A lithium-ion battery system and method based on monitoring strain forces

Abstract

The application discloses a lithium ion battery system and method based on monitoring strain force, and relates to the technical field of lithium ion battery state evaluation. Through obtaining multi-physical field monitoring data, a physical information module including a stress coupling electrochemical potential energy submodule, a temperature and pressure collaborative control diffusion dynamics submodule and a multi-physical field coupling thermal submodule is constructed; a data-driven compensation module is constructed by using a Kolmogorov-Arnold network; the outputs of the physical information module and the data-driven compensation module are adaptively fused based on a dynamic gating fusion mechanism; and SOC and SOH evaluation results are output. The application is based on an electric-thermal-force three-physical field coupling modeling framework, combines a physical enhanced hybrid Kolmogorov-Arnold network and a dynamic gating adaptive fusion, and realizes high-precision, high-robustness and high-explainability battery state online estimation.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery state assessment technology, and in particular to a lithium-ion battery system and method based on monitoring strain. Background Technology

[0002] Lithium-ion batteries are widely used in electric vehicles, energy storage systems, and portable electronic devices due to their high energy density and long cycle life. Accurately estimating the battery's state of charge (SOC) and state of health (SOH) is a core function of the battery management system (BMS), directly affecting the safety, reliability, and economy of the battery system.

[0003] Traditional SOC / SOH estimation methods primarily rely on monitoring electrochemical signals such as voltage, current, and temperature. These include Kalman filtering methods based on equivalent circuit models and traditional early warning methods based on voltage and temperature. However, these methods have inherent drawbacks: the electrochemical model parameters in Kalman filtering methods based on equivalent circuit models drift with battery aging and temperature changes, leading to decreased estimation accuracy. This is particularly true for lithium iron phosphate (LFP) batteries with flat open-circuit voltage (OCV) curves, where even small voltage errors can cause several times the error in the SOC estimate. Furthermore, traditional early warning methods based on voltage and temperature typically only detect anomalies after irreversible battery damage, resulting in insufficient warning time windows.

[0004] In recent years, researchers have begun to explore the use of strain signals for battery state assessment. For example, in a study published in the Journal of Southwest University (2025) by Zhu Jiangong et al., multidimensional signals such as voltage, current, and stress were used as features and input into neural networks such as LSTM and TCN, allowing the models to autonomously discover the implicit mapping relationship between mechanical signals and SOC / SOH. The study showed that after introducing stress, the error of the LSTM model decreased by 15.45%, and the error of the TCN model decreased by 45.88%.

[0005] However, existing strain-based methods are mostly purely data-driven models, lacking physical interpretability and failing to establish an explicit coupling relationship between strain signals and electrochemical processes. Furthermore, how to organically integrate mechanical signals into the existing electro-thermal estimation framework to achieve adaptive fusion of physical models and data-driven approaches remains an unsolved technical challenge. To address this, a lithium-ion battery system and method based on strain monitoring is proposed. Summary of the Invention

[0006] The main objective of this invention is to provide a lithium-ion battery system and method based on strain monitoring, which can effectively solve the problems in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for assessing the state of a lithium-ion battery system based on monitoring strain, comprising the following steps: Step S1: Acquire multi-physics field monitoring data of lithium-ion batteries during the charging and discharging process.

[0008] The multiphysics monitoring data includes at least current signals, voltage signals, temperature signals, and strain force signals.

[0009] Step S2: Construct a physical information module to establish an explicit physical mapping from the multi-physics monitoring data to the battery state variables.

[0010] The physical information module includes a stress-coupled electrochemical potential energy submodule, a temperature-pressure synergistic control diffusion dynamics submodule, and a multi-physics field coupled thermodynamics submodule. Specifically: The stress-coupled electrochemical potential energy submodule employs a modified Butler-Volmer equation and introduces a correction term for the stress-induced electrochemical reaction barrier. The correction term is expressed as: ,in This refers to the change in Gibbs free energy caused by stress. The stress coupling factor, which is SOC-dependent, is learned through the Kolmogorov-Arnold network. This is a stress signal; is the partial molar volume of lithium ions; F is the Faraday constant; It is Avogadro's constant; This is an overpotential.

