Stepwise Fast Charging and State-of-Health Estimation for High-Voltage Battery Systems

Introduction and Research Context

The transportation sector, being a significant contributor to global carbon emissions, has witnessed a substantial transformation driven by the proliferation of electric vehicles. This shift is largely a response to the pressing global challenges of fossil fuel depletion and escalating environmental pressures. While the rapid growth in the adoption of electric vehicles has effectively reduced reliance on traditional internal combustion engines, it has simultaneously brought into sharp focus critical bottlenecks related to the recharging of their energy storage systems. The discrepancy between the expanding fleet of new energy vehicles and the lagging development of charging infrastructure has resulted in substantial “charging anxiety” among consumers. This issue is a significant barrier to the further market expansion of electric vehicles. Consequently, advancements in fast charging technology that do not compromise safety or battery lifespan are of paramount importance.

This research centers on the development and optimization of stepwise fast charging strategies, specifically tailored for two distinct battery chemistries: sodium-ion and ternary lithium-ion batteries. These battery systems are crucial components for high-voltage battery packs used in modern electric vehicles. The investigation extends beyond the establishment of charging protocols. A core component of this work is a comprehensive, comparative analysis of battery degradation mechanisms under different charging regimes. Moreover, this study proposes feature engineering and data-driven models to accurately estimate the State of Health (SOH) of batteries undergoing these advanced fast-charging cycles. This integrated approach seeks to provide a holistic solution, addressing the need for speed, safety, and high-accuracy performance monitoring of high-voltage battery systems within the next-generation of electric vehicles.

Experimental Platform and Methodological Setup

To investigate the performance and degradation patterns of the batteries, rigorous experimental platforms were constructed for both sodium-ion and lithium-ion cells.

The experimental setup for sodium-ion batteries consisted of a Neware CT-4008 battery testing system, capable of high precision measurements, and a DGBEL BT-150C temperature chamber. The chamber maintained a constant ambient temperature of 25°C to eliminate the influence of thermal fluctuations on the test results. The study utilized commercial 18650 sodium-ion cells. These cells featured layered oxide cathodes based on nickel-iron-manganese, a hard carbon anode, and an electrolyte incorporating sodium hexafluorophosphate. These specific cells had a nominal capacity of 1.4 Ah and an operating voltage window of 1.5V to 4.0V.

For the lithium-ion battery platform, an Arbin-BT2000 system was employed. This system possessed the capability of high-resolution data acquisition and featured multiple independent channels for efficient experimentation. A larger DGBEL BTT-1000C constant temperature and humidity chamber was used to maintain stable testing conditions. The focus of the lithium-ion experiments was on commercial 18650 ternary lithium-ion cells. These cells use a nickel-cobalt-manganese cathode and a graphite anode. They had a nominal capacity of 1.5 Ah and a nominal operating voltage of 3.6V (cut-off voltages of 2.5V and 4.2V). Both systems incorporated specialized fixtures with four-wire sensing for precise voltage and current measurements. These high-precision instruments formed the foundation of our data generation, serving the subsequent analysis of the high-voltage battery cell behavior under various charging strategies.

Preliminary Characterizations, Screening, and Pre-Experimental Tests

Prior to initiating the long-term aging cycle tests, a rigorous selection process was conducted to ensure a high degree of consistency among the test cells. This step is critical for reducing experimental variance and ensuring the reliability of comparative studies. All cells were initially allowed to rest inside the temperature chamber to achieve electrochemical equilibrium. Their alternating current internal resistance was measured using a dedicated ACR meter, discarding cells with significant deviations from the normal distribution. A further capacity screening process was done, involving three full charge-discharge cycles at low current rates. The protocol involved a CC-CV charge at a 0.5C rate followed by a constant current discharge at the same rate. Only those cells exhibiting a discharge capacity variation of less than 0.5% over the three cycles were considered for further use. The detailed selection parameters are provided below.

Table 1: Initial cell screening specifications.

