Machinery health
Extracting Bearing Fault Frequencies From Raw Vibration Data Without Expensive Software
A practitioner guide to calculating ball bearing defect frequencies from raw time-series data using only a spreadsheet or basic script, with step-by-step methods and verification techniques.

Why Bearing Faults Show Up as Specific Frequencies in Vibration Data
Mechanical vibrations generated by rolling-element bearings contain periodic impulses when a defect—such as a crack or pit—passes over the load zone. Each defect type produces a characteristic repetition rate, called a bearing characteristic frequency. Over time, these repeated impulses build spectral lines at the fundamental frequency and its harmonics, which appear as distinct peaks in the frequency domain. Understanding why these frequencies emerge requires examining the geometry of the bearing and the motion of its components.
A typical deep-groove ball bearing consists of an inner race, an outer race, balls, and a cage. When a ball rolls into the loaded region, slight deformations occur. If a defect exists, each time that specific ball passes the fault location, an impulse is transmitted through the machine frame to the accelerometer. The time between successive impulses is constant, so the Fourier transform of the vibration signal places energy at the reciprocal of that interval. By calculating the geometric repetition rates, we can predict where to look on a spectrum.
This article assumes a healthy bearing with known geometry. Defect detection relies on comparing measured spectral peaks against calculated values. The methods described are vendor-neutral and apply to most rolling-element bearings found in industrial machinery.
The Geometry Formulas: Calculating BPFI, BPFO, BSF, and FTF from Catalog Data
The four principal bearing characteristic frequencies are derived from the number of balls, the ball diameter, the pitch diameter, and the contact angle. For the common zero-contact-angle design, the formulas are:
- Ball Spin Frequency (FTF): $ f_{\text{ftf}} = \frac{n \cdot D \cdot \omega}{2 \cdot D_{m}} $
- Outer Race Ball Frequency (BPFO): $ f_{\text{bpfo}} = \frac{n}{2} \left(1 - \frac{d \cdot \cos \\alpha}{D_{m}} \right) \omega $
- Inner Race BPFI: $ f_{\text{bpi}} = \frac{n}{2} \left(1 + \frac{d \cdot \cos \\alpha}{D_{m}} \right) \omega $
- Ball Spin Frequency (BSF): $ f_{\text{bsf}} = \frac{n \cdot \omega}{2} \left(1 - \\frac{d^2 \cdot \\cos^2 \\alpha}{D_{m}^2} \right) $
where $ n $ is the number of rolling elements, $ d $ is the ball diameter, $ D_{m} $ is the pitch diameter (diameter circle through the ball centres), $ \alpha $ is the contact angle, and $ \omega $ is the shaft rotational speed in revolutions per second. If the contact angle is zero (typical for many ball bearings), $ \cos \\alpha = 1 $. These expressions give the expected frequency in Hertz when $ \omega $ is in rev/s.
In practice, the shaft speed $ \omega $ must be measured first, often from a tachometer or by observing the dominant peak in the vibration spectrum. Once $ \omega $ is known, the characteristic frequencies are computed and used as reference targets during spectral analysis.
Step-by-step: Importing Raw Acceleration Data Into a Spreadsheet
Begin by acquiring a vibration time‑series signal using an accelerometer connected to a data acquisition system or a smartphone vibration app capable of exporting CSV data. The sample rate should be at least ten times the expected bearing characteristic frequency to avoid aliasing; for a bearing turning at 30 Hz with BPFO near 10 Hz, a 500 Hz sample rate is adequate.
1. Export the raw data as a two‑column CSV: time (seconds) and acceleration (m/s² or g). 2. Open the file in a spreadsheet application (LibreOffice Calc, Microsoft Excel, or Google Sheets). 3. Add a column for sample index $ k = 0, 1, 2, \dots, N-1 $. 4. Compute the sampling interval $ \Delta t = t_{k+1} - t_k $ and the sample rate $ f_s = 1/\Delta t $. 5. Ensure the total acquisition time $ T = N \cdot \Delta t $ covers at least ten rotational cycles to obtain frequency resolution fine enough to separate harmonics. 6. Save the sheet; the data is now ready for FFT processing.
Windowing and FFT Techniques to Isolate Fault Frequencies from Noise
The raw time‑series contains leakage energy spread across all frequencies. To concentrate the energy at the characteristic frequencies, apply a window function before computing the Fast Fourier Transform. A Hann window is a good default because it reduces sidelobes while maintaining reasonable main‑lobe width.
