Page 95 - EER-2-1

Explora: Environment
and Resource Artificial neural networks

often face a trade-off between computational speed and cycle life, are incorporated into the model, which is trained
accuracy. 14,15 Low-level lifetime models, while fast to set using extensive experimental data collected under various
up and execute, often rely on mathematical equations with operating conditions. The electrothermal component
many coefficients and may struggle to capture intricate employs a second-order Thévenin equivalent circuit
electrochemical degradation mechanisms. 16-19 Conversely, model (ECM), enhanced by advanced characterization and
high-fidelity models offer detailed analysis but are time- parameterization procedures.
consuming in both model setup and simulation. 20-23 Validation of the coupled model demonstrates its

To bridge this gap, there is a growing need for effectiveness in estimating both electrical and thermal
medium-fidelity lifetime models that balance accuracy performance across different stages of the cell’s lifespan.
and computational efficiency. Data-driven methodologies, The results revealed remarkable accuracy, with a maximum
particularly artificial neural networks (ANNs), have error of just 2% in voltage readings and 3% in temperature
demonstrated promise in significantly improving predictions during discharge. This comprehensive
modeling accuracy while maintaining reasonable modeling approach not only enhanced our understanding
computational demands. 24-26 ANNs have demonstrated of the long-term dynamics of Sony 3 Ah cells but also
high accuracy in various battery estimation tasks. 27-31 served as a computationally efficient tool for battery
32
Severson et al. demonstrated the potential of machine management systems and control strategies.
learning in predicting battery life cycles, achieving high
accuracy using early-cycle data. Building on this, Attia 2. Methods
et al. developed a closed-loop optimization system that 2.1. Battery feature
33
combined machine learning with Bayesian optimization
to optimize fast-charging protocols for LiBs efficiently. The batteries tested were cylindrical Sony 18650-type
In the realm of electrothermal modeling, Dai et al. lithium-ion cells (LiBs) with a capacity of 3 Ah (Sony,
34
proposed a coupled electrothermal-mechanical model that China). These cells feature a Li(NiMnCo)₁/₃O₂ cathode
provided insights into the multi-physics behavior of LiBs. and a graphite anode (G), with an average mass recorded at
Their work highlighted the importance of considering 46.6 g. Their nominal capacity and voltage were 3 Ah and
multiple physical domains in battery modeling. Similarly, 3.7 V, respectively. In this paper, the C-rate (C) is defined
Guo et al. developed a multiscale model that integrated as 1C = 3 A.
35
electrochemical, thermal, and aging effects, offering A Neware CT-4008 system (Neware, China) was
a comprehensive approach to battery performance employed to cycle the batteries, offering a voltage range
prediction. The integration of data-driven methods with of 25 mV – 5 V and a current range of 0.5 mA – 6 A.
physics-based models has also gained traction. Wu et al. This system was computer-controlled, and the charging/
36
combined a physics-based model with machine learning to discharging current was set based on the manufacturer’s
improve the accuracy of state-of-health estimation. Their specifications. To monitor the temperature of the batteries,
38
hybrid approach demonstrated superior performance a type-K thermocouple was affixed to the cells. Figure 1
compared to traditional methods. In a related study, Li illustrates the experimental setup for characterization.
et al. utilized a deep learning framework to enhance the
37
prediction of remaining battery lifespan, incorporating 2.2. Electrothermal characterization
both static and dynamic features of battery degradation. To parameterize the electrothermal model, several
Despite these advancements, the specific combination of characterization tests were conducted: (i) capacity test;
semi-empirical electrothermal modeling with ANN-based (ii) hybrid pulse power capability (HPPC); 39,40 (iii) open-
lifetime estimation for 18650-type cells remains relatively circuit voltage (OCV) (please send out + upload in GD)
unexplored. test; 41,42 (iv) thermal pulse test; 43,44 and high-current
45
Therefore, this study presents a comprehensive test. These tests were derived from international battery
electrothermal and lifetime model for Sony 3 ampere- standards 46,47 and were conducted at various temperatures,
hours (Ah) 18650 nickel manganese cobalt lithium-ion with a rest period of 3 – 4 h used to precondition the cells
cells, leveraging ANNs to predict the lifespan of these cells to the desired temperature (−10, 10, 25, 35, 45, or 60°C).
effectively. The proposed model integrates an electrothermal The capacity test involves performing full charge and
component with an ANN-based lifetime prediction discharge cycles at varying discharge C-rates (C/5, C/4,
approach, providing a holistic representation of cell C/3, C/2, 1C, and 2C) and temperatures (−10, 10, 25, 35,
behavior throughout its operational life. Key parameters, 45, and 60°C) to capture the data needed for modeling
such as state of charge (SoC), temperature, current, and battery capacity. During the test, the charging C-rate was

Volume 2 Issue 1 (2025) 2 doi: 10.36922/eer.7228