Data-Driven Sensitivity Framework for a MIMO Magnetic Levitation Actuator

A multiple-input, multiple-output planar actuator is proposed, utilizing mechanically driven stator magnet arrays to levitate a permanent magnet mover. A state of levitation and actuation is obtained by mechanically altering the orientation of the stator magnets to control the forces and torques on the mover. A challenge in the actuator is to deal with the incapability of magnetic-field amplitude control, which inherently follows from utilizing PMs in both the stator and mover. This leads to increased sensitivity and limits actuator controllability. To address this issue, a data-driven approach is introduced to mitigate peak sensitivities within the system. The problem is simplified into a two-body interaction model between a cylindrical stator magnet and a spherical mover magnet. The magnetic potential energy interaction is calculated with the equivalent surface charge model, after which a feedforward neural network is applied to approximate the magnetic potential energy to accelerate the model. Data generation involves simulating 106 randomized stator angles and observing the resulting magnetic potential energy on the mover. The mRMR feature selection method is applied for sensitivity analysis to the established dataset. Results indicate that an equilateral triangular stator topology with a rectangular mover minimizes sensitivity, which improves the stability and control potential of the actuator, while the mRMR technique is also shown to be an effective tool for selecting control parameters for a very large stator.