Vehicle traction battery performance degradation has become one of the most critical issues restricting the large-scale adoption of electric vehicles. Among various battery chemistries, lithium iron phosphate batteries are widely used in electric vehicles due to their high thermal stability, long cycle life, and relatively low cost. However, even LFP-based vehicle traction battery systems inevitably suffer from capacity fade and power decay during long-term operation. The degradation phenomena observed at the full-cell level, such as discharge capacity loss and impedance rise, originate from complex internal side reactions including lithium inventory consumption, active material structural deterioration, electrolyte decomposition, and interfacial film growth. A reliable diagnosis of these underlying degradation mechanisms is therefore of essential importance for battery health management, residual-life prediction, and safety assessment. Degradation of lithium-ion batteries can generally be attributed to two main categories: loss of lithium inventory and loss of active material. LAM may take place in the positive electrode, in the negative electrode, or in both electrodes simultaneously. In this work, I focused on the quantitative identification of LLI and LAM in a commercial lithium iron phosphate vehicle traction battery through both macroscopic electrical characterization and microscopic material analysis, and further developed a degradation-mode diagnosis method based on incremental capacity curves that is suitable for practical application.
The existing literature indicates that two major methodologies have been adopted for battery degradation analysis. The first methodology is non-destructive external-characteristic analysis, which infer degradation modes from measured voltage, current, and capacity data. This approach is convenient for onboard applications, but it typically relies on model assumptions and requires validation from direct physical observations. The second methodology is destructive post-mortem analysis, which disassembles aged cells and examines electrode materials with techniques such as scanning electron microscopy, X-ray diffraction, and half-cell testing. This approach can directly reveal aging mechanisms but inevitably destroys the investigated cell. Given the complementarity of these two approaches, I proposed in this thesis a combined micro–macro diagnostic framework. Macro-level degradation modes diagnosed from open-circuit voltage curves were quantitatively compared with micro-level measurements obtained from electrode half-cells and XRD characterization. After establishing the consistency between macroscopic and microscopic results, the study was further extended to IC curve-based diagnosis, which is more convenient for real vehicle traction battery management systems.
The research described in this thesis was conducted through systematic experiments and modeling. I first designed and performed a series of full-cell tests on commercial LFP cells at different states of health. The full-cell tests included capacity calibration, cycle aging, rate capability tests, and low-current reference charge/discharge tests. Then, cells at different SOH levels were disassembled in an argon-filled glove box. The harvested electrode sheets were subjected to scanning electron microscopy imaging and X-ray diffraction characterization. Some electrode sheets were assembled into half-cells with lithium metal counter electrodes to obtain electrode potential curves and half-cell capacity information. By comparing the micro-scale test results with macro-scale OCV-diagnosed degradation modes, I established a quantitative connection between degradation modes at the two levels. Finally, based on the validated degradation diagnosis model, I systematically investigated the influence of LLI and LAM on OCV and IC curves and proposed a practical IC-based degradation-mode diagnosis method.
The remaining parts of this article are organized as follows. The cell specifications and experimental platform are described first. The experimental procedures are then explained in detail, including full-cell characterization, cycling aging, electrode post-mortem analysis, and half-cell performance tests. A qualitative correlation between micro-scale features and macro-scale degradation is subsequently presented. After that, the OCV-based degradation diagnosis model is developed with coordinate transformation and quantitative calculation results. Micro-scale quantification of degradation modes is then discussed, including half-cell capacity analysis and XRD-based maximum lithium content calculation, together with the micro–macro comparison. Finally, the diagnosis method based on IC curves is presented, covering the influence of different degradation modes on IC features, the diagnostic formulas, and the adaptability analysis under different charge/discharge rates.
