ISO 20816 Vibration Severity Explained (and Where the Chart Falls Short)

In one line: ISO 20816 is the current standard for judging overall machine vibration severity: a single RMS velocity number sorted into four zones, from newly-commissioned to damage-risk. It's genuinely useful as a fleet-wide health check, but it evaluates one broadband number against a fixed limit, which is exactly the blind spot that lets early bearing faults hide in plain sight.

The bearing that's still "Satisfactory," until it isn't

A cooling-water pump on your monthly route reads 3.1 mm/s RMS. You check it against the ISO chart taped inside the panel door: Zone B, acceptable for unrestricted long-term operation, no action required. You move to the next asset on the list.

What the number doesn't show: three weeks ago, this same pump read 1.9 mm/s. It's climbing about 0.1 mm/s a week, and almost all of that increase sits at one specific frequency: an outer-race bearing defect, visible in seconds to envelope analysis, invisible in a broadband trend nobody is charting. The pump stays comfortably inside Zone B for five more weeks. Then, over two shifts, it's in Zone D.

Nothing about the standard was wrong here. The chart did exactly what it's built to do: classify one broadband number against a fixed severity scale. The failure is asking that number to carry more diagnostic weight than it was ever designed to hold.

What ISO 20816 is, and what it replaced

ISO 20816 is the current international standard for measuring and evaluating machine vibration, the direct successor to two older series: ISO 10816, which covered vibration on the bearing housing (casing vibration), and ISO 7919, which covered vibration on the rotating shaft. ISO 20816 folds both into one framework: Part 1 formally supersedes ISO 10816-1 and ISO 7919-1, and the rest of the series has migrated machine-type by machine-type since. If your plant still cites "ISO 10816," you're citing the predecessor, not a wrong standard: the zone concept, the mm/s units, and most numeric limits carried over largely unchanged.

Like its predecessor, ISO 20816 is a multi-part standard: there's no single "the ISO 20816 number." Part 1 sets general principles; Part 3 covers general industrial machines above 15 kW, 120-30,000 rpm (most pumps, fans, and motors on a plant floor); Part 4 covers gas turbines above 3 MW; Part 8 covers reciprocating compressors; other parts run from hydraulic machines to wind turbines. Each sets its own criteria, because a 3 MW turbine and a 30 kW pump don't share a "normal" vibration level.

What every part measures is the same underlying quantity: broadband RMS velocity, in mm/s, over roughly the 10-1000 Hz range (the classic 1x/2x/3x shaft-speed faults our guide to vibration analysis fundamentals covers in more depth). It's a deliberately coarse, single-number summary of "how much is this machine shaking, overall": exactly its strength, and, as you'll see, exactly its limit.

The four zones: A, B, C, D

ISO 20816 sorts a machine's broadband RMS velocity into four zones:

Zone A: vibration typical of a newly commissioned or freshly overhauled machine, closer to "as good as this machine gets" than a level to expect for years.
Zone B: acceptable for unrestricted long-term operation. Most healthy machines live here.
Zone C: not suitable for continuous operation. Treat it as a planning trigger: investigate, schedule corrective work.
Zone D: vibration severe enough that damage is likely. Immediate action.

The specific mm/s values separating A from B, B from C, C from D are not universal: they're set per machine group and per support type, which is the first place informal use of "the ISO limit" goes wrong.

Why the same reading means something different on a different machine

Zone boundaries under ISO 20816-3 (the part covering general industrial machines) depend on two things: the machine's size/power class, and how rigidly it's mounted.

As one concrete, widely-cited example: for large, rigidly-mounted industrial machines, published boundaries run roughly 2.3 mm/s (A/B), 4.5 mm/s (B/C), and 7.1 mm/s (C/D) RMS velocity. That 4.5 mm/s is the same "ISO threshold" referenced elsewhere on this blog as a typical bearing-alarm setpoint. Worth being precise about: it's the boundary between "acceptable" and "plan an inspection," for one specific group of machines.

A smaller industrial machine on the same rigid foundation gets tighter limits: roughly 1.4 / 2.8 / 4.5 mm/s is a commonly cited example. Flexibly-mounted machines (soft foundations, elevated structures, some fan and compressor skids) get looser limits than a rigid-mounted machine of the same size, because a flexible structure absorbs and re-radiates vibration differently.

The practical implication: "4.5 mm/s" isn't a universal alarm setpoint. It's correct for one group under one mounting condition. Apply it to a smaller motor or a flexibly-mounted fan and you'll either false-alarm constantly or miss real degradation: the exact trap the standard's group-specific limits are meant to prevent.

Where the chart falls short

None of this is a flaw in ISO 20816; it does exactly what a severity standard should: a defensible, reproducible pass/fail line across an entire fleet. The gaps show up when the chart gets used as a diagnosis instead of a classification. Four matter in practice:

One number hides where the energy is. Broadband RMS sums everything in the 10-1000 Hz band together. An early-stage bearing defect (a small spall on the outer race) adds a narrow spike of energy at one specific defect frequency, barely moving the overall total. The machine can sit inside Zone B for weeks while that narrow-band energy climbs in a part of the spectrum the overall number can't resolve.

