Detection of Anode Coating Defects in Batteries Electrode Production and their Effect on Cell Performance

In the different steps of the production process (see Fig. 5), various defects can occur (see Section 2.2). At first, active materials [6], binder and solvents are mixed to a slurry. As active materials, hard carbon (or graphite) and silicon are utilized for the anode side [14], the solvent is demineralized water. Nickel-Manganese-Cobalt (NMC) 622 or NMC 811 is currently the new standard for the cathode side in high energy density battery systems and as solvent N-Methyl-2-pyrrolidon is used [16]. The active materials are coated with a slot die in separate lines on a 6 microns copper foil for the anode and on a 12 microns aluminum foil for the cathode side. Multiple stripes of active material on the so called mother coil help to optimize production costs. After drying in a convective furnace, the solvents evaporate and porosity arises. To have an exact porosity level on the electrodes, a calender compacts the active battery material to the defined lamination strength [28, 29, 32]. The mother coil is slitted in single stripe daughter coils with the final width of the battery design. In the assembly line, a winder produces jelly rolls, combining anode and cathode, separated by a ceramic-coated polymer foil, the separator. The following process steps are kneading, beading, and laser-welding of copper and aluminum discs to have a tabless design with optimal electrical behaviour. Finally the jelly rolls are fitted in steel cans and the assembled battery is checked in a leakage test to impermeability. The last process steps are the filling of electrolyte, the formation of the battery cell and the sealing of the can. The Open-Circuit Voltage (OCV) is measured the first time and after aging of about two weeks, the OCV drop is measured again. The difference between these two results is characterising the quality of the cell.

There are two types of coating defects: bulk defects (e.g, layer thickness, layer width or shape of the boundary) or defects in the coating surface. This study focuses on the second. Common coating surface defects are given in Table 1. The defects investigated in this study are marked in bold and will explained in more detail in the following section. These defects were chosen because of their occurrence frequency and relevance to the battery production at BMW.

Pinholes are mostly occurring due to bursting bubbles in the slurry, which burst during the coating process or during drying [27]. Here, resulting holes can range through the whole coating thickness [5]. In a study by Bhamidipati et al. [5], various mechanisms that lead to air entrapment in slot-die coatings were investigated. It was found that a narrower coating gap results in smaller bubbles, while a higher band speed leads to larger bubbles [5].

Line defects describe elongated areas without coating and can occur in different forms. Horizontal line defects often appear in the direction of band movement, caused by agglomerates or other impurities that block the die gap [23]. Line defects in the direction of electrode movement can have varying widths, depending on the size of the blockage in the die gap.

Bild vergrößern

Particle contamination can have various sources in the different production steps. The contamination can already be present in the raw material, it can stem from wear of the machines, dust in the production environment, or the welding process [19, 23]. Metallic particle contamination is particularly critical because they are conductive and can puncture the separator if they are of sufficient size, potentially leading to a short circuit and even a thermal event. A thermal event refers to the overheating of the battery, during which energy is released uncontrollably within a few milliseconds, potentially triggering a battery fire. Therefore, particle contamination must be reliably detected through integrated NDT methods in the electrode manufacturing process [19].

In general NDT methods like ultrasound, X-rays, beta-rays, laser triangulation, optical systems, and thermography [2, 24, 25, 31] can monitor the quality after the coating and drying. For a review of the different sensing methods, please refer to Haghi et al. [17].

Optical camera systems for quality control are state-of-the-art in many production facilities today. Optical systems enable fast, consistent, and automated defect detection [25]. In contrast to infrared (IR) cameras they are sensitive in the visual range of the electromagnetic spectrum. Figure 6 shows the measuring principle of an optical line-scan camera system used in this study, which is particularly suitable for inspecting continuously moving materials. The images composed of individual lines are commonly analyzed using digital image processing or machine learning methods [25]. Different coating defects can be classified, and the coating parameters can be adjusted via a feedback loop if necessary [25].

In order to detect defects in the required micrometer range, a particularly high resolution of the camera is necessary. Furthermore, appropriate lighting is crucial for the quality of optical inspection using camera systems [27]. A brightness fluctuation detected by the camera indicates irregularity in the coating [27]. Various lighting scenarios increase the likelihood of detecting defects. Light-emitting diodes are mainly used as light sources, as they are highly energy-efficient and their illumination intensity is easily controllable. For particles in the range of 0.01-0.05 mm, line-scan cameras with 4096 pixels reach their limits due to high image noise [22].

