Battery multi-stress coupling accelerated life test and evaluation method under fast charging scenario

Abstract

The application belongs to the technical field of battery testing, and is used for solving the information lag in the prior art, which leads to sudden collapse of performance of the battery under stress impact in subsequent charge-discharge cycles, and makes the life prediction model based on macroscopic indicators produce serious deviation at key turning points. The application is a battery multi-stress coupling accelerated life test and evaluation method under fast charging scene, which comprises the following steps: obtaining a relaxation response signal collected in a relaxation stage after the Nth charge-discharge cycle of a target battery, wherein the relaxation response signal at least comprises a voltage time sequence; performing multi-scale fluctuation analysis on the voltage time sequence to extract a first fluctuation complexity feature for representing fluctuation disorder degree of the voltage time sequence at different time scales; and the application establishes a closed-loop feedback regulation mechanism based on comparison between a metastable state risk index and a dynamic risk threshold; and the system can adaptively reduce the charging stress of the next cycle according to the current metastable state level of the battery.

Description

Technical Field

[0001] This invention belongs to the field of battery testing technology, specifically a method for accelerated life testing and evaluation of batteries under fast charging scenarios with multi-stress coupling. Background Technology

[0002] With the rapid development of new energy vehicles and portable electronic devices, users' demand for charging speed is increasing, and fast charging technology has become an important development direction in the battery field. However, high-rate charging is often accompanied by complex coupling of multiple physical fields such as temperature, current, and mechanical stress, which accelerates the degradation of battery life. Therefore, how to accurately and efficiently evaluate the life characteristics of batteries under fast charging conditions has become a key issue in the field of battery testing.

[0003] Existing battery accelerated life testing methods typically employ cyclic charge-discharge cycles to simulate real-world usage conditions and assess battery health by monitoring macroscopic indicators such as capacity decay. In these tests, the relaxation phase (the period after charge-discharge cycles when the battery is at rest) is considered a "rest period." Current technologies generally only focus on the end-of-phase data of this phase, such as estimating the open-circuit voltage using the voltage at the end of the rest period, while ignoring the dynamic response during the relaxation process as noise or a transient state. This approach implicitly assumes that the relaxation process is a monotonic and predictable decay process, and its microscopic details do not contain valuable health information.

[0004] However, after a battery undergoes multiple stress couplings during the later stages of a fast-charging cycle, its internal electrochemical system and mechanical structure often undergo subtle, irreversible changes, such as the propagation of microcracks in the electrode material or the initial growth of lithium dendrites. These microstructural changes may not be immediately apparent in macroscopic capacity indicators, but they significantly affect the kinetics of the relaxation phase. Specifically, signals such as voltage and casing strain exhibit small, irregular fluctuations during relaxation, and the overall decay pattern deviates from the ideal exponential form. Existing testing methods, unable to detect and quantify the abnormal evolution of these microscopic dynamic characteristics during the relaxation phase, often give a normal assessment result even when the battery has entered a high-risk "metastable state." This information lag makes the battery highly susceptible to sudden performance collapse under stress shocks during subsequent charge-discharge cycles, causing severe deviations in lifetime prediction models based on macroscopic indicators at critical turning points.

[0005] Therefore, how to utilize the deep dynamic information contained in the relaxation stage to identify abnormal evolution trends of the battery's internal state before the macroscopic performance of the battery deteriorates, so as to improve the accuracy and safety of battery life testing under fast charging conditions, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a method for accelerated life testing and evaluation of batteries under multi-stress coupling in fast charging scenarios, in order to solve the problem that the existing technology has information lag, which makes the battery prone to sudden performance collapse under stress impact in subsequent charge and discharge cycles, causing the life prediction model based on macroscopic indicators to produce serious deviations at critical turning points.

[0007] The technical problem to be solved by this invention is: how to provide a method for accelerated life testing and evaluation of batteries under fast charging scenarios that can utilize the deep dynamic information contained in the relaxation stage to identify abnormal evolution trends of the internal state of the battery before the macroscopic performance deteriorates.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A method for accelerated life testing and evaluation of batteries under fast charging scenarios, applied to a battery testing system, includes the following steps:

[0010] Acquire the relaxation response signal of the target battery during the relaxation phase after the Nth charge-discharge cycle, wherein the relaxation response signal includes at least a voltage time series;

[0011] Multi-scale fluctuation analysis is performed on the voltage time series to extract the first fluctuation complexity feature, which characterizes the degree of disorder of fluctuations in the voltage time series at different time scales.

[0012] Relaxation kinetics analysis was performed on the voltage time series to extract the first relaxation time constant characteristic used to characterize the decay rate of the voltage time series;

[0013] Based on the change of the first fluctuation complexity feature relative to the preset benchmark complexity feature and the change of the first relaxation time constant feature relative to the preset benchmark time constant feature, the metastable risk index of the target battery after the Nth cycle is generated.

[0014] Based on the comparison between the metastable risk index and the dynamic risk threshold, the charging stress parameters of the target battery in the N+1th charge-discharge cycle are dynamically adjusted.

[0015] The present invention has the following beneficial effects:

[0016] 1. This invention breaks through the conventional thinking of traditional testing methods that regard the relaxation stage as a "quiet period". By using multi-scale sample entropy analysis, it extracts the fluctuation complexity features at different time scales from the voltage time series, transforming the tiny fluctuations that were originally ignored as noise into effective information for quantifying the degree of disorder in the battery's microstructure. This shift from "discarding data" to "mining features" enables the system to capture relaxation dynamic anomalies caused by early damage such as electrode microcrack propagation and lithium dendrite growth, providing a new data dimension for the refined assessment of battery health status.

