HAL Id: hal-01959233
https://hal.inria.fr/hal-01959233
Submitted on 18 Dec 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Optimisation de forme pour la conception des cavités de liner
Gilles Tissot, Gwénaël Gabard
To cite this version:
Gilles Tissot, Gwénaël Gabard. Optimisation de forme pour la conception des cavités de liner. CFA’18 - 14ème Congrès Français d’acoustique, Apr 2018, Le Havre, France. pp.1-30. �hal-01959233�
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 1/17
Shape optimisation for acoustic liners design
GillesTissot, GwénaëlGabard
Laboratoire d’Acoustique de l’Université du Mans
Congrès Français d’Acoustique Le Havre, Wednesday, 25 April 2018
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 2/17
Context
Acoustic liners:
Schematic representation.
Integrated in nacelles of aricrafts.
Constraints:
Low frequency absorption requires deep cavities.
Challenge to push away this limitation.
MACIAANR project: LAUM, SAFRAN.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 3/17
Context
Objective:
Optimise cavity shape of liners.
Make it resonate at lower frequencies.
Strategy:
Shape optimisation.
Based on frequency response.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 4/17
Outline
1 Governing equations
2 Shape optimisation
3 Numerical implementation
4 Results
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 5/17
Governing equations
Helmholtz equation:
∆p+k2p= 0 inΩ
∇p.n= 0 onΓs (free slip).
∇p.n=δvi−12 ∆Tp−δTk2(i−1)(γ−1)2 p
onΓw (walls)
∇p.n+ikp= 2ikeikL onΓZ (Impedance B.C.).
Γs
L
Bossart et al. (2003) Berggren et al. (2018)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 5/17
Governing equations
Helmholtz equation:
∆p+k2p= 0 inΩ
∇p.n= 0 onΓs (free slip)
∇p.n=δvi−12 ∆Tp−δTk2(i−1)(γ−1)2 p
onΓw (walls).
∇p.n+ikp= 2ikeikL onΓZ (Impedance B.C.).
δv= q 2
ωRea ; δT = q 2
ωReaP r.
∇f=∇Tf+∂f
∂nn;
∆f= ∆Tf+∂2f
∂n2 +∇T.n∂f
∂n
Γs
Γw
L
Bossart et al. (2003) Berggren et al. (2018)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 5/17
Governing equations
Helmholtz equation:
∆p+k2p= 0 inΩ
∇p.n= 0 onΓs (free slip)
∇p.n=δvi−12 ∆Tp−δTk2(i−1)(γ−1)2 p
onΓw (walls)
∇p.n+ikp= 2ikeikL onΓZ (Impedance B.C.).
Unitary amplitude of incoming plane wave.
Reflexion coefficient:
R= 1
LZ Z
ΓZ
p(x, L) dx−eikL
eikL.
Impedance: Z= 1 +R 1−R.
ΓZ
Γs
Γw
L
Bossart et al. (2003) Berggren et al. (2018)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 5/17
Governing equations
Helmholtz equation:
∆p+k2p= 0 inΩ
∇p.n= 0 onΓs (free slip)
∇p.n=δvi−12 ∆Tp−δTk2(i−1)(γ−1)2 p
onΓw (walls)
∇p.n+ikp= 2ikeikL onΓZ (Impedance B.C.).
Modify cavity shape for improving behaviour: Target impedance.
J(Ω, p) = 1 2
Z k2
k1
|Z(p, k)−ZT(k)|2dk
ΓZ
Γs
Γw
L
Bossart et al. (2003) Berggren et al. (2018)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 6/17
Shape optimisation
GeneralitiesInfinitesimal domain variations:
Reference domain:
Ω
Shape variation:
Ω +hδθ
h small.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 7/17
Shape optimisation
GeneralitiesCost functional: unconstrainedminimisation problem
J(Ω) = Z
Ω
f(x) dx Shape derivative: ∀δθ we have
(∇δθJ, δθ) = lim
h→0
J(Ω +hδθ)− J(Ω)
h =
Z
Γθ
δθ(x).n(x)f(x) dx.
