Process data is time-series data

A temperature reading is not a fact — it is a sequence of facts: 98.2 °C at 10:00:00.0, 98.4 °C at 10:00:00.5, and so on. Every instrument in a plant produces such a stream. Relational databases, designed for rows you update in place, are a poor fit: a historian stores values that are written once, queried by time range, and never updated.

This is why data historians exist: specialized time-series databases built for the volume, write rate, and query patterns of process data.

What a historian does

  • High-speed collection: thousands of tags per second from control systems, PLCs, and OPC servers.
  • Efficient storage: timestamp, value, and quality for each sample, structured for sequential access.
  • Time-range queries: retrieve "every sample of Tag X between 14:00 and 15:00" efficiently.
  • Trending and analytics: native support for calculating averages, totals, and deviation from setpoint over arbitrary windows.
  • Integration: standard connectors for OPC UA/DA, Modbus, and control-system vendor interfaces.

Compression and accuracy

Historians compress aggressively. Instead of storing every raw sample, most use swinging-door compression: a sample is stored only when the value moves outside a band around the last stored point. On a slowly changing temperature, millions of samples collapse to a handful of stored points — with the reconstruction error bounded by the band width.

Engineers must understand the trade-off: compression parameters define the accuracy of stored data. For regulatory or quality reporting, the compression band must be set tight enough to reconstruct values within the required tolerance. Never assume stored data is raw data.

Choosing and integrating a historian

Selection criteria: write throughput, query performance over long ranges, compression fidelity, connector coverage for your control systems, and the quality of its analytics API. Integration patterns:

  • Plant historian: collects from the control network via OPC UA, serving operators and process engineers.
  • Enterprise historian: aggregates multiple plant historians into one store for cross-site analytics.
  • Cloud time-series: replicates to a managed time-series service for long-term analytics and machine learning.

The common denominator is the tag model: a consistent naming and metadata scheme across plants is what makes cross-site analysis possible at all.

Key point: A historian's job is to make process data cheap to store and fast to query over time. Compression settings and tag naming are engineering decisions — not defaults.