Degradation mechanism detection for NMC batteries based on Incremental Capacity curves

The requirements and demands that batteries have to meet are constantly increasing. Therefore, effective control and management is crucial for ensuring a safe use of the battery while maintaining the best possible performance. Despite the fact that the diagnosis and prognosis of the State of Health (SoH) is absolutely essential in practical applications, they are not yet effectively implemented in Battery Management Systems (BMS). The main issue hampering the ability of BMSs to accurately track the SoH is the evolution of the voltage response of the cells with aging coupled with the path dependence of the degradation. The latter is a premier issue because it prevents the BMS to use preset look-up tables for SoC recalibration upon aging.


Introduction
The requirements and demands that batteries have to meet are constantly increasing.Therefore, effective control and management is crucial for ensuring a safe use of the battery while maintaining the best possible performance.Despite the fact that the diagnosis and prognosis of the State of Health (SoH) is absolutely essential in practical applications, they are not yet effectively implemented in Battery Management Systems (BMS).The main issue hampering the ability of BMSs to accurately track the SoH is the evolution of the voltage response of the cells with aging coupled with the path dependence of the degradation.The latter is a premier issue because it prevents the BMS to use preset look-up tables for SoC recalibration upon aging.
Path dependence is challenging to take in consideration because of the numerous degradation mechanisms that can occur throughout the life of a cell.It was shown that a multitude of parameters play a key role in degrading a Li-ion cell: depths of discharge (DoD), states of charge, applied currents, temperatures, calendar aging conditions, pulses… [1][2][3][4] In real life those parameters cannot be controlled and it is impractical to replicate all of them in a laboratory setting.
Operando degradation mechanisms detection is of huge importance in order to try to detect and quantify ageing in a cell.It will be a key feature as to predict degradation and SoH of a battery cell and thus on the ability of a BMS to handle path dependence.One attractive way to address this issue is to analyze battery degradation using differential voltage techniques.These techniques have been used by many researchers [5][6][7][8] in order to reveal battery degradation mechanisms and they were shown successful in quantifying the individual components of every degradation mechanism: the extent of Loss of lithium inventory (LLI), loss of active material (LAM) and kinetic changes.
In this work, we will use incremental capacity (IC) as the differential voltage technique.Additionally, we will be using the 'alawa toolbox developed by M. Dubarry et al. [5,[9][10][11] in order to study the impact of each individual component of the degradation.The main purpose of the toolbox is the intricate study of the degradation mechanisms by emulation of different scenarios.
In this regard, this study presents an extensive research of the degradation mechanism of NMC/Graphite cells cycled at different DoD based on the study of IC curves.From the simulations and the tested cells, different features of interest have been observed and selected [12], highlighting path dependence.

Cycling
Three high power 40Ah Li-ion cells containing Graphite (G) based anode and LiNi1-x-yMnyCoxO2 (NMC) based cathode have been used for this work.Cells were cycled at different DoDs, Cell A was cycled at 80% DoD, Cell B at 60% and Cell C at 30% with the charging protocol being the same.The ageing tests were performed in a temperature controlled environment at 25°C (ClimaTemperatur-System, CTS/T-40/200/Li and CTS/T-40/600/Li models, and Prebatem Selecta 80 and 150 L chambers with Peltier effect).Experiments were carried out using DIGATRON MCT (Multiple Cell Tester, MCT 100-06-10 ME and MCT 50-06-24 ME models) with BTS-600 battery data acquisition software.The cells were tested from the beginning of life (BoL) until the EoL (20% capacity loss, if reached).
Cycling was interrupted every 300 full equivalent cycles to perform reference performance tests (RPT).In this RPT, two tests were carried on the cells:  Capacity test: charge and discharge cycles (at current-constant voltage, CC-CV, and CC modes, respectively) at nominal conditions specified by cell manufacturer (1C-rate current and 25 ºC temperature).Three full charge-discharge cycles were performed in order to assess both reversible and irreversible capacity losses and check the repeatability of the results.
 Cell full charge/discharge test: Prior to the discharge of the cells are fully charged in CC-CV mode at 1C. Discharge were performed at a C/5 rate.This Crate was chosen as it was the fastest one that enabled observing the voltage plateaus in view of implementation in real application where low rate cycling is impracticle.Galvanostatic voltage profiles were used for examining electrode phase changes and understanding degradation phenomena.
The SoH was calculated from the capacity test according to equation (1) were Q is the capacity in Ah (1).