Descent direction:
J(Ω +hδθ) =J(Ω) +h Z
Γθ
δθ(x).n(x)f(x) dx+O(h2)
⇒δθ(x) =−f(x)n(x) ensures a descent direction! Indeed: J(Ω +hδθ) =J(Ω)−h
Z
Γθ
f2(x) dx+O(h2)
Allaire (2003,2004)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 7/17
Shape optimisation
GeneralitiesCost functional: unconstrainedminimisation problem
J(Ω) = Z
Ω
f(x) dx Shape derivative: ∀δθ we have
(∇δθJ, δθ) = lim
h→0
J(Ω +hδθ)− J(Ω)
h =
Z
Γθ
δθ(x).n(x)f(x) dx.
Descent direction:
J(Ω +hδθ) =J(Ω) +h Z
Γθ
δθ(x).n(x)f(x) dx+O(h2)
⇒δθ(x) =−f(x)n(x) ensures a descent direction! Indeed: J(Ω +hδθ) =J(Ω)−h
Z
Γθ
f2(x) dx+O(h2)
Allaire (2003,2004)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 7/17
Shape optimisation
GeneralitiesCost functional: unconstrainedminimisation problem
J(Ω) = Z
Ω
f(x) dx Shape derivative: ∀δθ we have
(∇δθJ, δθ) = lim
h→0
J(Ω +hδθ)− J(Ω)
h =
Z
Γθ
δθ(x).n(x)f(x) dx.
Descent direction:
J(Ω +hδθ) =J(Ω) +h Z
Γθ
δθ(x).n(x)f(x) dx+O(h2)
⇒δθ(x) =−f(x)n(x) ensures a descent direction!
Indeed: J(Ω +hδθ) =J(Ω)−h Z
Γθ
f2(x) dx+O(h2)
Allaire (2003,2004)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 8/17
Shape optimisation
Impedance matchingCost functional:
J(Ω, p) = 1 2
Z k2
k1
|Z(p, k)−ZT(k)|2dk
subject to
∆p+k2p= 0 + B.C.
Impedance: Z(p, k) = e−ikL−eikL+ ¯p
e−ikL+eikL−p¯ with p¯= 1 LZ
Z
ΓZ
pdl.
Bängtsson et al. (2003)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 8/17
Shape optimisation
Impedance matchingCost functional:
J(Ω, p) = 1 2
Z k2
k1
|Z(p, k)−ZT(k)|2dk
subject to
∆p+k2p= 0 + B.C.
We define the Lagrangian(constrained7→unconstrained)
L(Ω, p, λ) =J(Ω, p)−real Z k2
k1
λ,∆p+k2p
Ω dk
Bängtsson et al. (2003)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 8/17
Shape optimisation
Impedance matchingCost functional:
J(Ω, p) = 1 2
Z k2
k1
|Z(p, k)−ZT(k)|2dk
subject to
∆p+k2p= 0 + B.C.
We define the Lagrangian(constrained7→unconstrained)
L(Ω, p, λ) =J(Ω, p)−real Z k2
k1
λ,∆p+k2p
Ω dk
Derivative with respect to each variable:
(∇δλL, δλ) = (∇δpL, δp) = 0 (∇δθL, δθ) = 0⇒(∇δθJ, δθ)
Bängtsson et al. (2003)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 9/17
Impedance matching
Derivative with respect toλ: ∆p+k2p= 0.
Derivative with respect top: Adjoint equation
∆λ+ (k∗)2λ= 0 x∈Ω
∇λ.n= 0 x∈Γs
∇λ.n+δvi+ 1
2 ∆Tλ−δT(k∗)2(i+ 1)(γ−1)
2 λ= 0 x∈Γw
∇λ.n−ik∗λ= 2 LZ
eikL(Z(p, k)−ZT(k))
(e−ikL+eikL−p)¯2∗ x∈ΓZ.
Derivative with respect toΩ: For any displacement directionδθ, (∇δθJ, δθ) =
Z
Γθ
(δθ(x).n(x)) Z k2
k1
Real
−∇λ∗.∇p+k2λ∗p dkdx.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 9/17
Impedance matching
Derivative with respect toλ: ∆p+k2p= 0.
Derivative with respect top: Adjoint equation
∆λ+ (k∗)2λ= 0 x∈Ω
∇λ.n= 0 x∈Γs
∇λ.n+δvi+ 1
2 ∆Tλ−δT(k∗)2(i+ 1)(γ−1)
2 λ= 0 x∈Γw
∇λ.n−ik∗λ= 2 LZ
eikL(Z(p, k)−ZT(k))
(e−ikL+eikL−p)¯2
∗ x∈ΓZ.
