Signal Denoising Techniques for Fault Diagnosis Systems

Signal denoising is a critical prerequisite for reliable fault diagnosis across rotating machinery, gearbox assemblies, sensor calibration and other engineering systems. Spurious noise components—whether arising from mechanical vibration, electrical interference or environmental fluctuations—can obscure or distort the true signatures of incipient faults. Over the past decade, a suite of time–frequency and adaptive decomposition methods has been developed to enhance signal-to-noise ratio while preserving diagnostic features. Wavelet-based thresholding techniques exploit multiscale decomposition to isolate fault-related energy bands, with soft and hard threshold functions tailored to the noise statistics. Empirical and variational mode decomposition (EMD, VMD) adaptively separate intrinsic mode functions, often coupled with permutation or sample entropy criteria to distinguish noise-dominated modes from information-bearing components. Local mean decomposition (LMD) and its ensemble variants address mode mixing by introducing complementary noise or smoothing filters. Time-frequency peak filtering (TFPF) can further refine processed modes by applying selective windowing in the joint time–frequency domain. Recent trends emphasise hybrid algorithms that combine decomposition, entropy-based selection and optimisation-driven threshold tuning, improving robustness under non-stationary and non-Gaussian noise conditions. Such advances have delivered heightened sensitivity in bearing fault detection, gearbox composite fault extraction and high-G accelerometer calibration, laying the groundwork for real-time monitoring and predictive maintenance across industrial and transportation sectors.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Signal Denoising Techniques for Fault Diagnosis Systems publication trend

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

Technical terms

Wavelet thresholding: A multiscale signal decomposition method using threshold functions to suppress noise coefficients while retaining significant signal features.

Empirical mode decomposition (EMD): An adaptive technique that decomposes a signal into intrinsic mode functions based on local extrema and mean envelopes.

Variational mode decomposition (VMD): A non-recursive method that extracts band-limited modes by solving an optimisation problem to minimise mode bandwidth.

Local mean decomposition (LMD): A data-driven approach that separates a signal into product functions using local mean and envelope estimations.

Permutation entropy: A complexity measure that quantifies the disorder of a time series, used to distinguish noise-dominated components.

Time-frequency peak filtering (TFPF): A technique that enhances transient signal features by filtering along instantaneous frequency ridges in the time–frequency plane.

Intrinsic mode function (IMF): A component obtained via decomposition methods characterised by a narrowband oscillatory structure amenable to instantaneous frequency analysis.

Sample entropy: A statistic that measures the unpredictability of fluctuations in a time series, aiding in mode classification.

Signal-to-noise ratio (SNR): A metric expressing the relative strength of the desired signal components against background noise levels.

References

1. Wavelet Denoising Applied to Hardware Redundant Systems for Rolling Element Bearing Fault Detection. Journal of Dynamics Monitoring and Diagnostics (2023).
2. Application of an Improved Ensemble Local Mean Decomposition Method for Gearbox Composite Fault Diagnosis. Complexity (2019).
3. A Hybrid Algorithm for Noise Suppression of MEMS Accelerometer Based on the Improved VMD and TFPF. Micromachines (2022).