AI-based supercapacitor structure damage detection method and system

Abstract

This invention relates to the field of damage detection technology, and in particular to an AI-based method and system for detecting structural damage in supercapacitors. The method includes: acquiring multimodal response data; constructing operating condition labels by combining operating condition parameters; generating reference response data based on the operating condition labels; comparing the multimodal response data with the reference response data to construct a multimodal residual field; and performing structural damage identification by combining the unit topology diagram. This method helps to distinguish normal operating condition fluctuations caused by temperature fluctuations, changes in state of charge, pulse load changes, and equilibrium deviations from abnormal responses caused by structural damage, reducing the impact of operating condition coupling interference on the identification results, and making abnormal features related to structural damage more prominent. This improves the stability and accuracy of online structural damage identification, early localization, and damage type determination.

Description

Technical Field

[0001] This invention relates to the field of damage detection, and more particularly to an AI-based method and system for detecting structural damage in supercapacitors. Background Technology

[0002] In the application scenario of regenerative braking high-frequency pulse in vehicle-mounted hybrid energy storage supercapacitor modules, the actual operating environment is generally characterized by the coupled effects of multiple factors such as temperature fluctuations, changes in state of charge, equilibrium deviations, and frequent pulse load disturbances. This causes the external manifestations of internal structural damage to intertwine with fluctuations under normal operating conditions. Especially under the premise of not disassembling the module, not relying on laboratory-level in-situ characterization equipment, and not significantly increasing the operating burden of the entire vehicle, how to stably identify the actual structural damage characteristics caused by electrode delamination, loose current collector connections, local deformation of the diaphragm, and local instability of the electrolyte from the complex operating response has become a key problem that urgently needs to be solved for engineering applications in this scenario.

[0003] Currently, existing technologies typically perform performance diagnostics on supercapacitors by collecting operating parameters such as voltage, current, temperature, equivalent series resistance, capacitance, state of charge, and health state estimation results, combined with a state estimation model. This type of method is relatively easy to deploy and readily applicable to engineering scenarios, thus becoming a common online analysis tool for vehicle-mounted operation.

[0004] However, in the regenerative braking high-frequency pulse scenario of vehicle-mounted hybrid energy storage supercapacitor modules, most existing online diagnostic methods still rely on electrical or thermal characterization quantities such as equivalent series resistance, capacitance, temperature rise, and terminal voltage recovery curve. These characterization quantities are simultaneously affected by factors such as temperature, state of charge, equilibrium deviation, pulse amplitude, and sampling time. This makes it easy for normal fluctuations caused by changes in operating conditions and abnormal responses caused by structural damage to be mixed together. It is difficult to accurately extract the characteristic changes caused by real structural damage such as electrode delamination, loose connection, and local bulging in the early stage. Therefore, it can only make a rough judgment on the decline in health status and it is difficult to achieve online localization and early identification of structural damage. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing supercapacitor structural damage diagnosis technologies. On the one hand, these technologies primarily rely on macroscopic electrothermal parameters such as capacitance, temperature rise, and terminal voltage recovery curves for judgment, which are easily affected by temperature, state of charge, equilibrium deviation, and pulsed operating conditions, making it difficult to accurately distinguish between normal operating condition fluctuations and actual structural damage. On the other hand, detection methods that can more intuitively reflect structural evolution mostly rely on in-situ laboratory characterization or offline disassembly and analysis, which are difficult to adapt to the application requirements of online identification, early positioning, and engineering deployment in vehicle scenarios. Therefore, this invention proposes an AI-based supercapacitor structural damage detection method and system.

[0006] Firstly, the technical solution adopted by this invention is: an AI-based method for detecting structural damage in supercapacitors, comprising:

[0007] Acquire multimodal response data;

[0008] Based on the multimodal response data, the operating condition parameters are read, and operating condition labels are constructed according to the operating condition parameters. Reference response data is then generated based on the operating condition labels.

[0009] A multimodal residual field is constructed based on the multimodal response data and the reference response data;

[0010] Obtain the connection relationship data and thermal coupling relationship data between the units in the supercapacitor module, and construct the unit topology diagram based on the connection relationship data and the thermal coupling relationship data;

[0011] Structural damage identification is performed based on the multimodal residual field and the unit topology map to obtain the damage location and damage type;

[0012] The damage stage is determined based on the damage location and the damage type;

[0013] Based on the damage location, the damage type, and the damage stage, the safety pulse boundary parameters are determined.

[0014] Preferably, the specific steps for constructing the working condition label based on the working condition parameters include:

[0015] Based on the multimodal response data, energy storage level parameters, external temperature environment parameters, pulse intensity parameters, pulse duration parameters, and electrical connection initial consistency parameters are extracted.

[0016] The energy storage level parameter, the external temperature environment parameter, the pulse intensity parameter, the pulse duration parameter, and the electrical connection initial consistency parameter are written into the corresponding operating condition fields, and operating condition tags are constructed based on each operating condition field.

[0017] Preferably, the specific steps for generating reference response data based on the operating condition label include:

[0018] Pre-construct the target sample set and the first reference response generation model;

[0019] The first reference response generation model is trained based on the target sample set to obtain the second reference response generation model;

[0020] Based on the operating condition label, reference response data is generated using the second reference response generation model.

[0021] Preferably, the specific steps for constructing the multimodal residual field based on the multimodal response data and the reference response data include:

[0022] Based on the multimodal response data and reference response data, the electrical residual sequence, thermal residual sequence, and deformation residual sequence are determined.

[0023] The electrical residual sequence, thermal residual sequence, and deformation residual sequence are merged to form a multimodal residual field.

[0024] Preferably, the specific steps for constructing the unit topology graph based on the connection relationship data and the thermal coupling relationship data include:

[0025] Determine the connection edges and their connection weights based on the connection relationship data;

[0026] Determine the thermal coupling edges and their thermal coupling weights based on the aforementioned thermal coupling relationship data;

[0027] The node features corresponding to each unit are determined based on the multimodal residual field.

[0028] Based on the connecting edges and their connecting weights, thermally coupled edges and their thermally coupled weights, and node features, a unit topology graph is constructed.

[0029] Preferably, the specific steps for obtaining the damage location and damage type include:

[0030] Based on the connection edges, thermal coupling edges, and node features in the unit topology graph, propagation and aggregation are performed to obtain the aggregation features corresponding to each unit.

[0031] Based on the aforementioned polymerization characteristics, the electrical consistency determination results, mechanical consistency determination results, and thermal consistency determination results are determined respectively;

[0032] Based on the electrical consistency determination results, the mechanical consistency determination results, and the thermal consistency determination results, a joint determination is made to identify effective structural damage units;

[0033] The location and type of damage are determined based on the effective structural damage units.

[0034] Preferably, the specific steps for determining the damage stage based on the damage location and the damage type include:

[0035] Damage intensity values ​​are constructed based on the multimodal residual field;

[0036] The target unit is located based on the damage location, and the damage intensity value corresponding to the target unit is read.

