Figure. Long slender and compliant regions worsen the condition of the system matrix.
ProTOp has very advanced FEA solvers that can handle efficiently numerically difficult situations, typically appearing in topology optimization. In spite of that, care should be taken to improve numerical stability as much as possible in order minimize the possibility of errors and to minimize the computational effort.
Numerical stability of the FEA model should be as good as possible during the whole optimization process. To achieve this one has to:
When preparing the FEA model, one should try to get a rigid design which is supported as much as possible. Namely, long slender and compliant regions worsen the condition number of the system stiffness matrix. This causes the solver errors to increase and numerical stability decreases. This is also accompanied by an increase in computation time.
Figure. Long slender and compliant regions worsen the condition of the system matrix.
Care should be taken to run the optimization process in such a way that the appearance of disconnected orphan regions is minimized. Namely, unsupported and isolated orphan regions should actually cause the FEA analysis to fail. Numerical remedies are preventing this to happen, but numerical stability and computational time may still suffer. Modal eigenfrequency FEA is especially sensitive to this phenomena.
Figure. Appearance of disconnected orphan regions worsens numerical stability and increases the probability of failure.
NOTE. Disconnected orphan regions may be especially problematic in modal eigenfrequency analysis.