Outage and asset data: turning sensor signals into reliability decisions

# 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.View full definition → 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.View full definition →'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.View full definition → gets replaced next quarter or limps through another summer. This lesson follows those signals from the field to the decision.

The three data streams behind every outage

Utilities do not "see" the grid directly. They see data. Three systems dominate reliability decisions, and each answers a different question.

SCADA: what is happening right now

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.View full definition →'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.

Outage Management System (OMS): who is affected

The OMS translates equipment events into customer impact. It pulls from three inputs:

• Smart meter "last gasp" messages: many modern meters send a final signal when they lose power, like a digital flatline.
• Customer calls and app reports: still a real input, especially where meter coverage is thin.
• The connectivity model: a mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → of which meters sit downstream of which transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.View full definition →, fuse, and feeder.

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.View full definition → 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.View full definition → or its fuse, not 40 separate problems. This inference is the core value: it turns noise into a single dispatchable ticket.

Asset health data: how bad is it, really

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.View full definition → 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.

Fusing the streams during the storm

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.View full definition → 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.View full definition → failure with a bad health history needs a replacement unit, not just a repair crew.

Prioritizing crews: the dispatch math

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:

• Customer count: more affected customers, higher priority.
• Critical customers: hospitals, water pumping stations, emergency services get flagged automatically.
• Safety hazards: a downed live wire jumps the queue regardless of customer count.
• Estimated restoration effort: a fuse swap is faster than a transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.View full definition → replacement.

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))   # -> 13.3

The 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.View full definition → 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.View full definition → with 40 customers waits behind a downed wire and a hospital feeder, exactly as it should.

Feeding reliability metrics regulators watch

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.View full definition → causes gets recorded, and those records roll up into regulated performance metrics. Two are near-universal:

• SAIDI (System Average Interruption Duration Index): the average total minutes a customer is without power per year.
• SAIFI (System Average Interruption Frequency Index): the average number of interruptions a customer experiences per year.

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.

From single outage to capital plan

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.

Asset health scores drive capital priorities

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:

• Probability of failure: derived from the asset health index, age, DGA trends, and fault history.
• Consequence of failure: how many customers, whether critical loads, cost to replace, and safety exposure.

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.View full definition →, 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.View full definition → serving a few rural customers ranks low, even if it is old.

Why the storm data matters here

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.

The non-technical takeaway for decision-makers

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.View full definition → 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.

Where the data still falls short

Be realistic about limits. Three gaps are common:

• Model coverage: many older transformers have no online sensors. Their health is estimated from age and load, which is far weaker than live DGA.
• Connectivity errors: if the OMS mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → of which meter sits behind which transformertransformerA Transformer is a neural network architecture that uses self-attention to process sequences in parallel, powering most modern language and generative AI models.View full definition → is wrong, fault prediction fails silently. Keeping this model current is unglamorous and constant work.
• Data quality under stress: storms are when data floods in and when it is least reliable. Duplicate tickets, delayed meter messages, and manual overrides all creep in.

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.View full definition → as core infrastructure, not an IT afterthought.

Key Takeaways

• Three streams, three questions: SCADA tells you what happened, the OMS tells you who is affected, and asset health data tells you how serious the equipment condition is. Reliability decisions come from fusing all three.
• Dispatch is data-driven triage: outages are ranked by customer count, critical loads, safety, and repair effort, not by call order.
• Outage data is a compliance asset: it feeds SAIDI and SAIFI, which regulators tie to revenue and penalties, so accuracy has direct financial stakes.
• Capital follows risk, not noise: multiplying probability of failure by consequence of failure produces a defensible replacement priority list.
• Close the loop: feeding real field failure findings back into health models is what separates utilities that predict failures from those that only react to them.