A novel rolling bearing fault diagnosis method with a tunable spiral spring coupling device for fault feature enhancement

Abstract

Rolling bearing fault diagnosis is critical for rotating machinery safety, but weak fault characteristics are often buried in background noise, complicating feature extraction and fault diagnosis. This study proposes a novel diagnosis method integrating a tunable spiral spring coupling device (TSSCD) for feature enhancement and a convolutional neural network (CNN) for classification. The TSSCD comprises a spiral spring, coaxial repelling magnets, and a tuning mechanism. By adjusting the magnet distance, its resonant frequency is tuned to match the bearing’s fault characteristic frequency, amplifying weak fault signals while suppressing high-frequency noise. The theoretical model of the TSSCD, encompassing the spiral spring’s linear stiffness and magnetic repulsion, is established and experimentally validated. Results show its natural frequency tunable from 103 to 135 Hz, with good agreement between theory and experiments. The TSSCD was then mounted on a rolling bearing experimental rig to collect signals under four types of test bearings. TSSCD-enhanced and raw signals were input into a simplified CNN. The CNN achieved 99.51% average accuracy with enhanced signals, 3.17% higher than the 96.34% accuracy with raw signals. This method eliminates the need for complex traditional feature extraction, features easy TSSCD installation, and exhibits strong robustness, providing a new way to integrate physical vibration control with deep learning for high-precision fault diagnosis.