1. Introduction
For computational fluid dynamics (CFD) computations and fluid-structure interaction (FSI) computations, a flows simulation with a moving boundary are often encountered at various important problems such as stability analysis of bridges, aircraft optimizations, flutter calculations, and bioengineering simulations [
1,
2,
3,
4,
5]. The mesh deformation algorithm is required to adapt the CFD computational mesh with the motion of the boundary. With regards to the problems with large deformation of boundary, performance of mesh deformation methods often becomes a limiting factor to CFD simulations accuracy [
6]. Meanwhile, the computational mesh needs to be updated repeatedly during an unsteady simulation, bringing large computational cost. Accuracy and computational efficiency of mesh deformation algorithm is crucial for the previously mentioned problem [
7,
8].
Variety of mesh deformation methods have been developed in literature in terms of accuracy, efficiency, and robustness [
9]. These methods can be classified into two main categories: connectivity-based methods and node-based method. The spring analogy method [
10,
11] is a typical one of connectivity methods. The idea of the spring analogy method is to create a network of springs connecting all nodes in the mesh and the stiffness of springs is inversely proportional to the edge length [
12]. The whole mesh can be viewed as a spring system. It has been successfully applied to many unsteady and optimization problems. However, this method is relatively expensive in computational cost due to the necessity of total mesh connectivity information, and it may lead to an invalid mesh for large deformation [
13,
14]. Although several strategies have been introduced to improve the ability of the spring analogy method [
15,
16,
17], it is still difficult to guarantee both high efficiency and applicability. In the other classification of the node-based method, each node can be modified independently to its adjacent nodes, and then the deformation of mesh can be adjusted indiscriminately [
12]. This method allows for large deformation and does not require connectivity information between mesh nodes. The node-based method is useful for unstructured mesh due to those characteristics. Interpolation techniques are often introduced in a node-based method. Witteveen proposed the Inverse Distance Weighed (IDW) interpolation method [
18] in which the deformation of internal nodes depends on the weighted-averaging of the displacement of boundary nodes. Transfinite interpolation (TFI) developed by Gaitonode [
19] is a fast scheme used in a node-based method. Despite its efficiency, the deformed mesh quality is not guaranteed, especially the orthogonality of the mesh near the boundary. Liu proposed a coarse Delaunay graph with barycentric interpolation to propagate deformation from boundary nodes to internal ones [
20]. In addition, Luke et al. decreased computational cost of the previously mentioned method via tree-code optimization [
21]. Although the common node-based methods increase the efficiency of generation of adapted mesh, it does not perform well with complex mesh and is not capable of preserving the mesh orthogonality.
A well-established method based on radial basis functions (RBFs) proposed by Boer [
22] is a desirable approach for both structured and unstructured meshes. It can generally preserve the mesh quality with reasonable orthogonality near the boundary in mesh deformation [
23]. The research shows that different RBFs introduced could result in different results [
22,
24,
25]. Jokonsson et al. describe suggestions for choosing RBFs, and demonstrate that the computational cost will still grow as the mesh scale increases, even choosing the appropriate functions [
1]. Considering the high computational cost, several approaches are proposed with greedy algorithm [
21,
26,
27] or reducing the interpolated nodes [
28,
29,
30]. The improved RBFs method has better computational efficiency and can also preserve the mesh quality well. Various RBFs are compared in Reference [
22] and RBFs based on thin plate splines (TPS) shows good application prospects in mesh deformation, which is chosen as the mesh deformation method in this paper.
Another aspect in research of the mesh deformation algorithm is how to evaluate the quality of deformed mesh. To compare the different meshes quality after deformation, mesh quality metrics are introduced. Traditional mesh quality metrics are based on a set of Jacobian matrices, which contain information on basic element qualities such as size, orientation, shape, and skewness [
22]. We assume that the initial mesh is created with an optimal quality and the elements quality should be changed as little as possible after deformation. Both the angles and volume of the elements should be preserved. Several parametric evaluation criteria have been established in the past. The relative size metrics measuring the change in element size and skew metric measuring the skewness and distortion are often used in mesh quality evaluation. Knupp proposed other typical mesh quality metrics [
31] in which both the area and the angles of the element are measured. In commercial software like Pointwise [
32], area ratio, aspect ratio, equiarea skewness, and equi-angle skewness are often used in mesh quality evaluation. The equiarea skewness are represented as a ratio of the mesh element area to the optimum cell area. It only applies to triangles and tetrahedral elements. The equi-angle skewness is represented as the maximum ratio of the element included angle-to-angle of an equilateral element. The angle skewness applies to all element types and is available for domains and blocks. Existing mesh quality metrics are based on specific coordinates and connectivity information of the mesh, which we called mesh-based metrics. It will lead to higher computational costs. On the other side, no attention has been paid to the quality of the mesh deformation method, which can reveal the characteristics of the deformation algorithm itself. Therefore, an efficient method that can give a criterion of the ability of the mesh deformation algorithm should be introduced.