Incremental capacity curves
The operation mode of lithium ion batteries is based on the intercalation and deintercalation of lithium-ions between the anode and the cathode.The difference in the lithium electrochemical potential of the electrodes defines the potential of the cell.Thus, in order to estimate the state of the battery, a key parameter is the cell voltage.This parameter depicts electrochemical reactions and their evolution [5].In other words, the ability to analyse the cell voltage is the key to understand cell behaviour and degradation.
The variations on V vs. Q curves are usually pretty minute and hard to quantify.In order to ease the analysis, derivative methods, differential voltage and incremental capacity, could be used.For this work, as well as for the previous work regarding SoH estimation, incremental capacity technique is used.For this technique, the relationship between the two electrodes can be written as [7]: IC curves have already been used for ageing and degradation mechanisms detection in different battery technologies: LFP cells [5,13,14], LTO batteries [15], LCO [16] and NMC [8,17].They have been studied also for SoH estimation, like in NMC cells [18][19][20], LFP [21][22][23] and also composites NMC+LMO [23] cells.

Alawa Toolbox
Emulation of IC curves were performed using the 'alawa toolbox developed by M. Dubarry et al. [9,11,24].The toolbox enables simulation of a wide array of degradation scenarios with different anode and cathode materials, composite or not, and operating conditions, etc.Its main advantage is to give the possibility to mix and match all the possible degradations mechanisms in order to obtain the best fitting solution.
In this work, the exact replicate of the anode and cathode were not available.The toolbox was then calibrated on a standard G//NMC cell and used to visualize the trends of voltage changes induced by individual degradation modes.
In order to emulate the cell behaviour in the 'alawa toolbox, two parameters are necessary: the loading ratio between the negative and the positive electrode (LR) and the initial irreversible capacity loss of the negative electrode that compensates the SEI formation (OFS).These two parameters can typically be obtained from a discharge curve at C/25 at the beginning of life (BOL) [5].In this study, the C/25 discharge at the beginning of life was not performed and therefore the LR and OFS were estimated from the initial C/2 discharge.The best fit that could be obtained were LR = 1.15 and OFS= -5.

Cycling
The cells have been tested at 25°C at different depth of discharge levels (DoD).Cell A was cycled at 80% DoD, Cell B at 60% and Cell C at 30%.Cell A reached nearly an EoL state coming up to 82% capacity, cell B reached a 95% and cell C a 90% capacity.Cell B and cell C were not cycled to EoL because of time constraints.The cells degradation can be compared at three strategic points, BoL,95% SoH and 90% SoH (cells A and C only).The normalized capacity evolution of each cell can be seen at the left side of figure 1, the evolution of the IC curves obtained from the tested cells in terms of ageing are presented in the right side.In order to analyse ageing of the cells, the evolution of the IC curves with 5% increment in capacity loss are going to be studied.A clear evolution of the curves can be detected as the cells degrade and this evolution can be described from the changes of 4 main characteristics of the IC curves : two peaks (a) and (b) and two electrochemical features, an arch shape (c) and the intensity at 4V (d).The 2 peaks have their initial maximum of intensity at 3.6V (a) and 3.5V (b), respectively and this maximum of intensity seems to decrease and shift towards lower potentials with aging.The arch (c) appears to shift from 3.75 to 3.85V.The IC intensity at 4V, feature (d), seems to decrease with aging.

