Nonlinear Aeroelasticity for Limit-Cycle Oscillations Franco Mastroddi Daniele Dessi Luigi Morino Davide Cantiani Istituto Nazionale per Studi ed Esperienze di Architettura Navale Dipartimento di Ingegneria Meccanica ed Industriale Università ROMA TRE Bifurcation and Model Reduction Techniques for Large Multi- Disciplinary Systems University of
Summary of presentation 1. Normal Form (NF) approach for nonlinear (NL) (aeroelastic) systems experiencing bifurcations: highlights of theory for (aeroelastic) model reduction 2. NL analysis based on 3 rd order NF theory Appl.#1: LCO s control Appl.#2: Chaos control 3. NL analysis based on 5 th order NF theory Appl.#3: Gust response of a nonlinear aeroelastic system Appl.#4: Freeplay modeling: direct num. integration v.s. NF approach 4. Concluding remarks Page 2
Topics at a glance Aeroelastic System Fluid / Structure + Non - Linear 1st Level Physical Reduction 2nd Level Mathematical Reduction Modal Description / Finite State Aerodynamics Simpliflied Non - Linear Dynamics Complexity Our goal Reduced Complexity Page 3
Considered aeroelastic applications and relevance of NF Wing Panels in Supers. Flow Modal Model / / Finite State Aerodynamics Wing Typical Section (plunge, pitch, flap) in Subs. Unsteady Flow Piston theory for aerodynamics Nonlinearity Sources 1. Nonlinear pitch stiffness 2. Free-play in control surface 3-order Analysis Pitchfork Bifurcation Normal Form Reduction 5-order Analysis Knee Bifurcation Prevision of safe conditions flutter speed limits admissible gust PREDICT CONTROL Phenomenon intensity LCO amplitude Chaos behaviour Page 4
Authors papers related to the presentation Page 5
Some theoretical hightlights on NORMAL FORM (NF) approach for bifurcation -A tool for analysis of nonlinear (aeroelastic) systems with polynomial nonlinearities Ali H. Nayfeh, Method of Normal Forms, Wiley Series in Nonlinear Science A Reduced-Order Modeling for the Aeroelastic Stability of a Launch Vehicle IFASD 27 Stockholm 18-2 June 27 Page 6
Normal Form (I) Control parameter (e.g., flight speed) Nonlinear Aeroelastic Problem reduced to : Taking a Taylor expansion of around an equilibrium point the dynamical system can be rewritten as: nonlinear-term vector of type (generic NL s can be locally reduced in this form) : Assuming analytically dependent on, one obtains Similarily: Page 7
Normal Form (II) Setting the transformation (diagonalize the linear-ized part), one has ( diagonal matrix of eigenvalues ): where z = ²u The ordering parameter is introduced such that. Motivation: scaling the contributions of any terms in each equations on the base of the amplitude of the original state space variable. Balancing the nonlinear terms with the perturbation on the linear terms (condition for a local bifurcation) implies Page 8
Normal Form (III) Recasting equations in the new state space variable or The normal form method consists of simplifying the differential problem through the near identity coordinate transformation where have to be chosen so as to simplify the problem. Page 9
Normal Form (IV) Substituting ( ): Collecting terms of same order where NEW nl term DIFFERENCE OLD nl term Choosing for the same functional dependence as The following expressions are obtained for coefficients of (near)-resonance conditions Suitable small positive parameter defining the resonance conditions Page 1
Normal Form (V) RELEVANT ISSUE #1: opportunity of defining the near-identity transformation Given system ˆ : j j i 1 Pseudo-diagonal system N Near-Resonance condition ˆ ˆ Normal form y Λy g (2) () y Nonlinear transformation z y w (2) ( y) In the complex plane, the -points can be plotted for every nonlinear term (green for 3rd order terms, blue for linear terms); by enlarging the circle of radius more nonlinear terms are taken into account until a satysfying solution is reached. The radius is obtained by a trial and error procedure. ˆ Page 11
Normal Form (V) RELEVANT ISSUE #2: LCO solution analytically obtained (resonance conditions) Third order system Directly related to the linear part Directly related to the lnoninear part Consider the pair of complex conjugate equations given by the manifold Looking ANALYTIC for SOLUTION a solution FOR of LCO the type: Around a complex conjugated stable Complex eigenvalue The system becomes: Page 12
