Battery pack internal fault identification method and system

Abstract

The invention discloses a battery pack internal fault identification method and system, and belongs to the technical field of battery safety detection, and the method comprises the steps: obtaining a standard impedance signal, a signal residual value and a standard charging and discharging quality signal of a to-be-detected battery pack; based on the internal detection area of the to-be-detected battery pack, extracting an internal detection area temperature difference and a three-dimensional hot spot form map; performing feature extraction, multi-scale feature fusion and feature mapping conversion on the to-be-detected battery pack based on a preset improved YOLOv8 model to obtain visual feature data; and based on a preset fault diagnosis model, the standard impedance signal, the signal residual value, the standard charging and discharging quality signal, the temperature difference of the internal detection area, the three-dimensional hot spot morphological graph and the visual feature data are judged, and an internal fault result of the battery pack is obtained. According to the battery pack internal fault identification method provided by the invention, comprehensive fault identification can be carried out on the interior of the battery pack.

Description

Technical Field

[0001] This invention relates to the field of battery safety testing technology, and in particular to a method and system for identifying internal faults in a battery pack. Background Technology

[0002] With the acceleration of the global energy transition, electrochemical energy storage technology, a key technology supporting renewable energy consumption and grid peak shaving, is seeing its application scale continuously expand in smart grids, distributed energy systems, and electric vehicles. Among these technologies, lithium-ion battery packs have become the mainstream choice for electrochemical energy storage due to their high energy density and long cycle life. However, their complex internal structure makes them highly susceptible to thermal runaway during actual operation, issues such as cell aging, localized overheating, and abnormal charging and discharging. This can lead to fires, explosions, and other safety accidents, causing significant property damage and seriously threatening personnel safety. Therefore, achieving accurate detection of the internal state of battery packs and early fault warning is a core requirement for the safe and stable development of the electrochemical energy storage industry.

[0003] Under the current technological background, current battery pack testing technologies still have many problems and are difficult to meet actual safety testing needs. First, most testing technologies rely on single-parameter testing equipment, which is difficult to comprehensively cover the complex internal fault characteristics. For example, independent impedance testing devices mostly use offline measurement methods, which cannot capture the dynamic impedance changes of the battery pack in real time during charging and discharging, making it difficult to identify early cell degradation problems. Secondly, charge and discharge quality monitoring equipment generally suffers from insufficient accuracy in capturing high-frequency components, making it impossible to effectively identify the potential risks caused by current and voltage waveform distortion. Thirdly, thermal detection devices such as infrared thermometers can only detect the surface temperature of the battery pack. Due to the closed packaging structure, it is difficult to penetrate into the module to obtain the true hot spot distribution, and it is easy to miss the opportunity for fault intervention due to the lag in temperature detection. Moreover, the above-mentioned single-parameter testing methods can only reflect the local state of the battery pack and cannot form a complete fault diagnosis logic, which will lead to a high rate of missed and false diagnoses of early faults. Secondly, existing multi-parameter testing solutions are mostly simple superpositions of various independent devices, lacking systematic integration and data collaboration capabilities. That is, impedance testing, charge and discharge quality monitoring, temperature detection and other devices operate independently, and data is stored in different systems, making it impossible to achieve synchronous analysis and correlation judgment. Meanwhile, some detection devices are complex to operate, requiring specialized personnel to debug them individually. Furthermore, the probe structure is incompatible with the battery pack inspection ports, making it difficult to penetrate deep into the battery pack for on-site testing. This makes them unsuitable for the efficient inspection scenarios of large-scale energy storage power stations. More importantly, the risk assessment system of traditional detection solutions contains logical contradictions. Maintenance personnel lack reliable judgment criteria when facing complex faults, easily leading to misjudgments of warning levels and missing the opportunity for early intervention in thermal runaway. Summary of the Invention

[0004] The present invention aims to provide a method and system for identifying internal faults in a battery pack, so as to solve the above-mentioned technical problems, avoid the problems of missed or misjudged internal faults and delayed early warning caused by incomplete internal detection of the battery pack, and realize comprehensive fault identification of the internal battery pack.

[0005] To address the aforementioned technical problems, this invention provides a method for identifying internal faults in a battery pack, comprising: Acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test; Based on the internal detection area and the spatial distribution of preset detection points of the battery pack to be tested, the temperature difference and three-dimensional hot spot morphology map of the internal detection area are extracted. Based on the preset improved YOLOv8 model, feature extraction and multi-scale feature fusion are performed on the initial visual image inside the battery pack to obtain the first visual feature map, the second visual feature map and the third visual feature map. Then, feature mapping transformation is performed on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data. Based on a preset fault diagnosis model, the system performs multi-dimensional identification of standard impedance signals, signal residual values, standard charge and discharge quality signals, internal detection area temperature differences, three-dimensional hotspot morphology maps, and visual feature data to obtain internal fault results of the battery pack.

[0006] In the above scheme, by acquiring the standard impedance signal, signal residual value, standard charge / discharge quality signal, and initial internal visual image of the battery pack under test, the electrical and visual characteristic parameters of the battery pack can be extracted, providing data support for the determination of electrical and structural faults occurring inside the battery pack. Next, by extracting the temperature difference and three-dimensional hotspot morphology map of the internal detection area of ​​the battery pack under test based on the spatial distribution of preset detection points, the internal temperature distribution characteristics of the battery pack can be visualized, capturing the location, range, and morphological characteristics of abnormal temperatures inside the battery pack, providing an intuitive basis for determining thermal characteristic faults occurring inside the battery pack. Then, by using a preset improved YOLOv8 model to extract features and fuse multi-scale features from the initial internal visual image of the battery pack under test, the internal structural features of the battery pack can be extracted and fused, enhancing the edge, texture, and multi-scale semantic information expression of the fault structure. First, second, and third visual feature maps are obtained and feature-mapped to acquire visual feature data, providing highly recognizable visual feature data for determining structural faults occurring inside the battery pack. Finally, by using a pre-set fault diagnosis model, the system performs multi-dimensional identification of standard impedance signals, signal residual values, standard charge and discharge quality signals, internal detection area temperature differences, three-dimensional hotspot morphology maps, and visual feature data. This avoids the problems of incomplete internal detection of the battery pack, which could lead to missed or misjudged faults and delayed early warnings. The system can comprehensively cover all types of faults within the battery pack, obtain fault results, and achieve the identification of internal faults.

[0007] Further, acquiring the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test includes: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

[0008] In the above scheme, by acquiring the original impedance signal of the battery pack under test and performing low-pass filtering, power frequency interference and electromagnetic noise in the original impedance signal can be filtered out, eliminating the influence of non-feature interference on the impedance detection accuracy, and obtaining a purified impedance signal. Next, by using a preset exponential fitting model to perform feature fitting and residual quantization on the purified impedance signal, a fitted impedance signal and signal residual value reflecting the impedance characteristics of the battery pack can be obtained, which can be used as the basis for determining whether the impedance signal is abnormal in subsequent tests. Then, by performing phase correction on the fitted impedance signal, the phase offset error of the fitted impedance signal can be corrected, improving the calculation accuracy and obtaining a standard impedance signal that can be directly used for the electrical characteristic analysis of the battery pack. Finally, by acquiring the standard charge / discharge quality signal of the battery pack under test and the initial internal visual image of the battery pack, the battery pack charge / discharge electrical characteristic data and the original visual data of the internal structure are simultaneously collected, providing data support for subsequent charge / discharge fault determination and visual feature extraction.

[0009] Further, acquiring the standard charge / discharge quality signal of the battery pack under test and the initial internal visual image of the battery pack under test includes: The initial charge-discharge quality signal and the initial internal visual image of the battery pack under test are acquired, and the initial charge-discharge quality signal is de-DC processed to obtain a purified charge-discharge quality signal. Perform continuous wavelet transform on the purified charge and discharge quality signal to extract high-frequency feature data; The high-frequency feature data is reconstructed to obtain a standard charge / discharge quality signal.

