Anomaly Detection of Rotating and Oscillating Bearings using Autoencoder

Abstract

This paper explores the effectiveness of Autoencoders (AE) in detecting anomalies in both Fully Rotating Bearing (FRB) and Oscillating Bearing (OB) data, with an emphasis on early degradation detection. AE demonstrate greater sensitivity and precision in anomaly detection compared to Principal Component Analysis, a traditional statistical Predictive Maintenance method. The study utilizes a comprehensive and well-studied Kaggle dataset for FRB, along with a dataset from the German Aerospace Center for OB. AE built on these datasets enable the capture of degradation patterns for both operating modes.