9531425 Parker This Small Business Innovation Research Phase II project pertains to neural network methods for achieving Self-sensing active magnetic bearings (AMBs). AMBs employ collars of electromagnets to levitate rotors. The absence of mechanical contact and elimination of need for lubrication systems are advantages that make AMBs a candidate technology for the replacement of rolling-element and fluid-film bearings in many rotating machinery applications. Stabilization of the rotor requires that controllers be provided the instantaneous values of rotor displacement and coil current. Self-sensing AMBs exploit the fact that the current and voltage signals in the actuation coils can provide the information needed to sense rotor position. The fundamental reason for this is that the inductive coupling between rotor and stator is displacement-dependent. Phase I demonstrated the basic technical feasibility of self-sensing AMBs, even when the bearings are saturated magnetically. Static polynomial neural networks (PNN's) play a central role in computing displacement based on nonlinear functional relationships among current and voltage signal features. Phase II efforts will focus on perfecting and generalizing the technique, and on demonstrating the performance of PNN-based rotor position estimators operating in the closed-loop feedback control of an AMB rig. The use of sensorless magnetic bearings, although potentially beneficial wherever AMBs are employed, will be especially important in high-volume, low-cost applications (e.g., small pumps, air conditioning and refrigeration compressors, energy-storing flywheels for automotive applications, etc.) where the cost of sensing instrumentation can be prohibitive. In many applications, e.g., for reasons such as reliability, high temperatures, etc., conventional bearings cannot be utilized.