Optimization Algorithms
OptiStruct utilizes gradient-based optimization algorithms to solve the optimization problem.
- Method of Feasible Directions (MFD)
- Sequential Quadratic Programming (SQP)
- Dual Optimizer based on separable convex approximation (DUAL)
- Enhanced Dual Optimizer based on separable convex approximation (DUAL2)
- Large scale optimization algorithm (BIGOPT)
- Method of Moving Asymptotes (MMA)
MFD, SQP, DUAL, and DUAL2 are standard optimization algorithms. For further information, refer to relevant books and papers in the References section. Refer to Large Scale Optimization Algorithm (BIGOPT) for more information on BIGOPT.
During a run, the corresponding optimization algorithms are automatically selected by OptiStruct based on the optimization type. The DOPTPRM, OPTMETH parameter can be used to change the default optimization algorithms. The use of this parameter will override the defaults.
DUAL2 is an enhanced version of the proprietary CONLIN dual optimizer. Solution stability has been improved, reducing probability of failure in converging to a solution. The default optimizer for problems with a large number of design variables such as topology, free-size and topography is DUAL2. You can switch to DUAL or MMA in case you run into issues with the default DUAL2 method fails to converge for Topology, Free-Size, or Topography optimization.
MFD is the default optimizer for problems with a large number of constraints (but smaller number of design variables) such as size/gauge and shape.
MFD, SQP (primal methods) and BIGOPT are more suitable for size/gauge and shape since the approximate problem typically involves coupled terms, due to the advanced approximation formulation utilizing intermediate variables and responses.
If Equality Constraints are activated in size and shape optimization, then SQP is the default optimizer.
DOPTPRM,OPTIMOMP,YES can be used to turn on parallel computation for MFD, SQP, and DUAL2 algorithms.