The optical camera system used in this study is from the company Keyence [20]. To increase the POD, three different lighting directions are used: frontal lighting, side lighting from the left, and side lighting from the right. Figure 6 shows the setup for the line-scan camera with frontal lighting. For each lighting direction, a high-speed line-scan camera (CA-HL04MX) with a lens (CA-NPN20E) was mounted. The line-scan cameras, each with 4096 pixels per line, inspect the entire width of the electrode.

Besides the measurement system itself, the capabilities of the properties evaluation algorithm are evaluated in this study. The general detection concept is based on brightness differences, which are recognized using a digital image processing algorithm. Here, groups of pixels are tracked over the moving specimen, and if the brightness difference exceeds a threshold, a possible defect is detected [21]. The detections are further filtered based on their morphology. Different parameters are set for different defect types (e.g., pinholes and line defects). The image processing algorithm from Keyence was configured using reference samples with defects and the Keyence Vision Editor software. Detected defects by the manufacturers software are given in Fig. 13.

Active thermography allows for the use of various excitation sources [26]. The excitation generates a thermal response on the material’s surface, which correlates with its thermophysical properties or reveals near-surface defects [18]. This response is recorded using an IR camera. Common optical excitation methods include halogen lamps (used in lock-in thermography) and flashlights (used in pulse thermography). For in-line evaluation of electrodes, fast excitation with rapid heating is essential. Laser excitation is particularly well-suited for this purpose.

Bild vergrößern

Bild vergrößern

As stated in Kurfer et al. [22], thermography systems can detect particles smaller than 0.050 mm, making this approach effective for identifying surface defects on electrodes. In this study, a laboratory setup with a moving heat source was employed, and future studies will expand on this by incorporating a moving electrode with a fixed heat source. The setup is illustrated in Fig. 7. The optimal settings for the laser and the IR camera must be determined experimentally. These are the frame rate (30 Hz), laser power (195 W), and the scanning velocity of the laser (15 \(\frac{\text {m}}{\textrm {s}}\)). For each position on the specimen, images were captured until 1 s after the laser heated that specific point. The laser is directed onto the sample with a scanning mirror. The IR camera, with a resolution of 1280 x 1024 pixels, records the temperature response of the sample. The data is exported and analyzed using the DisplayIMG software from Edevis. The raw images are available in TIFF format. A 3D-printed frame was used as a sample holder to create production-like environmental conditions, as the electrode in the production facility is also suspended in air.

The evaluation of the thermography dataset was done using the flying spot gradient method. This procedure is based on the fundamental idea that defects lead to local changes in thermal properties, which are reflected in the temporal gradient of the temperature recording. In the generation of a flying spot gradient image, the maximum temperature gradient for each pixel is calculated and displayed, allowing all defects to be visualized simultaneously in a single image. In Fig. 8 (a), the image stack of a thermography measurement is shown. For each image, the maximum gradient was calculated, and the resulting image is displayed in (b). This image contains, on the one hand, line defects. On the other hand, cracks in the coating are visible, which stem from the transport of the specimen and were not possible to mitigate. In order to determine which defects were detected or not detected, digital image processing was used on the gradient images. An automated approach was chosen here to prevent intruding a bias from the operator in the evaluation. In the first step, a Canny edge filter was applied to the image. With subsequent morphological operations, the shape of the detected defects was refined and artifacts were removed. The final segmentation mask is shown in Fig. 8 (c).

Bild vergrößern

This study compares the two measurement systems by evaluating their POD. POD is a standardized approach [1, 2] used to evaluate and benchmark the performance of NDT methods. The MIL-HDBK-1823A handbook [3], a common reference in the aerospace industry, is used here along with its implementation for the analysis.

