Online multi-physics coupled intelligent sensing and early warning method and system for lithium-ion batteries

Abstract

This invention relates to an intelligent sensing and early warning method and system for lithium-ion batteries using online multi-physics coupling. The method includes: synchronously acquiring real-time multi-physics parameter data of the lithium-ion battery online; intelligently constructing a multi-physics coupling model and automatically marking error anomalies in the individual lithium-ion battery; automatically solving the multi-dimensional state parameters of each individual lithium-ion battery, calculating inconsistencies between individual lithium-ion batteries and automatically marking consistency anomalies; identifying outlier individual lithium-ion batteries and marking them as outliers; establishing a comprehensive evaluation system for multi-physics parameters, intelligently determining the normal, suspicious, or fault levels of each individual lithium-ion battery, and outputting the final abnormal battery identification result. This invention can effectively detect and capture subtle changes in the internal performance degradation of the battery, improving the accuracy of battery state assessment, state boundary management capabilities, and fault diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, and in particular to an online multi-physics coupled intelligent sensing and early warning method and system for lithium-ion batteries. Background Technology

[0002] Lithium-ion battery technology is crucial for my country's energy system's transition to a low-carbon model, and has become a core energy storage device for new energy vehicle power batteries, new power system energy storage batteries, and consumer electronics batteries. However, lithium-ion batteries have operating temperature and state limits; exceeding these limits can easily accelerate battery degradation and even cause fires and explosions. Therefore, it is essential to sense, calculate, and provide early warnings regarding the operating state of lithium-ion batteries.

[0003] Existing methods primarily rely on battery management systems (BMS) to estimate the battery's internal state by detecting basic parameters such as voltage, current, and temperature. However, this approach has significant limitations: it cannot detect changes in battery surface pressure and internal electrochemical behavior, making it difficult to capture subtle changes in battery performance degradation. This results in poor accuracy in battery state assessment, insufficient management of battery state boundaries, and a need to improve the accuracy of battery fault diagnosis.

[0004] Therefore, there is an urgent need to explore new battery sensing and early warning technologies to meet the requirements for the safe and reliable operation of lithium-ion batteries. Summary of the Invention

[0005] This invention addresses the shortcomings of existing lithium-ion battery sensing and early warning methods, such as poor accuracy in battery state assessment and insufficient reliability in fault diagnosis. It proposes an online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries, which can effectively detect and capture subtle changes in battery internal performance degradation, improving the accuracy of battery state assessment, state boundary management capabilities, and fault diagnosis accuracy. This invention also relates to an online multi-physics coupled intelligent sensing and early warning system for lithium-ion batteries.

[0006] The technical solution of the present invention is as follows:

[0007] A smart sensing and early warning method for lithium-ion batteries using online multiphysics coupling, characterized by the following steps:

[0008] Online detection steps for multi-physics parameters of a single battery: Based on the geometric characteristics of lithium-ion batteries, a collaborative acquisition method of electrochemical-mechanical-thermal-electrical multi-physics fields is adopted to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of the single lithium-ion battery in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing. Real-time multi-physics parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance.

[0009] Intelligent modeling and error anomaly labeling steps: Real-time multi-physics parameter data of each individual lithium-ion battery collected online are automatically aggregated into a dataset; a multi-physics coupling model is constructed based on the coupling relationship between electrochemical, mechanical, thermal, and electrical multi-physics fields. The input of the multi-physics coupling model is the real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current, and the output is the predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery; by comparing the predicted values ​​of surface pressure and electrochemical impedance output by the model with the corresponding real-time multi-physics parameter data of surface pressure and electrochemical impedance collected online in real time, the prediction deviation of surface pressure and electrochemical impedance is calculated. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data combined with the heat generation law of lithium-ion batteries; if the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically annotated with an error anomaly.

[0010] Intelligent State Estimation and Consistency Anomaly Labeling Steps: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, a multi-dimensional state estimation algorithm is used to automatically solve the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency. Based on the multi-dimensional state parameters, statistical analysis and entropy weight method are used to calculate the inconsistencies between individual lithium-ion batteries, obtaining quantitative results of range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weights. Individual lithium-ion batteries that exceed the corresponding inconsistency thresholds are automatically labeled as consistency anomalies.

[0011] Intelligent outlier detection and outlier labeling steps: Based on the real-time multi-physics parameter data of each individual lithium-ion battery collected online, a hybrid weighted outlier detection algorithm is used for automatic screening. High weight coefficients are assigned to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure change, electrochemical impedance, and current to highlight their contribution to outlier determination, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers.

[0012] Intelligent abnormal battery identification steps: Establish a comprehensive evaluation system of multi-physics parameters, and automatically fuse the results of error anomaly labeling, consistency anomaly labeling, outlier anomaly labeling, and the quantitative results of prediction deviation corresponding to error anomalies and inconsistency corresponding to consistency anomalies through evaluation matrix, multi-dimensional coupled evaluation function or decision tree model. Intelligently determine the normal, suspicious or fault level of each individual lithium-ion battery and output the final abnormal battery identification result.

[0013] Preferably, in the online detection step of multi-physics parameters of a single battery cell, when a piezoresistive sensor is deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing, the piezoresistive sensor is deployed near the battery tab, at the center of the battery pressure surface, and / or near the battery pressure surface; the amplitude and frequency range of the electrochemical impedance excitation signal are adaptively adjusted according to the capacity and impedance characteristics of the single lithium-ion battery cell.

[0014] Preferably, in the online detection step of multi-physics parameters of a single battery, when performing online synchronous acquisition of real-time multi-physics parameter data, the storage resource configuration is optimized according to the amount of real-time multi-physics parameter data generated, the data transmission rate is optimized according to the characteristic electrochemical impedance frequency of the lithium-ion battery, and the pressure acquisition sensitivity and minimum resolution are optimized according to the pressure change amplitude of the battery surface, so as to improve the efficiency of online synchronous acquisition of real-time multi-physics parameters.

[0015] Preferably, in the intelligent modeling and error anomaly labeling step, the multiphysics coupling model is: a discretized model obtained by reducing the order of a model constructed based on the Pade approximation method with the electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries as the core; or a model of electrical equivalent circuit-thermal equivalent circuit-mechanical equivalent circuit coupling based on the equivalent circuit principle; or a black box model obtained by training a neural network. The training process of the black box model is as follows: using the electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries to generate battery parameters under different working conditions, establishing a virtual dataset and dividing it into a training set and a test set, and using a temporal convolutional network-bidirectional gated unit neural network as the basic model for training. The virtual dataset is used for offline training of the black box model.

[0016] Preferably, in the intelligent modeling and error anomaly labeling step, the rationality verification of the predicted internal temperature value of the lithium-ion battery specifically involves: determining a reasonable range of internal temperature by combining real-time ambient temperature parameter data with the heat generation law of the lithium-ion battery, including the generation and conduction characteristics of ohmic heat, polarization heat, and reversible heat; the upper limit of the reasonable range is the sum of the ambient temperature and the maximum allowable temperature rise of the lithium-ion battery, and the lower limit is the difference between the ambient temperature and the minimum allowable temperature drop of the lithium-ion battery.

[0017] Preferably, in the intelligent state estimation and consistency anomaly labeling step, the multi-dimensional state estimation algorithm is one or more combinations of Kalman filtering algorithms, adaptive filtering algorithms, or machine learning algorithms; the corresponding thresholds for range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weights are all determined after statistical analysis and calibration based on the factory consistency parameters, cycle life test data, and multi-condition operation history data of the same batch of single lithium-ion batteries.

[0018] Preferably, in the intelligent outlier detection and outlier labeling step, the hybrid weighted outlier detection algorithm is:

[0019] Based on the DBSCAN outlier detection method, a parameter threshold determination mechanism is added. Individual lithium-ion batteries that exceed the preset parameter threshold are directly determined as outliers and do not need to participate in subsequent density clustering calculations.

[0020] Alternatively, based on the support vector machine classification and detection method, an expert weight vector can be introduced to assign differentiated weights to different multiphysics parameters, and outlier detection can be achieved by combining distance threshold judgment.

[0021] Preferably, in the intelligent abnormal battery identification step, when abnormal identification is achieved through a decision tree model, it specifically includes:

[0022] The results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, along with corresponding prediction bias and quantification results, are collected to construct a comprehensive feature vector. Based on historical operational data, cycle life test data, and simulation data, real fault labels are labeled, a decision tree model is trained, and the information gain of each feature is calculated to determine the optimal partitioning rule. The comprehensive feature vector is input into the trained decision tree model, and the anomaly fusion confidence score is output. Based on the confidence score, three anomaly levels are defined: confidence score < 0.5 corresponds to normal batteries, 0.5 ≤ confidence score < 0.8 corresponds to suspicious batteries, and confidence score ≥ 0.8 corresponds to faulty batteries.

[0023] An online multi-physics coupled intelligent sensing and early warning system for lithium-ion batteries is characterized by comprising, in sequence, an online detection module for multi-physics parameters of a single cell, an intelligent modeling and error anomaly labeling module, an intelligent state estimation and consistency anomaly labeling module, an intelligent outlier detection and outlier anomaly labeling module, and an intelligent abnormal battery identification module.