Cell Specifications and Experimental Platform
The vehicle traction battery investigated in this study was a commercial square-format lithium iron phosphate cell with a rated capacity of 20 Ah. The basic specifications of the cell are summarized in the following table.
| Property | Value |
| Nominal capacity | 20 Ah |
| Nominal voltage | 3.2 V |
| Internal resistance | 7 mΩ |
| Mass | 485 ± 15 g |
| Dimension | 130 × 70 × 27 mm |
All electrical tests were performed using an NBT5V20AC16-T battery test system with a voltage range of 0–5 V and a current range of 0–20 A. The temperature during all tests was maintained at 25 °C by placing the cells inside an ETE-GDJS-150L temperature chamber. For half-cell tests, a CT-4008Tn-5V10mA battery tester was used inside a BLC-150 half-cell temperature chamber. During the disassembly process, an argon-filled glove box was employed to prevent air exposure and unwanted side reactions of the electrode materials.
To investigate different aging levels, eight cells were divided into two groups. Cells #1 to #4 were used for post-mortem analysis in the fully charged state, while cells #5 to #8 were used for post-mortem analysis in the fully discharged state. Cells #9 and #10 were fresh cells representing the pristine condition. The SOH of each cell was defined as the ratio of the measured C/3 discharge capacity to the rated capacity.
| Cell number | #1 | #2 | #3 | #4 | #5 | #6 | #7 | #8 | #9 | #10 |
| SOH (%) | 93.47 | 89.12 | 86.77 | 79.41 | 96.54 | 90.61 | 83.63 | 79.76 | 102.3 | 99.1 |
Full-Cell Performance Tests
Capacity calibration experiments were conducted in accordance with the GB/T31486 standard. Each cell was repeatedly charged and discharged at C/3. If the measured discharge capacity of three consecutive tests differed by less than 2% from the average, the average value was adopted as the calibrated capacity. The cycle aging test was designed to obtain cells at different degradation levels. The aging protocol consisted of constant-current charging at 15 A until the voltage reached 3.65 V, followed by constant-voltage charging until the current dropped to 1 A. After a rest period, the cell was discharged at 15 A until the voltage reached 2.5 V. This cycle was repeated and the capacity was calibrated every 50 cycles until the target SOH was reached.
To evaluate the influence of charge/discharge rate on cell behavior, rate capability tests were performed at C/25, C/10, C/5, C/3, C/2, and 1C. The test results for cell #1 show that the discharge capacity decreases as the rate increases. This effect is primarily caused by polarization, which becomes more significant at high currents and leads to an earlier termination of discharge when the lower cut-off voltage is reached. The rate-dependent capacity difference implies that the state of charge at the end of discharge is also rate dependent. Therefore, if cells at different aging levels are charged or discharged at different rates before disassembly, the initial lithium distribution in the electrodes may differ.
To ensure a consistent baseline for electrode state before post-mortem analysis, a low-rate reference charge/discharge test was carried out. In this procedure, each cell was first fully discharged at C/3, then rested for 4 hours, charged at C/25 to the upper cut-off voltage, rested for another 4 hours, and finally discharged at C/25 to the lower cut-off voltage. This low-rate treatment minimized the influence of polarization and ensured that all cells reached nearly identical fully charged or fully discharged states before disassembly.
Half-Cell and Electrode Characterization Tests
After full-cell electrical characterization, the cells were disassembled inside an argon-filled glove box. For cells #1 to #4, the disassembly was conducted in the fully charged state; for cells #5 to #8, the disassembly was conducted in the fully discharged state. The harvested positive and negative electrode sheets were washed with dimethyl carbonate to remove residual electrolyte. Some electrode sheets were used directly for material characterization, while others were cut into small disks and assembled into CR2032 half-cells with lithium metal as the counter electrode.
For the positive half-cell, the upper and lower potential limits were set to 3.7 V and 2.4 V, respectively. For the negative half-cell, the limits were 1.5 V and 0.01 V. A small current was applied during the first formation cycle to build a stable solid electrolyte interphase layer and to fully wet the electrode. After the first cycle became stable, a low-rate charge/discharge test was performed to obtain equilibrium potential curves as a function of the state of lithiation.
