Many high fidelity analysis tools including finite-element analysis and computational fluid dynamics have become an integral part of the design process. However, these tools were developed for detailed design and are inadequate for conceptual design due to complexity and turnaround time. With the development of more complex technologies and systems, decisions made earlier in the design process have become crucial to product success. Therefore, one possible alternative to high fidelity analysis tools for conceptual design is metamodeling. Metamodels generated upon high fidelity analysis datasets from previous design iterations show large potential to represent the overall trends of the dataset. To determine which metamodeling techniques were best suited to handle high fidelity datasets for conceptual design, an implementation scheme for incorporating Polynomial Response Surface (PRS) methods, Kriging Approximations, and Radial Basis Function Neural Networks (RBFNN) was developed. This paper presents the development of a conceptual design metamodeling strategy. Initially high fidelity legacy datasets were generated from FEA simulations. Metamodels were then built upon the legacy datasets. Finally, metamodel performance was evaluated based upon several dataset conditions including various sample sizes, dataset linearity, interpolation within a domain, and extrapolation outside a domain.

This content is only available via PDF.
You do not currently have access to this content.