) Substituting for the Mi from (2.20) into (2.27), we obtain D4 w + hw = wxx yy 2wxy xy + wyy xx + I2 2 w (2.28) (2.27) (2.26)
Xiaopeng Zhao
21
Eliminating u and v from (2.6) yields the compatibility equation 2 e1 2 e2 2 e6 + 2 = y 2 x2 xy 2w xy

2
2w 2w x2 y 2
(2.29)
where e6 = 6 /2. Solving equation (2.20) for e1 , e2 , and e6 and substituting equation (2.26) into the outcome, we obtain e1 = e2 = 1 (yy xx ) , Eh (2.30) (2.31) (2.32)
1 ( yy + xx ) , Eh 1+ xy . Eh
e6 = Substituting (2.30)(2.32) into (2.29) yields
2 4 = Eh wxy wxx wyy .
(2.33)
Here equations (2.28) and (2.33) form the equations of motion in terms of w and .
2.1.3
Methods of Solution
Since exact solutions for the nonlinear equations are in general not available, one has to resort to purely numerical techniques, purely analytic techniques, or numerical-analytic techniques. The principal numerical methods available for dynamic deformations may be divided into nite dierences in space and time, nite element in space and nite dierence in time, and nite element representation in both space and time. These methods may be costly in computation time and may have diculty in yielding complicated system responses. The purely analytic methods are generally suitable for simple structures with simple boundary conditions. The temporal problem is usually solved by applying a perturbation method [Nayfeh, 1973, 1981, 1985, 1993; Nayfeh and Chin, 1999]. The spatial problem is usually solved by using the method of weighted residuals. According to Nayfeh and Mook [1979], the numerical-analytic approaches can be divided into three groups. In the rst group, the dependence of the solution on time is postulated in the form w (r, t) = (t) (r) ,
Xiaopeng Zhao
22
where is assumed to be harmonic. Then the method of harmonic balance is applied to obtain a nonlinear boundary-value problem for (r) . The second group assumes the following nonlinear matrix equation: Mq + Cq + K0 + K1 (q) + K2 q2 + + Kn (qn ) q = f , where M , C , and Kn are N N matrices and q and f are N 1 matrices. Then one solves for q using the method of equivalent linearization. In the third group, the displacement is expressed in the form
N
w (r, t) =
n=1
n (t) n (r)
where n (r) is the nth linear undamped natural modes of the system, which can be solved, for example, using the nite-element method. Then the orthogonality property of the n is utilized to generate a system of N coupled ordinary-dierential equations for the n . These equations can be solved using either a perturbation technique or a numerical method.
Xiaopeng Zhao
23
2.2
2.2.1
Dynamic Tools for Nonsmooth Systems

Dynamical systems
An n-dimensional dynamical system can be represented by a set of rst-order dierential equations of the form x = f (x) , (2.34)
where x and f are known as state and vector eld, respectively. If the vector eld f is continuous and has continuous partial derivatives throughout some region of state space, then there exists a unique solution satisfying the initial condition x (t0 ) = x0 for any arbitrary t0 and x0 in the region. The dynamical system (2.34) is said to be autonomous because the vector eld is not explicitly dependent on time t. Autonomous dynamical systems are characterized by the uniqueness of their solutions and by the fact that changes in the state depend only on the elapsed time and not on the absolute time. A time-dependent dynamical system can be converted to an autonomous one by the inclusion of the time variable as an additional state variable. (2.35)
2.2.2
Flows
Generally, a closed-form solution to (2.34) that satises a given initial condition is not available. Nevertheless, the basic existence and uniqueness result for an autonomous dynamical system implies the existence of a vector function (x, t), known as the ow function, that is as smooth as the vector eld such that (x, t) = f ( (x, t)) t and (x, 0) = x, (2.37) (2.36)
for any x and all t in some interval containing 0. The ow function represents the collection of state-space trajectories corresponding to solutions to the dynamical system based at dierent initial
Xiaopeng Zhao conditions.
24
In a smooth dynamical system, the sensitivity of the solution to changes in the initial conditions is given to lowest order by the Jacobian of the ow function x (x, t). From the properties of the ow function, it follows that x (x, t) must satisfy the linear initial-value problem x (x, t) = fx ( (x, t)) x (x, t) , t x (x, 0) = Id. (2.38) (2.39)
Equation (2.38) is known as the rst variational equation . Here, Id represents the identity matrix. In a piecewise smooth dynamical system, such as the model of the Mita actuator [Zhao et al., 2004b], it is necessary to consider discontinuous changes in x (x, t) associated with discontinuous changes in the state and/or vector eld. Suppose that a dynamical system discontinuity can be represented by a state-space discontinuity surface D given by the zero-level surface of an event function h (x) and an associated jump map g (x). We consider a solution trajectory based at the initial condition x0 that intersects D transversally at a point xin after an elapsed time tin , as shown in Figure 2.1. We denote by xout the image of the jump map applied at the point xin ; that is, xout = g xin . Finally, we denote by x1 the state on the solution trajectory based at xout after an additional elapsed time tout . It follows that x1 = g x0 , tin where h x0 , tin and hx x0 , tin f x0 , tin = hx xin f xin = 0. (2.41) = h xin = 0 (2.40) , tout ,
Trajectories based at initial conditions in a neighborhood of x0 reach a neighborhood of x1 in the xed time t = tin + tout . For a point x near x0 , the time of ight to the discontinuity surface is a small perturbation from the elapsed time tin . Thus, these trajectories can be described by the following map: g x, tin + , tout , (2.42)
Xiaopeng Zhao
25
x0
D
xin g x out x1
Figure 2.1: A ow passes a discontinuity surface. where is the time needed to reach the discontinuity surface from a point near xin . To nd the small time correction , we consider the function F (x, ) = h ( (x, )) . It follows that F xin , 0 = h xin = 0 and F xin , 0 = hx xin f xin = 0. (2.45) (2.44) (2.43)
It follows from the implicit function theorem that (x) is a unique-valued smooth function for x xin , such that (xin ) = 0 F (x, (x)) = h ( (x, (x))) 0, and x xin = h
x
(2.46) (2.47)
hx xin . xin f xin
Using the function (x), we dene a zero-elapsed-time discontinuity mapping D (x) = (g ( (x, (x))) , (x)) , such that D xin = xout . (2.49) (2.48)
Xiaopeng Zhao
26
To lowest order, changes in D (x) that result from changes in x away from xin are captured by the Jacobian Dx xin = gx xin + f (xout ) gx xin f xin hx xin . hx xin f xin
The map in Equation (2.42) can be written as D x, tin , tout
and, consequently, the sensitivity to changes in initial condition is, to lowest order, given by x (xout , tout ) Dx xin x x0 , tin , (2.50)
where the rst and third factors are given as solutions to the corresponding variational equations.
2.2.3
Periodic solutions
A solution trajectory that forms a closed curve in state space is called a periodic trajectory. It follows that there exists a nonzero quantity T, such that (x , t + T ) = (x , t) (2.51)
for any x on the trajectory and for all t. Periodic orbits are most conveniently studied through the introduction of a Poincar e section and its associated Poincar e map P (x). Specically, suppose that the periodic orbit intersects a Poincar e section given by the zero-level surface of the smooth event function h (x) transversally at a point x ; that is, h ( (x , T )) = h (x ) = 0 and hx (x ) f (x ) = 0. Now consider the scalar-valued function F (x, t) = h ( (x, t)) . Then F (x , T ) = 0 (2.53) (2.52)
Xiaopeng Zhao and
27
Ft (x , T ) = hx (x ) f (x ) = 0. It follows from the implicit function theorem (see Appendix A) that there exists a unique smooth function (x) T for x x , such that (x ) = T and F (x, (x)) = h ( (x, (x))) 0. (2.54)
Here, (x) is the elapsed time along trajectories based at initial conditions near x until an intersection with the Poincar e section. Using the above identity, it is possible to compute the derivatives of the function (x) at the point x , for example, x (x ) = hx (x ) x (x , T ) . hx (x ) f (x )
The existence of the function (x) allows us to dene a Poincar e map P as P (x) = (x, (x)) that maps points near x along the corresponding trajectory onto the Poincar e section given by the zero-level surface of h (x). Specically,
P (x ) = x is a xed point of the Poincar e map.
(2.55)
The local behavior of a dynamical system in the vicinity of a periodic orbit can be quantitatively described in terms of the sensitivity of the Poincar e map to changes of x away from the xed point x . To lowest order, this is given by the Jacobian of the Poincar e map evaluated at x Px (x ) = Id f (x ) hx (x ) x (x , T ) , hx (x ) f (x ) (2.56)
where x (x , T ) is obtained from the solution of the variational equations including discrete corrections due to discontinuities as discussed above . From the Hartman-Grobmann theorem, it follows that if all of the eigenvalues of Px (x ) lie within the unit circle in the complex plane, the xed point is asymptotically stable. If at least one eigenvalue lies outside the unit circle, the xed point is unstable. The crossing of an eigenvalue through the unit circle is typically associated with a bifurcation in system characteristics, such as the appearance of periodic trajectories with half the frequency (period doubling).
Xiaopeng Zhao
28
2.2.4
Continuation of periodic solutions
Given a Poincar e section and the associated Poincar e map P, the existence of a periodic solution to the original dynamical system for a particular choice of system parameters is equivalent to the existence of a xed point x, such that F (x, ) = P (x, ) x = 0. Suppose that F (x0 , 0 ) = 0 and that the Jacobian Fx (x0 , 0 ) = Px (x0 , 0 ) Id (2.59) (2.58)
def
(2.57)
is nonsingular (i.e., that Px (x0 , 0 ) has no eigenvalue equal to 1). From the implicit function theorem, it follows that there exists a unique smooth function () for 0 , such that (0 ) = x0 , F ( () , ) = P ( () , ) () = 0 and (0 ) = [Fx (x0 , 0 )]1 F (x0 , 0 ) . (2.62) (2.60) (2.61)
We may now implement the Newton-Raphson root-nding scheme for nding (1 ) for 1 0 using the initial guess (1 ) (0 ) + (0 ) (1 0 ) = (0 ) [Fx (x0 , 0 )]1 F (x0 , 0 ) (1 0 ) (2.63)
2.2.5
Continuation of grazing periodic solutions.
In the above analysis, we assumed that all intersections with state-space discontinuities are transversal, guaranteeing their persistence under variations in system parameters. Suppose instead that a periodic solution has been found, which achieves simple tangential contact with a state-space
Xiaopeng Zhao discontinuity D at a point x0 ; that is,
29
h (x0 ) = 0, hx (x0 ) f (x0 ) = 0, and f T (x0 ) hxx (x0 ) f (x0 ) + hx (x0 ) fx (x0 ) f (x0 ) = 0,
(2.64) (2.65)
(2.66)
where D is the zero-level surface of the event function h (x). We refer to the periodic trajectory as a grazing solution. Now introduce a Poincar e surface P given by the zero-level surface of the function H (x) = hx (x) f (x) and dene the function F (x, t) = H ( (x, t)) . Then, F (x0 , T ) = hx (x0 ) f (x0 ) = 0 and Ft (x0 , T ) = Hx (x0 ) f (x0 ) = f T (x0 ) hxx (x0 ) f (x0 ) + hx (x0 ) fx (x0 ) f (x0 ) = 0. (2.70) (2.69) (2.68) (2.67)
It follows from the implicit function theorem that there exists a unique smooth function (x) for x x0 , such that (x0 ) = T, F (x, (x)) = H ( (x, (x))) 0, and x (x0 ) = Hx (x0 ) x (x0 , T ) . Hx (x0 ) f (x0 ) (2.73) (2.71) (2.72)
We can now uniquely dene the Poincar e map P (x) = (x, (x)) (2.74)
Xiaopeng Zhao
30
on a neighborhood of the point x0 , such that P (x0 ) = x0 . (2.75)
It is clear that the grazing solution is not persistent under changes in a single system parameter, since two conditions must be satised for its existence. Instead, we must consider simultaneous changes in two system parameters, say and . Consider the function P (x, , ) x , F (x, , ) = h (x)
(2.76)
where we have explicitly included the parameters in the expression for the Poincar e map. It follows that F (x0 , 0 , 0 ) = Moreover, F(x,) (x0 , 0 , 0 ) = Px (x0 , 0 , 0 ) Id P (y0 , 0 , 0 ) Hx (x0 ) 0 P (x0 , 0 , 0 ) x0 h (x0 ) . (2.78) = 0. (2.77)
If this quantity is invertible, then the implicit function theorem guarantees the existence of unique smooth functions ( ) and m ( ) for 0 , such that (0 ) = x0, m (0 ) = 0 , F ( ( ) , m ( ) , ) = and P ( ( ) , m ( ) , ) ( ) h ( ( )) where (0 ) m (0 ) = F(x,) (x0 , 0 , 0 ) F (x0 , 0 , 0 ) = P (x0 , 0 , 0 ) 0
1
(2.79) 0 0 , (2.80) (2.81)
F (x0 , 0 , 0 ) , .
(2.82)
(2.83)
We may again implement the Newton-Raphson scheme for continuation of grazing periodic solutions under changes in the parameter .
Chapter 3
Mechanical Behavior of Electrically Actuated Microplates

3.1 Motivation
Electrostatically actuated microplates are the main component in micropumps, micromirrors, microphones, and many microsensors [Zengerle et al., 1992; Zhang et al., 2001; Hsu et al., 1998; Scheeper et al., 1999; Tilmans and Legtenberg, 1994]. Applications of micropumps spread into the elds of biotechnology, chemical, food, and mining industries. A micropump is used to manipulate very small and precise volumes. A microplate forms one side of a variable capacity air-gap capacitor, as shown in Figure 3.1. The electrostatic eld is created by applying a potential dierence between the microplate and a xed electrode. As the electrostatic force deforms the microplate, the electrostatic force itself changes with deection, which illustrates the electrical and mechanical coupling mechanism. The applied electrostatic load has an upper limit beyond which the mechanical restoring force of the microplate can no longer resist the electrostatic force, thereby leading to the collapse of the structure. This structural instability phenomenon is known as pull-in, and the critical voltage associated with it is called the pull-in voltage.
31
Xiaopeng Zhao
Chapter 3. Mechanical Behavior of Electrically Actuated Microplates
32
Stationary Electrode
v(t)
Microplate
Figure 3.1: A schematic of a micropump. The microplate restoring force is composed of three components: the plate resistance to bending, the membrane stiness due to the externally applied inplane loads, and midplane stretching due to the fact that the deection coupled with boundary constraints create inplane forces in the plate. Increasing the electrostatic force shifts the natural frequencies to lower values. On the other hand, increasing the mechanical restoring force shifts the natural frequencies to higher values. Analysis of MEMS devices is challenging because the classical structural dynamics methodology is not easily applicable to the types of forcing and nonlinearities encountered in MEMS. A common approach in the literature is to assume a linear relationship between the excitation force and the plate deection. The linear plate equation is then solved by numerical methods, such as the Galerkin method, the Rayleigh-Ritz method, and the nite-element method. However, the linear plate theory is only appropriate for small deections. When the plate deection is comparable to the thickness of the plate, which is the case in most microplates in typical MEMS devices, it produces erroneous results and one has to resort to a nonlinear plate theory [Timoshenko and Woinowsky-Krieger, 1959]. Francais and Dufour [1999] studied the static deection of a diaphragm under an electrostatic excitation. They attacked the linear plate equation using the Rayleigh-Ritz method and polynomials as basis functions. They developed an iterative procedure to solve for the deection at a given voltage. Their numerical results are in good agreement with their experimental measurements. Their analysis and results are limited to small deections only.
Xiaopeng Zhao
33
Ng et al. [2004] studied the dynamic characteristics of an electrically actuated microplate. The Laplace equation governing the potential dierence between the electrodes was solved using the boundary-element method (BEM). Geometric nonlinearities due to midplane stretching were included in the plate model, which was discretized using the nite-element method (FEM). The coupled electromechanical equations were then solved by iteration. They found qualitative dierences between the results of nonlinear analysis and those of linear analysis when the plate undergoes large deection. Because numerous variables are generated by FEM models, this approach makes it dicult to map the design space and identify the relationships among the design parameters [Younis et al., 2003]. In this chapter, we study the mechanical behavior of electrically actuated rectangular microplates through reduced-order modelling. Simply supported plates are treated rst because of their simplicity. Then the modeling is extended for plates with general boundary conditions.
3.2
3.2.1
Microplates with Simply Supported Boundary Conditions

Problem Formulation
We consider an electrically excited microplate with the geometry specied in Figure 3.2. For simplicity, we adopt the von K arm an nonlinear plate equations in terms of the transverse deection w and the stress function F , see Chapter 2 for derivation of these equations; that is, 4 w = p h h + (wxx Fyy + wyy Fxx 2wxy Fxy ) w D D D
2 4 F = E (wxy wxx wyy )
(3.1) (3.2)
where is the mass density and h is the plate thickness. The exural rigidity D is D= Eh3 12(1 2 ) (3.3)
where E is Youngs modulus and is Poissons ratio. The electric pressure p is approximated by p= 1 v (t)2 2 (d w)2 (3.4)
Xiaopeng Zhao
34
where d is the capacitor gap, v (t) is the applied voltage, and dierential operator 4 in Equations (3.1) and (3.2) is 4 = 4 4 4 +2 2 2 + 4 4 x x y y
is the dielectric constant. The
(3.5)
In this section, we consider the simply supported boundary conditions w = 0 and wxx + wyy = 0 at x = 0 and a w = 0 and wyy + wxx = 0 at y = 0 and b (3.6)
For a plate with immovable edges, the additional boundary conditions are [Chia, 1980] u = v = 0 at x = 0 and a u = v = 0 at y = 0 and b Alternatively the movable the inplane boundary conditions are [Chia, 1980] Fxy = 0 and Fyy = 0 at x = 0 and a Fxy = 0 and Fxx = 0 at y = 0 and b (3.9) (3.10) (3.7) (3.8)
For convenience, we introduce nondimensional variables (denoted by hats) w = where T = w , d = F , F Ed2 x = x , a y = y , b = t t T (3.11)
ha4 /D. Substituting Equation (3.4) into Equations (3.1)-(3.3) and dropping the
2 (1 2 ) 121 2 v (t)2 ( w, F ) + w 2 (1 w)2
hats, we obtain L(w) = (3.12) (3.13)
L(F ) = and
1 (w, w) 22
w = 0 and wxx + wyy = 0
at x = 0 and 1 (3.14)
w = 0 and wyy + wyy = 0 at y = 0 and 1
Xiaopeng Zhao
35
a y x yx z V
Stat
Mic
ropl ate
i o na
ry E lect rode
Figure 3.2: Geometry of a microplate-based capacitor. where the operators and L are dened as (f, g ) = fxx gyy + fyy gxx 2fxy gxy L=
4 4 4 2 4 + 2 + x4 x2 y 2 y 4
(3.15) (3.16)
The parameters appearing in Equations (3.12) (3.14) are = d 6 a4 (1 2 ) b , 1 = , 2 = a h Eh3 d3 (3.17)
A transverse displacement eld and a stress function satisfying the boundary conditions are assumed in the form w=
m,n
mn (x, y ) wmn (t)
(3.18)
1 F = (py x2 + px y 2 ) + mn (x, y )fmn (t) 2 m,n where mn = sin(mx) sin(ny ), and
(3.19)
mn = cos(mx) cos(ny )

(3.20)
m,n m=1 n=1
(3.21)
Xiaopeng Zhao
36
Here, px and py are constants equal to the average membrane pressure in the x- and the y -directions, respectively. For immovable in-plane constraints, px = py = 2 8(1 2 ) 2 8(1 2 )