The electrode potential curves obtained from the half-cells were fitted with nonlinear functions designed for different electrode chemistries. For the LFP positive electrode, two sets of hyperbolic tangent functions were used to represent the two phase-transition regions. The general form of the fitted positive electrode potential is
$$ U_p(x) = a_p + c_{p,1}\left[\exp\left(-\frac{x – x_{p,1}}{b_{p,1}}\right) – \exp\left(-\frac{x – x_{p,2}}{b_{p,2}}\right)\right] – f_{p,1}\tanh\left(\frac{x – d_{p,1}}{e_{p,1}}\right) – f_{p,2}\tanh\left(\frac{x – d_{p,2}}{e_{p,2}}\right), $$
where x is the lithiation fraction of the positive electrode and the parameters were obtained by nonlinear least-squares fitting. For the graphite negative electrode, three hyperbolic tangent functions were used because the negative electrode potential curve contains multiple staging phase transitions. The corresponding fitting equation is
$$ U_n(x) = a_n + c_{n,1}\left[\exp\left(-\frac{x – x_{n,1}}{b_{n,1}}\right) – \exp\left(-\frac{x – x_{n,1}}{b_{n,1}}\right)\right] – f_{n,1}\tanh\left(\frac{x – d_{n,1}}{e_{n,1}}\right) – f_{n,2}\tanh\left(\frac{x – d_{n,2}}{e_{n,2}}\right) – f_{n,3}\tanh\left(\frac{x – d_{n,3}}{e_{n,3}}\right). $$
The fitted curves agreed well with the measured half-cell potential curves. The maximum fitting error was less than 0.05 V over the entire lithiation range, confirming that the fitted functions could reliably describe the equilibrium potentials of both electrodes.
X-ray diffraction measurements were performed with a PANalytical X’Pert3 powder diffractometer. The scanning range was 10° to 80° with a step size of 0.013°. Using XRD, I analyzed the structural evolution of the electrode materials at different SOH values. For the positive electrode, the XRD patterns of FePO4 and LiFePO4 contained many diffraction peaks that were distributed over a wide angular range and showed considerable overlap. This overlap made it difficult to extract quantitative information through individual peak intensity changes. In contrast, the negative electrode XRD patterns were more concentrated and easier to interpret. In the fully charged state, the graphite electrode exists as a lithium-intercalated compound, and the characteristic peaks of LiC12 and LiC6 were clearly resolved. As the battery aged, the diffraction peak intensities of these lithium-carbon compounds visibly decreased, indicating that fewer lithium ions could be intercalated during each full charge. In the fully discharged state, only the graphite C diffraction peaks were observed. The intensity of the graphite peaks also decreased with aging, indicating loss of negative electrode active material.
For morphological observation, I used a Hitachi Regulus 8100 field-emission scanning electron microscope. SEM images of the negative electrodes from a fresh cell and from aged cells were compared. The graphite particles of the fresh electrode exhibited clear edges and relatively smooth surfaces. With increasing degradation, the particle edges became increasingly rough and an additional surface layer progressively covered the graphite. This observation was consistent with the growth of the SEI film, which consumes cyclable lithium and is linked to LLI.
Qualitative Correlation between Micro and Macro Features
Based on the above characterization results, I performed a qualitative correlation analysis between the micro-scale features and the macro-scale degradation states of vehicle traction battery cells. The SEM images revealed that the SEI film on the negative electrode gradually thickened as the SOH decreased. Since SEI formation consumes lithium ions from the active lithium reservoir, the thickening of the SEI layer is directly related to LLI. The XRD results of the fully charged anodes further showed that the intensities of the lithium-carbon compound peaks reduced with aging. This reduction is consistent with a decreasing amount of lithium that can be hosted by the negative electrode during each charge, again pointing to LLI as the dominant degradation mode in the LFP cell.
It is also worth noting that the graphite peak intensity in the fully discharged electrodes decreased with aging. This result indicates that the negative electrode active material itself was being lost during cycling. By contrast, the positive electrode did not exhibit significant capacity loss over the investigated aging range. The relatively stable structure of LiFePO4 during lithium extraction and insertion explains the absence of substantial positive electrode active material loss in this cell. Therefore, for the studied vehicle traction battery, the main degradation mechanisms appear to be LLI and LAM in the negative electrode.
