You are on page 1of 178
VIBRATIONS AND DYNAMIC STABILITY OF ROTOR BLADES by Victor Giurgiutiu, B.Se.(Eng.), ACGI. A thesis submitted for the degree of Doctor of Philosophy of the University of London and for the Diploma of Menbership of Imperial College. Department of Aeronautics, Imperial College of Science and Technology, ‘London SW7 2BY 1977 -1- ABSTRACT ‘The vibration and dynamic stability of rotating beams is ana~ lysed in a logically related manner, and novel methods of analysis are introduced to control and improve the accuracy of the numerical tools. The equations for free small vibrations of a uniform rotating beam are derived from the Newtonian principles and Timoshenko's theory of bending. The exact solution is obtained with the Frobenius method, and rotating frequencies and mode shapes are obtained. The analysis is extended, through a transfer matrix method, to nonuniform beams, and several practical configurations are studied. The nonlinear equations of motion required in stability analysis are derived from variational principles for uniform blades with root torsion and precone. The equations are completed by adding the noncon- servative aerodynamic forces for hover. Liapunov's theory of stability is employed, and the resulting equilibrium and perturbations equations are solved separately. A numerical shooting method is used to obtain accurately the nonlinear equilibrium solution which is used to obtain the coefficients of the perturbations equations. The Galerkin method is used to solve the perturbations equations, and the classic problens associated with selection of trial functions are overcome by the inclu- sion of boundary conditions in the Galerkin integrals, and the adoption of rotating mode shapes. The result is a low order complex eigenvalue problem whose roots yield Nyquist diagrams for determination of stability. Comparisons with published results illustrate the good convergence pro- perties of the method. The computer programs developed during this research are available on request. ACKNOWLEDGEMENTS ‘he author wishes to thank those many individuals without whom this research would not have been concluded, The financial support of the Dowty Group Ltd, and the help of its technical and managerial staff are gratefully acknowledged. Technical advice was also drawn from Westland Helicopters Ltd, and their kind assistance was highly valuable. The paternal advice of several well-known scientists helped on various occasions to put things on the right course, and author's gratefulness is especially directed towards Mr Adam Sobey, of Royal Aircraft Establishment, Farnborough. ‘The Aeronautics Department of Imperial College created the ideal environment for the research, The friendly cooperation of all the departmental staff, and the parental care of Dr Glyn Davies and Miss Miriam Pook cannot be over-praised. Consultation with members of the Mathematics Department helped on several occasions to overcome the problems of numerical analysis. ‘The close and scrutinous guidance performed with infinite patience and tact by Dr Roger Stafford is beyond praising in words. One can only say that the indestructible friendship forged over the three-year period is the human counterpart of the research results presented in this thesis. CHAPTER 1 CHAPTER 2 CHAPTER 3 CONTENTS INTRODUCTION 1.1 Modern Configurations in Rotor Systems Design 1.1.1 the articulated Rotor 1.1.2 the Hingeless Rotor 1.2 Discussion of Previous Research 1.2.1 The Rigid Blade Nodel 1.2.2 The Flexible Blade Model 1.3 Scope of Present Work GENERAL CONSIDERATIONS 2.1 Coordinate Systems and Transformations 2.2 Ordering Scheme 2.3 Second Order Strain Displacement Relations LINEAR VIBRATIONS 3.1 Derivation of Free Vibrations Equations by Newton's Second Law 3.1.1 Bending Vibrations 3.1.2 Torsion Vibrations 3.2 Solution by Frobenius Method 3.2.1 The Solution for Bending 3.2.2 The Solution for Torsion 3.3. Programming Considerations 3.3.1 transfer-watrix Method for Bending 3.3.2 Transfer-Natrix Method for Torsion 3.3.3 Parameter Tests 3.4 Applications to Practical Rotor Blades 12 18 26 30 35 39 42. 53 56 68. CHAPTER 4 NONLINEAR DYNAMIC STABILITY CHAPTER 5 CHAPTER 6 4.1 Derivation of Nonlinear Equations by Hamilton's Principle 4.1.1 Strain Energy Contributions 4.1.2 Kinetic Energy Contributions 4.1.3 Generalized Nonconservative Forces 4.1.4 Sunmazy of Partial Differential Equations of Motion and Boundary Conditions 4.2 Aerodynamic Loading 4.2.1 Quasi-Steady Strip theory 4.2.2. Velocity Components on a Rotor Blade 4.2.3. Final Bxpressions for Aerodynamic Loads 4.3. Stability in the Sense of Liapunov 4.3.1 Nondimensional Equations 4.3.2. the Perturbations Bxpansion 4.4 Selection of Configuration Parameters SOLUTION OF NONLINEAR EQUILIBRIUM EQUATIONS 5.1 The Equivalent First Order System 5.2 The Solution by Shooting Iterations 5.3 Numerical Results 5.4 Comparison with Galerkin Equilibrium 5.5 Generalization of Present Results SOLUTION OF LINEARIZED PERTURBATION EQUATIONS 6.1 Modal Expansions 6.2 Transformation of Perturbation Equations and Boundary Conditions 6.3. The Galerkin Matrices 6.4 The Quadratic Eigenvalue Problem 6.5 Programming Considerations 78 86 91 101 104 106 107 109 M5 116 CHAPTER 7 APPENDIX A APPENDIX B APPENDIX C APPENDIX D REFERENCES 6.6 Mumerical Results and Stability Plots 6.6.1 Torsionally Rigid Blade 6.6.2 Complete Lag-Flap-Pitch Freedon 6.6.3 Influence of Precone 6.6.4 Influence of Root Torsion 6.6.5 A Simplified Analysis 6.6.6 The Pitch-Lag and Pitch-Flap Coupling Coefficients DISCUSSION AND CONCLUSIONS Approximate Formulae for Frequencies of Rotating Uniform Beams Properties, Frequencies and Mode Shapes of Blade 2 Shooting Methods for Two-Point Boundary-Value Problems C.1 Initial-Value Problems and Boundary-Value Problems C.2 Numerical Methods for Initial-Value Problems .3. Newton-Raphson Iteration in n-Dimensions C.4 The Numerical Solution of B.V.P. by Newton~ Raphson Iteration C.5 Parallel Shooting ©.6 A General Algorithm ‘The Derivation of Equivalent Pitch-Lag and Pitch Flap Coefficients 127 136 140 141 160 im 173 2898, ay? a, ‘b(T) ,b(U) B BO) c(k) NOTATION blade structure cross section area, m* coefficient in the power series expansions airfoil lift curve slope, 2 per radian acceleration vector acceleration components number of blades kinetic and strain energy boundary terms reference frame which rotates with speed @ with respect to the stationary inertial frame 4x 4 matrix of beam functions Theodorsen's function blade chord, m airfoil profile drag coefficient matrix differential operators for boundary conditions arbitrary constant transfer matrix for the whole beam 469 = & E.T.B, FG) = (4 S} FG) displacement vector in the transfer matrix method Young's modulus, N/m? "Engineer's Theory of Bending’ dimensionless root torsion force vector in the transfer matrix method beam functions for vibrations analysis shear modulus, N/m? identity matrix unit vectors associated with undeformed beam coordinate system unit vectors associated with deformed beam coordinate system, "fixed" unit vectors with in direction of rotation of @ in. root torsion stiffness, Nm/rad reduced frequency, we/2V shear deflection coefficients for lag and flap length of bean seguent structural mass, gyroscopic and stiffness matrix differen tial operators aerodynamic mass, damping and stiffness matrix differential operators structural mass, gyroscopic and stiffness modal matrices z aerodynamic mass, damping and stiffness modal matrices overall modal matrices ‘order of ! point transfer matrix for the i-th junction inertial and aerodynamic forces inertial and aerodynamic moments column vector of perturbation modal generalized coordinates distance along the deformed elastic axis, m position vector of a generic point blade tip radius, m inertial frame transformation matrix relating deformed and undeformed blade coordinate systems velocity components of blade airfoil section with respect to the fluid, parallel to the x', y’ and z' axes, respec~ tively , m/sec displacements of the elastic axis in x,y,z directions, a transfer matrix for the i-th segment portions of v,w due to bending and shear lead-lag bending and flap bending generalized coordinates lag and flap mode shapes induced downwash velocity, m/sec K,Y,Z Ye an XyrFqr2, lil cael ze) = {aE} pe bv, bw, 54 yews Soe? Soe? ex, st fixed coordinate system, m direction of rotating x-axis, m undeformed coordinate system, m deformed coordinate system fixed to blade, m coordinates of a point in the undeformed blade, m coordinates of a point (which was at x,,y,,2, in the undeformed blade) in the deformed blade, m state vector precone angle, rad Lock number, 3p,acR/m for blade with uniform mass distri- bution Kronecker delta non zero values of boundary conditions after substitution of modal expansion column of perturbation variables non zero values of differential equations after substi- tution of modal expansion small parameter of the order of bending slopes engineering strain components blade cross section principle axes coordinates, m blade pitch angle, rad B= x/h 5% o 45, ,0, ve? on? Ox, 465 =10- equivalent kinematic pitch-flap and pitch-lag coupling parameters defined by bending-torsion structural coupling warp funetion dimensionless structural parameters dimensionless complex frequency dimensionless inertial parameters Poisson's ratio _ dimensionless axial coordinate (0 § & <1) dummy variables for integration structural density, kg/n? air density, kg/m? solidity be/nR engineering stress, N/m? elastic torsion deflection , rad torsion generalized coordinates torsion mode shapes dimensionless time, ot fundamental lead-lag, flap, and torsion natural frequencies e+ ct Od ) (gO) Our Od, ue oO -11- rotation vector rotor blade angular velocity, rad/sec afax or /9E 2/or a/ac or fay equilibrium and perturbation components of generalized coordinates and of aerodynamic loads circulatory aerodynamic term noncirculatory aerodynamic term frequency of a nonrotating beam frequency of uncoupled vibrations dimensionless quantity transpose of a square matrix denotes vector Other symbols are defined in the text. -12- CHAPTER ONE INTRODUCTION Ma MODERN CONFIGURATIONS IN ROTOR SYSTEMS DESIGN The history of helicopter development is marked by the continuous struggle of the designers to overcome the problems of unpleasant vibra~ tions and dynamic instabilities inherent in this type of aircraft. The concept of vertical take-off and landing capability achieved by placing the airscrew vertically occupied the minds of aviation pioneers from the beginning,and the first flight trials of helicopters were contemporary with the trials of fixed wing aircraft. These early attempts were most unencouraging, and in the following years the idea of helicopter flight was largely abandoned. ‘The Articulated Rotor ‘The introduction of hub hinges to eliminate the large bending stresses at the blade root and the rolling moment which arises in for- ward flight, made possible the first successful helicopter flights [1] by Cierva in early 1920's. The flapping hinge released the bending stresses at the root and resulted in large "centrifugal relief". Thus the steady configuration of the rotor is a coned position, such that the hinge moments created by the lift loads are balanced by the centrifugal force. A blade which is free to flap experiences large Coriolis moments in the plane of rotation, and a further hinge, called the drag or lag hinge, is provided to relief these moments. Lastly, the blade can be feathered about a third axis, usually parallel to the blade span, to enable the blade pitch angle to be changed. A fully articulated rotor will contain all three hinges (fig. 1.1), and is structurally sound and easy to stress. These features explain its popularity with the helicoper designers, a popularity which has lasted over half a century. However, a helicopter fitted with the simple articulated rotor of fig. 1.1 can be unstable, has unsatisfactory handling charac- teristics, and is subject to experiencing "ground resonance". Hence -13- Fig.1.1 Simple articulated rotor head Fig.1.2 Production articulated rotor head =14- modifications of the rotor have been introduced. A production articu- lated rotor as shown in fig. 1.2 contains several hinge offsets, requires heavy hinges to take the centrifugal axial load, and is fitted with complex dampers to prevent excessive levels of vibration. The bulkiness and complication of such rotors, and the complex servicing and maintenance requirements have led in recent years to major changes in rotor-system design concepts. 1.1.2 The Hingeless Rotor Improvements in blade design and the advancements of helicopter science in the fields of flight mechanics, dynamics and structural analysis, have enabled the designer to reconsider rotors which dispense with one or more hinges. The result is an overall improvement in helicopter performance, reduced manufacturing costs, and a mich simpli~ fied maintenance and inspection procedure. Both the blade and the hub designs were rethought, and in recent years a whole array of new rotor systems have come into existence. In blade design, the introduction of reinforced plastics construction, in replacement of the conventional metal blade, is a major trend. ‘The reinforced plastic materials offer a reduced weight to strength ratio, and thus a composite material blade may have much lower levels of static and dynamic stresses. The fatigue properties of reinforced plastic blades are excellent, allowing for a near infinite life due to very low ratios of working stress to fatigue stress. Other noteworthy advantages are that the rate of crack propagation is extremely slow (potentially 10 times slower than similar light alloy blades [2]); for glass composites progressive damage is characterized by stiffness reduction before failure is encountered; the vulnerability to handling damage and ballistics is lower; and production costs may be significantly reduced. In hub design, two major changes are underway. First, the conventional articulated system is being replaced by simpler configura~ tions, and secondly the metallic materials are begining to be replaced by reinforced plastic materials [3]. Fig. 1.3 shows a uniquely simple rotor hub configuration called "Triflex". Constructed of reinforced plastics, this hub contains no conventional hinge and all the three movements (pitch-change, lag and flap) are provided in a simple "wrist" made of thick glass-fibres embedded in an elastomer beam. However, present tendencies favour a rotor hub system in which the pitch changing Fig. 1.3 The Triflex rotor hub ring Fig. 1.4 Sketch of elastically uncoupled rotorshub Pitch change bearing Fig.1.5 Sketch of elastically coupled rotor hub -16- bearing is retained explicitly, whereas the lag and flap hinges are eliminated. Two concepts are possible, (a) elastically uncoupled, and (b) elastically coupled, rotor hubs. An elastically uncoupled rotor hub (fig. 1.4) permits a high degree of lag and flap elastic deformation in the hub region; thus it retains the beneficial properties of the articulated hub, e.g. blade root bending-stress relief. An elastically coupled rotor hub (fig. 1.5) has a very stiff hub, and very little root deformation is accomodated. The structural design of such a hub is more demanding, although it can draw on the experience of propeller designs. The practical realization of these two hub concepts are shown comparatively in figures 1.6, 1.7, 1.8. The MIR system (used in the 4 bladed MBB BO-105) is @ typical fully coupled rotor hub designed to carry glass reinforced plastic blades. The blade root will take and lag and flap movements, and, through pitch changes in the hub, the lag and flap deflections can become coupled. The elastically uncoupled Lynx system shown in fig. 1.7 allows lag deformation in a flexible member, the "dog-bone', situated outside the pitch bearing. The viscous damper which spans the dog-bone reduces the lag vibrations. The flap motion is permitted by the dog-bone itself, and by the hub, which is much more flexible in flap than in lag. The Starflex system shown in fig. 1.8 is an almost all-plastics unit of intriguing simplicity. Centrifugal forces are taken by a laminated rubber cone bearing on the rim of a hole close to the centre of the star-shaped hub. Pitch~change is accommodated in a simple bearing cushioned by elastomeric pads, which also dictate the lag frequency. Nothing needs lubrication, and the whole assembly can be inspected visually. Other rotor systems similar to those described above are also being considered, notably in the American helicopter industry. Compared with an articulated rotor, the hingeless rotors have several advantages. The mechanical simplification achieved by the elimination of hinges, dampers and blade stops, and the overall reduction in the number of component parts, lead to lower production and mainte~ nance costs and improve handling qualities. Also, the reduced profile gives some drag reduction. The large moments that can be transmitted to the rotor hub provide a higher control power which can exceed by a factor of four the control power of an articulated rotor [4]. The stability of the hingeless helicopter is improved in hover, but is deteriorated at high forward speeds. In view of these new rotor system ROTOR STAR Fig.1.6 The MBB BO-105 rotor head BLADE SLEEVE exTENSION SLEEVE neseavoR suse suseve Fig.1.7 The Lynx rotor head Fig.1.8 The Starflex rotor head ~19- configurations, new challenges are raised for the analyst, notably in the fields of rotor vibrations and stability. 1.2 DISCUSSION OF PREVIOUS RESEARCH The dynamics of hingeless helicopters generally involves coupling between the fuselage and rotor degrees of freedom (and possibly stability augmentation systems as well), but this is not always the case. Tests have shown that pure rotor blade instabilities can be identified even when complete hub freedom (fuselage motion) is present. Thus the vibrations and the stability of a single rotating blade with a fixed hub, is important for several reasons. In the first place, the dynamic behaviour of a single blade is the basic building block for the more complex problems, such as the mlti-blade rotor, or the complete rotor- fuselage system. For instabilities which depend on rotor fuselage coupling, such as air and ground resonance, a knowledge of pure rotor blade dynamics is the starting point. Finally, methods for improving the rotor blade stability are also likely to prove effective in reducing the succeptibility to coupled rotor-fuselage instabilities. The dynamic analysis of a single blade has developed alongside new design concepts, and at the present several methods of analysis are coexisting. A review of these methods,done in conceptual rather than historical order, will be attempted below. ‘The classical aeroelastic analysis of a hinged rotor system considered a blade rigid articulated near the hub. A set of at most 3 nonlinear differential equations in time are obtained by considering the lag, flap and pitch angular movements [5]. Simple strip aero- dynamics based on Theordorsen's theory [6], and approximate inflow based on blade element momentum theory, produce aerodynamic loads which form the right-hand side of these equations [1]. This simple model proved satisfactory for the analysis of most dynamics problems encountered with hinged helicopters, such as stability, vibrations and ground resonance [5]. The necessity of including the blade elasticity into the analysis, which become apparent for flutter calculations, was overcome with the inclusion of equivalent hinge springs. 