Thèse de doctorat présentée pour obtenir le grade de Docteur de l’École Polytechnique par Elena Vyazmina Bifurcations in a swirling flow* * Bifurcations d’un écoulement tournant Directeurs de thèse: Jean-Marc Chomaz et Peter Schmid 13 juillet 2010 1 Swirling flow Introduction → Swirling flow → Vortex breakdown → Applications → Classification → Problematic Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives A flow is said to be ’swirling’ when its mean direction is aligned with its rotation axis, implying helical particle trajectories. 2 Vortex breakdown: definition Introduction Rotating cylinder, fixed lid: S. Harris Free jet: Gallaire (2002) → Swirling flow → Vortex breakdown → Applications → Classification → Problematic Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives Vortex breakdown is defined as a dramatic change in the structure of the flow core, with the appearance of stagnation points followed by regions of reversed flow referred to as the vortex breakdown bubble. • • • Main Features: core of vorticity and axial velocity stagnation point reverse flow or “recirculation bubble” 3 Applications Introduction Combustion burner Tornado → Swirling flow → Vortex breakdown → Applications → Classification → Problematic Numerical method 2D vortex breakdown 3D vortex breakdown Aeronautics Active open-loop control Summary and perspectives 4 Vortex breakdown: classification Introduction Bubble or axisymmetric form Double helix form → Swirling flow → Vortex breakdown → Applications → Classification → Problematic Numerical method 2D vortex breakdown Faler & Leibovich (1977) Faler & Leibovich (1977) 3D vortex breakdown Active open-loop control Spiral form Cone form Summary and perspectives Faler & Leibovich (1977) Billant et al. (1998) 5 Problematic Pipe Introduction → Swirling flow → Vortex breakdown Experiments: Sarpkaya (1971), Faler & Leibovich (1978), Leibovich (1978,1983), Althaus (1990), Escudier & Zehnder (1982)… → Applications → Classification → Problematic Numerical method 2D vortex breakdown Theoretical and numerical investigations: Squire (1960), Benjamin (1962,1965,1967), Batchelor (1967), Escudier & Keller (1983), Keller et al. (1985), Beran (1989), Beran & Culick (1992), Lopez (1994), Wang & Rusak and coll. (1996, 1997, 1998, 2000, 2001, 2004), Buntine & Saffman (1995), Derzho & Grimshaw (2002), Herrada & Fernandez-Feria (2006)… 3D vortex breakdown Active open-loop control Summary and perspectives Open flow Experiments: Billant (1998) Numerical investigations: Ruith et al. (2003) – 2D; Ruith et al. (2002, 2003, 2004), Gallaire & Chomaz (2003), Gallaire et al. (2006) – 3D Theoretical investigations: not so many… 6 Problematic: open flow, “no” lateral confinement Introduction → Swirling flow Boundary condition allowing entrainment! → Vortex breakdown → Applications → Classification →Problematic Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives Governing parameters Re rcore u x 0 rcore , S u 0 (rcore ) ux 0 - the radius of the vortex core; ux0 - the inlet axial velocity; u 0 - the azimuthal velocity; 7 Overview • Introduction • Numerical method • 2D (axisymmetric) vortex breakdown • 3D vortex breakdown • Active open-loop control: effect of an external axial pressure gradient on 2D vortex breakdown • Summary and perspectives 8 Introduction Numerical method Flow configuration →DNS →RPM →Arc-length continuation 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives Numerical method • Flow configuration • Direct numerical simulations (DNS) • Recursive projection method (RPM) • Arc-length continuation 9 Flow configuration Introduction Numerical method →Flow configuration →DNS →RPM →Arc-length continuation R 