Voting model prediction of nonlinear behavior for double-circumferential-slot air bearing system

Abstract

Double-circumferential-slot air bearing (DCSAB) systems provide multidirectional supporting forces and have high stiffness, increasing the stability of instruments at high rotational speeds. However, DCSAB systems may exhibit chaotic motion because of a nonlinear pressure distribution within the gas film, supplied gas imbalances, or an inappropriate design. This study investigated the occurrence of nonperiodic motion in a DCSAB system by analyzing the dynamic response of systems with different rotor masses and bearing numbers. The dynamic trajectory, spectral response, bifurcation, Poincaré map, and maximum Lyapunov exponent were analyzed to identify chaotic behavior. Behavior was found to be highly sensitive to rotor mass and bearing number; the system exhibits chaotic behavior when the rotor mass has values in three intervals within 0.1–6.0 kg given a fixed bearing number of Λ = 3.8. To reduce the computational cost of predicting chaotic behavior, the maximum Lyapunov exponent was predicted using various machine learning models; a voting model combining random forest with XGBoost has the highest performance. The results can be used as a guideline for designing of DCASB systems for use in industrial applications.