# On-site balancing of centrifugal compressor rotors and impellers, where they operate

> A centrifugal compressor forgives no approximations. Speeds are high, the rotor is often flexible, the bearings run on oil film, and a significant part of the vibration comes from the gas, not from mass. We come out to measure and separate the causes: the running-speed component (vibration exactly at the rotation frequency, denoted 1x), the subsynchronous region (frequencies below running speed), blade-pass frequency, operating regime. We balance what can genuinely be balanced on site, and for everything else we hand over a report with numbers.

**In short:** Partly. On site, in the machine's own bearings, we balance overhung impellers on single-stage machines through an open inspection port, half-couplings, free shaft ends, drive-motor rotors, pulleys, and oil-cooler fans. Rotors of multistage compressors in horizontally split and barrel-type casings, as well as high-speed pinion shafts on integrally geared machines, are not balanced on site: the correction planes lie inside the flow path, the rotor behaves as flexible, and the residual-imbalance tolerance calls for specialised conditions. In these cases, our work on site is measurement and cause separation: imbalance, oil whirl, shaft misalignment, gear wear, surge, blade-pass pulsations. You get a report and a clear next step, not trial runs by guesswork.

Source: https://axiline.pt/en/equipment/on-site-balancing-centrifugal-compressor-rotors/  
Publisher: AXILINE · Vila Nova de Gaia, Portugal · +351 931 831 229 · axilinegeral@gmail.com

## Three rotor designs, three different answers

"Centrifugal compressor" on a site can mean at least three different machines, and the question of on-site balancing is settled separately for each one. We look not at the nameplate but at how many impellers sit on the shaft, how the casing is closed, and whether there's a speed increaser — a step-up gearbox — between the driver and the rotor.

It all comes down to one question: is there a correction plane you can reach without dismantling the flow path.

### Single-stage machine with an overhung impeller

The impeller sits outboard of the bearing support, and the cover comes off or an inspection port opens. The only group where on-site impeller balancing is realistic. Imbalance comes from product build-up, erosion, weld repair, or the impeller being repositioned on the shaft. Correction planes: the standard threaded holes in the main disc, a balancing ring, the impeller nut, the half-coupling at the opposite end.

### Multistage rotor with a horizontal split

Several impellers on a common shaft between two bearing supports. The upper half of the casing comes off, but the correction planes lie within the flow path, and the impellers are shrink-fitted on. They can't be touched without the manufacturer's procedure. On site, you're left with the half-coupling and the free shaft end, and that's far from always enough.

### Barrel-type casing

The rotor comes out only as the complete inner bundle, as a whole, through the end. There's no access to correction planes on site at all. A rotor like this is balanced under specialised conditions, usually as an assembly, with monitoring during run-up.

### Integrally geared machine

Impellers sit on the ends of pinion shafts, which are driven by the common bull gear of the speed increaser. The high-speed shaft turns at tens of thousands of rpm, and every measurement has the gear-mesh frequency sitting right next to it. Balancing a shaft like this isn't done on site. On site, we work with the low-speed line, the driver, and the coupling.

> Record the number of blades on each impeller and the speed increaser's gear ratio. Without these two numbers, a compressor machine's spectrum can only be read halfway: you won't be able to separate blade-pass frequency (number of blades times rotation speed) from gear-mesh frequency, and you'll mistake both for something else.

## Speed and tolerance: why the margin is tighter here

Force from an unbalanced mass grows with the square of angular speed. On a machine that spins several times faster than a fan, the same gram at the same radius loads the bearing support several times harder. Two practical things follow from that.

First: the residual-imbalance tolerance on these rotors is among the tightest, and it's taken not by analogy with the machine next door, but from the manufacturer's requirements and the applicable part of ISO 21940, current edition. How the balance quality grade is chosen is covered in a separate article.

Second: a trial weight is risky here. On a fan, you fit a trial mass with margin to spare and see what happens. On a high-speed rotor, a mass like that can push the level past the trip setpoint, overload the bearing, and damage the mounting point. We calculate the trial weight from the baseline level and the allowable force increment, not from habit.

- We choose the trial mass so the response reads confidently while the added load on the support stays within a reasonable fraction of the rotor's weight.
- We record the actual mounting radius, measured on the machine, not taken from the drawing.
- Mounting only by standard means: threaded holes, balancing screws and rings. Adhesive and tack-welding are out of the question on a high-speed rotor.
- We agree the vibration alarm and trip setpoints with your team before the first trial run.
- If the manufacturer limits the total weight mass or the number of holes that may be used, we work within that limit and note it in the report.