(2 m2 + n2 ) wmn
m=1 n=1
(3.22) (3.23)
( m2 + 2 n2 ) wmn
m=1 n=1
For movable in-plane constraints, px and py are zero [Timoshenko and Woinowsky-Krieger, 1959; Levy, 1942]. The coecient fmn is related to wmn as follows: fmn = 4(m2 1 + 1 n2 )2 brspq wrs wpq (3.24)
where the sum includes all products for which r p = m and s q = n. The coecient brspq is given by the expression brspq = 2rspq (r2 q 2 + s2 p2 ) (3.25)
where the positive sign applies for either the case r + p = m and s q = n or the case r p = m and s + q = n, and the negative sign applies otherwise [Timoshenko and Woinowsky-Krieger, 1959]. Substituting Equations (3.18) and (3.19) into Equation (3.12), multiplying the resulting equation by (1 w)2 rs , and integrating over the domain, we obtain the following system of nonlinear
Xiaopeng Zhao
37
ordinary-dierential equations:

M 1rs w rs 2
m,n i,j
M 2rsmnij w mn wij M 3rsmnijkl w mn wij wkl
+
m,n i,j k,l
+ K 1rs wrs 2
m,n i,j
K 2rsmnij wmn wij
+
m,n i,j k,l
K 3rsmnijkl wmn wij wkl

+ H 1rs wrs 2
m,n i,j
H 2rsmnij wmn wij
(3.26)
+
m,n i,j k,l
H 3rsmnijkl wmn wij wkl G1rsmnpq wmn fpq

m,n p,q
+2
m,n p,q
G2rsmnpqij wmn fpq wij G3rsmnijkl wmn fpq wij wkl

i,j k,l
m,n p,q i,j
rs = 0, F where the coecients are dened as M 1rs = 1/4

1 1 0 1
r, s = 1, 2, 3, . . . ,
M 2rsmnij = M 3rsmnijkl =
rs mn ij dxdy
1 0
(3.27)
rs mn ij kl dxdy
K 1rs = 4 (m2 + 2 n2 )2 M 1rs K 2rsmnij = 4 (m2 + 2 n2 )2 M 2rsmnij K 3rsmnijkl = 4 (m2 + 2 n2 )2 M 3rsmnijkl (3.28)
Xiaopeng Zhao
Chapter 3. Mechanical Behavior of Electrically Actuated Microplates H 1rs = 2 (px m2 + py n2 )M 1rs H 2rsmnij = 2 (px m2 + py n2 )M 2rsmnij H 3rsmnijkl = 2 (px m2 + py n2 )M 3rsmnijkl
1 1 0 1
38
(3.29)
G1rsmnpq = G2rsmnpqij =
pq )dxdy mn ( 1rs mn pq 2 2rs

1
0 1
pq ij )dxdy mn ( 1rs mn pq ij 2 2rs

1
(3.30)
G3rsmnpqijkl =
mn pq ij kl )dxdy ( 1rs mn pq ij kl 2 2rs

1 0 0 1
rs = 2 v (t)2 F where
rs dxdy
(3.31)
mn = sin(mx) sin(ny ) mn = mn = cos(mx) cos(ny ) mn = and 1 = 4 (q 2 m2 + p2 n2 ), 2 = 4 mnpq, =

2 (1 2 ) 121 2
(3.32) (3.33)
(3.34)
3.2.2
Static Response
To study the static deection of the microplate under a DC voltage, we let v (t) = Vp and set the time derivatives equal to zero in Equation (3.26). We obtain a set of innite number of nonlinear algebraic equations. Keeping only the lowest n modes yields an n-mode approximate solution. Because the basis functions mn (x, y ) form a complete set, as n approaches , the solution approaches the exact solution. To guarantee the accuracy of the numerical solution, we rst investigate convergence of the results. In Figure 3.3, we show variation of the maximum plate deection wmax = w(0.5, 0.5) with the electrostatic load using 1, 3, 8, and 15 mode approximations. All of the curves have a similar trend, and the results converge on the lower branch as the number of modes equals to 6. To further illustrate the inuence of the number of modes retained in the approximation, we show in Figures 3.4 and 3.5 the nondimensional electrostatic load and the maximum deection at
Xiaopeng Zhao

1 15 modes 0.8 8 modes 0.6 w max 0.4 0.2
39
2 modes 1 mode
25
50
75
100 Vp
125
150
175
Figure 3.3: Variation of the maximum nondimensional deection wmax with the nondimensional electrostatic load 2 Vp2 when = 1 and 1 = 2. pull-in calculated using 1 to 15 modes in the discretization. Both gures show that it is necessary to use at least 6 modes for convergence.
160
155 H V L PL
150
145
5
Number of Modes
10
15
Figure 3.4: Variation of the nondimensional electrostatic load at pull-in (2 Vp2 )P L with the number of modes when = 1 and 1 = 2. The ratio of the capacitor gap to the thickness of the microplate, denoted by 1 , was found to have a signicant impact on the plate response. According to linear theory, 1 has no eect on the response, as can be seen from Equations (3.12) and (3.13). Linear theory neglects midplane stretching (w, F ) in Equation (3.12), and hence does not need Equation (3.13). Therefore, linear
Xiaopeng Zhao

1 0.8 Hw maxLPL 0.6 0.4 0.2
40
5
Number of Modes
10
15
Figure 3.5: Variation of the maximum deection (wmax )P L at pull-in with the number of modes when = 1 and 1 = 2. plate theory generates a constant electrostatic load as well as a maximum deection at pull-in for any value of 1 . Linear theory produces these erroneous results because it fails to account for midplane stretching in the restoring force [Abdel-Rahman et al., 2002]. To study the inuence of midplane stretching, one has to resort to a nonlinear plate model, which accounts for the inplane movements of the middle plane. Two common inplane boundary conditions are movable and immovable edges [Timoshenko and Woinowsky-Krieger, 1959]. We calculate the electrostatic load and maximum deection at pull-in for various value of 1 using the linear and nonlinear models with movable and immovable inplane boundary conditions. As seen in Figures 3.6 and 3.7, the linear plate model produces a constant curve in both cases. The nonlinear plate model with movable inplane constraints produces almost linear curves. The nonlinear plate model with immovable inplane constraints produces highly nonlinear curves. The results of the nonlinear model with immovable constraints in Figures 3.6 and 3.7 are signicantly larger than those of the other two solutions because midplane stretching increases the stiness of plates and immovable boundary conditions amplies that eect. Because immovable inplane constraints are a more reasonable approximation for practical applications, we consider only this case in the remainder of this section. We show in Figures 3.8 and 3.9 the electrostatic load and maximum deection at pull-in for various values of the aspect ratio . As increases, the electrostatic load (2 Vp2 )P L at pull-in
Xiaopeng Zhao
41
140
H V L PL
120 100 80 60
0.6
0.8
1.2
1.4
1.6
1.8
Figure 3.6: Variation of the electrostatic load (2 Vp2 )P L with 1 at pull-in: (a) nonlinear plate model with immovable inplane constraints (+), (b) nonlinear plate model with movable inplane constraints ( ), and (c) linear plate model (solid line) when = 1. decreases, whereas the maximum deection (wmax )P L is unchanged. Figure 3.10 shows variation of wmax with 2 Vp2 . As increases, the plate stiness decreases. When 4.0, the plate deection is very close to that of an innitely long plate. Increasing beyond 4 has minimal impact on the stiness of the plate. Therefore, electrostatically actuated microplates with large aspect ratio may be regarded as innitely long plates.
3.2.3
Linear Mode Shapes and Corresponding Natural Frequencies
To study the linear vibration characteristics, we linearize the plate equations of motion around the deected position ws using the following expressions: w(x, y, t) = ws (x, y ) + u(x, y, t) F (x, y, t) = Fs (x, y ) + (x, y, t) where ws and Fs are given by L(ws ) =
2 (1 2 ) 2 Vp2 121 ( w , F ) + s s 2 (1 ws )2
(3.35) (3.36)
(3.37) (3.38)
L(Fs ) =
1 (ws , ws ) 22
Xiaopeng Zhao
42
0.65 0.6 Hw maxLPL 0.55 0.5 0.45 0.6 0.8 1 1.2
1.4
1.6
1.8
Figure 3.7: Variation of the maximum deection (wmax )P L at pull-in with 1 : (a) nonlinear plate model with immovable inplane constraints (+), (b) nonlinear plate model with movable inplane constraints ( ), and (c) linear plate model (solid line) when = 1. Substituting Equations (3.35) and (3.36) into Equations (3.12) and (3.13), using Equations (3.37) and (3.38), and keeping only the linear terms in u and , we obtain the following linear equations describing the free vibrations around ws : L(u) =
2 (1 2 ) 22 Vp2 121 ( w , ) + ( u, F ) + uu s s 2 (1 ws )3
(3.39) (3.40)
L() =
1 (ws , u) 2
We use the Galerkin method to discretize Equations (3.39) and (3.40) and solve the resulting algebraic eigenvalue problem for the fundamental natural frequency and the corresponding mode shape for various values of 1 . Figure 3.11 shows variation of the fundamental natural frequency
0 , with V 2 . For low 1 , normalized with respect to the natural frequency of the at plate 1 2 p
values of 1 , the fundamental natural frequency decreases as the electrostatic force increases and approaches zeros as pull-in develops. As 1 increases, the fundamental natural frequency increases for the same level of electrostatic forcing. At high values of 1 , for example 1 = 1.5 and 2.0, the fundamental natural frequency rst increases with the electrostatic force, then decreases, and eventually approaches zero, as predicted by Abdel-Rahman et al. [2002]. We normalize the mode shapes with respect to the maximum value. To visualize the mode shape, we plot the mode along the line y = 0.5. The rst mode shapes for a at square microplate
Xiaopeng Zhao
43
140 120 H V L PL
100 80 60 40 20 1 1.5 2 2.5 3 3.5 4
Figure 3.8: Variation of (2 Vp2 )P L with for 1 = 0.5 (+), 1 = 1.0 ( ), 1 = 1.5 ( ), and 1 = 2.0 (x). and a square microplate near pull-in are shown in Figures 3.12 and 3.13 for 1 = 0.5 and 1 = 2.0, respectively. The nearness of the deected position to pull-in is dened as 1 = 0.12. As 1 increases, the mode shape undergoes more pronounced changes with deection. In Figure 3.12, the dierence in the mode shape is insignicant, while Figure 3.13 shows a larger dierence.
3.3
3.3.1
Microplates with General Boundary Conditions

Problem Formulation
Again, we consider a capacitively actuated microplate with the geometry specied in Figure 3.2. We adopt the governing equations in terms of displacements to avoid complexity in the description of the general boundary conditions. Following the notations as in the book of Nayfeh and Mook [1979], Nayfeh [2000], and Nayfeh and Pai [2004], the dynamic analog of the von K arm an equations are 1 (1 )(uyy + vxy + wx wyy + wy wxy ) + uxx + wx wxx + (vxy + wy wxy ) = 0 2 1 (1 )(uxy + vxx + wx wxy + wy wxx ) + vyy + wy wyy + (uxy + wx wxy ) = 0 2 (3.41) (3.42)
Xiaopeng Zhao
44
0.7 0.65 Hw maxLPL 0.6 0.55 0.5
1.5
2.5
3.5
Figure 3.9: Variation of (wmax )P L with for 1 = 0.5 (+), 1 = 1.0 ( ), 1 = 1.5 ( ), and 1 = 2.0 (x). 1 2 1 2 1 2 4 h w Nxx wxx 2Nxy wxy Nyy wyy ux wxx wx wxx vy wyy wy wyy 12 2 2 1 2 1 2 (vy wxx + wy wxx + ux wyy + wx wyy ) (1 )(uy + vx + wx wy )wxy 2 2 1 = 2 wtt + f cp
(3.43)
where u(x, y ), v (x, y ), and w(x, y ) are the displacements in the x-, y -, and z -directions. The wave propagation speed and the electric forcing are c2 p = E (1 2 ) and f = (1 2 )V (t)2 , 2Eh(d w)2 (3.44)
where E is Youngs modulus, is Poissons ratio, h is the thickness of the plate, d is the capacitor gap, is the dielectric constant, V (t) is the applied voltage, and Nij is the applied force on the i-edge in the j -direction. The governing equations of motion (3.41)(3.43) are eighth-order partialdierential equations; therefore four boundary conditions are prescribed at each plate edge. For convenience, we introduce the following nondimensional variables (denoted by hats) u = au av w 2x 2y = t . , v = 2, w = , x = 1, y = 1, t 2 2d 2d d a b T (3.45)
Substituting equation (3.45) into equations (3.41)(3.43) and dropping the hats yields 1 (1 ) (uyy + vxy ) + 2 uxx + vxy 2 1 + (1 ) (wx wyy + wy wxy ) + 2 wx wxx + wy wxy = 0 2
(3.46)
Xiaopeng Zhao
45
=4.0 0.5 = 0.4
=3.0
wmax
0.3 =2.0 0.2 0.1 0 10 20 =1.0

30
40
50
Figure 3.10: Variation of the maximum deection with (2 Vp2 ) when 1 = 0.5.
1.4 1.2 1 1 0.8 0.6 0.4 0.2 0 25 50 75 1 =0.5 1 =1.0 1 =1.5 1 =2.0
100 Vp
125
150
175
Figure 3.11: Inuence of 1 on the normalized fundamental natural frequency. 1 (1 ) uxy + 2 vxx + vyy + uxy 2 1 1 + (1 ) (wx wxy + wy wxy ) + wy wyy + wx wxy = 0 2 wxxxx + 2 1 1 V (t)2 w + w + w xxyy yyyy tt 2 2 4 (1 w)2
(3.47)
1 1 2 = 30 Nxx wxx + 2 Nxy wxy + 2 Nyy wyy 1 1 1 1 1 2 + 121 ux + vy wxx + ux + vy wyy + (1 ) uy + vx wxy 2 1 2 1 1 2 1 1 2 1 2 2 + 121 wx + 2 wy wxx + wy + wx wyy + (1 ) 2 wx wy wxy 2 2 2 2
(3.48)
Xiaopeng Zhao
46
1 0.8 0.6 0.4 0.2 0 0.2 0.4 x 0.6 0.8 1
Figure 3.12: The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 0.5. The parameters appearing in equations (3.46)(3.48) are = and the time scale is set to T = a d 3 1 2 4 b , 0 = , 1 = , 2 = a a h h 8 Eh3 d3
ha4 16D .
(3.49)
Dropping the force and nonlinear terms in equation (3.48) yields wxxxx + 2 1 1 wxxyy + 4 wyyyy + wtt = 0. 2 (3.50)
Equation (3.50) and associated boundary conditions form an eigenvalue problem. Analytical solutions of this eigenvalue problem are only available for plates with at least one pair of opposite edges simply supported. The general eigenvalue problem can be solved using the nite-element method. In this Dissertation, we use the hierarchical nite-element method (HFEM) to solve this eigenvalue problem and the boundary-value problems associated with u, v, and w. Assuming that i (x, y ) is the ith eigenfunction, we write the transverse displacement eld satisfying the boundary conditions as
N
w=
i=1
i (x, y ) qi (t) .
(3.51)
Substituting equation (3.51) into equations (3.46) and (3.47) and considering the associated inplane boundary conditions yields a set of boundary-value problems for u and v. Using HFEM, we solve
Xiaopeng Zhao
47
1 0.8 0.6 0.4 0.2 0 0.2 0.4 x 0.6 0.8 1
Figure 3.13: The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 2.0. for u and v in terms of qi (t) in the form u = u (x, y, qi (t)) and v = v (x, y, qi (t)) . (3.52)
Multiplying both sides of equation (3.48) by (1 w)2 , substituting equations (3.51) and (3.52) into the outcome, and applying the Galerkin procedure, we obtain a set of nonlinearly coupled ordinary-dierential equations, which is the reduced-order model for the microplate.
3.3.2
Model Validation
Francais and Dufour [1999] measured the center deection of a fully clamped square microplate under various electrostatic actuations. In Figure 3.14, we compare the deection wmax at the center of the plate calculated using the reduced-order model with the experimental results of Francais and Dufour. There is good agreement. The macromodel shows robustness, being able to predict deections up to the pull-in voltage. The dots correspond to unstable equilibrium solutions and the solid line corresponds to stable equilibrium solutions calculated using the reduced-order model. We reinvestigate the simply supported plate problem using the present model. Figure 3.15 shows variation of the electrostatic pull-in load with the ratio 1 of the capacitor gap to the plate thickness. The results obtained using the analytical and present models are in excellent agreement. The maximum relative error is 0.5% obtained at 1 = 2. Variation of the maximum deection at
Xiaopeng Zhao

1 0.8 w max 0.6 0.4 0.2
48
+ + +
2 4
+
6 Vp
+
8 10 12
Figure 3.14: Comparison of wmax calculated using the reduced-order model with the experimental results (+) of Francais and Dufour [1999]. pull-in with 1 is shown in Figure 3.16. The maximum relative error is 0.6% obtained at 1 = 2.
140
120 100 80 60 0
0.5
1.5
Figure 3.15: Variation of the electrostatic pull-in load with 1 for a simply supported square microplate: analytical model developed in Section 3.2 (o) and present model (solid line).
3.3.3
Static Response
We investigate in this section the characteristics of fully clamped rectangular microplates, which are widely used in micropumps. We use the hierarchical nite-element method to calculate the natural frequencies and mode shapes of the undeected plates because they are not available analytically.
Xiaopeng Zhao
49
0.65 0.6 0.55 0.5
0 0.5
w max
1.5
Figure 3.16: Variation of the maximum deection at pull-in with 1 for a simply supported square microplate: analytical model developed in Section 3.2 (o) and present model (solid line). We rst perform a convergence study to determine the number of modes to be retained in the discretization in order to obtain accurate results. Then, we use the reduced-order model to perform a parametric study of the static and dynamic behaviors of these microplates. In Figure 3.17, we show variation of the maximum plate deection with the electrostatic load using 1, 3, 7, and 8 mode approximations. All of the curves have a similar trend, and the results converge on the stable branch as the number of modes increases.
1 0.8 w max 0.6 0.4 0.2 0 50 100 Vp 7 and 8 modes 3 modes
1 mode
150
200
Figure 3.17: Variation of the maximum deection wmax with the electrostatic load 2 Vp2 for a fully clamped square plate when 1 = 1.
Xiaopeng Zhao
50
In Figures 3.18 and 3.19, we show variations of the electrostatic load and maximum deection, respectively, at pull-in for various values of 1 , using linear and nonlinear plate theories. We note that 1 has no eect on the response obtained using the linear plate theory because it fails to account for midplane stretching, see the discussions in previous section. Consequently, the linear theory underestimates the device stability limits, predicting a constant electrostatic pull-in voltage and a constant maximum deection at pull-in. On the other hand, the nonlinear theory predicts a signicant impact of 1 on the plate response. The results obtained with the nonlinear model are signicantly larger than those obtained with linear theory because midplane stretching increases the stiness of the plate. When 1 is small, the plate deection is small compared to its thickness and the linear plate theory produces reasonable results. As 1 increases, the error resulting from using the linear plate theory increases exponentially. We note that 1 = 0.25 for the microplate built and used in the experiments of Francais and Dufour [1999]; therefore a linear plate model was able to predict reasonable results for this case.
+ 260 + + + + +
240 220 200 + ++ + 180 ++ ++++ 0 0.5 1 + + + +
1.5
Figure 3.18: Variation of the electrostatic pull-in load with 1 for a fully clamped square plate: nonlinear plate theory (+) and linear plate theory (solid line). Comparing Figures 3.15 and 3.18, we note that the pull-in voltage for a fully clamped square microplate is much larger than that for a simply supported square microplate for a given 1 . Further investigation revealed that the dierence in the trend of the predicted pull-in voltages of the two microplates is small when 1 varies from 0.1 up to 2.0, as shown in Figure 3.20.
Xiaopeng Zhao

