INDUSTRY FIELD GUIDE
Wind turbine sensor review records for AI
An automatically captured sample and a later reviewed record can refer to the same interval.
NAICS 221115 · Wind Electric Power Generation
An automatically captured sample and a later reviewed record can refer to the same interval. DOE’s Wind Data Hub publishes examples of reviewed turbine data. A wind operator can create a fictional trace to test whether a model reads data status and source revisions correctly. DOE Wind Data Hub · Reviewed turbine data example
Invent two traces, one with a reviewer flag and one with a missing sample. Give the model an earlier export and a reviewed copy. Ask which parts changed and whether a value was replaced, excluded, or never present. Include units and time window, and avoid using the task to diagnose equipment or infer settings.
Do not use live telemetry, equipment access, control details, asset identifiers, exact locations, or private generation figures. Synthetic records suffice.
Describe a sensor review benchmark
Original Fiund exercise
Track a value's review status without overwriting its origin
Artificial raw and reviewed traces demonstrate data-version interpretation. The example uses generic counts and contains no turbine telemetry or performance assessment.
| Record | Source | What it records |
|---|---|---|
| RAW-V1 | Original practice export | At 12:00 the value is 0; at 12:05 the observation is absent; at 12:10 the value is 5. |
| REVIEW-V2 | Reviewed copy | The 12:00 value is excluded with a review flag. The 12:05 observation remains absent. The 12:10 value is corrected to 4 with a revision note. |
| LINK-V2 | Version relation | V2 references V1's same practice interval and preserves each original timestamp. |
Question for the model
Which point was excluded, which was never present and which was corrected?
Show the reference answer and checks
The original 12:00 zero existed and was later excluded; it was not originally missing. The 12:05 observation was absent in V1 and remains absent. The 12:10 point changed from 5 to a reviewed value of 4. Keep raw value, reviewed value, presence and review action separate. The packet gives no equipment explanation for those data-review decisions.
- Calling an excluded original value a never-recorded observation.
- Treating a corrected value as the unmodified raw measurement.
- Collapsing all null outputs into one missingness category.
Fields to preserve
- Raw and reviewed version
- Preserve the lineage and original value.
- Review action
- Distinguish exclusion, correction and no supplied observation.
- Timestamp and reason reference
- Keep each decision linked to the exact point and review note.
Test another case
Supply a correction note for 12:15 with no matching raw point. The model should identify the unmatched correction instead of moving it to 12:10.
Use synthetic records only. Exclude equipment access, real identifiers, controls, locations and private generation figures. No maintenance or setting recommendation is made.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.