21.3.4. Guides for Convergence Tolerance

AutoDesign is a Sequential Approximate Optimizer, which uses meta-models sequentially. Figure 21.75 shows the optimization process of AutoDesign. First, it requires the initial DOE table including design table and analysis result table. Then, it constructs meta-models. The general numerical optimization algorithms solve the approximate optimization problem composed of the meta-models. The new analysis is performed for the approximate optimum. Finally, it checks the convergence criteria by using exact analysis results. If those criteria are not satisfied, the process is repeated by including the new design information to the DOE table.

Figure 21.75 Sequential Approximate Optimization Process in AutoDesign

• 21.3.4.1. Convergence Criteria
• 21.3.4.2. Premature Convergence in AutoDesign
• 21.3.4.3. Convergence Relaxation Control