Design and implementation of a low-cost battery management system using MLP-FCNN

Today, batteries used in many areas, especially electric vehicles, have gained great importance with the depletion of fossil fuels. The decrease in fossil fuel reserves has accelerated work on energy storage technologies and increased the need for the development of battery systems. Battery Management Systems (BMS) play a critical role in monitoring battery packs, which are formed by series-parallel connections of battery cells, in a safe, efficient and economical manner. The BMS monitors the battery pack in real-time, tracking important parameters such as overcharge/discharge status, state of charge (SOC) and cell balancing; thus ensuring effective use of battery life. Through these systems, users are protected from costs arising from battery failures and energy efficiency is increased. In this thesis study, an MLP-FCNN based and low-cost BMS has been designed and implemented for a sixteen-cell Lithium Iron Phosphate (LiFePO4) battery pack. The original contribution of this study is the successful real-time SOC estimation by customizing the Fully Connected Neural Network (FCNN) architecture appropriately for the project using the X-CUBE-AI library on the STM32F072RBT6 microcontroller with ARM® Cortex®-M0 based 128 kilobyte (kB) memory capacity with limited hardware. While many studies in the literature focus on platforms requiring high memory and processing power, in this thesis, deep learning has been made applicable in embedded systems at low cost by reducing model size and memory usage. The designed BMS measures cell voltages, current and temperature through the BQ76952 integrated circuit; it also performs passive cell balancing. The system offers high performance in terms of accuracy and cost, presenting an effective solution for industrial applications and low-budget prototypes.