# Outage and asset data: turning sensor signals into reliability decisions
At 2:14 a.m., a pole-mounted transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → on a suburban feeder sees its top-oil temperature spike as a windstorm pushes a tree limb into a nearby line. Within seconds, three data streams start telling a story: a SCADA relay trips, thousands of smart meters go silent, and the transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète →'s own health score ticks into the danger zone.
By dawn, that data will decide which crew goes where, in what order, and whether this transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → gets replaced next quarter or limps through another summer. This lesson follows those signals from the field to the decision.
Utilities do not "see" the grid directly. They see data. Three systems dominate reliability decisions, and each answers a different question.
SCADA (Supervisory Control and Data Acquisition) is the real-time nervous system of the grid. Sensors on substations, breakers, and feeders report status (open or closed), voltage, current, and alarms every few seconds.
When our transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète →'s upstream breaker trips, SCADA registers it immediately. That is the "something broke" signal. SCADA is fast and precise about equipment status, but it does not know how many customers just lost power.
The OMS translates equipment events into customer impact. It pulls from three inputs:
The OMS uses these to infer the likely fault location. If 40 meters behind one transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → go dark at once, the OMS predicts the transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → or its fuse, not 40 separate problems. This inference is the core value: it turns noise into a single dispatchable ticket.
The third stream is slower and often overlooked. Asset health data describes the condition of the equipment itself: transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → oil temperature, dissolved gas levels, load history, age, and past fault counts.
A common industry tool is Dissolved Gas Analysis (DGA), which measures gases produced when insulating oil breaks down under heat or arcing. Rising gas levels are an early warning of internal failure. Combined with load and thermal data, utilities compute an asset health index, usually a 0 to 100 score, that estimates how close a unit is to end of life.
Here is where data turns into decisions. During a major storm, no single stream is enough.
SCADA says a breaker tripped. The OMS says roughly 900 customers are out across four feeders. Asset health data flags that our transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → was already at a poor health score before the storm. Fusing these tells the operator something none of them could alone: this is not just an outage, it is a likely permanent equipment failure at a known weak point.
That changes the response. A tripped line that reclosed and restored itself needs no truck. A confirmed transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → failure with a bad health history needs a replacement unit, not just a repair crew.
During blue-sky days, dispatch is simple. During a storm with 200 simultaneous outages, it is a triage problem. Data drives the ranking.
Most utilities score each outage ticket on a blend of factors:
A simplified priority score might combine these into a single number the OMS updates in real time.
# Illustrative outage triage score (not a production formula)
def priority_score(customers, has_critical, is_safety_hazard, est_repair_hrs):
score = customers
if has_critical:
score += 500 # bump for hospitals, water, emergency
if is_safety_hazard:
score += 1000 # safety overrides almost everything
score = score / max(est_repair_hrs, 0.5) # favor quick, high-impact fixes
return round(score, 1)
# Our storm transformer: 40 customers, no critical loads,
# no live safety hazard, ~3 hours to replace
print(priority_score(40, False, False, 3)) # -> 13.3The point is not the exact formula. It is that reliability is now a ranked, data-drivendata-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.Voir la définition complète → queue, not a first-call-first-served list. Our transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → with 40 customers waits behind a downed wire and a hospital feeder, exactly as it should.
Every outage our transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → causes gets recorded, and those records roll up into regulated performance metrics. Two are near-universal:
Regulators in many jurisdictions tie these metrics to allowed revenue or penalties. A utility that lets SAIDI drift can face financial consequences. That makes accurate outage data a compliance issue, not just an operational one. The U.S. Energy Information Administration publishes national reliability data you can explore at the EIA electricity reliability page.
Note the data trap: if the OMS misattributes outages or crews close tickets sloppily, the metrics lie. Many utilities now audit their outage records specifically to keep SAIDI and SAIFI defensible.
Vérification des acquis
1. A breaker trip is registered instantly, but the system reporting it cannot tell how many customers lost power. Which system produced this signal, and what does its limitation reveal about its role?
2. When 40 meters behind a single transformer go dark simultaneously, the OMS predicts one fault at the transformer or its fuse rather than 40 separate problems. What core capability makes this inference possible?
3. Why does the OMS still treat customer calls and app reports as a real input despite the widespread deployment of smart meters?
4. A 'last gasp' message from a smart meter is best understood as which kind of signal?
5. Select ALL correct answers about the distinct roles of SCADA and the OMS in reliability decisions.
Sélectionnez toutes les réponses correctes.
6. Select ALL correct answers about the inputs the OMS uses to infer the likely location of a fault.
Sélectionnez toutes les réponses correctes.
The most valuable use of this data comes after the storm. Every failure is a data point about which assets to replace before they fail.
Utilities manage tens of thousands of transformers. They cannot replace them all, and regulators scrutinize how capital gets spent. So they rank replacement candidates by combining two dimensions:
Multiply the two and you get a risk score. High probability plus high consequence equals top of the replacement list. This is the core of what the industry calls asset performance management (APM).
Our storm transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète →, if it had a poor health score and sat on a feeder serving a clinic, would rank high. A healthy transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.Voir la définition complète → serving a few rural customers ranks low, even if it is old.
The storm did two things for the capital plan. First, it confirmed a real failure, which is stronger evidence than a model prediction. Second, it revealed which weak assets failed under stress, sharpening the health models for similar units.
Good utilities close this loop. Field crews record the actual condition of what they replace ("failed bushing," "moisture in oil") and feed it back. Over time, the health index gets more accurate because it learns from real failures, not just sensor thresholds.
You do not need to read DGA charts to use this. The key managerial question is: does our capital spending follow risk, or does it follow the squeaky wheel? Data-drivenData-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.Voir la définition complète → utilities can defend every replacement with a risk score. Others replace what failed most recently or what a senior engineer worries about. Regulators increasingly expect the former.
Be realistic about limits. Three gaps are common:
The best organizations treat data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → as core infrastructure, not an IT afterthought.