[0010] In the above scheme, by acquiring the initial charge / discharge quality signal and initial internal visual image of the battery pack under test, raw charge / discharge data and raw visual data of the internal structure of the battery pack are simultaneously collected. The initial charge / discharge quality signal undergoes DC-DC removal processing to eliminate fundamental DC bias interference, resulting in a purified charge / discharge quality signal, laying the data foundation for subsequent high-frequency feature extraction. Next, by performing continuous wavelet transform on the purified charge / discharge quality signal, high-frequency components can be accurately extracted, obtaining high-frequency feature data and providing characteristic basis for determining charge / discharge quality anomalies. Finally, by reconstructing the high-frequency feature data, key high-frequency abnormal features in the charge / discharge signal can be preserved, resulting in a standard charge / discharge quality signal that can be directly used for battery pack charge / discharge quality fault identification, improving the accuracy and effectiveness of charge / discharge quality detection.

[0011] Furthermore, the step of extracting the temperature difference and three-dimensional hotspot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack under test includes: Based on the internal detection area of ​​the battery pack to be tested, the temperature values ​​of several detection points are obtained, and the temperature difference of the internal detection area is calculated based on the temperature values ​​of the several detection points. Based on the spatial distribution of preset detection points and the temperature difference in the internal detection area, a three-dimensional hotspot morphology map is extracted.

[0012] In the above scheme, temperature values ​​at several detection points are obtained through the internal detection area of ​​the battery pack under test, and the temperature difference within the internal detection area is calculated. This yields temperature difference data that reflects the differences in temperature distribution within the battery pack, providing temperature characteristic basis for hotspot identification. Next, by pre-setting the spatial distribution locations of the detection points and the temperature differences within the internal detection areas, the temperature values ​​and spatial location information can be fused and mapped. The extracted three-dimensional hotspot morphology map can intuitively present the spatial outline, distribution range, and concentration of hotspots within the battery pack, providing visualized thermal characteristic data for determining thermal anomalies.

[0013] Further, the preset improved YOLOv8 model includes a backbone network unit, a neck network unit, and a head network unit; the feature extraction and multi-scale feature fusion of the initial battery pack internal visual image based on the preset improved YOLOv8 model to obtain a first visual feature map, a second visual feature map, and a third visual feature map, and feature mapping transformation of the first visual feature map, the second visual feature map, and the third visual feature map to obtain visual feature data, including: The initial internal visual image of the battery pack is corrected and filtered to obtain a standard internal visual image of the battery pack. Based on the backbone network unit, feature extraction is performed on the visual image inside the standard battery pack to obtain the initial bottom layer feature map, the initial middle layer feature map and the initial shallow layer feature map. Based on the neck network unit, multi-scale feature fusion is performed on the initial low-level feature map, the initial middle-level feature map and the initial shallow-level feature map to obtain the first visual feature map, the second visual feature map and the third visual feature map. Based on the head network unit, feature mapping transformation is performed on the first visual feature map, the second visual feature map, and the third visual feature map to obtain visual feature data.

[0014] In the above scheme, by correcting and filtering the initial visual image of the battery pack's interior, image distortion and noise interference can be eliminated, restoring the true visual information of the battery pack's internal structure and providing a clear standard visual image of the battery pack's interior for subsequent feature extraction. Next, feature extraction is performed on the standard battery pack's interior visual image through the backbone network unit. This enables multi-scale basic feature mining of the standard visual image, extracting deep semantic features, mid-level structural features, and shallow detail features of the battery pack's internal structure layer by layer, obtaining initial bottom-level feature maps, initial mid-level feature maps, and initial shallow feature maps, laying the foundation for multi-scale feature fusion. Then, multi-scale feature fusion is performed on the initial bottom-level feature maps, initial mid-level feature maps, and initial shallow feature maps through the neck network unit. This allows for deep fusion and enhancement of basic features at different scales, compensating for the information deficiencies of single-scale features, strengthening the multi-scale semantic and detail expression capabilities of the battery pack's internal fault structure, and obtaining first, second, and third visual feature maps. Finally, by performing feature mapping transformation on the first, second, and third visual feature maps through the head network unit, the fused multi-scale visual feature maps can be converted into visual feature data suitable for fault identification, completing the quantification and reconstruction of visual features and providing a highly recognizable basis for subsequent multi-dimensional identification.

[0015] Further, the neck network unit includes a first neck branch subunit, a second neck branch subunit, and a third neck branch subunit; the multi-scale feature fusion based on the neck network unit of the initial low-level feature map, the initial middle-level feature map, and the initial shallow-level feature map to obtain a first visual feature map, a second visual feature map, and a third visual feature map includes: Based on the first neck branch subunit, the initial bottom layer feature map, the initial middle layer feature map, and the initial shallow layer feature map are enhanced and fused to obtain the first enhanced feature map and the second enhanced feature map. Based on the second neck branch subunit, the first enhanced feature map, the second enhanced feature map and the initial low-level feature map are convolutionally fused to obtain the third enhanced feature map and the fourth enhanced feature map; Based on the third neck branch subunit, the second enhanced feature map, the third enhanced feature map and the fourth enhanced feature map are sequentially convolutionally expanded and spatially reconstructed to obtain the first initial visual feature map, the second initial visual feature map and the third initial visual feature map; Based on the third neck branch subunit, channel compression and feature reshaping are performed on the first initial visual feature map, the second initial visual feature map, and the third initial visual feature map to obtain the first visual feature map, the second visual feature map, and the third visual feature map.

[0016] In the above scheme, the initial low-level feature map, initial mid-level feature map, and initial shallow feature map are enhanced and fused through the first neck branch subunit. This enables the fusion and enhancement of features at different scales, integrating deep semantics, mid-level structure, and shallow detail information, compensating for the information loss of single-scale features, and obtaining the first and second enhanced feature maps, laying the foundation for subsequent multi-scale convolutional fusion. Next, the second neck branch subunit performs convolutional transformation and secondary fusion on the first, second, and initial low-level feature maps to further enhance the spatial correlation and semantic expressive ability of the features, optimize the fusion effect of multi-scale features, and obtain the third and fourth enhanced feature maps. Then, the third neck branch subunit sequentially performs convolutional expansion and spatial feature reconstruction on the second, third, and fourth enhanced feature maps, which expands the channel dimension of the fused features, reconstructs and enhances spatial texture and boundary information, improves the expressive dimension of the features, strengthens the spatial feature details of the internal fault structure of the battery pack, highlights key fault feature information, and obtains the first, second, and third initial visual feature maps. Finally, channel compression and feature reshaping are performed on the first, second, and third initial visual feature maps by the third neck branch subunit. This reduces the complexity of feature computation while retaining core fault feature information, resulting in first, second, and third visual feature maps that are adapted for subsequent feature mapping transformations.

[0017] Further, the enhancement and fusion of the initial low-level feature map, the initial middle-level feature map, and the initial shallow-level feature map based on the first neck branch sub-unit to obtain the first enhanced feature map and the second enhanced feature map includes: The initial low-level feature map is dynamically upsampled to obtain the reconstructed low-level feature map; The reconstructed bottom-layer feature map and the initial middle-layer feature map are spliced ​​and fused to obtain the first fused feature map; Perform feature enhancement on the first fused feature map to obtain the first enhanced feature map; The first enhanced feature map is dynamically upsampled to obtain the reconstructed enhanced feature map; The reconstructed enhanced feature map and the initial shallow feature map are spliced ​​and fused to obtain a second fused feature map; The second fused feature map is enhanced to obtain the second enhanced feature map.

[0018] In the above scheme, by dynamically upsampling the initial low-level feature map, the resolution of the initial low-level feature map can be adaptively restored, reconstructing the edge and detail information of the internal structure of the battery pack. This compensates for the low resolution and lack of detail in the initial low-level feature map, resulting in a reconstructed low-level feature map, laying the foundation for cross-scale feature fusion. Next, by concatenating and fusing the reconstructed low-level feature map and the initial mid-level feature map, the high semantic features of the low-level map and the structural features of the mid-level map can be integrated along the channel dimension, fusing feature information at different scales. This allows the feature map to simultaneously possess semantic expression and structural awareness capabilities, obtaining the first fused feature map. Then, by enhancing the first fused feature map, the features after cross-scale fusion can be deeply extracted and strengthened, improving the representational ability of the features, obtaining the first enhanced feature map, highlighting key feature information related to internal faults in the battery pack. Subsequently, by dynamically upsampling the first enhanced feature map, the feature map resolution can be adaptively increased again, further supplementing the feature detail information, allowing the fused features to better match the scale and detail features of the initial shallow feature map in subsequent iterations, obtaining the reconstructed enhanced feature map. Next, by concatenating and fusing the reconstructed enhanced feature map and the initial shallow feature map, the enhanced features, which integrate low-level and mid-level features, are combined with shallow detail features. This fully integrates the semantic and fine-grained detail information of multi-scale features, enriching the information dimension of the features and obtaining a second fused feature map. Finally, by enhancing the second fused feature map, the discriminability and stability of the features can be further improved, resulting in a second enhanced feature map that can be directly used for subsequent convolutional fusion.

