# On-site balancing of high-speed spindles and tool assemblies, where they operate

> At tens of thousands of rpm, an imbalance that nobody would notice on a fan turns into a force of tens of newtons and waviness on the workpiece. At the same time, the spindle itself can't be touched: the housing, the high-precision bearings, and their preload are the manufacturer's territory. We come out, measure vibration at the spindle head's bearing supports, and balance what can actually be removed and refitted: the tool holder, the collet, the chuck, the assembled tool. And we tell you right away where your problem ends with balancing and where it turns into a repair of the unit.

**In short:** Yes, but what we balance isn't the spindle unit itself — it's what turns on it: the tool assembly — the tool holder, collet, or chuck together with the cutting tool — and the spindle's standard balancing rings, if the factory built them in. We mount two accelerometers (vibration sensors) on the spindle head housing, radially at the front and rear bearing, and aim the laser phase sensor at a reflective mark on the spindle or on the tool holder's flange. We run the full cycle with an overall-level assessment up to roughly 12,000 rpm: the instrument computes vibration velocity RMS (the root-mean-square vibration level) across the 5–200 Hz band, and 200 Hz works out to exactly 12,000 rpm. Above that limit, we have two honest options. First: balance the rigid assembly at a reduced, stable rotation speed, then verify the result at operating speed by the running-speed component (vibration at the rotation frequency, denoted 1x) and the spectrum. Second: send the tool holder to a specialised machine built for tool assemblies. We never intervene in the spindle housing, the bearings, or their preload: that's the manufacturer's service work, not the balancer's.

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

## Why everything counts differently at tens of thousands of rpm

Centrifugal force from residual imbalance grows with the square of rotation frequency. Take a 3 kg assembly with a residual imbalance of 3.6 g·mm. At 8,000 rpm that's about 2.5 N — the machine simply won't notice it. At 20,000 rpm the very same assembly produces about 16 N of rotating force, applied to the front bearing and to the cutting edge. Nothing has changed except the speed.

A second factor comes into play after that. High-precision bearings run with a set preload, and it has a life of its own: it changes as the machine warms up, and it depends on the operating regime and on how long the spindle has been running. Until the preload has stabilised, the system's response to a trial mass keeps drifting, and the influence coefficient method — calculating correction weights from the machine's response to a trial mass — stops working. Hence the firm rule for a visit: we warm the spindle up at operating speed and balance only once the readings are stable.

- [x] Tool life has dropped under the same regimes as before, the edge chips instead of wearing evenly
- [x] Regular waviness has appeared on finishing operations, at a pitch equal to the feed per spindle revolution
- [x] The vibration level shown by the CNC system has risen at higher speeds, while it's unchanged at lower ones
- [x] Noise and vibration change with the tool: one assembly is quiet, another is noticeably worse
- [x] The picture changed after replacing the collet, the nut, the pull stud, or the extension
- [x] The front bearing runs hotter than usual, and the noise has changed in tone, not just in loudness

> If the level jumps within a single run, that's usually not imbalance but a lost insert, a shifted mass, or a clamp that's let go. Balancing would mask that, not fix it.

## What's actually rotating: the spindle, the tool holder, and the tool are one rotor

The spindle shaft itself is balanced at the factory and doesn't go out of balance in normal use. On a high-speed spindle, imbalance comes from outside, along with the tooling. The system being balanced is the spindle taper plus the tool holder plus the clamping unit plus the tool, and what needs balancing is exactly that assembly, in the angular position it actually works in.

### The taper and the type of fit

HSK clamps on both the taper and the face, SK and BT clamp on the taper alone, and there are polygonal shanks too. SK and BT have a separate source of imbalance — the pull stud: it's asymmetric by itself and is often not the same one from assembly to assembly. On HSK, cleanliness of the face and the condition of the clamping segments are critical: incomplete face contact produces runout that weights won't remove.

### The collet chuck

The nut with its wrench slots, the collet itself with its longitudinal slits, the mounting seat. Every part is asymmetric, and the result depends on what angular position you tightened the nut in. Hence the mandatory marking: the nut, collet, tool holder, and spindle go together mark-to-mark, or the balancing lasts only until the first strip-down.