Simulated curves
The main objective of this work is to determine and detect the degradation mechanisms that are occurring in the cells.In order to identity their individual impact on cell performance, simulations from the BoL state until a 90% SoH have been performed using the 'alawa toolbox.In this work we will focus only on thermodynamic changes and thus the kinetic factors will not be considered.As a result, 3 degradation mechanisms were studied in this work: Loss of Lithium Inventory (LLI), and the Loss of Active Material (LAM) on both electrodes at a delithiated state (the lithiated state is equivalent to the delithiated state plus some LLI).For sake of simpification, LAMdePE and LAMdeNE will be referred as LAMPE and LAMNE in the rest of this work.Figure 3   Starting with feature (a), which can be delimited as the most defined peak at approximately 3.6V.It can be detected that for both LAM scenarios the peak doesn't change significantly and only a slight intensity increase can be observed for LAMPE.In case of LLI (figure 2, case 1), the peak undoubtedly changes in both intensity, shape and position as it becomes less intense, broader and starts at higher voltages.
Something similar happens with feature (b) and the peak detected at 3.5 V.It can be easily recognized that in LAMNE (figure 2, case 2) and LAMPE (figure 2, case 3) this peak is not visible whereas, on the contrary, a new peak is appearing around 3.4 V when LLI is occurring (figure 2, case 1).This peak appears around 3.37V initially then moves towards higher voltages (3.42V) with further degradation.The intensity of the peak also changes, reaching higher intensity as the cell degrades.
The third detected feature refers to the arch, feature (c).This arch shape is not easily detected in any of the case of figure 2. In order to check the arch in the simulation curves, the data obtained for the discussion part is obtained from simulations at lower rates (not shown), in which the arch can be detected.The arch positon will be defined as the lowest intensity between 3.7 and 3.9 V.
The last characteristic to take into consideration is the incremental capacity intensity at 4V.In case of LLI (figure 2, case 1), the incremental capacity intensities do not change much with aging at 4V.The same can be observed with LAMNE (figure 2, case 2).In case of LAMPE (figure 2, case 3) the curves clearly reach progressively lower incremental capacity values upon degradation.

Features of Interests
The evolution of the features of interests was studied in order to detect the degradation mechanisms in the cells.The detection was based on the study of the path that each feature follows upon aging.The comparison of the cells at similar SoHs (BoL, 95% SoH and 90% SoH) is presented in figure 3.
The first figure on the left shows the BoL IC curves of the 3 tested cells.Overall the cells presented similar features which suggests that the cells were at a similar SoH at the beginning of the experiment.It has to be noted that Cell B presented a slightly lower intensity for feature (a), the other 3 feature stayed nearly identical.Different simulations were performed by the use of the 'alawa tool.The degradation mechanisms LLI, LAMNE and LAMPE were studied from a BoL state until a 90% SoH, checking the simulations every 1% SoH.Also all the different possible scenarios between the studied degradation mechanisms were considered: LLI+LAMPE, LLI+LAMNE, LAMPE+LAMNE and LLI+LAMPE+LAMNE.As a result, 7 degradation scenarios were investigated.
In the next section, the evolution of the features from the simulations are compared to the experimental ones.

Feature (a)
The peak at 3.6 V is the main feature towards degradation mechanism detection.As already detected and in order to have a careful study of it, two main properties will be analysed as the cell degrades: intensity and position.

 Intensity
The intensity of the peak at 3.6V is decreasing with aging.This trend can be compared with the simulations in order to gain insight on possible degradation scenarios in figure 4-a.The tested three cells have a very similar performance.Since the simulations were performed on a reference cell and not the exact replicate of the tested cells, we'll focus more on the exclusion of scenarios rather than an exact fit of the variations.The 3 tested cell stay at the middle of the graph between 100 and 120% of the initial peak intensity.As a result, three scenarios seems incompatible: LLI, LLI+LAMNE and LAMNE.The first two are excluded because they predict a larger peak intensity increase, and for third one, because it predicts a decrease of the intensity.