Parameter Sodium-ion cell Lithium-ion cell
Discharge capacity (mAh) 1370 – 1375 1505 – 1510
AC internal resistance (mΩ) 18.2 – 18.4 12.2 – 12.4

Following the screening process, a series of pre-experiments was executed to decode the fundamental electrochemical behavior of both new cells. High-rate charging tests were performed to map charge voltage profiles, charge efficiency, and temperature rise across different C-rates. A significant evaluation parameter derived from these tests was the “constant current ratio” (CCR), which quantifies the proportion of charge stored before transitioning to the const-voltage phase of a CCCV charge. In a series of new rate tests for the lithium-ion cell, the CCR was found to diminish as the charging current increased, providing key indicators for designing the stepwise charging protocol.

The Hybrid Pulse Power Characteristic test was another crucial preliminary step. The HPPC procedure involves applying a predetermined sequence of current pulses (charge and discharge) at various states of charge. By analyzing the voltage responses to these pulses, it becomes possible to determine key internal resistance parameters like the ohmic resistance $R_0$ and polarization resistance $R_{pol}$. The formulas established in a test are as follows:
$$ R_0 = \frac{V_2 – V_1}{I_{pulse}} $$
$$ R_{pol} = \frac{V_3 – V_2}{I_{pulse}} $$
where $I_{pulse}$ represents the applied current and $V_1$, $V_2$, and $V_3$ denote the voltages at specific instances (beginning, edge, and end of the pulse). The measure captured the direct current internal resistance as a function of the state of charge and played an instrumental role in crafting our custom charging strategy.

Design and Implementation of Stepwise Fast Charging Strategies

The formulation of the fast-charging strategies was guided by the principles of Mas’s law. This law suggests that the maximum acceptable charging current for a battery is not a constant. Rather, it is high when the battery is at low SOC and decreases as the SOC increases. This theoretical boundary is key to balancing charge speed against cell health. Charging with excessive current in the high-SOC region can result in lithium plating on the anode and rapid capacity degradation. Conversely, the direct current resistance is highest at the beginning of the charge, and a very high initial current could produce significant heat and strain.

A differential voltage analysis was considered for the determination of specific boundaries for the charging steps. The relationship between voltage and capacity can be expressed as:
$$ DV = \frac{dV}{dQ} $$
where $dV$ represents a small change in voltage, and $dQ$ represents the corresponding charge accumulation. This metric yields curves containing characteristic peaks that present the phase transition of the materials. The peaks are known indicators of the regions where a battery has a poor rate acceptance. Therefore, high-rate current should be decreased before hitting the main characteristic peak on the DV curve, as it is common in the lithiation of graphite. Based on these approaches and a careful study of the charge internal resistance, we have proposed two main types of stepwise fast charging for both sodium-ion and lithium-ion cells.

Voltage-Cutoff Stepwise Charging Strategies

Harnessing the data from the 0.1C charge tests and the DV analysis, a voltage-cutoff stepwise fast-charging strategy was designed. For the sodium-ion cell, the strategy utilized different charge currents across distinct voltage intervals. The charge process used a low current at the start, as was previously discussed, to address the issues of higher internal resistance. As the voltage gradually increased, the current was raised to a higher level within the plateau region of lower voltage. As the cell advanced to a high state of charge and met the characteristic peak, the current was deliberately lowered preemptively. The steps and current values are shown in the table below.

Table 2: Proposed 9-step voltage-cutoff (9V-MSCC) charge settings for a sodium-ion cell.

End voltage (V) 2.6 2.8 3.2 3.5 3.8 3.85 3.9 3.95 4.0
Charging Current (A) 1.0 1.4 2.0 1.3 1.2 1.1 0.9 0.7 0.5

Similarly, a more advanced voltage-cutoff strategy was tailored for the lithium-ion cells. This 10-step strategy defines not only a starting low current but also the highest rate possible on the very gentle voltage plateau. However, current reduction is activated much sooner. Upon hitting the voltage of 3.8V, a strategic reduction was started while the main graphitic peak is approached at 3.9V. A slow, taper charge was subsequently performed at the high-voltage/low-current region, again to mitigate the lithium plating and to maintain healthy cycling over 800 cycles.

Table 3: The proposed 10-step voltage-cutoff (10V-MSCC) charge profile for lithium-ion cell.