[0017] 2. This invention constructs a metastable risk index as an intermediate mapping quantity by integrating the change in the comprehensive amplitude of multi-scale entropy vectors with the change in the fast and slow relaxation time constants. This index is not a simple list of multiple original features, but a weighted combination based on physical meaning, which maps microscopic dynamic information from different sources and with different physical meanings into a dimensionless risk assessment space. This dimensionality upgrade from "multi-dimensional data" to "single index" enables the system to quantify the degree of battery deviation from its healthy state in a concise and effective manner, providing a clear basis for subsequent decision control.

[0018] 3. This invention establishes a closed-loop feedback control mechanism based on the comparison of metastable risk indicators and dynamic risk thresholds. By updating historical statistical thresholds and risk over-limit ratio adjustment rules in real time, the system can adaptively reduce the charging stress of the next cycle according to the current metastable level of the battery. This shift from "passive monitoring" to "active control" achieves a dynamic balance between accelerated testing efficiency and battery safety protection, avoiding sudden battery failure caused by risk accumulation in traditional fixed stress testing, and significantly improving the reliability and continuity of the testing process.

[0019] 4. This invention aligns metastable risk indicators with capacity decay sequences using a dynamic time warping algorithm, constructing a remaining lifetime prediction model based on risk indicators. This model utilizes the time difference between risk indicators and capacity decay to transform microstructural damage signals into forward-looking predictions of macroscopic performance degradation. This leap from "post-analysis" to "pre-warning" enables the system to identify lifetime inflection points before battery capacity drops, providing a scientific basis for optimizing test schemes and defining the safety boundaries of fast-charging strategies. It effectively solves the technical problem of a sharp drop in accuracy in traditional lifetime prediction models during the nonlinear aging stage. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the method according to Embodiment 1 of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In accelerated life testing of batteries under fast charging scenarios with multi-stress coupling, accurately assessing the battery's true health status after repeated high-rate charge-discharge cycles is crucial for ensuring the test's effectiveness. From the perspective of the intrinsic operating mechanism of the electrochemical system, after the battery completes charging and discharging, it enters a relaxation phase. The decay process of its physical quantities such as voltage and strain is essentially a macroscopic manifestation of the redistribution and eventual equilibrium of internal electrochemical and mechanical potential energy. The dynamic characteristics of this relaxation behavior, including the complexity of the fluctuations and the rate of decay, directly reflect the integrity of the electrode material's microstructure, the unobstructedness of lithium-ion diffusion channels, and the activity level of interfacial side reactions.

[0024] However, existing testing methods generally treat the relaxation phase as a "quiet period" or "rest period" between charge-discharge cycles, focusing only on the steady-state open-circuit voltage at the end of relaxation and ignoring the dynamic response during relaxation as noise or a transient state. This approach implicitly assumes that the relaxation process is a monotonous and predictable exponential decay process, and its microscopic details do not contain valuable health information, thus not needing to be included in the evaluation system. But this assumption often no longer holds after the battery has undergone long-term fast-charge cycles.

[0025] As the number of cycles increases, multi-stress coupling induces a series of irreversible microstructural changes within the battery, such as microcracks in electrode particles, lithium dendrites on the negative electrode surface, and compressive deformation of the separator pore structure. Although these micro-damages do not yet manifest as significant capacity degradation in macroscopic indicators, they directly disrupt the thermodynamic equilibrium path during the relaxation phase: the propagation of microcracks leads to localized stress concentration and slow release, causing minute irregular fluctuations in voltage and strain signals; the growth of lithium dendrites alters the local current distribution, resulting in additional fluctuations in the charge redistribution process; and the obstruction of diffusion channels gradually increases the relaxation time constant. These dynamic anomalies caused by micro-damage essentially constitute the precursory "metastable state" signal of the battery transitioning from a healthy state to a failure state.

[0026] Unfortunately, existing testing technologies lack effective mining of dynamic information during the relaxation phase, making it impossible to detect these microscopic-scale abnormal evolutions. Even when the battery has entered a metastable state, the system may still conclude that the battery is in normal condition based on macroscopic indicators such as steady-state voltage and capacity. This information lag makes the battery highly susceptible to sudden performance collapse under stress during subsequent fast-charging cycles, causing life prediction models based on traditional methods to deviate significantly at critical turning points, or even fail completely.

[0027] If the aforementioned issues are not addressed, accelerated life testing of fast-charging batteries will continue to face two challenges: First, the testing system cannot identify risks in the early stages of irreversible battery damage, leading to premature battery failure in subsequent tests, disrupting the testing process and wasting resources. Second, due to a lack of ability to capture precursors to metastability, life prediction models based on macroscopic data cannot accurately fit the transition from slow degradation to rapid failure, rendering the predictions worthless in the latter half of the battery's lifespan. Consequently, the reliability of test results will be systematically reduced, affecting the efficiency of fast-charging battery R&D iterations and the accurate definition of safety boundaries.

[0028] Example 1: As Figure 1 As shown, a method for accelerated life testing and evaluation of batteries under fast charging scenarios, applied to a battery testing system, includes the following steps:

[0029] Step S1: Obtain the relaxation response signal of the target battery during the relaxation phase after the Nth charge-discharge cycle. The relaxation response signal includes at least the voltage time series.

[0030] The relaxation response signal also includes the shell strain time series and / or surface temperature time series; the first fluctuation complexity feature is specifically a multi-scale entropy vector obtained by multi-scale sample entropy analysis of the voltage time series, and the multi-scale entropy vector contains the sample entropy values ​​of the voltage time series under multiple preset scale factors.