[0037] Pre-define and establish stage threshold groups corresponding to different damage types, and match the corresponding stage threshold groups according to the damage type;

[0038] The damage stage corresponding to the target unit is determined by comparing the damage intensity value with the stage threshold group.

[0039] Preferably, the specific steps for constructing the damage intensity value based on the multimodal residual field include:

[0040] Based on the multimodal residual field, electrical residual features, deformation residual features, and thermal residual features are extracted;

[0041] The electrical residual characteristics, deformation residual characteristics, and thermal residual characteristics are normalized to obtain the corresponding normalized characteristics;

[0042] The damage intensity value is obtained by weighted summarization based on the normalized features.

[0043] Preferably, the specific steps for determining the safety pulse boundary parameters include:

[0044] Pre-establish a set of baseline operating parameters and a table of correction coefficients;

[0045] The target unit is determined based on the location of the damage, and the location category of the target unit is determined;

[0046] Based on the operating condition label, the baseline operating parameters are read from the baseline operating parameter set, and the corresponding correction coefficients are read from the correction coefficient table based on the damage stage, the damage type, and the location category.

[0047] Based on the baseline operating parameters and the corresponding correction coefficients, the safety pulse boundary parameters are determined;

[0048] The operation control is performed and the safety pulse boundary parameters are updated based on the aforementioned safety pulse boundary parameters.

[0049] Secondly, the technical solution adopted by this invention is: an AI-based supercapacitor structural damage detection system, comprising:

[0050] The acquisition module is used to acquire multimodal response data;

[0051] The feature construction module is used to read operating parameters and construct operating labels based on the multimodal response data, generate reference response data based on the operating labels, construct a multimodal residual field based on the multimodal response data and the reference response data, and obtain the connection relationship data and thermal coupling relationship data between each unit in the supercapacitor module to construct a unit topology diagram.

[0052] The damage determination module is used to perform structural damage identification based on the multimodal residual field and the unit topology map, obtain the damage location and damage type, and determine the damage stage based on the damage location and the damage type.

[0053] The boundary determination module is used to determine the safety pulse boundary parameters based on the damage location, the damage type, and the damage stage.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] I. This invention generates reference response data based on operating condition labels, and then compares the measured response with the reference response data. This separates normal fluctuations caused by temperature changes, state of charge changes, and pulse intensity changes from abnormal deviations caused by structural damage. This helps to reduce the coupling interference of temperature fluctuations, state of charge changes, and pulse load changes on the response signal, making the abnormal features related to structural damage more prominent, thereby improving the stability and accuracy of online structural damage identification.

[0056] Second, this invention first performs residual processing on the original response based on reference response data, and then extracts damage features and constructs damage intensity values ​​based on the multimodal residual field. This is beneficial for retaining the abnormal distribution information of each unit, each time window, and each mode, while further compressing the dispersed abnormal features into quantifiable and comparable damage degree indicators. This reduces the interference of common components of the working condition on the subsequent identification results, further improves the prominence of structural damage-related abnormal features, and provides a unified quantitative basis for subsequent damage stage determination and safety pulse boundary parameter determination.

[0057] Third, this invention establishes connection edges based on the busbar connection table, establishes thermal coupling edges based on the temperature rise transfer ratio, and combines the electrical residual data, thermal residual data and deformation residual data corresponding to each unit under the current pulse window as node features and writes them into the unit topology graph. This is beneficial for simultaneously characterizing unit body anomalies, electrical connection propagation anomalies and thermal coupling propagation anomalies, thereby distinguishing between the body damage of a single unit and the coupling anomalies formed by the propagation of neighboring units, reducing the situation of judging the propagation response as body damage, and improving the accuracy of subsequent structural damage location and damage type identification.

[0058] Fourth, this invention first establishes a physical constraint diagram recognition model, then performs propagation calculations on time features based on connecting edges and thermal coupling edges, and determines effective structural damage units based on electrical consistency judgment results, mechanical consistency judgment results, and thermal consistency judgment results. It then outputs the target unit number, target abnormal area, and corresponding damage type, which helps to distinguish propagation anomalies from body anomalies and improves the accuracy of structural damage location and damage type identification.

[0059] Fifth, this invention first establishes a set of benchmark operating parameters and a correction coefficient table corresponding to ambient temperature, state of charge, damage stage, damage type, and target unit location category. Then, based on the damage location, damage type, and damage stage, it determines the allowable peak current, allowable pulse width, and allowable number of consecutive pulses. This facilitates the direct conversion of the structural damage results obtained from the preceding online identification into control boundaries that can be executed by the vehicle system, avoiding the limitation to laboratory characterization or offline analysis. This improves the practical application value and engineering deployment feasibility of early structural damage localization results in vehicle scenarios. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method of the present invention.

[0061] Figure 2 This is a comparison diagram of the effects of the present invention and the prior art. Detailed Implementation

[0062] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0063] Example 1:

[0064] To achieve the above objectives, please refer to Figure 1 This invention provides an AI-based method for detecting structural damage in supercapacitors, including:

[0065] Acquire multimodal response data;

[0066] Operating parameters are read based on multimodal response data, operating condition labels are constructed based on the operating parameters, and reference response data is generated based on the operating condition labels.

[0067] A multimodal residual field is constructed based on multimodal response data and reference response data;

[0068] Obtain the connection relationship data and thermal coupling relationship data between the units in the supercapacitor module, and construct the unit topology diagram based on the connection relationship data and thermal coupling relationship data;

[0069] Structural damage identification is performed based on multimodal residual fields and element topology diagrams to obtain damage location and damage type;

[0070] Determine the damage stage based on the location and type of damage;

[0071] Based on the damage location, damage type, and damage stage, the safety pulse boundary parameters are determined.

[0072] As an optional implementation method, the specific steps for acquiring multimodal response data include:

[0073] During the operation of the supercapacitor module, voltage acquisition modules, current acquisition modules, temperature acquisition modules, and micro-deformation acquisition modules are respectively deployed in the housing of each unit, the busbar connection area, and the surface of the module to obtain multi-modal response information of the supercapacitor module during regenerative braking; each unit housing refers to the external encapsulation housing of each unit in the supercapacitor module.

[0074] Specifically, the voltage acquisition module can use an isolated voltage sampling circuit to collect terminal voltage change data of each unit; the current acquisition module can use a Hall current sensor to collect module charging current change data; the temperature acquisition module can use a thermocouple or thermistor to collect temperature change data of each unit housing and busbar connection area; the micro-deformation acquisition module can use a resistance strain gauge, fiber optic grating sensor or capacitive displacement sensor, and the micro-deformation acquisition module is deployed on the surface of the unit housing or busbar connection area to collect local bulging, strain or displacement changes to form deformation response data.