[0024] The online detection module for multi-physics parameters of a single battery cell: Based on the geometric characteristics of lithium-ion batteries, it adopts a collaborative acquisition method of electrochemical-mechanical-thermal-electrical multi-physics fields to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of a single lithium-ion battery cell in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing, and real-time multi-physics parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance.

[0025] The intelligent modeling and error anomaly labeling module automatically aggregates real-time multi-physics parameter data of each individual lithium-ion battery collected online to form a dataset. Based on the coupling relationship between electrochemical, mechanical, thermal, and electrical multi-physics fields, a multi-physics coupling model is constructed. The input of this model is real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current. The output is the predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery. By comparing the predicted values ​​of surface pressure and electrochemical impedance output by the model with the corresponding real-time multi-physics parameter data of surface pressure and electrochemical impedance collected online, the prediction deviation of surface pressure and electrochemical impedance is calculated. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data and the heat generation law of the lithium-ion battery. If the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically annotated with an error anomaly.

[0026] The intelligent state estimation and consistency anomaly labeling module: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, it automatically solves the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency, through a multi-dimensional state estimation algorithm; and based on the multi-dimensional state parameters, it calculates the inconsistencies between individual lithium-ion batteries through statistical analysis and entropy weight method, and obtains the quantitative results of range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weight; and automatically labels the consistency anomalies of individual lithium-ion batteries that exceed the corresponding inconsistency thresholds.

[0027] The intelligent outlier detection and outlier labeling module: Based on the real-time multi-physics parameter data of each individual lithium-ion battery collected online, it uses a hybrid weighted outlier detection algorithm for automatic screening. It assigns high weight coefficients to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure change, electrochemical impedance, and current to highlight their contribution to outlier determination, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers.

[0028] The intelligent abnormal battery identification module establishes a comprehensive evaluation system of multi-physics parameters. It automatically integrates the results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, as well as the quantitative results of prediction deviation corresponding to error anomalies and inconsistency corresponding to consistency anomalies, through evaluation matrix, multi-dimensional coupled evaluation function or decision tree model. It intelligently determines the normal, suspicious, or fault level of each individual lithium-ion battery and outputs the final abnormal battery identification result.

[0029] Preferably, the online detection module for multi-physics parameters of a single battery cell includes a slave-controlled acquisition unit, an electrochemical acquisition and calculation unit, a pressure acquisition and conversion unit, and a signal communication unit.

[0030] The signal communication unit receives the signal detection command issued by the battery management system and forwards it to the slave control acquisition unit. The slave control acquisition unit activates the multi-physics field acquisition function, sending an enable signal to the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit, while simultaneously acquiring the voltage, current, and ambient temperature parameters of the tested lithium-ion battery cell. After receiving the enable signal, the electrochemical acquisition and calculation unit sends a frequency conversion excitation signal from the excitation source. This signal is amplified to the expected amplitude by the excitation signal amplifier and applied to the tested lithium-ion battery cell. The data processor acquires the response of the tested lithium-ion battery cell to the amplified excitation signal and calculates the electrochemical impedance. After receiving the enable signal, the pressure acquisition and conversion unit detects the surface pressure of the tested lithium-ion battery cell through the piezoresistive sensor and outputs the corresponding resistance signal. Combined with the sensor's nominal resistance, the signal comparator calculates the resistance change and then converts it into the surface pressure signal of the tested lithium-ion battery cell. The electrochemical impedance and surface pressure signals after signal acquisition and processing by the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit are sent to the slave control acquisition unit. Together with the voltage, current, and ambient temperature parameters acquired by the slave control acquisition unit, they are fed back to the signal communication unit as real-time multi-physics field parameter data. The signal communication unit then sends the data to the battery management system or other slave control acquisition units.

[0031] The technical effects of this invention are as follows:

[0032] This invention relates to an online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries. Its online detection step for multi-physics parameters of a single battery covers the core dimensions of electrochemistry, force, heat, and electricity through collaborative acquisition of multiple physics fields. Combined with the fixed-point layout based on the battery's geometric characteristics and the design of adaptive excitation signals, it ensures the targeted and comprehensive nature of parameter acquisition. The use of frequency conversion processing of electrochemical impedance and comparison and conversion methods of surface pressure improves the acquisition accuracy of key parameters, providing high-quality data input for subsequent modeling, estimation, and detection steps. It achieves synchronous online acquisition of multiple parameters, breaking the information limitations of single physics field parameters and laying a data foundation for capturing the complex coupled changes in battery state. Its intelligent modeling and... The error anomaly labeling step constructs a model based on the coupling relationship of multiple physics fields, accurately depicting the interaction law between parameters and improving the prediction reliability of key states such as internal temperature and surface pressure. Through real-time comparison of predicted and measured values ​​and verification of the rationality of internal temperature, it achieves automatic identification of error anomalies, captures potential battery problems reflected by model deviations in advance, and sets up an automated data collection and anomaly labeling process to reduce manual intervention, improve the real-time performance and efficiency of anomaly identification, and avoid the accumulation of potential faults. Its intelligent state estimation and consistency anomaly labeling steps comprehensively cover core parameters such as state of charge and health state through multi-dimensional state estimation algorithms, presenting a complete picture of the battery's operating state and breaking through the evaluation limitations of single state parameters. This system combines statistical analysis with entropy weighting for quantitative evaluation, enabling precise measurement of inconsistencies between individual cells. Multi-dimensional quantitative results enhance the scientific rigor of anomaly detection, allowing for targeted identification of cells with inconsistent performance. This prevents battery pack performance degradation or safety risks due to widening individual cell differences, extending the overall battery pack lifespan. Its intelligent outlier detection and anomaly labeling steps employ a hybrid weighted outlier detection algorithm for automatic screening. High-weight coefficients are assigned to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure mutation, electrochemical impedance anomaly, and current anomaly to highlight their contribution to outlier detection, strengthen the priority of identifying core risk factors, improve the accuracy of outlier anomaly identification, and enable rapid screening. The system can promptly detect sudden faults such as abnormal ambient temperature and sudden changes in surface pressure, shortening fault response time. Based on comprehensive screening of multiple physical field parameters, it avoids misjudgments caused by fluctuations in a single parameter, improving the stability and reliability of outlier identification. Its intelligent abnormal battery identification steps establish a multi-dimensional evaluation system and integrate three types of labeled results and quantitative data to achieve comprehensive integration of abnormal information, avoiding the one-sidedness of single-dimensional evaluation. It provides multiple fusion methods such as evaluation matrix and decision tree model to adapt to different application scenarios, improve the flexibility and applicability of the method, and output three-level judgment results of normal, suspicious, and faulty, achieving accurate differentiation of the degree of abnormality and providing a clear basis for the formulation of subsequent maintenance strategies.This invention presents an online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries. It constructs a full-process intelligent early warning system encompassing "acquisition-modeling-estimation-detection-fusion," achieving closed-loop management from parameter acquisition to anomaly detection, comprehensively covering potential risks throughout the battery's entire lifecycle. The design of multi-physics collaboration and multi-algorithm fusion overcomes the limitations of traditional single-dimensional early warning, significantly improving the comprehensiveness, accuracy, and real-time performance of anomaly identification, with recall and precision rates significantly superior to traditional methods. The automated process reduces reliance on manual labor, lowers maintenance costs, and effectively prevents safety accidents caused by battery failures by providing early warnings of battery anomalies, extending battery pack lifespan and improving the safety and economy of battery applications. Adaptive design to different individual battery characteristics and flexible fusion evaluation methods enhance the method's versatility, enabling its widespread application in various online monitoring scenarios for lithium-ion batteries.

[0033] By clearly defining the placement of piezoresistive sensors in areas of concentrated current stress, such as near the battery tabs and at the center of the pressure surface, key pressure change signals during battery operation can be accurately captured, improving the targeting and accuracy of pressure sensing. Furthermore, by adaptively adjusting the amplitude and frequency range of the electrochemical impedance excitation signal based on the individual battery capacity and impedance characteristics, acquisition errors caused by mismatch between the excitation signal and battery characteristics are avoided, ensuring the reliability of the electrochemical impedance parameters and providing high-quality multiphysics parameter inputs for subsequent modeling, state estimation, and other steps.

[0034] Furthermore, by optimizing storage resource configuration based on the amount of multi-physics parameter data generated, optimizing data transmission rate by combining characteristic electrochemical impedance frequency, and optimizing acquisition sensitivity and resolution based on pressure change amplitude, efficient resource utilization in the acquisition process is achieved, avoiding storage redundancy or transmission lag issues. At the same time, the acquisition adaptability of key parameters is improved in a targeted manner, significantly improving the overall efficiency of online synchronous acquisition of multi-physics parameters while ensuring acquisition accuracy, providing efficient data support for subsequent real-time modeling and anomaly labeling.

[0035] Furthermore, it provides three types of multiphysics coupling models to choose from: discretized models, equivalent circuit coupling models, and black-box models, to adapt to different application scenarios and accuracy requirements, thereby enhancing the flexibility and applicability of the method. Among them, the black-box model is trained on an offline virtual dataset and combines the strong temporal feature extraction capabilities of temporal convolutional networks and bidirectional gating units, which not only ensures the model's prediction accuracy but also avoids the resource consumption of online training. All three types of models closely adhere to the core of electrochemical-mechanical-thermal-electric coupling, accurately characterizing the interaction laws between parameters and providing a reliable model foundation for error anomaly labeling.

[0036] Furthermore, it is clarified that the verification of internal temperature rationality should be combined with the heat generation laws including ohmic heat, polarization heat, and reversible heat to comprehensively cover the core sources of battery heat generation and ensure that the determination of the reasonable range of internal temperature is scientific and accurate. The upper and lower limits of the range are defined by the difference between the ambient temperature and the maximum allowable temperature rise / fall, making the internal temperature verification standard clear and quantifiable, effectively avoiding abnormal misjudgments caused by internal temperature prediction deviations, further improving the accuracy of error anomaly labeling, and providing strong support for battery safety boundary management.

[0037] Furthermore, the specific types of multi-dimensional state estimation algorithms are limited to include Kalman filtering, adaptive filtering, and machine learning algorithms. This ensures both the diversity and adaptability of state parameter solutions and enhances the clarity of algorithm implementation. By determining the inconsistency threshold through factory parameters of batteries from the same batch, cyclic test data, and historical data under multiple operating conditions, and combining statistical analysis and calibration processes, the threshold setting is made more in line with actual application scenarios, significantly improving the scientificity and reliability of consistency anomaly labeling and effectively identifying potential differences between individual cells.

[0038] Furthermore, both hybrid weighted outlier detection algorithms have specifically optimized the outlier identification logic: DBSCAN, combined with a parameter threshold determination mechanism, directly identifies batteries exceeding the threshold as outliers, significantly improving the detection convergence speed; Support Vector Machine introduces expert weight vectors and combines them with a distance threshold to strengthen the contribution of key parameters and improve detection accuracy. The two algorithms each have their own focus, ensuring both the high efficiency of outlier anomaly identification and avoiding the limitations of a single detection method, comprehensively and quickly capturing various types of outlier batteries and improving the timeliness of anomaly warnings.

[0039] Furthermore, by constructing a comprehensive feature vector containing three types of annotation results—error anomaly annotation, consistency anomaly annotation, and outlier anomaly annotation—along with corresponding quantitative data, comprehensive integration of anomaly information is achieved, avoiding the one-sidedness of single-dimensional evaluation. A decision tree model is trained based on real fault labels, and the optimal partitioning rule is determined through information gain, giving the model strong classification ability and adaptability. By defining the confidence threshold of three levels of anomaly, the degree of anomaly can be accurately distinguished, clarifying the risk level of the faulty battery and providing a clear basis for subsequent maintenance strategy formulation, significantly improving the accuracy and practicality of anomaly battery identification.

[0040] This invention also relates to an online multi-physics coupled intelligent sensing and early warning system for lithium-ion batteries. Corresponding to the aforementioned online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries, this system can be understood as a system that implements the aforementioned online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries. The system includes, in sequence, a single-cell multi-physics parameter online detection module, an intelligent modeling and error anomaly labeling module, an intelligent state estimation and consistency anomaly labeling module, an intelligent outlier detection and outlier anomaly labeling module, and an intelligent abnormal battery identification module. These modules work collaboratively to achieve fully automated closed-loop management of the entire process: parameter acquisition, modeling and prediction, state estimation, outlier detection, and fusion identification. It can synchronously acquire multi-physics coupled parameters such as electrochemical impedance, surface pressure, voltage, current, and ambient temperature online, effectively detecting and capturing lithium-ion battery abnormalities. This system detects subtle changes in battery performance degradation, overcoming the limitations of traditional single-physical-parameter detection and enhancing the comprehensive analysis of battery internal performance evolution. Through the synergistic integration of a multi-physics coupling model, a multi-dimensional state estimation algorithm, and a hybrid weighted outlier detection algorithm, it significantly improves the accuracy of battery state estimation and the precision of anomaly identification, strengthens the management of battery state boundaries, and thus enhances the accuracy of battery fault diagnosis, effectively mitigating safety risks. The online multi-physics parameter detection module for individual cells can rationally arrange sensors according to the battery's geometry and current stress concentration characteristics, adapting to different capacities and types of lithium-ion batteries, enhancing the system's versatility and adaptability. It can output three levels of early warning results—normal, suspicious, and fault—based on the fusion of multi-source anomaly information, providing a clear basis for subsequent maintenance strategy formulation, balancing battery operational safety and maintenance economy. Attached Figure Description

[0041] Figure 1 This is a flowchart of the intelligent sensing and early warning method for lithium-ion batteries based on online multi-physics coupling according to the present invention.

[0042] Figure 2 This is a schematic diagram of electrochemical side reactions on the surface of the negative electrode of a lithium-ion battery.

[0043] Figure 3 This is a schematic diagram of the lumped parameter thermal model of a lithium-ion battery.

[0044] Figure 4 A schematic diagram of the electrode concentration gradient during the charging and discharging process of a lithium-ion battery.

[0045] Figure 5 This is a structural block diagram of the intelligent sensing and early warning system for lithium-ion batteries with online multi-physics coupling according to the present invention.

[0046] Figure 6 This is a preferred structural block diagram of the online detection module for multiple physical field parameters of a single cell in this invention. Detailed Implementation

[0047] The present invention will now be described with reference to the accompanying drawings.

[0048] This invention provides an online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries. First, based on the geometric characteristics of lithium-ion batteries, an online detection device for individual battery parameters using electrochemical-mechanical-thermal-electrical multi-physics coupling is designed. The magnitude of the electrochemical impedance excitation is designed based on the individual battery capacity and impedance range, and the placement of pressure sensors is designed based on the location of current stress concentration in the battery. Subsequently, the parameter information of each individual battery is aggregated into the battery management system via a CAN bus or other communication method. A multi-physics coupled battery model is established in the battery system. The error is calculated by comparing the model's predicted values ​​with the actual detected values. Batteries with errors exceeding a threshold are marked as P1. Furthermore, based on the battery's multi-physics parameters, battery state estimation is performed, followed by calculation and analysis of inconsistencies in individual cells. Cells exceeding the inconsistency threshold are marked as P2. Then, a hybrid weighted outlier detection method is designed within the battery management system to detect outliers in the multi-physics parameters of all cells, marking detected outliers as P3. Finally, a comprehensive evaluation standard for battery multi-physics parameters is established, combining P1, P2, and P3 values ​​to output the final identified abnormal cells. This effectively detects and captures subtle changes in battery internal performance degradation, significantly improving the accuracy of battery state assessment and boundary management capabilities, thereby enhancing the accuracy of battery fault diagnosis. Figure 1 As shown, it includes the following steps:

[0049] I. Online Detection Steps for Multi-Physical Field Parameters of Single Cell: Based on the geometric characteristics of lithium-ion batteries, an electrochemical-mechanical-thermal-electrical multi-physical field collaborative acquisition method is adopted to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of the single lithium-ion battery in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing. Real-time multi-physical field parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance.

[0050] Furthermore, when deploying piezoresistive sensors for fixed-point pressure sensing in areas of concentrated current in lithium-ion batteries, the preferred placement locations for the piezoresistive sensors can be near the battery tabs, the center of the battery pressure surface, and / or near the battery pressure surface. The amplitude and frequency range of the electrochemical impedance excitation signal are adaptively adjusted according to the capacity and impedance characteristics of the individual lithium-ion battery.

[0051] In addition, when performing online synchronous acquisition of real-time multi-physics parameter data, the storage resource configuration can be optimized based on the amount of real-time multi-physics parameter data generated, the data transmission rate can be optimized based on the characteristic electrochemical impedance frequency of lithium-ion batteries, and the pressure acquisition sensitivity and minimum resolution can be optimized based on the pressure change amplitude of battery surface, so as to improve the efficiency of online synchronous acquisition of real-time multi-physics parameters.

[0052] II. Intelligent Modeling and Error Anomaly Labeling Steps: The real-time multi-physics parameter data of each individual lithium-ion battery collected online are automatically aggregated to form a dataset. A multi-physics coupling model is constructed, centered on the coupling relationships between electrochemical, mechanical, thermal, and electrical multi-physics fields. The inputs to this model are real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current. The outputs are predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery. The predicted values ​​of surface pressure and electrochemical impedance output by the model are compared in real-time with the corresponding real-time multi-physics parameter data collected online to calculate the prediction deviations. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data and the heat generation law of the lithium-ion battery. If the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically labeled with an error anomaly. This step can also be called the anomaly battery labeling step based on the multi-physics coupling model.

[0053] Specifically, the real-time multi-physics parameter data of each individual lithium-ion battery can be aggregated into the battery management system via CAN bus or other communication methods. A multi-physics coupling model of the battery is established in the battery system. The error is calculated by comparing the model prediction value and the actual detection value. Batteries with errors exceeding the threshold are marked as P1.

