US 6,356,860 B1Grant
Method of grid generation
Issue Date:2002-03-12
•8 Claims
•10 Drawing Sheets
Abstract
The present invention provides a method of grid generation that uses the geometry of the problem space and the governing relations to generate a grid. The method can generate a grid with minimized discretization errors, and with minimal user interaction. The method of the present invention comprises assigning grid cell locations so that, when the governing relations are discretized using the grid, at least some of the discretization errors are substantially zero. Conventional grid generation is driven by the problem space geometry; grid generation according to the present invention is driven by problem space geometry and by governing relations. The present invention accordingly can provide two significant benefits: more efficient and accurate modeling since discretization errors are minimized, and reduced cost grid generation since less human interaction is required.
Metadata
Assignee
- Sandia Corporation
Inventor
- Daniel W. Barnette
Application Information
Application Number:US 09/169,308
Filing Date:1998-10-08
Priority Date:1998-10-08
Art Unit:7
Classifications
IPC:
G06F 1750G06F 760G06F 1710G06G 748G06G 750
Field of Search:
703 2703 9703 1703 7
Patent Drawings (10 sheets)
Description
Background of the Invention
[0003] This invention relates to the field of grid generation, specifically grid generation for computer modeling of governing relations over a problem space.
[0004] Computers can be used to model processes or relations occurring in a problem space. For example, computers can be used to model fluid or heat flow through a problem space such as a chamber, a mechanical part, or a portion of the world. The conventional approach involves the following steps:
[0005] 1. Represent by governing relations (or governing equations) the process or relations to be modeled.
[0006] 2. Construct a grid that represents the problem space and segregates it into cells appropriate to the resolution of model desired.
[0007] 3. Discretize the governing relations to the grid cell boundaries.
[0008] 4. Represent initial and boundary conditions.
[0009] 5. Solve the governing relations for the grid cells using the initial and boundary conditions.
[0010] 6. Iterate if necessary (for example, if the governing relations require iteration for convergence).
[0011] The governing relations matching many processes are continuous relations. Discretizing such governing relations can introduce discretization errors. Discretization errors can delay or even prevent convergence, making an accurate model more expensive or even impossible to obtain. They can also compromise the accuracy of the resulting model.
[0012] Generating a grid that represents the problem space also incurs substantial cost. Grid generation conventionally requires significant interaction and direction from a human user. Specialized grid generation expertise and complex grid generation tools are typically employed to help generate a grid that represents the problem space and is amenable to computer solution of the governing relations. See, e.g., Thompson et al., Numerical Grid Generation—Foundations and Applications, Elsevier Science Publishing Company, 1985; Carey, Computational Grids—Generation, Adaptation, and Solution Strategies, Taylor and Francis Publishing Company, 1997; Hoffman and Chiang, Computational Fluid Dynamics for Engineers, Engineering Education System, Wichita, Kans., USA, 1993. Correction of deficiencies in the grid typically requires expert analysis of the model results and another costly application of the grid generation expertise and tools.
[0013] The cost and deficiencies associated with current grid generation methods restrict the application of computers to model processes and relations. Accordingly, there is a need for grid generation methods that do not require intensive human user direction and that do not lead to modeling inaccuracies.
Summary of the Invention
[0014] The present invention provides a method of grid generation that uses the geometry of the problem space and the governing relations to generate a grid. The method can generate a grid with minimized discretization errors, and with minimal user interaction.
[0015] The method of the present invention comprises assigning grid cell locations so that, when the governing relations are discretized using the grid, at least some of the discretization errors are substantially zero. Conventional grid generation is driven by the problem space geometry; grid generation according to the present invention is driven by problem space geometry and by governing relations. The present invention accordingly can provide two significant benefits: more efficient and accurate modeling since discretization errors are minimized, and reduced cost grid generation since less human interaction is required.
[0016] As an example, grid generation according to the present invention can be practiced for solving Navier-Stokes fluid flow relations. A grid can be generated so that TE1discretization errors are zero using the problem space geometry and the Navier-Stokes relations. Further, a refined grid can be generated so that TE2discretization errors are zero using the problem geometry, the Navier-Stokes relations, and intermediate solutions. Grids appropriate to the Navier-Stokes relations and the problem space can accordingly be generated with little human interaction. Also, refined grids can be generated so that finer resolution is obtained where needed, and coarser resolution is used where sufficient.
[0017] Advantages and novel features will become apparent to those skilled in the art upon examination of the following description or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.