[0075] Among them, the regenerative braking pulse refers to the short-term, rapidly changing charging current load applied to the supercapacitor module during the regenerative braking energy recovery process of the vehicle. Because this charging current load rises and falls rapidly within a short period, the sampling frequency for voltage and current is preset to 5 kHz to fully preserve the details of the transient voltage drop and rapid current change process; since the thermal diffusion process is relatively slow, 50 Hz is sufficient to cover local temperature rise changes, therefore the sampling frequency for temperature is preset to 50 Hz; since the micro-deformation response speed is between the electrical and thermal responses, 500 Hz can accommodate the resolution requirements of shell bulging changes and sudden strain changes in the connection area, therefore the sampling frequency for micro-deformation is preset to 500 Hz.

[0076] After completing the data collection for each channel, the data for each channel is timestamped according to a unified time base.

[0077] Specifically, data from each channel is resampled using a uniform time step of 1 millisecond. For low-sampling-frequency channels, linear interpolation is used to fill in the time nodes, while for high-sampling-frequency channels, window averaging is used to map them to a uniform time axis. Then, in the aligned module charging current sequence, sudden increases in charging current are detected. Specifically, when the module charging current increases for three consecutive sampling points, and the current value corresponding to the third sampling point reaches or exceeds 3% of the rated peak charging current of the supercapacitor module, the corresponding first increasing sampling point is determined as the start time of each regenerative braking pulse. The response window is continuously truncated using the start time as the time base, with 50 milliseconds before the start time serving as the pulse pre-pulse threshold. The stabilization phase consists of a 200-millisecond truncation after the initial pulse loading phase and a 500-millisecond truncation after the pulse loading phase. The pre-pulse stabilization phase characterizes the stable baseline state before the pulse, the pulse loading phase characterizes the main charging process, and the pulse recovery phase characterizes the voltage recovery, thermal diffusion, and local deformation fall-off process after the pulse ends. After window truncation, the aligned channel data are aggregated according to the cell number to obtain multimodal response data corresponding to the same pulse event. The multimodal response data includes electrical response data, thermal response data, and deformation response data, which are used to characterize the electrical response, thermal response, and deformation response of the same cell within the same pulse window, respectively.

[0078] If the number of naturally occurring regenerative braking pulses is less than 20 within a continuous 10-minute operating cycle, the current effective pulse sample is deemed insufficient. At this time, within a low-load window where the vehicle's traction power is below 8% of the rated traction power and the module current fluctuation amplitude is below 5% of the rated peak current, the energy management system controls the bidirectional converter to apply diagnostic pulses to the supercapacitor module to supplement the effective pulse sample. To elicit a identifiable transient electrical response without significantly altering the vehicle's operating state, the peak current of the diagnostic pulse is set to 10% to 15% of the supercapacitor module's rated peak charging current. To cover transient voltage drop changes in the connection area and the initial response of local deformation, while avoiding excessive heat accumulation, the pulse width of the diagnostic pulse is set to 20 to 40 milliseconds, thus ensuring that additional disturbances remain within a controllable range.

[0079] This step obtains time-aligned multimodal response data in the manner described above, which helps to limit subsequent comparison objects to the same operating condition window and reduce artifacts caused by different sampling cycles.

[0080] As an optional implementation method, the specific steps of reading operating parameters based on multimodal response data, constructing operating condition labels based on the operating parameters, and generating reference response data based on the operating condition labels include:

[0081] Using the response window corresponding to the multimodal response data as a reference, the module terminal voltage at the end of the pre-pulse stabilization phase is read from the electrical response data, and the module terminal voltage is used as the energy storage level parameter; the average value of the module surface temperature sampling value within the pre-pulse stabilization phase is read from the thermal response data, and the average value is used as the external temperature environment parameter; the pulse peak current is extracted from the electrical response data, and the pulse peak current is used as the pulse intensity parameter; the pulse width is extracted from the electrical response data, and the pulse width is used as the pulse duration parameter; the resting voltage of each unit is read within the pre-pulse stabilization phase, the average value of the resting voltage of each unit is calculated, and the initial equilibrium state of the unit is determined based on the deviation of the resting voltage of each unit from the average value, and the initial equilibrium state of the unit is used as the initial consistency parameter of the electrical connection;

[0082] Specifically, let the resting voltage of the i-th unit during the pre-pulse stabilization period be... The average static voltage of each unit is If the total number of units is n, then the initial equilibrium state of the units is calculated according to the following formula: .

[0083] B is used to characterize the degree of voltage consistency among the units before the pulse is applied. The larger B is, the worse the initial equilibrium state of the units is.

[0084] Subsequently, the energy storage level parameter is written into the module terminal voltage field, the external temperature environment parameter is written into the module surface temperature field, the initial electrical connection consistency parameter is written into the cell initial equalization state field, the pulse intensity parameter is written into the pulse peak current field, and the pulse duration parameter is written into the pulse width field, thereby constructing a working condition label corresponding to the multimodal response data. In this embodiment, the working condition label consists of 5 fields: module terminal voltage field, module surface temperature field, cell initial equalization state field, pulse peak current field, and pulse width field. The module terminal voltage field is used to characterize the energy storage level, the module surface temperature field is used to characterize the external temperature environment, the cell initial equalization state field is used to characterize the initial electrical connection consistency, the pulse peak current field is used to characterize the current pulse intensity, and the pulse width field is used to characterize the current pulse duration.

[0085] As an optional implementation method, the specific steps for generating reference response data based on the operating condition label include:

[0086] Multimodal response data were collected in advance during the early stable operation test phase of the same type of supercapacitor module, and multiple pulse response windows were extracted in units of regenerative braking pulses as candidate samples. The candidate samples were then screened according to preset screening conditions to obtain the target sample set. The preset screening conditions included: based on the first 20 effective pulse response windows of the same type of supercapacitor module during the early stable operation phase, the average value of the equivalent series resistance, the average value of the temperature rise slope, and the average value of the local deformation peak were calculated respectively, and these were used as the initial references for the equivalent series resistance, the temperature rise slope, and the local deformation peak, respectively. Only when the drift of the equivalent series resistance relative to the initial reference of the equivalent series resistance of 5 consecutive pulses does not exceed 6%, the deviation of the temperature rise slope relative to the initial reference of the temperature rise slope does not exceed 10%, and the deviation of the local deformation peak relative to the initial reference of the local deformation peak does not exceed 12%, the corresponding candidate sample was retained in the target sample set.

[0087] Then, a first reference response generation model is constructed. The reference response generation model includes a label encoding layer, a feature mapping layer, and a multimodal output layer. The label encoding layer is used to encode the module terminal voltage field, module surface temperature field, unit initial equalization state field, pulse peak current field, and pulse width field in the operating condition label to obtain the operating condition features. The feature mapping layer is used to establish a nonlinear correspondence between the operating condition label and the response data based on the operating condition features. The multimodal output layer is used to output the reference electrical response data, reference thermal response data, and reference deformation response data, respectively.

