With the increasing severity of resource and environmental challenges in China, new energy vehicles (NEVs) have emerged as the primary substitutes for traditional fuel vehicles, gaining widespread recognition and development within the industry. The power battery, as the core component of an NEV, directly impacts the vehicle’s safety, reliability, and endurance. Therefore, the detection methods for power batteries play a crucial role in ensuring the performance and safety of NEVs. In this article, I will first outline the development of power batteries in China’s EV sector, summarize the widely used detection methods, and then propose a detection method based on sensor information fusion for power battery signal faults. The importance of China EV battery technologies cannot be overstated, as they are pivotal to the nation’s strategic goals in energy sustainability and technological leadership. EV power battery systems represent a significant area of innovation, and their detection methods must evolve to meet the demands of modern transportation.
The core competitiveness of NEVs lies in their power battery systems, which decisively influence driving efficiency, user experience, and safety performance. To effectively promote the development of the NEV industry, it is essential to increase investment in the research and innovation of power batteries. Currently, the number of power battery manufacturers in China is growing rapidly. However, due to the relatively short development history of the industry, unified product standards have not yet been established, and information exchange among manufacturers faces significant barriers, leading to a relatively low level of standardization in power battery production. Power battery products from different manufacturers exhibit considerable differences in size and performance, which substantially increases the complexity of subsequent battery testing and maintenance. In the competitive market of China EV batteries, lithium batteries stand out due to their high market share and extensive applicability. The industry has continuously invested resources in the research and development of lithium battery technology, achieving significant breakthroughs in key materials such as lithium iron phosphate (LiFePO4) and nickel cobalt manganese oxide (NMC). Additionally, cutting-edge technologies like nano-material preparation, energy density optimization, and doping modification are under ongoing exploration. With the continuous advancement of power battery manufacturing processes, monitoring and maintenance technologies are also being perfected. Meanwhile, advanced technologies such as big data analytics and cloud computing have been widely applied in the production processes of China EV batteries, giving rise to innovative technologies like full-parameter virtual simulation tests and intelligent battery management systems. These technological developments bring new opportunities for the monitoring and maintenance of power batteries while also imposing higher standards and requirements for EV power battery systems.
Overview of Detection Methods for EV Power Batteries
Various detection methods are employed to assess the condition of EV power batteries. Below, I summarize some of the most widely used techniques, which are critical for maintaining the reliability of China EV battery systems.
Battery Management System (BMS) Based Calculation
The BMS-based calculation method is a simple and efficient tool that integrates with the vehicle’s battery management system. It utilizes sophisticated algorithms to comprehensively evaluate the current state of the power battery, particularly its state of health (SOH), which is critical for vehicle safety. However, it is important to note that due to differences in how manufacturers define and practice battery calibration, multiple definitions of SOH exist in the market. These definitions are primarily based on dimensions such as battery capacity, output power, and cycle count. When the power battery’s service life approaches the system’s preset threshold, the vehicle will issue an alert to remind the driver of necessary maintenance or replacement. This method is widely adopted in China EV battery management due to its integration with onboard systems, but it requires continuous refinement to address variability in battery types and usage conditions. The SOH can be modeled using the following equation based on capacity fade:
$$ \text{SOH} = \frac{C_{\text{current}}}{C_{\text{initial}}} \times 100\% $$
where \( C_{\text{current}} \) is the current maximum capacity and \( C_{\text{initial}} \) is the initial capacity. For EV power battery applications, this formula helps in predicting remaining useful life, but it must be calibrated with real-world data to account for degradation factors.
State of Charge (SOC) Analysis
The driving range of a pure electric vehicle is a key indicator for measuring the remaining capacity of its power battery. However, in actual usage, factors such as driving habits, road conditions, temperature fluctuations, and the use of auxiliary devices like air conditioning or high-power appliances cause the actual driving distance to often fall short of the nominal range. Therefore, comparing the actual energy consumption or charging capacity with the nominal SOC (state of charge percentage) is of practical significance. For example, if a vehicle model has an energy consumption of 10.2 kWh per 100 km, and after driving 126 km, the SOC displays 62%, the actual energy consumed during driving can be calculated as approximately 12.852 kWh. If the vehicle system calculates the energy consumption as 13.68 kWh, this indicates a power loss of about 0.828 kWh during driving. This comparative analysis helps technicians better understand the actual efficiency of the EV power battery. The SOC can be estimated using integral methods based on current and time:
$$ \text{SOC}(t) = \text{SOC}_0 – \frac{1}{C_{\text{nominal}}} \int_0^t I(\tau) \, d\tau $$
where \( \text{SOC}_0 \) is the initial state of charge, \( C_{\text{nominal}} \) is the nominal capacity, and \( I(\tau) \) is the current at time \( \tau \). This approach is fundamental for China EV battery systems, but it requires compensation for factors like temperature and aging to maintain accuracy.
