A journal bearing performance prediction method utilizing a machine learning technique

Abstract

A predictive analytics methodology is presented, utilizing machine learning algorithms to identify the performance state of marine journal bearings in terms of maximum pressure, minimum film thickness, Sommerfeld number, load and shaft speed. A dataset of different bearing operation states has been generated by solving numerically the Reynolds equation in the hydrodynamic lubrication regime, for steady-state loading conditions and assuming isothermal and isoviscous lubricant flow. The shaft has been modelled with four different values of misalignment angle, lying within the acceptable operating range, as defined in the existing regulatory framework. The journal bearing was modelled parametrically using generic geometric parameters of a marine stern tube bearing. The lift-off speed was estimated for each loading scenario to ensure operation in the hydrodynamic lubrication regime and the effect of shaft misalignment on lift-off speed has been evaluated. The generated dataset was utilised for training, testing and validation of several machine learning algorithms, as well as feature selection analysis, in order to solve several classification problems and identify the various bearing operational states.