Vibration monitoring: route, baseline level, alarm thresholds and trend

One vibration-velocity reading tells you about a machine's condition today. A trend tells you about direction, and that gives you time: to order a bearing through normal channels instead of by courier on a Saturday, and to stop the machine during a planned window instead of on a Friday night. You don't need a fixed system with sensors on every bearing housing for that. You need a portable two-channel instrument, marked points, a table, and discipline in walking the route.

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

One number tells you about today. A trend tells you where things are heading

You put a sensor on an exhaust fan's bearing and see 3.4 mm/s. Now what? Almost nothing follows from that number alone. Maybe the machine has always been that way and will run another five years. Or maybe it was 1.4 a month ago, and it'll be 7 a month from now.

A trend removes that uncertainty. Twelve readings at the same point over a year turn an unknown into a line with a slope. And that slope is what you get for ten minutes of walking a route once a month: a lead time to prepare a repair.

From there it's simple planning arithmetic. A rolling-element bearing defect usually lives from a month to half a year, from the first noticeable sign to failure. Gear mesh in a gearbox deteriorates more slowly, from three months to a year. Unbalance from product build-up or impeller wear can sometimes build up its level within weeks. If you walk a machine's route once a month, you'll almost certainly catch a bearing on the rise and have time to order it the normal way.

A trend also has an underappreciated flip side: it protects you from unnecessary work. A level of 3.4 mm/s that hasn't changed in three years isn't a reason to stop the exhaust fan and pull the impeller. Stability itself is a result of the measurement.

A fixed system isn't needed for this. It's needed where a machine can't tolerate even a month-long gap between readings: a turbine, a large compressor, a critical unit with no standby.

Which machines go on the route, and how often to walk them

The temptation is understandable: put everything that spins on the route. That kind of route dies within two months. The walker can't keep up, starts skipping points, and the round eventually turns into a formality. Select machines deliberately, using clear criteria.

• What stops along with the machine. One exhaust fan stops the whole line. One of four supply fans stops nothing.
• How much an hour of downtime costs on that section. That figure will come in handy again when you're defending the monitoring program to management.
• Failure history. A machine that tears through bearings once a year goes on the route regardless of its power rating.
• Whether there's a standby unit, and how fast it comes online. A standby pump that starts within a minute takes off most of the urgency.
• Accessibility of the points while running. If you can't reach the bearing with the machine running, you have three honest options: build a service platform, bring out a stud with a pad to an accessible spot, or accept that this machine won't be on the route.
• Set the interval by criticality, not one blanket schedule. A working split: critical machines monthly, important ones quarterly, the rest twice a year.
• A benchmark for the interval: it should be no more than half the time your typical defect takes to go from the first noticeable sign to failure. Otherwise you'll see the defect exactly once, and right at the end.

This table is illustrative: a typical example of a route's structure, not anyone's actual plant section. H, V and A in the points column are measurement directions: horizontal, vertical and axial. RMS is the root-mean-square value, the standard way of averaging vibration velocity. The figures in the last two columns are calculated by the rule from the alarm-threshold section and tied to the machine's group. Calculate your own values from your own baseline levels.

Points and directions: set them once, and don't change them again

Mount the sensor on the bearing housing, as close to the bearing as possible, where the load actually passes through the metal. Not on the cover, not on the guard, not on the cladding, and not on the piping. Covers and sheet-metal cladding have natural frequencies of their own and live their own life: you'll be measuring the panel's vibration, not the bearing's condition.

There are three standard directions: horizontal-radial (H), vertical-radial (V) and axial (A). Horizontal is usually the most informative; axial matters where you expect misalignment, or where the rotor is overhung.

Here's the most important rule in the whole exercise. A reading taken at a different point or in a different direction creates a false change in the trend. You'll get either an alarm on a sound machine, or you'll miss a real defect because the new point 'reads' quieter. Catching that kind of substitution after the fact is nearly impossible: the table shows a plausible number, and there's nothing to argue with.

Mounting is also part of the point, not a free choice for whoever's walking the route. The mounting method caps the useful frequency band: a magnet on a flat, clean pad works up to roughly 1.5–2 kHz, a two-pole magnet higher, a stud up to 10 kHz and beyond. For overall level in the 10–1000 Hz band, a magnet is enough. A handheld probe isn't usable for a trend at all: repeatability is too low, and the scatter between two different people easily swamps a real rise in level.

