Systems, methods, and computer-readable media for diagnosing rapidly degrading battery cells

Abstract

A system and method for diagnosing rapidly deteriorating battery cells are disclosed. During at least one charging or discharging process, a direct current (DC) pulse is applied to the battery cell. The DC internal resistance (DCIR) is determined by measuring the change in voltage over time in response to the applied DC pulse. Ohmic resistance (R) is then excluded from the DCIR. o ) component to separate charge transfer resistance (R) ct ( ) component. A battery cell is diagnosed as rapidly deteriorating by predicting at least one state of harmlessness (SOH) in a battery cell that indicates rapid cell degradation.

Description

Technical Field

[0001] This disclosure relates to a technique for accurately diagnosing and mitigating rapidly deteriorating battery cells.

[0002] This application claims priority and benefit to U.S. Provisional Application No. 63 / 729,193, filed December 6, 2024, and U.S. Provisional Application No. 19 / 078,958, filed March 13, 2025, the disclosures of which are incorporated herein by reference. Background Technology

[0003] Batteries have become a key component in modern energy management systems. Battery technology allows for the storage of excess electricity during periods of low demand and the release of insufficient electricity during periods of high demand, thereby optimizing energy use (by reducing the limitations of solar and wind power), reducing reliance on fossil fuel-based power generation (such as gas turbines or internal combustion engine vehicles), and mitigating the effects of climate change by reducing greenhouse gas emissions.

[0004] As lithium-ion batteries (LIBs) become indispensable for many applications, ensuring their safety is paramount. Research in this field focuses on estimating or predicting the state of health (SOH) of LIBs, a term that typically describes a battery's capacity relative to its rated capacity over a given cycle. LIB capacity degradation occurs gradually due to side reactions during battery cycling. When a battery's SOH drops below 80%, it is considered to have reached its end-of-life (EOL), indicating that it is no longer safe to use the battery.

[0005] Methods for estimating or predicting the state of harmonics (SOH) of a battery cell (LIB) can generally be categorized into physics-based methods and data-driven methods. Physics-based methods include equivalent circuit models, which represent the deteriorating components of the battery as circuit components such as resistors and capacitors; empirical methods, which simplify complex degradation rates into empirical formulas; and electrochemical models (EM), which accurately simulate the electrochemical behavior inside the cell while aiming to manage computational costs. However, both equivalent circuit and empirical methods struggle to fully capture the complexities of degradation. While providing detailed insights, EM typically involves many unknown parameters and significant computational requirements. These complexities can lead to inaccuracies, especially when simulating scenarios with rapidly deteriorating cells.

[0006] Given these challenges, data-driven models are increasingly favored for predicting the lifetime characteristics of lifecycle bins (LIBs) that degrade rapidly under conditions such as fast charge-discharge and harsh operating environments. These models establish a relationship between the Health Index (HI) recorded during the initial charge-discharge cycle and future State of Health (SOH), enabling SOH prediction when new LIB data is available. HI is divided into Direct Health Index (DHI) and Refined Health Index (RHI). DHI is easily measured during cycling, while RHI requires separate post-cycle measurement.

[0007] Typical direct health indices (DHIs) such as voltage, current, and temperature (VIT) are not directly related to battery capacity. Therefore, estimating or predicting state of health (SOH) using these indices often requires complex deep neural networks. However, accumulating sufficient battery cycling data up to end-of-life (EOL) to effectively train these networks is expensive and challenging, limiting the availability of DHIs for accurate future SOH predictions. Furthermore, there are significant differences between typical cycling experiments conducted in the lab and actual charging / discharging patterns in electric vehicles (EVs). To overcome these data limitations, research utilizing transfer learning is emerging. Nevertheless, transfer learning has not consistently delivered satisfactory performance when battery composition or operating conditions change, highlighting the need for simpler models when cycling datasets are limited.

[0008] RHI (Reverse Activated Highness) generally correlates more strongly with SOH (State of Charge) than DHI (Depleted Highness), allowing for the use of simpler models, such as linear models, for lifetime prediction. For example, studies have demonstrated effective SOH prediction using simple models based on ratios such as dQ / dV. However, measuring these variables during actual electric vehicle (EV) operation is challenging, making it difficult for onboard battery management systems (BMS) to utilize them in real-time. Some studies have explored using capacity as an input variable for SOH estimation. Capacity trends are strong predictors of future battery performance and can be monitored during EV operation; however, accurately measuring capacity under real-world conditions is problematic because it requires cycling the state of charge (SOC) from 0% to 100% per cycle.

[0009] Therefore, while DHIs are easy to measure and commonly used, their weak direct correlation with SOH often requires complex modeling methods. Conversely, RHIs are strongly correlated with SOH, but real-time measurement in EVs presents significant challenges. Identifying input variables that are both easy to measure and closely related to SOH is crucial for enhancing user safety.

[0010] Internal resistance (IR) is another indicator highly correlated with capacity degradation, making it an important item in battery analysis. Recognizing the importance of IR, several studies have aimed to predict IR directly using DHI or RHI as input variables. Additionally, researchers have incorporated IR-based items into models used to estimate or predict SOH, typically using resistance measurements from electrochemical impedance spectroscopy (EIS). The value of EIS lies in its accurate differentiation of ohmic resistance (R... o ), charge transfer resistance (R) ct ) and polarization resistance (R p The ability to maintain and utilize the resistance term closely related to SOH (State of Health) is crucial. However, the practical application of EIS in EVs faces limitations because it cannot be performed during normal battery operation, presenting similar challenges to other RHIs in real-time EV diagnostics.

[0011] The background description provided herein is intended to present the overall context of this disclosure. Unless otherwise stated herein, the materials described in this section are not prior art to the claims of this application and are not, by virtue of their inclusion in this section, admitted to being prior art or suggestions of prior art. Summary of the Invention

[0012] Technical issues

[0013] This disclosure relates to providing a system and method that can utilize, by means of, a DC internal resistance (DCIR) (especially a charge transfer resistor (R)). ct The value of ) is used to separate the increase in resistance caused by sudden side reactions, and the future SOH of the battery cell is predicted based on the linearity and monotonicity between the increase in resistance and the state of health (SOH) of the battery.

[0014] Furthermore, the present invention relates to a system and method for diagnosing battery cells, which can be applied even under conditions where data is limited in the actual operating environment of electric vehicles, by applying a linear model with simple structure and sparsity (e.g., an elastic network (EN) regression model) instead of a complex deep neural network model.

[0015] These and other objects and advantages of this disclosure will become apparent from the following detailed description and from exemplary embodiments of this disclosure. Moreover, it will be readily understood that the objects and advantages of this disclosure can be achieved by the methods and combinations thereof shown in the appended claims.

[0016] Technical solution

[0017] According to one aspect, this disclosure relates to a system for diagnosing rapidly deteriorating battery cells, the system comprising: a controller including a processor and a non-transitory memory storing calculation instructions, the calculation instructions being configured, when executed by the processor, to: (a) apply a direct current (DC) pulse to the battery cell during at least one period of a charging process and a discharging process; (b) measure a change in voltage over time in response to the applied DC pulse to determine a DC internal resistance (DCIR); and (c) remove an ohmic resistance (R) from the DCIR. o ) component to separate the charge transfer resistance (R) of DCIR ct (d) By predicting at least one state of health (SOH) of a battery cell indicating rapid cell degradation, based on R ct The component analysis diagnosed the battery cell as rapidly deteriorating.

[0018] In some cases, step (d) includes: transferring R ct The components are input as features into the classifier to predict at least one SOH of the battery cell at 180 charge-discharge cycles and 270 charge-discharge cycles; and the battery cell is diagnosed as rapidly deteriorating when the difference between the predicted SOH at 270 charge-discharge cycles and the predicted SOH at 180 charge-discharge cycles is at least 2.5%.

[0019] In some cases, the classifier is a linear regression model that implements Elastic Network (EN) regression.

[0020] In some cases, the controller is configured to determine the DCIR at multiple timescales including at least one of a DC pulse of 0.1 seconds and 10 seconds or 30 seconds, and step (c) includes subtracting the DCIR measured at 0.1 seconds from the DCIR measured at at least one of 10 seconds or 30 seconds to exclude R. o Components and separation of R ct Quantity.

[0021] In some cases, the controller is configured to apply a DC pulse after one or more of the 0, 90, 180, or 270 charge-discharge cycles of the battery cell.

[0022] In some cases, the controller is configured to determine the DCIR at one or more time scales of the applied DC pulse, and one or more time scales include 0.1 seconds, 1 second, 5 seconds, 10 seconds, 15 seconds, 20 seconds, 25 seconds, 30 seconds, or 60 seconds.

[0023] In some cases, the controller is configured to determine the DCIR at one or more states of charge (SOC) of the battery cell, and one or more SOCs include 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%.

[0024] In some cases, the controller is configured to apply DC pulses during a hybrid pulse power characteristic (HPPC) test at multiple states of charge (SOC); and to determine multiple values ​​of DCIR from the HPPC test, wherein the HPPC test includes applying DC pulses with at least two different currents.