[0019] Further, the step of convolutionally fusing the first enhanced feature map, the second enhanced feature map, and the initial low-level feature map based on the second neck branch subunit to obtain the third enhanced feature map and the fourth enhanced feature map includes: Convolutional downsampling is performed on the second enhanced feature map to obtain the first convolutional feature map; The first convolutional feature map and the first enhanced feature map are concatenated and fused to obtain the third fused feature map; Perform feature enhancement on the third fused feature map to obtain the third enhanced feature map; Convolutional downsampling is performed on the third enhanced feature map to obtain the second convolutional feature map; The second convolutional feature map and the initial low-level feature map are concatenated and fused to obtain the fourth fused feature map; The fourth fused feature map is enhanced to obtain the fourth enhanced feature map.

[0020] In the above scheme, by performing convolutional downsampling on the second enhanced feature map, the scale of the second enhanced feature map can be compressed and the channel dimension adjusted to obtain the first convolutional feature map. Next, by concatenating and fusing the first convolutional feature map and the first enhanced feature map, the abstract semantic features of the second enhanced feature map and the fused feature information of the first enhanced feature map can be integrated in the channel dimension, further strengthening the correlation and complementarity of multi-scale features, enriching the information dimension of feature expression, and obtaining the third fused feature map. Then, by performing feature enhancement on the third fused feature map, the representational ability of the features can be improved, highlighting key feature information related to internal battery pack faults, filtering redundant feature interference, and obtaining the third enhanced feature map. Subsequently, by performing convolutional downsampling on the third enhanced feature map, the feature map can be scaled and abstracted again, focusing on core semantic features, and obtaining the second convolutional feature map. Next, by concatenating and fusing the second convolutional feature map and the initial low-level feature map, a fourth fused feature map is obtained. This integrates the abstract features enhanced by multiple fusions with the core semantic features of the initial low-level map, allowing the feature map to retain both the original semantic information of the low-level map and the enhanced feature information from multi-scale fusion, thus improving the integrity of the features. Finally, by enhancing the fourth fused feature map, the stability and discriminability of the features can be further improved, resulting in a fourth enhanced feature map that can be directly used for subsequent spatial feature reconstruction.

[0021] This invention provides a battery pack internal fault identification system, including an electrical vision data acquisition module, a thermal feature data extraction module, a visual feature data acquisition module, and a multi-dimensional fault identification module, specifically: The electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test. The thermal feature data extraction module is used to extract the temperature difference and three-dimensional hot spot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack to be tested. The visual feature data acquisition module is used to extract features and fuse multi-scale features from the initial internal visual image of the battery pack based on a preset improved YOLOv8 model, to obtain a first visual feature map, a second visual feature map and a third visual feature map, and to perform feature mapping transformation on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data. The multi-dimensional fault identification module is used to perform multi-dimensional identification of standard impedance signal, signal residual value, standard charge and discharge quality signal, internal detection area temperature difference, three-dimensional hot spot morphology map and visual feature data based on a preset fault diagnosis model, so as to obtain the internal fault results of the battery pack.

[0022] This invention provides a battery pack internal fault identification system. In practical applications, it only requires an electrical vision data acquisition module. By acquiring the standard impedance signal, signal residual value, standard charge / discharge quality signal, and initial internal visual image of the battery pack under test, it can extract the electrical and visual characteristic parameters of the battery pack, providing data support for determining electrical and structural faults occurring inside the battery pack. Next, a thermal feature data extraction module is used to extract the temperature difference and three-dimensional hotspot morphology map of the internal detection area based on the spatial distribution of the internal detection area and preset detection points of the battery pack under test. This visualizes the internal temperature distribution characteristics of the battery pack, capturing the location, range, and morphological characteristics of abnormal temperatures inside the battery pack, providing intuitive evidence for determining thermal characteristic faults occurring inside the battery pack. Then, a visual feature data acquisition module is employed. Using a pre-set improved YOLOv8 model, feature extraction and multi-scale feature fusion are performed on the initial internal visual image of the battery pack under test. This allows for the extraction and fusion of internal structural features, enhancing the expression of edge, texture, and multi-scale semantic information of faulty structures. First, second, and third visual feature maps are obtained and their features are mapped and transformed to acquire visual feature data. This provides highly recognizable visual feature data for identifying structural faults occurring within the battery pack. Finally, a multi-dimensional fault identification module is used. Through a pre-set fault diagnosis model, standard impedance signals, signal residual values, standard charge / discharge quality signals, internal detection area temperature differences, 3D hotspot morphology maps, and visual feature data are used for multi-dimensional identification. This avoids incomplete internal detection of the battery pack, preventing missed or misjudged faults and delayed early warnings. It comprehensively covers various fault types within the battery pack, acquiring fault results and enabling the identification of internal faults.

[0023] Furthermore, the electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test, including: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

[0024] In the above scheme, by acquiring the original impedance signal of the battery pack under test and performing low-pass filtering, power frequency interference and electromagnetic noise in the original impedance signal can be filtered out, eliminating the influence of non-feature interference on the impedance detection accuracy, and obtaining a purified impedance signal. Next, by using a preset exponential fitting model to perform feature fitting and residual quantization on the purified impedance signal, a fitted impedance signal and signal residual value reflecting the impedance characteristics of the battery pack can be obtained, which can be used as the basis for determining whether the impedance signal is abnormal in subsequent tests. Then, by performing phase correction on the fitted impedance signal, the phase offset error of the fitted impedance signal can be corrected, improving the calculation accuracy and obtaining a standard impedance signal that can be directly used for the electrical characteristic analysis of the battery pack. Finally, by acquiring the standard charge / discharge quality signal of the battery pack under test and the initial internal visual image of the battery pack, the battery pack charge / discharge electrical characteristic data and the original visual data of the internal structure are simultaneously collected, providing data support for subsequent charge / discharge fault determination and visual feature extraction. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a method for identifying internal faults in a battery pack, as provided in an embodiment of the present invention; Figure 2 An architecture diagram of a preset improved YOLOv8 model provided in an embodiment of the present invention; Figure 3 This is an architecture diagram of a battery pack internal fault identification system provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0029] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0034] See Figure 1To avoid incomplete internal testing of the battery pack, which could lead to missed or misdiagnosed faults and delayed early warnings, this embodiment provides a method for comprehensive internal fault identification of the battery pack. The flowchart of this method can be found in [link to flowchart]. Figure 1 ,include: Step S1: Acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test; Step S2: Based on the internal detection area of ​​the battery pack to be tested and the spatial distribution of preset detection points, extract the temperature difference and three-dimensional hot spot morphology map of the internal detection area; Step S3: Based on the preset improved YOLOv8 model, perform feature extraction and multi-scale feature fusion on the initial visual image inside the battery pack to obtain the first visual feature map, the second visual feature map and the third visual feature map, and perform feature mapping transformation on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data; Step S4: Based on the preset fault diagnosis model, perform multi-dimensional identification of standard impedance signal, signal residual value, standard charge and discharge quality signal, internal detection area temperature difference, three-dimensional hot spot morphology map and visual feature data to obtain the internal fault results of the battery pack.