It isn't normalized for speed or load. Vibration amplitude scales roughly with the square of shaft speed. A machine on a VFD running 600 to 1,800 RPM doesn't have one "normal" reading: it has a different normal at every speed. Evaluate it against one fixed chart and you'll either false-alarm at high speed or stay blind at low, regardless of which boundary you pick.

It evaluates severity, not cause. Zone C says a machine needs attention. It doesn't say whether that's imbalance, misalignment, looseness, or a bearing defect: four failure modes with different urgency and different repair scopes. Getting from "Zone C" to "replace the outer bearing on the drive end" still requires spectral analysis and, usually, a trained analyst.

Absolute zones matter less than your own trend, and the standard says so. ISO 10816-1 builds this in as a second, independent evaluation criterion: alongside Criterion I (absolute magnitude, the zones above), Criterion II covers a significant change from an established baseline, meaningful on its own even if the reading never leaves its zone. A pump that has always run at 3.8 mm/s and one that jumped from 1.5 to 3.8 mm/s last month are both "Zone B" under Criterion I. Only Criterion II (the one most informal monitoring programs skip) flags that the second pump is telling you something urgent.

What closes the gap: a learned baseline instead of a fixed one

Everything the chart is missing (frequency resolution, speed/load normalization, and genuine trend-awareness) is what modern condition-monitoring software adds on top of the standard, not instead of it.

A per-asset learned baseline replaces "one chart for every machine in this group" with "what does normal look like for this specific pump, at this speed", comparing today's reading to what this machine has done before. That's Criterion II, automated and running continuously instead of applied by hand on a monthly route.

Frequency-resolved detection does what envelope analysis does by hand: a narrow-band rise at a bearing's defect frequency gets flagged before it accumulates enough broadband energy to move the overall number. Speed/load normalization solves the VFD problem: a model trained on speed learns that a given mm/s reading is normal at one speed and abnormal at another.

None of this replaces ISO 20816. A fleet-wide, reproducible severity classification is still useful, arguably more useful once it isn't carrying your entire early-warning program by itself.

What this looks like with Prevly

Prevly is built to sit on top of ISO 20816, not argue with it: the standard's severity classification stays useful; the gaps above get closed by a model that learns each asset individually:

Read-only, on-premise by default. Prevly's edge agent subscribes to your existing vibration, temperature, and current sensors over read-only OPC-UA (no writes to your PLCs) and runs as an on-premise Docker deployment by default, so your data stays on your network.
A learned baseline instead of a fixed zone. LSTM-autoencoder anomaly detection is conformal-calibrated on each asset's own operating history, and it's speed- and load-aware: Criterion II, continuously, per machine.
A time estimate, not just a severity label. Once degradation is flagged, a LightGBM RUL model, validated on real NASA C-MAPSS data, estimates remaining useful life as a conformal prediction interval: a defensible range, not a false-precision number.
Attribution, not a black-box score. Every alert reports which signals drove it (Integrated Gradients for the deep anomaly model, SHAP for the gradient-boosted RUL model), so an engineer forms a hypothesis instead of trusting a number.
A prediction becomes a work order. Detection, RUL, and attribution feed a drafted work order that coexists with the CMMS you already run: Prevly feeds your system of record earlier and with more evidence than a zone chart alone.

Frequently asked questions

What is ISO 20816? ISO 20816 is the current international standard for measuring and evaluating mechanical vibration on industrial machines. It defines how to measure broadband RMS velocity (in mm/s) on bearing housings or shafts, and sorts that reading into four severity zones (A through D) that vary by machine type, size, and mounting.

What is the difference between ISO 10816 and ISO 20816? ISO 20816 is the successor standard. It supersedes ISO 10816 (bearing-housing/casing vibration) and folds in the former ISO 7919 series (rotating-shaft vibration) into one unified framework. The zone concept and most numeric limits carried over largely unchanged; what changed is scope, structure, and some methodology detail.

What are the ISO 20816 vibration zones? Four zones, best to worst: Zone A (newly commissioned or freshly overhauled condition), Zone B (acceptable for unrestricted long-term operation, where most healthy machines sit), Zone C (unsuitable for continuous operation; plan an investigation and corrective work), and Zone D (vibration severe enough that damage is likely; act immediately).

Is 4.5 mm/s a safe vibration level? It depends entirely on the machine. 4.5 mm/s RMS sits near the boundary between "acceptable" and "needs investigation" for large, rigidly-mounted industrial machines, but it's too high for a smaller motor and too conservative for some flexibly-mounted equipment. Treat it as a commonly cited reference point, not a universal alarm setting.

See what your chart is missing

ISO 20816 is a good pass/fail line, not a maintenance plan. The gap between "still Zone B" and "should have replaced that bearing three weeks ago" is exactly where AI-based condition monitoring earns its keep: a per-asset baseline, frequency resolution, and a work order before the number finally moves.

Request a Prevly demo and see what's developing behind your "satisfactory" reading.