Computational Fluid Dynamics

Computational fluid dynamics is one of the branches of liquid mechanics that uses numeric techniques and algorithms to unravel and analyze issues that involve liquid flows. PCs are used to perform the millions of calculations needed to simulate the interaction of liquids and gases with surfaces outlined by boundary conditions. Even with high speed supercomputers only approximate solutions can be accomplished in numerous cases. Continuing research may yield software that improves the accuracy and speed of complicated simulation eventualities like transonic or turbulent flows. First endorsement of such software is usually performed employing a wind tunnel with the last validation coming in flight test.

Background and history

A PC simulation of high speed air flow round the shuttle during re-entry. A simulation of the Hyper-X scramjet auto in operation at Mach-7 The basic basis of virtually all CFD issues is the Navier-Stokes equations, which outline any single-phase liquid flow. These equations can be made easier by removing terms describing viscosity to yield the Euler equations.
Further simplification, by removing terms describing vorticity yields the total potential equations. Eventually , these equations can be linearized to yield the linearized potential equations. Traditionally , techniques were first developed to resolve the Linearized Potential equations. Two-dimensional techniques, using conformal transformations of the flow about a cylinder to the flow about an airfoil were developed in the 1930s. The PC power available paced development of 3D strategies. The 1st paper on a practical 3D technique to unravel the linearized potential equations was released by John Hess and A.M.O. Smith of Douglas Aircraft in 1966.
This strategy discretized the outside of the geometry with panels, giving rise to this class of programs being called Panel Techniques . Their strategy itself was simplified, in that it didn't include lifting flows and therefore was generally applied to ship hulls and aircraft fuselages. The 1st lifting Panel Code ( A230 ) was described in a paper written by Paul Rubbert and Gary Saaris of Boeing Aircraft in 1968. In time, more complicated three dimensional Panel Codes were developed at Boeing ( PANAIR, A502 ), Lockheed ( Quadpan ), Douglas ( HESS ), McDonnell Aircraft ( MACAERO ), NASA ( PMARC ) and Analytical Strategies ( WBAERO, USAERO and VSAERO ).
Some ( PANAIR, HESS and MACAERO ) were higher order codes, using higher order distributions of surface singularities, while others ( Quadpan, PMARC, USAERO and VSAERO ) used single singularities on each surface panel. The benefit of the lower order codes was that they ran faster on the PCs of the time. Today, VSAERO has grown to be a multi-order code and is the most generally used program of this class. This program has been employed in the development of many submarines, surface ships, vehicles, helicopters, aircraft, and more lately air turbines. Its sister code, USAERO is an unstable panel technique which has also been used for modeling such stuff as high speed trains and racing yachts. The NASA PMARC code from an early version of VSAERO and a derivative of PMARC, named CMARC, is also commercially available.
In the two-dimensional realm, quite a good number of Panel Codes have been developed for airfoil research and design. These codes typically have a boundary layer research included, so that viscous effects can be modeled.
Professor Richard Eppler of the College of Stuttgart developed the PROFIL code, in part with NASA funding, which became available in the early 1980s. This was shortly followed by MIT Professor Mark Drela's Xfoil code. Both PROFIL and Xfoil incorporate two-dimensional panel codes, with joined boundary layer codes for airfoil research work. PROFIL uses a conformal metamorphosis strategy for inverse airfoil design, while Xfoil has both a conformal metamorphosis and an inverse panel system for airfoil design. Both codes are generally used. An intermediate step between Panel Codes and total potential codes were codes that utilised the Transonic Little Rumpus equations.
Particularly, the 3D WIBCO code, developed by Charlie Boppe of Grumman Aircraft in the early 1980s saw heavy use.
Developers next turned to total potential codes, as panel strategies couldn't work out the non-linear flow present at transonic speeds. The first outline of a method of using the actual potential equations was released by Earll Murman and Julian Cole of Boeing in 1970. Frances Bauer, Paul Garabedian and David Korn of the Courant Institute at Long Island School ( NYU ) wrote a sequence of two-dimensional actual potential airfoil codes that were commonly used, the most significant being named Program H. Another expansion of Progam H was developed by Bob Melnik and his group at Grumman Aerospace as Grumfoil. Antony Jameson, originally at Grumman Aircraft and the Courant Institute of NYU, worked with David Caughey to develop the crucial 3D Full Potential code FLO22 in 1975. Many actual potential codes appeared after this, finishing in Boeing's Tranair ( A633 ) code, which still sees heavy use. The very next step was the Euler equations, which guaranteed to provide more accurate solutions of transonic flows.
The methodology employed by Jameson in his three dimensional FLO57 code ( 1981 ) was employed by others to provide such programs as Lockheed's TEAM program and IAI / Analytical Methods' MGAERO program. MGAERO is unique in being a structured cartesian mesh code, while most other such codes use structured body-fitted grids ( with the exception of NASA's very successful CART3D code, Lockheed's SPLITFLOW code and Georgia Tech's NASCART-GT ).
Antony Jameson also developed the 3 dimensional aeroplane code ( 1985 ) which used unstructured tetrahedral grids. In the two-dimensional realm, Mark Drela and Michael Giles, then graduate scholars at MIT, developed the ISES Euler program ( essentially a collection of programs ) for airfoil design and research. This code first became available in 1986 and has been further developed to design, investigate and optimize single or multi-element airfoils, as the MSES program.
MSES sees wide use across the world.
A derivative of MSES, for the design and research of airfoils in a cascade, is MISES, developed by Harold "Guppy" Youngren while he used to be a graduate student at MIT. The Navier-Stokes equations were the final target of developers. Two-dimensional codes , for example NASA Ames' ARC2D code first emerged. A number of 3 dimensional codes were developed ( OVERFLOW, CFL3D are 2 successful NASA contributions ), leading to countless commercial packages.

