3.1.30Compressible Flow & Aerodynamics

Computational aerodynamics — panel method (intro), CFD overview

2,227 words10 min readdifficulty · medium

1. WHY do we need computational methods?

WHY Laplace? Incompressible V=0\Rightarrow \nabla\cdot\vec V=0. Irrotational V=ϕ\Rightarrow \vec V=\nabla\phi. Combine: (ϕ)=2ϕ=0\nabla\cdot(\nabla\phi)=\nabla^2\phi=0.

WHAT changes for real flow? Add viscosity (μ2V\mu\nabla^2\vec V term) and compressibility (ρ\rho varies) and you get the full Navier–Stokes equations, which are nonlinear — no superposition, must use CFD.


2. Elementary solutions (the LEGO bricks)

Because Laplace is linear, any sum of these is also a solution:


3. Panel method — DERIVATION from scratch

Step 1 — The physical condition we must enforce

The body is solid: no flow goes through the wall. So the velocity component normal to the surface is zero: Vn^=0on the body.\vec V \cdot \hat n = 0 \quad \text{on the body.} Why this step? It is the only boundary condition that defines the shape's effect — everything is built to satisfy it.

Step 2 — Discretise the surface

Replace the smooth contour with NN flat panels. Put a source sheet of unknown strength λj\lambda_j (constant) on each panel jj. Why this step? We can't find a continuous λ(s)\lambda(s) by hand, but NN unknown constants give a solvable linear algebra problem.

Step 3 — Total potential

ϕ(P)=V(xcosα+ysinα)+j=1Nλj2πpanel jlnrdsj\phi(P) = V_\infty(x\cos\alpha+y\sin\alpha) + \sum_{j=1}^{N}\frac{\lambda_j}{2\pi}\int_{\text{panel }j}\ln r\,ds_j Why this step? Superposition: freestream + every panel's contribution. The integral spreads λj\lambda_j over the panel.

Step 4 — Apply boundary condition at each control point

At the midpoint ("control point") of panel ii, demand zero normal velocity: Vcosβi+j=1Nλj2πIij=0,i=1NV_\infty\cos\beta_i + \sum_{j=1}^{N} \frac{\lambda_j}{2\pi}\,I_{ij} = 0, \qquad i=1\dots N where βi\beta_i is the angle between the freestream and the panel's outward normal, and Iij=nijlnrdsjI_{ij}=\dfrac{\partial}{\partial n_i}\int_j \ln r\,ds_j is the influence coefficient (normal velocity induced at panel ii by a unit source on panel jj).

Why this step? One equation per panel → NN equations for NN unknowns λj\lambda_j.

Step 5 — Solve the linear system

  Aλ=b  Aij=Iij2π,bi=Vcosβi\boxed{\;A\boldsymbol{\lambda} = \mathbf{b}\;}\qquad A_{ij}=\frac{I_{ij}}{2\pi},\quad b_i=-V_\infty\cos\beta_i Why this step? It's just a matrix inversion — the computer's job.

Step 6 — Post-process

Once λj\lambda_j known, compute the tangential velocity at each panel, then pressure via Bernoulli: Cp,i=1(Vt,iV)2C_{p,i} = 1 - \left(\frac{V_{t,i}}{V_\infty}\right)^2 Integrate CpC_p around the surface to get lift (need vortices/Kutta condition for lift) and moment.

Figure — Computational aerodynamics — panel method (intro), CFD overview

4. CFD overview — when panels are not enough

Aspect Panel method CFD (Navier–Stokes)
Discretises Surface only Whole volume
Equation Linear (Laplace) Nonlinear (N–S)
Viscosity / drag No (no skin friction) Yes
Shocks, separation No Yes
Cost Seconds Hours–days
Best for Early design, attached subsonic flow Final design, transonic/turbulent/separated

