THESE de DOCTORAT - cerfacs

THESE de DOCTORAT - cerfacs THESE de DOCTORAT - cerfacs

22.10.2014 Views

98 Chapter 6: Boundary conditions for low Mach number acoustic codes ( 1¯ρ ∂ ¯c 2 ∂x + ū ∂ū ¯ρ ∂x + ¯c2¯ρ γr ¯˙Q ¯ρ ¯p ) ( ∂ ∂ū ˆρ + ∂x ∂x + ū ∂ ) ( ∂ ¯p û + (γ − 1) ¯T ∂x 1¯p ∂x + ∂ ) ŝ − ˆϜ − jωû = 0 (6.5) ∂x ( ) ∂¯s ˆρ + ∂x û + ū ∂ ¯˙Q ∂x + (γ − 1) ŝ − r¯p ¯p ˆ˙Q − ˆ W k − jωŝ = 0 (6.6) This system of equations can be expressed as an algebraic linear system in the form ⎡ ⎤ ⎡ ⎤ ˆρ 0 ⎢ ⎥ ⎢ A ⎣û⎦ = ⎣ ˆϜ ⎥ ⎦ (6.7) ŝ r¯p ˆ˙Q + Wˆ k where ⎡ ∂ū ∂x + ū ∂ A = ⎢ ⎣ 1¯ρ ∂ ¯c 2 ∂x + ū¯ρ ∂x + ū ∂S S ∂ū ∂x + ¯c2¯ρ γr ¯˙Q ¯ρ ¯p ∂x − jω ∂ ∂x ∂ ¯ρ ∂x + ¯ρ ∂ ∂x + ¯ρ S ∂ū ∂x + ū ∂ ∂x − jω ∂¯s ∂x ∂S ∂x 0 (γ − 1) ¯T ( 1¯p ∂ ¯p ∂x + ∂ ∂x ū ∂ ∂x + (γ − 1) ¯˙Q¯p − jω ⎤ ) ⎥ ⎦ (6.8) The Quasi-1D Linearized Euler Equations (Eqs. 6.4, 6.5 and 6.6) are solved by a numerical tool called ‘SNozzle’. Within the computational domain, these equations are discretized by finite differences (FD) with a second order centered scheme. At the inlet, a first order FD downwind scheme is applied, whereas at the outlet a first order FD upwind scheme is used. A staggered grid arrangement (velocity fluctuations stored at the cell edges, density and entropy fluctuations stored at the cell centers) has been used in order to avoid the pressure field to be contaminated by the classical odd-even decoupling phenomenon. The Jacobi preconditioner is used to improve the quality of the matrix A (Eq. 6.8) and the linear system is solved by inverting the preconditioned matrix through the LU factorization. 6.3 The 1D linearized Euler equations for compact systems In order to validate the numerical too SNozzle, analytical solutions are derived for compact regions where the mean enthalpy or entropy changes. The starting point for this derivation are the 1D Euler equations for mass, momentum and total enthalpy.

6.3 The 1D linearized Euler equations for compact systems 99 ∂ρ ∂t + ∂ (ρu) = 0 ∂x (6.9) ∂ρu + ∂ ∂t ∂x (ρu2 ) = − ∂p ∂x + Ϝ (6.10) ∂ρh t + ∂ ∂t ∂x (ρuh t) = ˙Q + W k (6.11) Integrating over the 1D domain between positions x 1 and x 2 results in ∫ x2 x 1 ∫ ∂ρ ∂t dx + x2 ∫ x2 x 1 x 1 ∂ ∂x (ρu)dx = 0 ∂ρ ∂t dx + (ρu) ∣ ∣∣∣ x 2 x 1 = 0 (6.12) ∫ x2 x 1 ∂ρu ∂t dx + ∫ x2 x 1 ∫ x2 ∫ ∂ ∂p ∂x (ρu2 )dx + x 1 ∂x dx − x2 Ϝdx = 0 x 1 ∫ x2 x ∂ρu x 1 ∂t dx + 2 x 2 (ρu2 ) ∣ + p ∣ − F = 0 x 1 x 1 (6.13) ∫ x2 x 1 ∂ρh t ∂t dx + ∫ x2 x 1 ∂ ∂x (ρuh t)dx − ∫ x2 x 1 ∫ x2 x 1 ∂ρh t ∂t dx + (ρuh t) ∫ x2 ˙Qdx − W k dx = 0 x 1 x 2 ∣ − ˙q − W = 0 x 1 (6.14) where the following notations have been introduced: ∫ ˙Qdx = ˙q, ∫ Ϝdx = F and ∫ W k dx = W. After linearizing and recalling that h t = c p T t , Eqs. (6.12), (6.13) and (6.14) become ( ¯ρu ′ + ρ ′ ū ) ∣ ∣ ∣∣ x 2 ( p ′ + ρ ′ ū 2 + 2 ¯ρūu ′) ∣ ∣ ∣∣ x 2 [ cp ¯T t ( ¯ρu ′ + ρ ′ ū) + ¯ρū(c p T ′ + ūu ′ ) ] ∣ ∣ ∣∣ x 2 x 1 = − x 1 − F ′ = − x 1 − ˙q ′ − W ′ = − ∫ x2 x 1 ∂ρ ′ ∂t ∫ x2 x 1 ∫ x2 x 1 dx (6.15) ∂ ( ¯ρu ′ + ρ ′ ū ) dx (6.16) ∂t ∂ [ ρ ′ c p ¯T t + ¯ρ ( c p T ′ + ūu ′)] dx ∂t (6.17)

