O p tim ization o f P iezo electric A ctu a to r C onfigu ration on a F lexib le F in for V ib ration C ontrol U sin g G en etic A lgorith m s

Transcription

1 O p tim ization o f P iezo electric A ctu a to r C onfigu ration on a F lexib le F in for V ib ration C ontrol U sin g G en etic A lgorith m s by A ndrew R ader, B.E ng. A Thesis subm itted to the Faculty of G raduate Studies and Research in partial fulfilment of the requirem ents for the degree of M aster o f A pplied Science Ottaw a-c arleton Institute for M echanical and Aerospace Engineering D epartm ent of Mechanical and Aerospace Engineering Carleton University Ottaw a, Ontario, C anada July 2005 Copyright Andrew Rader, B.Eng.

2 1*1 Library and Archives Canada Published Heritage Branch 395 Wellington Street Ottawa ON K1A 0N4 Canada Bibliotheque et Archives Canada Direction du Patrimoine de I'edition 395, rue Wellington Ottawa ON K1A 0N4 Canada Your file Votre reference ISBN: Our file Notre reference ISBN: NOTICE: The author has granted a nonexclusive license allowing Library and Archives Canada to reproduce, publish, archive, preserve, conserve, communicate to the public by telecommunication or on the Internet, loan, distribute and sell theses worldwide, for commercial or noncommercial purposes, in microform, paper, electronic and/or any other formats. The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. AVIS: L'auteur a accorde une licence non exclusive permettant a la Bibliotheque et Archives Canada de reproduire, publier, archiver, sauvegarder, conserver, transmettre au public par telecommunication ou par I'lnternet, preter, distribuer et vre des theses partout dans le monde, a des fins commerciales ou autres, sur support microforme, papier, electronique et/ou autres formats. L'auteur conserve la propriete du droit d'auteur et des droits moraux qui protege cette these. Ni la these ni des extraits substantiels de celle-ci ne doivent etre imprimes ou autrement reproduits sans son autorisation. In compliance with the Canadian Privacy Act some supporting forms may have been removed from this thesis. While these forms may be included in the document page count, their removal does not represent any loss of content from the thesis. Conformement a la loi canadienne sur la protection de la vie privee, quelques formulaires secondaires ont ete enleves de cette these. Bien que ces formulaires aient inclus dans la pagination, il n'y aura aucun contenu manquant. i * i Canada

3 A bstract An efficient solution to slow crack growth caused by buffet-induced vibration of vertical tails on tw in-tail fighter airplanes is to use piezoelectric actuators and sensors to create an adaptive structure th a t actively reduces the buffeting. The optim ization of the sensor and actuator configuration is critical to ensure th at the structure will be effective in reducing the vibration for th e anticipated loading conditions. This thesis uses genetic algorithm s to solve two optim ization problems for vibration control of the first three modes of the fin. The first problem optimizes the position of a single pair of piezoelectric actuators on the fin. The second problem optimizes the activation of a pre-determ ined num ber of actuator pairs. The fitness functions for optim ization are determ ined from the Frequency Response Functions (FRFs) m easured at an accelerometer for the activation of each possible actuator. Individual modal and m ulti-m odal acceleration and displacement vibration control are considered. The advantage of this m ethod lies in the decoupling of the fitness function form ulation from the optim ization. In comparison to previous approaches, this allows optim ization on much more complex geometries where the derivation of an analytical fitness function is prohibitive or impossible. The optim ization results obtained through sim ulation are verified through a comparison with results obtained from an experim ental model of the fin. The agreement between results from sim ulation and experim ent dem onstrates the validity of the optim izations.

4 A cknow ledgm ents I would like to thank Professor Fred Afagh and Dr. Aghil Yousefi-Koma for all their insight and time. They allowed me the freedom to work indepently and pursue my own research, but were always available when I needed their experience, support, or when I needed additional m otivation to keep me on track. This thesis could not have been com pleted w ithout th e assistance provided by Professor Afagh in particular. I am grateful to th e N ational Research Council of C anada, and specifically to everyone at the Aeroacoustics and Structural Dynamics, Structures, M aterials, and Propulsion (SDSMP) laboratory of the Institute for Aerospace Research (IAR). Dr. Yousefi-Koma and Eric Chen constructed th e experim ental m odel of th e fin and assisted greatly in g a th e r ing the experim ental results. Dr. Yousefi-Koma and Dr. David Zimcik at the SDSMP laboratory contributed in w riting the journal and conference papers prepared as part of this research, as did Professor Afagh at C arleton University. A conversation with Professor David Amundsen in the School of M athem atics and Statistics at C arleton University was instrum ental in the development of the surface fitting genetic algorithm. I would like to acknowledge the generous financial support of the N atural Sciences and Engineering Research Council (NSERC), and the funding provided for the project by the Canadian D epartm ent of N ational Defence (DND). I also would like to thank my family and Helena Forbes for their unconditional support. iii

5 List of Sym bols FRF Frequency Response Function x, y C artesian co-ordinate system Strain d t V Piezoelectric charge constant M aterial thickness Applied voltage Strain a T T herm al expansion coefficient T em perature g E a rth s gravitational constant (9.81 m /s 2) P n z a d r I m Polynom ial order Num ber of coefficients in a polynomial Polynom ial surface m agnitude Polynom ial surface coefficient Actual F R F m agnitude Range param eter Size of population pool Num ber of individuals in a population P Random num ber between 0 and 1 7 Random num ber between -1 and 1 A M utation severity param eter