[0031] Before performing any cyclic testing, the system first performs initial calibration of the target battery to establish a benchmark for all subsequent comparisons. Specifically, a brand-new battery under test is connected to the test system and subjected to three standard constant current constant voltage charge-discharge cycles at a rate of 0.5C under standard ambient temperature (e.g., 25°C). After each charge, the battery enters a relaxation phase of 30 minutes (T_relax). During this phase, the system synchronously acquires the battery's terminal voltage signal V_ref(t), the strain signal ε_ref(t) at the center point of the casing, and the surface temperature signal Θ_ref(t) at a sampling frequency of 100Hz. These signals together constitute the benchmark relaxation response signal in the initial state. Subsequently, the system performs multi-scale sample entropy analysis on the benchmark voltage signal V_ref(t) to calculate the benchmark multi-scale entropy vector C_ref. This vector contains sample entropy values ​​at 20 scales from scale factor τ=1 to τ=20, used to quantify the intrinsic fluctuation complexity of the battery's relaxation process in its new state. Simultaneously, a double exponential curve fitting was performed on the reference voltage signal to obtain the reference fast relaxation time constant τ_{1,ref} and the reference slow relaxation time constant τ_{2,ref}. These two constants characterize the recovery rate of lithium-ion diffusion and electrochemical reaction in a healthy battery state. Furthermore, the system calculates the comprehensive amplitude ‖C_ref‖2 of the reference multi-scale entropy vector, which serves as the reference complexity feature for subsequent comparisons. At this point, the preset reference complexity features (including ‖C_ref‖2) and preset reference time constant features (including τ_{1,ref} and τ_{2,ref}) are constructed and stored in the non-volatile memory of the test system.

[0032] After initial calibration, the system enters the formal accelerated life testing process. After the Nth charge-discharge cycle (N being an integer greater than or equal to 1), the battery automatically enters the relaxation phase. Here, a charge-discharge cycle refers to a complete fast charge and subsequent constant-current discharge process. The system automatically triggers the data acquisition module to synchronously acquire the terminal voltage signal V_n(t), casing strain signal ε_n(t), and surface temperature signal Θ_n(t) during this relaxation phase at the same 100Hz sampling frequency as during initial calibration. To ensure that all key dynamic features during relaxation are captured, the acquisition duration for the relaxation phase is also set to 30 minutes. Thus, the system obtains the relaxation response signal after the Nth cycle, which at least contains the voltage time series {V_n(t_i)}, i=1,2,...,180000. It should be noted that although multiphysics signals are acquired in this step, the subsequent core feature extraction steps will primarily analyze the voltage time series. This set of raw time series data acquired in real time will serve as direct input for multi-scale fluctuation analysis in step S2 and relaxation dynamics analysis in step S3, laying the data foundation for the subsequent generation of metastable risk indicators.

[0033] Step S2: Perform multi-scale fluctuation analysis on the voltage time series and extract the first fluctuation complexity feature to characterize the degree of disorder of fluctuations in the voltage time series at different time scales;

[0034] The steps for extracting the first wave complexity feature include:

[0035] Determine multiple preset scale factors;

[0036] For each scale factor, the voltage time series is coarsened to generate a coarse-grained series at that scale.

[0037] Calculate the sample entropy for each coarse-grained sequence to obtain the sample entropy value under that scale factor;

[0038] Combine the sample entropy values ​​under all scale factors to generate a multi-scale entropy vector.

[0039] In step S1, the voltage-time sequence of the Nth relaxation cycle was obtained. (Where N=180000, sampling interval Δt=0.01 seconds) After that, the core task of step S2 is to extract the first fluctuation complexity feature from the sequence through multi-scale fluctuation analysis, which can quantify the degree of disordered evolution of the battery's internal microstructure at different time scales. In this embodiment, this feature is specifically represented as a multi-scale entropy vector, the construction process of which follows the algorithm framework of multi-scale sample entropy. This algorithm can effectively distinguish between deterministic structures and random fluctuations in the signal, and is particularly suitable for analyzing the complex dynamics generated by electrochemical-mechanical coupling during battery relaxation.

[0040] Before performing the specific calculations, the system has pre-set all the parameters required for multi-scale sample entropy analysis. The selection of these parameters is based on a deep understanding of the physical characteristics of the relaxation process and a trade-off of the statistical robustness of the algorithm: the embedding dimension m is fixed at 2 because for the dynamic reconstruction of the relaxation signal, two-dimensional embedding is sufficient to capture its main time dependencies; the similarity tolerance r is taken as 0.2 times the standard deviation of the current analysis sequence, and this ratio can better balance the sensitivity to small fluctuations and the ability to resist noise interference; the maximum scale factor τmax is set to 20, and this value is calculated based on the total duration of the relaxation phase (30 minutes) and the basic sampling rate (100 Hz), such that the time window corresponding to the maximum coarse-graining scale is approximately 20×0.01×20; actually, coarse-graining is to perform average downsampling on the original sequence, and the scale factor τ means combining consecutive τ points into a new point, so the equivalent sampling interval corresponding to the maximum scale factor 20 is 0.2 seconds, which can cover the dynamic range from the millisecond level to the tens of seconds level.

[0041] The specific implementation process is as follows: First, the system traverses all scale factors from τ = 1 to τ = 20. For each scale factor τ, a coarse-graining transformation is performed on the original voltage sequence {Vn(ti)} to generate a coarse-grained sequence {yj (τ)}. This transformation is achieved through the following formula

[0042] ;

[0043] where j is the index of the coarse-grained sequence. When τ = 1, the coarse-grained sequence is the original sequence itself.