OCV-Based Degradation Model
The open-circuit voltage of a lithium-ion cell is determined by the difference between the positive electrode potential and the negative electrode potential. During cell aging, LLI, positive active-material loss, and negative active-material loss each modify the matching relationship between the two electrodes, and consequently change the shape of the OCV curve in distinct ways.
To describe these changes quantitatively, I treated the positive and negative electrode states as two independent coordinate systems. The horizontal axis of the positive electrode system is the lithiation fraction of the positive electrode, and the vertical axis is the positive electrode potential. The negative electrode system is defined similarly. Degradation modes transform these coordinates in different manners. Loss of active material causes a shrinkage of the accessible lithiation range, while LLI changes the relative offset between the positive and negative electrode operating windows.
In the positive electrode coordinate transformation, the available active material capacity is denoted as C_APE, while the original capacity is C_PE. The transformation factor along the x-axis is
$$ T_{x,APE-PE} = \frac{C_{APE}}{C_{PE}} = \frac{C_{PE}-LAM_{liPE}-LAM_{dePE}}{C_{PE}}, $$
and the translation parameter introduced by lithium-containing positive electrode loss is
$$ R_{x,APE-PE} = \frac{LAM_{liPE}}{C_{PE}}. $$
In the negative electrode coordinate system, the analogous quantities are
$$ T_{x,ANE-NE} = \frac{C_{NE}-LAM_{liNE}-LAM_{deNE}}{C_{NE}}, $$
$$ R_{x,ANE-NE} = \frac{LAM_{liNE}}{C_{NE}}. $$
The relationship between the positive and negative electrode coordinate systems additionally depends on LLI because the amount of cyclable lithium determines how the positive and negative lithiation ranges overlap.
Since it is impossible to uniquely identify all the individual parameters solely from the OCV curve, the diagnostic variables were recombined into identifiable parameters. The first diagnostic variable was chosen as the available positive electrode capacity:
$$ X_1 = C_{PE} – LAM_{liPE} – LAM_{dePE}. $$
The second diagnostic variable was the available negative electrode capacity:
$$ X_2 = C_{NE} – LAM_{liNE} – LAM_{deNE}. $$
The third diagnostic variable described the relative offset between the positive and negative electrode lithiation scales and can be expressed as:
$$ X_3 = LAM_{liNE} – LAM_{dePE} + LLI. $$
The final variable X4 represented the negative electrode lithiation fraction at the fully discharged state. Using these variables, the reconstructed OCV curve of an aged cell can be written as
$$ U_{cell}(x_{ANE}) = U_p\left(1 – \frac{x_{ANE} X_2 + X_3}{X_1}\right) – U_n(x_{ANE}). $$
To link the model with experimental data, the relationship between the actual charge throughput and the normalized coordinate was established through
$$ x_{ANE}(i) = \frac{Q(i)}{X_2} + X_4, $$
where Q(i) is the charge throughput corresponding to the measured point, and X4 denotes the dimensionless negative electrode lithiation fraction at full discharge. I used the particle swarm optimization algorithm to minimize the root mean square error between the reconstructed OCV and the experimentally measured low-rate OCV curve:
$$ \min RMSE = \sqrt{\frac{1}{n}\sum_{i=1}^{n}\left[U_{cell}^{model}(x_{ANE,i}) – U_{cell}^{exp}(x_{ANE,i})\right]^2}. $$
The diagnostic results from the PSO procedure indicated that the available negative electrode capacity decreased progressively with aging, while the positive electrode available capacity remained almost constant over the investigated SOH range. The offset parameter X3 increased approximately linearly as the battery aged, suggesting that LLI played a dominant role in the early stage of degradation.
