On-site balancing or shop balancing on a machine: how to choose without overpaying in downtime

A fan is shaking, and you have two paths in front of you. Call in an engineer with an instrument and balance the wheel right on the frame, or remove the rotor and take it to a balancing machine. Both approaches work, and both sometimes end up wasted effort. The difference isn't in how accurate the instrument is — it's in what exactly you measure, and what you'll have to put back together afterward.

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

This is two different methods, not two work locations

On-site balancing means exactly this: the rotor stays in its own bearing housings, is spun by its own drive, at its own operating speed, standing on its own foundation. You mount sensors on the bearing housings and point the tachometer at a reflective marker on the shaft. The instrument measures the vibration of the housings and, from the response to a trial weight, calculates the correction for this specific assembly.

Shop balancing on a machine works differently. The rotor is removed, set on the machine's supports by its journals or mounted on an arbor, and spun by the machine's own drive. The supports are standardized, the speed is constant, and the procedure is the same for every rotor of that size. What you get is a characteristic of the rotor itself, independent of the machine it was pulled from.

All the other differences grow out of that. The Balanset-1A manual describes both applications directly: the instrument works as a portable kit for balancing in the machine's own bearings, and as a measuring system for soft-bearing (above-resonance) balancing machines.

Splitting the methods into 'field' and 'shop' isn't quite accurate. A balancing machine can sit right in your own shop, and you sometimes do field work in the next bay over. What actually distinguishes the methods is one thing: which bearings and at which speed the rotor turns.

You are measuring two different quantities

This is the main thing people confuse when choosing. On site, the instrument measures vibration velocity on the stationary parts — the bearing housings and supports — in mm/s RMS (root mean square). ISO 20816 was written for exactly this kind of measurement, and by it a machine is placed in zone A, B, C or D, from A, the best condition, to D, at which continued operation is not permitted. You are reducing the vibration of the machine.

On a balancing machine, the instrument determines the rotor's residual unbalance and expresses it as a specific value in g·mm/kg. That number is then compared against the tolerance for balance quality grade G from ISO 21940-11 — the smaller the number after the letter G, the tighter the tolerance. You are certifying the rotor.

There is a link between the two quantities, but it isn't a direct one. A rotor issued at grade G6.3 can go into a machine with shaft misalignment, a bent coupling half and a soft foot (a support that doesn't sit flush against the foundation) and still give you 8 mm/s. The reverse case is just as real: a machine brought into zone A on site can contain a rotor that wouldn't pass its grade G at all. Part of the correction compensated not for unbalance, but for quirks of the assembly.

• On-site balancing gives you a reduction in the running-speed component 1x — the vibration at the rotor's rotating frequency, which is what unbalance produces — in the real bearings, at the real temperature, with the real drive.
• Shop balancing gives you a confirmed residual unbalance for the rotor and a report stating the balance quality grade G.
• Neither one delivers the other result automatically.

Look at what's actually required of you on paper. 'Balance quality grade G2.5' is a requirement on the rotor, and it's satisfied on a balancing machine using a certified procedure. 'No more than 2.8 mm/s RMS in the 10–1000 Hz band at the bearing housings' is a requirement on the machine, and a balancing machine cannot satisfy it at all. Record the applicable part and edition of the standard, the measurement points, the frequency band and the operating mode: without that, an argument over acceptance is almost unavoidable.

Sources: ISO 20816-1:2016 · ISO 21940-11:2016 · Balanset-1A operation manual

A comparison by the criteria that actually decide it

The repeatability row matters more than it looks. On site, you're not fighting the instrument — you're fighting the operating mode: air leakage, temperature, belt tension, a drive speed that won't hold still. The instrument will honestly show you the scatter, and the balancing job will stall.

On-site balancing: seven conditions it needs to work

By time and money, the on-site method wins almost every time, so it's sensible to start there. But it has an entry price. Miss even one item, and you'll burn a shift with nothing to show for it.

The mounting deserves a word of its own — it isn't a formality. The correction calculation relies on a linear model: double the mass at the heavy spot, and you get roughly double the vibration. A unit that's simply sitting on the floor, not bolted down, breaks that linearity. Under strong vibration, the centrifugal force lifts the machine slightly, the stiffness of the system changes on the fly, and the influence coefficient stops being a constant number.