Description of the Figures
[0018] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
[0019] FIG. 1 is a diagram of a grid and coordinates according to the present invention.
[0020] FIG. 2(a,b) is an example of a grid generated according to the present invention.
[0021] FIG. 3(a,b) is an example of a grid generated according to the present invention.
[0022] FIG. 4 is a flow diagram according to the present invention.
[0023] FIG. 5 is a flow diagram according to the present invention.
[0024] FIG. 6 is a flow diagram according to the present invention.
[0025] FIG. 7ais a diagram of an apparatus according to the present invention.
[0026] FIG. 7bis a diagram of an apparatus according to the present invention.
Detailed Description of the Invention
[0027] The present invention provides a grid generation method that requires reduced human interaction and can provide increased solution speed and accuracy. Discretizing governing relations for solution on a grid can result in discretization error terms. According to the present invention, grid cell locations are assigned so that at least some of those discretization error terms are minimized. Since the grid cell locations are assigned based on the governing relations and the problem space, less human interaction is required than for conventional grid generation methods. Further, since the grid cell locations minimize discretization error terms, faster and more accurate solutions can be obtained.
[0028] The nature of the governing relations can affect what information is needed to assign grid cell locations according to the present invention. Some discretization errors can be dependent on problem space geometry and governing relations. The method of the present invention can generate grids that minimize these discretization errors before any solution is begun. Other discretization errors can further depend on the evolution of the solution. The method of the present invention accordingly can require information from initial solutions to generate grids that minimize these discretization errors. Details are provided below for practice of the present invention in conjunction with two-dimensional convection-dominated Navier-Stokes fluid flow equations, minimizing either TE1, TE2, or both, discretization errors. Those skilled in the art will appreciate other dimensionalities, other governing relations, and other discretization errors that are appropriate for use with the present invention.
[0029] Convection Discretization Error Terms
[0030] Two-dimensional convection terms in the Navier-Stokes equations can be written in transformed computational space as in Equation 1.
[0031] In Equation 1, f can be either the u or v covariant velocity, p is the density, and the Jacobian of the transformation is given by Equation 2.
[0032] J=xξyη−yξxη Equation 2
[0033] In Equation 1, U and V are the contravariant velocities as in Equations 3 and 4.
[0034] U=uyη−vxη Equation 3
[0035] V=vxξ−uyξ Equation 4
[0036] The independent variables ξ and η are the coordinate directions in computational space.
[0037] The discretization errors for convection terms can be arrived at by using second-order upwind differencing for the convection terms and second-order central differencing for the metric coefficients. The total discretization error can be expressed as in Equation 5.
[0038] TE=TE1+TE2+(higher order terms) Equation 5
[0039] In Equation 5, TE1is given by Equation 6.
[0040] In Equation 5, TE2is given by Equation 7.
[0041] Note that grouping is dependent on the order of the derivative f. Derivatives of f can be taken in physical space to allow more compact form of equations; they can be transformed to computational space when necessary. The analysis of higher order terms, while within the scope of the present invention, can be quite arduous. Accordingly, the discussion below concentrates on obtaining useful grid generation information from TE1and TE2.
[0042] TE1can be made zero as discussed below. The relations set forth allow a grid to be generated where TE1is zero without coupling to flow solver data. This allows the present invention to generate a grid before operation of the flow solver, similar to the sequencing required with conventional grid generation techniques.
[0043] TE2can be made zero as discussed below. The relations set forth allow a grid to be generated where TE2is zero by coupling grid generation to the flow solver through velocity gradients. The relations are 2ndorder, linear, hyperbolic, partial differential equations with variable coefficients. The algebraic results and the system of partial differential equations can form the basis of the method whereby the grid can be generated simultaneously with the development of the flow fields.
[0044] Grid Generation so TE1=0
[0045] The derivation of grid generation relations that provide a grid with TE1=0 can begin by setting Equation 6 to zero. One approach is to set the coefficients of fxand fyto zero independently. The resulting equations are set forth as Equations 8 and 9.
[0046] UXξξξ+Vxηηη=0 Equation 8
[0047] Uyξξξ+Vyηηη=0 Equation 9
[0048] If a distribution in x and y that satisfies Equations 8 and 9 can be constructed, then TE1 will be zero. Rewriting Equations 8 and 9 in matrix form yields Equation 10.
[0049] A nontrivial solution of Equation 10 can be obtained only if the determinant of the matrix is zero, as expressed in Equation 11.