[0044] After obtaining the coarse-grained sequence {yj (τ)｝}, the system further calculates the sample entropy value of this sequence. The calculation of sample entropy does not depend on the model assumption of the data itself, but measures the regularity degree of the sequence by statistically matching the probability of templates. Let the sequence length be L = ⌊N / τ⌋, and first construct a set of m-dimensional vectors:

[0045] ;

[0046] For each k, define the distance between the vector uk and all other vectors u l (l ≠ k) as the maximum value of the differences between the corresponding elements of the two:

[0047] ;

[0048] Count the number of all k that satisfy d kl < r (denoted as B kThen, calculate its ratio to the total number (L−m), and average it over all k to obtain B. Next, increase the dimension to m+1, repeat the above process, construct an m+1 dimensional vector, and count the number A of vectors whose distance is less than r. k Calculate the average ratio A. Finally, the sample entropy at this scale is defined as:

[0049] ;

[0050] The larger the value, the more disordered and complex the fluctuations of the sequence at the corresponding scale.

[0051] To improve computational efficiency, the system employs a fast algorithm and pre-sets the similarity tolerance r to 0.2 times the standard deviation of the current coarse-grained sequence, i.e., r = 0.2 × std({y j (τ) The standard deviation varies with different scales, so the similarity tolerance is adaptively adjusted, which ensures the comparability of entropy values ​​at different scales.

[0052] After the above processing, a sample entropy value C is obtained for each scale factor τ. n,τ Combining the entropy values ​​from all 20 scales forms the multi-scale entropy vector for the Nth cycle:

[0053] ;

[0054] This vector is the first fluctuation complexity feature mentioned in the claims. It's worth noting that each component of the vector has a clear physical meaning: the entropy values ​​at small scales (e.g., τ=1,2,3) mainly reflect the millisecond-level voltage fluctuations caused by the rapid insertion / extraction of lithium ions on the electrode particle surface and local current density fluctuations; the entropy values ​​at medium scales (τ=4~10) are related to the establishment and dissipation of the concentration gradient inside the electrode; while the entropy values ​​at large scales (τ=11~20) are more likely associated with the slow structural evolution of the electrode material, such as mechanical relaxation and microcrack propagation. Through joint analysis of these entropy values ​​at different scales, the system can comprehensively capture early signs of microdynamic instability during the transition of the battery from a healthy state to a metastable state.

[0055] For example, in a real-world test, the multi-scale entropy vector calculated for a battery in the early stages of cycling (the 10th cycle) was C. 10 =[0.12, 0.15, 0.09, 0.08, ... ], with small and stable entropy values ​​at each scale; however, in the later stages of the cycle (the 800th cycle), this vector becomes C. 800=[0.28, 0.31, 0.22, 0.19, ... ], where the small-scale entropy increase exceeds 100% for τ=1 and 2, while the large-scale entropy increase is relatively small. This pattern suggests that the initial growth of lithium dendrites may lead to drastic fluctuations in the local current distribution. This vector will be fully saved and passed to step S5 for comparison with the baseline complexity feature to generate a metastable risk index.

[0056] Step S3: Perform relaxation dynamics analysis on the voltage time series to extract the first relaxation time constant feature used to characterize the decay rate of the voltage time series;

[0057] The steps for extracting the first relaxation time constant feature include:

[0058] The voltage time series is fitted with a double exponential curve to obtain the fast relaxation time constant and the slow relaxation time constant, which are then used as the first relaxation time constant feature.

[0059] Step S3 aims to extract the first relaxation time constant characteristic, which can characterize the recovery rate of the electrochemical reaction and lithium-ion diffusion process inside the battery, by performing kinetic analysis on the voltage time series during the relaxation phase. According to electrochemical theory, the relaxation process of the battery after the cessation of charging and discharging is not an exponential decay of a single time constant, but the result of the superposition of multiple physical mechanisms: the fast relaxation process mainly corresponds to the redistribution of charge on the electrode surface and the discharge of the electric double layer, which usually occurs in the millisecond to second range; the slow relaxation process corresponds to the concentration polarization dissipation of lithium ions inside the electrode particles and in the electrolyte, which can last for tens of seconds or even minutes. In order to accurately separate and quantify these two dominant processes, this embodiment uses a double exponential curve fitting model to analyze the voltage time series.

[0060] Before starting the fitting process, the system first obtains the Nth cycle relaxation voltage time series acquired in step S1. In the process, all data points from the relaxation start time to the relaxation end time are extracted. The time axis is set with the relaxation start time as t=0 and the corresponding sampling time as t. i =(i-1)×Δt, where Δt=0.01 seconds. Since the relaxation period may be affected by transient interference at the moment of charge and discharge cutoff, in actual processing, the data points of the first 0.5 seconds (i.e. the first 50 sampling points) can be omitted to eliminate switching transient noise. However, this preprocessing is not necessary and can be dynamically adjusted according to the signal quality.

[0061] The double-exponential model used for fitting is as follows:

[0062] ;

[0063] Where V(t) is the terminal voltage at time t; V ∞This represents the steady-state open-circuit voltage after relaxation; A1 and A2 are the amplitude coefficients for the fast and slow relaxation processes, respectively. Typically, A1 + A2 is approximately equal to the charging cutoff voltage and V. ∞ The difference between them; τ1 and τ2 are the fast relaxation time constant and slow relaxation time constant, respectively, and satisfy τ1<τ2.

[0064] To obtain reliable fitting results, the system employs nonlinear least squares method for parameter estimation, specifically the Levenberg-Marquardt algorithm. This algorithm combines the stability of gradient descent with the damped Gauss-Newton method, making it suitable for fitting exponential nonlinear functions. Before iterative solution, four unknown parameters (V...) need to be calculated. ∞ Set reasonable initial values ​​for A1, A2, τ1, and τ2. The selection of initial values ​​directly affects the convergence speed and the accuracy of the results. This embodiment adopts the following heuristic strategy:

[0065] The average voltage within the last 1% time window of the relaxation phase is taken as V. ∞ The initial estimate;

[0066] Subtract V from the original voltage sequence ∞ We obtain the pure relaxation component V′(t) = V(t) − V ∞ ;

[0067] Taking the natural logarithm of V′(t), the logarithmic curve is not strictly linear due to the superposition of double exponents. However, in the later stage (when t is large), slow relaxation dominates, and the logarithmic curve is approximately linear. The negative reciprocal of its slope can be used as the initial value of τ2, and the intercept can be used as the initial value of lnA2.