Quantitative Degradation Diagnosis Based on OCV
Once the OCV parameter identification was completed, further derivation was needed to separate the lithium-containing and lithium-free portions of LAM. It was assumed that when active particles fracture and lose electrical contact, the lithiation fraction of the isolated portion is equal to the average lithiation fraction of the corresponding electrode over the operating range. Based on this assumption, the lithium-containing active material loss in the positive and negative electrodes can be expressed as
$$ LAM_{liPE} = \frac{x_{crk,P}^{min} + x_{crk,P}^{max}}{2} \left(C_{PE}^{fresh} – C_{PE}^{aging}\right), $$
$$ LAM_{liNE} = \frac{x_{crk,N}^{min} + x_{crk,N}^{max}}{2} \left(C_{NE}^{fresh} – C_{NE}^{aging}\right), $$
where the superscripts “fresh” and “aging” indicate the pristine and aged states, respectively, and the subscripts min and max denote the lithiation boundaries of the electrode. The lithium-free active material loss is then obtained as the difference between the total LAM and the lithium-containing fraction:
$$ LAM_{dePE} = LAM_{PE} – LAM_{liPE}, $$
$$ LAM_{deNE} = LAM_{NE} – LAM_{liNE}. $$
The capacity loss results diagnosed by the OCV method are summarized in the following table at selected SOH levels.
| SOH (%) | LLI loss (Ah) | LAMliNE loss (Ah) | LAMdeNE loss (Ah) | LAMliPE loss (Ah) | LAMdePE loss (Ah) |
| 96.54 | 0.22 | 0.06 | 0.06 | 0.01 | 0.01 |
| 93.47 | 0.85 | 0.36 | 0.31 | 0.02 | 0.01 |
| 89.12 | 1.63 | 0.78 | 0.66 | 0.03 | 0.02 |
| 86.77 | 2.28 | 1.12 | 0.98 | 0.04 | 0.02 |
| 83.63 | 2.79 | 1.47 | 1.31 | 0.05 | 0.03 |
| 79.76 | 3.38 | 1.74 | 1.52 | 0.05 | 0.03 |
| 79.41 | 3.40 | 1.82 | 1.55 | 0.05 | 0.03 |
The results clearly show that positive electrode capacity loss was negligible. The maximum LLI reached 3.40 Ah and the maximum negative electrode LAM reached approximately 2.83 Ah at the lowest SOH level. These results confirm that the primary degradation mechanisms of this vehicle traction battery consist of LLI and LAM in the negative electrode.
Micro-Scale Quantification of Degradation Modes
To verify the OCV-based diagnosis from an independent perspective, I characterized the harvested electrodes through half-cell tests. In a half-cell configuration, the lithium metal counter electrode provides an unlimited supply of lithium, so the measured capacity of the half-cell is mainly limited by the amount of active material in the working electrode. Therefore, half-cell capacity is suitable to characterize LAM. The test results showed that the positive electrode half-cell capacity was almost unchanged across the investigated SOH range. By contrast, the negative electrode half-cell capacity decreased approximately linearly with aging. This means that the loss of negative electrode active material was confirmed by the half-cell experiment.
Because the half-cell capacities were measured in units of mAh while the OCV-diagnosed LAM values were expressed in Ah, I used battery #5 as a reference to normalize both datasets. The half-cell capacity loss rate was calculated as
$$ r_{cap} = \frac{Cap_0 – Cap_i}{Cap_0} \times 100\%, $$
and the corresponding OCV-diagnosed LAM loss rate was
$$ r_{LAM} = \frac{LAM_0 – LAM_i}{LAM_0} \times 100\%, $$
where the subscript 0 refers to cell #5. The comparison between the two normalized loss rates is given below.
| Cell | #5 | #6 | #7 | #8 |
| Half-cell capacity loss rate (%) | 0 | 4.55 | 8.91 | 11.54 |
| OCV-diagnosed LAM (%) | 0 | 3.20 | 7.93 | 10.40 |
The maximum difference between the two methods was only 1.35 percentage points. This close agreement verified that the LAM diagnostic results obtained from the OCV model were accurate.