[0050] xξξξyηηη−xηηη=0 Equation 11
[0051] The matrix elements must be determined or specified. With four unknowns and two equations, two of the matrix elements must be specified unless all terms are set to zero. For example, grid distributions can be constructed such that only xξξξ and xηηη are zero, satisfying Equation 11. In this case, yηηη would be determined by Equation 9 with yξξξspecified (or vice versa). This would render part of the analysis closely coupled to the flow solver. The equations covering grid generation would then be as set forth in Equations 12 and 13.
[0052] xξξξ=xηηη=0 Equation 12
[0053] yηηη=−U/V yξξξ Equation 13
[0054] Another approach leads to an algebraic set of grid equations, and involves setting all grid derivatives to zero, as set forth in Equation 14.
[0055] xξξξ=xηηη=yξξξ=yηηη=0 Equation 14
[0056] Integration of xξξξ leads to the generic form of the solution in ξ as set forth in Equation 15.
[0057] In Equation 15, F1(η), F2(η), and F3(η) are functions of η only. Note that the quantity (ξ−1) is used in place of ξ with no loss in generality. Integrating xηηη, the generic form in is set forth in Equation 16.
[0058] Equations 15 and 16 represent a family of η and ξ curves. Each family has three ‘constants’ of integration, all of which can be determined from the boundary conditions listed in Table 1 for an arbitrary grid. The last condition consists of either specifying orthogonality at the left boundary or specifying a curve interior to the grid. Equations are set forth below for both conditions.
| TABLE 1 | |
|---|---|
| Computational Coordinates | Physical Coordinates |
| At ξ=1, 1≦η≦KMAX | x=x |
| At ξ=JMAX, 1≦η≦KMAX | x=x |
| At η=1, 1≦ξ≦JMAX | x=x |
| At η=KMAX, 1≦ξ≦JMAX | x=x |
[0059] The subscripts L, R, B, and T for x and y correspond to the Left, Right, Bottom, and Top of the grid as illustrated in FIG. 1. Indices (ξ,η)=(1,1) correspond to the point at which the Left and Bottom boundaries meet. Also note that xB′≡dxB(ξ)/dξ, yB′≡dyB(ξ)/dξ, etc. The term ΔS specifies the distance of the first grid point off the wall. Evaluating the constants in Equations 15 and 16 with the generic boundary conditions listed in Table 1, and imposing the orthogonality condition at the left boundary, leads to Equations 17 and 18.
[0060] The family of curves represented by Equations 17 and 18 can be combined to interpolate the entire region and form a computational mesh. The two equations are combined using Boolean Sum projectors. The final equation is given by Equation 19.
[0061] The analysis for y leads to Equations 20 and 21.
[0062] Combining Equations 20 and 21 using Boolean Sum projectors yields Equation 22.
[0063] The above grid distribution ensures that TE1will be zero. FIGS. 2(a,b) and 3(a,b) show example grids generated according to the above expressions of the present invention.
[0064] FIG. 4 is a flow diagram of a process according to the above expression of the present invention. The process begins with a definition of a surface grid defining the geometry of interest 41. The boundaries of the grid must be defined 42. Volume grid cells can then be generated algebraically 43 according to Equations 19 and 22. The governing relations can then be solved 44 using the surface and volume grids. Solution of the governing relations can converge more rapidly since the grid used does not introduce discretization errors.
[0065] Pseudocode that implements the above process for TE1=0 is listed in Table 2. Detailed programming for specific computers starting from the pseudocode is known to those skilled in the art. Grids can be output using the well-known PLOT3D format. See, e.g. Walatka et al., “PLOT3D User's Manual,” NASA Technical Memorandum 101067, March 1990. Grids can be visualized on a computer using any graphics program that reads in PLOT3D-formatted grids.
| TABLE 2 |
|---|
| This pseudocode outlines the process of generating a grid from the equations derived by setting T |
| generated is an algebraic grid. The four edge curves must be defined as well as the spacing of the first point off the |
| wall. The wall is assumed to be at eta=1 and is a function of xi. |
| dimension, parameter, and implicit statements; open input and output files. |
| perform any checks necessary to ensure grid fits within dimension statements mentioned above |
| call subroutines defining geometry and define boundaries of geometry of interest; |
[0066] Grid Generation so TE1=0
[0067] Grid generation equations that drive TE2to zero can be derived similarly to those for driving TE1to zero. The analysis can be significantly more complicated since TE2contains more derivatives and a cross derivative term. The derivation below focuses on arriving at equations for a grid distribution that are dependent on the flow solution, hence coupling the flow solver with the grid generation.
