Our powerful response surface methods include full second-order polynomial, kriging, non-parametric regression and neural network approaches. These serve to interpolate between the data points in multidimensional space. They can be visualized as a 2-D or 3-D description of the relationships between design variables and design performance.
ANSYS DesignXplorer can use the response surfaces as a reduced-order model. For example, while looking at optimization trade-offs, the algorithm can search the response surface to rapidly solve many thousands of samples. You can also probe the response surface or add design points at will.
Our adaptive response surface methods, such as Kriging with Auto Refinement or Sparse Grid, will actually refine until sufficient accuracy is achieved. These methods feature a convergence plot.