Several mapping algorithms are used when executing data transfers during system couplings. To assist in evaluating the quality of the mapping, a mapping summary is included in the System Coupling service log file (see System Coupling Service Log File (scLog.scl_, scLog_##.scl)). Note that summary data depends upon on the availability and relevance of specific information (for example, the number of nodes or area on the surface and/or target meshes) for each mapping algorithm.
Mapping is performed only at the start of the System Coupling simulation. Because of this, the mesh topology on the data transfer regions cannot change (that is, cannot be dynamically remeshed) during the simulation.
The two mapping algorithms used in System Coupling (discussed below) are Bucket Surface and General Grid Interface (GGI).
Data transfer algorithms are combinations of mapping/interpolation algorithms (discussed in the sections above) that are used in the System Coupling service.
Note that the fidelity of the data transferred to the target side of the interface is limited by the least-resolved side of the interface. For example, if the target side of the interface is significantly coarser than the source side, then only the large scale features of the source data will be captured in the data transfer. Similarly, if the target side of the interface is significantly finer than the source side, then the resulting target data will be a linearly interpolated representation of the data on the source side.
Note: A number of advanced controls for the data transfer algorithms are exposed via expert settings. For more information, see Expert Settings Related to Mapping in Expert Settings.
The Profile Preserving data transfer algorithm is the default
algorithm used by System Coupling when transferring non-conserved
quantities like displacements and temperatures. For this data transfer
algorithm, the Bucket Surface mapping algorithm is used to generate
mapping weights. In this algorithm, the mesh nodes on the target
side of the data transfer interface are mapped onto mesh elements
on the source side as illustrated in Figure 8: Mapping target node to source element for Profile Preserving
Data Transfer. Standard, weight-based
interpolation (resulting in the values shown) and
subsequent under-relaxation are used to evaluate the final data applied
on the target side of the interface.
Profile Preserving data transfer algorithm is the default algorithm used when transferring non-conservative quantities because of the profile-preserving nature of the mapping weights generated by the Bucket Surface algorithm.
The Conservative Profile Preserving data transfer algorithm is the default algorithm used by System Coupling when transferring conserved quantities like mass, momentum, and energy flows (for example, forces). For this data transfer algorithm, the General Grid Interface (GGI) mapping algorithm is used to generate mapping weights. Standard, weight-based interpolation and subsequent under-relaxation are used to evaluate the final data applied on the target side of the interface.
Conservative Profile Preserving data transfer algorithm is the default algorithm used when transferring conserved quantities because of the conservative nature of the mapping weights generated by the GGI algorithm. Resulting target values are locally (in the vicinity of each source and target element) conservative. If the source side of the interface is completely mapped to the target side of the interface, then the resulting target values are also globally conservative. If any portions of the source side of the interface are not mapped onto the target side, then the data transfer is not globally conservative. Note that any portions of the target side of the interface that are unmapped (that is, weights equal to zero) are automatically assigned a value of zero, which differs from the handling of unmapped nodes using the Profile Preserving data transfer algorithm.
The underlying ideas for this algorithm are presented in the book Computational Nonlinear Mechanics in Aerospace Engineering, American Institute of Aeronautics and Astronautics, edited by S. Atluri, ISBN 1563470446, Chapter 5, Fast Projection Algorithm for Unstructured Meshes by K. Jansen, F. Shakib, and T. Hughes, 1992. Specifically, the implementation of the Smart Bucket Algorithm as described in the chapter stated above is used in system coupling. This algorithm generates weights that are ideal for transferring the profiles of non-conserved quantities like stresses, displacement, temperature, and heat transfer coefficient from a source mesh to a target mesh. Since a complete description of the algorithm is available in the reference quoted above, only a brief overview of it is presented below.
The first step in the process of computing the mapping weights using the Smart Bucket Algorithm is to divide the mapping source mesh into an imaginary structured grid, with each grid section called a “bucket.” A 2D bucket is used to demonstrate this concept in Figure 9: Example of a Bucket Grid on a 2D Source Mesh. Similarly, a 3D bucket grid is generated for a 3D mesh, and this is what is used in System Coupling.
Next, each node on the data transfer regions of the target mesh is initially associated with a bucket. In System Coupling, data transfer regions consist of element faces from the 3D mesh. Two cases arise: buckets associated with the target node are either empty (without even one source element in it) or non-empty. For example, bucket A shown in Figure 9: Example of a Bucket Grid on a 2D Source Mesh is empty. Each case (empty and non-empty buckets) is discussed separately in the sections below.
Case 1: The bucket associated with a target node is non-empty
If the bucket associated to a given target node is non-empty, the mapping algorithm attempts to match each of the target nodes to one source element in the bucket.
First, each target node is checked to see if it is in the domain
of any of the source elements. This is done by looping through all
the source elements in that bucket and checking to see if the target
node is within their domain. For each source element in the bucket,
the vector element–local (or natural) coordinates (corresponding to the
vector of global coordinates of the target node,
) is found
by solving the set of nonlinear equations given by the isoparametric
mapping below:
(4) |
where is the matrix of linear shape functions
associated with the source element and
is the vector
of global coordinates of element–local node
. It is then checked to see if
lies within the domain of the source element based
on certain criteria discussed next.
For a four-noded quadrilateral source element, if the natural
coordinates corresponding to a target node satisfy the conditions
in Equation 5 below, the target node is said
to be exactly within the domain of the element.