98 Chapter 6: Boundary conditions for low Mach number acoustic co<strong>de</strong>s<br />

( 1¯ρ<br />

∂ ¯c 2<br />

∂x + ū ∂ū<br />

¯ρ ∂x + ¯c2¯ρ<br />

γr ¯˙Q<br />

¯ρ ¯p<br />

) (<br />

∂ ∂ū<br />

ˆρ +<br />

∂x ∂x + ū ∂ )<br />

(<br />

∂ ¯p<br />

û + (γ − 1) ¯T<br />

∂x<br />

1¯p ∂x + ∂ )<br />

ŝ − ˆϜ − jωû = 0 (6.5)<br />

∂x<br />

(<br />

)<br />

∂¯s ˆρ +<br />

∂x û + ū ∂<br />

¯˙Q<br />

∂x + (γ − 1) ŝ − r¯p ¯p<br />

ˆ˙Q − ˆ W k − jωŝ = 0 (6.6)<br />

This system of equations can be expressed as an algebraic linear system in the form<br />

⎡ ⎤ ⎡ ⎤<br />

ˆρ 0<br />

⎢ ⎥ ⎢<br />

A ⎣û⎦ = ⎣ ˆϜ<br />

⎥<br />

⎦ (6.7)<br />

ŝ<br />

r¯p<br />

ˆ˙Q + Wˆ<br />

k<br />

where<br />

⎡<br />

∂ū<br />

∂x + ū ∂<br />

A = ⎢<br />

⎣<br />

1¯ρ<br />

∂ ¯c 2<br />

∂x + ū¯ρ<br />

∂x + ū ∂S<br />

S<br />

∂ū<br />

∂x + ¯c2¯ρ<br />

γr ¯˙Q<br />

¯ρ ¯p<br />

∂x − jω<br />

∂<br />

∂x<br />

∂ ¯ρ<br />

∂x + ¯ρ ∂<br />

∂x + ¯ρ S<br />

∂ū<br />

∂x + ū ∂<br />

∂x − jω<br />

∂¯s<br />

∂x<br />

∂S<br />

∂x<br />

0<br />

(γ − 1) ¯T<br />

(<br />

1¯p<br />

∂ ¯p<br />

∂x + ∂<br />

∂x<br />

ū ∂<br />

∂x + (γ − 1) ¯˙Q¯p − jω<br />

⎤<br />

)<br />

⎥<br />

⎦ (6.8)<br />

The Quasi-1D Linearized Euler Equations (Eqs. 6.4, 6.5 and 6.6) are solved by a numerical tool<br />

called ‘SNozzle’. Within the computational domain, these equations are discretized by finite<br />

differences (FD) with a second or<strong>de</strong>r centered scheme. At the inlet, a first or<strong>de</strong>r FD downwind<br />

scheme is applied, whereas at the outlet a first or<strong>de</strong>r FD upwind scheme is used. A staggered<br />

grid arrangement (velocity fluctuations stored at the cell edges, <strong>de</strong>nsity and entropy fluctuations<br />

stored at the cell centers) has been used in or<strong>de</strong>r to avoid the pressure field to be contaminated<br />

by the classical odd-even <strong>de</strong>coupling phenomenon. The Jacobi preconditioner is used to<br />

improve the quality of the matrix A (Eq. 6.8) and the linear system is solved by inverting the<br />

preconditioned matrix through the LU factorization.<br />

6.3 The 1D linearized Euler equations for compact systems<br />

In or<strong>de</strong>r to validate the numerical too SNozzle, analytical solutions are <strong>de</strong>rived for compact<br />

regions where the mean enthalpy or entropy changes. The starting point for this <strong>de</strong>rivation are<br />

the 1D Euler equations for mass, momentum and total enthalpy.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!