groundup.ai
2/9/2026Mechanical seal failure rarely announces itself with a single clean frequency. It usually appears as a combination of friction-related high-frequency energy, changes around running speed, a rising temperature at the seal chamber, and a shift in load, developing while overall vibration stays inside its normal band. A single value cannot confirm it. Reliable diagnosis requires evaluating frequency content, thermal behaviour, operating state, and how the signal changes relative to the pump's own historical baseline.
A water utility team may therefore see: a seal weeping, then a pump tripped on leakage without the overall vibration report ever having flagged a problem.
The objective is not simply to detect a peak.
It is to understand the combination of signs that precede a seal letting go.
A mechanical seal keeps process fluid inside the pump where the shaft passes through the casing. Two flat faces, one rotating, one stationary, run together separated by a microscopically thin fluid film. When that film is lost or the faces are damaged, the seal leaks, and on a water asset that can mean anything from a nuisance drip to a lift-station containment event.
Seal failure is often a symptom of something upstream. Common drivers include:
This matters for diagnosis. The seal is where the damage shows, but the cause frequently lives elsewhere, which is exactly why single-symptom monitoring struggles with it.
Overall velocity RMS is an energy average across a broad band, typically 10 to 1000 Hz, dominated by low-frequency, high-displacement events like imbalance and misalignment.
Early seal distress is not that kind of signal.
When the fluid film breaks down and the faces begin to contact, the result is friction and rubbing. That produces high-frequency, low-amplitude energy, along with heat, and sometimes small stick-slip impulses. None of it moves a 10 to 1000 Hz velocity average very much.
So a pump can be running with a seal well into distress and still pass the overall check, because the earliest evidence is thermal and high-frequency, not low-frequency vibration.
Unlike a bearing, a seal has no single defect frequency. The evidence is spread across several signals, which is the whole reason it needs multi-feature reasoning.
Running Speed, 1X, and Harmonics
A rubbing or dragging seal can raise 1X and add harmonics if it loads the shaft.
For a pump at 1,450 RPM:
1X = 1,450 divided by 60 = 24.2 Hz
Harmonics at 2X, 3X and beyond may appear as rub develops.
Friction Band, High Frequency
Face contact and dry running raise a broadband high-frequency and acoustic-emission signature, similar in band to cavitation but with different behaviour.
Temperature
Lost lubrication film generates heat at the seal faces, so a rising seal-chamber temperature is often the earliest and clearest sign.
Load and Current
Added drag from face contact can show as a small change in motor current or load.
No one of these confirms a seal on its own. The pattern across them does.
Read the signals together and a picture forms.
The confirming clue is behavioural. Because seal distress is often driven by suction and operating conditions, the friction and thermal signature frequently rises when the pump runs dry, loses suction, or operates off its best efficiency point. If the evidence builds across friction, temperature, and running-speed behaviour at once, you are watching a seal degrade, not a bearing.
Not necessarily.
Friction-band energy and rub harmonics are strong clues, but other conditions share the neighbourhood.
Maintenance teams should also ask:
This context is what separates a failing seal from a bearing fault, a cavitation problem, or a misalignment-driven rub.
Condition | Common signal clues | Where to look |
|---|---|---|
Seal failure | Rising friction-band and acoustic energy, seal-chamber temperature rise, possible rub harmonics | High-frequency and acoustic bands, temperature, load |
Bearing defect | Discrete defect-frequency tones, BPFO, BPFI, BSF | Enveloped high-frequency spectrum |
Cavitation | Broadband high-frequency floor tracking NPSH and suction | High-frequency and enveloped bands, acoustic emission |
Misalignment | Elevated 1X and 2X, axial and radial | 1X, 2X, direction across the coupling |
Looseness | Multiple running-speed harmonics | Depends on location |
These are diagnostic patterns, not absolute fault rules. In practice seal failure and its upstream causes often appear together, which is why the combination matters more than any single line.
Seal failure is read primarily through broadband friction energy and seal-chamber temperature, not the overall vibration value.
Overall vibration compresses a complex pump signal into one number, useful for general monitoring but blind to the friction and thermal changes a seal produces first.
Consider the following illustrative example.
Monitoring Stage | Friction-band energy | Seal-chamber temperature | Overall velocity (10 to 1000 Hz) |
|---|---|---|---|
Baseline | 1.00 | 1.00 | 1.00 |
Stage 1 | 1.35 | 1.10 | 1.02 |
Stage 2 | 1.85 | 1.28 | 1.05 |
Stage 3 | 2.55 | 1.52 | 1.10 |
Stage 4 | 3.40 | 1.85 | 1.20 |
Illustrative normalised data only. These are not customer measurements.
Friction-band energy and seal-chamber temperature climb well ahead of overall velocity, which remains inside the normal band.
At Stage 2:
Overall velocity = 1.05 times baseline
while:
Friction-band energy = 1.85 times baseline and temperature is climbing
A traditional overall alarm would still call the pump normal. But the seal is already losing its film, and the leak is coming.
Seal condition has to be read against operating state, because in water utilities that state rarely holds still.