For the quantitative characterization of LLI from the micro-scale perspective, I used the XRD results of the fully charged negative electrodes. Since the graphite active material can host a certain amount of lithium, the lithium content intercalated during a full charge may be treated as the maximum extractable lithium content in one cycle. The XRD peak intensities of LiC6 and LiC12 were used to quantify the amount of lithium-carbon compound. Based on the normalization assumption that the total graphite content is unity, the lithium content in the negative electrode was calculated as
$$ LiC_{x,aging} = \frac{1}{x_1+x_2}\left(\frac{x_1}{6} + \frac{x_2}{12}\right), $$
where x1 and x2 are the diffraction peak intensities of LiC6 and LiC12, respectively. The maximum lithium content decrease measured from the XRD data is shown in the following table for the investigated SOH values.
| SOH (%) | 96.54 | 93.47 | 89.12 | 86.77 | 83.63 | 79.76 | 79.41 |
| XRD lithium content decrease (%) | 2.11 | 6.44 | 10.85 | 14.03 | 17.46 | 23.20 | 24.96 |
Although the decreasing trend of the XRD-derived lithium content agreed qualitatively with the LLI evolution, the absolute value was considerably higher than the OCV-diagnosed LLI. The maximum difference reached about 7.78 percentage points. The reason is that in the above normalization the total graphite content was implicitly fixed to unity. However, the half-cell test results showed that the negative electrode active material was indeed lost during aging. The denominator in the normalization therefore should consider the remaining fraction of the active material.
To correct this effect, I used the negative-electrode half-cell capacity measured at the discharged state to quantify the reduction of the active material. Because cells #1 to #4 were disassembled in the fully charged state and cells #5 to #8 were disassembled in the discharged state, an interpolation was applied to estimate the remaining active material for the charged-state cells. The corrected lithium content was then expressed as
$$ LiC’_{x,aging} = \frac{1}{1 – LAM_{NE}’}\left(\frac{x_1}{6} + \frac{x_2}{12}\right)\frac{1}{x_1+x_2}, $$
where LAM_NE’ is the normalized negative-electrode active material loss derived from half-cell capacity. The corrected maximum lithium content variation was then compared with OCV-diagnosed LLI.
After applying this correction, the maximum difference between the XRD-based lithium content decrease and the OCV-diagnosed LLI was only 1.68 percentage points. This result confirms that the OCV diagnosis model produced reliable LLI estimates. The good agreement between the micro-scale and macro-scale calculations further suggests that a direct quantitative link can be established between the XRD diffraction characteristics of electrode materials and the degradation modes of a vehicle traction battery.
Influence of Degradation Modes on OCV and IC Curves
Although the OCV-based diagnosis method is accurate, its computational procedure is rather involved for vehicular applications. In practical battery management systems, measured charging voltage and current data are often transformed into incremental capacity curves that amplify subtle changes in the voltage profile. To build a practical diagnosis tool, I first analyzed how individual degradation modes change the shape of the OCV and IC curves.
For this purpose, I constructed synthetic cells with prescribed degradation levels by varying LLI, LAMliNE, and LAMdeNE individually, as well as in combination. The specific values used for these constructed cells are provided in the following table.
| Constructed cell no. | LAMliNE (Ah) | LAMdeNE (Ah) | LLI (Ah) |
| 1 | 0 | 0 | 0 |
| 2 | 0.25 | 0.25 | 0.5 |
| 3 | 0.5 | 0.5 | 1 |
| 4 | 0.75 | 0.75 | 1.5 |
| 5 | 1 | 1 | 2 |
| 6 | 1.25 | 1.25 | 2.5 |
| 7 | 1.5 | 1.5 | 3 |
| 8 | 1.75 | 1.75 | 3.5 |
The OCV simulation results showed that LLI mainly narrowed the accessible lithiation range of the electrodes without substantially altering the shape of the individual electrode potential curves. In contrast, LAMdeNE produced a comparatively small effect on the full-cell OCV curve. LAMliNE reduced the usable negative electrode capacity and simultaneously reduced the amount of lithium that can participate in the charge/discharge reaction.