Sources: [ISO 21940-11:2016](https://www.iso.org/standard/54074.html)

## Flexible rotors and critical speeds

The influence coefficient method, which underlies any on-site balancing, requires a repeatable response. The rotor has to react the same way to the same mass from run to run. A rigid rotor operates with margin below the first critical speed — the rotation rate at which the rotor enters resonance and bends: the deflection shape doesn't change, and the correction found for it holds.

A centrifugal compressor rotor often operates above the first critical speed. It bends, and the deflection shape depends on speed. A weight that removes vibration at operating speed can raise the amplitude when passing through the critical speed on the next start. A rotor like this needs modal balancing: several planes and several speeds, not two trial runs on site.

The coast-down settles it — a measurement taken as speed drops freely after the driver is switched off. We record the amplitude and phase (angular position) of the running-speed component during run-up and coast-down, locate the critical speeds, and check the margin to the operating regime. If a peak sits close by and the phase at operating speed shifts from run to run, there won't be a stable result on site, and we'll say so the same day. A detailed look at how flexible rotors behave is in the article on critical speed.

> We also look separately at thermal bow. Hot gas and uneven heating of the shaft change the running-speed component as the machine comes up to operating conditions. We take the measurement only once the thermal state has settled, and we record the temperature. A cold machine and a warmed-up machine are two different objects, and comparing their numbers is pointless.

Sources: [ISO 21940-12:2016](https://www.iso.org/standard/50429.html)

## Journal bearings: what they do to your numbers

Almost all these machines run on journal bearings. The oil film damps vibration, and the bearing housing transmits less to the outside than what the shaft actually does inside the clearance. The absolute numbers on the housing look modest even when the shaft is moving noticeably.

This has little effect on the balancing itself: the amplitude and phase of the running-speed component read fine from the housing. It has a strong effect on condition assessment. If the machine has standard proximity probes — non-contact sensors that track the shaft's movement within the clearance — we compare our numbers with the shaft's relative displacement, and we take the limits from the applicable part of ISO 20816 for your machine type, power, speed, and bearing type. The zones for housing vibration and for shaft displacement are different scales, and they can't be mixed.

- [x] Oil temperature and pressure at the bearing inlet, grade and viscosity, condition of the oil cooler.
- [x] Bearing clearance and pad preload, if the bearing is a tilting-pad type.
- [x] Load on the bearing. An underloaded bearing is a classic setup for oil-film instability.
- [x] Axial position of the rotor and condition of the thrust bearing.
- [x] Bearing housing temperature and the allowable limit for a magnetically mounted accelerometer.
- [x] Standard monitoring channels: proximity probes, phase sensor, the active setpoints.

Sources: [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) · [ISO 281:2007](https://www.iso.org/standard/38102.html)

## The subsynchronous region: where imbalance stops

Imbalance lives exactly at the rotation frequency. Everything below it has nothing to do with balancing, and on a centrifugal compressor the subsynchronous region is densely populated. We sort through it before we ever reach for weights.

| What the measurement shows | What it usually is | What's done about it |
| --- | --- | --- |
| A peak at roughly 0.4 to 0.5 of rotation frequency, that tracks speed as it changes | Oil whirl: instability of the oil film in a journal bearing | Clearance, viscosity, oil temperature and pressure, bearing load. Balancing doesn't help |
| The subsynchronous peak has stopped tracking speed, locked near the first critical speed, amplitude is rising | Oil whip: the whirl has locked onto the rotor's bending mode | Shutdown and investigation together with your team. This is a dangerous condition, not a reason to fit weights |
| A subsynchronous component at roughly 0.7 to 0.9 of rotation speed, appearing as flow is reduced | Rotating stall in the flow path or the diffuser | Operating point, guide vanes, recirculation. Not fixed with weights |
| Very low-frequency oscillations in pressure and flow, the machine "breathes," flow reverses | Surge: instability of the compressor-and-network system | Anti-surge control, operating regime, piping. Has nothing to do with imbalance |
| A peak at the number of blades times rotation speed, and its harmonics | Blade-pass frequency and gas pulsations, tight impeller-to-diffuser clearance, header acoustics | Operating regime, clearances, discharge piping supports |
| A peak at the speed increaser's gear-mesh frequency with sidebands | Gear wear or a gear defect | Gear diagnostics. We carry out balancing separately for each shaft |
| The running-speed component dominates, phase repeats from run to run, level doesn't depend on the regime | Imbalance | Balancing, if an accessible correction plane exists |

> The last row is the most useful one. Vibration that's indifferent to the valve, the flow, and the pressure is mechanical, and that's what we reduce with weights. Anything that changes along with the process regime is worked through the process side. How to read a spectrum step by step is laid out in our article on the vibration spectrum.