[0025] In some cases, at least two different currents include at least two of 0.5C, 1C, 1.5C, 2.0C, or 2.5C.

[0026] In some cases, battery cells are configured to power electric vehicles, and the controller is integrated with the electric vehicle's battery management system (BMS).

[0027] In some cases, battery cells are configured to store renewable electricity generated by solar or wind power in order to reduce reliance on fossil fuel-based power generation and mitigate the effects of climate change.

[0028] According to another aspect, this disclosure relates to a method for diagnosing rapidly deteriorating battery cells, the method comprising the steps of: (a) applying a direct current (DC) pulse to the battery cell during at least one period of a charging process and a discharging process; (b) measuring the change in voltage over time in response to the applied DC pulse to determine the DC internal resistance (DCIR); and (c) removing ohmic resistance (R) from the DCIR. o ) component to separate the charge transfer resistance (R) of DCIR ct (d) By predicting at least one state of health (SOH) of a battery cell indicating rapid cell degradation, based on R ct The component analysis diagnosed the battery cell as rapidly deteriorating.

[0029] According to another aspect, this disclosure relates to a non-transitory computer-readable medium having computer-executable instructions stored thereon, the computer-executable instructions causing the controller to perform operations for diagnosing rapidly deteriorating battery cells in response to execution of a processor of the controller, the operations including: (a) applying a direct current (DC) pulse to the battery cell during at least one period of a charging process and a discharging process; (b) measuring the change in voltage over time in response to the applied DC pulse to determine the DC internal resistance (DCIR); and (c) removing ohmic resistance (R) from the DCIR. o ) component to separate the charge transfer resistance (R) of DCIR ct(d) By predicting at least one state of health (SOH) of a battery cell indicating rapid cell degradation, based on R ct The component analysis diagnosed the battery cell as rapidly deteriorating.

[0030] It should be noted that the technical effects obtainable through this disclosure are not limited to those described above, and other effects not mentioned herein will be clearly understood by those skilled in the art based on the following description.

[0031] Beneficial effects

[0032] According to one aspect of this disclosure, the system and method can utilize a DC internal resistance (DCIR) (especially a charge transfer resistor (R)). ct The values ​​of the resistance values ​​are used to separate the increase in resistance caused by rapid side reactions, and the future SOH of the battery cell is predicted based on the linearity and monotonicity between the increase in resistance and the state of health (SOH).

[0033] Furthermore, according to one aspect of this disclosure, the system and method can be effectively applied even with limited available data in the actual operating environment of electric vehicles by using linear models with simple structure and sparsity (such as elastic network (EN) regression models) instead of complex deep neural network models.

[0034] The effects of the present invention are not limited to those described above, and other effects not mentioned will be clearly understood by those skilled in the art based on the description of the claims. Attached Figure Description

[0035] The accompanying drawings illustrate exemplary aspects of this disclosure and, together with the following detailed description, are intended to provide a further understanding of the technical spirit of this disclosure. However, this disclosure should not be construed as being limited to the drawings.

[0036] Figure 1 This is a table showing a detailed overview of the HPPC tests performed during each RPT according to one aspect of this disclosure, indicating the number of DCIR measurements performed in each test.

[0037] Figure 2 This is a graph showing the SOH degradation rate of the battery cell for each 90-cycle interval according to one aspect of this disclosure.

[0038] Figure 3 This is a table showing the distribution of SOH degradation rate of the battery cell under normal pressure and high pressure during the last 90 cycles, according to one aspect of this disclosure.

[0039] Figure 4This is a graph showing the average degradation rate of both Class 0 and Class 1 cells according to one aspect of this disclosure at 0, 90, 180 and 270 cycles.

[0040] Figure 5 It is a table showing the proportional distribution of Class 0 and Class 1 cells under varying pressure, temperature and charging modes according to one aspect of this disclosure.

[0041] Figure 6 This is a diagram illustrating the HPPC test procedure performed with 2.5C and 1C long pulses at SOC=80%, 50%, and 20% according to one aspect of this disclosure.

[0042] Figure 7a This is a graph showing the voltage changes and the resistance components associated with each change when a pulse current is applied, according to one aspect of this disclosure.

[0043] Figure 7b It is shown that R according to one aspect of this disclosure ct and R p With R o The separate tables.

[0044] Figure 8 This illustrates the SOH according to one aspect of this disclosure. 270 The average regression performance of the linear regression model across five random seeds is presented in a table as a function of the l1 ratio.

[0045] Figure 9 This illustrates the SOH measured for each 90-cycle interval for both Class 0 and Class 1 cells according to one aspect of this disclosure. 270 A table showing the correlation between DCIR and DCIR.

[0046] Figure 10 This is a table showing the state of charge (SOC) distribution of the top 20 DCIRs with Pearson and Spearman coefficients > 0.5 according to one aspect of this disclosure.

[0047] Figure 11 This is a table showing the first 20 DCIRs according to one aspect of this disclosure, wherein the values ​​are adjusted by subtracting an initial 0.1s DCIR to separate values ​​exceeding R. o . contributions.

[0048] Figure 12 This illustrates the normalization term Ratio according to one aspect of this disclosure. avg The definite diagram.

[0049] Figure 13a and Figure 13bThis illustrates different ambient temperatures (T) according to one aspect of this disclosure. amb Ratio in normal pressure cells during charging mode avg A graph comparing the decrease in average SOH.

[0050] Figure 14a and Figure 14b This illustrates the ratio under pressure variation according to one aspect of this disclosure. avg A graph comparing the decrease in average SOH with the decrease in T amb =10℃ and T amb 'Power' mode cell at 45℃.

[0051] Figure 15 This is a flowchart illustrating a method for diagnosing rapidly deteriorating battery cells according to one aspect of this disclosure.

[0052] Figure 16 This is a diagram illustrating a system including a controller for diagnosing rapidly deteriorating battery cells according to one aspect of this disclosure.

[0053] Figure 17 This illustrates one aspect of the disclosure. Figure 16 A diagram showing the integration of the controller in various applications. Detailed Implementation

[0054] In the following, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Before this description, it should be understood that the terminology used in this specification and the appended claims should not be construed as limited to its general and dictionary meanings, but should be interpreted based on the meanings and concepts corresponding to the technical aspects of the present disclosure, on the basis of allowing the inventors to appropriately define the terminology for the best interpretation.

[0055] Therefore, the description presented herein is merely a preferred example for illustrative purposes and is not intended to limit the scope of this disclosure. It should be understood that other equivalent substitutions and modifications may be made thereto without departing from the scope of this disclosure.

[0056] Ordinal terms such as “first” and “second” can be used to distinguish one element from another among various elements, but are not intended to limit elements by terminology.

[0057] Throughout the specification, when a section is referred to as “comprising” or “including” any element, it means that the section may further include other elements without excluding other elements, unless otherwise expressly stated.

[0058] Furthermore, throughout the specification, when one part is referred to as "connected" to another part, it is not limited to the case where they are "directly connected," but also includes the case where they are "indirectly connected" with another element inserted between them.

[0059] Given the aforementioned limitations, DC internal resistance (DCIR) has emerged as a viable alternative, which can be measured during charging and discharging when a constant current (CC) is applied. DCIR is typically calculated by analyzing the voltage change relative to the current change over a specific duration (such as the first 0.1s or 1s) during a pulse test at a given SOC in the Mixed Pulse Power Characteristic (HPPC) test of the Reference Performance Test (RPT).

[0060] Unlike EIS, which requires alternating current, DCIR has the advantage of being measurable during driving because it uses direct current in a specific direction. Furthermore, EV chargers typically set a termination SOC level, thus providing a degree of SOC uniformity immediately after charging. Additionally, factors that might affect DCIR, such as the c-rate and temperature, tend to be consistent across the same driver, making them meaningful input variables for direct use. It is also inferred that the SOC just before charging (after discharging) is similar and uniform among drivers. It is noteworthy that charge transfer resistance (R) is closely related to capacity degradation. ct The terms did not change significantly within the 20% to 80% SOC range, indicating minimal impact on the accuracy of the regression model.

[0061] Using DCIR presents challenges, primarily in effectively separating the resistance component, as DC is only applied during cycling and the sampling time can be relatively long. Furthermore, researchers must infer which resistance term is dominant based on the measurement timescale, since different degradation mechanisms influence capacity decay under varying operating conditions. While relatively few studies have focused on using DCIR-based terms as input variables for SOH estimation, there are noteworthy cases. Sun et al. (Z. Sun, W. He, J. Wang, X. He, State of health estimation for lithium ion batteries with deep learning approach and direct current internal resistance, Energies 17(11)(2024)2487) developed a model using Long Short-Term Memory (LSTM) to estimate SOH using DCIR without separating the resistance component. Conversely, Van et al. (CNVan, DTQuang, Estimation of soh and internal resistances of lithium ion battery based on LSTM network, International Journal of Electrochemical Science 18(6)(2023)100166.) utilized R after the application of a pulsed current (in the first few microseconds) o The fact that it can be measured immediately and quickly, using ohmic resistors (R) derived from DCIR. o To estimate SOH, studies typically either do not differentiate between different resistance terms of DCIR over time scales, or focus only on R, which is easier to measure. o It is understood that no studies have yet investigated the use of charge transfer resistance (R) estimated via DCIR in rapidly deteriorating cells. ct ) is used as the main input variable. Under the condition of slow LIB degradation, R in the electrolyte o R at the electrolyte-electrode interface ct The rate of increase is comparable, making the traditional method effective. However, in cases where R deteriorates rapidly due to accelerated side reactions, ct R o The more rapid increase in resistance complicates accurate lifetime prediction of these cells using conventional methods. Furthermore, the resistance increases not only due to side reactions but also with factors such as ambient temperature (T0). amb The resistance varies with operating conditions such as pressure, making it challenging to predict lifetime characteristics by only considering resistance changes attributable to side reactions.

