Developers of FEA software have proposed several methods for estimating error and improving the accuracy of results by using various adaptive processes for refining meshes.

The most widely used methods are the h and padaptive refinement. Each has its tradeoffs. But combining the two into an advanced form improves accuracy and processing time while eliminating many of the tradeoffs.
"In an ideal world, the accuracy of FEA results is determined by the user knowing the exact solution for a particular mode," says Tomi Mossessian, CEO, PlassoTech Inc., Encino, Calif. "But the solution is not available in most cases, so a process has to be established for reaching a specified accuracy. The automated procedure for improving results that reach the required level of accuracy is called an adaptive process," he says.
The most common way to check the effectiveness of an adaptive approach is the rate of convergence. "It shows how fast the solver is approaching an exact solution as the user increases the discretization degrees of freedom," says Mossessian. The compute efficiency of the method is also an important factor because users want to expend the least amount of time and memory to get the most accurate results.
Mossessian has experimented with several methods to show the pros and cons of each. For example, the uniform mesh refinement method is one of the simplest for improving FEA results. "It creates a finer mesh by uniformly decreasing the element size throughout the model. This increases the number of elements until the results stop substantially changing. That is, the solution has converged within a certain percentage," he says.
Although the simplest method, it is also the least efficient because a uniformly finer mesh of a 3D model substantially increases the number of elements to process. "For example, a 1m cube containing cubic elements of 10 cm in lengths would have 1,000 elements, by reducing the element size to 5 cm, the number of elements increases to 8,000. The growth is cubic," says Mossessian. Even in thin solids the growth is quadratic. "The uniform mesh refinement method is too slow and resource intensive to be practical in most reallife design situations," he says.
An hadaptive method improves on uniform mesh refinement. Hadaptive refines the mesh only in areas containing a high number of errors. The increase in the number of elements that must be processed is substantially less than when uniformly refining the mesh. Technical papers often refer to these methods as hadaptive.
The technique typically uses an error estimate over the elements at a given solution step and refines the mesh by a practical factor only in the higherror regions and then resolves the problem. The adaptive mesh refinement and resolving continues until reaching the specified accuracy. For structural analysis, typically this would be highest stress in the model. Although the hadaptive method is more efficient than uniform refinement, for many cases it does not have an optimum convergence rate and requires complete or partial remeshing of the model in addition to resolving at each cycle, which is a timeconsuming process," says Mossessian.
The padaptive method does not change the mesh or number of elements. Instead, it increases the order of the polynomial approximation used within each element. This eliminates the need for remeshing and only requires resolving the problem at each pcycle until the required accuracy is reached.
"Similar to the hadaptive method, error estimates are used with different points of the mesh. Polynomial order is increased within regions of high error until results reach the userdefined accuracy. In terms of convergence rate, this approach is superior to the traditional hadaptive method. However, when sharp stress concentrations are present, this advantage deteriorates. In addition, by substantially increasing the porder, it becomes compute expensive," he says.
The best method for efficiently reaching convergence is a combination of both methods, the hpadaptive method. It provides the mostefficient technique for controlling the FEA approximation errors. Rather than increasing the p order indefinitely or just relying on pure mesh refinement through the hadaptive method, the hpmethod uses the padaptive within each hrefinement step. This is the most difficult method to implement and is offered by few FEA packages.
The implementation of FEA solvers based on the hpadaptive method in 3G.author, an FEA designanalysis program, allows calculating accurate results in the mostefficient manner. This allows a versatile definition of convergence criteria for both global and local parameters and provides a number of advanced attributes with respect to accuracy, efficiency, and robustness.
A key attribute is an efficient pbased solver using a state of the PCCG ( preconditioned conjugate gradient) technique tied to adaptive mesh refinement. In addition, accuracy can be controlled so it is easily tailored for global or local result parameters, including stresses, displacement and temperature. Localized results can include or exclude any geometric entity, such as faces, edges, and vertices.
In addition, advanced functions allow controlling accuracy and efficiency when solving assemblytype models.
For users needing complete control over meshing, Mossessian says, it is available on top of the hpadaptive method. "Users more familiar with their model behavior can accelerate the adaptive process by choosing a finer initial mesh over any area of interest, such as parts, faces, edges and vertices, using 3G.author's global and local mesh control capabilities.
MAKE CONTACT:
PlassoTech Inc.,
www.plassotech.com