Feasibility-Constrained Surrogate Modeling for Parameter Optimization

The patent applies preliminary action by filtering parameter sets before they are used for optimization. A feasibility identifier evaluates parameter sets in advance to mark them as technically feasible, non-feasible, or erroneous. This preliminary filtering ensures that only valid parameter sets are used in the optimization process, preventing the generation of useless parameters and improving both efficiency and reliability.

The patent applies the copying principle by creating a simplified surrogate model that replicates the behavior of the complex computerized simulation model. The surrogate model is trained on a sample of parameter sets and can evaluate new parameter sets rapidly without requiring full FEA or CFD simulations. This copying approach maintains evaluation accuracy while dramatically reducing computation time and resource consumption.

The patent applies local quality by treating different parameter sets differently based on their validity. Instead of uniformly using all parameter sets, the feasibility identifier assigns different qualities to parameter sets: technically feasible ones are used for training the surrogate model, while non-feasible and erroneous ones are excluded. This selective approach ensures the model learns only from valid data, maintaining optimization accuracy while still utilizing sufficient training data.