Technicalities

The most basic consideration in CFD is how one treats a regular liquid in a discretized fashion on a P. C. . One strategy is to discretize the spatial domain into little cells to form a volume mesh or grid, and then apply an appropriate algorithm to unravel the equations of motion ( Euler equations for inviscid, and Navier-Stokes equations for viscous flow ). Additionally, such a mesh can be either irregular ( as an example composed from triangles in 2D, or pyramidal solids in 3D ) or regular ; the distinguishing characteristic of the previous is that each cell must be stored separately in memory. Where shocks or discontinuities are present, high res schemes like Total Modification Reducing ( TVD ), Flux Corrected Transport ( FCT ), Fundamentally NonOscillatory ( ENO ), or MUSCL schemes are wanted to avoid phony oscillations ( Gibbs phenomenon ) in the solution. If one selects not to continue with a mesh-based methodology, a number of possibilities exist, particularly :

Smoothed particle hydrodynamics ( SPH ), a Lagrangian system of solving liquid issues,
Spectral techniques, a method where the equations are projected onto basis functions like the round harmonics and Chebyshev polynomials,
Lattice Boltzmann techniques ( LBM ), which simulate an equivalent mesoscopic system on a Cartesian grid, rather than sorting out the macroscopic system ( or the genuine miniscule physics ).

It is feasible to immediately solve the Navier-Stokes equations for laminar flows and for turbulent flows when all the important length scales can be fixed by the grid ( a Direct numeric simulation ). Generally however, the range of length scales suitable to the difficulty is bigger than even today's colossally parallel PCs can model. In cases like these, turbulent flow simulations need the arrival of a turbulence model. Reynolds-averaged Navier-Stokes equations ( RANS ) and Large eddy simulations ( LES ) the formulation, with the k-949 ; model or the Reynolds stress model, are 2 strategies for working with these scales.
In several examples, other equations are solved concurrently with the Navier-Stokes equations. These other equations can also include those describing species concentration ( mass transfer ), chemical reactions, heat transfer, and so on. More complicated codes allow the simulation of more complicated cases concerning multi-phase flows ( e.g. Liquid / gas, solid / gas, liquid / solid ), non-Newtonian liquids ( like blood ), or chemically reacting flows ( like combustion ).

Methodology

In all these approaches identical elementary process is followed.

During preprocessing
1. The geometry ( physical bounds ) of the difficulty is outlined.
2. The volume occupied by the liquid is split into discrete cells ( the mesh ). The mesh could be uniform or non uniform.
3. The physical modeling is defined - as an example, the equations of motions + enthalpy + radiation + species conservation
4. Boundary conditions are outlined. This involves specifying the liquid behavior and properties at the limits of the issue. For temporary issues, the opening conditions are also outlined.

The simulation is started and the equations are solved iteratively as a steady-state or brief.
Eventually a postprocessor is utilized for the research and visualisation of the ensuing solution.

Discretization strategies

The steadiness of the selected discretization is generally established numerically instead of analytically as with simple linear issues.
Special care must also be brought to guarantee the discretization handles discontinuous solutions gracefully. The Euler equations and Navier-Stokes equations both, contact surfaces and admit shocks.
Some of the discretization strategies being used are:

Finite volume Method( FVM ). This is the "classical" or standard approach used most frequently in commercial software and research codes. The ruling equations are solved on discrete control volumes. FVM recasts the PDE's ( Partial Differential Equations ) of the N-S equation in the conservative form and then discretize this equation. This guarantees the conservation of fluxes thru a selected control volume. Though the general solution will be conservative in nature there is not any guarantee that it's the exact solution. Likewise this technique is delicate to deformed elements which can forestall convergence if such elements are in urgent flow regions. This integration approach yields a technique that's intrinsically conservative ( i.e. Quantities like density remain physically pointed ) : Where Q is the vector of preserved variables, F is the vector of fluxes ( see Euler equations or Navier-Stokes equations ), V is the cell volume, and is the cell surface area.