1.2.1 the Rigid Blade Model Early work on hingeless rotors followed the methods for arti- -19- culated rotors. A rigid blade was considered hinged at the root where several springs were attached to simulate the blade flexibility. Flap, lag-flap, and lag-flep-torsion displacements can be modelled with one, two and three hinges, respectively. Such an approximation reduces the analysis of hingeless rotors from one of partial differential equations to one involving only ordinary differential equations. The solutions then become simpler to obtain, and the results are somewhat easier to interpret in physical terms, The flap-lag stability studies by Young [7] and Hohenemser [8] revealed some of the complexities of nonlinear aerodynamic and inertial coupling. Ormiston and Hodges [9] extended the double-hinge model by distributing the flexibility between the inboard and the outboard sides of the pitch bearing (fig. 1.9). This permits a variable amount of elastic coupling, and the studies for hover showed that this is important for stability. This analysis was further extended for forward flight by Peters [10]. Recently, Reichert [11] reports using a complete rigid blade model (fig. 1.10) which incorporates all three degrees of freedom. ALL such analyses of rigid rotors yield sets of one, two, or three nonlinear, time dependent ordinary differential equations. Often, expedient crude approximations were used to simplify the equations and obtain a solution [7], but the relevance of such solutions to design is doubtful. The nonlinear equations can of course be solved by direct time integration [8], [12], [13]. In principle the stability can be judged from the time history traces. However, the large computational, requirements prohibit extensive parameter studies, and often the transient response is obscured by the forced response [10] which can drift consi-~ derably during one or two rotor revolutions. One analytic approach to the problem of nonlinear oscillations is the method, of perturbations [14]. This method proceeds in several steps that can be identified in Tong's [15] analysis of flap-lag stabi- lity in hover. The perturbed motion about an assumed equilibrium position (static equilibrium in [15]) is investigated. First the linear stability boundaries are derived by solving the equations of the first approximation. Next the complete response is derived with asymptotic perturbations expansions of the dependent variables. Multiple time scales are used to eliminate the secular terms. Thus stable, or unstable, limit cycles are identified, showing that stable, limited non~ linear response can take place in regions of linear instability. Tong's analysis is open to improvement and extensions, but to date no attempt =20- Fig.1.9 Arrangement of flap and lead-lag springs of rotor blade and hub for simulating variable elastic coupling Blade torsion Lag Flap Control system flexibility 2 SP Fig.1.10 Rigid model of a rotor blade -flap, lag, and torsion angles -21- has been made to continue it by using the exact nonlinear equilibrium position, or to extend it to lag-flap-torsion analysis. The Linear stability boundary (via Nyquist diagrams) is as far as most other analy ses would go due to the complexity of the nonlinear equations which systematically include many structural, configurational, and aerodynamical parameters [9], [10]. 2.2 The Flexible Blade Model The adoption of a flexible blade model [16] constituted a major advance in stability analysis. This gives a good representation of blade curvatures, but with large increase in mathematical complexity, since a solution has to be found to two or more, linear or nonlinear, integro-partial differential equations describing the blade motion. Actual rotor blades have a complex structure consisting of several segments of different and possibly nonuniform (e.g. tapered) properties (fig. 1.11). They present several configurational features that may be relevant to design studies, and precone, droop, torque offset and sweep are shown in fig. 1.12, The blade cross section itself can present a complicated configuration, and various other offsets (mass, tension and aerodynamic) from the elastic centre can also be present. The linear differential equations of coupled lag-flap-torsion motion, derived by Houbolt and Brooks [17] include most of the fore- mentioned parameters. The Euler theory for beams, and the St. Venant theory for torsion were used in the analysis of a preconed, pretwisted, nonuniform blade of symmetric cross section, with mass and tension offset, and under unspecified aerodynamic loading. These equations have constituted the basis for most of the present day work on linear (and nonlinear) vibrations and stability of hingeless rotors. By assuming zero loading, the Houbolt and Brooks equations are made to describe the free linear vibrations of an arbitrary rotor blade. Various methods for solution are available. By applying the Galerkin Method with standard nonrotating mode shapes, Yntema [18] derived the uncoupled frequencies and mode shapes of a rotating uniform cantilever. Whereas in [19] analytical solutions were developed for the same problem, and by combining low stiffness expansions and high stiffness expansions simple empirical closed form expressions were derived for the frequencies of this elementary beam. To tackle nonuniform blades, Myklestad [20] devised a transfer matrix method which was applied to the simplified equations of a lumped mass, segmented beam idealization. Isakson and =22- Fig.1.11 Primary elements of a typical hingeless rotor blade Sie view Fig.1.12 Configuration parameters of a hingeless rotor blade Eisley [21] extended Myklestad's method to derive the coupled frequencies and mode shapes of twisted nonuniform blades. The inclusion of shear deformation and rotary inertia made the transfer matrix method a more precise tool in rotor blade design [11] at higher frequencies. ‘The development of numerical integration techniques for two~ point boundary value problems [22] permitted a direct integration approach to the Houbolt and Brooks equations. Wadsworth and Wilde [23] illustrated the shooting method of solution for frequencies and mode shapes of flap vibrations of uniform and nonuniform beams. Whereas King [24] reported the direct integration of the complete equations for coupled lag-flap-pitch vibrations of actual blades. Finite differences solutions of the coupled equations are also possible, and Hunter [25] applied the integration matrix method to find the natural frequencies of bending vibrations of twisted propeller blades. Up to 40 collocation points were used with integrating matrices of different degrees, This method was extended by White [26] to obtain the frequencies and mode shapes of coupled bending-torsion vibrations of twisted nonuniform rotor blades, using 20 collocation points, The numerical integration tech- niques are best suited for blades with smoothly varying properties, but this is not always the case with production blades. Nonlinear extensions to Houbolt and Brooks equations were first attempted on reduced systems, e.g. lag-flap [16], or lag-flap-root torsion [27]. Several nonlinear structural terms were identified, mainly through second order axial strain effects, but the results were neither complete, nor conclusive. Finally, Hodges and Dowell [28] developed a complete second order extension to the Houbolt and Brooks equations, through second order approximation to the strain tensor, and based on a consistent ordering scheme. Both Hamiltonian and Newtonian derivations were used, since they offer accuracy and insight, respectively. ‘An even more comprehensive set of equations, which apply to unsynmetric blade sections and include first order Timoshenko corrections to Euler's theory, was derived through Hamiltonian methods by Sobey [29]. e fixed wing aerodynamics, the rotor blade aerodynamics cannot be separated from the blade motion, and this renders the problem an order of magnitude more difficult. To calculate the aerodynamic loads it is necessary to know the local components of the airflow at any station along the blade, and this in turn requires a knowledge of the air velocity induced by the lift of the blade. A survey of present day literature on rotor blade loads prediction [30], [31] indicates that in a24= recent years considerable research was invested in the subject. Complicated pressure fields and nonlinear aerodynamics have been consi= dered and new problems were raised, such as airfoil dynamic stall, blade- vortex interaction, tip-vortex induced dynamic stall, transient Mach number effects, yawed and reversed flow. Remarkable results have been reported (Ward and Young, [32]) of solving part of these problems by numerical techniques. But direct time integration requires considerable computing effort and, for reasons already stated, is unsuitable for stability analysis. Hence most investigations into blade stability use one of the available two dimensional linear strip theories reviewed in [33]. Such fixed wing strip theories are normally extended to rotor blade calculations by expressing the local flow velocities in terms of blade motions and of uniform induced velocity (blade element momentum theory [34]). In this way a linear, quasi-steady aerodynamic operator can be derived based on Theodorsen's theory [16]. Inclusion of several second order angle of attack terms due to bending slope can lead to a nonlinear aerodynamic operator [27]. Using Greenberg's [35] correction for pulsating flow, and a rotation matrix consistent to second order, a complete quasi-steady aerodynamic loading was derived for hover [36]. Frequency dependence and compressibility effects can also be considered [33]. The resulting aerodynamic loads form the right-hand side of the nonlinear equations of motion , and hence the required ingredients for a nonlinear stability analysis are specified. The stability equations are complicated sets of integro-partial differential equations in time and space, and their solution poses a challenge to the analyst. One approach is to eliminate the space variable through the application of Galerkin method, and to treat the resulting ordinary time differential equations with an analytic method. Friedman and Tong [37] employed this approach to study the stability in hover and forward flight of a simplified, torsionally rigid blade. By applying the Galerkin method (one mode in lag and flap, respectively) the equations were reduced to two ordinary nonlinear equations in time, similar to the rigid blade equations. The perturbations about an equilibrium configuration were studied using the method of multiple time scales, and linear stability boundaries, and limit cycles were obtained, The Galerkin method usually requires more than one mode for convergence [38], and the size of the transformed problem is usually large. The additional complexities involved in retaining some of the forementioned design parameters has discouraged further development of =25= the analytic solution methods. Directing their efforts towards fully establishing the correct differential equations and aerodynamic loading, subsequent investigations were generally limited to finding the linear stability boundaries. Nyquist diagrams [39] and Lag-damping plots were derived [16] for a torsionally rigid uniform blade via a Galerkin transformation (with several standard cantilever modes) and a linearized perturbation analysis. For the values of the parameters studied, the stability of soft inplane configurations was found to increase with pitch angle, and with (positive or negative) precone. Emphasis was placed upon the importance of nonlinear terms, omission of which was shown to result in an overestimated stability. The same Galerkin transformation was extended to solve the complete nonlinear equations [28], with standard hover aerodynamics [40], and with improved hover aerodynamics and elastic coupling parameter [36]. The mechanism of torsion-bending coupling, which dominates the stability of these blades, was explained in terms of nonlinear structural terms, and equivalent pitch-lag and pitch-flap coupling coefficients were derived. The lag damping of soft inplane configurations at zero precone was found larger than for similar tor~ sionally stiff blades, but a new type of instability was reported at positive angles of precone and at low pitch. The convergence properties of the Galerkin methods were also studied, and ref. [36] reports that accurate results require at least five standard cantilever modes in the modal expansion. Since this would lead to large matrices (3N = 15, size 30 for complex eigenvalues), a method of modal condensation, based on coupled free-vibration eigenvectors, was used. Extensions of this method to analyse the effects of pretwist, droop and pitch-link flexibility (root torsion) were reported without details in [41]. However the Hodges and Ormiston [9] elastic coupling parameter is meaningless for twisted blades. Extensive stability studies were also done by Friedman et al. for torsionally rigid uniform blades, with or without root torsion. Several parameters, notably structural coupling, were at first omitted from the idealization [27]. Numerical results were obtained for hover and for forward flight [42],[43] and the influence of various design parameters was studied [44],[45]. A factor of major importance for stability, especially in forward flight, is the equilibrium position and the effect of various trim conditions was investigated in [46]. Throughout these studies the Galerkin method with uncoupled rotating -26- bending modes (one in each direction) was used (apparently no convergence studies were attempted). For forward flight the Floquet interpretation of Liapunov's stability [39] was employed. Numerical time integration was also used occasionaly to check, or complement, the perturbations stability results. In the most recent studies the aero~ dynamics is refined by considering different strip theories, and using frequency dependant airloads [33]. A nonlinear flutter type analysis is executed for forward flight with a torsionally flexible blade, for which one rotating mode shape is used in each of the three degrees of freedom. However, the use of an approximate equilibrium solution (Linear equili- brium for hover) obscures the validity of the results. Stability of nonuniform rotor blades was treated by White [26]. The dimensionless equations for nonuniform torsionally rigid blades were transformed with the Galerkin method based on the coupled free vibration mode shapes of the same blade. The perturbation method was applied and linear equilibrium solutions, and then the stability plots, were derived in the usual way. Investigation of convergence showed that accurate results could be obtained using only the first flap and first lag modes. 1.3 SCOPE OF PRESENT WORK Because of the way many methods approximate internal and centrifugal forces, as well as geometry, practical vibrations analysis often require a large number of segments or divisions, thus leading to large matrices or extensive computation [11], [25]. Hence the develop- ment of analytic methods for rotating beam vibrations appears to be an attractive alternative. Beam functions will be developed for the free uncoupled lag, flap and pitch vibrations of uniform rotating beans. ‘The differential equations with the Timoshenko corrections [48] for bending will be solved with the Frobenius method [47]. Hence the numerically exact solution for rotating uniform blades will require no more conceptual effort to formulate than the solution for corresponding stationary beams. The assumption of steady harmonic response leads to inary. differential equations for. which a. power series expansion is assumed and recursion formulae are developed. Elementary, albeit lengthy algebra will reduce the general power series to two, or four, explicit polynomials multiplied by constants determined from the boun- dary conditions. These polynomials are treated like any known functions, and for nonrotating beams they reduce to trigonometric and hyperbolic sine and cosine. The transfer matrix method [57] can then be used to -27- treat blades with stepwise nonuniform properties. mforcement of boundary conditions will lead to a 2 x 2 determinant for the frequency equation, and frequencies and mode shapes can be found in the usual way. The main feature of the stability analyses previously discussed [36], [33] was the use of the Galerkin method to eliminate the space dependence prior to application of perturbations expansion. This approach, though simple, has the disadvantage of seeking the stability of a system in which errors have already been introduced through the modal expansion used in the Galerkin transformation, Here the stabi~ lity analysis will be treated in a different way. The nonlinear differential equations are first expanded about the unknown equilibrium position with the perturbation method. The ordinary nonlinear diffe~ rential equations for equilibrium, and the linear partial differential equations for perturbations results from Liapunov's first order inter pretation of stability [14]. Then the equilibrium and perturbation equations will be solved separately. The importance of having the correct equilibrium solution has already been pointed out [46] and a numerically exact solution to the nonlinear equilibrium equations can be obtained through a shooting method [22], Herein this method is applied to uniform blades, but it can readily be extended to nonuniform blades. The convergence of the shooting method will be investigated, and an algorithm will be developed based on an incremental increase of the pitch angle and Newton-Raphson iteration [49]. The availability of a numerically exact equilibrium position permits a precise evaluation of the coefficients in the perturbation equations, and a comparison with coefficients obtained through Galerkin solutions will be illustrated. A nonlinear aerodynamic operator will also be adopted. Through a transformation of axes the local flow velocities can be evaluated, and substitution into Greenberg's [35] strip theory will yield the aerodynamic loads. Applying the ordering scheme, second order aero~ dynamic terms will be retained, consistent with retention of second order elastic terms. perturbation equations. The type of mode shapes used in the modal expansion determines the convergence of the method and two aspects are important, the effect of rotation [19] and the correct inclusion of boundary conditions. Previous researchers have generally used either many nonrotating modes [16], [36], or one rotating mode [33] for each elastic degree of freedom. For existing rotor blades the effect of rotation on bending is much stronger than on torsion [50]. Hence, to ensure rapid convergence, rotating bending mode shapes, and nonrotating torsion mode shapes will be used here. These mode shapes, corresponding to the free vibration mode shapes of the system at zero pitch will be derived for a uniform blade using the methods developed for vibration analysis, The use of coupled mode shapes [26] is not attractive for parameter studies because of the need for extensive mode recomputation. The Galerkin method in the strict sense requires that the mode shapes satisfy exactly the boundary conditions [38], but this can not be easily achieved here due to the interaction of nonlinear terms in the root torsion boundary condition. By applying the Galerkin method in the extended sense [51] this difficulty can be avoided through adding the boundary residuals to the differential equations residuals. To reduce the order of differentiation (and hence the numerical error) integration by parts will be applied, and the boundary terms cancel. ‘The error is reduced further by inclusion of root-torsion in the deri- vation of torsion mode shapes, and quick convergence of the method is expected. The usual assumption of harmonic motion leads to a small quadratic eigenvalue problem to be solved by standard techniques. After the extraction of complex roots, the regions of instability are estimated from Nyquist diagrams. The inclusion in this analysis of torsion flexibility and root torsion, with the associated nonlinear terms, is expected to strongly influence the stability boundaries, since the non~ Linear torsion-bending coupling plays an important role in rotor blade dynamics [52]. In Chapter 2 the general configuration of the rotor blade is defined. After a brief presentation of the coordinate systems, the transformations and the ordering scheme ate developed for use in the non- linear derivations. Second order strain-displacement relations are developed for long, slender beans. In Chapter 3 the linear vibrations theory of the rotor blade is developed, and the uncoupled bending and torsion vibrations are studied. o Linea differential equations ave derived: from Rewtonian.onin for bending and torsion, independently. Frobenius series solutions are derived, and results for actual rotor blades are obtained through a transfer matrix method. Spoke diagrams, mass balancing, and mode shapes are illustrated. In Chapter 4 the nonlinear dynamics of the rotor blade is =29- treated. The nonlinear coupled differential equations of motion are derived from Hamiltonian principles on the lines of [28]. Addition of Greenberg's unsteady aerodynamics [36], and elimination of axial dis~ placement, yields a set of integro-partial differential equations. Application of perturbation method splits the dimensionless equations into equilibrium and perturbation sets, and these are solved separately for a uniform beam. In Chapter 5 solutions to the nonlinear differential equations for the equilibrium position are obtained with the shooting method. Typical equilibrium configurations are illustrated, and comparison is made with results with Galerkin solutions using a simplified set of equations. Full details of the shooting method are given in Appendix C, In Chapter 6 the linearized perturbation equations are solved with the extended Galerkin method, and Nyquist diagrams are obtained for several blade configurations. Convergence speed is examined and compared with that of other methods. The mechanism of torsion-bending is explained, and a comparison is given between available stability analyses. =30- CHAPTER TWO GENERAL CONSIDERATIONS 2a COORDINATE SYSTEMS AND TRANSFORMATIONS As a result of some pioneering work on coordinate transfor~ mations by Peters [28] the derivation of both linear and nonlinear equations for a rotor blade can be systematized to a high degree. In this section the coordinate systems and transformations are developed in some detail with a notation similar to that of ref [28]. The orthogonal axes system X,Y,Z and agsociated unit vectors 1, 5, & (fig. 2.1) are fixed in an inertial frame@. The orthogonal axes x, y, 2 are fixed in a reference frame @which rotates with res~ pect toQ at constant angular velocity sk. Point 0, a common fixed point to ®and B, is located at the root of the beam. The plane con- taining X, x,, ¥ and y is called the reference plane, or plane of rotation, The undeformed beam lies inclined at the precone angle @,. out of the reference plane (fig. 2.2). The x axis lies along the beam elastic axis (locus of shear centres), The orthogonal axes x, y, 2, and the corresponding unit vectors t, j, & are also fixed in®. Bending deflections of the beam are defined by the displacements u, v, and w of the elastic axis parallel to x, y, 2 coordinates, respectively. A blade~fixed orthogonal system x', y', and 2' is defined in the cross~ section with axes y', and 2" parallel to the section principal axes n, t, and with the origin at the shear centre. The cross section is assumed to be symmetric with respect to the naxis. The system x', y', 2! moves with the blade as the blade undergoes bending displacenents, torsional displacements, and pitch angle rotations. The projection of the blade wn before and after deformation in rode section in hey, 2 plane Ls fig. 2.3. Before deformation, the blade principal axes are rotated with respect to the undeformed coordinates by the pitch angle, 8. After deformation, the elastic axis is displaced by u, v, w and the blade twisted through the angle 9. Although the deformed y", 2" axes do not lie exactly in the y, z plane, their projection in that plane is shown in fig. 2.3. -31+ Figure 2.1 Undeformed coordinate systems E___.»