End Voltage (V) 3.6 3.8 3.9 4.0 4.1 4.12 4.14 4.16 4.18 4.2
Current (A) 2.0 5.0 3.5 2.5 2.0 1.5 1.25 1.0 0.75 0.5

Capacity-Cutoff Stepwise Charging Strategies

Along with the voltage-based method, we designed a battery of capacity-cutoff stepwise tests. In these tests, state-of-charge thresholds determined the current values. For the sodium-ion cells, the state of charge was split into equal segments, dividing the full charge process into either three, five, or ten distinct steps. These are named 3SOC-MSCC, 5SOC-MSCC, and 10SOC-MSCC, respectively. The experiments with constant current were set to mimic previous charging constraints.

A similar set of protocols was implemented for the lithium-ion cells. These comprised three distinct 3-stage strategies (3SOC-MSCC-1, -2), three distinct 5-stage strategies (5SOC-MSCC-1, -2, -3) and one 10-stage strategy (10SOC-MSCC). As provided in the methodology, this broad matrix of charging protocols was tested. A boost charging method (BC-CCCV) was added as a common fast charge alternative: this is a specific two-step strategy starting at high-current then dropping to lower current at a threshold voltage. In this design, we aimed at analyzing the impact of step numbers and current levels on battery capacity retention upon long-term cyclability. These experiments offer direct comparisons to the constant current constant voltage baseline charge.

Charge Time Comparison

The time to perform a fast charge from 10% to 80% state of charge is critical for the electric vehicle user and was used as a key criterion for evaluating our protocol. For the control group, the battery was charged using a 1C constant current with constant voltage (CCCV) finish, while a 3A, 2C CCCV protocol was used for lithium. By the very essence of the stepwide fast charge, the proposed strategies resulted in large time savings when compared to the conventional CCCV baseline. The data below is taken from the very first cycles of the two high-voltage battery cells.

Table 4: Comparative charging duration of SIB and LIB across different strategies. This data was taken for the 10% – 80% SOC range.

Charging Strategy Time [min:sec] Time saved [min:sec] Percent reduced
Sodium-1C CCCV 45:26 – –
3SOC-MSCC (Na) 41:54 3:32 7.78%
5SOC-MSCC (Na) 41:11 4:15 9.35%
10SOC-MSCC (Na) 43:00 2:26 5.36%
9V-MSCC (Na) 38:10 7:16 16.0%
Lithium-3A CCCV 21:03 – –
3SOC-MSCC-2 (Li) 21:15 -00:12 -0.95%
5SOC-MSCC-2 (Li) 20:36 0:27 2.14%
BC-CCCV (Li) 20:16 0:47 3.72%
10SOC-MSCC (Li) 21:16 -0:13 -1.03%
10V-MSCC (Li) 21:09 -0:06 -0.47%

The results demonstrated that the voltage-cutoff strategies provided a remarkable reduction in charging time. Specifically, the 9V-MSCC strategy for sodium-ion battery provided a 16% improvement for the cell. In lithium-ion cells, all protocols managed to keep a high charging speed. Although the 10V-MSCC was slightly slower than CCCV, it produced a far superior outcome in cycle life, as it was tailored to age the cell gracefully.

Analysis of Aging Cycles and Post-Mortem Characterization

To validate the long-term effectiveness of our proposed charging strategies, two separate long-duration cycle aging tests were executed. Sodium-ion cells were cycled 150 times, while lithium-ion cells underwent 800 cycles. To monitor degradation, we performed intermediate capacity checks at regular intervals using low-rate currents to establish reliable performance benchmarks (every 50 cycles for Na and every 100 cycles for Li). The exact charge and discharge protocols of these reference performance tests involve a 0.5C rate for charge and discharge. The loss in capacity was measured against the original first cycle, providing the main basis for the health state (SOH) estimation:
$$ SOH_Q = \frac{Q_{current}}{Q_{rated}} \times 100\% $$
where $Q_{current}$ represents the current maximum available capacity, and $Q_{rated}$ the nominal capacity at the beginning of life. This gave a clear mapping of battery health.

