Vibration came back after balancing: why, and what to check

An exhaust fan was balanced in March. Vibration dropped from 9 to 1.8 mm/s, the report was signed, the machine was closed up and forgotten. In May the operator complains of rumbling again, and a reading gives 6 mm/s. The question isn't whether the balancing was done badly. The question is what changed over those two months, and how you tell, without taking the machine apart.

Updated 27 August 2026 · by AXILINE · Vila Nova de Gaia

Two different situations that get called by the same word

First, separate two situations, because they're treated differently.

The first: vibration after balancing didn't drop at all, or barely dropped. That's a misdiagnosis — the vibration was never coming from unbalance, and that case is covered in a separate article on why balancing sometimes doesn't help: resonance, misalignment, looseness, oil whirl, hydraulics, an electromagnetic cause.

The second situation, the one this piece is about: the machine genuinely came within tolerance. You saw the drop with your own eyes, the check run confirmed it, overall vibration settled into zone A or B (condition zones under ISO 20816: A is the level of a new machine, B is fit for long-term operation). The machine ran for weeks or months, and the vibration climbed back up.

The difference matters. Unbalance is a distribution of mass around the rotor. Balancing fixes it at the moment of the visit, and it holds exactly as long as the mass, the rotor's geometry, and the stiffness of the supports stay the same. The moment any of that shifts, the weights calculated back in March stop being correct.

Which gives the first practical takeaway: don't start with weights. Start with the previous balancing report and one new reading: overall vibration, and the amplitude and phase of 1x (phase being its angle relative to the shaft marker) at the same points, in the same direction, at the same speed. Comparing the two vectors — amplitude and phase together — will tell you more than another trial run.

If there's no report, you're working blind and losing half the information. In that case, at least record the current numbers in full: points, directions, speed, operating mode, overall vibration, and the amplitude and phase of 1x. That becomes the baseline for next time.

How exactly the vibration came back: a table of clues

The shape of the return tells you more about the cause than the absolute number on the display. A gradual rise over weeks, a jump within one run, and a step after a repair are three different stories, and you can tell them apart before you even pick up the instrument.

Three questions to ask first. When did it start. Did it rise gradually or jump. What was happening with the machine and the product during that period.

The table gives you a hypothesis, not a conclusion. Causes like to come in pairs: the wheel picked up build-up, and the foundation settled at the same time. So finding the first sign isn't a reason to stop checking.

Sources: ISO 13373-3:2015 · ISO 281:2007

Gradual rise: the wheel built up again, or wore down

The two most common return scenarios look similar on a graph and call for completely different fixes. Both produce a rise in 1x specifically, with a stable or slowly creeping phase. You tell them apart by whether they're reversible.

Build-up is reversible. On an exhaust fan, a fan handling wet dust, a mulcher, a screw conveyor, material settles unevenly on the impeller: sometimes in a pocket at a blade root, sometimes on the disc. You get a genuine unbalance that grew out of the process, not out of the rotor. The tell is simple: wash the wheel and take a reading again. If the vibration drops back to close to the post-balancing level, it was build-up.

Erosion is not reversible. An abrasive stream, cavitation, or hot gases carrying particles strip metal off the blades unevenly. The process is slower than build-up, runs over months, and no amount of washing reverses it. On the edges you'll see waviness, thinning, and in places pitting and undercuts at the root. Here, balancing becomes a routine maintenance operation on a symptom, not a fix.

Let's be direct about where the method stops working. A worn-down wheel can be balanced indefinitely, but each correction only lasts until the next measurable increment of wear, and you end up visiting the machine more and more often. At some point it's cheaper to restore or replace the impeller. The signal for that switch: the interval between balancing jobs keeps shrinking, and the mass of the correction weights keeps growing from one visit to the next.

A separate note on cracks. A crack in a blade, at the root or along a weld, also produces a slowly rising 1x, and it too can be masked with weights. Don't do that. Masking the vibration from a cracked wheel means the machine keeps running with a growing crack until it fails. Before any repeat balancing on a wheel that runs with abrasive material or in a hot environment, inspect the blade roots and welds.

• Build-up: rises over hours, days, weeks; correlates with product and moisture; washing brings vibration back down to close to the earlier low level.
• Erosion: rises over months; washing has no effect; the correction weight mass grows from visit to visit.
• A crack: gradual rise, often with a phase change; vibration is unstable between runs; inspection reveals a line at a blade root or along a weld.
• Build-up breaking off in one piece: a jump in vibration with no human intervention; a clean patch is visible on the disc or blade against the surrounding build-up.

Wash the wheel evenly and completely. Partial cleaning is a cause of unbalance on its own, and a reading afterward is mandatory, even if the machine was running smoothly before the wash.

Sources: ISO 13373-5:2020

A sharp jump: the correction weight came loose

If the vibration rose not gradually but within a single run, the first thing to check is the weights. This is the most common cause of a sudden relapse, and the most frustrating one, because it's entirely within our own control.