[0068] Rearranging Equation 7 to have factors that multiply the contravariant velocity terms U and V yields Equation 23.
[0069] U└xξxξξfxx+(yξxξξ+xξyξξ)fxy+yξyξξfyy┘+V└xηxηηfxx+(yηxηη+xηyηη)fxy+yηyηηfyy┘ Equation 23
[0070] Setting the coefficients of U and V to zero and rearranging yields equations 24 and 25.
[0071] (xξfxx+yξfxy)xξξ+(xξfxy+yξfyy)yξξ=0 Equation 24
[0072] (xηfxx+yηfxy)xηη+(xηfxy+yηfyy)yηη=0 Equation 25
[0073] Setting the coefficients to zero in Equation 24 yields Equations 26 and 27.
[0074] xξfxx+yξfxy=0 Equation 26
[0075] xξfxy+yξfyy=0 Equation 27
[0076] Multiplying Equation 26 by xξ, Equation 27 by yξ, and subtracting removes the cross derivative term, as in Equation 28.
[0077] xξ2fxx−yξ2fyy=0 Equation 28
[0078] An analogous result for Equation 25 is shown in Equation 29.
[0079] xη2fxx−yη2fyy=0 Equation 29
[0080] Equation 30 shows the matrix form.
[0081] Setting the determinant of the matrix to zero gives a non-trivial solution, as shown in Equation 31. In Equation 31, J is the Jacobian in Equation 2.
[0082] xξ2yη2−xη2yξ2=(xξyη+xηyξ)(xξyη−xηyξ)=(xξyη+xηyξ)J=0 Equation 31
[0083] Further analysis of the above equations allows solution for xξη and yξη, as in Equations 32 and 33. A detailed exposition of the analysis is given by Barnette in “A Mathematical Basis for Automated Structured Grid Generation with Close Coupling to the Flow Solver,” AIAA 16thApplied Aerodynamics Conference, 1998, incorporated herein by reference.
[0084] xξη=Ψ1xξ+Ψ2xη Equation 32
[0085] yξη=Ψ1yξ+Ψ2yη Equation 33
[0086] In Equations 32 and 33 the velocity derivatives are expressed in computational space, and the coefficients represented by Ψ are given by Equations 34 and 35.
[0087] If only Equations 32 and 33 equations are considered, it must be assumed that some initial grid points close to the geometry of interest are supplied, for example by a user. An initial solution can obtained on this partial grid. Equations 32 and 33 are known to be hyperbolic, 2nd-order, linear partial differential equations with variable coefficients. These equations may be finite differenced to provide equations for predicting the next layer of grid points as determined by the velocity gradients of the initial solution. By iterating back and forth between the grid generation algorithm, which adds more grid point layers, and the solution of the governing equations, which provides updated velocity gradients for the variable coefficients of Equations 32 and 33, computational grids can be generated simultaneously with the solution. This method of grid construction ensures accuracy by annihilating the dominant convective truncation error terms.
[0088] FIG. 5 is a flow diagram of a process according to the above expression of the present invention. A surface grid defining the geometry of interest is known, for example by definition by a user 51. Initial volume grid cells near the surface are generated, for example by definition by a user 52 or according to the TE1expression of the present invention as described above. An initial solution of the governing relations can then be generated using the initial volume grid cells 53. Velocity gradients on the existing grid cells can then be determined 54. Additional layers of grid cells can then be determined from the velocity gradients and Equations 32 and 33. The above process can be repeated until a desired terminal condition is reached, for example until a solution to the governing relations converges 56. The iterative solution-grid construction process can encourage faster, more accurate solution by minimizing TE2truncation errors.
[0089] Pseudocode that implements the above process for TE2=0 is listed in Table 3. Detailed programming for specific computers starting from the pseudocode is known to those skilled in the art. Grids can be output using the well-known PLOT3D format. See, e.g. Walatka et al., “PLOT3D User's Manual,” NASA Technical Memorandum 101067, March 1990. Grids can be visualized on a computer using any graphics program that reads in PLOT3D-formatted grids.
| TABLE 3 |
|---|
| This pseudocode generates a grid such that TE2=0 in the higher-order convective truncation such that these error |
| terms will be zero. |
| dimension, parameter, and common statements go here |
[0090] Grid Generation so TE1=TE2=0
[0091] The convection truncation error analyses above provide equations that can be used for automated grid generation. One set of equations for constructing grids is uncoupled from the flow solver; the other set, coupled. This permits an initial grid to be generated that ensures the leading convection truncation error term is zero. The initial grid is assumed to be of minimal construction and rather arbitrarily placed, with for example five grid points off the wall. Once a flow solver has been run on the initial grid for several time steps and possibly some prominent features of the flow have begun to emerge, the non-converged solution can be fed into the coupled grid generation equations to intelligently redistribute the grid points and possibly enhance flow field convergence. This redistribution ensures the second leading truncation error term to be zero. A layer of grid points can then be added to the previous outer layer of grid points, the process of which is discussed below. The algorithm then iterates between the flow solver and the grid construction until the flow solver indicates convergence and no more outer-layer grid points need be added.