6 w\, W2, UJ3 ui, u>2, ^ 3 q RME M E M odal weighting factors N atural frequencies Size of population for m ultiple actuator configuration optim ization Relative m ean effectiveness M inimum effectiveness v

7 Table of C ontents A bstract ii A cknow ledgm ents iii List of Sym bols iv Table o f C ontents vi List o f Tables x List o f Figures xii 1 Introduction Thesis Overview M otivation Active Tail Buffet Suppression P r o je c t V ibration Control of Individual M o d e s V ibration Control of M ultiple M o d e s Acceleration vs. Displacement C o n tro l B a c k g ro u n d Active M a te r ia ls Piezoelectric A ctuators A daptive S tr u c tu r e s Genetic A lgorithm s vi

8 1.3.5 Com bination of Finite Element M ethods with Genetic Algorithm s O b je c tiv e s F lexible Fin O verview and D ata E xtraction Finite Element M ethods O v e rv ie w Fin C o n fig u ra tio n s Experim ental Set-up vs. Finite Element M o d e l A ccelero m eters Finite Element Model for A ctuator Configuration O ptim ization Finite Element M ethod D ata Acquisition FEM Elements and Configuration Therm al A n a lo g y P atran and Frequency Response S im u la tio n N astran and Post Processing R e s u lts Excel Files and Selection of Peak F R F O ptim ization o f th e P osition o f a Single Pair o f A ctuators Approach: Polynomial Surface F itting to O btain a Continuous Fitness Function Suitability of Three-Dimensional Polynomial S u r f a c e s Fitness Function for Polynomial Surface F ittin g Convergence of S o lu tio n Genetic Algorithm for Surface F i t t i n g Genetic Algorithm S te p s Factors Affecting Convergence Polynomial Surface F itting R e s u lts First M o d e Second M o d e Third M o d e Genetic Algorithm for A ctuator Positional O p tim iz a tio n Fitness Function for Positional O p tim iz a tio n vii

9 3.4.2 Genetic Algorithm S te p s Positional O p tim iz a tio n M ultiple A ctuator G rouping O ptim ization Approach: Direct Peak F R F M axim ization Frequency Response Function Results First M o d e Second M o d e Third M o d e Genetic Algorithm for Grouping O p tim izatio n Fitness F u n c tio n Genetic Algorithm S te p s Assessing C onfigurations A ctuation A u th o r ity M inim um Effectiveness (ME) and R elative M ean Effectiveness (RM E) O ptim ization Types A ctuation M ax im izatio n M inimization of Num ber of A c t u a to r s Trade-off S tu d y Configurations for A ctuation M a x im iz a tio n First M o d e Second M o d e T hird M o d e M ulti-m odal A c c e le ra tio n M ulti-m odal Displacement Exam ple of a Trade-off S t u d y Verification o f Sim ulation through E xperim entation Experim ental D ata A c q u is itio n Test Bed Specifications viii

10 5.1.2 Experim ental Results Scaling of Experim ental R e su lts Comparison of Experim ental Results w ith Sim ulation Com parison w ith R esults from FEM m odel in E xperim ental Configuration (12 A c tu a to rs ) Comparison with Results from M ultiple A ctuator Configuration Optim ization (C hapter 4 ) Comparison with Results from Single A ctuator Positional O ptim ization (C hapter 3 ) C onclusions and R ecom m ations C o n c lu sio n s Recom mations List o f R eferences 99 A ppix A Surface F ittin g M atlab C ode 102 A ppix B Single A ctuator M atlab O ptim ization C ode 144 A ppix C G rouping O ptim ization M atlab C ode 168 ix

11 List of Tables 2.1 X and Y Accelerometer Positions on the Sm art Fin along with Corresponding Finite Element N o d e Number of Coefficients for a Polynomial Surface by Polynomial O rder Values of the Range Param eters for Each Polynomial O r d e r Approxim ate Number of G enerations per Second by Polynomial O rder on a 2000 MHz Pentium P C Effect of Polynomial Order on E r r o r Effect of the Initial Population Pool Size on Convergence Effect of M utation Probability on C o n v e rg e n c e Effect of M utation Severity on C o n v erg en ce Best Single 9th Order Surface and Sub-surface Coordinates and Error for the Second M o d e Optim um A ctuator Position w ith Different W eighting F a c to r s Exam ple of a Forced M utation (FM) Trigger During reproduction Selected A ctuation M axim ization Configurations for the First Mode Selected A ctuation M axim ization C onfigurations for th e Second M ode Selected A ctuation M axim ization Configurations for the T hird Mode Selected A ctuation M axim ization C onfigurations for th e M ulti-m odal Acceleration C o n tr o l Selected A ctuation M axim ization C onfigurations for th e M ulti-m odal Displacement C o n t r o l Comparison of N atural Frequencies from Simulation and Experim ental Model 83 x