The POD quantifies the probability of identifying a defect with a specific characteristic, such as its area, using the selected measurement system. Several factors impact the POD, including physical constraints (like temperature) and the evaluation methodology employed. There are two primary types of POD analyses: hit/miss POD and \(\hat{a}\) vs. a POD. The hit/miss POD is used in this study since the hits and misses of the NDT systems were estimated manually, representing a binary output. The POD was calculated in this study aswhere \(\mu \) and \(\sigma \) are the mean and standard deviation of the dataset. \(\Phi \) is a linkage function [3]. In this study, the logistic linkage function was appliedThe parameters of these equations were estimated using maximum likelihood estimation. To derive the confidence bounds for the resulting curve, the likelihood-ratio method is employed, which analyzes a likelihood surface surrounding the maximum likelihood value. This process computes POD curves that correspond to the specified confidence interval, helping to define the upper and lower limit curves. Generally, a 95% confidence bound is used for this analysis. The accuracy of these confidence limits is affected by several factors, including sample size and the distribution of defect sizes relative to the actual POD curve [33]. A schematic POD curve is shown in Fig. 9. Several characteristic values can be estimated using a POD curve to compare the performance of NDT systems. In this study the \(a_{90}\) and the \(a_{50}\) values are used.

Bild vergrößern

To ensure reproducibility, defects were intentionally introduced into the carbon coating on the copper anode. The artificial defects were created in a controlled manner as described below. These methods were developed in Weinzierl [34]. To generate the pinhole and line defects, which are meant to simulate process deviations, a 3D-printed setup was used. This is shown in Fig. 10. Needles with the following diameters were used for this experimental setup: 0.15 mm, 0.25 mm 0.35 mm, 0.4 mm, and 0.5 mm. The distance of the needles to the coating surface can be precisely adjusted with two screws and nuts. As part of the experimental setup, the crossbar was positioned at a distance of 137 mm before the first drying segment within a roll-to-roll coating line. The substrate and liquid coating are moving through the plant at constant speed. To create defects, the crossbar is manually lowered so that all needles touch the coating surface and thus displace the material applied to the substrate. It is crucial that all needles penetrate completely into the coating without damaging the foil. The length of the line defect was controlled by holding the crossbar in the lowest position for a set time. Pinhole defects were created by pressing the needles at a predetermined position into the coated surface, with the coating process being stopped during this procedure.

Bild vergrößern

For a meaningful \(\hat{a}\) vs. a POD analysis, at least 30 samples with defects are necessary while at least 60 data points are required for a hit/miss POD analysis [33]. Due to problems in transporting the specimens 60 data points could only be reached for the line defects.

Table 2 lists the number of samples produced for each defect type. The line defects were executed with a length of approximately 10-20 cm per sample. For evaluation, the line defects were divided into different sections, as shown in Fig. 13. It was then assessed whether the line defect was detected in each individual section. Consequently, the number of data points used for line defects is significantly higher than for other defect types. The same specimens were investigated using both the optical system and thermography. However, some defects were damaged during transport to the thermography setup. Therefore, the data size is slightly smaller for the thermography evaluation.

To estimate the actual size of the pinhole and line defects, reference measurements were conducted using a 3D profilometer (Keyence VR-6200). The cross-sectional area of the defects was chosen as the characteristic value, which depends on the depth and width of the defects. In Fig. 11, a profilometer measurement of a pinhole defect is shown.

The resulting measured defect areas and their correlation with different needle sizes are shown in Fig. 12 for the line defects in (a) and for the pinholes in (b). For the line defects, only the largest needle size of 0.5 mm leads to significantly larger defect areas. For the pinholes, the correlation between needle size and defect area is more clear. Additionally, the measured defect areas vary considerably for a given needle size. This could be due to the fact that, with larger needle diameters, a minimal increase in the force during defect creation leads to significantly larger defects. This stresses the importance of conducting the POD study based on the sizes estimated with the reference profilometer measurement.

Metallic particles could not be used to evaluate the impact of particle contamination on cell performance, as the risk of damaging the calander rollers in the next process step would have been too high. Therefore, plastic particles made of polyvinyl chloride were used for the cell testing, as this material is softer and does not leave dents in the calender rollers. The experimental design of the cell tests is given in Table 3.

