A multi-scale vibration feature fusion method for condition identification of air bearings

Abstract

Air bearings provide non-contact support through a pressurized gas film and are used in high-speed, high-precision, and clean operating conditions. Their condition is sensitive to supply pressure, film clearance, restrictor behavior, rotor eccentricity, and external disturbances. Since real air-bearing fault samples are difficult to obtain in batches, this paper proposes a mechanism-sensitive multi-scale stability-weighted feature fusion method, termed MS-SWFF. Time-, frequency-, and time-frequency-domain features are extracted from bearing-seat vibration signals and grouped according to air-bearing response mechanisms. Feature weights are calculated from class discriminability, cross-condition stability, and redundancy, and the weighted vector is classified by an SVM. A rotor-bearing fault test rig is used for surrogate validation. MS-SWFF achieves 82.8% accuracy in the same-condition task and 71.2% average accuracy in cross-speed tasks. The moderate improvement over direct concatenation shows that stability-weighted fusion helps identify rotating-support vibration states, although real air-bearing validation is still required.