### Shrink-fit and hydraulic holders

More symmetrical by design than a collet, and they usually give a smaller initial imbalance. But a shrink-fit holder is sensitive to the cleanliness of the bore and to how deep the tool is seated, while a hydraulic holder is sensitive to leftover lubricant in the chamber. Set tool overhang against a stop, not by eye: changing the overhang changes both the imbalance and the stiffness.

### The tool itself

A single-insert mill, a boring head with an adjustable bit, a stepped tool, an extension. These are asymmetric bodies by design, and their imbalance isn't going anywhere. Assemblies like these must be balanced, and always together with the specific tool holder they'll actually run on.

### Test-rig spindles

Spin-test and drive shafts on test rigs, spindle heads on run-in rigs and balancing machines. These are usually a rigid rotor (one that doesn't bend at operating speed) of moderate length, with standard correction planes at the end faces. Our format suits them best, provided speed can be held stable.

> The rule is simple: it's the assembly that gets balanced, not the part. Take it apart and reassemble it without marking, and you've lost the result.

## Where the line falls between balancing the tool and repairing the spindle

This is the main question we're called out for most often. Balancing only reduces the running-speed component of vibration — the part that repeats exactly once per revolution. Everything else isn't fixed with weights. The first diagnostic run exists precisely to separate these two zones before anyone pays for weights.

| What's visible on the machine | What the instrument shows | Whose territory this is |
| --- | --- | --- |
| The level steps up when the tool holder is fitted, and is low on the empty spindle | 1x dominates, phase repeats from run to run within a few degrees | Balancing the assembly, done on site |
| Different assemblies produce different levels, and the worst ones are always the same ones | 1x changes along with the tooling, the rest of the spectrum is clean | Balancing and screening the sets, done on site |
| The level is already high on the warmed-up, empty spindle | 1x is present, but not tied to the tooling | An internal cause: rotor fit, traces of past repair. Manufacturer's service |
| Noise at every regime, the front bearing runs hot | Peaks at high frequencies not related to running speed as multiples, rising acceleration levels | Spindle bearings. Repair the unit, balance afterwards |
| Taper and face runout is visible with a dial indicator on the stopped spindle | 1x plus poor phase repeatability | Fit geometry: cleaning, lapping, taper repair |
| Amplitude and phase drift as the machine warms up at constant speed | Drift with no change in regime | Thermal state and preload. Don't balance |
| A sharp rise over a narrow speed range, falling off above it | A narrow peak, phase flips by roughly 180° | Resonance in the spindle head, cover, or bed. Structure and mounting |

> We don't get into the spindle housing. We don't strip the unit down, adjust the preload, replace the bearings, or drill anything in either the taper or the housing. If we find a worn-out bearing, we hand over the numbers, the spectrum, and a report, and push the balancing to after the repair. Residual imbalance is an added rotating load on the bearings, and a bearing-life calculation to ISO 281 won't account for it unless you build it in. How to read a bearing's spectrum is covered in a separate article on bearing diagnostics from vibration.

Sources: [ISO 13373-3:2015](https://www.iso.org/standard/40840.html) · [ISO 281:2007](https://www.iso.org/standard/38102.html)

## What we do on site, and what we don't

The line isn't drawn by preference, but by the measurement band and how stable the regime is. The instrument computes vibration velocity RMS across the 5–200 Hz band, and 200 Hz corresponds to 12,000 rpm. Up to that figure, we run the full cycle: diagnostics, trial runs, correction, verification, and an overall-level assessment.

Above 12,000 rpm, rigid-rotor physics takes over: residual imbalance is a property of the assembly, not of the operating regime. So it can be balanced at a reduced, stable rotation speed, where the response is linear and repeats. We then check the running-speed component and the spectrum at operating speed. This works on spindles with a variable-frequency drive, where intermediate speeds can be set and held. If that's not possible, there's only one honest answer: the tool holder is balanced on a specialised machine for tool assemblies, and we do diagnostics of the unit and a report.

