الفهرس | Only 14 pages are availabe for public view |
Abstract Over the past decade, deep reinforcement learning has greatly impacted the field of continuous control, from mastering simple games to controlling multiple actuators in a robot doing complex tasks. Deep reinforcement learning agents are capable of finding optimal control policies without a model of the underlying system. On the other hand, in the field of rotordynamics, vibration control has been previously achieved almost exclusively using classical control theories that rely on modeling the system and diagnosing the cause of vibration. Vibration control of rotating machinery is crucial to prevent failures and allow machines to operate in dynamic vibration conditions or near their critical speeds. We propose using model-free deep reinforcement learning to control multiple sources of vibrations in a system supported by Smart Electro-Magnetic Actuator Journal Integrated Bearings (SEMAJIB). SEMAJIB is a smart bearing that integrates a journal bearing for load carrying and an electromagnetic actuator for control purposes. Journal bearings are excellent load carriers; however, they introduce some instabilities known as oil whirl and oil whip due to the movement of oil. In this work, we demonstrate the ability of the proposed deep reinforcement learning controller in finding successful control policies for stabilizing the system and reducing the synchronous vibration caused by the rotor’s unbalance. Our proposed controller is evaluated on a simulated and physical test rig with both unbalance and oil whip vibration. The proposed controller is able to balance the system with unbalance vibration reduction of up to 93%. The controller is able to completely eliminate oil whip vibration with a vibration reduction of up to 99%. In a system with both vibrations, the proposed controller reduced the total vibration by 85%. |