A method for discriminating early signs of thermal runaway of a sodium-ion battery and a health management system

Abstract

The application provides a sodium ion battery thermal runaway early sign discrimination method and health management system, and relates to the technical field of artificial intelligence computer vision. The discrimination method specifically comprises the following steps: acquiring a two-dimensional image sequence of a battery surface, separating a global motion compensation reference and a small residual displacement field, and outputting a suspected abnormal local coordinate and an environmental common mode vibration frequency spectrum; eliminating background noise at the suspected abnormal local coordinate in combination with the environmental common mode vibration frequency spectrum, and extracting a characteristic deformation frequency; determining an orthogonal excitation frequency based on the characteristic deformation frequency and the environmental common mode vibration frequency spectrum, injecting a micro-current pulse sequence into the battery, synchronously extracting an excited main response displacement signal and a dynamic terminal voltage response signal, and calculating a cross-modal phase lag angle; generating an intrinsic lag alarm threshold based on an environmental absolute temperature and a health state scalar, and comparing the cross-modal phase lag angle with the intrinsic lag alarm threshold to diagnose early thermal runaway or identify mechanical artifacts.

Description

Technical Field

[0001] This application relates to the fields of health management and artificial intelligence computer vision technology, and in particular to a method for identifying early signs of thermal runaway in sodium-ion batteries and a health management system. Background Technology

[0002] With the rapid development of new energy vehicles and large-scale energy storage systems, sodium-ion batteries have been widely used due to their abundant resources and excellent performance over a wide temperature range. However, with the increase in battery system energy density and the increasing complexity of real-world application environments, the safety hazards of power batteries are becoming increasingly prominent. High-precision hazard prediction and risk assessment of early signs of battery thermal runaway has become a core element in ensuring the safe and reliable operation of new energy systems. Therefore, the development of advanced automotive-grade intelligent monitoring devices and systems is an urgent industrial need.

[0003] Compared to traditional lithium-ion batteries, sodium-ion batteries possess unique physical and electrochemical characteristics. The insertion and extraction of sodium ions into and out of the hard carbon anode triggers highly nonlinear micro-expansion and contraction within the battery cell. More critically, in the very early stages of thermal runaway (such as localized micro-short circuits or SEI film decomposition), trace amounts of gas are typically generated. This early gas production causes drastic changes in the impedance and mechanical damping of the gas-liquid-solid multiphase interfaces within the battery. These mechanical response characteristics caused by structural degradation occur significantly earlier than the rapid rise in the battery's macroscopic temperature. Therefore, highly sensitive online performance testing and online non-destructive testing of individual battery cells are crucial for achieving very early warning systems.

[0004] Currently, existing health management systems (BMS) primarily rely on the passive acquisition and data processing of one-dimensional scalar parameters such as voltage, current, and surface temperature. However, the changes in these electrical parameters are extremely subtle in the very early stages of thermal runaway, easily masked by normal parameter fluctuations caused by high-rate charging and discharging in actual operating conditions. This makes it difficult to extract effective early warning features from massive amounts of industrial big data. By the time existing intelligent sensing systems detect a significant abnormal temperature rise, thermal runaway has often already entered a violent chain reaction phase. Traditional passive electrical indicator discrimination methods suffer from severe lag and are insufficient to meet the demands of modern new energy industries for system-level fault prediction and health management.

[0005] To compensate for the limitations of single electrical monitoring, some existing technologies attempt to introduce basic external monitoring equipment to measure battery expansion vibrations. However, in complex energy storage cabinet or vehicle-mounted operating scenarios, battery systems are subjected to high-frequency environmental vibrations and alternating external stresses for extended periods. Traditional computer vision and image processing technologies are prone to signal aliasing when faced with vehicle bumps or cabinet fan resonance. Existing pattern recognition algorithms often fail to accurately extract features from complex global common-mode mechanical interference, easily misinterpreting normal mechanical vibrations as deformation caused by thermal runaway, resulting in a very high false alarm rate.

[0006] More seriously, existing battery safety monitoring methods are often unidirectional and operate in isolation. There is a lack of in-depth cross-validation and dynamic constraint mechanisms between external mechanical vibration analysis and internal electrochemical polarization state. Existing monitoring systems cannot achieve cross-modal data fusion, nor can they use visually extracted physical parameters to guide and calibrate the active electrical diagnostics of the BMS. This technological fragmentation prevents existing safety algorithms from forming an effective digital signal processing closed loop, making it difficult to robustly identify the extremely subtle early signs of thermal runaway in sodium-ion batteries, and also hindering the implementation of efficient edge computing and spatial computing optimization on vehicle-side chips with limited computing power. Summary of the Invention

[0007] This invention provides a method for identifying early signs of thermal runaway in sodium-ion batteries, the method comprising: Two-dimensional image sequences of the battery surface are acquired, the global motion compensation reference and the small residual displacement field are separated, and the coordinates of suspected abnormal local areas and the common mode vibration spectrum of the environment are output. At suspected abnormal local coordinates, background noise is eliminated by combining the environmental common-mode vibration spectrum, and characteristic deformation frequencies are extracted; Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, a micro-current pulse sequence is injected into the battery, the excited principal response displacement signal and the dynamic terminal voltage response signal are extracted simultaneously, and the transmodal phase hysteresis angle is calculated. Intrinsic hysteresis alarm thresholds are generated based on ambient absolute temperature and health status scalars. The cross-modal phase hysteresis angle is compared with the intrinsic hysteresis alarm thresholds to diagnose early thermal runaway or identify mechanical artifacts.

[0008] The process of separating the global motion compensation benchmark from the micro residual displacement field includes: locking the static benchmark anchor point on the external module frame in the two-dimensional image sequence, performing feature extraction and trajectory tracking; establishing a two-dimensional affine transformation model that includes rotation, translation, scaling and shear deformation, solving the global affine transformation matrix and translation vector in real time to form a global motion compensation benchmark characterizing global mechanical vibration; tracking the actual observed coordinates of the visual label on the battery surface, subtracting them from the theoretical compensation coordinates calculated after substituting them into the global motion compensation benchmark, and obtaining the micro residual displacement field stripped of the influence of global mechanical vibration.

[0009] To achieve vibration measurement and data processing, the system outputs suspected abnormal local coordinates and environmental common-mode vibration spectra, including: obtaining the Cauchy strain tensor by spatially differentiating the small residual displacement field; calculating the spatial gradient magnitude of the Cauchy strain tensor; extracting the centroid coordinates of connected regions with spatial gradient magnitudes greater than a preset calibration threshold as suspected abnormal local coordinates; selecting a sliding time window to extract the signed components of the translation vector containing alternating oscillation characteristics; performing fast Fourier transforms after mean reduction processing to synthesize the environmental common-mode vibration spectrum in two-dimensional space.

