Lithium-ion batteries, particularly LiFePO4 batteries, are widely adopted in electric vehicles and energy storage systems due to their high energy density, long cycle life, and safety features. However, as these batteries age, they become susceptible to internal faults such as micro-short circuits, which pose significant safety risks. A micro-short circuit refers to a minor internal short where the separator is compromised, allowing direct contact between the positive and negative electrodes, leading to continuous energy dissipation. If undetected, this can escalate to severe internal short circuits, thermal runaway, and potential fires or explosions. Therefore, developing effective diagnostic methods for micro-short circuits in battery packs is critical for ensuring operational safety and reliability. In this article, I propose a novel diagnostic approach based on the relative charging time of individual cells within a LiFePO4 battery pack, which leverages routine charging data without requiring specialized equipment or disruptive testing.
Existing methods for detecting micro-short circuits in batteries often involve dedicated apparatus or complex model-based techniques. For instance, some approaches use pressure-based devices to measure insulation resistance between electrodes, but these are typically applicable only to single cells and require physical manipulation. Others rely on equivalent circuit models to estimate internal short-circuit resistance by quantifying self-discharge currents between charging cycles. However, these methods depend on threshold settings that may not be accurate for aged batteries, and they can be computationally intensive. To address these limitations, my method focuses on analyzing the charging behavior of battery packs during normal operation, specifically by examining the time each cell takes to reach the charge cutoff voltage relative to a reference cell. This approach is non-intrusive, easy to implement, and suitable for online monitoring in real-world applications.
The core principle of my diagnostic method stems from the characteristic behavior of a micro-short circuit in a LiFePO4 battery. When a cell develops a micro-short, it continuously loses energy, causing it to require more charging time to reach the same voltage level compared to healthy cells. In a series-connected battery pack, charging stops when the first cell hits the upper voltage limit, leaving other cells undercharged. By comparing the charging curves of all cells against the first cell to finish, I can derive a metric called the relative charging time, which captures the additional time needed for each cell to fully charge. Over multiple charging cycles, the relative charging time for a micro-shorted cell will progressively increase, while it remains stable for normal cells. This trend forms the basis for fault detection and localization.
To formalize this, let me define the relative charging time for a cell in a battery pack. During the nth charging cycle, denote the time when the first cell reaches the cutoff voltage as $$t_{n,0}$$. For any other cell j that has not yet reached the cutoff voltage at this moment, if it could be charged individually, its time to reach the cutoff voltage would be $$t_{n,j}$$. The relative charging time is then calculated as:
$$\Delta t_{n,j} = t_{n,j} – t_{n,0}$$
In practice, since individual charging is not feasible, I estimate $$t_{n,j}$$ by using the charging voltage curve of the first cell as a reference. Specifically, I map the end-of-charge voltage of cell j onto the reference curve to find the corresponding time, yielding an approximate relative charging time. This approximation is valid under consistent charging conditions and minimal cell-to-cell variations.
The diagnostic process involves tracking changes in $$\Delta t_{n,j}$$ across consecutive charging cycles. I define a parameter $$K_{n-n-1,j}$$ to quantify the relative change:
$$K_{n-n-1,j} = \frac{|\Delta t_{n,j} – \Delta t_{n-1,j}|}{\Delta t_{n,j}}$$
For a healthy LiFePO4 battery cell, $$K_{n-n-1,j}$$ tends to be small and consistent over time, whereas for a micro-shorted cell, it increases due to the growing $$\Delta t_{n,j}$$. To identify anomalies, I employ a statistical tool called the box plot (or box-and-whisker plot), which highlights outliers in the $$K_{n-n-1,j}$$ values across all cells in the pack. Cells with $$K_{n-n-1,j}$$ values outside the interquartile range are flagged as potential fault candidates. However, since high internal resistance can also cause fluctuations in charging time, I integrate direct current internal resistance (DCIR) analysis to refine the diagnosis. The DCIR for each cell is computed from voltage measurements during charging initiation:
$$R_{k,j} = \frac{U_{2-k,j} – U_{1-k,j}}{I_k}$$
where $$U_{1-k,j}$$ is the resting voltage before the kth test, $$U_{2-k,j}$$ is the voltage 10 seconds after charging starts, and $$I_k$$ is the charging current. By checking the consistency of DCIR values across cells, I can discriminate between micro-shorts and resistance-related anomalies. The overall workflow includes data selection, relative charging time calculation, outlier detection via box plots, and DCIR validation, culminating in pinpointing the faulty cell.
