Finding the Average Rotation
Geometrically Linear Case
If an arbitrary reference point ‘p’ has a certain translation up and rotation θp, then the average displacement of any other point can, under the assumption of rigid body motion, be written as
The rigid body motion parameters up and θp can be determined using a least squares fit. The quadratic error is
(3-248)
It is possible to solve the least squares problem analytically by minimizing R2. In that process, the first of the three integrals above is unimportant, since it does not involve the unknowns. The second integral can be simplified to
Similarly, the third integral can be expanded to
To minimize this function, it should be differentiated with respect to all unknowns, and each equation set equal to zero. The first equation is obtained from
The next two equations can be obtained by using index permutation. The fourth equation reads
Finally, the fifth and sixth equations are obtained by using index permutations. This gives the following system of equations on matrix form:
This system of equations can now be directly solved.
Geometrically Nonlinear Case
In a geometrically nonlinear setting, the cross product must be replaced by a general rotation matrix, R. The rotation matrix is now a nonlinear function of the rotation vector, θp. However, another parameterization of the rotation using a quaternion representation is chosen in this case. For more information about the quaternion representation, see the theory for Rigid Connector. The average displacement field can now be written as
The rotation matrix is a function of the four quaternion degrees of freedom, symbolically collected in the vector qp.
The quadratic error is, as before
(3-249)
Using a weak form of the least squares problem,
This constitutes a set of nonlinear equations for determining the average rigid body motion in terms of up and qp. The Jacobian of this system will, in general, cause a full coupling to the displacement field u and it is also nonsymmetric. Since the structural mechanics problem in itself often has a symmetric stiffness matrix, adding a few unsymmetric contributions to the problem can affect the solution time and memory consumption significantly.
If, however, the displacement field is considered as known, then a small, symmetric system of equations for up and qp can be solved separately. From an implementation point of view, the nojac() operator is then applied to u in the weak formulation. This approach is a much more efficient, but requires a special solver strategy where the displacement field is solved first.
Reference Plane
When the warping displacement is evaluated with respect to a reference plane, the quadratic error R2 in Equation 3-248 is evaluated only at the selected point:
(3-250)
Symmetry Type
When a symmetry type is set as either symmetry or antisymmetry, the reference point ‘p’ is projected onto the symmetry plane. Then, the translation up and rotation θp have the following constraints for symmetry:
and the following constraints for antisymmetry: