Device and method for predicting safety index of retired power battery

Abstract

The invention discloses a decommissioned power battery safety index prediction device and a decommissioned power battery safety index prediction method, which form a'test-analysis-application 'closed loop through cross-scale parameter acquisition of a hardware module, nonlinear feature learning of a modeling module and engineering decision support of an interaction module, and solve the problem of high risk of missed judgment of traditional thermal parameter evaluation. And the safety evaluation accuracy of the retired battery is improved from 72% to 96.25%. The hardware test module can realize high-precision synchronous acquisition of microcosmic parameters such as dynamic internal resistance and charge transfer resistance and thermal parameters such as self-heating initial temperature and thermal runaway critical temperature, and ensures the accuracy and integrity of data. The data processing module can realize cross-capacity and cross-system (lithium iron phosphate / ternary) battery safety level quantitative evaluation through Z-score standardization processing in combination with a BSI formula and an SVM model of an RBF kernel function. The scheme has the advantages of nondestructive testing, shortened flow time by 60% and low cost, and a standardized tool is provided for large-scale safety evaluation of the retired battery.

Description

Technical Field

[0001] The present invention relates to the technical field of cascade utilization of power batteries for new energy vehicles, and in particular to a device and method for predicting the safety index of retired power batteries based on microscopic electrochemical parameters and machine learning. Background Art

[0002] With the rapid development of the new energy vehicle industry, the scale of power battery retirement has gradually expanded. The cascade utilization of retired automotive power batteries is an important task for my country to implement the strategy of building a strong country in new energy vehicles, and it is also the main way to maximize the value of batteries throughout their life cycle. However, a major problem currently facing the cascade utilization of power batteries is that the safety of retired power batteries is difficult to accurately assess. During the entire life cycle of power batteries, the safety hazards caused by defect amplification, accumulation of micro-abuse, etc. are complex and diverse, and battery degradation has a significant impact on battery safety, and the internal change state is poor. Traditional detection methods have the disadvantages of low detection efficiency, low accuracy, high cost, and damage to batteries, and cannot meet the safety assessment needs of large-scale retired power battery cascade utilization. Summary of the Invention

[0003] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is to provide a device and method for predicting the safety index of retired power batteries, which can achieve high-accuracy, high-detection efficiency, low-cost, and non-destructive safety level assessment of retired power batteries, and provide safety guarantees for the cascade utilization of retired power batteries.

[0004] To achieve the above objectives, the present invention provides a device for predicting the safety index of retired power batteries, comprising:

[0005] The hardware testing module uses testing equipment to obtain the microscopic electrochemical parameters and thermal safety parameters of retired batteries, providing raw data support for safety assessments;

[0006] The data processing and modeling module extracts features and standardizes the raw test data, and builds a security assessment model through machine learning algorithms to achieve the conversion from data to security index and evaluate the security level;

[0007] The result output and interaction module displays the safety assessment results and provides decision support to assist in formulating strategies for the reuse of retired batteries;

[0008] The raw data collected by the hardware test module is transmitted to the data processing and modeling module, and a standardized feature vector is generated after feature extraction; the data processing and modeling module simultaneously runs BSI formula calculation and SVM model prediction: the BSI formula is based on thermal parameters and capacity normalization to output safety levels; the SVM model classifies the feature vector through a kernel function and optimized parameters, and outputs a safety / unsafe label; the final result is summarized to the result output and interaction module, and a report and decision suggestion are generated in combination with a historical case library.

[0009] As a further improvement of the application, the hardware test module comprises:

[0010] Microscopic electrochemical parameter testing instruments are used for high SOC pulse charge and discharge testing and electrochemical impedance testing microscopic parameter acquisition.

[0011] An accelerated rate calorimeter (ARC) tester is used to obtain thermal stability and thermal runaway related parameters, and to obtain a self-heating initiation temperature , a thermal runaway critical temperature , ;

[0012] A temperature control device is used to keep the power battery at a constant temperature or maintain a stable temperature rise rate.

[0013] A data acquisition and control unit synchronously acquires electrochemical and thermal parameters to generate a timestamp-aligned raw data set.

[0014] As a further improvement of the application, the data processing and modeling module comprises:

[0015] A feature extraction and preprocessing unit performs data processing to convert the raw data of the hardware test module into a standardized feature vector, and eliminates noise and dimension differences.

[0016] A safety index calculation engine calculates a safety index BSI based on thermal parameters and capacity normalization, provides a basic risk level assessment, and outputs a BSI value and level.

[0017] A machine learning model and training unit constructs a nonlinear classification model through a SVM algorithm to improve the accuracy of high-risk battery identification.

[0018] As a further improvement of the application, the result output and interaction module comprises:

[0019] A safety level visualization interface converts the safety assessment result into intuitive charts, labels key risk parameters, and supports interactive analysis.

[0020] A decision support database provides historical case comparison and echelon utilization strategy recommendation to support intelligent decision-making.

[0021] The interactive control and report generation unit supports user operations, data export and report generation, improving system usability.

[0022] The present invention also discloses a method for predicting the safety index of retired power batteries, comprising:

[0023] S100, micro electrochemical parameter test, and extract micro electrochemical parameters, micro electrochemical parameters include dynamic internal resistance , polarization recovery time , charge transfer resistance , Warburg coefficient , polarization voltage difference , phase transition peak intensity of ICA curve , peak area of ​​ICA curve, abnormal peak area of ​​DV curve ;

[0024] S200, safety performance index acquisition and calculation:

[0025] Obtaining the self-heating onset temperature of retired power batteries through Accelerating Rate Calorimeter (ARC) testing , thermal runaway critical temperature , maximum temperature , total heat release energy of thermal runaway ( ) four parameters;

[0026] Battery safety indicators It is described by the following formula:

[0027] ;

[0028] S300, model construction:

[0029] S310, taking the microscopic electrochemical parameters obtained in S100 as characteristic parameters, and taking the microscopic electrochemical parameters obtained in S200 as characteristic parameters 、 、 The three parameters are used as evaluation indicators of battery safety to construct a least squares support vector machine model;

[0030] Choose Gaussian radial basis function (RBF) as the kernel function:

[0031] ;

[0032] S320, based on the RBF kernel function, the classification decision function of SVM is:

[0033]

[0034] Safety Category Label : The BSI level of S200 is mapped into two categories, =1 corresponds to "safe", =-1 corresponds to "unsafe";

[0035] S400, model training:

[0036] Collect the microscopic electrochemical parameters and accelerated calorimetry (ARC) test results of retired power batteries to construct a training dataset (XT rain ,yT rain ), where XT rain is the feature matrix, yT rain is the corresponding category label (safe or unsafe); XT rain The feature matrix contains N groups of samples, each group of samples is a 9-dimensional vector ( , , , , , 、 、 、 ), yT rain Labels are verified by BSI grade manual labeling or thermal runaway testing;

[0037] The SVM model is trained using the Sequential Minimal Optimization (SMO) algorithm to solve the model parameters by minimizing the following optimization problem:

[0038]

[0039] Subject to the following constraints:

[0040]

[0041] Among them, C is the penalty factor, which controls the degree of penalty for misclassification; i It is a slack variable that allows a certain degree of misclassification;

[0042] Solve the above optimization problem by Lagrange multiplier method and get the Lagrange multiplier And the bias term b, the final classification decision function is:

[0043]

[0044] Input features x: dimensional microscopic electrochemical parameters extracted from S100 test ( , , , , ), normalized by Z-score;

[0045] S500, safety level prediction:

[0046] The microscopic electrochemical parameters of the battery to be evaluated are input into the trained SVM model, which outputs a safety label and maps the risk according to the grade table based on the BSI value calculated by S200;

[0047] When BSI ≤ threshold and safety label = +1, confirm the BSI risk level;

[0048] When BSI>threshold or safety label=-1, it is judged as high risk.