¥____4 Figure 2.3 Rotor blade cross section before and after =32- Several coordinate transformations are used, including the simple transformations from I, J, K to 7, 3, &. A more difficult son eee : ae transformation is one fron 2, j, K to i", jt, #'. A concise derivation, and second order results are given below. (Note that the distance along the deformed elastic axis is also denoted by r, in addition to x's) Let [ be the transformation between the undeformed blade coordinate system 7, 7, &, and the deformed blade system i", 3", Kk’: t t t Z + t tli (2.1) ® ® In terms of Euler angles (fig. 2.4), T may be expressed as cosBcose cosésing sing g =| ~sindsinBoost costeose cosfsind an ~costsing -sintsinBsind ~cosBsinScost ~sindcost, cosfcosé +sindsing ~singsinBcosd The Euler angles are taken in the order T, B, They uniquely define the orientation of the blade principal axes with respect to the undeformed coordinate system. The Euler angles must be related to vw, >. As shown in figure 2.4, the angles % and B can be easily expressed in terms of v and w (sin B=w', sin = v'/vi-w® , etc.). ‘The determination of the third Euler angle 6, however requires the formulation and solution of a differential equation for T, as follows. Consider a small rotation @ dr of the blade-fixed system, which occurs as x goes through the increment dr. The vector components of the rate of rotation jb.can be identified as the torsional rotation rate, oj = C49)", and the beuding curvatures and uw, as shown in fig. 2.5. ik This infinitesimal rotation can be written in terms of T, giving the equalities Figure 2.4 Deformation and Euler angles. Figure 2.5 Transformation geometry. =34- bye b tb aloo alg] = a , i = PTY] = i e ra ®t nyt 2 i i 1 +} 3 =ryi ¥ & and hence the differential equation wa (2.3) The differential equation (2.3) can be solved by applying the identity rl sgt led as (2.4) Formally there are nine equations; however there are only three independent ones due to the usual orthonormality conditions on the transformation matrix. Substitution of (2.2) into (2.4) yields + Esin B. (6+$)* and substituting expressions for Euler angles Identifying derived from figure 2.5 yield, after some manipulations, He 2 F yitutar _/? _wiiwttvtar ° 2 [aa ane yh - o+e-] (2.5) oat av aw =35- ‘This is a general result which includes root torsion, fixed and variable pitch, and pretwist, To second order (2.5) may be written as a x T= 046, 6 <6 -f vivax (2.6) ° Thus, from (2.6) and (2.2), T is expressed to second order as 1- w je + np wt -[v'cos (8+) +w'sin(6+4)] cos (8+s4v'w') (1 sin(o+9) (I- S-) os yt sp wt [v'sin(6+9)-w'eos(8+4)] -sin(or}ev'w') (I- S-) — cos(o+d) (1- 5) an The deformed beam is shown in figure 2.6 with force and moment resultants acting on the face of a cross-section, At any point along the beam, x" is tangent to the deformed elastic axis, and the y' and z’ axes are identical to the n and t axes, respectively. Stress resultants and moments are subscribed with x", y’, z' to associate them with the deformed beam; for example, M,, is a moment about the x' axis. Vector components can be transformed from one coordinate system to another. For example, a resultant moment vector if may be written as Ho- wie ase ik (2.8) oe % hn n en Hoe athe mT ee (2.9) in terms of components measured in the undeformed or deformed coordinate spectively. elati colton t aM, systems, respectively. The relation between My, My, My and M,N, M, are determined by using the unit vector relations of (2.1) and taking appropriate dot products of (2.8) or (2.9). Other vectors, such as resultant force V and acceleration 4, obey the same transformation laws. ee: ORDERING SCHEME When deriving nonlinear equations one can encounter a con~ siderable number of higher-order terms which are small and may be discarded. If such terms are neglected within a large system of equa~ tions, care must be exercised that the terms retained consitute self- adjoint structural and inertial operators, This ensures symmetric -36- sultant forces and moments. Figure 2. -37- stiffness and mass matrices and an antisymmetric gyroscopic matrix in the modal equations. A systematic set of guide lines (ordering scheme) is adopted to determine which terms to retain and which to ignore. Several ordering schemes were proposed [15], [44], [46] and that of ref. [28] is employed below. A small parameter, ¢, of the order of magnitude of the bending slopes, is introduced. The orders of magnitude of the other constants and variables are expressed in terms of ¢ and are presented in Table 2.1. ‘The dimensionless axial deflection u/R was taken to be of the same order as the square of v/R and w/R, true for most blades. The elastic twist 4 is a small angle in the sense that sing and cos ¢ = 1*. The axial coordinate x is of the order of R and the lateral coordinates n, and ¢, are of the same order as the chord ¢, and thickness, t, respectively. Both ¢ and t are assumed to be at the same order of magnitude as v and w. The warp function is taken to be of the order ¢ times t, so that the actual warp displacement will be an order of magnitude less than the axial displacement u. Within the energy expression, terms of order c? are ignored with respect to unity. Thus, if the largest terms of the energy expression are O(c), then all terms of O(c*) are retained (first order terms), all terms of O(e5) are also retained (second order terms), and generally terms of O(c®) are discarded. There are conditions, however, under which certain O(e°) terms should be retained, and these are mainly associated with torsion. This inconsistency stems from the implied assumption that c and t, and hence n, and ¢, are of the same magnitude, where in fact, for most applications, tv c/10. This feature is a shortcoming of the present ordering scheme, and the situations when third order and higher terms are retained will be pointed out in the text. The physical quantities involved in the strain energy and the Kinetic energy are fundamentally different, and a standard of equivalence has to be introduced. Assuming EA/m0°R? = 0(¢*), this difficulty was overcome, and thus the ordering scheme can be consistently applied to both strain energy and kinetic energy terms. In the application of the above guidelines to a Hamiltonian derivation of nonlinear equations, it is important that the ordering be done within the total energy context, or equivalently, within the virtual work expressions (e.g. the v equa~ * However the sum (049) will be used in the derivation of the equations and the small angle expansion in terms of ¢ will be done only at a later stage. =38- TABLE 2.1 Assumed orders of magnitude of physical quantities Re ae 2 x oay | 0?) ff nb. eee ee oe) | Rog oe) a 4 oe) | o = oe) R aafan anfoe - won, Ae oe | a = oe) a Ym Rag = od) ik 7 Oe) ee oe’ 6 (aerodynamic = O(c) ma?R? forces only) EL, EL, cr —.— = 0(1) mR" m?R* mR kK, k, m, A 1 poate muemey 0 = oa ~39- tions times the év). Ordering differently in one equation than in another, without regard to this consideration can lead to the previously mentioned symmetry problems, and introduces spurious dissipative and circulatory terms. Also, the scheme implies that the same order terms should be retained in the v, w,$ equations, but terms of one order less should be retained in the u equation. 2.3 SECOND ORDER STRAIN DISPLACEMENT RELATIONS The development of a nonlinear strain-displacement relation is central to both the Newtonian and the Hamiltonian methods for developing a nonlinear system of equations. The use of this relation, together with the generalised Hooke's law, permits the strain energy, the force resultants, and the moment resultants to be expressed in terms of defor- nation quantities. Second-order nonlinear strain-displacement rela~ tions were derived in ref, [28] from a general standpoint and then reduced to second order. This proved that definitions of strain based on the deformed length increment, dr, or on the original length incre- ment, dx, become identical. A simpler derivation can result if second order approximations are introduced from the onset, as shown below. The strain tensor may be expressed in terms of T, and ¥, [53] the vector positions of the same point on the deformed and undeformed blade, res- pectively, as x Fen xe] | et fe, de, - a fe +10) at, dr, - ae, 2[dx dn de] fam 48ne| | 27 (2.10) sym. foe dc where dr, dn, dt are the increments along the deformed elastic axis and two cross-sectional axes, respectively, The vector position of a generic point on the undeformed beam is given by (x,y,z) with respect to the unit vectors 7, J, & where (x,0,0) is the elastic axis. The corresponding point on the deformed beam is given by (a) the position of the deformed elastic axis (x+tu,v,w) with respect to the undeformed plade system 7, 3, &, and (b) the position of the point relative to the elastic axis [-A(®+¢)',n,¢] with respect to the deformed blade system i, 3', &. the quantity -A(o+p)' represents an axial position where A is the warp function; 4(0,0) = 0. The warp function 4, and the section coordinate n,z may change under a general deformation, but they are taken constant in a second order analysis, Also, the elastic axis increment dr is equal to dx, to second order, and hence d/dr = d/dx = ( )" =40- Thus xt (org) H-@th fiw fect 5 : Quy w 5 where T is the transformation matrix given in (2.7), and relates the deformed axes (7", }', R') and the undeformed axes (i, 3, E). Before deformation, t= t, Q.12) and using (2.11) x -n8" =-GFRH | oferty] a ; (2.13) ° 5 where 1 0 0 | = |0 cose sino : (2.14) 0 -sine cose The position vector differentials are given by Lut =A(8+4)" A (94g) "ae= (A, ant dz) (044)" vt faret™'ar} on +T an w c ar eB, eA, (2.15) n cali where TY can be expressed using (2.3) as 0 0 a t Tr Tl 9 4] (2.16) ~, 4 0 Substituting (2.16) into (2.15) and expressing in terms of deformed -41- coordinates gives COG)" ~A (048) "2O, antadcy(040) " ar, a faet an w 0 & a (2.17) The dt, differential is dizectly deduced from (2.13) and (2-14) in terms of undeformed coordinates as dr 1 oO oO whoMdr- (A dnth-dg) 8? at,- TH | o}+ 0 cose sine an (2.18) 0 0 -sind cos® ap Substituting (2.17) and (2.18) into (2.10) allows solution for the components of the strain tensor. Substituting v', w' and © (see (2.6)) for uj, ©; and w;, respectively, and deleting higher order term yields 2 coo — Ag" + (n? + 02)(81H1 + S—) v"[ncos(9#}) - sin(e+é)] - w"[nsin(e+g) + ccos(0+9)] (2.198) Seq 7 TUE HALO 19b) bye 7 (NADY (2.190) The assumption of uniaxial stress, valid for long slender beans (rq 7 Seg" %q¢ * 9) was invoked and though ¢,, is not identical zero, it is [28] two orders of magnitude lower than the shear strains ¢,. and eqcr Equations (2,19) are exact to second order. Although the ¢ term in (2.19a) is of fourth order according to the ordering scheme, it leads to a tension term previously identified in [17] that contributes to the elastic torque M,", and hence it is retained. =42- CHAPTER THREE LINEAR VIBRATIONS There is a fundamental difference between the dynamics of stationary and rotating beams, This difference lies in the Coriolis forces and the centrifugal force field which produces varying axial force, and adds to the bending and twisting moments. Correct inclusion of these effects leads to complicated differential equations [17]. To gain insight into the dynamics of rotating beams, the linear equations for uncoupled bending and torsion motions are derived from Newtonian principles. 3.1 DERIVATION OF FREE VIBRATIONS EQUATIONS BY NEWION'S SECOND LAW Attention is restricted to a uniform rotor blade at zero angle of pitch, zero pretwist, and zero precone. ‘The centres of shear and mass are assumed to coincide. Under these assumptions the general configuration presented in figures 2.1-2.3 simplifies, and the coordinate systems (x,,Y42), (%y,2) Gton,t) coincide, as shown in fig. 3.1, The strain-displacement relations (2.19) are taken to first order; however the shear and rotary inertia correction [48] to the Euler's Theory of Bending (henceforce ETB) are included. The total displacements v, w, are split into contributions due to bending (subscript b) and due to shear (subscript s) as vorvyytyy + WoT Wye tw, G1) Thus the strain-displacement relations become ec UTS vy = wey BA) gry yt 4 - + Dees, : ay 7 TBO TS 3.2) (y-BWeew . ez Be: 8 The first terms in the shear strain components ¢,,, ¢,, Tepresent the TORSION wv 3.1 OPientation and motion of rotating uniform blade. aad effect of torsion, the second terms the effect of shear. The material is considered elastic, obeying Hooke's law Sy = Beye? yy = yy? ge 7 Ege G3) 3.1.1 Bending Vibrations By combining (3.2) and (3.3) and integrating over the cross section, the stress resultants for Timoshenko beams with axial load are obtained. The sign convention for the stress resultants is shown in fig. (3.2.a, b) and the stress resultants expressions are To Uf oh = Bhat (34a) = TY ays = GkyAvy (3.4b) TY aa = Ck Aw (3.40) Ho LI Sygeth = By (3.4a) 7 ID 94 = ELyy (3.4) where ky and k, are the shear deflection coefficients for the section. Considerations of force and moment equilibrium for the diffe rential elements in figure 3.2 result in the following equations Coes = 0 sy + (vii +p, = 0 SL + (iw) +p, = 0 (335) (a) (b) Fig.3.2 Differential elments for (a) lag, (b) flap, and (c) torsion. =46- where p,, Py» P, and dy, 4, represent the applied forces and monents per unit length. For free vibrations the loads are solely inertial and are obtained as follows. Consider an element of mass pdA initially at position (x,y,z). After deformation it will be (2.11) at the vector position acho 0 + + 2 t zy J Hl v fecty ; where the transformation matrix given by (2.7) is reduced to first order terms, 1 vb “ = [oy A 0 : a ° 1 Thus the deformed vector position is tu-yw!- x tu = yy) -2w) 12TH] vey 5 (3.6) wee The acceleration vector is given by the usual expression +0 x@xz,) @.7) where @ = sik is the rotation vector for the moving frame Oxyz. Hence the acceleration components are obtained as == yi ~ aul ~ 208 + a(n + u yeh - a) v + 20(8 ~ yep ~ ait) - n2(v + ¥) (3.8) =47~ The inertial forces p,, Ps P,» and the moments q,, q, are now evaluted in a D'Alembert sense as 1 p, = -ff pada = ~ mu + amy + m2(x + u) 2 HT as av - Imo + mo2v vy + ST 94,08 -Jf pada = aw, m=ff oda (3.9) a a a 1 a, 7 Sf oayaa = - aK + oy, % =i J pytaa e ' 2 1 -ff vajza = a (up + omf) i a 2 TM OM) 0. (100) ‘The first equation controls the axial motion, the second the lead-lag motion, and the third the flap motion. The axial and lead-lag motion are coupled through Coriolis terms. Generally, the lowest axial frequency is greater than ten times the rotor speed, hence the axial motion, and its time derivatives will be ignored. The centrifugal force resultant T is assumed to be inde~ pendent of time, and is found by simple integration as T(x) = fmn2(R2 - x2) Gel) =48- where the steady state axial displacement in (3.10.2) was ignored. The only difference between flap and lag vibrations is now reduced to the lateral effect of centrifugal force in the latter. This de-stiffening y-component of centrifugal force is shown underlined in (3.10.b). Hence a unified analysis is developed to cover both motions, and terms peculiar to lag are deleted when flap is considered. (a) Lag Vibrations ‘The uncoupled equations of motion for lag vibrations become sy + Cav")! - ov + mv =0 (3.124) Be avy) = 0 (3.12b) Substituting (3.4.b) and (3.4.e) into (3.12a) and (3.12b) respectively, and using the definition of total displacement (3.1), leads to the equation of motion ~ m, EIvi" +m(v ~ av) - (Iv')! ma [nv - 92) = Cav)" m2 my ee Mod . - “ae OO sy)" ~ m(v ~ 2928 + ayy + [EG - 92) 1") “e (3.13) The motion is assumed to be steady and harmonic, and hence the time dependence is eliminated with the usual substitution of vibration analyis caer be. 3.14) It is convenient to use the following dimensionless coefficient a= mtha?/EL, (rotation speed) A= metu2/EL (Erequency) (3.15) és EL, /Gk Ao? (shear deflection coefficient) e= ke 122 (rotatory inertia coefficient) 2 =49- where 6 and © are small quantities for practical geometries. The axial coordinate is made dimensionless, £ = x/!, and the differential. operator is redefined as ( )' = 8/38. The dimensionless axial force is defined by (E) = TGe)/mn?22 (3.16) but the deflection, v is kept in dimensional form. Hence substitution of (3.14) and (3.15) into (3.13) yields a fourth-order variable coeffi- cient ordinary differential equation for the mode shape, (1 + Gary" + 36arv™ - [1 ~ eS(Ata)] fatty! + Qta)v] + [le + 8 + edar) (A + a) - a(t ~ 362") Jv" = 0. (3.17) Care must be taken with the boundary conditions, for often only the bending component of slope can be specified at the boundary. Combining (3.15), (3.12b) and (3.4), the bending slope can be obtained in terms of total displacement py | S(itsar)v" + 2620c'v" + [1452 (tata) |v (3.18) ce BEL = e6(Ata)] Differentiating (3.4.b) and using (3.12), the moment can also be obtained in terms of total displacement 2 [C1 + éat)v" + 6artv' + 6(Ata)v] (3.19) Similarly the shear force is obtained from (3.12b) “EL, Sy = (C14 dat)v + 26at'v" + [(e+8) ta) tad] ¥"} y Bp -e8 (An 13[I-e6(+a)] (3.20) Note that displacement, slope, moment and shear retain their dimensions, as this form is more convenient when matching boundary conditions between discontinuous segments. () Flap Motion The flap equations are deduced from the lag equations by sub- stituting the appropriate constants and variables, and deleting those -50- terms connected with the lateral component of centrifugal force, (itéar)w" + 36atw' = [1-e8 (Ata) ] Cor'w'4hw) + [54+e(1+8at) (Ata) = a(t-36e")Jw" = 0, 3.21) 84, (issan)w!! + 262er'w" + [1452 +a") Jw" (3.22) ol afi - e6(+a)] -ET, Mw = —Z [(1 + Sor)w" + gartw" + S207] , (3.23) ¥ we “EL, 8, = 2 CCitsaxyw + 2oartw" + [8hteG4a) + abe" IW") - 23[1-e6 (Ata) ] (3.24) In practical applications, the blade is often much more flexible in flap than in lag and the Timoshenko corrections to Euler theory are unnecessary for flap vibrations. Then (3.21)-(3.24) reduce to the well-known forms wi" = apy" - ap'w' - aw = 0, dwi/ax = w'/2 (3.25) sonny? 