Skip to content
Meet us at CIGRE Paris Session 2026, August 23–28
Eneryield
All news
Insights

Preventing Substation Fires with AI Fault Prediction

Rare, but catastrophic

Substation fires can take out supply for tens of thousands of customers, destroy assets worth millions and endanger personnel. They are rare events — which is exactly why purely statistical maintenance schedules don’t protect against them.

Fires announce themselves electrically

Most substation fires start from a small set of electrical failure modes: insulation breakdown, overheating connections, partial discharge and evolving earth faults. Each of these produces characteristic signatures in current and voltage waveforms — often weeks before thermal runaway:

  • Partial discharge shows up as repetitive micro-transients clustered around voltage peaks.
  • Degrading insulation produces intermittent earth faults that self-clear in milliseconds — invisible to protection relays, visible in raw waveforms.
  • Loose or corroded connections modulate current with characteristic patterns as contact resistance varies with temperature.

Why reactive and calendar-based maintenance fall short

Traditional approaches inspect substations on a fixed schedule and respond to failures after they happen. But the failure modes behind substation fires don’t follow the calendar — a joint can degrade from healthy to critical between two inspection rounds, and a reactive strategy by definition arrives after the damage. Predictive fault analytics inverts this: the substation’s own disturbance recordings are screened continuously, so the maintenance visit happens when the data says the risk is rising — not when the schedule says it’s due.

Why measurement resolution matters

These signatures live in the details. At standard sampling rates they blur into noise; at megahertz resolution they are unmistakable. This is why our Härryda Energi deployment streams waveforms at 2 MHz — millions of measurements per second — and why the models behind IntelliView® are trained to recognize the precursors, not just the failures.

From detection to prevention

When the platform flags an evolving fault with its root cause and location, a planned inspection replaces an emergency. In the best case, nobody outside the maintenance team ever knows how close the grid came to a very bad day.

See IntelliView® in action

A 30-minute demo on real grid data — and a conversation about your network.