9.5. Interpolation Algorithms

Several interpolation algorithms are available for use with supported material data tables (defined via the TB command).

Simple linear interpolation is the default for 1-D and 2-D grid data. To obtain meaningful results for multidimensional non-grid data, it is necessary to use one of the multidimensional interpolation algorithms.

To select an interpolation algorithm or to switch between algorithms, issue the TBIN command.

The following topics for interpolation algorithms are available:

9.5.1. Simple Linear Interpolation

Simple linear interpolation can be used for one- and two-dimensional grid data:

9.5.1.1. Simple Linear Algorithm (One-Dimensional)

To demonstrate the interpolation of data in a sparsely defined grid, consider the results of a two-dimensional interpolation at a temperature of 100 and a sliding distance of 0.40. In this case, the program performs only a one-dimensional interpolation because the defined temperature value (100) lies directly on the defined grid field. For this case, the program obtains a friction coefficient value of 0.45 based on the following calculations:

Equation (1)

where

Equation (2)

Substituting the tabular values

x = 0.4,x0 = 0.2,x1 = 0.5
y0 = 0.35y1 = 0.5 

Substituting these values into Equation (2):

Equation (3)

and solving for the interpolated values using Equation (1), we obtain

Equation (4)

9.5.1.2. Simple Linear Algorithm (Two-Dimensional)

Two-dimensional interpolation requires data provided in the proper order. If UF01 and TEMP are the two field variables, for example, input the data in the following format:

  UF01 = 100 UF01 = 200 UF01 = 300 UF01 = 400
TEMP = 0 1234
TEMP = 10 5  6
TEMP = 20 7 89

If the TBFIELD command for TEMP is issued first, all data points with TEMP held constant and UF01 varying are issued. Then TEMP is modified again and data points with UF01 changing are issued.

Consider the case where a true two-dimensional interpolation is required at a temperature of 180 and a sliding distance of 0.40. The program performs three different linear interpolations to determine the property value within the grid. When performing two-dimensional interpolation, the program always interpolates first along the two relevant rows of the grid (Temperature in this case), then between the rows.

In this example, the program performs the first interpolation at a temperature of 100 and a sliding distance of 0.4, yielding the result of 0.45 (as shown in Equation (4)).

The program performs the second interpolation for a temperature of 200 and a sliding distance of 0.4. In this case, we find that

x = 0.4,x0 = 0.2,x1 = 0.5
y0 = 0.2y1 = 0.14 

Substituting these values into Equation (2):

Equation (5)

and solving for the interpolated values using Equation (1), we obtain

Equation (6)

Finally, the program performs a third interpolation between the temperature value of 100 and 200 at a sliding distance of 0.4.

t = 180,t0 = 100,t1 = 200
y0 = 0.45y1 = 0.16 

Substituting these values into Equation (2):

Equation (7)

and solving for the interpolated values using Equation (1), we obtain

Equation (8)

9.5.2. Multidimensional Interpolation

Three algorithms are available for performing interpolation when two or more field variables are input:

Unlike simple linear interpolation, you can input a random collection of data points where the material property was experimentally evaluated. It is not necessary to input data in any specific grid format (although you can still do so), but you must provide enough data to cover the region where you intend to interpolate

Because the data is a random sampling, the algorithm creates an n-dimensional rectangular bounding box, and the queries outside of the bounding box are projected to fall on the surface or edge of the bounding box. The projection is done by calculating a normal to each axis until it finds a position within the bounding box.

In the following figure, the stars represent the data points input that you provide, and the blue stars represent queries outside of the bounding box:

The figure shows how the data points are projected first in one dimension and then in the second dimension. This method was extrapolated and implemented for multiple dimensions.

Nonlinear interpolation can produce slightly different results from that of linear interpolation.

For material data tables that use TBPT-based input (such as TB,PLASTIC), stresses are first evaluated at the neighboring strain values provided in the TBPT input using the specified multidimensional interpolation algorithm, then linearly interpolated between the neighboring values to maintain the stress strain data C0.

9.5.2.1. Linear-Multivariate Algorithm

Linear-multivariate is a local algorithm that uses points near the query location to perform the interpolation process. The algorithm can have discontinuities based on the selection of the neighboring points used for the interpolation process.