• The point sits on the bearing housing, not on the cover, guard or piping.
• The pad is cleaned of paint and rust for the magnet's contact patch, and won't get painted over the next time the shop gets repainted.
• The point is marked and labeled: machine number, point number, direction.
• The axial direction is marked with an arrow, so the next person doesn't mix it up with radial.
• The mounting method is recorded on the machine's card and stays the same across every round.
• The quantity and band are always the same: for overall level, that's mm/s RMS in the 10–1000 Hz band.
• The order of points on the route is fixed, and isn't changed 'for convenience.'
• The machine's card has a photo of every point with the sensor mounted.

Baseline level: the reference point everything else is measured from

The baseline level is taken on a sound machine. Not one that's currently running, but one you've just finished work on: the mechanics checked, the fasteners torqued, the shafts aligned, the rotor balanced if needed, the bearings new or known to be good. At that moment the machine is in the best condition it will ever be in. That's what you record as the zero reference.

Take the baseline at the operating mode, and record that mode alongside the numbers. Taking one reading isn't enough: take two or three in a row, removing and remounting the sensor each time, and look at the scatter. The scatter shows the inherent noise of your own procedure, and tells you which change in the trend already means something and which one is still within the margin of the sensor mount.

The baseline level is specific to the point, not to the machine. The motor bearing and the fan bearing on the same unit give different numbers, sometimes several times over. Two identical pumps off the same rack will also give different baseline levels, because the foundations, piping and connections are different for each.

Revisit the baseline level only when the machine itself changes: after an overhaul, a rotor replacement, reinstallation on a different foundation, rework of the supports, or a change in operating mode. Don't overwrite the old numbers when you set new ones — put them side by side with the repair date. Comparing baseline levels before and after a repair immediately shows how successful the repair actually was.

Now the honest part. If a machine has been running for ten years and you've never seen it in sound condition, you have no baseline and nowhere to get one from. In that case, take the current reading as a provisional reference point, flag it in the table as provisional, and assess the machine by the absolute zones plus the slope of the line. You'll get a real baseline at the first repair.

Three alarm levels, and why absolute standards fail without a baseline

The A–D zones under ISO 20816 give you an absolute frame: A for a new or overhauled machine, B for long-term operation, C for limited operation, D not permitted. It's a frame you need — without it, you can't defend your decisions to management or to a contractor. But it describes a group of machines, not your specific one. Check separately which exact part and edition applies: individual parts have restrictions by power, speed and machine type. We cover the zones and groups in more detail in a separate piece on vibration standards.

Look at how an absolute standard breaks down without a baseline. A pump with a baseline of 0.9 mm/s falls in a zone that allows up to 2.8 mm/s. An alarm threshold set 'by the book' at the B/C boundary will stay silent until the level has tripled. You'll find out about the problem at the very end of the curve, right where there's nothing left to plan. The opposite case isn't any better: a machine with a baseline of 2.6 mm/s will trip that same threshold on every round, and within a month it just gets switched off.

Which gives you the working rule. Set alarm thresholds off the baseline level, and keep the absolute zones as a ceiling: you can lower a threshold below that ceiling, but you can't raise it above.

Check units and detector type separately. Comparing a peak value against a threshold stated in RMS is easier to do by accident than it sounds, and the mismatch comes out as a multiple. An instrument that defaults to ips also trips people up: 1 ips ≈ 25.4 mm/s, and a modest-looking 0.28 ips on the screen is already 7.1 mm/s.

Sources: ISO 20816-1:2016

What to record at every reading

A number with no context is nearly useless in a trend. Six months on, you're looking at a jump from 1.6 to 2.9 mm/s and can't answer a simple question: did a defect grow, or did someone just close a damper that day. Context gets written down at the moment of the reading — you can't reconstruct it afterward.

Speed heads this list. Unbalance force grows as the square of rotation frequency, so even a few percent change in speed noticeably moves the level. On a drive with a variable-frequency converter, a trend with no recorded actual speed turns into a set of random numbers. Without it you also can't convert the spectrum into multiples of running speed, or tell what's 1x and what's 2x.