(5) |
where and
are the components of the vector of natural coordinates
. However, if the natural
coordinates do not satisfy the conditions in Equation 5 but do satisfy the ones in Equation 6 below, then the target node is in the
domain of the source element but only within the specified tolerance
(also known as element edge tolerance). The value of tolerance is
exposed in the System Coupling UI as one of the expert settings. See
the description of
ProfMapEdgeTol in the
section Expert Settings Related to Mapping in Expert Settings.
(6) |
This concept is explained with the help of Figure 10: A Quadrilateral Source Element in the Natural Coordinate Space wherein a quadrilateral source element is shown along with two different target nodes, one of which satisfies Equation 5, and other that satisfies Equation 6.
Similarly, for a three-noded triangle element, the conditions listed in Equation 7 below, are used to check if a target node is exactly within the domain of the element:
(7) |
And the conditions in Equation 8 below
will determine if the target node is within the domain but up to a
tolerance .
(8) |
Now that target nodes are determined to be in the domain of specific source elements, each node must be paired with only one source element. In both of the cases above (four-noded quadrilateral and three-noded triangle), it is possible that a target node occurs (either exactly or within a tolerance) in more than one source element’s domain. The finite element interpolation of the nodal solution requires each target node to be paired with only one source element.
To satisfy this requirement, the target node is consequently paired with that source element for which the gap is minimized. The gap is defined as the Euclidean distance between the target node and its projection onto/into a source element. In some cases, such as when candidate source elements are co-planar, the gap values may be identical and an alternate approach is required to pair the target node with one source element. Under these conditions, only the source elements with identical (and minimized) gaps are considered. The target node may be exactly in the domain of any of these source element, or it will be in their domain within a tolerance. Preference is given to the last source element for which the target node is exactly in its domain. If the target node is only in the different domains within a tolerance, then the last candidate source element is used.
Once the target node is paired with a source element, mapping weights are computed by evaluating the finite element shape functions associated with the paired source element at the target node.
If no target node-to-source element match is found in a non-empty bucket, then the target node is reported as being unmapped. It is important to note, however, that mapping weights are still evaluated for such nodes using the Bucket Surface Algorithm. Specifically, all unmapped target nodes are simply mapped to the nearest source node in the bucket and the target node is assigned the solution value corresponding to that source node.
Note: Significant ‘gap’ distances between successfully-mapped target nodes and source elements may occur. For information about how to have mapped nodes with ‘gap’ distances larger than a specified tolerance be reported as unmapped, see Expert Settings Related to Mapping in Expert Settings.
Case 2: The bucket associated with a target node is empty
If the bucket initially found for the target node is empty, then the closest non-empty bucket is found and the same procedure as highlighted in Case 1 is followed so that each target node is mapped to one source element and mapping weights are calculated.
Unmapped Nodes
With the Bucket Surface algorithm, there are two types of target nodes that can be reported as unmapped: nodes that do not fall within a bucket (these are “unmatched nodes”), and nodes that do fall within a bucket, but that do not meet the gap tolerance (these are "gap nodes”). Unmatched nodes are mapped to the nearest source node in the bucket and the target node is simply assigned the solution value corresponding to that source node. Unmatched nodes are always reported as unmapped in the SCL file. Gap nodes are within a bucket, and so are mapped to the source nodes like all of the other mapped nodes and given a value accordingly. Gap nodes are reported as mapped in the SCL file. The gap tolerance and the reporting of gap nodes in the SCL file can be modified using Expert Settings.
The underlying ideas for this algorithm are presented in the article on Three- Dimensional Navier Stokes Predictions of Steady-State Rotor/Stator Interaction with Pitch Change, 3rd Annual Conference of the CFD, Society of Canada, Banff, Alberta, Canada, Advanced Scientific Computing Ltd, by P.F. Galpin, R.B. Broberg and B.R. Hutchinson, June 25-27, 1995. This algorithm generates weights that are ideal for transferring conserved quantities such as mass, momentum and energy flows.
In this algorithm, each element face on both the source and the target sides is first divided into n integration point (IP) (sub-) faces, where n is the number of nodes on the face. The three-dimensional IP faces are then converted into a two-dimensional quadrilaterals made up of rows and columns of pixels. Pixels from the converted quadrilaterals on the source and target sides are intersected, creating a number of overlapped areas called ‘control surfaces.’ Mapping weight contributions are evaluated for each control surface based upon the associated source and target element face areas and the pixel intersections. Final mapping weights for each of the target (or source) nodes are evaluated by accumulating these control surface contributions.
If no control surfaces are created (for example, when no polygon intersection between mapping source and target exists), then mapping weights are identically zero and nodes and elements on the target (or source) side of the interface are reported as being unmapped.
As an example, consider the schematic shown in Figure 11: General Grid Interface Mapping that corresponds to a typical interface between the source (sending) and target (receiving) sides. In the schematic, the control surfaces resulting from the intersection of all IP faces on the interface (labeled with an ‘X’), are shown. For example, the IP faces S1 and S2 on the source side intersect with the IP faces R1 and R2 on the target side creating areas A1, A2, and A3 on the control surface. In this case, the mapping weight contributions for the target IP face R1 (and associated target node) that are associated with the source IP faces S1 and S2 (and nodes) are respectively given by:
(9) |
and
(10) |