An Adjoint-Based Data-Driven Inverse-Free Iterative Feedforward Control Method

Feedforward control is widely used in control systems since it can achieve high tracking performance by effectively compensating for known disturbances before they affect the system. For traditional feedforward control methods, the performance improvement highly depends on the model quality of the system model and the accuracy of the model-inversion. However, on one hand, in practice, the modeling error is inevitable, especially for precision motion systems with complex dynamics, on the other hand, the nonminimum phase (NMP) systems often lead to a problem that plant inversion is unstable and non- causal. Therefore, this paper proposes an approach that combines the benefits of data-driven feedforward control and the gradient descent method. This integrated approach aims to address challenges related to laborious model identification and unstable plant inversion simultaneously. The main idea is to replace the model with dedicated experiments on the system and to avoid calculating plant inversion by applying the gradient descent method to the learning process. The simulation and experimental results show that the algorithm can achieve the optimal point-to-point tracking performance without relying on model-inversion.