[0035] In this embodiment, an internal fault identification device is provided, including a main control unit, an impedance detection unit, a charge / discharge quality detection unit, a thermal detection unit, and a visual inspection unit. Before identifying internal faults in the battery pack, installation and debugging are required: When installing the impedance detection unit, the oxide layer of the aluminum busbar must be avoided. If there are oxide marks on the surface, lightly sand them with fine sandpaper before clamping the high-voltage insulated alligator clips onto the positive and negative aluminum busbars, ensuring that the jaws are in full contact with the aluminum busbars. The signal cable is passed through a waterproof connector into the equipment cabinet and connected to the corresponding interface of the main control unit. When installing the charge / discharge quality detection unit, the opening and closing current clamp is slowly opened and closed, and the main circuit cable of the battery pack is inserted, ensuring that the current clamp completely wraps the cable without deviation. Ordinary alligator clips are used for clamping. Holding the positive and negative aluminum busbars, maintain a certain distance from the alligator clips of the impedance unit to avoid signal crosstalk. Communicate with the main control unit via the CAN bus, with a matching terminating resistor connected to the end of the bus. When installing the thermal detection unit and the vision detection unit, fix the robotic arm base to a flat structure outside the battery pack inspection port with bolts to ensure a firm and stable installation. Adjust the robotic arm's motion trajectory using an independent controller, drive the dual probes at the end of the arm to extend into the module, adjust the trajectory parameters to ensure the probes do not touch the battery cells, and save the detection trajectory adapted to the current battery pack. After all units are installed, close the power supply, and the main control unit automatically sends self-test commands to other units to check the communication status and data acquisition function of each unit. After confirming there are no abnormalities, it enters the ready state.

[0036] The impedance detection unit, charge / discharge quality detection unit, thermal detection unit, and vision detection unit are connected to the main control unit via CAN bus or Ethernet for data interaction and command transmission. The main control unit includes an MCU, a storage subunit, and a communication subunit. The MCU uses a high-performance processor supporting multi-channel synchronous data processing. The storage subunit uses an SDRAM+FLASH combination for high-speed data caching and long-term storage. The communication subunit supports RS-232, CAN bus, and RJ45 Ethernet interfaces to meet data transmission needs in different scenarios. The impedance detection unit acquires the standard impedance signal and signal residual value of the battery pack under test, the charge / discharge quality detection unit acquires the standard charge / discharge quality signal, and the vision detection unit acquires the initial internal visual image of the battery pack. This allows for the extraction of electrical and visual characteristic parameters of the battery pack, providing data support for determining electrical and structural faults within the battery pack. The vision detection unit includes a flexible vision probe subunit and an image transmission subunit. The probe diameter is less than or equal to 6mm, with external wall insulation (insulation level greater than or equal to 1500V), a focal length of 3-20cm, and a built-in LED supplementary light. The thermal detection unit and the visual detection unit share a robotic arm. The end of the robotic arm integrates an infrared probe subunit and a flexible visual probe subunit (15mm apart to avoid detection interference). The other end is connected to an independent controller, which can preset the motion trajectory to adapt to the inspection port of different sizes of battery packs.

[0037] Next, a thermal detection unit is used to extract the temperature difference and three-dimensional hot spot morphology map of the internal detection area of ​​the battery pack under test by using the spatial distribution of the internal detection area and preset detection points. This allows for visualization of the internal temperature distribution characteristics of the battery pack, capturing the location, range, and morphological characteristics of abnormal internal temperatures, and providing an intuitive basis for determining thermal characteristic faults occurring inside the battery pack.

[0038] Then, by using a preset improved YOLOv8 model to extract features and fuse multi-scale features from the initial internal visual image of the battery pack to be tested, the internal structural features of the battery pack can be extracted and fused, enhancing the edge, texture and multi-scale semantic information expression of the fault structure. The first visual feature map, the second visual feature map and the third visual feature map are obtained and feature mapping transformation is performed to obtain visual feature data, providing highly recognizable visual feature data for the determination of structural faults occurring inside the battery pack.

[0039] Finally, a pre-set fault diagnosis model is used to perform multi-dimensional identification of standard impedance signals, signal residual values, standard charge / discharge quality signals, internal detection area temperature differences, three-dimensional hotspot morphology maps, and visual feature data. First, the abnormality level of the feature data is determined by category (minor abnormality S, moderate abnormality M, severe abnormality H). Then, based on the built-in abnormality level determination rule library and Table 1, the pre-processed four types of feature data can be directly received, and the level determination results and risk probabilities are output. This allows for the determination of abnormality levels for impedance, charge / discharge quality, temperature, and visual features, identification of abnormal combination types, and avoidance of internal battery pack inspections. To address the issues of incomplete coverage leading to missed or misjudged faults and delayed early warnings within the battery pack, this system comprehensively covers all types of faults within the battery pack. It matches relative risk coefficients and calculates fault probability values: Fault Probability Value = Relative Risk Coefficient × 100%. Based on the fault probability value, it triggers corresponding early warning levels, obtains the internal fault results of the battery pack, and achieves fault identification. Ultimately, it triggers precise early warnings and intervention measures: a yellow warning for 30%-50% faults, an orange warning for 50%-80% faults, and a red warning for >80% faults. Warning signals are output through the communication subunit of the main control unit. The preset fault diagnosis model, trained and optimized using 400 sets of lithium-ion battery pack fault experimental data, is embedded in the main control module, possessing multi-module combined fault identification capabilities, with a call response time of less than or equal to 50ms.

[0040] The anomaly levels of the various feature data are classified as follows: Slight anomaly (S): The signal residual value reaches 80%-95% of the preset residual threshold; the standard charge and discharge quality signal has an unstable power peak in the high frequency band; the temperature of the detection point in the three-dimensional hot spot morphology map is 60-69℃; and there is a slight offset of the wiring terminal in the visual feature data. Moderate anomaly (M): The signal residual value exceeds the preset residual threshold, the standard impedance signal has the risk of overcharging, the standard charge and discharge quality signal forms a stable power band, the peak fluctuation is less than or equal to ±5%, the temperature of the detection point in the three-dimensional hot spot morphology map is 70-79℃, and there are slight bulges or a small amount of electrolyte residue in the visual feature data. Severe anomaly (H): The slope of the standard impedance signal changes from negative to positive, indicating a precursor to thermal runaway; the standard charge / discharge quality signal is in a stable power band and the current and voltage waveform distortion of the standard charge / discharge quality signal exceeds 10%; the temperature of the detection point in the three-dimensional hotspot morphology diagram is greater than or equal to 80℃; and there are obvious bulges or a large amount of electrolyte residue in the visual feature data.

[0041] Table 1. Relative Risk Coefficients of Abnormal Combinations The specific early warning and intervention measures are as follows: Yellow warning: The main control unit pushes a special verification instruction to the host computer, clarifies the abnormal data type and strongly correlated monitoring unit, and prompts staff to conduct on-site verification within 24 hours, record the abnormal change trend, and temporarily suspend active intervention; Orange Alert: Send instructions to the Battery Management System (BMS) to reduce the charging / discharging current to 60% of the rated value; start the module-level cooling fan and increase its speed to 120% of the rated value; increase the coolant flow rate of the liquid-cooled module to 110% of the rated value; push abnormal data combination charts to guide on-site personnel in troubleshooting, and record abnormal data every hour until the probability of abnormality is less than 50%. Red Alert: The main circuit of the battery pack is cut off within 100ms, triggering the main contactor to disconnect; the liquid cooling module starts the emergency cooling mode, and the coolant flow rate is increased to 150% of the rated value; the air cooling module starts the backup fan; an audible and visual alarm is triggered, and abnormal data combinations and location information are pushed to the safety management platform to prompt personnel to evacuate to a safe area; after the abnormality is resolved, the battery cell voltage and internal resistance are manually checked and no abnormalities are found before restarting.

[0042] Further, acquiring the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test includes: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

[0043] In this embodiment, the impedance detection unit includes an excitation source subunit, a data acquisition circuit subunit, a DC blocking subunit, and an impedance calculation subunit. The excitation source subunit outputs a 40Hz-500Hz sinusoidal excitation signal. The data acquisition circuit subunit is a four-wire unit employing a high common-mode rejection ratio differential operational amplifier paired with a switched-capacitor filter, providing a variable center frequency and extremely narrow bandwidth to effectively suppress interference. The excitation source subunit features a linear amplification structure, outputting a sinusoidal wave with low harmonic content and incorporating DDS waveform generation technology for high frequency stability. The data acquisition terminals of the data acquisition circuit subunit are two high-voltage resistant insulated alligator clips, requiring no welding or bolt fixing, and can directly clamp the positive and negative aluminum busbars of the battery pack, adapting to the installation requirements of aluminum busbars of different thicknesses. It also possesses excellent insulation performance, meeting the safety requirements of high-voltage detection scenarios.

