A field engineer’s guide to sensor selection, spectrum interpretation, and building a monitoring program that catches machine faults before they become shutdowns
Most rotating machinery gives you a warning before it fails. The problem is that the warning shows up as vibration long before it shows up as noise, heat, or smoke, and by the time a fault is audible or visible, the repair bill has usually gone up by an order of magnitude. I’ve walked into plants after a gearbox seized or a fan bearing shed its rollers, and in almost every case the vibration signature had been drifting for weeks. Vibration condition monitoring exists to catch that drift early, while it’s still a bearing you can swap out on a Tuesday afternoon rather than a production line you’re rebuilding over a weekend.
This guide walks through how vibration monitoring actually works: what the sensors are measuring, how to choose between an accelerometer and a velocity sensor, how to read the data once you have it, and how a continuously monitored asset behaves differently to one that only gets checked on a route once a month.
Every piece of rotating equipment vibrates. A perfectly balanced, perfectly aligned machine with healthy bearings still produces a small, stable vibration signature at its running speed. Condition monitoring is the practice of measuring that signature over time and watching for the changes that indicate wear, imbalance, misalignment, looseness, or bearing damage before those problems cause a failure.
The practice is built on a simple idea: as a mechanical fault develops, it changes how the machine shakes. Amplitude climbs, new frequencies appear, and the shape of the vibration signal shifts in ways that map back to specific fault types. Standards such as ISO 10816 (now largely superseded by ISO 20816) give plants a way to benchmark overall vibration severity against machine class and mounting type, so a reading isn’t just a number, it’s a number with a defined meaning.
Vibration severity is most commonly reported as RMS velocity in millimetres per second (mm/s), because velocity correlates well with fatigue damage across the typical fault-frequency range of industrial machinery.
A rotating shaft generates forcing frequencies tied directly to its running speed. If that shaft is out of balance, you’ll see a strong peak at exactly 1x the running speed. Misalignment between a motor and pump typically shows up at 2x running speed. Bearings have their own fault frequencies, calculated from the geometry of the rolling elements, races, and cage, which is why a vibration analyst can often tell you which bearing component is failing without opening the housing.
None of this is visible from a single overall vibration reading. It only becomes visible once you break the signal down into its component frequencies, which is where spectrum analysis comes in (more on that below).
This is the first fork in the road for anyone specifying a monitoring system, and it trips up a lot of people who assume one sensor type is simply “better” than the other. They measure different physical quantities and are suited to different jobs.
An accelerometer, almost always piezoelectric in industrial use, measures acceleration (g) and, through the electronics in an IEPE or charge-mode sensor, can be integrated to derive velocity or displacement. Its frequency range typically stretches from a few Hz up into the tens of kilohertz, which makes it the only practical choice for detecting high-frequency bearing defects and gear mesh faults.
A velocity sensor, whether it’s an older electrodynamic (seismic) design or a modern piezo-velocity unit, outputs a signal proportional to velocity (mm/s) directly. Traditional electrodynamic velocity sensors are self-generating, meaning they don’t require external excitation power, and they’ve historically been the workhorse for balance-of-plant equipment monitored against ISO velocity severity charts. Their usable frequency range is narrower and their low-frequency response is often better than a comparable accelerometer.
If the asset runs at high speed, has rolling-element bearings, or includes gearing, I’d default to an accelerometer every time, because bearing and gear mesh fault frequencies live well above what a velocity sensor can resolve. If you’re monitoring slower, heavier industrial equipment such as large pumps, fans, or motors where the plant already trends overall vibration against ISO 10816/20816 velocity limits, a velocity sensor (or an accelerometer with onboard integration to output velocity) does the job without over-specifying the hardware.
In practice, most modern programs standardise on IEPE accelerometers with built-in integration circuits, because you get the high-frequency bearing detection of an accelerometer and a velocity output for ISO benchmarking from a single sensor. This is the direction most new installations take unless there’s a specific reason to keep a legacy velocity sensor in place.
Raw vibration data arrives as a time waveform, which is genuinely difficult to interpret by eye once more than one fault frequency is present. Two views make it usable:
A single spectrum tells you the current state of the machine. A trend of spectra over weeks or months tells you the story, and the story is almost always more useful than the snapshot. A slowly climbing peak at a specific bearing defect frequency is a maintenance work order in the making; a spike that appears overnight is a call to the shift supervisor.
Each fault type has a recognisable fingerprint in the spectrum:
Why this matters for maintenance planning Knowing which fault frequency is trending up means the maintenance team can order the correct bearing or coupling ahead of the shutdown, instead of opening the machine to find out.
Why this matters for maintenance planning
Knowing which fault frequency is trending up means the maintenance team can order the correct bearing or coupling ahead of the shutdown, instead of opening the machine to find out.