[0049] As a further improvement of the present invention, S100 includes:

[0050] S110, performing a high SOC pulse charge and discharge test on the retired power battery to obtain a high SOC pulse charge and discharge test curve;

[0051] Parameter extraction: dynamic internal resistance , polarization recovery time ;

[0052] S120, performing a high SOC electrochemical impedance test on the retired power battery to obtain a high SOC electrochemical impedance test curve;

[0053] Parameter extraction: Charge transfer resistance , Warburg coefficient , characterizes the interfacial reaction activity and ion diffusion rate;

[0054] S130: Perform a charge voltage-discharge voltage test on the retired power battery to obtain a charge voltage-discharge voltage curve;

[0055] Parameter extraction: polarization voltage difference ;in is the charging voltage, is the discharge voltage;

[0056] S140, perform incremental capacity analysis test on retired power batteries to obtain ICA curve;

[0057] Parameter extraction: differential capacity dQ / dV-V (ICA curve) phase transition peak intensity , peak area, identifying electrode material phase change and active material loss;

[0058] S150, performing a differential voltage analysis (DV) test on the retired power battery to obtain a DV curve;

[0059] Parameter extraction: Abnormal peak area .

[0060] As a further improvement of the present invention, S130 performs a charge voltage-discharge voltage test on the retired power battery, including:

[0061] Charge the retired power battery with constant current to the cut-off voltage and record the average voltage during the charging process After standing still, discharge at a constant current until the cut-off voltage is reached, and record the average voltage during the discharge process. ;

[0062] Converted energy efficiency: ;

[0063] Energy efficiency η refers to the ratio of battery discharge energy to charging energy, reflecting the energy conversion efficiency of the battery during charging and discharging. The higher η is, the smaller the energy loss of the battery during charging and discharging, and the better the performance.

[0064] Add η to optimize the calculation formula of BSI:

[0065]

[0066] The value of is used to replace BSI for judgment.

[0067] As a further improvement of the present invention, S150 performs differential voltage analysis (DV) testing on retired power batteries, including:

[0068] Charge and discharge control: Perform charge and discharge tests on retired power batteries at a constant current, covering the full SOC range;

[0069] Data acquisition: Sampling rate ≥ 10 Hz, recording voltage-capacitance (VQ) raw data;

[0070] Differential processing: After smoothing and filtering the VQ data, calculate the differential voltage: ;

[0071] Curve drawing: Generate dV / dQ-Q curves (DV curves) and analyze abnormal peaks in specific voltage ranges;

[0072] Abnormal peak area extract:

[0073] Locating abnormal areas: Identifying peaks and valleys that deviate from normal morphology within characteristic voltage ranges;

[0074] Area calculation: perform numerical integration on the abnormal peaks and calculate the area under the peaks. The unit is normalized area. The formula is: .

[0075] As a further improvement of the application, S500 further comprises: storing the input parameters, BSI values and decision results into a historical case library; when new data is added to the case library, triggering S400 model retraining to update parameters.

[0076] As a further improvement of the application, in S200, the capacity conversion formula is used to obtain :

[0077]

[0078] Battery safety index formula optimization:

[0079] .

[0080] The beneficial effects of the application are:

[0081] The application solves the three defects of traditional methods through four-step closed loop of data fusion→model learning→dynamic decision→feedback optimization: microscopic defect missed detection (S100 supplement), capacity difference interference (S200 normalization solution), nonlinear correlation misjudgment (SVM model of S300 / S400 solves), which makes the accuracy of retired battery safety evaluation from 72% to 96.25%. The hardware test module can realize high-precision synchronous acquisition of microscopic parameters such as dynamic resistance and charge transfer resistance, and thermal parameters such as self-heating starting temperature and thermal runaway critical temperature, ensuring the accuracy and integrity of the data. The data processing module can realize cross-capacity, cross-system (lithium iron phosphate / ternary) battery safety grade quantitative evaluation through Z-score standardization processing combined with BSI formula and SVM model of RBF kernel function. The scheme has the advantages of nondestructive testing, process time reduction of 60% (single battery evaluation time from 48 hours to 19 hours) and low cost, providing a standardized tool for large-scale safety evaluation of retired batteries, and effectively promoting the safety guarantee and industrial application of retired power battery gradient utilization. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 is a system structure schematic diagram of a retired power battery safety index prediction device;

[0083] Figure 2 is a schematic diagram of the cooperative relationship of S100-S500 in the retired power battery safety index prediction method;

[0084] Figure 3 is a flowchart of the retired power battery safety index prediction method. DETAILED DESCRIPTION

[0085] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0086] Example 1

[0087] See also Figure 1 The retired power battery safety index prediction device of this embodiment includes:

[0088] Hardware testing module, which uses testing equipment to obtain microscopic electrochemical parameters of retired batteries, including , , , , , 、 、 、 ; and thermal safety parameters, including self-heating onset temperature , thermal runaway critical temperature 、 、 , providing original data support for safety assessment;

[0089] The data processing and modeling module extracts features and standardizes the raw test data, and builds a security assessment model through machine learning algorithms to achieve the conversion from data to security index and evaluate the security level;

[0090] The result output and interaction module visualizes the safety assessment results and provides decision support to assist in formulating strategies for the reuse of retired batteries;

[0091] The raw data collected by the hardware test module (such as pulse voltage drop value ΔV, ARC thermal parameters) is transmitted to the data processing and modeling module, and after feature extraction, a 9-dimensional standardized feature vector (such as , , , , , 、 、 、 );

[0092] The data processing and modeling module runs BSI formula calculation and SVM model prediction simultaneously:

[0093] The BSI formula outputs a safety level (e.g., Level II) based on normalized thermal parameters and capacity;

[0094] The SVM model classifies the feature vectors using the kernel function (RBF) and optimized parameters (C=50,σ=2.0) and outputs a safe / unsafe label ( = ±1);

[0095] The final results are summarized in the result output and interaction module, and combined with the historical case library to generate visual reports and decision-making recommendations (such as cascade utilization scenario matching).

[0096] Through cross-scale parameter acquisition (micro + macro) of the hardware module, nonlinear feature learning (SVM) of the modeling module, and engineering decision support of the interactive module, a closed loop of "testing-analysis-application" is formed to solve the problem of high-risk omissions in traditional thermal parameter assessment, and increase the accuracy of retired battery safety assessment from 72% to 96.25%.

[0097] Preferably, the hardware testing module includes:

[0098] Micro electrochemical parameter test instrument, used for high SOC pulse charge and discharge test, electrochemical impedance test micro parameter acquisition; execute micro electrochemical parameter test process, obtain 、 、 、 、 , resolution up to 0.1mΩ ( ) and 0.01 (normalized );

[0099] Accelerated Calorimetry (ARC) tester, to obtain parameters related to thermal stability and thermal runaway, and to obtain the self-heating starting temperature , thermal runaway critical temperature 、 , temperature test error ≤ 2℃, Accuracy ±5%; Contains thermal runaway monitoring sensor, responsible for real-time collection of temperature and heat data during thermal runaway (sampling rate ≥ 10Hz, capture 、 、 and );

[0100] Temperature control device, used to maintain a constant temperature or a stable heating rate for the power battery. ARC test chamber can be used (heating rate adjustable from 1°C / min to 10°C / min, temperature accuracy ±1°C);

[0101] The data acquisition and control unit synchronously collects electrochemical and thermal parameters, generates time-stamped raw data sets, and stores them in CSV / Excel format. A multi-channel data acquisition card (sampling rate ≥ 100Hz, 16-bit ADC) can be used.