[0011] The thermo-pressure co-controlled diffusion dynamics submodule employs a stress-corrected diffusion coefficient tensor. , is represented as: ,in Pre-exponential factors; It is the diffusion activation energy; Hydrostatic stress; Let k be the diffusion activation volume; k be the stress-diffusion coupling coefficient; R be the gas constant; T be the temperature; I be the unit tensor; and establish the thermo-baric coupled diffusion equation satisfied by the lithium ion concentration field: , where c is the lithium ion concentration; This is the source term for electrochemical reactions.

[0012] The multiphysics coupled thermal submodule employs an extended Bernardi heat generation model, which adds a stress dissipation heat dissipation term in addition to irreversible heat, reversible heat, and Joule heating. , is represented as: ,in For von Mises equivalent stress; Plastic strain rate; For electrode volume; The mechanical energy to heat energy conversion coefficient; The stress dissipation heat dissipation term is assigned to the thermal balance equations for the core temperature state and the surface temperature state of the battery.

[0013] Step S3: A data-driven compensation module is constructed using a Kolmogorov-Arnold network (KAN) to learn the complex nonlinear residual relationships that the physical information module failed to describe.

[0014] Specifically: The Kolmogorov-Arnold network uses B-spline basis functions for parameterization of univariate functions: ,in For B-spline basis functions; These are the learnable spline coefficients; and These are the coefficients of the linear and bias terms; The Kolmogorov-Arnold network comprises at least three independent subnetworks: a first KAN for learning the stress coupling factor, a second KAN for learning the strain-SOC sensitivity coefficient, and a third KAN for learning the aging compensation amount.

[0015] Step S4: Based on the dynamic gating fusion mechanism, according to the prediction confidence of the physical information module under the current operating conditions, adaptively fuse the output of the physical information module and the output of the data-driven compensation module to obtain the fused battery state estimate.

[0016] Specifically: The dynamic gating fusion mechanism includes: Construct a gated network with input features including the current physical model prediction state, input signal, observable output signal, and physical model uncertainty index; The uncertainty index of the physical model Defined as: ,in This is the measured output vector; For measurement functions; For prior estimation of the physical model; To measure the trace of the noise covariance matrix; The gated network outputs a fusion weight vector. The final battery state estimate Represented as: ,in This is element-wise multiplication; These are estimates from the physical model; The compensation amount output by the Kolmogorov-Arnold network.

[0017] Step S5: Based on the fused battery state estimate, output the state of charge and health assessment results of the lithium-ion battery.

[0018] Preferably, the physical information module further includes a mechanical model, which incorporates the total strain. Decomposed into reversible strain and irreversible strain ; The evolution of reversible strain satisfies: The strain-SOC sensitivity coefficient is obtained by learning from the Kolmogorov-Arnold network; The evolution of irreversible strain satisfies: This is the cumulative coefficient of irreversible strain; This is the critical strain threshold. It is the activation energy.

[0019] Preferably, the loss function used to train the Kolmogorov-Arnold network includes a physical consistency constraint term, which includes: Energy conservation constraint Defined as: , where N is the number of samples; This is the correction amount for the SOC predicted by the KAN network; Coulomb efficiency is taken as 0.98-0.999 during charging and 1 during discharging. Let be the current at time k; The sampling time interval; This is the maximum battery capacity; For estimated health status; Thermodynamic consistency constraint Defined as: ,in Temperature estimated for the physical model; This is the temperature correction amount predicted by the KAN network; The temperature of the physical model at time k-1; This refers to the change in temperature. Total heat generation rate; Mechanical consistency constraint Defined as: ,in V represents the reversible strain correction predicted by the KAN network; V is the total electrode volume. This represents the partial molar volume of lithium ions. The concentration change field predicted by the KAN network; The total loss function is a weighted sum of the three physical consistency constraint terms and the prediction error term.

[0020] Preferably, the method further includes a recursive filtering update step: using an extended Kalman filter framework to perform time and measurement updates on the physical information module, wherein the state vector of the physical information module includes: state of charge, polarization voltage, core temperature, surface temperature, reversible strain, irreversible strain, maximum internal stress, health status, and aging characteristic parameters.

