The first measurement is worth as much as your ability to repeat it

The baseline is a condition, not a number of measurements

Baseline means one thing only: the data collected when the machine’s operation is known to be acceptable and stable. The definition, restated in our own words, is in ISO 13373-1:2002 clause 7.2.1 and, nearly identical, in ISO 17359:2018 clause 8.10. Full text available from UNI, the Italian standards body.

Many people expect a prescription like twelve measurements in thirty days. There isn’t one. None of the four standards cited here, ISO 13373-1, ISO 17359, ISO 20816-1, and ISO 20816-3, sets a number of samples or a minimum duration. They set a condition: normal and stable operation, after run-in, with the operating conditions documented.

Why a condition is more serious than a number. A number you can satisfy even by measuring twelve times a machine that is already running badly. The condition, no: it forces you to know what state the machine was in, and to know how to write it down. If you can’t describe the baseline’s conditions, that baseline is useless.

On a new or freshly overhauled machine, the vibration changes over the first days. That’s run-in, not degradation: the baseline is taken afterwards, at operating temperatures and at the normal conditions of speed, load, flow, and pressure. Those values are recorded together with the measurement, not from memory.

A machine with multiple states wants multiple baselines

A machine that works across several regimes doesn’t have one normal vibration. It has several, one per state. ISO 13373-1 and ISO 17359 say so explicitly: with multiple operating states, a baseline may be needed for each state.

An inverter-driven fan at 40% and the same fan at 90% are two machines, from the measurement’s point of view. ISO 20816-3:2022, clause 4.5, requires measuring in all the conditions the machine works in for extended periods, and considers the highest value representative of severity.

The mistake that hollows out the program. If you mix different states into a single baseline you get an average that describes no real state. Then the system sounds when the machine has merely changed its way of working. ISO 13373-1 warns of exactly this: variation due to a change in conditions is easily mistaken for an incipient fault.

Two evaluation criteria, not one

The evaluation rests on two distinct criteria. ISO 20816-1:2016 separates them in clause 6.3.2, absolute level, and clause 6.3.3, change relative to a reference. The second sees what the first does not: a Group 2 machine on a rigid support can go from 1.6 to 2.6 mm/s while staying in zone B. The change, however, says something has shifted, and can indicate damage developing before zone C is reached.

Zone	Group 2, rigid support	How to read it
A	up to 1.4 mm/s	typical of a machine just placed in service
B	up to 2.8 mm/s	can run like this long-term, without restrictions
C	up to 4.5 mm/s	should not run like this continuously
D	above 4.5 mm/s	can end up damaging the machine

These are broadband RMS velocities, Group 2 on a rigid support. Group 2 covers medium-size machines with a nominal power above 15 kW and up to and including 300 kW; for electrical machines, the shaft height applies as an alternative, from 160 mm inclusive to below 315 mm. Classification is based on the nominal power of the rotating machine, not on the output delivered by the plant. The standard applies between 120 and 30,000 rpm, and other classes have other numbers. These are guidelines built on experience, not a report card: manufacturer and customer can agree on different values. One requirement, though, the standard expresses as an obligation, with a shall: two measurements are only comparable if taken at the same point, with the same orientation, and under roughly equal operating conditions, at steady state. That holds within the standard, which remains voluntary in application.

The change criterion: 25% of what

The 25% threshold quoted everywhere is not 25% of the baseline, nor of the previous measurement. It is 25% of the boundary between zone B and zone C.

The math. Group 2 on a rigid support: B/C boundary at 2.8 mm/s. 25% of 2.8 is 0.7 mm/s. ISO 20816-3:2022, clause 6.3, flags that change as significant, whether upward or downward. It is an absolute quantity: it changes with the machine’s class, not with your baseline.

Our own example, same class. A baseline of 0.5 mm/s rising to 1.3: change of 0.8, threshold exceeded. A baseline of 2.0 rising to 2.4: change of 0.4, below threshold. In percentage terms they would be +160% and +20%, but the criterion is worked in mm/s. The 25% remains a guideline from a note: the standard admits other values based on experience, and requires investigating evident changes even below the threshold.

Alarm and trip: what the standard says, what is practice

The alarm signals a value reached or a significant change: in general the machine can keep working while you investigate. The trip indicates the level beyond which continuing can cause damage, and calls for immediate action or shutdown.

Widespread practice, not a normative rule. Alarm at the B/C boundary and trip at the C/D boundary is common practice, not the rule the standard writes. A trip at 1.5 times the C/D boundary also circulates: that factor exists, but it concerns run-up and coast-down transients on rotating parts, a different context.

What the standard actually says about thresholds, in practice.

• Alarm above the baseline by 25% of the upper limit of zone B, and in any case not beyond 1.25 times that limit (ISO 20816-3:2022, clause 6.5.2).
• With a low baseline the alarm can fall below the B/C boundary: that is anticipated, not an error.
• New machine with no baseline: start from experience with similar machines or from the agreed acceptance values, then establish the baseline and correct the threshold. After a major overhaul, it’s redone.
• The trip limit concerns mechanical integrity, depends on the design, and normally not on the baseline. It falls in zone C or D, and it is recommended that it not exceed 1.25 times the upper limit of zone C (clause 6.5.3).
• ISO 17359:2018 recommends reviewing the criteria over time (clauses 8.9 and 11): the initial thresholds are a starting point.

Who looks at the data, and how you notice the silence

Programs don’t fail on the measurement. They fail on four questions with no written answer: who looks at the data, how often, what is done when a threshold trips, who stops the machine. Answer with the names of people, before buying anything.

The enemy is the false alarm: a few weeks of useless notifications and the system gets ignored, then the real warning goes unnoticed. Three defenses: compare only measurements taken under similar conditions; apply a statistical filter before the change criterion, as a note in ISO 20816-3:2022 suggests for machines with variable vibration behavior; validate the quality of the measurement. ISO 17359 builds this check into its flow: poor reading, failed transducer, machine stopped, vibration from a neighboring machine.

The data that stops arriving. A dead sensor and a healthy machine look alike: neither reports anything. ISO 17359 accounts for the case, but on how to watch over the instrument through time the standards don’t go into detail: that work is yours. Look at the age of the last measurement, not just the value: data sitting still for days is as suspicious as a high value. Cross-check with running hours or electrical draw. And every now and then go and touch the sensor: mountings come loose.

Startup checklist

To decide before attaching the first sensor.

1. Which machines are in and why: by criticality, not by convenience of access.
2. Which failures you expect, and whether they have a measurable symptom. If they don’t, monitoring is not the answer.
3. Where you measure: points on the bearings, marked and named in writing.
4. How you mount the instrument: method, calibration, and sensitivity never change.
5. Which operating states you distinguish, and what you record: at least speed and load.
6. When run-in ends and the baseline starts.
7. Which initial thresholds you use, where they come from, when you review them.
8. Who looks at the data, how often, what they do if a threshold trips.
9. Who stops the machine, and who stands in for them.
10. How you notice that an instrument has stopped talking.