In my work on electric vehicle battery pack reliability, I have repeatedly observed that the electric vehicle battery pack is one of the most safety-critical and cost-sensitive assemblies in a new energy vehicle. Across vehicle-level fault distributions, the electric vehicle battery pack accounts for a very high proportion of failures. These failures can spread quickly, can trigger thermal runaway within a short time, and can create serious safety incidents. I therefore treat fault prediction and preventive maintenance of the electric vehicle battery pack as a central engineering problem rather than as a routine after-sales task.
The dominant root cause that I identify in my analysis is inconsistency among individual cells inside the electric vehicle battery pack. This inconsistency appears as voltage differences, capacity offsets, internal resistance increases, and temperature gradients. When a vehicle operates under heavy loads, frequent starts and stops, or continuous day-and-night operation, the inconsistency of the electric vehicle battery pack deteriorates further, and the fault process accelerates. Conventional maintenance systems based mainly on periodic service and corrective repair cannot predict early hidden faults, cannot diagnose them in advance, and cannot allocate maintenance resources precisely.
For fault diagnosis and prediction algorithms applied to the electric vehicle battery pack, I group existing methods into three categories: knowledge-based methods, physical-model-based methods, and data-driven methods. Knowledge-based methods depend heavily on expert experience and have weak generalization when unknown faults appear. Physical-model-based methods often rely on high-precision electrochemical models and have relatively poor real-time performance. Data-driven methods, especially deep learning and machine learning, are a major research focus and show good engineering applicability in multi-fault identification of the electric vehicle battery pack. However, they still suffer from high false-alarm rates, high training costs, and weak coupling between fault prediction and maintenance decisions.
In this article, I first analyze the fault mechanism and evolution characteristics of the electric vehicle battery pack. I then build a fault prediction system and formulate a graded preventive maintenance strategy. Finally, I verify the strategy through engineering application cases. My goal is to provide a practical reference for fault prediction and preventive maintenance of the electric vehicle battery pack.
1. Fault Mechanism and Evolution Characteristics of the Electric Vehicle Battery Pack
1.1 Fault Representation and Mechanism
In my analysis, cell-to-cell inconsistency is the core factor in faults of the electric vehicle battery pack. It is mainly represented by voltage differences, capacity offsets, internal resistance increases, and temperature gradients. Voltage differences among cells appear already in the early fault stage. The relationship between voltage deviation range and fault stage is summarized in Table 1.
Table 1. Voltage inconsistency grades and corresponding fault development
| Consistency grade | Voltage standard deviation interval | Fault development degree | Corresponding maintenance strategy |
|---|---|---|---|
| Stage I | sigma less than or equal to sigma1 | Good consistency, no performance degradation | Execute normal charge and discharge management, optimize thermal management and charge-discharge strategy |
| Stage II | sigma1 less than sigma less than or equal to sigma2 | Mild consistency deviation, no substantial cell performance damage | Execute balanced charge and discharge maintenance, correct cell voltage deviation |
| Stage III | sigma2 less than sigma less than or equal to sigma3 | Some cells have substantial performance damage, moderate consistency deterioration | Execute battery regrouping optimization, replace severely degraded cells |
| Stage IV | sigma greater than sigma3 | Severe cell performance deterioration, high fault and thermal runaway risk | Replace the entire battery pack and regroup, eliminate safety hazards |
I define the voltage deviation of cell i at time t as follows:
$$ \Delta V_i(t)=V_i(t)-\bar{V}(t) $$
where V_i(t) is the terminal voltage of cell i and \bar{V}(t) is the average voltage of the electric vehicle battery pack. The standard deviation of cell voltages, which I use as a primary consistency index, is:
$$ \sigma_V(t)=\sqrt{\frac{1}{N}\sum_{i=1}^{N}\left(V_i(t)-\bar{V}(t)\right)^2} $$
Capacity inconsistency is equally important. I express the remaining capacity of cell i as:
$$ Q_i(t)=Q_{0,i}\left(1-\alpha_i t^{\beta_i}\right) $$
where Q_{0,i} is the initial capacity, \alpha_i is the degradation coefficient, and \beta_i is the degradation exponent. The capacity inconsistency within the electric vehicle battery pack is then:
$$ \Delta Q_i(t)=Q_i(t)-\bar{Q}(t) $$
Internal resistance growth is another early indicator. I use a simplified linear or square-root model:
$$ R_i(t)=R_{0,i}+\gamma_i t^{\lambda_i} $$
where R_{0,i} is the initial internal resistance, and \gamma_i and \lambda_i are aging parameters. The resistance deviation is:
$$ \Delta R_i(t)=R_i(t)-\bar{R}(t) $$
Temperature gradient is both a cause and a consequence of inconsistency. I define the temperature deviation as:
$$ \Delta T_i(t)=T_i(t)-\bar{T}(t) $$
From an electrochemical perspective, I attribute degradation and failure of the electric vehicle battery pack to three major mechanisms: positive electrode instability, lithium dendrite growth, and electrolyte decomposition. These mechanisms interact and eventually lead to capacity fade, power fade, and thermal runaway. Table 2 summarizes the mechanisms, indicators, and consequences.