12 5.2 M aterial Properties of Aluminum and Piezoelectric M a te ria l Accelerometer Inform ation for the Flexible Fin Test B e d C om parison of E xperim ental and Sim ulation R esults for th e F irst M ode of the Sm art F i n C om parison of E xperim ental and Scaled Sim ulation R esults for th e Second Mode of the Sm art F i n C om parison of E xperim ental and Scaled Sim ulation R esults for th e T h ird Mode of the Sm art F i n A ctuator Ranking from Most Effective to Least For Individual M odal Control A ctuator Ranking from Most Effective to Least For M ulti-m odal Control Comparison of A ctuator Ranking of M ultiple A ctuator O ptim ization with Experim ental Results for Individual Modal Control using Equivalent A ctuators Comparison of A ctuator Ranking of M ultiple A ctuator O ptim ization with Experim ental Results for M ulti-m odal Control using Equivalent A ctuators. 93 xi

13 List of Figures 1.1 Leading Edge Vortices of an F /A -18 at a High Angle of A ttack [ 1 ] F/A -18 Vertical Tail w ith Bonded Piezoelectric A ctuators [ 2 ] F irst M ode Dwell R eduction for Sim ulation (Left) and E xperim ent (Right) [3] The Fin Mode S h a p e s Geom etry of the Sm art Fin in Experim ental C onfiguration Finite Element Model of the Sm art Fin in Experim ental Configuration [4] Accelerometer Positions on the Sm art F i n The 48 A ctuator Positions from the Finite Element Model of the Sm art Fin A ctuator Modelling using 25 HEX20 Solid Elements [4] Exam ple of Excel Results File for Extraction of Peak F R F s Peak F R F at Accelerometer 1 with the A ctuator at each of the 48 Possible P o sitio n s Typical SRSS Error (left) and Normalized M ean Error (right) vs. G eneration P l o t s Calculation of Range Param eter r o Calculation of Range Param eter r i Calculation of Range Param eter V Exam ple of Crossover at the Second A llele Comparison of Convergence R ate W ith and W ithout the Im m igration O perator C om parison of th e D iscrete F R F Peak M agnitudes and th e Best 3rd O rder Polynomial Surface for the First M o d e xii

14 3.8 Com parison of the Discrete F R F Peak M agnitudes and the Best Single 9th O rder Polynomial Surface for the Second M o d e The A ctuator Test Area Subdivisions The First (a) and Second (b) Sub-surfaces for the Second M o d e The T hird (a) and Fourth (b) Sub-surfaces for the Second M o d e Com parison of the Discrete FR F Peak M agnitudes (a) and the Combined Subsurfaces (b) for the Second M o d e Comparison of the Discrete F R F Peak M agnitudes and the Best 8th Order Polynomial Surface for the T hird Mode The Optim um A ctuator Pair Position for the First Mode The Optim um A ctuator Pair Position for the Second M o d e The Optim um A ctuator Pair Position for the Third M o d e The Optim um A ctuator Pair Position for Acceleration C o n t r o l The Optim um A ctuator Pair Position for Displacement C o n tro l Combined M ulti-m odal Polynomial with W eighting Factors to Equalize Mean FR F M a g n itu d e s Discrete FR F Peak M agnitudes for the First M o d e Discrete FR F Peak M agnitudes for the Second M o d e Discrete FR F Peak M agnitudes for the Third Mode A ctuator Numbering S c h e m e A ctuation M aximization Configurations for the First M o d e A ctuation M aximization Configurations for the Second M o d e A ctuation M aximization Configurations for the Third M o d e M inim um Effectiveness (ME) and R elative M ean Effectiveness (RM E) versus Num ber of A ctuators for Acceleration C o n tro l A ctuation M axim ization Configurations for M ulti-m odal A cceleration C ontrol A ctuation M axim ization C onfigurations for M ulti-m odal A cceleration C ontrol Minimum Effectiveness (ME) and Relative M ean Effectiveness (RM E) versus Num ber of A ctuators for Displacement C o n tr o l xiii

15 4.12 A ctuation M axim ization Configurations for M ulti-m odal D isplacem ent C ontrol A ctuation M axim ization Configurations for M ulti-m odal D isplacem ent C ontrol Flexible Fin with Piezoelectric A ctuators [3 ] The Experim ental Test Bed [ 3 ] A Typical F R F vs. Time Plot xiv

16 C hapter 1 Introduction Buffet-induced vertical tail vibration is a significant problem, particularly for twin-tail fighter jets which require frequent inspection to prevent catastrophic failure [1]. In order to reduce the crack growth rate and therefore reduce the reliance on inspection, one of two possible approaches is generally adopted. The first involves stiffening the structure, partially to reduce the severity of vibration, but prim arily to strengthen the structure in order to reduce crack growth. The second, and potentially more efficient approach, is to use an adaptive structure to actively reduce the buffeting in order to reduce crack growth and increase the lifespan of a vertical tail fin. This thesis focuses on optim izing the configuration of piezoelectric actuators on a flexible vertical tail fin in order to actively reduce induced buffeting displacem ent and acceleration a t any point on the fin. 1.1 T h esis O verview C hapter 1 describes the active tail buffet suppression project, conducted in part at the N ational Research Council of Canada, which was the prim ary m otivation for this research. An overview of th e project is presented, and recent developm ents are expounded. Acceleration and displacement vibration control and the relative effects of each are described as they relate to a reference point on th e fin. A background of active m aterials, active stru c tures, genetic algorithm s, and some combinations of finite element m ethods with genetic algorithm s in current literatu re are presented. 1