[0054] The multiphysics coupling model can be: a discretized model obtained by reducing the order of a model built on the core electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries using the Pade approximation method; a model based on the equivalent circuit principle that couples electrical equivalent circuits, thermal equivalent circuits, and mechanical equivalent circuits; or a black-box model trained using neural networks. The training process for the black-box model is as follows: battery parameters under different operating conditions are generated using the electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries; a virtual dataset is established and divided into training and testing sets; a Temporal Convolutional Network (TCN)-Bidirectional Gated Recurrent Unit (Bi-GRU) neural network is used as the basic model for training; and the virtual dataset is used for offline training of the black-box model. Furthermore, the rationality of the predicted internal temperature of the battery is verified based on real-time ambient temperature parameter data and the heat generation law of the lithium-ion battery. Specifically, the rationality verification of the predicted internal temperature of the lithium-ion battery is as follows: by using real-time ambient temperature parameter data and combining the heat generation law of the lithium-ion battery, including the generation and conduction characteristics of ohmic heat, polarization heat, and reversible heat, a reasonable range of internal temperature is determined; wherein, the upper limit of the reasonable range is the sum of the ambient temperature and the maximum allowable temperature rise of the lithium-ion battery, and the lower limit is the difference between the ambient temperature and the minimum allowable temperature drop of the lithium-ion battery.

[0055] (1) Construction of training and test datasets

[0056] To train a black-box model based on TCN–Bi-GRU, a multi-condition dataset containing input-output correspondences is first constructed. The data can be derived from experimental test data of actual batteries under different operating conditions. The input to the black-box model is the battery terminal voltage. Battery current and ambient temperature The model outputs EIS data of battery electrochemical impedance spectroscopy. and battery surface pressure Based on the input and output of the black-box model, construct the representation of the battery training set u and the label y:

[0057]

[0058] (2) TCN–Bi-GRU cascade

[0059] Based on the training set constructed above, a black-box network architecture model with the structure of input → TCN module → Bi-GRU module → fully connected output layer is built.

[0060] 1) Input layer

[0061] The input features at each time step are three-dimensional vectors:

[0062]

[0063] The resulting sequence shape is , where L is the number of time steps.

[0064] 2) TCN module

[0065] One-dimensional causal convolution and dilated convolution are used to convolve the temporal dimension, stacking multiple layers of convolution-activation-residual units to extract multi-scale temporal features, resulting in a product of the same length. Feature sequences:

[0066]

[0067] 3) Bi-GRU module

[0068] Will Input a bidirectional GRU network and obtain the forward hidden state at each time step. and reverse hidden state The result obtained by piecing together:

[0069]

[0070] Take the last time step Features:

[0071]

[0072] 4) Output layer

[0073] Will Input one or two fully connected layers:

[0074]

[0075] in These are the model's predicted values. This is the output mapping function of the battery black-box model, and its general expression is:

[0076]

[0077] Where W and b are the parameters to be fitted.

[0078] (3) Loss function design

[0079] To simultaneously optimize the prediction accuracy of internal temperature and surface pressure, a multi-output mean square error (MSE) loss is employed:

[0080]

[0081] in It is an adjustable weight.

[0082] (4) Abnormal battery calibration

[0083] 1) Simulation results based on black-box model Compared with the measured results Calculate the error rate between simulation results and measured results. The calculation formula is as follows:

[0084] ))

[0085] Where i represents the i-th battery cell, and g() represents the processing function for impedance data at different frequencies based on EIS sampling results at the same time, and its expression is:

[0086]

[0087] 2) If the error rate If the error is greater than 0.05, the cell is labeled as P1, and the error rate is recorded. .

[0088] III. Intelligent State Estimation and Consistency Anomaly Labeling Steps: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, a multi-dimensional state estimation algorithm is used to automatically solve for the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency. Based on these multi-dimensional state parameters, statistical analysis and entropy weighting are used to calculate the inconsistencies between individual lithium-ion batteries, obtaining quantitative results for range inconsistency, variance inconsistency, and entropy weighting inconsistency based on expert weights. Individual lithium-ion batteries exceeding their respective inconsistency thresholds are automatically labeled as consistency anomalies. In other words, this step estimates the battery state based on the real-time multi-physics parameter data of each individual lithium-ion battery, then calculates and analyzes the inconsistencies of individual lithium-ion batteries, and marks batteries exceeding the inconsistency threshold as P2.

[0089] Battery state estimation refers to the calculation of the battery's state of charge (SOC), state of health (SOH), state of energy (SOE), ohmic internal resistance (Ro), and battery impedance (Zw) at a specific frequency using a multiphysics coupling model. Multidimensional state estimation algorithms can be one or more combinations of Kalman filtering algorithms, adaptive filtering algorithms, or machine learning algorithms. Specific state estimation methods include using a discretized model of the electrochemical-thermal-mechanical coupling model to estimate the ion concentrations at the positive and negative electrodes, thereby calculating the battery's SOC, and calculating the maximum ion concentrations at the positive and negative electrodes to calculate the battery's SOH. Other methods include establishing a state-space equation based on an equivalent circuit model and using a volumetric particle filter for joint estimation of battery SOC-SOH-SOE. The inconsistencies in individual lithium-ion batteries include range inconsistencies, variance inconsistencies, and inconsistencies based on entropy weighting with expert weights. Furthermore, the thresholds for range inconsistency, variance inconsistency, and entropy weighting inconsistency based on expert weights are all determined through statistical analysis and calibration based on the factory consistency parameters, cycle life test data, and multi-condition operation history data of the same batch of single lithium-ion batteries.

[0090] This step can be described as abnormal battery labeling based on battery SOX:

[0091] (1) Battery state estimation based on multiphysics parameters

[0092] 1) Construction of a multiphysics model for batteries

[0093] ① Calculation of solid-phase lithium-ion concentration

[0094] A spherical coordinate system is established with the center of the spherical particles of the positive and negative electrode active materials as the origin. According to Fick's law, the coordinate system... The expression for the solid-phase lithium ion concentration at a given location is shown in the following formula:

[0095]

[0096] Subscript This indicates a solid phase region. or These correspond to negative and positive pole particles, respectively; This refers to the concentration of lithium ions in the solid phase. The lithium-ion diffusion coefficient of the positive and negative electrodes in the solid phase region; For time.

[0097] An ideal spherical particle is centrally symmetric, therefore there is no flux at the particle center, and the boundary conditions at the particle center satisfy the following equation:

[0098]

[0099] At the surface of spherical particles, the current density (i.e., the total net current per unit volume of electrode) of the electrochemical reaction occurring at the solid / liquid interface is determined by the rate of change of lithium-ion surface concentration. Therefore, the boundary conditions at the surface of spherical particles satisfy the following equation:

[0100]

[0101] in, The equivalent radius of the positive and negative polar spherical particles. The volumetric current density for lithium-ion insertion or extraction; It is Faraday's constant. It is the active specific surface area (the total surface area of ​​active material per unit volume).

[0102] For an ideal spherical particle, the active specific surface area can be approximated by the following formula:

[0103]

[0104] in, This represents the volume fraction of the active materials for both the positive and negative electrodes (the volume ratio of the active material to the electrode). Considering that the electrode also contains substances such as binders, therefore... The value is less than the volume fraction of the solid phase.

[0105] When considering the SEI film thickening and lithium plating side reactions at the negative electrode, in addition to the volume current density of lithium delithiation and lithium intercalation, the current of the side reactions must also be considered. Therefore, at the negative electrode:

[0106]

[0107] Since the side reaction processes of the positive electrode active material are not considered, the total current in the positive electrode material is entirely formed by the delithiation and lithium insertion processes.

[0108] ② Calculation of liquid-phase lithium ion concentration

[0109] The movement of lithium ions in an electrolyte is influenced by two factors: firstly, the presence of a lithium ion concentration gradient in the liquid phase leads to diffusion; secondly, the migration of lithium ions under an electric field. The lithium ion concentration in the liquid phase satisfies the following equation:

[0110]

[0111] Subscript or These correspond to the positive and negative electrodes and the diaphragm, respectively. This represents the concentration of lithium ions in the liquid phase. This represents the volume fraction of the liquid electrolyte.

[0112] The effective diffusion coefficient of lithium ions in the liquid phase can be calculated by the following formula:

[0113]

[0114] in denoted as the diffusion coefficient of lithium ions in the liquid phase. is the Bruggman coefficient, which is the pore structure coefficient of media such as electrodes and separators, and is used to correct the liquid phase transport performance of lithium ions in tortuous porous materials.

[0115] It should be noted that when considering side reactions such as SEI film formation, the positive and negative electrodes... It will decline as the aging process intensifies and needs to be corrected.

[0116] Considering that there is no lithium-ion flux in the current collector in the P2D model, the boundary condition for the liquid-phase lithium-ion concentration is readily known as follows:

[0117]

[0118] ③ Calculation of solid-state potential

[0119] The solid-phase potential distribution in the positive and negative electrode materials is shown in the following equation:

[0120]

[0121] in For solid-state potential distribution, The equivalent conductivity in a solid material is... Similarly, it also needs to be corrected by the Bruggman coefficient, the expression of which is shown in the following formula:

[0122]

[0123] in The solid-phase conductivity of the electrode material.

[0124] Considering that the charge flux at the connection between the positive and negative electrodes and the current collector corresponds to the total battery current, the boundary condition at the current collector is as follows:

[0125]

[0126] in Here, the charging direction is defined as the positive direction, where the current is the charge. This represents the total area of ​​the electrodes.

[0127] Considering that there is no charge flux at the connection between the positive and negative electrodes and the membrane, the boundary conditions at the membrane can be obtained as follows:

[0128]

[0129] ④ Calculation of liquid phase potential

[0130] In the P2D model, the liquid phase potential distribution of a lithium-ion battery satisfies Ohm's law and the theory of concentrated solutions, as shown in the following expression:

[0131]

[0132] in The liquid phase potential, Corresponding to the ideal gas constant, T corresponds to the battery temperature. The effective conductivity of the electrolyte. The ion activity coefficient corresponding to the electrolyte.

