Case study 03 · Asset management
A Fortune 500 utility trained asset-health models on the signals its substations were already producing — dissolved-gas readings, loading, thermal profiles, maintenance history — to flag developing transformer faults while there was still time to plan, not react.
An anonymized account of work led by Corvana's founding team with a Fortune 500 utility in New England. Outcomes are rounded; customer identity is withheld.
A power transformer rarely fails without warning — the warnings just live in places nobody reads together: a slow drift in dissolved-gas analysis, a loading pattern that changed last summer, a thermal signature that stopped matching its twin in the next bay, a maintenance history with a gap where a test should have been.
When one of them does fail unplanned, the bill compounds fast: the replacement itself, emergency crews, switching and load transfers, regulatory reporting — and if the asset is critical enough, customer outages that end up in front of the commission. The fleet was aging faster than the replacement budget, and the maintenance program was running on calendar intervals, not condition.
Models that read what the substations were already saying, and a workflow that treated every prediction as a claim for an engineer to judge:
The models never touched operations — they read from historians and test systems, above the operational boundary, and wrote to a watch list, not to any control. Every recommendation was a surfaced candidate with its evidence attached; asset engineers stayed the named decision-makers on every test, deferral, and replacement, and the model's miss rate was tracked openly against outcomes.
developing faults flagged early enough in the first year to plan the response instead of reacting to a failure.
avoided cost across those events versus unplanned failure — replacement logistics, emergency response, and outage exposure.
fewer routine truck rolls as calendar-based checks shifted to condition-based ones on the covered fleet.
The quieter change was in the rate-case conversation: a maintenance program that could show why each dollar went where it did — condition, evidence, engineer sign-off — is a program a commission can follow.
Outcomes from the founding team’s engagement are rounded and anonymized. Model scope, validation approach, and baseline definitions are available in a review conversation.
A fixed-fee, six-week Proof of Value replays your substation history — the faults already on the record are the baseline — and reports what the early-warning models would have flagged, and when, with every action decided by your engineers.
Book a review