The corresponding IC curves were obtained by numerical differentiation of the OCV curves with appropriate smoothing. In the simulated IC curves, three characteristic peaks can be observed and are denoted here as Peak ①*II, Peak ②*II, and Peak ⑤*II. The influence of LLI and LAM on these IC features was analyzed separately. Increasing LLI caused the intensity of the high-potential peak, i.e. Peak ①*II, to decrease, while the middle peak, Peak ②*II, remained almost unchanged. Increasing LAMliNE caused both Peak ②*II and Peak ①*II to decrease. Increasing LAMdeNE, however, increased Peak ①*II while decreasing Peak ②*II. When LAMdeNE and LAMliNE acted simultaneously, their opposite influences on Peak ①*II largely canceled each other out. This cancellation explains why, under multi-mechanism aging, the observed change of Peak ①*II is mainly governed by LLI, while Peak ②*II is mainly governed by negative-electrode LAM.
I further compared the peak area and peak height of Peak ②*II and Peak ①*II under different degradation modes. The results showed that the peak height and peak area of Peak ②*II declined with increasing negative-electrode LAM, while LLI had little effect on this middle peak. For Peak ①*II, the peak height and peak area decreased with LLI but were almost insensitive to the combined effect of LAMliNE and LAMdeNE because of the aforementioned cancellation. These observations form the basis of IC-based degradation-mode quantification.
IC-Based Degradation Diagnosis Method
Because the IC curve can be computed from routine charging data in real applications, I derived quantitative formulas that relate the change in peak area to degradation modes. The negative-electrode active-material loss is calculated from the area of Peak ②*II as
$$ LAM_{NE}^{IC} = \frac{Area_{B,fresh} – Area_{B,aging}}{Area_{B,fresh}} \times 100\%, $$
where Area_B,fresh and Area_B,aging are the Peak ②*II areas for the pristine cell and the aged cell, respectively. The loss of lithium inventory is calculated from the change in Peak ①*II area relative to the fresh cell capacity:
$$ LLI^{IC} = \frac{Area_{C,fresh} – Area_{C,aging}}{Q_{fresh}} \times 100\%, $$
where Q_fresh is the measured capacity of the pristine cell. In this formula, the denominator is the full-cell capacity instead of the peak area of the fresh cell. This modification reflects the fact that LLI shifts the entire chemical window rather than compressing a single phase-transition region, and leads to a more accurate estimate of LLI.
I applied these formulas to the C/25 full-cell data of all aged LFP cells. The IC-based LLI and negative-electrode LAM results were compared with those obtained from the OCV method. For the peak-area-based approach, the LLI maximum difference between the two methods was only 1.79 percentage points, and the LAM maximum difference was 1.62 percentage points. These values demonstrate that the IC-based method can provide degradation diagnoses that are consistent with the validated OCV method.
Using the IC diagnostic procedure, I further analyzed the peak area and peak height as degradation indicators. It was found that both the peak area and the peak height were correlated with degradation. However, the peak area was more representative of the actual compression of the curve and yielded more stable results, especially for negative-electrode LAM diagnosis. Therefore, the peak area was selected as the main diagnostic feature in the proposed method.
Influence of Discharge/Charge Rate and Rate Correction
For on-board vehicle traction battery diagnosis, the practical operating current is generally much higher than C/25. Before applying the IC-based diagnostic method at higher rates, I experimentally measured IC curves of the same LFP cells at C/10, C/5, C/3, C/2, and 1C. The test results showed that as the current rate increased, the peak positions of the IC curves shifted toward higher voltages during charging and toward lower voltages during discharging due to increased polarization. The peak intensities decreased with increasing rate, and at higher rates some peaks even merged together.
To quantify the adaptability of the diagnostic method, I used C/10, C/5, and C/3 data to calculate LLI and LAM. The negative-electrode LAM results at these rates followed the same overall dependence on SOH as the C/25 results. However, the LLI results exhibited a systematic negative bias with respect to the C/25 baseline. The maximum absolute errors of LLI at C/10, C/5, and C/3 were 3.25%, 4.77%, and 6.07%, respectively.