[0021] Secondly, the present invention provides a lithium-ion battery system based on strain monitoring for implementing the above method, comprising: A multi-physics sensing unit is used to collect multi-physics monitoring data in real time during the charging and discharging process of lithium-ion batteries; A physical information module, the input of which is connected to the multiphysics sensing unit, the physical information module comprising: The stress-coupled electrochemical potential energy submodule is used to establish an explicit physical mapping between stress signals and electrochemical reaction kinetics; The temperature-pressure co-controlled diffusion dynamics submodule is used to establish a lithium-ion diffusion model under the coupling effect of temperature and stress. The multiphysics coupled thermal submodule is used to build a heat generation model that includes the contribution of stress dissipation. A data-driven compensation module, whose input is connected to the multiphysics sensing unit, is constructed using a Kolmogorov-Arnold network to learn the residual relationships that the physical information module fails to describe. The dynamic gating fusion unit has a first input terminal connected to the output terminal of the physical information module and a second input terminal connected to the output terminal of the data-driven compensation module. It is used to adaptively fuse the output of the physical information module and the output of the data-driven compensation module according to the prediction confidence of the physical information module under the current operating conditions to generate a fused battery state estimate. The state assessment output unit, whose input is connected to the output of the dynamic gating fusion unit, is used to output the state of charge and health status assessment results of the lithium-ion battery based on the fused battery state estimate.

[0022] Thirdly, the lithium-ion battery system state assessment method based on strain monitoring provided by this invention is applied in at least one of the following scenarios: (a) Online real-time estimation of state of charge and / or state of health in the vehicle battery management system; (b) Early warning of thermal runaway in lithium-ion battery energy storage systems; (c) Battery aging status diagnosis and remaining life prediction based on strain force signals; (d) Intelligent battery state assessment integrated with implanted or surface-mount fiber Bragg grating sensors. Beneficial effects

[0023] Compared with the prior art, the present invention has the following beneficial effects: This invention establishes a three-physics coupling modeling framework of electro-thermal-mechanical fields, explicitly coupling mechanical signals (strain / pressure) into the electrochemical and thermodynamic models through three major physical modules: the improved Butler-Volmer equation, the stress-corrected diffusion coefficient, and stress dissipation heat dissipation, thereby achieving a more complete description of the complex physical processes inside the battery.

[0024] This invention employs the Kolmogorov-Arnold network (KAN) as a data-driven compensation module. Its structure, based on the summation of univariate functions, is naturally suitable for learning physical residuals and has higher parameter efficiency and interpretability compared to the traditional multilayer perceptron (MLP).

[0025] This invention dynamically adjusts the fusion weights of the physical model and data-driven compensation by calculating the prediction confidence of the physical model in real time. When the physical model has high accuracy, it takes the lead; when the physical model fails (such as under extreme conditions or in the later stages of aging), the data-driven module automatically compensates, thus achieving adaptive optimal estimation across the entire operating range.

[0026] This invention introduces energy conservation, thermodynamic consistency, and mechanical consistency constraints into the loss function, ensuring that the model predictions do not violate fundamental physical laws even under conditions not covered by training data, thus significantly improving the generalization ability and safety of the method. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the state assessment method for lithium-ion battery systems based on strain monitoring according to the present invention. Figure 2 This is a schematic diagram of the overall architecture of the Kolmogorov-Arnold network of this invention; Figure 3 This is a schematic diagram of the structure of a lithium-ion battery system based on strain monitoring according to the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] Example 1: This embodiment provides a lithium-ion battery system state assessment method based on strain monitoring, which can be used for online real-time estimation of SOC and SOH in vehicle battery management systems.

[0030] Test subject: Square lithium iron phosphate (LFP) battery module with a rated capacity of 50Ah.

[0031] Sensor configuration: Current sensor: Hall effect type, measurement range ±200A, accuracy ±0.5% Voltage sensor: Differential amplifier type, measurement range 0-5V, accuracy ±0.1% Temperature sensor: PT1000 thermistor, attached to the battery surface and positive and negative terminals, accuracy ±0.5°C. Strain sensors: resistive strain gauges attached to the center of the battery surface (measurement range ±2000με) and fiber Bragg grating (FBG) strain sensors (embedded inside the cell core). Data acquisition frequency: 100Hz.

