Every EV scooter startup in India knows the numbers. Build the scooter for 80,000 rupees, sell it for 1,20,000, and then spend 15,000 to 25,000 per vehicle in warranty repairs over the next two years. Service centers are overflowing. Spare parts are backordered. Riders are frustrated. And social media complaints are destroying brand trust faster than marketing can rebuild it.
The root cause is not bad engineering. Most EV scooters are well designed. The problem is that OEMs have no idea what is happening to the vehicle after it leaves the factory. The only telematics most scooters have is GPS for the mobile app. GPS tells you where the scooter is. It tells you nothing about how the scooter is doing.
Self-diagnostic AI changes this completely. Instead of waiting for the rider to complain or for a component to fail under warranty, the scooter tells you what is going wrong before anyone notices.
The warranty cost problem for EV OEMs
Let us look at the numbers that keep EV founders up at night. A typical EV scooter OEM selling 10,000 units per year spends 15 to 25 crore rupees annually on warranty and service. The top three cost drivers are motor controller replacements (25 percent of warranty spend), battery module replacements (35 percent), and brake system repairs (15 percent).
Here is what makes it worse. Most of these failures develop over weeks or months. A motor winding that overheats under certain load conditions does not fail on day one. It degrades gradually until one day the controller throws a fault code and the scooter stops. By then, the motor is damaged and needs replacement.
If the OEM had known that the motor was running 12 degrees hotter than normal three weeks ago, a simple firmware adjustment or a service visit to check the bearing alignment could have prevented a 4,000 rupee motor replacement.
This is not hypothetical. This is the exact pattern we see in EV fleet telematics deployments where OEMs who monitor vehicle health in real time spend 30 to 40 percent less on warranty than those who wait for complaints.
What self-diagnostic AI actually means
Self-diagnostic AI is not a single feature. It is a layer of intelligence built into the scooter telematics unit (TCU) that continuously monitors component health and detects problems early. Here is what it monitors and why.
Motor temperature profiling. The TCU reads motor winding temperature and correlates it with current draw, speed, ambient temperature, and ride duration. A healthy motor has a predictable thermal profile under any given load condition. When the actual temperature starts deviating from the expected profile, the AI flags it. Common causes: bearing friction increasing, coolant flow reduced, winding insulation degrading, or controller PWM duty cycle drifting.
Battery cell imbalance detection. The BMS reports individual cell voltages to the TCU. Self-diagnostic AI does not just check if cells are within spec. It tracks how cell voltages diverge during charging and discharging over time. A healthy pack has cells that stay within 20mV of each other. When one cell starts drifting, the AI detects it weeks before the BMS balancing algorithm cannot keep up, and months before the rider notices reduced range.
Brake wear pattern analysis. Regenerative braking data combined with mechanical brake activation frequency tells you exactly how the brakes are wearing. If the rider uses mechanical brakes more than expected (suggesting regen is weak), or if braking force drops for the same lever pressure (suggesting pad wear), the AI flags it. This prevents the dangerous scenario where a rider discovers worn brakes in traffic.
Controller health monitoring. Motor controller failures are the most expensive single warranty item. The AI monitors controller temperature, switching frequency patterns, and output current waveforms. A MOSFET starting to degrade shows subtle changes in switching characteristics long before it fails completely.
The architecture: from sensor to service alert
Here is how self-diagnostic AI works end to end in an EV scooter.
The scooter already has sensors. Every EV scooter has a BMS with cell voltage and temperature sensors, a motor controller with current and temperature sensors, and throttle and brake sensors. You do not need to add new hardware. You need to read the data that is already there and do something intelligent with it.
The telematics control unit (TCU) is the brain. This is a small compute module connected to the scooter CAN bus. It reads data from the BMS, motor controller, and other ECUs every 100 milliseconds. Modern TCUs based on ARM Cortex-A processors or even ESP32 modules have enough compute to run lightweight ML models on-device.
Edge inference runs on the TCU. The self-diagnostic models run locally on the TCU. They do not need a cloud connection to detect problems. A TinyML model for anomaly detection uses under 200KB of memory and processes inference in under 50 milliseconds. The model compares current sensor patterns against learned baselines for that specific vehicle.
Cloud aggregation provides fleet intelligence. The TCU sends diagnostic summaries to the Akran IQ cloud platform via cellular or WiFi (when the scooter is parked near a charger). The cloud does not get raw sensor data. It gets health scores, anomaly alerts, and trend data. This keeps data costs low while giving the OEM a complete fleet health picture.
The service alert reaches the right person. When the AI detects a developing issue, the platform generates a service alert with the vehicle ID, the problem description, the severity, and the recommended action. This goes to the OEM service dashboard and optionally to the rider via the mobile app.
Motor temperature: the key section
Motor failures account for a quarter of all warranty costs. Here is exactly how the Akran IQ platform monitors motor temperature and predicts failure before the rider notices.
During the first 500 kilometers, the system learns the thermal fingerprint of each motor. It maps motor temperature against a matrix of conditions: speed, current draw, ambient temperature, ride duration, and terrain gradient (derived from GPS altitude changes). This creates a multi-dimensional baseline unique to that vehicle.