[0088] Subsequently, the operating condition labels in the target sample set are used as input samples, and the electrical response data, thermal response data, and deformation response data corresponding to each operating condition label are used as target output samples. The AI-based reference response generation model is then trained under supervision to obtain the second reference response generation model.

[0089] After training is completed, the current operating condition label is input into the second reference response generation model to generate reference response data. The reference response data includes reference electrical response data, reference thermal response data, and reference deformation response data that are consistent with the time window of the multimodal response data. These data are used to characterize the baseline response that a unit that meets the preset screening conditions should exhibit under the corresponding operating condition.

[0090] This step generates reference response data based on the operating condition labels, and then compares the measured response with the reference response data. This separates the normal fluctuations caused by temperature changes, state of charge changes, and pulse intensity changes from the abnormal deviations caused by structural damage. This helps to reduce the coupling interference of temperature fluctuations, state of charge changes, and pulse load changes on the response signal, making the abnormal features related to structural damage more prominent, thereby improving the stability and accuracy of online structural damage identification.

[0091] As an optional implementation, the specific steps for constructing a multimodal residual field based on multimodal response data and reference response data include:

[0092] First, the electrical response data and the reference electrical response data are subtracted point by point on the same time axis to obtain the electrical residual sequence; the thermal response data and the reference thermal response data are subtracted point by point on the same time axis to obtain the thermal residual sequence; the deformation response data and the reference deformation response data are subtracted point by point on the same time axis to obtain the deformation residual sequence.

[0093] Subsequently, the electrical residual sequence, thermal residual sequence, and deformation residual sequence are merged using the element number as the first index, the time window as the second index, and the modal category as the third index to form a multimodal residual field. The modal categories include electrical modes, thermal modes, and deformation modes. The multimodal residual field is used to characterize the degree of deviation of each element from the baseline response characterized by the reference response data under the corresponding operating conditions, and serves as input data for subsequent structural damage identification.

[0094] Furthermore, after constructing the multimodal residual field, damage feature quantities are extracted based on the multimodal residual field; among them, the relative residual of equivalent series resistance is extracted based on the electrical residual sequence, the relative residual of local strain gradient is extracted based on the deformation residual sequence, and the relative residual of thermal diffusion time constant is extracted based on the thermal residual sequence.

[0095] Specifically, the ratio of transient voltage drop to charging current in the pulse loading segment of the electrical response data is determined as the equivalent series resistance, and the difference between the measured equivalent series resistance and the reference equivalent series resistance is divided by the reference equivalent series resistance to obtain the relative residual of the equivalent series resistance; the ratio of strain difference to sampling distance between adjacent sampling points on the surface of the unit shell in the deformation response data is determined as the local strain gradient, and the difference between the measured local strain gradient and the reference local strain gradient is divided by the reference local strain gradient to obtain the relative residual of the local strain gradient; the temperature decay time constant obtained by fitting the thermal response data in the pulse recovery segment is determined as the thermal diffusion time constant, and the difference between the measured thermal diffusion time constant and the reference thermal diffusion time constant is divided by the reference thermal diffusion time constant to obtain the relative residual of the thermal diffusion time constant.

[0096] Then, the relative residuals of the equivalent series resistance, local strain gradient, and thermal diffusion time constant were normalized to obtain three normalized features, and the damage intensity value A was calculated using a weighted summation method. The damage intensity value is a comprehensive quantitative result obtained by further extraction and summarization based on the multimodal residual field, used to characterize the degree of structural damage, and serves as the quantitative basis for subsequent damage stage determination and the determination of safety pulse boundary parameters.

[0097] The formula for calculating the damage intensity value is: ;

[0098] in, This represents the relative residual of the normalized equivalent series resistance. This represents the relative residual of the normalized local strain gradient. This represents the relative residual of the normalized thermal diffusion time constant. Since in vehicle-mounted hybrid energy storage scenarios, issues such as loose connections, electrode delamination, and localized thermal instability must be considered, and contact degradation in the connection area and electrode delamination have a greater impact on early safety, higher weights are given to the electrical residual and deformation residual. For example, the three weights are 0.38, 0.37, and 0.25 respectively, and the total weight sum is 1.

[0099] This step involves first performing residual processing on the original response based on reference response data, and then extracting damage features and constructing damage intensity values ​​based on the multimodal residual field. This helps to compress the scattered abnormal features into quantifiable and comparable damage degree indicators while retaining the abnormal distribution information of each unit, time window, and modality. This reduces the interference of common components of the working condition on the subsequent identification results, further improves the prominence of structural damage-related abnormal features, and provides a unified quantitative basis for subsequent damage stage determination and safety pulse boundary parameter determination.

[0100] As an optional implementation, the specific steps for acquiring connection relationship data and thermal coupling relationship data between units in a supercapacitor module, and constructing a unit topology diagram based on the connection relationship data and thermal coupling relationship data, include:

[0101] First, the busbar connection record of the supercapacitor module is obtained, and the connection sequence and adjacency relationship of each unit formed by the busbar are determined according to the corresponding relationship of each unit number recorded in the busbar connection record, thus obtaining the connection relationship data; wherein, the connection relationship data is used to characterize the electrical connection sequence and adjacency relationship between each unit.

[0102] Specifically, each unit is defined as a node, and connection edges between nodes are established based on the busbar connection record; when two units are directly adjacent via the same busbar, a connection edge is established between them, and the connection weight is set to 1; when there is one intermediate unit between two units, a connection edge is established between them, and the connection weight is set to 0.5.

[0103] For example, if unit 1, unit 2, and unit 3 are connected sequentially via the same busbar, then a connection edge with a weight of 1 is established between unit 1 and unit 2, a connection edge with a weight of 1 is established between unit 2 and unit 3, and a connection edge with a weight of 0.5 is established between unit 1 and unit 3. In this way, the busbar connection structure can be transformed into connection relationship data that can be used for graph structure representation.

[0104] Secondly, under preset thermal excitation conditions, a thermal load is applied to the supercapacitor module, and temperature change data of each unit during the thermal load process is collected by a temperature acquisition unit to obtain temperature rise response data. The preset thermal excitation conditions refer to controlled heating conditions that allow heat to transfer between units within the module. In this embodiment, the preset thermal excitation conditions are: an ambient temperature maintained between 25 and 30 degrees Celsius, and a constant thermal power input of 5 to 8 watts continuously applied to the target unit for 30 seconds. The temperature rise response data is used to characterize the temperature rise amplitude and temperature change rate of each unit under the thermal excitation conditions.

[0105] Based on the temperature rise response data of each unit, the proportion of temperature rise transfer to other units relative to the target unit after thermal excitation of the target unit is calculated to obtain thermal coupling relationship data; among them, the larger the temperature rise transfer ratio, the higher the degree of thermal influence between the corresponding units.