[0032] See Figure 1 The execution process includes: Step S1: Multiphysics Data Acquisition The following data were collected simultaneously under constant current constant voltage (CCCV) charging and dynamic stress testing (DST) conditions: Current I k ; Terminal voltage V t,k ; Surface temperature T surf,k ; Total strain ε total,k ; Surface stress σ surf,k .

[0033] Step S2: Construct and initialize the physical information module The system state vector is defined as: x=[SOC,V1,V2,T] core ,T surf ,ε rev ,ε irr ,σ max [SOH] T Initial settings: SOC0=100%, all polarization voltages are initially 0, and the core temperature is initially equal to the ambient temperature (25℃).

[0034] The three sub-modules of the physical information module are constructed as follows: (1) Stress-coupled electrochemical potential energy submodule A modified Butler-Volmer equation is used, incorporating a correction term for the stress-induced electrochemical reaction barrier. , is represented as: ,in This refers to the change in Gibbs free energy caused by stress. The stress coupling factor, which is SOC-dependent, is learned through the Kolmogorov-Arnold network. This is a stress signal; is the partial molar volume of lithium ions; F is the Faraday constant; It is Avogadro's constant; This is an overpotential.

[0035] The improved Butler-Volmer equation is expressed as: ;in Stress-dependent activation energy .

[0036] In this embodiment, the stress coupling factor The basis functions were obtained through offline learning using the first KAN network, employing B-spline basis functions (order k=3, number of nodes M=10).

[0037] (2) Temperature and pressure coordinated control of diffusion dynamics submodule The temperature-pressure co-control diffusion dynamics submodule employs a stress-corrected diffusion coefficient tensor. And establish the thermo-pressure coupled diffusion equation satisfied by the lithium ion concentration field; Among them, the diffusion coefficient tensor , is represented as: ,in Pre-exponential factors; It is the diffusion activation energy; Hydrostatic stress; R is the diffusion activation volume; k is the stress-diffusion coupling coefficient; R is the gas constant; T is the temperature; I is the unit tensor. The thermo-pressure coupled diffusion equation is expressed as: , where c is the lithium ion concentration; This is the source term for electrochemical reactions.

[0038] In this embodiment, the diffusion equation is spatially discretized (10 nodes) using the finite difference method, and the time discretization uses the implicit backward Euler scheme to ensure numerical stability.

[0039] (3) Multiphysics Coupled Thermal Submodule The multiphysics coupled thermal submodule employs an extended Bernardi heat generation model, which adds a stress dissipation heat dissipation term in addition to irreversible heat, reversible heat, and Joule heat. , is represented as: ,in For von Mises equivalent stress; Plastic strain rate; For electrode volume; The mechanical energy to heat energy conversion coefficient; The stress dissipation heat dissipation term is assigned to the thermal balance equations for the core temperature state and the surface temperature state of the battery.

[0040] Step S3: Construct the KAN data-driven compensation module Construct three independent KAN networks, including: (1) Stress coupling factor KAN: Input SOC single variable, output γ.

[0041] In this embodiment, a single-layer network structure with 5 B-spline basis functions is used.

[0042] (2) Strain sensitivity KAN: Input [SOC, T] two-dimensional vector, output .

[0043] In this embodiment, a network structure with two layers of KAN, the first layer having a width of 5 and the second layer having a width of 1 is adopted.

[0044] (3) Aging compensation KAN: Input A four-dimensional vector, output .

[0045] In this embodiment, a three-layer KAN network structure with a width of [4, 10, 20, 1] is used.

[0046] The B-spline basis functions of the KAN network employ an adaptive grid refinement strategy: the initial number of nodes is 10, and the number of nodes is dynamically increased according to the training error, up to a maximum of 30.

[0047] Step S4: Dynamic Gating Fusion The overall integration process for this step is as follows: Figure 2 As shown, the details are as follows: First, the extended Kalman filter (EKF) framework is used to update the state of the physical information module: Time update (prediction): ; ; in This includes the SOC update equation (based on the modified Butler-Volmer equation), the polarization voltage update equation (based on the RC circuit), the temperature update equation (based on the thermal model), and the strain / stress update equation (based on the mechanical model). It is the Jacobian matrix, obtained through sign differentiation or finite difference.