17 C hapter 2 describes, in detail, the different configurations of the finite element and experim ental models of the flexible fin th at were used in this work. The m ethodology used to extract the frequency response results in term s of the frequency response function (FRF) from the finite element model using PATRAN and NASTRAN is discussed in detail. The experim ental setup and m ethodology for F R F extraction is described. C hapter 3 describes the optim ization of the position of a single pair of piezoelectric actuators on the fin by using a genetic algorithm to fit the m odal FR Fs onto continuous surfaces in order to obtain a continuous fitness function. The derivation of this fitness function is described in detail. A complete exam ination of all steps of the genetic algorithm is a prim ary focus of the chapter. Particular attention is paid to the initial population pool and the m utation operator developed in this study to particularly suit three dimensional polynomial surface fitting. Factors th a t affect convergence such as population pool size, m utation probability, and m utation severity are examined in order to determ ine the optim al param eters which are later used in the genetic algorithm. The results of polynomial surface fitting are then presented for each mode. The second mode F R F was not well suited for representation as a single polynomial surface, but this problem was overcome by dividing the examined area of the fin into four neighbouring subregions, each of which provided good representations. A second genetic algorithm th at was used to determ ine the optim al position of the actuator is then discussed in detail. Finally, the optim um piezoelectric ac tu a to r position is determ ined for several vibration control schemes. C hapter 4 describes th e optim ization of th e configuration of up to 48 individual piezoelectric actuators on the fin by direct m axim ization of the FR F for either an individual mode or m ulti-m odal acceleration or displacement control. The derivation of the fitness function is examined for all vibration control schemes. The different types of optim ization possible in term s of trade-off studies, actuation maximization, and m inim ization of num ber of actuators through application of the methodology are highlighted. The genetic algorithm used for optim ization of the actuator configuration is discussed in detail, w ith emphasis on reproduction and m utation, including forced m utation, a unique feature of the algorithm. The FR F results for the first three modes are presented, and key measures of optim ization

18 3 effectiveness, namely actuation authority, minimum effectiveness (ME), and relative mean effectiveness (RME) are defined. The optim ized configurations for individual m odal control and m ulti-m odal acceleration and displacem ent control are presented, and an example of a trade-off study between actuator weight and actuation authority is presented. C hapter 5 presents a verification of the finite element simulation results by comparison to optim ization results obtained from an experim ental model of the fin. The experim ental results are presented along with a m odal scaling used to correlate the results to the simulation results. Comparisons of the experim ental results with sim ulation results obtained from the finite element model of the fin in experim ental, single actuator position optim ization, and m ultiple actuator optim ization configurations are conducted. Chapter 6 emphasizes the m ajor contributions to advancing the state of knowledge in the field and conclusions th at can be inferred from the research, and provides recom m ations for future directions of research. 1.2 M o tiv a tio n This research was conducted at the N ational Research Council of C anada and Carleton U niversity in order to com plem ent a p roject to develop an adaptive control system to actively reduce buffet induced tail vibrations as part of a m ultinational research project. An active control system has been developed for the flexible fin presented in this thesis in the Aeroacoustics and Structural Dynamics, Structures, M aterials, and Propulsion (SDSMP) laboratory of the Institute for Aerospace Research (IAR) at the O ttaw a branch of the National Research of C anada [3]. This control system was developed using evenly distributed pairs of piezoelectric actuators such as in the experim ental configuration presented in Chapter 5 in this thesis. The efficiency of such a control system can be vastly improved by instead optimizing the actuator configuration. The purpose of the research described in this thesis is to present a novel approach to such an optim ization and investigate the possible problems associated w ith th e methodology.

19 1.2.1 A c tiv e T ail B u ffet S u p p ressio n P r o je c t Buffet induced tail vibrations such as those shown in Figure 1.1 occur on high performance twin-tail aircraft such as the F/A -18 and F-15 when unsteady pressures associated with separated flow, or vortices, excite the vibration modes of the vertical tail fin structural assemblies [1]. This is a significant problem, particularly for the F/A -18 aircraft operated by Canada, Australia, and the U nited States, which requires frequent inspection to prevent catastrophic failure caused by crack growth. There have been numerous studies on using an active control system in order to reduce the buffet loads on both scaled models and actual aircraft [5] [6] [7]. An alternative m ethod of vibration control to piezoelectric actuators is to use active rudder control [8]. Results of studies using piezoelectric actuators for vibration control have shown th at a 60% reduction of first mode bing am plitude can be acheived, resulting in a significant extension of the lifespan of a tail fin. O ther studies have been conducted by Boeing, NASA Langley, the Air Force Research Laboratory (AFRL), and Daim ler Chrysler, to highlight a few [9] [10]. Figure 1.1: Leading Edge Vortices of an F/A -18 at a High Angle of A ttack [1]

20 P roject O verview A joint project between Canada, Australia, and the United States was initiated in 1997 under T he Technical C ooperation P rogram (T T C P) to determ ine th e suitability of an active control system to reduce the buffet induced tail vibrations on a full-scale F/A -18. In C anada, this involved th e C anadian D epartm ent of N ational Defence (DND) and th e National Research Council of C anada (NRC). The Canadian contribution was the development of an active control system using piezoelectric actuators [2], The control scheme was able to provide a reduction of up to 60% in buffet vibration am plitude in the best case, and of 10% in the worst case [11], If incorporated into the aircraft, this control system would double the durability of the fin. Figure 1.2 shows the full scale test of the control system w ith bonded piezoelectric actuators. Figure 1.2: F/A -18 V ertical Tail w ith Bonded Piezoelectric A ctuators [2] System Identification A prim ary challenge in the development of an active control system is system identification, or the extraction of a dynamic model for a flexible structure with piezoelectric transducers.