+
51
0.6 + + 0.56 w max + + 0.52 + 0.48 ++ + ++ 0 0.5 + + + + + + +
1.5
Figure 3.19: Variation of the maximum deection at pull-in with 1 for a fully clamped square plate: nonlinear plate theory (+) and linear plate theory (solid line).
3.3.4
Linear Mode Shapes and Corresponding Natural Frequencies
Next, we study the linear vibration characteristics of the microplate around its deected positions. Figure 3.21 shows variation of the fundamental natural frequency 1 , normalized with respect to
0 , with the electrostatic load V 2 . For low values of , the natural frequency of the at plate 1 2 p 1
the fundamental natural frequency decreases as the electrostatic force increases and approaches zero as pull-in develops. As 1 increases, the fundamental natural frequency increases for the same level of electrostatic forcing. At high values of 1 , the fundamental natural frequency rst increases with the electrostatic force, then decreases, and eventually approaches zero. For ease of comparison, the mode shapes are normalized with respect to the maximum value. To visualize the mode shapes, we plot the modes along the line y = 0. The rst mode shapes around the at position and near pull-in are shown in Figures 3.22 and 3.23 for 1 = 0.5 and 1 = 2.0, respectively. The nearness of the deected position to pull-in is dened as 1 = 0.1. For both cases, there are no signicant changes in the mode shapes as the microplate deects and approaches pull-in.
Xiaopeng Zhao

+ 260 + + + + + +
52
240 220 200 + ++ + 180 ++ ++++ 0 0.5 1 + + +
1.5
Figure 3.20: Comparison of the electrostatic load at pull-in obtained for a fully clamped square plate (+) with that obtained for a simply supported square plate (solid line) for various value of 1 . The pull-in loads for the simply supported plate have been shifted upward by 120 for convenience of comparison.
3.4
Summary
Many MEMS devices use electrically actuated microplates as actuation components, which often undergo deections comparable to their thicknesses; therefore nonlinear plate equations are necessary because of the geometric nonlinearities caused by midplane stretching. We rst developed the reduced-order model for a rectangular microplate with simply supported boundary conditions. We investigated convergence of the solution with the number of modes retained in the Galerkin approximation. For the plate considered in this Dissertation, a 6-mode approximation produces fairly good results up to pull-in. The nonlinear plate theory predicts that the electric load and maximum deection at pull-in are signicantly larger than the traditional stability limit obtained from linear plate theory. Theses results indicate the importance of including the deection distribution and geometric nonlinearities in the analysis to avoid underestimation of the stability limit. We found that both the electrostatic load and the maximum deection at pullin increase with 1 (the gap to plate thickness ratio). Increasing the aspect ratio of the plate decreases its stiness. When the aspect ratio is larger than 4, the microplate may be regarded as an innitely long plate. The linear vibration problem was solved around the deected position for
Xiaopeng Zhao

1 0.8 0.6 1 0.4 0.2 0 50 100 1 =0.5 1 =1.5
53
150 200 Vp
1 =1.0
1 =2.0 250 300
Figure 3.21: Variation of the normalized fundamental natural frequency 1 with the electrostatic load 2 Vp2 for various values of 1 . various parameters. For high values of 1 , the fundamental natural frequency rst increases with the electrostatic force, then decreases, and eventually approaches zero. The mode shapes around the deected positions are slightly dierent from those around the initially at position for small values of 1 . The dierence increases as 1 increases and reaches quantitatively signicant values at 1 = 2. We then extended the reduced-order modelling to microplates with rectangular geometry and any classical boundary conditions. The linear undamped mode shapes are computed numerically using the hierarchical nite-element method and used as basis functions in the Galerkin procedure to generate the reduced-order model. Convergence of the solutions is achieved as the number of modes retained in the Galerkin approximation increases. The developed model is validated against experimental results and analytical results obtained from the previous model. The validated model is used to conduct a parametric study of the static and dynamic characteristics of a fully clamped microplate. Again, it is found that linear plate theory underestimates the plate stiness and therefore underestimates the stability range. The linear vibration problem around the deected congurations is solved for various parameters. Mode shapes of the microplate do not exhibit signicant changes as the plate deects even when it approaches pull-in. This justies the use of at-plate mode shapes as basis functions.
Xiaopeng Zhao
54
1 0.8 0.6 0.4 0.2 0 -1 -0.5 0 x 0.5 1
Figure 3.22: The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 0.5. The developed reduced-order models are robust and provide an eective and accurate design tool for microplate-based MEMS devices.
Xiaopeng Zhao
55
1 0.8 0.6 0.4 0.2 0 -1 -0.5 0 x 0.5 1
Figure 3.23: The rst mode shape for a at plate (solid) and near pull-in (dashed) for = 1 and 1 = 2.0.
Chapter 4
Modeling and Simulation of Impact Microactuators

4.1 Motivation
Precise displacement control and manipulation is required in microscopes, electromechanical interferometers, optical devices, X-Y stages, assembly of micromachines, nanoscale data storage, and micro surgery [Kim and Kim, 2002; Reiley et al., 1995; Cusin et al., 2000; Jungnickel et al., 2002; Breguet and Clavel, 1998; Bergander and Clavel, 2000; Higuchi, 1984; Higuchi. et al., 1990; Ohmichi et al., 1997; Morita et al., 1999; Liu and Higuchi, 2001; Jiang et al., 2000; Lee et al., 1993; Daneman et al., 1996; Saitou et al., 2000; Mita et al., 2003]. Surface adhesion forces, such as electrostatic, van der Waals, and surface tension become signicant when the characteristic size is less than a millimeter [Saitou et al., 2000; Fearing, 1995]. Therefore, manual assembly of micro components is dicult due to sticking, which causes inaccuracy in positioning. Microactuators and microsystems have been extensively studied for precise positioning. Based on their working principles, microactuators can be classied as inchworm motors [Cusin et al., 2000; Jungnickel et al., 2002], stick-slip actuators [Breguet and Clavel, 1998; Bergander and Clavel, 2000], and impact actuators [Higuchi, 1984; Higuchi. et al., 1990; Ohmichi et al., 1997; Morita et al., 1999; Liu and Higuchi, 2001; Jiang et al., 2000; Lee et al., 1993; Daneman et al., 1996; Saitou et al., 2000; Mita et al.,
56
Xiaopeng Zhao
Chapter 4. Impact Microactuators
57
2003]. Recently, impact microactuators have attracted a lot of attention due to ease of fabrication, capability of batch processing, robustness to environmental perturbations, high accuracy, and high-power output. Two types of impulsive forces are used for actuation: impulsive inertial forces due to the rapid expansion of actuation elements [Higuchi, 1984; Higuchi. et al., 1990; Ohmichi et al., 1997; Morita et al., 1999; Liu and Higuchi, 2001] and mechanical collisions between dierent parts [Lee et al., 1993; Daneman et al., 1996; Saitou et al., 2000; Mita et al., 2003]. The impacts and/or rapid expansions can be achieved electromagnetically [Higuchi, 1984], piezoelectrically [Higuchi. et al., 1990; Morita et al., 1999; Liu and Higuchi, 2001; Saitou et al., 2000], photothermally [Ohmichi et al., 1997], or electrostatically [Lee et al., 1993; Daneman et al., 1996; Mita et al., 2003]. Higuchi and co-workers have demonstrated a series of micromotors driven by impulsive forces. Their impact drive mechanism consists of three major components: the main body, the inertial weight, and the actuation part. The weight is connected to the main body through the actuation part. A strong inertial force is generated when the actuation part expands or contracts rapidly. The impulsive inertial force overcomes the static friction on the main body and causes the main body and the weight to move in opposite directions. When the main body comes to a stop due to dynamic frictional force, the actuator slowly restores to its original length and gets ready for the next impact. Repeating these fast and slow operations of the actuation part generates repetitive step motion of the micromotor. Higuchi et al. [1984; 1990; 1997] have successfully fabricated the micro impact mechanism with electromagnetic, piezoelectric, and thermal actuation. Pisano and co-workers are the rst to present microactuators based on mechanical impacts [Lee et al., 1993; Daneman et al., 1996]. Each of these microvibromotors consists of two major components: the converter and the motor. Converters are active resonating structures which store and transfer input energy. Rotors are passive components actuated by friction and impact. Impulsive forces are generated by oblique impacts between converters and rotors. Electrostatic comb resonators are chosen as the converters. Two types of angular microvibromotors are demonstrated by Lee et al. [1993]. The frequency range is 1020 kHz and the rotating speed is as high as 60, 000 rpm. A linear microvibromotor is demonstrated by Daneman et al. [1996]. The actuator has an average step size of 0.27 m and a speed of more than 1000 m/s. The design of Daneman
Xiaopeng Zhao
58
et al. was modied by Saitou et al. [2000]. The modied microactuator is selectively driven by shaking the substrate with a piezoelectric actuator. Mita et al. [2003] fabricated a micromachined impact actuator driven by electrostatic forces. The impulsive force is generated by collisions between a silicon micromass and a stopper. The overall size of the actuator is 3 3 0.6 mm3 . A step-wise motion of 10 20 nm at 100 V was generated. A speed of 2.7 m/s was achieved at 200 Hz. Microactuators based on impact can generate linear or rotational motion without mechanical linkages, such as ball screws and complicated clamping systems. However, impacts make the dynamical system nonsmooth, which is a source of strong nonlinearity. The study of nonsmooth dynamical systems is challenging and one cannot just use traditional dynamical system tools to understand their behaviour. An impact actuator usually consists of an oscillator, which, under the nominal working conditions, impacts a rigid wall and thereby produces large impulsive forces to cause the actuator to move. For convenience, the nominal working condition is called on in this Dissertation. When the driving force on the oscillator is not strong enough, there are no impacts between the oscillator and the rigid wall; therefore the actuator is not moving, which is called o. Increasing the driving force to a critical level, the oscillator impacts with the rigid wall with zero impacting velocity, which is called a grazing impact. Grazing therefore separates the on and o states of a microactuator and nding an orbit that achieves grazing impact enables one to predict the minimum eort needed to turn on the microactuator. Several authors have studied the modelling of impact microactuators [Jiang et al., 2000; Lee et al., 1993; Daneman et al., 1996; Saitou et al., 2000; Mita et al., 2003]. However only direct numerical simulation or very simplied analysis has been implemented. The nonlinear characteristics associated with impacting systems have not been addressed properly. Nor has the transition between o and on been studied. The coarse analysis is unable to provide a comprehensive understanding of impact microactuators, which makes the design, optimization, and control of such systems dicult and computationally expensive. So far, research on microactuators lacks comprehensive modelling and a serious eort to understand nonlinear phenomena in such devices.
Xiaopeng Zhao
59
Using the methodology presented in Chapter 2, we studied in detail the electrostatically actuated impact actuator reported by Mita and associates. We examine the inuence of system parameters, such as the excitation frequency and amplitude, on the dynamics of the actuator. The developed model is also suitable for parameter optimization and control strategy studies.
4.2
Mitas Actuator
q2
stopper
k d
electrodes
m2
stopper
c m1 q1
Figure 4.1: The schematic of the impact microactuator. A two-degree-of-freedom schematic model of the microactuator reported by Mita and associates is shown in Figure 4.1. The system consists of a movable block of mass m2 , which is connected to a frame of mass m1 by a linear spring-damper. The stoppers and the electrode are rigidly xed to the frame. The movable mass acts as one of the electrodes. The frame rests on the horizontal ground. Friction between the frame and the ground is modelled using Coulomb friction during slip and Amontons law during stick. We denote the coecient of static friction by s and that of dynamic friction by d . When a driving voltage v (t) is applied between the electrodes, the movable block is accelerated toward the stoppers until an impact occurs with the stoppers. In the analysis below, we assume
Xiaopeng Zhao
60
that the impact impulse is large enough to overcome the static friction between the frame and the ground for all impact velocities. As a result, an impact produces a small displacement of the frame. When a periodically varying voltage is applied, there are repeated impacts and the frame moves by a certain amount during each cycle, thereby producing the needed displacement over a period of time. The dynamics of the microactuator can be decomposed into distinct phases separated by the occurrence of impacts and the associated onset of slip as well as the subsequent cessation of slip through an instantaneous transition to stick. Specically, we introduce the state vector x=
T
q1 u1 q2 u2
(4.1)
where q1 and u1 are the displacement and velocity, respectively, of the frame relative to the ground; q2 and u2 are the displacement and velocity, respectively, of the movable block relative to the frame; and = t is the phase of the sinusoidally varying driving voltage V (t) = Vamp sin(t). During stick, the equations of motion can then be written as 0 0 dx = fstick (x) = u2 dt 2 1 Vamp sin2 m kq2 cu2 (dq2 )2 2 Here, =
1 2 0 A,
(4.2)
where
is the permittivity of free space, A is the overlap area, and d is the
zero-voltage gap between the electrodes. These equations of motion are valid as long as hfront (x) = q2 < 0, hback (x) = q2 + > 0, hstick+ (x) = kq2 + cu2 and hstick (x) = kq2 + cu2
2 sin2 Vamp s N < 0, (d q2 )2 2 sin2 Vamp + s N > 0, (d q2 )2
(4.3) (4.4) (4.5)
(4.6)
where N is the normal reaction from the ground. We assume that gravity is the only external force, in which case N = (m1 + m2 )g , where g 9.81 m/s2 is the acceleration of gravity.
Xiaopeng Zhao
61
During slip, the equations of motion can be written as dx = fslip (x) = dt u1

1 m1 F
, 1 m1 F
u2
1 m2
2 Vamp
(4.7)
sin2
(dq2 )2
kq2 cu2
where F = kq2 + cu2
2 Vamp sin2 (dq2 )2
d N
and we use the negative sign when u1 > 0 and the
positive sign when u1 < 0. Again, the corresponding equations of motion are valid as long as hfront (x) = q2 < 0, hback (x) = q2 + > 0, and hslip (x) = u1 = 0. (4.10) (4.8) (4.9)
At the moment that contact is established between the movable block and the stoppers, hfront (x) or hback (x) is equal to zero. Assuming an inelastic collision with a coecient of restitution e and using conservation of momentum, the function that maps the state immediately prior to impact to the state immediately after impact is given by the jump map gimpact (x) = q1 u1 +
(1+e)m2 m1 +m2 u2 T
q2 eu2
(4.11)
The transition from slip to stick occurs as the velocity of the frame relative to the ground becomes zero; that is, as hslip (x) equals zero. Although there is a discontinuous change in the vector eld as a result of this transition, there is no associated instantaneous change of state; that is, gstick (x) = x. (4.12)
The dynamical system derived here is piecewise smooth due to the presence of impacts and friction. As suggested above, the dynamics can be divided into two dierent phases. During the stick phase, only the movable block is in motion relative to the ground and the stationary frame, as described by fstick (x). In contrast, during the slip phase, the movable block is in motion relative
Xiaopeng Zhao
62
to the frame and the frame is in motion relative to the ground, as described by fslip (x). The integration procedure is shown in Figure 4.2, where the associated impact map is described by gimpact (x). Here, the events are detected by computing the values of the event functions hfront , hback , hstick+ , hstick- , and hslip .
Integrate Phase 1 until event Impact Y Impact Map Integrate Phase 2 until event N
Sliding Stopped
Figure 4.2: Flowchart of the algorithm. Direct numerical integration generates results qualitatively consistent with those reported by [Mita et al., 2003]. A sinusoidal voltage of 100 V at 1 Hz is applied in the experiments [private communication] and the average displacement is found to be 20 nm/impact. Parameter values reported by Mita et al. are used for the numerical simulation. The parameter values not reported are tted to obtain a displacement per impact of approximately 20 nm. Figure 4.3 shows the relationship between the displacement and voltage. We note that there are two impacts in one excitation cycle, again consistent with the experimental observations.
Xiaopeng Zhao
63
Applied Voltage
Displacement per impact ~20 nm
3000
4000
5000
6000
7000
8000
t (microseconds)
Figure 4.3: Step displacement of the microactuator under very low frequency inputs.
4.3
Parametric Study
In the numerical results reported below, we have normalized mass, length, time, and voltage by m2 , d, m2 /k , and V0 , respectively, where V0 is some characteristic voltage. The nondimensional
system parameter values used for numerical computations are m1 = 5, m2 = 1, k = 1, c = 0.04, d = 1, = 0.5, e = 0.8, s = 0.4, d = 0.27, and = 1. Typical steady-state time histories of the displacements q1 and q2 are shown in Figure 4.4. Here, the movable block hits the frame once every cycle corresponding to a so-called periodic one-impact solution.
4.3.1
Switch on and o mechanism
The system possesses two types of recurrent solutions: those with impacts and those without impacts. In the absence of forcing, the only recurrent motion available to the impact microactuator is the trivial equilibrium state. For suciently small, but nonzero, values of Vamp , the movable mass exhibits a periodic oscillation of the same period as that of the forcing term (note that the voltage input appears squared in the forcing term; that is, the angular frequency of the forcing term equals 2 ) but without impacting the front or back stopper. Indeed, as the forcing term is always in the
Xiaopeng Zhao
-3
64
11 10
x10
q1
9 8 7 Stopper Position 0.4
q2
-0.4 150 160 170
180
190
200
Figure 4.4: Steady-state time histories of q1 and q2 when = 0.5 and f = 0.24. positive direction, the oscillatory motion of the movable mass is shifted in the direction of positive values of q2 . It follows that, under increasing values of Vamp , impacting motions that impact with q2 = will occur before impacting motions that impact with q2 = . We therefore focus on recurrent motions that only impact with q2 = and, consequently, for which the relative velocity between the frame and the substrate is positive during sliding episodes. It is interesting to study the transitions between impacting and nonimpacting solutions that occur under changes in system parameters as this has implications on the function of the actuator. We say that the actuator is on when the exhibited asymptotic dynamics involve repeated impacts and o otherwise. The focus of this section is on the mechanism for switching the actuator on and o. Three dierent types of switch on and o mechanisms are found for the microactuator. To visualize the switch on and o scheme, a Poincar e section [Nayfeh and Balachandran, 1995] is chosen to correspond to the event function hPoincar e (x) = q2 . One example of type I switch on and o mechanism is shown in Figure 4.5 for = 0.5. We note that both of the nonimpacting and impacting branches of periodic solutions are obtained from the continuation method rather than direct numerical simulation. Figure 4.5 shows variation of the velocity u2 of mass m2 on the Poincar e section with respect to the applied voltage Vamp . When Vamp = 0.16, the voltage is not strong enough to drive the microactuator. Here, the frame is held still by static friction, and the
Xiaopeng Zhao
65
amplitude of the motion of the movable block is less than . As the voltage is slowly increased, the maximum displacement of the movable block increases, see Figure 4.6. At a critical voltage, a grazing periodic solution is established as zero-velocity contact occurs between the movable block and the stoppers. As seen in the diagram, a further increase in Vamp results in a transition of the asymptotic dynamics to an impacting solution with relatively large impact velocity. On the other hand, when Vamp = 0.24 the system exhibits an impacting periodic solution with one impact per period. Reducing the voltage decreases the impact velocity, see Figure 4.6, and therefore the step movement of the microactuator. When a critical voltage is reached, an eigenvalue of the Jacobian of the Poincar e map corresponding to the impacting solution is equal to 1. This corresponds to a cyclic-fold bifurcation and no impacting solution exists if the voltage is reduced further. Instead, a further reduction in Vamp results in a transition of the asymptotic dynamics to a nonimpacting solution with relatively small amplitude. Figures 4.5 and 4.6 show the presence of hysteresis when the system is switched on and o.
0.48 0.46 0.44 0.42 0.4 Impacting orbit CF G
u2
0.38 0.36 0.34 0.32 0.3 0.16 0.17 0.18 0.19 0.2 0.21 0.22 0.23 0.24 Nonimpacting orbit
Vamp
Figure 4.5: Type I switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: G indicates a grazing bifurcation and CF indicates a cyclicfold bifurcation. Figure 4.7 shows an example of the type II switch on and o mechanism for = 0.485. When Vamp = 0.171, the frame is held still by static friction and the movable block oscillates with an amplitude smaller than . Again a grazing periodic solution is established at a critical voltage. In
Xiaopeng Zhao