The picture on the instrument is recognizable. The 1x amplitude jumps, and the 1x phase shifts by tens of degrees, because a specific mass at a specific point has dropped out of the vector sum. Often the new 1x vector turns out close in length to what you were correcting for last time, and points roughly opposite the installed weight. That's practically a signature.

How a weight comes loose. A tack weld at two points instead of a full seam cracks from vibration and thermal cycling. An unlocked bolt backs out. A magnetic or adhesive weight slides on a heated surface. A weight welded onto a dirty or painted surface comes off along with the paint. A weight placed on a blade edge gets sheared off by an abrasive stream.

There's also a softer variant: the weight is still there, but has partly worn or burned away. In that case the rise comes in steps rather than a jump, and the mass on the scale won't match the report.

• Locate every installed weight by the report and the markings. If there's no marking, that's the first thing worth fixing on this visit.
• Remove each weight and weigh it. Compare against the report: a discrepancy of more than a few percent already accounts for part of the rise.
• Check the weld or fastening. A full weld seam around the contour, not two tack points. A bolt with a nut and a locking feature, not a self-tapping screw.
• Prepare the surface: clean it down to bare metal, degrease it, no paint or dirt under the weight.
• Keep the same radius the correction was calculated at. A weight moved closer to the axis works less effectively, and the vibration will come back partially even at the check run.
• Place the weight where the flow won't shear it off and the housing won't catch it. On a wheel handling abrasive material, drilling or a weight on the leeward side of the blade works better.
• Mark permanent plane and position references on the rotor and record them in the report. The next check will take minutes.

If you're working in fixed-position mode, record not just the mass but the position number too: Z1, Z4 and so on. That removes any argument about exactly where the weight was installed when a different person comes to the machine six months later.

Not 1x that's rising: bearing, fasteners, foundation

Here's the key fork in the road. Look at two numbers side by side: overall vibration in mm/s RMS (root mean square) in the 10–1000 Hz band, and the 1x amplitude. If it's mainly 1x that rose, you're dealing with the rotor's mass. If overall rose more than 1x did, unbalance has nothing to do with it, and rebalancing won't get you anything.

Check repeatability before drawing any conclusions. Run the machine twice at the same speed and compare the 1x vector. A phase discrepancy of a few degrees is normal. A discrepancy of tens of degrees means the system is nonlinear, and any correction calculation will be unstable no matter how carefully you compute it.

Sources: ISO 281:2007 · ISO 13373-3:2015

The machine became a different one: mode, speed, heat, mounting

A separate class of relapse, where the rotor itself is entirely fine. You balanced the machine for one set of conditions, and now it's running under a different one. The unbalance hasn't changed — everything around it has.

Thermal behavior and nonlinear response are especially characteristic of flexible rotors operating above their first critical speed. For them, the applicable part of the standard and the balancing method are different from those for rigid rotors. Check which class your machine falls into before carrying familiar approaches over to it.

Sources: ISO 21940-12:2016

What to do when it comes back: measure, don't reach for weights

The temptation is understandable: the instrument is already there, the influence coefficients — the machine's response to a trial weight, recorded last time, used to calculate the correction — are saved, no trial runs needed, the add-on is calculated in a minute. That's exactly why blind rebalancing is so common, and so often harmful. It masks the indicator and leaves the cause running.

The sequence that closes out most cases in one visit looks like this.

1. 01

   Pull the previous balancing report

   You need the speed, mode, measurement points and directions, overall vibration before and after, the amplitude and phase of 1x at each bearing, the mass and angular position of the installed weights, the installation radius, and the influence coefficients. Without this data you're doing a fresh balancing job from zero, not investigating a relapse.

2. 02

   Repeat the reading under the same conditions

   Same points, same direction, same sensor mounting, same speed, same mode. A sensor stud or magnet on a clean, flat spot. A reading taken at a different mode can't be compared with the earlier one: an argument about the cause becomes unresolvable.

3. 03

   Compare overall and 1x separately

   If mainly 1x rose, work with the rotor's mass: build-up, wear, a weight that came loose. If overall rose more than 1x, work with the mechanics: bearing, looseness, rubbing, mode. This is the first and most useful fork, and it takes a minute.

4. 04

   Compare the 1x vector against the report, not just the amplitude

   If the amplitude rose but the phase stayed nearly the same, mass is changing gradually in the same angular sector: that's build-up or wear. If the phase shifted by tens of degrees, a mechanical event happened: a weight, flaked-off build-up, a chipped edge, a rotated fit. These call for different fixes.

5. 05

   Go over the mechanics by hand

   Torque on the foot and support bolts, soft foot, play, the fit of the wheel on the shaft, grout and anchor condition, belt and pulley condition, fit runout, signs of rubbing inside the housing, bearing temperature, lubrication. Inspect the wheel: build-up, cracks at blade roots and welds, edge condition, and whether the weights are intact. Part of the cases get resolved right here, without instruments.