8, = nrw" 23 y 3.1.2 Yorsion vibrations By combining the strain-displacement relations (3.2) with Hooke's law (3.3), and integrating over the cross section, the stress resultants for torsion are obtained as ir {fo ga = Bau! ww TT Weg *,y]0k + JI o%22)0 64a + KGEAu'g! (3.26) where -51- oe x IL 0 + 22)dA . The first integral in the expression for My is associated with St. Venant torsion theory. The second integral represents the untwisting effects of the components of longitudinal stress acting normal to the elastic axis with moment arm ¥y2#z2 [17]. Second order warping effects [55] are ignored. Consideration of force and moment equilibrium for the diffe rential element in fig. (3.2c) yields T+p, = 0 (3.27) M+a = 0 ‘The derivation of inertial force and moment p,, 4, is similar to that for bending. To first order in @ the rotation matrix (2.11) is given by 1 0 ° 0 1 (3.28) o 6 1 Ignoring warping, the deformed vector position of a point is Fo= @+uitGy-wy+getar. (3.29) 1 The acceleration vector (3.6.6) results in the 3 components 4292) — 92 (xtu) a. = 2 #200 - 92 (y-29) Hence the inertial loading becomes 4 = Sfoada =~ mu + mg? (xu) (3.30) 52+ = -ffo-ajz, + ay dd = -8 ffoly+z2)aa - 02ffo(y2z22aa v1 * 9% = mk2p ~ mk? - k? Dg (3.31) ‘n ny Say 2 1 where k2 = 2 42 Ae The first term in a, is due to torsional inertia, whereas the second term is associated with centrifugal untwisting effects. Combining (3.27) and (3.30) yields the equations of motion* = mi + Imoy + ma%(x tu) = 0 (3.32) WGI" ~ R2CTG) + mk2G + mOP(K2 - KA DG = 0 (3.33) bs 2 Equation (3.32) is the same as (3.10a) derived in the bending analysis, and the same assumptions regarding axial motion are invoked, As for bending, dimensionless equations are constructed introducing the non- dimensional parameters below: a me ca (rotation speed) a a w? (frequency) A= KARE (ratio of structural to inertial radii of gyration) woos e/a? (torsional inertia) Wi = a e Poe (centrifugal untwisting) * For non zero pitch angle a similar derivation yields the corresponding equation GIg" = K2(THNY' + ke} + mA (K2 -K? Joos20§ = —mH2(k2 -K2 ) Linde a mn 2 my ‘my “n,?2 =53+ The dimensionless axial force is given by (3.16). Hence (3.33) becomes gh + mua rg")' + Ou = any) ¢ = 0 (3.34) 3.2 SOLUTIONS BY FROBENIUS METHOD Unlike stationary beam equations, the equations for rotating beams do not admit simple analytical solutions. To obtain exact solu- tions the method of Frobenius [47] will be employed. Since (3.17), (3.21) and (3.34) are linear real, ordinary differential equations, the solution can be expressed in terms of real positive integer powers of the dependent variable. Thus the solution is assumed in the form . n+p x = Jae, (3.35) where the dependent variable X(£) may be either v(E), w(E) or $(E). It is necessary to write the axial force as (E) = atbetce? . (3.36) For discontinuous beams, or beams with concentrated inertias, constants a, b, and c, will also be discontinuous, Substitution of (3.35) and (3.36) into (3.17), (3.21) and (3.34) generally yields the solution. The constant exponent p in (3.35) has to be determined from the indicial equation [47]. Inspection of (3.17) and (3.21) indicates that these equations have no singular points. Thus, for the bending solutions, p will be zero. Inspection of the torsion equation (3.34), indicates the possibility of regular singular points, but for practical multi-segmented configurations these singular points are outside the range of interest, 0 < & <1. In the case of a singlesegmented uni- form blade, built-in at the root, it can be shown [19] that (3.34) has a regular singularity at € = 1, and, through a simple transformation of independent variable, yields the Legendre equation. The series solution of Legendre equation is well established [56] and the exponent p-is zero. ‘Thus, for all the three differential equations under con~ sideration the series solution has the form xe) = Dave (3.37) neo =54- 3.2.1 The Solution for Bending After substitution of (3.25) and (3.36) into (3.17), coeffi- cients of like powers of £ are collected and the following recurrence formula is obtained for lag motion. ‘ Sb (43) A, aed +3 (taba, yg = {8(a4a) - aa + e(Ata) (Ita8a) A int on + (a#3)(nt2)08e} Tapp eaaay tH ~ £6040)] (nt1)2abA_., + [Atotn(ntloc]A x a = - neo. (3.38) TEDDY” Substitution of (3.37) and (3.36) into (3,21) yields a similar expression for flap motion which is obtained from (3.38) by deletion of the terms underlined. Inspection of(3.38) shows that only four coefficients can be chosen arbitrarily, and these are taken to be A, through A,. Hence four independent functions are defined as follows: e Led + aleS + ales + ¥,(E) = 1+ aje* + aleS + aleS +... , 265 4 Aze6 + FA) + AES + azeo te, (3.39) - ae5 + A3e8 + FACE) = + aged + ages ey = HES 4 ahes #,() tated + ages eo... ‘The coefficients are all obtained from (3.38) by using the following initial values: For F, (6) set Neer alae and (3.40) 3 oe Al = Ad = and AZ = a= Al eo curd Oo to =55- Then (3.38) is used to generate the remaining terms*. The general solution of (3.17) and (3.21) can now be written ve) = FC) + CoP, (E) + GSP A(E) + C.F, (E) , G41) where constants C; are determined from four boundary conditions at the ends. Inspection of (3.38) shows that at £ = 1 only two terms need be examined to establish convergence asb(nt3)A,,, — ade(nt2)A. Auag "7 TGUAYGGEY 7 TIRSTSYGRES * (Strongly convergent terms), = C20)? a = BOO, (3.42) To ensure rapid convergence the terms aSb and aéc must be small, and for most rotor ‘blades log (aé) <-I. For very large rotation speeds aS can always be made smaller by subdividing the blade into two or more segments and thus keep the product 29 small; this was found necessary for aeroplane propellers. When boundary conditions (3.18)-(3.20) are vsed, derivatives up to third order are required and results similar to (3.42) are obtained. In the degenerate case of no rotation, and zero shear defor mation and rotatory inertia (6 (3.38) reduces to © = a= 0), the recurrence formula aA, Aus GGG * © Using (3.40), functions (3.39) reduce to cos 8.x+cosh Bx sin 8.x + sinh 6.x z , ar (3.43) cos 8)x + cosh Bx ~sin Bx + sinh 6.x (ge)? : (9) 3/3 * The superscripts on A, were introduced to denote that the coefficients for each function are independently generated. Thus A} # A}, AZ # Ad, etc. -56- The denominators can be conbined with the constants C, in (3.41), and the usual non-rotating beam functions [48] are recovered. 3.2.2 The Solution for Torsion Upon substitutionof (3.36) and (3.37) into (3.34), the follow- ing general recurrence formula is obtained +1) 2 Ay + +1) Apac Au = ony, (ott) 2mads,,, + [n(at1) Ayoe + An omy 9], Aye (3.44) a (a#2) (n#1) (1 + Anca) where A, and A, are undefined constants to be determined from the boundary conditions. Hence the general solution has the form Sey = OFC) + CF p(E) (3.45) where the functions F,(€) and F,(¢) are determined with the recurrence 0, and A, relation by making A, = 1, Ay 0, A, = 1, respectively = 14 alee + ales + alah +... (3.46) = 5 + Aze? + Ze? + age F, E+ age? + aZed + azet +... For the limiting case of zero rotation speed, the above (3.46) degene- rate to the well-known expressions [48] [rar fake cos\V aux) > sin( (3.47) IDERATIONS 3.3 PROGRAMMING Ct 3.3.1 Transfer-Natrix Method for Bending Beam functions F;(€) and their derivatives can be treated as known functions by constructing a FORTRAN function sub-program of (3.39) using (3.40) and (3.38). For a simple cantilever, deflection and slope are equated to zero at the clamp as are moment and shear at the free end. The result is four simultaneous homogeneous equations, and for constants j to-be non-zero the determinant must-vanish. ~Hence a simple interval halving method is used to search a predefined range for the discrete real values 1; which give zero determinant. The actual natural frequency w; is then obtained from (3.15). Actual rotor blades are seldom uniform, and often consist of a root region followed by a relatively uniform blade having perhaps one -57- or more balanceweights. In most cases this can be idealized as an assembly of n segments, resulting in a 4n x 4n determinant, When the determinant becomes large, the value near a root also remains large, e.g. 10 exp 30 for one 12x 12, Hence it is convenient both numer: cally and formally to employ a transfer matrix method (our notation is similar to that of Pestel and Leckie [57]). By placing deflection, bending slope, moment and lateral force in a column vector (usually termed a state vector), they can be related to the undetermined constants ©, by a4 x 4 matrix B, 28) = BCE) a (3.45) ay, Le we M,V}o, a= (ys Cy Cy, cyt . where V is the resultant lateral force given by “EI, ye = {C+ Sar)v"t + 26actv" a 23 [1-e6(Ata)] + [CetStedat) (A4a) - a(t-de") Jv" (3.46a) EL vi =S8, +e = ——2—— ((1 + éar)w" + 26ar'w" a8 23[1-c8( 40) ] + [8A + (etedar) (ta) - a(r-Se") Jw'} (3.46b) av, ee 2 Bag Ta ROT (3.47) By = MIV= FG)» Bg o7 Wt FO] the notation [v= F;(G)] denotes that throughout the expression v(«) and its derivatives are replaced by P(E). Consider a beam composed of n segments as shown in Fig. 3.3. Tf 2; (0) and 2, (1) are the state vectors at the beginning and end of the i-th segment, then 2,0) = BC) By) 2,0) = (3.48) -58- Mat Mi a i element (a) Bending iP torsional inertia it element iat (6) Torsion Pig.3.3 Generalized rotating beam (a) bending, (b) torsion ~59- where U, is the transfer matrix for the i-th segment. A concentrated mass may be included at the interface between two segments as shown in Fig. 3.3.a, and the transfer matrix from segment i to segment i+! for lag motion is 1 0 0 0 0 1 0 0 24400 = 2y2,(), By = (3.49) oO o 1 0 hesouea For flap vibrations the underlined terms are deleted. Hence the state vector at any point can be expressed in terms of the root or initial vector zp(0); at the tip, zq(1) = De), = B= (3.50) ‘The boundary conditions are now imposed and (3.50) is partitioned in displacement and force parts. For a cantilevered blade, 5 (Dt , = , (3.51) ay bea Par do (Ds 1M, (0), 5 (0) 3 Pea Pee The second of (3.51) yields the frequency equation [eel 0, and afore~ mentioned search technique isolates the eigenvalues 4;. For every A, the mode shape can be constructed by the sequential application of (3.48) and (3.49) starting with the root state vector zp = {0, fg} where fp is the solution of D.-fp = 0. Normalisation of the resulting mode can be enforced in the usual way. Care must be taken that the centrifugal force is properly represented. For the i-th segment, rE) = ap + betes? , Os bel , a, = Lampe jey + 85/2) * Mis (3.52) =60- Coefficients (3.15) may also vary from segment to segment, and overall convergence was improved when segment length 2; (3.15). t, was used in 3.3.2 franefer Matrix Method for Torsion Figure (3.3.b) shows an assenbly of uniform segments of different properties, and additional concentrated torsional inertias. © boundary conditions with root-torsion K, are enforced. The state vector for transfer over the i-th element is taken to be 2,0) = (0) 9M (I Osesl (3.53) where 4, (£) and M, (€) are the rotation and the torsion moment within the i-th element. “ The rotation ¢, is given by the solution (3.45). The torsion moment (3.26) is expressed in dimensional form as n= Dos caer (3.54) Hence the state vector can be expressed in terms of the unspecified constants as a, dy (3.55) where B,(e) Be) = (3.56) SS (reece) Awd Ft (E) (144 (2) ANDES Hence the transfer over the i-th element is written (c.f. (3.48)) as A) , (0) (3.57) The i-th inertia participates with a torsion moment due to inertial and centrifugal forces as My = M; [Rfu? ~ Ki. N19 where the K} and Kf. of gyration K, , Ky 2 2 Re 4K? 2 KR = Re TR TR Ba = BOR eens i seen Hence continuity of displacement and equilibrium of torsion moments give the transfer expression for the i-th inertia (c.f. (3.49) 1 0 (3.58) 41 = Bz), 2y2K2, 92: M, (RZu 2,0) 1 The combined transfer from the left-hand end of the i-th element to the left-hand end of the next is given by 2441 = Biz.) (3.59) Repeated application of (3.59) enabled one to obtain the tip state vector 2, in terms of root state vector zp, > (3.60) results in Hence for zero tip moment the following frequency equation must be satisfied Deg * Pesky o . G.61) -62- A search technique is used to extract the eigenvalues of (3.61), and the mode shapes are constructed by the sequential application of (3.59) starting with an assumed root vector 2, = es » 103}, Normalisation of the resulting mode shape can then be enforced if required. For the Limiting case of uniform stationary beam with root torsion the frequency equation takes the simple form KR eee Atan A Gr Oo, (3.62) and the associated mode shape is given by 4 = Ob where A, is a root of (3.62). FORTRAN programs were constructed [65] to determine frequencies and mode shapes of multisegmented blades, and results for tests and typical rigid rotor blades are illustrated. 3.3.3. Parameter Tests To evaluate the method and the general behaviour of rotating beans, a study was made of a cantilevered beam having a symmetric Ae section with mt*/ET = thus a! = 9, and xf = w and the Timoshenko corrections were given typical values 4 sionless frequency x4 vas found for the first six modes as shown in ‘The dimensionless rotational speed a’ was varied to 100 and the dimen~ Fig. 3.4, At high rotational speeds it was necessary to subdivide the blade into two or three segments. For comparison, it is useful to rewrite equation (3.17) for this bean, (1+ 6n2r)v"" + 36n2rv" + [(et6)u? + (ct8)0? + 92(36r"=c) Jv" = arty! = (wr+n2)v = 0 (3.63) where products ¢6 have been ignored and the underlined terms are deleted for flap. “63+ E&=Solution by Euler's theory T=Solution with Timoshenko's corrections 500 Dimensionless frequency, X? s 8 N Ss Ss 100 0 20 40 60 80 100 Dimensionless rotation speed, cx!!? Fig.3.4 Dimensionless plot of frequency versus speed for a uniform cantilever, -64- Except for small rotational speeds, the frequency increased linearly with 2? as predicted by the dominant terms of (3.63). When the frequency is much greater than the rotational speed, uw? >> 02, the lag and flap curves are virtually identical, This can be seen in (3,63), the differences between flap and lag are terms of order 27/w*, The similarity of the two motions is seldom seen in practice as ET, and EI, usually differ by an order of magnitude or more. For stationary beams (1 = 0), both shear and rotatory inertia influence frequency identically through the classical term (é+c)02v" which reduces the ETB frequency. However, for rotating beams there are additional shear correction terms chiefly dependent on the product gr, and in the lower modes (9%t >> w?) these terms dominate. Due to the coupling between shear deflection and rotation speed the first lag frequency has a special behaviour, and at high speeds corrections larger than 10% can be found (fig. 3.5). In the higher modes the classical term (S+e)w?v" dominates and for small @ the additional shear correction terms ($9%r, ete.) reduce the total correction (-25% for u,). But as Q increases the additional terms add to the classical correction (about 17% for An 60). 5 ata Several approximative expressions are used in practice to modify the frequencies of a non-rotating beam for rotation speed effects, in the form wt Re, where K is the well-known Rayleigh Southwell coefficient. For uniform beams, empirical expressions were derived by Peters [19] from an analy- tic treatment of the differential equations. These expressions (Appendix A) were "bench tested" against the present exact solutions, and the percentage errors are shown in figures 3.6 and 3.7, for bending and torsion respectively. These errors are generally small. The 0, and %» ©, whereas at normal operating speeds give results within 3% of the composite expressions for bending fit asymptotically at both & exact solution. The Galerkin formula for torsion fits adequately even low stiffness blades. But it cannot account for root torsion, which can decrease considerably the frequency of practical high stiffness blades. 100 90 80 NONDIMENSIONAL ROTATION SPEED Q TIMOSHENKO. 20 30 40 50 60 70 10 10 =~ 8 g 2 © 4 N 8 ot A ONSNO RA TVNOISNIWIGNON Fig.3.5 Variation of first lag frequency with speed Wa Error 3% Rotation speed, 8, rad/sec est of Peters'sexpression for bending [ 19], 33.516. "NI Error 4% 3% 2% 1% 4 ee Q=2., Low stiffness, Oa” 2.28 ——.— High stiffness, pig /=5-68 Ws, 20 40 100 Rotation speed, Q, rad/sec Fig.3.7 Bench t of Peters' [19]expression for torsion. ~68- 3.4 APPLICATIONS TO PRACTICAL ROTOR BLADES To test the general applicability of the method the helicopter main rotor blade shown in Fig. 3.8 was analyzed; the values shown are similar to actual blades. The first test was to generate the classic spoke diagram, and two segments were used to represent the blade without the Ikg added mass. The results for uncoupled lag and flap are shown in Fig. 3.9, There is no fundamental difference between this blade and the test curves in Fig. 3.45 however, the high lag stiffness means that only a small portion of the curves in Fig. 3.4 are shown in Fig. 3.9. For this blade and its frequency-speed range, the correction to ETB is almost entirely due to the classical term. The addition of small balancing weights on a blade can shift frequencies considerably. At the operating speed, Fig. 3.9 shows a potentially dangerous 5% resonance of the second lead-lag mode. Hence two identical segments are used for the aerodynamic portion, their interface being the location of the Ikg addedmass. The position of the 1kg mass sas varied from r = 1 m to the tip, and the resulting mass balancing diagram is shown in Fig. 3.10. The first mode frequencies decrease as expected, but the influence of the added mass on the higher modes is more complex. The increase of frequency is due to the added stiffening effect of centri- fugal force which depends on the relative position of the I kg mass. Thus the points of 2ero frequency change are not node points of the eigenvector, but points where the decrease in w; due to added mass is matched by the added centrifugal force stiffening. Figure 3.10 indi- cates two preferred positions, point A and the tips the former would probably be selected for fatigue and other reasons. By using two or more weights, it is possible to shift frequencies nearly independently. Another helicopter main rotor is shown in fig. 3.11. For this configuration, the blade is not built in at the root, but is attached to a complex hub structure, vhich has considerable bending flexibility. Hence the analysis is performed for the whole configuration, by consi- dering a number of different segments to model the complete structure, from the centre of the hub, through the pitch bearing and the "dog-bone", to the tip of the blade. For torsion a root spring , modelling the control system flexibility, was attached at the end of the pich bearing. The properties of these segments are listed in Appendix B, and were w OSE w009e w 0071 “059° a wy b4S= w w/By L=u | hy, O4L=W Zz | zz wy Z0- yur Zurn 4081-7719 ZW NY 06 = “TF NW 9 =v449 zune 9 | gurny su=*13 | 1, a simple hingeless rotor blade. Pig.3.8 Blade " 4 iy 3 Operating speed Frequency, w, rad/sec Rotation speed, Q,rad/sec Fig.3.9 Spoke diagram for Blade 1. Frequency, rad /sec 215 210 oS a oS 3S 120 a 10 46 4a 32 30 | wh Blade '~ 2 2. 4 §. Radial position of added mass, m s balancing diagram for Blade 1. GOH TOINGD HOLId 3av 18 4OLoe (3NOG -900 INGISNILX3 3AITS JOVIG ( ONTEY3A HOLId) 3ATI7S 30v7E anh 40104 Soe yo a complex hingeless rotor system. 3 cI o -73- used to compute the natural frequencies and the corresponding mode shapes. The frequencies are presented in fig. 3.12, and fig. 3.13 compares non-rotating and rotating modes. The bending slope, bending moment, shear force, and torque are also shown. The full details of mode shapes for lag, flap and torsion up to the third frequency were collected in Appendix B. A study of these mode shape diagrams gives further insight into the influence of centrifugal force upon blade bending and torsion. ‘The sharp variations in slope and curvature, present in all plots, are due to the sharp changes in properties near the root. For example, due to its high bending flexibility, the dog- bone takes 80% of the total first mode lag bending (Fig. 3.134). For higher modes these percentages are different. The distribution of bending moment varies radically between nonrotating and rotating con~ figurations. For lag, the bending monent at half radius drops with rotation speed from 30% of total, when stationary, to 15% at full speed (Pig. 3.13.a). In flap the decrease is from 31% to less than 1% (Pig. 3,13.b) which illustrates the centrifugal relief of soft inplane blades. By comparison the torsional node is not visibly influenced by rotation speed due to its high stiffness characteristics (Fig. 3.13.c). The amount of root torsion was chosen such as to match the nonrotating torsional frequency given in [50], and Fig. 3.13.c illustrates how root~ torsion modifies the mode shapes. For the first rotating flap mode most of the bending takes place in the first fifth of the blade. However this is not true for higher modes (Appendix B) where significant bending can also take place in the outer blade portions. One particular advantage of the exact solutions and transfer matrices is the ability to calculate higher derivatives accurately. Most other methods which employ finite differences [25], or interpo- lation functions (e.g. Galerkin [17] or finite element solutions [29]), obtain moments and shear by numerical differentiation of the mode shapes, and the results are well-known to be inaccurate [54]. A second advan- tage is that higher frequencies are computed with equal accuracy. and ease. Apparently the method used for the Blade 2 in [50] runs into difficulty at the second lag frequency. FREQUENCY ihe ) S$ \ 100 200 300 400 ROTOR SPEED (RPM) F=Flap Present analysis Lelag —-—- Predicted by[50] T = Torsion x ~~ Measured [50 ] Fig.3.12 Spoke diagram for Blade 2. é fro | 3 6 € 4 = 8, o O Fr Bb RB 4 5 t 2 # Radius(m) Radius(m) 320) 4 2300 = y F160 & 200 5 ee § 5 = a 60. & 100 o o o 5 o ff DB 2 Radius (m Radius (m) fa) LAG MODEI. ~~ Stationary Fig.3.13 Mode shapes for Blade and (c) torsion. — Operating (241.89 rad /sec } 2, (a) lag, (b) flap, 32 & = N & 10. g 5 6 af 3 a 8 5. OF RR Ke Radius (m) Radius (m ) 160 z = 2 : é 5 Radius (m) Fig.3.13 continued. (eo) a ese Radius (m) FLAP MODE. >= Stationary — Operating (N= 4t-89rad /sec ) zo g § 0.60 3 & & Be 0-40- e g 2 & 0.20 4 a4 0 1 2 3 : ‘ Radius (m)} (c ) TORSION MODE 1. built-in, Ky =o p= 363 rad /sec ~ 12004 8 “. —-—root-torsion, Ky =2639 Nm 3 . Wy = 131 rad /sec & = 800+ © 2 & Ss e 400- o4 1 2 3 ; Z Radius (m) Fig.3.13 concluded, -78- CHAPTER FOUR NONLINEAR DYNAMIC STABILITY 4a DERIVATION OF NONLINEAR EQUATIONS BY HAMILTON'S PRINCIPLE The equations of motion and boundary conditions for a slender homogeneous cantilever beam rotating at constant speed are obtained from Hamilton's principle. The simple representation used here assumes piece wise constant mass and stiffness properties, a symmetric blade section with no structural and inertia offsets, zero pretwist, and a small precone angle. Euler bending and St. Venant torsion theories are employed. The derivation differs from [28] or [22] in that a some~ what simpler blade is assumed, new higher order torsion terms are retained and root torsion flexibility is included. Hamilton's principle may be expressed as £2 { [s(u-r) - swat = 0 (4.1) t i where U is the strain energy, T is the kinetic energy, and W is the virtual work of the external forces. Thus expressions for 6U, 6T and 6W are derived and combined to yield the nonlinear differential equations of motion. 