21 Using a reduced m odel experim ental flexible fin, a theoretical m ethod of system identification using finite element modelling has been developed at the SDSMP laboratory at the NRC of C anada to model a dynamic system and to obtain its Frequency Response Function (FRF) [4]. A fi synthesis technique is then used to m atch a transfer function to the FR F plot. Experim ental system identification was also carried out using the experim ental test bed described in C hapter 5 of this thesis. A state space model produced the best fit to the experim ental data, and resulted in good agreement between the two m ethods. This was an essential part of the development of a control system for the flexible fin. Control System A feedback control scheme for the experim ental configuration of the flexible fin presented in this thesis using 12 pairs of actuators was developed at the SDSMP laboratory [3]. Using the fj, synthesis technique, the transfer function of the system was derived from the FR F data. Based on this model, an active control system was implem ented by using the Real- Time W orkshop and xpc TargetBox of M athw orks. Results dem onstrated th a t active control systems could reduce the vibration significantly both in narrow band and broadband frequency ranges, as shown in Figure 1.3. Control on Control on Time [sec] Time [sec] (a) Sim ulation (b) Experim ent Figure 1.3: F irst M ode Dwell R eduction for Sim ulation (Left) and E xperim ent (R ight) [3]

22 1.2.2 V ib ra tio n C o n tro l o f In d iv id u a l M o d es In many operations, buffeting is only able to excite a particular vibrational mode of the vertical tail fin. In this case, it is efficient to design a control scheme to reduce the vibration of th at mode, and to place actuators where they are best suited to control th a t mode. This results in the best possible position of actuators and most effective control system, but if unexpected frequencies are encountered, it may not be able to adequately control the resulting vibration. Alternatively, it is possible to distribute actuators over the fin. This would allow some reduction of any vibrations th a t could possibly be encountered, however efficiency would suffer, som etim es drastically, deping on th e vibration m ode V ib r a tio n C o n trol o f M u ltip le M o d es If it is not possible to identify a single mode of concern, but it is possible to identify a frequency range th a t is likely to be encountered in operation, it is possible to optim ize an actuator configuration for m ulti-m odal control. This entails identifying the corresponding vibration modes, and tailoring a system specifically to control either the acceleration or displacement (or a combination of each) for those modes. This results in a more versatile system, with a corresponding loss of actuation efficiency. It is, however, still much more efficient th an a random or even distribution of actuators. It is inevitable th a t the wider the range of conditions a system is optim ized for, the less effective the system is for each individual condition. Thus an understanding of the likely operational conditions for the system is critical A c c e le r a tio n v s. D isp la cem en t C ontrol This research used both acceleration and displacement, m easured w ith accelerometers, as indepent bases for optim ization of actuator configuration. Either can be im portant factors for vibration of the flexible fin. Buffet induced tail vibration is usually associated with activation of the first and second mode, and crack growth rate is prim arily dictated by first m ode vibrational displacem ent. However, th e second and th ird m odes are w ithin

23 the excitation range th a t is likely to be encountered in operation and are quite significant, particularly when considering acceleration. 1.3 B ackground The research presented in this thesis falls into several fields, including active m aterials, piezoelectric actuators, flexible active structures, genetic algorithms, and the com bination of a genetic algorithm with finite element m ethods A c tiv e M a teria ls Active m aterials incorporate physical properties th at can be controlled by external stimulus, allowing them to be used in control systems where the m aterial properties are selectively adjusted to achieve a specific purpose. Properties such as viscosity (in the case of m agneto and electro-rheological fluids) or strains (in the case of m agnetostrictive and electrostrictive m aterials or shape memory alloys) are typically controlled by electrical, magnetic, or tem perature stimulus. Stimuli th at can be rapidly adjusted such as an electric or m agnetic field are particularly relevant, since they can be used in real tim e control systems P ie z o e le c tr ic A c tu a to r s Piezoelectrics are the m ost common form of electrostrictive m aterials th a t can generate an electric field when they are strained. As a result of this direct piezoelectric effect they can be used as sensors detecting straining of a host structure. Conversely, when subjected to an applied electric field, employing their converse piezoelectric effect, these m aterials can be used as actuators to strain the host structure. As discussed later on in this thesis, both direct and converse piezoelectric effects are considered to be linearly proportional in the present study, i.e., the induced strain is assumed to be linearly proportional to the applied electric field intensity, changing signs as the sign of the electric field changes. The piezoelectric effect has been observed in natural quartz crystals, semicrystalline polyvinylidene, some ceramics, and hum an bone [12], Either lead zirconium titan ate (PZT) ceramics or polyvinyl fluorides

