Multivariate Signal Decomposition Techniques for Fault Diagnosis

Multivariate signal decomposition techniques have become indispensable for the early detection and diagnosis of faults in complex engineering systems. By harnessing data from multiple sensors, these methods adaptively separate non-stationary and nonlinear signals into physically meaningful components, revealing subtle fault signatures hidden within background noise. Key approaches include empirical and variational mode decompositions, dynamic mode decomposition, synchrosqueezed transforms and sliding singular spectrum analysis, often extended to handle multichannel data with enhanced mode alignment. Tensor-based frameworks such as high order singular value decomposition further capture multidimensional correlations, improving feature extraction for classification or prognostics. Collectively, these tools support real-time, robust condition monitoring across industries ranging from rotating machinery and automotive engines to civil infrastructure and energy generation.

Research from Nature Portfolio

Recent comparative analyses have systematically evaluated leading data-driven decomposition algorithms for non-stationary signals in both single- and multi-channel contexts. The investigations assessed variants of empirical mode decomposition, variational mode decomposition, synchrosqueezed transforms and sliding singular spectrum analysis on synthetic benchmarks and real-world measurements. Emphasis was placed on the mode-alignment property in multivariate signals, the influence of parameter choices on noise robustness and decomposition fidelity, and the trade-offs between adaptive flexibility and computational cost. These findings provide practitioners with a decision framework for selecting and tuning decomposition methods tailored to specific diagnostic applications, underscoring that the optimal choice depends on signal characteristics and fault types rather than a one-size-fits-all solution.

Multivariate Signal Decomposition Techniques for Fault Diagnosis publication trend

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

Technical terms

Multivariate Signal Decomposition: The process of separating multichannel measurements into constituent components that reveal underlying system dynamics or fault signatures.

Intrinsic Mode Function (IMF): A nearly monocomponent function derived from adaptive decomposition, representing a single oscillatory mode within the signal.

Empirical Mode Decomposition (EMD): A data-driven algorithm that iteratively extracts IMFs by identifying local extrema and constructing envelope means.

Variational Mode Decomposition (VMD): An optimisation-based method that decomposes signals into band-limited modes by minimising a cost function in the frequency domain.

Dynamic Mode Decomposition (DMD): A linear-algebraic framework that computes spatio-temporal coherent structures by analysing sequential data snapshots.

High Order Singular Value Decomposition (HOSVD): A tensor generalisation of matrix SVD that captures correlations across multiple dimensions in multivariate data.

Mode Alignment: The enforcement of consistent modal decomposition across different channels to enable direct comparison and fusion of extracted features.

References

1. Multi-Fault Diagnosis of Rolling Bearings via Adaptive Projection Intrinsically Transformed Multivariate Empirical Mode Decomposition and High Order Singular Value Decomposition. Sensors (2018).
2. A Novel Fault Diagnosis Method for Diesel Engine Based on MVMD and Band Energy. Shock and Vibration (2020).
3. Multivariate Dynamic Mode Decomposition and Its Application to Bearing Fault Diagnosis. IEEE Sensors Journal (2023).
4. Data-driven nonstationary signal decomposition approaches: a comparative analysis. Scientific Reports (2023).