[0133] The calculation method and and Similar, such as:

[0134]

[0135] in This represents the conductivity of the electrolyte.

[0136] Considering that there is no charge flux in the liquid phase at the interface between the battery and the current collector, the boundary condition for the liquid phase potential is as follows:

[0137]

[0138] ⑤ Delithiation and Lithium Intercalation

[0139] Lithium insertion / extraction occurs at the electrolyte / electrode interface, and the current density generated by the lithium insertion / extraction reaction on the particle surface can be calculated using the Butler-Volmer equation:

[0140]

[0141] in This represents the electrode reaction exchange current density during the lithium insertion / extraction process. This represents the reaction overpotential during the lithium insertion / extraction process. The anode transfer coefficient, Let be the cathode transfer coefficient, and:

[0142]

[0143] The expression:

[0144]

[0145] in: The reaction rate parameter for the lithium insertion / extraction process. This is the reference concentration of lithium ions in the electrolyte. The maximum lithium intercalation concentration corresponding to the positive and negative electrode materials. This corresponds to the concentration of lithium ions on the surface of the spherical particles of the positive and negative electrode materials.

[0146] Reaction overpotential equation:

[0147]

[0148] in, The equilibrium potential of the corresponding electrode reaction is related to the stoichiometric ratio of lithium ions on the particle surface; The equivalent resistance of the SEI film.

[0149] ⑥ Electrochemical side reactions

[0150] like Figure 2 As shown, this invention primarily considers two electrochemical side reactions: SEI film formation on the surface of graphite anode materials and lithium plating on the anode surface. During the first cycle of a lithium-ion battery, an initial SEI film forms on the surface of the anode material. This initial SEI film prevents side reactions from occurring, thus benefiting battery cycle performance, lifespan, and safety characteristics. However, due to the porous structure of the SEI film, electrolyte can still diffuse through it to the anode surface, consuming lithium ions during the reaction process. This leads to continuous thickening of the SEI film, increasing its resistance, and consequently, reducing usable capacity and increasing internal resistance. As the number of battery cycles increases, SEI film growth becomes one of the main factors contributing to the aging of commercial lithium-ion batteries. Lithium plating is also a major factor in lithium-ion battery aging. Because the equilibrium potential of the lithium insertion / extraction reaction process in graphite anode materials is close to that of the lithium plating process, lithium plating may occur under extreme conditions such as high current charging and low temperature charging, resulting in a loss of usable lithium ion reserves and a decrease in battery capacity. The lithium dendrites generated during the lithium plating process may also block the separator or even puncture the separator, leading to an internal short circuit, which may bring the risk of internal short circuit or even thermal runaway in the battery.

[0151] The composition of the SEI film is very complex, and it can be considered to be formed by the reaction of ethylene carbonate (EC) with lithium ions. The main components of the SEI membrane:

[0152]

[0153] Assuming the growth of the SEI film is irreversible, and that the SEI formation process is influenced by the EC concentration on the surface of the negative electrode particles and the electrochemical reaction kinetics, it follows the Tafel equation:

[0154]

[0155] in, The reaction rate parameters corresponding to the SEI film formation process, The corresponding electrolyte concentration on the surface of the negative electrode particles. =0.5 is the transfer coefficient for the SEI formation reaction. This is the equilibrium potential for the SEI film formation reaction. For simplicity, we can take... for vs .

[0156] The electrolyte concentration on the surface of the negative electrode particles satisfies:

[0157]

[0158] in, The diffusion coefficient corresponding to the electrolyte in the SEI membrane, This corresponds to the electrolyte concentration inside the battery.

[0159] Corresponding SEI film thickness on the negative electrode surface With lithium plating thickness The sum, assuming the SEI film reactants and deposited lithium are uniformly distributed on the surface of the negative electrode particles, is expressed as:

[0160]

[0161] Overpotential of the SEI reaction:

[0162]

[0163] From the above formula, we can obtain:

[0164]

[0165] The SEI film growth equation can be obtained from the laws of conservation of mass and Faraday's law:

[0166]

[0167] in, The molar mass of the corresponding SEI product components, The average density corresponding to the SEI film.

[0168] Lithium plating at the negative electrode mainly occurs under conditions such as low-temperature charging, high-rate charging, and high SOC charging. Assuming the lithium plating process is irreversible (the deposited lithium will not react again to form an SEI film or be oxidized back to lithium ions), and the equilibrium potential of the lithium plating reaction... for vs Current density during lithium plating:

[0169]

[0170] in, The corresponding reaction rate constant for the lithium plating process, and Lithium plating reaction transfer coefficient corresponding to the battery electrode

[116] The values ​​are 0.3 and 0.7 respectively; the equilibrium potential for the lithium plating reaction. Pick vs The overpotential for the lithium plating reaction is as follows:

[0171]

[0172] The thickness of lithium metal deposits also satisfies the laws of conservation of mass and Faraday's law:

[0173]

[0174] in, These correspond to the molar mass and density of lithium, respectively.

[0175] The above analysis shows that the negative electrode surface deposit consists of an SEI film and deposited lithium; since the resistance of the SEI film is much greater than that of the deposited metallic lithium, the resistance of the negative electrode surface deposit can be obtained as follows:

[0176]

[0177] in, The SEI film state corresponding to the initial state. The average conductivity of the corresponding SEI film.

[0178] As the aging process progresses, the amount of deposits on the negative electrode surface increases, and the liquid phase volume fraction decreases. The porosity expression considering the effects of aging is:

[0179]

[0180] in, The negative electrode porosity corresponding to the initial state.

[0181] ⑦ Thermal Model

[0182] To obtain the battery temperature and correct the P2D model parameters, a battery heat generation and heat transfer model is introduced; thus, the calculation accuracy is improved by coupling the P2D and thermal models. In practical engineering applications, for simplicity, the battery is often equivalent to a uniform heat-generating body, so a lumped parameter model can be used for simulation.

[0183] like Figure 3 As shown, the lithium ion concentration C, potential U, and electrical potential in P2D are... Changes in parameters such as these affect battery heat generation, driving the heat generation model to generate a total heat generation q. The heat generation q and the ambient temperature of the battery heat transfer environment are also factors. Determines the average temperature of the battery Temperature changes in the heat generation and heat transfer model will affect the corresponding temperature-sensitive parameters in P2D.

[0184] (a) Battery heat generation

[0185] The main sources of heat generation in batteries are entropy change heat (reversible heat) from electrochemical reactions, ohmic heat from the Joule effect, and polarization heat caused by polarization. Among these, ohmic heat and polarization heat are irreversible heat.

[0186] Ohm heat The expression:

[0187]

[0188] Polarization heat The expression is as follows:

[0189]

[0190] Among them, subscript Corresponding to various electrochemical reactions, For different electrochemical reactions, the net current per unit volume is... Equilibrium potential corresponding to different electrochemical reactions.

[0191] Reversible heat The expression is as follows:

[0192]

[0193] in, The equilibrium potential corresponding to the lithium insertion / extraction electrochemical reaction of the positive and negative electrode materials.

[0194] Total heat generated by the battery The expression is as follows:

[0195]

[0196] (b) Battery heat transfer

[0197] The energy balance equation of a battery:

[0198]

[0199] Where m and C p Differentiate between battery quality and average specific heat capacity.

[0200] In practical applications, the heat generated by the battery can be transferred in various ways. However, when the battery is placed in a constant temperature chamber for testing, it is actually in a wind-cooled environment, and the temperature meets the following requirements:

[0201]

[0202] in The convective heat transfer coefficient is... This represents the effective surface area of ​​the battery. For the sake of simplicity, the ambient temperature is used in this invention. Taken from the set temperature.

[0203] By combining the above two equations, we can obtain the battery temperature change rate:

[0204]

[0205] ⑧ Battery mechanical force model

[0206] Mechanical damage to electrode materials is also an important factor affecting the aging process of lithium-ion batteries. During charging and discharging, lithium ions diffuse within the electrode materials, which can cause the electrodes to expand or contract. For example, graphite anode materials undergo about 10% volume deformation after complete lithiation, and cathode materials also experience varying degrees of volume deformation.

[0207] like Figure 4 As shown, the charging and discharging process causes differences in the concentration of lithium ions within the electrode material, forming a concentration gradient. The diffusion behavior of lithium ions within the electrode leads to uneven concentration distribution, which in turn causes material deformation, generating diffusion-induced stress. This may result in surface cracks or even fracture of the electrode material, leading to loss of electron / ion pathways and thus loss of active material.

[0208] Similar to the P2D model, the positive electrode active material is equivalent to an isotropic ideal sphere, which allows for the calculation of the stress and strain field distribution induced by lithium-ion diffusion. In a spherical coordinate system, the tangential stress-strain relationship in the electrode is expressed as follows:

[0209]

[0210] In this model, the subscripts r and θ correspond to the radial and tangential dimensions, respectively; ε represents strain; σ represents stress; v is Poisson's ratio; Ω is the partial molar volume of the solute; and E is the Young's modulus of the electrode material. It is also assumed that the elastic properties of the spherical particles are independent of the lithium intercalation concentration in the material.