24 (PVDF) are typically used for piezoelectric actuators. The actuators used in this research were pairs of thin PZT ceramic plates th a t could be easily bonded to the surface of the flexible fin. However, in the finite element model, a therm al analogy was employed to simulate the piezoelectric effect. Piezoelectric ceramics have high actuation response in term s of acceleration or displacement per volt due to the high Young s m odulus inherent in the m aterial, being able to develop strains on the order of 0.1% [13]. A high bandw idth (on the order of 100 khz) and a large tem perature range (-20 to 200) allows them to be used in a wide range of dynamic control applications [13]. The principle drawbacks are a high density and brittleness [12], Poly crystalline m aterials composed of piezoelectric crystals are m ade piezoelectric by a process called poling. For ceramics, this involves heating the m aterial to just below its Curie tem perature and then subjecting the m aterial to a strong electric field to align the electric dipoles in the ceramic with the direction of the applied field, commonly referred to as the poling direction. The m aterial is then cooled, locking the dipoles in their aligned orientation. Once the electric field is removed, the ceramic is left with a net dipole moment in the poling direction th at is now considered to be the prim ary direction th a t can be strained when subjected to an electric field. An opposite electric field results in opposite strain. A limit is imposed on the strain th a t can be developed because too high an applied electric field will cause the m aterial to repole. The prim ary direction is always in the same direction as one of the common axes of each crystal. The remaining two crystal axes rem ain oriented random ly about the aligned axis, thus m aking poled ceramics transversely isotropic m aterials with respect to the poling direction A d a p tiv e S tru ctu res In general, an adaptive structure consists of a host structure with a system of sensors and actuators directed by a controller th at is capable of modifying the response of the combined system in real tim e in a desired fashion to environm ental and operational excitations [14]. In the case of the flexible fin, it refers to the combination of piezoelectric actuators and sensors, and th e control system to control vibration of fin. The sensors m easure th e response

25 10 of the fin and the actuators apply forces and moments to counteract the vibration. The configuration of these sensors and actuators is critical to the ability to reduce vibration for anticipated operational conditions (excitation frequencies) G e n e tic A lg o rith m s A genetic algorithm is an optim ization technique inspired by evolution in nature, whereby a D arw inian survivahof-the-fittest approach is com bined w ith an efficient inform ation exchange system [15]. In nature, performance is gauged by the ability of an organism to survive; in a genetic algorithm, performance is m easured by a problem specific user defined fitness function [16]. A genetic algorithm incorporates characteristics of both random and directed search techniques. A large initial population of tria l solutions is random ly generated and the best solutions, subject to a fitness function, are kept while the others are discarded. The type of genetic algorithm used deps on the problem. In m any genetic algorithm s binary coding is used, where the individuals (trial solutions) are represented as strings consisting of 0 and 1 bits (alleles) [17]. O ther genetic algorithm s use continuous variables to represent individuals. A genetic algorithm is able to locate near-optim um solutions to large and difficult search problem s quite rapidly. Unlike m any conventional optim ization techniques, m any points in the search space are considered sim ultaneously and random choice is used to guide the search [15]. A genetic algorithm is especially applicable to surface fitting because it does not require any prior information about the function (such as its derivatives), and, given a sufficiently large initial population and num ber of generations for convergence, will always find the best solution. For example, increasing the initial population size reduces the chance of getting stuck at local minima. W hile the population is initially random ly chosen, chance is used efficiently to converge to the final solution. In all genetic algorithms, the operators of selection, reproduction, and m utation are applied sequentially on an initial random population. Some genetic algorithm s incorporate other features, such as an im m igration operator, or crossover before, after, or instead of reproduction. Reproduction and crossover are quite simple operations, yet they give genetic algorithm s m ost of their searching power. T hey efficiently combine and exchange

26 11 inform ation from the fittest individuals in a generation, ensuring th at the next generation will be at least as fit, if not more fit. This gives genetic algorithm s the ability to converge to a solution over m any generations. Following reproduction, m utation is applied, which enhances the ability of a genetic algorithm to find a near-optim um solution by introducing new alleles into the population. This can allow the final solution to incorporate information th a t was not included in the initial population. Although m utation is critical to introduce new information, it should be noted th at it occurs with a small probability and also th at most of the m utations are detrim ental. The m utations th at do not improve the solution are subsequently removed in th e selection process. A very sm all num ber of beneficial m u tations occur, but these contribute vital inform ation to the final solution. The fact th at most m utations are detrim ental means th a t too high a m utation rate will slow convergence by detrim entally changing the fittest individuals of the population too often; however, too low a m utation ra te will also slow convergence by not introducing sufficient m utations C o m b in a tio n o f F in ite E le m en t M e th o d s w ith G e n e tic A lg o rith m s O ptim ization of the placement or grouping of piezoelectric actuators and sensors using genetic algorithm s has been examined in the literature but these studies have consistently used an analytical solution in order to derive the fitness (cost) function for optim ization. For exam ple, W ang, Quek, and Ang [18] used a genetic algorithm com bined w ith an analytical approach to optimize the positions of actuators and sensors on flexible composite plates. Kwak, Heo, and Han [19] proposed a technique for dynamic modelling a rectangular plate w ith arbitrarily oriented piezoelectric sensors and actuators based on the Rayleigh- Ritz m ethod. Although a genetic algorithm was subsequently employed to determ ine the optim al positions of a pair of sensors and a pair of actuators sim ultaneously fixed on the plate, their m ethod was also based on an analytical approach. Sheng and K apania [20] introduced th e com bination of a finite elem ent m odel (FEM) together w ith a genetic algorithm solver, although an analytic form ulation was still employed. Lopes, Steffen, and Inm an [21] again used an analytical approach in order to optim ize th e position of a c tu a tors and sensors. T he m ost critical (and often m ost challenging) feature of optim ization,

