Vibration Limits per ISO 20816: Zones A, B, C, D and How to Use Them

The report reads "4.2 mm/s." On its own, that number means neither "good" nor "bad": four things give it meaning, and they almost always go unrecorded. Let's break down exactly what ISO 20816 assesses (ISO 10816 used to play that role), what's behind zones A, B, C, and D, why tables found online contradict each other, and where they get mixed up most often.

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

What ISO 20816 Assesses, and in What Units

The standard answers one question: is a machine's vibration, measured on its non-rotating parts, acceptable. Non-rotating parts means bearing housings, bearing pedestals, covers, and the bed. Not the shaft. Shaft vibration is picked up with non-contact displacement probes and assessed against different documents, usually for turbomachinery on plain bearings.

The ISO 20816 series consists of several parts. One is general, one covers industrial machines measured in situ, and the rest cover large steam and gas turbines, hydro units, reciprocating compressors, gearboxes. A plant mechanic almost always needs just the first two. Check the series' composition and part editions before citing them in a document.

This kind of assessment fixes five things in place, and you can't change them to suit yourself.

• The quantity: root mean square (RMS) of vibration velocity, mm/s. Not peak, not peak-to-peak, not acceleration.
• Frequency band: 10–1000 Hz for general-purpose machines.
• Point: the bearing housing or pedestal. Sensor on a rigid mount — stud or magnet on a clean, flat pad.
• Direction: radial, horizontal and vertical, and axial where specified. The direction stays the same from one reading to the next.
• Condition: steady-state running speed and operating load.

The word "broadband" is the key one here. A single number lumps everything together at once: unbalance, misalignment, loose fasteners, gear-mesh frequencies, hydraulics. The standard doesn't ask where the vibration comes from — it answers whether there's a lot of it or a little. The spectrum (the breakdown of vibration by frequency) shows the cause, and that's the subject of a separate article on reading a spectrum. An ISO 20816 assessment also doesn't confirm the quality of a rotor's balancing: we covered the difference between the three separate tolerances in the article on the three tolerances in balancing.

Sources: ISO 20816-1:2016

Zones A, B, C, D: What Each One Means for Your Decision

The standard deliberately avoids a single pass/fail line. It splits the scale into four zones, and each zone isn't a rating — it's a prescribed action.

Zone boundaries aren't a cliff edge. 4.4 and 4.6 mm/s against a 4.5 boundary mean practically the same machine condition, especially since sensor mounting quality alone easily accounts for a 10–15% spread. A zone exists to help you pick an action, not to argue over tenths of a unit.

Zone Table by Machine Classes I–IV

This is the classic table every mechanic knows by heart: it gets printed into procedures and pinned to the wall in the reliability department. Classes are split by power, and for large machines, by foundation type as well.

These numbers are given in mm/s RMS for the 10–1000 Hz band and come from the previously used ISO 10816 classification. Use them as a working guideline for your own condition monitoring. Before contractual acceptance, open the applicable part of ISO 20816 in its current edition and check its scope: ISO 20816-3 addresses industrial machines above 15 kW at rotation speeds of 120–30,000 rpm, and it contains exceptions by machine type. For small fans, screw conveyors, mulchers, and belt-driven units, check applicability separately. For reciprocating machines the general table doesn't work at all: the harmonics normal for their back-and-forth motion would push a healthy unit into zone C or D, and separate documents exist for machines like these.

Sources: ISO 20816-1:2016

Rigid or Flexible Foundation: The Row People Mix Up

The last two rows of the table differ by one word, yet the numbers are twice apart. This is exactly where mistakes happen constantly, and the mistake is costly either way: you either rework a perfectly fine machine, or you sign off as normal a level that's already harming the bearings.

Here's the physics. A flexible foundation deforms on its own at the rotation frequency and soaks up part of the vibration energy. The bearing housing moves more noticeably as a result, but the dynamic load on the bearing itself ends up lower. A rigid foundation behaves the opposite way: the housing barely moves, and the full force goes straight into the bearing and the foundation. Hence a conclusion that seems counterintuitive: the limit for a flexible foundation is higher, not lower.