$\frac{\partial}{\partial t}\iiint Q\, dV + \iint F\, d\mathbf{A} = 0,$

Finite Element Method ( FEM ). This technique is preferred for structural research of solids, but is also applicable to liquids. The FEM formulation needs special care to guarantee a conservative solution. The FEM formulation has been evolved to be used with the Navier-Stokes equations. Though in FEM conservation needs to be sorted, it is way more stable than the FVM approach. Subsequently it's the new direction in which CFD is moving. Typically stability / robustness of the solution is better in FEM though for some cases it would take more memory than FVM techniques. In this technique, a weighted residual equation is created : where Ri is the equation residual at a factor apex i, Q is the conservation equation voiced on a factor basis, Wi is the weight factor and Ve is the volume of the component.

$R_i = \iiint W_iQ\,dV^e$

Finite difference system. This technique has historic significance and is easy to program. It is presently only utilized in few specialized codes. Modern finite difference codes use an inserted boundary for handling complicated geometries making these codes highly effective and correct. Other ways to respond to geometries are using overlapping-grids, where the solution is interpolated across each grid. Where Q is the vector of preserved variables, and F, G, and H are the fluxes in the x, y, and z directions respectively.

$\frac{\partial Q}{\partial t}+ \frac{\partial F}{\partial x}+ \frac{\partial G}{\partial y}+ \frac{\partial H}{\partial z}=0$

Boundary part strategy. The boundary occupied by the liquid is split into surface mesh.
High-resolution schemes are get used where shocks or discontinuities are present. To capture pointy changes in the solution needs the utilization of 2nd or higher order numeric schemes that don't introduce false oscillations. This generally requires the applying of flux limiters to make sure that the solution is total difference lessening.

Turbulence models

Turbulent flow produces liquid interaction at a huge range of length scales. This problem suggests that it is needed that for turbulent flow regime calculations must try to take this into account by modifying the Navier-Stokes equations. Failure to do so may lead to an unstable simulation. When answering the turbulence model there exists a trade-off between accuracy and speed of computation.

Direct numerical simulation

Direct numeric simulation ( DNS ) captures all the applicable scales of turbulent motion, so no model is required for the littlest scales. This approach is highly costly, if not intractable, for complicated issues on modern computing machines, therefore the requirement for models to represent the tiniest scales of liquid motion.

Reynolds-averaged NavierStokes

Reynolds-averaged Navier-Stokes ( RANS ) equations are the very old approach to turbulence modeling. An ensemble version of the ruling equations is solved, which introduces new apparent pressures known as Reynolds tensions. This adds a second order tensor of unknowns for which diverse models can provide various levels of closure. It's a common myth the RANS equations don't apply to flows with a time-varying mean flow because these equations are 'time-averaged'. Actually, statistically unstable ( or non-stationary ) flows can similarly be handled. This is frequently called URANS.
There's nothing embedded in Reynolds averaging to prevent this, but the turbulence models used to shut the equations are valid only so long as the time scale of these changes in the mean is huge compared to the time scales of the turbulent motion containing the majority of the energy.
RANS models can be split into 2 broad approaches:

Boussinesq conjecture this technique includes the use of an algebraic equation for the Reynolds tensions which include deciding the turbulent viscosity, and depending on the level of class of the model, solving transport equations for judging the turbulent kinetic energy and dissipation. Models include k-949 ; ( Spalding ), Mixing Length Model ( Prandtl ) and 0 Equation ( Chen ). The models available in this approach are commonly referred to by the amount of transport equations they include, for instance the Mixing Length model is a "Zero Equation" model because no transport equations are solved, and the k-949 ; on the other hand is a "Two Equation" model because 2 transport equations are solved.
Reynolds stress model ( RSM ) This approach tries to really solve transport equations for the Reynolds strains. This implies arrival of many transport equations for all of the Reynolds pressures and therefore this approach is far more expensive in CPU effort.

Large eddy simulation

Massive eddy simulations ( LES ) is a strategy in which the smaller eddies are filtered and are modeled employing a sub-grid scale model, while the bigger energy carrying eddies are simulated. This technique often needs a more refined mesh than a RANS model, but a far rougher mesh than a DNS solution.