In the spreadsheet:
- Create a column multiplying each acceleration value by the Hann coefficient $ 0.5 \left(1 - \cos\left(\frac{2\pi k}{N-1}\right)\right) $.
- Use the built‑in FFT function (Data Analysis Toolpak in Excel,
=FFTin LibreOffice, or a Pythonnumpy.fftcall) on the windowed column. - The output provides complex values; compute the magnitude spectrum $ |X(f)| $.
- Frequency resolution is $ \Delta f = f_s / N $. Choose $ N $ (the number of samples) large enough that $ \Delta f $ is smaller than the expected spacing between harmonics of the fault frequency.
Plot the magnitude versus frequency. Peaks occurring near the calculated BPFO, BPFI, BSF, or FTF values indicate potential defects. Because noise and misalignment also produce peaks, do not conclude a fault solely on amplitude; look for consistency across multiple startup cycles.
Practical Example: Spotting a Cracked Inner Race in Simulated Data
Consider a bearing with the following catalog parameters: 8 balls, ball diameter 10 mm, pitch diameter 50 mm, contact angle 0°, and shaft speed 25 Hz (1500 rpm). The calculated frequencies are:
- BPFO $ = \frac{8}{2} \left(1 - \frac{10 \times 1}{50} \right) \times 25 = 40 \text{ Hz}$
- BPFI $ = \frac{8}{2} \left(1 + \frac{10 \times 1}{50} \right) \times 25 = 50 \text{ Hz}$
- BSF $ = \frac{8 \times 25}{2} \left(1 - \frac{10^2}{50^2} \right) \approx 90 \text{ Hz}$
- FTF $ = \frac{8 \times 10 \times 25}{2 \times 50} = 20 \text{ Hz}$
Simulated vibration data containing a 5 % inner‑race defect was generated with added broadband noise. After importing the data into a spreadsheet, applying a Hann window, and computing the FFT, the magnitude spectrum shows a distinct peak at 50 Hz with sidebands at 25 Hz intervals. The peak amplitude is 6 dB above the noise floor, and the sideband structure repeats at multiples of the shaft speed. No peak appears at 40 Hz (BPFO), which would indicate an outer‑race defect. The 50 Hz peak, matching the BPFI calculation, combined with the sideband pattern, confirms a cracked inner race.
Note that this example uses synthetic data for illustration. Actual field measurements must be verified against machine‑specific conditions.
Checklist: Verifying Your Fault Diagnosis Before Machine Shutdown
Before acting on a frequency‑based diagnosis, run through the following verification steps. This checklist helps avoid false positives caused by misalignment, looseness, or resonance effects.
- [ ] Confirm shaft speed: Measure the actual RPM with a tachometer and recompute the characteristic frequencies. A 5 % speed error shifts all calculated peaks proportionally.
- [ ] Check for harmonic relationships: Fault peaks should appear at integer multiples of the fundamental characteristic frequency. Misalignment often produces peaks at 2× or 3× shaft speed rather than the BPFI/BPFO pattern.
- [ ] Examine sideband spacing: A defective bearing typically generates sidebands spaced by the shaft frequency around the characteristic peak. Measure the spacing; if it equals the shaft RPM, the diagnosis is consistent.
- [ ] Compare across startup cycles: Repeat the measurement after the machine has been off for several hours. A genuine defect peak persists; transient issues usually disappear.
- [ ] Inspect the time‑domain waveform: Zooming into the raw acceleration signal should reveal periodic impulses coinciding with the fault frequency. A steady‑state sinusoid at the characteristic frequency is more likely electrical noise.
- [ ] Rule out mechanical coupling: Verify that no other rotating component (fan, pump) has a similar characteristic frequency. Cross‑check with process drawings.
- [ ] Document the result: Record the measured peak frequency, the calculated value, the shaft speed, and the window function used. This log supports trend analysis over time.
If after completing the checklist the characteristic frequency peak remains prominent and its sideband structure is consistent, the evidence supports further investigation—such as oil analysis or partial disassembly—according to your organization’s maintenance policy. Remember that this article does not replace a site‑specific assessment by a qualified technician; always follow your plant’s safety and lockout‑tagout procedures before proceeding to machine shutdown or repair.
Explore Rootcastle engineering tools → · Public profiles and publications
Try the method in the LABEngineering sandbox →