6. 06

   Check whether the conditions changed

   Actual speed against the report. Damper position. Product and moisture. The variable-frequency drive's setpoint. What was done to the machine over the intervening period, per the repair log. If conditions differ, explain the difference first, calculate weights second.

7. 07

   Only now decide about balancing

   The cause is unbalance again, the mechanics are sound, the conditions match — balance. With saved influence coefficients, that's done without trial runs, in one run and one correction. The cause is something else — fix it, and do the balancing after the repair, because the influence coefficients will have changed anyway.

If you're adding weights to the same machine twice in a quarter, you're not balancing — you're maintaining the cause. Stop and work out exactly what's changing the mass or the stiffness.

Prevention: so a relapse stops being a surprise

A vibration relapse can't be banned outright, but it can be seen coming, letting you plan a shutdown instead of an emergency. It costs one reading a week and one table.

What matters isn't just the magnitude, but the trend. Absolute criteria for overall vibration in mm/s RMS give you a condition zone. Deviation from a sound machine's baseline gives you a process. A several-fold rise from baseline over a month deserves attention even if the number still sits inside zone B.

• Take a baseline reading right after balancing: overall vibration and 1x at each bearing, in three directions, with phase and speed. This is your reference point, and without it there's no trend to build.
• Start a trend. One reading every week or two, at the same mode, the same points and directions. A graph over weeks will show the shape of the rise, and the shape points to the cause.
• Set yourself a response threshold, not just an alarm threshold. For example: 1x doubling from baseline means inspecting the wheel at the next stop.
• Derive the impeller cleaning interval from your own trend, not from someone else's recommendation. Half the time it takes the machine to go from baseline to the response threshold works as a sensible interval.
• Fasten correction weights for the long haul: a full weld seam instead of tack points, a locking bolt instead of a self-tapper, a clean, degreased mounting spot, the same radius as in the calculation.
• Mark permanent references for the correction planes and fixed positions right on the rotor, and enter them in the report.
• Document the influence coefficients for every machine. With an unchanged system, rebalancing comes down to one run with no trial weights.
• Keep not just the conclusions but the raw data too: time waveforms and spectra. Six months on, they're what will tell you whether a defect is growing or the reading was simply taken differently.
• Link the vibration log to the repair and operating-mode logs. Half of all mysterious relapses turn out to be explained by an entry that says 'changed the coupling' or 'raised the setpoint by 10%.'

Use the numbers and zones as a working reference. Condition by overall vibration is assessed under ISO 20816 (formerly ISO 10816), and rotor balance quality by grade G under ISO 21940-11 (formerly ISO 1940-1). Check the applicable part and edition against your own machine, and for contractual acceptance, record the part and edition of the standard, the measurement points, the frequency band, the operating mode and the type of supports.

Sources: ISO 20816-1:2016 · ISO 21940-11:2016

What AXILINE does when vibration comes back

We're the engineers who design and build Balanset instruments and use them for field balancing ourselves. So on a repeat call, we come out with the instrument and the previous report, not with weights.

The first step is measuring and comparing. Same points, same directions, same speed, same mode. We look at what exactly rose: the running-speed component, or the whole of overall vibration. We compare the 1x vector against the previous one, by amplitude and by phase. We check repeatability between runs, look at the spectrum and the time waveform, go over the mechanics, and inspect the impeller and the weights. Only after that do we say whether balancing is needed, or a repair.

You'll get a straight answer even if you don't like it. A worn-down wheel — we'll say balancing is going to become a routine operation, and suggest repairing the wheel. A crack at a blade root — we'll decline to balance. A wrecked bearing or a settling foundation — we'll name the order of work, with balancing coming last, not first.

When the cause really is unbalance, we balance the rotor on site, in its own bearings, with no disassembly: in one or two planes, splitting the weight across fixed positions, and if there's nowhere to weld, we calculate a drilling correction instead. With saved influence coefficients, a repeat visit takes one run with no trial weights. We document the result with a before-and-after report, so you have a baseline for the next comparison.

The instrument can be taken on yourself, to run the machines on your own. The kit includes two accelerometers, a laser phase and speed sensor, a two-channel USB module with preamplifiers, integrators and an ADC, and Windows software: overall vibration, 1x amplitude and phase, an FFT spectrum, the time waveform, simultaneous two-channel acquisition, a polar diagram, tolerance calculation by grade G, trim balancing, balancing from saved influence coefficients, recalculating weights for other planes, and a results archive for reports. The Balanset-1A is also used as the measuring system on soft-bearing balancing machines. Consulting support is included, including going through your own readings and spectra with you.

Sources: Balanset-1A operation manual · Balanset-1A manufacturer specification

Frequently asked questions

Related content