A new parallel multi-harmonic neural network adaptive filtering hybrid control algorithm for helicopter active vibration control under variable working conditions

Helicopter vibration primarily originates from high-order harmonic vibrations induced by rotor loads. Complex external operating conditions and rotor variable-speed technologies necessitate active vibration control systems capable of adapting to frequency-varying vibrations. Conventional Filtered-x Least Mean Square (FxLMS) algorithms employ constant single convergence factors, making it difficult to adapt to variable working conditions. This paper proposes a new parallel multi-harmonic neural network adaptive filtering hybrid control algorithm to suppress multi-harmonic vibrations under variable working conditions. Bandpass filters are adopted to separate multiple harmonics for parallel control. Leveraging neural networks' strong approximation and memory capabilities, the proposed algorithm conducts reinforcement learning-guided supervised training across multiple environments, significantly enhancing training efficiency and the network's adaptability to frequency changes. In addition, state normalization and denormalization methods are adopted to eliminate the influence of vibration amplitude variation on the network control performance. Concurrently, adaptive filtering ensures the steady-state performance of the algorithm. An active vibration control system for helicopters equipped with electromagnetic inertial actuators has been established. The simulation and experimental results show that compared with the conventional FxLMS algorithm, the proposed control algorithm has a better convergence speed and similar multi-harmonic control effect. Meanwhile, the results of various variable working condition experiments also verify that the proposed algorithm has better adaptability and robustness, thereby effectively adapting to the active vibration control of helicopters with variable rotor speeds.