Predictive Maintenance for EV Chargers

Relay wear: rising ON-current relative to session start indicates contact resistance building up. Sessions that used to start at 32A now settle at 30A; the relay is degrading. Left alone, it welds.

RCD nuisance trips: repeated transient trips without a persistent fault cluster before hard failures of the RCD module or the load side it protects. Count them per week; when the count climbs, schedule maintenance.

Connector temperature drift: rising temperature across similar session profiles suggests connector wear or contamination. Left alone, it hits thermal cut-out and then physical damage.

Comms flapping: WebSocket reconnect rate is a leading indicator of network-side degradation, not just charger-side faults. Rising rate warrants investigation before it becomes total loss.

Detection of these signatures across a fleet, per-charger health scoring based on the pattern, alerts routed to your operations team before the charger fails a session. Fault analytics with root cause classification when something does fail.

This is fault analytics: it reads the telemetry, classifies the pattern, and tells you what to do. It is not oracle-grade prediction. If a vendor claims otherwise, ask them to demonstrate on chargers they didn't train on.

Real prediction, 'this charger will fail in 12 days', needs labelled failure data across enough vendors, climates and duty cycles to generalise. That data does not exist yet in Indian conditions, which is why RIOD runs an open research program.

If you operate a charging network and would like to contribute anonymised failure data in exchange for early model access, the research program page describes how. This is the honest path forward.

• Labelled fault events with timestamps and root cause
• Charger vendor, model, and firmware version at time of fault
• Session duty cycle, energy delivered and session count history
• Climate: ambient temperature, humidity, dust exposure
• Site electrical profile: voltage stability, harmonics, phase imbalance
• Repair outcome and time-to-restore for each event

• Analytics finds known patterns in telemetry after the fact.
• Prediction estimates future failure probability with a defensible confidence.
• Most 'predictive maintenance' features shipped today are analytics rebranded.

• Too many false positives waste technician time and erode trust in the alerts.
• False negatives hurt uptime and drive angry driver reports.
• Models need continuous feedback from the field to stay calibrated.