Normal Form (VI) (the other equations are slaves Consider the other eigenvalues Stable Complex eigenvalue Stable Real eigenvalue Final analytical Solutions Page 13
Normal Form (V) RELEVANT ISSUE #3: othe stability of LCO can be discussed (Hopf theorem) Page 14
Aeroelastic applications background: linear case Flow Speed System parameter Unsteady Load Fluid (linear) Gust Load Excitation mechanism Structure (linear) Elastic Displacement Initial Values Page 15
Background: nonlinear case (1) Flow Speed > < Critical speed Unsteady Load Fluid (linear) Structure (NONlinear) Nonlinear structural stiffness arising from large displacement gradients Freeplay in the control surfaces Elastic Displacement UNStable Limit Cycles Hopf (pitchfork) bifurcation: subrcritical-detrimental flutter Hopf (pitchfork) bifurcation: supercritical-benign flutter Page 16
Physical model Unsteady potential incompressible unbounded 2D flow hypothesis Nonlinear structure with torsional and bending modes of vibration. Uniform and constant flow horizontal velocity + vertical deterministic gust Vibrating structure with nonlinear (soft) torsional spring Aerodynamic augmented states DOFs Heave & Pitch Page 17
Methematical model: A simple (very used ) aeroelastic model: Typical Section With Control Surface A three degree of freedom aeroelastic typical section with a trailing edge control surface NONLINEAR TERMS Page 18
APPLICATION #1 based on NF theory: control the LCO amplitude and taming of explosive flutter Nonlinear aeroelastic system with a nonlinear control Then the previous Hopf analysis can be repeated with a new closed-loop nonlinear coefficient: with Nonlinear feedback Condition for stable limit cycle in closed loop conditions NB: it is always possible to choose such as to satisfy previous Eq. it is possibile to control the LCO amplitude by a nonlinear feedback Page 19
APPLICATION #1: LCO nonlinear control Linear system unstable response envelope Unstable linear system nonlinearly controlled: LCO envelope Unsteable linear system with nonlinear control Different LCO amplitude with different values for non-linear gain Different LCO amplitude with different values for non-linear gain Ref.: Morino, L., Mastroddi, F., Limit-cycle oscillation control with aplication to flutter, The Aeronautical Journal, Nov. 1996. Page 2
APPLICATION #2 based on NF theory: periodic, quasi periodic, or chaotic solution by transforming the original problem with different Periodic orbits Harmonic Limit Cycle Oscillations Quasi Periodic orbits Non-periodic Oscillations Strange Attractors Chaos Page 21
APPLICATION #2 based on NF theory: periodic, quasi periodic, or chaotic solution by transforming the original problem with different Classic nonlinear Panel flutter Simply supported panel in a supersonic flow with dynamic pressure a buckling load R x and structural stabilizing nonlinearities PDE system solved a la Galerkin assuming for the solution with Page 22
APPLICATION #2 based on NF theory: periodic, quasi periodic, or chaotic solution by transformin the original problem with different Nonlinear Panel flutter = =1 =5 The greater is the assumed radius, the less simply harmonic the solution is the NF is able to identify the NL terms which are responsible of chaotic solution Ref.: Morino, M., Mastroddi, F., Cutroni, M.,``Lie Transformation Method for Dynamical System having Chaotic Behavior, Nonlinear Dinamics, Vol. 7, No.4, June 1995, pp. 43-428. Page 23
Some theoretical hightlights on (fifth order) NORMAL FORM (NF) approach -A tool for nonlinear analysis of aeroelastic system with polynomial nonlinearities -. For more complicated bifurcation. Precritical instabilities. Ali H. Nayfeh, Method of Normal Forms, Wiley Series in Nonlinear Science Ref: Dessi, D., Morino, L., Mastroddi, F., ``A Fifth-Order Multiple-Scale Solution for Hopf Bifurcations," Computers and Structures, Vol. 82, 24, pp. 2723-2731. Page 24
Fifth-order NF Normal Form (very brief details) Considering the fourth order terms the following equations need to be simplified by the NF procedure: Again the NF-method consists of searching for a new state space coordinates through the near-identity transformation The transformed dynamical system will be in the form By using the coordinate transformation Page 25
