← All notes & connection map

FVM

Finite Volume Method

Created on: 2025-01-18
Tags: #FVM
Course: SU2


Motivation

  • To convert the PDE (Navier Stokes Equation) to a system of linear equations and solve it using a linear solver.

Finite Volume Method

  • Steady Incompressible N-S Equations \(\nabla \cdot (UU) = -\frac{1}{\rho} \nabla p + \nabla \cdot (\nu \nabla U) + g\)
  • Discretize the equation using a generic 3D polyhedral cell.
  • In the second order FVM, the flow variable vary linearly across the cells.
  • The values are computed at the cell centroid.
  • Each cell will have M neighbors due to M faces of the 3D polyhedral cell.

Integration

  • We integrate the N-S equation across the cell \(\int_{V}\left[ \nabla \cdot (UU) + \frac{1}{\rho} \nabla p - \nabla \cdot (\nu \nabla U) - g \right]\,dV = 0\)
  • We can integrate each term separately.

Source Term

  • Constant source terms (like gravity) are easy \(\int_{V} [g]dV = gV_{p}\)
  • The gravity force is added to the right hand side (to the $B$ vector).
  • Linear source terms require more care.
  • Assume that the source terms $SU$ was added to the Navier Stokes equations. \(\int_{v} [SU]\,dV = S_{p} \int_{V} U dV = S_{p} \int_{V} (U_{p}+(x-x_{p})*(\nabla U_{p}))dV = S_{p}U_{p}V_{p}\)
  • We could either add $-S_{p}V_{p}$ to the $M$ matrix or add $S_{p}U_{p}V_{p}$ to the $B$ matrix.

Implicit and Explicit Treatment

  • Adding the source term to the $M$ matrix is called implicit treatment.
  • Adding the source term to the $B$ matrix is called explicit treatment.
  • The treatment of source term is dependent on the sign of the source term as it will either improve the $M$ matrix dominance (if negative) or reduce (if positive).

Convection Term

  • These terms contain the divergence operator ($\nabla \cdot$).
  • The divergence theorem is used to convert the volume integrals into surface integrals. \(\int_{S} F \cdot \hat{n}\, dS\)
  • The convection term is written as \(\int_{V} [\nabla \cdot UU]\,dV = \int_{S} [U(U\cdot \hat{n})]\,dS\)
  • The dot product of the velocity, the unit normal vector and the surface element is the volume flow rate out of the surface.
  • The other velocity outside the bracket $(U)$ is the unknown we are solving for.
  • The velocity is transported by the volume flux.
  • The surface integral into a sum over a finite number of faces. \(\int_{S} [U(U\cdot \hat{n})]\,dS = \sum_{i=1}^M \int_{S} [U_{i}(U_{i}\cdot \hat{n}_{i})]\,dS_{i}\)
  • We can approximate as the value at the face center $f$. \(\sum_{i=1}^M \int_{S} [U_{i}(U_{i}\cdot \hat{n}_{i})]\,dS_{i} \approx \sum_{i=1}^M [U_{fi}(U_{fi}\cdot \hat{n}_{fi})]\,S_{i}\)
  • We need the value at the face centers $(f)$ and we only have the centroid values $(P,N)$.
  • Interpolation is needed.

FVM1

  • Variety of face interpolation schemes to calculate $U_{f}$:
    • Upwind
    • Second-order/linear Upwind
    • Central Differencing
    • QUICK
  • The convection term is now \(\int_{V} [\nabla \cdot UU]\,dV \approx \sum_{i=1}^M U_{fi} F_{fi}\)
  • The convection terms leads to diagonal and off-diagonal terms.
  • This is a reflection of the cell connectivity.

Face Interpolation Schemes

Linear/Central Differencing

  • Linear variation between the centroids.
  • Second order accurate (due to the linear variation).
  • Scheme is unbounded (prone to oscillations).
  • Central differencing is used for the diffusion term.
  • Should never be used for the convection term in a RANS solver.
  • Although, it can be used for the convection term in LES. \(\psi = \frac{|x_{f}-x_{P}|}{|x_{N}-x_{P}|}\,\,\, \text{such that}\,\,\, \psi \in[0,1]\) \(\phi_{f} = \psi\,\, \phi_{N} + (1-\psi)\,\, \phi_{P}\)

    Upwind Differencing

  • Depends on the direction of the mass flux \(F_{f} = \rho_{f} A_{f}(U_{f}\cdot \hat{n})\)
  • $F_{f}>0$ indicates the mass flow out of the cell.
  • $F_{f}<0$ indicates the mass flow into the cell.

upwindDiff

  • The value of $\phi$ on the cell face ($\phi_{f}$) is given by: \(\phi_{f} = \begin{cases} \phi_{p} & F_{f}>0\\ \phi_{N} & F_{f}<0 \end{cases}\)
  • The $\phi$ is constant between the cell centroid and the cell face.
  • It is a first-order accurate. It alters the flow properties and hence not accurately calculated.
  • But it is the most stable scheme for convection dominated flows.

Linear Upwind Differencing

  • Popular scheme
  • More accurate than upwind differencing (Second Order Accurate) \(\phi_{f} = \begin{cases} \phi_{p}+(\nabla \phi)_{P} \cdot r & F_{f}>0\\ \phi_{N}+(\nabla \phi)_{N} \cdot r & F_{f}<0 \end{cases}\)
  • We use the gradient $\nabla \phi$ to improve the accuracy of the extrapolation.
  • The variation between the cell centroid and the face is linear.
  • It is popular for the convection term but not for the diffusion term.
  • However, we often have to limit the gradient $(\nabla \phi)_{P}$ to prevent local maxima/minima.
  • Gradient Limiter ($\varphi$) and $\varphi \in[0,1]$.

Advanced Discretization Schemes

  • Combination of Linear/Central Differencing $(\phi_{CD})$ for accuracy and Upwind Differencing for stability ($\phi_{UD}$). \(\phi_{f} = \Psi\,\,\phi_{OD} + (1-\Psi)\,\,\phi_{CD}\)
  • $\Psi$ is a blending function to switch between schemes.
  • The schemes pick clear ways to choose $\Psi$.

Linked from

  • No other published notes link here yet.