[0068] Subtract the estimated slow relaxation component A2e from V′(t) -t / τ2 We obtain the residual sequence dominated by fast relaxation, then take its logarithm and fit it to obtain the initial values ​​of τ1 and A1.

[0069] The above initial value estimation ensures that the algorithm can converge quickly.

[0070] After setting the initial values, the system constructs the objective function as the sum of squared residuals between the measured voltage and the model prediction, and iteratively updates the parameters until the convergence condition is met (e.g., the change in the sum of squared residuals between two consecutive iterations is less than the preset tolerance of 10). -6 It is worth noting that the fitting of the relaxation process needs to eliminate possible data anomalies. For example, when there are obvious measurement noise spikes in the voltage sequence, the system can use a robust weighting method based on the absolute deviation of the median to automatically reduce the weight of outliers during the iteration process.

[0071] After nonlinear least squares fitting, the system finally obtains the fast relaxation time constant τ1,n and slow relaxation time constant τ2,n of the Nth cycle, along with the amplitude coefficient A. 1,n A 2,n and steady-state voltage V ∞,n These time constants are output together. They form the basis for calculating the second and third relative changes in subsequent steps. For example, in a typical test, a battery fits τ at the 100th cycle. 1,n =12.3 seconds, τ 2,n =153.7 seconds, and τ at the 800th cycle 1,n Increased to 18.9 seconds, τ 2,n The time constant increased to 287.2 seconds. This significant increase in time constant indicates impaired lithium-ion diffusion within the battery or alterations in the electrode interface properties, which is a key indicator of metastability risk. The τ extracted in step S3... 1,n and τ 2,n This characteristic, which serves as the first relaxation time constant, is passed to step S5 to participate in the fusion calculation of the metastable risk index.

[0072] Step S4: Based on the change of the first fluctuation complexity feature relative to the preset benchmark complexity feature and the change of the first relaxation time constant feature relative to the preset benchmark time constant feature, the metastable risk index of the target battery after the Nth cycle is generated by fusing them.

[0073] The steps involved in generating the metastable risk indicators include:

[0074] Calculate the first relative change in the comprehensive magnitude of the first fluctuation complexity feature relative to the comprehensive magnitude of the preset baseline complexity feature;

[0075] Calculate the second and third relative changes of the fast relaxation time constant and slow relaxation time constant in the first relaxation time constant characteristic relative to the preset reference fast relaxation time constant and the preset reference slow relaxation time constant, respectively;

[0076] Based on the preset first fusion weight, second fusion weight, and third fusion weight, the first relative change, second relative change, and third relative change are weighted and summed to obtain the metastable risk index.

[0077] The core task of step S4 is to fuse the first fluctuation complexity feature extracted in step S2 with the first relaxation time constant feature extracted in step S3 to generate a dimensionless index (metastable state risk index) that can comprehensively characterize the current internal state of the battery. This fusion process is not a simple numerical superposition, but a weighted combination based on physical meaning, aiming to map microscopic dynamic information of different dimensions into a unified risk assessment space.

[0078] Before performing the specific fusion calculations, the system first needs to complete a crucial "black box deconstruction" step, namely, determining the preset first fusion weight, second fusion weight, and third fusion weight in claim 5. These weights are not arbitrarily assigned but are learned from historical data through an offline calibration process. Specifically, the system collected test data from at least 50 batteries of the same type throughout their complete lifecycle, including the multi-scale entropy vector, relaxation time constant, and corresponding actual capacity decay for each cycle from the initial healthy state to 80% capacity decay. Using a multiple linear regression method, with the capacity decay rate as the target variable and the rate of change of the comprehensive amplitude of the multi-scale entropy vector, the rate of change of the fast relaxation time constant, and the rate of change of the slow relaxation time constant as input features, a linear regression model was fitted. The standardized regression coefficients of each input feature in this model, after normalization, become the fusion weights α, β, and γ. This calibration process ensures that the weights objectively reflect the actual contribution of different features to the battery aging process and satisfy α + β + γ = 1. For example, in a set of typical calibration results, the weight α of the multi-scale entropy feature is 0.45, the weight β of the fast relaxation time constant is 0.30, and the weight γ of the slow relaxation time constant is 0.25, indicating that in this type of battery, the fluctuation complexity change in the relaxation stage has the most significant impact on the aging process.

[0079] After completing the weight calibration, the system begins calculating the relative changes of each item in the current loop. First, the comprehensive magnitude of the first fluctuation complexity feature is calculated. For the multi-scale entropy vector C_n of the Nth loop generated in step S2, its comprehensive magnitude is defined as the Euclidean norm of the vector:

[0080] ;

[0081] This composite amplitude condenses the sample entropy values ​​across 20 scales into a single scalar, characterizing the overall fluctuation and disorder of the relaxation process. The system simultaneously reads the composite amplitude ||C|| of the benchmark multi-scale entropy vector stored during initial calibration. ref ||2, and then calculate the first relative change:

[0082] ;

[0083] When ΔC n A value greater than 1 indicates that the disorder level of the current relaxation fluctuation is higher than that of the initial state;

[0084] Next, the relative change in the relaxation time constant is calculated. The system obtains the fast relaxation time constant τ for the Nth cycle from step S3. 1,n and slow relaxation time constant τ 2,n And read the reference fast relaxation time constant τ stored during initial calibration. 1,ref and the reference slow relaxation time constant τ2,ref The second and third relative changes are defined as follows:

[0085] ;

[0086] A ratio greater than 1 for these two values ​​means that the relaxation process is slowed down, i.e., electrochemical kinetics are hindered.

