The vibration signal from a bearing housing contains the fingerprints of every fault present in the machine. The challenge is separating them. FFT converts the time-domain signal into the frequency domain where each fault type occupies its own predictable frequency, making overlapping fault signatures readable independently.
Step 1: The Vibration Source
Every rotating machine generates vibration from the cyclic forces produced by its motion. An imbalanced rotor creates a centrifugal force rotating at shaft speed. A bearing with an outer race defect generates an impact each time a rolling element passes over the defect at the BPFO frequency. A meshing gear pair produces a sinusoidal force at the gear mesh frequency.
These forces transmit through the machine structure to the bearing housing surface, where they arrive combined and modified by the structural dynamics of the machine. The housing surface is where the sensor measures them.
Step 2: The Accelerometer and Its Physics
A piezoelectric accelerometer contains a crystal (quartz or ceramic) that generates an electrical charge proportional to the mechanical force applied to it. When the sensor is attached to a vibrating surface, the surface acceleration compresses and relaxes the crystal at the vibration frequency.
The output is a voltage signal proportional to acceleration, measured in g, where 1g = 9.81 m/s². ICP (Integrated Circuit Piezoelectric) sensors include a built-in charge amplifier powered by a constant current source from the data collector, eliminating the need for a separate charge amplifier at the measurement point.
Sensor mounting affects accuracy significantly. A stud-mounted sensor maintains its rated frequency response. A magnet-mounted sensor introduces a resonance peak at a lower frequency and attenuates signals above it. Hand-held probes are unreliable above a few hundred Hz.
Step 3: Signal Integration
Acceleration is the second derivative of displacement with respect to time. Integration converts the acceleration signal to velocity (one integration) or displacement (two integrations).
Velocity, expressed in mm/s or in/s RMS, is the standard for overall machine health assessment per ISO 10816. It correlates well with the fatigue damage done to machine structures at audio frequencies (10 to 1,000 Hz). Acceleration is preferred at higher frequencies, above 1,000 Hz, for bearing fault detection because bearing impacts are brief, high-frequency events where acceleration amplitude is more pronounced than velocity.
Most analyzers perform digital integration after sampling the raw acceleration signal.
Step 4: Analog-to-Digital Conversion and Sampling
The continuous analog voltage from the accelerometer is sampled at discrete time intervals by the data collector's ADC (analog-to-digital converter). The sampling rate must exceed twice the highest frequency of interest, per the Nyquist theorem. For analysis to 5,000 Hz, sampling at 12,800 samples per second is common practice.
The digital time waveform produced by this process is the raw data for all subsequent analysis. Anti-aliasing filters in the data collector prevent high-frequency content above the Nyquist limit from folding back into the spectrum as false low-frequency peaks.
Step 5: The Fast Fourier Transform
The FFT algorithm transforms the time-domain waveform into the frequency domain. The result is a spectrum showing vibration amplitude at each discrete frequency across the measured range.
Frequency resolution equals 1 divided by the measurement duration. A 1-second measurement provides 1 Hz resolution; a 4-second measurement provides 0.25 Hz resolution. Higher resolution requires longer measurement time, which matters when distinguishing closely spaced fault frequencies at low shaft speeds.
Windowing functions, such as Hanning and flat-top, reduce spectral leakage at the edges of the measurement window. Hanning is the general-purpose default. Flat-top provides more accurate amplitude readings at the cost of poorer frequency resolution, making it useful for calibration checks.
Step 6: Spectrum Interpretation
With fault frequencies calculated from the machine's known parameters, the analyst examines the spectrum for peaks at those locations and compares their amplitude to baseline values. Each fault type has a characteristic frequency derived from the machine geometry:
BPFO and BPFI are the ball pass frequency for the outer and inner race respectively. N is the number of rolling elements, d is the rolling element diameter, D is the pitch circle diameter, and theta is the contact angle. These values come from the bearing manufacturer's specification sheet.
Amplitude at a fault frequency above baseline is the diagnostic signal. Sidebands around the fault frequency (peaks spaced at shaft speed on either side) indicate modulation, which typically means the defect is progressing and becoming visible in the time waveform as well.
| Fault Type | Primary Frequency | Axial vs Radial | Harmonics Expected |
|---|---|---|---|
| Imbalance | 1x RPM | Radial dominant | Low |
| Misalignment | 2x RPM | Axial dominant | 2x, 3x present |
| Outer race defect | BPFO | Radial | Harmonics + sidebands |
| Inner race defect | BPFI | Radial | Sidebands at 1x RPM |
| Gear mesh | GMF | Radial | GMF harmonics + sidebands |
| Looseness | 0.5x, 1x, 1.5x RPM | Radial and axial | Sub-harmonic series |
Step 7: Trending Over Time
A single spectrum reading provides limited diagnostic value without context. The same bearing running at -20 dB at its BPFO frequency tells the analyst nothing without knowing what the reading was six months ago.
Trending means comparing successive measurements from the same point, at the same machine load, with the same sensor mounting, over time. Consistent measurement conditions are a prerequisite; load variation alone can change vibration amplitude substantially without any change in machine condition.
A rising amplitude at BPFO over four quarterly measurements indicates a developing outer race defect. A stable amplitude over the same period is not a defect. The rate of rise matters as much as the absolute level: a bearing that doubles its BPFO amplitude in two weeks is in a different failure stage than one that doubles it over two years.
Time Waveform Analysis
The FFT averages the signal over the measurement window, which can hide intermittent or impulsive events. The time waveform, amplitude plotted against time, reveals patterns the spectrum cannot.
Impulsive events: Bearing defect bursts and gear tooth impacts appear as spikes in the waveform. The repetition rate of these spikes identifies which component is generating them.
Modulation: Amplitude varying at a lower frequency indicates a defect at one frequency being amplitude-modulated by another, often shaft speed. This produces the sideband patterns visible in the spectrum.
Truncation: Signal clipping from overrange causes flat-topped peaks in the time waveform and introduces broadband harmonic distortion in the spectrum. Any truncated waveform invalidates the measurement and requires resampling at a higher range setting.
High kurtosis (the ratio of peak amplitude to RMS amplitude) in the time waveform indicates impulsive content. A kurtosis value above 4 is a threshold many practitioners use as a flag for bearing or gear defects worth investigating in the spectrum.
Continuous FFT Without the Manual Steps
Tractian's sensors sample continuously and run FFT analysis on connected assets, trending fault frequency amplitudes automatically without manual spectrum interpretation.
See Mechanical Monitoring