[0106] For example, if the target unit is denoted as unit i, and the other units are denoted as unit j, the temperature rise of the target unit at the end of 30 seconds is denoted as... The temperature rise of other units at the end of 30 seconds is recorded as follows: Then the temperature rise transfer ratio of unit j relative to unit i Determine using the following formula: The thermal coupling edges and their weights between units are determined based on thermal coupling relationship data. Specifically, when the temperature rise transfer ratio... At that time, a first thermal coupling edge is established between element i and element j, and the thermal coupling weight is set to 0.7; when the temperature rise transfer ratio At that time, a second thermal coupling edge is established between element i and element j, and the thermal coupling weight is set to 0.35; when the temperature rise transfer ratio When the temperature rise is proportional, a third thermal coupling edge is established between unit i and unit j, and the thermal coupling weight is set to 0.15; when the temperature rise is proportional, no thermal coupling edge is established between unit i and unit j.

[0107] Subsequently, based on the multimodal residual field, the target unit is located according to the unit number, the target time interval is located according to the current pulse window, and the electrical residual data, thermal residual data, and deformation residual data of the target unit within the target time interval are read according to the electrical category, thermal category, and deformation category, respectively. Then, the electrical residual data, thermal residual data, and deformation residual data are combined in the order of electrical residual data first, thermal residual data in the middle, and deformation residual data last to obtain the node features of the corresponding node.

[0108] For example, if the electrical residual data of unit 2 in the current pulse window is 6-dimensional, the thermal residual data is 4-dimensional, and the deformation residual data is 5-dimensional, then the 6-dimensional electrical residual data, the 4-dimensional thermal residual data, and the 5-dimensional deformation residual data are combined in sequence to obtain the 15-dimensional node features of the corresponding node of unit 2.

[0109] Finally, the connection edges and their connection weights, the thermal coupling edges and their thermal coupling weights, and the node characteristics of each node are written into the graph structure to construct the unit topology graph. The unit topology graph is used to characterize the electrical connection propagation path and the thermal coupling influence path within the module; among them, nodes are used to characterize the comprehensive residual state of each unit under the current pulse window, connection edges are used to characterize the electrical connection propagation relationship, and thermal coupling edges are used to characterize the thermal influence propagation relationship.

[0110] This step establishes connection edges based on the busbar connection table, establishes thermal coupling edges based on the temperature rise transfer ratio, and combines the electrical residual data, thermal residual data, and deformation residual data corresponding to each unit under the current pulse window as node features and writes them into the unit topology graph. This is beneficial for simultaneously characterizing unit body anomalies, electrical connection propagation anomalies, and thermal coupling propagation anomalies, thereby distinguishing between the body damage of a single unit and the coupling anomalies formed by the propagation of neighboring units, reducing the number of cases where the propagation response is judged as body damage, and improving the accuracy of subsequent structural damage location and damage type identification.

[0111] As an optional implementation, the specific steps for performing structural damage identification based on multimodal residual fields and element topology diagrams to obtain damage location and damage type include:

[0112] First, a physical constraint graph recognition model is established. This model includes a temporal feature extraction layer, a graph propagation layer, a consistency determination layer, and a result output layer. Specifically, the temporal feature extraction layer extracts node features of the target element within the current pulse window and the two preceding consecutive pulse windows, combining them in chronological order to obtain temporal features. The graph propagation layer performs propagation calculations on the temporal features of each element based on connecting edges and thermally coupled edges to obtain aggregated features. The consistency determination layer determines the electrical, mechanical, and thermal consistency determination results based on the aggregated features. The result output layer determines the damage location and damage type based on the electrical, mechanical, and thermal consistency determination results.

[0113] Subsequently, the multimodal residual field and the element topology graph are input together into the physical constraint graph recognition model.

[0114] Specifically, the time feature extraction layer reads the node features of the target unit in the current pulse window and the previous two consecutive pulse windows, and combines them in chronological order to obtain the time features of the target unit; similarly, it obtains the time features of adjacent units that are electrically connected to the target unit, as well as the time features of adjacent units that are thermally coupled to the target unit.

[0115] Taking target unit i as an example, the aggregation features of target unit i The calculation formula is as follows: In the formula, The aggregation features of target unit i are represented; Represents the original node features of target unit i; Represents the original node features of unit j adjacent to target unit i; This represents the set of adjacent units that are electrically connected to the target unit i. This represents the set of adjacent elements that have a thermal coupling relationship with the target element i; This represents the connection weight between target cell i and its neighboring cell j; This represents the thermal coupling weight between target unit i and its adjacent unit j.

[0116] The first item is 0.5. The second item is used to preserve the abnormal characteristics of the target unit itself. The third term is used to introduce anomalous features of adjacent units j that are electrically connected to target unit i, in order to characterize the influence of electrical connections propagated via the busbar. This is used to introduce anomalous features of neighboring unit j that has a thermal coupling relationship with target unit i, in order to characterize the thermal coupling effect formed by heat transfer.

[0117] For example, if the electrical residual in the node features of unit 4 is large, and the adjacent units 3 and 5 also have obvious thermal residuals, then after the above propagation calculation, the aggregation feature of unit 4 not only includes its own anomaly, but also the propagation influence information from units 3 and 5.

[0118] After obtaining the aggregation characteristics, the consistency determination layer determines the electrical consistency determination results, mechanical consistency determination results, and thermal consistency determination results, respectively. Specifically, when the relative residual of contact resistance is greater than or equal to 0.15 and the relative residual of transient voltage drop is greater than or equal to 0.12, the electrical consistency determination result is recorded as valid; when the relative residual of local strain gradient is greater than or equal to 0.18, the mechanical consistency determination result is recorded as valid; when the relative residual of thermal diffusion time constant is greater than or equal to 0.2 and the relative residual of local temperature difference is greater than or equal to 0.15, the thermal consistency determination result is recorded as valid. Then, a joint determination is made based on the above three determination results: when at least two of the three consistency determination results are valid, the corresponding unit is determined as an effective structural damage unit; when fewer than two of the three consistency determination results are valid, the corresponding unit is not determined as an effective structural damage unit.

