Multi-objective neural network optimization for PID control tuning of axial active magnetic bearing systems

Abstract

This bachelor’s thesis studies the closed-loop behavior of an axial active magnetic bearing (AMB) system by using a simplified single-input single-output model and a PID controller combined with a low-pass filter. The work focuses on the controller parameters Kp, Kd, and Tf, while the integral time constant Ti is kept fixed, and examines how they affect system stability, robustness, and transient response. The nominal closed-loop behavior is first evaluated by using the sensitivity function. Then, the effects of Kp, Kd, and Tf are analyzed through sensitivity plots, pole maps, and step responses. The results show that these controller parameters influence the closed-loop response in different ways and affect both robustness and transient performance. A Pareto-based controller parameter selection is also carried out in the (Kp, Kd, Tf) space by using the peak sensitivity Ms and the overshoot as performance indicators. Three representative controller candidates are selected from the Pareto front and compared with the nominal controller. The results show that different controller parameter choices lead to different trade-offs between robustness and transient performance. Overall, this thesis shows that a simplified axial AMB model can provide useful insight into controller tuning and closed-loop performance. The work also provides a basis for future multi-objective tuning and more detailed AMB control studies.