- We take on: tool assemblies on machining centres and milling machines where speed is controllable
- We take on: spindle heads and drive shafts on test rigs with a rigid rotor
- We take on: one-off diagnostics, when you need to know whether it's imbalance, bearings, or resonance
- We don't take on: stripping down the spindle, adjusting preload, taper repair
- We don't take on: certifying tooling to a G grade for acceptance — that's a job for a balancing machine with its own measurement setup
- We don't take on: a spindle whose speed can't be held stable — influence coefficients won't reproduce there
- We don't take on: a machine in resonance at operating speed — the result won't hold

> The choice between on-site work and a shop machine is covered in a separate article. If the on-site format doesn't suit your job, we say so before the invoice, not after.

## How the visit goes

1. **Agreement before the visit** — Machine and spindle type, operating speed and adjustment range, taper type, what's currently mounted on the spindle, what defect is visible on the workpiece. This is where we decide whether we're going for the full cycle or for diagnostics.
2. **Measurement points** — We clean off mounting spots and fit two accelerometers on the spindle head housing, radially, at the front and rear bearing, on bare metal, not on a cover or a panel. We keep the direction the same from measurement to measurement.
3. **Phase mark** — We attach a reflective mark to the spindle or to the tool holder's flange — exactly one — and check speed against the instrument's reading. Using the mark, the instrument ties the vibration to the shaft's angle of rotation — that's the phase. On a belt drive, the mark goes on whichever shaft we're balancing, or it's easy to mix up the spindle's running-speed component with the motor's in the spectrum.
4. **Warm-up and baseline measurement** — We warm up at operating speed, usually for twenty minutes or more, until the readings stabilise. We record the overall level, the 1x amplitude and phase at both bearings, the spectrum, and the time waveform. Two runs in a row show whether the phase repeats.
5. **Correction planes** — The machine is stopped and locked out. We look for where mass can physically go: the spindle's balancing rings, the tool holder's adjustment screws or sockets, end-face holes. We number the positions in the direction of rotation and measure the actual radius.
6. **Trial weight** — We fit a weighed mass and mount it just as securely as the permanent one will be. The run is valid if 1x changes by 20–30% in amplitude or 20–30° in phase. On a rigid, high-speed assembly, a fraction of a gram is enough to achieve this, so accurate scales matter.
7. **Calculating and fitting the correction** — The software outputs the mass and angle for each plane, and in fixed-position mode, the socket or screw number directly. Where there's nowhere to add mass, it calculates drilling instead. The angle's zero reference is always wherever the trial weight was fitted.
8. **Verification, marking, coefficients** — A verification run at the same regime and the same points, with one more iteration with a small add-on weight if needed. We also check at two more of the machine's operating regimes. We mark the angular position of the tool holder, collet, and nut, and save the unit's influence coefficients.

> We work with the Balanset-1A, a portable two-channel vibration analyser and balancer: two accelerometers, a laser phase sensor keyed to the reflective mark, a two-channel USB module, and software on a laptop. Balancing in one and two planes using the influence coefficient method, overall vibration and 1x with phase, speed, spectrum, time waveform, fixed positions, drilling calculation, tolerance by G grades, archiving and reports. With the saved coefficients, the next touch-up balance goes through without trial runs — there's a separate article on trim balancing about that.

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

## Where to put the mass, and how to mount it

On a high-speed spindle, the choice of correction points leaves almost no freedom. Anything the design doesn't provide for is out of the question immediately: at these speeds, centrifugal force tears off anything that isn't held by a thread or by its own shape.

### Standard balancing rings

Many spindles have two rings with weights that rotate around the circumference, or stacked weight segments in an annular groove. This is a legitimate correction plane, engineered by the manufacturer for operating speed. We use these first.

### The tool holder's sockets and screws

Tool holders often have threaded holes for balancing screws, or annular grooves. Mass is fitted at a known radius, torqued to the instructions, with thread locker if needed.