[0102] Furthermore, the microscopic electrochemical parameter testing instrument includes:

[0103] High SOC pulse charge and discharge tester, such as ArbinBT2000, can perform high SOC pulse charge and discharge tests on retired power batteries. It supports 1C pulse current with an accuracy of ±0.1A.

[0104] Electrochemical impedance spectrometer, such as Solartron1260, is used to perform high SOC electrochemical impedance testing on retired power batteries, with a frequency range of 100Hz~100kHz and an AC amplitude of 5mV.

[0105] A charge-discharge cycle tester is used to perform charge-discharge voltage testing on retired power batteries. It can be used with Arbin Instruments BT2000 series or Neware BTS series battery test systems, supporting 0.2C-0.5C constant current charge and discharge with a cutoff voltage accuracy of ±10mV. Data from the charge-discharge cycle tester is collected by the Data Acquisition and Control Unit, and then used to perform incremental capacity analysis (ICA) and differential voltage analysis (DV) tests on retired power batteries.

[0106] Furthermore, the data acquisition and control unit includes:

[0107] The synchronous acquisition unit uses a high-precision clock synchronization module (such as a GPS clock or oven-controlled crystal oscillator, with an accuracy of ±10μs) to ensure that the timestamp error of electrochemical parameters (such as pulse voltage ΔV) and thermal parameters is less than 1ms, avoiding analysis bias caused by cross-module data asynchrony. It supports at least 16 channels of synchronous acquisition (8 channels of electrochemical parameters + 8 channels of thermal parameters), meeting the real-time monitoring requirements of multiple physical quantities (voltage, current, temperature, and heat) of retired batteries.

[0108] The data preprocessing unit features a built-in sliding window filtering algorithm (window size 5-10 points) that automatically identifies and removes spike noise (such as voltage glitches caused by sudden current changes) during pulse testing, improving raw data quality. The ADC range is automatically switched (e.g., 0-10V / 0-1V) based on the signal amplitude. The 16-bit ADC ensures a voltage resolution of 15μV (at the 10V range), enabling high-precision acquisition of microvoltage signals (such as polarization recovery voltage).

[0109] The real-time control unit, with a built-in state machine controller, automatically switches test modes (e.g., from pulse test to impedance test) according to preset procedures and supports user-defined scripts (e.g., Python / Matlab interfaces). It also monitors voltage / current / temperature thresholds in real time and automatically triggers power-off protection when safety limits are exceeded (e.g., voltage > 4.3V or temperature > 80°C), preventing battery overcharge and over-discharge.

[0110] Preferably, the data processing and modeling module includes:

[0111] The feature extraction and preprocessing unit performs data processing, converting the raw data of the hardware test module into a standardized feature vector to eliminate noise and dimensional differences. , , , , , 、 、 、 , , eliminate noise interference and generate a 9-dimensional standardized feature vector ( , , , , , 、 、 、 , ). Includes: signal smoothing algorithm, (Savitzky-Golay filter, window 5 / order 2). Z-score normalization engine, ( mean μ = 40 mΩ, standard deviation σ = 15 mΩ);

[0112] Safety index calculation engine, based on thermal parameters and capacity normalization to calculate the safety index BSI, providing basic risk level assessment. Output BSI value and level, support cross-capacity battery normalization (error <5%). Includes: BSI formula calculator, (support capacity conversion ), risk level mapping table, (level I to level V threshold fixed configuration);

[0113] The machine learning model and training unit uses the Support Vector Machine (SVM) algorithm to build a nonlinear classification model, improving the accuracy of high-risk battery identification. The Smooth Matrix Operator (SMO) algorithm solves the Lagrange multiplier to generate a classification decision function, achieving a 96.25% high-risk battery identification rate. This includes: an SVM model engine (RBF kernel function, σ=2.0, C=50) and a 5-fold cross-validation module (optimizing parameters to achieve an F1 score ≥ 0.94).

[0114] Preferably, the result output and interaction module includes:

[0115] The safety level visualization interface converts safety assessment results into intuitive charts, annotates key risk parameters, and supports interactive analysis. It includes risk heat map dashboard, real-time parameter dashboard, and multi-dimensional trend analysis. The risk heat map dashboard uses ECharts Sankey Diagram to display parameter associations; the real-time parameter dashboard displays dynamic internal resistance. , abnormal peak area , BSI level dynamic badges (such as green shield for level II and red alert for level IV), etc. Multi-dimensional trend analysis is used to obtain the correlation curve between BSI value and cycle number (including SVM prediction of the trend of the next 50 cycles); parameter deviation between the current battery and the new battery of the same type (such as A 20% drop is highlighted in red).

[0116] Decision support database, providing historical case comparison and cascade utilization strategy recommendation to support intelligent decision-making. It includes database and cascade utilization strategy library. The database stores multi-dimensional cases, and each case uses a 9-dimensional feature vector ( , , , , , 、 、 、 The cascade utilization strategy library prioritizes successful application cases of the same system and capacity (e.g. 8Ah lithium iron phosphate recommended case ID=567, 200 cycles without abnormalities).

[0117] The interactive control and report generation unit supports user operations, data export and report generation, improving system usability.

[0118] See also Figure 2-Figure 3 The method for predicting the safety index of retired power batteries in this embodiment includes:

[0119] S100, micro electrochemical parameter test:

[0120] S110, perform a high SOC pulse charge and discharge test on the retired power battery to obtain a high SOC (SOC>90%) pulse charge and discharge test curve;

[0121] Parameter extraction: dynamic internal resistance , polarization recovery time , reflecting the electrode conductivity and electrolyte transport capacity; ΔV refers to the voltage drop value under the action of pulse current in high SOC pulse charge and discharge test, the unit is volt (V), I is the pulse current (unit: A), polarization recovery time It refers to the time (unit: s) required for the battery voltage to recover from the lowest point to 90% of the initial value after the pulse current stops.

[0122] S120: Perform high SOC electrochemical impedance spectroscopy (EIS) testing on retired power batteries to obtain high SOC (SOC>90%) EIS curves. The impedance test was performed using a Solartron 1260 electrochemical workstation with an AC amplitude of 5mV and a frequency range of 100Hz-100kHz.

[0123] Parameter extraction: Charge transfer resistance (semicircle diameter in high-frequency region), Warburg coefficient (the slope of the oblique line in the low-frequency region) characterizes the interfacial reaction activity and ion diffusion rate. It is obtained through high SOC (SOC>90%) electrochemical impedance spectroscopy (EIS) testing. In the impedance spectrum, the diameter of the semicircle in the high-frequency region (100Hz~100kHz) is , the unit is milliohm (mΩ).

[0124] Warburg coefficient Describes the diffusion impedance of lithium ions in the electrode material and reflects the migration ability of ions in the solid electrode. The unit is In the low frequency region of the electrochemical impedance spectrum (<10Hz), the impedance spectrum shows a 45° slope, and its slope is similar to Related, through the formula , The unit of ω (angular frequency) is radian / second (rad / s), and its relationship with the test frequency f is ω=2πf. In the low frequency region (f<10Hz), and It is a linear relationship and can be extracted by the slope .