4a Strain Energy Contributions The usual expression for strain energy in terms of engineering stresses and strains defined in (2.19) is toe + 2 4 exSace * Saen an * ce See aNAea + AKL OZ Sy» 2d R vas io. oA where the last term represents the strain energy in the root torsion spring. The first variation is -79- au Pl aoe + ny + Mer Sey aNALEK + KyF0,) 56 (5) (4.3) where By Eh Syn * ene Oye * Ces (4.4) and Seg = Sul + visu" + wtbwt + (n24e2)g" Sp! ~ ASG = [ncos(9#9) - cein(o+s) ](6v" + W'S) = [nsin(o+4) + ceos(o+4) ](6w" = v"S9) 5) E+ EE be, n= A, D8 Since the strains (and stresses) are composed of the sum of terms of order e? and 3, any product of stress and strain consist of terms of O(e*), (5), and O(e®), Consistent with the ordering assumption (e2<<1), terms of O(c%) are neglected because of the presence of 0(e*) terms. In terns of stress resultants and moments, the strain energy variation becomes R U = f (youl + vtout + wiawt) + M59" + DM, cos (ors) + M, sin(o44) ](Sv" + w"69) + [H,ysin(o+g) ~ M, cos (od) ] (Su! ~ v"69) Hox + Fo) Mo) ees where the stress resultants and moments are defined by = ff oande = BA (ut + (4.6) ae = - + gto (ne? Meet SE LO, ~ 8g) * Hn? 46°) ands (4.7) 'y paetg! (ut + 212 = GF 8+ Baki6) (u > (4.7) ~R0- = Jf Sogande = BLL, fvsin(o+4) ~ weos(6+9)], (4.70) a ™ a wr = -[f no, ands EL, [v"cos(0#¢) + w"'sin(9+4)]. (4.74) A To simplify the analysis second order warping effects and coupling between warping and bending were ignored. Two higher order terms were retained inspite of the ordering scheme, the torsion term in the axial force (single underline) and the coupling between axial force and torsion (double underline). The former will contribute to the gyro- scopic matrix, the latter is essential for the effect of rotation on torsional vibrations. The section integrals appearing in equation (4.7) are defined as follows. BA = ff Ednde (axial stiffness) A Jf Be?anae (flatwise stiffness) A EL, = Jf Bn?dndc (chordwise stiffness) (4.8) A 2c 2h 45 2d) orsional stiffness: a {foto ecto Bsc Banae ce 1 stiffness) Ak? = f{(n2+2%)ande, (polar moment of area) These integrals are to be evaluated only over the portions of the blade cross section that are structurally effective. Integration by parts of the strain energy yields R uf ou + ¥,ov + Yow + Y,S0dx + b(U), (4.9) “Wy! (4.10a) = [H,,cos (ory) + M,,sin(o+9)]" - (Wyav") (4.100) -81+ DM, ysin(o+$) ~ My ycos (044) ]" - Maw)! (4.100) x, = -Q1,4)' - WDE sin(ore) - M1008 (+4) ] = wh EM, 1eos(044) + My, 9in(o+9)] (4.104) jr R BCU) = Vudu] + Wye vt DH, .c08 (ore) + Msin(o49) ]"I6>] ° ° R \R + Di,veos (ord) +H, sin(o44) 18 le +L Mrsin(org) ~ Lve0n or4) Tom" | \* + (aw! = Df rsin(ete) ~ ¥, ,08(9*4)] 286 | R + HS] Ky (oy 280) * ant) Tt is generally essential to include b(U) in (4.9) when using Galerkin solution method. 4.1.2 Kinetic Energy Contributions In deriving the kinetic energy contributions it is more con- venient to employ the coordinates (x,,y,,2,) used in (2.11). The deformed vector position F, = (x,,¥,.%,) is given by (2.11), which for zero pretwist reduces to x, =x + U- 2" ~ v"[neos (+4) ~ ssin(Or4)] ~ w' [nsin(o+4) + ceos(or9)], yy, =v + neos(ors) - csin(O+4) (4.12) w+ nsin(o+g) + teos(O+s) - In expanding (2.11) second order terms in the displacements were ignored. ‘The first of (4.12) can be reexpressed in the simple form xox tur ad ov vy) - ww) (4.13) ‘the velocity of the generic point in the blade with respect to the inertial frame is -82- set nk x, 14) where 3/at is the derivative in the rotating frame fig. 2.1 and R= ogo + sit. The second term in (4.14) is the velocity contributed by the rotating coordinate system, ras z > * xi, = -ay,T + (ox, - 22 + : he 9K xT, = oy, T+ (x, - 218,05 + Oye (4.15) Thus the total velocity is given by ene re rs 17 WPT + Gy + oe, - mB OT + Gy + 9y,8 DE 4.16) ‘The kinetic energy is the volume integral of }pV W and its variation is simply 2 ea ot = f ff ov 8¥ andzax 4.17) o A where 4 ee Bene + 8 = (6k, ~ v8y,)T + (69, + 98x, - 98652))5 +(e, + 98,57) (4.18) The integrand of (4.17) thus becomes Wa¥ = 6%. = oh by, - Ay, 6%, - 02 + o2x bx, - 9% 2 1 7 8, 8y, — AY, 6%, ~ My, Sy, + MF xpd, - Ox /B 82, 5, - we + 92,62 - ; + oxyaF, ~ 07218, dx, + 215% 82, - 028,65 + OF 8x, - 0,862, + 5,85, + 0y, 82 OF oy ~ OF, 8, 82, +H, 8F, + 74,8 Ly, + 9y;8,.58, + 02,8,.57) + pe (4.19) According to the variational method, (4.19) must be integrated between two arbitrary points in time, t, and t,. The initial and final values t | t and combining ‘various terms in (4.17) and (4.19) yields = ag | °: ; ; (outs 0%] :°) are eaken as zero. Hence integrating by parts R er =f ff 2,6u + 2 6v + Z ow + 2,86 + Zev" + 2.8" + 2,186" )pdndedx oo (4.20) where -93- Wx + 206 9 fue(y,-v)] ~ + $Czy-w) + 208, - OLE -'Y,-v) H(z -W)] = 0B x - ¥- 3 - bo, Z 078 x ~ 206, eG, =02x(y,-v)w! + 92x(z,-w)v" — a2 fve(y,-v) H(z.) + ¥(z,-w) 1 1 1 1 1 (4.21) (y-~) ~ 98K -w) ~ SL? # 7) Zi y= (nx + 200) (y,-v) ~(92x + 290) (247%) “ox. 4 ‘These expressions have been truncated consistent with the ordering scheme (c2< (4.29) * A full account can be found in [28]. =86- in which the last term represents the root-torsion contribution. This approach to the root-torsion of cantilever blades will prove advantageous when applying the Galerkin method of solution, and natural boundary conditions, like (4.29), will be included in the process. 4.2 AERODYNAMIC LOADING Friedman and Yuan [33] present a concise review of the classic strip theories pertinent to rotor blade aeroelastic stability. These ari (a) Theodersen's classical incompressible unsteady aero- dynamics; (b) Loewy's modification of Theordorsen's fixed wing aero- dynamics which accounts approximately for the effect of returning wake of the rotors (c) unsteady two dimensional fixed wing aerodynamics with compressibility corrections; (a) unsteady two dimensional rotary wing aerodynamics which contains compressibility corrections in addition to the effects of the returning wake. ‘The above theories are all based on the simple blade model shown in fig. 4.1 which incorporates vertical and torsion oscillations and uniform incoming flow. This is adequate only for fixedwing aeroclasticity. A rotor blade performs simultaneously involved flap, lag and pitch motions, and hence the model, and the theories, mist be modified to include: (1) the time dependant velocity variation due to lag motion, which adds to the constant oncoming flow; (2) the constant pitch setting, which adds to the oscillatory angle of attack due to elastic torsion; (3) the constant inflow through the rotor disc, which adds to the vertical oscillation. Implementation of these modifications leads to complicated analytical expressions for the aerodynamic loads, which can be substituted in the right hand sides of equations (4.27). These modifications are applied to Theodorsen's theory, as extended by Greenberg [35] and as presented in [36]. Quasi-steady conditions are assumed. ~87- 4.2.1 Quasi-Steady strip Theory In Theodorsen's theory [6] a two dimensional airfoil is assumed to be pivoted about an axis which may be distinct, in general, from the aerodynamic centre. The airfoil is pitched at an angle ¢(t) to the free stream flowing at constant velocity V. The airfoil is vertically displaced with velocity h(t) positive downward as shown in fig. 4.1.a. Greenberg [35] has extended Theodorsen's theory for pulsating free stream velocity v(t). The lift and moment are expressed as the sum of circulatory and non-circulatory parts L = ly + lye (4.30) Moz Mot Me» where Lgs Tyge Mgs Myg can be expressed in terms of h, €, V and their derivatives, and the constants p,, a, c. It should be noted that € is the angular position of the airfoil with respect to space; é and & are the angular velocity and angular acceleration of the airfoil. For a rotor blade airfoil, the total fluid velocity U is expressed in terms of components U, and U, in the principal axes systen as shown in fig. 4.1.b. Hence the angle of attack is defined as a = tan (0/09) - The following expressions are set up to relate the Greenberg theory with the rotor blade model, Ups -h-ve , P vo: We « v OT (4.31 Thus the 1ift and moment components can be expressed, for quasi-static condition and coincidence of aerodynamic and elastic axis, as Pac ee . , ot aU US a3 (4.32) Pyae2 SMe GG Gee The noncirculatory lift is taken to act normal to the chord line, and the circulatory lift is taken to act normal to the resultant blade velocity U as shown in fig. 4.2.1. An aerodynamic profile drag force =88- (a) classical aeroelasticity [6] (b) rotor blade dynamics Fig.4.1 Geometry of Blade Motion (a) classical aeroelasticity [6] (b) rotor blade dynamics Fig.4.2 Orientation of Aerodynamic Load Components -89- per unit length, acting parallel to the resultant blade velocity, is included based on a constant profile drag coefficient cy , 0 pase 2 — (4.33) The forces Lg, Lyg and D, shown in fig. 4.2.a, are resolved upon the section principal axes to yield the forces $ and T as shown in fig. 4.2.b. This is achieved through simple geometric projections. 4.2.2 Velocity Components on a Rotor Blade ‘The velocity components U, and U, are next expressed in terms of blade bending and torsion deflections, v, w and ¢, as follows. First the velocity vector U is written in the reference coordinate system x,y,z as tray tu ze UT UR 4.34) B= G- mts G+ OT + Ow, +H + v8 DK =U, where v, is the induced inflow velocity. The induced inflow is taken to be steady and uniform along the blade radius, and equal to the value of nonuniform inflow given by the blade element momentum theory at the 0.75R [34]. The blade angle at x = 0.75R is set equal to the blade collective pitch plus the equilibrium elastic twist radial station x $o at x= 0.75R. Thus Ving = 8enl6+e,(0-75R)} 0 32 [4 +B 49(0.75R)| = | (4.35) ‘The unit vectors 7, }, &, and 1", J", &" of the coordinate systems x, y, z and x', y', 2', respectively, can be related by the coordinate trans— formation T given by (2.2). Therefore, the blade velocity components in the deformed x", y', 2! coordinate system are u, v Up Tee (4436) ® Thus using equations (4.36) and (2.2) yields, to first order in 6, the expressions for velocity components = ote x = ce (0+p+ f viw'dx) + vy + W- (OHE) + av(6, ote") (4.37) ° ‘The derivation of angular velocity component about the x' axis, &, is similar to the derivation of u; (2.4), But here T is differentiated with respect to time, and not to space. Thus é can be expressed [36] e = $+ 96,. +0) (4.38) 4.2.3 Final Expressions for Aerodynamic Loads Substitution of (4.37) and (4.38) into (4.32) and (4.33), followed by projection of Lygs Les Myc» Mo upon Oy" and Oz! axes, and rotation of resulting T and $ forces through rotation matrix T yields the final expressions for the aerodynamic loads: ca fev; (O48) - [ax 52 + Corer, 1 + [uz ~ ox(o49)]0) (4.39a) Cay, + HCO y+ [vu "as) = ota 8") +e yew") + [2ax(ate) — vy] ~ si (4.39b) (4.39) Nonlinear rate product terms ww, J? and W2 were neglected since they do not contribute to a linearised stability theory. Also, all 0(c3) terms, except those contributing to lead-lag and torsion damping (double underlined) are neglected (including © terms). “91+ 4.3 STABILITY IN THE SENSE OF LIAPUNOV The stability of the solution of nonlinear differential equations is analysed here in the sense of Liapunov [14]. An equili- brium solution of the system is assumed to exist, and the stability of small perturbation motions about this equilibrium state, is investi- gated. For illustration consider the single degree of freedom non- linear system: = £@) , (4.40) for which the equilibrium solution x,, satisfies fe) = O- (4.41) Making the substitution, x(t) = x, + Ax(t), into (4.40), and expanding £(x, + dx(t)) in Taylor series about x,, yields M(x.) ak = £(x,) + dxf (x,) + (ax)? Neglecting the higher order terms, and using equation (4.41) leads to the equation of the first approximation, ak = ajox(e) a, = EMG.) (4.42) Equation (4.42) is linear in the perturbation variable Ax. Liapunov's theorem states that provided a, #0, which is frequently satisfied in physical problems, the information obtained from the linear equations of the first approximation is sufficient to give a correct answer to the question of stability of the non-linear system. The solution of (4.42) is dx = Ce“I", Hence, according to Liapunov, if a, <0, the equilibrium is stable; if a, > 0 it is nstable; 1 finally, if a, = 0 the equation of the first approximation is not applicable and higher order terms are required. ‘These simple ideas are easily extended to multi-degree of freedom systems generally described by a system of first order diffe~ rential equations Since any higher order system of differential equations can be reduced to a first order system of a larger size, the Liapunov theorem of stability can be applied to the nonlinear differential equations (4.27) 92+ for a rotor blade. 4.3.1 Nondimensional Equations Before proceeding any further with the stability analysis, it is convenient to cast the differential equations in nondimensional form. ‘The following nondimensional parameters are used. EA Kh K+ aye + ns a we co , - a, aneer t i 2 EL Kn = aa Mt eS EL, KR = =i; =; mazer + f= ori (4.43) Ajcos?6 + Aysin?é BOOT Hy thy = (iy ~ u,)sin2e in? 2 Aysin2o + A,cose Hy Tyg = hy - hy singe Hy = (iy ~ wy eos20 = 2 - a2 Ty, = @3 - A2)cos2e By replacing the independent variables x and t by & and y, the differential operators are redefined as ( )' = 3/36, () = 3/3¥. The unknowns u, v, and w become @ = u/R, ¥ = v/R, = w/R, and ¢ and 6 were nondimensional from onset. Using small angle expansions in $, (4.27) becomes (4.44a) 120)" + Ty, Gay” =93- <0 4 28 = (4.460) +28 v = T-B Ue (4.440) wacrgtyt + BB (wt = v2) + yyw = Ad + ub + Tyg (4.44d) (4.44e) oO (4.446) This set of five nonlinear partial differential equations in u, V, w and + can be reduced to a set of 3 nonlinear integro-differential equations in the V, W, ¢ by the following process. Integrate (4.44a) to obtain 1 : vos f G+ Wee, - (4.45) & Substitute in (4.44e) and write 1 ~ peel ce Way (4.46) Integrate with respect. to £ and differentiate with respect to ¥ to yield Ble 5 5 ff vaeae, - [Gt vt + ww" + ag’ ae, (4.47) or ° vals Though not essential, it is convenient to discard the first term in “94- (4.47) since K~ O[10°]. This is equivalent to postulating that the rotor blade is inextensible for perturbation bending deflections. Substituting (4.45) and (4.47) into the set (4.44) yields -, 7! s — “hw { (E+ Qv)de]' + Ay : epee ; - ~ Bw ev 2 J (iv! + wiv + Ae'o dE, = T (4.48a) 1 : -[w" [Ce + Ww)de]" & ++ Bye Ty Bock (4.48b) Gee Me Gia — G02) 4 F ~ALo ee +e aN = V2) Twa = gh + F “2 o tubt wes Bw (4.480) 1 : + Af E+ WA] oe) + ALE (g) = 0 (4.48d) “Ly The third order term 2v in the torsion equation (4.48c) was discarded in [28], but is retained here. The nondimensional equations (4.48) are nonlinear, integro- partial differential equations with variable coefficients in & and contain the essential ingredients for stability analysis of the simple cantilever rotor blade configuration studied here. The nondimensional nonconservative forces L,, L,, Hy were, so far, unspecified. Henceforth + it is assumed that the aerodynamic loading (4.39) applies. In non- dimensional form the loads are + OF I + + Dv, > (+4) 6] (4.49a) ee £ = = G (rev, + 62 (0+ e+] vitae) = sv(B, + W') +E + (4.496) B= +t (-2 68) ene (- 5) (4.49e) = santo + ¢,6-75)1 2 [0 + Ble + 90-75)! - 1] (4.494) 4.3.2 The Perturbation Expansion For stability analysis, (4.48) will be linearised for small perturbation motions about an equilibrium operating condition by writing the dependent variables as vEW) Va (6) + AvCE SY) WEw) = HCE) + wesw) (4.50) oy) bo (6) * Ad(EW) Two sets of equations result. The equilibrium equations are a set of coupled nonlinear integro-ordinary differential equations in the space variable, and are the equivalent of (4.41). Their solution will be the equilibrium configuration defined by V,(E), W,(E)s ¢,(E)- The other set, whose coefficients depend upon V,, W,+ $,5 is a set of linear integro-partial differential equations in time and space variables, and is the equivalent of (4.43). They are henceforth referred to as the perturbation equations. The equilibrium and perturbation equations are derived by employing first order expansions of nonlinear products, such as 1 1 : TPG, +a, = Gravy fe, + wwe, & & 2 Fe +a f mae, ‘The nonlinear terms in (4.48) generate nonlinear equilibrium cerms (single underline) and linear perturbation terms (double underline). ‘The equilibrium equations are given below, -wl a Oy i, -52)" 12 a a - 7 4,¢0) * AyF aco) 7° The equilibrium loads in (4.51) are oe re vat ‘in ¢ = 4a. = t [ % See aal ea ¥; (0r9.)6 ] in XO, + e2ECor > + me - (8, + wD 6 [in * ELE: J VoNoe] ~ olBne * MS +e wiel Ge * ¥)3 sen(o) cr + Joly? - 1y 8 10 ‘The perturbation equations take the following form - EE wy) aw a = F946" + Ha" + Cedi" + Wag" - av av ~ 28 Ww (4.51a) (4.51b) (4.51e) (4.514) (4.52a) (4.52b) (4.52c) (4.524) -97- + ' eee eee ee . wd) - 2 f Gisavt + aaw + Ag a6)48, = of (4.53a) sm Kaa! 4 Th aay" +E av + Taw + Ty (a0 + Vine) a) wf 2avae,) + Ty (@jaw" + ag)" + tie - J + 28, Av = of, (4.53b) = (4.53c) A a “Uh + Doi Mola” MYcon! 28886, = 0 (4.534) Similar algebra yields the perturbation loads; ¢ Ss = sh pes at, = £ -0¥,,00 - [2 Pe + OHH d7, 0 fh + [2y,, — (ore yeldw dy (4.54a) _ 5 — ae ai, = Eero + i Copa + whav' yds] ~ ELV ABT + (8, +8) AF] + wie} Ai + [26(049,) ~ Viyldv ~ caw + SE cah-$ aw) (4.546) Zo Seah). (G54) ~99- a 7 senle + 6,075) 2 [fr + Blo + #,¢-75)|)# - 1] (4.544) The solution of equilibrium and perturbation equations, (4.51) and (4.53), will be obtained in the next two chapters. Since the equilibriun solution is used to form the coefficients of the pertur- bation equations, special attention will be given to obtaining an accurate nonlinear equilibrium solution. The solution of perturbation equations will be obtained with the Galerkin method, and to this purpose it is convenient to cast the equations in matrix form by defining the following matrix differential operators. 