INDUSTRY FIELD GUIDE
HVAC diagnostics data for evaluating AI fault review
A technician may see two alarms that look alike while the system’s operating conditions point to different causes.
NAICS 238220 · Heating and air-conditioning contractors
A technician may see two alarms that look alike while the system’s operating conditions point to different causes. The U.S. Department of Energy has published material on test datasets and frameworks for automated fault detection and diagnostics. Scale publicly lists HVAC among field-service interests for de-identified operational records. Those are signals for different opportunities. A controlled fault benchmark can demonstrate a method, while a records program may require actual authorized work histories. DOE fault detection datasets · Scale Data Partnerships
Pair a sensor trace with the technician’s evidence
Use a simulator or approved training bench. Create a case with an apparent abnormal reading under one operating mode, then a second case where the same reading is normal given a different mode. Include timestamps, units, control mode, data gaps, the corroborating measurement and the reviewer’s verified cause. Ask whether the buyer needs detection, a ranked list of possible causes or a safe request for more information. Keep the verified outcome hidden during evaluation.
Keep live controls and client sites out of the pilot
Use training data with documented rights. Keep building layouts, occupancy patterns, credentials, customer identities and proprietary controls out of the first example. Follow the organization’s safety practices and exclude work on live systems from a staged shoot.
Start with one alarm, two plausible explanations and a verifiable distinction. Describe a dataset
Original Fiund exercise
Interpret the same artificial reading under two stated modes
A detached toy rule and invented observations test mode-sensitive interpretation. The demonstration units and ranges are not HVAC settings or diagnostic thresholds.
| Record | Source | What it records |
|---|---|---|
| RULE-H1 | Original exercise rule | Mode Alder's practice interval is 4–6 units; mode Birch's is 8–10. These ranges apply only to the exercise. |
| OBS-H2 | First artificial observation | The value is 5 units with documented mode Alder. |
| OBS-H3 | Second artificial observation | The value is 5 units with documented mode Birch. |
| OBS-H4 | Incomplete observation | The value is 5 units, but its operating-mode reference is absent. |
Question for the model
How does each observation compare with the supplied toy rule, and which one lacks the needed context?
Show the reference answer and checks
H2 is within Alder's stated practice interval. H3 is outside Birch's interval. H4 cannot be classified under the mode-specific rule because its mode is missing. The same number supports different comparisons when context changes. None of these comparisons establishes a real equipment fault, cause or corrective action.
- Labeling every value of 5 alike without checking its mode.
- Inferring a missing mode from the value that would make it appear normal.
- Turning an exercise-range mismatch into a real equipment diagnosis.
Fields to preserve
- Observation time and identity
- Keep each value with its own context.
- Documented mode
- Preserve the state needed to choose the rule.
- Rule version and unit
- Retain the specific artificial comparison basis.
- Reviewed outcome
- Keep any independent cause finding separate from the range check.
Test another case
Change the exercise rule version for Birch while leaving the observations fixed. The answer must follow the applicable rule rather than remember the first label.
Use a simulator, approved detached data or paper records. Exclude real controls, building layouts, occupancy patterns, credentials, customer records and live-system work.
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.