Active Magnetic Bearing PID Tuning With Model-Free Analysis Based Modified Hebb Learning Rule

Active Magnetic Bearings (AMBs) are critical for high-speed rotating machinery, yet their PID controller tuning remains heavily reliant on manual calibration and precise system models, limiting efficiency and adaptability. This paper proposes a novel model-free PID tuning framework that integrates a modified Hebb learning rule with adaptive disturbance rejection to overcome these limitations. The methodology employs a two-phase approach: During static levitation, heuristic offline calibration guided by directional analysis of displacement-current correlations optimizes initial P and D parameters to ensure baseline stability. In rotational operation, an adaptive notch filter (ANF) isolates synchronous disturbances from sensor signals, enabling online Hebb learning adaptation of PID weights to suppress vibration. Experimental validation on a 9.05 kg rotor system demonstrates the framework's efficacy. Key innovations include: (1) A Hebb learning rule with heuristic initialization for PD weight updates; (2) ANF-based synchronous disturbance extraction to decouple learning from broadband noise; (3) A hybrid offline-online tuning strategy eliminating model dependency. Results show up to 20% reduction in displacement vibration amplitude at 2000 rpm under steady state compared to conventional PID, confirming the method's superiority in stable regimes, while acceleration transients reveal ANF phase-delay limitations. This work establishes a practical, model-free pathway for autonomous AMB control, merging dynamic adaptability with operational safety.