5. Common mistakes (Steel-man + fix)


6. Active recall

Recall Quick self-test (hide answers)
  • What PDE governs ideal flow, and why is its linearity essential? → Laplace; allows superposition of elementary solutions.
  • What boundary condition fixes the panel strengths? → Zero normal velocity (flow tangency).
  • Why does a source-only panel solution give zero lift? → No circulation.
  • One reason CFD beats panels? → Captures viscosity/separation/shocks.
What PDE governs steady incompressible irrotational flow?
Laplace's equation 2ϕ=0\nabla^2\phi=0.
Why is linearity of Laplace's equation crucial for panel methods?
It lets us superpose elementary solutions (stream + sources + vortices).
What boundary condition does the panel method enforce at each control point?
Flow tangency — zero velocity normal to the surface, Vn^=0\vec V\cdot\hat n=0.
What are the unknowns solved for in a source-panel method?
The source strengths λj\lambda_j on each panel, from Aλ=bA\boldsymbol\lambda=\mathbf b.
What is an influence coefficient IijI_{ij}?
The normal velocity induced at panel ii by a unit-strength source on panel jj.
Why can't a source-only panel method predict lift?
It produces no circulation Γ\Gamma; by Kutta–Joukowski L=ρVΓ=0L'=\rho V_\infty\Gamma=0.
What extra ingredient gives a panel method lift?
Vortex panels plus the Kutta condition (smooth trailing-edge flow) to fix Γ\Gamma.
How do you get pressure from panel velocities?
Bernoulli: Cp=1(Vt/V)2C_p=1-(V_t/V_\infty)^2.
What equations does CFD solve?
The (nonlinear) Navier–Stokes equations over a volume mesh.
Name the CFD discretisation methods.
Finite Volume, Finite Difference, Finite Element.
Why is Finite Volume preferred for aerodynamics with shocks?
It conserves mass/momentum/energy exactly across cell faces.
What is d'Alembert's paradox?
Inviscid flow predicts zero drag on a body — drag needs viscosity.
What two checks validate a CFD result?
Mesh independence and comparison with experiment.
How does panel-method cost scale with N panels?
Roughly N2N^2N3N^3 (dense influence matrix).
Recall Feynman: explain to a 12-year-old

Imagine you want to know how wind flows around a toy car. Instead of solving impossible math, you cover the car's outline with lots of tiny "tape strips". Each strip can gently blow or suck air. You adjust how hard each strip blows until the air just slides along the car and never goes through it. Add up all the gentle blows and the steady wind, and you get the whole airflow! That's the panel method. When the car is fast or sticky air matters (friction, swirling wakes), tape strips aren't enough — so we instead fill the whole room of air with a grid of tiny boxes and track wind in every box. That bigger, slower computer job is CFD.


Connections

Concept Map

inviscid irrotational incompressible

is linear

enables

elementary solutions

summed to build

wraps surface in

carry unknown

enforces

gives

add viscosity and compressibility

no superposition needs

captures

Velocity field PDEs

Laplace equation

Superposition allowed

Panel method

Stream source vortex doublet

N flat panels

Source strengths

No through-flow V dot n = 0

Linear system for lambda

Navier-Stokes nonlinear

CFD volume mesh

Turbulence shocks viscosity

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, aerodynamics mein air ka flow PDE se chalta hai, aur real wing ke liye yeh haath se solve karna almost impossible hai. Isliye hum computer ko kaam dete hain. Do popular tareeke hain. Pehla — panel method. Idea simple hai: body ki surface ko chhote-chhote flat panels mein todo, har panel par ek "source" laga do jiski strength λj\lambda_j unknown hai. Phir ek hi condition lagani hai — air body ke andar nahi ghusni chahiye, yaani normal velocity zero (Vn^=0\vec V\cdot\hat n=0) har control point par. Yeh N equations dega N unknowns ke liye, matrix Aλ=bA\lambda=b solve karo, bas ho gaya. Yeh fast hai kyunki Laplace equation linear hai, toh hum freestream + sources + vortices ko superpose kar sakte hain.

Ek important baat: sirf sources se lift nahi milta, kyunki lift ke liye circulation Γ\Gamma chahiye (Kutta–Joukowski: L=ρVΓL'=\rho V_\infty \Gamma). Iske liye vortex panels add karte hain aur Kutta condition lagate hain (trailing edge par flow smoothly nikle). Aur ek trap: panel method drag predict nahi karta — d'Alembert paradox, kyunki yeh inviscid hai, viscosity (friction) ko ignore karta hai.

Jab flow fast ho (transonic, shocks), ya separation/turbulence ho, tab panel method kaafi nahi. Tab CFD use karte hain — pure volume of air ko mesh (chhote boxes) mein todo aur full Navier–Stokes equations solve karo. Usually Finite Volume method, kyunki yeh mass-momentum-energy ko har cell face par exactly conserve karta hai (shocks ke liye zaroori). CFD slow aur mehnga hai, par sab kuch capture karta hai — friction, wake, shock sab.

80/20 funda yaad rakho: early design aur attached low-speed flow ke liye panel method 1% cost mein 80% answer de deta hai. CFD tab nikaalo jab viscosity, separation ya shocks dominate karein. Aur CFD ka result hamesha mesh independence aur experiment se validation ke baad hi maano — sirf residual girne ka matlab "correct" nahi hota.

Go deeper — visual, from zero

Test yourself — Compressible Flow & Aerodynamics

Connections