Data-driven system identification and optimal control of an active rotor-bearing system

Both diagnostics and control in any piece of rotating machinery is getting more and more attention as computational powers allow storage of big data sets, and we are able to perform approximation, feature extraction and classification of big data. Such transfer from traditional diagnostics and control to machine learning and decision making also makes new ways of system identification based on artificial neural network architectures more attractive as they are more flexible in representing non-linear dynamics. Data-driven linear system identification or non-linear system identification based on artificial neural networks when coupled with optimal control obtained using either traditional or machine learning approaches result in a powerful tool for rotor-bearing system modeling and control that allows using identified model for predictive analysis as well. In the present paper such approach is presented for a rotor system on two journal fluid-film bearings, one of the bearings is equipped with a set of two actuators for feeding pressure control that allows adjustments of rotor position in the bearing. Such adjustments are aimed at decreasing power loss due to friction, so the system is more energy efficient. The presented approach of system identification and modeling can be applied to any similar rotor system.