Self-learning models based on artificial neural networks can be used to analyze the state of health of Li-ion batteries and predict their lifetime.

Lithium-ion (Li-ion) batteries have become the most promising solution for energy storage in various applications, from laptops and smartphones to electric vehicles and grid storages. The cost of Li-ion batteries is however still significant, making the battery lifetime critical for reaching profitability in many applications.

Due to the complexity of these electrochemical devices there is currently a lack of knowledge on the aging processes in the Li-ion battery, which again restricts the development of accurate lifetime prediction models. To overcome this challenge, accelerated aging tests in combination with data-driven models –extrapolating the aging test results to real life conditions – are needed to provide a lifetime model that can be used to develop operational strategies.

Post doc Mohsen Vatani (IFE) are developing such mathematical and data-driven aging models for Li-ion batteries in MoZEES. One of these models are the artificial neural network (ANN). ANN is a mathematical tool with the adaptability and self-learning ability to represent complex nonlinear models of systems that often lack scientific background knowledge.  The main idea behind the ANN concept is to imitate the neural network of the human brain. Each ANN is made up of some neuron sets which learn through examples and generalizes without any prior knowledge about the data nature and interactions.

In his work, two different commercially available types of Li-ion battery cells (NCA and LFP chemistry) are cycled under different conditions (temperature, starting state of charge, depth of discharge, and discharge current rate). The capacity loss of the cells is measured with regard to the amount of the charge that has passed through the cells (total Ampere hours). The data from these accelerated cycling tests are then used to develop the ANN-based model for each cell chemistry, which then can predict one-step ahead state of health of the cells cycled under different conditions.

In a recently published paper presented at the “IEEE PES Innovative Smart Grid Technologies Conference Europe”, the ability of ANN to predict lifetime and analyze the state of health of Li-ion batteries are validated. The model for instance confirms that battery cells have the highest degradation rate when they are charged in low ambient temperatures.