IMAC 2022 - A Gaussian Process Regression Reduced Order Model of Geometrically Nonlinear Structures
This paper presents a new reduced order modeling methodology for geometrically nonlinear structures based on Gaussian Process Regression (GPR). Specifically, the reduced order model (ROM) is in the form of polynomials in the modal coordinates of the dominant modes of the structure. Only bending modes are included; any membrane motions are captured implicitly. The proposed approach generates a GPR model for each ROM coefficient, which is able to capture how each ROM parameter changes due to variations in FEM (e.g. design parameters or uncertainties). The GPR-ROM approach also provides an estimate of the variability of each ROM coefficient, which provides confidence bounds for the ROM (due to inaccuracies in the training data or model form uncertainty) and also provides a means to filter out any ROM coefficients that are uncertain with respect to the load scaling factor. This has the potential to address a major limitation of the implicit condensation and expansion (ICE) method [Hollkamp & Gordon, JSV, 2008].
by K. Park and M.S. Allen, IMAC-XL, Orlando, Florida, 2022