[0087] After obtaining the three relative changes, the system calculates the metastable risk index for the Nth cycle according to the weighted summation rule:

[0088] ;

[0089] This indicator is a dimensionless positive real number. In a healthy battery state, since all relative changes are close to 1, the MSRI value is also close to 1. As the battery ages, these indicators gradually deviate from the baseline, and the MSRI value monotonically increases. Its magnitude directly reflects the degree to which the battery deviates from its healthy state. For example, in a certain actual test, the calculated ΔC value for a battery at the 200th cycle... 200 =1.21、Δτ 1,200 =1.15、Δτ 2,200 =1.09, substituting the values ​​into the calibration weights α=0.45, β=0.30, and γ=0.25, the MSRI is calculated. 200 =0.45×1.21+0.30×1.15+0.25×1.09=1.165; and in the 600th cycle, the relative changes increased to 1.68, 1.53, and 1.47 respectively, then MSRI 600 =0.45×1.68+0.30×1.53+0.25×1.47=1.578, this upward trend accurately reflects the cumulative process of internal damage in the battery.

[0090] It is worth noting that the MSRI calculation results will serve as a key intermediate variable, simultaneously supporting two subsequent processes: firstly, it is directly input into step S5 for comparison with the dynamic risk threshold to adjust the charging stress of the next cycle in real time; secondly, it is stored in the historical database, forming the input features of the lifetime prediction correction model together with the capacity decay sequence, achieving collaborative prediction of remaining lifetime. This design allows a single indicator to simultaneously perform the dual functions of real-time regulation and long-term prediction, reflecting the intensive nature of the system architecture and the integrity of the data flow.

[0091] Step S5: Based on the comparison between the metastable risk index and the dynamic risk threshold, dynamically adjust the charging stress parameters of the target battery in the N+1th charge-discharge cycle.

[0092] The dynamic risk threshold is dynamically generated based on the statistical distribution of historical metastable risk indicators, specifically including:

[0093] Obtain the historical metastable risk indicator sequence for the current loop and previous loops with a preset number of iterations;

[0094] Calculate the mean and standard deviation of the historical metastable risk indicator series;

[0095] The dynamic risk threshold is updated based on the mean and standard deviation.

[0096] The steps for dynamically adjusting charging stress parameters based on comparison results include:

[0097] If the metastable risk index is less than or equal to the dynamic risk threshold, then the charging stress parameter of the target battery in the N+1th cycle is maintained at the preset basic stress parameter.

[0098] If the metastable risk index is greater than the dynamic risk threshold, the charging stress parameter is reduced proportionally according to the degree to which the metastable risk index exceeds the dynamic risk threshold, and the reduced charging stress parameter is not lower than the preset minimum stress threshold.

[0099] The core of step S5 lies in comparing the metastable risk index MSRI_n calculated in step S4 with the currently effective dynamic risk threshold, and making a real-time decision on the charging stress parameters for the next cycle based on the comparison result. This decision-making mechanism enables the test system to adaptively adjust the acceleration stress according to the current health state of the battery, ensuring test efficiency while avoiding sudden battery failure due to excessive aggressiveness.

[0100] Prior to this step, the system has maintained a dynamic risk threshold, which is generated according to the method of claim 6 based on the historical MSRI sequence up to the previous cycle (N-1th cycle). Specifically, let the historical MSRI sequence of the most recent K cycles (e.g., K=20) be {MSRI_{N-20},...,MSRI_{N-1}}, the system calculates the mean μ and historical standard deviation σ of this sequence, then the dynamic risk threshold currently used for the comparison in the Nth cycle is defined as: MSRI th =μ+3σ; where the multiplier 3 is the default value, but in subsequent step S7, this multiplier may be adjusted to other values ​​based on the statistical data of the reference battery. It should be noted that, in order to ensure the timeliness of the threshold, the system updates μ and σ once with the latest data after each loop, thereby providing a dynamic benchmark for the next comparison.

[0101] In this step, the charging stress parameter specifically refers to the charging rate in the (N+1)th charge-discharge cycle, denoted as the base charging rate C_base. This value is preset by the test scheme, for example, 2C. Simultaneously, the system presets a minimum stress threshold, which in this example is set to 0.5 times the base charging rate, i.e., 0.5C, to ensure that even with extremely high risks, the charging stress will not fall below the lower limit required for safety testing.

[0102] The comparison logic and adjustment rules are as follows:

[0103] If MSRI_n≤MSRI_th, it indicates that the metastability risk of the current battery is still within an acceptable range. The test system maintains the charging rate of the N+1th cycle unchanged, i.e., C_rate,n+1=C_base.

[0104] If MSRI_n > MSRI_th, it indicates that the battery has entered a high-risk state, and subsequent charging stress needs to be reduced immediately to avoid failure. In this case, the system calculates an adjustment factor f_n based on the degree of risk.

