TinyML based monitoring and diagnosis system to enable the European battery digital passport

The European Union mandates comprehensive battery digital passports for EV and industrial batteries above 2 kWh capacity starting February 2027. This paper presents a TinyML-based monitoring system that enables real-time battery parameters estimation to support digital passport requirements using voltage (V), current (I), and temperature (T) measurements. Two neural network architectures were evaluated: an Artificial Neural Network (ANN) with 6.5K parameters and a Convolutional Neural Network (CNN) with 24.5K parameters. Both models were quantized to 8-bit precision and deployed on an STM32F413ZHT6 microcontroller with ESP32-WROOM-32E for cloud connectivity. Experimental validation was performed using a case study of State of Charge estimation for a 20 Ah Lithium Iron Phosphate battery. The ANN model achieved superior performance with a Mean Absolute Error below 5.3% and a Root Mean Square Error below 6.7%, while consuming only 6.36 kB of flash memory and 0.35 kB of scratch memory. The quantized models showed minimal accuracy degradation compared to full-precision versions. The proposed system provides a cost-effective solution for continuous battery monitoring, supporting regulatory compliance while enabling predictive diagnostics and enhanced safety management throughout the battery lifecycle.