[0062] To address these challenges, this system and method utilize DCIR values, particularly in conjunction with charge transfer resistance (R). ct DCIR values ​​were used to predict the future state of health (SOH) of rapidly degraded cells. Separating the sudden rise in resistance due to rapid side reactions is feasible, and these resistance increases exhibit high linearity and monotonicity along with subsequent capacity decay. To this end, an EV-based dataset was used, and criteria for rapid capacity decay were derived from previous studies. 79 DCIR values ​​were calculated from each RPT (Real-Time Per Test), where Hybrid Pulse Power Characteristic (HPPC) testing was performed at three different SOC levels. The linear correlation between DCIR and SOH in each RPT cycle was calculated and compared between rapidly degraded and normally degraded cells. Based on these analyses, representative linear and sparse models were constructed using the Elastic Network (EN) method, where DCIR-based values ​​were taken as input. This model achieved high prediction accuracy, with a root mean square error (RMSE) of approximately 1.3%, which outperforms more complex neural network-based models. The choice of a linear model over complex deep neural networks reflects the limited data available for rapidly degraded cells in real-world EV scenarios, thus highlighting the importance of simplicity in the modeling approach. Furthermore, by employing a sparse model that selects a limited number of input variables, the information content of each variable can be determined, and the information content that best predicts SOH can be identified.

[0063] Therefore, this disclosure describes a system and method for diagnosing rapidly deteriorating lithium-ion battery cells by utilizing DC internal resistance (DCIR) as a diagnostic indicator. This system and method can identify significant deviations in the DCIR pattern, indicating rapid degradation of the battery cell, even before significant capacity decay becomes apparent. DCIR can be measured at multiple state-of-charge (SOC) levels and time scales to extract the resistive component, such as ohmic resistance (R). o ), charge transfer resistance ( ct ) and polarization resistance ( p The research focuses on ct The above is closely related to the capacity degradation caused by rapid side reactions.

[0064] Methodology

[0065] This section reviews the experimental dataset derived from electric vehicle driving patterns used in previous research by Lee et al. (J. Lee, S. Jegal, M. Chung, S. Kim, MJ Kim, DWKoh, JHLee, Electric vehicle pattern-based battery cycling dataset and its application in predicting rapid degradation, chemrxiv.org (2024).), outlines the rapid degradation criteria established from existing knowledge, and describes the DCIR term extracted from each RPT cycle. Additionally, a health state (SOH) term for the final cycle is introduced. 270 The EN model for regression of ).

[0066] Dataset Overview and Standards for Identifying Rapidly Degrading Battery Cells

[0067] The master dataset used in the experiments disclosed herein comprises cycle data from 143 LIB cells subjected to various operating conditions. These LIB cells have nickel-cobalt-manganese (NCM) positive electrodes and graphite negative electrodes. The discharge protocol follows typical EV driving modes, incorporating regenerative braking and rest periods to closely simulate real-world EV conditions. Charging is performed using a strategy alternating between fast and slow modes, typical of EV charging systems. Specifically, operating conditions are conducted at three ambient temperatures (T0, T ... amb =45, 25, 10℃), three charging modes (fast charging marked "Power", "Normal" (specifically, the average C-rate in "Power" mode, with a slightly higher SOC range than "Normal" mode, and the same maximum C-rate), and slow charging marked "Lifetime") and two applied pressures (4, 40 kgf / cm²). 2 The RPT was performed at 0, 90, 180, and 270 charge-discharge cycles. During each RPT, the galvanostatic intermittent titration technique (GITT) was initially used to measure the cell capacity via a 2.5V CC discharge test.

[0068] Subsequently, HPPC tests were conducted at 80%, 50%, and 20% SOC, as follows: Figure 1The details are shown in Table 1. Table 1 is a detailed overview of the HPPC tests performed during each RPT, indicating the number of DCIR measurements performed in each test. A total of 79 DCIR measurements were obtained per RPT. For each SOC, discharge currents of 2.5C and 1C were applied with two different pulses, allowing DCIR to be calculated based on these varying discharge currents. Additionally, five consecutive 20-second pulse tests were performed at 50% SOC. For clarity, within the HPPC framework, the 100-second test is referred to herein as the long pulse test, and the 20-second test as the short pulse test.

[0069] Based on the described dataset, this paper will briefly review the criteria established by Lee et al. in their previous study for identifying rapidly deteriorating cells (classified as 'Class 1' cells). Initially, to assess the presence of rapid degradation in the dataset, for the 134 cells analyzed, such as Figure 2 The figure shows the percentage of degradation for each 90-cycle interval. Figure 2 The percentage of state-of-the-art (SOH) degradation of the tested cells within each 90-cycle interval is shown. Each box plot spans from the first quartile (Q1) to the third quartile (Q3), where outliers (below 1.5 times Q1 minus the interquartile range (IQR) or above 1.5 times Q3 plus IQR (where IQR = Q3 - Q1)) are represented by points. The numbers within each box plot represent the average degradation value within each interval.

[0070] Clearly, significant degradation typically occurs during the first 90 cycles, primarily due to the stabilization process, with most cells stabilizing between 90 and 180 cycles. Beyond this point, while many cells exhibit their lowest degradation rates between 180 and 270 cycles, a significant increase in degradation rate is observed in some cells during the final 90 cycles. This pattern suggests rapid degradation, or an “inflection point” in the degradation curve, which previous work has indicated is more likely to occur under certain operating conditions, such as high voltage, extreme temperatures, and fast charging.

[0071] A standard for identifying Class 1 cells was established based on the degradation rate observed under different pressures during the last 90 cycles. Figure 3 This shows the effect under normal pressure (4 kgf / cm). 2 ) and high pressure (40 kgf / cm) 2The distribution of state-of-the-art (SOH) degradation rate of cells during the last 90 cycles is shown. It is generally known that high-voltage conditions, which accelerate degradation due to various side reactions, significantly affect cell degradation. Data shows that under normal pressure, only about 8% of cells degrade by more than 2.5% in the last 90 cycles, while under high voltage, this percentage increases significantly to 37%. Based on this observation, cells with degradation exceeding 2.5% in the last 90 cycles are classified as Class 1 cells, indicating rapid capacity decay, while other cells are designated as Class 0.

[0072] To verify whether this standard accurately captures the behavior of rapidly deteriorating battery cells, such as Figure 4 As shown, the average degradation rate of Class 0 and Class 1 cells was analyzed at 0, 90, 180, and 270 cycles. The results show that Class 0 cells exhibit an almost linear degradation pattern. Conversely, Class 1 cells show a significant acceleration in degradation, particularly in the last 90 cycles, thus indicating a distinct 'inflection point' in the degradation curve. This difference confirms the effectiveness of the applied criteria in identifying cells prone to rapid degradation.

[0073] Based on the defined criteria, the distribution of Class 0 and Class 1 cells in a total of 133 cells was examined to analyze the conditions under which each classification occurred. Figure 5 Table 2 shows the proportion of Class 0 and Class 1 cells under varying pressure, temperature, and charging modes, illustrating the percentage of each class under various operating conditions. Under normal pressure, most cells are classified as Class 0, while under high-temperature conditions (T) during power charging... amb At 45℃, only 6 out of 18 cells were identified as Class 1. Conversely, under high voltage, in power mode, T amb All cells at 45°C were classified as Class 0, while the other two experimental settings showed a high incidence of Class 1 cells (50% and 67%, respectively). Overall, only 18 out of 134 cells were classified as Class 1. Despite the intentional creation of conditions favorable for rapid degradation, the proportion of cells rapidly deteriorating under normal operating conditions remains low, consistent with existing knowledge. In previous studies, Class 0 and Class 1 cells were successfully classified with a recall of approximately 0.99 using only data from the first three cycles. In this current study, our goal is to use a simple model to identify DCIR-based characteristics that make Class 1 cells diagnosable, performing lifetime regression on these few Class 1 cells, which are even rare under real-world conditions.