Sodium-Ion Battery Results

The long-term cycling performance affirmed the advantage of the stepwise charging strategies over the standard CCCV charging. Analysis of the capacity fade and direct current resistance confirmed that the way current is managed during charge is crucial. The voltage-based strategy that avoided large DC resistance at the low SOC side and also reduced current before high peaks of the differential voltage produced the slowest degradation.

After completing 150 cycles, a clear hierarchy was established. As outlined, the 9V-MSCC strategy was the most effective in preserving battery lifetime. Health states improved by 150 cycles were shown to reach 92.3% under 9V-MSCC, which was higher than the other capacity-cutoff charging methods. Mean values of SOH remained at 85.7% for 3SOC-MSCC, but improved when the SOC segmentation was refined. The 5-stage and 10-stage have a similar improvement, reaching 90.7% and 90.9%, respectively. The most severe degradation took place during the baseline CCCV cycles.

Incremental capacity analysis provided deeper insight into the aging. We can obtain the incremental capacity by plotting:
$$ \frac{dQ}{dV} = \frac{I}{dV/dt} $$
through which phase transitions in the electrodes can be observed. The IC curves after cycling are assessed by the heights and locations of the main peaks. Loss in peak intensity directly reflects decay of active material. The IC data showed that the voltage-cutoff charging group had sharper, less shifted peaks, indicating less material loss. This aligns with the capacity fade results.

Table 5: Nonlinear degradation evaluation results for sodium-ion cells after 150 cycles.

Charging Strategy SOH, [%] ACIR (mΩ) DCR at 3.2V
1C Baseline CCCV 84.9 27.81 116.03
3SOC-MSCC 85.7 27.62 113.07
5SOC-MSCC 90.7 25.63 105.63
10SOC-MSCC 90.9 25.41 103.55
9V-MSCC 92.3 23.51 98.52

Based on these results, we could conclude that the 9V method maintained its excellent characteristics by actively preventing battery stress. This strategy acknowledges a high internal resistance at very low state of charge, lowering current to avoid current-induced stress. Then, there is a high-current stage that’s beneficial for fast charge. Near upper SOC’s, the charging current is dramatically reduced to prevent lithium plating and to make it more consistent.

Lithium-Ion Cell Results

The 800-cycle aging test on lithium-ion cells provided an even more dramatic insight into the differences between the fast charging protocols. The initial capacity states and their decreases under different cycle regimes are summarized below. The conventional 3A CCCV protocol, while offering convenience, produced a substantial capacity fade which is primarily associated with the continuous high power delivery at both the low and high SOC zones of the battery. At the opposite end, our proposed 10V-MSCC demonstrated remarkable capacity retention.

A key finding in cycle life data is that the number of steps was a critical parameter for the lifetime. The conventional 3-stage method, when combined with too high early-stage current, can cause unnecessary aging, producing similar or worse effects compared with the standard CCCV protocol under specific conditions. On the other hand, the 5-stage method provided improved protection to the battery, and 10-stage offered limited incremental advantage. The distinction was much less significant compared with the voltage cutoff strategies.

Table 6: The resulting State of Health (SOH) for the Li-ion cells under various charge strategies after 800 cycles.

Charging Strategy SOH [%] ACIR after aging [mΩ]
3A CCCV (Baseline) 64.1 21.32
3SOC-MSCC-1 (high current) 60.1 21.53
3SOC-MSCC-2 74.9 19.76
5SOC-MSCC-1 76.7 19.95
5SOC-MSCC-2 78.6 19.65
5SOC-MSCC-3 79.5 19.17
10SOC-MSCC 79.7 19.08
BC-CCCV 75.4 21.11
10V-MSCC 89.5 18.57

The exceptional result of the 10V-MSCC underlines the importance of shaping the current at low states of charge and at high-voltage boundaries near the charge cut-off. By mitigating the stresses from a rapidly increasing direct current resistance, our cell underwent the least degradation compared to CCCV or the capacity-defined step charging protocols. It is crucial to mention that the SOH gain of 25.4 percentage points above the baseline at the same total test duration is considerable for high-voltage battery system operation.

