抄録
In most of engineering optimization problems, objective functions and constraint functions are not defined in explicit and sometime satisfying all of constraints are more difficult than improving objective function due to severe requirements to the problem. On the other hand, most of optimization algorithms assume that initial design point satisfy all of constraints (feasible design) or even if it is not so, the violations of constraints are not so large. And this consideration lack of existence of feasible design often make the optimization works in failure. So, it is necessary to assess feasible region before starting optimization in such cases. We propose non-linear discriminant analysis as such an assessment method. Discriminant analysis is one of common and widely used statistical method as one of major data-mining methods.