### Metal removal by drilling

Where there's nothing to add mass to, the software calculates the hole diameter and depth at a given radius. Drilling is done only in the body of the tool holder, and only in the zone the manufacturer permits. We never touch the taper, the mounting face, or the spindle housing, under any circumstances.

### The automatic balancing device

Grinding spindles sometimes have a standard device that compensates for drift in the wheel's imbalance, within its own range. It doesn't replace the initial balancing: if the device has run to its limit, the imbalance has exceeded that range, and it's removed with weights first.

### What never belongs on a spindle

A welded plate, modelling clay, a magnetic weight, a washer under a random bolt. At tens of thousands of rpm, that's a projectile. We mount the trial weight exactly as securely as the permanent one, or the very first run will be the last one.

> We have separate articles on methods for mounting correction masses and on safety during on-site work. On the visit, we agree the machine lockout procedure and the setup mode with the person responsible for the equipment before the first run.

## Tolerance: why it comes down to milligrams

Balance quality grades under ISO 21940-11 set the specific residual imbalance, calculated as the G grade multiplied by 9549 and divided by speed. The higher the speed, the smaller the allowed residual, and at tens of thousands of rpm it drops to values that ordinary scales can't catch. Below is an example for a 3 kg tool assembly with a 25 mm correction radius.

- The target residual 1x in the instrument's software is a number in mm/s that you set yourself. It's an internal procedure criterion, and it doesn't replace the other two
- Overall vibration in mm/s RMS (RMS is the same as the root-mean-square level) assesses the machine's condition. Record the applicable part and edition of ISO 20816, the measurement points, and the regime separately — there are qualifications there for machine tools
- Residual imbalance in g·mm/kg by G grade is already a characteristic of the rotor, not of the machine, and is confirmed by its own measurement setup

| Speed | Grade | Specific residual, g·mm/kg | Allowable mass at a 25 mm radius |
| --- | --- | --- | --- |
| 12,000 rpm | G 2.5 | 1.99 | about 0.24 g |
| 20,000 rpm | G 2.5 | 1.19 | about 0.14 g |
| 20,000 rpm | G 1 | 0.48 | about 57 mg |
| 24,000 rpm | G 1 | 0.40 | about 48 mg |
| 30,000 rpm | G 0.4 | 0.13 | about 15 mg |

> For tool assemblies, the tolerance is often set not as a G grade but as a maximum residual imbalance in g·mm at the tool holder's stated maximum rotation speed. Pin down exactly which one applies to you in the work order, along with the part and edition of the standard. How the G grade is chosen is covered in a separate article. There's just one practical takeaway: masses are weighed on accurate scales, every part of the assembly goes back in the same angular position, and dirt in the taper devalues any figure.

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

## What residual imbalance actually costs you on this kind of machine

On a slow-running rotor, imbalance means noise and wear. On a high-speed spindle, it first ruins the workpiece and eats through the tool, and only after that does it get to the bearings.

- The tool's cutting edge takes a varying load once per revolution. Instead of even wear, you get chipping, and tool life drops sooner than expected
- Regular waviness appears on the surface, at a pitch equal to the feed per spindle revolution. This is what tells imbalance apart from chatter (self-excited vibration during cutting), whose pitch has nothing to do with speed
- Dimensions stop holding on finishing passes, forcing you to add passes and cut the feed rate
- The front bearing takes on a constant rotating load that isn't in the original life calculation
- The preload works under a heavier load, the unit runs hotter, and its geometry drifts faster
- The taper and the clamping unit take on an alternating load, the fit wears, runout grows, and the cycle closes in on itself

> The reverse also holds: an assembly balanced to a tight grade removes this load entirely. But only the part of it that comes from mass. Chatter, taper wear, and a bearing defect stay exactly where they were.

## Price, what you get, and how to order

You get the engineers who design and manufacture Balanset instruments coming out to you, doing the on-site balancing themselves. Base in Vila Nova de Gaia near Porto, we travel across all of Portugal. There's also the Balanset-1A OEM version without the case, if you want to build the measuring chain into your own rig or machine and balance tooling in-house.

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 assemblies, the number of correction planes, the time needed for warm-up and runs, and how far away the site is. The calculator on the site gives an exact figure.