The objective of this paper is to develop a quality metric method for unstructured and structured mesh. The RBFs mesh deformation method based on generalized TPS is introduced, and is evaluated by this metric evaluating method. Different from traditional mesh-based metrics, this quality metric method is calculated based on the mapping information between the initial and deformed mesh, so it is named mapping-based metrics in this paper. The mapping-based metrics are based on the deformation principle of continuum mechanics [
33,
34], which is especially suitable for nonlinear large deformation problems. Without the requirement of element connectivity information, this method possesses high efficiency.
The rest of this paper is allocated as follows.
Section 2 introduces the methodology of RBFs’ mesh deformation method based on generalized TPS.
Section 3 demonstrates the mapping-based metrics we proposed in detail. In
Section 4, three 2-D mesh cases with structured and unstructured mesh are performed and the numerical results of the mapping-based metrics are presented. A comparison between mapping-based metrics, mesh-based metrics, and a geometric calculation directly are also been shown to illustrate the accuracy of the metrics evaluation method proposed in this paper. Finally, conclusion are described in
Section 5.
2. RBFs Method Based on Generalized TPS
Radial basis functions (RBFs) interpolation can be applied to obtain the displacement of the internal fluid mesh nodes based on the deformation of the structural boundary. The basic form of RBFs interpolation can be expressed as:
where
is the interpolation function, describing the displacement in the whole domain.
are the centers in which the values are known. In this research, the coordinates are of the
th boundary node,
is the total quantity of boundary nodes,
is the centers in which the values are unknown. In this research, the coordinates are of the random internal fluid mesh node,
is the weight coefficient of the
th RBF,
is the given RBF, and
represents the Euclidean distance of vector
and
. In 3D space, components of
and
are given as
and
, and distance
has the following formulation:
The weight coefficients
are determined by the interpolation conditions:
where
are the actual displacements of the boundary nodes. With a
matrix containing the evaluation of the basis function:
The weight coefficients vector
can be obtained:
The values of internal mesh nodes displacement
can be derived by calculating the interpolation Equation (1) after calculating the coefficients vector
:
Where
is the coordinates of the
th internal mesh node.
Several RBFs can be selected in literature, which are appropriate for this research. They can be divided in two categories: compact functions and global functions [
6,
33]. The compact function is generally scaled with a support radius
to control the compact support. The mesh nodes inside a circle (2D) or sphere (3D) with radius
around a boundary node are influenced by the corresponding displacement. The global functions are not equal to zero outside a certain radius
and operate on the whole interpolation space. Two categories RBFs are shown by Boer [
22], and the frequency used formulations are presented in
Table 1 and
Table 2.
We chose the thin plate spline (TPS) as the RBFs in this research, which generates the highest mesh quality of adapted mesh [
25].
3. Mapping-Based Metrics
The quality of deformed mesh is evaluated by the mesh quality metrics. Most of the present research studies give quality metrics that require the topology of the initial mesh or deformed mesh. It is based on a set of Jacobian matrices containing information on mesh element qualities such as size and skewness. In this section, an efficient mesh quality metrics generation method based on a mesh deformation algorithm is introduced.
3.1. Mapping Relationships
In order to describe the motion and deformation of an element, the initial configuration
is selected as the reference configuration and
as the current configuration deformed. A nonsingular single-valued mapping
is established, as shown in
Figure 1.
is set as the global coordinate system. The position of the point
in the reference configuration is
and its image point
with the deformation process in the current configuration is
.
and
are coordinates in
. The mapping of the point can be written as:
As shown in
Figure 1, the position of two adjacent points in the reference configuration is given by
and
. The distance vector between two adjacent points is
. In the current configuration, the distance vector between two adjacent points after mapping is
. The transformation between
and
is investigated as:
where
is the deformation gradient tensor.
Assuming the displacement vector of point is
, then it easily follows from Equation (8) that:
Given a displacement gradient tensor
, the formulation of distance can be expressed as:
It is assumed that are the unit base vectors of the reference configuration. Any vector in the reference can be expressed as . Images of unit base vectors in the current configuration is and the images of vector can be expressed as .
According to continuum mechanics [
34], the relationships between coordinates of two vectors in the reference coordinates are:
Therefore, the image of the vector can be written as:
Therefore, the mapping
for the vectors is:
Based on Equation (14), the coordinates of image vector according to image base vectors is the same as the coordinates of the initial vector according to initial base vectors of the reference configuration. Therefore, the mapping can be measured according to the change of unit base vectors of initial configuration over the computational domain.