[0125] S130: Perform a charge voltage-discharge voltage test on the retired power battery to obtain a charge voltage-discharge voltage (UU) curve.

[0126] Parameter extraction: polarization voltage difference ;in is the charging voltage, is the discharge voltage;

[0127] Obtained through the charge voltage-discharge voltage (UU) test, the specific steps are as follows:

[0128] Charge the retired power battery at a constant current of 0.5C to the cut-off voltage (e.g. 4.2V for ternary battery and 3.65V for lithium iron phosphate), and record the average voltage during the charging process. After standing for 30 minutes, discharge at a constant current of 0.5C to the cut-off voltage (e.g. 2.8V for ternary battery and 3.0V for lithium iron phosphate), and record the average voltage during the discharge process. ;

[0129] Energy efficiency: ;

[0130] , discharge capacity (unit: Ah), that is, the amount of electricity released during the discharge process;

[0131] , average discharge voltage (unit: V), that is, the average voltage during the discharge process;

[0132] , charging capacity (unit: Ah), that is, the amount of electricity input during the charging process;

[0133] , average charging voltage (unit: V), that is, the average voltage during the charging process;

[0134] Energy efficiency η refers to the ratio of battery discharge energy to charging energy, expressed as a percentage (%), reflecting the energy conversion efficiency during the battery charging and discharging process.

[0135] Energy conversion capacity assessment: The higher η, the less energy loss during charge and discharge, and the better the performance. New batteries typically have an energy efficiency of over 95%. Retired batteries with η < 90% may have significant internal degradation (e.g., loss of active material and increased interfacial impedance).

[0136] Safety state association: Energy efficiency decline is often accompanied by increased polarization and heat generation. For example, when η=85% of a retired lithium iron phosphate battery, its internal impedance increases by 30% compared with η=95%, and the critical temperature of thermal runaway is A 12°C drop in temperature will result in a 0.15-0.2 decrease in the BSI. η can be used to optimize the BSI calculation formula:

[0137] ;

[0138] Coefficient selection:

[0139] 10: Amplify the impact when η deviates from 0.85;

[0140] 0.85: When η=85%, the BSI drops by 0.15-0.2, which is the threshold value and defines the energy efficiency safety critical value.

[0141] Source: Failure analysis of 1,200 retired batteries

[0142]

[0143] After trimming Can replace BSI.

[0144] S140. Perform an incremental capacity analysis (ICA) test on the retired power battery to obtain an ICA curve.

[0145] Parameter extraction: differential capacity dQ / dV-V (ICA curve) phase transition peak intensity , peak area, to identify the phase change of electrode materials and the loss of active substances.

[0146] Test method: Perform a 0.2C constant current charge and discharge test on retired power batteries, and record the voltage-capacity data throughout the test (sampling rate ≥ 10Hz). After smoothing the QV data (such as Savitzky-Golay filtering), calculate the differential capacity dQ / dV and draw the dQ / dV-V curve.

[0147] Phase transition peak intensity It is the height of the characteristic phase change peak in the ICA curve, in units of A / V (or dimensionless after normalization), reflecting the intensity of the phase change of the electrode material. Extraction method: locate the peak in a specific voltage range (such as lithium iron phosphate 3.45V±0.1V, ternary battery 4.0V±0.1V), and take the maximum value as .

[0148] Peak area is the integrated area of ​​the characteristic phase transition peak, expressed in A·h / V (or dimensionless after normalization). It is positively correlated with the total amount of active substances involved in the phase transition. Extraction method: Numerical integration (e.g., trapezoidal method) is performed over the peak interval to calculate the area under the peak.

[0149] The electrochemical activity of the active material is proportional to the new battery =1.0 (normalized), when the active substance loss is 10%, ≈0.9;

[0150] Peak area decay: The peak area is proportional to the total amount of active material. The peak area of ​​a new battery = 1.0 (normalized). When the active material loss is 20%, the peak area ≈ 0.8.

[0151] when When the peak area decreases by 15% and the critical temperature of battery thermal runaway decreases by 20%, The average temperature dropped by 12℃, and the total heat release energy of thermal runaway Increase 5kJ, resulting in safety index Decreased by 0.1-0.2.

[0152] Engineering application threshold

[0153] .

[0154] S150: Perform a differential voltage analysis (DV) test on the retired power battery to obtain a DV curve.

[0155] The DV curve of a normal battery shows smooth and regular peaks and valleys in the characteristic voltage range, corresponding to the phase change process of the electrode material; when the battery has microcracks, impurities mixed in or the active material is unevenly distributed, the DV curve will have abnormal peaks and valleys, and the degree of defect can be evaluated by quantifying the abnormal peak area.

[0156] Test steps

[0157] Charge and discharge control: Perform charge and discharge tests on retired power batteries at a constant current of 0.2C, covering the full SOC range (e.g. 2.8V-4.2V for ternary batteries);

[0158] Data acquisition: Sampling rate ≥ 10 Hz, recording voltage-capacitance (VQ) raw data;

[0159] Differential processing: After smoothing and filtering the VQ data (such as Savitzky-Golay filtering), the differential voltage is calculated: ;

[0160] Curve drawing: Generate dV / dQ-Q curves (DV curves) and analyze abnormal peaks in specific voltage ranges.

[0161] Abnormal peak area extract:

[0162] Locating abnormal areas: Identifying peaks and valleys that deviate from normal morphology in characteristic voltage ranges such as 3.8V±0.1V for ternary batteries and 3.6V±0.1V for lithium iron phosphate batteries;

[0163] Area calculation: Perform numerical integration (such as the trapezoidal method) on the abnormal peaks and calculate the area under the peak. The unit is normalized area (dimensionless). The formula is:

[0164] Abnormal peak area , detect internal defects such as microcracks and impurities;

[0165] Microcrack Detection Principle

[0166] Impact of crack formation: Microcracks are generated in the electrode material due to volume expansion / contraction during the cycle, which increases the contact resistance between the active material and the current collector, and an additional peak appears on the DV curve at the corresponding phase change voltage (such as 3.8V for ternary batteries);

[0167] Characteristic performance: Normal battery DV curve has no obvious peak at 3.8V. When microcracks exist, there will be an abnormal peak with a width of ≥0.2Q (capacity) and a height of ≥0.05mV / Ah. >0.03.

[0168] Principle of impurity mixing detection

[0169] Impurity influence: Metal particles (such as Fe, Cu) or electrolyte decomposition products mixed in during the production process can cause local micro-short circuits, resulting in abnormal fluctuations in the DV curve in the non-characteristic voltage range;

[0170] Characteristic performance: abnormal peak position is not fixed, peak shape is sharp (width ≤ 0.1Q), area >0.02, accompanied by a shortening of the charging voltage platform.

[0171] Relationship with battery safety:

[0172] when When the interface impedance increases due to microcracks, the critical temperature of thermal runaway may be Lowering 15℃, the total heat release energy of thermal runaway Increase 8kJ;

[0173] Micro-short circuits caused by impurities will cause a sudden rise in local temperature. For every 0.01 increase in the abnormal peak area of ​​the DV curve, the probability of thermal runaway increases by 12%.