Loss of Stability for Nonlinear Aeroelastic Systems Eigenvalue crossing of the imaginary axis (linear analysis) Pitchfork/Hopf bifurcation: supercritical (benign flutter) and subcritical (explosive flutter) Beyond the Hopf bifurcation: knee bifurcation (precritical LCO s) Page 26
Aeroelastic interpretation Unsteady Load Fluid (linear) Turning point Flow < Speed Flow Speed > Flutter < Flutter speed speed Structure (NONlinear) Elastic Displacement Wind tunnel model with free-play Hopf (knee) bifurcation: Wind tunnel model with transonic effects subrcritical to supercritical UNStable & Stable Stable Limit Cycles Limit Cycles Page 27
Physical Sources of Structural Nonlinearities in the Fixed Wing Model Freeplay in the control surfaces - bilinear stiffness due to loosely connected structural components (Conner et al., 1997) Nonlinear structural stiffness arising from large displacement gradients (Lee, B.H.K., et al., 1989) Approximation with Polinomial Nonlinearities 8 states variables (aerodynamic ones included) Page 28
Nonlinear Analysis: 3 order v.s. 5 order NF using = M ( ) c 3 3 No 3 order NF analysis is not able to describe the special kind of bifurcation The NF-method has to be extended to fifthorder to capture at least the qualitative behavior of the aeroelastic LCO in the case of knee bifurcations. Page 29
Nonlinear Analysis: 5 order NL using > M ( ) c 3 3 To improve the accuracy, more equations and terms have to be added. Including the near-resonant terms, an extended (in the sense of Center-Manifold) NF-method is implemented Page 3
bifurcation curve.35 A detail on this analysis: exact.3 Normal 5th order Normal 3rd order M( ) c3 3 LCO amplitude.25 exact Normal 5th order Normal 3rd order.2 exact Normal 5th order.15 Normal 3rd order.1 Third order and fifth order normal form (CM) provide a satisfactory approximation.5.2 1 LCO, IC>aLC 2 3 4 5 6 7-3 x 1.15 LCO, IC<aLC.15.1.1.5 LCO, IC>aLC.3.2.1 Time History.5 Time History ( ) -.5 3rd order approximation 5th order approximation numerical -.1 -.1 -.5 -.15 -.2 -.2 2 -.1 -.15 5 3rd order approximation 5th order approximation exact 1 15 2 25 t(s) -.3 3 35 4 45 2.1 2.2 5 2.3 1 2.4 15 5 2 2.5 t(s) 25 t(s) 3rd order approximation 5th order approximation 2.6 exact 2.7 2.8 2.9 3 35 4 45 21 5 Page 31
As sub-product of this analysis: identified a parameter moving the system from Hopf to knee-bifurcation M ( ) c 3 3 Plunge Bifurcation Diagrams for different elastic axis positions a h Fisically: the elastic axis position evealed to havea clear influence on the type of bifurcation, subcritical knee-like bif. or supercritical pitchfork bif. Page 32
NF METHODOLOGY/APPROACH - The nonlinear analysis for gust-excited problem Dessi, D., Mastroddi, F., A nonlinear analysis of stability and gust response of aeroelastic systems, Journal of Fluids and Structures, 24 (3), p.436-445, Apr 28 Page 33
CASE A: Initial conditions x A( U ) x f ( x) g( ;U ) no gust excitation x() x THIS IS THE CASE PRESENTED BEFORE! Page 34
CASE B: APPLICATION #3 No initial conditions x() discrete (=time impulsive) gust excitation x A( U) x f( x) g( ; U ) Page 35
CASE B: Gust excitation with no IC In the case of the gust problem, the system dynamics is excited only by the discrete gust profile with zero initial conditions x A( U) x f( x) g( ;U ) x( ) It is assumed a discrete gust of the wave form w ( ) G w 2 1 Cos G where: Gust intensity Gust gradient.5.4.3.2 v w -.5.1 -.1 -.1 2 4 6 8 1 12 1 2 3 Page 36
2D gust-parameter space Gust intensity Gust analysis Fix the gust gradient G to a value.4 Define a region in the plane of the parameters and consider a matrix of numerical simulations for each value of the grid nodes 4.925 4.93 4.935 Speed.2 -.1 Introduce the logarthmic damping coefficient between consecutive peaks in the amplitude modulated periodic solution.6.4 n 1 log x (max) i ( n n 1 ) 1 x (max) i n ( n ).2 -.2 14 15 16 17 18 19 Page 37
Gust analysis For each run, consider the logarithmic damping for n N, and obtain the response surface N N U,w ; G The level curve N can be determined by interpolation between the grid nodes It represents the set of parameters (gust intensity for a given flight speed) that leads the state-vector to be very close to a stable or unstable periodic solution (LCO) Z Line of critical gust intensity Undamped oscillations Y X The level curve does not depend on the final time step N Damped oscillations.4 N.3 Gust intensity.2 Primary critical speed Secondary critical speed.1 4.925 4.93 Speed -.1 4.935 Page 38