[0119] After identifying the effective structural damage units, the result output layer further outputs the damage location and damage type. The damage location includes the target unit number and the target abnormal region. The target abnormal region includes the busbar connection area, the shell surface bulge area, the temperature rise concentration area, the internal stress area of ​​the unit, and the internal interface area of ​​the unit. Specifically, when the electrical consistency judgment result is valid and the abnormal peak corresponds to the busbar connection area, the damage location is determined to be the busbar connection area of ​​the target unit, and the damage type is determined to be current collector connection looseness. Current collector connection looseness is used to characterize the abnormal contact state in the electrical connection path of the target unit. When the mechanical consistency judgment result is valid and the abnormal peak corresponds to the shell surface bulge area, the damage location is determined to be the shell surface bulge area of ​​the target unit, and the damage type is determined to be shell bulge. Shell bulge is used to characterize the local bulging abnormality of the external encapsulation structure of the target unit. When the thermal consistency judgment result is valid and the abnormal peak corresponds to the temperature rise concentration area, the damage location is determined to be the temperature rise concentration area of ​​the target unit, and the damage type is determined to be electrical... Local instability of the electrolyte is used to characterize abnormal thermal diffusion and dielectric distribution within the target unit. When both electrical and mechanical consistency judgments are met, and the anomaly is concentrated in the stress area within the unit, the damage location is determined to be the stress area within the target unit, and the damage type is determined to be electrode delamination. Electrode delamination is used to characterize abnormal electrode layer bonding within the target unit. When both mechanical and thermal consistency judgments are met, and the anomaly is concentrated in the interface area within the unit, the damage location is determined to be the interface area within the target unit, and the damage type is determined to be local diaphragm deformation. Local diaphragm deformation is used to characterize abnormal local deformation of the diaphragm structure within the target unit.

[0120] For example, when the relative residual of the contact resistance of unit 3 is 0.19, the relative residual of the transient voltage drop is 0.16, and the abnormal peak is located in the busbar connection area, the busbar connection area of ​​unit 3 can be identified as the damage location, and the damage type can be identified as loose current collector connection; when the relative residual of the local strain gradient of unit 5 is 0.21, the relative residual of the thermal diffusion time constant is 0.24, and the abnormality is concentrated in the internal interface area of ​​the unit, the internal interface area of ​​unit 5 can be identified as the damage location, and the damage type can be identified as local deformation of the diaphragm.

[0121] This step involves first establishing a physical constraint diagram identification model, then performing propagation calculations on time features based on connecting edges and thermal coupling edges, and determining effective structural damage units based on electrical consistency, mechanical consistency, and thermal consistency judgment results. It then outputs the target unit number, target abnormal region, and corresponding damage type, which helps distinguish propagation anomalies from intrinsic anomalies, improves the accuracy of structural damage location and damage type identification, and provides a direct basis for subsequent damage stage determination and safety pulse boundary parameter determination.

[0122] Example 2:

[0123] This embodiment 2 further provides an improved solution based on embodiment 1. Embodiment 1 has achieved online identification of structural damage to supercapacitor modules under the combined effects of temperature fluctuations, changes in state of charge, equilibrium deviations, and high-frequency pulse disturbances, and outputs the damage location and type. However, in actual vehicle-mounted regenerative braking applications, simply obtaining the damage location and type is insufficient to directly support vehicle operation control. This is because different damage types have significantly different impacts on peak current withstand capability, pulse duration capability, and continuous pulse withstand capability at different stages of evolution. Without further stage determination and boundary quantization, the control strategy may be too conservative or the boundary contraction may be insufficient, affecting the module's operational safety and energy recovery efficiency. This embodiment introduces damage stage determination, safety pulse boundary parameter back-calculation, and a self-correcting closed-loop update mechanism based on embodiment 1, achieving a closed-loop optimization effect from structural damage identification to operational boundary output and subsequent adaptive correction under operating conditions. This further solves the defects of the identification results being difficult to directly convert into executable control boundaries and the easy drift of boundary parameters during long-term operation. The specific implementation method is as follows:

[0124] As an optional implementation, the specific steps for determining the damage stage based on the damage location and damage type include:

[0125] After obtaining the damage location and damage type in Example 1, the target unit is located based on the damage location, and the corresponding damage intensity value of the target unit is read. Then, the corresponding stage threshold group is matched according to the damage type, and the damage intensity value is compared with the matched stage threshold group to obtain the damage stage. The damage stage is used to characterize the degree of evolution of structural damage, including the early damage stage, the development stage, and the risk stage.

[0126] Specifically, corresponding stage threshold groups are pre-set for different damage types. Each stage threshold group is calibrated based on stable operating samples, accelerated aging samples, and historical pulse samples of the same type of supercapacitor module, thus matching the stage division boundaries corresponding to different damage types with their actual evolution patterns. After calibration, the same damage type is determined using the corresponding fixed stage threshold group. The damage location is used to locate the target unit, the damage type is used to match the stage threshold group corresponding to that type of damage, and the damage intensity value is used to characterize the degree of damage to the target unit under the current pulse window. These three factors together serve as the basis for determining the damage stage.

[0127] Specifically, taking the damage intensity value as normalized to the range of 0 to 1 as an example, when the damage type is electrode delamination, if the damage intensity value is greater than or equal to 0.3 and less than 0.51, it is determined to be in the early damage stage; if the damage intensity value is greater than or equal to 0.51 and less than 0.73, it is determined to be in the development stage; if the damage intensity value is greater than or equal to 0.73 and less than or equal to 1, it is determined to be in the risk stage.

[0128] When the damage type is loose current collector connection, if the damage intensity value is greater than or equal to 0.25 and less than 0.46, it is determined to be in the early damage stage; if the damage intensity value is greater than or equal to 0.46 and less than 0.69, it is determined to be in the development stage; and if the damage intensity value is greater than or equal to 0.69 and less than or equal to 1, it is determined to be in the risk stage. Because this type of damage has a significant impact on transient voltage drop and contact resistance in its early stages, its stage threshold group enters the judgment range earlier than electrode delamination.

[0129] When the damage type is local electrolyte instability, if the damage intensity value is greater than or equal to 0.28 and less than 0.49, it is determined to be in the early damage stage; if the damage intensity value is greater than or equal to 0.49 and less than 0.71, it is determined to be in the development stage; if the damage intensity value is greater than or equal to 0.71 and less than or equal to 1, it is determined to be in the risk stage.

[0130] When the damage type is local deformation of the diaphragm, if the damage intensity value is greater than or equal to 0.27 and less than 0.47, it is determined to be in the early damage stage; if the damage intensity value is greater than or equal to 0.47 and less than 0.69, it is determined to be in the development stage; and if the damage intensity value is greater than or equal to 0.69 and less than or equal to 1, it is determined to be in the risk stage. Local deformation of the diaphragm is used to characterize the local deformation anomalies of the diaphragm structure inside the target unit.

[0131] When the damage type is shell bulging, if the damage intensity value is greater than or equal to 0.32 and less than 0.52, it is determined to be in the early damage stage; if the damage intensity value is greater than or equal to 0.52 and less than 0.74, it is determined to be in the development stage; if the damage intensity value is greater than or equal to 0.74 and less than or equal to 1, it is determined to be in the risk stage. Shell bulging is used to characterize local bulging anomalies in the external packaging structure of the target unit.

[0132] For example, when the damage location of unit 3 is determined, and its damage type is loose current collector connection, and the damage intensity value of unit 3 is read as 0.58, since 0.58 is greater than or equal to 0.46 and less than 0.69, the damage stage of unit 3 is determined to be the development stage. Similarly, when the damage type of unit 5 is local electrolyte instability, and the damage intensity value is 0.74, since 0.74 is greater than or equal to 0.71 and less than or equal to 1, the damage stage of unit 5 is determined to be the risk stage.