Sources: [ISO 13373-3:2015](https://www.iso.org/standard/40840.html)

## How the visit goes

1. **Limits of intervention agreed before the visit** — We gather the data: machine type, speed of each shaft, gear ratio, number of blades, bearing type, whether proximity probes are fitted, manufacturer's requirements. This is also where we settle what's actually allowed to be touched. If the answer is "nothing," we go out for diagnostics, and you know that in advance.
2. **Baseline measurement at steady-state conditions** — Two accelerometers on the bearing supports, magnetically mounted on a cleaned, flat spot, direction radial-horizontal and the same in every run. The laser phase sensor is aimed at the reflective mark on the shaft we're actually balancing. On a machine with a speed increaser, a mark on the wrong shaft makes the whole measurement worthless.
3. **Run-up and coast-down** — We record the 1x amplitude and phase as speed changes, locate the critical speeds and any resonances in the frame and piping, and assess the margin to the operating regime.
4. **Separating the causes** — We compare the overall level with the running-speed component, and work through the subsynchronous region, the second harmonic together with axial vibration, blade-pass and gear-mesh frequencies, and the high-frequency bearing zone. If imbalance isn't confirmed, we stop here. That's a normal and frequent outcome.
5. **Balancing what's accessible** — If imbalance is confirmed and a correction plane is accessible, we proceed with a trial weight using the influence coefficient method: from the machine's response to the trial weight, the software calculates the correction mass and angle. One plane means a baseline run and one trial run; two planes mean a baseline run and two trial runs, plus a verification run. Every run means a shutdown, depressurising, and opening up access.
6. **Verification run and trim** — We repeat at the same regime and the same temperature. If the level has dropped but the target value hasn't been reached, the software calculates an add-on to the weights already fitted. We save the influence coefficients: the next trim will go through without repeat trial runs.
7. **The report** — We hand over the before-and-after numbers for each bearing support, spectra and the time waveform, measurement conditions, weight locations and masses, plus a separate list of what balancing won't fix, prioritised by urgency.

> We don't start trial runs until we've confirmed that the running-speed component accounts for most of the level and its phase repeats. On a high-speed rotor, a trial weight fitted by guesswork isn't a wasted hour — it's a risk.

Sources: [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/)

## What's actually balanced on site

Below are the planes we work with on a compressor unit without intervening in the flow path. The list is shorter than we'd like, but every entry is a working option.

- The overhung impeller of a single-stage machine, through an open inspection port or removed cover, using the disc's standard threaded holes.
- The half-coupling and its balancing bolts. After a coupling repair, the source of a high running-speed component is often located right here.
- The free shaft end and the standard balancing ring, if the design provides for them.
- The drive motor's rotor. A separate machine with its own access and its own planes, and balancing it sometimes solves the whole problem on its own.
- The pulley and flywheel on belt-driven units.
- The oil-cooler fan and the motor cooling fan. A small thing that reliably ruins the numbers across the whole frame.

> Balancing is the last step, not the first. First the fasteners and soft foot (the machine's foot not sitting flush against the frame), then shaft alignment, then a repeat measurement. Shaft misalignment is the cause, shaft alignment is the fix, and this order can't be changed.

Sources: [Balanset-1A manufacturer specification](https://vibromera.eu/product/balanset-1/)

## Where we stop and say so out loud

A significant share of centrifugal compressor rotors are balanced not on site but under specialised conditions, on balancing equipment and to the manufacturer's procedure, often including a spin test up to operating speed. This isn't us being cautious — it's physics and regulation. We flag this before the visit, not after.

On site, in these cases, we're useful in a different way. A measurement with amplitude, phase, spectrum, and coast-down answers the main question: is this imbalance at all. It often turns out that it isn't, and in that case a removed rotor would have gone to the shop for nothing.

- A multistage rotor in a horizontally split casing, and any rotor in a barrel-type casing: correction planes are inside the flow path.
- High-speed pinion shafts on an integrally geared machine.
- A flexible rotor above the first critical speed: modal balancing at several speeds is needed.
- The manufacturer prohibits intervention, and there are no standard balancing points on the rotor.
- The rotor is damaged: a bent shaft, a crack, severe impeller erosion, or a destroyed seal. Repair comes first.
- The machine can't be brought to a stable operating regime for a verification run, or the speed drifts.
- The cause is process-related: surge, rotating stall, blade-pass pulsations, operation far from the design point.
- An explosion-hazard zone or a hot environment where safe access to the plane during operation isn't possible.

> What can be resolved on site versus only in the shop is something we covered in a separate article on choosing between on-site and shop balancing. For compressor machines, the dividing line runs through plane access and rotor behaviour, not machine size.