[0010] Extracting characteristic deformation frequencies includes: acquiring two-dimensional micro-deformation components at suspected abnormal local coordinates in a local high-frequency focusing mode and calculating their high-frequency two-dimensional spatial spectrum; constructing an adaptive spectrum suppression mask based on energy normalization based on the maximum value of the environmental common-mode vibration spectrum; multiplying the high-frequency two-dimensional spatial spectrum with the adaptive spectrum suppression mask frequency by frequency and performing dynamic filtering to eliminate the frequency response excited by broadband mechanical vibration of the road surface, thus obtaining the micro-characteristic spectrum; extracting the frequency point corresponding to the absolute maximum amplitude in the micro-characteristic spectrum and locking it as the characteristic deformation frequency.

[0011] Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, and a micro-current pulse sequence is injected into the battery. This includes: taking the characteristic deformation frequency as the center frequency, finding the frequency point corresponding to the minimum value of the energy amplitude of the environmental common-mode vibration spectrum within the preset effective resonant frequency offset range, and establishing it as the optimal orthogonal excitation frequency; continuously injecting a sinusoidal AC reference excitation current signal containing the optimal orthogonal excitation frequency into the target single cell, which is converted into electromechanical resonant stress with the same frequency as the optimal orthogonal excitation frequency.

[0012] Using digital signal processing algorithms, the stimulated principal response displacement signal and dynamic terminal voltage response signal are extracted synchronously to calculate the cross-modal phase hysteresis angle. This includes: aligning the timestamps of the visual acquisition sequence with the timestamps of the current output; projecting the extracted micro-deformation components onto the principal vibration direction to obtain a signed one-dimensional stimulated principal response displacement signal, and simultaneously acquiring the dynamic terminal voltage response signal; generating orthogonal and in-phase reference signals with the same frequency as the reference excitation current signal; performing cross-modal digital phase-locked amplification operations on the one-dimensional stimulated principal response displacement signal and the dynamic terminal voltage response signal respectively within a set integer-cycle phase-locked integration time window to extract the electro-resonance displacement amplitude, the initial phase angle of the mechanical response, and the electrical polarization phase angle; introducing a pre-calibrated hardware-level cross-modal synchronization delay constant to calculate the fixed hardware phase offset; performing differential calculations on the initial phase angle of the mechanical response, the electrical polarization phase angle, and the fixed hardware phase offset; and outputting the cross-modal phase hysteresis angle after adding a phase winding reduction algorithm.

[0013] To achieve fault prediction and health management of battery systems, an eigenhysteresis alarm threshold is generated based on ambient absolute temperature and health status scalars. This includes: synchronously reading the ambient absolute temperature and current health status scalars of individual target cells; calculating a nonlinear dynamic baseline by combining pre-calibrated baseline hysteresis angle constant, material viscous flow activation energy parameters, ideal gas constant, reference absolute temperature, and aging coupling gain coefficient; and introducing the upper limit of the hysteresis of the forced vibration system as a physical extreme constraint to prevent missed alarms. The nonlinear dynamic baseline and the upper limit of the hysteresis of the forced vibration system are compared with a minimum amplitude limit to generate the eigenhysteresis alarm threshold.

[0014] The comparison is used to confirm early thermal runaway or identify mechanical artifacts, including hazard prediction and pattern recognition logic: if the cross-modal phase hysteresis angle is greater than or equal to the intrinsic hysteresis alarm threshold, it is determined to be an early gas-generating thermal runaway diagnosis, and a true positive alarm command is output; if the absolute value of the cross-modal phase hysteresis angle is less than or equal to a phase shift threshold close to zero, and the electro-resonance displacement amplitude is greater than or equal to a preset abnormal amplitude threshold, it manifests as undamped dry friction or elastic slapping, and is diagnosed as a vehicle mechanical component artifact.

[0015] As an edge computing power optimization mechanism for intelligent monitoring devices, after identifying mechanical artifacts, a closed-loop mask feedback step to prevent false alarms is also included: when a vehicle mechanical component artifact is confirmed at the target point coordinates, the two-dimensional spatial mask matrix used for the underlying global calculation is reconstructed, and the mask elements within the connected domain of the artifact are corrected to zero; the updated two-dimensional spatial mask matrix is ​​fed back to the global real-time monitoring loop, and the Hadamard product mask filtering operation of element-wise scalar multiplication is forcibly performed on the subsequently extracted small residual displacement field to remove the possibility of triggering anomalies in the spatial region strain gradient.

[0016] The present invention also provides a health management system, the system comprising: Environmental common-mode vibration spectrum extraction module: acquires two-dimensional image sequences of the battery surface, separates the global motion compensation reference from the small residual displacement field, and outputs the coordinates of suspected abnormal local areas and the environmental common-mode vibration spectrum; Deformation frequency extraction module: At suspected abnormal local coordinates, background noise is eliminated by combining the environmental common-mode vibration spectrum, and characteristic deformation frequencies are extracted; Phase calculation module: Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, a micro-current pulse sequence is injected into the battery, the excited main response displacement signal and the dynamic terminal voltage response signal are extracted simultaneously, and the cross-modal phase hysteresis angle is calculated. Judgment module: Based on the ambient absolute temperature and health status scalar, the intrinsic hysteresis alarm threshold is generated. The cross-modal phase hysteresis angle is compared with the intrinsic hysteresis alarm threshold to diagnose early thermal runaway or identify mechanical artifacts.

[0017] This application proposes a method for identifying early signs of thermal runaway in sodium-ion batteries and a health management system for new energy vehicles and large-scale energy storage systems under complex operating conditions. As an advanced intelligent monitoring device and intelligent sensing system, this invention overcomes the interference of strong mechanical vibration environments such as vehicle driving bumps and cabinet fan vibrations on weak physical signals. Through cross-modal multi-coupling analysis, it achieves precise monitoring of battery safety status in harsh environments with extremely low signal-to-noise ratios, greatly improving the system's safety assurance capabilities.

[0018] This invention innovates in the fields of computer vision and image processing. The system acquires visual label image sequences from the surface of battery packs and performs high-precision feature extraction using pattern recognition and image analysis techniques. Facing global common-mode mechanical interference, instead of relying on traditional absolute deformation thresholds, it employs dynamic image filtering and image enhancement techniques to separate global displacement from minute non-rigid residual deformations in real time. This image recognition and processing effectively identifies suspected anomalous regions with high strain gradients at the target recognition level, avoiding the aliasing of visual information with global environmental noise, and providing a highly accurate spatial reference for subsequent target detection.