[0211] For an equivalent spherical particle of an active material, the values ​​of its radial and tangential strains at the radial coordinate r are:

[0212]

[0213] Where μ corresponds to radial displacement.

[0214] Considering that the diffusion rate of lithium ions in the electrode material is much lower than the elastic deformation rate, the diffusion process is regarded as a quasi-static equilibrium problem, and its equation is as follows:

[0215]

[0216] Since the radial stress on the surface of the electrode particles is zero, and the radial tangential stress at the center of the sphere is approximately equal, the boundary conditions of the static equilibrium equations are satisfied as follows:

[0217]

[0218] Based on the above derivation, the radial and tangential stresses of the electrode particles can be solved, such as:

[0219]

[0220] (2) Battery SOC calculation based on multiphysics model

[0221] After inputting the EIS and pressure detection data into the above model to solve for the parameters, the lithium-ion concentration of the solid-state battery can be obtained, and then the SOC of the battery can be calculated. The calculation method is as follows:

[0222]

[0223] in It is the average lithium concentration in the negative electrode solid phase at a certain moment. Defined as the lithium concentration in the negative electrode solid phase at 0% SOC. Defined as the lithium concentration in the negative electrode solid phase at 100% SOC.

[0224] (3) Abnormal battery identification based on SOC

[0225] Considering the inconsistency of SOC (State of Charge) of individual battery cells within a pack or module, this invention uses a normal distribution method to evaluate and calibrate batteries with abnormal SOC. The basic method is as follows:

[0226] First, the SOC values ​​of all individual cells in the module or battery pack at the same time are statistically analyzed to form a set of individual cell SOC values.

[0227]

[0228] in This represents the number of battery cells within a module or battery pack. Statistical analysis is performed on this set to calculate its mean and standard deviation:

[0229] ,

[0230] Based on this, the distribution of monomeric SOC is approximated as following a normal distribution. This is used as a benchmark to assess the degree of SOC deviation of a single battery cell. For each battery cell, its SOC deviation can be expressed as:

[0231]

[0232] When a certain single substance deviates Exceeding the preset abnormal threshold If the cell is found to have an abnormal SOC (State of Charge), it is marked as P2. The battery error rate is then calculated. The calculation formula is as follows:

[0233]

[0234] IV. Intelligent Outlier Detection and Outlier Labeling Steps: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, a hybrid weighted outlier detection algorithm is used for automatic screening. High-weight coefficients are assigned to key sensitive parameters such as abnormal ambient temperature, voltage residual, sudden surface pressure changes, abnormal electrochemical impedance, and abnormal current to highlight their contribution to outlier detection, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers. This step can be understood as designing a hybrid weighted outlier detection method in the battery management system to perform outlier detection on the multi-physics parameters of all batteries, and marking the detected outlier batteries as P3.

[0235] Specifically, the hybrid weighted outlier detection algorithm can be: adding a parameter threshold determination mechanism to the DBSCAN outlier detection method, directly determining individual lithium-ion batteries that exceed the preset parameter threshold as outliers without participating in subsequent density clustering calculations, thereby accelerating the convergence of the outlier detection algorithm; or, introducing expert weight vectors to the support vector machine classification detection method, assigning differentiated weights to different multiphysics parameters (i.e., assigning different weights to the degree of difference of different parameter values ​​through expert weights), and combining it with distance threshold determination to achieve outlier detection, thereby improving detection efficiency and accuracy.

[0236] This intelligent outlier detection and outlier labeling process can be termed an outlier battery labeling process based on model parameter analysis:

[0237] (1) The parameters calculated by the multiphysics coupling model in the previous step are used to form an eigenvector:

[0238] in, For feature dimension, Indicates the first The feature vector of each battery cell.

[0239] (2) Introduce expert weights and perform feature weighting processing.

[0240] To highlight the feature dimensions in expert experience that have a more significant impact on battery anomaly identification, this invention introduces an expert experience weight vector:

[0241]

[0242] in Indicates the first The importance level of each feature in anomaly detection.

[0243] We construct weighted features by performing a weighted transformation on the feature vector using expert weights:

[0244]

[0245] This weighting process can effectively improve the contribution of highly sensitive features (such as temperature anomalies, voltage residuals, and pressure mutations) in the abnormal battery identification process.

[0246] (3) Abnormal battery identification based on support vector machine

[0247] After obtaining the weighted feature vectors, this invention uses a Support Vector Machine (SVM) classifier for training. SVM constructs an optimal classification hyperplane to achieve linear or nonlinear distinction between normal and abnormal battery cells. Its optimization objective function is:

[0248]

[0249] Meet the conditions

[0250] in, This indicates whether the sample is normal or abnormal; This represents the penalty factor, used to adjust the classification margin and the cost of misclassification; : Slack variables; The number of battery samples used in training.

[0251] After training, a support vector machine model with weights w and biases b is obtained for anomaly detection.

[0252]

[0253] when Considered a normal battery, when If so, it is determined to be an abnormal battery.

[0254] Next, the battery flag P3 that is identified as faulty is identified, and the offset is calculated. Its expression is:

[0255]

[0256] V. Intelligent Abnormal Battery Identification Steps: A comprehensive evaluation system for multi-physics parameters is established. Through an evaluation matrix, multi-dimensional coupled evaluation functions, or a decision tree model, the results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, along with the quantitative results of prediction deviations corresponding to error anomalies and inconsistencies corresponding to consistency anomalies, are automatically fused. This intelligently determines the normal, suspicious, or fault level of each individual lithium-ion battery and outputs the final abnormal battery identification result. This step establishes a comprehensive evaluation standard for battery multi-physics parameters, combining P1, P2, and P3 values ​​to output the final identified abnormal batteries.

[0257] The comprehensive evaluation standard for this parameter consists of an evaluation matrix based on the values ​​of P1, P2, and P3, a multi-dimensional coupled evaluation function designed with P1, P2, and P3 as benchmarks, and a decision tree model built with P1, P2, and P3 as benchmarks. The evaluation matrix quantifies the weights of P1, P2, P3, and their corresponding deviations in matrix form, directly calculating the comprehensive score. The multi-dimensional coupled evaluation function constructs mathematical functions (such as weighted summation, exponential functions, etc.) containing P1, P2, and P3, outputting the fusion result. The decision tree model achieves classification fusion through feature partitioning (such as thresholds for P1, P2, and P3, and deviation ranges), belonging to a data-driven fusion method.

[0258] Specifically, when using a decision tree model to identify anomalies, the process includes: collecting the results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, along with the corresponding prediction bias and quantification results, to construct a comprehensive feature vector; labeling real faults based on historical operating data, cycle life test data, and simulation data, training the decision tree model, calculating the information gain of each feature, and determining the optimal partitioning rule; inputting the comprehensive feature vector into the trained decision tree model, outputting the anomaly fusion confidence score, and dividing the anomaly into three levels based on the confidence score: confidence score < 0.5 corresponds to normal batteries, 0.5 ≤ confidence score < 0.8 corresponds to suspicious batteries, and confidence score ≥ 0.8 corresponds to faulty batteries.

[0259] (1) According to an embodiment of the present invention, a comprehensive feature vector for decision tree determination is first constructed for each lithium-ion battery cell. The feature vector includes the annotation results from three angles and the corresponding deviations, specifically:

[0260]

[0261] (2) Generation of candidate partition sets for each attribute feature

[0262] Based on the accumulation of historical data, and utilizing historical operational data, experimental data, or simulation data, some individual battery cells are manually or systematically labeled to obtain their true fault state labels. ,in: Normal battery. Suspicious battery Faulty battery. The combined feature vector is paired with the true label to form the training sample set for the decision tree:

[0263]

[0264] in This represents the number of samples used in the training process.

[0265] (3) Define the node sample set and category entropy

[0266] During the decision tree construction process, the current set of samples to be split is denoted as... ,remember The number of samples in each category is The corresponding category is Then the node The class probability is

[0267]

[0268] in For set Total number of samples.

[0269] Define nodes The information entropy is:

[0270]

[0271] (4) Generation of candidate partitioning method based on single feature

[0272] For any feature This invention expresses them uniformly as a set of candidate partitioning rules:

[0273]

[0274] Each of them It is a child node after partitioning, defined as:

[0275]

[0276] in For the sample The value taken on feature A; Let A be a certain "range of values" for feature A.

[0277] (5) Calculate conditional entropy and information gain

[0278] For a given candidate partition, suppose the current sample set is... Divided into several subsets Then the conditional entropy after partitioning is defined as:

[0279]

[0280] The information gain of feature A is:

[0281]

[0282] in: The entropy before partitioning; The weighted entropy is calculated based on this feature and threshold. Information gain reflects the "purity improvement" brought about by this partition.

[0283] (6) Select the best feature as the partitioning rule for the current node.

[0284] Calculate the information gain for all six features:

[0285]

[0286] Select the feature with the highest information gain Its corresponding optimal partition set The decision rules that constitute the current node.

[0287] At this point, node partitioning can be performed.

[0288] (7) Recursively constructing lower-level nodes

[0289] Repeat steps (1) through (7) until the information gain is less than the minimum requirement.

[0290] (8) Faulty battery identification

[0291] The products generated in steps S1 to S4 The vector is input into the decision tree to classify each battery cell and identify faulty batteries.