10rcore 2D vortex breakdown x0 20rcore 3D vortex breakdown Active open-loop control Summary and perspectives The numerical simulations are based on the incompressible time-dependent axisymmetric Navier-Stokes equations in cylindrical coordinates (x,r,) [ L] rcore , [ ] 1, [U ] u x 0 , [T ] rcore . ux0 10 Flow configuration R 10rcore x0 20rcore Grabowski profile (matches experiments of Mager (1972)) u x 0 (0, r ) 1, uniform flow ur 0 (0, r ) 0, u 0 (0, 0 r 1) Sr (2 r 2 ), u 0 (0,1 r ) S / r. Grabowski & Berger (1976) Flow configuration: open lateral boundary R 10rcore x0 20rcore Traction-free n 0 Boersma et al. (1998) Ruith et al. (2003) ur ( x, R) 0, r u x ur ( x, R ) ( x, R) 0, x r u u ( x, R) ( x, R) 0. r r Flow configuration: open outlet boundary R 10rcore x0 20rcore Convective outlet conditions (steady state) Ruith et al. (2003) u x u ( x0 , r ) C x ( x0 , r ) 0, t x ur ur ( x0 , r ) C ( x0 , r ) 0, t x u u ( x0 , r ) C ( x0 , r ) 0. t x u x ( x0 , r ) 0, x ur ( x0 , r ) 0, x u ( x0 , r ) 0. x Direct Numerical Simulation (DNS) Introduction Code adapted from the code developed by Nichols, Nichols et al. (2007) Numerical method →Flow configuration →DNS Mesh: • clustered around centreline in radial direction Hanifi et al. (1996) →RPM →Arc-length continuation 2D vortex breakdown Discretization: • sixth-order compact-difference scheme in space 3D vortex breakdown Active open-loop control Summary and perspectives Timestepping method: • fourth-order Runge-Kutta scheme in time • • • computation of the predicted velocity computation of pressure from the Poisson equation correction of the new velocity 14 Recursive Projection Method (RPM) Steady solutions with b.c.can be found by the iterative procedure: un+1=F(un), where F(un) is the “Runge-Kutta integrator over one time-step” F determines the asymptotic rate of the The dominant eigenvalue of the Jacobian J u convergence of the fixed point iteration || us u n 1 ||| | max || us u n || RPM: method implemented around existing DNS alternative to Newton! • Identifies the low-dimensional unstable subspace of a few “slow” eigenvalues • Stabilizes (and speeds-up) convergence of DNS even onto unstable steady-states. • Efficient bifurcation analysis by computing only the few eigenvalues of the small subspace. Even when the Jacobian matrix is not explicitly available (!) Recursive Projection Method (RPM) Reconstruct solution: un+1 = p+q=PN(p,q)+QF n n +1 Initial state un • Treats timestepping routine as a “black-box” DNS evaluates un+1=F(un) Picard iterations Subspace Q =I-P Subspace P of few slow & unstable eigenmodes un+1 =F(un) • Recursively identifies subspace of slow eigenmodes, P F(un) • Substitutes pure Picard iteration with DNS Newton iterations Newton method in P no Convergence? yes Steady state us Shroff et al. (1993) Picard iteration in Q = I-P • Reconstructs solution u from sum of the projectors P and Q onto subspace P and its orthogonal complement Q, respectively: u = PN(p,q) + QF Arc-length continuation Continuation of a branch of steady solution with respect to the parameter : • S F(u,)=0, where in our case • We assume that the solution curve u() is a multi-valued function of • At c • Pseudo – arc length condition F u, c ) det 0 u Newton iterations u u s 0 s s T • Full system F (u, ) 0 u u s 0 s s T RPM procedure: – Picard iteration in Q – Newton in other Introduction Numerical method Axisymmetric vortex breakdown →Transcritical bifurcation (inviscid) 2D (axisymmetric) vortex breakdown →Viscous effect →Resolution test • Transcritical bifurcation (inviscid) 3D vortex breakdown • Viscous effects Active open-loop control • Resolution test Summary and perspectives J. Kostas 18 