AutoDesign provides 4 meta-models such as Simultaneous Kriging, Radial Basis Function with Gaussian, Radial Basis Function with Quadratics, and Conservative Response Surface.
Guide 1: As Kriging and Radial Basis Function (RBF) method fundamentally include regression model, Conservative RSM is not recommended. It is only academic model for students to study RSM.
Guide 2: In RBF methods, multi-quadrics gives more stable result than Gaussian. Thus, RBF with multi-quadrics is recommended as the default method.
Guide 3: Kriging method is more sensitive to the sampling quality than RBF. So, it is not good with ISCD-1 & ISCD-2.
Guide 4: If RBF method fails to converge to the solution, then switch the meta-model with kriging and re-optimize it with ‘Get from simulation history’. Before re-optimization, the feasibility should be checked. If all analysis results of SAOs and the initial sampling results do not satisfy the constraints, then, the problem may be non-feasible or very difficult problem. For this case, relax the limitation of constraints or convergence tolerances and try to re-optimize.
Figure 1 Guideline for the selection of Meta-Model method