Match historian data to the batch

Choose the read that answers the question.

PI Web API provides recorded, interpolated and summary reads. The vendor workbook demonstrates startTime/endTime, a maxCount limit for recorded data and an interval for interpolation. It also shows Timestamp, Value, Good, Questionable and Substituted in a response. Preserve those fields when they are returned; selecting only timestamps and values would discard quality context.

Start with a read-only request for one resolved stream and one closed phase window. Save the source server, point or AF attribute path, resolved WebId, absolute UTC bounds, action, all explicit query parameters and retrieval time. Resolve the same displayed tag against both the source system and equipment: V-4 temperature and V-5 temperature are different observations.

Choose recorded data for inspecting archived events, an interpolated series for an explicitly chosen reporting grid, or a summary for a specified statistic. An hourly reporting grid is a calculation choice. It does not show that the historian acquired a physical measurement on every hour. For a summary, record the installed service’s calculation basis, boundary treatment and bad-value behavior rather than relying on an omitted default.

Source: OSIsoft PI World 2020: Using PI Web API, exercises 2–3.

Separate missing acquisition from compressed history.

An archive can legitimately contain irregularly spaced events. Do not label every long interval between recorded events as missing data, and do not assume the last good value remains trustworthy indefinitely. Use the source status transitions, acquisition diagnostics and the approved interpolation or maximum-age rule to establish which time is usable.

The fictional example explicitly adopts a step-held rule: each listed value applies until the next listed boundary. At 09:30 a bad status interrupts that rule; at 09:40 usable data resumes. A linearly interpolated calculation would be a different model and must be reproduced separately. Never draw a continuous line across a bad interval simply to obtain 100% coverage.

Save the 1,080 °C·min numerator and the 50-minute usable denominator with 21.6 °C. Dividing by the full hour would produce 18 °C and would effectively give the missing interval a zero contribution. Store the excluded duration and reason so a reviewer can distinguish a partial statistic from a complete phase assessment.

Make neighboring phases and late corrections reproducible.

Resolve the batch-to-equipment relationship and the source occurrence ID before extracting data. In this example the review window is [09:00, 10:00) UTC. A source read may include values outside these bounds to construct a boundary estimate. Retain those supporting values but do not count them as observations belonging to a neighboring phase.

Use an absolute interval for a closed batch, not a moving expression such as the last hour. Record the source timezone and offset before converting. Around a daylight-saving clock change, the same local time can occur twice; a timezone-free timestamp leaves the association unresolved.

A later query can return corrected source values or a revised phase end. Retain the earlier extract, the new retrieval, and the changed source/query evidence. A new retrieval time alone does not imply a new process observation. Recalculate affected summaries as a visible revision and leave the previous review linked to the evidence it actually used.

Prove the retrieval finished before calculating coverage.

Before running a scheduled collection, demonstrate read access through the site-approved network route and record which service account can read the chosen stream. A cloud Script needs a reachable, approved interface or transferred file; configuring a private plant address does not create connectivity.

For the first run, inspect the complete source response, limits, continuation behavior and per-item failures. If a response reaches its requested cap, establish whether more data remains using the installed interface’s documented continuation or a verified subdivision of the window. Do not infer completeness from HTTP 200 or from a plausible first page.

Keep a collection receipt containing query identity, extract revision, expected streams/windows, retrieved counts, completion evidence and source File reference. Distinguish retrieval complete from evidence adequate: all four intervals in this example can arrive successfully while the ten-minute quality gap still prevents a full-window conclusion.

Put the calculation beside the batch decision.

A practical Seal template links a batch to equipment, a phase occurrence and an extract File. Keep the observed value, unit, statistic, usable duration, full duration and calculation revision in typed fields, with separate fields for retrieval state and review outcome. This gives the team a concise batch view while preserving the underlying evidence.

Configure the mapping Script to update its imported fields and keep operator explanations and review decisions under their assigned ownership. A useful first delivery is one complete phase package whose source intervals, result and unresolved gap can be inspected together. Increase the tag population and cadence only after this package reconciles at the intended volume.

Vendor references for this workflow

• OSIsoft PI World 2020: Using PI Web API, exercises 2–3. Vendor training illustrates recorded, interpolated and summary reads and timestamp/value/quality fields. Use the installed PI Web API help for version-specific parameters; the interval arithmetic below is our fictional receiving example.

Test it with your data.

Use these cases to agree and test the connection’s behaviour. They are proposed acceptance checks, not completed tests or automatic connector features.

Check your system’s connection options.

Confirm the installed version, available exports and permissions. These pages cover the source interface and link to setup instructions.

• AVEVA PI System
• Emerson DeltaV
• Local Files

For configuration: Seal Scripts and local-file collection.