To evaluate methodologically the influence of the generated defects on cell performance of commercial high-energy cells, cylindrical cells were assembled with the prior produced electrodes. The cell format which was used, is an internal prototype standard of BMW, with dimensions of 95 mm in height and 46 mm in diameter. This cell design targets a specific energy of around 300 \(\frac{\text {Wh}}{\textrm {kg}_{\textrm {Cell}}}\) and a volumetric energy density of approximately 800 \(\frac{\text {Wh}}{\textrm {L}_{\textrm {Cell}}}\) for each cell. As anode active material, a composition of graphite with a silicon content of less than 10 \(\text {weight}\)-% was used. The cathode active material consisted of NMC with a nickel content of over 85 atomic-%. The cut-off voltage limits of the full-cell were set to 2.8 V and 4.2 V. The capacity ratio of the negative electrode to the positive electrode, also referred to as the N/P ratio, was around 1.05 in these cells. For the electrolyte, linear and cyclic carbonates were used as solvents, fluoroethylene carbonate was added for the formation of a solid electrolyte interphase, and LiPF\(_{6}\) served as the conducting salt.

The assignment of all cells produced in the experiment to the respective defect type is shown in Table 4. All cells underwent a defined formation process. During the activation cycle, charging was carried out at a rate of C/10 with constant current and voltage, discharging was done with constant current. Subsequently, the cells were charged to 50% of their capacity at a charge rate of C/5. This was followed by a discharge pulse at a 1C rate for 30 seconds under constant current conditions. The cells were then further discharged to a state of charge of 30% at a rate of C/5. Cells with pinholes and particle defects were subjected to a life-cycle test, during which the discharge rate was maintained consistently at C/2 at an ambient temperature of 25\(^\circ \)C.

Bild vergrößern

Bild vergrößern

At the beginning and the end of the lifetime test, a comprehensive cell test was conducted. During the cycling, regular checks were carried out every 75 cycles to monitor the state of health of each cell, with comprehensive and minor examinations being alternated. In the charging process, the cell was charged up to a predetermined cut-off voltage at the specified C-rate and then continued to be charged at constant voltage until the current dropped to C/50. Discharging occurred at a constant current under a defined C-rate until the predetermined minimum voltage was reached. This test cycle was repeated until the End-of-Life criterion was met. Cells with line defects were examined in a separate test, with the cycles being identical to those described above. For safety reasons the number of cycles for these cells was limited to 200 and took place in a specially equipped safety bunker. During these tests, interim checks were conducted every 50 cycles.

Bild vergrößern

Bild vergrößern

The images captured by the optical camera were analyzed using the Keyence software. To identify line defects, the line defect filter in the Keyence software was applied. The results of this analysis are shown in Fig. 13 (a). On the left side of the figure, the line drawn with a 0.5 mm needle is displayed, while on the right, the line created with a 0.15 mm needle is shown.

As evident in Fig. 13 (a), smaller needle sizes were detected either incompletely or not at all by the software, especially for the two smallest needle sizes of 0.15 mm and 0.25 mm. For detecting line defects, Lines were evenly distributed horizontally across the image, appearing in green in Fig. 13 (a) on the right. A manual assessment was conducted to verify if the green lines aligned with the actual defects.

For detecting pinhole defects, the point recognition filter in the Keyence software was applied. The pinhole defects in Fig. 13 (b) increase in size from left to right. In the two rows of pinhole defects, three out of five defects were successfully recognized. The smallest defects were indistinguishable from the noise in the images.

Particles were detected using the same algorithms in the manufacturers software as those for the pinhole defects. In the image shown in Fig. 13 (c), the smallest particle, with a size of 0.15 mm, was not detected (marked in the image). Larger particles, with sizes of 0.30 mm and 0.35 mm, were recognized by the software. The particle sizes were verified using a digital microscope.

An example of pinholes is shown in Fig. 14 (b). Not all defects were recognized by the image processing algorithm applied to the thermography images. The leftmost point represents a needle size of 0.5 mm, while the rightmost point represents a needle size of 0.15 mm. Three rows of pinhole defects were introduced in the specimen. In the bottom row, no defects were detected here in the evaluation.

In Fig. 14 (c), 10 particles with a size of 0.50 mm are shown in both the thermography gradient image and the optical image. All particles of this size were successfully detected. The vertical lines in the processed IR gradient image result from a software error, which is not expected to impact the evaluation.

The POD curves for the line defects are shown in Fig. 15 (a) for the optical system and (b) for thermography. It can be seen that thermography achieves a higher POD overall. Specifically, the \(a_{90}\) value is 0.16 \(\text {mm}^2\) for thermography, compared to 0.52 \(\text {mm}^2\) for the optical system. The largest defects were captured by both systems. The number of investigated defects is higher for the optical system than for the thermography system, which would lead to narrower confidence bounds. With the optical system, 63.6% of the defects were detected, whereas with thermography, 66.5% were detected.

