Sensor & lab-data cleanup
Upload a CSV from your test bench, battery cycler, or sensor log. Get denoised channels, filled dropout gaps, and flagged anomalies back — in minutes. No filter to design, no model to train. A self-tuning alternative to Butterworth, Savitzky-Golay, wavelet and Kalman when they break on real, non-stationary data. Remove noise from sensor data, fill gaps in irregular time series, and despike outliers across every column, in one pass.
Clean a whole file free — every column, no channel limits · CSV in, CSV out.
Live demo — no sign-up
Drag a CSV here or
One small file, no account, no card — processed in memory and never stored.
Want your whole dataset, more files, or anomaly detection? Sign up free — no card →
In → Out
Battery cycler exports — Arbin, Neware, Maccor, BioLogic. Bench DAQ logs, climate-chamber runs, process and sensor telemetry. Any numeric time-series as CSV or TSV — every column, however many.
Denoised channels in three latency modes, dropout gaps filled from correlated channels, spikes and outliers flagged, and anomaly events with a root-cause hint — as a clean CSV you download.
Why it’s painful today
Deleting spikes and interpolating gaps cell by cell costs hours per file, and no two people do it the same way — your cleanup isn’t reproducible.
pandas + scipy works, but you rewrite it for every cycler format and every project, and picking the filter parameters is the hard part.
A Butterworth cutoff or a Savitzky-Golay window that’s right for the quiet part smears the transients — on non-stationary data a single setting is wrong somewhere.
How it works
Drop the file — a cycler export, a bench log, a sensor trace. The live demo needs no account.
Self-tuning denoise, gap-fill from correlated channels, and anomaly flags across every column — offline-grade, no parameters to pick.
Get a clean CSV plus a before/after view and a list of what was changed and what couldn’t be recovered.
Evidence
dB RMSE-reduction vs the noisy input; higher is better, negative means the filter made it worse than doing nothing. On hard, non-stationary signals every classical baseline goes negative at high SNR — the self-tuning output stays strongly positive.
| Hard signal · SNR | Best classical filter | Low-latency | Offline |
|---|---|---|---|
| Multi-tone vibration · 30 dB | −6.5 dB | +8.3 dB | +12.5 dB |
| Chirp (moving spectrum) · 20 dB | −11.7 dB | +5.3 dB | +0.0 dB |
Full tables, all eight signal classes and the protocol — on the Benchmarks page →
How it compares
| Approach | Effort | Reproducible | Cost |
|---|---|---|---|
| Excel by hand | Hours per file | No | Your time |
| Python (pandas/scipy) | Rewrite per format | If you version it | Your time + skill |
| Enterprise data platform | Onboarding | Yes | Seats, quotes |
| nlsys — upload a CSV | Minutes | Deterministic | Free demo, then by volume |
FAQ
CSV or TSV, including cycler exports from Arbin, Neware, Maccor and BioLogic, bench DAQ logs, and any numeric time series.
It removes noise and fills gaps while preserving the underlying trend — capacity fade, for instance. A do-no-harm criterion backs off where the signal is already clean, and every change is listed.
The live demo cleans one small file, no account. Sign up free — no card — to clean bigger files and more of them. You’re billed only by the volume of data processed, never by the number of channels: every column of your file is cleaned.
Demo files are processed in memory and never stored. Account data is processed on our own servers, stays your property, and is deleted on request. See the Privacy Policy.
Try it
A whole file cleaned — every column, no channel limits. Try the live demo above with no account, or sign up free to clean your full dataset.
Guides
Savitzky-Golay vs moving average vs Kalman — how each works and where each fails.
Arbin, Neware, Maccor, BioLogic — format gotchas and de-spiking without killing the fade curve.
Five methods — forward-fill, interpolation, spline, correlated-channel, flagging — and when each breaks.
Hampel and friends — remove spikes without deleting real events.
The tabular angle: what it means, the Python way, and the no-code way.