Fault Diagnosis Techniques for Rolling Element Bearings

Fault diagnosis for rolling element bearings underpins the reliability and efficiency of rotating machinery across industries such as power generation, transport and manufacturing. Traditional approaches rely on vibration-based monitoring, extracting time-domain statistical indicators (rms, kurtosis), frequency-domain spectra and envelope analysis to detect characteristic fault frequencies associated with defects on the inner race, outer race or rolling elements. Advances in signal processing have introduced time-frequency representations (wavelet transform, Hilbert–Huang transform, multisynchrosqueezing) and adaptive decomposition methods (empirical mode decomposition, variational mode decomposition, singular value decomposition) to isolate weak impulsive signatures in noisy environments and under variable load conditions. Parallel developments in machine learning have leveraged classical classifiers (support vector machines, decision trees) trained on hand-crafted features, while deep learning architectures (convolutional neural networks, autoencoders) are increasingly applied directly to raw or pre-processed vibration signals. Hybrid frameworks that combine signal-decomposition techniques with neural networks have demonstrated improved sensitivity and robustness in incipient-fault scenarios. Real-time implementation and integration with condition-based maintenance systems extend the global impact of these methods, enabling early detection and prognosis that reduce downtime, extend component life and enhance safety.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has introduced an improved singular value decomposition packet algorithm that restructures the trajectory matrix to suppress mode mixing and enhance extraction of bearing-fault impulses in vibration signals. Another study proposed a hybrid framework combining SVD-based denoising with a self-adaptive time-reassigned multisynchrosqueezing transform, optimising parameters via a permutation-entropy metric to reveal weak fault features under strong background noise, as demonstrated on high-speed train axle bearings. In parallel, integration of SVD with a one-dimensional convolutional neural network has been shown to automatically learn and highlight discriminative features from raw bearing signals, achieving superior classification performance on both simulated and experimental datasets.

Fault Diagnosis Techniques for Rolling Element Bearings publication trend

The graph below shows the total number of articles in fault diagnosis techniques for rolling element bearings across all publications each year (not limited to Nature Index journals).

Technical terms

Rolling element bearing: A mechanical component using rolling elements (balls or rollers) to support radial and axial loads while minimising friction between moving parts.

Singular value decomposition (SVD): A matrix factorisation technique that separates a signal into orthogonal components ordered by energy, used for noise reduction and feature enhancement.

Time-frequency representation (TFR): A signal-analysis method that describes how spectral content evolves over time, examples include wavelet transform and multisynchrosqueezing.

Envelope spectrum analysis: A demodulation technique that extracts amplitude variations (envelopes) of high-frequency components to reveal periodic impulses caused by bearing defects.

Convolutional neural network (CNN): A deep-learning model employing convolutional layers to automatically learn hierarchical features from raw or transformed input signals.

References

1. A Novel Bearing Fault Diagnosis Methodology Based on SVD and One‐Dimensional Convolutional Neural Network. Shock and Vibration (2020).
2. A Hybrid SVD-Based Denoising and Self-Adaptive TMSST for High-Speed Train Axle Bearing Fault Detection. Sensors (2021).
3. Bearing Fault Diagnosis Method Based on Improved Singular Value Decomposition Package. Sensors (2023).