All case studies

Predictive maintenance

Covering 90% of failures by focusing on 60% of the chargers

At an Indian fleet charging hub, R-One ranked every DC charger by failure risk, and showed which 20% could safely be left alone.

Client
Indian fleet operator
Region
India
Estate
DC fleet charging hub
  • 90%

    of failures came from the 60% of chargers R-One ranked highest for the next 7 days

  • 2%

    of failures came from the quietest 20% of chargers

  • 7–10x

    more likely to hit a real failure than picking chargers at random

The challenge

The hub’s DC chargers were down 3.5% of the time on average. Small on paper, but a fleet plans charging around schedules, so even short outages are felt.

Maintenance was fix-on-fail, so each failure meant an operational impact before it meant a repair.

Some faults were hard to see: chargers that fault again and again, and sessions that never start, without a clear outage to point at.

A charger counted as down when it reported Faulted or Unavailable, or went quiet for 45 minutes.

What R-One did

  1. Read the operating record

    R-One worked from data the operator already had: charger messages and fault history across several charger makes, usage and work orders. There was nothing new to install.

  2. Surface the invisible faults

    Quiet problems that never cause an outage on their own, but wear a charger down, were flagged separately from the alarms the CMS already raises.

  3. Score every charger on two horizons

    Each charger got a failure risk for the next 7 days and for the next 21 days, sorted into five tiers from very high to very low.

  4. Test it on data the model had not seen

    The model learned from most of the history and was scored on a held-out part it had never seen, so the results reflect what it would have said in advance.

What it found

Ranked by risk for the next 7 days, the top 60% of chargers accounted for 90% of failures. Over the longer 21-day view, the top 20% carried 52%.

Failures concentrate in a few chargers

  • Very high risk10% of chargers, 19% of failures.
  • High risk10% of chargers, 19% of failures.
  • Medium risk40% of chargers, 52% of failures.
  • Low risk20% of chargers, 10% of failures.
  • Very low risk20% of chargers, 2% of failures.
Share of chargers and share of failures in each risk tier, next 7 days. Held-out test data.

Far better than picking at random

  • Picking chargers at random9%
Chance that a flagged charger fails within the window. R-One is 7x to 10x better than random selection.

30%

of downtime tied to invisible faults

R-One’s invisible-fault engine had detected the underlying problem behind 30% of downtime. 55% of its alerts lined up with real downtime.

What this changes

  • Look first where the failures are

    Most of the estate’s failures sat in a defined majority, and the ranking says which chargers come first.

  • Leave the quietest 20% alone

    They produced 2% of failures, so scheduled checks there can wait.

  • Move from fix-on-fail to fix-before-fail

    A forecast gives operations time to plan the repair around the fleet’s schedule.

These results are a back-test on the operator’s own history: R-One scored data it had not been trained on, and the outcomes were checked against what actually happened.

See what R-One finds in your network.

Share your charger history and we will show you your own risk ranking.