CF Impacting orbit Nonimpacting orbit G
66
maximum displacement
0.5 0.45 0.4 0.35 0.3 0.15
impact velocity
Nonimpacting orbit 0.1 0.05 0 0.16 0.17 CF Impacting orbit G 0.18 0.19 0.2 0.21 0.22 0.23 0.24
Vamp
Figure 4.6: Type I switch on and o mechanism: variations in the maximum displacement of the movable block and the impact velocity under changes in the applied voltage: G indicates a grazing bifurcation and CF a cyclic-fold bifurcation. contrast to type I, a narrow region of chaotic impacting orbits with low impact velocity is found for further increases in Vamp . On the other hand, when Vamp = 0.177, the system exhibits an impacting periodic solution with one impact per period. When the voltage is slowly decreased, the asymptotic dynamics undergo a period-doubling cascade to chaos. When the voltage is further decreased, the chaotic region shrinks toward the nonimpacting periodic solution at grazing. Type II does not show hysteresis. An example of type III switch on and o mechanism is shown in Figure 4.8 for = 0.45, where the chaotic band is obtained by integrating the system dynamics for 5000 periods and plotting data from the last 500 periods. At low voltages, the microactuator is o. As the voltage is slowly increased, the oscillation amplitude of the movable block increases. However, in contrast to types I and II, a cyclic-fold bifurcation occurs at the critical voltage Vamp 0.2356 prior to the occurrence of grazing. Further increases in Vamp beyond the critical value causes the system dynamics to jump to a chaotic impacting solution. When the voltage is decreased, the chaotic impacting solution persists until it touches an unstable nonimpacting solution at Vamp 0.2355 in state space.This sudden loss of the chaotic attractor is referred to exterior crisis [Nayfeh and Balachandran, 1995]. On the other hand, when Vamp = 0.29, the system exhibits an impacting periodic solution with
Xiaopeng Zhao
G

Impacting Chaos
67
0.46 0.458 0.456 0.454
PD
u2
0.452 0.45 0.448 0.446 0.444 0.171 Nonimpacting Orbit Period One Impact Orbit
0.172
0.173
0.174
0.175
0.176
0.177
Vamp
Figure 4.7: Type II switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: G indicates a grazing bifurcation and PD indicates a perioddoubling bifurcation. one impact per period. When the voltage is slowly decreased, the system dynamics become chaotic through a period-doubling cascade. The dierent behavior regimes are shown in the Vamp parameter space in Figure 4.9. The solid line denotes the switch on boundary and the dotted line denotes the switch o boundary. When > 0.489, the switch on and o mechanism is type I. When 0.480 < < 0.489, the system undergoes type II switch on and o mechanism. Type III mechanism is experienced when < 0.480. A chaotic band follows the switch on voltage in both of type II and type III. Both of type I and type III have hysteresis. To produce accurate and predictable displacement, it is necessary to operate the microactuator in a periodic manner. Because of the existence of chaos in type II and type III, excitation frequencies in the type I range are favorable. Moreover, type I and type II mechanisms are generic bifurcations in impact dynamical systems [Foale and Bishop, 1994]. Type III mechanism is due to the nonlinearity of the electrostatic forces. Using numerical simulation, Daneman et al. [1996] investigated the transient response of a linear microvibromotor under discrete impulses. They found that the response time can be signicantly reduced by driving the impact arms sinusoidally with an amplitude just shy of impact
Xiaopeng Zhao
68
PD
u2
CF
Nonimpacting Orbit
Vamp
Figure 4.8: Type III switch on and o mechanism: variations in u2 at the Poincar e section q2 = 0 under changes in the applied voltage: CF indicates a cyclic fold bifurcation and PD indicates a period-doubling bifurcation. and then imposing an additional dc voltage pulse to produce impact. This algorithm works best in the region I of the Vamp parameter space based on our analysis of the on-o schemes.
4.3.2
Voltage responses and frequency responses
The steady-sate response of a vibromotor is best described by its voltage and frequency responses; they were studied, respectively, by Daneman et al. [1996] and Saitou et al. [2000] using numerical simulations. However, numerical simulations are computationally expensive and easily fail when multiple solutions coexist. On the other hand, using a continuation method [Nayfeh and Balachandran, 1995] combined with the root-nding technique presented in this Dissertation, one is able to easily trace a steady periodic orbit and study its stability. Figure 4.10 shows variation of the displacement of the frame m1 with the applied voltage when the frequency is kept xed at = 0.5, where the solid line represents stable periodic one-impact orbits and the dashed line represents unstable ones. The period-one impacting orbit loses stability at Vamp 0.556 through a period-doubling bifurcation. When the voltage is further increased, the system enters chaotic motions through a route of period-doubling sequence. Chaotic grazing,
Xiaopeng Zhao
69
0.4
0.35
Vamp
0.3
III
Switch On Boundary
II
0.25 Switch Off Boundary 0.2 0.4 0.42 0.44 0.46 0.48 0.5 0.52
Figure 4.9: Variation of switch on and o voltages with respect to the excitation frequency. chatter, and sticking phenomena also exist in that area. On the other hand, reducing the voltage decreases the movement of the frame. When a critical voltage is reached, the periodic one-impact orbit encounters a cyclic-fold bifurcation and no impacting orbits exist if the voltage is further reduced. This part can be better viewed in the enlarged plot. When increasing the voltage, the nonimpacting orbit reaches grazing at Vamp 0.227. A slight disturbance can drive the system out of the nonimpacting orbit into an impacting orbit with relatively large impact velocity. This on-o scheme is actually another representation of the type I switch on scheme shown in Figures 4.5 and 4.6. Keeping the applied voltage xed at Vamp = 0.18, we study the frequency response shown in Figure 4.11. To compare the velocity of the frame as the frequency is changing, we dene the average velocity as the product of the displacement of the frame in one period and the excitation frequency. At low frequencies, the microactuator is o. As the frequency is slowly increased, a cyclic-fold bifurcation is encountered at the critical frequency 0.4783. Increasing the frequency slightly beyond the critical value causes the system to jump to a chaotic impacting orbit. When the frequency is decreased, the chaotic impacting orbit persists until it touches an unstable nonimpacting orbit at 0.47825. We note that the contact between the chaotic band and the unstable nonimpacting orbit occurs in state space, which can be best viewed on a Poincar e map, such as
Xiaopeng Zhao
-4
70
x10 8 7 6 5
CF PD
4 3 2 1 0 0.1 0.15 0.2 0.25 0.3
0.35 0.4
0.45 0.5
0.55 0.6
Vamp
Figure 4.10: Voltage response when = 0.5: PD indicates a period-doubling bifurcation, CF indicates a cyclic-fold bifurcation, and G indicates a grazing bifurcation. Figure 4.8. The right side of the chaotic band is connected to a periodic impacting orbit, which achieves maximum velocity in the frame when = 0.5037. The average velocity decreases if the frequency is further increased until a cyclic fold bifurcation is reached at = 0.5046, beyond which no impacting orbits exist. The unstable impacting orbit coincides with a nonimpacting orbit at a grazing point when = 0.4932. The jump from an impacting orbit to a nonimpacting orbit through a cyclic-fold bifurcation was observed by Saitou et al. [2000] using numerical simulations.
4.4
Summary
We have studied the electrically driven impact microactuator proposed by Mita et al. The system is modelled as a system of two coupled rigid bodies subjected to a periodically varying excitation voltage. Transitions between impacting (on) orbits and nonimpacting (o) orbits were studied. Three dierent on and o switching mechanisms were identied, all of which are generic for impact actuators. High resolution and large travel range are both necessary for a microactuator to be used for precise positioning. The resolution of an impact microactuator is determined by the size
Xiaopeng Zhao
-5
71
x10 3.5 3 2.5 2 1.5 1 0.5

G CF
0
CF
0.47
0.475
0.48
0.485
0.49
0.495
0.5
0.505
0.51
Figure 4.11: Frequency response when f = 0.18: CF indicates a cyclic-fold bifurcation and G indicates a grazing bifurcation. of the step movement. The travel range is essentially limited by the speed of the microactuator. A large step size increases the speed but decreases the resolution. Therefore, there is a trade-o between high resolution and speed and the right combination depends upon the application at hand. The study of voltage and frequency responses reveals that irregular behavior exists for a certain parameter region. The plethora of nonlinear phenomena observed clearly brings out the need to understand the dynamics thoroughly and characterize the role of nonlinearities in the system. The present study also emphasizes the need for design optimization. The presented methodology is general and can be used for device optimization, control, and simulation of various microactuators based on impulsive forces.
Chapter 5
Local Analysis of Grazing Bifurcations

5.1 Motivation
Analysis and design of a proper functioning impact microactuator, such as that developed by Mita et al. 2003, rely on an understanding of the possibly dramatic changes in system response associated with grazing bifurcations. As shown in the previous chapter, there exists a periodic nonimpacting oscillation of the movable block with zero-relative-velocity contact with the frame for some critical parameter values. In a state-space description of the dynamics of the microactuator, such zerorelative-velocity contact corresponds to a grazing contact between a state-space trajectory and a discontinuity surface, representing the sudden changes in the velocities of the frame and the movable block that result from an impact. In contrast to periodic trajectories in smooth systems, the local description in the vicinity of a grazing trajectory is well-known to be non-dierentiable with dramatic implications to the stability of the grazing trajectory and to its persistence under further parameter variations [Brogliato, 1999; Foale and Bishop, 1994; Shaw and Holmes, 1983]. Indeed, as there is no advance warning of this instability, any local description must account for the nonsmooth character of the ow near the grazing trajectory. The local dynamics in the vicinity of a grazing trajectory can be analyzed through the introduction of a discontinuity mapping that i) captures the local dynamics in the vicinity of the grazing contact, including variations in time-of-ight to the discontinuity and the impact mapping; ii) can
72
Xiaopeng Zhao
Chapter 5. Local Analysis
73
be entirely characterized by conditions at the grazing contact; iii) is nonsmooth in the deviation from the point of grazing contact; and iv) can be studied to an arbitrary order of accuracy [Nordmark, 1997]. Properly formulated, the discontinuity mapping thus introduces the correction to the otherwise smooth dynamics that is due to the brief interaction with the discontinuity [di Bernardo et al., 2000; Fredriksson and Nordmark, 1997, 2000; Molenaar et al., 2001; Nordmark, 1997]. In this chapter, we consider grazing bifurcations in the Mita microactuator using the discontinuity mapping approach. In particular, we show how the three transition scenarios, identied in the preceding chapter, correspond to co-dimension-one bifurcations and derive local mappings that accurately describe the dynamics in the vicinity of the grazing trajectory. Moreover, we identify the boundaries of the regions corresponding to the co-dimension-one bifurcation scenarios with co-dimension-two bifurcation points and apply the discontinuity-mapping approach to formulate approximate mappings that capture the local dynamics on open neighborhoods of these boundary points. Specically, in Section 5.2 we summarize the bifurcation scenarios for the Mita microactuator. In Section 5.3, we formulate the discontinuity-mapping methodology in the context of the impact microactuator. The local mappings derived in the vicinity of the two co-dimension-two bifurcation points are given in Section 5.4 together with numerical simulations showing the agreement between the local mapping and the results of numerical simulations of the piecewise smooth system. A concluding discussion is presented in Section 5.5.
5.2
Transition Scenarios
We have numerically investigated representative transitions between impacting and nonimpacting recurrent solutions that occur under changes in system parameters. Here, we review the dierent transition scenarios found previously as well as describe additional scenarios that occur under parameter variations through isolated points in the two-dimensional parameter space given by Vamp and .
Xiaopeng Zhao
74
For convenience, we use the following notation: D= x hD (x) = hfront (x) = 0 fstick (x) hD x (x) = u2 > 0 fstick (x) hD x (x) = u2 = 0 fstick (x) hD x (x) = u2 < 0 .
def
(5.1) (5.2) (5.3) (5.4)
D+ = x D D0 = x D D = x D
As per the previous discussion, for suciently small values of Vamp , the recurrent motion is contained within the region hD (x) < 0. Since gimpact maps D+ to D , trajectories that reach D+ experience an instantaneous jump to D (as the incoming velocity u2 > 0 is changed to an outgoing velocity eu2 < 0). By a grazing periodic trajectory, we refer to a periodic trajectory on which there exists a locally unique point x D0 , because such a trajectory experiences no jump in state upon reaching D. As discussed in Chapter 2, we can use a Newton method to numerically locate parameter values for and Vamp for which grazing periodic trajectories exist. Indeed, over the region of interest (0.46 0.51), we nd a curve = (, Vamp )
Vamp = Vamp
in parameter
space, such that there exists a grazing periodic trajectory of the fundamental forcing period for every choice of parameters on . We now consider recurrent dynamics of the impact microactuator on some neighborhood in parameter space of arbitrary points on the grazing curve . To visualize the results of the numerical study, we introduce a Poincar e section P corresponding to the zero-level surface of the event function
0 hP (x) = fstick (x) hD x (x) = u2 for u2 decreasing. Therefore, P D is contained in D . In the
absence of impacts, points on P correspond to local maxima in the value of q2 along system trajectories (since q 2 = u2 ). In the presence of impacts, trajectories that reach D+ jump across P without intersecting P . We represent such a crossing by the virtual point of intersection with P of the corresponding forward trajectory segment in the absence of the jump in velocity, as suggested in Figure 5.1. Figures 5.2 5.4 show three distinct bifurcation scenarios corresponding to transversal oneparameter variations across some selected points on . Further numerical studies establish the existence of three distinct regions on , referred to here as I , II , and III , for which the bifurcation scenarios in Figures 5.2 5.4 are representative.
Xiaopeng Zhao
75
D+ P
D0
Figure 5.1: Intersections of nonimpacting and impacting trajectories with the Poincar e section P . The nonimpacting trajectory reaches its local maximum in q2 at the intersection. The impacting trajectory reaches P virtually by neglecting the existence of the discontinuity surface D. As shown in Figure 5.2, variations in Vamp across I are associated with a discontinuous transition of the asymptotic motion from a nonimpacting to an impacting periodic trajectory as Vamp is
; and from an impacting to a nonimpacting periincreased above the grazing parameter value Vamp sn1 < V odic trajectory as Vamp is decreased below Vamp amp corresponding to a cyclic-fold bifurcation. sn1 , V For Vamp Vamp amp , three distinct periodic trajectories (one nonimpacting and two impacting)
coexist. The coexistence of multiple attractors implies the possibility of parameter hysteresis in the long-time response of the impact actuator. As shown in Figure 5.3, variations in Vamp across II are associated with a continuous transition of the asymptotic dynamics from a nonimpacting periodic trajectory to an impacting chaotic
. In contrast to the transiattractor and back as Vamp is increased above and decreased below Vamp
tion across I , the nonimpacting trajectory persists beyond the grazing contact (albeit with impacts and sliding episodes), but experiences a discontinuous change in stability as one of its eigenvalues jumps to . Under further increases in Vamp , this unstable impacting periodic trajectory regains
pd stability via a period-doubling bifurcation at Vamp . Moreover, as suggested in the magnied portion
of Figure 5.3 (b), the impacting chaotic attractor undergoes an inverse period-doubling cascade to
g im
ct pa
Xiaopeng Zhao
76
penetration
SN
0
Vamp Vamp
Figure 5.2: A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across I (G=grazing contact, SN=saddle-node bifurcation). an impacting period-two trajectory that connects to the branch of period-two trajectories born at the period-doubling bifurcation through a grazing bifurcation. The continuous nature of the transitions between system attractors in the case of variations across II implies that parameter
pd hysteresis cannot occur in the interval Vamp < Vamp < Vamp .
penetration
PD
penetration
PD
0
Vamp
Vamp
Vamp
(a)
(b)
Figure 5.3: A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across II (G=grazing contact, PD=period-doubling bifurcation). Here, the black regions correspond to impacting chaotic attractors. An enlargement near the period-doubling bifurcation in (a) is shown in (b). As shown in Figure 5.4, variations in Vamp across III are associated with a discontinuous
Xiaopeng Zhao
77
transition of the asymptotic dynamics from a nonimpacting periodic trajectory to an impacting

sn2 > V ; and from an impacting chaotic attractor chaotic attractor as Vamp is increased above Vamp amp cr < V sn2 , where V cr to a nonimpacting periodic trajectory as Vamp is decreased below Vamp amp amp < Vamp ,
corresponding to the disappearance of the basin of attraction of the chaotic attractor in a global
, V sn2 , three distinct periodic trajectories (two nonimpacting crisis bifurcation. For Vamp Vamp amp
and one impacting) coexist. Of these, the unstable impacting trajectory originates in the grazing
sn2 . It eventually stabilizes via a period-doubling contact and persists as Vamp is increased beyond Vamp pd bifurcation at Vamp . Of the nonimpacting trajectories, the small-amplitude motion is stable, while the large-amplitude motion (that experiences grazing contact at Vamp and collides with the chaotic cr ) is unstable. Again, the coexistence of multiple attractors implies the possibility attractor at Vamp
of parameter hysteresis in the long-time response of the impact actuator.