[0105] ;

[0106] This factor reflects the linear penalty for relative risk excess. However, to avoid over-penalization leading to test stagnation, the adjustment factor is limited to above 0.5, i.e.:

[0107] ;

[0108] Therefore, the charging rate for the (N+1)th cycle is determined as follows:

[0109] ;

[0110] For example, suppose the current dynamic threshold MSRI_th = 1.3 and the base charging rate C_base = 2C. If the MSRI_n = 1.25 is calculated in the Nth cycle, since 1.25 ≤ 1.3, the next cycle will still charge at 2C. If MSRI_n = 1.5, the relative amount exceeding the threshold is (1.5 - 1.3) / 1.3 ≈ 0.154, the adjustment factor f_n = 1 - 0.154 = 0.846, which is greater than the lower limit of 0.5, so C_rate,n+1 = 2C × 0.846 ≈ 1.69C. If MSRI_n is as high as 2.0, then (2.0 - 1.3) / 1.3 ≈ 0.538, f_n = 1 - 0.538 = 0.462, which is lower than the lower limit of 0.5, so f_n = 0.5, and C_rate,n+1 = 1C.

[0111] After making the above decision, the system sends the newly determined charging rate C_rate,n+1 to the charge / discharge test unit for executing the (N+1)th charging cycle. Simultaneously, this decision result, along with MSRI_n, is recorded in the test log for subsequent analysis and model correction. It is worth noting that this step only changes the charging stress parameters; the discharging parameters and other test conditions remain unchanged to ensure the comparability of the test results.

[0112] Through this dynamic stress adjustment mechanism, the testing system achieves real-time response to battery status: maintaining accelerated stress when the battery is healthy to shorten the test cycle, and actively degrading when the risk increases to protect the battery, thereby achieving a dynamic balance between efficiency and safety in accelerated life testing.

[0113] It also includes a lifetime prediction correction step:

[0114] Obtain the historical metastable risk indicator sequence and historical capacity decay sequence up to the current cycle;

[0115] Dynamic time warping is performed on the historical metastable risk index sequence and the historical capacity decay sequence to eliminate the time lag between the two sequences and generate an aligned capacity decay sequence.

[0116] Based on the aligned capacity decay sequence and historical metastable risk index sequence, a nonlinear relationship model between capacity decay rate and metastable risk index is fitted.

[0117] Based on the metastable risk index and nonlinear relationship model of the current cycle, predict the remaining number of cycles required for the target battery to reach the preset failure threshold.

[0118] After the dynamic adjustment of the charging stress for the (N+1)th cycle is completed in step S5, the system enters the lifespan prediction and correction step. This step aims to dynamically predict and correct the remaining lifespan of the battery using accumulated historical monitoring data. This prediction process does not simply extrapolate the capacity decay curve, but rather constructs a collaborative prediction model that can provide early warnings of lifespan inflection points by exploring the deep nonlinear correlation between metastable risk indicators and capacity decay.

[0119] To support this prediction model, the system first maintains two core data sequences: one is a historical metastable risk index sequence from the first cycle to the current Nth cycle, which is generated and stored after each cycle in step S4; the other is the corresponding historical capacity decay sequence, where the capacity decay value for each cycle is calculated by the ratio of the measured discharge capacity to the initial calibrated capacity for that cycle, specifically 1 minus the ratio of the current capacity to the initial capacity. These two sequences together constitute all the input data for lifetime prediction.

[0120] Because changes in metastable risk indicators often precede capacity decay in time—meaning the deterioration of the battery's internal microstructure manifests before the decrease in macroscopic capacity—there is a phase lag between the two sequences. Directly using the original sequences for modeling would prevent the model from accurately capturing the true causal relationship between risk indicators and capacity decay. To address this issue, this step first introduces a dynamic time warping algorithm to align the two sequences. The core idea of ​​this algorithm is to allow the two sequences to scale non-linearly along the time axis, finding an optimal matching path that minimizes the cumulative distance between corresponding points on the path. In practice, the system constructs a distance matrix where each element represents the absolute difference between a point in the risk indicator sequence and another point in the capacity decay sequence. Then, a dynamic programming method is used to search for the optimal path from the start to the end of the sequence. Along this optimal path, the system remaps the original capacity decay sequence into a new aligned sequence, ensuring that each capacity decay value in the new sequence precisely matches the corresponding risk value in the risk indicator sequence in time. After dynamic time warping and alignment, each fluctuation of the risk indicator can accurately correspond to the capacity decay response it induces, thereby eliminating the modeling error caused by response lag.

[0121] After obtaining the aligned capacity decay sequence, the system further constructs a nonlinear relationship model between the capacity decay rate and the metastable risk index. According to electrochemical aging theory, when a battery enters the accelerated aging stage, the capacity decay rate and the degree of internal damage often exhibit an exponential correlation. Therefore, this embodiment uses an exponential differential equation model to describe this relationship, where the capacity decay rate (capacity loss per unit cycle) is proportional to an exponential function of the current metastable risk index. To apply this continuous model to discrete cyclic data, the system uses data points from the most recent preset number of cycles (e.g., the most recent 50 cycles) for parameter fitting. For each cycle involved in the fitting, the system calculates an approximate value of the capacity decay rate for that cycle, typically taking the difference between the capacity decay values ​​after alignment of two adjacent cycles. Then, these approximate capacity decay rates and the corresponding metastable risk index values ​​are substituted into the exponential model, and the model is transformed into a linear form through logarithmic transformation. Finally, the least squares method is used to estimate two key parameters in the model: the base decay coefficient and the risk amplification coefficient. These two parameters jointly determine the accelerating effect of the risk index on subsequent capacity decay.

[0122] After obtaining the model parameters, the system begins predicting the remaining battery life. The prediction starts with the capacity decay value of the current cycle, with the preset failure threshold typically set at 20% capacity decay (i.e., battery health dropping to 80%). During the prediction process, the system uses the latest metastable risk indicator as a conservative estimate of the risk indicator for subsequent cycles, employing a step-by-step recursive approach. With each additional cycle, the system updates the capacity decay value based on the current capacity decay value and the decay increment calculated from the current risk indicator and model parameters. This recursive process is repeated until the predicted capacity decay value reaches or exceeds the preset failure threshold; the number of recursive steps taken at this point represents the predicted number of remaining cycles.