The linear-multivariate algorithm is best used for parameter spaces of Order 3 and above. It is faster than other interpolation algorithms for a large selection of data points and is highly accurate when given a sufficient distribution of points. (More data points results in better accuracy.)

To evaluate the interpolation values prior to the finite element solution, issue the *GET command. For more information, see *GET General Items, Entity = TBTYPE in the Mechanical APDL Command Reference and Evaluating Interpolation Algorithm Results.

For more information, see Linear Multivariate Interpolation in the ANSYS Composite PrepPost User's Guide.

9.5.2.2. Radial-Basis Algorithm

Radial-basis interpolation is a global algorithm that is C1 continuous, especially useful for higher-dimensional scatter data.

Interpolation of the parameters is done by solving the following radial-basis function Z = f(X1, X2, X3… XO) at all input data points:

where N is the number of data points and O is the number of free variables (or the order of the interpolation). Input data is (xj,1, xj,2, …, zj) where j varies from 1 to N.

The unknown values are ai (where i varies from 1 to N) and c. Z is the unknown and Xj refers to the query point. The equation is evaluated for all data points provided in the input to calculate the ai and c values.

This algorithm is computationally intensive. Large data sets can affect performance adversely.

9.5.2.3. Nearest-Neighbor Algorithm

Given a query point, the nearest-neighbor algorithm searches the supporting point with the smallest Euclidean distance to the query point and returns the corresponding dependent value.

Implementation of the nearest-neighbor function is expressed as:

where is the set of all supporting points and is the query point.

Use the nearest-neighbor algorithm only for very large scatter-data sets when other algorithms cannot be used.

9.5.3. Evaluating Interpolation Algorithm Results

The following example input shows how to input and evaluate the interpolation algorithm via the *GET command:

! Example: evaluating interpolation of material data tables
/prep7

! Use a six X six X six grid
xpts=6
ypts=6
zpts=6

! 3D Grid is unnecessary. The grid below is created for demonstration
! purposes only.

! Setting up a TB,ELAS property with linear-multivariate interpolation algorithm
tb,elas,1
*DO,x,1,xpts,1
*DO,y,1,ypts,1
*DO,z,1,zpts,1
xloc=(x-1)/(xpts-1)
yloc=(y-1)/(ypts-1)
zloc=(z-1)/(zpts-1)
tbfield,xcor,xloc
tbfield,ycor,yloc
tbfield,zcor,zloc
E=(1+xloc**2+yloc**2+zloc**2)*1e6
tbdat,,E,0.0
*ENDDO
*ENDDO
*ENDDO
tbin,algo,lmul

*dim,ucalc,,101
*dim,zcalc1,,101

! Issue the *GET command to evaluate the value on
! a line across the data
*DO,x,1,101,1
myval=(x-1)/100
ucalc(x)=myval
*get,velas,elas,1,tbev,1,1,3,0.33,0.33,myval
zcalc1(x)=velas
*ENDDO

*stat,ucalc,1,101
*stat,zcalc1,1,101
finish

Figure 9.1:  Evaluating Interpolation for Linear Data with a 6x6x6 Grid

Evaluating Interpolation for Linear Data with a 6x6x6 Grid

Figure 9.2:  Evaluating Interpolation for Quadratic Data with a 6x6x6 Grid

Evaluating Interpolation for Quadratic Data with a 6x6x6 Grid

9.5.4. Material Model Support for Interpolation

Simple linear interpolation is supported for all material data tables.

Linear-multivariate and nearest-neighbor interpolation are supported for these material data tables:

Material Model TB,Lab Value
Coefficient of thermal expansion CTE
Mass density DENS
Elasticity ELASTIC
Hyperelasticity HYPER
Nonlinear plasticity PLAS
Porous media PM

Radial-basis interpolation is supported for all materials listed in Table 9.1: Predefined Field Variables and Supported Material Data Tables.

9.5.5. Reference

For further information about multidimensional field-variable interpolation, consult this reference:

  1. Amidror, Isaac. “Scattered Data Interpolation Methods for Electronic Imaging Systems: A Survey .” Journal of Electronic Imaging. 11:2 (2002): 157-176.


Release 18.2 - © ANSYS, Inc. All rights reserved.