penetration
penetration
G C
PD
SN
Vamp
Vamp
Vamp
Vamp
(a)
(b)
Figure 5.4: A schematic bifurcation scenario shows the switching between impacting motions (solid for stable and dotted for unstable) and nonimpacting motions (dashed for stable and dash-dotted for unstable) under variations in Vamp across III (G=grazing contact, SN=saddle-node bifurcation, PD=period-doubling bifurcation, C=global crisis). Here, the black regions correspond to impacting chaotic attractors. An enlargement near the period-doubling bifurcation in (a) is shown in (b).
From these scenarios, we nd it useful to collect in a diagram of parameter space, the grazing curve , the locus of the saddle-node bifurcations that occur when crossing I , the locus of the period-doubling bifurcations that occur when crossing II and III , the locus of the saddle-node bifurcations that occur when crossing III , and the locus of the global crisis of the chaotic attractor that occurs when crossing III . As can be seen from Figure 5.5, the saddle-node bifurcation curve associated with I and the period-doubling bifurcation curve appear to terminate at the boundary
Xiaopeng Zhao
78
between I and II and to be tangent to at this point. Similarly, the crisis and saddle-node bifurcation curves associated with III terminate at the boundary between II and III and are tangent to at that point.
(!) Vamp Vamp
10 5
0:34 0.30
(a)
cr (!) Vamp
8 4 0
(b)
pd Vamp sn 1 Vamp
II
I 0.8 0 ! 10 4
Vamp
0.26 0.22 0:18 III
-0.8
(!) Vamp Vamp
sn 2 (!) Vamp pd (!) Vamp
10 5
2 1 0
sn 2 Vamp cr Vamp
II
sn 1 (!) Vamp
(c)
0.46
0.48
!
0.50
II III -0.4 0 0.4 ! 10 3
Figure 5.5: A collection of bifurcation curves in the parameter space, where represents the grazing
sn1 represents the locus of saddle-node bifurcations when crossing ; V pd represents the curve; Vamp amp I sn2 represents the locus of locus of period-doubling bifurcations when crossing II and III ; Vamp cr represents the locus of the global crisis of the saddle-node bifurcations when crossing III ; Vamp
chaotic attractor when crossing III . Panels (b) and (c) show enlargements of the neighborhoods of the boundary point between I and II and the boundary point between II and III , respectively. Additional transition scenarios may be found by considering transversal one-parameter variations across the boundary points between I and II and between II and III , respectively, see Figure 5.6. As seen in Figure 5.6 (a) and in contrast to the transitions across I or II , crossing the boundary point between I and II is associated with a continuous transition of the asymptotic dynamics from a nonimpacting periodic trajectory to an impacting periodic trajectory and back as
. The transition in Figure 5.6 (b), on the other Vamp is increased above and decreased below Vamp
hand, shows signicant similarities with the transition across II , although here the grazing trajectory is nonhyperbolic. As the additional transition scenarios occur only near isolated points in the Vamp parameter space, we refer to the corresponding grazing bifurcations as co-dimension-two, thereby distinguishing them from the co-dimension-one bifurcations associated with passage across
Xiaopeng Zhao I , II , or III .
79
penetration
penetration
PD
Vamp
Vamp
Vamp
Vamp
(a) = 0.4878
(b) = 0.4803
Figure 5.6: Schematic bifurcation scenarios associated with the switching between impacting motions and nonimpacting motions at the boundary between I and II in Panel (a) and those at the boundary between II and III in Panel (b) (G=grazing contact, PD=period-doubling bifurcation). Here, solid curves correspond to stable periodic motions and dashed curves to unstable periodic motions. The black regions correspond to impacting chaotic attractors.
We note that bifurcation scenarios similar to those observed here in the case of I and II have been previously documented in piecewise smooth dynamical systems, for example, in linear impact oscillators [Foale and Bishop, 1994]) and in friction oscillators [Dankowicz and Nordmark, 1999]. The transition scenario associated with III is due to the nonlinearity of the electrostatic forces and is a consequence of the dynamic pull-in phenomenon [Seeger and Boser, 2003; M. Varghese and Senturia, 1997] in electrostatically attracting microelectrodes. The goal of the following sections is to develop closed-form maps that approximate the dynamics near grazing trajectories in state and parameter space, which allow accurate predictions of the long-time responses without the need to simulate the original piecewise smooth system of dierential equations. Indeed, as simulations of recurrent motions that involve many near-zero-velocity impacts require very high numerical accuracy and long integration times, such closed-form maps can be enormously benecial.
Xiaopeng Zhao
80
5.3
Discontinuity mappings
We wish to associate a Poincar e mapping P with the Poincar e section P introduced above. Ignore, for a moment, the jump map associated with the discontinuity D and assume that the dynamics are governed entirely by the vector eld fstick and constrained to a constant-q1 slice of the submanifold S corresponding to the zero-level surface of the event function hS (x) = u1 . Suppose that for some
ref , the forward trajectory based at a point xref P intersects P values = ref and Vamp = Vamp
transversally after some time tref ; that is,

ref hP stick xref , tref ; ref , Vamp
=0
(5.5)
and
ref ref ref ref hP ,i stick x , t ; , Vamp i ref fstick stick xref , tref ; ref , Vamp
< 0,
(5.6)
where stick is the smooth ow corresponding to the vector eld fstick . Here, the latter condition corresponds to the requirement that u2 be decreasing; that is, the acceleration of the movable mass relative to the frame is negative. Now introduce the function F (x, t, , Vamp ) = hP (stick (x, t; , Vamp )) . It follows that
ref F xref , tref , ref , Vamp =0
(5.7)
(5.8)
and
ref Ft xref , tref , ref , Vamp < 0.
(5.9)
According to the implicit function theorem, there exists a unique smooth function (x, , Vamp )
ref , such that dened on a neighborhood of xref , ref , Vamp ref xref , ref , Vamp = tref
(5.10)
and F (x, (x, , Vamp ) , , Vamp ) 0. (5.11)
Xiaopeng Zhao
81
In other words, (x, , Vamp ) is the time of ight from x back to P . A smooth Poincar e mapping
ref Psmooth can now be dened on a neighborhood of xref , ref , Vamp by the expression
Psmooth (x, , Vamp ) = stick (x, (x, , Vamp ) ; , Vamp ) .
(5.12)
If we reintroduce the nontrivial jump map gimpact associated with D, expression (5.12) is still valid as long as hD (x) 0. If, instead, hD (x) > 0, we recognize that the point x corresponds to a virtual point of intersection; as per the denition in the previous section, it cannot actually be reached by the piecewise smooth dynamical system. We again dene the Poincar e mapping P by the above formula, but include an initial correction to account for the virtual nature of the initial point x for the impacting ow. To this end, we consider the Poincar e mapping P dened by P (x, , Vamp ) = Psmooth (D (x, , Vamp ) , , Vamp ) , (5.13)
where the discontinuity mapping D maps x to some point on P in such a way that the subsequent dynamics respect those of the corresponding actual trajectory. For more discussion of the concept of discontinuity mappings and their derivation, we refer the reader to Dankowicz and Nordmark [1999], di Bernardo et al. [2000], Fredriksson and Nordmark [1997], Molenaar et al. [2001], and Nordmark [1997]. We restrict our attention to points x P near x P D, where x corresponds to the point of tangential contact of a grazing periodic orbit with the discontinuity surface D. To arrive at an expression for D, we consider the trajectory segments shown in Figure 5.7. Here, an incoming trajectory (solid) in S governed by the vector eld fstick reaches the discontinuity surface D at a point xin D+ , experiences a jump to a point gimpact (xin ) D , ows under the vector eld fslip+ until reaching the stick manifold S at a point xout , and then continues to ow in S under the vector eld fstick . The dashed trajectory segments correspond to a ow in S governed by the vector eld fstick from xin forward in time until reaching P at a point x0 ; and from xout backward in time until reaching P at a point x1 . The sought correction to the smooth ow given by stick is then obtained by mapping x0 to x1 , as this correctly accounts for the eects of the jump map and the subsequent sliding episode. Thus, given an initial point x on P , such that hD (x) > 0, we dene D as the composition of the following steps:
Xiaopeng Zhao
82
D
gimpact
xb
xin
P
S
xout x1 x0
Figure 5.7: Trajectories associated with the discontinuity mapping D. Here S stands for the stick manifold, D is the discontinuity surface, and P is the Poincar e section. We note that S , D, and P are all 4dimensional hyper-surfaces in a 5dimensional space. 1. Flow for a time t1 < 0 with the vector eld fstick until reaching D: ow from x0 to xin ; 2. Apply the jump map gimpact : map from xin to xb ; 3. Flow for a time t2 > 0 with the vector eld fslip+ until reaching S : ow from xb to xout ; 4. Flow for a time t3 < 0 with the vector eld fstick until reaching P : ow from xout to x1 . To arrive at a functional expression for D, we seek to express the ow times in terms of the corresponding initial conditions in state space and the corresponding parameter values.
, there exists a point x , such that Suppose, in particular, that for = and Vamp = Vamp
hD (x ) = 0,
i hP (x ) = hD ,i (x ) fstick x , , Vamp = 0,
(5.14) (5.15) (5.16) (5.17)
hS (x ) = 0,
i hS ,i (x ) fslip+ x , , Vamp < 0,
Xiaopeng Zhao and
83
j a x , , Vamp = hP ,j (x ) fstick x , , Vamp , , V ) f i hD ( x x amp ,ij stick fj = stick x , , Vamp < 0. D i +h,i (x ) fstick,j x , , Vamp
def
(5.18)
Clearly, x P S is a point of simple grazing contact with D corresponding to a local maximum in hD along a trajectory segment of the vector eld fstick based at x . Step 1. Suppose that hD (x) > 0 and consider the function E (1) (x, y, t, , Vamp ) = t hD (x) hD (stick (x, t; , Vamp )) t hP (x) y. t2 (5.19)
Since hP (x) = 0 for x P , E (1) (x, y, t, , Vamp ) = 0 implies that the time of ight from the hD (x) level surface to the hD (x) y 2 level surface is t. Expanding hD (stick (x, t; , Vamp )) in a Taylor series for t 0 yields hD (stick (x, t; , Vamp )) 1 D 2 3 = hD (x) t hD ,t (x, , Vamp ) + h,tt (x, , Vamp ) t + O t , 2 where
D i P hD ,t (x, , Vamp ) = h,i (x) fstick (x, , Vamp ) = h (x) , D (x) f i h ( x , , V ) amp ,ij stick f j (x, , Vamp ) . hD ,tt (x, , Vamp ) = stick D i +h,i (x) fstick,j (x, , Vamp )
(5.20)
(5.21) (5.22)
Specically,
hD ,t x , , Vamp = 0, hD ,tt x , , Vamp = a x , , Vamp .
(5.23) (5.24)
We then rewrite Equation (5.19) as 1 (x, , Vamp ) + O (t) y. E (1) (x, y, t, , Vamp ) = t hD 2 ,tt (5.25)
Xiaopeng Zhao It follows that
84
E (1) x , 0, 0, , Vamp =0
(5.26)
and Et
(1) x , 0, 0, , Vamp =
1 . a x , , Vamp 2
(5.27)
The implicit function theorem now implies the existence of a unique smooth function (1) (x, y, , Vamp )
on an open neighborhood of x , 0, , Vamp , such that (1) x , 0, , Vamp =0
(5.28)
and E (1) x, y, (1) (x, y, , Vamp ) , , Vamp 0. It follows from the denition of E (1) that t1 = (1) x, hD (x), , Vamp . (5.29)
As the smooth function (1) (x, y, , Vamp ) is implicitly dened by Eq. (5.29), we can com pute arbitrary partial derivatives of (1) at x , 0, , Vamp using implicit dierentiation and demanding that all of the partial derivatives of E (1) must vanish at x , 0, , Vamp . For
example, it is found that (1) x , 0, , Vamp = 0, x (1) 2 x , 0, , Vamp = , y a x , , Vamp (1) x , 0, , Vamp = 0, (1) x , 0, , Vamp = 0. Vamp
It follows that (1) (x, y, , Vamp ) can be approximated near x , 0, , Vamp as
(5.30) (5.31) (5.32) (5.33)
(1) (x, y, , Vamp ) = y Now let
2 + h.o.t. a x , , Vamp
(5.34)
D1 (x, y, , Vamp ) = stick x, (1) (x, y, , Vamp ) ; , Vamp .
(5.35)
Xiaopeng Zhao
85
Since stick is smooth, we can compute arbitrary partial derivatives of D1 at x , 0, , Vamp .
We write x as x =
q1
(5.36)
and assume that the deviation of x0 , shown in Figure 5.7, from x is x = where q2 > 0. It follows that
fstick x , , Vamp = T
q2
(5.37)
0 0 0 acc2
(5.38)
and
a x , , Vamp = acc2 ,
(5.39)
where acc2 =
sin2 ( ) Vamp
m2 (d )
k . m2
(5.40) to the desired order in the
Step 1 is completed by expanding D1 x0 ,

deviation from x , 0, , Vamp . For example,
hD (x0 ), , Vamp
xin = D1 x0 ,
hD (x0 ), , Vamp hD (x0 ), , Vamp ; , Vamp hD (x0 ), , Vamp

q2 2acc 2
= stick x0 , (1) x0 ,
(1) x0 , = x0 fstick x , , Vamp
+ O (2)
T
0 q1
2 acc2 q2 +
+ O (2) ,
(5.41)
where O (n) refers to terms of the form (q2 )n1 /2 ()n2 ( )n3 (Vamp )n4 , for which n1 + n2 + n3 + n4 = n. Step 2. Since gimpact it can be expanded to the desired order in the deviation from x , thus
Xiaopeng Zhao
86
completing Step 2. Specically, we have xb = gimpact (xin ) = gimpact (x ) + gimpact,x (x ) (xin x ) q 1 (1+e)m2 2 acc q 2 2 m + m 1 2 = + O (2) , e 2 acc2 q2 + 2 q2 acc2 where 1 0 0 1 gimpact,x (x ) = 0 0 0 0 0 0 0 0 1 0 0 0
(1+e)m2 m1 +m2
(5.42)
0 e 0
0 0 , 0 0 1
(5.43)
as can be obtained from Equation (4.11). Step 3. Now consider the function E (2) (x, t, , Vamp ) = hS (slip+ (x, t; , Vamp )) , where slip+ is the ow corresponding to the vector eld fslip+ . Then,
E (2) x , 0, , Vamp = hS (x ) = 0
(5.44)
(5.45)
and E,t
(2) i x , 0, , Vamp = hS ,i (x ) fslip+ x , , Vamp < 0.
(5.46)
The implicit function theorem again implies the existence of a unique smooth function
(2) (x, , Vamp ) on an open neighborhood of x , , Vamp , such that (2) x , , Vamp =0
(5.47)
and E (2) x, (2) (x, , Vamp ) , , Vamp 0. (5.48)
Xiaopeng Zhao
87
It follows from the denition of E (2) that t2 = (2) (x, , Vamp ).
As the smooth function (2) (x, , Vamp ) is implicitly dened by Eq. (5.48), we can com pute arbitrary partial derivatives of (2) at x , , Vamp using implicit dierentiation and demanding that all of the partial derivatives of E (2) must vanish at x , , Vamp . For ex-
ample,
hS (2) ,x (x ) , x , , Vamp = S x h,x (x ) fslip+ x , , Vamp
(5.49) (5.50) (5.51)
(2) x , , Vamp = 0, (2) x , , Vamp = 0. Vamp

It follows that (2) (x, , Vamp ) can be approximated near x , , Vamp as
(2) (x, , Vamp ) = Now let
hS ,x (x ) (x x ) hS ,x (x ) fslip+ x , , Vamp
+ h.o.t.
(5.52)
D2 (x, , Vamp ) = slip+ x, (2) (x, , Vamp ) ; , Vamp .
(5.53)
Since slip+ is smooth, we can compute arbitrary partial derivatives of D2 at x , , Vamp .
Step 3 is completed by expanding D2 (xb , , Vamp ) to the desired order in the deviation from
x , , Vamp . For example, fslip+ x , , Vamp =
0 acc1 0 acc2 acc1
(5.54)
where acc1 =
sin2 Vamp d N k . 2 m1 m1 m1 (d )
(5.55)
Xiaopeng Zhao It follows that xout = D2 (xb , , Vamp )

= xb fslip+ x , , Vamp
88
= =
+ O (2) hS ,x (x ) fslip+ x , , Vamp q1 0 (1+e)m2 2 acc2 q2 acc1 m1 +m2 (1+e)m2 2 acc2 q2 m1 +m2 + O (2) 0 acc1 acc acc e 2 acc2 q2 2 1 2 q2 + acc2 q1 0 + O (2) (5.56) 1 m2)+acc2 m2 (1+e) acc1 (e macc 2 acc2 q2 1 (m1 +m2 ) m2 (1+e) 2 q2 + 2 acc q 2 2 acc2 acc1 (m1 +m2 )
hS ,x (x ) (xb x )
Step 4. Finally, consider the function E (3) (x, t, , Vamp ) = hP (stick (x, t; , Vamp )) , Then,
E (3) x , 0, , Vamp = hP (x ) = 0
(5.57)
(5.58)
and E,t
(3) x , 0, , Vamp = a x , , Vamp > 0.
(5.59)
The implicit function theorem again implies the existence of a unique smooth function
(3) (x, , Vamp ) on an open neighborhood of x , , Vamp , such that (3) x , , Vamp =0
(5.60)
and E (3) x, (3) (x, , Vamp ) , , Vamp 0. It follows from the denition of E (3) that t3 = (3) (x, , Vamp ). (5.61)
Xiaopeng Zhao
89
As the smooth function (3) (x, , Vamp ) is implicitly dened by Eq. (5.61), we can com pute arbitrary partial derivatives of (3) at x , , Vamp using implicit dierentiation and demanding that all of the partial derivatives of E (3) must vanish at x , , Vamp . For ex-
ample,
hP (3) ,x (x ) , x , , Vamp = P x h,x (x ) fstick x , , Vamp
(5.62) (5.63) (5.64)
(3) x , , Vamp = 0, (3) x , , Vamp = 0. Vamp

It follows that (3) (x, , Vamp ) can be approximated near x , , Vamp as
(3) (x, , Vamp ) = Now let
hP ,x (x ) (x x ) hP ,x (x ) fstick x , , Vamp
+ h.o.t.
(5.65)
D3 (x, , Vamp ) = stick x, (3) (x, , Vamp ) ; , Vamp .
(5.66)
Since stick is smooth, we can compute arbitrary partial derivatives of D3 at x , , Vamp .
Step 4 is completed by expanding D3 (xout , , Vamp ) to the desired order in the deviation from
x , , Vamp . It follows that
x1 = D3 (xout , , Vamp )
= xout fstick x , , Vamp 4 x4 out (x ) acc2 hP ,x (x ) (xout x ) hP ,x (x ) fstick x , , Vamp T
+ O (2)
= xout =
0 0 0 acc2
m1 (1+e) (m1 +m2 )
+ O (2)
T
0 0 + q1
q2 2 acc2
+ O (2)
(5.67)
Now combining the above steps yields
Xiaopeng Zhao
90
x1 = D (x0 , , Vamp ) Id = D D g hD (x0 ), , Vamp 3 2 impact D1 x0 , Id = x + 0 0 0 0 m1 (1+e) 2q2 acc2 (m1 +m2 ) The composite map for q2 > 0 can be written as
when q2 = hD (x0 ) 0 , , Vamp

T
when q2 = hD (x0 ) > 0 when q2 0 . (5.68)
+ O (2) when q2 > 0
P (x0 , , Vamp ) = Psmooth (D (x0 , , Vamp )) 0 0 m1 (1+e) q2 q2 q2 q2 q2 = x + Psmooth 2 , (m1 +m2 ) acc2 + Psmooth, + Psmooth, + Psmooth,Vamp Vamp 0 P m1 (1+e) q2 2 smooth, (m1 +m2 ) acc2 + Psmooth, + Psmooth, + Psmooth,Vamp Vamp
+O (2)
We note that although Psmooth are smooth on an open neighborhood of x , , Vamp , D and P are not smooth on an open neighborhood of x , , Vamp due to the substitution of
hD (x).
While Psmooth results in no change in q1 , the inclusion of the jump mapping gimpact and the associated sliding episode yields a discrete change in q1 described by the discontinuity mapping D and, consequently, by the composite Poincar e map P. As the system is invariant under variations in q1 and as, for low-velocity impacts P (x) P S , below we consider the reduced system obtained by restricting attention to the third and fth components of P.
5.3.1
Near-grazing Dynamics
A periodic impacting orbit near the grazing orbit with one impact per period corresponds to a xed point of the composite Poincar e map P. To obtain an analytical expression for the location of such a xed point in terms of and Vamp , consider the introduction of a small parameter such that and Vamp are both O () . 1,
Xiaopeng Zhao
91
q2 Now suppose that Psmooth , (x ) = O (1). Then, a consistent expansion for the position of a
xed point of P is given by the ansatz q2 = 2 q2