• Date and time of the reading.
• Actual speed from the tachometer, not the number off the nameplate.
• Operating mode and load: throughput, damper or valve position, pressure, motor current, how many pumps are running in parallel.
• Bearing-housing temperature. A pyrometer is cheap and gives you a second, independent channel of information.
• Who took the reading and with what: instrument, sensor, mounting method.
• What you saw and heard: an oil leak, a whistle, a knock, a new crack in a frame weld, a swaying pipe.
• A note recording any skipped point, with the reason. A skip nobody knows about is worse than a skip that's on record.

The level rose. What to do before reaching for weights

A tripped threshold is not a work order for balancing. It's a message saying 'go investigate.' Balancing only reduces the running-speed component, and if what rose isn't that, weights won't help, and sometimes make things worse.

1. Step 1

   Retake the reading on site

   Before raising an alarm, repeat the reading. Check that the sensor sits firmly on its pad, the cable is intact, the value isn't 'wild' — like 999 mm/s — and the signal isn't saturated. A bad reading that makes it into the database smears the trend and feeds false alarms for months.

2. Step 2

   Check the mode against the baseline

   Speed, load, damper position, ambient temperature. A good chunk of the 'rise' gets explained right here, and you don't need to go further.

3. Step 3

   Compare overall vibration and 1x

   Did the running-speed component 1x rise, and does it account for most of overall vibration? The unbalance hypothesis is alive. Did overall rise while 1x stayed put? Look for looseness, a bearing, hydraulics. We've written a separate piece on these three quantities and the role of phase.

4. Step 4

   Pull the archived spectrum and compare

   This is where the habit of saving the spectrum, not just the number, pays off. Look at what exactly went up: 1x, 2x, a comb of harmonics, or the high-frequency region. How to read that picture is covered in our piece on reading the spectrum.

5. Step 5

   Check the mechanics by hand

   Torque on the foundation bolts, soft foot (a machine support that doesn't sit flush against the frame and twists the casing when tightened), play, the condition of the supports and frame, bearing temperature, lubricant condition. This costs nothing and resolves a good share of cases.

6. Step 6

   Only now choose an action

   Balancing, shaft alignment, bearing replacement, or work on support stiffness. The decision rests on the whole picture, not on one number that crossed a threshold.

If the 1x amplitude and phase drift by more than 10–15% during the reading, don't put that reading into the trend. The machine may be running near resonance (where the rotation frequency coincides with a natural frequency of the structure), and the number doesn't mean anything in that state.

Seven mistakes that kill a trend

How a two-channel instrument with a laser tachometer, and AXILINE's engineers, help

For a route that actually works, a two-channel instrument, a laser tachometer, a laptop and a table are enough. In vibrometer mode, the Balanset-1A shows overall vibration-velocity RMS, and the amplitude and phase of the running-speed component, plus rotor speed, on two channels at once. Alongside that, the time waveform and FFT spectrum. All of it gets saved, so six months later you pull up not just a number but a picture to compare it against.

Two channels at once give you something you lose with sequential single-sensor readings. You capture both bearings in one run, at one mode, at one speed, with no question of whether the machine was running the same way twenty minutes earlier.

The laser tachometer, reading a reflective marker, closes two jobs at once. Speed in the reading is always the actual value. And you get the phase of the running-speed component, which in a trend often says more than amplitude does: if the level holds steady but the phase has shifted 40°, something changed in the mass distribution or in the support.

Let's be direct about the boundaries. The Balanset-1A is a balancing instrument with a full vibrometer mode, not a route data logger. It has no route database, no automatic point-by-point thresholds, and it doesn't build the trend for you. You keep the trend yourself: in a table, in a round log, in a maintenance record system. For a section with ten to forty machines, that's more than enough. If you have hundreds of points and need automatic upload to a database with trends and reports, you need a route data logger — a different class of equipment, and a different price.

AXILINE's engineers are the same people who design and build Balanset instruments and use them for field balancing themselves. We can come out after a repair and take baseline levels, mark and label the points, help set thresholds by machine group, work through a tripped alarm, and explain what a rising peak means. We can supply the instrument and train you to use it, and you walk the route yourself from there. Consulting support stays with us.

Sources: ISO 20816-1:2016 · ISO 13373-3:2015 · ISO 13373-5:2020 · Balanset-1A operation manual · Balanset-1A manufacturer specification

Frequently asked questions

Related content