Which gives you the rule: fix a faulty machine first, and only then balance it. Balancing does not substitute for repair. And if the repair requires disassembly anyway, the question of 'on site or in the shop' answers itself in the shop's favor.

• The rotor can be safely run up to its operating speed: with the cover removed or through an access window, with the danger zone guarded.
• You have access to the correction plane with the machine stopped: blades, a disc, a coupling half, a balancing ring, bolt holes.
• The speed is stable and repeats from start to start. A belt drive is acceptable, but slip will corrupt the phase — the angular reference of the vibration to the shaft marker.
• The machine can tolerate three to four starts and stops: the initial run, one or two trial runs, a check run.
• The unit is rigidly bolted to its foundation, the bolts are torqued, there is no soft foot.
• The operating speed doesn't fall inside a resonance zone of the structure, frame or piping.
• The operating mode repeats: the same damper position, the same throughput, the same temperature.

Count the number of starts in advance. One correction plane requires two runs before the calculation: the initial run and one trial run. Two planes require three: the initial run and two trial runs, one per plane. On top of that comes a check run and, if needed, one or two trim runs — for adding a small refining weight.

When on-site balancing is impossible or off the table

Here's the honest list. Fall into one of these, and field work either won't produce a result at all, or will produce one for a single operating mode, and not for long.

There's a separate case that looks like unbalance but isn't: aerodynamic and electromagnetic unbalance. Aerodynamic force grows in proportion to angular velocity, while the centrifugal force of a correction weight is proportional to its square. That means compensation is only accurate at the exact speed you balanced at, and a mismatch shows up at any other speed. The same goes for electromagnetic forces in an electric motor. A balancing machine doesn't solve this problem either — it simply doesn't see it.

Sources: ISO 21940-12:2016 · ISO 13373-5:2020 · ISO 21940-11:2016

A balancing machine in the shop: what it gives you and what it costs in return

A balancing machine buys you repeatability. The same supports every time, the same speed every time, the operator works to one procedure, and the machine's influence is excluded. For a series of identical rotors, this settles everything: set up once, then work from saved influence coefficients and spend one run instead of three.

The second reason is purely technical. Balancing machines come in two types, and it isn't a matter of taste. Soft-bearing machines have compliant supports, for example on flat springs: the natural frequency of the supports is usually 2–3 times lower than the speed of the rotor being balanced, and the support vibration is measured with accelerometers. Hard-bearing machines have rigid supports with a natural frequency 2–3 times above the rotor's speed, and they use force sensors. The advantage of a hard-bearing setup is that you can balance at low speeds already, down to 400–500 rpm: the structure and foundation are simpler, and throughput and safety are higher.

In return, a balancing machine demands a lot.

• Removal and reassembly. Taking off the coupling half, impeller or wheel, and the bearings, then, after reassembly, almost always shaft alignment.
• Rigging and transport. Slinging, knocks to mounting surfaces, lost keys and seals.
• A machine sized for the mass and dimensions. The rotor has to fit the machine by mass, diameter, length and speed range, and its journals or arbor have to seat on the machine's supports.
• The arbor and its eccentricity — the offset of the arbor's axis from the rotation axis. Arbor runout adds its own error. It's removed by index balancing: the rotor is flipped 180° in the arbor, an extra run is made, and the software calculates the actual rotor unbalance without the arbor's contribution.
• A speed that isn't the operating speed. That's acceptable for a rigid rotor, not for a flexible one.

The main caveat, which usually goes unmentioned. A rotor issued at balance quality grade G goes back in, and the machine's vibration stays high. The cause isn't the rotor: shaft misalignment after reassembly, an interference fit, a bent coupling half, a soft foot, frame resonance. So a vibration reading at the bearing housings after the rotor goes back in is mandatory, not optional.

Sources: Balanset-1A operation manual · ISO 21940-11:2016

The hybrid path: decide from data, not in advance

In practice, the choice is rarely made on paper. The right sequence is different: measure on site first, then decide on removal. One visit with an instrument is cheaper than any wrong disassembly.

1. 01

   Measure at the operating mode

   Sensors on the bearing housings, a laser tachometer on the reflective marker. Record overall vibration (the total level across all frequencies), 1x with phase, and the spectrum (the breakdown of vibration by frequency) at both bearings. Repeat the reading: a number obtained once won't hold up a decision.

