Electrical & Computer Engineering: Faculty Publications

CNN-Based Detection of Bearing Race Defects from Variable Reluctance Sensor Signals

Document Type

Article

Publication Date

4-20-2026

Publication Title

SoutheastCon 2026 Proceedings

DOI

10.1109/SoutheastCon63549.2026.11476701

Abstract

This paper presents a non-intrusive framework for detecting and classifying bearing outer-race defects using the existing inductive magnetic reluctance (MR) gear-tooth sensor signal available in equipment such as positive-displacement flowmeters operating in hazardous locations. Rather than installing accelerometers or other added instrumentation, often impractical in HazLoc due to certification and retrofit constraints, we exploit fault-induced amplitude and phase disturbances in the near-sinusoidal tooth-passage waveform. We transform 20 sec recordings into time-frequency spectrogram images and classify bearing conditions using transfer-learned CNNs (GoogLeNet and ResNet-50). The novelty is the repurposing of an already-installed MR speed/flow sensor for bearing diagnostics in HazLoc settings, coupled with a spectrogram-to-CNN transfer-learning pipeline tailored to MR tooth-passage signals. Results demonstrate high classification accuracy across multiple defect conditions and operating speeds.

Comments

Georgia Southern University faculty member, Rami J. Haddad co-authored, "CNN-Based Detection of Bearing Race Defects from Variable Reluctance Sensor Signals."

Copyright

This work is archived and distributed under the repository's Standard Copyright and Reuse License (opens in new tab). End users may copy, store, and distribute this work without restriction. For all other uses, permission must be obtained from the copyright owners or their authorized agents.

Link to Full Text

Share

COinS