Per-asset-class anomaly models calibrated on real distribution signatures
Generic vibration thresholds fail for distribution field assets. Fieldiq's models are trained separately per asset class — distribution transformer, line recloser, and step voltage regulator — on the electrical and thermal signatures specific to each.
How the anomaly models work
Each asset class has a dedicated model architecture trained on the signal characteristics of that equipment type — because distribution transformer winding vibration is nothing like recloser contact mechanics.
-
Multivariate time-series baseline
Each model ingests multiple sensor streams simultaneously — temperature, vibration, and load-current — and learns the normal co-variation pattern for that asset class under varying operating conditions.
-
Thermal aging model
IEEE C57.91 thermal aging calculations are embedded into transformer models — ambient temperature and load cycling are accounted for in failure probability scoring, not treated as noise.
-
Pilot calibration period
During the first 30 days of a pilot, Fieldiq's models learn the specific signature of your assets at your site. Anomaly thresholds are set relative to observed baseline, not a population average.
-
Asset health index scoring
Each asset receives a composite health index score (0–100) combining anomaly severity, asset age, load criticality, and geographic outage impact — enabling fleet prioritization, not just individual alerts.
-
Recloser wear modeling
Operation-count thresholds are a blunt instrument. Fieldiq's recloser model incorporates load magnitude at each operation and ambient temperature to compute actual contact wear accumulation.
-
DGA proxy signals
While Fieldiq does not perform dissolved gas analysis directly, the thermal and vibration signatures it monitors serve as early proxy indicators for the combustible gas conditions that DGA catches later.
Model performance on validation data
| Asset class | Precision | Recall | Avg lead time |
|---|---|---|---|
| Distribution Transformers | ~94% | ~89% | 3.2 weeks |
| Line Reclosers | ~92% | ~87% | 2.8 weeks |
| Step Voltage Regulators | ~91% | ~85% | 2.4 weeks |
Metrics from held-out validation sets. Results vary by asset age, sensor placement, and data quality. Not a performance guarantee.
Models calibrated on your assets, not population averages
Every Fieldiq deployment starts with a calibration period on your specific assets and service territory conditions. Contact us to discuss what that looks like for your distribution fleet.