Table 2. Electrochemical degradation mechanisms of the electric vehicle battery pack
| Mechanism | Main process | Observable indicator | Consequence for the electric vehicle battery pack |
|---|---|---|---|
| Positive electrode instability | Positive electrode material reconstruction, active material loss, transition metal dissolution | Capacity fade, voltage plateau shift, increased impedance | Reduced usable energy, increased inconsistency, accelerated aging |
| Lithium dendrite growth | Lithium plating on the negative electrode, dendrite formation and growth | Sudden voltage drop, abnormal coulombic efficiency, local hot spots | Internal short circuit risk, thermal runaway trigger, safety hazard |
| Electrolyte decomposition | Electrolyte reduction and oxidation, SEI film overgrowth, gas generation | Increased internal resistance, pressure rise, capacity loss | Power fade, thermal instability, shortened service life |
Among these mechanisms, I find that lithium dendrite growth on the negative electrode and excessive SEI film evolution are the main causes of thermal runaway in the electric vehicle battery pack. When local lithium plating occurs, the affected cell can develop an internal short circuit, and the heat generated can propagate to neighboring cells. Because the electric vehicle battery pack is a tightly coupled system, a local fault can rapidly become a pack-level fault.
1.2 Fault Evolution Characteristics
The formation of an electric vehicle battery pack fault is not caused by a single factor. In its evolution, the core factors that lead to failure can run through research and development, manufacturing, assembly, and use stages. In the research and development stage, if the energy management strategy is defective, abnormal temperature may appear during use, and inconsistency among cells will be aggravated. In the manufacturing stage, if material and production process standards are low, cell-to-cell inconsistency will also increase. In the assembly stage, if welding or matching quality and precision are poor, contact resistance differences will appear, and local cell aging will cause inconsistency in the electric vehicle battery pack. In the use stage, if normal operating temperature or charge-discharge rate cannot be maintained, cell degradation will accelerate.
For special vehicle types such as urban sanitation vehicles and mountain logistics vehicles, high-frequency start-stop operation and heavy-load climbing occur often. If these conditions are not properly addressed, they will accelerate the occurrence and development of faults in the electric vehicle battery pack. Table 3 summarizes the evolution path across lifecycle stages.
Table 3. Lifecycle evolution of faults in the electric vehicle battery pack
| Lifecycle stage | Potential adverse factor | Effect on the electric vehicle battery pack | Long-term consequence |
|---|---|---|---|
| Research and development | Defective energy management strategy, insufficient thermal design | Abnormal temperature, uneven current distribution | Accelerated inconsistency and early degradation |
| Manufacturing | Low material standards, unstable process control | Initial capacity and resistance differences among cells | Poor initial consistency, faster aging |
| Assembly | Poor welding quality, low matching precision | Contact resistance differences, local hot spots | Local aging and pack-level inconsistency |
| Use | Abnormal temperature, high charge-discharge rate, over-discharge | Increased lithium plating risk, electrolyte decomposition | Capacity fade, power fade, thermal runaway risk |
| Special operating conditions | High-frequency start-stop, heavy-load climbing, continuous operation | High dynamic stress and thermal stress | Rapid fault development in the electric vehicle battery pack |
I therefore conclude that fault prediction for the electric vehicle battery pack must consider not only the present state but also the cumulative stress history. A single measurement of voltage or temperature is insufficient. The evolution of inconsistency is a time-dependent process, and it must be modeled with multi-dimensional data.
