Predicting Li-ion Battery SOC, SOH and RUL with Nuvoton M55M1 Edge AI MCU
The Nuvoton M55M1, featuring Arm Cortex-M55 CPU and Ethos-U55 NPU with 110 GOPS compute, enables real-time on-device inference of LSTM and CNN models to accurately predict battery SOC, SOH, and RUL — delivering intelligent battery management for EVs, energy storage, and consumer electronics.
Background
Battery health is critical across modern electronics and electric vehicles, yet accurately knowing a battery's true state remains a challenge. Why does a phone shut down with charge remaining? Why do EV ranges fall short of expectations? The answer lies in the difficulty of precisely measuring battery state. Nuvoton's M55M1 AI MCU addresses this with powerful edge inference, enabling accurate prediction of State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL).
Edge AI Models
Edge AI estimation feeds raw charging voltage curves directly into models without relying on manual expertise. M55M1 supports:
- LSTM (Long Short-Term Memory): Ideal for battery time-series data, LSTM retains memory of hundreds of prior samples to capture long-term aging trends.
- CNN (Convolutional Neural Network): Treats voltage-capacity curves as images to extract subtle local features — such as micro-peaks on IC curves invisible to the human eye.
M55M1 Key Advantages
- Powerful AI Processing: Arm Cortex-M55 CPU + Ethos-U55 NPU at 220MHz delivers 110 GOPS, running INT8 quantized models locally with <100ms latency.
- Rich Data Acquisition: Dual 12-bit 5 MSPS SAR ADCs and dual 12-bit 1 MSPS DACs for high-precision voltage, current, and temperature sensing; UART/SPI/I2C interfaces for seamless BMS integration.
- Low Power: 1.6V–3.6V operating voltage, -40°C to +105°C range; on-device inference eliminates cloud dependency and reduces system power consumption.
Technical Implementation
- Data Acquisition: Real-time ADC sampling with built-in DSP filtering and feature extraction.
- Model Deployment: TensorFlow Lite Micro support; SOH prediction error within ~4% on NASA public datasets.
- Real-time Feedback: Continuous SOH calculation fed to BMS; triggers alerts or adjusts charge/discharge strategy when SOH drops below 80%.
Applications
- Smart BMS: Core controller for real-time battery health monitoring and lifespan prediction.
- EV & Energy Storage: Accurate health assessment and optimized charge strategies to extend battery life.
- Consumer Electronics: Precise battery management to help users avoid habits that accelerate degradation.
Learn More
Interested in edge AI solutions for battery SOC, SOH, and RUL estimation? Contact Nuvoton to learn more.
Content sourced from Nuvoton MCU official WeChat account.