27 12 particularly with a genetic algorithm, is the derivation of a fitness function. A consistent lim itation of existing approaches is th at they require the derivation of a closed-form fitness function for optim ization, thus reducing and lim iting the complexity of problems th a t can be solved. In contrast, the advantage of the optim ization approach presented in this thesis lies in the simplicity of the fitness function and the applicability of the technique to any configuration of sensors and actuators on a structure with basically any degree of geometric complexity. For this research, a finite element model of the fin was used to develop the fitness function from the frequency response function (FRF). A genetic algorithm was then used to optimize the configuration of a selected num ber of piezoelectric actuators or the position of a single actuator on the fin. The decoupling of d ata acquisition from the optim ization process, elim inating the need for derivation of a closed-form solution, is the m ain advantage of the present technique in comparison to previous approaches th a t are reported in the existing literature. It can therefore be employed for complex geometries where analytical solutions cannot be obtained, or where derivation of such a solution would be prohibitively difficult. This m ethod is somewhat reminiscent of th a t employed by Cook and Crossley [22], who optim ized the position of sm art actuator placement for aircraft maneuvering using a genetic algorithm. However, a m ajor difference lies in the derivation of the fitness function; in th at study PM ARC (a fluid dynamics software package) was used to determ ine fitness of the population at each generation as the algorithm converged, instead of initially specifying a num erical description of th e population fitness. 1.4 O b jectives There were three m ain objectives of the research described in this thesis: the optim ization of a single pair of piezoelectric actuators on the fin using a continuous fitness function, the optim ization of the configuration of the fin with multiple actuators, and verification of the sim ulation results by comparison w ith results obtained from an experim ental model of the fin. To accom plish th e first objective, a genetic algorithm was used for polynom ial surface

28 13 fitting of the FR F in order to obtain a continuous fitness function. To accomplish the second objective, the fitness function was derived directly from the FR F m easured from the finite element model of the fin which had 48 possible pairs of actuators, completely covering the examined area w ithout overlap. The final objective was accomplished by the comparison of an optim ization where the fitness function was determ ined using the finite element model w ith an optim ization where the fitness function was obtained by direct m easurem ent of the FR Fs from an experim ental model of the fin with 12 pairs of actuators. One of the m ain advantage of the m ethod presented in this thesis is th a t it can be exted to practically any geometry or sensor and actuator configuration. Provided a structure can be modelled using finite element m ethods, the configuration of sensors and actuators on it can be optimized. This is especially useful for complex structures for which th e derivation of a closed form fitness function is im possible or prohibitively complex.

29 C hapter 2 F lexible Fin O verview and D ata E xtraction The integration of actuators and sensors with the host structure often makes it prohibitively complex, if not impossible, to develop a m athem atical model for a complex sm art structure. It is often possible to adopt one of two alternative approaches. The structure can in some cases be reduced to an equivalent experim ental apparatus on which actual actuators and sensors can be applied. M odal and other m ethods of vibration analysis can be applied to this model, and the results can be scaled to correspond to the actual structure. A control system can then be developed and tested on the experim ental apparatus. A much simpler and usually reasonably accurate approach is to use a finite element model of the structure to extract vibrational information and develop a suitable control system. In this thesis, both the experim ental and finite element approaches are adopted to investigate the problem of optimizing the configuration of piezoelectric actuators on a scaled model of a flexible aircraft fin. 2.1 F in ite E lem en t M eth o d s O verview Finite element m ethods are often used to model and predict the dynamic response of a structure w ith integrated sensors and actuators. An excellent early example of this is the study by Rahm oune et. al [23], where the electromechanical coupling effect of piezoelectric m aterials was used to establish a finite element model of a flexible plate with bonded piezoelectric sensors and actu ato rs. A t th e N ational Research Council of C anada, a finite 14

30 15 element model of the sm art fin discussed in this thesis was used as one of the bases for the development of the active control for the fin [4]. 2.2 F in C onfigurations The same flexible fin model was used for all parts of this research, although the actuator configurations were modified to accomplish the objectives described in C hapter 1. The same basic finite element model of the fin, th a t represented the experim ental apparatus, was used for all optim izations. The experim ental configuration will therefore be described first, although the results it yielded are discussed in C hapter 5 while results from the finite element model optim izations are discussed in C hapters 3 and 4. The general shapes of the first three modes are shown in Figure 2.1. (a) F irst Mode (b) Second Mode (c) T hird Mode Figure 2.1: T he F in M ode Shapes E x p erim en ta l S et-u p v s. F in ite E le m en t M o d el The experim ental model of the flexible fin th at was developed at the ASDSMP laboratory of the N ational Research Council of C anada was used for the purposes of this study. It consisted of a flexible aluminum fin of 1 mm thickness cantilevered at the fin root between a pair of solid alum inum blocks. Twelve lead zirconium titanate (PZT) thin ceramic plate actuators (BM500 from SensorTech Systems, Inc.) were bonded to each side of the fin [4].

