# Ohmforge Vibration Spectrum (accelerometer CSV)

> Accelerometer CSV in, vibration spectrum out: peak frequencies, amplitudes, rpm and RMS. Finds the frequencies a machine shakes at from accelerometer samples — a washing machine during spin, an unbalanced fan, a pump. Paste or upload a CSV with a time column and one to three acceleration columns (for example a recording from Ohmforge's Pocket instruments or a phone sensor app). You get the spectrum, the strongest peaks with amplitude and the matching rotation speed in rpm, and RMS acceleration per axis.

Part of usefulweb.app (https://usefulweb.app/). Part of [Ohmforge](https://ohmforge.usefulweb.app/) (ownerApp: ohmforge). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://ohmforgevibrationspectrum.usefulweb.app/ (interactive tool at https://ohmforgevibrationspectrum.usefulweb.app/tool)
- JSON API: POST https://ohmforgevibrationspectrum.usefulweb.app/api/v1/calculate with a JSON object of inputs (schema: https://ohmforgevibrationspectrum.usefulweb.app/api/v1/tool)
- OpenAPI: https://ohmforgevibrationspectrum.usefulweb.app/openapi.json
- MCP (Streamable HTTP): https://ohmforgevibrationspectrum.usefulweb.app/mcp - tool `ohmforge_vibration_spectrum`

## Inputs
- file (file): CSV file - One row per sample: time, then 1–3 acceleration columns; header optional. Up to 200 000 rows.
- csv (textarea): Or paste the samples - Used when no file is uploaded. Comma, semicolon, tab or space separated.
- timeUnit (select): Time unit
- accelUnit (select): Acceleration unit
- fftSize (select): FFT size - Larger = finer frequency resolution but fewer averages (noisier).
- peaks (number): Peaks to list

## Method
- The samples are resampled to a uniform grid at the median sample interval (linear interpolation, because phone sensor events arrive with jitter) and each axis' mean is removed — that takes gravity and sensor offsets out.
- Each axis gets a Welch spectrum (Hann window, 50 % overlap) with the same FFT as Ohmforge's capture analyzer and the Soundscape audio analysis; the combined spectrum is √(x² + y² + z²) per frequency bin. Amplitudes are the peak amplitude of a sine (window gain corrected), so a 0.5 m/s² sine reads 0.5 m/s².
- Peaks are local maxima that stand out from the median of their neighbourhood, refined by a parabola through three bins. rpm = frequency × 60 assumes one disturbance per turn (imbalance); blade-pass and gear frequencies are multiples of the rotation speed.
- Only frequencies below half the sample rate can be seen (Nyquist); phones sample at about 50–500 Hz. Linear interpolation lowers amplitudes close to the Nyquist frequency.

## Limits
- Phone accelerometers are not vibration sensors: their sample rate, resolution and mounting limit what they can show. Use the results to compare and find frequencies, not to judge machine condition against ISO 10816/20816 limits.