- Before-and-after measurement at the same regime: speed, overall level in mm/s RMS, 1x amplitude and phase at both bearings
- Spectrum and time waveform, showing what dropped and what remained
- Masses, angular positions or position numbers, and the actual radii of the fitted weights
- A mechanical assessment: runout, bearing condition from the spectrum, fasteners, signs of resonance
- Saved influence coefficients for the unit, and a marking diagram on the equipment itself

- [x] Machine type, spindle model, operating speed and adjustment range
- [x] Taper and clamping type: HSK, SK, BT, collet chuck, shrink-fit holder
- [x] What's currently mounted on the spindle, and which assemblies are raising questions
- [x] What defect you're seeing on the workpiece, and from what point it appeared
- [x] Whether there are standard balancing rings or an automatic balancing device
- [x] Current vibration figures, if you have monitoring, and photos of the unit and the correction zone

> If the first run shows that the level isn't coming from mass, we won't start balancing. You'll get the measurements, the spectrum, and a report to take to the manufacturer's service department.

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

## Frequently asked questions

**We have a motorized spindle rated at 24,000 rpm. Can you do anything on site?**

The running-speed component on a spindle like that sits at 400 Hz — outside the 5–200 Hz band the instrument uses to compute vibration velocity RMS. So we don't run the classic full cycle with an overall-level assessment at operating speed. Two workarounds apply. First: we balance the assembly at a reduced, stable frequency where the response repeats — for a rigid rotor, residual imbalance doesn't depend on speed. We then check the running-speed component and the spectrum at operating speed. Second: the tool holder with the tool is balanced on a specialised machine for tool assemblies, and on site we do diagnostics of the unit and a vibration assessment. Which one applies to you depends on the drive type and on whether intermediate speeds can be set and held.

**Should you balance the spindle or the tool holder?**

Start with the tool holder and the whole tooling assembly. Run a simple test: measure vibration on the warmed-up, empty spindle, then with the empty tool holder, then with the tool holder and the tool fitted. If the level steps up each time tooling is added, it's a balancing job and gets solved in a single visit. If the level is already high on the empty spindle, balancing the tool holder won't remove it, and the conversation turns to the rotor fit, the taper, and the bearings.

**Does every tool holder in the magazine need balancing?**

No. Balance the assemblies that actually run at the higher speeds and on finishing operations, and the ones where the tool is asymmetric by design: boring heads, single-insert mills, long extensions. Roughing assemblies at moderate speeds usually don't need it. It's more sensible to pick out a dozen critical sets, balance them, and mark the angular position of every part, than to chase the entire magazine.

**There's an automatic balancing device on the grinding spindle. What do you add, then?**

The standard device compensates for slow drift in the wheel's imbalance: dressing, wear, absorbed coolant. It works within its own compensation range and doesn't replace the assembly's initial balancing. If the device has run to its limit, the imbalance has exceeded that range, and it needs to be removed with weights first. It also won't do anything about taper runout, a bearing defect, or resonance in the spindle head. We measure what's left after it's done its job, and tell you exactly what's causing the residual.

**Vibration rises as the machine warms up and doesn't come back down. Is this imbalance?**

Probably not. Drift in the amplitude and phase of the running-speed component at constant speed and rising temperature points to the unit's thermal behaviour: a change in preload, thermal bow, rubbing. Balancing in a state like that would lock in one point in the warm-up curve and make another one worse. We take a series of measurements as the machine warms up and decide based on that. We balance only once the amplitude and phase are stable on the warmed-up spindle.

**What balance quality grade should be specified for a tool assembly?**

The grade is set by the customer commissioning the work, not by the contractor. A practical benchmark under ISO 21940-11 is G 1 for machine tool spindles, G 0.4 for precision grinding spindles, and G 2.5 for less critical tooling. Bear in mind that for tool assemblies the tolerance is often specified differently: as a maximum residual imbalance in g·mm at the tool holder's stated maximum rotation speed. Record both the value and the applicable part and edition of the standard in the work order, or acceptance will turn into an argument.