1 0 0 an oenO O30) D 0 -F 0 (4.55a) o oO v oO o 0 _ ot 1g 1 o& -2[Wf(..nae,]'—2ferc.o tae, | -2fagc rae, angen Dae, . ° ° ° | -26 pe ro a 1 ol Gab) + 28, | 0 0 1 n2ne, f Coedde | 0 0 E (4.55b) =99- cons )essy, | | -orperi, | 0 | oe * & (4.55¢) (4.554) -100~ 0 0 “EVE S sy = > >» ae-E (8, #4) In) det EF & 0 oO oO (4.55e) ) shows where the dependent In the expressions above the symbol ( variables must be inserted, and when the symbol is missing, the variables will be inserted at the end. The system (4.53) can be expressed in concise form using matrices (4,55) and the column vector SEW) = (VCE), AWCE,W), 46(E,0) } (4.50) 0 o olfa] fo o 0 we 0 o 0} jaw) +]o 0 0 aw) 78 (4.57a) se ftom Foe ° cos) -1ol- 44 SELECTION OF CONFIGURATION PARAMETERS Before ending the general description of stability, a few conments are made in order to explain clearly the choice of the blade configuration parameters used in the numerical examples to come. These consist of the structural and inertia parameters EI, sR", EL, /n?R*, ms2R" and nd the geometric and aero Ca/anPR, Hyg Ky/Bs Ky Meg,» and KA/GS, and the geometric and se dynamic parameters c/R, B05 0» as a, andy. Herein only one control parameter, the pitch angle @, will be used to define the rotor operating condition (i.e. rotor thrust in hover). Its range is 0 < 8 <5. Values of these parameters will be chosen to facilitate com parison with other stability results in the literature [36,41]. For the aerodynamic and geometric parameters the Lock number y = 5.0 and solidity 6 = .10 are widely used values. Since o = be/aR, the choice of a four bladed rotor (b = 4) yields the chord ratio c/R= 1/40. The Precone angle 6,., which has an important influence on stability, will be given small values in the range -.05 < §,,< .10. The section lift curve slope a, and drag coefficient Sd, are given the values a = 2n, Sd,/a = 0.01. The structural and inertial parameters mainly determine the natural frequency of the blade, which in turn serve as rotorblade con- figuration parameters. However, for an untwisted cantilever beam with uniform mass and stiffness, the structural and inertial parameters can be expressed explicitly in terms of the uncoupled natural frequencies for the nonrotating condition, i.e. k 2 z g Berea ease Na ues) Beech os Sun, mR BP nr By’ makeRe yO? where 8, = 1,875104068712, and y, = 1/2, are standard parameters, and the subscript NR refers to fequencies of the nonrotating beam. In most design checks, the nonrotating frequencies are specified, as they may be determined by clamped-free vibration tests. Conversely, for most predesign studies, the rotating frequencies are specified. In this case, the rotating frequencies are used to find the nonrotating frequencies, through iterations. The bending rotating frequencies are determined with the exact method described in Chapter 3. The torsional rotating frequency, less sensitive to the effect of rotation, is determined with the approximate formula given in [19]. The torsional -102- frequency depends also on the intensity of torsion-tension coupling, which is proportional to k%/k2, and the inertia ratio Baya” These parameters, together with the mass radius of gyration k,/B and the root torsion K,R/GI axe fixed independently. Therefore any choice of the fundamental nonrotating frequencies, and of the values for ki/k2, k, /k, » 1 and k,/R, will yield the appropriate values of EI, /ni®R', ET, ma°R', cs /mo?R", for the equations of motion. Results for two practical con— figurations proves that the configuration parameters vary in a narrow band for soft inplane* blades. These are presented in the second and third columns of Table 4.1, The values for the hypothetical blade [36, 41] lie somewhere in between, as shown by the first colum of Table 4.1. Unless otherwise specified, these latter values will be also adopted here. + According to their dimensionless lag frequency, rotor blades can be (a) stiff inplane when w, > 1, or (b) soft inplane when a, <1. Only soft inplane blades are Yonsidered here. -103- TABLE 4.1 Configuration Parameters for Soft Inplane Rotor Blades Hypothetical blade | Practical Blade 1| Practical Blade 2 ref. [36,41] (fig. 3.8) (fig. 3.11) ry 0.576 0.623 0.4230 VNR 0.425 0.177 0.2440 1,908,4.732 3.500 8.6000 0.700 0.750 0.6580 1.150 1.130 1.0900 we 2.5,5.0 | 3.000 3.3000 | | | | 7 | | | (gf) 1,500 | 0.6185, | 1.0000 x, /R 0.025 0.0136 0.0148 kK Mk 0.000 0.0400 0.1450 ‘a, 1% =(kyR/GD) 0,1, 10,6 6.5000 0.4700 c/R 1/40 = .785 0.1350 0.1850 * Torsional frequency of non-rotating blade without root-torsion ** Torsional frequency of rotating blade including root-torsion -104~ CHAPTER FIVE SOLUTION OF NONLINEAR EQUILIBRIUM EQUATIONS In the previous chapter, the equilibrium and perturbation equations were derived from the general nonlinear integro-partial diffe~ tential equations of motion, The equilibrium equations are a set of nonlinear differential equations in space variable £, as described by equations (4.51) and (4.52). Since the solution of these equations, i.e, the equilibrium configuration v,(6), (6), ¢,(£), contribute to the coefficients of the perturbation equations (4.53), (4.54), it is important that a high accuracy equilibrium solution is obtained, for the stability results to be meaningful. Previous researchers have used the modal expansion method of solution, with one mode [46] or with several modes [36], [41]. Such an approach could only approximately describe the solution, Here a numerically exact technique is used, the Shooting Method. This numerical method can be applied to linear and nonlinear two-point boundary value problems, and to eigenvalue problems. The general form of the shooting method is presented in Appendix C. Here it is used to solve the nonlinear equilibrium equations (4.51), (4.52). The solution is derived for a range of parameters, and a comparison is made with solutions obtained by the method of modal expansion. 1 THE EQUIVALENT FIRST ORDER SYSTEM The system of equations (4.51), (4.52) are first transformed into an equivalent system of ten first order equations in the form of equation (C.13) in Appendix C. £0) The vector y is defined as "4s Oo9 Ohh G1) whereas the right hand side expression f(x,y) should be obtained by -105- formally solving for vs"(E), w,"(E) and (5). The algebraic difficul~ ties associated with a formal solution can be avoided by using a numeri-~ cal process. This was done here, At the time of evaluating Wi", wi", 9" all the lower derivatives are numerically known, and the diffe~ rential equations (4.51) can be conveniently written as 414 = ay, (5.2) Ap z a2 = 2 Mito ay ° a, 7 + vse tv) a = 42 6.3) 82 893, a = 86 = hy wm 9) + 2h, am gt — a EE a Gey 2h pe 21%o #6 * *ia¥o'®o Moa * MG) poe -106- The nonlinear terms in (5.3) were underlined. The algebraic system is solved numerically for ¥"", W" nonlinear right hand side is required, and thus » oN every time an evaluation of the 5.2 THE SOLUTION BY SHOOTING ITERATIONS The complete Algorithm is described in Appendix C, The range of integration is taken x, = 0, x, = 1, with the meeting point arbi- 1 ‘The boundary conditions specified by (4.51.4) trarily chosen as r = are enforced as ¥(0) = (05 0, Oy Gyr 0 0% Oy Os 4,55 5) YC) = {06s 075 05 0, Og» Gg» 0, 0, O49 0} where % = ( IME is the root torsion contribution to the boundary conditions, The unknown parameters Os (i=l, «4. , 10), are obtained by a Newton-Raphson iteration. An arbitrary initial guess o° is required for starting the process, the simplest choice being the null vector o° = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0} (5.6) With the guess (5.6) convergence is not always guaranteed. To overcome convergence difficulties, two schemes were devised. One is based on forced linearization of the equations by removing all the nonlinear terms underlined in (5.3). Then a pseudo-solution o” is expediently obtained based on the guaranteed convergence of a linear system. This pseudo- solution o” is then used as initial guess wich the fully nonlinear equations, and convergence is usually obtained. A better method is based on the influence of pitch angle © onto the nonlinear character of equations (5.3). (@ does not explicitly appear in (5.3), but it is contained in the calculation of parameters Ty» Ty, Typ, ete.) At zero pitch, 0 = 0, the equations become less nonlinear, and it was found that convergence with a null initial guess -107- (5.6) can always be obtained, Then the angle ® is increased in small steps 49;, with the rule ®, = 6, + A8,. At every value 9, the iterations are started with the solution 4-1 of the previous step, and, when convergence is obtained, the solution o; will be used to start the following step, Theoretically, any value of the parameter @ can be attained provided A@ is sufficiently small. Solutions for @ up to 0.5 radians have been satisfactorily obtained with a8 = 0.05 rad, for the values of parameters investigated. This method, resembling a com- bination of the Picard method in @ with a Newton-Raphson iteration at every 8;, was used to generate the nonlinear equilibrium solutions which are presented next. 5.3 NUMERICAL RESULTS A soft inplane configuration was chosen for numerical calcu- lations, The values of configuration parameters are those given in the first column of Table 5.1, These values are representative of actual blades, and permit a direct comparison with the results given in [36] and [41]. Since the results in this work were obtained with a direct numerical integration method, whereas the results of [36] and [41] were obtained with a Galerkin modal expansion method, a direct comparison of these two methods is informative. In fig. 5.1 the equilibrium configuration for lag-flap-torsion deflection of a uniform blade built-in at the shaft ie presented. The variation with pitch angle @ of tip lag, flap and torsion dimensionless deflections is shown. Increasing the angle 8 increases the aerodynamic steady forces acting on the blade, and generally increases the magnitude of equilibrium deflections. At zero pitch, the only load is the drag force, and hence the lag deflection retains a small value, whereas flap and pitch deflections vanish. At higher angles 8, the flap loading increases as the lift increases, and significant flap deflection can be attained. Due to the structural coupling in bending (bending about non-principal axes), the lag deflection increases as well. The equilibrium torsional deflection ¢|.., depends on a com bination of inertial and structural moments. Piret, the inertial moment, known as "centrifugal untwisting moment" to propeller designers, dominates the torsional moments at low angles of pitch, This term produces a negative equilibrium torsional deflection, Second, the structural torsion moments influence the equilibrium deflection at higher 6's, and tend to decrease its value. In the torsion equation, -108- 20} x 16 12 —+='> FIXED ROOT, few. = — + FIXED ROOT, APPROX,(36]. ROOT TORSION, f= 5 04 & 3 PITCH ANGLE, 8,RAD rip 03 with pitch angle Fig.5.1 Varlation_of equilibrium deflections +796 at zero precone,@, =.576, 425, By MR NR -109- the bending-torsion structural coupling terms are nonlinear products of the second derivatives of the bending deflections. Thus parameters influencing V" and Wi similarly influence the equilibrium torsion deflection, At high values of @ the V" and W% are large, and hence the nonlinear bending-torsion coupling becomes more important. The equilibrium deflections calculated by [36] are also plotted in fig. 5.1. For flap and lag deflections there is a close agreement between the two sets of results. However, for torsion the comparison of the two methods is poor, the present method giving an average twice as much tip deflection. The effect of root torsion on equilibrium deflections is also shown in fig. 5.1. The main effect is to increase the torsional deflection, and as much as 25% can be reached at maximum pitch. However, even this augmented torsion deflection represents only a small fraction of the total pitch angle, and hence the corresponding reduction of the bending deflections, V,|i,+ ¥|rip , is very small. The radial distribution of deflections, with and without root torsion, is shown in fig. 5.2. The only significant difference is found in the distribution of 4, where the root torsion produces a shift of the curve of elastic deformation to acconmodate the finite value of root rotation, ‘The influence of precone on the equilibrium deflection is shown in fig. 5.3, The preconed blade has a significant flap deflection even at zero pitch angle, due to the centrifugal force loading. The corresponding lag deflection is however unchanged, since at zero pitch angle there is no structual coupling in bending. At positive pitch angle, the effect of positive precone is to decrease all the deflections, whereas negative precone has the opposite effect. 5.4 COMPARISON WITH GALERKIN EQUILIBRIUM Several researchers [36], [41] have used a modal expansion (Galerkin) method to solve the nonlinear equilibrium system (4.51), (4.52). ‘This method will be sketched here. Assume vi) = 2 Mey(e ao ena eet ona, oF a Be #5(8) -L1o~ 20 10 —— — FIXED ROOT, feo —— Foor TORSION, f= 5 0 o DIMENSIONLESS RADIUS, § 0 eraser eres ‘0 10 4 0 10 +204 30 Fig.5.2 The radial distribution of equilibrium deflections with and without root torsion, @, =.576, w, 425 IR 7 bs NR Opn” +795 Bo, 2h 08. Tip 01 *olrip --02 Fig.5.3 V all- — =~ Be == 05 rod —— Bre = 10 rad = Pitch angle,@, rad ~ ~ x SS i ~ SN ~ ~ riation of tip equilibrium def: angle for a blade with preco: ow, =4.76, faa Onn -112- and upon substi ution into the equilibrium equations, a nonlinear algebraic system in unknown coefficients Vj, W. oi? *o is obtained. This system can be solved with the Newton-Raphson itera~ tion, After the algebraic solution is obtained for V,;, Wy;s ®, pags the equilibrium solution is assembled with (5.7). Two factors determine the accuracy of this method. First, the nonlinear algebraic system is solved by iterations, and second, the solution is expressed as a truncated series expansion. The first factor is common to the shooting method as well, and accuracy control is easily attained in the iteration process. However, the accuracy of the truncated series respresentation is outside direct control. This latter factor is a major shortcoming of the Galerkin method. However stringent the accuracy of the nonlinear iteration may be, the accuracy of the result will not go beyond the accuracy of the modal expansion. By contrast, the shooting method guarantees that the differential equa~ tions are exactly represented in the nonlinear algebraic system, and when solution is obtained, its accuracy is entirely controlled by the iteration acceptance test. This will be illustrated below with results obtained on a reduced system, the linear equilibrium equations for lag-flap motion. (The use of a linear system eliminates the need for Newton-Raphson iteration.) Fig. 5.4, shows the radial distribution of bending curvatures, ‘Vi, WY, as obtained with an increasing number of standard cantilever modes (N = 1,3,5) by the Galerkin method, and the exact distribution obtained by the shooting method. The exact solution predicts a high concentration of bending curvature towards the root; this is poorly represented with a low number of modes. Even for N = 5, the Galerkin method still departs by an average 10% (16% peak). Next the following integrals were computed tye PRO, VE as, Tet vi, 4 i VE aS) dt,...5N (5.8) where 5, 9; are weighting functions arbitrarily taken as the bending and torsion mode shapes of a standard cantilever. (the significance of these integrals will become apparent in the next chapter). Figure 5.5 shows the percentage error that can be found in these integrals versus the number of trial functions taken in the modal expansion. Convergence is rapid only for the first integrals, and for the higher integrals convergence is very poor. An unfavourable combination of weighting -1Lae 604 —+:- Nat 4 —'- N23) Galerkin —--N=5 Exact, present analysis 40 0 2 4 6 8 1.0 Dimensionless radius, & Fig.5.4 The radial distribution of equilibrium bending curvatures, WW. +6, W, =-4, Wy =00, f=00 NR WHR Our 50% (a) ERROR -50% 50% () ERROR -50% ® Fig.5.5 The error in modal representation (5.6) of equilibrium deflections measured as weighted integrals (5.7) (a) lag, (b) flap. -115- functions can amplify the error, and this accounts for the zigzaging behaviour of the curves in fig. 5.5. The above examples illustrate that the error introduced in the second derivatives of equilibrium solution (v¥ and WN) by the modal representation can propagate, and even get amplified, when certain weighted integrals are performed. Such errors can bear heavily on the stability results since the coefficients of the perturbations equations depend on the equilibrium solution and on its derivatives in a complex manner, as will be shown in the next chapter. Thus the use of shooting methods, which offer controlled accuracy, is advocated. 565 GENERALIZATION OF PRESENT RESULTS The solutions of the rotor blade equilibrium equations, as discussed above, are only an example from a larger class of problems that can be solved by shooting methods. Compared with alternative solution methods, such as Galerkin method, the shooting method not only offers controlled accuracy, but has certain advantages for the analyst. The method requires a minimum of problem analysis and program prepara~ tion, It is relatively easy to implement shooting methods on the computer using standard subroutines for the numerical integration of 0.D.E., solutions of linear algebraic equations, or even multipurpose shooting-method packages, as above. With a properly written code only one subroutine need be altered from problem to problem, the one in which the right-hand side of the system of differential equations, written in a standard form, is entered. More complex rotor blade equations could thus be easily tackled, and non-uniform rotor blade configurations could be analysed. The inclusion of uniformly varying properties will require a modification of the differential equations, whereas step changes in properties can be handled with a segmented model. A transfer expression can be set up on the segment boundary, and numerical inte gration can proceed from root to tip as before. -116- CHAPTER SIX SOLUTION OF LINEARIZED PERTURBATION EQUATIONS In the previous chapter, the equilibrium configuration of the blade was obtained by solving the equilibrium equations (4.51) and (4.52) with a very accurate numerical technique. Given the equili- brium solution VO» WCE) and ¢,(6) the coefficients of the pertur— bation equations (4.53) can be computed,and the perturbation equations can be solved for the perturbation motions AV(E,¥), AW(E,W) and A$, (Es). The Galerkin method of solution [60] was adopted for this purpose. 6.1 MODAL EXPANSIONS Separability of variables and a modal expansion are assumed in the form - Now av(ésy) = J avze’Yv, (5) 1 es ea ae aw(E,~) = J AWje"ws(E) A = tia (6.1) 1 Noy Ag(EW) = J bose ¥e, CE) 1 The choice of mode shapes is particularly important for the convergence speed of the method. Previous researchers [36] have used the standard mode shapes for bending and torsion of a nonrotating canti- lever beam. Such mode shapes are easy to handle as they have simple analytic expressions, but generally a large number of them (at least N= 5) is required for convergence. Better convergence may be obtained with modes that more closely represent the dynamics of a rotating beam. Hence the v;(£) and w;(£) mode shapes are the free vibration modes of a rotating cantilever beam. However the 9; (£) mode shapes are the free vibration modes of a nonrotating cantilever (with root torsion) as these were shown in Chapter 3 to be practically identical to rotating. torsion -1L7- modes. The mode shapes satisfy the following differential equations and boundary conditions*. 0 , v,(0)=v; (0)=0, vi =v) =} (0)=0, (6.2) 'f (O)=w'(0)=0. Aso + woe Oued $")-£4¢5) era) = 0. The solution to (6.2) vas already derived in Chapter 3, and the mode shapes were used here in numerical form, By including the effects of rotation on bending modes, and of root flexibility on torsion, it was hoped that a rapid convergence of the Galerkin process would be achieved. 6.2 TRANSFORMATION OF PERTURBATION EQUATIONS AND BOUNDARY CONDITIONS The modal expansion (6.1) is substituted into the perturbation equations (4.56) and in the boundary conditions (4.57) since application of the Galerkin method in the extended sense [38], [61] requires that the boundary conditions not satisfied by the mode slapes be added into the formulation. The boundary conditions on v,, w, at both ends, and the tip (E=1) boundary condition on 4; are automatically satisfied, hence only the root-torsion boundary condition appears in (4.57). After substitution of modal expansion, differential equations (4.56) and boundary conditions (4.57) are not exactly satisfied, since the repre- sentation of the solution is only approximate. The residuals, admit~ tedly small, are given for the differential equations by N | = YT tema)? + (ogre + ggg | apy (6.3) r % Equations (6.2) are related to the full equations (4.53), from which they can be derived, and the bending modes v,,w, represent the free bending vibrations of the system (4.53) at 2éropitch angle. -118- and for the boundary conditions by o| = (6.4) These residuals are then orthogonalized with respect to the modes shapes v;, w;, $;, 28 follows 1 Lvjeyae sO) 1,2, 0058 (6.5) 1 f gegas + 9,06, = 0 and the time dependence e*” cancels through. Thus (6.5) becomes a system of algebraic linear equations which can be expressed concisely using structural and aerodynamic stiffness, damping, and mass matrices Coyle and Kyo Gyslly)- nine submatrices, corresponding to the nine elements of the integro- These matrices are each blocked up into differential operators (4.56) and (4.57). Every submatrix is generated from one operator element by letting the modal subscripts vary from | to N. 6.3 THE GALERKIN MATRICES To reduce the order of differentiation, and hence the numerical error, integration by parts is applied. Adequate use of the clamped- free boundary conditions on V,, W,, vj, W;,, and the root-torsion-free boundary conditions on ¢, and ¢,, leads to certain simplifications as follows. The structural stiffness matrix, generated by the weighted integration of (4.55.c), contains terms involving the derivatives of the equilibrium axial force which become tae, viae, and similarly ~119- A 1 ~F4jO + A,f6; |] o}ae. In the Last expression the boundary residual corresponding to ky. in (6.4) (underlined) cancelled with boundary terms resulting from inte~ gation by parts. Other terms in the structural stiffness matrix involving higher order of differentiation are reduced using the fact that mode shapes satisfy (6.2). Whereas the terms involving equili- brium deflections are reduced by repeated integration by parts, as for example . 