After baseline learning, the system runs a continuous comparison. Every 10 seconds, it predicts what the motor temperature should be given the current operating conditions. If the actual temperature exceeds the prediction by more than a dynamic threshold (typically 5 to 8 degrees Celsius, adjusted for confidence), it increments an anomaly counter.
A single anomaly is noise. Urban riding with frequent stop-and-go can cause occasional spikes. But when the anomaly counter crosses a trend threshold (typically 50 events in 7 days), the system generates a predictive alert: "Motor thermal anomaly detected. Estimated severity: moderate. Recommended action: inspect motor bearing alignment and coolant flow."
The rider has not noticed anything yet. The scooter still works fine. But in 3 to 6 weeks, without intervention, that motor would overheat during a long ride and either trigger a thermal cutoff (leaving the rider stranded) or cause permanent winding damage (requiring a full motor replacement under warranty).
By catching it early, the OEM schedules a 30-minute service visit instead of a 4,000 rupee motor replacement. Multiply this across 10,000 vehicles and the savings are massive.
Brake wear prediction in practice
Brake problems are dangerous and expensive. Traditional approach: wait for the rider to complain about weak braking or for a service technician to inspect pad thickness during a scheduled service visit. By then, the pads might be metal-on-metal, the rotor is scored, and the repair bill triples.
Self-diagnostic AI monitors brake health continuously using data already available on the CAN bus. Regenerative braking energy recovery rate. When regen efficiency drops, it means either the motor controller regen calibration is off or the battery is too full to accept charge. Both are actionable.
Mechanical brake activation frequency and force. If the rider is pressing the brake lever harder for the same deceleration, pad thickness is decreasing. If brake activation frequency increases while regen recovery stays constant, the rider is compensating for weaker regen by using mechanical brakes more.
Brake temperature (if equipped). Some scooters have brake disc temperature sensors. Rising baseline temperature indicates pad glazing or caliper drag.
The AI combines these signals into a brake health score that degrades gradually. When the score drops below a threshold, the OEM gets an alert and can proactively reach out to the rider: "Your scooter is due for a brake inspection. Book a free service visit." The rider is impressed. The OEM avoids a warranty claim. Everyone wins.
The business case for OEMs
Here is the math that matters to an EV scooter OEM.
Adding self-diagnostic AI costs approximately 500 to 1,200 rupees per vehicle. This covers the TCU hardware upgrade (if the existing unit cannot run ML models), the platform license for cloud aggregation, and the cellular data plan for diagnostic uploads.
The savings per vehicle over a 2-year warranty period are 5,000 to 10,000 rupees. This comes from catching 60 to 70 percent of major failures early and converting them from expensive replacements into cheap preventive service visits. Motor replacements drop by 40 percent. Battery module replacements drop by 25 percent (because cell imbalance is caught early). Brake system repairs drop by 50 percent.
For an OEM selling 10,000 scooters per year, that is 5 to 10 crore rupees in annual warranty savings against a 50 lakh to 1.2 crore rupee investment. The ROI is 5x to 10x in the first year alone.
But the real value goes beyond cost savings. Proactive service builds brand loyalty. A rider who gets a call saying "we detected a potential issue and have already ordered the part for your free service visit" becomes a brand evangelist. Compare that to a rider stranded on the road with a dead scooter waiting for roadside assistance.
What OEMs need to implement this
If you are an EV scooter OEM and want to add self-diagnostic AI to your fleet, here is what you need.
A telematics control unit with enough compute for edge ML. If your current TCU is a basic GPS tracker, it needs an upgrade. An ARM Cortex-A class processor or a capable microcontroller like the ESP32-S3 with CAN bus interface is sufficient.
CAN bus access to BMS, motor controller, and brake ECU data. Most scooters already have this internally. The TCU just needs to be connected to the CAN bus and configured with the right DBC file for your vehicle.
A cloud platform for fleet aggregation and service workflow. The Akran IQ platform provides the complete stack: device management, data processing, diagnostic dashboards, and service alert workflows. We handle the infrastructure so your team can focus on building scooters.
ML model training and deployment. We work with your engineering team to build the diagnostic models using data from your test fleet. The models are trained on your specific motor, battery, and brake configurations. Once validated, we deploy them to your production fleet via OTA updates.
The typical deployment timeline from kickoff to production fleet rollout is 6 to 8 weeks. Talk to our team to start the conversation.
Beyond warranty: the data advantage
Self-diagnostic AI gives OEMs something they have never had before: real-world usage data at scale. You learn how riders actually use the scooter, which components degrade fastest in which conditions, and where your next-generation design should improve.
This data feeds back into R&D. If you discover that motor temperatures are consistently higher in coastal cities (salt air corroding bearing seals), you can specify better seals in the next hardware revision. If you find that battery degradation accelerates in vehicles used for food delivery (high cycle count, deep discharges), you can offer a heavy-duty battery option for commercial riders.
The OEMs that will dominate the Indian EV scooter market in the next five years are not the ones with the cheapest vehicle. They are the ones who know their vehicle best after it leaves the factory. Self-diagnostic AI is how you get there.