• Compare the rotation frequency with the system's first natural frequency in the measurement direction: if the natural frequency is above the running speed, the foundation is rigid; if below, it's flexible.
• Take three readings in a row: the bearing housing, the frame nearby, the foundation under the frame. If the frame vibrates comparably to the housing while the foundation barely does, the foundation is behaving as flexible.
• Look at what the machine physically sits on: springs, rubber isolators, and a welded frame on four points are almost always a flexible arrangement.
• Check anchor-bolt torque and soft foot before assigning a foundation type.
• Record the foundation type you settled on in the report along with the reasoning. That's half of any future dispute with the customer, settled in advance.

If a table you've found sets a stricter limit for a flexible foundation than for a rigid one, the table is wrong. Don't sign a report against it, and don't put it into a spec. And separately: loose fasteners or a cracked frame turn a nominally rigid foundation into a flexible one, but the tolerance doesn't loosen because of that. That's a defect, not a design choice.

Why Tables Found Online Don't Match

You find two tables of "ISO limits," and the numbers in them differ. Neither is made up. The fact is, two different classification schemes exist, and both circulate online under the same name.

The first scheme is classes I–IV from the older general classification: power first, then foundation type for large machines. That's the one given above, and it's the one that shows up in plant procedures most often.

The second scheme comes from the part of the standard covering industrial machines measured in situ. There, machines are split into groups by power and shaft-center height, and within each group, boundaries are set separately for rigid and for flexible supports. The numbers come out different.

Boundaries in mm/s RMS, 10–1000 Hz band, given as a working guideline. Notice the pattern: here too, flexible supports get the higher numbers. The practical takeaway is simple. For internal condition monitoring, pick one scheme and don't change it for years, or the trend becomes incomparable. For contractual acceptance, use whichever table is printed in the applicable part and edition of the standard, and copy its row into the report verbatim.

Sources: ISO 20816-1:2016

The 10–1000 Hz Band: Three Places Where the Overall Level Goes Blind

The band isn't a formality tucked into the header of a report. It directly determines what the instrument counts and what it throws away. Three consequences are worth keeping in mind at all times.

Being upfront about our own instrument's band. The Balanset-1A measures vibration velocity RMS in the 5–200 Hz band, range 0.02–80 mm/s, speed 100–100,000 min⁻¹, phase measurement error ±1°. For balancing that's exactly what's needed: the running-speed component and lower harmonics fall inside the band, and for slow-turning machines the 5 Hz lower limit catches 1x better than the standard 10 Hz. But this band doesn't match 10–1000 Hz, and the overall level doesn't account for the high-frequency portion. If a report is going to acceptance, state the instrument's actual band right next to the number. That way the figure can be correctly compared against someone else's.

Why Trend Matters More Than the Absolute Number

Two machines. The first held 1.2 mm/s for six months, then climbed to 2.6 over the last month. The second has sat at 5.5 mm/s, unchanged, for five years. Formally, the first is in zone B and the second in zone C. It's the first one you need to deal with.

The reason is that an absolute number compares your machine against a generic fleet of machines in its class, while a trend compares the machine against itself. The second comparison is always more precise: it already accounts for the design, the installation, and the quirks of this particular foundation. That's why the standard gives two assessment criteria, not one: the magnitude of broadband vibration and its change.

Working guidelines for setpoints — the thresholds a warning and a protective trip are configured against. The alarm level is typically set as the baseline plus about 25% of zone B's upper boundary, and usually no higher than 1.25 times zone B's upper boundary. The trip level sits in zone C or D and usually doesn't exceed 1.25 times zone C's upper boundary. Check the specific rules and coefficients against the current edition of the applicable part.

A change of about 25% of zone B's upper boundary is considered significant even when the machine stays within its own zone. And a drop in level is also a signal, not a reason to relax: that's what a correction weight coming loose looks like, or a broken blade, a sensor magnet that's come unstuck, or a changed operating condition.

• A baseline only makes sense with an unchanged point, direction, band, and condition. Change any one of them, start the trend over, and note it in the archive.
• Three directions at every bearing housing: horizontal, vertical, axial. A reading from the wrong point produces a false change in the trend, which means either a false alarm or a missed defect.
• Check a wild value on the spot: 250 mm/s where you'd normally see 3 mm/s is more often a broken cable or a magnet not making contact than an actual failure. Remeasure — only record it if the value repeats.
• Don't raise the setpoint just to make the instrument stop complaining. That's a guaranteed way to miss a real defect.

Tie the measurement interval to how fast a defect develops. Unbalance develops over weeks and months, rolling-bearing defects usually over one to six months, gear meshing over three to twelve. From this comes a practical rule: the interval between readings should be no more than half the expected time from the first sign to failure.