Axisymmetric vortex breakdown: review Pipe flow • Non uniqueness of the solution u (r ) on the parameter S 0 core ux0 • Hysteretic behavior • Theory of Wang and Rusak for a finite domain Critical swirl Stability of the inviscid solution Viscous effect Beran & Culick (1992) Open flow ? Transcritical bifurcation (inviscid) open flow Base flow : Grabowski inlet profile q0(r)=(ux0(r),ur0(r),u0(r)) Small disturbance analysis q(x,r)=q0(r) +eq1(x,r)+…, q1(x,r)=(ux1(x,r),ur0(x,r),u0(x,r)) of Euler equations equation for the radial velocity ur1: Analytical solution: separation of variables ur1(x,r)=sinpx/2x0)F(r) ODE for F=F(r) and W=S2 2u 0 d ru 0 ) d 1 d rF ) p 2 2 W 2 F 0, dr r dr 4 x0 r u x 0 dr F (0) 0, dF ( R) 0 dr Eigen value problem on W W1=S12- the “critical swirl” . Solution q1 determined up to a multiplicative constant q1= Aq’1 Vyazmina et al. (2009) Viscous effects: asymptotics of an open flow Introduction Wang & Rusak (1997) showed in a pipe: regular expansion is invalid near W1S12 Vyazmina et al. (2009): non-homogeneous expansion for open flow Numerical method Axisymmetric vortex breakdown WW1+eW’, e2’, with W’=O(1), ’=O(1) →Transcritical bifurcation (inviscid) Linearization of Navier-Stokes →Viscous effect →Resolution test Three-dimensional vortex breakdown Active open-loop control Summary and perspectives q(x,r)=q0(r)+ e q1(x,r)+ e 2 q2(x,r) + … q1= Aq’1 Fredholm alternative e : L ur1=0 e 2: ur†1 | s 0 L ur2=s(q1,q0), Amplitude equation: A2M1+AW’M2+’ W1M3=0, with M 1 ur†1 | s 1 , M 2 ur†1 | s 2 , M 3 ur†1 | s 3 21 Viscous effect: asymptotics of an open flow Introduction A2M1+AW’M2+’ W1M3=0, Numerical method Axisymmetric vortex breakdown A W ' M 2 →Transcritical bifurcation (inviscid) W ' M 2 ) 2M1 2 4 ' M 1M 3 , | W ' | 2 M 1M 3 | M2 | ' →Viscous effect →Resolution test 3D vortex breakdown Active open-loop control Summary and perspectives Sc2 1 W1 2 Sc2 2 W1 2 M 1M 3 | M2 | M 1M 3 | M2 | , Obtain solution q1= Aq’1 22 Viscous effects: numerical simulations Re=1000 Importance of the resolution for high Re Resolution N1: NR =127; Nx =257 Other resolutions: N2=2N1; N3=3N1; N4=4N1 Point C: comparison N1 and N4 ? • • • Point A: N1 error 4 % N2 error 0.7 % N3 error 0.2 % • • • Point B: N1 error 2.5 % N2 error 0.4 % N3 error 0.1 % • • • Point C: N1 error 8 % N2 error 1 % N3 error 0.2 % Viscous effect, Re=1000: second bifurcation ? Introduction Numerical method Axisymmetric vortex breakdown →Transcritical bifurcation (inviscid) →Viscous effect →Resolution test 3D vortex breakdown Active open-loop control Summary and perspectives 26 Introduction Numerical method 2D vortex breakdown Three-dimensional vortex breakdown →Mathematical formulation →Spiral vortex breakdown Three-dimensional vortex breakdown • Mathematical formulation • Spiral vortex breakdown Active open-loop control Summary and perspectives Lim & Cui (2005) 27 3D vortex breakdown: short review Introduction Numerical method 2D vortex breakdown Three-dimensional vortex breakdown →Mathematical formulation →Spiral vortex breakdown Active open-loop control Summary and perspectives Spiral vortex breakdown has been observed • Experimentally: Sarpkaya (1971), Faler & Leibovich (1977), Escudier & Zehnder (1982), Lambourne & Bryer (1967) • DNS: Ruith et al. (2002, 2003) Transition to helical breakdown: sufficiently large pocket of absolute instability in the wake of the bubble, giving rise to a self-excited global mode Gallaire et al. (2003, 2006) 28 3D vortex breakdown: mathematical formulation Introduction Numerical method 2D axisymmetric state U U x ( x, r ),U r ( x, r ),U ( x, r ) ) is stable to axisymmetric perturbations 2D vortex breakdown Three-dimensional vortex breakdown 3D