[0292] This invention also relates to an online multi-physics coupled intelligent sensing and early warning system for lithium-ion batteries, which corresponds to the online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries described above. It can be understood as a system that implements the aforementioned online multi-physics coupled intelligent sensing and early warning method for lithium-ion batteries. Figure 5 As shown, the system includes, in sequence, a single-cell multi-physics parameter online detection module, an intelligent modeling and error anomaly labeling module, an intelligent state estimation and consistency anomaly labeling module, an intelligent outlier detection and outlier anomaly labeling module, and an intelligent abnormal battery identification module.

[0293] The online detection module for multi-physics parameters of a single battery cell: Based on the geometric characteristics of lithium-ion batteries, it adopts a collaborative acquisition method of electrochemical-mechanical-thermal-electrical multi-physics fields to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of a single lithium-ion battery cell in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing, and real-time multi-physics parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance.

[0294] The intelligent modeling and error anomaly labeling module automatically aggregates real-time multi-physics parameter data of each individual lithium-ion battery collected online to form a dataset. Based on the coupling relationship between electrochemical, mechanical, thermal, and electrical multi-physics fields, a multi-physics coupling model is constructed. The input of this model is real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current. The output is the predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery. By comparing the predicted values ​​of surface pressure and electrochemical impedance output by the model with the corresponding real-time multi-physics parameter data of surface pressure and electrochemical impedance collected online, the prediction deviation of surface pressure and electrochemical impedance is calculated. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data and the heat generation law of the lithium-ion battery. If the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically annotated with an error anomaly.

[0295] The intelligent state estimation and consistency anomaly labeling module: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, it automatically solves the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency, through a multi-dimensional state estimation algorithm; and based on the multi-dimensional state parameters, it calculates the inconsistencies between individual lithium-ion batteries through statistical analysis and entropy weight method, and obtains the quantitative results of range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weight; and automatically labels the consistency anomalies of individual lithium-ion batteries that exceed the corresponding inconsistency thresholds.

[0296] The intelligent outlier detection and outlier labeling module: Based on the real-time multi-physics parameter data of each individual lithium-ion battery collected online, it uses a hybrid weighted outlier detection algorithm for automatic screening. It assigns high weight coefficients to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure change, electrochemical impedance, and current to highlight their contribution to outlier determination, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers.

[0297] The intelligent abnormal battery identification module establishes a comprehensive evaluation system of multi-physics parameters. It automatically integrates the results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, as well as the quantitative results of prediction deviation corresponding to error anomalies and inconsistency corresponding to consistency anomalies, through evaluation matrix, multi-dimensional coupled evaluation function or decision tree model. It intelligently determines the normal, suspicious, or fault level of each individual lithium-ion battery and outputs the final abnormal battery identification result.

[0298] Furthermore, the preferred structure of the online detection module for multiple physical field parameters of a single battery cell is as follows: Figure 6 As shown, the system includes a slave acquisition unit, an electrochemical acquisition and calculation unit, a pressure acquisition and conversion unit, and a signal communication unit. The slave acquisition unit receives control commands from the signal communication unit, initiates the multi-physics acquisition function, performs signal calculation, and sends the multi-physics parameters of the lithium-ion battery to the signal communication unit. The signal communication unit receives signal detection commands from the EMS battery management system or other slave acquisition units and forwards them to the slave acquisition unit. It also sends the multi-physics parameters acquired by the slave acquisition unit to the EMS battery management system or other slave acquisition units. The electrochemical acquisition and calculation unit also... The EIS (Electrochemical Impedance Acquisition and Calculation) unit is responsible for receiving electrochemical impedance acquisition commands from the slave acquisition unit and sending excitation current signals. This includes an excitation signal amplifier to amplify the excitation signal, then acquiring the voltage signal, calculating the electrochemical impedance of the lithium-ion battery under an excitation current signal at a given frequency, and designing the magnitude of the electrochemical impedance excitation based on the single-cell capacity and impedance range. The pressure acquisition and conversion unit acquires the voltage signal value corresponding to the battery surface pressure through a piezoresistive sensor, then calculates the resistance value corresponding to the pressure sensor voltage division by comparing it with the nominal resistance, and then calculates the battery surface pressure value based on this resistance value.

[0299] The specific workflow is as follows:

[0300] (1) The battery management system issues a signal acquisition request (or signal detection command);

[0301] (2) After receiving the signal detection command issued by the battery management system, the signal communication unit forwards it to the slave control acquisition unit;

[0302] (3) The multi-physics field acquisition function is started by the slave acquisition unit, and the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit are enabled (i.e., the enable signal is sent), while the voltage, current and ambient temperature parameters of the lithium-ion battery cell under test are acquired simultaneously.

[0303] (4) After receiving the enable signal, the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit will detect the tested lithium-ion battery cell. The detection methods of the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit are as follows:

[0304] a) Electrochemical acquisition and calculation unit: After receiving the enable signal, the excitation source emits a variable frequency sinusoidal excitation signal with an amplitude of 5-10mV and a frequency range of 0.1Hz-1kHz. This signal is amplified and adjusted to the expected amplitude by the excitation signal amplifier and applied to the lithium-ion battery cell under test. Subsequently, the response of the lithium-ion battery cell under test to the amplified excitation signal is acquired by the data processor, and the electrochemical impedance is calculated.

[0305] b) Pressure acquisition and conversion unit: After receiving the enable signal, the surface pressure of the tested lithium-ion battery cell is detected by the piezoresistive sensor and the corresponding resistance signal is output. Combined with the nominal resistance of the sensor, the resistance change is calculated by the signal comparator, and then the surface pressure signal of the tested lithium-ion battery cell is converted based on the preset calibration curve.

[0306] (5) After the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit complete the signal acquisition and processing of the tested lithium-ion battery cell, the acquired and processed electrochemical impedance and surface pressure signals are sent to the slave acquisition unit. They are then integrated with the voltage, current and ambient temperature parameters acquired by the slave acquisition unit and fed back to the signal communication unit as real-time multi-physics parameter data. The signal communication unit then forwards the data to the battery management system of other slave acquisition units.

[0307] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention patent.

Claims

1. An online multi-physical field coupling lithium-ion battery intelligent sensing and calculation early warning method, characterized in that, Includes the following steps: Online detection steps for multi-physics parameters of a single battery: Based on the geometric characteristics of lithium-ion batteries, a collaborative acquisition method of electrochemical-mechanical-thermal-electrical multi-physics fields is adopted to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of the single lithium-ion battery in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing. Real-time multi-physics parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance. Intelligent modeling and error anomaly labeling steps: Real-time multi-physics parameter data of each individual lithium-ion battery collected online are automatically aggregated into a dataset; a multi-physics coupling model is constructed based on the coupling relationship between electrochemical, mechanical, thermal, and electrical multi-physics fields. The input of the multi-physics coupling model is the real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current, and the output is the predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery; by comparing the predicted values ​​of surface pressure and electrochemical impedance output by the model with the corresponding real-time multi-physics parameter data of surface pressure and electrochemical impedance collected online in real time, the prediction deviation of surface pressure and electrochemical impedance is calculated. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data combined with the heat generation law of lithium-ion batteries; if the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically annotated with an error anomaly. Intelligent State Estimation and Consistency Anomaly Labeling Steps: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, a multi-dimensional state estimation algorithm is used to automatically solve the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency. Based on the multi-dimensional state parameters, statistical analysis and entropy weight method are used to calculate the inconsistencies between individual lithium-ion batteries, obtaining quantitative results of range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weights. Individual lithium-ion batteries that exceed the corresponding inconsistency thresholds are automatically labeled as consistency anomalies. Intelligent outlier detection and outlier labeling steps: Based on the real-time multi-physics parameter data of each individual lithium-ion battery collected online, a hybrid weighted outlier detection algorithm is used for automatic screening. High weight coefficients are assigned to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure change, electrochemical impedance, and current to highlight their contribution to outlier determination, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers. Intelligent abnormal battery identification steps: Establish a comprehensive evaluation system of multi-physics parameters, and automatically fuse the results of error anomaly labeling, consistency anomaly labeling, outlier anomaly labeling, and the quantitative results of prediction deviation corresponding to error anomalies and inconsistency corresponding to consistency anomalies through evaluation matrix, multi-dimensional coupled evaluation function or decision tree model. Intelligently determine the normal, suspicious or fault level of each individual lithium-ion battery and output the final abnormal battery identification result.

2. The online multi-physics coupled lithium-ion battery smart sensing and calculation early warning method according to claim 1, characterized in that, In the online detection step of multi-physics parameters of a single battery cell, when a piezoresistive sensor is deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing, the piezoresistive sensor is deployed near the battery tab, at the center of the battery pressure surface, and / or near the battery pressure surface. The amplitude and frequency range of the electrochemical impedance excitation signal are adaptively adjusted according to the capacity and impedance characteristics of the single lithium-ion battery cell.

3. The online multi-physics coupled lithium-ion battery smart sensing and calculation early warning method according to claim 1, characterized in that, In the online detection step of multi-physics parameters of a single battery, when performing online synchronous acquisition of real-time multi-physics parameter data, the storage resource configuration is optimized according to the amount of real-time multi-physics parameter data generated, the data transmission rate is optimized according to the characteristic electrochemical impedance frequency of the lithium-ion battery, and the pressure acquisition sensitivity and minimum resolution are optimized according to the pressure change amplitude of the battery surface, so as to improve the efficiency of online synchronous acquisition of real-time multi-physics parameters.