1 LG) db = vi (4Q¥5 pevuy 1 a f vydgvyaes 1 J vyQge "as = 1 = Jue Thus the structural stiffness matrix K, takes the simpler form (6.6.0) where the maximum degree of differentiation is two, though differentia~ tion up to the fourth degree was used in (4,55-d), Syumetry of K,, which can be easily verified for most terms, confirms that the original differential operator was self adjoint. The aerodynamic stiffness matrix K, generated from (4.55.e) is shown in (6.6.b). ‘The structural gyroscopic matrix C, is generated from (4.55.b) and it contains simple weighted integrals of the precone angle, and more complicated double integrals involving the equilibrium position. -120- (er9"9) EnQatly + Sa!@yy Fe face Hey fatty 's Car, fata tany bg } faehat : Li ee efaFapr = (+z) ey. Wy (a79"9) ° ° 0 | apfa(n - Baka fa apfanetaale f - i oe 1 as gS 7 feat p Zapl gp (Zs) fmcZ4y alg layin © | eeetet eer apio,a% J ‘aplap ayhacayealaclayta ff aplapZayfacZayoabactayta sp 1 ap'oMtaata f- ° ° 1 -122- The double integrals are processed through integration by parts and using the properties of area integration as shown in the example below. 1 1 7 i: 1 §%@ WG) [vj G4eQ] = EY HEY fv g gy 1 1 - [vp py Epes, {vj G@)46, ° g = = JJ VEG PW, )¥5 (645 6p & 1 1 =- f vj CEp)aby LEP ae, 1 ey e { vj (Ee, LM d¥E Cy) ae “LL, EDM EGE PHE a” 1 In the expressions above the 01 11 integration areas A, and A, are as defined in fig. 6.1. a 5 1.0 ' fig. 6.1 Integration areas for double integrals The caicellation of the time dependent boundary residual g,, in (6.4) is achieved in the calculation of the following term 1 1 1 - Ny #06045 CED] 2v5 (E46) " — 26; (0) 65 (0) ! vj (6p) 1 fu 1 1 = W2Ag: (E45 CE) i v;(Ep)d8,| + 24 f otesae, f 1 E a 1 -123- 1 5) J vj Gp)aby . 1 Ea)dbo = alae 1 1 = 2ng, (04500) J ° The final expression for the gyroscopic matrix(6.7.a)shows the expected antisymetry of a non-dissipative velocity proportional matrix operator. In previous derivations which ignored the effect of torsion on the axial strain [28], [36], the (1,3) term of C, did not appear, and to obtain antisymmetry the (3,1) term was ignored. This inconsistency was overcome in the present derivation, which yields a full antisym~ metric gyroscopic matrix. ‘The remaining matrices have simpler expressions. The aero dynamic damping matrix C, results by direct weighted integration of ‘A (4.55.c) and is presented in (6.7.b). The structural and aerodynamic mass matrices are both diagonal in the form vi ° 0 5 1 wo fas 0 “8; 0 > 0 0 w955 o 0 0 1 My = G fae} o oO}. o o ° i= 1,2, 0.5,N- (6.8) If orthonormal modes are used, the mass matrices need not be integrated and take the simple expressions Mo = [é,, 8;, ud;, ys imp: ) M Ue, 1i Mii] xt -£6.. i onecke: £fo -£6,, 0) , where 6,, is the Kronecker delta. 6.4 THE QUADRATIC EIGENVALUE PROBLEM ‘The matrix formulation of equation (6.5) Leads to the quadratic eigenvalue problem (49) ° 0 Zaplapclay faye. Fay f6 ve | | ey ee. apiale’s fz + 1 + 0 0 S Zgpl tar®ay°nclayfaT ; aplap(“ayin@ayinc'ay'a fz ° r t apiatng fe %sp!ap[ (a) Fac!) fan rv c ¥ Zsplap@ayfo@aycoctayta ffve- | aplap@ayfacaySactayta ffz- Zayfaclayta aoe ffe- -125- (we79) a(Coseyae + Fayia Faca-y ta z - “FaCPoe0)—] Fa Fang + 3(Cbse)-12a -126- 2+ C+ HE = 0 (6.9) where {AV BV ys s+ oy AVyys BN BN) 5 os y Aly 50,588), 6+4, 860}, mM, Ma 7 May = co Sy ES us ¢ The complex eigenvalues 1; are extracted by the method described in [62]. The system (6.9) is first rewritten as » = : (6.10) There, with the notations q = {r Ar} and 0 I ne the quadratic eigenvalue problem (6.3) is expressed as a standard eigen- value problem, Aq = 2g, of double the size, The real matrix A, is searched for complex eigenvalues, ~ 3, + ig, with an algorithm based on reduction to upper Hessenberg form by stabilized elementary similarity transforms, followed by QR eigenvalue extraction. This algorithm is available in the computer library of mathematical routines [63]. 6.5 PROGRAMMING CONSIDERATIONS The solution of the perturbation equations by the Galerkin method was coded in FORTRAN and a brief description of the program is given below. The mode shapes v;, W;, $;, i= 1, ... , N, used in the Galerkin transformation are calculated previous to running of the program with the methods of Chapter 3, and they satisfy exactly the differential equations (6.2). The equilibrium shapes V,, Wo, $, are calculated with the shooting method described in Chapter 5. They are numerically exact solutions to the equilibrium equations (4.51), (4-52). Along with the mode and equilibrium shapes, their derivatives up to the order required in the differential equations are also generated. Thus the values of -127- Oyedts dele, JAS and 4,548 as calculated at twenty-one equidistant points in the interval Osé<1, are all stored on a mass storage device, and are independently acces~ sable by the Galerkin program. After calculating some configuration and operation parameters, the Galerkin program enters the Gauss quadrature loop for integrating the structural matrices Mj, C, and K,. (If orthonormal mode shapes are used, the integration of M, is bypassed and M, is filled in directly). During the integration, the values of v.,wi.$;,¥ysW51$5» and their deri- vatives have to be evaluated at the Gauss points, and this is done by interpolation with Aitken's method of successive linear interpolations [63]. The linear integrals over the interval 0 < £ ¢ 1, and the area integrals over the area A, (fig. 6.1) are calculated with a I-dimensional variable-order Gauss-Legendre quadrature, and with 2-dimensional variable-order Gauss-Radau quadrature, respectively. ‘The aerodynamic matrices My, C,, Ky are calculated in a similar way. Then the total mass, damping and stiffness matrices, M, ¢, K, are obtained by sub tracting the aerodynamic matrices from the mass matrices. Finally, the complex roots of the system are extracted with the method explained in section 6.4, Stability is evaluated from plots of the complex roots (lyquist diagrams). The program was constructed on a modular basis, and is open to further refinements, For example, if rotating torsion mode shapes were to be used instead of non-rotating mode shapes, the change would easily be implemented by modifying the modes~bank, and redefining the appropriate integration by parts in the stiffness matrix K,. 6.6 NUMERICAL RESULTS AND STABILITY PLOTS 6.6.1 Torsionally Rigid Blade It is useful for understanding the general blade stability to first review the simpler stability of a torsionally rigid blade. The locus of roots for such a blade at zero precone are presented in fig. 6.2. These results were derived from three sources: (a) a flap- lag nonlinear analysis with nonrotating modes, similar to [16]; (b) the results of [36] which were derived from a complete analysis with non- ~128- 6=0. 2 Sad F112 1.10 FLAP +7-06 1.04, 1.02 7h (72 LAG wr nonrotating mode standard aerodynamics nonrotating modes rotating mode rotating modes or new aerodynamics nonrotating modes Fig.6.2 Locus of roots for torsionally rigid blade, @. VWI ot o 7 =.425, B70» VNR -129- rotating modes; and (c) the present analysis based on bending rotating modes. The analyses of parts (b) and (c) above wore made to handle a torsionally stiff blade by setting the torsional stiffness to a very large value, and making null the torsion equilibrium and mode shape inputs. Figure 6.2 shows that the configuration is generally stable, but the lag damping is much lower than the flap damping. At zero pitch, when the modes are uncoupled, the flap mode is strongly damped by the lift forces. Correspondingly, the lag mode derives its damping only from the much lesser drag forces. As pitch angle increases the coupling between the lag and flap modes effects a transfer of damping from flap into the lag motion. The differences between various types of analyses are pertinent. The inclusion of aerodynamic stiffness and mass matrices, and of second order angle of attack terms ("new aerodynamics") significantly influences the stability plots. The stability plots obtained with nonrotating modes shift considerably with the number of modes, at least five modes being required to ensure consistency. The present analysis with rotating modes converges much quicker, and good results were obtained with only a few modes (usually 1, maximum 3). 6.6.2 Complete Lag-Flap-Pitch Freedom The lag-flap stability characteristics are altered vhen torsional flexibility of the blade is introduced, due to the influence of equili- brium dependent nonlinear terms in the torsion equation. Figure 6.3 shows the root locus for lag-flap-pitch stability. As expected the effect of torsion frequency is larger when @, is lover. The flap mode damping is reduced, whereas the lag mode damping is significantly increased. The results obtained with the present method are also com pared with results obtained with five nonrotating modes in ref [36]. Again, the present method converges more rapidly, and good results are obtained with as few as 3 modes. 6.6.3 Influence of Precone The configurations studied so far were all stable. An example of blade instability is furnished by the stability analysis of preconed configurations. In the stability analyses of torsionally rigid blades [16] the effect of negative and positive precone was found to be stabilizing. But a complete lag-flap-pitch analysis gives a different result. Since a preconed blade has non zero equilibrium deflections el =130- vers * 2.6 P. ITCH \ 1 i 24 ' By 225 29 S 12 ” 1.0 ~~" BRESENT ANALYSIS 4 —— hd —+—N=5, NONROTATING MODES 136] 8 8:Q__9 25, rad Lac z Fig.6.3 Effect of torsion flexibility on the root locii, o 576, ©, =.425, @, NR pe’ -131- even at zero pitch (fig. 5.3), the weight of torsion~bending nonlinear terms is increased. Fig. 6.4 compares the lag stability plots of a soft inplane blade at precone angle 6, = .1 resulting from several analysis. Curve B was obtained with a torsionally rigid analysis and is stable for all 0's. Curve C corresponds to a torsionally flexible blade and is unstable for 0 <0 < .13. This new type of instability which occurs in torsionally flexible configurations is mainly due to the nonlinear torsion-bending coupling [36]. 6.6.4 | Influence of Root Torsion Since root torsion increases the magnitude of pitch deflection, it is expected that a torsionally flexible blade with root torsion will show more instability of the type described in 6.6.3, then the same torsionslly flexible blade built in at the root. This is illustrated in fig. 6.4, curve D. The region of instability, which for built-in blade extends up to = .13 rad, is increased by the inclusion of root torsion and covers the range 0 <@< .305, 6.6.5 A Simplified analysis Also plotted in fig.6.4 are results from a simplified analysis based on the linear complete equations [17] that can be used in a linear flutter analysis [6]. For the present program it can be easily veri- fied that the system of nonlinear equations reduces, when all the non- linear terms are removed, to the system of linear equations [17]. Inspection of equations (4.53) shows that all the nonlinear terms from the perturbations equations depend directly on the equilibrium solution, and hence their elimination can be easily attained with a simple change of input. That is, all the equilibrium modes are set to zero and the Galerkin program is run to generate a standard linear flutter analysis, and curve A was obtained in this way. Comparison of curves A through D indicates that the simplified analyses can be grossly in error when estimating stability of unstable configurations. However a simplified analysis may suffice for a stable configuration. This is shown in fig. 6.5 for an unpreconed blade for which the effect of nonlinear terms is less accute. 6.6.6 The Pitch-Lag and Pitch-Flap Coupling Coefficients Sone simplified analyses which ignored the torsional equation [9] have attempted to account for torsion effects in an approximate way 132 72 STABLE UNSTABLE “05 0% +03 “02 “Or “01 02 —— — A, complete linear equations. —+— 38, reduced nonlinear equations, lag-flap. C,D, complete nonlinear equations. Fig.6.4 Prediction of lag instability by different 133 — — — A, complete linear equations, — -— B, lag-flap nonlinear equations, C, complete nonlinear equations. €1 25 | Pitch 24 23 12 L/S Flap 14 Fig.6.5 Effect of analysis upon the roots locii of a stable configuration, ©, =.576, @,, =.425, = WR Bp 71 +908, Boots f=, 0-0 - .5 rad. -134- by introducing two scalar parameters to represent independently: pitch- lag and pitch-flap coupling. Thus a fictitious pitch perturbation is defined as = oav+ 48 o,AV + 8.0m , and extensive parameter studies in [64] showed that (a) positive 0, result in instability, and (b) stability margin increases with increas- i ative 8. ing neg: : In some practical blades 0, and 8, can result from actual kine~ matic links, but generally their origin lies with the structural non~ linear bending-torsion coupling. Hence approximate expressions for 8, and 0, can be derived as shown in [36] and in Appendix D. To check the usefulness of these expressions, scaled values of &, were plotted in fig. 6.6 for the same blade as treated in fig. 6.2 through 6.5. For a stable configuration such as §,, = 0, the pitch-Lag coupling is negative and increases with ©. This correlates well with fig. 6.3, and hence confirms conclusion (b) above. For an unstable configuration such a8 8, * +1, the piteh-Lag coupling @, is positive over a range of pitch angles, but this range does not exactly correspond to those of curves C and D in fig. 6.4, The regions of instability are overesti- nated, whereas the effect of root torsion is underestimated. These observations confirm that the equivalent pitch-lag coupling coefficient can predict the possibility of instability, but this prediction is only approximate for flexible blades. -135- “4 == 34 N N ~ ~ Qype 1, rad “2 S N 1 UNSTABLE . \ \ \ 8, rad oF aN ‘ \ —14 | N \ STABLE — —— Built-in, few Root torsion, f=5 Fig.6.6 Variation of equivalent pitch-lag coupling coefficient with pitch angle, @, =.576, @,, #.425, a NR “NR Dy =1.908. . On ~136- CHAPTER SEVEN DISCUSSION AND CONCLUSIONS The linear differential equations for uncoupled vibrations of rotating Timoshenko beams were shown to be significantly more difficult to solve than those for vibrations of stationary beams. A semi-analytic solution technique was developed for solving these equations exactly. The application of Frobenius power series method leads to bending and torsion rotating beam functions, and thus the computation of frequencies and mode shapes is simplified to the level of classic bean analysis. Parameter studies with a test beam showed that, with the exception of first lag frequency, the shear deflection and rotatory inertia correc- tions are small, do not appreciably vary with rotation speed, and are generally significant only for the higher frequencies (e.g. 4.3% in the third frequency). However, coupling between Timoshenko corrections and rotation speed is pertinent in the first lag frequency, and large reductions were obtained for high speeds. Though this effect amounts to only 1% for practical rotor speeds, large fans and some turbines may be significantly affected. The analysis of nonuniform blades was undertaken with the transfer matrix method applied in the standard way, but with rotating beam functions. Thus step-wise changes in properties are treated exactly, and practical rotor blades were analysed. The increase in frequency caused by rotation was found to be very large for flap, mode- rate for lag (124%), and small for torsion (6%). The percentage increase was much lower for higher modes, though the absolute change vas significant. The influence of rotation speed was generally less obvious on modal deflections than on frequencies, whereas bending moments and shear showed major differences produced by the radially varying centrifugal field. For the first flap mode this accounted for a 300% inerease in the root moment. Detailed computation of mode shapes showed that for complex blades the radial variation of stiffness and mass properties is reflected with fidelity by slope and curvature patterns, but not by the modal deflection. This illustrates the advan- -137- tage of the transfer matrix method over methods using numerical differen tiation. Point masses were also included in the analysis,and a method for dynamic balancing of rotor blades was illustrated. For one of the blades considered, a 200% increase in frequency separation and safe departure from resonance was achieved with the addition of only | kg at an optimum radial position. For stability analysis the Hamiltonian derivation of lag-flap- pitch nonlinear differential equations of motion was employed, and application of the ordering scheme led to second order expressions. Due to a difficiency in the ordering scheme some third order torsion terms were also retained. This derivation differs from [28] and [41] by using a simplified blade model and introducing root-torsion as a boundary condition. An additional torsion term was retained in the axial strain expression to restore consistency to the gyroscopic operator. Quasi-steady aerodynamic loads were specified including aerodynamic mass and stiffness terms [35] and second order bending effects on the local angle of attack [36]. The stability was investigated with the perturbation method which, unlike [33], [36], et al., was applied to the actual differential equations. This ensured exact derivations of nonlinear integro-ordinary differential equations for perturbations. One advantage of this approach is that several effects which would be associated with a direct nodal approach, such as parametric excitation or intermodal coupling, are not encountered, Another advantage is that equilibrium and pertur- ation equations can be solved independently, and through a good choice of methods propagation of error can be minimized . A numerically exact solution was obtained for equilibrium with the shooting method. Hence the coefficients of the perturbation equations were determined accurately. Alternative use of a Galerkin modal solution would have resulted, for a small number of modes (less than 5), in significant error being introduced in these coefficients and propagated into the stability results [36]. The perturbation equations resemble the simple linear equations of damped vibrations, and complex eigenvalues were expediently obtained with the extended Galerkin method [51]. After explicit inclusion of the torsion boundary residual in the orthogonalization process its elimination was achieved through integration by parts. Convergence was improved when rotating mode shapes were used for bending, as gene- -138- rated by the vibration analysis, and specification of root flexibility on the torsion mode shapes decreased the error associated with this unsatisfied boundary condition. Several stability studies were performed with uniform shaft~ fixed blades in hover, which were made representative of actual blades by matching the first lag, flap and torsion frequencies. The results for a torsionally rigid blade confirmed the small lag-stability margin of soft inplane blades at zero pitch [16]. The strong effect of non~ linear aerodynamics, and the improved convergence properties of the present method were illustrated. For torsionally flexible blades a shift in the stability locii took place, and a new instability was observed. This instability occurs at low pitch angles, and is due to nonlinear pitch-lag structural coupling which can become positive for positively preconed blades. Addition of root torsion to the blade model extended this instability to a larger range of pitch angles. In two correlated plots, the root loci, as predicted by several simplified analyses and the present analysis, were superimposed. Thus an overall view was obtained of the effect of developments and refine~ ments in the stability analysis, and the following conclusion can be dravn. For stable configurations in which stability depends largely upon the aerodynamic damping and lag-flap coupling, good predictions can be obtained from a simplified linear analysis. However stability of soft inplane configurations largely depends on the magnitude of non- linear torsion-bending coupling, and in such cases adequate answers can only be found through a complete nonlinear analysis. Torsionally