Predictive analytics to prevent outages

Maintenance scheduled by the calendar, not by condition

A small utility still has to maintain equipment across its whole service territory, and it has to do it without the staff of a large one. Equipment histories, sensor telemetry, and location context all existed, but they sat apart from one another. Deciding what to inspect next meant reasoning across those sources by hand, and a fixed maintenance interval treats a healthy asset and a degrading one the same way. Problems that could have been caught early became outages instead.

Two kinds of model over one asset base

Fulton Ring built the forecasting on the utility’s existing time-series, asset, and spatial data. Time-series models look for the drift and instability in sensor history that tend to precede a failure. Spatial models add the context a single sensor cannot see, including how an asset’s surroundings and its neighbors are behaving. Together they produce a risk score for each asset rather than one global threshold applied to everything.

We delivered this as a service, which is what made it viable at this size. Fulton Ring ran the pipeline and the models, so the utility got current forecasts without hiring a data science team to keep them working.

A ranked queue, not an automated decision

Predictions ranked assets for review. A high score meant inspect sooner, not that a failure was certain. Maintenance specialists kept the decision, because they carried the safety responsibility and they knew the equipment. That boundary is what made the forecasts usable: a planner can act on a ranked queue in a way nobody can act on an unexplained alarm.

The purpose of the work was fewer unplanned outages and steadier service. We have not published figures for avoided outages or cost, because the ones we could stand behind are the client’s to release.