The rate dependence of the LLI estimate was approximately linear, which enabled me to introduce rate-correction factors. Taking the C/25 result as the reference, the correction factors were determined from the slopes of the LLI-versus-capacity-loss relationships at different rates. The resulting correction factors are summarized in the table below.
| Rate | C/25 | C/10 | C/5 | C/3 |
| Correction factor | 1 | 1.19 | 1.26 | 1.47 |
After applying these rate-correction factors, the corrected LLI values at different rates were brought into close agreement with the C/25 reference. The maximum error of the corrected LLI was reduced to 1.68 percentage points. This result confirms the robustness of the IC-based degradation-mode diagnosis method for vehicle traction battery applications under moderate charge/discharge rates.
In practical vehicle applications, the IC diagnosis procedure can be structured as follows. The battery management system first samples the current, voltage, and time data during a charging event. The capacity throughput between two voltage points is calculated and the IC curve is constructed using the polynomial center-smoothing method. The peak areas of the relevant characteristic peaks are then extracted. Using the diagnostic formulas described above, the negative-electrode LAM and LLI are obtained. Because the charging rate during real operation is not necessarily equal to the reference rate, the rate-correction factor is applied to obtain the final LLI estimate. If accurate OCV data become available during long rest periods, the correction factors can be updated online, further improving the adaptability of the method.
Conclusions
In this thesis, I investigated the degradation modes of a commercial lithium iron phosphate vehicle traction battery through a combination of macroscopic electrical diagnostics and microscopic material characterization. The main contributions and findings can be summarized as follows.
First, I established a complete test matrix for cells at different SOH values and combined full-cell electrical testing with half-cell electrochemical testing, scanning electron microscopy, and X-ray diffraction analysis. The SEM results showed that the SEI film thickness increased with aging, while the XRD results showed that the intensities of lithium-carbon diffraction peaks decreased with aging. These observations provide qualitative micro-scale evidence for LLI and LAM in the negative electrode.
Second, I developed an OCV-based degradation diagnosis model by deriving the coordinate transformation relationships between the positive and negative electrode equilibrium potential curves. The model quantitatively separated LLI from positive and negative active-material losses. The results indicated that positive electrode active-material loss was negligible, while LLI and negative-electrode LAM increased with aging. The maximum capacity losses from LLI and negative-electrode LAM reached 3.40 Ah and 2.83 Ah, respectively.
Third, I proposed a method to calculate the maximum lithium content of the negative electrode from XRD diffraction peak intensities of LiC6 and LiC12. Because the initial normalization assumed a fixed graphite content, the calculated lithium decrease overestimated LLI. I then corrected the lithium content calculation by accounting for negative-electrode active-material loss measured from half-cell capacities. The corrected XRD-based lithium-content decrease agreed with the OCV-diagnosed LLI within 1.68 percentage points. The negative-electrode LAM diagnosed by the OCV method agreed with the half-cell capacity loss rate within 1.35 percentage points.
Fourth, I analyzed the influence of individual degradation modes on OCV and IC curves. It was found that Peak ①*II of the IC curve was mainly governed by LLI because the opposite effects of LAMliNE and LAMdeNE canceled each other out. Peak ②*II was found to be mainly controlled by negative-electrode LAM. Based on these observations, I proposed quantitative IC-based diagnostic formulas that use the peak areas of the two characteristic peaks. The LLI and negative-electrode LAM calculated from the IC curves matched the OCV-based results with maximum differences of 1.79% and 1.62%, respectively.
Finally, I evaluated the applicability of the IC diagnostic method at higher charge/discharge rates. The LLI estimates exhibited a systematic negative dependence on the current rate. A rate-correction factor was introduced to compensate for this effect. After rate correction, the maximum LLI error decreased to 1.68%, demonstrating that the proposed IC-based method can accurately and robustly diagnose the degradation modes of a vehicle traction battery under practical operating conditions.
Future research may be directed toward improving the accuracy of IC curve extraction for asymmetric peaks, validating the proposed diagnostic method on dynamically varying current profiles from real electric vehicles, and developing online correction algorithms that can automatically update the rate-correction factors based on field data.