Vibrations in gas turbines: how to prevent failures and outages

Imbalance and misalignment: assess before taking action

Monitoring makes it possible to identify mechanical and operational degradation through vibrations in gas turbines. One of the biggest mistakes in diagnosing rotating equipment is automatically associating a particular frequency with a specific failure.

For example, a dominant component at 1× RPM may be caused by unbalance, but it may also result from other dynamic conditions. The diagnosis should take into account amplitude, phase, measurement direction, speed, load, behavior during start-up and shutdown, and the response of other measurement points.

In an imbalance analysis, the phase takes on particular importance. When the behavior is consistent with an improper mass distribution, a balancing procedure may be considered later, but that decision must first be supported by dynamic evidence.

Misalignment requires an equally careful evaluation. An angular or parallel misalignment can generate additional forces on the coupling and bearings and alter the vibration components of the drive train. However, it should not be diagnosed solely based on the presence of a second harmonic.

The correct sequence is:

vibration pattern → hypothesis → supplementary measurement → inspection → diagnosis → intervention.

This methodology prevents the replacement of components without first determining the physical mechanism causing the anomaly.

Bearing failures and early detection

Bearings can develop localized defects in the outer race, inner race, rolling elements, or cage. When these defects begin to cause repetitive impacts, the vibration signal may contain information related to the characteristic frequencies of each component.

In the early stages, acceleration analysis and signal processing techniques can provide information that is not necessarily evident from an overall RMS speed value. The acceleration envelope, for example, can be used to highlight high-frequency impacts associated with certain bearing defects. Recent literature continues to explore diagnostic methods that combine temporal and spectral features to improve fault identification.

But vibration should not be analyzed in isolation.

Suspected bearing failure can be confirmed by:

• Temperature trends.
• Lubricant analysis.
• Presence of wear particles.
• Changes in load.
• Changes in rotational speed.
• Temporal signal behavior.

The combination of variables increases the reliability of the diagnosis.

This principle is particularly important in gas turbines, where work on a high-speed component can result in significant downtime, as well as costs associated with disassembly, inspection, and repair.

Vibration diagnostic table

A practical maintenance tool involves correlating the observed pattern with failure hypotheses and supplementary measurements.

Observed pattern	Possible mechanism	Evidence that must be verified
1× Dominant RPM	Imbalance, among other possibilities	Phase, amplitude, speed, and rotor behavior
Harmonic components	Misalignment, play, or other nonlinearities	Phase, radial/axial alignment, and mechanical inspection
Characteristic frequencies of bearings	Localized defect	Envelope, acceleration, temperature, and lubrication
Sharp increase near a certain speed	Resonance or critical speed crossover	Start/stop, phase, and structural response
Load-dependent vibration	Operational or aerodynamic phenomenon	Process conditions, load, and thermal parameters
Gradual shift in the trend	Evolving degradation	Historical data, spectra, and auxiliary variables

Vibrations, critical speed, and rotor dynamics

Vibrations in gas turbines should be analyzed as a rotor-bearing-structure system, not as a set of independent components.

During startup and shutdown, the rotor passes through different speed ranges. At certain points, it may approach the system’s natural frequencies, causing an amplification of the vibrational response.

These phenomena are particularly important in high-speed machines because a small excitation force can generate a significant response when a resonance condition exists.

For this reason, the vibration curves during startup and shutdown provide information that cannot always be obtained during steady-state operation.

The analysis may include:

• Amplitude versus speed.
• Phase versus speed.
• Identification of critical speeds.
• Rotor behavior during acceleration.
• Behavior during deceleration.
• Relative shaft displacement.
• Bearing response.

Rotor dynamics is precisely a specialized field that studies phenomena such as unbalanced masses, whirl, stability, and the system’s vibrational behavior.

Technological advances are shifting condition monitoring from periodic campaigns to systems capable of continuously collecting data.

In a continuous monitoring architecture, data can be analyzed automatically to identify deviations from the machine’s normal behavior.

This makes it possible to develop a strategy based on three levels:

detection → diagnosis → prognosis.

Is there a deviation from the normal condition?

The diagnosis asks:

What mechanism might be causing it?

And the prognosis seeks to answer:

How might the condition progress, and when should we intervene?

This final stage represents one of the greatest challenges of predictive maintenance. It is not enough to detect an anomaly; it is necessary to determine its significance for the asset’s availability and reliability.

Modern systems can combine vibration, temperature, speed, load, and other operational variables to construct condition indicators. Current research is also exploring machine learning techniques for diagnostics and remaining life estimation, although these models must be carefully validated in the face of changes in speed, load, sensors, and operating conditions.

Machinery protection is not predictive maintenance

This difference warrants special attention. The purpose of a machinery protection system is to prevent a hazardous condition from causing further damage. In industrial applications, standards such as API 670 establish requirements for systems that monitor variables such as radial shaft vibration, housing vibration, axial position, speed, and other critical variables.

Condition monitoring serves a different purpose.

It aims to detect changes in the machine’s behavior before it becomes necessary to activate an emergency protection system.

Therefore, the two systems are complementary:

Protection:
hazardous condition → alarm/trip → equipment protection.

Condition monitoring:
change in behavior → diagnosis → maintenance planning.

Confusing these two functions can lead to a flawed strategy. Waiting until a trip condition is triggered to detect a deterioration means forfeiting much of the value that predictive maintenance offers.

Standards and criteria for evaluating vibration

Vibration evaluation must be performed using criteria appropriate for the type of machine, measurement location, and operating condition. The family establishes principles for the measurement and evaluation of vibrations in different types of machinery. The series covers various configurations and applications, including specific provisions for gas turbines.

However, a vibration limit does not in itself constitute a diagnosis.

The fact that a machine is within a limit does not necessarily mean that no degradation exists. Similarly, exceeding a reference value does not automatically identify the cause of the anomaly.

The evaluation criteria must be combined with:

• Machine history.
• Operating conditions.
• Trends.
• Spectrum.
• Phase.
• Rotor characteristics.
• Bearing design.
• Operational experience.
• Manufacturer’s recommendations.

In other words, the number is an indicator of a condition; the diagnosis requires context.

Digitization is changing vibration analysis

The growing availability of sensors and computing power makes it possible to analyze large volumes of data that previously required manual intervention.

Artificial intelligence can be used to identify patterns, detect anomalies, and help prioritize assets that require inspection. However, there is one essential requirement: data quality.

An incorrectly acquired signal can lead to an incorrect diagnosis, regardless of how sophisticated the algorithm is. That is why techniques designed to identify invalid data before feeding it into diagnostic models are becoming increasingly important. Recent research in industrial vibration monitoring has demonstrated the importance of detecting signals contaminated by sensor failures, shutdown transients, or abnormal acquisition conditions before performing automatic diagnosis.

How can we prevent failures and outages in gas turbines?

A robust program should begin with a baseline of the turbine’s performance. From there, trends can be identified and deviations detected.

Subsequently, the program must include periodic or continuous analysis—depending on the asset’s criticality—spectrum reviews, monitoring of operational variables, and correlation with physical inspections.

Frequently Asked Questions (FAQs)

References

1. https://www.sciencedirect.com/science/article/pii/S2949822826004570
2. https://www.sciencedirect.com/topics/engineering/rotor-dynamics
3. https://www.istec.com/what-does-api-standard-670-state-about-overspeed-protection-systems/
4. https://arxiv.org/abs/2608.00480