[0048] Measurement Update (Correction): ; ; Measurement function This includes output voltage (calculated based on open-circuit voltage and polarization voltage), surface temperature (directly output from the state), total strain (reversible + irreversible), and surface stress (calculated through elastic constitutive relations).

[0049] Then, calculate the uncertainty index of the physical model: ;in This is the measured output vector; For measurement functions; For prior estimation of the physical model; This measure is used to measure the trace of the noise covariance matrix; this metric reflects the degree of deviation between the physical model predictions and the measured data.

[0050] eigenvectors Input gating network, output fused weights ; In this embodiment, the gated network uses a three-layer feedforward neural network with a hidden layer dimension of 32 and an activation function of ReLU.

[0051] KAN compensation amount calculate: ;in It is a multiple-input multiple-output (MIMO) KAN network with an output dimension of 9.

[0052] Final fusion estimate: Final battery state estimate Represented as: ,in This is element-wise multiplication; These are estimates from the physical model.

[0053] Step S5: Output evaluation results from The estimated values ​​of SOC and SOH are extracted and output to the vehicle controller via the CAN bus. Simultaneously, the estimation results are stored in local non-volatile memory for subsequent long-term aging trend analysis.

[0054] Training process: In this embodiment, the KAN network is trained using a historically acquired cyclic aging dataset of 174 batteries (including different temperatures, rates, and depths of discharge). The loss function used to train the Kolmogorov-Arnold network includes a physical consistency constraint term, which includes: Energy conservation constraint Defined as: , where N is the number of samples; This is the correction amount for the SOC predicted by the KAN network; Coulomb efficiency is taken as 0.98-0.999 during charging and 1 during discharging. Let be the current at time k; The sampling time interval; This is the maximum battery capacity; For estimated health status; Thermodynamic consistency constraint Defined as: ,in Temperature estimated for the physical model; This is the temperature correction amount predicted by the KAN network; The temperature of the physical model at time k-1; This refers to the change in temperature. Total heat generation rate; Mechanical consistency constraint Defined as: ,in V represents the reversible strain correction predicted by the KAN network; V is the total electrode volume. This represents the partial molar volume of lithium ions. The concentration change field predicted by the KAN network; The total loss function is a weighted sum of the three physical consistency constraint terms and the prediction error term.

[0055] In this embodiment, SOC prediction is taken as an example: Total loss function Designed as follows: ; where the prediction error term Represented as: , For a true SOC; These are the weighting coefficients; In this embodiment, take This was determined through cross-validation.

[0056] Engineering basis for weight selection: The order reflects the hierarchy of importance of the three constraints in actual battery operation. Energy conservation is the "first principle" of the battery, and thermodynamic constraints are naturally satisfied under normal operating conditions (only taking effect under abnormal conditions), so they are given the lowest weight to avoid over-regularization.

[0057] The Adam optimizer was used with an initial learning rate of 0.001, a batch size of 64, and 200 training epochs. The learning rate was decayed to 0.5 times its original value every 50 epochs.

[0058] Experimental results: To verify the effectiveness of the method of this invention, a comparison was made with existing methods on the same test dataset: (1) SOC estimation accuracy

[0059] (2) Precision of SOH estimation After 500 cycles of aging experiments, the mean absolute error (MAE) of SOH estimation by the method of this invention is 1.87%, while the MAE of the traditional EKF method is 4.23%.

[0060] (3) Early warning lead time In the overcharge-induced thermal runaway experiment, the method of this invention, based on the anomaly detection of strain signals, issues an early warning on average 57 seconds earlier than the traditional temperature threshold method; in the internal short circuit experiment, it issues an early warning on average 31 seconds earlier.

[0061] (4) Small sample learning ability When the amount of training data is reduced to 25% of the complete dataset, the SOC estimation RMSE of the method of this invention is 1.34%, while the RMSE of the pure data-driven LSTM method rises to 3.87%, indicating that physical constraints significantly enhance the model's small sample generalization ability.