[0123] Based on the detailed description of the above embodiments, this scheme constructs a metastable risk index that can characterize the degree of instability of the battery's internal microstructure by mining the multi-scale fluctuation characteristics and relaxation dynamics features implicit in the voltage time series during the relaxation stage. This index integrates the complexity changes and time constant drift of the relaxation process, realizing the quantitative capture of early abnormal signals during the transition of the battery from a healthy state to a failure state. On this basis, by introducing a dynamic risk threshold based on the statistical distribution of historical risk indices, the system can adaptively adjust the charging stress parameters of subsequent cycles according to the current metastable level of the battery. When the risk is low, it maintains accelerated stress to ensure test efficiency, and actively degrades to avoid sudden failure when the risk increases. At the same time, by using a dynamic time warping algorithm to align the risk index with the capacity decay sequence, a remaining lifetime prediction model based on the risk index is established, realizing early warning of nonlinear aging behaviors such as capacity drop. This technical solution transforms the dynamic information of the relaxation phase, which is neglected in traditional testing, into a quantifiable risk assessment indicator. It fundamentally solves the problem that existing methods cannot detect metastable drift due to information lag, significantly improving the accuracy and safety of battery life testing under fast charging conditions. It provides an adaptive control mechanism that balances efficiency and reliability for battery acceleration testing, and helps to more accurately define safety boundaries and optimize fast charging strategies during the battery R&D stage.

[0124] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0125] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0126] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for accelerated life testing and evaluation of batteries under fast charging scenarios, applied to battery testing systems, characterized in that... Includes the following steps: Acquire the relaxation response signal of the target battery during the relaxation phase after the Nth charge-discharge cycle, wherein the relaxation response signal includes at least a voltage time series; Multi-scale fluctuation analysis is performed on the voltage time series to extract the first fluctuation complexity feature, which characterizes the degree of disorder of fluctuations in the voltage time series at different time scales. Relaxation kinetics analysis was performed on the voltage time series to extract the first relaxation time constant characteristic used to characterize the decay rate of the voltage time series; Based on the change of the first fluctuation complexity feature relative to the preset benchmark complexity feature and the change of the first relaxation time constant feature relative to the preset benchmark time constant feature, the metastable risk index of the target battery after the Nth cycle is generated. Based on the comparison between the metastable risk index and the dynamic risk threshold, the charging stress parameters of the target battery in the N+1th charge-discharge cycle are dynamically adjusted.

2. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The relaxation response signal further includes the shell strain time series and / or the surface temperature time series; the first fluctuation complexity feature is specifically a multi-scale entropy vector obtained by performing multi-scale sample entropy analysis on the voltage time series, and the multi-scale entropy vector contains the sample entropy values ​​of the voltage time series under multiple preset scale factors.

3. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The steps for extracting the first fluctuation complexity feature include: Determine multiple preset scale factors; For each scale factor, the voltage time series is coarsened to generate a coarse-grained series at that scale. Calculate the sample entropy for each coarse-grained sequence to obtain the sample entropy value under that scale factor; The multi-scale entropy vector is generated by combining the sample entropy values ​​under all scale factors.

4. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The extraction steps for the first relaxation time constant feature include: The voltage time series is fitted with a double exponential curve to obtain a fast relaxation time constant and a slow relaxation time constant, which are then used as features of the first relaxation time constant.

5. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The steps for generating the metastable risk indicators include: Calculate the first relative change in the comprehensive magnitude of the first fluctuation complexity feature relative to the comprehensive magnitude of the preset benchmark complexity feature; Calculate the second and third relative changes of the fast relaxation time constant and the slow relaxation time constant in the first relaxation time constant feature relative to the preset reference fast relaxation time constant and the preset reference slow relaxation time constant, respectively; Based on the preset first fusion weight, second fusion weight, and third fusion weight, the first relative change, second relative change, and third relative change are weighted and summed to obtain the metastable risk index.

6. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The dynamic risk threshold is dynamically generated based on the statistical distribution of historical metastable risk indicators, specifically including: Obtain the historical metastable risk indicator sequence for the current loop and previous loops with a preset number of iterations; Calculate the mean and standard deviation of the historical metastable risk index series; The dynamic risk threshold is updated based on the mean and standard deviation.

7. The method for accelerated life testing and evaluation of batteries under fast charging scenarios according to claim 1, characterized in that, The step of dynamically adjusting the charging stress parameters based on the comparison results includes: If the metastable risk index is less than or equal to the dynamic risk threshold, then the charging stress parameter of the target battery in the N+1th cycle is maintained at the preset basic stress parameter. If the metastable risk index is greater than the dynamic risk threshold, the charging stress parameter is reduced proportionally according to the degree to which the metastable risk index exceeds the dynamic risk threshold, and the reduced charging stress parameter is not lower than the preset minimum stress threshold.

8. The method for accelerated battery life testing and evaluation under fast charging scenarios according to any one of claims 1-7, characterized in that, It also includes a lifetime prediction correction step: Obtain the historical metastable risk indicator sequence and historical capacity decay sequence up to the current cycle; The historical metastable risk index sequence and the historical capacity decay sequence are dynamically time-warped to eliminate the time lag between the two sequences and generate an aligned capacity decay sequence. Based on the aligned capacity decay sequence and the historical metastable risk index sequence, a nonlinear relationship model between the capacity decay rate and the metastable risk index is fitted. Based on the metastable risk index of the current cycle and the nonlinear relationship model, the remaining number of cycles required for the target battery to reach the preset failure threshold is predicted.

Images