2. 02

   Work out the 1x share

   If 1x accounts for most of the overall vibration, unbalance is plausible. If overall vibration is several times higher than 1x, balancing won't remove it, and the question of 'on site or in the shop' isn't about you yet.

3. 03

   Check the mechanics and resonance

   Fasteners, soft foot, play, the fit of the wheel on the shaft, bearing condition. Change the speed and watch the amplitude and phase of 1x. Resonance rules out both kinds of balancing at once.

4. 04

   Try it on site

   One trial weight shows whether the system responds linearly. A 20–30% change in 1x amplitude, or a 20–30° change in phase, means the influence coefficient can be calculated. The instrument itself will tell you whether the trial weight is adequate or needs to be increased.

5. 05

   Read the signs against on-site work

   The correction doesn't reduce vibration. The result doesn't repeat from run to run. Implausibly large weights are called for. Phase drifts at a constant speed. That's data now, not guesswork.

6. 06

   Decide on the balancing machine

   Now you have grounds for disassembly: specific system behavior and specific numbers. And a clear list of what needs to be measured while the rotor is out.

There's a side benefit to this order: even if the rotor ends up going to the shop, you'll still have 'before' readings at the real operating mode and the real points. Without them you can't prove things got better after the repair, and you won't notice if the vibration comes back a month later.

Sources: ISO 13373-3:2015 · Balanset-1A manufacturer specification

Economics: compare total cost, not the price of the labor

The most common mistake looks like this: compare the cost of a site visit with the cost of shop balancing and take whichever is cheaper. The two numbers aren't comparable, because the second one covers only part of the work.

Build your own list for each option and fill it in with your own numbers. We're deliberately not putting figures here: they depend on your equipment, your rigging rates, and what an hour of downtime costs in your process.

• Hours of downtime for the unit or the line, multiplied by your cost per hour of downtime.
• Removal and installation: people, time, lifting equipment, a work area for disassembly.
• Transport both ways and the risk of damage in transit.
• Consumables that don't survive disassembly: gaskets, seals, fasteners, sometimes bearings.
• Shaft alignment after reassembly. That's a separate job, and without it the vibration will come back.
• A standby unit: do you have one, or does the process stop?
• The likelihood of a repeat visit or a repeat disassembly, if the cause turns out not to be unbalance.

There's a flip side too: don't cut corners on disassembly where it's actually needed. A rotor with a crack, blade erosion or runout, balanced on site, will come back to you. An uncompensated centrifugal force is a dynamic load on the bearings, and their service-life calculation under ISO 281 is based on exactly that applied load. Unbalance left 'until the next shutdown' gets paid for in bearings and an unplanned stop.

Sources: ISO 281:2007

How to decide for your own case, and what AXILINE can help with

The short rule: until proven otherwise, start on site. The proof comes from a single visit, and that same proof becomes the grounds for disassembly if it's actually needed.

AXILINE's engineers design and build Balanset instruments and use them to balance on site themselves. We come out and balance rotors in their own bearings at operating speed: fans, exhaust fans, impellers, pulleys, spindles, crushers, pumps. The kit includes two accelerometers, a laser phase sensor, a two-channel USB module, and Windows software that calculates the solution in one and two planes, plots a spectrum and a polar diagram, and keeps an archive for the reports.

You get before-and-after readings at the same points, spectra, residual 1x and a report. If our data says the rotor needs to come out, we'll say so, instead of putting weights on damaged geometry. And if you balance rotors in-house on a regular basis and want to do it yourself, that same Balanset-1A works as the measuring system for a soft-bearing (above-resonance) balancing machine: here we supply the instrument and consulting support on how to use it.

• Go through the seven conditions from the on-site balancing section. If all of them check out, it's too early to remove the rotor.
• Clarify which document is actually being required of you: mm/s on the bearing housings, or a balance quality grade G on the rotor.
• Work out the total cost of both paths, including downtime and shaft alignment after reassembly.
• Keep the 'before' readings either way, even if the rotor goes to the shop.

If the rotor is flexible and there's no suitable balancing machine nearby, the honest answer is this: on-site balancing using rigid-rotor methods isn't the solution here, and we won't promise it is. The first step is to measure and find out whether the rotor is actually operating above its first critical speed. Often the assumption of flexibility doesn't hold up, and the job turns out to be an ordinary one.

Sources: Balanset-1A manufacturer specification · Balanset-1A operation manual · ISO 21940-12:2016

Frequently asked questions

Related content