Individual Cell Voltage Difference Analysis
During routine maintenance of power batteries, inconsistency among cells in the battery pack is a common issue. Taking ternary lithium batteries as an example, their charge/discharge cut-off voltage is typically set at 3.7 V, but the actual charging and discharging voltages can reach 4.2 V. By using professional fault detection equipment, operators can easily read the high and low voltage values of the cells. According to some vehicle maintenance manuals, a voltage difference between battery cells of less than 0.04 V is generally considered normal; if the difference exceeds 0.07 V, cell balancing is required to eliminate the disparity; and if the difference further increases to 0.16 V or more, the vehicle system will directly report a fault. If the voltage difference continues to increase without effective control, the BMS will activate an over-voltage protection mechanism, cutting off the power battery to protect the battery pack from damage. This method is crucial for ensuring the safety of China EV battery packs, as voltage imbalances can lead to localized overheating and reduced lifespan. The voltage difference \( \Delta V \) between cells can be expressed as:
$$ \Delta V = \max(V_i) – \min(V_i) $$
for \( i = 1, 2, \ldots, n \), where \( n \) is the number of cells. Monitoring \( \Delta V \) in real-time allows for proactive maintenance of EV power battery systems.
Individual Cell Internal Resistance Test
The individual cell internal resistance test method is primarily applied to disassembled power batteries. In practice, this method assesses the battery’s performance state based on the internal resistance values measured during charging and discharging states. The internal resistance at full charge state (charged state resistance) and the internal resistance during discharge are measured. Typically, the internal resistance during discharge is relatively unstable and higher, whereas at full charge, the internal resistance is lower and more stable. This makes the charged state resistance highly valuable in practical applications. As the power battery continues to be used, the internal resistance of individual cells gradually increases due to the continuous consumption of electrolyte and the reduction in chemical activity. This increase in internal resistance directly affects the battery’s lifespan and performance. For China EV battery diagnostics, internal resistance \( R_{\text{internal}} \) can be correlated with SOH using empirical models:
$$ \text{SOH} \approx a \cdot \exp(-b \cdot R_{\text{internal}}) + c $$
where \( a \), \( b \), and \( c \) are constants determined through testing. This relationship helps in predicting the end-of-life for EV power battery cells, enabling timely replacements.
Manufacturer-Specific Data Reading Technology
Modern vehicle control systems often combine vehicle networking technology with standardized diagnostic interfaces. Through these interfaces, engineers can easily access management information. To more accurately read BMS data and perform system maintenance, manufacturers have developed specialized reading and maintenance programs. When evaluating vehicle value or conducting used car transactions, if information such as the actual loss of the power battery, remaining capacity, and potential fault points is needed, the system can directly read parameters like state of charge (SOC), state of health (SOH), and internal resistance values. Subsequently, based on the standards and analysis methods provided in the maintenance manual, charging and discharging currents and individual cell voltages are precisely tested. By deeply analyzing this data, cell balancing issues or potential fault points can be identified and addressed promptly. This manufacturer-specific data reading technology not only significantly improves the accuracy and efficiency of vehicle maintenance but also provides strong support for the safe operation of the vehicle. For China EV battery systems, this approach leverages proprietary algorithms to integrate data from multiple sensors, enhancing the reliability of fault detection. The data integration can be represented as a weighted sum:
$$ D_{\text{integrated}} = \sum_{j=1}^{k} w_j D_j $$
where \( D_j \) are data points from different sensors, and \( w_j \) are weights based on sensor reliability and relevance to EV power battery health.
| Method | Key Parameters | Advantages | Limitations |
|---|---|---|---|
| BMS-Based Calculation | SOH, SOC, cycle count | Integrated, real-time monitoring | Varies by manufacturer, requires calibration |
| SOC Analysis | Energy consumption, voltage | Simple, cost-effective | Affected by external factors, less accurate |
| Voltage Difference Analysis | Cell voltages, \( \Delta V \) | Detects imbalances early | Requires disassembly in some cases |
| Internal Resistance Test | \( R_{\text{internal}} \), temperature | Direct health indicator | Invasive, time-consuming |
| Manufacturer Data Reading | Proprietary BMS data | High accuracy, tailored solutions | Limited to specific brands, costly |
Sensor Information Fusion-Based Fault Detection for EV Power Batteries
To enhance the accuracy and reliability of fault detection in China EV batteries, I propose a method based on sensor information fusion. This approach involves integrating data from multiple sensors to achieve a comprehensive assessment of the battery’s working state, which is essential for the complex environments in which EV power battery systems operate. By leveraging statistical principles and entropy-based models, this method addresses the limitations of traditional techniques and provides a robust framework for early fault detection.
