A physics-based and data-driven framework providing a guideline for state estimation in active magnetic bearing-supported rotors

This study proposes a low-cost methodology for response estimation in rotor systems supported by active magnetic bearings (AMBs). The methodology forms a framework that integrates multidisciplinary physics-based and data-driven modules, enabling reliable and computationally efficient estimations in various operating conditions (OCs). The developments involve a time-domain simulation based on full rotordynamic finite element (FE) modeling, and rigid body-, Kalman filter (KF)-, and machine learning (ML)-based estimations, to make use of their capabilities in suitable OCs and to highlight their weaknesses. Model and data reductions and inclusion of physical information are incorporated into the framework. As a case study, the methodology is implemented for the estimation of unmeasured translational displacements at the actuator locations. Time- and frequency-domain validations with noisy signals prove that the rigid body and KF estimations are unable to track the system dynamics and diverge from the reference simulation in the vicinity of the critical speeds, where ML provides considerably more precise estimates. The effectiveness of multiple approaches in the framework is concluded with an estimation guideline. Modular developments in the framework are proposed for future studies.