(2)
+ O(3 )
(5.69) (5.70)
= (1) + O(2 )
A xed point of P corresponding to a periodic impacting orbit with one impact per period must then satisfy the equations
q2 0 = Psmooth , (x ) (1) = Psmooth , (x ) q2 (1) + c1 q2 + Psmooth , (x ) (1) q2 + Psmooth + c2 , (x ) (2) (2)
+ O 2 , + O 2 ,
(5.71) (5.72)
where = m1 (1 + e) m1 + m2 2 < 0, a0 (5.73) (5.74) (5.75)
q2 q2 c1 = Psmooth , (x ) + Psmooth,Vamp (x ) Vamp , c2 = Psmooth , (x ) + Psmooth,Vamp (x ) Vamp ,
that is,
(2) q2 (1)
= =
q2 1 Psmooth , (x ) c1 + Psmooth, (x ) c2 q2 Psmooth , (x ) q2 Psmooth , (x )
(5.76) (5.77)
q2 Psmooth , (x ) c1 + Psmooth, (x ) c2
q2 Since < 0, it follows that a xed point exists for 1 Psmooth , (x ) c1 + Psmooth, (x ) c2 > 0 q2 q2 q2 when Psmooth , (x ) > 0 and for 1 Psmooth, (x ) c1 + Psmooth, (x ) c2 < 0 when Psmooth, (x ) <
0. This behavior is illustrated in Figure 5.8. q2 If, instead Psmooth , (x ) = O ( ), a consistent expansion for the position of a xed point of P is given by the ansatz q2 = q2 + O 2 1 = ( 2 ) + (1) + O 3/2
(1)
(5.78) (5.79)
Xiaopeng Zhao
92
b>0
penetration penetration
0
b<0
G
G
0
(a)
(b)
Figure 5.8: Fixed points of the composite Poincar e map P corresponding to period-1 nonimpacting orbits (solid), period-1 impacting orbits with a single impact per period (dashed). Here, penetration is dened as sign (q2 )
q2 Psmooth , (x ) c2 . q2 |q2 |, b = Psmooth , (x ), and c = 1 Psmooth, (x ) c1 +
A xed point of P corresponding to a periodic impacting orbit with one impact per period must then satisfy the equations q2 =
(1)
2 2 1 Psmooth (x ) Psmooth (x ) (1) , , q + ( 2 ) 2 2 1 1 (1) (1) +c1 + a11 q2 + a12 q2 ( 2 ) + a13 ( 2 ) q q
+ O(3/2 ) (5.80)
1 ( 2 ) + (1) = Psmooth , (x ) +
(2) q2 + Psmooth , (x ) (1)

1
(5.81)
1
(1) + c + a q Psmooth 2 21 2 + a22 , (x )
(1) 2
q2 ( 2 )
(1)
+ O(3/2 ) (5.82)
+a23
1 ( 2 )
where a11 , a12 , a13 , a21 , a22 ,and a23 are constants. It follows that 0 = a q2 + b q2 c, ( 2 ) =
1
(1)
(1)
(5.83) (5.84)
(1) 1 Psmooth , (x ) 1
q2 ,
2 2 a21 + a22 + a2 q2 + c2 23
(1)
(1) = where
(5.85)
Xiaopeng Zhao
93
a = 1 a11 a12 a13 2 2 , b=

q2 Psmooth ,
(5.86)
1
(x )
1 Psmooth , (x )
(5.87) (5.88)
c = c1 ,
= Psmooth , (x ) 1 Psmooth, (x ) 1
.
c a
(5.89) 0 and
Since
q2 0, it follows that i) if
c a
(1)
b a
> 0, there exists one xed point near x when

b a
2
no xed point near x when

c a
< 0, whereas ii) if
< 0, there exists one xed point near x when

c a
) > 0 and two xed points near x when (b/a 4
0 (cf. Figure 5.9).

b a
penetration
penetration
b a
>0
<0
G
2
0
(a)
c a
) (b=a 4
0
(b)
c a
Figure 5.9: Fixed points of the composite Poincar e map P corresponding to period-1 nonimpacting orbits (solid), period-1 impacting orbits with a single impact per period (dashed). Here, penetration is dened as sign (q2 ) |q2 |.
q2 From the analysis in the case when Psmooth , (x ) = O (1) it is well established that a coq2 q2 dimension-two bifurcation occurs when Psmooth , (x ) switches sign. Since Psmooth, (x ) may be
computed from the nonimpacting dynamics, it is possible to identify various co-dimension-two grazing bifurcation points on the grazing curve in parameter space.
5.4
Numerical results
We now apply the discontinuity-mapping approach to yield explicit numerical expressions for the composite Poincar e mapping P (x, , Vamp ) on neighborhoods of selected grazing periodic trajectories as found numerically using the Newton approach described previously. For convenience, we
Xiaopeng Zhao
94
let x = x + x, = + , and Vamp = Vamp + Vamp , and use (x, , Vamp ) as local
coordinates. The grazing periodic trajectory at the co-dimension-two point at the boundary between I and II corresponds to
0.487809, Vamp 0.171067 and x T
0 0.5 0 5.496544
(5.90)
Following the methodology in Section 5.2, we nd Psmooth (x, , Vamp ) x = 0 . 936621 q + 0 . 775788 V 0 . 174538 2 amp + (0.111825 q2 0.0103004 + 0.960382 Vamp 0.627897 ) q2 (0 . 103166 + 0 . 14153 V + 0 . 669646 ) amp 2 2 +3 . 45355 V 1 . 52077 3 . 21475 V amp amp = 0 . 142633 q + 0 . 825198 + 0 . 98176 V + 5 . 82623 2 amp + (0.394304 q 0.148787 + 2.22656 V 0 . 441188 ) q 2 amp 2 + (0 . 0138705 2 . 25913 V 0 . 53266 ) amp 2 +4.23226 Vamp 1.82448 2 7.41318 Vamp and D (x, , Vamp ) x = Id when q2 0 0.237147 q2 = , (5.92) when q2 > 0 1 . 55767 q 0 . 189248 q + 2 2 + (0.206574 1.21057 V 3.1932 ) q
amp 2
(5.91)
where we have omitted terms of order three or higher in
q2 , , , and Vamp .
Figures 5.10, 5.11, and 5.12 show the predictions of the local map P = Psmooth D on the bifurcations associated with variations in Vamp for select values of corresponding to the schematic transition scenarios across I , II , and the boundary point between I and II as described in Section 5.2. While the predicted results are in excellent agreement with the simulated
Xiaopeng Zhao
103
95
0 q2
0.4 eigenvalues
imaginary part
-0.4
0 real part 0 0.5 Vamp

103
0.5 Vamp
103
(a)
(b)
Figure 5.10: Comparison of the bifurcation results predicted from the local map with those obtained through numerical integration when is at the boundary between I and II . Panel (a) shows variation of q2 with the applied voltage, where the impacting motions are in black and the nonimpacting motions are in gray. We note that the predicted results are indistinguishable from the numerical results. Panel (b) shows the dependence of the real and imaginary parts of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (solid) and as predicted by the local mapping (dashed). data for the original dynamical system, we recognize that the former were obtained at a fraction of the computing time required to generate the latter. The local map shown above is valid on an open neighborhood of the co-dimension-two point at the boundary between I and II . It follows that it should be able to capture the saddle-node bifurcation curve associated with I and the period-doubling bifurcation curve associated with II . In fact, the explicit nature of the local map allows the determination of the bifurcation curves as Taylor expansions in the perturbation . Specically, we nd that the saddle-node bifurcation curve is given by Vamp = 0.224982 + 17.7774 2 + O 3 , > 0 and the period-doubling curve is given by Vamp = 0.224982 + 1557.35 2 + O 3 , < 0. (5.94) (5.93)
Figures 5.13 (a) and (b) compare the predicted bifurcation points using the local map and the predicted bifurcation curves using the truncated Taylor expansions, respectively, with the actual
Xiaopeng Zhao
104
96
eigenvalues
0.8
50 20 0
1.6 1.0 0.4
(b)
0.4
q2
(a)
eigenvalues
0.4 0.2 0 -2
(c)
0 -2 -1 Vamp 0
104 1
-1 104 0 Vamp
Figure 5.11: Comparison of the bifurcation results predicted from the local map with those obtained through numerical simulation when = 0.4888. Panel a) shows variation of q2 with the applied voltage, where the impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, and dashed curves correspond to unstable periodic motions. Panels (b) and (c) show the dependence of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (dots for stable and circles for unstable) and as predicted by the local mapping (solid for stable and dashed for unstable). bifurcation points obtained from direct numerical simulation of the original dynamical system. Again, it is clear that the local map provides a very accurate description of the dynamics in the vicinity of the grazing periodic trajectory in state and parameter space. The grazing periodic trajectory at the co-dimension-two point at the boundary between II and III corresponds to
0.480271, Vamp 0.176796 and x T
0 0.5 0 5.286519
(5.95)
Xiaopeng Zhao
97
Again, following the methodology in Section 5.22, we nd Psmooth (x, , Vamp ) x = 0 . 94297 q + 0 . 0512257 + 0 . 809064 V + 0 . 00715561 2 amp + (0.138282 q2 + 0.0544217 + 1.24226 V amp 0.431478 ) q2 (0 . 105538 0 . 426753 V + 0 . 888627 ) amp 2 2 1.16325 V +4 . 24501 V 2 . 28863 amp amp = 0.192815 q2 + 0.826809 + 1.5879 Vamp + 6.00327 + (0.724366 q + 3.29973 V 2 amp 0.342789 0.136173 ) q2 (0.0565448 + 2.32779 V amp + 1.15147 ) 2 2 +7.2304971806356110 Vamp 3.61189 6.14775 Vamp and D (x, , Vamp ) x = Id when q2 0 0.238028 q2 = , when q > 0 1 . 58753 q 0 . 204934 q + 2 2 2 + (0.219773 1.92188 V 3.30549 ) q
amp 2
(5.96)
(5.97) where we have again omitted terms of order three or higher in q2 , , , and Vamp .
Figures 5.14 and 5.15 show the excellent agreement between the predictions of the local map P = Psmooth D and the result of direct numerical simulations of the original dynamical system. Again, we may choose to use the local map to predict the locus of the saddle-node and global crisis bifurcations. Figure 5.16 shows predicted bifurcation points and corresponding bifurcation points obtained using the simulated data. The observation made previously regarding computation time applies again. Indeed, while the predicted bifurcation values were obtained in a matter of minutes, the simulated data required several days of computations. Using the predicted data as initial conditions for the numerical simulation, the computation time was reduced to a couple of hours.
Xiaopeng Zhao
98
Figure 5.17 shows the largest Lyapunov exponent for the chaotic attractors in Figure 5.12 as predicted by the local map.
5.5
Summary
In the previous sections, we formulated discrete maps that capture the dynamics of a model microelectromechanical impact actuator in the vicinity in state and parameter space of the transition between nonimpacting and impacting asymptotic dynamics. Specically, using a discontinuitymapping-based approach, we were able to achieve excellent agreement between the results of computationally-expensive direct numerical simulations of the impacting dynamics and the computationally cheap predictions from the associated discrete maps. Indeed, the discrete maps faithfully describe the local dynamics while reducing the computation time by several orders of magnitude. As such the methodology provides an ecient tool in the analysis and design of proper function in impact microactuators. While the analysis was applied to the Mita actuator, only nominal modications would be required to derive similar local maps to other impact microactuators. Although we have only shown explicit formulae for the local map on neighborhoods of the two co-dimension-two points, similar maps can easily be derived on neighborhoods of the co-dimensionone points. It is interesting to observe that the predictions of such local maps when truncated at rst order in q2 , , , and Vamp agree qualitatively with the numerical simulations for small perturbations near the co-dimension-one points, but that the second-order terms are necessary to capture the bifurcation behavior near the co-dimension-two points.
Xiaopeng Zhao
99
103
1 0.5 q2 0
103
1 0.5 q2 0
-0.5 0 0.02 0.04 p V amp 0.06
-0.5 0 0.02 0.04 p V amp 0.06
(a)
0 eigenvalues eigenvalues 0
(b)
-20
-0.2 0 0.02 p 0.04 0.06 V amp
-40 0 0.02 p0.04 0.06 V amp
(c)
(d)
Figure 5.12: Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.485. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors. Panels (c) and (d) show the dependence of the eigenvalues of the periodic impacting orbit on the applied voltage as obtained from numerical simulations (dots for stable and circles for unstable) and as predicted by the local mapping (solid for stable and dashed for unstable).
Xiaopeng Zhao
100
103
(! ) V amp Vamp
103
(! ) V amp Vamp
0.8 0.4 0 -1 0 ! 1
0.8 0.4 0 -1 0 ! 1
103
103
(a)
(b)
Figure 5.13: Bifurcation curves obtained numerically (Panels (a) and (b)), compared to those predicted from the local map (a)), and those predicted from the Taylor expansions (b)). In both panels, triangles, circles, and + correspond to the numerically obtained grazing, period-doubling, and saddle-node bifurcation points, respectively; the solid lines, dashed lines, and dotted lines correspond to the predicted grazing, period-doubling, and saddle-node bifurcation points, respectively.
103
4 2 q2 0
103
4 2 q2 0
-2 0 0.04 0.08 p V amp 0.12
-2 0 0.04 0.08 p V amp 0.12
(a)
(b)
Figure 5.14: Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.4803. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors.
Xiaopeng Zhao
101
103
0.4 q2 0
103
0.4 q2 0
-0.4 0 0.04 0.08 p V amp 0.12
-0.4 0 0.04 0.08 p V amp 0.12
(a)
(b)
Figure 5.15: Comparison of the bifurcation results obtained through numerical simulation (Panel (a)) with those predicted from the local map (Panel (b)) when = 0.4793. The impacting motions are in black, the nonimpacting motions are in gray, solid curves correspond to stable periodic motions, dashed curves correspond to unstable periodic motions, and black regions correspond to chaotic attractors.
-4
x10
Vamp Vamp (!)
2 1.5 1 0.5 0
-5
x10
-4
Figure 5.16: Comparison of predicted bifurcation points with those obtained using the simulated data (solid line - numerically obtained grazing bifurcation points, dots - predicted grazing bifurcation points, + signs - numerically obtained crisis points, circles - predicted crisis points, * signs numerically obtained saddle-node bifurcation points, triangles - predicted saddle-node bifurcation points.
Xiaopeng Zhao
102
0.5 0.4 0.3 0.2 0.1 0 0.02 0.04 p V 0.06

Figure 5.17: Dependence of the largest Lyapunov exponent of the chaotic attractors on the applied voltage as obtained from the local mapping.
Chapter 6
Bifurcations and Chaotic Motions of Post-Grazing Dynamics

6.1 Motivation
As shown in previous chapters, the system possesses two types of recurrent solutions: those with impacts and those without impacts. Transitions between nonimpacting and impacting asymptotic dynamics occur in three dierent mechanisms. In each case, the post-grazing dynamics settle down to a stable impacting periodic motion, which persists for certain parameter values near those corresponding to the grazing periodic motion. It is also shown that, near grazing bifurcations in parameter space, a larger excitation voltage corresponds to a larger impact velocity of the impacting motion and thus a larger step displacement of the microactuator. It is therefore the interest of this chapter to study how this impacting periodic motion evolves under further changes of system parameters. In the following, we rst review the type I switching on -o mechanism when = 0.5 and then discuss the subsequent plethora of bifurcations. It has been found that nonimpacting and impacting motions coexist for certain voltages, see Figure 6.1. To visualize the transition between nonimpacting and impacting motions, we show in Figure 6.2 the dierence between the maximum and minimum displacements sampled per forcing period. When V = 0.18, the system has a nonimpacting periodic motion, whose maximum 103
Xiaopeng Zhao
0.4
Chapter 6. Post-Grazing Dynamics

0.4
104
0.4 0 -0.4
-0.4 0 0.5
-0.4 -0.3 0 0.5
-0.4
-0.4
0.5
(a)
0.5
(b)
0.4 0 -0.4 0.4
(c)
-0.5 -0.4 0 0.5
-0.4
0.5
-0.4 -0.4
0.5
(d)
(e)
(f)
Figure 6.1: Phase portraits in the x2 , x2 space: nonimpacting periodic orbit at (a) V = 0.18, (b) V = 0.2; periodic grazing orbit at (c) V = 0.2272; impacting periodic orbit at (d) V = 0.18, (e) V = 0.21, (f) V = 0.24. We note that the rigid stopper is placed at x2 = 0.5. displacement is less than = 0.5, see Figure 6.1 (a). When the voltage is slowly increased, this periodic motion persists with larger maximum displacement, see Figure 6.1 (b), until a critical voltage is reached, where a grazing periodic solution is established as zero-velocity contact occurs between the movable block and the stopper, see Figure 6.1 (c). A further increase in V results in a transition of the asymptotic dynamics to an impacting solution with relatively large impact velocity. For example, when V = 0.24 the system exhibits an impacting periodic solution with one impact per period, as shown in Figure 6.1 (f). When the voltage is reduced slowly, the impacting periodic motion persists, see Figures 6.1 (d) and 6.1 (e), until a critical voltage is reached, where an eigenvalue of the Jacobian of its associated Poincar e map (see Chapter 2) equals 1. This corresponds to a saddle-node bifurcation and no impacting solution exists if the voltage is reduced further. Instead, a further reduction in V results in a transition of the asymptotic dynamics to a nonimpacting solution with relatively small amplitude. Figure 6.2 shows that impacting and
nonimpacting periodic motions coexist between the voltages corresponding to SN and G. It is followed from Figures 6.1 and 6.2 that increasing the driving voltage increases the impact velocity and therefore the step movement of the microactuator. Indeed, as the voltage is further increased above V = 0.24, the impacting periodic motion persists with an increasing impact velocity. It is thus desirable to operate the microactuator at higher voltages to achieve larger step displace-
Xiaopeng Zhao
105
0.95
Displacement (max-min)
SN
0.9 0.85 0.8 0.75 0.7 0.18 0.19 0.2
Vamp
0.21
0.22
0.23
0.24
Figure 6.2: One example of the transition between nonimpacting and impacting motions. Here, the dashed line stands for nonimpacting periodic motions, the solid line stands for stable impacting periodic motions, and the dotted line stands for the unstable impacting periodic motions: G indicates a grazing bifurcation and SN indicates a saddle-node bifurcation. ments. However, various bifurcations ruin the period-one motion and make the displacements of the system dicult to predict. In the following sections, we show a plethora of bifurcations, such as grazing, chaos, crises, chatter, and sticking for V > 0.55.
6.2
Bifurcation Scenarios
A bifurcation diagram for the range of excitation voltage 0.55 < V < 0.88 is shown in Figure 6.3. To distinguish dierent periodic impacting motions, we denote the motion with p impacts occurring in q forcing periods of / as P (p, q ). A period-q motion corresponds to q xed points in Figure 6.3 (a), which shows the dierence between the maximum and minimum displacements x2 of mass m2 sampled per period of forcing. Figure 6.3 (b) shows the velocity x2 of m2 at impact, which indicates the number of impacts p in one period. For example, a P (2, 1) motion at V = 0.75 corresponds to 1 xed point in Figure 6.3 (a) and 2 xed points in Figure 6.3 (b). As seen from Figure 6.3, a P (1, 1) motion exists for V = 0.55. When V is increased, the P (1, 1) motion undergoes a period-doubling bifurcation at V 0.555 and becomes a P (2, 2) motion. The P (2, 2) motion then loses stability through a grazing bifurcation at V 0.58 and becomes a P (3, 2) motion, which is followed by a period-doubling route to chaos. We note that discontinuous changes occur at a few places in Figure 6.3. These correspond to the discontinuous type of grazing
Xiaopeng Zhao
106
0.85 0.8 0.75 0.7 0.65 0.55 0.6 0.65 0.7 Vamp 0.75 0.8 0.85
(a)
1.2
Impact velocity
1.0 0.8 0.6 0.4 0.2 0 0.55 0.6 0.65 0.7 Vamp 0.75 0.8 0.85
(b) Figure 6.3: Bifurcation diagram for the impact microactuator when = 0.5: (a) shows the difference between the maximum and minimum displacements x2 of mass m2 sampled per period of forcing, and (b) shows the velocity x2 of mass m2 at impact. bifurcation, where nonimpacting and impacting motions coexist for certain range near grazing bifurcations. For more detailed discussions of grazing bifurcations, we refer the reader to Foale and Bishop [1994] and Nordmark [1991]. Figure 6.4 shows the corresponding phase portraits in the x2 , x2 space. For 0.6 V 0.78, the system encounters various bifurcations, such as grazing and period-doubling and chaotic attractors with periodic windows. We also note that a chaotic attractor is destroyed through an exterior crisis at V 0.718. An enlargement of Figure 6.3 in the range 0.78 V 0.815 is shown in Figure 6.5. Here, the bifurcation diagram starts with a P (5, 2) motion at V = 0.78. A route of a sequence of period-doubling bifurcations to chaos is clearly seen. The chaotic attractor is divided into parts by periodic windows appearing at many places. Particularly, we note a sudden jump between two dierent branches of P (3, 1) motions at V 0.808. A bifurcation diagram near this range shows that these two branches are connected to each other through a branch of unstable P (3, 1) motions,
Xiaopeng Zhao
0.8 0.4 0 -0.4 -0.8 -0.3 -0.1