[0074] Feature extraction

[0075] This section outlines the method used to derive multiple resistance terms from DCIR measurements obtained during HPPC testing. The HPPC testing process... Figure 6As shown in the diagram, long-pulse tests of 2.5C and 1C were performed at SOC=80%, 50%, and 20%. Additionally, after the long-pulse tests, five consecutive short-pulse tests (20s) were performed at SOC=20%. From SOC=100%, a given cell was discharged at 0.33C until it reached SOC=0%. Specifically, the discharge process was paused at SOCs of 80%, 50%, and 20% to measure DCIR. At these SOCs, 2.5C and 1C CC pulse currents were applied, and the DCIR at each SOC was measured based on the voltage change during that period. Specifically, HPPC included tests performed within 100 seconds (long-pulse test) and 20 seconds (short-pulse test). The division of the measurement timescale after measurement allows for targeting specific resistance components, each affected differently by the pulse duration.

[0076] Figure 7a and Figure 7b It is presented to show the voltage changes when a pulsed current is applied and the resistance terms associated with each change, and features extracted based on the resistance terms are introduced. Figure 7a The diagram shows the voltage changes and occurrence times caused by each resistor term when a pulsed current is applied for 100 seconds. Figure 7b The timescale of the features extracted from each pulse test and the proportion of each resistance term are shown. Only short pulse tests of 20s were performed, so the four terms after 30s were not extracted.

[0077] exist Figure 7a In the process, when a pulsed current is applied, the main reason is due to the ohmic resistance (R). o A rapid voltage drop was observed, and the ohmic resistance (R) o This occurs within the first few microseconds and typically represents the most significant drop observed throughout the entire HPPC test. This is followed by a drop in charge transfer resistance (Rc). ct The associated voltage drop originates from the electrochemical reaction at the electrode-electrolyte interface. After tens of seconds, the polarization resistance (R) p The fact that the mass transfer limitation becomes apparent reflects the limitations of mass transfer.

[0078] Based on this fundamental understanding, nine different time scales (t) were established. s It is designed to capture R with varying amplitudes. o R ct and R p ,like Figure 7b The details are as follows, to identify features closely related to the SOH regression from these components. The table on the left shows the R-values ​​extracted from the voltage / current difference at 0.1, 1, 10, 30, and 100 seconds of the initial application of the pulse current. o The item. To alleviate R oThe main effect on the total voltage drop is calculated by subtracting the DCIR of 0.1 seconds from the DCIR measured at 1, 10, 30, and 100 seconds, resulting in four additional DCIR values, as shown in the table on the right. This is compared to the actual R... o The effect lasts for 0.1 seconds, but the first 0.1 seconds are excluded because the sampling time is set to 0.1 seconds. However, in the case of rapid degradation, R... ct Significant changes within seconds, expected to be able to extract R ct and R p Significant values ​​in this aspect, even if insignificant changes in the first 0.1 seconds are excluded.

[0079] For short-pulse testing at 50% SOC, five specific DCIR measurements were utilized: including R o The effect is reflected in the direct readings at 0.1 seconds, 1 second, and 10 seconds, as well as the DCIR calculated by subtracting 0.1 seconds from the 1-second and 10-second measurements to focus on the long-term resistance change (R). ct R p The adjustment value is then calculated. Therefore, for each RPT test, 6 long-pulse tests are performed (for 2.5C and 1C, 2 at SOC=80%, 50%, and 20%) and 5 short-pulse tests are performed (5 consecutively at SOC=50%), resulting in 6×9=54 features from the long-pulse tests and 5×5=25 features from the short-pulse tests, thus generating 79 DCIR-based features for each RPT used for lifetime regression.

[0080] Modeling framework

[0081] This section introduces EN as a DCIR-based measurement for SOH. 270 The regression model is predicted, and the rationale for choosing Entropy Normalization (EN) as the modeling method is discussed. EN is characterized by utilizing a linear sparse model that leverages the linear relationship between input and output variables. EN is considered sparse because it selectively uses informational input variables through regularization terms, thereby reducing model complexity and mitigating the risk of overfitting.

[0082] EN is a linear sparse model that effectively combines features from both LASSO and RIDGE regressions to enhance model robustness and feature selection. Mathematically, it is represented as follows:

[0083] In this formula, y represents the dependent variable vector, X is the matrix of input variables, and w is the coefficient to be optimized. It is a hyperparameter that controls the overall strength of normalization, and l1 balances the LASSO (l1 norm) and RIDGE (l2 norm) components.

[0084] Using the L1-norm LASSO component to zero out the coefficients of less important input variables promotes model sparsity and emphasizes key features. Utilizing the L2-norm RIDGE component addresses overfitting by reducing multicollinearity among input variables. By adjusting l1 in the range of 0 to 1, the model can vary from pure RIDGE (when l1=0) to pure LASSO (when l1=1). This flexibility allows the model to leverage the advantages of both approaches, depending on the data and desired outcome. Increasing l1 drives more coefficients to zero, simplifying the model, while the L2-norm component helps balance the coefficients to reduce the chance of overfitting, which is crucial when dealing with multicollinearity or multiple predictor variables. EN tuning involves adjusting hyperparameters. And l1. The optimal settings for these parameters are usually found through a trial-and-error process, aiming to achieve the best balance between model complexity and prediction accuracy. If EN is tuned correctly, it allows for analysis of which input variables are most influential by observing which coefficients remain non-zero under various settings of l1, providing insights into the variables that most significantly affect the dependent variable.

[0085] Based on the characteristics of EN, a model was chosen for the study for several reasons. Firstly, due to the actual scarcity of Class 1 cells and the need for a model that can effectively utilize the vehicle's BMS, a linear model was selected. The expected resistance variable is compared to the VIT variable and the SOH. 270 The more direct correlation makes them more suitable for lifetime prediction in linear modeling scenarios. Regarding feature selection for the model, as mentioned above, 79 DCIRs are used as input variables. These DCIR measurements include various resistance terms, such as R... o R ct and R p Using a dense model with all these input variables can lead to overfitting, complicating the interpretation of which DCIRs are most relevant to battery life characteristics. Conversely, sparse models not only mitigate the risk of overfitting but also enhance model interpretability by highlighting key input variables via non-zero weight coefficients. The sparsity of the EN model is particularly valuable because it allows analysis of the importance of the measured resistance term across different time scales and SOCs, thereby enhancing the model's relevance and applicability to real-world EV settings. This approach supports a deeper understanding of which aspects of DCIR measurements best predict SOH. 270 This allows us to explain how the model can be implemented in actual EV operations. The SOH developed using these insights is discussed further below. 270 Results and implications of regression models.

[0086] Experimental results

[0087] This section describes the SOH of DCIR based on Class 1 cells. 270The results of the linear regression model were examined, and patterns of input-output correlation and highly correlated DCIR terms were investigated under conditions of rapid degradation. This analysis aims to further elucidate why DCIR serves as a robust input variable under conditions of rapid degradation during EV operation.

[0088] Regression results

[0089] Figure 8 Table 3 shown depicts the SOH 270 The average regression performance of the linear regression model on five random seeds, as a function of the l1 ratio, was obtained using RHI-based DCIR measurements from Class 1 cells in each RPT cycle.

[0090] A significant change in test set regression performance with l1 ratio was observed at RPT=0 iterations. Conversely, for RPT=90, 180, and 270 iterations, performance remained relatively consistent regardless of the l1 ratio, with these later iterations generally outperforming the initial RPT=0 iterations in the test set results.

[0091] Based on these results, instances with the lowest test set RMSE are identified and analyzed, taking into account the number of variables used (affected by the l1 ratio). For example, SOH caused by DCIR from cycle 0. 270 The test set RMSE was approximately 2.26%, with an average of only 11.0 DCIR variables (out of 79) used as effective inputs. Although the significantly smaller l1 ratio (l1=0.9) was optimal in the 180-cycle case compared to the 0-cycle case, an average of 24 DCIR variables were utilized in the 180-cycle case. This indicates that the EN model did not effectively capture relevant DCIR values ​​in the initial 0-cycle scenario, suggesting a need for further investigation into the predictive power of early-cycle DCIR measurements.

[0092] On the other hand, using DCIR from the 90th, 180th, and 270th cycle points significantly improved SOH on all datasets. 270 Predictive performance, particularly achieving mean errors of only 1.25%, 1.30%, and 1.00% respectively on the test set. Notably, for the best-case scenario, EN assigned non-zero weights to 72.50, 24.00, and 75.75 DCIRs on average in each corresponding RPT cycle. This is consistent with the results of the variable analysis, which showed that a large number of DCIRs exhibited similar performance to SOH after the cyclic process. 270 Its linearity and monotonicity.

[0093] In terms of all L1 ratios and overall regression performance considering the selected input variables, the models based on 90, 180, and 270-cycle DCIR significantly outperformed the 0-cycle model, thereby improving prediction accuracy by approximately 50% or more while utilizing an average of twice the DCIR value. This highlights the increased predictive power and relevance of DCIR measurements taken in later cycles when accurately estimating battery health.