2. Construction of a Fault Prediction Technology System for the Electric Vehicle Battery Pack
2.1 Data System and Feature Engineering
Data are the foundation of fault prediction for the electric vehicle battery pack. In my design, the data sources mainly include three parts. The first part is data obtained from the battery management system during vehicle operation, which is also the basis for model training. The second part is electrochemical data collected during experiments. The third part is environmental data obtained from vehicle-mounted sensors. Table 4 lists the data sources and their roles.
Table 4. Data sources for fault prediction of the electric vehicle battery pack
| Data source | Typical variables | Role in prediction | Challenge |
|---|---|---|---|
| Battery management system | Cell voltage, pack current, state of charge, state of health, temperature | Primary training data and real-time monitoring | Noise, missing values, sampling inconsistency |
| Experimental electrochemical data | Capacity, internal resistance, electrochemical impedance spectroscopy, coulombic efficiency | Mechanism verification and label generation | Laboratory conditions differ from real driving |
| Vehicle-mounted sensors | Ambient temperature, humidity, vibration, vehicle speed, acceleration | Operating context and stress quantification | Sensor drift and installation variation |
After collecting fault prediction data, I perform preprocessing before feature extraction. The preprocessing includes outlier removal, missing-value imputation, time synchronization, and normalization. I use robust scaling because the electric vehicle battery pack data often contain heavy-tailed distributions:
$$ x’_{j}=\frac{x_j-\mathrm{median}(x)}{\mathrm{IQR}(x)} $$
where IQR(x) is the interquartile range. For missing values, I use linear interpolation for short gaps and model-based imputation for longer gaps:
$$ \hat{x}_t = x_{t-1} + \frac{t-(t-1)}{(t+1)-(t-1)}(x_{t+1}-x_{t-1}) $$
Feature extraction is performed in three dimensions. In the time domain, I extract charge-discharge voltage slope, instantaneous temperature rise rate, and cell voltage difference to determine parameter trends. In the frequency domain, I extract dominant frequency amplitude and energy density indicators to help identify early weak faults. In the statistical domain, I extract cell voltage mean, standard deviation, and information entropy to quantify the dispersion degree of consistency in the electric vehicle battery pack. Table 5 summarizes these features.
Table 5. Feature extraction domains for the electric vehicle battery pack
| Domain | Feature examples | Physical meaning | Fault sensitivity |
|---|---|---|---|
| Time domain | Voltage slope, temperature rise rate, cell voltage difference, charge-discharge duration | Trend and rate of parameter change | High for early degradation and abnormal heating |
| Frequency domain | Dominant frequency amplitude, spectral energy, harmonic ratio | Dynamic response and weak fault signatures | High for internal short circuit and connection faults |
| Statistical domain | Mean, standard deviation, information entropy, skewness, kurtosis | Dispersion and distribution shape of cell states | High for inconsistency of the electric vehicle battery pack |
To remove redundant information and avoid overfitting, I use the Pearson correlation coefficient for feature selection. The Pearson coefficient between feature X and fault label Y is:
$$ r_{XY}=\frac{\sum_{i=1}^{n}(X_i-\bar{X})(Y_i-\bar{Y})}{\sqrt{\sum_{i=1}^{n}(X_i-\bar{X})^2}\sqrt{\sum_{i=1}^{n}(Y_i-\bar{Y})^2}} $$
I retain features with strong correlation to faults of the electric vehicle battery pack and remove features with high mutual correlation. This step reduces model complexity and improves generalization.
2.2 Construction of Fault Prediction Algorithm Models
I analyze two methods to adapt to different fault prediction scenarios: time-series degradation trend prediction and fault classification prediction. For time-series degradation trend prediction, I use an LSTM model as the basic framework. The fitting effect and prediction stability of this model have been verified in real vehicle tests. To reduce training cost and false-alarm rate, I use Bayesian optimization to tune the hyperparameters of the LSTM model. For simultaneous prediction of multiple typical faults in the electric vehicle battery pack, I use a fault classification prediction method based on an improved random forest algorithm. I use grid search and cross-validation to optimize the key parameters of the original random forest algorithm, thereby improving its performance metrics and engineering applicability.