Derivative with respect toΩ: For any displacement directionδθ, (∇δθJ, δθ) =
Z
Γθ
(δθ(x).n(x)) Z k2
k1
Real
−∇λ∗.∇p+k2λ∗p dkdx.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 9/17
Impedance matching
Derivative with respect toλ: ∆p+k2p= 0.
Derivative with respect top: Adjoint equation
∆λ+ (k∗)2λ= 0 x∈Ω
∇λ.n= 0 x∈Γs
∇λ.n+δvi+ 1
2 ∆Tλ−δT(k∗)2(i+ 1)(γ−1)
2 λ= 0 x∈Γw
∇λ.n−ik∗λ= 2 LZ
eikL(Z(p, k)−ZT(k))
(e−ikL+eikL−p)¯2
∗ x∈ΓZ.
Derivative with respect toΩ: For any displacement directionδθ, (∇δθJ, δθ) =
Z
Γθ
(δθ(x).n(x)) Z k2
k1
Real
−∇λ∗.∇p+k2λ∗p dkdx.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Generalities Impedance matching 10/17
Impedance matching
Algorithm:
Solve iteratively:
Solve Helmholtz direct ⇒ p(k) (frequency response). Solve Helmholtz adjoint ⇒ λ(k) (frequency response). Optimality condition ⇒ ∇δθJ.
Move shape with linesearch algorithm(Armijo backtracking). Iterate until convergence.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 11/17
Numerical implementation
Finite elements with immersed boundary:
XFEMfor integration methods on cut elements (GETFEM++).
Level set:
ψ= 0 x∈Γθ ψ >0 x∈Ω ψ <0 x /∈Ω.
Transport equation for moving the shape:
∂ψ
∂t +vn.∇ψ= 0.
with
v=−∇δθJ.n. (Regularised by Sobolev Gradient) nis the outward wall normal direction. SUPG discretisation.
Moës et al. (1999) Osher et al. (2001) Protas et al. (2004)
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 12/17
Results
No viscosity: ZT = 0,k= [0.2 : 0.6]
Initial condition. Converged.
Coloured by∇δθJ.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 12/17
Results
No viscosity: ZT = 0,k= [0.2 : 0.6]
−20
−10 0 10 20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
ReactanceIm(Z).
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 13/17
Results
No viscosity: ZT = 0,k= [0.15 : 0.25]
Initial condition. Converged.
Coloured by∇δθJ.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 13/17
Results
No viscosity: ZT = 0,k= [0.15 : 0.25]
−20
−10 0 10 20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
ReactanceIm(Z).
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 14/17
Results
Viscosity: ZT = 1.0,k= [0.2 : 0.6]
Initial condition. Converged.
Coloured by∇δθJ.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 14/17
Results
Viscosity: ZT = 1.0,k= [0.2 : 0.6]
−20
−10 0 10 20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
0 0.2 0.4 0.6 0.8 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
ReactanceIm(Z). ResistanceRe(Z).
0 0.02 0.04 0.06 0.08 0.1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α
k
Initial Final
Absorptionα= 1− |R|2.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 15/17
Results
Viscosity: ZT = 1.0,k= [0.15 : 0.25]
Initial condition. Converged.
Coloured by∇δθJ.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion
Inviscid problem Viscous problem 15/17
Results
Viscosity: ZT = 1.0,k= [0.15 : 0.25]
−20
−10 0 10 20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
0 0.2 0.4 0.6 0.8 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z
k
Initial Final
ReactanceIm(Z). ResistanceRe(Z).
0 0.02 0.04 0.06 0.08 0.1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α
k
Initial Final
Absorptionα= 1− |R|2.
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 16/17
Conclusion
Summary:
Shape optimisation for impedance matching.
Viscous model: compromise between Helmholtz and full linearised Navier-Stokes.
Perspectives:
Optimise from efficient/realistic configuration.
Different cost functional (Penalty to initial guess, absorption, . . . ).
Context Outline Gov. equations Shape optimisation Implementation Results Conclusion 17/17