Li-Ion Batteries Releasable Capacity Estimation with Neural Networks on Intelligent IoT Microcontrollers

Abstract

Lithium-Ion (Li-Ion) batteries are gaining remarkable popularity, due to their chemical ability to maximize battery life while increasing power energy density. These rechargeable batteries are widely used in mobile computing devices, such as smartphones and smartwatches, and automotive systems, such as hybrid and electric vehicles. The estimation of the releasable capacity allows the computation of the State of Health (SoH) of a battery, e.g. a measure of its functionality in energy storage and delivery, which is a fundamental parameter for the battery health monitoring. Several prognostic analysis approaches use machine learning algorithms, such as Support Vector Machines (SVMs), Random Forest regression and Artificial Neural Networks (ANNs). In this paper, we compare different machine learning algorithms for predicting maximal releasable capacity of Li-Ion batteries by analysing accuracy versus complexity, with special focus on implementing ANN on resource constrained microcontrollers (MCUs). In particular, Forward Neural Networks (FNNs) and Recurrent Neural Networks (RNNs) are compared. These approaches are applied on the Litium-Ion battery prognostic datasets made available by the National Aeronautics and Space Administration (NASA). Complexity is profiled on STM32 microcontrollers (MCUs) by using the toolset X-CUBE-AI, which automatically converts pre-trained ANNs and generates optimized and validated ANSI C code for STM32.