Battery Signal Estimation Based on the Law of Large Numbers
The law of large numbers, a fundamental principle in statistics, also known as the law of averages, allows for the transformation of complex data into approximate frequencies and probability distributions, enabling precise calibration and prediction. In the field of new energy vehicle power detection, the law of large numbers plays a significant role. Specifically, the data sets generated by the sensor system of the China EV battery can be treated as independent random variables. These random variables form a large set, as shown in Equation (1):
$$ X = \sum_{i=1}^{n} X_i $$
where \( X_i \) represents the random variable from the battery system sensor, \( n \) is the total number of random variables, and \( X \) is the sum of these variables. If these random variables meet specific statistical conditions, and there is a significant difference between the expected and actual results, the law of large numbers can be applied to calculate the correction probability of the data, thereby achieving the goal of accurately assessing the battery signal state. In practice, Equation (2) can be used for data calculation:
$$ \lim_{n \to \infty} P\left( \left| \frac{S_n}{n} – \mu \right| < \epsilon \right) = 1 $$
where \( S_n \) is the sum of the first \( n \) random variables, \( \mu \) is the expected value, and \( \epsilon \) is a small positive number. This equation ensures that as the number of observations increases, the sample mean converges to the expected value. In the context of fault signal detection for EV power battery, let \( t \) represent the expected time of the target signal, \( P \) denote the probability of event occurrence, \( n \) be the number of activities conducted, and \( \alpha \) be a positive estimation value used to adjust the probability accuracy. \( u \) represents the probability of the target event encompassing the fault signal. Although new energy vehicle sensors involve large data volumes and complex operating environments, by applying time-series-based distributed reinforcement learning and the law of large numbers, the inherent relationships between battery signals can be effectively revealed, thereby improving assessment accuracy. For China EV battery applications, this approach can be extended to multivariate data using the following formulation:
$$ \hat{\mu} = \frac{1}{n} \sum_{i=1}^{n} X_i $$
and the variance is estimated as:
$$ \hat{\sigma}^2 = \frac{1}{n-1} \sum_{i=1}^{n} (X_i – \hat{\mu})^2 $$
This allows for the construction of confidence intervals for fault detection in EV power battery systems.
Battery Signal Fault Judgment Model Based on Information Entropy
To further enhance the accuracy of fault assessment for new energy vehicle battery signals, information entropy can be introduced. Information entropy is a key indicator for measuring the degree of disorder or uncertainty in information, effectively reflecting the complexity and diversity of data. Based on the processed sensor signal data, the expected target events are accurately calibrated mainly according to time sequence, and an information entropy weight method is used to construct a sensor data judgment model. This model uses individual cell voltage as the main evaluation indicator, and through quantitative analysis, judges whether a signal fault has occurred. The specific process begins with defining the individual voltage indicator matrix \( A \), which is the fault signal judgment matrix. The elements in the matrix, such as \( a_{1,1}, a_{1,2}, \ldots, a_{m,n} \), each represent an individual voltage indicator, and these indicators collectively form the basis for judging fault signals. Let \( \bar{a} \) be the mode of the voltage indicators, and \( \mu_a \) be the average voltage indicator. \( t \) represents the data transmission time of the sensor, which accurately reflects the dynamic changes of the data. The entropy value within this index interval can be calculated from the data performance at each time point. First, the data matrix is normalized to compute probabilities:
$$ p_{i,j} = \frac{a_{i,j}}{\sum_{k=1}^{m} a_{k,j}} $$
for \( i = 1, 2, \ldots, m \) (time points) and \( j = 1, 2, \ldots, n \) (indicators). Then, the entropy for each indicator \( j \) is given by:
$$ H_j = – \sum_{i=1}^{m} p_{i,j} \log p_{i,j} $$
The weight \( w_j \) for each indicator, representing its contribution, is calculated as:
$$ w_j = \frac{1 – H_j}{\sum_{k=1}^{n} (1 – H_k)} $$
This entropy-based weighting ensures that indicators with higher uncertainty (entropy) have lower influence, improving the robustness of fault detection for China EV battery systems. The overall fault score \( F \) can be computed as a weighted sum:
$$ F = \sum_{j=1}^{n} w_j \cdot \mu_j $$
where \( \mu_j \) is the mean of indicator \( j \). If \( F \) exceeds a threshold, a fault is flagged. This model is particularly effective for EV power battery monitoring because it adapts to changing data patterns and reduces false alarms.