Post-Mortem Scanning Electron Microscopy Analysis

To further validate the distinct degradation patterns observed through electrical measurements, selected lithium-ion cells underwent post-cycling disassembly and post-mortem scanning electron microscopy. After completion of the 800 cycles, the cells were fully discharged and opened in an argon-filled glove box. The negative electrodes and corresponding separator sheets were retrieved and prepared for microscopic inspection, providing micro-structural evidence of electrode performance.

The SEM images revealed significant differences in the condition of the anode surface and the separator microstructure between test groups. The graphite anode from our 10V-MSCC test still had well-formed flakes which presented clean edges, indicating minimal structural degradation. In a stark contrast, the electrode from the standard CCCV procedure showed a rather rough, uneven surface having several regions of thick deposits. These were most likely lithium dendrites, and their presence was a sign of lithium plating that results from too-large charge currents in high SOC regions. The separator retrieved from the cell charged under 10V-MSCC maintained a porous, highly organized network of fibers, ensuring good electrolyte wetting as well as minimal impedance. In contrast, the separator from the CCCV-cycled cell was largely clogged with reaction products. It was characterized by severe blockage of its pores and breaks to the fibrous network. This underscores the significant degree of internal strain and negative side reactions on the high-voltage battery cells tested with the basic charging profile.

Estimation of the State of Health Under Fast-Charging Conditions

An accurate SOH evaluation is essential for battery management systems to ensure uninterrupted and secure operation of a high-voltage battery pack. A frequently used but overly simplistic method uses only total capacity to indicate health. For our stepwise fast-charging protocols, we proposed to use a wider range of features available from the charging data. These features offer a real-time snapshot of health as it progresses.

The first step of this estimation approach is identifying a robust set of health factors based on the planned charge curves. During the charge process, the features altered in a predictable fashion. Analysis across cycles showed a significant and consistent differentiation among curves, especially during the initial three current steps. We defined new “ladder factors” using the charging time or the charge delivered per-voltage-step during these early stages. These ladders are strongly linked to internal aging.

Feature Engineering for SOH Estimation

For the sodium-ion battery, where our test procedure lasted 150 cycles, the voltage-capacity profiles at several cycle intervals showed a clear dependence on age. The distinguishing aspect of these profiles is that each charge step has a smaller capacity window when the battery has aged, leading to a shift of the charging curve on the capacity axis. The first three step capacities, until the main differential voltage peak, responded more to the cell age. For this reason, charging time at each of the first three steps were employed as “fast-charge specific health indicators”.

This can be scaled and transferred to the lithium-ion battery, where we conducted a more extensive test. For this battery, we extracted seven features from the early part of the charge. These features were not only those steps’ charged capacities but also included temperature at the end of the steps and the rate of voltage increase for one case. Our goal was to establish a health factor set that is sensitive to aging. The selected features are listed in the table below.

Table 7: Selected health factor set for lithium-ion battery SOH estimation.

Feature Definition
F1 Charge capacity of the first step (at 3.6 V).
F2 The battery temperature at the end of the first step.
F3 Charge capacity of the second step.
F4 The battery temperature at the end of the second step.
F5 The voltage increase rate of the second step.
F6 Charge capacity of the third step.
F7 The battery temperature at the end of the third step.

To determine from which charging steps our SOH estimator can start making accurate predictions, we constructed two test scenarios. In the **Scenario 1**, the model’s inputs are only the first three steps’ health features (F1 to F7 in the table). In **Scenario 2**, the model also used the data from the subsequent steps (fourth, fifth, and sixth), adding more features to better predict SOH. To verify the quality of these features, we quantified their linear correlations with the actual states of health.

Given that the scale and monotonicity of battery fading may vary, the Pearson correlation coefficient was used to evaluate the monotonic relationship for the sodium-ion battery. The analysis proved a robust association of the constructed features to the SOH levels, with each value exceeding 0.9. This was further proof of the reasonableness of the health factor selection.

Estimation Models for Sodium-Ion Battery SOH

The sodium-ion dataset was small, composed of 150 cycles with data points provided per charge step. To predict SOH from this limited but information-rich dataset, we leveraged a hybrid prediction algorithm. This approach coupled a back-propagation neural network with a particle swarm optimization algorithm. The primary task of PSO was to optimize initial architecture weights and biases of the back propagation network, overcoming the common pitfalls of gradient-based methods like slow convergence and the chance to settle at a local optimum.