[0094] Compared to a 2D convolutional neural network (CNN) model using VIT data from Lee et al., using SOH 270 The model performing on the first 90 cycles of the data yielded a mean test set RMSE of 7.52% without oversampling and 1.73% with oversampling. Conversely, the model utilizing only the DCIR measured at the 90th cycle as the input variable demonstrated superior performance, achieving a mean test set RMSE of 1.29%, representing a 25.4% improvement compared to the oversampling case. Furthermore, the simpler structure of this model, unlike more complex 2D CNN models, highlights its feasibility for integration into automotive BMS. These results underscore that while Class 1 cells exhibiting rapid and non-linear degradation in SOH typically require complex modeling methods, especially when data availability is limited, it is possible to identify cells exhibiting rapid and non-linear degradation in SOH. 270 Variables with high linearity and monotonic correlation were used to construct an efficient state of harmonics (SOH) for these cells. 270 Predictive Modeling. This approach allows for the development of satisfactory predictive models using a minimal number of battery cells, thereby simplifying the modeling process and enhancing its applicability in real-world scenarios.

[0095] Correlation analysis

[0096] This section provides a detailed analysis of the DCIR values ​​as an input variable for cells undergoing rapid degradation, and uses Pearson and Spearman correlation analysis to compare DCIR with SOH from each RPT cycle. 270 The linear correlation between them. Subsequently, this section compares the performance of cells across the two classes with SOH. 270 The highly correlated DCIR pattern is analyzed, and a method for diagnosing abrupt capacity decay based on changes in the DCIR pattern is proposed.

[0097] Figure 9 Table 4 shows the DCIR and SOH measured for Class 0 and Class 1 cells in each 90-cycle interval. 270 The correlation between them was calculated. Pearson and Spearman correlations were calculated to assess the SOH. 270 And the linearity and monotonicity of DCIR. Table 4 shows the relationship with SOH. 270The number of DCIRs exhibiting strong linear (Pearson) and monotonic (Spearman) correlations, with coefficients exceeding 0.5, is shown. For Class 0 cells, regardless of RPT cycles, DCIR consistently shows a correlation with SOH. 270 The negligible linearity or monotonicity suggests that constructing a robust linear regression model based solely on DCIR measurements to predict SOH in these cells is not feasible. 270 This will be challenging. Conversely, for Class 1 cells, while the DCIR measured during the initial 0-cycle RPT (before the start of cycling) has little significance for lifetime prediction, the DCIR obtained after 90, 180, and 270 cycles shows similarities to SOH. 270 The strong linearity and monotonicity of the DCIR are evident. Over 50% of the 79 measured DCIRs were identified as significant features (Pearson or Spearman coefficients > 0.5). This substantial correlation explains the effectiveness of the DCIR-based EN model in predicting the SOH of Class 1 cells. 270 The effective performance of DCIR highlights its importance as a predictor in scenarios involving rapid degradation.

[0098] To further analyze the importance of DCIR, the top 20 DCIRs were identified, defined by having Pearson and Spearman coefficients higher than 0.5 after cycling (hereinafter referred to as the top 20 DCIRs). First, the state of charge (SOC) distribution of these top 20 DCIRs was analyzed, such as... Figure 10 As shown in Table 5, the SOC measurements of these important DCIRs show very small variations, indicating that differences in SOC do not significantly affect the identification of important DCIRs. This observation is consistent with existing knowledge that the most critical DCIR term R in Class 1 cells is... ct There was no significant change within the SOC range of 20% to 80%.

[0099] In terms of measurement timescale, 19 of the first 20 DCIRs exceeded the initial 0.1-second (Pearson) measurement. Specifically, excluding the 0.1-second DCIR measurements, 7 had a total measurement time of 10 seconds, and 5 lasted for 30 seconds. Similarly, based on the Spearman coefficient, 18 of the first 20 DCIRs excluded the initial 0.1-second measurement, of which 6 each had a total measurement time of 10 seconds and 30 seconds.

[0100] This analysis supports an understanding of R. oThe prediction of SOH for Class 1 cells was not particularly relevant. Furthermore, the total time of 10 seconds and 30 seconds after subtracting the initial 0.1 seconds was identified as the most informative timescale for measuring DCIR. Notably, the 30-second measurement, not included in the short-pulse test (SOC=50%) and occurring only six times per RPT, was represented by the Pearson coefficient five times among the top 20 variables, and by the Spearman coefficient for all six. This suggests that measuring DCIR within a 30-second period may be particularly valuable, as it potentially captures R... ct and R p Both provide a more comprehensive understanding of the resistance that affects battery health.

[0101] Analysis of DCIR mode under various operating conditions

[0102] This section analyzes the DCIR measured under various operating conditions in different RPT cycles, compares the average DCIR values ​​of the corresponding cells, and discusses their consistency with insights from previous studies and their implications for understanding LIB capacity degradation. Figure 11 Table 6 shows the top 20 DCIRs categorized by measurement duration, where values ​​are adjusted by subtracting the initial 0.1s DCIR to isolate contributions beyond ohms. The most relevant timescales for the life characteristics of Class 1 cells are those measured at 30-0.1 seconds and 10-0.1 seconds. Generally, selecting an appropriate measurement timescale for DCIR is crucial for effectively utilizing this data for battery diagnostics.

[0103] This analysis focuses on DCIR patterns at 30–0.1 and 10–0.1 seconds under various operating conditions and explores how these specific timescales relate to SOH. 270 The data were correlated and their usability in diagnosing rapid degradation was evaluated. Additionally, data from short-pulse tests at 50% SOC were excluded because utilizing five consecutive short-pulse tests in real-world EV scenarios is likely impractical, and these measurements are correlated with SOH. 270 The correlation is poor.

[0104] To address the variability of DCIR values ​​under different operating conditions and measurement time scales, a normalization term Ratio is introduced. avg This metric helps to fairly compare the magnitude of average DCIR values ​​under different conditions. Figure 12 Ratio is shown avgThe calculation involves, for example, analyzing two IR values ​​under three operating conditions (in practice, using 12 DCIR values ​​measured on timescales of 30–0.1 seconds and 10–0.1 seconds), first determining the maximum value in each set of DCIRs, and then calculating the ratio of each DCIR to that maximum value under each operating condition. IR1 Ratio IR2 Then calculate the average of these ratios under all conditions (Ratio). IR And expressed as Ratio avg It is used as a standardized measurement to compare DCIR under different operating conditions.

[0105] Use Ratio avg As a baseline, the following discussion examines the DCIR patterns related to operating conditions and their implications for capacity degradation, particularly under normal pressure scenarios. This discussion involves observing the operating conditions of some Class 1 cells under normal pressure scenarios, examining how their resistance distribution deviates from typical patterns, and discussing how the DCIR patterns shift under high-pressure conditions, where Class 1 cells exhibit a higher proportion. These patterns can potentially serve as early indicators of rapid capacity degradation. This discussion helps to clarify why Class 1 cells exhibit similar characteristics to SOH (State of Harmonic Hiatus). 270 The highly linear and monotonic relationship highlights the diagnostic potential of DCIR in predicting rapid degradation events in lithium-ion batteries.

[0106] Normal pressure conditions

[0107] Figure 13a and Figure 13b It shows the way Figure 13a ambient temperature (T) amb ) and through Figure 13b The charging mode in the battery, and the ratio in the normal pressure battery under different operating conditions. avg Comparison with the decrease in average SOH. Figure 13a and Figure 13b This shows the ratio for each RPT cycle under normal stress conditions where Class 1 cells are almost nonexistent. avg And capacity decay (100-SOH (%)). Figure 13a and Figure 13b It also highlighted the key operating conditions T amb The changes in charging modes are discussed in order to illustrate the general modes based on operating conditions.

[0108] From T amb Initially, the increase in resistance and the decrease in capacity were observed to exhibit almost opposite trends. Typically, resistance increases with T due to the decrease in ion mobility within the battery. amb The capacity decreases and increases; at the same time, the capacity decays with Tamb The decrease is exacerbated, mainly due to the more pronounced lithium coating. Conversely, the capacity decay associated with solid electrolyte interface (SEI) formation increases with T. amb It increases with increasing. This complex interaction makes purely T-based... amb Determining a direct relationship between resistance change and capacity decay is challenging. It is noteworthy that in the dataset, within the normal pressure group, T... amb At 10°C, the sharp capacity decay caused by the lithium plating is minimized. However, most Class 1 cells (6 out of 8 under normal stress conditions) exhibit this degradation at T... amb Significant capacity decay was observed at 45°C, indicating that this temperature setting resulted in the highest average capacity decay during the last 90 cycles. This observation is consistent with the idea that SEI formation is exacerbated at higher temperatures (primarily occurring during the early cycles). In the initial 90 cycles, at T... amb =10℃ and T amb =Ratio exists between 25℃ avg Significant differences; however, T amb =25℃ and T amb The difference between 45°C and 45°C is negligible. As the cycle continues, Ratio... avg In T amb =25℃ and T amb The widening difference between T and 45°C indicates that during the initial 90 cycles... amb The deviation from the typical pattern at 45°C may indicate potential further capacity degradation.