Route-based monitoring, where a technician walks the plant with a handheld data collector on a weekly, monthly, or quarterly schedule, has been the backbone of condition monitoring programs for decades, and it still has a place for lower-criticality assets. Its limitation is timing: a bearing that starts spalling can go from a barely detectable defect to a seized shaft in days, well inside the gap between two scheduled route visits.
Continuous or near-continuous monitoring, using permanently mounted sensors feeding a data acquisition system either by cable or wireless telemetry, closes that gap. Data streams in constantly (or at short, automated intervals), so an alarm can fire the moment a fault frequency crosses a threshold rather than whenever the next scheduled walk happens to fall. For critical, high-consequence, or fast-developing-fault assets, that difference in timing is the whole value proposition.
The mechanism is straightforward: trended data plus defined alarm thresholds convert an unknown future failure into a known, schedulable repair. A rising bearing defect frequency triggers a work order weeks or months ahead of failure, giving the plant time to order parts, plan the outage around production, and replace one bearing instead of a bearing, a shaft, and whatever else it damages on its way out.
The financial case is rarely subtle once you compare the two paths side by side: a planned bearing swap during a scheduled maintenance window against an unplanned line stoppage, expedited freight on replacement parts, and lost production while the plant scrambles to diagnose a failure it didn’t see coming. Most reliability programs justify the entire cost of a monitoring system on avoiding a single unplanned failure of a critical asset.
A working system is a chain, and each link needs to match the others or the data at the end is only as good as the weakest component:
Getting this right usually comes down to matching sensor specification to data acquisition channel count and sample rate at the design stage, rather than bolting sensors onto whatever DAQ hardware happens to be on hand. This is the stage where working with a supplier who can spec the sensor and the data acquisition hardware together, rather than selling one in isolation, tends to save a plant from expensive rework. Applied Measurement Australia, for instance, supplies both the accelerometers and gyroscopic sensors and the DAQ systems that sit behind them, so the sensor’s frequency range, sensitivity, and output are already confirmed to suit the acquisition hardware before anything is mounted on a bearing housing.
A practical selection checklist looks like this:
Because these decisions interact (a sensor’s frequency range is only as useful as the DAQ system’s sample rate, and a wireless sensor is only as good as the battery life it offers between site visits), it’s worth working through the specification with an application engineer rather than sourcing sensor and DAQ hardware from separate catalogues. If you’re scoping a program from scratch, the AMA team can walk through data acquisition system selection alongside sensor specification so the two are matched before anything ships.
It depends on asset criticality and how fast a fault can develop. Route-based collection every 2–4 weeks suits lower-criticality equipment; critical assets with fast-developing fault modes, such as high-speed gearboxes, generally justify continuous or near-continuous monitoring.
It’s machine-class and mounting-type specific rather than a single universal number. ISO 20816 defines severity zones (A through D) by machine group, and a reading is only meaningful when compared against the correct zone for that asset.
Yes. Misalignment typically produces a strong peak at 2x running speed, often with elevated axial vibration, which is one of the more distinctive and reliably diagnosed fault signatures.
IEPE (constant-current) accelerometers need a small constant-current power source, usually supplied by the signal conditioner or DAQ input. Charge-mode accelerometers don’t need excitation power but require a charge amplifier to condition the signal.
Wired sensors generally offer higher sample rates and no battery to manage, at the cost of cable installation. Wireless sensors cut installation cost and time on hard-to-reach assets but trade off sample rate and depend on battery life and network coverage.
Vibration condition monitoring works because machines tell you what’s wrong before they stop working, provided you’re measuring the right quantity, at the right frequency range, and often enough to catch a fault while it’s still developing. Getting the sensor choice right between an accelerometer and a velocity sensor, reading the spectrum rather than just the overall number, and closing the gap between route-based checks and continuous monitoring are the three levers that turn a monitoring program from a data-logging exercise into a genuine reduction in unplanned downtime.
If you’re specifying a new monitoring installation or replacing sensors on an existing one, Applied Measurement Australia supplies the precision accelerometers, velocity sensors, and data acquisition systems to build the complete chain, and their application engineers can help match sensor specification to your asset’s fault frequencies before you buy. Get in touch to talk through a monitoring program for your plant.
Contact Applied Measurement: Phone: (03) 98745777 Email: sales@appliedmeasurement.com.au Location: 24a/49 Corporate Blvd, Bayswater VIC 3153 Hours: Monday to Thursday 09:00 – 17:00 Friday 09:00 – 16:00
Your email address will not be published. Required fields are marked *
Δ
Leave your details below and a dedicated Applied Measurement team member will be in touch to discuss your requirements.