SCADA data quality check
Look at the data itself before building a predictive maintenance or power forecasting model. Drop your CSV file; we check the timestamps and every channel and report what we find in plain language. The file never leaves your computer.
The file is read only in your browser; it is not uploaded to any server.
Comma, semicolon or tab separated CSV; first row with column names. Timestamps: 2025-10-26 02:00, 26.10.2025 02:00 or Unix time. Decimal commas are recognised. Up to 200,000 rows and 40 numeric channels are checked.
01
How to use it
A
Export a period of data from your SCADA, historian or data logger as CSV.
B
Drop the file into the box; timestamps and every channel are checked within seconds.
C
Review the findings, copy the report and share it with your team; flag problematic periods before modelling.
02
What is checked?
Timestamps: record interval, duplicate and out-of-order records, gaps and the number of missing records, duplicate or missing hours on daylight saving time changes, and whether time zone information is present.
Every numeric channel: fill rate, runs where the value does not change (frozen), physically impossible values in channels recognised by name (wind speed, direction, temperature, humidity, speed) and values far from the usual distribution. Rows where all channels are zero at the same time are counted separately.
03
Why before the model?
SCADA systems are built for operations, not for analysis. A gap filled with zeros tells the model "the machine produced nothing"; a frozen temperature sensor hides real heating; data recorded in local time shifts by an hour when joined with maintenance logs.
A model trained before these problems are dealt with learns data errors, not faults. For a detailed explanation and the approach we recommend, see our note on SCADA data quality.
04
What do the findings mean?
Not every finding is a problem. Zero output during stops, constant power under grid curtailment or a channel that looks frozen during maintenance can reflect real operating states. The tool cannot tell them apart; that takes matching with maintenance and alarm logs.
We recommend flagging problematic records rather than deleting them: clean periods can be selected for training, and loss calculations can state plainly which part of the data could not be assessed.
FAQ
- Is my file uploaded anywhere?
- No. The file is read and checked by code running in your browser; it is neither sent to a server nor stored. Closing the page discards the data.
- Which formats can it read?
- CSV and TXT files separated by commas, semicolons, tabs or pipes, with column names in the first row. Timestamps can be ISO (2025-10-26 02:00), day.month.year (26.10.2025 02:00) or Unix time; decimal commas from European Excel exports are recognised.
- I have a large file; what are the limits?
- Up to 50 MB, 200,000 rows and 40 numeric channels are checked; anything beyond is cut off and the report says so. For 10-minute data that is roughly 3.8 years of one turbine.
- How are the physical limits set?
- From the channel name: general limits are applied to channels whose names mention wind speed, direction, temperature, humidity or speed (e.g. direction 0–360°). If your channels are named differently, this check is skipped; for limits specific to your equipment, get in touch.
Let's get your data ready for predictive maintenance
We turn existing SCADA and sensor data into a health score per machine and early warnings before failure. Let's talk about your data and goals in a free discovery call.