rigid nonlinear analysis which ignores torsion-bending coupling was not found satisfactory for most of the configurations studied, i.e. soft inplane blades with torsion frequency less than 50. Since the foregoing stability results were obtained on a simplified blade, the conclusions have a limited relevance to the stabi- lity of actual blades. However, in the method of analysis care was taken throughout to assure that it contains ell the required ingredients for extension to nonuniform blades. This could be achieved by simple modifications to the FORTRAN as follows. qa The piecewise uniform segmented configuration employed for vibration analysis should be retained throughout. (2) The shooting method employed to obtain the equilibrium position should be modified to integrate segment by segment from root to tip. -139- @) ‘The Galerkin integrals should be performed segment by segment using the equilibrium position from step (2) and the mode shapes from step (1). ‘Thus the assembled matrices would repre- sent exactly the segmented blade properties. ‘The extraction of complex roots and stability plots then proceeds as before with no change. An automated search for flutter boundaries can be set up as an outside loop on the present analysis. For practical work on real helicopters further research and development on several aspects should be considered. For pitched blades lag and flap motions are of the same order, while chordwise deflections are an order of magnitude less than flatwise deflections. Hence an ordering scheme consistent for all pitch angles would result from the adoption of section principal axes in the derivation of the equations. Consideration of material damping is best achieved by choosing visco~ elastic integral operator stress-strain laws, which result in frequency dependent stiffness and damping terms, Improvements in the aerodynamic operator would result from introducing the effect of compressibility and fully unsteady coefficients as tabulated data [33]. For forward flight aerodynamics, the problem of periodic loading could be treated with the Floquet extension to Liapunov's theory [10]. -140- APPENDIX A APPROXIMATE FORMULAE FOR FREQUENCIES OF ROTATING UNIFORM BEAMS Cantilever uniform beams built in at the hub admit several approximate formulae for frequencies. For bending an empirical fit to certain expansion solutions yields [19] for the nth mode Fpap = 1 + Wig = ACQn-1) + aR +E where Uy, is the nonrotating (lag or flap) frequency and bar over quantities represents nondimensionalization with respect to rotation speed 2. Quantities A, and 8, are listed below:- n 1 2 3 4 >a 8 1.87510 4.69409 7.85476 10.9955 }(2n-1) A, 1419334 6.47823 17.8603, 36.9553 For torsion the use of the following Galerkin formula (modified for kg/k,, #1) is recommended [19] = 1 = & ML y+ 45 @n-1)2] . ‘Toston ~ “ye * | * ALG +z @n-1)*] -141- APPENDIX B PROPERTIES, FREQUENCIES AND MODE SHAPES OF BLADE 2 The rotor blade presented in fig. 3.11 has complex configu- rations, The blade characteristics [24] and the results of vibration analysis quoted in Chapter 3 are presented in this Appendix. Table B.1 lists the geometrical, structural and inertial characteristics of the 28 segments which constitute the blade model. SI units are used through- out, and Imperial equivalents are also given in places. ‘The first three mode shapes and frequencies for lag, flap and torsion are presented in the following figures for zero and operating speeds (f= 41.89 rad/sec). Figures B.1 through B.6 show the lag modes, figures B.7 through B.12 the flap modes. The torsional modes, with and without root torsion are shown in figures B.13 through B.18, and B.19 through B.24, respectively. The value of root-torsion spring was fixed at K, = 2.639 KNm/rad in order to match the first torsion fre~ quency uy = 125 rad/sec at zero speed. Without root-torsion, the torsion frequency would be 338 rad/sec. -142- 399°0|c-9z oros jos'z 418 sez jos-eos | svisz |evo-ofz"t ” s79°0 9°92 tus joz"e ocr ost josreze | ever |tyo-oj9rr e1 yas-o lores $*6z|s9"t ez sy foctezt | say |eeorols-e z iey'0|8°st ores |szre ast ss jorssr | s*7s —fororojy-0 u tty"0|8°81 eres jou” oat so forrser | s*9 — Joeorojz*t o Lyy"0f9° Lt eros |sere oe Sor jos*ssz | 19°66 — |oto-o} yo 6 ceno|z-2t erye|sery 19 ser joctoee | 9°ver jocorojz*t 8 Lov"0 0°91 erse|sery ez ozt jovsiz | 1°92 — fozo-o] 80 “ ta¢*0]2°St e°i {ort 6zz | og) | virze Zit — j€z0°0] 6"0 9 yoe-olerot 9-91 |c6ro oot as |y9ror es |jssovoje*z s g0c-olorzt 8°21 }00"1 18% oor | y991 ss jors-ojorz ’ ssz"o]o"ot 9-6 jort 169 oz [serv ors |tso-ojorz € y0z"0|0"8 gr izloz"t siti] o6e | serait ors |isorolorz z est‘o|o"9 ersz|sy"t zat] seo. |asviz sc | irorolart I zitofyry w ay8yler/at | mor} er-zatgor | wna [uresqrgor} mu furesaror | @ | er] @ | er “| z 2 ‘ yy u 9 evn "1a (avy “ta % x “on “89s Z pele 105 Soyazedozd Buypuog —[°€ TIEVE -143- y16ry |s*e6r 8z90"]e0t0"|se"a |ev'0 jevsz | sre — orsns ost fee's ork | soto|s-9 8% 69279 |o° ca 0€40"}9010"|s9"y Joz'o Jerez | st6 fortes Oca oi st | ovo jo-ee & 608°¢ Jovost | sz90"|z600"|62"9 [sero j6roe | gror forsee ouz gsviz SL | ofz*t |ovos | 92 sez Joroor | tszo"|1910"] 91" Jovro [arte | ort — jorvo6 ste p6rzz ove | s0s-o|o-oz | sz | 1€0°z Joro8 szgo"|orno'|19°9 |ze°0 lovee | stir jort96 see ferzz ore | soroloror % | } S291 j0"¥9 | szio'|zere |tv7o jevey | stor forvezt| ogy Joarez oro | 6zz*o love & | 96€°1 jorss €€10"|99°6 give | or9e jorset| 00. [et-oy orvt | czr-ojors w 692" 1 Joos oeto"| ory t98 | oro —jovvezt] oe” oeves o-0z | 94070 Jove 1 e61't force | e190°| oz10°| 09°61 seyt| 00s jovozet} 09% Jossznt | sere | o0ro|z"€ oz | | 660° ere” | ozeo"} 070 | oz*9s ges] s*zet jovtg6 ze jos" ose over | ecorolert 6t | 990°t |orzy y9v0"] 0°0 jogt9zjos*t jorost| 2799 jorstz SL Jovrosz | "26 | 120"0|a"z at | $66°0|z"6£ 9900") »900"/on"er /sevo jovtt | gre — fevwr s fsevot ors | vezro}s-or a |1ze-0}y°ez 9600°|8600"/ 01°91 06"0 jevoe | Lt01— [z04 at [oztor ory | ocoroyz"t | ou | |tesrolzeez | oz" Ze Joo'z lovoty sm | orsey ertst | ezo-ofero | | ost | | | 8990 |e"9z wo] ow | w/8afer/at| oa wre] wa pral eae [asap] = [als fa x ton 0g a | £9 corp 718 (ara) Pontrqueg 1" SISVL -144- a g 1 \¢ t | |e |e 2 | [es 18° eta ie 8 3 | [18 Vor Tee too Son tgs eae lara | > RADIUS—> || > RADIUS Leno-LA0 nave 1. 17-738 RRO/SEC [esooune none 1.” arTeeneorsec | aio > noneNT Vesa Rae Sao Roo Sa | nee oa Sa ie Ro LEAD-LAG HODE 1. 7.7 ea 17-738 RRO/SEC o HODE 1. Fig. 3.1 First lag mode of Blade 2 at St=0 -145- oeri—> >—RADIUS> LeRO-LAe MODE 2. 278.084 RAD/SEC >RADIUS—> LERO-LAG NODE 2+ 276-084 RAD/SEC 4 >seno.sLare—> LEAD-LAG NODE 2~ p—RAOLUS—> LEAD-LAG NODE 2 © 276 .084 RAO/SEC >—RADIUS—> 276.064 RAO/SEC Fig. B.2 Second lag mode of Blade 2 at {2 =0 -146- 8 $ pnnotus— Leno-L00 noe 2.” see.o79.anovsec | t | 1a i" 3 i et i i | snno1us— y—naotus— Leno-LAo moves.” s42.373.an0/sec LeAd-uR0 nove? $42,979 RAO/SEC Fig. B.3 Third lag mode of Blade 2 at &=0 -147- ba bs z 12 los Ta | Sates an er ee | ee eos = > vAD/ SE —— LEAD-LAG MODE 1. 27.568 RAD/SEC LEAD-LAG NODE 1 27-868 RAD/SEC_ not > nonent ‘Vlas iee nae a0 oe 0 Tao eas Stas ae S00 >—RADIUS—> > RADGUS—> LEAD-LAG NODE 1 27.868 RAO/SEC LEAD-LAG NODE 1. 27.888 RAO/SEC Fig. B.4 First lag mode of Blade 2 at Q=41.89 rad/sec -14a~ > BENO.SLoPE—> | —RADIUS—> | LEAD-LAG NODE 2. 290.804 RAO/SEC >—RADIUS—> LEAD-LAG NODE 2. 280. | RAO/SEC wot > wiot—> nomen sheer —ravtus—> LEAO-LAG NODE 2. 290.904 RAO/SEC > Rrotus—> 10 MODE 2+ 280.804 nAg/seC Fig. B.5 Second lag mode of Blade 2 at S=41.89 rad/sec -149- 1 gt ' | 3 + > RADIUS > RaDLUs—> LEAD-LAG MODE 3. _—S65.721 RAD/SEC | ean-tag none 3. 585.721 RAD/SEC 2g0.o > RaDtus—> »-Raotus—> LEAD-LAG NOoE 3. Se5-721 RRO/SEC__ | LeRD-LAG ODED 565.721 RAD/SEC Fig. B.6 Third lag mode of Blade 2 at Q=41.89 rad/sec -150- {os ¥ = - i | —_— 8 1? te Ss 1 g 3 1s | Taree Sago Bow Boos oa sto—oo > Rotus—> oRnotus-» FLAP MODE t “10.231 RAD/SEC “0 MOMENT >RAOIUS—> 10.231 RAO/SEC | FLAP MODE 1 | FLAP Mooe 1. 10.281 RAD/SEC t 1 ae ee ae Sean ea = Raotus—> FLAP NODE 10.231 RAO/SEC Fig. B.7 First flap mode of Blade 2 at % =0 ~151- t i 5 1 pnno1Us—> ruse nooce. —"serrsanovste ——__d | sige noe e. >—RAo1Us—> 52.795 RAD/SEC > RADIUS—> >—RAoTUs—> LFLmP nove 2. 82.775 RAO/SEC FLAP Mave 2. 52.775 RAO/SEC Fig. B.8 Second flap mode of Blade 2 at S=0 -152- oer >—seno.sLore—> RADIUS I »—RADLUS—> FLAP nove 9. __182.810 RRO/SEC_ FLAP_HOOE 3. 132.310 RAD/SEC >—Raptus—> 132.310 RAO/SEC, RADIUS [evap nooes. —_""WSesto novsee Fig. B.9 Third flap mode of Blade 2 at Q=0 -153- te 8 a ) ~haowus—> “fume noo.” 45-600 aR0/SEC = 4 doe a eB »aADTUS—> go RAD/SEC % x 12 + Tee eee Sao a pangs FLAP noOE 1 45 -600_RRO/SEC ee ot > sHEAR FLAP nOOE 1. »—Rrotus—> 45.600 RAD/SEC Fig. B.10 First flap mode of Blade 2 at 2=41.89 rad/sec -154- eri beno.sLore—> >—RAoIUs—> > RADIUS» FLAP MODE 2. 110.278 RAD/SEC FLAP nove 2. 110-278 RAO/SEC noe =y0-00_ gto 4p.00 —sHeaR 30.00 ~420.00 >—RADIUS—> > —RADIUS—> _FLAP MODE 2. 110.278 RAO/seC FLAP MODE Fig. B.11 Second flap mode of Blade 2 at Q=41.89 rad/sec FLAP HODES. __—_—194 698 RAO/SEC —Raotus—> > seno.sLore—> o—ano1us—> FLAP RODE 3. 194-698 RRO/SEC wit —nonenT FLAP KODE RADIUS 194 .688 RAO/SEC Fig. B.12 Tird flap mode of mot shear »—RAoLUs—> FLAP NODE 9. 194 .698 RAO/SEC Blade 2 at 241.89 rad/sec -156- Shed “a yh te Btw eid ‘Ste mien snows < —snon 3 1 ra ra i i Pa —sni0H x,, or x < x,. A boundary-value problem (henceforth BVP) for the same above system (C.1) is obtained by requiring that the dependent variables satisfy conditions at two (or more) distinct points. Thus if x, and x, are two extreme points, then general 2-point boundary conditions (linear) are specified as fuGe,) + BuGx,) = a (c.3) where A and B are n-th order matrices with constant coefficients and a is a constant n-vector, and solution is sought for x, < x < x). Solution of the initial-value problem has received a large interest in the past and existence,uniqueness and continuity properties are well fundamented [22]. For £(x,u) continuous on the domain x, ¢ x ¢ x)» ju] < © and satisfying the Lipschitz condition with some constant K, that is [£(x,u) - £Gu¥)| < Klu-y| for all (x,y) and (x,y) in R, then the initial value problem -162- wh = £660)» has a unique solution u = u(x,u,) defined on the interval [x,,x,], and this solutionis Lipschite-continuous in u,, uniformly in x. Solution of the boundary-value problem is based on the solution of the related initial value problem uo = £Gy), uy) = 5 (6.4) where g is a n-vector to be determined. In terms of the solution u(x,s) of the problem (C.4) we define a system of n equations = [As + Bu(x,s)] ~ (c.5) Clearly if 5 = 9 is a root of this equation, we expect that u u(x,8*) (C.6) is a solution of the boundary-value problem (C.3). In fact we now have [22], for £(x,y) continuousand Lipschitzian, as many solutions as there ro are distinct roots § of equation (C.5). Under certain condi~ tions listed in [22], pp 16-18,equation (C.5) has a unique root and then the boundary value problem has a unique solution even in the general nonlinear case. ‘he evaluation of the algebraic equation (C.5) and the search for its root forms the basis of what is known as "shooting methods". Some initial guess is assumed for the initial condition s = s, and solution of the related IVP is obtained through an IVP technique. The solution is evaluated at x = x, and a value is found for g(s,). In general g(s,) #0 and a correction és is applied, usually using Newton~ Raphson method. Then the process of integration is restarted with 41 7 85 + $8 and the whole procedure is repeated until g(s,) is suffi ciently close to zero. c.2 NUMERICAL METHODS FOR INITIAL VALUE PROBLEMS Out of the large variety of integration schemes for IVP, the two widely accepted,Kutta-Merson and Adam-Bashforth, are sufficient for illustrating the method. For numerical integration, the x-dimension is devided up into ij} The initial point x, is the first point of the net and the value of the function discrete intervals {h;} contained between net points -163- U, = u(x,) is assumed known, Then a set of values (U,} mist be con- structed at the nodes to represent the behaviour of the function u (x), and for convergence, consistency and stability of the numerical method the net function (U,) must become a close approximation tou (x) for (h,} sufficiently smati. The Kutta-Merson process is a modification of the 4th order one-step Runge-Kutta scheme suitable for automatic interval adjustment. Writing y, = the Kutta-Merson process is described by: ie yon Ut yn UD 2s 7 7) - , +4ay,) eee 0 * 2 Ps vg =U, +d be(x.sU.) + Sn.e(x, +4ayy,) +b hte, + hay,) =U 3 ees ies see ie) GRE o jet4 ajtl For a one-dimensional system, the error is estimated as follows [59] : if the interval is small enough, so that we can represent f(x,y) by the linear approximation f(x,y) = ax + by tc, Merson [59] showed that the error in y, is - 1a n5y) ean 20 (v) + and that in y, » So that a good estimate of the error in the computed 1 Peis is ~ 795 b’y py): This value can be used to adjust the interval automatically, usually by Y5» the accepted value of y at the end of the step, is ©, = a factor of two in a binary machine. For a n-dimensional system the error is calculated in each component and then the overall error is estimated as ‘The effect of non-linearities may overestimate the smallness -164- of the interval necessary to achieve the required accuracy; though underestimation is also theoretically possible, so far it has not been reported. The Adams~Bashforth process is an explicit multi-step scheme of the predictor-corrector type. Unlike Kutta-Merson, it requires values of the function at previous lattice points and hence it is non- self-starting. The general algorithm is described by the equations h * ; + of D5: Viet i ~ 98 5-3) ~ Predictor s ~ 598;., + 378, (c.8) 2] ~ Corrector = apne * - = Uy tay EGS UF) + 198; - 58 +E £0) 3ahy ‘The "Predictor" equation is used to estimate a value for the function * at the end of the interval, U;,,, which is then substituted into the jorrector" to obtain the refined value U,,,- The process can be used * and iteratively until sufficient agreement between is met. Ujar ond Ui (Wowever one pass usually suffices). The values U,, Ups 5 cannot be calculated by this algorithm and a starting scheme must be employed. Either a Runge-Kutta scheme, or @ Taylor expansion, can be used about the starting point x,. The predictor-corrector method has the advan- tage of controlled accuracy; however, previous steps have to be stored, it requires a separate starting procedure, and interval modification is not straightforward. Either of the above schemes, or any other more refined ones, can be used to solve the Initial Value Problem related to the Boundary Value Problem, starting with assumed initial conditions U, = s at the x, end of the interval and ending with an estimate for the function at the other end x, During the iteration for the exact initial conditions s* numerous passes from x, to x, will be performed, and it is advanta- geous to choose an efficient high order scheme in order to minimize computer time. c.3 NEWTON-RAPHSON ITERATION IN n-DIMENSIONS The numerical methods for solving nonlinear boundary-value problems lead to the problem of solving a nonlinear system of equations. Here we shall discuss the application of Newton-Raphson method to this -165- purpose [49]. Recall the set of algebraic equations (C.5) a(s) = 0 where $= {815855 +++ 8h 87 {gy (8,5595 +++ 9,)) If there is an approximate solution s to (C.5), a better approximation may be found by solving the set of linear equations Oo, (C.9) which can be written in the form G58 = -e(s) (c.10) where G, is the Jacobian matrix (0g,/28;}. The Jacobian can be calcu lated by direct differentiation of the expressions for the g,. Tf these are very complicated, or if § does not appear explicitly (our case), it can be easily estimated from the difference expressions ag, Se Day (8 9892585984 (14E Bj yee 8g) — BC] 98Q907298)] (cy 21085 (14C)) 9854p 098,) “BC, 9899 Tew, [By Spr8arro98 ip 8,0 where ¢ are arbitrary small perturbations in s, The first method required n+l evaluations of & the second requires 2n+1, since g(s) is also required in (C.10). Then s+ 6s is regarded as a better approximation than s if ||g(stés)||? < ||g(s)||2*. Tf this is not satisfied then the incement és is helved once or more tines as necessary until a value is reached for which g + 26g is a better approximation than s (see ref. [49],pp 256). Successive applications of this method results in either a point for which ||g||? is a minimum, or a solution of ||g||? = 0. 4 ‘THE NUMERICAL SOLUTION OF B.V.P. BY NEWION-RAPHSON ITERATION For the general nonlinear boundary value problem * Bucledian norm: ||g|| = Je?- -166- yl = £Guy), x Sx (13) with the Linear 2-point boundary conditions, Ay(x.) + ByG,) a = 0, (C14) solve the related initial value problem ut = £(xu), Rex su) = (415) with the arbitrary guess s” = ¢,. The integration from x, to x, is performed numerically with an IVP technique, and the value of the function at the x = x, is estimated as u(x,,8”). Substitution in the boundary condition(¢.14) gives BCs) = As’ + Bu(x 8°) - @ (o.16) Applying the Newton-Raphson method the Jacobian at point s is estimated as G(s”) = ag(s”)/as (C.17) with the forma (C.1I). For this purpose additional n integrations have to be performed with the perturbed starting values * = {s,s” vay, v v ae = (Yo83o vos Sh posh (te: 8hap9 cee SIs Geb2seeem + (C18) Once the Jacobian is evaluated the correction és” is the solution of the linear system G(s)ss¥ = -g(s”), and the new estimate gY*! = s¥% 4s” is used for a new integration, and the process is repeated. For s¥”! close enough to s” the Jacobian need not be re-evaluated and the process can be very quick. For a large difference between gY*! and s¥ the Jacobian must be recalculated for each iteration, and in general (nti) IVP's must be evaluated for each pass of a BVP iteration. For linear BVP the Newton-Raphson iteration produces the exact solution in one pass only, as shown below. Consider the linear differential equation = R@etfG@, x exe x (C.19) -167- Then 8) = As + Buys) and ag(s) aux, >8) GS) = Geo 7 AT BT (c.20) Differentiating (C.19) with respect to § the differential equations au! ou aux, »8) = kD) Gs) >a TE (.21) aux, 8) which is integrated from x, to x) to generate 3e . Since (C.22) 2u(x, 98) - is independent of s, ——t—— has a uniform value for all g, and so does G(s). Hence one Newton-Raphson step is sufficient to reach the exact solution’. ‘The above proof also illustrated an alternative way of finding the Jacobian, which can be applied to the general nonlinear BVP. Differentiating (C.13) with respect to s ' af Guu) a _— Gp (c.22) aus) and integrating from x = x,, with initial condition —=2— = I, we u(x, 98) 2 obtain at x= x, the required value——j— . This is substituted into (C.20) and yields G(s) directly. The process uses n integrations iy : +: i . (one for each column of 5) and computation-wise is equivalent to process (C.11). However it is more accurate, though it requires formal diggerentiation of f(x,y) with respect to u. c.5 PARALLEL SHOOTING Since an initial value problem can be integrated both forwards and backwards, there is no obvious reason for our integration of (C.13) to start always from x,. Tt could just as well start with a guess s at x, and be integrated backwards towards x,. A unified approach, which has the added advantages that all given boundary conditions are * In practice, for high accuracy work, two or three iterations may be needed to eliminate computational errors. -168- satisfied from the outset, and prevents integration instability by shortening the interval, is to estimate the unknown boundary conditions at both ends and integrate inwards to a meeting point. Changes can then be made to all the boundary values to make the two branches of the curve fit together. Tf (u;), and (u;), denote the values of u;(x,8) at the meeting point found by integration from right and left respec~ tively, then, ay) = Gude - Gd, Gon 1s oe a). (C.23) The vector g will now contain all 2n. values, But often boundary con~ ditions are imposed as specified values for the dependent variables (total n), and in this restrictive case the size of s becomes n. When the related IVP has solutions which increase rapidly, it is necessary to subdivide the interval [x,,x,] into even more subinter— vals, and integration can be done either with assumed conditions at x, and all intermediate points, and matching at the intermediate points and at x The size of the problemincreases with the number of subdivisions, and for m such divisions the size is in general n(m+l). c.6 ‘A GENERAL ALGORITHM We here shall attempt a generalization of the boundary value problem for the case when parameters other than boundary values are to be determined. These parameters may be eigenvalues, coefficients in the differential equations, length of the range, etc. The differential equation yl = £6yy,0) (.24) is to be solved on the interval

You might also like