[0044] The acquisition circuit subunit is equipped with a high common-mode rejection ratio differential operational amplifier and an ultra-narrow bandwidth switched capacitor filter. The high-voltage resistant insulated alligator clips in the acquisition circuit subunit acquire the original impedance signal of the battery pack under test, which contains 50Hz power frequency interference, electromagnetic noise and phase shift. The DC blocking subunit performs a 5th order Butterworth low-pass filter (cutoff frequency 10Hz) to filter out power frequency interference and electromagnetic noise in the original impedance signal. At the same time, it filters out the DC component of the battery pack, retains only the AC response signal, eliminates the influence of non-characteristic interference on the impedance detection accuracy, and obtains a purified impedance signal.

[0045] Next, an impedance calculation subunit is employed, integrating a high-speed ADC converter and a digital signal processing chip. Through a preset exponential fitting model, it performs feature fitting and residual quantization on the purified impedance signal, obtaining a fitted impedance signal and signal residual values ​​that reflect the battery pack's impedance characteristics. These values ​​are used as the basis for subsequently determining whether the impedance signal is abnormal. The measurement range of the impedance calculation subunit is... Measurement accuracy reaches Level. The residual quantization process employs exponential fitting residual analysis, optimizing preset exponential fitting model parameters (such as fitting coefficients and attenuation coefficients) by minimizing the sum of squared residuals to obtain the optimized exponential fitting model and signal residual values.

[0046] Then, by performing phase correction on the fitted impedance signal using the least squares method to fit the phase shift (error ≤ 0.5°), the phase shift error of the fitted impedance signal can be corrected, the calculation accuracy can be improved, and the calculation accuracy of both the real and imaginary parts of the impedance can be less than or equal to 0.2m. This yields a standard impedance signal that can be directly used for analyzing the electrical characteristics of the battery pack.

[0047] Finally, by acquiring the standard charge and discharge quality signals of the battery pack under test and the initial internal visual image of the battery pack, the electrical characteristic data of the battery pack charge and discharge and the original visual data of the internal structure were collected simultaneously, which can provide data support for subsequent charge and discharge fault determination and visual feature extraction.

[0048] Further, acquiring the standard charge / discharge quality signal of the battery pack under test and the initial internal visual image of the battery pack under test includes: The initial charge-discharge quality signal and the initial internal visual image of the battery pack under test are acquired, and the initial charge-discharge quality signal is de-DC processed to obtain a purified charge-discharge quality signal. Perform continuous wavelet transform on the purified charge and discharge quality signal to extract high-frequency feature data; The high-frequency feature data is reconstructed to obtain a standard charge / discharge quality signal.

[0049] In this embodiment, a hinged current clamp and two ordinary alligator clips are configured. The current clamp passes through the main circuit cable of the battery pack through its hinged structure, allowing for installation without cutting the cable and avoiding damage to the original circuitry of the battery pack. The alligator clips are used to collect voltage signals, working in conjunction with the current clamp to obtain the complete electrical parameter data required for power spectrum analysis. The charge / discharge quality detection unit includes a voltage sensor subunit, a current sensor subunit, a current transformer subunit, and a signal conditioning subunit. The voltage sensor subunit has a maximum measurement range of 0-1000V, the current sensor subunit has a maximum measurement range of 0-600A, a sampling rate greater than or equal to 1ksps, a voltage detection accuracy less than or equal to 1mV, a current detection accuracy less than or equal to 1mA, and a high-frequency component detection error less than or equal to 5mV / 5mA.

[0050] The current transformer subunit conforms to the principle of electromagnetic induction, with an accuracy of ±0.1%, a temperature drift of less than or equal to 10ppm / ℃, and strong anti-electromagnetic interference capability. The signal conditioning subunit includes a filter circuit, a high-speed AD converter, and a high-frequency component extractor, capable of signal denoising, amplification, and high-frequency feature extraction. The voltage detection accuracy is less than or equal to 1mV, the current detection accuracy is less than or equal to 1mA, and the high-frequency component detection error is less than or equal to 5mV / 5mA. The initial charge and discharge quality signal of the battery pack under test, including the fundamental DC bias of the battery pack, is obtained through the voltage sensor subunit, the current sensor subunit, and the current transformer subunit. The initial internal visual image of the battery pack is obtained through the visual inspection unit. The original charge and discharge data and the original visual data of the internal structure of the battery pack are collected simultaneously. The signal conditioning subunit uses the first-order derivative wavelet transform method to perform DC removal processing on the initial charge and discharge quality signal to eliminate the fundamental DC bias interference in the initial charge and discharge quality signal, obtaining a purified charge and discharge quality signal, laying the data foundation for subsequent high-frequency feature extraction.

[0051] Next, the signal conditioning subunit selects the db4 wavelet basis function to perform a three-level continuous wavelet transform on the purified charge and discharge quality signal. This enables the accurate extraction and calculation of characteristic coefficients of high-frequency components within 1kHz in the purified charge and discharge quality signal, thereby obtaining high-frequency characteristic data and providing characteristic basis for judging charge and discharge quality anomalies.

[0052] Finally, by reconstructing the high-frequency feature data based on the waverec function, the key high-frequency abnormal features of the unstable power peak lasting 10-30ms in the charge and discharge signal can be retained, resulting in a standard charge and discharge quality signal that can be directly used for battery pack charge and discharge quality fault identification, thus improving the accuracy and effectiveness of charge and discharge quality detection.

[0053] Furthermore, the step of extracting the temperature difference and three-dimensional hotspot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack under test includes: Based on the internal detection area of ​​the battery pack to be tested, the temperature values ​​of several detection points are obtained, and the temperature difference of the internal detection area is calculated based on the temperature values ​​of the several detection points. Based on the spatial distribution of preset detection points and the temperature difference in the internal detection area, a three-dimensional hotspot morphology map is extracted.

[0054] In this embodiment, the thermal detection unit includes an infrared probe subunit and a signal conversion subunit. The infrared probe subunit employs VOx thin-film sensing technology, with a resolution of 256×192, a temperature measurement range of -20℃ to 350℃, and an accuracy of ±2℃. A robotic arm carrying a flexible probe extends through the inspection port into the internal detection area of ​​the battery pack under test via the infrared probe subunit to obtain temperature values ​​at several detection points on the surface of the battery cells. The temperature difference within the internal detection area is calculated, providing temperature difference data that reflects the temperature distribution differences within the battery pack, thus providing temperature characteristic data for hotspot identification.

[0055] Next, combining the encrypted acquisition frequency during the charging and discharging phase with the spatial distribution of preset detection points and the temperature difference within the internal detection area, and through blackbody furnace zoning via a signal conversion subunit, the infrared signal is converted into a digital signal. This allows for the fusion mapping of temperature values ​​and spatial location information, and the extracted three-dimensional hotspot morphology map can intuitively present the spatial outline, distribution range, and concentration of hotspots within the battery pack, providing visualized thermal characteristic data for determining thermal anomalies. The spatial distribution of the preset detection points can be layered according to the cell body, tabs, and heat dissipation channels.

[0056] The specific state correspondence of the three-dimensional hotspot pattern diagram is as follows: Normal and uniform heating state: No obvious hot spots, temperature difference in internal detection area is less than or equal to 3℃, and the outline of the three-dimensional hot spot morphology map has no local protrusions and is uniformly distributed. Low hot spot consistency deviation: In the 3D hot spot morphology map, there are 1-3 local adjacent detection points with higher temperatures. The hot spot outline is scattered / small clusters, with a distribution range of 0.3-1cm². The temperature difference between the detection point with higher temperature and the surrounding area is 3-8℃. Aging / degradation cell heating status: The temperature of the detection point corresponding to a single cell is relatively high. The hot spot outline of the three-dimensional hot spot morphology map is diffuse within the range of a single cell, with a distribution range of 0.5-2cm² and blurred boundaries. Structural fault-induced concentrated heat generation: The temperature of the detection point corresponding to the tab / connector suddenly increases, the hot spot outline in the three-dimensional hot spot morphology diagram is dotted / lined, the distribution range is ≥0.5cm², and the temperature difference between the detection point with high temperature and the surrounding area is ≥8℃.