The LSTM cell used in my model is governed by the following equations:
$$ f_t=\sigma\left(W_f[h_{t-1},x_t]+b_f\right) $$
$$ i_t=\sigma\left(W_i[h_{t-1},x_t]+b_i\right) $$
$$ \tilde{C}_t=\tanh\left(W_C[h_{t-1},x_t]+b_C\right) $$
$$ C_t=f_t\odot C_{t-1}+i_t\odot \tilde{C}_t $$
$$ o_t=\sigma\left(W_o[h_{t-1},x_t]+b_o\right) $$
$$ h_t=o_t\odot \tanh(C_t) $$
where f_t, i_t, and o_t are the forget, input, and output gates, C_t is the cell state, h_t is the hidden state, \sigma is the sigmoid function, \tanh is the hyperbolic tangent function, and \odot denotes element-wise multiplication. For the electric vehicle battery pack, the input sequence x_t includes voltage, current, temperature, state of charge, and derived inconsistency features.
Bayesian optimization treats hyperparameter tuning as a sequential decision problem. I define the objective as:
$$ x^*=\arg\max_{x\in\mathcal{X}} f(x) $$
where f(x) is the validation performance of the LSTM model, and \mathcal{X} is the hyperparameter space. I use a Gaussian process surrogate model and an acquisition function to balance exploration and exploitation. This approach reduces the number of training runs and improves the reliability of the electric vehicle battery pack fault prediction model.
For multi-fault classification, the random forest model combines multiple decision trees. The final prediction is determined by majority voting:
$$ \hat{y}=\arg\max_{c}\sum_{k=1}^{K} I\left(h_k(x)=c\right) $$
where K is the number of trees, h_k(x) is the prediction of the k-th tree, and I is the indicator function. I improve the random forest by optimizing the number of trees, maximum depth, minimum samples per leaf, and maximum features. I also use class weighting to handle imbalanced fault categories in the electric vehicle battery pack.
Table 6 summarizes the two prediction models and their optimization methods.
Table 6. Fault prediction models for the electric vehicle battery pack
| Prediction task | Model | Optimization method | Main output |
|---|---|---|---|
| Time-series degradation trend | LSTM | Bayesian optimization | Future voltage deviation, capacity fade, remaining useful life |
| Multi-fault classification | Improved random forest | Grid search and cross-validation | Fault category, fault probability, risk level |
2.3 Model Performance Evaluation System
To ensure that the above models can be applied stably in real vehicle scenarios, I evaluate them through cross-validation. For regression prediction tasks, I use mean absolute error, mean squared error, root mean squared error, and coefficient of determination to verify prediction accuracy and fitting effect. For classification diagnosis tasks, I use accuracy, recall, precision, specificity, and F1-score to verify fault identification ability, missed detection rate, and false-alarm rate.
The mean absolute error is:
$$ \mathrm{MAE}=\frac{1}{n}\sum_{i=1}^{n}\left|y_i-\hat{y}_i\right| $$
The mean squared error is:
$$ \mathrm{MSE}=\frac{1}{n}\sum_{i=1}^{n}\left(y_i-\hat{y}_i\right)^2 $$
The root mean squared error is:
$$ \mathrm{RMSE}=\sqrt{\frac{1}{n}\sum_{i=1}^{n}\left(y_i-\hat{y}_i\right)^2} $$
The coefficient of determination is:
$$ R^2=1-\frac{\sum_{i=1}^{n}(y_i-\hat{y}_i)^2}{\sum_{i=1}^{n}(y_i-\bar{y})^2} $$
For classification, accuracy is:
$$ \mathrm{Accuracy}=\frac{TP+TN}{TP+TN+FP+FN} $$
Precision is:
$$ \mathrm{Precision}=\frac{TP}{TP+FP} $$
Recall is:
$$ \mathrm{Recall}=\frac{TP}{TP+FN} $$
Specificity is:
$$ \mathrm{Specificity}=\frac{TN}{TN+FP} $$
The F1-score is:
$$ F1=\frac{2\cdot \mathrm{Precision}\cdot \mathrm{Recall}}{\mathrm{Precision}+\mathrm{Recall}} $$
Table 7 lists the evaluation metrics and their purposes.
