Intelligent robust controllers for tribotronic conical fluid film bearings

Abstract

The article presents the results of the development of means for intelligent robust control of the parameters of a tribotronic rotor-support system with a tapered bearing with a removable bush. The proposed controller is implemented on the basis of deep Q-network reinforcement learning (DQN) synthesized on the basis of a numerical model of a rotor support system. The control strategy involved simultaneous control of the shaft position and friction in the lubrication layer. Methods for control synthesis are presented for both a deterministic system and a system with stochastic parameters. In the latter case, a controller synthesis technique is proposed that takes into account uncertainties in the system at the training stage. Testing of the resulting controllers shows the better ability of a controller trained to take into account uncertainties to cope with variable loads, as well as predict possible changes in the system and proactively transfer the system to more advantageous states.