[0019] In the field of mechanical vibration analysis and measurement, this invention achieves high-fidelity extraction of minute electromechanical resonance signals. The system extracts the intrinsic characteristic deformation frequency by performing frequency domain transformation and signal pattern recognition on the local minute residual displacement field. Subsequently, the system injects customized orthogonal alternating current pulses based on this characteristic frequency, transforming the imperceptible early electrochemical decay into a mechanical resonance response of the same frequency. Utilizing a cross-modal digital lock-in amplification algorithm, the system performs vibration measurement and resonance detection in the frequency domain, effectively eliminating mechanical vibration noise in a wideband environment. This mechanism achieves a precision inspection effect similar to online non-destructive testing of structures, accurately acquiring the cross-modal phase hysteresis response reflecting changes in internal gas-liquid mixed-phase damping.

[0020] At the level of industrial big data processing, fault prediction, and health management, this system constructs a multi-dimensional state-coupled nonlinear dynamic intrinsic threshold. The system integrates real-time industrial big data such as current ambient absolute temperature and battery health status for in-depth state analysis, effectively offsetting temperature drift interference caused by slowed electrochemical polarization. Based on the calculated cross-modal phase hysteresis angle, the system can perform high-confidence hazard prediction and risk assessment, not only confirming true positives for very early thermal runaway but also reversing the identification of mechanical structure artifacts such as loose fasteners. This intelligent system-level data processing greatly improves the accuracy of fault prediction and prevents data omissions and false alarms caused by extreme temperatures or long-term aging.

[0021] In summary, this invention deeply integrates image processing, mechanical vibration analysis, and fault prediction technologies to form a highly robust industrial control software logic detection system. By achieving local computational power optimization at the edge computing level through the final cross-modal high-dimensional discrimination result, the ineffective consumption of system computing resources is avoided. This invention transforms visual deformation and electrical polarization into a unified cross-modal intrinsic state indicator, providing a highly exclusive online material performance testing and safety monitoring method for power batteries and energy storage systems, possessing extremely high industrial practical value. Attached Figure Description

[0022] Figure 1 This is a waveform diagram showing the separation of small residual displacements, where... Figure 1 (a) Original observation coordinate sequence, Figure 1 (b) Macroscopic motion compensation benchmark, Figure 1 (c) Extracted small residual displacement field; Figure 2 Frequency domain orthogonal mask and characteristic deformation frequency extraction spectrum, where... Figure 2 (a) Environmental common-mode vibration spectrum, Figure 2 (b) Local microscopic spectrum, Figure 2 (c) Adaptive spectral suppression mask, Figure 2 (d) High-purity micro-feature spectrum; Figure 3 This is a diagram illustrating the convergence process of the modal digital phase-locked amplification of the present invention, wherein, Figure 3 (a) is a graph showing the change in the principal response displacement signal under strong environmental noise. Figure 3 (b) is a graph showing the variation of the orthogonal excitation reference current signal. Figure 3 (c) Convergence curve of cross-modal orthogonal integral Figure 4 This is a flowchart illustrating the early signs of thermal runaway in sodium-ion batteries according to the present invention. Detailed Implementation

[0023] This embodiment provides a method for identifying early signs of thermal runaway in sodium-ion batteries and a health management system.

[0024] This embodiment is applied to battery pack systems under continuous mechanical vibration conditions, preferably new energy vehicles (such as electric passenger cars and new energy heavy trucks) in motion or energy storage cabinets with high-power fans and mechanical vibrations. In this application scenario, the battery pack as a whole and its internal structural components are subject to wide-frequency mechanical vibration interference (e.g., low-frequency vibrations of 1Hz to 50Hz caused by road bumps). The spatial displacement amplitude of these mechanical vibrations is much greater than the local deformation caused by early trace gas production or solid-liquid phase transitions inside the battery, and the two overlap in some frequency domains. This system is configured to decouple common-mode vibrations and extract features in real time in the strong vibration environment without interrupting vehicle operation or energy storage cabinet operation.

[0025] The system's hardware structure mainly includes: a multimodal vision acquisition module, an adaptive active electro-excitation health management system (BMS), and a cross-modal central processing unit. The physical layout and functional limitations of each hardware component are as follows: The system monitors a battery pack composed of multiple individual cells. Preferably, the individual cells are sodium-ion batteries using hard carbon negative electrodes. A low-cost visual tag is attached to the surface of the casing of each individual cell, preferably in the peripheral edge area of ​​the explosion-proof valve on the top cover or in a deformable area of ​​the side wall. The visual tag includes a high-contrast disordered speckle pattern or a high-frequency moiré pattern, the feature size of which is configured to adapt to the pixel resolution of the visual acquisition module to provide two-dimensional texture variation features. Preferably, the visual tag is made of a high-temperature resistant, electrolyte-corrosion-resistant acrylic polymer material and is rigidly attached to the battery casing using a hot-pressing process to ensure that the displacement of the tag is strictly equivalent to the mechanical deformation of the battery casing.

[0026] The multi-mode vision acquisition module includes a monocular miniature camera array. The miniature camera array is fixed to the inside of the battery pack's top cover or to the support structure of the module frame, with the lens's field of view facing downwards or to the side, covering the visual label of the battery group.

[0027] The aforementioned fixing method ensures that the coordinate system of the miniature camera array remains consistent with the relative motion of the battery pack casing, thus providing a relative reference for subsequent decoupling of global displacement. The multi-mode vision acquisition module has the ability to dynamically switch operating modes via hardware registers, specifically configured as follows: Global low-frame scanning mode: Global image acquisition is performed across the entire field of view at a set first frame rate (preferably 1fps to 5fps) to extract a large-scale environmental vibration baseline. Local high-frequency focusing mode: Based on external instructions, a specific region of interest (ROI) within the field of view is locked, and the acquisition frame rate of that region is instantly increased to a second frame rate (preferably 120fps to 240fps or higher), while acquisition of non-ROI regions is paused. This allows for the acquisition of high-frequency optical flow timing signals in local areas without significantly increasing transmission bandwidth and computational resource consumption.

[0028] The adaptive active electrical excitation health management system (BMS) includes, in addition to traditional passive monitoring circuits for voltage, current, and temperature, an adaptive active micro-current excitation circuit is substantially integrated.

[0029] The adaptive active microcurrent excitation circuit is connected to the two poles of each individual cell within the battery pack. This excitation circuit integrates a digital frequency synthesizer (DDS) and a programmable power amplifier module. It is configured to receive orthogonal excitation frequency and amplitude commands from the cross-modal central processing unit and, based on these commands, instantaneously inject a microcurrent pulse sequence containing a specific AC frequency (preferably with a modulation range of 10Hz to 1000Hz) into the target individual cell. This microcurrent pulse sequence is configured to actively excite electromechanical resonance at the solid-liquid-gas mixed-phase interface through changes in the electrochemical impedance within the cell.