[0109] Regarding charging modes, the relationship between resistance rise and capacity decay was not shown using T. amb A clear opposite trend was observed. The increase in resistance was more pronounced in 'Power' mode compared to 'Normal' and 'Lifetime' modes, indicating that the charging mode has a significantly smaller impact on resistance than T. amb Regarding capacity degradation, the fast-charge modes ('Normal' and 'Power') showed more degradation than the 'Lifetime' mode. Specifically, during the first 180 cycles, the 'Normal' mode cells exhibited more capacity degradation, but in the last 90 cycles, the 'Power' mode cells showed greater degradation, potentially indicating impending rapid degradation. This mode suggests a higher resistance ratio during 0-180 cycles. avg The 'power' mode cells have already shown early signs of rapid capacity degradation in the future.

[0110] This analysis examines changes in the resistance mode and their correlation with capacity decay under specific conditions: Condition 1, at T amb =45℃ 'Power' mode, and condition 2, in T amb=10℃ 'normal' mode. These conditions represent typical operating scenarios observed under normal and high-voltage conditions, where Class 1 cells were primarily found. Each scenario is compared with T... amb A comparison was made between the high-voltage conditions in 'power' mode at 45°C (where no Class 1 cells were present) to analyze how resistance changes under similar conditions and its implications for capacity decay.

[0111] T was discussed amb The impact at each pressure setting. Figure 14a and Figure 14b This shows the ratio under pressure changes. avg Compared with the average decrease in SOH, it is concentrated in T amb =10℃ and T amb Battery in 'power' mode at 45°C. This comparison is based on a limited set of operating conditions in a high-voltage scenario. Figure 14a This illustrates a normal pressure scenario, indicating that at lower temperatures (T... amb The higher resistance at T (=10℃) is attributed to reduced ion mobility, a natural cyclic phenomenon unrelated to capacity decay caused by side reactions. Therefore, at T amb Throughout the entire cycle at 10℃, Ratio avg 2.3% higher 5.9%, while capacity decay shows the opposite trend.

[0112] on the contrary, Figure 14b This shows that for high voltage conditions, the resistance rises and the capacitance decays at T. amb The temperature is higher at 10°C. It's worth noting that the ratio... avg In T amb =10℃ and T amb The difference between 45℃ and 35℃ has widened significantly to 3.6%. The 8.6% deviation suggests that this deviation may be due to acute side effects rather than just operating conditions. The significant increase in DCIR during 0 to 180 cycles, coupled with significant capacity decay during 180 to 270 cycles, indicates that DCIR is a valuable input for diagnosing potential future sharp capacity decay rather than simply estimating the current cycle's SOH.

[0113] High voltage condition - charging mode effect

[0114] This section explores the impact of charging modes on cell performance under different pressure conditions. Figure 14a and Figure 14b The Ratio in response to pressure changes is shown. avg The average decrease in SOH is concentrated in T amb A battery in 'Power' and 'Normal' modes at 45°C. Figure 14aIt shows the normal pressure condition, while Figure 14b A high-voltage scenario is shown, taking into account the limited operating conditions under high-voltage conditions.

[0115] Under normal stress, 'power' mode cells, which typically operate at a higher c-rate over a wider SOC range, exhibit a greater increase in resistance compared to 'normal' mode cells. However, the difference in capacity degradation between the two modes is not significant. This suggests that while 'power' mode may cause higher resistance due to increased stress on the battery, it does not necessarily translate into a correspondingly higher rate of capacity degradation.

[0116] Under high-pressure conditions, a significant reversal of the DCIR pattern occurs starting from the 90-cycle mark. This shift can indicate a sharp decline in side reactions, deviating from the typical pattern observed in 'normal' mode cells before the 90-cycle mark, leading to significantly greater capacity decay in these cells. Notably, capacity decay in 'normal' mode cells shows a temporary stabilization between 90 and 180 cycles, followed by rapid degradation within 180 to 270 cycles. This pattern suggests that the predictability of DCIR remains reliable, as the resistance in 'normal' mode is consistently higher than that in 'power' mode since the beginning of this phase. These observations highlight the complex interactions between charging mode, pressure conditions, and thermal environment influencing the electrochemical behavior and degradation patterns of lithium-ion cells. The varying effects on resistance and capacity decay underscore the importance of considering these factors when assessing battery health and lifespan in practical applications.

[0117] This analysis explored Class 1 cells exhibited prior to the final cycle (particularly those involving R). ct How do those items' DCIR measurements behave with SOH? 270 The cells exhibit high linearity and monotonicity. In scenarios dominated by Class 0 cells, such as under normal pressure conditions, typical patterns of resistance rise are analyzed based on operating conditions, and it is noted that these are generally not closely related to capacity decay. However, for Class 1 cells, deviations from these typical patterns are observed, suggesting that such variations can indicate an increase in resistance due to rapid side reactions.

[0118] Of particular note is the inversion or increase in DCIR gap within the pattern itself, which, depending on operating conditions, is closely related to rapid degradation. Since these changes manifest before actual capacity degradation becomes apparent, it demonstrates the feasibility of using only DCIR measurements to diagnose rapidly deteriorating cells and predict their lifespan characteristics. This approach highlights the predictive power of DCIR as a tool for early detection of potential faults and estimation of battery life under varying operating stresses.

[0119] Discussion of experimental results

[0120] Therefore, this system and method advantageously deploy a lifetime regression model for rapidly degrading cells (Class 1 cells) using DCIR, a metric that can be easily measured during the charge-discharge process of battery cells in EVs and other applications. Considering the rarity of rapidly degrading cells during typical EV operation and the impracticality of equipping automotive BMSs with complex neural network-based models, this system and method opt for a simpler regression model using EN. This approach demonstrates superior performance compared to neural network models using VIT data, where many DCIR variables play a significant role.

[0121] This disclosure focuses on DCIR measurement and last-cycle health status (SOH). 270 The linearity and monotonicity between (R0 and R0). It is worth noting that, based on charge transfer resistance (R0... ct The terms for Class 1 cells show very high linearity and monotonicity, highlighting the need to exclude ohmic resistance (R) when diagnosing rapidly deteriorating cells and predicting their lifespan. o The importance of ).

[0122] This disclosure also examines R-based systems across different RPT cycles. ct The study investigated the changes in resistance over time and correlated these changes with capacity under various operating conditions. For Class 0 cells (slow degradation), there was no discernible relationship between resistance increase and capacity decay over time, and distinguishing between resistance changes due to operating conditions and those due to side reactions was challenging. However, Class 1 cells (rapid degradation) exhibited a strong correlation between resistance increase and capacity decay, thus confirming the possibility of diagnosing rapid degradation and predicting future lifespan based on deviations from typical patterns under specific operating conditions.

[0123] Methods for rapid diagnosis of degraded battery cells using DCIR

[0124] Figure 15 This is a flowchart illustrating a method 100 for diagnosing rapidly deteriorating battery cells using DCIR according to one aspect of this disclosure. Method 100 can be performed on one or more battery cells in a stationary battery system (such as a battery pack or module or a battery energy storage system (BESS) including a battery rack), an electric vehicle (e.g., an electric car, truck, boat, or aircraft), or other electronic device (e.g., a mobile phone or laptop computer). Method 100 can be a computer-implemented method executed by a controller (e.g., a computing device).

[0125] In step 102, a direct current (DC) pulse can be applied to the battery cell during at least one of the charging and discharging processes. For example, a DC pulse can be applied to the battery cell during the discharging process by performing an HPPC test. The HPPC test may include a test lasting 100 seconds (long pulse test) and 20 seconds (short pulse test). An example of the HPPC test procedure is shown in... Figure 6 The diagram shows that, at different SOCs including 80%, 50%, and 20%, discharge currents of 2.5C and 1C were applied in two different pulses (see also the section on...). Figure 1 (See Table 1 for description). In some cases, the applied discharge current may include two or more of 0.1C, 0.2C, 0.3C, 0.4C, 0.5C, 0.6C, 0.7C, 0.8C, 0.9C, 1.0C, 1.5C, 2.0C, 2.5C, and 3.0C, but this disclosure is not limited thereto. In some cases, pulses are applied at a State of Charge (SOC) of 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 100%, but this disclosure is not limited thereto. In some cases, five 20-second pulse tests may be performed consecutively at a 50% SOC. In some cases, DC pulses may be applied at or after 0, 90, 180, or 270 charge-discharge cycles, but this disclosure is not limited thereto.

[0126] In step 104, the voltage change over time (dV / dt) is measured in response to the applied DC pulse to determine the DCIR. For example, Figure 7a The diagram shows the voltage variation over time and occurrence time of each resistance component when a long pulsed current is applied for 100 seconds. When a pulsed current is applied, the voltage variation is primarily due to the ohmic resistance (R). o A rapid voltage drop was observed, and the ohmic resistance (R) o This rapid voltage drop occurs within the first few microseconds (e.g., <0.1 seconds) and typically represents the most significant drop observed throughout the entire HPPC test. This rapid voltage drop is followed by a decrease in charge transfer resistance (R0). ct The associated voltage drop originates from the electrochemical reaction at the electrode-electrolyte interface. After tens of seconds, the polarization resistance (R) p The limitations of mass transfer become apparent.

