The 6 Pillars of Data Quality for the PI System

PI data quality is broader than a good or bad status. A useful quality model considers several dimensions that determine whether data is fit for operational use.

Six practical dimensions are validity, completeness, freshness, consistency, contextual accuracy, and traceability.

1. Validity

The value should use an expected data type, state, and plausible range.

Validity checks can include bad PI states, impossible values, and invalid codes.

A valid numeric value can still be wrong if it comes from the wrong source.

2. Completeness

Expected data should be present for the required time range.

Check for gaps, missing laboratory samples, incomplete backfills, and missing asset attributes.

Completeness requirements depend on the use case.

3. Freshness

The latest value should be current for the signal type.

A fast process signal and a monthly inspection value need different freshness thresholds.

4. Consistency

Data should follow consistent units, naming, metadata, and modeling rules.

Inconsistent engineering units or duplicate signals can create conflicting interpretations even when each value is individually valid.

5. Contextual accuracy

The value must be connected to the correct asset, measurement, and process meaning.

This includes correct AF mapping, correct source tag, and correct calculation logic.

This dimension is critical in industrial systems because a plausible value mapped to the wrong equipment can be more dangerous than an obvious bad state.

6. Traceability

Teams should be able to determine where an important value came from, how it was transformed, and what uses it.

Traceability includes lineage, ownership, and relevant change history.

Apply the dimensions by risk

Not every tag needs the same set of checks.

Use stricter controls for data that supports critical operations, safety, environmental reporting, maintenance decisions, or enterprise KPIs.

Avoid one composite score without evidence

A single quality score can be useful for summary, but users should be able to see the findings behind the score.

The objective is not to hide complexity. It is to make quality understandable and actionable.

These six dimensions provide a practical framework for evaluating whether PI data is reliable enough for its intended use.