John Pasquarette | Cutsforth

For decades, vibration analysis has been central to turbomachinery condition monitoring. For the mechanical faults it is designed to detect—imbalance, misalignment, looseness and bearing or gear defects—it remains a proven and effective tool.

In many programs, though, vibration and condition monitoring have become synonymous. Vibration is just one measurement domain of the many that matter to a reliability program. Turbomachinery reliability is outgrowing single-modality monitoring, and the programs that have recognized this and adopted a multiphysics approach are able to catch more failure modes earlier and diagnose them with higher confidence.

What Vibration Catches & What It Does Not

Vibration analysis quantifies mechanical severity, identifies fault types from characteristic defect frequencies and tracks progression over time. It tells whether a condition is worsening or stable. None of that is in dispute. The problem is what sits outside the vibration domain.

Electrical degradation is one example. Stator winding faults, rotor bar cracks, insulation breakdown and arcing develop electrically, often weeks to months before they produce a measurable mechanical signature. Lubrication chemistry is another. Oil debris, viscosity loss and water ingress show up in oil analysis before the bearing being affected begins to vibrate. Thermal anomalies are another example. Frictional heating and electrical hot spots appear instantly in infrared, and vibration may not respond until the damage is advanced.

There is also a timing problem inside vibration itself. For programs relying on monthly or quarterly route-based monitoring, gaps persist. Bearing damage accelerates in the final 10% of service life, which means periodic sampling routinely misses the diagnostic window where intervention is cheapest.

A machine can read “healthy” on the last route while degradation may already be accelerating This can result in false confidence. A vibration trend stays flat while electrical insulation degrades, lubrication contaminates or thermal balance shifts. The first signal the team sees is the failure itself.

| IMAGE 1: Multiphysics measurement comparison table: six measurement domains and the failure modes each is built to catch. No single domain covers the entire asset; the value is in the combination. (Images courtesy of Cutsforth)

The Retiring Expert

Most condition monitoring programs that work today were built on a small group of experienced analysts who “just know” how to read vibration data. They can tell a real fault from a transient resonance, a sensor problem from a machine problem. Tacit expertise from 20 years of pattern recognition on the same fleet of assets is the irreplaceable bedrock of a plant’s reliability program.

A large portion of that generation is currently retiring. The replacement workforce is technically capable but does not have two decades of accumulated judgment on the same equipment. Programs that depended on the grizzled veteran are now exposed, which can result in more false positives reaching the work-management system, more real faults missed in the noise and less confidence in every diagnostic call. This is not solvable by hiring harder. The knowledge transfer window is closing, and waiting for the next generation to develop the same instincts on the same equipment is not a strategy.

The Multiphysics Approach

Multiphysics monitoring does not mean more sensors or more data. It means the right combination of sensors, chosen so each measurement domain compensates for the blind spots of the others (Image 1).

Confidence, Not Sensor Count

The value of multiphysics is not in the number of signals. It is in what the signals do for each other. The benefits compound across several vectors:

More failure modes covered: Each domain captures different physical effects of degradation.
Earlier detection: ESA and EMI lead vibration by weeks to months on the failure modes they are sensitive to.
Higher diagnostic confidence: When an anomaly appears in one domain, cross-verification in another tells whether it is real or environmental noise. This means fewer false alarms and fewer missed warnings.
More accurate AI models: Holistic data from multiple domains gives AI models greater context to train and execute models.

The goal is not more data. It is about getting the right data.

A Specific Example: Medium-Voltage Motor

Consider a medium-voltage motor. Industry-standard failure mode and effects analysis (FMEA) identifies 22 distinct failure modes for this asset class. Most failure modes can be detected by more than one sensor type. So, the question is not whether a sensor can see a failure mode, but how strongly. The chart in Image 2 scores diagnostic strength on a scale of 1 to 9 across six sensor families. Equally important are the blind spots. Every measurement type misses some failure modes entirely. Broad, reliable coverage requires multiple technologies working together.

2 Stories From the Field

On a distillation column at a continuous-process chemical operation, a vacuum pump was instrumented with vibration sensors at three positions, plus a differential pressure sensor across the suction-side filter. Vibration analysis showed classic cavitation signatures: vane-pass harmonics, an elevated noise floor at impeller excitation frequencies and a rising true-peak impact trend. Vibration alone pointed at the pump. The natural next action was to pull it.

The differential pressure sensor told a different story. Filter pressure was climbing. Correlated with input from operations, the team traced the cavitation upstream to misconfigured valving causing moisture intake. The fix was not a pump rebuild, but a valve adjustment. Cavitation diminished within days. Vibration named what was happening at the pump. Process data named why.

The same plant ran a vertical boiler feedwater pump monitored with vibration, temperature, head pressure and discharge pressure. Vibration showed misalignment through dominant 2x running speed with harmonics. Discharge pressure tracked seal wear over time, and the operations team learned to use it as a scheduling trigger for seal replacements rather than reacting to leaks.

What the program did not have was electrical signature analysis on the motor driving that pump. The motor eventually tripped on over-amperage and went to the shop—likely with rotor bar damage and electrical eccentricity that ESA tracks weeks to months before an over-amperage event. Vibration plus process data caught the pump-side story. The missing electrical domain measurement was the one that would have caught the motor-side story earlier. The cost of a missing modality is rarely zero.

| IMAGE 2: Diagnostic strength of six sensor families against the 22 failure modes identified for a medium-voltage motor. Strong (≥7), moderate (4-6) and weak (1-3) scores reflect signal-to-noise for each pairing; white space shows blind spots. No single technology covers more than 11 of 22 failure modes.

| IMAGE 3: Vibration spectrum from the vacuum pump at the distillation column. Harmonics of vane pass and an elevated noise floor read as classic cavitation—a textbook call to pull the pump. Differential pressure data told a different story.

The AI Question

AI is being pitched into reliability as the answer to the retiring-expert problem. The premise is real: pattern-matching across multiple signals can find subtle changes faster and more consistently than even an experienced engineer.

But AI is only as good as the data it sees. On a single measurement domain, it has a ceiling—it can find subtler vibration patterns, but it cannot see what vibration does not measure. Multiphysics data has more inputs, more cross-verification and a more accurate root-cause identification.

There is a difference between analytics that name the failure mode and explain the reasoning, and dashboards that produce anomaly scores nobody can act on. With the right measurement architecture in place, analytics can extract real value from it.

Vibration is not going away. It remains essential for the mechanical faults it was built to find. But on the pumps, compressors and turbines that matter most—whose unplanned failure cascades through the operation—single-modality monitoring may no longer be enough.

The right combination of measurement domains finds more failure modes, finds them earlier and supports more confident diagnosis. That changes the planned-to-reactive ratio of the maintenance organization, which tips the balance between reacting to failures and planning around them

John Pasquarette is director of marketing at Cutsforth. For more information, visit cutsforth.com.

In This Issue

Cover Series Pump System Optimization