[0174] Engineering applications and threshold standards

[0175] Typical defect thresholds

[0176]

[0177] Actual case verification

[0178] Case 1: After a ternary battery is cycled 1200 times, the DV curve appears at 3.8V =0.03 abnormal peak, disassembly found that the density of micro cracks in the positive electrode material reached 25 / mm², and the thermal runaway test showed Reduce to 160℃ (normal ≥180℃);

[0179] Case 2: Lithium iron phosphate battery =0.028, accompanied by =50mΩ, BSI=0.38 (Level IV), warning of risks 4 cycles in advance, identifying defects earlier than traditional thermal parameter evaluation.

[0180] S200, safety performance index acquisition and calculation:

[0181] The ARC test was conducted with a heating rate of 1°C / min and an initial temperature of 25°C. The self-heating starting temperature T1, thermal runaway critical temperature T2, maximum temperature T3, and total heat release energy of the retired power battery were obtained. ) four parameters;

[0182] Among them, self-heating starting temperature (T1): the temperature at which the battery starts to self-heat. The higher T1 is, the better the thermal stability of the battery;

[0183] Thermal runaway critical temperature (T2): The temperature at which the battery temperature rise rate reaches 1°C / min. The higher T2, the safer the battery.

[0184] Maximum temperature (T3): the highest temperature recorded during thermal runaway;

[0185] Total heat release energy of thermal runaway (H Total );

[0186] Battery safety indicators are described by the following basic formula:

[0187]

[0188] in, is the heat release energy normalized to the battery capacity of 10Ah. For lithium iron phosphate batteries, =24kJ, for ternary battery, H ref =40kJ;

[0189] By dividing the thermal stability parameter ( 、 ) and the severity of thermal runaway ( ) to eliminate the influence of the maximum temperature difference of different batteries and obtain the relative thermal stability index;

[0190] The nonlinear characteristics of the exponential function are used to enhance the suppression of high heat release energy. Much greater than When the exponential term approaches 0, the BSI is significantly reduced, which intuitively reflects the logic that "the higher the heat release energy, the less safe the battery is."

[0191] Versatility across battery types: Pass The classification setting and capacity conversion make BSI applicable to retired batteries of different systems (lithium iron phosphate / ternary) and different capacities, meeting the needs of large-scale assessment;

[0192] S300, model construction:

[0193] S310, taking the microscopic electrochemical parameters obtained in S100 as characteristic parameters, and taking the microscopic electrochemical parameters obtained in S200 as characteristic parameters 、 、 The three parameters are used as evaluation indicators of battery safety to construct a least squares support vector machine model;

[0194] Choose Gaussian radial basis function (RBF) as the kernel function:

[0195]

[0196] in:

[0197] , : Input vector, representing the microscopic electrochemical parameter characteristics of retired batteries (such as , , , , etc.);

[0198] : The square of the Euclidean distance between two eigenvectors measures the degree of parameter difference between different battery samples; the smaller the distance, the higher the sample similarity;

[0199] σ is a parameter that controls the width of the Gaussian function. The smaller σ is, the higher the model complexity is. σ determines the influence range of the sample point in the high-dimensional space:

[0200] The smaller σ is, the steeper the function curve is, the model is only affected by neighboring samples, and the complexity is high (prone to overfitting);

[0201] The larger σ is, the flatter the function curve is, distant samples will also affect the classification, and the model has strong generalization ability (possibly underfitting);

[0202] The formula maps the nonlinear classification problem in low-dimensional space to high-dimensional space, transforming it into a linear classification problem and resolving the nonlinear correlation between microscopic electrochemical parameters and safety levels. The similarity of eigenvectors is calculated using a Gaussian function, enabling the model to capture implicit correlations between parameters (such as the negative correlation between charge transfer resistance and the critical temperature of thermal runaway). Optimizing σ to 2.0 through 5-fold cross-validation balances model complexity and generalization capabilities on a retired battery dataset, ensuring evaluation accuracy.

[0203] S320, based on the RBF kernel function, the classification decision function of SVM is:

[0204]

[0205] in:

[0206] is the Lagrange multiplier (only support vectors are non-zero). For support vectors that are non-zero, it reflects the contribution of the support vector to the classification decision, and the optimal value is determined through training;

[0207] is the class label of the support vector (+1 = safe or -1 = unsafe);

[0208] It is the support vector, the key sample point closest to the classification hyperplane, which determines the location of the classification boundary. Its number is much smaller than the total number of samples, reflecting the "sparseness" advantage of SVM.

[0209] b is the bias term, which adjusts the position of the classification hyperplane to optimize the classification boundary;

[0210] sgn (sign function): outputs the classification result (+1 or -1) according to the positive or negative value in the brackets, corresponding to the security level judgment;

[0211] Formula through kernel function Calculate the similarity between the battery to be evaluated and the support vector, perform classification after weighted summation, and realize the identification of complex safety patterns; rely on only a small number of support vectors instead of all data, greatly reducing the computational complexity, and suitable for large-scale rapid evaluation of retired batteries; support vectors correspond to battery samples with typical safety characteristics, which can be used to analyze the common parameter characteristics of high-risk batteries (such as high charge transfer resistance + low );

[0212] S400, model training

[0213] Collect the microscopic electrochemical parameters and accelerated calorimetry (ARC) test results of retired power batteries to construct a training dataset (XT rain ,yT rain ), where XT rain is the feature matrix, yT rain is the corresponding category label (safe or unsafe); XT rain The feature matrix contains N groups of samples, each group of samples is a 9-dimensional vector ( , , , , , 、 、 、 ), yT rain The labels are verified by manual annotation or thermal runaway testing at the BSI level. The training dataset contains N = 1500 samples, including 750 each of lithium iron phosphate and ternary batteries, with a capacity range of 5-20Ah.

[0214] The SVM model is trained using the Sequential Minimal Optimization (SMO) algorithm to solve the model parameters by minimizing the following optimization problem:

[0215]

[0216] Subject to the following constraints:

[0217]

[0218] Among them, C is the penalty factor, which controls the degree of penalty for misclassification; i It is a slack variable that allows a certain degree of misclassification;

[0219] w (weight vector): determines the weight of each feature parameter on the classification, and optimizes the key safety parameters (such as 、 ) obtains a higher weight; each element of the weight vector w corresponds to the importance of the microscopic electrochemical parameters in S100, such as The larger the corresponding element value, the stronger the influence of this parameter on the classification decision.

[0220] b (bias term): Same as the bias in the decision function, adjusts the position of the classification hyperplane;

[0221] (Slack variable): allows some samples to violate the classification interval constraint (i.e. misclassification), ≥0 indicates the degree of misclassification of the i-th sample.

[0222] C (penalty factor), controls the degree of penalty for misclassification:

[0223] The larger C is, the less the model tolerates misclassification and tends to improve the accuracy of the training set (possibly overfitting);

[0224] The smaller C is, the model allows more misclassification and focuses more on maximizing the classification interval (possibly underfitting);

[0225] C=50 is the optimal penalty factor, balancing classification accuracy and generalization ability. C=50 was determined through 5-fold cross-validation, achieving a balance between accuracy (96.25%) and F1 score (0.94) on a test set of retired batteries, avoiding overfitting (e.g., when C=100, the test set accuracy dropped to 92%) or underfitting (e.g., when C=10, the accuracy dropped to 88%).

[0226] In the formula:

[0227] It is a regularization term that suppresses the weight vector from being too large, avoids model overfitting, and improves generalization ability;

[0228] To quantify the cost of misclassification and balance it with the regularization term, we used the optimal value of C = 50 to achieve a balance between classification accuracy and generalization ability on the retired battery dataset.

[0229] The function interval from the sample to the classification hyperplane is required to be at least 1- , ensuring the reliability of classification boundaries, Control the maximum misclassification allowed; Allowing the margin to shrink (i.e. misclassification), =0, the sample strictly meets the classification interval requirements.