Detached eddy simulation

Detached eddy simulations ( DES ) is an alteration of a RANS model in which the model switches to a subgrid scale formulation in regions fine enough for LES calculations.
Regions near solid limits and where the turbulent length scale is less than the maximum grid dimension are allotted the RANS style of solution. As the turbulent length scale surpasses the grid dimension, the regions are solved using the LES mode. So the grid resolution for DES isn't as demanding as pure LES, thus significantly cutting down the price of the computation.
Though DES was at first compounded for the Spalart-Allmaras model ( Spalart et al, 1997 ), it can be implemented with other RANS models ( Strelets, 2001 ), by reasonably modifying the length scale which is explicitly or unconditionally concerned in the RANS model. So while Spalart-Allmaras model based DES acts as LES with a wall model, DES based on other models ( like 2 equation models ) behave as a hybrid RANS-LES model. Grid generation is more complex than for an easy RANS or LES case thanks to the RANS-LES switch.
DES is a non-zonal approach and offers a single smooth speed field across the RANS and the LES regions of the solutions.

Vortex Method

The Vortex strategy is a grid-free technique for the simulation of turbulent flows. It uses vortices as computational elements, mimicking the physical structures in turbulence. Vortex techniques were developed as a grid-free strategy that wouldn't be restricted by the basic smoothing effects related to grid-based techniques. To be practical vortex strategies require means for quickly computing velocities from the vortex elements to paraphrase they need the answer to a selected form of the N-body problem ( in which the motion of N objects is tied to their mutual influences ). A long-sought discovery came in the latter 1980's with the development of the Fast Multipole Method ( FMM ), an algorithm which has been announced as one of the top 10 advances in numeric science of the twentieth century. This discovery cleared the path to practical computation of the velocities from the vortex elements and is the root of successful algorithms.

Software primarily based on the Vortex methodology offer the engineer a new means for solving troublesome liquid dynamics problems with minimal user intervention. All that is needed is blueprint of problem geometry and setting of boundary and first conditions. Among the important advantages of this modern technology;

It is nearly grid-free, therefore junking much iteration related to RANS and LES.
All issues are treated identically. No modeling or calibration inputs are needed.
Time-series simulations, which are vital for correct research of acoustics, are possible.
The little scale and Large scale are meticulously simulated at the same time.

Vorticity Confinement technique

The Vorticity Confinement method ( VC ) is an Eulerian system, renowned for the simulation of turbulent wakes. It employs a solitary-wave like approarch to supply stable solution with no numeric spreading.
VC can capture the little scale features to over as few as two grid cells.
Inside these features, a nonlinear difference equation is solved vs finite difference equation.
VC has similarities to surprise capturing strategies, where conservation laws are satisfied, so the necessary integral quantities are correctly computed.

Two phase flow

The modeling of two-phase flow is still in development. Different strategies have been suggested. The Volume of fluid technique gets lots of attention recently, but Level set and front tracking are also valuable approaches the majority of these strategies are either good in maintaining a pointy interface or at preserving mass. This is critical since the analysis of the density, viscosity and surface stress in primarily based on the values averaged over the interface.

Solution algorithms

Discretization in space produces a system of normal differential equations for unstable problems and algebraic equations for steady issues. Implicit or semi-implicit techniques are typically used to integrate the normal differential equations, manufacturing a system of ( often ) nonlinear algebraic equations. Applying a Newton or Picard iteration produces a system of linear equations which is nonsymmetric in the shadow of advection and unfixed in the shadow of incompressibility. Such systems, especially in 3D, are often too massive for direct solvers, so iterative strategies are used, either still techniques like successive overrelaxation or Krylov subspace techniques.
Krylov techniques like GMRES, usually used with preconditioning, operate by minimizing the residual over successive subspaces generated by the preconditioned operator. Multigrid is particularly popular , both as a solver and as a preconditioner, due to its asymptotically perfect performance on many issues. Standard solvers and preconditioners are efficient at reducing high-frequency elements of the residual, but low-frequency elements often need many iterations to reduce. By operating on multiple scales, multigrid decreases all elements of the residual by similar factors, leading to a mesh-independent number of iterations. For unfixed systems, preconditioners like unfinished LU factorization, addition Schwarz, and multigrid perform poorly or fail completely, so the issue structure need to be used for effective preconditioning. The standard techniques typically employed in CFD are the Straightforward and Uzawa algorithms which exhibit mesh-dependent convergence rates, but latest advances based primarily on block LU factorization mixed with multigrid for the ensuing definite systems, have led to preconditioners which deliver mesh-independent convergence rates.