Project Ideas for Edge AI in Automotive Industry: BMS + EV Charging
🔋 Project Idea 1: Edge AI-Based Battery SoH Estimation and Failure Prediction System
🎯 Goal:
To build a real-time SoH (State of Health) monitoring system for EV batteries that predicts battery degradation, identifies thermal or voltage imbalances, and prevents sudden failures — especially in fleet or high-mileage vehicles.
🧠How Edge AI Helps:
Unlike traditional BMS logic that uses look-up tables or fixed thresholds, Edge AI learns the unique degradation pattern of each battery pack over time. It correlates temperature, charge/discharge cycles, and load history to estimate remaining battery life and predict early signs of failure, running directly on the ECU or edge controller inside the vehicle.
📥 System Requirements – Input Values Needed
| Sensor / Value | Source | Purpose |
|---|---|---|
| Cell voltage & current | BMS or shunt sensors | Identify cell-level imbalance |
| Temperature of cells/modules | Thermistor or RTD | Detect hotspots and thermal drift |
| SoC history and cycle count | BMS log | Model degradation per usage |
| Charging/discharging power | HV line current sensors | Understand load impact on aging |
| Vehicle speed & terrain data | CAN bus or GPS | Optional for usage stress correlation |
🧠Edge AI Model Used
| Functionality | AI Model Type |
|---|---|
| SoH Estimation | LSTM Regression or Gradient Boosted Trees (GBT) |
| Degradation Pattern Learning | Time-series CNN on voltage/temp cycles |
| Thermal Runaway Prediction | Random Forest or Anomaly Detector (Autoencoder) |
| Remaining Useful Life (RUL) | Survival Analysis Model (Cox, Weibull) |
âš¡ Project Idea 2: Edge AI Smart EV Charger for Load Scheduling & Tariff Optimization
🎯 Goal:
To develop an AI-enabled smart EV charging controller that optimizes charging time based on real-time tariffs, grid conditions, and vehicle SoC, while also coordinating across multiple chargers in a fleet or residential setup.
🧠How Edge AI Helps:
The edge controller predicts the best time to charge the vehicle considering the current load, solar availability, and time-of-use (TOU) tariff. It uses reinforcement learning to manage multiple chargers in a cost-efficient and grid-friendly manner, while ensuring that vehicles are charged before their departure schedule.
📥 System Requirements – Input Values Needed
| Sensor / Value | Source | Purpose |
|---|---|---|
| Vehicle SoC | EV CAN or OCPP handshake | Base charging target |
| Real-time energy price (TOU) | Utility API | Optimize cost |
| Site load (total and per EVSE) | Smart meter / CTs | Prevent overloading |
| Solar generation (if available) | Inverter or irradiance sensor | Balance grid and renewable power |
| User schedule / departure time | User input / fleet server | Prioritize based on urgency |
🧠Edge AI Model Used
| Functionality | AI Model Type |
|---|---|
| Charging Time Optimization | Reinforcement Learning (DQN or PPO) |
| Load Forecasting & Peak Shaving | LSTM or Prophet time-series model |
| Multi-EV Charger Coordination | Multi-Agent Reinforcement Learning (MARL) |
| Priority Scheduling | Decision Tree + Constraint Solver |
🛡 Project Idea 3: Driver Behavior-Based Safety Risk Estimation System (Active Safety AI)
🎯 Goal:
To create an onboard safety monitoring system that tracks driver behavior in real time, detects risky actions (e.g., harsh braking, aggressive steering), and warns the driver or logs the event — supporting active safety and insurance risk profiling.
🧠How Edge AI Helps:
Edge AI models use IMU data, camera inputs, and vehicle signals to detect dangerous patterns before accidents happen. It runs on the vehicle’s local edge processor (IVI/ADAS ECU) without needing a cloud connection, ensuring low latency feedback and high reliability even offline.
📥 System Requirements – Input Values Needed
| Sensor / Value | Source | Purpose |
|---|---|---|
| Accelerometer & gyroscope data | IMU or smartphone sensor | Detect sudden braking, cornering |
| Brake, throttle, and steering angle | CAN signals | Analyze control behavior |
| Speed and GPS location | CAN + GPS module | Monitor driving context and zones |
| In-cabin camera feed (optional) | Driver-facing camera | Detect drowsiness, distractions |
| Weather/road condition (optional) | API or external sensor | Risk context enhancement |
🧠Edge AI Model Used
| Functionality | AI Model Type |
|---|---|
| Risky Behavior Detection | 1D-CNN on IMU + CAN features |
| Driving Pattern Classification | SVM or Decision Tree trained on labeled events |
| Drowsiness / Distraction Detection | CNN (Vision Model) with face landmarks |
| Risk Score Estimation | Multi-Input Regression model combining all modalities |