[0092] The process of intelligently adding more grid points to the previous grid's outer layer is the method by which the grid is constructed or, perhaps more appropriately, built up. This allows the user to begin the flow solution with nothing more than the surface grid describing the geometry about which the flow field is to be calculated, and an initial coarse grid rather arbitrarily constructed with the use of algebraic methods such as those given by Equations 19 and 22. The equations used to construct, or add, the new outer layer are given by Equations 32 and 33. These equations are hyperbolic. A proper finite difference stencil allows the equations to predict a grid level, i.e. the new outer level, above those points at which the velocities and their gradients have been determined from the previous iteration. Development of this process is straightforward to those skilled in the art or may be learned by practice of the invention.
[0093] FIG. 6 is a flow diagram of a process according to the above expression of the present invention. A surface grid defining the geometry of interest is known, for example by definition by a user 61. Initial volume grid cells near the surface are generated, for example by definition by a user 62. Additional grid points can then be determined 63 according to the present invention (as expressed in Equations 19 and 22). Determining the additional grid points according to Equations 19 and 22 ensures that TE1=0, thereby minimizing errors introduced by discretization. An initial solution of the governing relations can then be generated using the initial volume grid cells 64. If the solution has not yet converged, then velocity gradients on the existing grid cells can then be determined 66. Additional layers of grid cells can then be determined 67 from the velocity gradients and Equations 32 and 33. The above process can be repeated until a desired terminal condition is reached, for example until a solution to the governing relations converges 65. The iterative solution-grid construction process can encourage faster, more accurate solution by minimizing TE2truncation errors.
[0094] Apparatus
[0095] The present invention also comprises apparatus adapted to practice the method described above. FIG. 7ashows an apparatus 800 according to the present invention. A processor 801 connects with an input/output subsystem 802 and storage subsystem 803. Apparatus 800 also comprises means for representing, in storage subsystem 803, a representation of a surface grid defining the geometry of interest 804a, means for generating volume grid points 804baccording to Equations 19 and 22, and means for solving the governing equations using the surface and volume grids 804c. For example, processor 800 can comprise a workstation such as contemporary workstations from Sun, IBM, HPUX, and Silicon Graphics, with local storage such as disk and semiconductor memory, input/output such as a keyboard, display, and network connection. The means for representing, in storage subsystem 803, a representation of a surface grid defining the geometry of interest 804a, means for generating volume grid points 804baccording to Equations 19 and 22, and means for solving the governing equations using the surface and volume grids 804c, can comprise computer software, hardware, network downloads, or combinations thereof. Apparatus 800 can also comprise a multiprocessor computer such as the Intel Paragon.
[0096] FIG. 7bshows an alternate apparatus 810 according to the present invention. A processor 811 connects with an input/output subsystem 812 and storage subsystem 813. Apparatus 810 also comprises means for representing, in storage subsystem 813, a representation of a surface grid defining the geometry of interest 814a, means for representing, in storage subsystem 813, a representation of initial layers of volume grid points 814b, means for solving the governing equations using existing grid points 814c, means for determining velocity gradients using existing grid points 814d, means for generating volume grid points 814eaccording to Equations 32 and 33, and means for iterating until a desired terminal condition is reached 814f. For example, processor 810 can comprise a workstation such as contemporary workstations from Sun, IBM, HPUX, and Silicon Graphics, with local storage such as disk and semiconductor memory, input/output such as a keyboard, display, and network connection. The means for representing, in storage subsystem 813, a representation of a surface grid defining the geometry of interest 814a, means for representing, in storage subsystem 813, a representation of initial layers of volume grid points 814b, means for solving the governing equations using existing grid points 814c, means for determining velocity gradients using existing grid points 814d, means for generating volume grid points 814eaccording to Equations 32 and 33, and means for iterating until a desired terminal condition is reached 814f, can comprise computer software, hardware, network downloads, or combinations thereof. Apparatus 810 can also comprise a multiprocessor computer such as the Intel Paragon.