[0030] The cross-modal central processing unit is located on the main control board inside the vehicle domain controller or battery pack, and it establishes data connections with the multi-mode vision acquisition module and BMS through the vehicle high-speed bus (such as CAN-FD or vehicle Ethernet).

[0031] Preferably, the cross-modal central processing unit integrates a hardware-level Precise Time Protocol (PTP) clock synchronization module. This clock synchronization module is configured to align the timestamps of the two-dimensional image sequence frames output by the multi-modal vision acquisition module with the timestamps of the BMS performing adaptive active microcurrent excitation and voltage / current sampling at the microsecond level. This hardware-level time alignment is a prerequisite for subsequent frequency-domain quadrature phase-locked loop extraction and calculation of the cross-modal phase hysteresis angle.

[0032] This embodiment provides a method for identifying early signs of thermal runaway in sodium-ion batteries, specifically disclosing the implementation details of step S1 (environmental common-mode vibration background and extraction of suspected abnormal local areas).

[0033] In high-vibration scenarios such as moving vehicles or energy storage cabinets with large cooling fans, the core technical challenge of the system is that the overall mechanical vibration amplitude of the battery pack is much greater than the surface physical bulging amplitude caused by the early trace gas generation of individual battery cells. Furthermore, the battery pack casing and module frame supporting the camera are not ideal rigid bodies under continuous alternating stress and will experience low-frequency torsion and bending. If the absolute displacement of feature points is directly extracted from the image sequence, the results will be completely overwhelmed by global environmental vibration noise.

[0034] Therefore, step S1 in this embodiment is based on the principles of mechanics and relative kinematics. Through continuous mathematical tensor operations, the global motion is stripped away in real time, and the minute non-rigid deformations that belong purely to a single cell are purified. The specific steps are as follows: S1.1: Separation of global affine motion and minute deformation The cross-modal central processing unit acquires the sequence of two-dimensional images continuously output by the multimodal vision acquisition module in global low-frame scanning mode. The time variable is defined as... The unit is seconds ( ).

[0035] Global affine motion background solution: In the two-dimensional image sequence, the system locks the battery module frame. Fixed connectors, such as module fastening bolts or external support ribs, serve as static reference anchor points. It is important to note that these reference anchor points are spatially derived from the module's outer frame, maintaining strict isolation from the deformable surfaces of the individual cells. This anchor point selection mechanism ensures that the subsequent affine model only fits the torsion and vibration of the outer frame, without absorbing or offsetting the local nonlinear bulging characteristics of the individual cells.

[0036] Defined at the initial time , No. The spatial coordinate vectors of the static reference anchor points are: At the current moment The spatial coordinate vector is All coordinate parameters are calibrated to the unit of length, millimeter (mm). ).

[0037] To eliminate low-frequency torsional and vibration interference of the battery pack frame under external forces, a general two-dimensional affine transformation model incorporating rotation, translation, scaling, and shear deformation is established. The global affine transformation matrix is ​​solved in real time using the least squares method. By minimizing the objective function Solve for the current time. Global affine matrix (Unitless) and translation vector ( The above parameters together constitute the motion compensation benchmark characterizing the current vehicle's mechanical vibration and frame torsion.

[0038] Calculation of small residual displacement fields: For visual labels on the surface of individual cells, the Digital Image Correlation (DIC) algorithm is used to track speckle patterns on arbitrary surfaces. Initial coordinates at time At the current moment Actual observed coordinates The initial coordinates Substitute into the affine transformation model and calculate the theoretical compensated coordinates if the point only moves with the environment. : Subtracting the actual observed coordinates from the theoretical compensated coordinates by vector subtraction yields the small residual displacement field vector completely isolated from the effects of environmental vibration and frame deformation. : in, for The residual displacement field excludes vehicle body bumps and module bending under stress, and only characterizes the minute expansion or deformation of the individual battery cells due to changes in internal stress.

[0039] S1.2: Parameter Output To ensure the spatial accuracy and frequency domain anti-interference capability of subsequent local high-frequency feature phase-locked extraction, step S1.2 must output cascaded constraint parameters in two dimensions based on the small residual displacement field and the global motion compensation benchmark: Output parameter 1: Local coordinates of suspected anomalies The deformation caused by the early trace gas production forms a small dome shape. Although the absolute displacement at the peak is large, the deformation gradient is small, while the stress concentration is most significant at the edge of the dome. Therefore, the residual displacement field vector... By performing spatial differentiation, the Cauchy strain tensor is obtained. : Further calculation of the spatial gradient magnitude of the strain tensor : extract The centroid coordinates of connected regions that exceed a preset calibration threshold are encapsulated as suspected abnormal local coordinates. This parameter will be used as the sole spatial constraint input to steps S2 and S3, instructing the multi-mode vision acquisition module to focus its computing power and high-frequency acquisition mode on the target area corresponding to this coordinate.

[0040] Output parameter 2: Environmental common-mode vibration spectrum To address the issue of subsequent active excitation signals being swallowed up by the broadband vibration of the road surface, this system must accurately quantize and output the global vibration characteristics of the current vehicle in the frequency domain without distortion.

[0041] Extracting the translation vector components containing alternating oscillation characteristics in step S1.1 and Select the current time. The sliding time window before To avoid the fundamental frequency masking effect and DC component distortion caused by absolute rectification, the mean value of the components within the sliding window is first calculated. and Subsequently, Fast Fourier Transform (FFT) was performed on the mean-reduced AC components to calculate the single-axis amplitude spectrum. and : in, For frequency parameter ( ); and for Further synthesizing the environmental common-mode vibration spectrum in two-dimensional space. : This spectrum avoids nonlinear truncation and accurately maps the low-frequency mechanical resonance frequency distribution of the battery pack under the influence of the current road surface (such as continuous corrugated or asphalt road). Environmental common-mode vibration spectrum. Once generated, it will be sent to the Health Management System (BMS) as a forced anti-interference frequency domain mask parameter to ensure that the active microcurrent excitation frequency output by the BMS can avoid interference. The high-energy interference frequency band in the frequency domain achieves strict cross-modal frequency domain orthogonality.

[0042] Next, this embodiment provides the implementation details of step S2 (deformation feature extraction) in the above-mentioned method for identifying early signs of thermal runaway in sodium-ion batteries.

[0043] In the current strong vibration of the moving vehicle, although step S1 locks the suspected abnormal local coordinates through global decoupling, However, the system still faces a problem at this point: is the local anomaly of the high strain gradient due to the real nonlinear physical expansion caused by early gas production inside the battery (true positive), or is it simply due to the local modal resonance and 'tympanic membrane effect' artifact (false positive) that is inevitably generated by the coupled excitation of the thin-walled shell or conductive connector of the battery under the broadband vibration of the road surface and the alternating thermal stress of high-rate charging and discharging?

