Ultra-precision control of dual-axis scanning mechanism based on PSO-RBF neural network

Broadband ultra-high resolution spectrometers require arc-second level grating positioning accuracy to achieve picometer to femtometer spectral resolution. Traditional control approaches exhibit significant limitations when confronting system nonlinearities, parameter uncertainties, and external disturbances. This work presents an adaptive PID control methodology integrating Particle Swarm Optimization (PSO) with Radial Basis Function (RBF) neural networks. The approach establishes a pitch axis dynamic model incorporating friction, gravitational effects, and transmission backlash, then implements an RBF network architecture enabling real-time parameter adaptation. An enhanced PSO algorithm performs global network parameter optimization, creating a dual-stage control framework combining "offline global optimization with online local adaptation." Simulation results demonstrate that the proposed method achieves 0.7-1.0 arc-second angular positioning accuracy-representing substantial improvements in step response characteristics compared to conventional PID and RBF-PID controllers. Under ± 30% system parameter variations, the controller maintains excellent positioning performance with markedly reduced disturbance recovery times. External disturbance testing reveals a disturbance rejection ratio of -25.31 dB, with substantially diminished steady-state accuracy degradation. These findings validate the control strategy's effectiveness for grating precision positioning applications, offering a novel solution for high-precision spectral measurement technology while providing valuable insights for other precision mechanical systems requiring arc-second level positioning accuracy.