Active hybrid journal bearings with lubrication control: Towards machine learning

Abstract

The study proposes an approach for simulating and designing rotating machines with water lubricated active hybrid journal bearings with the load capacity is created by combination of hydrostatic and hydrodynamic effects. Lubricant supply pressure in the bearing is adjusted in each feeding channel by a controlled servovalve. Adjustable hydrostatic forces in the bearing influence the journal position and allow reject various disturbances in the system. The ANN-based model of the rotor-bearing system was used to synthesise different types of optimal feedback controllers of the rotor position. The advanced solution of the control tasks is a machine learning based model predictive controller that has considerable robustness margins under realistic operating conditions of noisy signals and uncertainties compared with the tested PID and LQG controllers. Also an approach to reduction the energy consumption in such bearings is offered by driving the system into the optimal state where the viscous friction torque is minimal. The presence and attainability of such a state are confirmed by the results of theoretical calculations and experiment.