Decoupling control for tri-axial standard vibrator based on optimized neural network inverse model

Abstract

To ensure the calibration accuracy of tri-axial vibration sensors, the tri-axial vibrator needs to output low-distortion and low-coupling vibration signals. However, due to the complexity of the motion decoupling structures, significant residual coupling still remains during vibration generation, which impacts calibration accuracy. For this purpose, this paper proposes a decoupling control method based on an optimized neural network inverse model to suppress inter-axis motion coupling and improve the precision of output vibration signals. First, the training data for neural network model identification was optimized using a uniform design strategy, overcoming issues such as low computational efficiency and overfitting in conventional sampling strategy. The identified inverse model of the leaf-spring-type tri-axial standard vibrator was then integrated before the original system to achieve precise decoupling of motion along each axis. Finally, simulation and experimental results show that, compared to fuzzy PID and conventional neural network inverse model control, the optimized inverse model control more effectively suppresses inter-axis motion coupling in the leaf-spring-type tri-axial standard vibrator.