4. The online multiphysics coupled intelligent sensing and early warning method for lithium-ion batteries according to any one of claims 1 to 3, characterized in that, In the intelligent modeling and error anomaly labeling steps, the multiphysics coupling model can be: a discretized model obtained by reducing the order of a model constructed based on the Pade approximation method with the electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries as the core; or a model of electrical equivalent circuit-thermal equivalent circuit-mechanical equivalent circuit coupling based on the equivalent circuit principle; or a black-box model obtained by training a neural network. The training process of the black-box model is as follows: battery parameters under different operating conditions are generated using the electrochemical-mechanical-thermal-electrical coupling relationship of lithium-ion batteries, a virtual dataset is established and divided into a training set and a test set, and a temporal convolutional network-bidirectional gated unit neural network is used as the basic model for training. The virtual dataset is used for offline training of the black-box model.

5. The intelligent sensing and early warning method for lithium-ion batteries based on online multiphysics coupling according to claim 4, characterized in that, In the intelligent modeling and error anomaly labeling step, the rationality verification of the predicted internal temperature value of the lithium-ion battery is specifically as follows: by using real-time ambient temperature parameter data, combined with the heat generation law of the lithium-ion battery including the generation and conduction characteristics of ohmic heat, polarization heat, and reversible heat, the reasonable range of internal temperature is determined; the upper limit of the reasonable range is the sum of the ambient temperature and the maximum allowable temperature rise of the lithium-ion battery, and the lower limit is the difference between the ambient temperature and the minimum allowable temperature drop of the lithium-ion battery.

6. The online multiphysics coupled intelligent sensing and early warning method for lithium-ion batteries according to any one of claims 1 to 3, characterized in that, In the intelligent state estimation and consistency anomaly labeling steps, the multi-dimensional state estimation algorithm is one or more combinations of Kalman filtering algorithms, adaptive filtering algorithms, or machine learning algorithms; the corresponding thresholds for range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weights are all determined after statistical analysis and calibration based on the factory consistency parameters, cycle life test data, and multi-condition operation history data of the same batch of single lithium-ion batteries.

7. The online multiphysics coupled intelligent sensing and early warning method for lithium-ion batteries according to any one of claims 1 to 3, characterized in that, In the intelligent outlier detection and outlier labeling steps, the hybrid weighted outlier detection algorithm is as follows: Based on the DBSCAN outlier detection method, a parameter threshold determination mechanism is added. Individual lithium-ion batteries that exceed the preset parameter threshold are directly determined as outliers and do not need to participate in subsequent density clustering calculations. Alternatively, based on the support vector machine classification and detection method, an expert weight vector can be introduced to assign differentiated weights to different multiphysics parameters, and outlier detection can be achieved by combining distance threshold judgment.

8. The online multiphysics coupled intelligent sensing and early warning method for lithium-ion batteries according to any one of claims 1 to 3, characterized in that, In the intelligent abnormal battery identification step, when anomaly identification is achieved through a decision tree model, the specific steps include: The results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, along with corresponding prediction bias and quantification results, are collected to construct a comprehensive feature vector. Based on historical operational data, cycle life test data, and simulation data, real fault labels are labeled, a decision tree model is trained, and the information gain of each feature is calculated to determine the optimal partitioning rule. The comprehensive feature vector is input into the trained decision tree model, and the anomaly fusion confidence score is output. Based on the confidence score, three anomaly levels are defined: confidence score < 0.5 corresponds to normal batteries, 0.5 ≤ confidence score < 0.8 corresponds to suspicious batteries, and confidence score ≥ 0.8 corresponds to faulty batteries.

9. An online multi-physics coupled intelligent sensing and early warning system for lithium-ion batteries, characterized in that, It includes a single cell multi-physics parameter online detection module, an intelligent modeling and error anomaly labeling module, an intelligent state estimation and consistency anomaly labeling module, an intelligent outlier detection and outlier anomaly labeling module, and an intelligent abnormal battery identification module connected in sequence. The online detection module for multi-physics parameters of a single battery cell: Based on the geometric characteristics of lithium-ion batteries, it adopts a collaborative acquisition method of electrochemical-mechanical-thermal-electrical multi-physics fields to generate an electrochemical impedance excitation signal that adaptively matches the capacity and impedance range of a single lithium-ion battery cell in real time. Piezoresistive sensors are deployed in the current stress concentration area of ​​the lithium-ion battery for fixed-point pressure sensing, and real-time multi-physics parameter data of the lithium-ion battery, including electrochemical impedance, surface pressure, ambient temperature, voltage, and current, are collected online synchronously. Among them, the electrochemical impedance is obtained by frequency conversion processing, amplification, response acquisition, and calculation of the electrochemical impedance excitation signal, and the surface pressure is obtained by comparing and converting the pressure sensing signal collected by the piezoresistive sensor with the nominal resistance. The intelligent modeling and error anomaly labeling module automatically aggregates real-time multi-physics parameter data of each individual lithium-ion battery collected online to form a dataset. Based on the coupling relationship between electrochemical, mechanical, thermal, and electrical multi-physics fields, a multi-physics coupling model is constructed. The input of this model is real-time multi-physics parameter data of ambient temperature, lithium-ion battery voltage, and current. The output is the predicted values ​​of the internal temperature, surface pressure, and electrochemical impedance of the corresponding individual lithium-ion battery. By comparing the predicted values ​​of surface pressure and electrochemical impedance output by the model with the corresponding real-time multi-physics parameter data of surface pressure and electrochemical impedance collected online, the prediction deviation of surface pressure and electrochemical impedance is calculated. Simultaneously, the rationality of the predicted internal temperature value is verified based on the real-time ambient temperature parameter data and the heat generation law of the lithium-ion battery. If the prediction deviation of any parameter exceeds a preset threshold, or the predicted internal temperature value exceeds a reasonable range, the individual lithium-ion battery is automatically annotated with an error anomaly. The intelligent state estimation and consistency anomaly labeling module: Based on real-time multi-physics parameter data of each individual lithium-ion battery collected online, it automatically solves the multi-dimensional state parameters of each individual lithium-ion battery, including state of charge, state of health, state of energy, ohmic internal resistance, and battery impedance at a specific frequency, through a multi-dimensional state estimation algorithm; and based on the multi-dimensional state parameters, it calculates the inconsistencies between individual lithium-ion batteries through statistical analysis and entropy weight method, and obtains the quantitative results of range inconsistency, variance inconsistency, and entropy weight method inconsistency based on expert weight; and automatically labels the consistency anomalies of individual lithium-ion batteries that exceed the corresponding inconsistency thresholds. The intelligent outlier detection and outlier labeling module: Based on the real-time multi-physics parameter data of each individual lithium-ion battery collected online, it uses a hybrid weighted outlier detection algorithm for automatic screening. It assigns high weight coefficients to key sensitive parameters such as abnormal ambient temperature, voltage residual, surface pressure change, electrochemical impedance, and current to highlight their contribution to outlier determination, thereby quickly identifying outlier individual lithium-ion batteries and labeling them as outliers. The intelligent abnormal battery identification module establishes a comprehensive evaluation system of multi-physics parameters. It automatically integrates the results of error anomaly labeling, consistency anomaly labeling, and outlier anomaly labeling, as well as the quantitative results of prediction deviation corresponding to error anomalies and inconsistency corresponding to consistency anomalies, through evaluation matrix, multi-dimensional coupled evaluation function or decision tree model. It intelligently determines the normal, suspicious, or fault level of each individual lithium-ion battery and outputs the final abnormal battery identification result.

10. The online multiphysics coupled intelligent sensing and early warning system for lithium-ion batteries according to claim 9, characterized in that, The online detection module for multi-physics parameters of a single cell includes a slave-controlled acquisition unit, an electrochemical acquisition and calculation unit, a pressure acquisition and conversion unit, and a signal communication unit. The signal communication unit receives the signal detection command issued by the battery management system and forwards it to the slave control acquisition unit. The slave control acquisition unit activates the multi-physics field acquisition function, sending an enable signal to the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit, while simultaneously acquiring the voltage, current, and ambient temperature parameters of the tested lithium-ion battery cell. After receiving the enable signal, the electrochemical acquisition and calculation unit sends a frequency conversion excitation signal from the excitation source. This signal is amplified to the expected amplitude by the excitation signal amplifier and applied to the tested lithium-ion battery cell. The data processor acquires the response of the tested lithium-ion battery cell to the amplified excitation signal and calculates the electrochemical impedance. After receiving the enable signal, the pressure acquisition and conversion unit detects the surface pressure of the tested lithium-ion battery cell through the piezoresistive sensor and outputs the corresponding resistance signal. Combined with the sensor's nominal resistance, the signal comparator calculates the resistance change and then converts it into the surface pressure signal of the tested lithium-ion battery cell. The electrochemical impedance and surface pressure signals after signal acquisition and processing by the electrochemical acquisition and calculation unit and the pressure acquisition and conversion unit are sent to the slave control acquisition unit. Together with the voltage, current, and ambient temperature parameters acquired by the slave control acquisition unit, they are fed back to the signal communication unit as real-time multi-physics field parameter data. The signal communication unit then sends the data to the battery management system or other slave control acquisition units.

Images