perturbations? →Mathematical formulation →Spiral vortex breakdown • Base flow is axisymmetric and stable to 2D perturbations Active open-loop control • Since the base flow is independent of time and azimuthal angle, the perturbations are Summary and perspectives u u( x, r )eim iwt , p p( x, r )eim iwt , where m – azimuthal wavenumber, w - complex frequency; the growth rate sRe(-i w ) the frequency Re(-i w ) 29 Spiral vortex breakdown: non-axisymmetric mode m=-1 S=1.3 growth rate vs Re Ruith et al. (2003) solved fully nonlinear 3D equations Re=150, S=1.3, m=-1 Introduction Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Effect of the external pressure gradient →Theoretical expectations • Theoretical expectations →Numerical results • Numerical results Summary and perspectives 31 An imposed pressure gradient: review for a pipe • Batchelor (1967): in a diverging pipe solution families have a fold as the swirl increased. • Numerically Buntine & Saffman (1995) showed the existence of bifurcation where two equilibrium solutions exist in a certain range of swirl below this limit level. • Asymptotic analysis of Rusak et al. (1997) of inviscid flow due to Rusak et al. (1997) the pipe convergence or divergence. • Converging tube Leclaire (2006) Introduction Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control →Theoretical expectations →Numerical results Summary and perspectives Leclaire (2010) 32 Pressure gradient: Theoretical expectations Carrying out the similar non-homogeneous asymptotic analysis with two competitive small parameters: and using dominant balance (e2’, =e2 ’) we obtain the amplitude equation in the form A2M1-AW’M2+’ W1M3- ’ M4=0, M4 did not calculated, since there is not analytical solution for the adjoint problem. A W ' M 2 2 4M1 ' M 3 ' M 4 ) 2M1 Sc2 1 W1 2 Sc2 2 W1 2 c W ' M 2 ) M3 M4 M 1 M 3 M 4 ) | M2 | M 1 M 3 M 4 ) | M2 | , | W ' | 2 M 1 ' M 3 ' M 4 ) | M2 | Schematic bifurcation surface , , Pressure gradient: bridging the gap Introduction Numerical method Schematic bifurcation surface 2D vortex breakdown 3D vortex breakdown Active open-loop control →Theoretical expectations →Numerical results Summary and perspectives 35 Pressure gradient: numerical results Re=1000 N3 N1 N2 N1 Does the steady solution exist down to =0? No, in the case Re=1000 N3 N2 N3 N3 N3 N3 N2 N3 Favorable pressure gradient delays vortex breakdown Introduction Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives Summary and perspectives →Summary →Perspectives 37 Summary • 2D: Bifurcation due the viscosity: numerical and theoretical analysis. • 3D: 2D stable solution is unstable to 3D perturbations. Spiral vortex breakdown, m = -1. • 2D: external negative pressure gradient can delay or even prevent vortex breakdown; – Bifurcation with respect to S and is more complex than a double fold Perspectives Introduction • Numerical method to find vortex breakdown-free state at S >Sc 2 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives 1 W Computations at higher Reynolds numbers Re ~ O • Asymptotic analysis with two competitive parameters and , determine the adjoint mode numerically • Compute 3D global modes of the adjoint Navier-Stokes linearized around the axisymmetric vortex breakdown state. Proceed sensitivity analysis • The slow convergence along the vortex breakdown branch →Summary →Perspectives Investigation of the stability of the solution 39 Perspectives: Supercritical Hopf bifurcation Introduction Numerical method 2D vortex breakdown 3D vortex breakdown Active open-loop control Summary and perspectives →Summary →Perspectives 40 Hopf bifurcation and period doublings perspectives Chaotic dynamics ? Merci pour votre attention! 42