[0127] In step 106, the ohmic resistance (R) can be eliminated from DCIR. o ) component to separate the charge transfer resistance (R) of DCIR ct ) components. For example, Figure 7bThe diagram shows the resistance components extracted at various timescales for each pulse test, along with the proportion of each resistance component. Five different timescales (t) are established. s To capture R at different sizes o R ct and R p : 0.1 seconds, 1 second, 10 seconds, 30 seconds, and 100 seconds. To reduce R... o The main effect on the total voltage drop was determined by measuring the DCIR values ​​(representing R) at 1 second, 10 seconds, 30 seconds, and 100 seconds. ct and R p Subtract the DCIR value measured at 0.1 seconds (representing R) o To determine the additional DCIR value.

[0128] In step 108, by predicting the SOH of the battery cell indicating rapid cell degradation, it is possible to base it on R ct The component diagnoses the battery cell as rapidly deteriorating (e.g., "Class 1" cell). For example, R ct The components can be used as feature inputs into the classifier to predict the state of charge (SOH) of the battery cell at 180 charge-discharge cycles and 270 charge-discharge cycles.

[0129] A battery cell can be diagnosed as rapidly degraded when the difference between the predicted State of Health (SOH) at 270 charge-discharge cycles and the predicted SOH at 180 charge-discharge cycles is at least approximately 2.5%. While many cells exhibit their lowest degradation rates between 180 and 270 cycles, a significant increase in degradation rate is observed in some cells during the last 90 cycles. For example, as discussed regarding... Figure 4 As described, cells that do not degrade rapidly (e.g., “Class 0” cells) exhibit an almost linear degradation pattern. In contrast, cells that degrade rapidly (e.g., “Class 1” cells) show a significant acceleration in degradation, especially in the last 90 cycles, thus indicating a distinct ‘inflection point’ in the degradation curve.

[0130] In some cases, the classifier is a linear regression model employing Elastic Network (EN) regression, characterized by its ability to utilize the linear relationship between input and output variables—a linear sparse model. For example, see [link to article on Elastic Network (EN) regression]. Figures 8 to 11 The description is as follows. In some cases, the classifier is based on a machine learning model, such as a neural network model, support vector machine (SVM), random forest classifier, gradient boosting model, or other supervised learning algorithm. These models can be trained on historical battery cycling data to learn R... ct With SOH 180 and SOH 270The choice of classifier type can depend on factors such as the size of the available training data, the computational resources of the battery management system, and the required prediction accuracy.

[0131] A system for rapid diagnosis of degraded battery cells using DCIR

[0132] Figure 16 This is a diagram illustrating a system 200 for diagnosing rapidly deteriorating battery cells using DCIR, according to one aspect of this disclosure. System 200 may include a controller 202, which can implement... Figure 15 Method 100 is described. In some cases, controller 202 may be integrated with BMS 208, which manages the charging, discharging, and monitoring of one or more battery cells 210. In some cases, controller 202 and BMS 208 are a single device. In some cases, controller 202 may include BMS 208, or vice versa. In some cases, controller 202 may be external to BMS 208 and may communicate with BMS 208 via circuitry, wired connection, or wireless connection.

[0133] The controller 202 may include one or more processors 204 (i.e., processing modules) configured to execute program instructions stored in memory 206 (i.e., memory modules). In this respect, one or more processors 204 of the controller 202 may execute any of the various models, methods, processes, steps, or algorithms described throughout this disclosure, for example, relative to... Figure 15 Steps 102 to 108 of method 100 are described. For example, controller 202 can apply a DC pulse to battery cell 210 by instructing BMS 208 to apply a DC pulse (e.g., using a current regulator to set a constant current). Controller 202 can measure the change in voltage over time (dV / dt) and determine DCIR by instructing BMS 208 to measure the change in voltage over time (e.g., using a voltage sensor). Controller 202 can isolate the R of DCIR. ct Components and exclude ohmic resistance (R) o ) components, for example, by means of later time scales (such as 10 seconds and 30 seconds (R ct and R p Subtract the value in the 0.1-second timescale (R) from the determined DCIR. o The controller 202 can determine the DCIR by predicting the SOH of battery cells that indicate rapid cell degradation, for example, by using R ct The components are used as feature inputs to the classifier based on R. ctThe component diagnoses battery cell 210 as rapidly deteriorating (e.g., a "Class 1" cell). Furthermore, controller 202 can be configured to receive data, including but not limited to voltage over time, current over time, SOC, SOH, cell temperature, cell pressure, C-rate, charging mode, charge-discharge cycle count, power output, charging duration, discharging duration, and pulse duration data.

[0134] Controller 202 (i.e., computing device) may include a desktop computer, mainframe computer system, workstation, server computer, embedded system, graphics computer, parallel processor, mobile device, tablet computer, head-mounted display, wearable computer, or any other computer system (e.g., networked computer). One or more processors 204 of controller 202 may include any processing element known in the art. In this sense, one or more processors 204 may include any microprocessor-type device configured to execute algorithms and / or instructions, such as application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), parallel processor, graphics processing unit (GPU), central processing unit (CPU), microcontroller unit (MCU), digital signal processor (DSP), system-on-a-chip (SoC) processor, programmable logic controller (PLC), logic circuit, electronic processor, or other chipset. It should be further appreciated that the term "processor" may be broadly defined to cover any device having one or more processing elements that execute program instructions from non-transitory memory 206. Furthermore, the steps described throughout this disclosure may be performed by a single controller 202 or, alternatively, by multiple controllers. For example, controller 202 and BMS 208 may be the same controller or multiple controllers. Additionally, controller 202 may include one or more controllers arranged in a common housing or multiple housings. In this way, any controller or combination of controllers can be individually packaged as a module suitable for integration into system 200.

[0135] Memory 206 may include any storage medium known in the art suitable for storing program instructions executable by one or more associated processors 204. For example, memory 206 may include non-transitory memory media. By another example, memory 206 may include (but is not limited to) read-only memory, random access memory, magnetic or optical storage devices (e.g., disks), magnetic tape, solid-state drives, etc. It should also be noted that memory 206 may be housed together with processor 204 in a common controller housing. In some cases, memory 206 may be remotely located relative to the physical location of processor 204 and controller 202. For example, one or more processors 204 of controller 202 may access remote memory (e.g., a server or cloud) accessible via a network (e.g., the Internet, intranet, and the like).

[0136] The systems and / or methods of this disclosure can be implemented as computer programs stored in memory 206. Any data, information, models, metrics, graphs, statistics, inputs, outputs, values, variables, or parameters described in this disclosure can be stored in memory 206. Computer programs (also referred to as programs, program instructions, software, software applications, scripts, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as standalone programs or as modules, components, subroutines, objects, or other units suitable for use in a computing environment. Computer programs may, but do not necessarily, correspond to files in a file system. Programs can be stored as a portion of a file containing other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinating files (e.g., a file storing one or more modules, subroutines, or code portions). Computer programs can be deployed to execute on a single controller or on multiple controllers located at a single site or distributed across multiple sites and interconnected via a communication network.

[0137] To provide interaction with the user, embodiments of the subject matter described herein can be displayed on a user interface (UI), such as a display device like a CRT (cathode ray tube), LCD (liquid crystal display), LED (light-emitting diode), OLED (organic light-emitting diode), micro LED, mini LED, plasma, DLP (digital light processing), QLED (quantum dot LED), or an electronic ink monitor for displaying information to the user. Input devices such as keyboards and / or pointing devices (e.g., mouse, touchpad, touchscreen, capacitive touchscreen, resistive touchscreen, stylus, digital pen, game controller, control panel, joystick, gamepad, remote control, gesture control device, eye-tracking device, facial recognition system, and / or trackball) can be used to provide input to the computer. Input can also be provided via wearable devices such as smart glasses, smartwatches, VR and / or AR goggles, rings, or other body-worn sensors. While visual or graphical user interfaces are described herein, this disclosure contemplates screen readers or voice control systems and other auditory systems as user interfaces, including natural language processing systems, speech recognition systems, and / or text-to-speech systems. This disclosure also envisions multimodal interactions that combine two or more input methods simultaneously or sequentially, such as touch input combined with gesture control.

[0138] In some cases, the battery cell 210 may have one or more of the following battery types (i.e., chemical properties): lithium-ion, lithium nickel manganese cobalt (NMC), lithium iron phosphate (LFP), lithium sulfur (Li-S), solid-state, sodium-ion (e.g., sodium sulfur (NaS), sodium nickel chloride), potassium-ion, aluminum-ion, lead-acid, silicon anode, or combinations thereof, but this disclosure is not limited thereto. Other battery types may include lithium titanate (LTO), magnesium-ion, calcium-ion, organic radical batteries, and solid polymer electrolyte batteries.

[0139] Figure 17 This is a diagram illustrating the integration of system 200 300 in various applications according to one aspect of this disclosure.

