How to Detect Gas Turbine Bearing Wear and Blade Fouling from Vibration Data?

Why Is Gas Turbine Vibration Monitoring So Difficult?

Gas turbines produce some of the most complex vibration signatures in industrial machinery. A typical two-shaft gas turbine — like the GE 7F.05 (Frame 7), Siemens SGT-800, or Mitsubishi M501GAC — generates vibration energy from multiple simultaneous sources.

A Cat III or Cat IV vibration analyst can interpret this spectrum manually, but there are fewer than 5,000 certified Cat III/IV analysts worldwide, and most power plants do not have one on staff. The result: turbines are monitored with simple overall amplitude thresholds (typically per ISO 10816 or API 616) that miss early-stage bearing wear, compressor fouling, and combustion instability.

Canary Edge replaces the need for analyst expertise by learning the expected spectral shape across all frequency bands for each specific turbine. The JEPA model ingests the full spectrum — not just overall amplitude — and flags deviations from the learned multi-dimensional baseline.

How Does Blade Fouling Affect Turbine Vibration and Efficiency?

Compressor blade fouling is the most common performance degradation mechanism in gas turbines, reducing power output by 2-5% and heat rate by 1-3% before any vibration alarm triggers. On a 200 MW combined-cycle unit burning natural gas at $3/MMBtu, a 3% heat rate increase costs approximately $500K-$1M per year in excess fuel.

Fouling deposits (salt, oil mist, dust, hydrocarbon films) accumulate unevenly across compressor stages, creating mass imbalance and aerodynamic asymmetry. The vibration signature of fouling is subtle: a gradual increase in 1x amplitude (imbalance component) of 0.1-0.3 mils over weeks, plus a broadband rise in blade-pass frequency harmonics as airfoil profiles degrade.

Traditional monitoring catches fouling only when it progresses to the point of significant imbalance — typically a 1x amplitude increase of 1.0+ mils. By then, the fouling has hardened (baked-on deposits in the hot section) and requires an offline water wash or abrasive cleaning, which means a 12-24 hour planned outage.

Canary Edge detects fouling onset by tracking the ratio of blade-pass energy to shaft-speed energy. As fouling progresses, this ratio shifts in a characteristic pattern. Early detection enables online water washing (available while running, takes 30-60 minutes, no production loss) before deposits harden — saving both the offline outage cost and the months of excess fuel burn.

How Does Canary Edge Handle Multi-Shaft Turbine Vibration?

Multi-shaft gas turbines — where the gas generator and power turbine operate at independent speeds — present a challenge that breaks simple threshold monitoring. The two shafts interact: gas generator speed changes affect power turbine vibration through aerodynamic coupling, and power turbine load changes affect gas generator bearing loads through torque reaction.

Canary Edge addresses this by building a unified baseline model that incorporates data from all bearing positions simultaneously. Rather than setting independent alarm thresholds for each bearing, the JEPA model learns the expected correlation structure between:

• Gas generator bearing vibration (bearings #1 and #2)
• Power turbine bearing vibration (bearings #3 and #4)
• Shaft speeds (N1 and N2)
• Exhaust gas temperature spread
• Compressor discharge pressure

When a single bearing begins to degrade, the model detects the deviation in the context of all other measurements. This produces far fewer false alarms than per-channel thresholds — the model understands that vibration increases during load ramps are normal, while a bearing vibration increase at steady load is anomalous.

For GE Frame 7 units with the Mark VIe control system, Canary Edge reads all vibration channels via OPC-UA from the turbine control panel. For Siemens SGT-800 units, integration uses the SPPA-T3000 historian interface. No additional sensors or wiring are required if the existing protection system is already collecting vibration data.

What Does Combustion Instability Look Like in Vibration Data?

Combustion dynamics — pressure oscillations inside the combustor — are a leading cause of hot-section component damage in modern dry-low-NOx (DLN) gas turbines. Combustion instability manifests as vibration energy in the 100-1,000 Hz range, often with discrete tonal components at combustor acoustic resonant frequencies.

On a GE 7F DLN-2.6 combustion system, combustion dynamics typically appear at 150-400 Hz. On Siemens SGT-800 with the 3rd-generation DLE combustor, the primary dynamics band is 200-600 Hz. These frequencies overlap with structural resonances and blade-pass harmonics, making manual interpretation difficult.

Canary Edge isolates combustion dynamics from structural vibration by learning the normal spectral contributions in the 100-1,000 Hz band at each operating point. When combustion dynamics energy increases relative to the baseline — indicating a combustor tuning drift, fuel nozzle degradation, or pilot flame instability — the system generates an alert.

Early detection of combustion dynamics prevents cascading damage to transition pieces, first-stage turbine nozzles, and combustor liners — components that cost $200K-$2M to replace and require a major inspection outage.

How Does Canary Edge Integrate with Power Plant Historians?

Power generation facilities typically have deep historian infrastructure already in place. Canary Edge integrates as a layer on top of existing systems:

For plants running OSIsoft PI — which represents roughly 60% of large thermal power plants in North America — the integration takes less than a day. Canary Edge reads vibration and process tags from PI, runs inference, and writes anomaly scores and diagnostic tags back to PI. Operators view results in PI Vision, PI ProcessBook, or any PI-connected visualization tool.

The key advantage: no new monitoring interface to learn. Anomaly trends appear alongside the process data that operators already watch, in the tools they already use.