The detailed workflow of the PSO-BP algorithm is shown in Figure below:
1. **Data Preparation:** We normalize our extracted features and split them into training and testing subsets. In this work, two different split ratios were investigated to evaluate model robustness: 70% for training versus 30% for testing, and a stricter 50/50 split.
2. **Network Definition:** We configured the BP architecture. In this structure, the input layer holds as many nodes as health indicators. We have 6 neurons allocated for the input, along with one hidden layer. Through a formula for estimating the hidden layer dimension:
$$ m = \sqrt{n+p} + a $$
where $m$ is the number of hidden nodes, $n$ and $p$ are numbers of input and output nodes, and $a$ is an integer from 0 to 10. The final output layer consists of a single node that represents the estimated SOH.
3. **PSO Initialization:** We initialize a population of particles each encoding a solution of network weight and bias values. Their quality is evaluated through the mean-squared error between the output from the forward pass and the known training target SOH:
$$ f = \frac{1}{N} \sum_{i=1}^{N} \sqrt{\sum_{d=1}^{D} (y_{id} – \hat{y}_{id})^2} $$
Here, $N$ is the number of training samples, $D$ is the output dimension, $y$ values are predicted outputs and $\hat{y}$ are real SOH values.
4. **Velocity and Position Update:** Each particle updates a velocity to guide its search direction to maintain a balance between exploring new regions and converging upon high-quality solutions:
$$ v_{id}^{t+1} = w v_{id}^{t} + c_1 r_1 (p_{id}^{t} – x_{id}^{t}) + c_2 r_2 (p_{gd}^{t} – x_{id}^{t}) $$
$$ x_{id}^{t+1} = x_{id}^{t} + v_{id}^{t+1} $$
where the optimal personal solution $p_{id}$ and the optimal global solution $p_{gd}$ at generation $t$ are updated at each iteration.
5. **Terminate and Evaluate:** Upon completion, this process yields optimized network parameters and metrics of the tested model are calculated.

The state estimates have been quantified through the mean average error and root mean square error:
$$ MAE = \frac{1}{n} \sum_{i=1}^{n} |y_i – \hat{y}_i| $$
$$ RMSE = \sqrt{\frac{1}{n} \sum_{i=1}^{n}(y_i – \hat{y}_i)^2} $$

Using the created data, we trained this PSO-BP estimator. To establish its merit, we benchmarked its predictions against results of other representative machine learning algorithms, namely standard BP, support vector machines, extreme learning machines, standard convolutional neural network, and a genetic algorithm optimized network (GA-BP).

Table 8: Comparison table of SOH performance (RMSE and MAE) using PSO-BP and other methods under different training sets and test groups for the sodium-ion cell.

Estimator Training Ratio RMSE MAE
PSO-BP 70% 0.610 0.450
PSO-BP 50% 0.678 0.583
GA-BP 70% 0.873 0.647
GA-BP 50% 0.910 0.739
BP 70% 1.326 1.113
CNN 70% 2.150 2.076
ELM 70% 2.571 2.024
SVM 70% 1.445 1.220

Compared to the other methods, PSO-BP produced a clear reduction in both RMSE and MAE for SOH. The optimization in the initial weights had a pronounced effect on yielding high precision and robustness. The RMSE of 0.610 in a 70% training setup is around 0.5% of the full capacity, demonstrating high precision. Even when trained with a smaller dataset, using a 50% training split, the proposed PSO-BP algorithm maintained a small prediction error of well under 1%, demonstrating its high reliability for health assessment.

Estimation Models for Lithium-Ion Battery SOH

The lithium-ion dataset provided 800 cycles and enabled testing of more complex deep learning architectures. We employed a hybrid neural network that first uses a CNN feature extractor which learns key health characteristic patterns. These extracted feature maps are then fed into a Long Short-Term Memory module. The recurrent nature of the LSTM enables modeling dependencies present through the sequence of charging cycles. It can learn the underlying temporal evolution of battery fading. This architecture is highly applicable to modeling SOH accurate over long-term storage life.

