Data-Based Settling-Time Optimization for Linear Feedback Control Systems Using Global Extremum Seeking

Abstract

High-performance control designs are indispensable in high-end industrial applications. At the same time, tuning a controller for optimal performance only on the basis of model knowledge is generally hampered by model uncertainty, unknown disturbances, and variation in the dynamics between systems of the same make due to manufacturing tolerances. Data-driven control methods facilitate system-specific controller tuning in an automated fashion while taking these aspects into account through measured performance data. This article presents a data-based extremum-seeking approach for the optimization of transient system performance in terms of settling time. A novel cascaded global optimization approach tackles the problem that the settling time depends discontinuously on controller parameters. In addition, it ensures that the resulting controller designs have guaranteed closed-loop stability and robustness margins. The effectiveness of the proposed approach in optimizing transient system behavior is shown experimentally in an industrial case study on a wire bonder system. Herein, it is also shown how to achieve improved performance uniformly over a range of setpoint designs and for position-dependent dynamics.