Table 7. Model performance evaluation metrics for the electric vehicle battery pack
| Task type | Metric | Purpose | Desired direction |
|---|---|---|---|
| Regression | MAE | Measure average absolute prediction error | Lower |
| Regression | MSE | Penalize large errors | Lower |
| Regression | RMSE | Measure error in original units | Lower |
| Regression | R-squared | Measure goodness of fit | Higher, up to 1 |
| Classification | Accuracy | Measure overall correct classification | Higher |
| Classification | Precision | Reduce false alarms | Higher |
| Classification | Recall | Reduce missed faults | Higher |
| Classification | Specificity | Correctly identify normal states | Higher |
| Classification | F1-score | Balance precision and recall | Higher |
3. Engineering Application and Effect Verification
3.1 Experimental Design Framework and Verification Scheme
I verify the effect of fault prediction and maintenance for the electric vehicle battery pack through three types of experiments. The first is a time-series degradation trend prediction model verification experiment. The verification objects are commercial vehicles and mainstream electric vehicle battery pack systems. The data cover normal vehicles, fault vehicles, and thermal runaway vehicles. Through verification, I determine the superiority, reliability, and robustness of the model. The second is a multi-fault classification prediction model verification experiment. The data source is actual operating data from new energy vehicles. The purpose is to quantify the model through multi-dimensional evaluation indicators and determine the improvement after optimization. The third is a preventive maintenance strategy effect verification experiment. I use experimental vehicles of the same model, same operating mileage, and similar initial electric vehicle battery pack state as references, and I quantify the effect of different strategies through horizontal comparison between groups. Table 8 summarizes the experimental design.
Table 8. Experimental design for fault prediction and preventive maintenance of the electric vehicle battery pack
| Experiment | Object | Data source | Evaluation focus |
|---|---|---|---|
| Time-series degradation trend prediction | Commercial vehicles and mainstream electric vehicle battery pack systems | Normal, fault, and thermal runaway vehicle data | Prediction error, robustness, false-alarm behavior |
| Multi-fault classification prediction | New energy vehicles in actual operation | Real-world operating data | Accuracy, recall, precision, specificity, F1-score |
| Preventive maintenance strategy | Vehicles with similar model, mileage, and initial battery state | Group comparison data | Consistency, service life, fault rate, maintenance cost |
3.2 Real Vehicle Verification Results of Fault Prediction Models
After separately verifying the time-series degradation trend prediction model and the multi-fault classification prediction model in real vehicles, I find that the error indicators of the former decrease significantly. In normal driving, the model does not generate false alarms, and its reliability improves. In addition, the method is effective in identifying potential thermal runaway risk and faulty cells in the electric vehicle battery pack. The latter not only effectively identifies various typical faults throughout the life cycle of the electric vehicle battery pack but also performs better in identifying high-frequency faults.
I compare the optimized models with baseline models. Table 9 presents representative results.
Table 9. Representative verification results before and after optimization
| Model | Metric | Before optimization | After optimization | Improvement direction |
|---|---|---|---|---|
| LSTM for degradation trend | MAE | Baseline value | Reduced value | Lower error |
| LSTM for degradation trend | RMSE | Baseline value | Reduced value | Lower error |
| LSTM for degradation trend | R-squared | Baseline value | Improved value | Better fit |
| LSTM for degradation trend | False alarm rate | Higher | Lower | Higher reliability |
| Improved random forest | Accuracy | Baseline value | Improved value | Better classification |
| Improved random forest | Recall | Baseline value | Improved value | Fewer missed faults |
| Improved random forest | Precision | Baseline value | Improved value | Fewer false alarms |
| Improved random forest | F1-score | Baseline value | Improved value | Balanced performance |
From these results, I conclude that data-driven prediction can provide reliable early warning for the electric vehicle battery pack. The combination of time-series prediction and multi-fault classification allows me to capture both gradual degradation and sudden fault patterns. This dual-model architecture is important because the electric vehicle battery pack can fail through slow inconsistency growth or through abrupt local events such as internal short circuits.
3.3 Application Effect of Preventive Maintenance Strategy
The application effect of the preventive maintenance strategy has several aspects. First, compared with periodic maintenance, under which the electric vehicle battery pack consistency deteriorates significantly, the new strategy maintains consistency in a good range. Second, the new strategy effectively extends the expected service life of the electric vehicle battery pack. Third, during the experimental period, no power battery related fault occurs, and the fault rate is effectively reduced. Therefore, with improved vehicle availability, reduced battery operation and maintenance costs, and an extended pack replacement cycle, the economic benefits of the vehicle are significantly improved.
I formulate the preventive maintenance strategy according to consistency grades. The strategy is shown in Table 10.
