You can control the optimizer by modifying the value of the following parameters:
the scheme, if multiple objective functions are defined
You have the choice of multi-compromise or multi-priority (see Objective Functions for details).
the mode of optimization
There are two modes available for the optimization:
full optimization
For this mode, the optimizer searches for the optimum.
computation of sensitivities only
For this mode, the optimizer does not search for the optimum, but instead computes only the sensitivities for the initial configuration. This is useful before starting a full optimization, in order to identify the most influential design variables. If certain sensitivities are found to have negligible impact, the associated design variables or the optimization functions can be removed from the calculation.
the method for computing the sensitivities
Two methods exist to compute the sensitivities of the optimization functions with respect to the design variables:
analytical
For most problems, Ansys Polyflow is able to compute the sensitivities of the optimization functions with respect to the design variables analytically. It is recommended that you use this method to save CPU time.
finite difference
The optimizer can apply an infinitesimal perturbation to the design variables (that is, relative values of
) in order to compute the sensitivities by finite
difference. This method is useful if Ansys Polyflow is unable to do
it using the analytical method. Such a circumstance can arise when
UDFs are used to make the parameters vary depending on the
coordinates, or if a discretization method (that is, the routine
that computes the local matrix associated with an equation) does not
have all the derivatives; incorrect or even vanishing sensitivities
can result. Therefore, the finite difference method would be
preferable.
the maximum number of solutions
This value sets a limit to the number of solutions calculated by the main F.E.M. solvers. If the optimum has not been reached by this number of solutions, then the optimization process is stopped after the current LS method is complete.
the maximum number of iterations to reach a minimum
This value acts as a criterion for convergence, and corresponds to the maximum number of times
is calculated as part of the LS method algorithm (as
described in The Line Search (LS) Method. the number of iterations before resetting the direction
This value is used to determine when Equation 36–21 should be used rather than Equation 36–16 during the FR method algorithm (as described in The Fletcher-Reeves (FR) Method).
the precision for the design variables (default
) This precision value is used to check the convergence of the design variables during the LS method.
the precision for constraints (default
) This value contributes to the calculation of tolerance
for the ALM method algorithm (see The Augmented Lagrange Multiplier (ALM) Method for details). The precision for
constraints is a dimensionless quantity. the precision for the objective function (default
) This value contributes to the calculation of tolerance
for the ALM method algorithm (see The Augmented Lagrange Multiplier (ALM) Method for details). The precision for the
objective function is a dimensionless quantity. the initial value for the penalty coefficient
This initial value is noted as
in the description of the ALM method algorithm (see The Augmented Lagrange Multiplier (ALM) Method for details). the penalty coefficient multiplier
This multiplier is noted as
in Equation 36–10 of the ALM method algorithm (see The Augmented Lagrange Multiplier (ALM) Method
for details). the maximum value of the penalty coefficient
This value is noted as
in the discussion of the ALM method algorithm (see The Augmented Lagrange Multiplier (ALM) Method for details).