0.8 0.4 0 -0.4 -0.8
0.1 0.3 0.5
107
-0.3
-0.1
0.1
0.3
0.5
(a)
0.8 0.4 0 -0.4 -0.8 -0.3 -0.1 0.1 0.3 0.5 0.8 0.4 0 -0.4 -0.8 -0.3
(b)
-0.1
0.1
0.3
0.5
(c)
(d)
Figure 6.4: Numerically computed phase portraits of mass m2 : (a) P (1, 1) motion at V = 0.55, (b) P (2, 2) motion at V = 0.575, (c) near grazing P (2, 2) motion at V = 0.58, and (d) P (3, 2) motion at V = 0.58. as seen in Figure 6.6. Here, three dierent P (3, 1) motions coexist.
6.3
Chatter and Sticking Motions
Figure 6.7 shows the time history and phase portrait of m2 for a P (3, 1) motion when V = 0.82. Here three impacts occur in every period, with one of them having signicantly larger impact velocity than the other two. Such behavior, often referred to as chatter [Budd and Dux, 1994], occurs because the overall forcing attracts m2 towards the stoppers after it is bounced away from them. A chattering motion has a sequence of low-velocity impacts occurring in nite time. If the sequence is long enough, the impacting velocity tends to zero and the movable block sticks to the stoppers. A chatter is called complete chatter if the sequence leads to sticking, otherwise it is called incomplete chatter. The development of chatters to sticking as the excitation voltage is increased
Xiaopeng Zhao
108
0.78 0.76 0.74 0.72 0.7 0.68 0.78 0.785 0.79 0.795 V 0.8 amp 0.805 0.81
(a)
1.0
Impact velocity
0.8 0.6 0.4 0.2 0 0.78 0.785 0.79 0.795 Vamp 0.8 0.805 0.81
(b) Figure 6.5: An enlargement of Figure 6.3 in the range 0.78 V 0.815. can be observed through the series of time histories shown in Figure 6.8. In Figure 6.9, we present an enlargement of the last part of Figure 6.3; it shows an increase in the number of chatters and a route of chatter to sticking. Here, the motion starts with a P (3, 1) chatter at V = 0.81 and encounters a grazing bifurcation at V 0.835 before it becomes chaotic at V 0.838. Then, the motion is stabilized onto a P (4, 1) chatter, which in turn becomes P (5, 1) after a grazing bifurcation and a narrow chaotic band. This pattern repeats until the chatter sequence is complete and achieves sticking at V 0.872. In Figure 6.9, we indicate by cn the voltage at which the P (n, 1) motion originates. Specically, we found c4 = 0.83950, c5 = 0.85520, c6 = 0.86318, c7 = 0.86728, and c8 = 0.86944. Interestingly, we can associate a constant with the above sequence dened as = cn+1 cn 0.5. cn cn1 (6.1)
6.4
Frequency-Response Curve
Wagg and Bishop [2001] studied a two-degree-of-freedom impact oscillator, where the equations of motion are piece-wise linear. They found that the resonant frequency of the impacting dynamics is
Xiaopeng Zhao
109
0.77 Displacement (max-min)
SN
0.76
SN
0.75
0.74 0.808
0.809
Vamp
0.81
Figure 6.6: Bifurcation diagram showing two stable branches of P (3, 1) motions connected through an unstable branch of P (3, 1) motions. Here, the solid lines stand for stable motions, the dotted line stands for unstable motions, and SN indicates a saddle-node bifurcation. dierent from that of the nonimpacting dynamics. We assume for a moment that the frame m1 is held still with respect to the ground and the stoppers are removed from the frame, and study the nonimpacting dynamics of the microactuator using Equation (4.2). The frequency-response curve of the nonimpacting dynamics exhibits a softening behavior with its peak bent to the left [Nayfeh and Mook, 1979; Nayfeh and Balachandran, 1995], as seen in Figure 6.10. On the other hand, the frequency-response curve of the impacting dynamics shows a hardening behavior with the peak bent to the right, as seen in Figure 6.11.
6.5
Summary
We studied the post-grazing dynamics of the Mita microactuator. Using the previously developed methodology, we examined the piece-wise smooth dynamics and the associated bifurcations under changes of certain system parameters. A plethora of bifurcations, such as period-doubling bifurcation, saddle-node bifurcation, grazing bifurcation, chaos, and crises, are identied using direct numerical simulation as well as a root-nding technique based on the Newton-Raphson algorithm. The grazing bifurcations may introduce jumps in the bifurcation diagram when the applied volt-
Xiaopeng Zhao

1.5
110
0.5 0.3
0.5
0.1 -0.1 180 190 200
-0.5 -0.3 -0.1 0.1 0.3 0.5
(a)
(b)
Figure 6.7: Time history (a) and phase portrait (b) of m2 representing a P(3, 1) chattering motion at V = 0.82. We note that the three impact velocities are positive in (b). An enlargement of the boxed area is shown in Figure 6.8 (a). age is slowly varied. We found that a sequence of chatter dynamics develops into sticking when the applied voltage is slowly increased. The pattern of chatter-to-sticking is illustrated using the bifurcation diagram and a series of time histories. We found a constant associated with the chatter sequence. A study of the frequency-response curves reveals that impacts not only change the resonant frequency of the system but also show a hardening behavior, whereas the nonimpacting dynamics show a softening behavior. The numerical results bring out the importance of understanding the underlying dynamics to ensure a robust and consistent device operation. The present study also emphasizes the need for feedback control, especially for parameter ranges where chaotic attractors exist.
Xiaopeng Zhao
111
(a)
(b)
(c)
(d)
(e)
(f)
Figure 6.8: Time histories of the displacement x2 at excitation voltages showing the transition from chatter to sticking. Here, (a) shows an enlargement of the boxed area in Figure 6.7 (a) and (b) (f) show zooms near the maximum displacement of the corresponding time history to show the increasing number of chatters and sticking: (a) P (3, 1) chattering motion at V = 0.82, (b) P (4, 1) chattering motion at V = 0.85, (c) P (5, 1) chattering motion at V = 0.86, (d) P (6, 1) chattering motion at V = 0.864, (e) P (7, 1) chattering motion at V = 0.868, and (f) sticking motion at V = 0.872.
Xiaopeng Zhao
112
0.76 0.74 0.72 0.7 0.68 0.81
c4
c5
c6 c7 c8
0.82
0.83
0.84 Vamp0.85
0.86
0.87
0.88
(a)
1.2
Impact velocity
1.0 0.8 0.6 0.4 0.2 0 0.81 0.82 0.83 0.84 0.85 Vamp 0.86 0.87 0.88
(b) Figure 6.9: Bifurcation diagram showing a route of chatter to sticking. Here, cn indicates the voltage at which the P (n, 1) chattering motion originates.
1.2 Displacement (max-min)
0.9
0.6
0.3 0.45 0.47 ! 0.49 0.51
Figure 6.10: Nonimpacting frequency-response curve showing variation of the maximum minus minimum displacements with forcing frequency, when V = 0.175.
Xiaopeng Zhao
113
0.04
G
Displacement (max-min)
0.96
SN
Impact velocity
SN
0.02
0.956
0
0.952 0.49 ! 0.498
0.49
0.498
(a)
(b)
Figure 6.11: Impacting frequency-response curves when V = 0.175. The dashed line corresponds to nonimpacting periodic motions, the dotted lines correspond to unstable impacting P (1, 1) motions, and the solid line corresponds to stable impacting P (1, 1) motions; SN stands for saddle-node bifurcation, and G stands for grazing bifurcation.
Chapter 7
Conclusion and Recommendation for Future Work

