Multiscale-Kolmogorov–Arnold Network Surrogate Modeling Approach for Dynamic Stiffness Prediction of a Heavy-Duty Gas Turbine

Abstract

The flexibility of the casing plays an important role in the dynamic response of the gas turbine rotor. The effect of the support parameters on the dynamic stiffness of the supporting components must be considered. While finite element method analyses remain valuable, their time-intensive nature, particularly in model definition, necessitates the search for more time-efficient methods. Motivated by the need for fast, manageable solutions with repeatable configurations within defined design parameters, this paper introduces a novel multiscale Kolmogorov–Arnold network (MKAN) model to predict the dynamic stiffness of gas turbine casings. Unlike conventional methods, the MKAN model is an innovative, simplified alternative for predicting dynamic stiffness. The effectiveness of the MKAN model in predicting dynamic stiffness is validated using test set data, compared to common models. Additionally, the casing’s support stiffness and damping ratio parameters were identified using the trained MKAN combined with particle swarm optimization.