[0057] Further, the preset improved YOLOv8 model includes a backbone network unit, a neck network unit, and a head network unit; the feature extraction and multi-scale feature fusion of the initial battery pack internal visual image based on the preset improved YOLOv8 model to obtain a first visual feature map, a second visual feature map, and a third visual feature map, and feature mapping transformation of the first visual feature map, the second visual feature map, and the third visual feature map to obtain visual feature data, including: The initial internal visual image of the battery pack is corrected and filtered to obtain a standard internal visual image of the battery pack. Based on the backbone network unit, feature extraction is performed on the visual image inside the standard battery pack to obtain the initial bottom layer feature map, the initial middle layer feature map and the initial shallow layer feature map. Based on the neck network unit, multi-scale feature fusion is performed on the initial low-level feature map, the initial middle-level feature map and the initial shallow-level feature map to obtain the first visual feature map, the second visual feature map and the third visual feature map. Based on the head network unit, feature mapping transformation is performed on the first visual feature map, the second visual feature map, and the third visual feature map to obtain visual feature data.

[0058] In this embodiment, the visual detection unit employs a fisheye distortion correction algorithm. Through calibration experiments, intrinsic parameters, distortion parameters, and projection matrices are obtained to correct the initial visual image of the battery pack's interior. Then, a 3×3 median filter is performed to eliminate image distortion and noise interference, restoring the true visual information of the battery pack's internal structure and providing a clear standard visual image of the battery pack's interior for subsequent feature extraction. Next, the backbone network unit extracts features from the standard battery pack's interior visual image. This enables multi-scale basic feature mining of the standard visual image, extracting deep semantic features, mid-level structural features, and shallow detail features of the battery pack's interior structure layer by layer, obtaining initial bottom-level feature maps, initial mid-level feature maps, and initial shallow feature maps, laying the foundation for multi-scale feature fusion.

[0059] Then, a dynamic upsampling operator and a general inverted bottleneck search block are introduced into the neck network unit. Multi-scale feature fusion is performed on the initial low-level feature map, initial mid-level feature map, and initial shallow feature map through the neck network unit. This enables deep fusion and enhancement of basic features at different scales, compensating for the information deficiencies of single-scale features and strengthening the multi-scale semantic and detailed representation capabilities of the internal fault structure of the battery pack, obtaining first, second, and third visual feature maps. Finally, feature mapping transformation is performed on the first, second, and third visual feature maps through the head network unit. This transforms the fused multi-scale visual feature maps into visual feature data suitable for fault identification, completing the quantification and reconstruction of visual features and providing a highly discriminative basis for subsequent multi-dimensional identification.

[0060] Further, the neck network unit includes a first neck branch subunit, a second neck branch subunit, and a third neck branch subunit; the multi-scale feature fusion based on the neck network unit of the initial low-level feature map, the initial middle-level feature map, and the initial shallow-level feature map to obtain a first visual feature map, a second visual feature map, and a third visual feature map includes: Based on the first neck branch subunit, the initial bottom layer feature map, the initial middle layer feature map, and the initial shallow layer feature map are enhanced and fused to obtain the first enhanced feature map and the second enhanced feature map. Based on the second neck branch subunit, the first enhanced feature map, the second enhanced feature map and the initial low-level feature map are convolutionally fused to obtain the third enhanced feature map and the fourth enhanced feature map; Based on the third neck branch subunit, the second enhanced feature map, the third enhanced feature map and the fourth enhanced feature map are sequentially convolutionally expanded and spatially reconstructed to obtain the first initial visual feature map, the second initial visual feature map and the third initial visual feature map; Based on the third neck branch subunit, channel compression and feature reshaping are performed on the first initial visual feature map, the second initial visual feature map, and the third initial visual feature map to obtain the first visual feature map, the second visual feature map, and the third visual feature map.

[0061] In this embodiment, as Figure 2 As shown, Figure 2 As shown, `up` is the dynamic upsampling operator, `Concat` is the channel-dimensional concatenation operation block, `C2f` is the feature enhancement block, `Conv` is the convolution operator, `UIB` is the inverted bottleneck search block, and `Detect` is the detection head. By enhancing and fusing the initial low-level feature map, initial mid-level feature map, and initial shallow feature map through the first neck branch subunit, features at different scales can be fused and enhanced, integrating deep semantics, mid-level structure, and shallow detail information to compensate for the information loss of single-scale features, obtaining the first and second enhanced feature maps, laying the foundation for subsequent multi-scale convolutional fusion. Next, the second neck branch subunit performs convolutional transformation and secondary fusion on the first enhanced feature map, second enhanced feature map, and initial low-level feature map to further strengthen the spatial correlation and semantic expressive ability of the features, optimize the fusion effect of multi-scale features, and obtain the third and fourth enhanced feature maps.

[0062] Then, an inverted bottleneck search block (UIB) is introduced in the first neck branch subunit. The second, third, and fourth enhanced feature maps are sequentially convolved and spatially reconstructed through the third neck branch subunit. This enables channel dimension expansion of the fused features, reconstruction and enhancement of spatial texture and boundary information, improvement of feature expression dimension, enhancement of spatial feature details of the internal fault structure of the battery pack, highlighting of key fault feature information, and acquisition of the first, second, and third initial visual feature maps.

[0063] Subsequently, the third neck branch subunit performs pointwise convolutional channel expansion operations on the first, second, and third initial visual feature maps using the inverted bottleneck search block (UIB) to enhance the feature representation dimension. Then, the inverted bottleneck search block (UIB) employs depthwise separable convolution to reconstruct the expanded feature maps spatially, enhancing the ability to express internal structural texture and boundary information. Following this, channel compression and feature reshaping are performed, reducing feature computation complexity while preserving core fault feature information. Finally, when the input and output feature dimensions are consistent, the original structural information is preserved through residual connections using the detection head (Detect), improving feature stability and yielding the first, second, and third visual feature maps adapted for subsequent feature mapping transformations. The geometric errors of the first, second, and third visual feature maps obtained after feature mapping transformation are all less than or equal to 0.5%, preserving key structural features such as cell bulging and loose screws, with a single-frame inference time of less than or equal to 15ms.

[0064] Further, the enhancement and fusion of the initial low-level feature map, the initial middle-level feature map, and the initial shallow-level feature map based on the first neck branch sub-unit to obtain the first enhanced feature map and the second enhanced feature map includes: The initial low-level feature map is dynamically upsampled to obtain the reconstructed low-level feature map; The reconstructed bottom-layer feature map and the initial middle-layer feature map are spliced ​​and fused to obtain the first fused feature map; Perform feature enhancement on the first fused feature map to obtain the first enhanced feature map; The first enhanced feature map is dynamically upsampled to obtain the reconstructed enhanced feature map; The reconstructed enhanced feature map and the initial shallow feature map are spliced ​​and fused to obtain a second fused feature map; The second fused feature map is enhanced to obtain the second enhanced feature map.

[0065] In this embodiment, as Figure 2As shown, a dynamic upsampling operator `up` is introduced in the first neck branch subunit. By dynamically upsampling the initial low-level feature map, the resolution of the initial low-level feature map can be adaptively restored, reconstructing the edge and detail information of the internal structure of the battery pack. This compensates for the low resolution and lack of detail in the initial low-level feature map, resulting in a reconstructed low-level feature map, laying the foundation for cross-scale feature fusion. Next, the reconstructed low-level feature map and the initial mid-level feature map are concatenated and fused using the channel-dimensional concatenation operation block `Concat`. This integrates the high semantic features of the low-level map with the structural features of the mid-level map along the channel dimension, fusing feature information at different scales. This allows the feature map to simultaneously possess semantic expression and structural awareness capabilities, obtaining the first fused feature map.

[0066] Then, feature enhancement is performed on the first fused feature map using the feature enhancement block C2f. This enables deep extraction and enhancement of the cross-scale fused features, improving their representational ability and obtaining the first enhanced feature map, which highlights key feature information related to internal faults in the battery pack. Subsequently, the first enhanced feature map is dynamically upsampled using the dynamic upsampling operator up, which adaptively increases the feature map resolution again, further supplementing the feature detail information. This allows the fused features to better match the scale and detail features of the initial shallow feature map in subsequent steps, resulting in the reconstructed enhanced feature map.

