Neural-Network-Based Active Vibration Control of Rotary Machines

This paper presents a novel approach for modeling and mitigating vibrations in rotary machines during transient operations, such as start-up and shutdown. We propose using two time-series artificial neural networks (ANNs)—Long Short-Term Memory (LSTM) and Time-Delay Neural Network (TDNN)—to model lateral vibrations. We utilized the normalized singular values of the Hankel matrix of the system to derive a reduced-order model, which was then used to generate training data for the neural networks. These networks were trained with experimental data collected from a laboratory test rig under various asymmetric loading conditions. The trained LSTM and TDNN networks were validated with real data in the presence of measurement noise. Subsequently, we employed the TDNN to develop an active vibration control algorithm based on the nonsingular terminal sliding mode control (NTSMC) technique. Finally, we evaluated the stability, robustness, and effectiveness of the proposed closed-loop controller using the laboratory test rig.