[0062] Example 2: This embodiment is basically the same as Embodiment 1, except that: The strain signal is acquired using an embedded fiber Bragg grating (FBG) sensor array. During the cell winding process, three FBG sensors are embedded at the interfaces of the positive electrode, negative electrode, and separator, respectively, to achieve distributed measurement of internal strain. Each FBG sensor has 5 grating points etched on it, with an axial spacing of 20 mm, for a total of 15 measurement points.

[0063] After the FBG sensor was implanted, the battery's capacity retention rate after 1000 cycles was 93.74%, which was not significantly different from the 94.12% without the implanted battery, indicating that the impact of the embedded sensing solution on battery performance is negligible.

[0064] Distributed strain data enables the physical information module to build a more accurate local stress-diffusion coupling model: the stress state at different spatial locations is monitored independently, and the stress gradient term in the diffusion equation is... It can be directly calculated instead of relying on model estimation, which significantly improves the accuracy of the thermo-baric co-diffusion dynamics submodule.

[0065] Example 3: This embodiment is basically the same as Embodiment 1, except that: The method is deployed in the battery management system of a large-scale energy storage power station.

[0066] System Scale: A 20-foot containerized energy storage system comprising 10 battery clusters, totaling 2016 cells. Each battery cluster is equipped with a local BSU (Battery Management Slave Unit) responsible for acquiring multiphysics data, including strain signals; a central BMU (Battery Management Master Unit) runs the PEHKAN algorithm of this invention, outputting SOC / SOH estimates and safety status for each cell.

[0067] Data from six months of actual operation shows that, compared to traditional passive management strategies based on voltage balancing, the method of this invention increases the overall available capacity of the system by approximately 8.2% and reduces the number of protective shutdowns triggered by battery inconsistencies by 67%. Simultaneously, through real-time analysis of strain signals, three cells with abnormally rising internal resistance were identified in advance, preventing potential thermal runaway incidents.

[0068] Example 4: This example is basically the same as Example 1, except that: The method described herein is applied to the optimization of charging strategies in fast charging scenarios for electric vehicles. During 350kW high-power fast charging, traditional methods employ conservative charging current limits (typically not exceeding 2C) to avoid lithium plating. The method of this invention, by real-time monitoring of the reversible strain evolution during charging, determines whether the electrode material has entered the lithium plating risk zone (characterized by a sudden change in the slope of the strain-SOC curve) and dynamically adjusts the charging current.

[0069] Test results show that, under the premise of ensuring lithium plating safety, the charging current can be increased to a maximum of 3.2C, reducing the charging time from 10% to 80% SOC from 22 minutes to 15 minutes, improving the charging efficiency by about 32%, while keeping the battery temperature rise within the safety threshold.

[0070] Comparative example: To highlight the technical effects of the present invention, the following comparative experiments were conducted: Comparative Example 1: The traditional EKF method is used, which only uses voltage and current signals, without strain information and KAN compensation.

[0071] Comparative Example 2: Using a pure data-driven LSTM method, input voltage, current, temperature and strain signals, without a physical model and without dynamic gating fusion.

[0072] Comparative Example 3: An electro-thermal coupling model and Kalman filtering method are used to introduce temperature information but without mechanical signal coupling.

[0073] Comparative Example 4: The method of the present invention is used, but the dynamic gating fusion mechanism is turned off (i.e., fixed weight fusion).

[0074] The experimental results are shown in the table below:

[0075] Experimental results show that the method of this invention is the best in all accuracy indicators, especially under low temperature and small sample conditions (73% and 73% improvement respectively compared to Comparative Example 1). Although the computation time is increased compared to the traditional EKF, it is still within the allowable range of BMS real-time computation (100Hz sampling interval corresponds to 10ms / step, and 16.8ms / step of this method is basically feasible, and can be further compressed to within 12ms through code optimization).

[0076] The lithium-ion battery system state assessment method based on strain monitoring provided by this invention can be widely applied to: Electric vehicle battery management system; Battery operation and maintenance system for energy storage power stations; Monitoring of power batteries for drones and electric aircraft; Intelligent power management for portable electronic devices; Rapid screening and sorting for battery reuse.

