Systems and methods of using multiple surrogate-based parameter selection of anisotropic kernels in engineering design optimization

摘要

Methods of conducting design optimization of a product using multiple metamodels are described. First and the second metamodels are configured with common kernel function. Kernel width parameter is the output or result of the first metamodel while the second metamodel requires a set of substantially similar kernel width parameters defined a priori. Further, the second metamodel is configured with an anisotropic kernel. First and second metamodels are trained in two stages. In the first stage, kernel width parameters are obtained by fitting known responses (obtained in numerical simulations) into the first metamodel with one or more prediction trends. Additional kernel width parameter set is derived by algebraically combining the obtained kernel width parameters. The second metamodel is then trained by cross-validating with known responses using N trial sets of metamodel parameter values including the kernel width parameter values determined in the first stage along with various combinations of other parameters.

主权项

1. A method used in conducting an engineering design optimization of a product using multiple metamodels, said method comprising:
receiving a set of design variables, objectives and constraints for designing and optimizing a product in a computer system having an engineering design optimization application module installed thereon; choosing a plurality of design of experiments (DOE) samples in a design space defined by the design variables, said each DOE sample corresponding to a unique combination of the design variables; obtaining respective numerically-simulated responses of the DOE samples by conducting a computer aided engineering analysis of each of the DOE samples; selecting first and second metamodels for approximating responses in the design space, the first metamodel and the second metamodel having a common correlation function with a set of kernel width parameters being determined in the first metamodel as a result and defined a priori in the second metamodel; obtaining a plurality sets of kernel width parameter values in a first stage of a metamodel training procedure by fitting the obtained numerically-simulated responses into the first metamodel with one or more prediction trends; deriving additional set of kernel width parameter values by algebraically combining said plurality sets of kernel width parameter values; determining a minimum error set of metamodel parameter values to be used in a trained second metamodel in a second stage of the metamodel training procedure from N trial sets of metemodel parameter values that includes said plurality sets and said additional set of kernel width parameter values obtained in the first stage along with various combinations of other parameters of the second metamodel, the minimum error set being the set having minimum error in a cross-validation procedure using the obtained numerically-simulated responses, wherein N is a positive integer; and obtaining one or more optimized designs of the product based on the set of design variables, objectives and constraints using approximated responses from the trained second metamodel.