[0140] In some cases, system 200 can be integrated with electric vehicle 302. Electric vehicle 302 can be, for example, a car, truck, semi-truck, van, bus, train, subway, trolleybus, tram, forklift, scooter, bicycle, motorcycle, watercraft, small boat, ship, ferry, aircraft, fixed-wing aircraft, helicopter, drone, robot, construction vehicle, or other electric vehicle. For example, when electric vehicle 302 is a car, method 100 can be implemented during the discharge process, such as when driving or when supplying power to the grid using a vehicle-to-grid (V2G) system, or during the charging process.

[0141] In some cases, system 200 can be integrated with stationary battery system 304, such as a battery energy storage system (BESS). A BESS may include multiple battery racks, each comprising multiple battery modules or battery packs. A suitable exemplary battery rack may be, for example, the TR1300 (model ERT5422CN201) manufactured by LG Energy Solution.

[0142] In some cases, system 200 may be integrated with electronic devices 306, such as cellular phones, laptops, tablets, smartwatches, fitness trackers, smart glasses, augmented reality headsets, virtual reality headsets, game consoles, portable gaming devices, digital cameras, camcorders, smart speakers, smart displays, smart home devices, wireless earbuds, wireless headphones, power tools, smart appliances, uninterruptible power supplies (UPS), energy storage systems, or other battery-powered electronic devices.

[0143] In some cases, in response to diagnosing battery cell 210 as rapidly deteriorating (e.g., "Class 1"), an alarm indicating rapid deterioration at the corresponding battery rack can be generated. In some cases, the alarm can be displayed on a display or screen (e.g., an LCD monitor). The alarm may include information identifying the corresponding battery cell and may be sent to the user (e.g., a computing device, cellular phone, or pager) via email, SMS, pager, or other notifications.

[0144] In some cases, the alarm includes instructions to take action to mitigate or replace rapidly deteriorating battery cells. In some cases, the action may be automated, causing it to execute automatically. In some cases, the action may be performed manually (e.g., by one or more members of the maintenance team). In some cases, the action may include at least one of the following: checking battery cell connections, disconnecting battery cells, discharging battery cells, or replacing battery cells.

[0145] The present disclosure has been described in more detail above with reference to the accompanying drawings and aspects. However, the configurations described in the drawings or the aspects in the specification are merely aspects of the present disclosure and do not represent all the technical ideas of the present disclosure. Therefore, it should be understood that various equivalent substitutions and variations covered by the claims may exist in place of them at the time of filing this application.

[0146] (See attached image labels)

[0147] 200: System

[0148] 202: Controller

[0149] 204: Processor

[0150] 206: Memory

[0151] 100: Method

[0152] 208: Battery Management System (BMS)

[0153] 210: Battery cell

[0154] 300: Systems integrated with various application areas

[0155] 302: Electric vehicles

[0156] 304: Fixed Battery System

[0157] 306: Electronic devices

Claims

1. A system for diagnosing rapidly deteriorating battery cells, the system comprising: The controller includes a processor and a non-transitory memory storing computation instructions, which, when executed by the processor, are configured to perform the following steps: (a) Applying a direct current (DC) pulse to the battery cell during at least one period of the charging and discharging process; (b) Measure the change in voltage over time in response to an applied DC pulse to determine the DC internal resistance (DCIR). (c) By removing the ohmic resistance (R) from the DCIR o ) component to separate the charge transfer resistance (R) of the DCIR. ct ) portion; as well as (d) By predicting at least one state of health (SOH) of the battery cell indicating rapid cell degradation, based on the R ct The component diagnosed the battery cell as rapidly deteriorating.

2. The system according to claim 1, wherein, Step (d) includes the following steps: R ct The components are input as features into a classifier to predict at least one state of harm (SOH) of the battery cell at 180 charge-discharge cycles and 270 charge-discharge cycles; and The battery cell is diagnosed as rapidly deteriorating when the difference between the predicted SOH after 270 charge-discharge cycles and the predicted SOH after 180 charge-discharge cycles is at least 2.5%.

3. The system according to claim 2, wherein, The classifier is a linear regression model that implements Elastic Network (EN) regression.

4. The system according to claim 1, wherein, The controller is configured to determine the DCIR at multiple time scales of the DC pulse, the multiple time scales including at least one of 10 seconds or 30 seconds and 0.1 seconds. Step (c) includes subtracting the DCIR measured at 0.1 seconds from the DCIR measured at at least at 10 seconds or 30 seconds to exclude the R. o Components and separation of the R ct Quantity.

5. The system according to claim 1, wherein, The controller is configured to apply the DC pulse after one or more of the 0, 90, 180, or 270 charge-discharge cycles of the battery cell.

6. The system according to claim 1, wherein, The controller is configured to determine the DCIR at one or more time scales of the applied DC pulse, wherein the one or more time scales include 0.1 seconds, 1 second, 5 seconds, 10 seconds, 15 seconds, 20 seconds, 25 seconds, 30 seconds, or 60 seconds.

7. The system according to claim 1, wherein, The controller is configured to determine the DCIR at one or more states of charge (SOC) of the battery cell, wherein the one or more SOCs include 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%.

8. The system according to claim 1, wherein, The controller is configured to: The DC pulse is applied during a hybrid pulse power characteristic (HPPC) test at multiple states of charge (SOC); and multiple values ​​of the DCIR are determined from the HPPC test. The HPPC test includes applying the DC pulses having at least two different currents.

9. The system according to claim 8, wherein, The at least two different currents include at least two of 0.5C, 1C, 1.5C, 2.0C, or 2.5C.

10. The system according to claim 1, wherein, The battery cells are configured to power an electric vehicle, and the controller is integrated with the electric vehicle's battery management system (BMS).

11. A method for diagnosing rapidly deteriorating battery cells, the method comprising the following steps: (a) Applying a direct current (DC) pulse to the battery cell during at least one period of the charging and discharging process; (b) Measure the change in voltage over time in response to an applied DC pulse to determine the DC internal resistance (DCIR). (c) By removing the ohmic resistance (R) from the DCIR o ) component to separate the charge transfer resistance (R) of the DCIR. ct ) portion; as well as (d) By predicting at least one state of health (SOH) of the battery cell indicating rapid cell degradation, based on the R ct The component diagnosed the battery cell as rapidly deteriorating.

12. The method according to claim 11, wherein, Step (d) includes the following steps: R ct The components are input as features into a classifier to predict at least one state of harm (SOH) of the battery cell at 180 charge-discharge cycles and 270 charge-discharge cycles; and The battery cell is diagnosed as rapidly deteriorating when the difference between the predicted SOH after 270 charge-discharge cycles and the predicted SOH after 180 charge-discharge cycles is at least 2.5%.

13. The method according to claim 12, wherein, The classifier is a linear regression model that implements Elastic Network (EN) regression.

14. The method according to claim 11, wherein, The DCIR is determined at multiple time scales of the DC pulse, including at least one of 10 seconds or 30 seconds and 0.1 seconds. Step (c) includes subtracting the DCIR measured at 0.1 seconds from the DCIR measured at at least at 10 seconds or 30 seconds to exclude the R. o Components and separation of the R ct Quantity.

15. The method according to claim 11, wherein, The DC pulse is applied after one or more of the 0, 90, 180, or 270 charge-discharge cycles of the battery cell.

16. The method according to claim 11, wherein, The DCIR is determined at one or more time scales of the applied DC pulse, wherein the one or more time scales include 0.1 seconds, 1 second, 5 seconds, 10 seconds, 15 seconds, 20 seconds, 25 seconds, 30 seconds, or 60 seconds.

17. The method according to claim 11, wherein, The DCIR is determined at one or more states of charge (SOC) of the battery cell, wherein the one or more SOCs include 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%.

18. The method according to claim 11, wherein, The DC pulse is applied during a hybrid pulse power characteristic (HPPC) test at multiple states of charge (SOC); and multiple values ​​of the DCIR are determined from the HPPC test, wherein the HPPC test includes applying the DC pulse having at least two different currents.

19. A non-transitory computer-readable medium storing computer-executable instructions, the computer-executable instructions causing the controller to perform operations for diagnosing rapidly deteriorating battery cells in response to execution by a processor of a controller, the operations comprising the following steps: (a) Applying a direct current (DC) pulse to the battery cell during at least one period of the charging and discharging process; (b) Measure the change in voltage over time in response to an applied DC pulse to determine the DC internal resistance (DCIR). (c) By removing the ohmic resistance (R) from the DCIR o ) component to separate the charge transfer resistance (R) of the DCIR. ct ) portion; as well as (d) By predicting at least one state of health (SOH) of the battery cell indicating rapid cell degradation, based on the R ct The component diagnosed the battery cell as rapidly deteriorating.

20. The non-transitory computer-readable medium according to claim 19, wherein, Step (d) includes the following steps: R ct The components are input as features into a classifier to predict at least one state of harm (SOH) of the battery cell at 180 charge-discharge cycles and 270 charge-discharge cycles; and The battery cell is diagnosed as rapidly deteriorating when the difference between the predicted SOH after 270 charge-discharge cycles and the predicted SOH after 180 charge-discharge cycles is at least 2.5%.

Images