Given
, the length of vector
can be formulated as:
The length of vector
can obtained by:
According to Equation (11), it easily follows that:
where
is the green strain tensor.
Equation (17) can be given in a components format by:
For a 2-D case, consider specified vectors
and their image vectors
. The area of the parallelogram formed by the two vectors is computed as follows:
And its image is given by:
The angle between two vectors is related to the dot product. Let
be the angle between
and
be the angle between
, then:
The mapping quality of the deformation algorithm is related to the variety of the unit base vectors in the reference configuration as mentioned before. Therefore, the mapping-based size metrics and skew metrics of the mesh deformation algorithm of a 2-D mesh are established to measure the area change and angle change based on the mapping relationship of the unit base vectors.
3.2. Size Metrics
Size metrics is to judge whether the deformed elements are too large or too small compared to the initial elements. According to Equation (20), denote
as the mapping-based size metric of area change, and:
where
. With deformation information, the mapping-based size metrics can be calculated with Equations (6) and (11).
means that the deformed element has the same area as the original one.
means that the deformed element is degenerate.
In a traditional method [
6,
22], mesh-based size metrics measuring area change could be expressed as:
where
is the ratio of the area of the deformed elements to the area of the initial elements. Coordinates of each deformed and initial elements and calculation of the area are required.
3.3. Skewness Metrics
Skewness metrics is to analyse distortions of the deformed elements compared to the initial elements. According to Equation (22), denote
as the mapping-based skewness metric of angle change, and:
where
are items in a green strain tensor
.
means that the deformed element has equal angles with the origin reference one.
means that the deformed element is degenerate.
In a traditional method [
6,
22], mesh-based skewness metrics, which could be also called shape metrics, could be expressed as:
The definition of variables in the expression and calculation process can be found in Reference [
30]. The evaluation requires coordinates of mesh elements and determinant calculation, which brings a computational burden.
Based on the above description, the quality metrics of the generalized TPS mesh deformation algorithm used in this paper can be evaluated. The specific steps are as follows:
Step 1. Calculate the deformation of mesh nodes over the computational domain based on the RBFs method with Equation (6).
Step 2. Calculate the displacement gradient tensor
according to Equation (11), based on a central difference method. With the adjacent three mesh nodes
, a difference item
of the central node
in a displacement gradient tensor
can be calculated as:
where
are displacement components of the mesh node
and
are coordinates of the initial mesh of node
.
Step 3. Calculate the green strain tensor
according to Equation (17).
Step 4. Calculate the mapping-based size metrics of the mesh deformation method according to Equation (23).
Step 5. Calculate the mapping-based skewness metrics of the mesh deformation method according to Equations (25) and (26).
It should be noted that, in step 2, quality evaluation of mesh nodes on a fluid boundary is not valuable because the deformation of the fluid boundary is not clear. A difference item of boundary nodes can be equal to adjacent nodes in the fluid field inside. The mapping-based metrics only require the coordinates of initial mesh nodes and displacement of the mesh nodes as a description of step 1, which can be easily obtained from a mesh deformation process. Therefore, it can be regarded as an efficient method and can reveal the characteristic of the mesh deformation algorithm itself without information of deformed mesh.
5. Conclusions
In this paper, the mapping-based metrics are established to evaluate the CFD mesh quality with a deforming algorithm. With no grid-connectivity information required, implementation of this method is relatively simple and efficient compared to traditional mesh-based metrics. RBFs’ method based on generalized TPS is tested with three test cases including structured and unstructured mesh. This method can handle large deformations of a mesh caused by translation, rotation, and airfoil transform, and it generates meshes with high quality after deformation. Mapping-based metrics are used to analyze the characteristics of this algorithm. A new point for mesh quality evaluation based on continuum mechanics is proposed.
Compared with a geometrical method, mapping-based metrics show high accuracy for an angle change and area change of a specific element. Meanwhile, when it is used to give the quality of deformed mesh, mapping-based metrics do not need the information of deform mesh. The method proposed here is efficient to calculate the quality of the mesh deformation process, especially for the dynamic cases to check if there are degradable problems. Meanwhile, the method has a high potential to be used in commercial software, which only needs the initial mesh coordinates and deformation vectors with uncomplicated calculations. The display of the results is consistent with existing software processing. Although the method in this paper is deduced by generalized TPS, the approach is consequently suitable for any other RBFs.
Finally, it should be noted that, because three 2-dimensional test cases are used, the calculation time and computing resource consumption is not different enough between a mesh-based metrics method and a new method proposed. In the next research, the mapping-based metrics method will be extended to a more complex problem in a 3-dimensional space, and the improvement of efficiency is expected.