Model Robustness and Sensitivity

Consider the following:
  • Results need not match exactly, which allows for variance.
    • SW (algorithm, for example) and HyperWorks variations (processors, RAM, for example).
  • Include sensitivity by checking model robustness.
    • Qualify a model as a reference model for further studies.
    • Define a sensitivity corridor.
  • Parameters to use: Random Nodal Noise.
  • Changes in physical parameters in sensitivity studies.
    • Collect sensitivity details so results remain in the sensitivity range/corridor.
  • Customer specific validation criteria (engineering criteria for validation).
    • Include the physics of the problem.

HVVH Radioss Sensitivity



Figure 1.

Check if the model can be considered a reference model for further study. Random noise/sensitivity analysis can be a base run with different seed values. Compare across Radioss versions using the sensitivity corridor.

Create and compare (m x n number of solver runs across different options):
  • Sensitivity corridor.
  • Response curves (time vs. response).
  • Statistical details.


Figure 2.


Figure 3.

HVVH OptiStruct Sensitivity



Figure 4.


Figure 5.

Verify and Validate Field Data/Proving Ground Data/Time History Data



Figure 6.