 Voltage
Figure 4-b presents the evolution of the position of the peak upon aging for the experimental data as well as different degradation scenarios.According to this figure, cells A and C behave in a different way than of the simulations with a more marked shift towards higher voltage.This is likely induced by the differences between the reference cell and the actual cell.In any case, scenarios predicting an increase of the voltage, LLI and LLI+LAMNE, appear incompatible.

Feature (b) peak
The second feature to take into consideration is the appearance of a small peak at 3.5V.Both characteristics, voltage and incremental intensity values will be taken into consideration.

 Intensity
The peak at 3.5V of the tested cells reaches higher incremental capacity values with aging.The tested three cells have a very similar performance.According to the different simulations, LLI and LLI+LAMNE scenarios seems incompatible.

 Voltage
Figure 5-b presents the evolution of the position of the peak upon aging for the experimental data as well as different degradation scenarios.It has to be noted that Cell B and Cell C overlaps.In the same way as in the intensity, the incompatible scenarios are pure LLI and LLI+LAMNE.

Feature (c)
From the analysis of the changes of arch positon on the tested cells, the arch shifts towards higher voltages and that its intensity is decreasing.

Feature (d)
The last feature to consider is the incremental capacity intensity at 4V. Figure 7 shows that the three tested cells share the same evolution with an intensity decrease.According to our comparison with simulated scenarios, the less likely to occur are LAMNE, LLI and LLI+LAMNE.

Summary of the observations
In table 1, a summary of the excluded paths is presented.From here, the most likely scenarios list can be compiled.
For this study, 7 scenarios were considered: LLI, LAMPE, LAMNE LLI+LAMPE, LLI+LAMNE, LAMPE+LAMNE and LLI+ LAMPE+LAMNE.Compiling this information, it seems that most likely scenario for the degradation are LLI+LAMPE and LLI+ LAMPE+LAMNE.However, LAMNE was excluded as a possible degradation mode for all the features, we are therefore considering LLI+LAMPE as the best candidate.

Path dependence: Cell A vs. Cell C
As established in the previous section, the most likely degradation path involves LLI+LAMPE.A quantification analysis showed that the trends observed on the experimental data at 90% SoH are best represented with around 5% LLI and 5% LAMPE, figure 8 as it matches the evolution of features (a), (b), (c) and (d) Overall, this simulation is matching the overall behavior of the 3 cells.Nonetheless, some differences in the aging process, and thus some path dependence of the degradation was observed even though the cells reached the same SoH, 90%.This is especially visible between Cell A that experienced deed discharges (80% DoD) and Cell C that experienced shallow discharges (30%DoD).According to their respective evolution of the features of interest (Section 4.1) and the simulated scenarios (Section 4.2), we can assess that the observed differences between the cells were likely introduced by some additional LAMPE on the cells that performed deep discharges.

5-Conclusions
The developed work aims to detect and quantify the degradation which is happening in NMC/Graphite cells.Three cells have been tested from a BoL state at different DoDs: 80, 60% and 30%.IC curves have been used for the identification of the possible degradation modes in the 3 cells with the help and use of the 'alawa toolbox.
From the simulated scenarios in the 'alawa toolbox and the tested cells comparison, four different characteristics have been detected as the key parameters for degradation mechanisms detection.From these 4 features, a deep comparison between the tested cells and the paths conformed by the different simulated scenarios has been done.All curves have been studied separately in order to compare the degradation with simulated scenarios.A final best fitting curve has been obtained giving as an estimate of the cell degradation mechanisms which involve around 5 % of both LLI and LAMPE after a 10% capacity loss.
Additionally, some path dependence was observed and the cell that performed the shallowest cycling (30%) DoD seems to have experienced less LAMPE.
Management research area at IK4-IKERLAN.He has directed several PhD students and has many publications on electrochemistry.