31 16 A schematic of the model is shown in Figure 2.2. The m aterial properties of the components and the natural frequencies and dam ping ratios of the integrated structure are discussed in C hapter 5. A finite element model of the integrated fin with accelerometers and actuators positioned in the same locations as in the experim ental apparatus was used for comparison of experim ental and sim ulation results. This equivalent finite element model is presented in Figure Accelerometer Piezoelectric Actuator * Figure 2.2: Geom etry of the Sm art Fin in Experim ental Configuration A ccelero m eters For both the experim ental and finite element models, five accelerometers were used to m easure the dynamic response of the integrated sm art structure. These were positioned so

32 17 Figure 2.3: Finite Element Model of the Sm art Fin in Experim ental Configuration [4] th a t they could m easure relatively high responses from the first three dynamic modes of the sm art fin. The positions of these accelerometers are shown in Figure 2.4, and are listed in Table 2.1. All coordinates refer to th e centroid of each accelerom eter. Table 2.1: X and Y Accelerometer Positions on the Sm art Fin along with Corresponding Finite Elem ent Node Accelerometer X(mm) Y(mm) Finite Element Node Num ber The accelerometer positions were chosen in order to be well suited to measure the response of all modes, specifically, accelerometer 1 was positioned at the fin tip trailing edge, as far as possible from the node lines associated with the lower fundam ental modes. As such, it was well positioned to record a substantial response when the fin was excited at all

33 18 *2 Figure 2.4: Accelerometer Positions on the Sm art Fin three of the first fundam ental frequencies, and the response m easured by it was used for the subsequent optim ization of actuator position for each mode. The other accelerometers were prim arily used for reference and verification of data, and no optim ization was performed using the data they recorded. Accelerometer 2 was positioned midway from root to tip at the leading edge, and as such, it was well positioned to record specifically the response of the fin when excited at the second fundam ental frequency. Similarly, accelerometers 3, 4, and 5 were well positioned to record the response at the first and third, second, and second and fourth fundam ental frequencies, respectively. In general, a complex frequency response function contains both the m agnitude and phase of a steady state response [24], For the purposes of this work, since the m agnitude was the prim ary concern, the F R F m agnitude was taken as the acceleration recorded at accelerometer 1 divided by the input voltage in order to norm alized th e response F in ite E le m en t M o d el for A c tu a to r C o n fig u ra tio n O p tim iz a tio n For both the single and multiple actuator configuration optim ization, the same basic finite element model with 48 possible actuator locations (6 rows of 8 actuators) was used, Figure

34 In both cases, and for each vibration mode, the resulting F R F at accelerom eter 1 was m easured once for each of the 48 possible actuator positions. However, in the case of the single piezoelectric actuator optim ization problem, a single pair of actuators was bonded sequentially at each of the 48 different positions, whereas in the case of the actuator configuration optim ization problem, each actuator was sequentially activated from the 48 bonded actuators th a t entirely covered the pre-determ ined region for all runs. Actuator Test Area 200mm (X) x 150 mm (Y) Single Actuator Figure 2.5: The 48 A ctuator Positions from the Finite Element Model of the Sm art Fin 2.3 F in ite E lem en t M eth o d D a ta A cq u isition A basic FEM model of the integrated sm art fin th at was available at the ASDSM P laboratory of the National Research Council was used in this study. This model was developed using th e MSC P a tra n software package and MSC N astran was employed as th e finite element model frequency response solver. A therm al load analogy was used to sim ulate the piezoelectric effect because it is not represented in P atran as is the case with m ost finite element software packages. The following specific assum ptions are incorporated into the FEM model: - a perfect bond between the host structure and the actuators, - negligible change of m ass and stiffness due to th e bonding agent, and

35 20 - incorporation into the model of a general structural dam ping ratio of 3.4% based on experim ental results. This dam ping ratio was determ ined by trial and error by equalizing the first mode response from the experim ental and FEM models. This assum ption led to a requirem ent to scale the results for the second and third modes in C hapter F E M E le m en ts an d C onfig u ra tio n Shell elements and solid elements were employed in P atran for the alum inum plate and actuators, respectively. The actuator solid elements were connected at their base to nodes on the alum inum plate shell elements. Each actuator was modelled as 25 solid 5 mm x 5 mm x 1 m m HEX20 (20 nodes) elem ents in PATRAN; an exam ple is shown in Figure 2.6. The alum inum fin was modelled within the actuator test area as 5 mm x 5 mm square QUAD8 shell elements of 1 mm thickness, each having nodes at their corners as well as mid-side nodes, for a total of 8 nodes per shell element. Outside the actuator test area, the elements varied in size and shape, as shown in Figue 2.5. It was critical th a t the face of each solid element was consistent in size and shape with the shell elements inside the actuator test area. However, the size and shape of the elements outside this area was unim portant and the resulting variation in elem ent geom etry was necessary to create th e fin shape T h erm a l A n a lo g y Since only a few commercial software packages such as ANSYS are capable of piezoelectric modelling, a therm al analogy was employed in N astran to model the piezoelectric actuators. For low frequency ranges and low applied voltages, the induced stra in is given by: e = y A F, (2.1) where d^i is the piezoelectric charge constant for the prim ary poled direction. A V and t represent the applied voltage, and thickness of the piezoelectric actuators, respectively. Although there are strains in the perpicular direction, the focus in this work is on strain