## Report, price, and preparation

We keep condition assessment and balancing result separate. The software says "within tolerance" when the residual running-speed component is below the set target value. The unit's condition is assessed against the applicable part of ISO 20816, with the edition, points, frequency band, regime, and bearing type all on record. The quality of the rotor balancing is assessed by residual imbalance and balance quality grade under ISO 21940. These are three different tolerances, and we don't mix them.

Vibration diagnostics with a report costs 300 EUR per unit, balancing adds from 250 EUR, and the minimum invoice per visit is 500 EUR. The final figure depends on the number of rotors and planes, how remote the site is, and the time needed for access. The calculator on the site gives an exact figure, and we confirm it after a short conversation. Base in Vila Nova de Gaia near Porto, we travel across all of Portugal. The work is done by engineers who design and manufacture Balanset instruments and do the on-site balancing themselves.

- [x] Machine type: single-stage overhung, multistage with a split casing, barrel-type casing, integrally geared.
- [x] Speed of each shaft and the speed increaser's gear ratio.
- [x] Number of blades on the impellers.
- [x] Type of bearing supports, and whether standard proximity probes with setpoints are fitted.
- [x] Current vibration figures, measurement points and directions, history over recent months.
- [x] Manufacturer's requirements on intervening in the rotor and on weight-mounting locations.
- [x] Operating regime: flow rate, suction and discharge pressure, anti-surge valve position, gas temperature.
- [x] Photos of the machine showing the bearing supports, coupling, inspection port, and driver.
- [x] The window in which the machine can be shut down, and the site address.

> If the data and photos show that the job can't be solved on site, we'll say so in writing beforehand. An unnecessary visit isn't worthwhile for either side, and a trip for a single measurement is justified only when that measurement genuinely settles the question.

Sources: [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) · [ISO 21940-11:2016](https://www.iso.org/standard/54074.html)

## Frequently asked questions

**Can a multistage centrifugal compressor rotor be balanced on site?**

Practically no. The impellers sit inside the casing and are shrink-fitted, the correction planes lie within the flow path, and on a barrel-casing machine there's no access at all. Rotors like this are balanced under specialised conditions to the manufacturer's procedure. On site, we measure, locate the critical speeds on coast-down, separate out imbalance, oil whirl, shaft misalignment, and process causes, and balance what's accessible from outside: the half-coupling, the free shaft end, the drive motor's rotor.

**We have a machine with a speed increaser, and vibration on the high-speed shaft. What can be done?**

A high-speed pinion shaft isn't balanced on site: speeds run into the tens of thousands of rpm, the tolerance is tight, and there's no access to the impeller. We do something else instead. We fit a mark and a phase sensor to the specific shaft, tie every line in the spectrum to its rotation frequency, and separate the gear-mesh frequency with its sidebands from the running-speed component and from blade-pass frequency. From the results, you learn whether it's imbalance, gear wear, or a bearing. We go ahead and balance the low-speed line and the driver at the same time.

**There's a subsynchronous peak at around 0.45 of rotation speed in the spectrum. Is this imbalance?**

No. Imbalance lives exactly at the rotation frequency. A peak at roughly 0.4 to 0.5 of rotation speed on a machine with journal bearings is oil whirl, an instability of the oil film. The causes lie in clearance, oil viscosity and temperature, supply pressure, and bearing load. Balancing doesn't remove it, and on top of that it masks bearing degradation. Check separately whether the peak tracks speed. If it has locked near the first critical speed and is growing, that's already oil whip, and the machine needs to be shut down.

**Vibration increases when we reduce flow. Will balancing help?**

Almost certainly not. Dependence on the operating regime is a sign of a process cause. Closer to the stability limit, rotating stall appears in the flow path and the diffuser, and beyond that, surge, with flow reversal and pressure oscillations throughout the system. What works here is the operating point, anti-surge control, and the piping. Imbalance is indifferent to the valve: it produces the same running-speed component and the same phase at any regime.

**How many shutdowns are needed and how long will it take?**

One correction plane requires a baseline run and one trial run; two planes require a baseline run and two trial runs, plus a verification run. Between runs, the machine is shut down, depressurised, and access to the plane is opened so the weight can be repositioned. A diagnostic visit without balancing fits into a single run with coast-down. The real schedule is set by your shutdown procedure and how long it takes to open the port, not by the measurement itself, so we agree the window in advance.

**What level do you bring vibration down to, and how is that confirmed?**

We set the target value for the running-speed component before work begins and record it in the report. The machine's overall condition is assessed against the applicable part of ISO 20816, with a fixed edition, points, frequency band, and regime, while the quality of the rotor balancing is assessed by residual imbalance and balance quality grade under ISO 21940. Promising a specific number in advance would be dishonest: it's limited by bearing condition, access to the planes, proximity to the critical speed, and the contribution of process causes. We tell you what's achievable after the baseline measurement and coast-down.