Fourier neural operator for flow-induced rotordynamics force prediction and application to a SCO2 magnetic bearing-rotor system

This study presents a novel approach for rotordynamic fluid–structure interaction (FSI) models via the use of a Fourier Neural Operator (FNO) in high-speed rotors supported by canned magnetic bearings (MB) immersed in supercritical carbon dioxide (SCO2). Calculating the nonlinear fluid forces in the canned MB gaps is computationally expensive due to iterative SCO₂ property evaluation and heat transfer coupling. The proposed methodology to address this issue includes the following key contributions: (1) The FNO surrogate model achieves a four-order reduction in computation time with a mean squared error of 0.01. (2) An efficient method is introduced for generating input–output image data using a 3D Reynolds-based SCO2 film model. (3) The feasibility of computing full rotordynamic and control systems, including nonlinear FSI forces, is demonstrated. (4) The models are validated against literature and are useful to predict rotordynamic instability speed in SCO2 turbomachinery.