Basin of attraction: critical gust intensity Gust intensity N,w ; U w / The branch of the curve N G for which N across it, ( c ) ( c ) represents the critical gust intensity w w (U ) characterized by the property that w w w w ( c ) ( c ) the solution is undamped the solution is damped.5 Upper basin of attraction of stable LCO.4.3 Lower basin of attraction of stable LCO.2.1 Basin of attraction of fixed point solution 4.926 4.928 4.93 4.932 4.934 4.936 4.938 Speed Thus, a basin of attraction for the solution has been identified in the space of physical parameters. This analysis has to be repeated for several gust gradients G Page 39
and what happens if the nonlinearities are not polynomials?... For example: How is important to correctly model the discontinuity of a freeplay? (i.e., avoiding polynomials) Page 4
Force/Moment Force/Moment APPLICATION #5: physically more realistic modeling freeplay for Freeplay/Hysteresis Nonlinearity DOMAIN SUB-DIVISION: Zone 1 & Zone 3 15 1 5 zone 2 zone 1 Preload zone 3-5 -1 15 1 hysteresis zone 2,, c -15-4 -3-2 -1 1 2 3 4 displacement/rotation Zone 2 & Zone 4 5-5 Ampl. zone 1 Preload zone 3 zone 4-1,, c -15-4 -3-2 -1 1 2 3 4 displacement/rotation Page 41
frequency (Hz) LCO amplitude Freeplay Results: direct numerical integration.3 Bifurcation graphic Amplitude of LCOs increases with flow speed and activated between the two linear stability limit.2.1 -.1 -.2 8 characteristic frequency -.3 5 1 15 2 25 3 flow speed (m/s) 7 6 5 4 3 2 1 First frequency max PSD Second Frequency max PSD (linear) Flutter speed flap disconnected (linear) Flutter speed flap connected Second Frequency characteristic 3 times the first -1 5 1 15 2 25 3 speed flow (m/s) Page 42
How to apply Normal Form in this case? (1) Moment/Force LCO Amplitude.3 Approximation Freeplay must be approximated by a polinomial nonlinearity in order to perform the NF analysis.2.1 -.1 -.2 freeplay cubic.1.5 -.5 cubic approximation Vs freeplay -.1 1 2 3 4 5 6 7 8 9 t(s) cubic approximation ZOOM freeplay.1.5 -.3 -.4-5 -4-3 -2-1 1 2 3 4 5 Rotation/Displacement x 1-3 The cubic approximations give satisfactory results for the amplitude and the frequency of LCOs. However the transitory is quite different -.5 -.1 7 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 71 t(s).3.2 Bifurcation graphic cubic freeplay All the polynomial modeling for the freeplay discontinuity works locally well for LCO (around bifurcation point) but.1 -.1 -.2 -.3 5 1 15 2 25 3 flow speed (m/s) Page 43
How to apply Normal Form in this case? (2) all the equations must be used ( <4) to obtain a correct approximation for LCO x 1-3 5 Approximation via Normal Form 4 3 2 1 freeplay cubic approximation Normal Form 5th order -1-2 -3-4 -5 43.2 43.25 43.3 43.35 43.4 43.45 43.5 43.55 43.6 t(s) Page 44
CONCLUDING REMARKS 1/2 (from ourown experience in nonlinear aeroelasticity) Once a system can be mathematically modeled by nonlinear first order differential equation with polynmial nonlinearities (many aeroelastic systems can be modelled so), NF approach may represent a local powerfull tool o To obtain the analytic solution around a Hopf bif. o To reduce the system size to a restricted set equations (Center Manifold theorem) o To study the LCO stability in precritical condition (with a higher order analysis (basin of attraction defined with I.C. or with suitable input) o To find a nonlinear feedback to tame linear and nonlinear oscillations o To identify the nonlinear contributions responsible of chaotic behavior Page 45
CONCLUDING REMARKS 2/2 If system nonlinearities are not in a polynomial form (it is the case of the freeplay modeling) o A polynomial approximation of the nonlinearity could be used and the NF approach can efficiently capture the nonlinear behavior of the system if near-resonace terms are included o An extension of the NF theory should be developed (something is existing like Lie Transformation) in this case Comment The freeplays nonlinearities can be trivially identified but not so easily analyzable by NF approach Polynomial nonlinearity are (not so-trivially) analyzable by NF approach but are not identifieable by actual measurements at all Page 46