Vibration analysis for rotating equipment uses a fast Fourier transform (FFT) to break a vibration signal into its component frequencies, then trending those amplitudes over time to catch faults weeks before failure. On a typical industrial fan the 1x running-speed peak reveals unbalance, 2x indicates misalignment, and rising high-frequency energy above 2,000 Hz warns of bearing wear; each fault has a measurable signature you can track against an ISO 10816 zone limit of 4.5 mm/s.
Why FFT Beats a Single Overall Reading
A handheld meter that reports only overall RMS velocity hides the cause of the problem. FFT separates the signal into a spectrum where each mechanical fault appears at a known frequency multiple of shaft speed. A pump at 1,450 rpm has a running speed of 24.2 Hz, so a peak at 24 Hz is unbalance, one at 48 Hz is misalignment, and one near 120 Hz can be a vane-pass problem. Separating these lets you plan the correct repair instead of guessing, which typically cuts unplanned downtime on HVAC plant by 25 to 40 percent.
The Core Fault Frequencies
Most rotating faults fall into a small set of frequency bands. Knowing them lets a single spectrum tell the whole story. The table below maps the common signatures to their usual cause and the action they trigger on a fan or pump in the 7.5 to 315 kW range.
| Frequency band | Typical fault | FFT signature | Action threshold |
|---|---|---|---|
| 1x running speed | Unbalance | Single tall peak at shaft Hz | Rebalance above 2.8 mm/s |
| 2x running speed | Misalignment | Strong 2x, sometimes 3x | Align at next stop |
| Bearing defect freq. | Bad bearing | Modulation above 2,000 Hz | Replace before 7 mm/s |
| Vane pass (5-10x) | Blade or vane issue | Cluster near BPFO/ BPFI | Inspect impeller |
How to Take a Repeatable Measurement
Mount the accelerometer on the same clean, machined point every time, usually on the bearing housing in both radial and axial directions. Sample at 25,600 Hz to capture up to about 10,000 Hz, which covers rolling-element bearing defects. Set the FFT lines to 3,200 or more so a narrow bearing tone is not smeared across two bins. Record three values per point: overall velocity in mm/s, the 1x amplitude, and the highest high-frequency peak, then store them against the date so the next reading is directly comparable.
Trending: The Part That Prevents Failure
A single spectrum is a snapshot; the trend is the early-warning system. Plot RMS velocity each month and fit a slope. A fan that climbs from 1.5 to 3.8 mm/s over four readings is degrading even though each point sits inside zone B, and the slope predicts when it will cross the 4.5 mm/s repair limit. Most sites set a warning at a 2x rise over the baseline and a shutdown at 7.1 mm/s. Trending this way typically finds a failing bearing 6 to 10 weeks before it seizes.
When to Use FFT vs When to Use a Simpler Check
- Choose FFT spectrum analysis when the machine is critical, runs continuously, and a single stoppage costs more than the sensor kit (roughly USD 1,200 to 3,500).
- Choose overall RMS-only trending when the asset is small, spare, or you only need a pass/fail against the 4.5 mm/s limit.
- Choose a monthly route with both when you manage 20 to 200 machines and want one dataset for planning.
- Choose continuous online monitoring when the fan is in a remote plant room and a failure stops production.
Common Mistakes That Hide Faults
Comparing readings taken at different load points is the biggest error: a fan at 60 percent flow vibrates differently than at 100 percent, so always trend at the same duty. Another is ignoring the axial direction, where misalignment and coupling faults show first. Finally, do not average a bad bearing's modulation into the overall number; the high-frequency envelope needs its own alarm, usually set 12 dB above baseline.
Choosing the Right Hardware and Skill
A 100 mV/g ICP accelerometer with a magnetic base is enough for route work, while online nodes suit permanent installs. The analyser should compute FFT to at least 12,800 lines and export CSV for your CMMS. Training matters: a Level I ISO 18436-2 analyst reads the spectrum correctly the first time, avoiding the 30 percent false-alarm rate seen with untrained users.
Keruite manufactures this equipment at an 80,000 m² facility in Dezhou, China, with an in-house AMCA-210-style test laboratory; every unit is balanced to G2.5 and performance-tested before shipment, so baseline vibration data ships with the unit.
Establishing a Reliable Baseline
A trend is only as good as its first point, so take a baseline spectrum within the first month of commissioning while the machine is known clean, and record the date, load, and the three key values against a labelled route point so the next technician stands in the same spot. A baseline taken at 40 percent load is useless for comparison at full load, so note the duty clearly on the work order. Sites that skip this step spend 20 to 30 percent more time arguing whether a reading is real, because they have no reference to compare against and end up replacing parts that were never faulty.
Setting Alarm Limits That Trigger Work
Beyond the ISO zones, set two site alarms: a warning at a 2x rise over baseline that schedules inspection, and an action at the 4.5 mm/s zone C boundary that plans repair at the next stop. The warning catches the slow drift that a single reading hides, and the action prevents the zone D excursion above 7.1 mm/s where failure is imminent. Review these limits every year as machines age, because a 10-year-old fan may need a tighter warning at 1.5x to keep the same lead time for planning.
Case Study: A 90 kW Cooling Tower Fan
One plant tracked a 90 kW cooling tower fan monthly and watched the 1x velocity climb from 1.8 to 4.1 mm/s over five months while the overall value stayed under the zone B line. The trend slope predicted crossing 4.5 mm/s within six weeks, so the team rebalanced the impeller during a planned stop and avoided an unplanned trip that modelling valued at 18 to 24 hours of lost cooling plus a USD 1,500 emergency repair. The FFT programme paid for the sensor kit within the first avoided failure.
FAQ
What is the first vibration frequency to watch on a centrifugal fan?
On most centrifugal fans the dominant fault frequency is 1x running speed, which points to unbalance. A peak at 2x suggests misalignment, while bearing defects appear at high-frequency bands above 2,000 Hz. Trend 1x weekly so small unbalance growth is caught before the unit reaches the 4.5 mm/s ISO 10816 alarm limit.
How often should I take vibration readings for trending?
For critical fans and pumps run continuously, collect spectra every 30 days and compare against the baseline. Less critical equipment can be trended every 90 days. The interval matters less than consistency: a fixed schedule lets the slope of RMS velocity reveal a degrading bearing weeks before a 7 mm/s shutdown level is reached.
Can FFT vibration analysis replace a normal visual inspection?
No. FFT identifies the fault type and severity, but a walk-round still finds loose guards, oil leaks and belt dust that sensors miss. Use vibration trending to plan the work and the visual check to confirm it. Together they typically extend mean time between failures by 30 to 40 percent on rotating HVAC plant.
What RMS velocity level starts a repair action?
ISO 10816-3 places most medium fans in zone B up to 2.8 mm/s and zone C up to 4.5 mm/s, where you plan repair at the next stop. Above 7.1 mm/s is zone D and immediate action is required. Always judge the trend slope, not a single reading, because a 4.5 mm/s value rising fast is worse than a stable 4.8 mm/s.