Optimization of Radial Electrodynamic Bearing Using Artificial Neural Network

This article focuses on the prediction of essential bearing characteristics and optimization of electrodynamic bearing (EDB). Initially, a sensitivity analysis was conducted, manipulating key design parameters to assess their impact on electric pole frequency ( $\omega $ ), stiffness (k), and damping (c). Subsequently, the data derived from the sensitivity analysis was employed as input for training an artificial neural network (ANN) model. The ANN model was developed and trained with six inputs using various algorithms and different hidden neuron configurations to forecast essential bearing characteristics. Three distinct artificial neural network models (for c, k, and $\omega $ ) were created. Notably, Bayesian Regularization with 10 hidden neurons exhibited superior performance, demonstrating the least average error. In the final stage, the ANN model was utilized to optimize the EDB through the Bonobo optimization algorithm in MATLAB. The optimization results were validated using COMSOL Multiphysics, where essential bearing characteristics were determined by fitting an analytical model to simulation outcomes. These outcomes were then compared with the ANN model predictions, affirming the applicability of ANN models in both predicting and optimizing EDB performance.