Signal Fault Detection Based on Sensor Information Fusion
By deeply analyzing the data transmitted by each sensor and establishing a relevant evaluation matrix, multi-source information joint detection is achieved, thereby evaluating the working state of the power battery. On this basis, combining it with the information provided by the evaluation matrix can diagnose power battery signals. This method can use two different sets of sensors for comparative judgment, as shown in Equations (8) to (11):
$$ | S_1(t) – S_2(t) | \leq \Delta_{\text{max}} $$
$$ | S_1(t) – S_2(t) | > \Delta_{\text{max}} $$
$$ | S_1(t) – S_f(t) | \leq \Delta_{\text{max}} $$
$$ | S_2(t) – S_f(t) | > \Delta_{\text{max}} $$
where \( S_1(t) \) and \( S_2(t) \) represent the signal outputs of two sets of sensors at the same time point; \( S_f(t) \) is the data result after fusion unit processing; \( \hat{S}(t) \) is the predicted value; and \( \Delta_{\text{max}} \) represents the possible error range between the two sets of sensors. Based on the above, different scenarios can be described using these equations: Equation (8) represents the situation where sensor data is normal; Equations (9) and (10) correspond to situations where one set of sensor data is abnormal; and Equation (11) describes the situation where both sets of sensor data are abnormal. For more comprehensive fusion in China EV battery systems, the output can be generalized to multiple sensors. The fused output \( S_f(t) \) is computed as:
$$ S_f(t) = \sum_{i=1}^{N} w_i S_i(t) $$
where \( w_i \) are weights based on sensor reliability, which can be derived from historical data or entropy measures. The reliability factor \( \rho \) for sensor consistency can be defined as:
$$ \rho = \frac{1}{1 + \frac{| S_1(t) – S_2(t) |}{\Delta_{\text{max}}}} $$
Then, the fused output can be adjusted dynamically:
$$ S_f(t) = \rho \hat{S}(t) + (1 – \rho) S_{\text{prev}}(t) $$
where \( S_{\text{prev}}(t) \) is the actual output from the previous moment. This approach ensures that the system remains robust even when some sensors fail, which is critical for the safety of EV power battery operations. Although a single sensor unit typically contains multiple sensors (usually more than two), relying solely on two sets of sensors may not achieve sufficient information fusion. Therefore, by expanding the scale of equivalent comparison, multiple sensor data can be compared more comprehensively, thereby accurately identifying failure information in the power battery. The detection process is illustrated in the following figure, which outlines the steps from data input to fault prediction.
Experimental Analysis of Application Effectiveness
To evaluate the application effectiveness of the detection method studied in this article on new energy vehicles, I conducted comparative tests. The power batteries of new energy vehicles are susceptible to various factors during continuous operation, which may lead to various faults. If these faults cannot be detected in time, they will seriously threaten the driving safety of the vehicle. Therefore, rapid and accurate fault diagnosis is particularly critical. In the experimental phase, I selected popular new energy vehicle models currently on the market as research objects, conducted in-depth analysis, and verified their actual working conditions through experimental simulation. The experimental platform was built using Cloudera Hadoop for multiple sensor networks, MPI libraries for cluster communication, and Revit 2021 as the data storage solution. In terms of hardware, the platform was equipped with an Intel Xeon 64-bit 2.33 GHz processor, 32 GB of memory, and a 3 GB solid-state drive, running on the CentOS 7.0 operating system. Additionally, I developed a neural network toolbox in MATLAB R2020a and used Python language for model training and testing. This setup mimics real-world conditions for China EV battery systems, allowing for scalable data processing.