[0077] This method improves the SOC estimation accuracy of lithium-ion batteries from 2%-5% of traditional methods to within 1%, and the SOH estimation accuracy from 3%-6% to within 2%. It also extends the thermal runaway warning time from several seconds to minutes (average 30-60 seconds), which is of great value for improving battery system safety, extending service life, and reducing total lifecycle costs. The proposed physical-enhanced hybrid Kolmogorov-Arnold network structure is not only applicable to lithium-ion batteries but can also be extended to the state assessment of other complex systems involving multi-physics coupling, such as fuel cells, supercapacitors, and solid-state batteries, demonstrating broad versatility and industrialization prospects.

[0078] Example 5: See Figure 3 The present invention also provides a lithium-ion battery system based on strain monitoring that can be applied to actual production, comprising: A multi-physics sensing unit is used to collect multi-physics monitoring data in real time during the charging and discharging process of lithium-ion batteries; The physical information module, whose input is connected to the multiphysics sensing unit, includes: The stress-coupled electrochemical potential energy submodule is used to establish an explicit physical mapping between stress signals and electrochemical reaction kinetics; The temperature-pressure co-controlled diffusion dynamics submodule is used to establish a lithium-ion diffusion model under the coupling effect of temperature and stress. The multiphysics coupled thermal submodule is used to build a heat generation model that includes the contribution of stress dissipation. The data-driven compensation module has its input connected to the multiphysics sensing unit. The data-driven compensation module is constructed using a Kolmogorov-Arnold network to learn the residual relationships that the physical information module fails to describe. The dynamic gating fusion unit has a first input terminal connected to the output terminal of the physical information module and a second input terminal connected to the output terminal of the data-driven compensation module. It is used to adaptively fuse the output of the physical information module and the output of the data-driven compensation module according to the prediction confidence of the physical information module under the current operating conditions to generate a fused battery state estimate. The state assessment output unit, whose input is connected to the output of the dynamic gating fusion unit, is used to output the state of charge and health assessment results of the lithium-ion battery based on the fused battery state estimate.

[0079] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the state of a lithium-ion battery system based on strain monitoring, characterized in that, Includes the following steps: Acquire multi-physics field monitoring data of lithium-ion batteries during charging and discharging processes, wherein the multi-physics field monitoring data includes at least current signals, voltage signals, temperature signals, and strain force signals; A physical information module is constructed to establish an explicit physical mapping from the multi-physics field monitoring data to the battery state variables. The physical information module includes a stress-coupled electrochemical potential energy submodule, a temperature-pressure co-controlled diffusion dynamics submodule, and a multi-physics field coupled thermal submodule. A data-driven compensation module is constructed using a Kolmogorov-Arnold network to learn complex nonlinear residual relationships that the physical information module fails to describe; Based on the dynamic gating fusion mechanism, the outputs of the physical information module and the data-driven compensation module are adaptively fused according to the prediction confidence of the physical information module under the current operating conditions to obtain the fused battery state estimate. Based on the fused battery state estimate, the state of charge and health status assessment results of the lithium-ion battery are output.

2. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The stress-coupled electrochemical potential energy submodule employs a modified Butler-Volmer equation and introduces a correction term for the stress-induced electrochemical reaction barrier. The correction term is expressed as: ,in This refers to the change in Gibbs free energy caused by stress. The stress coupling factor, which is SOC-dependent, is learned through the Kolmogorov-Arnold network. This is a stress signal; is the partial molar volume of lithium ions; F is the Faraday constant; It is Avogadro's constant; This is an overpotential.

3. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The thermo-pressure co-controlled diffusion dynamics submodule employs a stress-corrected diffusion coefficient tensor. , is represented as: ,in Pre-exponential factors; It is the diffusion activation energy; For hydrostatic stress; R is the diffusion activation volume; k is the stress-diffusion coupling coefficient; R is the gas constant; T is the temperature; I is the unit tensor. And establish the thermo-baric coupled diffusion equation satisfied by the lithium ion concentration field: , where c is the lithium ion concentration; This is the source term for electrochemical reactions.

4. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The multiphysics coupled thermal submodule employs an extended Bernardi heat generation model, which adds a stress dissipation heat dissipation term in addition to irreversible heat, reversible heat, and Joule heating. , is represented as: ,in For von Mises equivalent stress; Plastic strain rate; For electrode volume; The mechanical energy to heat energy conversion coefficient; The stress dissipation heat dissipation term is assigned to the thermal balance equations for the core temperature state and surface temperature state of the battery.

5. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The dynamic gating fusion mechanism includes: Construct a gating network with input features including the current physical model prediction state, input signal, observable output signal, and physical model uncertainty index; The uncertainty index of the physical model Defined as: ,in This is the measured output vector; For measurement functions; For prior estimation of the physical model; To measure the trace of the noise covariance matrix; The gated network outputs a fusion weight vector. The final battery state estimate Represented as: ,in This is element-wise multiplication; These are estimates from the physical model; The compensation amount output by the Kolmogorov-Arnold network.

6. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The physical information module also includes a mechanical model, which will include the total strain. Decomposed into reversible strain and irreversible strain ; The evolution of reversible strain satisfies: The strain-SOC sensitivity coefficient is obtained by learning from the Kolmogorov-Arnold network; Irreversible evolution satisfies: This is the cumulative coefficient of irreversible strain; This is the critical strain threshold. It is the activation energy.

7. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The loss function used to train the Kolmogorov-Arnold network includes a physical consistency constraint, which includes: Energy conservation constraint Defined as: , where N is the number of samples; This is the correction amount for the SOC predicted by the KAN network; Coulomb efficiency is taken as 0.98-0.999 during charging and 1 during discharging. Let be the current at time k; The sampling time interval; This is the maximum battery capacity; For estimated health status; Thermodynamic consistency constraint Defined as: ,in Temperature estimated for the physical model; This is the temperature correction amount predicted by the KAN network; The temperature of the physical model at time k-1; This refers to the change in temperature. Total heat generation rate; Mechanical consistency constraint Defined as: ,in V represents the reversible strain correction predicted by the KAN network; V is the total electrode volume. This represents the partial molar volume of lithium ions. The concentration change field predicted by the KAN network; The total loss function is a weighted sum of the three physical consistency constraint terms and the prediction error term.

8. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The Kolmogorov-Arnold network uses B-spline basis functions for parameterization of univariate functions: ,in For B-spline basis functions; These are the learnable spline coefficients; and These are the coefficients of the linear and bias terms; The Kolmogorov-Arnold network comprises at least three independent subnetworks: a first KAN for learning the stress coupling factor, a second KAN for learning the strain-SOC sensitivity coefficient, and a third KAN for learning the aging compensation amount.

9. The method for assessing the state of a lithium-ion battery system based on strain monitoring according to claim 1, characterized in that, The method further includes a recursive filtering update step: using an extended Kalman filter framework to perform time and measurement updates on the physical information module, wherein the state vector of the physical information module includes: state of charge, polarization voltage, core temperature, surface temperature, reversible strain, irreversible strain, maximum internal stress, health status, and aging characteristic parameters.

10. A lithium-ion battery system based on strain monitoring, for implementing the method as described in any one of claims 1-9, characterized in that, include: A multi-physics sensing unit is used to collect multi-physics monitoring data in real time during the charging and discharging process of lithium-ion batteries; A physical information module, the input of which is connected to the multiphysics sensing unit, the physical information module comprising: The stress-coupled electrochemical potential energy submodule is used to establish an explicit physical mapping between stress signals and electrochemical reaction kinetics; The temperature-pressure co-controlled diffusion dynamics submodule is used to establish a lithium-ion diffusion model under the coupling effect of temperature and stress. The multiphysics coupled thermal submodule is used to build a heat generation model that includes the contribution of stress dissipation. A data-driven compensation module, whose input is connected to the multiphysics sensing unit, is constructed using a Kolmogorov-Arnold network to learn the residual relationships that the physical information module fails to describe. The dynamic gating fusion unit has a first input terminal connected to the output terminal of the physical information module and a second input terminal connected to the output terminal of the data-driven compensation module. It is used to adaptively fuse the output of the physical information module and the output of the data-driven compensation module according to the prediction confidence of the physical information module under the current operating conditions to generate a fused battery state estimate. The state assessment output unit, whose input is connected to the output of the dynamic gating fusion unit, is used to output the state of charge and health status assessment results of the lithium-ion battery based on the fused battery state estimate.

Images