The full power of TDengine, now free forever for up to 5,000 tags.

Explore More

Can Engineers Query Historian Data in Natural Language?

Juno Qiu

November 12, 2026 /

Yes. A natural-language interface can let an engineer ask a question about accessible historical data in ordinary language, then return a value, table, or explanation. The system still needs a way to identify the right equipment and retrieve the data. The engineer needs to check what was actually returned.

Suppose an operations engineer asks, “What was the highest temperature in Oven O-12 during the night shift?” That question has a clear measurement, but “night shift” may mean different hours at different sites. It also matters which oven zone and temperature sensor are intended, which unit is used, and whether all readings arrived.

Turn the operating question into a precise request

A useful question identifies the equipment, measurement, time range, and calculation. For example:

For Oven O-12, what was the maximum temperature recorded by the Zone 2 sensor between 10 p.m. on November 3, 2026, and 6 a.m. on November 4, 2026, in the site’s local time? Show the unit and identify any missing data in that period.

This is a hypothetical request. Whether an application can report missing intervals or show supporting details depends on its design and the data it can access. If it cannot answer part of the question, an engineer should not assume that the missing detail was checked.

For other questions, specify the operation needed. A maximum temperature, an hourly average, and the most recent reading serve different purposes. Include an operating state when it matters: an average across oven warm-up and steady production may not describe either condition well. If the measurement is named differently in the historian, agree on the correct tag or asset attribute before interpreting the answer.

How does a natural-language query reach historical data?

The details vary by system. A typical workflow interprets the question, identifies accessible data and context, constructs a query, and returns a result. The interface may hide the database syntax from the engineer, but a query can still be generated and executed in the background.

TDengine IDMP’s AI Chat documentation describes a workflow that generates SQL to query accessible TDengine data and returns an answer. It can also return charts or tables for some questions. The associated asset model provides context such as element and attribute descriptions; see data contextualization.

That description does not mean an AI chat can directly query any third-party historian. The required data must be available through a supported, configured path, the user must have appropriate access, and the necessary history must be retained. IDMP’s LLM connection documentation also describes the AI connection required for Chat. Check the actual deployment and source access before relying on a particular workflow.

Check the answer against the question

A well-worded answer may still use the wrong asset or a different interval. Before using it in an operating review, check:

  • Asset and measurement: Is this the intended oven and Zone 2 temperature signal, rather than a similarly named oven or another zone’s sensor?
  • Time range and time zone: Does the interval match the shift, including the correct calendar date when it crosses midnight?
  • Calculation and unit: Did the system return the maximum, an average, or a sampled value? Is the result in the unit the team expects?
  • Data coverage: Were there gaps, delayed records, or periods when the sensor was unavailable? A maximum over incomplete data may miss the actual peak.
  • Operating context: Was the oven warming up, in steady production, or stopped? Is the result consistent with the question being asked?

Where the application makes the source or calculation available, inspect it. If an authorized trend or a known query is available, compare the answer with that reference. The exact evidence shown to users depends on the product and configuration. If the data cannot be checked or an important part of the request remains ambiguous, refine the question or ask the historian team for help.

What if the answer raises another question?

A peak temperature may tell an engineer when to look more closely, but it does not explain why the peak occurred. A useful follow-up might ask for oven state and related measurements around the same period. Those results are evidence for an investigation, not confirmation of a cause.

Natural-language queries can make a historian question easier to express. Their usefulness still depends on accessible data, clear definitions, and the engineer’s ability to verify the answer. For the documented TDengine workflow, consult the AI Chat guide and confirm the permissions and data available in your environment.

Related Articles