Figure 1 :
Figure 1: Capacity evolution of the tested cells on the left side, and evolution of IC curves obtained from the tested cells in terms of ageing on the right side.
shows the resulting IC curves obtained from the ʻalawa toolbox for a G//NMC cell with LR = 1.15 and OFS = -5 and a discharge rate of C/5.
The second figure presents the IC signature of the 3 cells at 95% SoH.The intensity of feature (a) decreased for all the cells, by 14% for cells B & C and by 9% for cell A. The intensity of feature (b) increased with aging.Again, cells B and C presented a similar evolution of 3%.Cell A presented an increase of 0.5%.The voltage of feature (b) presented a shift to lower voltages.Cell A turned 10mV to a lower voltage position, on the contrary cells B and C shifted slightly less, moving 1mV and 6mV respectively.Feature (c) focuses on the arch shape between 3.7 and 3.9V.The arch intensity lowered by 6% with aging and shifted to higher voltages.The voltage shift was different for all cells, 21mV for cell A, 31 mV for cell B, and 54 mV for cell C. Finally, the intensity of feature (d) decreased by 2.5% for all the cells.Lastly, in the figure showing a 90% SoH only cell A and C were available.The intensity of feature (a) decreased further for all the cells, by 4% for cell A and by 1.2% for cell C. The intensity of feature (b) kept increasing with aging, 1% for Cell C and 1.4% for Cell A. The voltage of feature (b) also shifted to higher voltages.Cells A shifted by 7mV and cell C by 4mV.Feature (c) intensity decreased by 1-2% for both cells.In terms of voltage, cell A shifted again towards higher voltages by 21 mV and cell C by 47 mV.Finally, the intensity of feature (d) decreased for all the cells, by 3% for cell A and 2% for cell C.

Figure 3 :
Figure 3: IC curves at different SoH for the three tested cells a) BOL, b) 95% SoH and c) 90% SoH.

Figure 4 :
Figure 4: Normalize intensity and peak voltage evolution for feature (a).

Figure 5 :
Figure 5: Normalize intensity and peak voltage evolution for feature (b).

Figure 6 -
Figure 6-a compares the variations of arch position for both the experimental data and the simulations.According to this comparison, the cells behave very similarly and the less likely scenarios to explain the variations are: LAMNE, LLI and LAMPE as they predict a decrease of the intensity. Voltage According to the measured evolution of the arch voltage, it can be said that cells B and C share the same trend and cell A behaves differently.Cell A arch voltage reached lower values compared to the other two cells.Looking at the different scenarios, pure LLI scenario is unlikely due to the higher predicted values.LAMNE and LAMPE+LAMNE scenarios also seems incompatible due to the low predicted voltage values.

Figure 6 :
Figure 6: Normalize intensity and peak voltage evolution for feature (c).

Feature
(a) and (b) evolution seems to indicate that LLI and LAMNE are unlikely.Feature (c) evolution allows to exclude LLI, LAMNE, LAMPE and LAMPE +LAMNE.Finally, feature (d) points towards LLI, LAMNE and LLI+LAMNE being incompatible.

Figure 8 :
Figure 8: Best fit obtained by mixing LLI and LAMPE by using the 'alawa toolbox on the left side, and evolution of IC curves obtained from the tested cells in terms of ageing on the right side.
Prof. Dr. Eng.Joeri Van Mierlo leads the MOBI -Mobility, Logistics and automotive technology research centre (http://mobi.vub.ac.be).A multidisciplinary and growing team of 60 staff members.Prof. Van Mierlo was visiting professor at Chalmers University of Technology, Sweden (2012).He is expert in the field of Electric and Hybrid vehicles (batteries, power converters, energy management simulations) as well as to the environmental and economical comparison of vehicles with different drive trains and fuels (LCA, TCO).

Table 1 :
Summary of the excluded paths for each feature