[0044] Relying on traditional visual absolute deformation thresholds can easily misjudge mechanical resonances excited at specific frequency bands of the road surface (such as those over speed bumps or washboard roads) as thermal runaway. Therefore, step S2 constructs a dynamic orthogonal suppression model in the frequency domain to remove mechanical artifacts and purify the true electrochemical physical characteristics. The specific steps are as follows: S2.1: Spatial Dimensionality Reduction Focusing and Wideband Timing Acquisition: The cross-modal central processing unit receives the suspected abnormal local coordinates sent in step S1. To capture high-frequency dynamic responses that include minute stress releases and changes in interface damping, the system instructs the multi-mode vision acquisition module to switch its operating mode to a local high-frequency focusing mode.

[0045] Within the field of view, the system not only captures... The region of interest (ROI) is centered, while forcibly retaining a high-frequency tracking window for at least one of the static reference anchors described in step S1.

[0046] Increase the capture frame rate of the area to the second frame rate. (Preferably 120fps to 240fps, and meets the following requirements) To conform to the Nyquist sampling law and cover all possible high-frequency mechanical vibrations and excitation frequency bands.

[0047] Set the high-frequency acquisition duration window to (unit: Within this time window, the system synchronously acquires the high-frequency displacement of the reference anchor point and substitutes it into the affine transformation model to calculate the displacement under this high-frequency sampling system. The tiny residual displacement field vector of the global frame motion was stripped at the target point. It consists of signed components in two orthogonal spatial dimensions: in, and All units are By directly reusing residual components with positive and negative directions, the fundamental frequency masking and harmonic aliasing distortion caused by absolute value rectification are avoided from the source.

[0048] To extract pure deformation features, the mean values ​​of the two components within the high-frequency time window are calculated separately. and Then, mean-reduction processing is performed to obtain the zero-mean micro-deformation component. and ( ): S2.2: Dynamic background noise cancellation based on wideband baseline: The cross-modal central processing unit processes the aforementioned micro-deformation components respectively. and Using Fast Fourier Transform (FFT), calculate the single-axis spectrum under high-frequency broadband conditions. and ( ): And synthesized into a high-frequency two-dimensional spatial spectrum of the suspected lesion area. : Meanwhile, because the reference anchor point was tracked at a high frequency in S2.1, the system... The high-frequency environmental common-mode vibration spectrum covering a wide frequency range is synchronously updated under sampling. ( This ensures that subsequent frequency domain cancellation operations are performed on the same frequency coordinate axis.

[0049] To remove local mechanical resonance artifacts, an adaptive spectral suppression mask based on energy normalization is constructed. First calculate Normalize the maximum value: Subsequently, a dynamic spectral mask was constructed. : in, The preset dimensionless penalty coefficient (preferred value range is...) If a certain frequency If the overall vibration energy of the road surface is high, then It decays exponentially and approaches Conversely, if a certain frequency band is completely free from road vibration interference, then Approaching .

[0050] Local small spectrum With dynamic spectral mask Frequency-by-frequency multiplication yields the high-purity, small-feature spectrum after background removal. ( ): This invention adaptively removes all frequency responses excited by broadband mechanical vibrations of the road surface (such as local modal resonances excited by a thin-walled shell at a specific vehicle speed). At this point, The remaining spectral energy in the battery can only originate from the weak electrochemical decay inside the battery and the spontaneous deformation of the gas-liquid mixed microbubbles generated during gas production at extremely low frequencies or specific frequency bands.

[0051] S2.3: Output cascade initial features: Cross-modal central processing unit in high-purity micro-feature spectrum In the process, extract the frequency point corresponding to the absolute maximum amplitude: This frequency point parameter is defined as the deformation characteristic frequency. , ( It should be noted that although the global expansion of hard carbon sodium storage is a slow deformation process, the early localized trace gas production and solid-liquid phase transition will generate discrete stress release pulses similar to acoustic emission at a microscale, thus exhibiting unique broadband micro-resonance peaks in the frequency domain. This refers to the concentrated response frequency that represents this type of intrinsic physical change.

[0052] Once extracted, it will be used as phase-locked information for cross-modal linkage across the entire system and forcibly sent to the Health Management System (BMS). In subsequent S3, the modulation frequency of the active excitation electrical pulse generated by the BMS will be based on this... For constraints. Because pass By avoiding the dominant frequency band of environmental vibration, the electromechanical resonance signal subsequently injected by the BMS is ensured to enter the blank area of ​​the vehicle road vibration spectrum.

[0053] This embodiment provides implementation details of step S3 (orthogonal excitation based on environmental spectrum) in the method for identifying early signs of thermal runaway in sodium-ion batteries, specifically disclosing steps S3.1 and S3.2.

[0054] In current applications involving strong mechanical vibrations (such as bumpy roads), the system extracts the characteristic deformation frequency through step S2. In this embodiment, step S3 orthogonally avoids the detection frequency from environmental noise, introduces a phase-locked amplification mechanism across the modal dimension, and utilizes strict clock alignment to use the excitation of a specific frequency as the sole reference for visual demodulation, thereby suppressing non-co-frequency environmental interference to the maximum extent. The specific steps are as follows: S3.1: Frequency Domain Orthogonal Avoidance and Custom Excitation The cross-modal central processing unit simultaneously receives the environmental common-mode vibration spectrum output in step S1. and the characteristic deformation frequency purified in step S2 .

[0055] Although The characteristic peak value is the purified peak value, but in order to ensure that the subsequent actively excited electrical signal avoids interference from the current global vibration frequency band of the road surface to the greatest extent, the system uses... With the center frequency as the center frequency, within the preset effective resonant frequency offset range Inside, searching The minimum energy amplitude is the frequency point corresponding to the weakest frequency domain gap in environmental vibration. This frequency point is established as the optimal orthogonal excitation frequency. ( ): It should be noted that sodium-ion batteries typically exhibit low quality factor and wideband characteristics due to damping changes and resonance caused by gas production and gas-liquid mixing. Therefore, as long as the bandwidth parameter... Set within the half-power bandwidth of this intrinsic response, even if the excitation frequency... Slightly deviated from the peak The system can still generate a sufficiently strong effective electromechanical response, while achieving absolute orthogonal isolation from environmental disturbances.

[0056] Sure Subsequently, the adaptive active microcurrent excitation circuit in the cross-modal central processing unit's health management system (BMS) continuously injects a sequence of sinusoidal alternating current pulses into the target cell. The current time variable for the pulse injection is defined as... ( The reference excitation current signal output by the BMS. (unit This can be expressed as: in, For constant microcurrent amplitude parameters (unit: ).