The POD curves for the pinhole defects are given in Fig. 16 for the optical system and thermography system in (a) and (b), respectively. For the optical system, a clear trend is visible, where larger defects are more easily detected by the NDT methods. This trend is not observed with the thermography system. For the pinhole defects and the thermography evaluation, no statistically significant POD curve could be estimated. As shown in Fig. 16 (b), even larger defects (primarily located in the bottom row) were missed by the thermography evaluation.

Bild vergrößern

Bild vergrößern

Reasons for this could include the evaluation procedures (such as the gradient and the edge detection methods) or physical limitations of the thermography system. In contrast to the line defects, the pinhole defects have one less dimension, which alters heat dissipation and results in lower contrast. The generally larger confidence bounds for the pinhole defects can be explained by the smaller dataset size compared to the line defects. With the optical system, 81.1% of defects were detected, whereas with the thermography system, 59.5% were detected. The \(a_{50}\) and \(a_{90}\) values were provided only for the optical system and are smaller here than for the line defects. However, it should be noted that the sample size, with 37 samples, is relatively small, which limits the representativeness of the POD analysis.

Bild vergrößern

Bild vergrößern

For defects representing particle contamination, the nominal size of the particles was used to estimate the POD curves. The POD curves are shown in Fig. 17 (a) for the optical system and in (b) for the thermography system. The optical system detected almost all defects, with a 97.9% detection rate. Only one of the smallest particles was missed. The \(a_{50}\) and \(a_{90}\) values are close to each other for the optical system. For the thermography system, more particles were missed, including some of larger size. In total, 65.9% of the particles were detected.

It should be noted that the dataset size for the pinhole and particle defects is below the recommended 60 data points for a hit/miss evaluation [33].

The summary of results of different defect types depending on the NDT method is shown in Table 5.

To assess the methodologies used for understanding the criticality of defects on the performance of Li-Ion battery cells, it is necessary to correlate anomalies in cell performance with the effects on the electrodes. During the lifetime testing, a systematic comparison of the cells with a reference is conducted. Key performance indicators, including capacity retention over the full life cycle and maximum discharge temperatures during interim checks, are thoroughly examined in this process. Firstly the cells are evaluated for their capacity throughout their entire life cycle, with a particular focus on detecting any decrease in specific discharge capacity compared to previous measurements. Figure 18 presents a direct comparison of the specific discharge capacities of cells that have been modified with pinhole defects on the anode as described in Table 4 the unchanged reference cells. This comparison is based on the relative decrease in capacities compared to their initial capacities.

The initial capacities of the reference cells were 32.8 Ah with a standard deviation of 0.11 Ah, while the cells with pinhole defects had an initial capacity of 33.05 Ah with a standard deviation of 0.09 Ah, and the cells with particle defects had an initial capacity of 33.12 Ah with a standard deviation of 0.07 Ah. The capacity differences are caused by process variations during the manufacturing of the reference cells, which resulted in the installation of slightly less electrode material, as evidenced by a lower total weight after formation, compared to the modified cells. In Figs. 18, 19 and 20 the regularly appearing declines in specific capacity retention are attributed to the interim checks. Additionally, the negative slope of the curves is indicative of the cells’ cyclic aging. In this analysis, the magnitude of the decline is deemed negligible as it reflects expected behaviour. The primary focus is on identifying anomalies that become apparent when comparing the variations across the individual curves. The analysis clearly indicates that the pinhole defects do not have a significant impact on capacity retention of the cells. Furthermore, no premature cell failures were observed during the testing period. Figure 19 illustrates the capacity trend of the cells with particle defects.

It is evident that Cell 12, which contains large particles at a low frequency, shows a deviation from the capacity loss of the reference cells after 340 cycles, with a measured capacity drop of 0.71 Ah. Cell 14, which is a cell with large particles at a low frequency, shows an abrupt drop in capacity over 0.9 Ah after 449 cycles. In addition, Cell 7, containing small particles at a high frequency, experienced a complete failure after few cycles due to overheating during an interim check whereas the data is not shown here. Figure 20 illustrates the specific capacities for cells with line defects in comparison to reference cells. The initial capacities are recorded at 33.19 Ah for Cell 17 and 33.26 Ah for Cell 18.

