A lithium-ion (Li-ion) battery is an advanced battery technology that uses lithium ions as a key component of its electrochemistry. During a discharge cycle, lithium atoms in the anode are ionized and separated from their electrons. The lithium ions move from the anode and pass through the electrolyte until they reach the cathode, where they recombine with their electrons and electrically neutralize. The lithium ions are small enough to be able to move through a micro-permeable separator between the anode and cathode. In part because of lithium’s small size, Li-ion batteries are capable of having a very high voltage and charge storage per unit mass and unit volume. Over time, Li-ion batteries degrade, leading to a serious loss of their effective capacity. Notably, the degradation process of individual Li-ion batteries varies and depends on different factors that are often difficult to identify. This has been partly attributed to the nonlinearity of the degradation profiles of most Li-ion batteries, where the capacity of the batteries sharply drops beyond a certain threshold, leading to potential catastrophic failure. This is of great significance in safety critical applications like in the case of electric cars, where the information on the health of batteries could be used to optimize the system operations.
Without accurate prediction of the State of Health (SoH) of these batteries, the alternative solution is to practice preventive maintenance that involve replacing the batteries before reaching their end of service life. This approach is inappropriate as it is costly and results in more waste. Two primary indicators that are widely used in quantifying the health of Li-ion batteries are the SoH and the State of Charge (SoC) of the battery. For instance, when the SoH of a battery reaches a threshold of about 70 – 80%, the battery is deemed unsuitable for use in safety critical systems due to the strong decline in the battery capacity after surpassing the threshold value.
Whereas SoC can be generally predicted with high accuracy, accurate prediction of SoH remains a big challenge. With the growing demand for Li-ion batteries, the need for accurate methods/tools for efficient and accurate SoH predictions has attracted significant research attention. The proposed methods can be grouped into three: physics-based models that use the chemistry and physical knowledge behind the degradation of batteries, data-driven models that use machine learning techniques to model the complex dynamics of the batteries based on the available operation data, and hybrid models that comprise the characteristics of the two.
Herein, Dr Elisa Y.M. Ang and Dr. Yew Chai Paw from the Singapore Institute of Technology developed an intuitive linear prediction model for the accurate prediction and estimation of SoH of Li-ion batteries. The predictive algorithm only relied on the voltage-time discharge and temperature profiles to predict the current SoH of the battery. The model was validated by comparing it with the existing open-source battery data. The work is currently published in the, Journal of Energy Storage.
Results showed that the proposed algorithm provided an efficient and easy method for training the predictive health model and enhancing its accuracy in predicting the SoH of Li-ion batteries. The main advantage of the predictive algorithm lies in its feature design and data preparation, as it was based on a straightforward regression model that is much easier to train and use. It is relatively simple and demonstrates the ability to achieve the required accuracy with high computational efficiency. Moreover, the trained algorithm only needed one cycle of discharge data to accurately predict the current SoH of the battery with a root mean square error of 1%. A simplified version of the predictive algorithm that relied only on voltage-time discharge was also developed. The reduction of the required features allowed testing to be conducted with a wider dataset. The simplified model was able to predict SoH of battery with root mean square error of 12%.
In summary, an efficient predictive algorithm based on a linear regression model was proposed for accurate prediction of the SoH of Li-ion batteries with higher performance comparable to those in the published literature. Although the application of the model was limited to constant current discharging profiles, it demonstrated that preprocessing the data used in training the model is the key to improved SoH prediction. Data preprocessing entails cleansing and normalization of the raw measured data to retain only the key features rich in the desired information. In a statement to Advances in Engineering, Dr Elisa Ang said that the simple, intuitive and computationally effective model requires only the measured voltage data and could prove invaluable in predictive maintenance of batteries in a multitude of applications.
Ang, E., & Paw, Y. (2021). Efficient linear predictive model with short term features for lithium-ion batteries state of health estimation. Journal of Energy Storage, 44, 103409.