[0230] The SMO algorithm iteratively optimizes the Lagrange multipliers of two samples , gradually solving the optimal w and b, which is suitable for efficient training of large-scale retired battery datasets.

[0231] Solve the above optimization problem by Lagrange multiplier method and get the Lagrange multiplier And the bias term b, the final classification decision function is:

[0232]

[0233] Input features x: dimensional microscopic electrochemical parameters extracted from S100 test ( , , , , ), after Z-score standardization; when Z-score standardization, The mean μ=40mΩ and the standard deviation σ=15mΩ.

[0234] Safety Category Label :The BSI level of S200 is mapped into two categories, BSI ≤ 0.7 corresponds to "safe" ( =+1), BSI>0.7 corresponds to "unsafe" ( =-1);

[0235] Support Vector : The number of key samples screened out after training (such as the characteristics of retired batteries with microcracks) accounts for 5%-10% of the total samples, reflecting the sparsity advantage of SVM;

[0236] By integrating the characteristic matrix X of electrochemical parameters, the formula improves the recognition rate of high-risk batteries from 78% of traditional SVM to 96.25%.

[0237] S500, safety level prediction:

[0238] a) Microscopic parameter input: The microscopic electrochemical parameters of the battery to be evaluated are input into the trained SVM model, and the model outputs a safety label (+1 = safe / -1 = unsafe);

[0239] b) BSI level mapping: Based on the BSI value calculated by S200, map the risk according to the level table;

[0240] c) Collaborative decision-making (the threshold is set according to the battery type, such as 0.6 for lithium iron phosphate and 0.7 for ternary):

[0241] When BSI ≤ threshold and safety label = +1, confirm the BSI risk level;

[0242] When BSI>threshold or safety label=-1, it is judged as high risk; even if BSI is at a low risk level, if the model label is unsafe, high risk is prioritized.

[0243] d) Data feedback: storing input parameters, BSI values, and decision results in the historical case library;

[0244] e) Model optimization: When new data is added to the case library (e.g., more than 100 items), the S400 model is triggered to retrain and update parameters.

[0245] Risk level table:

[0246] .

[0247] Combine Figure 2 , summarized as follows:

[0248] (1) Data collection collaboration (S100+S200);

[0249] The hardware test module performs S100 micro-test and S200 thermal test simultaneously:

[0250] Pulse charge and discharge (S110) are triggered synchronously with ARC test, and the timestamp error is less than 1ms;

[0251] The generated microscopic parameters and thermal parameters are automatically merged into a 9-dimensional feature vector;

[0252] (2) Model training collaboration (S300+S400)

[0253] The SVM model defined by S300 receives the microscopic parameters of S100 as input features and the BSI value of S200 as the label source:

[0254] BSI ≤ 0.7 → label 'safe' (y = +1);

[0255] BSI>0.7→label ‘unsafe’ (y=-1);

[0256] S400 uses a 9-dimensional feature vector normalization matrix during training;

[0257] (3) Decision-making and execution collaboration (S500)

[0258] The S500 receives two types of input simultaneously:

[0259] Input 1: original BSI value output by S200;

[0260] Input 2: microscopic parameters of S100 real-time test (normalized by Z-score);

[0261] Perform dual-channel decision making:

[0262] Channel A: BSI value is mapped to level IV risk;

[0263] Channel B: Micro-parameters are input to the SVM model to output safety labels;

[0264] Final judgment rules:

[0265] If BSI ≤ 0.6 and label = safe → confirm BSI level;

[0266] If BSI>0.6 or label=unsafe→determined as high risk";

[0267] (4) Closed-loop optimization collaboration

[0268] The decision results of S500 and actual battery application data are stored in the historical case library, triggering S400 model retraining: SVM parameters are automatically updated after 100 sets of data are added each month, and model version management records changes in accuracy.

[0269] Preferably, in S200, the capacity conversion formula can be used to obtain :

[0270]

[0271] Battery safety indicator formula optimization:

[0272]

[0273] This ensures that BSI calculations for batteries of different capacities are comparable;

[0274] Engineering application cases

[0275] Case 1: Comparison of lithium iron phosphate batteries with different capacities

[0276] Battery A: 10Ah, =24kJ, =24kJ, the exponential term is exp(−2424)= ≈0.368;

[0277] Battery B: 5Ah, =12kJ, =24×105=12kJ, the exponential term is exp(−1212)= ≈0.368;

[0278] although Different, but the index terms are the same after normalization, and the BSI calculation results are comparable.

[0279] Case 2: Comparison between lithium iron phosphate and ternary batteries

[0280] 10Ah lithium iron phosphate battery: =24kJ, exponential term ≈0.368;

[0281] 10Ah ternary battery: =40kJ, exponential term ≈0.368;

[0282] When the heat release energy per unit capacity of batteries of different systems is the same, the BSI index items are consistent, reflecting that their thermal runaway risks are comparable, which is consistent with actual safety characteristics.

[0283] It is the core parameter in the BSI formula for achieving "unified assessment across capacities and material systems". By linearly correlating the exothermic energy baseline value with the battery capacity, it eliminates the interference of capacity differences on safety assessments, making BSI a quantifiable and comparable safety indicator, and providing a standardized tool for large-scale safety assessments of retired batteries.

[0284] Example 2

[0285] Retired lithium iron phosphate battery (8Ah, 1000 cycles, safe status)

[0286] S100, micro-electrochemical parameter test, all electrochemical tests are carried out in a constant temperature box at 25℃±1℃

[0287] S110, pulse test

[0288] Conditions: SOC = 95%, 1C pulse current (8A), duration 10s;

[0289] Voltage drop ΔV=0.12V, dynamic internal resistance =15mΩ;

[0290] Polarization recovery time = 25s, (the time required for the voltage to recover from the lowest point to 90% of the initial value).

[0291] S120, Impedance Test

[0292] Equipment: Solartron 1260 electrochemical workstation, AC amplitude 5 mV, frequency range 100 Hz-100 kHz;

[0293] Parameter extraction:

[0294] The semicircle diameter in the high frequency region (100Hz~100kHz) corresponds to the charge transfer resistance =30mΩ;

[0295] Calculate the Warburg coefficient of the 45° slope in the low frequency region (<10Hz) =0.02Ω⋅ (pass );

[0296] S130, UU test

[0297] Process: 0.5C constant current charge to 3.65V, stand for 30 minutes and then 0.5C constant current discharge to 3.0V;

[0298] Parameter extraction:

[0299] Average charging voltage =3.5V, average discharge voltage =3.3V, polarization voltage difference ΔV=0.2V;

[0300] Charging capacity , discharge capacity ,Energy efficiency:

[0301] η= ;

[0302] S140, ICA test

[0303] Conditions: 0.2C constant current charge and discharge, sampling rate 10Hz, using OriginPro2021 software for Savitzky-Golay filtering (window 5, order 2), and differential calculation using the built-in algorithm;

[0304] Parameter extraction:

[0305] Phase transition peak intensity in the range of 3.45V±0.1V =0.8 (normalized), peak area =0.85 (normalized). Both are higher than the safety threshold ( ≥0.8, peak area ≥0.85).

[0306] S150, DV test

[0307] Process: 0.2C constant current charge and discharge covers 2.8V-3.65V, sampling rate 10Hz, and the differential voltage is calculated using the central difference method on the smoothed VQ data:

[0308] , where i is the data point index.