[0067] Next, the reconstructed enhanced feature map and the initial shallow feature map are concatenated and fused using the channel-dimensional concatenation block Concat. This integrates the enhanced features that fuse low-level and mid-level features with shallow detail features, fully fusing the semantic and fine-grained detail information of multi-scale features, enriching the information dimensions of the features, and obtaining the second fused feature map. Finally, the second fused feature map is enhanced using the feature enhancement block C2f, which further improves the discriminativeness and stability of the features, resulting in a second enhanced feature map that can be directly used for subsequent convolutional fusion.

[0068] Further, the step of convolutionally fusing the first enhanced feature map, the second enhanced feature map, and the initial low-level feature map based on the second neck branch subunit to obtain the third enhanced feature map and the fourth enhanced feature map includes: Convolutional downsampling is performed on the second enhanced feature map to obtain the first convolutional feature map; The first convolutional feature map and the first enhanced feature map are concatenated and fused to obtain the third fused feature map; Perform feature enhancement on the third fused feature map to obtain the third enhanced feature map; Convolutional downsampling is performed on the third enhanced feature map to obtain the second convolutional feature map; The second convolutional feature map and the initial low-level feature map are concatenated and fused to obtain the fourth fused feature map; Feature enhancement is performed on the fourth fused feature map to obtain the fourth enhanced feature map.

[0069] In this embodiment, the second enhanced feature map is downsampled using the convolution operator Conv, which compresses its scale and adjusts its channel dimensions to obtain the first convolutional feature map. Next, the first convolutional feature map and the first enhanced feature map are concatenated and fused using the channel-dimensional concatenation block Concat. This integrates the abstract semantic features of the second enhanced feature map with the fused feature information of the first enhanced feature map along the channel dimension, further strengthening the correlation and complementarity of multi-scale features and enriching the information dimension of feature expression to obtain the third fused feature map. Then, the third fused feature map is enhanced using the feature enhancement block C2f, which improves the feature representation capability, highlights key feature information related to internal battery pack faults, and filters out redundant feature interference to obtain the third enhanced feature map.

[0070] Subsequently, the third enhanced feature map is downsampled using the Conv convolution operator, which further compresses and abstracts the feature map, focusing on core semantic features to obtain the second convolutional feature map. Next, the second convolutional feature map and the initial bottom-level feature map are concatenated and fused using the channel-dimensional concatenation block Concat to obtain the fourth fused feature map. This integrates the abstract features enhanced by multiple fusions with the core semantic features of the initial bottom layer, allowing the feature map to retain both the original semantic information of the bottom layer and the enhanced feature information from multi-scale fusion, thus improving feature integrity. Finally, the fourth fused feature map is enhanced using the feature enhancement block C2f, further improving the stability and discriminability of the features, resulting in a fourth enhanced feature map that can be directly used for subsequent spatial feature reconstruction.

[0071] This embodiment provides a battery pack internal fault identification system, such as Figure 3 As shown, it includes an electrical vision data acquisition module, a thermal feature data extraction module, a visual feature data acquisition module, and a multi-dimensional fault identification module, specifically: The electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test. The thermal feature data extraction module is used to extract the temperature difference and three-dimensional hot spot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack to be tested. The visual feature data acquisition module is used to extract features and fuse multi-scale features from the initial internal visual image of the battery pack based on a preset improved YOLOv8 model, to obtain a first visual feature map, a second visual feature map and a third visual feature map, and to perform feature mapping transformation on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data. The multi-dimensional fault identification module is used to perform multi-dimensional identification of standard impedance signal, signal residual value, standard charge and discharge quality signal, internal detection area temperature difference, three-dimensional hot spot morphology map and visual feature data based on a preset fault diagnosis model, so as to obtain the internal fault results of the battery pack.

[0072] This embodiment provides a battery pack internal fault identification system. In practical applications, it only requires an electrical vision data acquisition module. By acquiring the standard impedance signal, signal residual value, standard charge / discharge quality signal, and initial internal visual image of the battery pack under test, it can extract the electrical and visual characteristic parameters of the battery pack, providing data support for determining electrical and structural faults occurring inside the battery pack. Next, a thermal feature data extraction module is used to extract the temperature difference and three-dimensional hotspot morphology map of the internal detection area based on the spatial distribution of the internal detection area and preset detection points of the battery pack under test. This visualizes the internal temperature distribution characteristics of the battery pack, capturing the location, range, and morphological characteristics of abnormal temperatures inside the battery pack, providing intuitive evidence for determining thermal characteristic faults occurring inside the battery pack.

[0073] Then, a visual feature data acquisition module is employed. Using a pre-set improved YOLOv8 model, feature extraction and multi-scale feature fusion are performed on the initial internal visual image of the battery pack under test. This allows for the extraction and fusion of internal structural features, enhancing the expression of edge, texture, and multi-scale semantic information of faulty structures. First, second, and third visual feature maps are obtained and their features are mapped and transformed to acquire visual feature data. This provides highly recognizable visual feature data for identifying structural faults occurring within the battery pack. Finally, a multi-dimensional fault identification module is used. Through a pre-set fault diagnosis model, standard impedance signals, signal residual values, standard charge / discharge quality signals, internal detection area temperature differences, 3D hotspot morphology maps, and visual feature data are used for multi-dimensional identification. This avoids incomplete internal detection of the battery pack, preventing missed or misjudged faults and delayed early warnings. It comprehensively covers various fault types within the battery pack, acquiring fault results and enabling the identification of internal faults.

[0074] Furthermore, the electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test, including: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

[0075] In this embodiment, the impedance detection unit includes an excitation source subunit, a data acquisition circuit subunit, a DC blocking subunit, and an impedance calculation subunit. The excitation source subunit outputs a 40Hz-500Hz sinusoidal excitation signal. The data acquisition circuit subunit is a four-wire unit employing a high common-mode rejection ratio differential operational amplifier paired with a switched-capacitor filter, providing a variable center frequency and extremely narrow bandwidth to effectively suppress interference. The excitation source subunit features a linear amplification structure, outputting a sinusoidal wave with low harmonic content and incorporating DDS waveform generation technology for high frequency stability. The data acquisition terminals of the data acquisition circuit subunit are two high-voltage resistant insulated alligator clips, requiring no welding or bolt fixing, and can directly clamp the positive and negative aluminum busbars of the battery pack, adapting to the installation requirements of aluminum busbars of different thicknesses. It also possesses excellent insulation performance, meeting the safety requirements of high-voltage detection scenarios.

[0076] The acquisition circuit subunit is equipped with a high common-mode rejection ratio differential operational amplifier and an ultra-narrow bandwidth switched capacitor filter. The high-voltage resistant insulated alligator clips in the acquisition circuit subunit acquire the original impedance signal of the battery pack under test, which contains 50Hz power frequency interference, electromagnetic noise and phase shift. The DC blocking subunit performs a 5th order Butterworth low-pass filter (cutoff frequency 10Hz) to filter out power frequency interference and electromagnetic noise in the original impedance signal. At the same time, it filters out the DC component of the battery pack, retains only the AC response signal, eliminates the influence of non-characteristic interference on the impedance detection accuracy, and obtains a purified impedance signal.

[0077] Next, an impedance calculation subunit is employed, integrating a high-speed ADC converter and a digital signal processing chip. This subunit performs feature fitting and residual quantization on the purified impedance signal using a preset exponential fitting model. This yields a fitted impedance signal and signal residual values ​​that reflect the battery pack's impedance characteristics, which are then used as the basis for determining whether the impedance signal is abnormal. The impedance calculation subunit has a measurement range of 100mΩ-300mΩ and a measurement accuracy at the mΩ level. The residual quantization process employs exponential fitting residual analysis. By minimizing the sum of squared residuals, the preset exponential fitting model parameters (such as the fitting coefficient and attenuation coefficient) are optimized to obtain the optimized exponential fitting model and signal residual values.

