Data-Driven Quadratic Optimization for Tracking Problems with Application to a Tray Indexing System

Abstract

Iterative Feedback Tuning (IFT) algorithm explores the correlation between the cost function and the parameters of the controller. It introduces one data-driven approach to derive the gradient of the cost function with respect to the controller parameters and provides an iterative numerical method for the controller parameters optimization. However, classical IFT algorithms require three experiments to be conducted at one iteration, and one special experiment is included in the three experiments. That means the reference signal is changed during the iteration because of the special experiment, which restricts the application of the IFT algorithm in many areas. Besides, in many process control applications with long duration, IFT algorithm is much more time-consuming than many other tuning algorithms. In this paper, a novel IFT algorithm is proposed. For each iteration, no special experiment is required, which means parameters of the controller can be tuned when the system is running normally. Simulation results on a tray indexing system demonstrate the high performance and applicability of this algorithm.