Data-Driven ILC Based on RBFNN for Contour Control of Nonlinear Time-Varying Systems

To address the issues of slow convergence speed and low repetitive control accuracy in batch processing of time-varying systems, a data-driven iterative learning control algorithm is proposed. Firstly, the dynamic cross-feature fusion mechanism extracts significant trajectory features, which are fed into the Bidirectional Long Short-Term Memory (BiLSTM) network to learn optimized initial control inputs from historical tracking paths. This hybrid prediction model facilitates swift transfer of historical experience, enabling rapid target trajectory tracking. Secondly, by leveraging the strong fitting ability of RBF neural networks for nonlinear functions, a dynamic linearization process between input and output is established to perform data-driven control of the system, reducing the impact of system time-varying on repetitive control accuracy. Finally, the algorithm is verified in a multi-axis engraving machine system. Compared with existing algorithms, the control accuracy is improved by 67.52 %, demonstrating the effectiveness and superiority of the proposed algorithm.