Vibration Analysis Guide for Predictive Maintenance
July 8, 2025 · Dr. Raj Patel
Almost every rotating machine announces its own failure weeks or months before it stops. A bearing spalls, and the first sign is not a squeal — it is a faint, high-frequency ripple repeating at a mathematically predictable rate. An imbalance develops, and a 1× running-speed component grows in the spectrum while the machine hums on normally. Vibration analysis reads those signals: measure how the machine vibrates, decompose it into constituent frequencies, and match each to a specific mechanical defect. Done continuously, it converts “the pump failed unexpectedly” into “the bearing was trending down for six weeks and we changed it in a planned window.”
The Physics in One Paragraph
A healthy machine vibrates at predictable frequencies: 1× running speed from residual imbalance, small harmonics from normal dynamics, and low-level broadband from friction and flow. Defects add their own signatures — imbalance peaks at exactly 1×, misalignment at 2× and 3×, looseness as many harmonics of 1×, and a failing bearing as a family of peaks at the geometric frequencies of its balls, cage, inner race, and outer race. The analyst’s job is to compare the measured spectrum against these expectations and watch how the pattern changes over time.
FFT Fundamentals
The enabler is the Fast Fourier Transform, which takes the time-domain waveform and decomposes it into a spectrum of frequency components — turning a signal that looks like messy noise into a histogram of distinct, attributable peaks.
Three settings govern whether that spectrum is useful:
- Sampling rate and bandwidth. The accelerometer’s range must capture the frequencies of interest. Bearing and gear-mesh signatures are high — often 2 to 10 kHz — requiring high sample rates; imbalance and misalignment are low-frequency, visible at a few times running speed.
- Resolution. The frequency spacing between adjacent lines decides whether you can separate a fault peak from a near neighbor. A 4,000-line spectrum over 3,000 Hz resolves 0.75 Hz — enough to split the sidebands of a slight speed variation, not to split two close bearing peaks.
- Windowing and averaging. A Hanning window reduces spectral leakage from peaks; averaging several spectra suppresses the random noise that would mask a small defect peak.
The Bearing Defect Frequencies
Rolling-element bearings are the most common source of vibration-induced failures, and their defect frequencies are computed from geometry and speed. The four characteristic frequencies are:
- BPFO (ball pass frequency, outer race): where most spalling starts, because the outer race takes the load in the loaded zone.
- BPFI (ball pass frequency, inner race): inner-race faults.
- BSF (ball spin frequency): defects on the rolling elements themselves.
- FTF (fundamental train frequency, cage): cage wear or lubrication problems.
A worked example: a motor at 1,800 RPM with a bearing of 8 balls and a ball-to-pitch-diameter ratio of 0.17. BPFO = (N/2)(1 − d/D)(RPM/60) = 4 × 0.83 × 30 ≈ 100 Hz; BPFI = (N/2)(1 + d/D)(RPM/60) = 4 × 1.17 × 30 ≈ 140 Hz. A peak at 100 Hz with sidebands spaced at running speed means the outer race is spalling — and the peak’s amplitude trend tells you if the spall is days or months old.
This is why vibration analysis feels like a superpower: the technician never sees the bearing; the frequency is the diagnosis.
Envelope Analysis: Finding the Buried Signal
Early bearing damage produces only low-energy impacts — the defect peak sits buried in the noise floor. Envelope (demodulation) analysis filters the high-frequency carrier to reveal the low-frequency repetition rate of those impacts, which is the bearing defect frequency. That is why a pump with a developing spall shows a clean BPFO peak in the envelope spectrum while the raw acceleration spectrum still looks healthy — and why modern programs catch bearing failures at the “schedule it” stage instead of the “it just failed” stage.
Severity Standards: What Amplitude Means
For general-purpose machinery, ISO 10816 (now ISO 20816) defines severity bands in terms of RMS velocity (mm/s) on the bearing housing. The bands depend on machine class — a small pump versus a large turbine — but the shape is consistent:
| RMS velocity (mm/s) | Interpretation |
|---|---|
| < 2.3 | Good / acceptable |
| 2.3 – 4.5 | Satisfactory, review trend |
| 4.5 – 7.1 | Unsatisfactory, plan corrective action |
| > 7.1 | Unacceptable, act immediately |
Velocity emphasizes the mid-frequency band where imbalance and misalignment live, making it the right metric for overall machine health; acceleration and its high-frequency envelope are the right domain for bearing and gear diagnostics. A good program reports both.
Route-Based vs. Continuous Monitoring
Vibration programs fall into two complementary models:
- Route-based (portable). A technician walks a route with a handheld analyzer on a schedule — weekly for critical assets, monthly for the rest. It is inexpensive and comprehensive, but point-in-time: a defect that develops between visits is missed until the next round.
- Continuous (online). Permanently mounted accelerometers stream to the platform 24/7, which does the FFT and trend analysis automatically and alerts when a spectrum deviates. It catches the fault the day it appears, essential for unstaffed or high-value assets.
The economics decide the split: continuous sensors on the pumps, fans, compressors, and motors whose unplanned failure would stop production; route-based for the rest.
Building a Vibration Program
- Inventory and baseline. Measure every critical machine in a known-good state to establish its normal spectrum and severity baseline.
- Choose the split. Continuous on the top 10 to 20 percent of assets by criticality; route-based on the rest.
- Configure alerts on trends, not absolutes. A 1× peak that grows 30 percent in two months is actionable long before ISO severity says “unacceptable.”
- Tie alerts to work orders. The diagnosis in the alert should produce a planned repair with the right part stocked — that is the point of the exercise.
- Track the payoff. Log each vibration-detected failure against the cost of the unplanned version — the audit trail justifies the program next budget cycle.
Vibration analysis pays for itself the first time it schedules a repair instead of suffering one. The machine still fails eventually — but it fails on your terms, in a maintenance window, with the spare part on the shelf. Integrar IoT brings the continuous accelerometer feeds onto the same asset-monitoring platform as the motor power draw and process data, so a bearing fault can be correlated with the energy and throughput signals of the machine that hosts it.