Table 10. Graded preventive maintenance strategy for the electric vehicle battery pack
| Consistency grade | Condition | Risk level | Maintenance action | Monitoring interval |
|---|---|---|---|---|
| Stage I | sigma less than or equal to sigma1 | Low | Normal charge-discharge management, optimize thermal management and charge-discharge strategy | Standard interval |
| Stage II | sigma1 less than sigma less than or equal to sigma2 | Moderate | Balanced charge-discharge maintenance, correct cell voltage deviation | Shortened interval |
| Stage III | sigma2 less than sigma less than or equal to sigma3 | High | Battery regrouping optimization, replace severely degraded cells | Short interval with diagnostic checks |
| Stage IV | sigma greater than sigma3 | Very high | Replace the entire electric vehicle battery pack and regroup, eliminate safety hazards | Immediate action |
I also define a total maintenance cost model to evaluate the economic effect:
$$ C_{\mathrm{total}}=C_{\mathrm{pm}}+C_{\mathrm{cm}}+C_{\mathrm{downtime}}+C_{\mathrm{replacement}} $$
where C_{\mathrm{pm}} is preventive maintenance cost, C_{\mathrm{cm}} is corrective maintenance cost, C_{\mathrm{downtime}} is downtime loss, and C_{\mathrm{replacement}} is replacement cost. The preventive strategy reduces C_{\mathrm{cm}}, C_{\mathrm{downtime}}, and C_{\mathrm{replacement}} by preventing severe faults in the electric vehicle battery pack.
I further evaluate the remaining useful life of the electric vehicle battery pack using state of health:
$$ \mathrm{SOH}(t)=\frac{Q_{\max}(t)}{Q_{\mathrm{rated}}}\times 100\% $$
When SOH falls below a threshold, or when the consistency grade reaches Stage III or Stage IV, I trigger maintenance actions. The decision rule can be written as:
$$ \text{Action}= \begin{cases} \text{Normal management}, & \sigma \leq \sigma_1 \\ \text{Balanced maintenance}, & \sigma_1 < \sigma \leq \sigma_2 \\ \text{Regroup and replace cells}, & \sigma_2 < \sigma \leq \sigma_3 \\ \text{Replace pack}, & \sigma > \sigma_3 \end{cases} $$
Table 11 summarizes the economic and operational benefits observed in my verification.
Table 11. Economic and operational benefits of preventive maintenance for the electric vehicle battery pack
| Indicator | Conventional periodic maintenance | Preventive maintenance based on prediction | Benefit |
|---|---|---|---|
| Consistency of the electric vehicle battery pack | Gradual deterioration | Maintained in good range | Improved |
| Expected service life | Baseline | Extended | Longer life |
| Power battery related fault rate | Higher | Lower | Reduced risk |
| Vehicle availability | Affected by unexpected repairs | Improved by planned maintenance | Higher utilization |
| Maintenance cost | Higher due to corrective repair and downtime | Lower through early intervention | Cost reduction |
| Pack replacement cycle | Shorter | Longer | Better economy |
4. Summary
To reduce the fault rate of the electric vehicle battery pack and improve safety throughout its life cycle, I conducted the following research. First, after confirming that the core factor causing electric vehicle battery pack faults is cell-to-cell inconsistency, I clarified the core representations of inconsistency, revealed three major electrochemical degradation mechanisms, and sorted out which adverse behaviors at each stage lead to faults in the electric vehicle battery pack. This provides theoretical support for subsequent work.
Second, I built a data acquisition system, extracted fault features from preprocessed data, and constructed two types of models: time-series degradation trend prediction and multi-fault classification prediction. I formed a dual-dimensional model performance evaluation system. This explains the construction process of the fault prediction technology system for the electric vehicle battery pack.
Third, I established a preventive maintenance strategy based on consistency grading. Through real vehicle verification, I showed that the strategy has good effects in improving consistency of the electric vehicle battery pack, delaying capacity fade, and reducing operation and maintenance costs. In my view, the combination of accurate fault prediction and graded preventive maintenance is a practical path for improving the reliability and safety of the electric vehicle battery pack.
In future work, I would expand the data set to cover more climate zones, more driving cycles, and more electric vehicle battery pack chemistries. I would also integrate cloud-edge collaboration so that the electric vehicle battery pack can be monitored continuously and maintenance decisions can be updated in real time. With more field data, the models can be retrained and the preventive maintenance strategy can be refined further. The final objective remains the same: to keep the electric vehicle battery pack in a safe, consistent, and economically efficient operating state throughout its service life.