Abstract
Nonlinear factors such as inertia and friction in the feed system of CNC machine tools, along with non-minimum phase characteristics, can induce significant tracking errors and directly degrade the contour accuracy of the machined workpiece. Conventional inverse-model-based feedforward control relies on the explicit computation of transfer function poles and zeros, which suffers from poor numerical stability in high-order systems and poses difficulties for engineering implementation. To address the above issues, this paper proposes a preview-based partial inverse compensation feedforward control method. This method is based on the hybrid response model architecture, which decouples the system dynamics into a basic response model characterizing the dominant linear characteristics and a friction response model characterizing the nonlinear friction characteristics. An approximate inverse model is constructed only for the basic response model, and the SRCZ method is proposed to eliminate the unstable poles of the inverse model without explicitly computing poles and zeros, thereby providing a stable initial solution for subsequent iterative optimization of the inverse model. Furthermore, by leveraging the finite-preview trajectory information available in the CNC system, an iterative learning concept is introduced to iteratively optimize the parameters of the stable inverse model obtained using the SRCZ method, significantly improving the accuracy of the inverse model. Meanwhile, the predictions from the friction response model are incorporated into the compensation target to counteract the effects of nonlinear dynamics. The results of tracking and machining experiments conducted on two CNC machine tools demonstrate that, compared with the existing feedforward control methods, the maximum absolute tracking error is reduced by 26% to 65%. The actual contour error of the workpiece is reduced from 5