We conducted the experiment on two different scenarios: one used only data from the first three charge stages as inputs, and the second used data from the first six stages. Because the battery data is heavily nonlinear and monotonic, we used the Spearman correlation metric. Similar to the prior correlation analyses, feature relevance was excellent. The advanced capacity and thermal metrics show values over 0.9, validating our choice of the model inputs. In the table below, we have benchmarked the CNN-LSTM hybrid. We compared this SOH estimator with PSO-BP and standard LSTM models.

Table 9: Comparison of SOH estimation evaluation indexes for lithium-ion cell under different HF scenarios.

Estimator HF Scenario RMSE MAE
CNN-LSTM Seven feature inputs 0.929 0.829
CNN-LSTM Extended feature input 0.856 0.681
PSO-BP Seven feature inputs 1.304 0.988
PSO-BP Extended feature input 0.848 0.571
LSTM Seven feature inputs 2.360 1.946
LSTM Extended feature input 1.975 1.529

In the comparative analysis, CNN-LSTM and PSO-BP achieved similar performance when the extended feature set was evaluated. However, for a reduced feature set, our CNN-LSTM model clearly outperformed the PSO-BP, giving a strong advantage by being nearly four times as accurate with low RMSE. It demonstrates the model’s capacity to extract critical patterns from scarce data by its convolutional layer. The key result is that choosing only the first three step data is effective for model predictions. This enables an SOH estimation as soon as 13 minutes from charge initiation. It reduces the requirement of complete data while making real-time health monitoring feasible. This may ensure secure operation of EV battery systems without significant off-line testing.

Conclusions and Insights into Fast Charging and Battery Health

This research provides a comprehensive examination of stepwise fast charging strategies and associated health estimation for modern high-voltage battery systems. Through this rigorous and systematic workflow, we were able to draw conclusions that may have significant impacts on the future of EV battery technology.

Conventional CCCV fast charging protocols are often plagued by a need to balance speed and longevity. Our developed charging strategies utilized key insights into DV curve characteristics and DC internal resistance. Those aspects, which vary by SOC, are closely tied to a battery’s tolerance to charging power. From the design stage to long-term aging experiments that included 150 cycles on sodium-ion cells and 800 aging cycles on lithium-ion cells, a clear difference was noted. The result was that the voltage-cutoff, stepwise charging strategies, such as 9V-MSCC and 10V-MSCC for high-voltage battery systems, were powerfully superior.

We found direct evidence that:
* The high voltage-based cell 10V-MSCC can significantly increase the battery lifetime. Specifically, by the end of study, the SOH stood at 89.5 %, over 25% higher than its CCCV counterpart, with faster total charge times. The post-mortem SEM results effectively tied electrical performance to the micro-structure state, showcasing that our tested control cell had been able to minimize lithium plating and structural damage.
* For capacity-defined stepwise charging (MSCC), a moderate step number (5) was more advisable. However, an over-segmentation (e.g., ten steps) does not offer additional gain as it tends to prolong charge duration. A step increase from 3 to 5 gave a positive effect, yet from 5 to 10 was close to marginal improvements.
* The highly important advantage of voltage-cutoff strategy lies in its self-correcting nature. Since our voltage threshold does not change as the battery ages, it does not face the common problem of SOC-based transitions, where estimates can lose accuracy when battery health fades.

Finally, we developed highly accurate and test-adapted SOH estimation models. For the dynamic fast charge data, we introduced “ladder factors” based on the early charge portion. Both correlation analysis and the performance of our models verified what was conceptually strong: the earlier portion of the charge is a great indicator. The PSO-BP algorithm was successfully applied to sodium-ion cell monitoring, with RMSE often below 0.7%. In parallel, the CNN-LSTM model proved to be the best fit for the long-term lithium-ion data sequences. This architecture allowed reliable predictions even when only parts of the early charge cycles were used. Using these advanced tools, high-voltage battery systems can achieve robust and real-time SOH tracking with exceptional confidence while improving performance.

Scroll to Top