[0057] As the alternating current pulse sequence flows through the battery, it is converted into a signal with the base frequency through a linear motor coupling mechanism. Electromechanical resonance stress of the same frequency is applied to the battery casing.

[0058] Sodium ion insertion and extraction at hard carbon anodes are accompanied by a significant partial molar volume effect. When the implantation frequency is... sinusoidal microcurrent According to Faraday's law, the interfacial sodium ion flux and the instantaneous current are synchronous, and through the partial molar volume effect, this is linearly converted into a synchronous micro-strain on the shell surface. The visual tag is rigidly attached via thermoforming, and its displacement is equivalent to the shell deformation; therefore, this strain is mapped to a change in the tag texture. Although this micro-strain is small in scale, it is significant within the entire phase-locked window period. Internal coherent integral, processing gain approximately By combining sub-pixel DIC and spectral mask to orthogonally avoid environmental vibrations, it can be robustly extracted.

[0059] S3.2: Local high-frequency acquisition and cross-modal phase-locked loop: System setting phase-locked integration time window To avoid self-aliasing errors, mandatory constraints are applied. It is a positive integer multiple of the excitation period, that is, it satisfies ,in It is a positive integer.

[0060] Within this time window, the multi-mode vision acquisition module, based on the hardware-level time protocol (PTP) clock synchronization module, ensures that the timestamp of the vision acquisition sequence matches the timestamp of the BMS current output. Maintain alignment. The vision system remains in a localized, high-frequency focusing mode, continuously tracking the target. Two-dimensional micro-deformation components are extracted, and their covariance matrix is ​​calculated. The eigenvector corresponding to the largest eigenvalue is extracted as the principal vibration direction. The two-dimensional micro-deformation is projected onto this principal direction to obtain a signed one-dimensional excited principal response displacement signal. ( ).

[0061] A cross-modal digital lock-in amplification algorithm is executed to generate two dimensionless orthogonal reference signals with the same frequency as the excitation current: In-phase reference signal: Orthogonal reference signal: Calculate the displacement signal of the stimulated principal response within the time window Integral of in-phase components Integral with orthogonal components : because If the integer period condition is met, the product term of its own frequency can be extracted, thus generating... The high-frequency harmonic terms are canceled out. More importantly, for deviations... Any globally deterministic environmental vibration frequency (like (the suspension sway), its spectral leakage residual in the orthogonal integral operation will be approximately... The proportion decreased rapidly. By ensuring The system can suppress extremely severe road vibration interference to a very low level, thereby extracting the intrinsic electromechanical response in an environment with extremely poor signal-to-noise ratio.

[0062] The cross-modal central processing unit ultimately uses in-phase and quadrature components to calculate the electro-resonance displacement amplitude after excluding environmental interference. ( ) and the initial phase angle of the mechanical response ( ): The phase angle The degree of hysteresis directly reflects the damping abrupt change caused by gas-liquid mixing inside the battery, and is a core diagnostic parameter for determining the early stage of thermal runaway.

[0063] In complex automotive applications, the system not only faces interference from road surface mechanical vibrations but also inevitably operates within a drastically fluctuating ambient temperature field. At low temperatures, the solid-phase ion diffusion and charge transfer impedance within a healthy sodium-ion battery significantly increase, slowing down the normal electrochemical polarization establishment process. In this situation, the mechanical stress generated by polarization itself produces a normal and significant intrinsic electrochemical hysteresis. If the initial phase angle of the mechanical response output in step S3.2 is directly used… As the final discrimination parameter, the system may produce false positives (false alarms) in temperature because it cannot distinguish between normal electrochemical hysteresis caused by low temperature and gas-liquid mixed-phase damping hysteresis caused by thermal runaway gas production.

[0064] S3.3: Cross-modal hysteresis calculation The cross-modal central processing unit injects a reference excitation current signal into the instruction BMS. The same phase-locked integration time window Internally, the dynamic terminal voltage response signal at both ends of the target cell collected by the BMS is acquired synchronously. (unit ).

[0065] To ensure the absolute reference uniformity of the cross-modal phase system, the system reuses the dimensionless orthogonal reference signal with the same frequency as the excitation current generated in step S3.2: In-phase reference signal: Orthogonal reference signal: Dynamic terminal voltage response signal A digital phase-locked loop (PLL) integral operation, completely isomorphic to that of the vision unit, is performed. This operation extracts the in-phase response while suppressing low-frequency or DC voltage fluctuation interference caused by high-rate charging and discharging in actual operating conditions. The in-phase voltage component is then calculated. With orthogonal voltage components : Furthermore, the specific injection frequency was calculated. The polarization phase angle below ( ): The phase angle This mainly represents the baseline of phase deflection caused by electrochemical kinetics inside the battery at the current temperature and charge state.

[0066] Assuming that dynamic electromechanical strain is primarily driven by local electrochemical polarization overpotentials, then the phase of the driving force for the mechanical response is equivalent to the electrochemical polarization phase angle. Therefore, the mechanical phase extracted visually. In reality, it is a linear superposition of polarization phase, structural mechanical damping delay, and fixed deviation of system hardware links.

[0067] To extract pure structural delay features, the system introduces a calibrated cross-modal synchronization delay constant. This represents the circuit delay between the BMS current output and the exposure trigger of the miniature camera, with dimensions of . Calculate its fixed hardware phase offset at the current excitation frequency. ( ): The cross-modal central processing unit performs differential calculations on the three phase parameters mentioned above. This ensures that the final phase is within the four-quadrant domain. It has a unique and definite physical meaning, avoiding quadrant drift and undefined values ​​caused by large-angle subtraction, and is forcibly included. The phase winding reduction algorithm ultimately outputs the cascaded evaluation parameter—the cross-modal phase hysteresis angle. ( ): In low-temperature environments, healthy sodium-ion batteries may experience a slowdown in electrochemical reactions, leading to... Significantly increased, Increase accordingly; by subtracting The system adaptively offsets this extremely significant electrochemical temperature drift, greatly narrowing the false alarm range of the system over a wide temperature range.

[0068] When early abnormal micro-gas production occurs inside the target single-cell battery (such as electrolyte side reaction vaporization), the originally dense solid-liquid multiphase interface evolves into a solid-liquid-gas system containing microbubbles. Due to the extremely high compressibility of the gas, the local bulk modulus in this region drops sharply, and the physical viscous damping increases nonlinearly and rapidly. This purely structural degradation causes the mechanical stress wave (effect) to lag significantly behind the electrical polarization driving force (cause), leading to... Occurrence and A highly disproportionate unidirectional surge. At this point, the calculated... This will exhibit an abnormal jump far exceeding the normal mechanical temperature drift tolerance range. Through the calculation in step S3.3, this application utilizes the linear superposition and phase reduction rules to internalize the traditionally isolated visual deformation and electrical polarization into a subtractive isomorphic closed-loop physical quantity, making... It becomes an intrinsic state indicator that eliminates background noise.