[0133] This step locates the target unit based on the damage location, matches the stage threshold group corresponding to the damage type's evolution law based on the damage type, and compares the damage intensity value with the matched stage threshold group to obtain the damage stage. This helps to advance the identification of the preceding structural damage from identifying what kind of damage it is to identifying the extent to which the damage has developed. As a result, the determination of the subsequent safety pulse boundary parameters can be directly based on the common constraints of the target unit, damage type, and damage stage, improving the pertinence and reliability of the subsequent control boundary determination.

[0134] As an optional implementation, the specific steps for determining the safety pulse boundary parameters based on the damage location, damage type, and damage stage include:

[0135] A baseline operating parameter set is pre-established based on stable operating samples of the same type of supercapacitor module, and a correction coefficient table is established based on accelerated aging samples and historical pulse samples. Specifically, the baseline operating parameter set is established by grouping stable operating samples according to ambient temperature range and state of charge range, and statistically analyzing representative values ​​of peak current, pulse width, and number of consecutive pulses within each group to obtain the baseline peak current, baseline pulse width, and baseline number of consecutive pulses for the corresponding ambient temperature range and state of charge range. The representative values ​​are obtained using one of the following: average value, weighted average value, and median value. The correction coefficient table is established by classifying accelerated aging samples and historical pulse samples according to damage stage, damage type, and target unit location category, and calculating the proportional relationship between the safe operating boundary and the corresponding baseline operating parameters for each category to obtain the stage correction coefficient corresponding to the damage stage, the type correction coefficient corresponding to the damage type, and the location correction coefficient corresponding to the target unit location category.

[0136] After obtaining the damage location, damage type, and damage stage, the target unit is located based on the damage location, and the position category of the target unit is determined according to the unit arrangement order of the target unit number in the supercapacitor module; among them, the target units located at both ends of the unit arrangement are determined as edge positions, and the target units located in the middle of the unit arrangement are determined as center positions.

[0137] Subsequently, based on the ambient temperature and state of charge corresponding to the current operating condition label, the corresponding reference peak current, reference pulse width, and reference continuous pulse count are read from the reference operating parameter set; then, based on the damage stage, damage type, and location category, the stage current correction coefficient, type current correction coefficient, location current correction coefficient, stage width correction coefficient, type width correction coefficient, location width correction coefficient, as well as the stage count correction coefficient, type count correction coefficient, and location count correction coefficient are read from the correction coefficient table.

[0138] The allowable peak current is determined based on the reference peak current, stage current correction factor, type current correction factor, and position current correction factor; the allowable pulse width is determined based on the reference pulse width, stage width correction factor, type width correction factor, and position width correction factor; and the allowable number of consecutive pulses is determined based on the reference number of consecutive pulses, stage number correction factor, type number correction factor, and position number correction factor.

[0139] Specifically, allow peak current Allowable pulse width and the number of consecutive pulses allowed Determine them respectively using the following formulas: ; ; ;

[0140] In the formula, Indicates the reference peak current. Indicates the reference pulse width. Indicates the number of reference continuous pulses; , , These represent the current correction factor, width correction factor, and number correction factor corresponding to the damage stage, respectively. , , These represent the current correction factor, width correction factor, and number correction factor corresponding to the damage type, respectively. , , These represent the current correction factor, width correction factor, and number correction factor corresponding to the target cell location category, respectively.

[0141] For example, when the target unit is located in the middle of the module, the damage type is local electrolyte instability, and the damage stage is the risk stage, first read the reference peak current, reference pulse width, and reference number of consecutive pulses corresponding to the current ambient temperature range and state of charge range from the reference operating parameter set. Then, read the stage correction coefficient corresponding to the risk stage, the type correction coefficient corresponding to local electrolyte instability, and the position correction coefficient corresponding to the middle position of the module from the correction coefficient table. Finally, substitute them into the above formulas to obtain the allowable peak current, allowable pulse width, and allowable number of consecutive pulses.

[0142] The safety pulse boundary parameters include the allowable peak current, allowable pulse width, and allowable number of consecutive pulses, which are used to characterize the pulse operation boundary that can be applied to the target unit under the corresponding damage state.

[0143] This step involves first establishing a set of baseline operating parameters and a table of correction coefficients corresponding to ambient temperature, state of charge, damage stage, damage type, and target unit location category. Then, based on the damage location, damage type, and damage stage, it determines the allowable peak current, allowable pulse width, and allowable number of consecutive pulses. This facilitates the direct conversion of the structural damage results obtained from the preceding online identification into executable control boundaries for the vehicle system, avoiding reliance on laboratory characterization or offline analysis. This enhances the practical application value and engineering deployment feasibility of early structural damage localization results in vehicle scenarios.

[0144] As an optional implementation, the specific steps for performing operation control and updating the safety pulse boundary parameters based on the safety pulse boundary parameters include:

[0145] Current limiting control, pulse width limiting control, and continuous pulse count limiting control are performed based on safety pulse boundary parameters. Among them, current limiting control is used to limit the actual pulse current from not exceeding the allowable peak current, pulse width limiting control is used to limit the actual pulse width from not exceeding the allowable pulse width, and continuous pulse count limiting control is used to limit the number of continuous pulses from not exceeding the allowable number of continuous pulses.

[0146] After executing the above control, the multimodal response data after execution is continuously collected, and the actual peak current, actual pulse width and actual number of continuous pulses after execution are read to obtain the execution result data; at the same time, the corresponding operating condition label is reconstructed based on the multimodal response data after execution.

[0147] Subsequently, based on the executed multimodal response data, corresponding working condition labels, and execution result data, the working condition mapping parameters and damage stage determination threshold parameters in the reference response generation model are slowly updated. Slow updates refer to performing parameter updates once every 50 pulse windows, with each update using an update step size of 0.001. The working condition mapping parameters characterize the correspondence between working condition labels and reference response data, while the damage stage determination threshold parameters characterize the determination boundary between damage intensity values ​​and damage stages.

[0148] Specifically, the working condition mapping parameters are updated based on the multimodal response data after execution, the corresponding working condition labels, and the execution result data; the damage stage determination threshold parameters are updated based on the redefined damage location, damage type, damage stage, and execution result data.

[0149] After the parameter update is completed, the reference response data is regenerated based on the updated reference response generation model. The multimodal residual field is reconstructed based on the multimodal response data after execution and the updated reference response data. Then, the damage location, damage type and damage stage are re-determined based on the multimodal residual field and the element topology diagram. Finally, the safety pulse boundary parameters are re-determined based on the re-determined damage location, damage type and damage stage.