The data indicates that Cell 18 experienced a significant drop in specific capacity by 0.96 Ah after just 100 cycles. Furthermore, the decline in specific capacity is more pronounced in cells with defects compared to reference cells within the observed cycles. A linear fit of the trend lines reveals a slope of -0.03 for Cell 17, -0.15 for Cell 18, and -0.02 for each of the reference cells, suggesting a steeper degradation rate for the defective cells. The thermal behaviour of the cells was closely monitored during interim checks, with the maximum recorded discharge temperatures for cells with particle contamination and pinhole defects presented in Diagrams Figs. 21 and 22. It should be noted that the temperatures of the cells with line defects were not fully recorded, and therefore they are not included in this evaluation. During the interim checks, the charging current was maintained at C/3, while the discharge current was alternated between 1C and 2C.

The data presented in Fig. 21 reveals that cells with defects display a temperature spread, with a maximum recorded temperature of 42.7\(^\circ \)C for Cell 1 and a minimum of 35.8\(^\circ \)C for Cell 10 for the 1C discharge pulse.

Bild vergrößern

Figure 22 shows that the thermal behaviour observed during the 1C pulse is also evident in the 2C pulse. However, for cells with particle defects, temperatures reach up to 61.9\(^\circ \)C, with the lowest temperatures recorded at 47.8\(^\circ \)C. The mean temperatures across the three interim checks are summarized in Tables 6 and 7. For this analysis, data from the initial check, a check after approximately 170 cycles, and a check after around 340 cycles were utilized, as these checks provided a sufficient data basis. Cells that did not reach these cycle counts were excluded from the investigation. For the calculation of the mean values for the cells with pinhole defects, 6 cells could be included, while for the reference cells, 7 cells were used in the analysis. These different sample sizes result from the completeness of the available data sets. A t-test analysis shows that the mean values of the temperature distributions differ significantly between the cell groups. The variances of the data were previously tested using the method of Levene [7] and found to be statistically equal. It is important to note that not all cells were able to complete the third interim check with the 2C discharge, and therefore, the data from this assessment point was excluded from the analysis.

Table 6 presents the mean values of the thermal behaviour during a 1C pulse interim check. It indicates a trend where reference cells exhibit the lowest maximum temperatures, while cells with pinhole defects show elevated mean temperatures, and those with particle defects show the overall highest mean temperature with 39.83\(^\circ \)C.

Bild vergrößern

Bild vergrößern

From Table 7, it is evident that the temperatures of the reference cells are the lowest during the 2C discharge cycle at 52\(^\circ \)C. The mean values of the temperatures of the cells with pinhole defects lie between those of the reference cells with 54\(^\circ \)C to 56 \(^\circ \)C and the cells with particle contamination with 56\(^\circ \)C to 57\(^\circ \)C. Furthermore, it is observed that the temperatures of the cells with pinhole defects during the 2C discharge decrease with increasing cycle counts, whereas the temperatures of the cells with particle defects increase. The maximum temperatures of the reference cells, on the other hand, remain nearly constant over the test period. Through the opening of Cell 9 with large, high-frequency pinhole defects, it was found that the originally introduced defect pattern on the anode was no longer identifiable within the jelly roll. Lithium plating, which was assumed to be the primary cause for this type of defect, was only found to be distributed non-specifically along the entire anode length. One possible explanation for this is that the pinhole defects are so heavily compressed during the calendering process that the material displaced to the side refills the defects. This suggests that the mechanical deformation of the anode during the production process effectively mitigates the impact of the intentionally introduced pinhole defects and masks their presence in the final cell structure. Disassembling Cell 7, which was equipped with small particles at a high frequency and had been removed from the test after 168 cycles due to high temperatures during an interim check, the particles were successfully identified on the anode as Fig. 23 shows.

Additionally, a significant fold was discovered in the separator, which in this instance is presumed to be the primary source of the defect as can be seen in Fig. 24.

When inspecting Cell 10, which contained small particles at high frequency and had completed 253 cycles, a defect was observed in the area of the anode where the particles were distributed. This defect was also apparent on the separator, as can be seen in Fig. 25. Apart from these identified issues, no additional defects were visible on the electrodes after the cell was opened.