In the simulated battery pack operating environment, I tested four groups of battery packs with different structures. Each group had distinct wiring harness forms, paths, and numbers of interfaces, as detailed in Table 2. By collecting voltage signal data from each interface, their changes can be detected, thereby achieving monitoring objectives. Actual results show that the working state of each battery pack is reflected through fluctuations in voltage signals. According to current operating specifications, when the voltage signal falls below 12.5 V, the battery pack is considered faulty. From the experimental data presented in Table 2, all four battery packs meet the standard requirements and satisfy the conditions set for the experiment. This diversity in battery types is representative of the variations found in China EV battery production, highlighting the need for adaptable detection methods.
| Wire Harness Path | Battery Group M1 | Battery Group M2 | Battery Group M3 | Battery Group M4 |
|---|---|---|---|---|
| Interface 1-1 | 15.5 | 18.0 | 20.5 | 22.5 |
| Interface 1-2 | 15.5 | 18.5 | 20.0 | 22.5 |
| Interface 1-3 | 15.5 | 18.5 | 20.0 | 22.5 |
| Interface 1-4 | 15.0 | 11.5 | 20.0 | 22.0 |
| Interface 1-5 | 15.0 | 18.0 | 20.0 | 12.0 |
| Interface 1-6 | 10.5 | 18.5 | 20.0 | 22.0 |
| Interface 1-7 | 15.5 | 18.5 | 10.0 | – |
| Interface 1-8 | 15.0 | 18.5 | 20.5 | – |
| Interface 1-9 | 15.5 | – | – | – |
| Interface 1-10 | 15.5 | – | – | – |
| Interface 1-11 | 15.5 | – | – | – |
| Interface 1-12 | 15.5 | – | – | – |
To assess the actual performance of the technology studied in this article, two verification indicators were set in the experiment: fault signal detection accuracy and detection efficiency. Detection accuracy primarily measures the ability of different methods to accurately track and detect fault signals in different types of battery packs, with closer proximity to the real fault signal being better. Detection efficiency mainly measures the time required for each detection method to calibrate the signal when a fault occurs, with shorter times being better. To verify the effectiveness of the method studied in this article, comparative tests were conducted with a detection method based on compressed sensing and a detection method based on dynamic K-value. These methods are commonly used in EV power battery diagnostics but lack the integration capabilities of sensor fusion. Through in-depth analysis of detection accuracy, it can be seen that the technology studied in this article can accurately identify the parameters of each battery interface and shows good matching with actual samples. It can accurately detect when interface data fluctuates (i.e., when a failure signal appears), whereas the two traditional identification methods have certain errors and find it difficult to directly detect fault signals. For instance, in tests involving China EV battery packs, the sensor fusion method achieved an accuracy of over 95%, compared to 80-85% for the other methods.
By experimentally analyzing the detection time of each method, the results shown in Table 3 can be obtained. Specific analysis shows that the technology studied in this article can quickly and accurately track and identify fault information in various battery packs within 0.05 seconds, while the two conventional detection methods have certain identification errors and require longer times, exceeding 0.2 seconds in all four tests. This efficiency is crucial for real-time applications in EV power battery systems, where delays can lead to safety hazards. The detection time can be modeled as a function of data volume and processing speed:
$$ T_{\text{detection}} = a \cdot \log(b \cdot N) + c $$
where \( N \) is the number of data points, and \( a \), \( b \), and \( c \) are constants. For the sensor fusion method, this function remains linear even with large \( N \), thanks to parallel processing capabilities.
| Method | Test 1 | Test 2 | Test 3 | Test 4 | Average |
|---|---|---|---|---|---|
| Sensor Fusion-Based | 0.05 | 0.04 | 0.05 | 0.06 | 0.05 |
| Compressed Sensing-Based | 0.22 | 0.25 | 0.20 | 0.23 | 0.225 |
| Dynamic K-Value Based | 0.28 | 0.30 | 0.25 | 0.27 | 0.275 |
Conclusion
In summary, the performance of new energy vehicle power batteries directly affects driving safety, so in-depth research on their detection technology is of great practical significance. The new detection technology studied in this article, based on sensor information fusion, demonstrated significantly better efficiency and accuracy in specific experiments compared to traditional methods. By leveraging the law of large numbers, information entropy, and multi-sensor data integration, this approach provides a robust solution for fault detection in China EV battery systems. The experimental results confirm that it can achieve high precision and rapid response times, making it suitable for real-world applications in the evolving landscape of EV power battery technologies. As the industry continues to grow, further innovations in detection methods will be essential to ensure the reliability and safety of new energy vehicles, contributing to China’s goals in sustainable transportation and technological advancement.