[0150] This step involves first executing control based on the safety pulse boundary parameters, then slowly updating the working condition mapping parameters and damage stage judgment threshold parameters based on the multimodal response data, corresponding working condition labels, and execution result data. After completing the parameter update, the reference response data is regenerated, the multimodal residual field is reconstructed, and the damage location, damage type, and damage stage are redefined, thereby redetermining the safety pulse boundary parameters. This helps to gradually adjust the safety pulse boundary parameters according to the actual service status of the module and reduce the accumulation of misjudgments caused by the offset of the judgment boundary.

[0151] Example 3:

[0152] This embodiment provides an AI-based supercapacitor structural damage detection system, which includes:

[0153] The acquisition module is used to acquire multimodal response data;

[0154] The feature construction module is used to read operating parameters and construct operating labels based on multimodal response data, generate reference response data based on the operating labels, construct a multimodal residual field based on the multimodal response data and reference response data, and obtain the connection relationship data and thermal coupling relationship data between the units in the supercapacitor module to construct the unit topology diagram.

[0155] The damage determination module is used to perform structural damage identification based on multimodal residual fields and element topology diagrams, obtain damage location and damage type, and determine the damage stage based on damage location and damage type.

[0156] The boundary determination module is used to determine the safety pulse boundary parameters based on the damage location, damage type, and damage stage.

[0157] like Figure 2 As shown, compared with the prior art, the embodiments of the present invention exhibit higher relative performance in terms of anti-interference of operating condition fluctuations, purification of abnormal features, unit-level localization, damage type differentiation, control boundary output, and long-term online self-adaptation. The embodiments of the present invention achieve operating condition decoupling through operating condition labeling and reference response generation, achieve abnormal information purification through multimodal residual fields, achieve unit-level localization and type determination through unit topology diagrams and structural damage identification, and further transform the identification results into executable operating boundaries of the vehicle system through damage stage and safety pulse boundary parameter determination. Finally, the self-calibrating closed loop enables the safety pulse boundary parameters to be gradually adjusted according to the actual service status of the module.

[0158] 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 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 claimed invention.

Claims

1. An AI-based method for detecting structural damage in supercapacitors, characterized in that, include: Acquire multimodal response data; Based on the multimodal response data, the operating condition parameters are read, and operating condition labels are constructed according to the operating condition parameters. Reference response data is then generated based on the operating condition labels. A multimodal residual field is constructed based on the multimodal response data and the reference response data; Obtain the connection relationship data and thermal coupling relationship data between the units in the supercapacitor module, and construct the unit topology diagram based on the connection relationship data and the thermal coupling relationship data; Structural damage identification is performed based on the multimodal residual field and the unit topology map to obtain the damage location and damage type; The damage stage is determined based on the damage location and the damage type; Based on the damage location, the damage type, and the damage stage, the safety pulse boundary parameters are determined.

2. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for constructing the operating condition label based on the operating condition parameters include: Based on the multimodal response data, energy storage level parameters, external temperature environment parameters, pulse intensity parameters, pulse duration parameters, and electrical connection initial consistency parameters are extracted. The energy storage level parameter, the external temperature environment parameter, the pulse intensity parameter, the pulse duration parameter, and the electrical connection initial consistency parameter are written into the corresponding operating condition fields, and operating condition tags are constructed based on each operating condition field.

3. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for generating reference response data based on the operating condition label include: Pre-construct the target sample set and the first reference response generation model; The first reference response generation model is trained based on the target sample set to obtain the second reference response generation model; Based on the operating condition label, reference response data is generated using the second reference response generation model.

4. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for constructing the multimodal residual field based on the multimodal response data and the reference response data include: Based on the multimodal response data and reference response data, the electrical residual sequence, thermal residual sequence, and deformation residual sequence are determined. The electrical residual sequence, thermal residual sequence, and deformation residual sequence are merged to form a multimodal residual field.

5. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for constructing the unit topology diagram based on the connection relationship data and the thermal coupling relationship data include: Determine the connection edges and their connection weights based on the connection relationship data; Determine the thermal coupling edges and their thermal coupling weights based on the aforementioned thermal coupling relationship data; The node features corresponding to each unit are determined based on the multimodal residual field. Based on the connecting edges and their connecting weights, thermally coupled edges and their thermally coupled weights, and node features, a unit topology graph is constructed.

6. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for obtaining the damage location and damage type include: Based on the connection edges, thermal coupling edges, and node features in the unit topology graph, propagation and aggregation are performed to obtain the aggregation features corresponding to each unit. Based on the aforementioned polymerization characteristics, the electrical consistency determination results, mechanical consistency determination results, and thermal consistency determination results are determined respectively; Based on the electrical consistency determination results, the mechanical consistency determination results, and the thermal consistency determination results, a joint determination is made to identify effective structural damage units; The location and type of damage are determined based on the effective structural damage units.

7. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for determining the damage stage based on the damage location and the damage type include: Damage intensity values ​​are constructed based on the multimodal residual field; The target unit is located based on the damage location, and the damage intensity value corresponding to the target unit is read. Pre-define and establish stage threshold groups corresponding to different damage types, and match the corresponding stage threshold groups according to the damage type; The damage stage corresponding to the target unit is determined by comparing the damage intensity value with the stage threshold group.

8. The method for detecting structural damage in a supercapacitor according to claim 7, characterized in that, The specific steps for constructing the damage intensity value based on the multimodal residual field include: Based on the multimodal residual field, electrical residual features, deformation residual features, and thermal residual features are extracted; The electrical residual characteristics, deformation residual characteristics, and thermal residual characteristics are normalized to obtain the corresponding normalized characteristics; The damage intensity value is obtained by weighted summarization based on the normalized features.

9. The method for detecting structural damage in a supercapacitor according to claim 1, characterized in that, The specific steps for determining the safety pulse boundary parameters include: Pre-establish a set of baseline operating parameters and a table of correction coefficients; The target unit is determined based on the location of the damage, and the location category of the target unit is determined; Based on the operating condition label, the baseline operating parameters are read from the baseline operating parameter set, and the corresponding correction coefficients are read from the correction coefficient table based on the damage stage, the damage type, and the location category. Based on the baseline operating parameters and the corresponding correction coefficients, the safety pulse boundary parameters are determined; The operation control is performed and the safety pulse boundary parameters are updated based on the aforementioned safety pulse boundary parameters.

10. An AI-based supercapacitor structural damage detection system, applicable to the supercapacitor structural damage detection method according to any one of claims 1 to 9, characterized in that, include: The acquisition module is used to acquire multimodal response data; The feature construction module is used to read operating parameters and construct operating labels based on the multimodal response data, generate reference response data based on the operating labels, construct a multimodal residual field based on the multimodal response data and the reference response data, and obtain the connection relationship data and thermal coupling relationship data between each unit in the supercapacitor module to construct a unit topology diagram. The damage determination module is used to perform structural damage identification based on the multimodal residual field and the unit topology map, obtain the damage location and damage type, and determine the damage stage based on the damage location and the damage type. The boundary determination module is used to determine the safety pulse boundary parameters based on the damage location, the damage type, and the damage stage.

Images