Spectral Energy Redistribution for Mechanical Fault Diagnosis

53 concepts39 papers

If you hold a failing household blender, you can instantly feel it shake. In an industrial setting, measuring this overall shaking—known as global vibration intensity or root mean square (RMS) velocity—is a common way to tell that a machine is breaking down. However, as Lian Bo Shi points out, global intensity metrics lack the specificity required to isolate the exact mechanical source of a fault.

To truly diagnose a complex machine, we cannot just look at the total amount of energy it radiates; we must look at how that energy is distributed across the frequency spectrum. When mechanical components degrade, they fundamentally change the system's physical dynamics, causing energy to migrate from normal operational frequencies into highly specific, predictable frequency bands. Understanding this spectral energy redistribution is the key to modern, non-invasive machine diagnostics.

Why Faults Shift Energy Upward

In a perfectly healthy rotating machine, energy is concentrated in low-frequency bands that correspond directly to the primary rotational speed of the motor or shaft. The movement is smooth and cyclic.

However, when a component begins to fail—such as a bearing developing a pit or a gear tooth cracking—this smooth motion is interrupted. Every time the damaged component makes contact, it creates a sharp, impulsive impact. In the domain of physics, a perfectly sharp impact acts like a mathematical delta function, injecting transient energy across a broad spectrum of high frequencies.

Healthy Operation: Smooth, low-frequency cyclic motion
[───〰〰〰───] ──> Energy concentrated in primary rotational frequencies

Faulty Operation: Sharp, periodic impacts (cracks, pits)
[───⚡───⚡───] ──> Energy redistributes into high-frequency bands

Consequently, mechanical wear manifests as a distinct migration of energy from low-frequency ranges to mid- and high-frequency regions. For example, Rongsheng Zhu et al. demonstrated that when vertical centrifugal pumps experience structural faults, their vibration energy shifts dramatically out of the lower operational bands. By modeling these frequency-energy distributions, Jianhua Luo et al. showed that the ratio of high-frequency energy to total energy serves as a direct, quantitative proxy for physical wear, allowing engineers to catch internal component degradation long before a catastrophic failure occurs.

Mapping Frequency Bands to Physical Modes

Why does segmenting the spectrum work so well? It works because different physical properties of a machine govern different frequency ranges.

  • Low-frequency bands are typically dominated by macroscopic, inertial effects, such as overall structural alignment, unbalance, or total winding displacement.
  • High-frequency bands are governed by localized stiffness, microscopic surface friction, and capacitive properties.

By slicing a signal's frequency response into distinct bands, engineers can map specific spectral deviations to precise physical failure modes. As Guobin Wang et al. demonstrated in high-voltage power equipment, deviations in low-frequency bands correlate with physical winding displacements, while high-frequency anomalies point directly to localized insulation failures like inter-turn short circuits.

To automate this insight, Yang Jianwei proposed mathematical abstractions called energy eigenvectors. Instead of forcing a diagnostic system to process a massive, noisy time-domain waveform, the waveform is decomposed into discrete frequency bands. The energy in each band is calculated and compiled into a low-dimensional vector:

$$\mathbf{E} = [E_{\text{band}_1}, E_{\text{band}_2}, \dots, E_{\text{band}_n}]$$

This vector acts as a unique feature signature. Because different mechanical faults redistribute energy into different higher-order bands, these eigenvectors create a clear spatial separation in machine learning models, making it simple to differentiate between distinct types of mechanical wear.

Isolating Signatures in High-Noise Environments

In real-world factories or power plants, the faint signature of a failing bearing or gear is often drowned out by the roar of surrounding machinery. To find the fault, engineers must isolate the exact frequency bands where the anomalous energy is concentrated.

One primary tool for this is Fast Spectral Kurtosis (FSK). Kurtosis measures the "peakedness" or impulsiveness of a signal. By calculating kurtosis across a grid of localized frequency bands, FSK pinpoints exactly which frequencies contain the sharp, faulty impacts while ignoring steady background noise. Lan Chen et al. leveraged this technique to analyze strain signals in complex planetary gearboxes, successfully isolating the precise frequency characteristics that differentiate a sun gear crack from an inner ring crack.

Alternatively, when a fault causes a high-frequency structural resonance to ring out periodically, researchers use demodulated envelope analysis. This technique strips away the high-frequency carrier wave and focuses entirely on the repetition rate of the energy bursts. The authors of a study on mine gearbox fault diagnosis showed that by suppressing random noise and enhancing periodic impulses, this approach reveals a clean "harmonic comb" in the frequency spectrum. This comb serves as an unmistakable, definitive signature of gear failure, even in extreme industrial environments.

Predicting Failure via Spectral Energy Regression

Once the critical, fault-sensitive frequency bands are isolated, the goal shifts from mere detection to long-term prognosis: When will the machine fail?

Log-Energy
  │                                     /  Failure Threshold
  │                                    / 
  │                       *   *   *   /  <-- Projecting trend line
  │               *   *              /
  │       *   *                     /
  │   *                             |
  └───┴───┴───┴───┴───┴───┴───┴───┴─┴──── Time
     Now                            Estimated RUL

Instead of waiting for the overall vibration to cross a generic danger threshold, Hani Beirami et al. developed a method known as spectral energy regression. This approach continuously tracks the log-energy contained strictly within the fault-isolated frequency bands over time. Because the energy in these targeted bands grows predictably as a physical defect worsens, applying linear regression to the energy trend allows the system to project forward and calculate exactly when the energy will breach a failure threshold. This transforms raw vibration data into a highly accurate estimate of a machine's Remaining Useful Life (RUL).

Key Foundational Work

Reliable Control of Ship-Mounted Satellite Tracking Antenna (2010) — contributed 1 concept elements to this direction.
Automatic Page Turning Machine for Various Books (2005) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_centroid_frequency_analysis_081b65a9,
  title = {centroid frequency analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/centroid-frequency-analysis/rdr_081b65a912dd4ad998393ec652e4b751},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_centroid_frequency_analysis_081b65a9_v1,
  title = {centroid frequency analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/centroid-frequency-analysis/rdr_081b65a912dd4ad998393ec652e4b751/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}

See something wrong? Report a mistake