What to Record in the Report for Contractual Acceptance

A dispute over vibration almost always turns out to be a dispute not about the numbers, but about what exactly was measured. Settle this in advance, before the site visit. One agreed-upon sheet saves a week of back-and-forth correspondence.

• Standard part and edition: not "per ISO" — a specific part with a year.
• Table and row: machine class or group, power, support type, with the reasoning behind the choice.
• Measurement points: each bearing housing listed separately, with a sketch or photo and unambiguous labels.
• Directions: horizontal, vertical, axial, and the same from one reading to the next.
• Parameter and band: vibration velocity RMS in mm/s, the instrument's actual band.
• Sensor mounting method: stud, magnet, prepared pad.
• Condition: speed, load, temperature, process state.
• Target zone: A or B, and what each party does if the reading lands in C.
• Before-work and after-work values, taken at the same points.
• As a separate line item, the residual 1x running-speed component (vibration at the rotor's rotation frequency) and residual unbalance, if the contract requires a balance quality grade.

Matching "before" and "after" points matters more than instrument accuracy. Shift the sensor by 100 mm, or take the reading from the other side of the housing, and the work's result becomes impossible to prove.

You've Landed in Zone C or D. What to Do

A high number doesn't tell you what to do by itself. It only tells you it's time to dig in. The sequence that saves money runs like this: measurement first, then a decision, then repair.

1. 01

   Verify the reading is trustworthy

   Sensor mounting, direction, band, machine condition. Compare against the history for this same point. Half of all "critical" levels disappear at this step.

2. 02

   Break the level down into its components

   Compare overall vibration against the 1x running-speed component and look at the spectrum. If 1x dominates, it's unbalance. A noticeable 2x (vibration at twice the rotation frequency) combined with elevated axial vibration more often means misalignment, and shaft alignment is the cure. A comb of harmonics with a raised noise floor points to looseness and rubbing. Peaks higher up in frequency that aren't multiples of the running speed usually point to bearings.

3. 03

   Rule out resonance and mechanical issues

   Anchor-bolt torque, soft foot, play, the condition of the frame and foundation. Check whether the running speed coincides with the structure's natural frequency. Balancing gives an unstable result in resonance — there's a separate article on that.

4. 04

   Balance on site if 1x dominates

   A run for baseline vibration, a trial weight in the first correction plane (a rotor cross-section where weights are mounted), a trial weight in the second plane if there are two, calculating the correction, and a verification run. The trial weight must change the 1x amplitude by 20–30% or the phase by 20–30°, or the influence coefficient (the machine's measured response to a trial weight, from which the instrument calculates correction weights) will come out unreliable.

5. 05

   Record a new baseline

   The same points, the same direction, the same band, the same condition. This is the starting point for the next trend, and without it the whole job turns into a one-off service call.

On-site balancing moves a machine into zone A or B when the running-speed component accounts for most of the level. If 1x accounts for less than half, weights will only shave a few percent off the overall level. We've covered the cases where balancing is powerless separately.

If There's No Time to Sort Out Zones and Supports

An ISO 20816 assessment looks like plain arithmetic right up until you have to decide whether the foundation is flexible, which part of the standard to apply, what to do about a slow-turning machine, and why the two tables you found give different numbers. From there it becomes engineering work, and it usually goes faster in hands that have done it before.

AXILINE does on-site dynamic rotor balancing in situ and vibration diagnostics. We arrive with a two-channel measurement system, record overall vibration and the running-speed component with phase at every bearing housing, show you the spectrum, separate unbalance from misalignment, looseness, and resonance, and balance in one or two planes in the machine's own bearing housings, with no rotor removal. What you get out of it is a report with points, band, condition, and before-and-after values, showing which zone the machine landed in and why.

A second option, if this kind of work is a regular thing for you: get the instrument for yourself. The Balanset-1A is a kit of two accelerometers, a laser phase sensor, a two-channel USB module, and Windows software. It measures speed, vibration velocity amplitude and phase, shows the spectrum and time waveform, calculates weights and tolerance against balance quality grades, works from saved influence coefficients, and keeps an archive for reports. The engineers who design and manufacture these instruments do their own on-site balancing with them, so the consulting support comes from practice, not from a brochure.

Tell us what machine you have, its power, its speed, what it sits on, and what numbers you got. We'll tell you whether it's worth a site visit, whether balancing will help, and what zone makes sense as a target.

Frequently asked questions

Related content