[0309] Parameter extraction: Abnormal peak area in the range of 3.6V±0.1V =0.025, which is lower than the lithium iron phosphate microcrack warning threshold (>0.03).

[0310] S200, thermal parameters and BSI calculation

[0311] ARC test: heating rate 1°C / min, initial temperature 25°C; ℃ (self-heating starting temperature), ℃ (critical temperature at a temperature rise rate of 1℃ / min), ℃ (maximum temperature), =20kJ (total energy released by thermal runaway).

[0312] Capacity conversion: =24× =19.2kJ.

[0313] BSI calculation: BSI= ≈1.23×0.35≈0.43 (level II, low risk);

[0314] S300, actual verification and safety assessment

[0315] Secondary utilization scenario: Home energy storage system (5kWh energy storage module), 200 charge and discharge cycles, operating temperature 25°C to 40°C; operating performance:

[0316] The charge and discharge efficiency is stable at 85%-87%, with no abnormal voltage fluctuations;

[0317] Infrared thermal imaging testing showed that the temperature difference on the battery surface was ≤5°C, with no local overheating.

[0318] After 200 cycles, the battery was immediately disassembled and the positive electrode was observed using a scanning electron microscope (SEM). No obvious microcracks were found. The electrolyte residue test showed that 85% remained (the new battery is 100%), which is consistent with the BSI assessment results.

[0319] Example 3

[0320] Ternary retired battery (12Ah, 1500 cycles, high risk)

[0321] S100, micro-electrochemical parameter test, all electrochemical tests are carried out in a constant temperature box at 25℃±1℃

[0322] S110, pulse test

[0323] Dynamic internal resistance =18mΩ(ΔV=0.216V,Pulse current =18mΩ);

[0324] Polarization recovery time .

[0325] S120, Impedance Test

[0326] Equipment: Solartron 1260 electrochemical workstation, charge transfer resistance =80mΩ (semicircle diameter in high frequency area, exceeding threshold value 60mΩ);

[0327] Warburg coefficient

[0328] S130, UU test

[0329] Polarization voltage difference ΔV=0.3V( );

[0330] Energy efficiency η=92%, η= ×100%≈92%).

[0331] S140, ICA test

[0332] Use OriginPro2021 to perform Savitzky-Golay filtering (window 5, order 2);

[0333] Phase transition peak intensity (normalized, lower than the ternary battery safety threshold of 0.75);

[0334] Peak area = 0.7 (normalized, below the safety threshold of 0.8).

[0335] S150, DV test

[0336] Central difference calculation , where i is the data point index.

[0337] Abnormal peak area =0.06 (3.8V±0.1V range, exceeding the microcrack warning threshold of 0.025).

[0338] S200, thermal parameters and BSI calculation

[0339] Heating rate 1℃ / min, initial temperature 25℃, obtain ℃ (self-heating starting temperature), ℃ (critical temperature at a temperature rise rate of 1℃ / min), ℃ (maximum temperature), =30kJ (total energy released by thermal runaway).

[0340] Capacity conversion: ==48kJ.

[0341] BSI calculation: BSI= ≈1.26×0.535≈0.67 (Level III, moderate risk); but due to and , the model triggers a high-risk warning. Even if the BSI is at Level III, if the model label is unsafe, the high-risk is prioritized to avoid conflicts between the rating table and the model decision.

[0342] S300, Model Evaluation and Disassembly Verification

[0343] Input the feature vector X=[18,80,0.3,0.7,0.06,140,200,270,48] into the trained SVM model (C=50,σ=2.0) and output the classification label =-1 (unsafe), judged as high risk. Model weight analysis shows and The weight of the index is as high as 45%, triggering a high-risk warning.

[0344] Used as a power source for low-speed vehicles in cascades, disassembled and tested after 300 cycles:

[0345] Scanning electron microscopy (SEM) observation of the positive electrode showed a microcrack density of 30 per mm² (less than 5 per mm² for new batteries);

[0346] The electrolyte residual test showed 65% remaining (a new battery is 100%);

[0347] The retest of thermal runaway showed that T2 was actually 180°C (20°C lower than the original test), confirming the high risk.

[0348] Summary of technical advantages

[0349] .

[0350] Comparative Example

[0351] Traditional thermal parameters are evaluated separately (using the lithium iron phosphate 8Ah in Example 2 as an example)

[0352] 1. Evaluation Objects and Methods

[0353] Evaluation object: retired lithium iron phosphate battery (8Ah, 1000 cycles, safe state) similar to that in Example 2.

[0354] Traditional methods:

[0355] Only the thermal parameters were obtained using Accelerating Rate Calorimeter (ARC) testing. , , , , ignoring the microscopic electrochemical parameters of S100 (such as 、 、 etc.), and no capacity normalization was performed (directly using =24kJ, not converted to ).

[0356] 2. BSI calculation process and results

[0357] ARC test thermal parameters: ℃ (self-heating starting temperature), ℃ (critical temperature at a temperature rise rate of 1℃ / min), ℃ (maximum temperature), =20kJ (total energy released by thermal runaway).

[0358] Traditional BSI calculation (not converted to capacity):

[0359] BSI= (Level II, low risk);

[0360] 3. Actual risks and assessment bias

[0361] Actual state of microscopic parameters (data from Example 2):

[0362] mΩ (≤50mΩ, safe), (≤0.03, no microcracks), (≥0.8, normal), all microscopic parameters are within the safety threshold, and the actual risk is consistent with the BSI level.

[0363] Causes of deviation in traditional methods:

[0364] Uncapacitated: Traditional methods do not use =24× =19.2kJ, resulting in an inflated BSI value of 0.11 (error 25.6%), but the actual risk was not concealed because the battery was in good condition.

[0365] Lack of micro-parameter support: cannot pass , Quantifying internal degradation (e.g., 85% electrolyte residue) using other parameters only relies on thermal parameter evaluation, which theoretically may miss high-risk situations (e.g., the ternary battery scenario in Example 3).

[0366] Results comparison and technical advantages

[0367]

[0368] Conclusion: Traditional methods rely solely on thermal parameters and do not normalize capacity. Although they can correctly assess the risk level in some normal batteries, they suffer from large calculation errors and are unable to identify microscopic degradation hazards. Examples 2 and 3 address the above issues through the following innovations:

[0369] Capacity normalization: Pass Eliminate the interference of capacity difference, and the BSI calculation error is less than 5%;

[0370] Micro-parameter fusion: through , Equalizing electrode degradation, the identification rate of high-risk batteries increased from 72% to 96.25%;

[0371] Machine learning modeling: Using SVM to capture nonlinear correlations between parameters, avoiding the linear evaluation limitations of traditional formulas, and providing more accurate safety guarantees for the cascade utilization of retired batteries.

[0372] It should be noted that, unless otherwise specified, the technical or scientific terms used in this application should have the common meanings understood by those skilled in the art to which this application belongs.

[0373] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A retired power battery safety index prediction device, characterized in that: include: The hardware testing module uses testing equipment to obtain the microscopic electrochemical parameters and thermal safety parameters of retired batteries, providing raw data support for safety assessments; The data processing and modeling module extracts features and standardizes the raw test data, and builds a security assessment model through machine learning algorithms to achieve the conversion from data to security index and evaluate the security level; The result output and interaction module displays the safety assessment results and provides decision support to assist in formulating strategies for the reuse of retired batteries; The raw data collected by the hardware testing module is transmitted to the data processing and modeling module, and a standardized feature vector is generated after feature extraction; The data processing and modeling module simultaneously runs the BSI formula calculation and SVM model prediction: the BSI formula outputs the safety level based on thermal parameters and capacity normalization; The SVM model classifies feature vectors through kernel functions and optimization parameters, and outputs safe / unsafe labels; the final results are summarized in the result output and interaction module, and combined with the historical case library to generate reports and decision recommendations.

