Sunday, February 12, 2017

Adaptive Convergence

You can control the relative accuracy of a solution in two ways.

  • You can use the meshing tools to refine the mesh before solving, or 
  • you can use convergence tools as part of the solution process to refine solution results on a particular area of the model.
Through its convergence capabilities, the application can fully automate the solution process, internally controlling the level of accuracy for selected results. You can seek approximate results or adapted/converged results.

You can control convergence to a predefined level of error for selected results. In the calculation of stresses, displacements, mode shapes, temperatures, and heat fluxes, the application employs an adaptive solver engine to identify and refine the model in areas that benefit from adaptive refinement. The criteria for convergence is a prescribed percent change in results. The default is 20%.

You can continue to refine the mesh based on a specific solution result. When you pick a result (Equivalent Stress, Deformation, Total Flux Density, etc.), indicate that you want to converge on this solution. You pick a value and the solution is refined such that the solution value does not change by more than that value.

To add convergence, click the result you added to your solution; for example, Equivalent Stress, Total Deformation, or Total Flux Density. If you want to converge on deformation, right-click on Total Deformation and select Insert> Convergence. In the Details View, you can specify convergence on either the Minimum or Maximum value. Additionally, you can specify the Allowable Change between convergence iterations.

General Notes
  • Adaptive convergence is not supported for orthotropic materials.
  • Adaptive convergence is not supported for solid shell elements (the SOLSH190 series elements).
  • Adaptive convergence is not valid for linked environments where the result of one analysis is used as input to another analysis.
  • Low levels of accuracy are acceptable for demonstrations, training, and test runs. Allow for a significant level of uncertainty in interpreting answers. Very low accuracy is never recommended for use in the final validation of any critical design.
  • Moderate levels of accuracy are acceptable for many noncritical design applications. Moderate levels of accuracy should not be used in a final validation of any critical part.
  • High levels of accuracy are appropriate for solutions contributing to critical design decisions.
  • When convergence is not sought, studies of problems with known answers yield the following behaviors and approximated errors:
  • At maximum accuracy, less than 20% error for peak stresses and strains, and minimum margins and factors of safety.
  • At maximum accuracy, between 5% and 10% error for average (nominal) stresses and elastic strains, and average heat flows.
  • At maximum accuracy, between 1% and 5% error for average stress-related displacements and average calculated temperatures.
  • At maximum accuracy, 5% or less error for mode frequencies for a wide range of parts.
  • When seeking highly accurate, Converged Results, more computer time and resources will be required than Manual control, except in some cases where the manual preference approaches highest accuracy.
  • Given the flexible nature of the solver engine, it is impossible to explicitly quantify the effect of a particular accuracy selection on the calculation of results for an arbitrary problem. Accuracy is related only to the representation of geometry. Increasing the accuracy preference will not make the material definition or environmental conditions more accurate. However, specified converged results are nearly as accurate as the percentage criteria.
  • Critical components should always be analyzed by an experienced engineer or analyst prior to final acceptance.
  • For magnetostatic analyses, Directional Force results allow seeking convergence based on Force Summation or Torque as opposed to other results converging on Maximum or Minimum values.
  • Adaptive convergence is not valid if a Periodic Region or Cyclic Region symmetry object exists in the model.
  • Adaptive convergence is not valid if an imported load object exists in the environment.