Construction of device performance models using adaptive interpolation and sensitivities.

摘要

The performance of a device, for certain design parameters, can be modeled using finite element analysis (FEA); however, this can be computationally complex and time consuming. But because this performance can be modeled as a continuous function over the parameter ranges, it is efficient to combine FEA with interpolation to rapidly estimate the performance for any variables. While there exists many interpolation techniques, FEA can provide sensitivities at little extra cost, so these methods should take advantage of this information. In this thesis, a new adaptive interpolation scheme is proposed which uses radial basis functions (multiquadrics) while making use of sensitivities. Results demonstrate the greater accuracy of the new scheme compared to a previous multiquadric algorithm without sensitivities. Test cases include artificial functions, and an example combining FEA and the interpolation of the reflection coefficient for a rectangular waveguide containing a partial height metallic post.