2. The retired power battery safety index prediction device according to claim 1, characterized in that: The hardware testing module includes: Micro-electrochemical parameter testing instrument, used for high SOC pulse charge and discharge testing, electrochemical impedance testing and micro-parameter collection; Accelerated Calorimetry (ARC) tester, to obtain parameters related to thermal stability and thermal runaway, and to obtain the self-heating starting temperature , thermal runaway critical temperature 、 ; Temperature control device, used to maintain the power battery at a constant temperature or a stable heating rate; The data acquisition and control unit synchronously collects electrochemical and thermal parameters and generates raw data sets with time stamp alignment.

3. The retired power battery safety index prediction device according to claim 1, characterized in that: Data processing and modeling modules include: Feature extraction and preprocessing unit, which performs data processing, converts the raw data of the hardware test module into standardized feature vectors, and eliminates noise and dimensional differences; Safety index calculation engine, which calculates the safety index (BSI) based on thermal parameters and capacity normalization, provides basic risk level assessment, and outputs BSI value and level; The machine learning model and training unit use the SVM algorithm to build a nonlinear classification model to improve the accuracy of high-risk battery identification.

4. The retired power battery safety index prediction device according to any one of claims 1 to 3, characterized in that: Result output and interaction modules include: A security level visualization interface converts security assessment results into intuitive charts, annotates key risk parameters, and supports interactive analysis; Decision support database, providing historical case comparison and tiered utilization strategy recommendations to support intelligent decision-making; The interactive control and report generation unit supports user operations, data export and report generation, improving system usability.

5. A method for predicting the safety index of retired power batteries, characterized in that: include: S100, micro electrochemical parameter test, and extract micro electrochemical parameters, micro electrochemical parameters include dynamic internal resistance , polarization recovery time , charge transfer resistance , Warburg coefficient , polarization voltage difference , phase transition peak intensity of ICA curve , peak area of ​​ICA curve, abnormal peak area of ​​DV curve ; S200, safety performance index acquisition and calculation: Obtaining the self-heating onset temperature of retired power batteries through Accelerating Rate Calorimeter (ARC) testing , thermal runaway critical temperature , maximum temperature , total heat release energy of thermal runaway ( ) four parameters; Battery safety indicators It is described by the following formula: ； S300, model construction: S310, taking the microscopic electrochemical parameters obtained in S100 as characteristic parameters, and taking the microscopic electrochemical parameters obtained in S200 as characteristic parameters 、 、 The three parameters are used as evaluation indicators of battery safety to construct a least squares support vector machine model; Choose Gaussian radial basis function (RBF) as the kernel function: ； S320, based on the RBF kernel function, the classification decision function of SVM is: ； Safety Category Label : The BSI level of S200 is mapped into two categories, =1 corresponds to "safe", =-1 corresponds to "unsafe"; S400, model training: Collect the microscopic electrochemical parameters and accelerated calorimetry (ARC) test results of retired power batteries to construct a training dataset (XT rain ,yT rain ), where XT rain is the feature matrix, yT rain is the corresponding category label (safe or unsafe); XT rain The feature matrix contains N groups of samples, each group of samples is a 9-dimensional vector ( , , , , , 、 、 、 ), yT rain Labels are verified by BSI grade manual labeling or thermal runaway testing; The SVM model is trained using the Sequential Minimal Optimization (SMO) algorithm to solve the model parameters by minimizing the following optimization problem: ； Subject to the following constraints: ； Among them, C is the penalty factor, which controls the degree of penalty for misclassification; i It is a slack variable that allows a certain degree of misclassification; Solve the above optimization problem by Lagrange multiplier method and get the Lagrange multiplier And the bias term b, the final classification decision function is: ； Input features x: dimensional microscopic electrochemical parameters extracted from S100 test ( , , , , ), normalized by Z-score; S500, safety level prediction: The microscopic electrochemical parameters of the battery to be evaluated are input into the trained SVM model, which outputs a safety label and maps the risk according to the grade table based on the BSI value calculated by S200; When BSI ≤ threshold and safety label = +1, confirm the BSI risk level; When BSI>threshold or safety label=-1, it is judged as high risk.

6. The method for predicting the safety index of retired power batteries according to claim 5, characterized in that: S100, including: S110, performing a high SOC pulse charge and discharge test on the retired power battery to obtain a high SOC pulse charge and discharge test curve; Parameter extraction: dynamic internal resistance , polarization recovery time ; S120, performing a high SOC electrochemical impedance test on the retired power battery to obtain a high SOC electrochemical impedance test curve; Parameter extraction: Charge transfer resistance , Warburg coefficient , characterizes the interfacial reaction activity and ion diffusion rate; S130: Perform a charge voltage-discharge voltage test on the retired power battery to obtain a charge voltage-discharge voltage curve; Parameter extraction: polarization voltage difference ;in is the charging voltage, is the discharge voltage; S140, perform incremental capacity analysis test on retired power batteries to obtain ICA curve; Parameter extraction: differential capacity dQ / dV-V (ICA curve) phase transition peak intensity , peak area, identifying electrode material phase change and active material loss; S150, performing a differential voltage analysis (DV) test on the retired power battery to obtain a DV curve; Parameter extraction: abnormal peak area .

7. The method for predicting the safety index of retired power batteries according to claim 6, characterized in that: The S130 performs charge voltage and discharge voltage tests on retired power batteries, including: Charge the retired power battery with constant current to the cut-off voltage and record the average voltage during the charging process After standing still, discharge at a constant current until the cut-off voltage is reached, and record the average voltage during the discharge process. ; Converted energy efficiency: ; Energy efficiency η refers to the ratio of battery discharge energy to charging energy, reflecting the energy conversion efficiency of the battery during charging and discharging. The higher η is, the smaller the energy loss of the battery during charging and discharging, and the better the performance. Add η to optimize the calculation formula of BSI: ； The value of is used to replace BSI for judgment.

8. The method for predicting the safety index of retired power batteries according to claim 6, characterized in that: The S150 performs differential voltage analysis (DV) testing on retired power batteries, including: Charge and discharge control: Perform charge and discharge tests on retired power batteries at a constant current, covering the full SOC range; Data acquisition: Sampling rate ≥ 10 Hz, recording voltage-capacitance (VQ) raw data; Differential processing: After smoothing and filtering the VQ data, calculate the differential voltage: ; Curve drawing: Generate dV / dQ-Q curves (DV curves) and analyze abnormal peaks in specific voltage ranges; Abnormal peak area extract: Locating abnormal areas: Identifying peaks and valleys that deviate from normal morphology within characteristic voltage ranges; Area calculation: perform numerical integration on the abnormal peaks and calculate the area under the peaks. The unit is normalized area. The formula is: .

9. The method for predicting the safety index of retired power batteries according to any one of claims 5 to 8, characterized in that: S500 also includes: storing input parameters, BSI values, and decision results in a historical case library; when new data is added to the case library, triggering S400 model retraining and updating parameters.

10. The method for predicting the safety index of retired power batteries according to any one of claims 5 to 9, characterized in that: The capacity conversion formula is used in S200 to obtain : ； Battery safety indicator formula optimization: 。

Images