7.1 Conclusions
In this dissertation, we investigated the modeling and simulation methodologies for microelectromechanical systems (MEMS) driven by electric forces. In the rst part, we developed a reduced-order model to study the mechanical response of electrically excited microplates, which are the actuation components in many micropumps, microsensors, and microswitches. Plates with rectangular shape and various boundary conditions were considered. We used the nonlinear von K arm an plate equations to account for the geometric nonlinearity due to large deection of the microplates. We considered the case where the gap between the exible electrode (microplate) and the xed electrode is small compared to the in-plane dimensions. We then used the Galerkin procedure to reduce the nonlinear partial-dierential equations of motion into a system of nonlinearly coupled ordinary-dierential equations. As numerical examples, we studied the static response and vibration characteristics of a simply supported microplate and a fully clamped microplate. In both examples, we investigated the convergence of the solution as the number of modes retained in the Galerkin approximation increases. Comparison between the numerical results and available experimental data validates the developed model. We 114
Xiaopeng Zhao
Chapter 7. Conclusions
115
found that the pull-in voltage and the maximum deection at pull-in computed with the nonlinear plate theory are signicantly larger than those computed with the linear plate theory. This brings out the need for nonlinear analysis when the plate deection is comparable to its thickness. The reduced-order model provides an eective and accurate design tool for microplate-based MEMS. In the second part, we considered impact microactuators whose work principle is based on impacts and stick-slip transitions. Impact microatuators have the ability to achieve precise positioning without large strokes and high actuation voltages and thus have been proposed for applications in microscopes, assembly of micromachines, nanoscale data storage, and micro surgeries. The state of the art is still based on trial and error and lacks a comprehensive modeling and simulation approach. We presented a methodology for modeling and numerical simulation of such devices. A rigid multibody approach was employed, where impacts and stick-slip transitions are assumed to be instantaneous. One advantage of this method is that wave eects during impacts or during stick-slip contact are neglected. However, the rigid multibody approach makes the equations of motion nonsmooth. A continuation method was developed to nd system parameters corresponding to periodic grazing orbits. Further, a root-nding technique was developed to nd periodic impact motions. Using the proposed modeling and analysis tools, we studied the Mita microactuator as an example. We identied three dierent switching on -o mechanisms, corresponding to dierent types of grazing bifurcations. Each of these on -o mechanisms can be generalized to any impacting system with similar nonlinearities. As grazing bifurcations introduce dramatic changes in system behavior without advance warning, the design and control of impact microactuators rely on an understanding of the critical transition between the asymptotic nonimpacting and impacting system dynamics and the associated changes in system behavior. To study the local characteristics of grazing bifurcations, we formulated discrete maps that i) capture the dynamics in the vicinity of grazing contact; ii) can be entirely computed by conditions at the grazing contact; iii) are nonsmooth in the deviation from the point of grazing contact; and iv) can be studied to any arbitrary order of accuracy. Results of the local analysis are in excellent agreement with numerical simulations, with the computation time reduced by several orders of magnitude. Intuitively, a large driving voltage would produce larger displacements in the impact microac-
Xiaopeng Zhao
116
tuator. However, numerical simulations reveal that the impacting dynamics undergo various bifurcations when the driving voltage is increased beyond the voltage corresponding to nonimpactinggrazing periodic motion. We found that grazing bifurcations change the periodicity of the impacting motions and introduce hysteresis in the system response when the driving voltage is changed quasi-statically. We identied a series of low-velocity impacts, known as chatter. The pattern of chatter-to-sticking was illustrated using a bifurcation diagram and a series of time histories. We also found a constant associated with the chatter sequence. The frequency-response study revealed that impacts not only change the resonant frequency of the system but also change the system response from softening to hardening. The analysis of impact actuators brings out the importance of understanding the underlying dynamics as to ensure a robust and consistent device operation. The current study also emphasizes the need for device optimization and feedback control, especially for the parameter ranges where chaotic attractors exist.
7.2
Recommendation for Future Work
The current research can be improved and extended in several ways. The following are suggested for future work:
1. When the microplate is deected, the frequencies of certain modes may become commensurate, thereby requiring the investigation modal interactions. 2. Many MEMS devices employ multi-physical elds. The reduced-order model should therefore be extended to take into account thermal loads, squeeze-lm damping, and structure-uid interaction. These reduced models can be further used for optimal design of microsensors, micropumps, and microswitches. The inuence of the shape of the xed electrode on the pull-in voltage and vibration characteristics should also be studied. 3. Experimental work on various impact microactuators needs to be conducted to provide a test bed for the developed modeling and simulation methodology. 4. Control strategies need to be developed to reduce the response time, to change the grazing bifurcation from the discontinuous type to the continuous type or vice versa, and to eliminate
Xiaopeng Zhao
117
or suppress the unwanted chaotic motions in the impact microactuator. Further, various combination of system parameters need to be studied to achieve optimal design. The modeling and simulation of other types of linear and angular microvibromotors need to be investigated. 5. Tools and methodologies developed for the impact microactuator can be used to study other impacting systems. An example is the intermittent impact of the micro-cantilever with substrate in an atomic force microscope.
Bibliography
http://www.coventorware.com. http://www.intellisense.com. E. M. Abdel-Rahman, M. I. Younis, and A. H. Nayfeh. Characterization of the mechanical behavior of an electrostatically actuated microbeam. Journal of Micromechanics and Microengineering, 12:759766, 2002. R. C. Ackerberg. On a nonlinear dierential equation of electrohydrodynamics. Proc. Roy. Soc. A, 312:129140, 1969. J. Adolfsson, H. Dankowicz, and A. Nordmark. 3d passive walkers: Finding periodic gaits in the presence of discontinuities. Nonlinear Dynamics, 24:205229, 2001. L. Azrar, E. H. Boutyour, and M. Potier-Ferry. Nonlinear forced vibrations of plates by an asymptotic numerical method. Journal of Sound and Vibration, 252(4):657674, 2002. N. S. Bardell. The application of symbolic computing to the hierarchical nite element method. International Journal for Numerical Methods in Engineering, 28:11811204, 1989. N. S. Bardell. Free vibration analysis of a at plate using the hierarchical nite element method. Journal of Sound and Vibration, 151(2):263289, 1991. R. Benamar. Nonlinear Dynamic Behavior of Fully Clamped Beams and Rectangular Isotropic and Laminated Plates. PhD thesis, University of Southampton, 1990. R. Benamar, M. M. K. Bennouna, and R. G. White. The eects of large vibration amplitudes
118
Xiaopeng Zhao
Bibliography
119
on the mode shapes and natural frequencies of thin elastic structures. part II: Fully clamped rectangular plates. Journal of Sound and Vibration, 164:399424, 1993. Breguet J. M. Schmitt C. Bergander, A. and R. Clavel. Micropositioner for microscopy applications based on stick-slip eect. Proceedings of the 2000 Micromechatronics and Human Science, pages 213216, 2000. T. Bourouina and J. P. Grandchamp. Modeling micropumps with electrical equivalent networks. Journal of Micromechanics and Microengineering, 6:398404, 1996. J. M. Breguet and R. Clavel. Stick and slip actuators: design, control, performances and applications. Proceedings of the 1998 Micromechatronics and Human Science, pages 8995, 1998. B. Brogliato. Nonsmooth Mechanics. Springer-Verlag, New York, 1999. E. J. Brunelle and S. R. Robertson. Vibrations of an initially stressed thick plate. Journal of Sound and Vibration, 45:405416, 1976. Chris Budd and Felix Dux. Chattering and related behavior in impact oscillators. Philosophical Transactions: Physical Sciences and Engineering, 347(1683):365389, May 16 1994. K-S Chen, K-S Ou, and L-M Li. Development and accuracy assessment of simplied electromechanical coupling solvers for MEMS applications. Journal of Micromechanics and Microengineering, 14:159169, 2004. C. Y. Chia. Nonlinear Analysis of Plates. McGraw-Hill, New York, 1980. C. K. Chiang, C. Mei, and C. E. Gray Jr. Finite element large-amplitdue free and forced vibrations of rectangular thin composite plates. Journal of Vibration and Acoustics, 113:309315, 1991. A. Cozma and R. Puers. Electrostatic actuation as a se-testing method for silicon pressure sensors. Sensors and Actuators, 60:3236, 1997. P. Cusin, T. Sawai, and S. Konish. Compact and precise positioner based on the inchworm principle. Journal of Micromechanics and Microengineering, 10:516521, 2000.
Xiaopeng Zhao
Bibliography
120
M. J. Daneman, N. C. Tien, O. Solgaard, A. P. Pisano, K. Y. Lau, and R. S. Muller. Linear microvibromotor for positioning optical components. IEEE Journal of Microelectromechanical Systems, 5(3):159165, 1996. H. Dankowicz and A. B. Nordmark. On the origin and bifurcations of stick-slip oscillations. Physica D, 136:280302., 1999. H. Dankowicz, P. Piiroinen, and A. B. Nordmark. Low-velocity impacts of quasiperiodic oscillations. Chaos, Solitons and Fractals, 14:241255, 2002. M. di Bernardo, C. J. Budd, and A. R. Champneys. Normal form maps for grazing bifurcations in n-dimensional piecewise-smooth dynamical systems. Physica D, 160:222254, 2000. J. G. Eisley. Large amplitude vibration of buckled beams and rectangular plates. AIAA Journal, 2:22072209, 1964. W. F. Faris, E. M. Abdel-Rahman, and A. H. Nayfeh. Mechanical behavior of an electrostatically actuated micropump. Proc. of 43rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2002. R. S. Fearing. Survey of sticking eects for micro parts handling. In Proceedings of International Conference on Intelligent Robots and Systems, pages 212217, 1995. C. A. J. Fletcher. Computational Galerkin Methods. Springer Verlag, New York, 1984. S. Foale and R. Bishop. Bifurcations in impacting systems. Nonlinear Dynamics, 6:285299, 1994. O. Francais and I. Dufour. Dynamic simulation of an electrostatic micropump with pull-in and hysteresis phenomena. Sensors and Actuators A, 70:5660, 1998. O. Francais and I. Dufour. Normalized abacus for the global behavior of diaphragms: pneumatic, electrostatic, piezoelectric or electromagnetic actuation. Journal of Modeling and Simulation of Microsystems, 2:149160, 1999. M. H. Fredriksson and A. B. Nordmark. Bifurcations caused by grazing incidence in many degrees of freedom impact oscillators. Proceedings of the Royal Society of London, series A, 453:12611276, 1997.
Xiaopeng Zhao
Bibliography
121
M. H. Fredriksson and A. B. Nordmark. On normal form calculations in impact oscillators. Proceedings of the Royal Society of London, series A, 456:315330, 2000. J. Hadian and A. H. Nayfeh. Modal interaction in circular plates. Journal of Sound and Vibration, 142:279292, 1990. T. Higuchi. Application of electromagnetic impulsive force to precise positioning tools in robot system. Proceedings of the 2nd International Symposium of Robotics, pages 144148, 1984. T. Higuchi., Y. Yamagata, K. Furutani, and K. Kudoh. Precise positioning mechanism utilizing rapid deformations of piezoelectric elements. IEEE Conference on Microelectromechanical Systems, pages 222226, 1990. F. B. Hildebran, E. Reissner, and G. B. Thomas. Notes on the foundations of the theory of small displacements of orthotropic shells. NASA Technical Note No. 1833, 1949. P. C. Hsu, C. H. Mastrangelo, and K. D. Wise. A high sensitivity polysilicon diaphragm condenser microphone. MEMS Conference, 1998. C. Huang and B. E. Sandman. Large amplitude vibrations of a rigidly clamped circular plate. International Journal of Non-Linear Mechanics, 6:451468, 1971. A. P. Ivanov. Bifurcations in impact systems. Chaos, Solitons and Fractals, 7(10):16151634, 1996. T. Y. Jiang, T. Y. Ng, and K. Y. Lam. Optimization of a piezoelectric ceramic actuator. Sensors and Actuators: A, 84:8194, 2000. U. Jungnickel, D. Eicher, and H. F. Schlaak. Miniaturized micro-positioning system for large displacements and large forces based on an inchworm platform. 8th International Conference on New Actuators, pages 684687, 2002. C. H. Kim and Y. K. Kim. Micro xy-stage using silicon on a glass substrate. Journal of Micromechanics and Microengineering, 12:103107, 2002. E-R Konig and G. Wachutka. Analysis of unstable behavior occurring in electro-mechanical microdevices. In MSM Conference 1999, pages 330333, 1999.
Xiaopeng Zhao
Bibliography
122
G. C. Kung and Y. H. Pao. Nonlinear exural vibration of a clamped circular plate. Journal of Applied Mechanics, 39:10501054, 1972. A. P. Lee, D. J. Jr. Nikkel, and A. P. Pisano. Polysilicon linear microvibromotors. Proc. 7th Int. Conf. Solid-State Sensors and Actuators, 1(2):4649, 1993. R. I. Leine, D. H. Van Campen, and B. L. Van de Vrande. Bifurcations in nonsmooth discontinuous systems. Nonlinear Dynamics, 23(2):105164, 2000. A. W. Leissa. Vibration of Plates. Acoustical Society of America, 1993. S. Levy. Bending of rectangular plates with large deections. Technical report, NACA, 1942. L. Librescu. The thermoelastic problem of shells and plates approached by eliminating the lovekirchho hypothesis. Studii si Cercetari de Macanica Aplicata, 21:351365, 1966. L. Librescu. The elasto-kinetic problems in the theory of anisotropic shells and plates-II: Theory of plates. Revue Roumaine des Sciences Techniques. Serie de Mecanique Appliquee, 14:637654, 1969. L. Librescu and J. N. Reddy. A few remarks concerning several rened theories of anisotropic composite laminated plates. International Journal of Engineering Science, 27:515527, 1989. Y. T. Liu and T. Higuchi. Precision positioning device utilizing impact force of combined piezopneumatic actuator. IEEE Transactions on Mechatronics, 6(4):467473, 2001. D. W. Lobitz, A. H. Nayfeh, and D. T. Mook. Nonlinear analysis of vibrations of irregular plates. Journal of Sound and Vibration, 50:203217, 1977. D. Sauer M. Varghese, R. Amantea and S. D. Senturia. Resistive damping of pulse-sensed capacitive position sensors. In Tech. Dig. 1997 Intl. Conf. Solid-State Sensors and Actuators (Transducers 97), pages 11211124, June 1997. C. Malhaire, M. Le Berre, M. Febvre, D. Barbier, and P. Pinard. Eect of clamping conditions and built-in stresses on the thermopneumatic deection of SiO2 /Si membranes with various geometries. Sensors and Actuators A, 74:174177, 1999.
Xiaopeng Zhao
Bibliography
123
C. Mei. Finite element displacement method for large amplitude free exural vibrations of beams and plates. Computers and Structures, 3:163174, 1973. R. D. Mindlin. Inuence of rotatory inertia and shear on exural motions of isotropic elastic plates. Journal of Applied Mechanics, 18(3138), 1951. M. Mita. Personal communication. 2003. M. Mita, M. Arai, S. Tensaka, D. Kobayashi, and H. Fujita. A micromachined impact microactuator driven by electrostatic force. IEEE Journal of Microelectromechanical Systems, 12(1):3741, 2003. J. Molenaar, J. G. de Weger, and W. Van de Water. Mappings of grazing impact oscillators. Nonlinearity, 14(2):301321, 2001. T. Morita, R. Yoshida, Y. Okamoto, M. K. Kurosawa, and T. Higuchi. A smooth impact rotation motor using a multi-layered torsional piezoelectric actuator. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 46(6):439144, 1999. M. V. V. Murthy. Nonlinear character of response in stretched strings. Journal of the Acoustical Society of America, 38:461471, 1965. H. C. Nathanson, W. E. Newell, R. A. Wickstrom, and J. R. Davis. The resonant gate transistor. IEEE Trans. on Electron Devices, 14:117133, 1967. A. H. Nayfeh. Perturbation Methods. John Wiley Interscience, New York, 1973. A. H. Nayfeh. Introduction to Perturbation Techniques. John Wiley Interscience, New York, 1981. A. H. Nayfeh. Problems in Perturbation. John Wiley Interscience, New York, 1985. A. H. Nayfeh. Method of Normal Forms. John Wiley Interscience, New York, 1993. A. H. Nayfeh. Nonlinear Interactions: Analytical, Computational, and Experimental Methods. Wiley, New York, 2000. A. H. Nayfeh and B. Balachandran. Applied Nonlinear Dynamics: Analytical, Computational, and Experimental Methods. Wiley, New York, 1995.
Xiaopeng Zhao
Bibliography
124
A. H. Nayfeh and C. M. Chin. Perturbation Methods with Mathematica. Dynamics Press, Virginia, 1999. A. H. Nayfeh and D. T. Mook. Nonlinear Oscillations. Wiley, New York, 1979. A. H. Nayfeh and P. F. Pai. Linear and Nonlinear Structural Mechanics. Wiley, New York, 2004. T. Y. Ng, T. Y. Jiang, H. Li, K. Y. Lam, and Reddy. A coupled eld study on the non-linear dynamic characteristics of an electrostatic micropump. Journal of Sound and Vibration, 273: 9891006, 2004. N-T Nguyen, X. Huang, and T. K. Chuan. MEMS-micropumps: A review. Transactions of the ASME, 124:384392, 2002. A. B. Nordmark. Non-periodic motion caused by grazing incidence in an impact oscillator. Journal of Sound and Vibration, 145:279297, 1991. A. B. Nordmark. Universal limit mapping in grazing bifurcations. Physical Review E, 55(1):266 270, 1997. K. Oh and A. H. Nayfeh. High- to low- frequency modal interactions in a cantilever composite plate. Journal of Vibration and Acoustics, 120(2):579587, 1998. O. Ohmichi, Y. Yamagata, and T. Higuchi. Micro impact drive mechanisms using optically excited thermal expansion. IEEE Journal of Microelectromechanical Systems, 6(3):200207, 1997. J. A. Pelesko. Electrostatic eld approximations and implications for MEMS devices. In ESA 2001 Proceedings, pages 126137, 2001. J. A. Pelesko and X. Y. Chen. Electrostatic deections of circular elastic membranes. Journal of Electrostatics, 57(1):112, 2003. C. Rajalingham and R. B. Bhat. Inuence of electric eld on diaphragm stability and vibration in a condenser microphone. Journal of Sound and Vibration, 211(5):819827, 1998. T. C. Reiley, L. S. Fan, and H. J. Mamin. Micromechanical structures for data storage. Journal of Microelectronic Engineering, 27:495498, 1995.
Xiaopeng Zhao
Bibliography
125
P. Ribeiro and M. Petyt. Nonlinear vibration of plates by the hierarchical nite element and continuation methods. International Journal of Mechanical Sciences, 41:437459, 1999. P. Ribeiro and M. Petyt. Nonlinear free vibration of isotropic plates with internal resonance. International Journal of Nonlinear Mechanics, 35:263278, 2000. O. L. Roufaeil and D. J. Dawe. Raleigh-ritz vibration analysis of rectangular mindlin plates subjected to membrane stresses. Journal of Sound and Vibration, 85:263275, 1982. M. T. A. Saif, B. E. Alaca, and H. Sehitoglu. Analytical modeling of electrostatic membrane actuator for micropumps. Journal of Microelectromechanical Systems, 8:335345, 1999. K. Saitou, D. A. Wang, and S. J. Wou. Externally resonated linear microvibromotor for microassembly. IEEE Journal of Microelectromechanical Systems, 9(3):36346, 2000. M. Sathyamoorthy. Nonlinear Analysis of Structures. CRC Press, Boca Raton, Florida, 1997. P. R. Scheeper, A. G. H. van der Donk, and P. Bergveld. A review of silicon microphones. Sensors and Actuators A, 44:111, 1999. R. Schmidt. Rened nonlinear theory of plates with transverse shear deformation. Industrial Mathematics, 27:2338, 1977. Joseph I. Seeger and Bernhard E. Boser. Charge control of parallel-plate, electrostatic actuators and the tip-in instability. Journal of Microelectromechanical Systems, 12(5):656671, 2003. S. W. Shaw and P. J. Holmes. A periodically forced piecewise linear oscillator. Journal of Sound and Vibration, 90(1):129155, 1983. Y. Shi and C. Mei. A nite element time domain modal formulation for large amplitude free vibrations of beams and plates. Journal of Sound and Vibration, 193:453464, 1996. S. Sridhar, D. T. Mook, and A. H. Nayfeh. Non-linear resonances in the forced response of plates, part i: Symmetric response of circular plates. Journal of Sound and Vibration, 41:359373, 1975. S. Sridhar, D. T. Mook, and A. H. Nayfeh. Nonlinear resonances in the forced responses of plates, part II: Asymmetric response of circular plates. Journal of Sound and Vibration, 59:159170, 1978.
Xiaopeng Zhao
Bibliography
126
S. H. Strogatz. Nonlinear Dynamics and Chaos. Perseus Books, Massachusetts, 1998. G. I. Taylor. The coalescence of closely spaced drops when they are at dierent electric potentials. Pro. Roy. Soc. A, 306:423434, 1968. T. L. Teng, C. C. Liang, and C. C. Liao. Nonlinear forced vibration analysis of rectangular plates by fourier series method. Computational Mechanics, 23:17, 1999. H. A. Tilmans and R. Legtenberg. Electrostatically driven vacuum-encapsulated polysilicon resonators: Part ii. theory and performance. Sensors and Actuators A, 45:6784, 1994. S. P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells. McGraw-Hill, New York, 1959. G. W. Vogl and A. H. Nayfeh. A reduced-order model for electrically actuated clamped circular plates. In ASME International 19th Biennial Conference on Mechanical Vibration and Noise, page DETC2003/VIB48530, Chicago, Illinois, 2003. D. J. Wagg and R. Bishop. Chatter, sticking and chaotic impacting motions in a two degree of freedom impact oscillator. International Journal of Bifurcation and Chaos, 11(1):5771, 2001. R. G. White. Eects of nonlinearity due to large deections in the resonance testing of structures. Journal of Sound and vibration, 16:225267, 1971. N. Yamaki. Inuence of large amplitudes on exural vibrations of elastic plates. Zeitschrift f&udi. r angewandte Mathematik und Mechanik, 41:501510, 1961. F. Yang. Electromechanical instability of microscale structures. Journal of Applied Physics, 92(5): 27892794, 2002. M. I. Younis, E. M. Abdel-Rahman, and A. H. Nayfeh. A reduced-order model for electrically actuated microbeam-based mems. Journal of Microelectromechanical Systems, 12(5):672680, 2003. Y. Y. Yu. Vibrations of Elastic Plates. Springer, New York, 1996. R. Zengerle, A. Richter, and H. Sandmaier. A micro membrane pump with electrostatic actuation. Micro Electro Mechanical Systems Conference, 1992.
Xiaopeng Zhao
Bibliography
127
X. M. Zhang, F. S. Chau, C. Quan, Y. L. Lam, and A. Q. Liu. A study of the static characteristics of a torsional micromirror. Sensors and Actuators A, 90:7381, 2001. X. Zhao, E. M. Abdel-Rahman, and A. H. Nayfeh. Mechanical behavior of an electrically actuated microplate. In ASME International 19th Biennial Conference on Mechanical Vibration and Noise, page DETC2003/VIB48531, Chicago Illinois, 2003. X. Zhao, E. M. Abdel-Rahman, and A. H. Nayfeh. A reduced order model for electrically actuated microplates. Journal of Micromechanics and Microengineering, 14:900906, 2004a. X. Zhao, H. Dankowicz, C. K. Reddy, and A. H. Nayfeh. Modelling and simulation methodology for impact microactuators. Journal of Micromechanics and Microengineering, 14:775 784, 2004b. R. C. Zhou, D. Y. Xue, and C. Mei. Finite element time domain-modal formulation for nonlinear utter of composite panels. American Institute of Aeronautics and Astronautics Journal, 32: 20442052, 1994. J. Zhu, S. K. Datta, and A. H. Shah. Modal representation of transient dynamics of laminated plates. Composite Engineering, 5(12):949963, 1995. O. C. Zienkiewicz. The Finite Element Method. Butterworth-Heinemann, Boston, 2000.
Appendix A
The Implicit Function Theorem

Suppose that F (x, y) is a C k function (for some k > 1) into Rm in some region containing the point (x0 , y0 ) Rn+m , such that F (x0 , y0 ) = 0 and that the m m matrix A = Fy (x0 , y0 ) is invertible. Then, it follows that there exists a unique C k function y (x) on a neighborhood of x0 , such that y (x0 ) = y0 and F (x, y (x)) 0.
128
Appendix B
Hierarchical Finite Element Method

The main dierence between the nite-element method and the classical Rayleigh-Ritz method lies in the choice of admissible functions [Zienkiewicz, 2000; Bardell, 1989, 1991]. The usual niteelement methods divide the domain of interest into a number of convex subdomains. The solution is then locally approximated by piecewise smooth admissible polynomial functions over each individual subdomain. The admissible polynomial functions usually have a low degree of freedom. The most common way to improve the accuracy of the nite-element approximation is to rene the nite-element mesh while keeping the degree of the approximation polynomials xed. As an alternate method, the hierarchical nite-element method increases the degree of the approximating polynomial functions while keeping the mesh size xed. One advantage of this method, namely the feasibility to model a plate as one hierarchical nite-element, has attracted the author to adopt this method. Using hierarchical displacement functions, we can write the middle-plane displacements as
pi pi u u Nj qj , j =1 po v v Nj qj , j =1
u= where
v=
w=
j =1
w w Nj qj ,
u [Nj ] = [g1 ( )g1 ( ) g1 ( )g2 ( ) . . . gpi ( )gpi ( )] w [Nj ] = [f1 ( )f1 ( ) f1 ( )f2 ( ) . . . fpo ( )fpo ( )]
129
Xiaopeng Zhao
Appendices
130
Here, gi and fi are the in-plane and out-of-plane shape functions, where pi and po are the number of in-plane and out-of-plane shape functions, respectively. The rst two linear in-plane functions and the rst four cubic out-of-plane functions used in the conventional FEM are retained and listed as follows 1 (1 ) 2 1 g2 ( ) = (1 + ) 2 1 3 1 f1 ( ) = + 3 2 4 4 1 1 1 1 f2 ( ) = 2 + 3 8 8 8 8 1 3 1 3 f3 ( ) = + 2 4 4 1 1 1 1 f2 ( ) = + 2 + 3 8 8 8 8 g1 ( ) = For the higher-order in-plane and out-of plane functions, we use the following set of shape functions derived from Rodrigues form of Legendre polynomials [Bardell, 1989]:
int(r/21)
gi =
n=0 int(r/21)
(1)n (2r 2n 5)!! r2n1 2n n!(r 2n 1)! (1)n (2r 2n 7)!! r2n1 2n n!(r 2n 1)!
(r > 2)
fi =
n=0
(r > 4)
where r!! = r(r 2) . . . (2or1), 0!! = (1)!! = 1, and int is a function that outputs the integer part of its argument. We note that all of the higher-order in-plane functions have zero displacement and non-zero slope at each end of the element and all of the higher-order out-of-plane functions have both zero displacement and zero slope at each end of the element. Therefore, the higher-order functions represent an internal displacement eld.
Vita
Xiaopeng Zhao was born in Shanxi, China on April 19th, 1974. He earned his Bachelors degree in 1996 and Masters Degree in 1999 both from the Engineering Mechanics Department in Tsinghua University, Beijing, China. He joined the Ph.D. program in Engineering Science and Mechanics Department at Virginia Tech, Blacksburg, Virginia in August 1999. Since then, he has been working under the supervision of Dr. Ali Nayfeh in the modeling and simulation of various MEMS devices, and their control and optimization.
131

Dissertation

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Dissertation

Hochgeladen von

Copyright:

Verfügbare Formate

Modeling and Simulation of MEMS Devices

Doctor of Philosophy in Engineering Mechanics

August 2, 2004 Blacksburg, Virginia

Modeling and Simulation of MEMS Devices

2 Literature Review 2.1

Dynamic Tools for Nonsmooth Systems . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.2.1 Dynamical systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2.2 2.2.3 2.2.4 2.2.5

3 Mechanical Behavior of Electrically Actuated Microplates 3.1 3.2 Motivation

Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Linear Mode Shapes and Corresponding Natural Frequencies . . . . . . . . . 41

4 Modeling and Simulation of Impact Microactuators 4.1 4.2 4.3 Motivation

5 Local Analysis of Grazing Bifurcations 5.1 5.2 5.3 Motivation

Transition Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Discontinuity mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 5.3.1 Near-grazing Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 103

7 Conclusion and Recommendation for Future Work 7.1 7.2

Bibliography A The Implicit Function Theorem B Hierarchical Finite Element Method

A schematic of an air-gap capacitor. . . . . . . . . . . . . . . . . . . . . . . . . . . .

Variation of switch on and o voltages with respect to the excitation frequency. . . . 69

A collection of bifurcation curves in the parameter space, where represents the

Challenges in MEMS Modeling

Electric Actuation and Pull-in Instability

Electrically Actuated Microplates

Impacts and Friction

Nonlinear Analysis of Plates

Chapter 2. Literature Review

Chapter 2. Literature Review

Chapter 2. Literature Review

are the Jaumann strains.

= e1 zwxx , = e2 zwyy , = 6 2zwxy , =

where the midplane strain components e1 , e2 , and 6 are 1 2 e1 = ux + wx , 2 1 2 e2 = vy + wy , 2 6 = uy + vx + wx wy . (2.6)

Chapter 2. Literature Review

Substituting (2.9) and (2.10) into (2.2) yields T =

where {I0 , I1 , I2 } {1, z, z 2 }dz.

We note that I1 = 0 due to the homogeneous assumption.

Chapter 2. Literature Review

Substituting (2.5) and (2.6) into (2.3) yields =

11 (ux + wx wx zwxx ) + 22 (vy + wy wy zwyy )

+ 12 (uy + vx + wx wy + wy wx 2zwxy ) dAdz =

N1 (ux + wx wx ) + N6 (uy + vx + wx wy + wy wx ) + N2 (vy + wy wy ) M1 wxx M2 wyy 2M6 wxy dA =

(N1x + N6y ) u + (N2y + N6x ) v

+ M1xx + M2yy + 2M6xy + (N1 wx )x + (N2 wy )y + (N6 wx )y + (N6 wy )x w dA +

Chapter 2. Literature Review

At (x, y ) = (0, 0), (a, b), (a, 0), (0, b) : w = 0 or M6 = 0 (2.19)

Chapter 2. Literature Review

Chapter 2. Literature Review

Eliminating u and v from (2.6) yields the compatibility equation 2 e1 2 e2 2 e6 + 2 = y 2 x2 xy 2w xy

e6 = Substituting (2.30)(2.32) into (2.29) yields

2 4 = Eh wxy wxx wyy .

Chapter 2. Literature Review

Chapter 2. Literature Review

Dynamic Tools for Nonsmooth Systems

Xiaopeng Zhao conditions.

Chapter 2. Literature Review

Chapter 2. Literature Review

hx xin . xin f xin

Chapter 2. Literature Review

The map in Equation (2.42) can be written as D x, tin , tout

Xiaopeng Zhao and

Chapter 2. Literature Review

P (x ) = x is a xed point of the Poincar e map.

Chapter 2. Literature Review

Continuation of periodic solutions

Continuation of grazing periodic solutions.

Xiaopeng Zhao discontinuity D at a point x0 ; that is,