Dry running and lost suction are among the fastest ways to destroy a seal, and both are tied to operating state. A friction level that is normal at full flow can be an alarm when the wet well is low. Without operating-state separation, a system either misses the developing seal problem or raises false alarms. Operating-state classification is what makes the comparison meaningful.
The same seal reads normal at its best efficiency point and alarming under dry-running or low-suction conditions, which is why operating state has to be part of the baseline.
Two nominally identical pumps do not produce identical signatures. Their behaviour differs because of:
That means a single universal friction or temperature threshold cannot tell the full story.
A pump-specific baseline changes the question from:
Is this friction or temperature level generally high?
to:
Is this abnormal for this particular pump, at this operating state?
That distinction is decisive when a seal degrades gradually.
Anomaly detection answers:
Is something different?
Fault diagnosis tries to answer:
What could be causing the difference?
Maintenance decision support goes one step further:
What should be inspected or adjusted next?
For a seal, an anomaly on a friction band is only the start. The value comes from confirming it is the seal rather than a bearing, tying it to a suction or dry-running condition, and pointing at the fix before the pump leaks.
The seal-failure workflow can be simplified into six stages.
This is a much broader process than generating an alarm.
Seal failure is one of the root causes the Groundup.ai Asset Library identifies, alongside bearing outer-race and roller faults, gear failure, coupling failure, poor lubrication, and looseness. That matters, because establishing the baseline and separating the signals is where most programs stop. Recognising the pattern and naming the cause is where the Asset Library comes in.
Because the Asset Library holds the confirmed signatures of these faults across fleets and sectors, the system does not just show a raised friction band and a warm seal and leave the interpretation to whoever is on shift. It recognises the pattern, distinguishes a failing seal from a developing bearing fault or a lubrication problem, estimates how much life remains against the P-F interval, and points the operator at the specific fix, including the upstream cause loading the seal.
That is the difference between an alert and #CognitiveMaintenance: reading the data, and understanding it.
This is not theory for the water sector. Groundup.ai was selected by PUB, Singapore's National Water Agency, as the sole provider from an open tender of 15 respondents, and awarded a ~S$4 million contract to deploy #CognitiveMaintenance across PUB's critical rotating equipment.
What the deployment monitors is exactly the combination this article describes. It analyses tri-parameter machine health data across sound, vibration, and thermal signals, the same three signatures a failing seal moves: friction and rubbing in the sound and vibration, and the lost fluid film in the temperature. The assets are the pumps and rotating equipment that keep clean water flowing.
It is a signal. The world's most essential infrastructure is moving past predictive alerts and toward machines that reason, diagnose, and guide.
Yes. Seal failure is within the set of root causes Groundup.ai identifies, because the evidence lives in signals the system already captures and the Asset Library already recognises.
Rather than relying on a single threshold, the system evaluates multiple features together:
For example: a raised friction band alone provides limited context.
But:
a raised friction band, a climbing seal-chamber temperature, rub harmonics, no discrete bearing tones, and deviation from this pump's baseline provides a strong pattern for a seal-failure diagnosis.
Data and signals identify where to investigate. Physical inspection confirms the seal condition.
What are the signs of mechanical seal failure in pump data?
Rising friction-band and acoustic energy, a climbing seal-chamber temperature, and sometimes 1X rub harmonics, usually developing together while overall vibration stays normal.
Does high overall vibration mean the seal is failing?
No. Overall velocity often stays normal during early seal distress because the first evidence is friction and heat, not low-frequency vibration.
How is seal failure different from a bearing fault in the data?
Bearing faults produce discrete defect-frequency tones under enveloping. Seal distress produces a broadband friction signature with a thermal rise, and often traces back to an upstream cause.
What usually causes a pump seal to fail?
Dry running, cavitation and suction problems, misalignment or shaft deflection, bearing wear, contamination, and excess heat. The seal is often where an upstream problem shows up.
Why does operating state matter?
Dry running and lost suction destroy seals quickly and are tied to operating state, so the same friction level can be normal at full flow and an alarm at low level.
Can Groundup detect seal failure?
Yes. Seal failure is one of the root causes the Groundup.ai Asset Library identifies, using friction, thermal, and running-speed features together and comparing them with the pump's own baseline. Physical inspection remains the final confirmation.
Seal failure does not show up where most pump routes look.
Sometimes the overall vibration is: perfectly normal, while the seal is already: losing its film and heading for a leak
The strongest clue is rarely a clean peak. It is friction energy and temperature climbing together, tied to a suction or dry-running condition, and it only means something when you know the pump's own baseline and operating state.
Effective condition monitoring therefore has to move beyond: Is the vibration high?
and toward:
Because Cognitive Maintenance is not about collecting more pump data. It is about turning that data into a maintenance decision, before the seal lets go.
Bergabunglah dengan Wall of ❤️ kami — bagikan dan tag kami!
Kami menyoroti para pemimpin yang mendorong inovasi dan kinerja, memberi Anda visibilitas sambil membentuk masa depan AI
Artikel Terkait