[0069] Step S4: Final discrimination and artifact space correction based on phase hysteresis angle.

[0070] In step S3.3, the system outputs the photoelectric cross-mode phase hysteresis angle to decouple electrochemical polarization interference through cross-mode subtraction. In actual vehicle service life of several years, the system still needs to overcome the viscosity coefficient of the electrolyte inside the sodium-ion battery, the mechanical flexibility of the separator, and the Young's modulus of the shell material. It will itself undergo slow nonlinear drift with the decline of the current ambient absolute temperature and the battery health status.

[0071] If a fixed absolute phase threshold is used for a one-size-fits-all alarm judgment, the normal increase in mechanical damping due to battery aging or extreme cold weather can easily exceed the fixed threshold, causing false alarms in the system. Conversely, if the threshold is raised simply to avoid false alarms, early trace gas production may be missed.

[0072] Synchronously read the ambient absolute temperature of the current target cell from the BMS. and current health status scalar .

[0073] Because the viscosity damping of the electrolyte and polymer materials inside the battery exhibits an Arrhenius exponential relationship with temperature, and because long-term aging leads to internal structural fatigue and The cross-modal central processing unit exhibits a strong correlation and constructs the following nonlinear dynamic baseline model to calculate the intrinsic hysteresis alarm threshold in real time under the current state. ( ): in, To calibrate this model of battery at the reference temperature and in brand new condition ( Normal baseline hysteresis constant ( ) ); To calibrate the reference absolute temperature constant; The activation energy parameter (J / mol) for viscous flow of the material was calibrated through a temperature chamber experiment. The ideal gas constant (J / (mol)) K)); The dimensionless aging coupling gain coefficient; The upper limit of the hysteresis of the forced vibration system is preferably defined as follows: .

[0074] Due to the output It has been entangled and bound to Within the domain, if the Arrhenius exponent term is drastically amplified beyond [a certain value] under extremely cold conditions... This would result in the system never being able to meet the alarm conditions. Therefore, an upper limit based on extreme values ​​is introduced. This ensures that the bounded true phase features can be compared with a reasonable threshold, preventing underreporting vulnerabilities at low temperatures.

[0075] The cross-modal central processing unit will output the signed cross-modal phase hysteresis angle. The dynamic intrinsic hysteresis alarm threshold generated in step S4.1 Compare the values ​​and combine them with the resonant displacement amplitude output from the middle. Perform the following physical classification diagnosis within the full parameter space: True positive result diagnosis, early diagnosis of uncontrolled gas production and heat: If the conditions are met: When trace amounts of gas are generated inside the battery due to early-stage thermal runaway side reactions, the dense solid-liquid multiphase interface evolves into a highly compressible gas-liquid-solid three-phase mixture. At this point, viscous damping undergoes an exponential abrupt change. This pure structural degradation causes the mechanical response (effect) to lag significantly behind the polarization driving force (cause), exhibiting a large positive phase hysteresis far exceeding the normal temperature drift and aging tolerance range. Once this condition is met, the system immediately outputs an early-stage thermal runaway diagnosis alarm command and triggers the BMS safety intervention strategy.

[0076] False positive detection, automotive-grade mechanical artifact judgment: If the following conditions are met: and .

[0077] in, For a phase shift threshold close to zero (e.g.) ), As a preset abnormal amplitude threshold, when a visual sticker peeling or external fasteners loosening are selected as target points, although they are subjected to the same frequency as the electrical pulse via step S3.1 ( Driven by weak electromechanical stress, but because it is detached from the damping drag of the viscous liquid phase fluid inside the battery casing, it exhibits undamped dry friction or elastic tapping. Therefore, in Phase-locked amplitude at the location The size of the object is abnormally large due to the lack of constraints, but its mechanical response phase is extremely close to the driving force phase (i.e., it exhibits near-zero hysteresis), which leads the system to diagnose it as a mechanical component artifact.

[0078] Normal decay maintenance: If the conditions are met: Or the amplitude is so small that it does not have an effective signal-to-noise ratio. The system determines that the current response is within the normal temperature drift or aging tolerance range, with no risk of thermal runaway, and does not generate an alarm. It only records the current data into the system to maintain normal cyclic monitoring in subsequent cycles.

[0079] To avoid the system repeatedly wasting limited local high-frequency computing power on confirmed mechanical artifacts during the vehicle's subsequent thousands of kilometers of driving, this step constructs a spatial shielding feedback mechanism.

[0080] When the target coordinates are confirmed After mechanical artifacts are detected, the cross-modal central processing unit will trigger the two-dimensional spatial mask matrix of the global computation layer in step S1. Dynamic updates.

[0081] Initially, all elements of the mask matrix across the entire field of view are... When the update is triggered, at coordinates Centered on the mask elements within the connected region identified as artifacts, the mask elements are corrected to... (: After the update is complete, the mask matrix is ​​fed back to the beginning of step S1. In subsequent global real-time monitoring loops, the residual displacement field vector extracted in step S1 is used... Hadamard mask filtration will be enforced: The spatial region marked as an artifact has a corresponding mask value. , It will always output a zero vector, thus eliminating the triggering of strain gradient anomalies (generating new suspected lesions). (possibly).

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0084] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying early signs of thermal runaway in sodium-ion batteries, characterized in that, The method includes: Two-dimensional image sequences of the battery surface are acquired, the global motion compensation reference and the small residual displacement field are separated, and the coordinates of suspected abnormal local areas and the common mode vibration spectrum of the environment are output. At suspected abnormal local coordinates, background noise is eliminated by combining the environmental common-mode vibration spectrum, and characteristic deformation frequencies are extracted; Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, a micro-current pulse sequence is injected into the battery, the excited principal response displacement signal and the dynamic terminal voltage response signal are extracted simultaneously, and the transmodal phase hysteresis angle is calculated. Intrinsic hysteresis alarm thresholds are generated based on ambient absolute temperature and health status scalars. The cross-modal phase hysteresis angle is compared with the intrinsic hysteresis alarm thresholds to diagnose early thermal runaway or identify mechanical artifacts.

2. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 1, characterized in that, The process of separating the global motion compensation benchmark from the micro residual displacement field includes: locking the static benchmark anchor point on the external module frame in the two-dimensional image sequence, performing feature extraction and trajectory tracking; establishing a two-dimensional affine transformation model that includes rotation, translation, scaling and shear deformation, solving the global affine transformation matrix and translation vector in real time to form a global motion compensation benchmark characterizing global mechanical vibration; tracking the actual observed coordinates of the visual label on the battery surface, subtracting them from the theoretical compensation coordinates calculated after substituting them into the global motion compensation benchmark, and obtaining the micro residual displacement field stripped of the influence of global mechanical vibration.

3. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 2, characterized in that, To achieve vibration measurement and data processing, the system outputs suspected abnormal local coordinates and environmental common-mode vibration spectra, including: obtaining the Cauchy strain tensor by spatially differentiating the small residual displacement field; calculating the spatial gradient magnitude of the Cauchy strain tensor; extracting the centroid coordinates of connected regions with spatial gradient magnitudes greater than a preset calibration threshold as suspected abnormal local coordinates; selecting a sliding time window to extract the signed components of the translation vector containing alternating oscillation characteristics; performing fast Fourier transforms after mean reduction processing to synthesize the environmental common-mode vibration spectrum in two-dimensional space.

4. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 1, characterized in that, Extracting characteristic deformation frequencies includes: acquiring two-dimensional micro-deformation components at suspected abnormal local coordinates in a local high-frequency focusing mode and calculating their high-frequency two-dimensional spatial spectrum; constructing an adaptive spectrum suppression mask based on energy normalization based on the maximum value of the environmental common-mode vibration spectrum; multiplying the high-frequency two-dimensional spatial spectrum with the adaptive spectrum suppression mask frequency by frequency and performing dynamic filtering to eliminate the frequency response excited by broadband mechanical vibration of the road surface, thus obtaining the micro-characteristic spectrum; extracting the frequency point corresponding to the absolute maximum amplitude in the micro-characteristic spectrum and locking it as the characteristic deformation frequency.

5. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 1, characterized in that, Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, and a micro-current pulse sequence is injected into the battery. This includes: taking the characteristic deformation frequency as the center frequency, finding the frequency point corresponding to the minimum value of the energy amplitude of the environmental common-mode vibration spectrum within the preset effective resonant frequency offset range, and establishing it as the optimal orthogonal excitation frequency; continuously injecting a sinusoidal AC reference excitation current signal containing the optimal orthogonal excitation frequency into the target single cell, which is converted into electromechanical resonant stress with the same frequency as the optimal orthogonal excitation frequency.

6. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 5, characterized in that, Using digital signal processing algorithms, the stimulated principal response displacement signal and dynamic terminal voltage response signal are extracted synchronously to calculate the cross-modal phase hysteresis angle. This includes: aligning the timestamps of the visual acquisition sequence with the timestamps of the current output; projecting the extracted micro-deformation components onto the principal vibration direction to obtain a signed one-dimensional stimulated principal response displacement signal, and simultaneously acquiring the dynamic terminal voltage response signal; generating orthogonal and in-phase reference signals with the same frequency as the reference excitation current signal; performing cross-modal digital phase-locked amplification operations on the one-dimensional stimulated principal response displacement signal and the dynamic terminal voltage response signal respectively within a set integer-cycle phase-locked integration time window to extract the electro-resonance displacement amplitude, the initial phase angle of the mechanical response, and the electrical polarization phase angle; introducing a pre-calibrated hardware-level cross-modal synchronization delay constant to calculate the fixed hardware phase offset; performing differential calculations on the initial phase angle of the mechanical response, the electrical polarization phase angle, and the fixed hardware phase offset; and outputting the cross-modal phase hysteresis angle after adding a phase winding reduction algorithm.

7. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 1, characterized in that, To achieve fault prediction and health management of battery systems, an eigenhysteresis alarm threshold is generated based on ambient absolute temperature and health status scalars. This includes: synchronously reading the ambient absolute temperature and current health status scalars of individual target cells; calculating a nonlinear dynamic baseline by combining pre-calibrated baseline hysteresis angle constant, material viscous flow activation energy parameters, ideal gas constant, reference absolute temperature, and aging coupling gain coefficient; and introducing the upper limit of the hysteresis of the forced vibration system as a physical extreme constraint to prevent missed alarms. The nonlinear dynamic baseline and the upper limit of the hysteresis of the forced vibration system are compared with a minimum amplitude limit to generate the eigenhysteresis alarm threshold.

8. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 6, characterized in that, The comparison is used to confirm early thermal runaway or identify mechanical artifacts, including hazard prediction and pattern recognition logic: if the cross-modal phase hysteresis angle is greater than or equal to the intrinsic hysteresis alarm threshold, it is determined to be an early gas-generating thermal runaway diagnosis, and a true positive alarm command is output; if the absolute value of the cross-modal phase hysteresis angle is less than or equal to a phase shift threshold close to zero, and the electro-resonance displacement amplitude is greater than or equal to a preset abnormal amplitude threshold, it manifests as undamped dry friction or elastic slapping, and is diagnosed as a vehicle mechanical component artifact.

9. The method for identifying early signs of thermal runaway in sodium-ion batteries according to claim 8, characterized in that, As an edge computing power optimization mechanism for intelligent monitoring devices, after identifying mechanical artifacts, a closed-loop mask feedback step to prevent false alarms is also included: when a vehicle mechanical component artifact is confirmed at the target point coordinates, the two-dimensional spatial mask matrix used for the underlying global calculation is reconstructed, and the mask elements within the connected domain of the artifact are corrected to zero; the updated two-dimensional spatial mask matrix is ​​fed back to the global real-time monitoring loop, and the Hadamard product mask filtering operation of element-wise scalar multiplication is forcibly performed on the subsequently extracted small residual displacement field to remove the possibility of triggering anomalies in the spatial region strain gradient.

10. A health management system, characterized in that, The system includes: Environmental common-mode vibration spectrum extraction module: acquires two-dimensional image sequences of the battery surface, separates the global motion compensation reference from the small residual displacement field, and outputs the coordinates of suspected abnormal local areas and the environmental common-mode vibration spectrum; Deformation frequency extraction module: At suspected abnormal local coordinates, background noise is eliminated by combining the environmental common-mode vibration spectrum, and characteristic deformation frequencies are extracted; Phase calculation module: Based on the characteristic deformation frequency and the environmental common-mode vibration spectrum, the orthogonal excitation frequency is determined, a micro-current pulse sequence is injected into the battery, the excited main response displacement signal and the dynamic terminal voltage response signal are extracted simultaneously, and the cross-modal phase hysteresis angle is calculated. Judgment module: Based on the ambient absolute temperature and health status scalar, the intrinsic hysteresis alarm threshold is generated. The cross-modal phase hysteresis angle is compared with the intrinsic hysteresis alarm threshold to diagnose early thermal runaway or identify mechanical artifacts.

Images