To validate my method, I conducted experiments on a retired LiFePO4 battery pack from an electric vehicle application. The pack consisted of 13 modules connected in series, with each module comprising 6 soft-pack LiFePO4 cells in parallel. Each cell had a nominal capacity of 50 Ah and a voltage range of 2.7 V to 3.6 V. Testing was performed using a high-power battery cycler within a temperature-controlled chamber set at 45°C to accelerate aging and simulate harsh conditions. The pack underwent repeated charge-discharge cycles: discharging at 300 A until any module hit 2.7 V, followed by charging at 300 A until any module reached 3.6 V. Data from three separate test runs were collected, focusing on charging phases where the pack behaved normally.
Data selection is crucial for reliable analysis, as variations in charging capacity and end voltages can obscure trends. I applied three criteria to filter the charging data: (1) adjacent charging capacities must differ by less than 2 Ah to ensure consistency; (2) voltage differences for each module between consecutive charges must be within ±10 mV to minimize noise; and (3) the module that first reaches the cutoff voltage should remain the same across selected cycles to avoid基准 shifts. The table below summarizes the charging cycles retained after filtering:
| Test Run | Selected Charging Cycles (Cycle Numbers) |
|---|---|
| 1 | 6–9 |
| 2 | 11–17 |
| 3 | 20–25, 27 |
For each selected cycle, I identified the reference module (the first to finish charging) and computed $$\Delta t_{n,j}$$ for all other modules. As an example, consider a charging curve where module 10 reaches 3.6 V first; its voltage profile serves as the reference for estimating $$\Delta t_{n,j}$$ for modules 1–9 and 11–13. The $$K_{n-n-1,j}$$ values were then derived and aggregated across cycles. The box plot analysis of these values revealed several outliers, primarily in modules 2 and 6. To determine whether these anomalies stemmed from micro-shorts or high internal resistance, I calculated the DCIR for each module over the three test runs. The results are presented in the following table:
| Module Number | Average DCIR (mΩ) – Test 1 | Average DCIR (mΩ) – Test 2 | Average DCIR (mΩ) – Test 3 |
|---|---|---|---|
| 1 | 0.85 | 0.88 | 0.91 |
| 2 | 0.82 | 0.83 | 0.89 |
| 3 | 0.79 | 0.81 | 0.84 |
| 4 | 0.80 | 0.82 | 0.86 |
| 5 | 0.83 | 0.85 | 0.88 |
| 6 | 0.95 | 0.98 | 1.02 |
| 7 | 0.90 | 0.92 | 0.95 |
| 8 | 0.81 | 0.84 | 0.87 |
| 9 | 0.84 | 0.86 | 0.90 |
| 10 | 0.91 | 0.93 | 0.96 |
| 11 | 0.80 | 0.83 | 0.85 |
| 12 | 0.82 | 0.85 | 0.88 |
| 13 | 0.79 | 0.82 | 0.84 |
From the DCIR data, module 6 consistently exhibits the highest resistance, which explains its anomalous $$K_{n-n-1,j}$$ values. A box plot of DCIR across modules confirmed no extreme outliers, but module 6’s resistance is notably elevated. To isolate the effect of micro-shorts, I excluded modules with DCIR above the 75th percentile (modules 6, 7, and 10) and re-analyzed the $$K_{n-n-1,j}$$ values for the remaining modules. The subsequent box plot highlighted module 2 as the most frequent outlier, with $$K_{n-n-1,j}$$ deviations recurring across multiple cycles. This pattern aligns with the expected behavior of a micro-shorted LiFePO4 battery cell, where energy loss leads to progressively longer relative charging times.
Further evidence came from the voltage profiles during testing. In the later stages, the entire battery pack failed to charge or discharge properly, with module 2 hitting voltage limits abruptly. This indicates rapid capacity fade in module 2, characteristic of an advancing micro-short circuit. Thus, my diagnostic method successfully identified module 2 as the faulty unit, corroborating the actual failure observed in the experiments. The integration of relative charging time analysis and DCIR filtering proved effective in distinguishing micro-shorts from other aging effects like resistance growth.
The advantages of this method are manifold. It operates solely on standard charging data, making it suitable for integration into battery management systems (BMS) without additional hardware. The computational load is minimal, relying on basic arithmetic and statistical tools, which facilitates real-time implementation. Moreover, it is scalable to large LiFePO4 battery packs used in grid storage or electric vehicles, as it processes each cell independently within the pack framework. To enhance robustness, future work could incorporate machine learning algorithms to automate outlier detection or adapt to varying operating conditions such as temperature fluctuations and load profiles.
In conclusion, I have developed and validated a diagnostic technique for detecting micro-short circuits in LiFePO4 battery packs based on relative charging time dynamics. By monitoring changes in this parameter and cross-verifying with internal resistance measurements, the method accurately locates faulty cells while minimizing false positives. This approach contributes to the safety and longevity of lithium-ion battery systems, offering a practical solution for preventive maintenance. As the adoption of LiFePO4 batteries expands in renewable energy and transportation, such diagnostic tools will become increasingly vital for ensuring reliable and hazard-free operation.