Process historian basics
How process historians store years of time-series data: sampling and exception reporting, compression (deadband, swinging door), timestamps and quality, retrieval with interpolation and aggregates.
What a historian stores
For each tag: timestamp, value, quality — millions of values per day for a medium plant. Storing every sample of every tag forever is impossible, so historians reduce data while keeping the shape of the signal.
Data reduction
| Stage | Method | Effect |
|---|---|---|
| Collection | Sampling period / on change | Defines the time resolution |
| Exception | Store only if the value moved more than a deadband since the last stored point | Removes noise and steady values |
| Compression | e.g. swinging-door: discard points that lie within a tolerance band of a straight line | Keeps slopes and peaks with few points |
Deadbands are set per tag in engineering units: 0.1 °C for a temperature, 1 % of range for a flow. Too wide loses information; too narrow stores noise. Digital states are stored on change.
Timestamps and quality
- Prefer source timestamps (from the PLC/RTU) over server receive time for sequence-of-events analysis.
- Store quality: a gap or "bad" period is information too — never interpolate silently across a communication outage.
- Use UTC internally; convert to local time only for display.
Retrieval
- Raw values as stored.
- Interpolated values at regular intervals (for calculations and reports).
- Aggregates over intervals: average (time-weighted), min, max, count, duration in state.
Time-weighted average differs from the arithmetic mean of stored points when the points are irregular — use the historian's functions, not Excel averages over raw exports.
Sizing hints
Estimate: tags × average stored points per day × bytes per point (compressed). Plan retention (e.g. raw 2 years, aggregates 10 years) and backups from the start.
Connections
Open in Knowledge Graph →Loading the map…
Sources
Updated 2026-10-02