View article

Robust feature extraction is essential for the automation of predictive diagnostics. For years, features have been implemented and documented in the literature. However, the preprocessing of the raw vibration data is ambiguous and a comparison of the statistical features applied to Mechanical Diagnostics Test Bed (MDTB) data does not yet exist.

The focus of this paper is on features calculated on the residual and difference signals. The residual features analyzed were NA4 and NA4*, while the difference features analyzed were M6A, M8A, and FM4. Various preprocessing parameters were used in an attempt to optimize the features. These features were evaluated for the gearbox run-tofailure accelerometer data acquired on the Mechanical Diagnostics Test Bed at the Pennsylvania State University Applied Research Laboratory (ARL).