Flexible Rotor Dynamic Balancing Method Based on a Multi-Layer LSTM Network

Transient dynamic balancing of flexible rotors is of significant practical value in engineering. Traditional dynamic balancing requires repeated trial weights and start–stop cycles, resulting in inefficient balancing. This paper proposed a multi-layer LSTM (Long Short-Term Memory) network-based method for transient dynamic balancing of flexible rotors. Firstly, a dynamic model of the rotor bearing system is established using the finite element method. Secondly, leveraging the relationship between transient unbalance force and quantity, a formula using the force zero-crossing to identify unbalance magnitude and phase is derived, and a multi-layer LSTM architecture is established to predict the unbalance force from the rotor dynamic response. Finally, a complete transient dynamic balancing solution based on the multi-layer LSTM network is developed. Simulation verification is conducted on a three-disc rotor, and the identification errors of unbalance phase and eccentricity are both less than 5%. After balancing, the maximum amplitude of the rotor decreases by about 93%. Additional verification is carried out on the rotor test rig. The experimental results show that the multi-layer LSTM flexible rotor transient dynamic balancing method can significantly reduce unbalance vibration: the transient amplitude of the rotor decreases by 70.22%, and the steady-state amplitude decreases by 69.3% after balancing.