Can a charger predict its own failure?
We think the answer is yes, for some failure modes, some of the time. Nobody in the industry can honestly claim more than that yet, including us. This page describes what RIOD is researching, what we have learned so far, and the partners we are looking for to take it further.
Reactive maintenance means the truck rolls after the damage.
An EV charger that fails in the field costs far more than its repair bill. The bay earns nothing while it is down, the driver who found it dead does not come back, and in most Indian deployments the failure is discovered by a customer, not the operator.
The industry's answer is "predictive maintenance", usually presented as a solved feature on a pricing sheet. From what we see in real deployments, it is not solved. Charger failures in India have causes that global failure models were never trained on: 170 to 270V grid swings, 90%+ monsoon humidity, 55 degree C enclosure temperatures, dust ingress, and connector handling that no European duty cycle anticipates.
We build the hardware, the firmware, and the cloud layer in-house, which means we can see every layer where a failure announces itself in advance. That is the research opportunity.
The data.
RIOD chargers stream OCPP events, internal temperatures, relay cycle counts, RCD trip events, voltage and current waveform summaries, and session-level anomalies from deployments across India. Our self-resettable RCD hardware also gives us something unusual: a labelled record of every trip event and whether it was transient or persistent.
Where we are. The program is at the instrumentation stage across participating sites. Early observations will be shared with partners as findings become defensible.
What we do not know yet. Which signatures generalise across charger models and vendors. How much lead time is real versus artefact. Whether prediction accuracy survives the noise of sites with poor earthing and unstable supply, which is exactly where prediction is most valuable.
Four kinds of partner, four different reasons to join.
Charge Point Operators
You have the failure history; we have the instrumentation. Partners get early access to the models, instrumented monitoring on their network, and first claim on any resulting product. We get labelled failure data across more sites, vendors, and climates than any single network owns.
Academic & research institutions
Reliability engineering, power electronics, and applied ML groups. Real field data (anonymised), hardware access, co-authorship on anything publishable. Standing student project scopes available.
Component manufacturers
Relay, contactor, connector, and power module makers who want field-degradation data on their parts in Indian conditions, in exchange for accelerated-ageing test collaboration.
Insurers & financiers
Charger uptime is becoming an underwritten risk. If you price warranties, breakdown cover, or infrastructure loans, failure prediction changes your models, and your claims data improves ours.
Five steps. No fee to participate.
Scoping call
What data you have, what failure modes hurt you most, what a win looks like.
Data agreement
Anonymisation, ownership and publication rights settled in writing up front.
Instrumentation
Your sites get monitoring firmware or retrofit sensing, depending on hardware.
Quarterly review
Findings shared with all partners. Negative results included.
First rights
Partners deploy validated models before anything becomes a commercial product.
Frequently asked, answered honestly.
Is this a product I can buy today?
No, and be wary of anyone selling one. What you can buy today is fault analytics: detection, diagnosis, and automated ticketing when something has already gone wrong, which our Operations Suite does. Prediction before failure is what this program exists to make real.
What do partners get that others will not?
Early model access, influence over which failure modes we prioritise, instrumented visibility into their own network, and first deployment rights. Later customers get a product; partners get to shape it.
Who owns the results?
Models trained on pooled data are RIOD's, deployed first to partners. Your raw data stays yours, is anonymised before pooling, and the agreement in step 2 puts that in writing before anything moves.
We run non-RIOD chargers. Can we still join?
Yes, and we specifically want mixed fleets. OCPP-level data alone carries useful signal, and cross-vendor coverage is exactly what makes the models generalise.
Will you publish the research?
Where partners agree, yes. Field reliability data from Indian deployments barely exists in the literature, and publishing it serves the whole industry, including us.
Talk to the team doing the work.
Write with a line about your network or lab. An engineer replies, not a sales sequence.