[0097] The particular sizes and equipment discussed above are cited merely to illustrate particular embodiments of the invention. It is contemplated that the use of the invention may involve components having different sizes and characteristics. It is intended that the scope of the invention be defined by the claims appended hereto.
Claims
We claim:
1. In a computer prediction of fluid flow within a problem space, wherein the fluid flow is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the fluid flow, comprising dividing the problem space into grid cells by:
a) Generating a grid having grid cell locations assigned according to Equation 6 with TE1 set to zero, wherein the grid cell locations are assigned according to Equations 19 and 22;
b) Using the computer to model said fluid flow using said grid, comprising:
i) Using the computer to predict velocity derivatives in a subset of said grid;
ii) Generating a refined grid adjacent said subset having grid cell locations assigned so that TE2 errors resulting from discretization of the fluid flow model are substantially zero;
iii) Using the computer to model said fluid flow using said refined grid.
2. The method of claim 1, wherein said step of generating a refined grid comprises generating a refined grid having grid cell locations assigned according to according to Equation 7 with TE2 set to zero.
3. The method of claim 1, wherein said step of using the computer to model said fluid flow comprises repeating steps ii) and iii), using velocity derivatives from step iii) in step ii), until a terminal condition is reached.
4. In a computer prediction of the behavior of a system, wherein the system is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the behavior, comprising dividing the problem space into grid cells, wherein grid cell locations are assigned according to Equation 6 with TE1 set to zero, wherein the grid coordinates are assigned according to Equations 19 and 22.
5. In a computer prediction of the behavior of a system, wherein the system is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the behavior, comprising dividing the problem space into grid cells, wherein grid cell locations are assigned according to Equation 7 with TE2 set to zero, wherein the grid coordinates are assigned according to Equations 32 and 33.
6. In a computer prediction of the behavior of a system, wherein the system is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the behavior, comprising dividing the problem space into grid cells by:
a) Generating an initial grid having grid cell locations assigned so that TE1 errors resulting from discretization of said governing relations are substantially zero, wherein the initial grid cell locations are assigned according to Equations 19 and 22;
b) Using the computer to generate an initial solution of the governing relations for an initial subset of the said initial grid;
c) Generating a subsequent grid from said initial solution having grid cell locations assigned so that TE2 errors resulting from discretization of said governing relations are substantially zero;
d) Using the computer to generate a subsequent solution from said initial solution and said subsequent grid; p1 e) Repeating steps c) and d), using said subsequent solution for said initial solution and said subsequent grid for said initial grid, until a terminal condition is reached.
7. In a computer prediction of fluid flow within a problem space, wherein the fluid flow is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the fluid flow, comprising dividing the problem space into grid cells by:
a) Generating an initial grid having grid cell locations assigned so that TE 1 errors resulting from discretization of said governing relations are substantially zero;
b) Using the computer to generate an initial solution of the governing relations for an initial subset of the said initial grid;
c) Generating a subsequent grid from said initial solution having grid cell locations assigned so that TE2 errors resulting from discretization of said governing relations are substantially zero, wherein the subsequent grid cell locations are assigned according to Equations 32 and 33;
d) Using the computer to generate a subsequent solution from said initial solution and said subsequent grid;
e) Repeating steps c) and d), using said subsequent solution for said initial solution and said subsequent grid for said initial grid, until a terminal condition is reached.
8. In a computer prediction of fluid flow within a problem space, wherein the fluid flow is represented in the computer by a set of governing relations applied to a problem space, a method for generating a grid to for use in a finite element prediction of the fluid flow, comprising dividing the problem space into grid cells by:
a) Generating a grid having grid cell locations assigned according to Equation 6 with TE1 set to zero, wherein the grid cell locations are assigned according to Equations 19 and 22;
b) Using the computer to model said fluid flow using said grid.
Patent Citations (2)
Non-Patent Literature (3)
- Lin et al.; “Automatic grid refinement and higher order flux discretization for diffusion modelin”; IEEE Trans. Comp. Aided Design of Int. Circuits; pp. 1209-1216, Aug. 1993.Applicant
- Edwards: “Elimination of adaptive grid interface errors in the discrete cell centered pressure equation”; J. Computational Physics; pp. 356-372, Jul. 1996.Applicant
- Eiseman, “Automatic Structured Grid Generation,” in Computatiional Fluid Dynamics Review 1996.Applicant