Evaluating Battery Health for Electric Vehicles Using Electrochemical Impedance Spectroscopy Measurements

Lithium-ion batteries (LIBs) are a critical product in today’s economy. Because of their use in many electronic devices that power the globe, their production and lifespan help ensure that society functions. Recent studies have examined ways to evaluate battery health, with some suggesting alternative energy sources like lithium metal batteries and others showing how analytical techniques can help improve monitoring of lithium-ion batteries (1,2).

Recently, a team of researchers from three universities (University of Huddersfield, Yanshan University, and Teesside University) collaborated on a new model to improve the accuracy of state of health (SOH) estimation in LIBs. The team proposed a model that overcomes the challenge of overlapping semicircles in electrochemical impedance spectroscopy (EIS) measurements (3).

Electrochemical impedance spectroscopy (EIS) applies an alternating current (AC) signal to an electrochemical system and measures the response over a frequency range, typically from 0.01 Hz to 100 kHz. This technique analyzes electrical properties, providing insights into charge transfer, diffusion, and resistance in electrochemical processes (3).

SOH estimation of LIBs is important in making sure electric vehicles are safe and perform as expected. Because LIBs degrade over time, it becomes difficult to extract meaningful data from EIS, a technique used to evaluate battery aging (1–3). One of the main obstacles has been the overlapping of EIS semicircles, which introduces uncertainties in identifying electrochemical processes during battery degradation (3). Zuolu Wang and his team sought to overcome these uncertainties by combining advanced techniques to extract influential health indicators (HIs) (3).

In their model, the researchers integrated the distribution of relaxation time (DRT). The DRT is an effective method for interpreting EIS data that can discern pertinent information about the electrochemical processes (4). This is the first step in a multi-faceted process that includes data refinement and feature extraction using an autoencoder and Spearman rank correlation analysis, methods that reduce the noise and dimensionality of the extracted HIs (3).

Using the autoencoder is important to the researcher’s model, because it performs a critical function that makes the model more accurate. The autoencoder helps the model filters out irrelevant information, which allows the device to focus on features that are directly linked to battery aging (3). Spearman rank correlation analysis further refined these HIs by identifying the ones most closely associated with battery degradation processes (3). As a result, this layered approach ensures that only the most influential features are used to estimate battery SOH, leading to more reliable predictions (3).

The team's model was rigorously tested under varying conditions to see if the model can maintain its accuracy in different environments. The researchers took four different LIBs and evaluated them at three distinct temperatures—25 °C, 35 °C, and 45 °C (3). The results showed that their cascade feedforward neural network optimized by a genetic algorithm (GA-CF) model outperformed other popular machine learning models, including multilayer perceptron (MLP), long short-term memory (LSTM), and support vector regression (SVR) (3). Not only did the new model achieve higher accuracy in terms of R-score, but it also produced lower root mean square error (RMSE) and mean absolute error (MAE), key metrics that validate the model's performance (3).

Current machine learning models often struggle with the nonlinearity of battery degradation, especially under fluctuating operational conditions. This model demonstrated in the study developed by Wang's team overcomes some of the main obstacles by capturing these nonlinear relationships with better precision (3).

Moreover, the model’s robustness suggests that it can be generalized to other battery systems and environments. This flexibility makes it highly promising for widespread application in electric vehicles (EVs), where accurate SOH estimation can lead to better safety mechanisms and more efficient energy management systems (3).

Looking ahead, the research team plans to expand their work by validating the model under additional operational conditions and exploring the integration of transfer learning to further reduce dependency on historical cycling data (3). This advancement would bring the model closer to practical applications, particularly in industries where battery life management is critical.

References

1. Wetzel, W. New AI-Based Model Boosts Accuracy in Estimating Lithium-Ion Battery Aging. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/new-ai-based-model-boosts-accuracy-in-estimating-lithium-ion-battery-aging (accessed 2024-10-21).
2. Wetzel, W. Lithium Metal Batteries and the Critical Function of Solid Electrolyte Interphases: An Interview with Lauren Marbella of Columbia University. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/lithium-metal-batteries-and-the-critical-function-of-solid-electrolyte-interphases-an-interview-with-lauren-marbella-of-columbia-university (accessed 2024-10-21).
3. Zhao, X.; Liu, S.; Li, E.; et al. A hybrid intelligent model using the distribution of relaxation time analysis of electrochemical impedance spectroscopy for lithium-ion battery state of health estimation. J. Ener. Stor. 2024, 84A, 110814. DOI: 10.1016/j.est.2024.110814
4. Subotic, V.; Hochenauer, C. Analysis of solid oxide fuel and electrolysis cells operated in a real-system environment: State-of-the-health diagnostic, failure modes, degradation mitigation and performance regeneration. Prog. Ener. Combustion Sci. 2022, 93, 101011. DOI: 10.1016/j.pecs.2022.101011