[0078] Then, by performing phase correction on the fitted impedance signal using the least squares method to fit the phase shift (error ≤ 0.5°), the phase shift error of the fitted impedance signal can be corrected, improving the calculation accuracy and ensuring that the calculation accuracy of both the real and imaginary parts of the impedance is less than or equal to 0.2mΩ, thus obtaining a standard impedance signal that can be directly used for the analysis of the battery pack's electrical characteristics. Finally, by acquiring the standard charge-discharge quality signal of the battery pack under test and the initial internal visual image of the battery pack, the battery pack's charge-discharge electrical characteristic data and the original visual data of its internal structure are simultaneously collected, providing data support for subsequent charge-discharge fault determination and visual feature extraction.

[0079] To more intuitively and fully illustrate how the battery pack internal fault identification method and system provided in this embodiment can avoid the problems of missed or misjudged internal faults and delayed early warnings caused by incomplete internal battery pack detection, and to achieve the identification of internal battery pack faults, the following two embodiments are provided for specific explanation: When only one unit has an anomaly level of H, Table 1 shows a relative risk coefficient of 0.9 and a failure probability of 90%, triggering a red alert. At this time, the main circuit of the battery pack is immediately cut off, the liquid cooling module starts emergency cooling mode, an audible and visual alarm is triggered, and the abnormal data combination and location information are pushed to the safety management platform, prompting personnel to evacuate to a safe area. After the anomaly is cleared, manual inspection reveals that the current transformer shield of the charging and discharging unit is damaged. After replacement, the device is restarted. It can only be restarted after 24 hours without any abnormalities.

[0080] When one unit has an anomaly level of H and another unit has an anomaly level of S, Table 1 shows a relative risk coefficient of 0.92 and a failure probability of 92%, triggering a red alert. At this time, the main circuit of the battery pack is immediately disconnected, the liquid cooling module starts emergency cooling mode, an audible and visual alarm is triggered, and the abnormal data combination and location information are pushed to the safety management platform, prompting personnel to evacuate to a safe area. After the anomaly is cleared, the staff tightens the main circuit terminals, and the temperature returns to normal. However, if two units still have an anomaly level of S, Table 1 shows a relative risk coefficient of 0.55 and a failure probability of 55%, triggering an orange alert. Current reduction operation is initiated, and after maintaining current reduction operation for a period of time without any abnormalities, normal current is restored.

[0081] Through the above embodiments, the battery pack internal fault identification method and system provided in this embodiment can comprehensively capture various fault characteristics inside the battery pack, realize accurate early fault identification and graded early warning, effectively avoid thermal runaway accidents, and ensure the safe and stable operation of the battery pack.

[0082] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the functions of the system as described above.

[0083] It is understood that the above system embodiments correspond to the method embodiments of the present invention, and can implement the battery pack internal fault identification method provided by any of the above method embodiments of the present invention.

[0084] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0085] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying internal faults in a battery pack, characterized in that, include: Acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test; Based on the internal detection area and the spatial distribution of preset detection points of the battery pack to be tested, the temperature difference and three-dimensional hot spot morphology map of the internal detection area are extracted. Based on the preset improved YOLOv8 model, feature extraction and multi-scale feature fusion are performed on the initial visual image inside the battery pack to obtain the first visual feature map, the second visual feature map and the third visual feature map. Then, feature mapping transformation is performed on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data. Based on a preset fault diagnosis model, the system performs multi-dimensional identification of standard impedance signals, signal residual values, standard charge and discharge quality signals, internal detection area temperature differences, three-dimensional hotspot morphology maps, and visual feature data to obtain internal fault results of the battery pack.

2. The method for identifying internal faults in a battery pack according to claim 1, characterized in that, The acquisition of the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test includes: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

3. The method for identifying internal faults in a battery pack according to claim 2, characterized in that, The acquisition of the standard charge / discharge quality signal and the initial internal visual image of the battery pack under test includes: The initial charge-discharge quality signal and the initial internal visual image of the battery pack under test are acquired, and the initial charge-discharge quality signal is de-DC processed to obtain a purified charge-discharge quality signal. Perform continuous wavelet transform on the purified charge and discharge quality signal to extract high-frequency feature data; The high-frequency feature data is reconstructed to obtain a standard charge / discharge quality signal.

4. The method for identifying internal faults in a battery pack according to claim 1, characterized in that, The step of extracting the temperature difference and three-dimensional hotspot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack under test includes: Based on the internal detection area of ​​the battery pack to be tested, several temperature values ​​at detection points are obtained, and the temperature difference of the internal detection area is calculated based on the several temperature values ​​at detection points. Based on the spatial distribution of preset detection points and the temperature difference in the internal detection area, a three-dimensional hotspot morphology map is extracted.

5. The method for identifying internal faults in a battery pack according to claim 1, characterized in that, The preset improved YOLOv8 model includes a backbone network unit, a neck network unit, and a head network unit. Based on the preset improved YOLOv8 model, feature extraction and multi-scale feature fusion are performed on the initial visual image inside the battery pack to obtain a first visual feature map, a second visual feature map, and a third visual feature map. Feature mapping transformation is then performed on the first, second, and third visual feature maps to obtain visual feature data, including: The initial internal visual image of the battery pack is corrected and filtered to obtain a standard internal visual image of the battery pack. Based on the backbone network unit, feature extraction is performed on the visual image inside the standard battery pack to obtain the initial bottom layer feature map, the initial middle layer feature map and the initial shallow layer feature map. Based on the neck network unit, multi-scale feature fusion is performed on the initial low-level feature map, the initial middle-level feature map and the initial shallow-level feature map to obtain the first visual feature map, the second visual feature map and the third visual feature map. Based on the head network unit, feature mapping transformation is performed on the first visual feature map, the second visual feature map, and the third visual feature map to obtain visual feature data.

6. The method for identifying internal faults in a battery pack according to claim 5, characterized in that, The neck network unit includes a first neck branch subunit, a second neck branch subunit, and a third neck branch subunit; the multi-scale feature fusion based on the neck network unit of the initial low-level feature map, the initial middle-level feature map, and the initial shallow-level feature map to obtain a first visual feature map, a second visual feature map, and a third visual feature map includes: Based on the first neck branch subunit, the initial bottom layer feature map, the initial middle layer feature map, and the initial shallow layer feature map are enhanced and fused to obtain the first enhanced feature map and the second enhanced feature map. Based on the second neck branch subunit, the first enhanced feature map, the second enhanced feature map and the initial low-level feature map are convolutionally fused to obtain the third enhanced feature map and the fourth enhanced feature map; Based on the third neck branch subunit, the second enhanced feature map, the third enhanced feature map and the fourth enhanced feature map are sequentially convolutionally expanded and spatially reconstructed to obtain the first initial visual feature map, the second initial visual feature map and the third initial visual feature map; Based on the third neck branch subunit, channel compression and feature reshaping are performed on the first initial visual feature map, the second initial visual feature map, and the third initial visual feature map to obtain the first visual feature map, the second visual feature map, and the third visual feature map.

7. The method for identifying internal faults in a battery pack according to claim 6, characterized in that, The enhancement and fusion of the initial low-level feature map, initial middle-level feature map, and initial shallow-level feature map based on the first neck branch subunit to obtain the first enhanced feature map and the second enhanced feature map includes: The initial low-level feature map is dynamically upsampled to obtain the reconstructed low-level feature map; The reconstructed bottom-layer feature map and the initial middle-layer feature map are spliced ​​and fused to obtain the first fused feature map; Perform feature enhancement on the first fused feature map to obtain the first enhanced feature map; The first enhanced feature map is dynamically upsampled to obtain the reconstructed enhanced feature map; The reconstructed enhanced feature map and the initial shallow feature map are spliced ​​and fused to obtain a second fused feature map; The second fused feature map is enhanced to obtain the second enhanced feature map.

8. A method for identifying internal faults in a battery pack according to claim 6, characterized in that, The step of convolutionally fusing the first enhanced feature map, the second enhanced feature map, and the initial low-level feature map based on the second neck branch subunit to obtain the third and fourth enhanced feature maps includes: Convolutional downsampling is performed on the second enhanced feature map to obtain the first convolutional feature map; The first convolutional feature map and the first enhanced feature map are concatenated and fused to obtain the third fused feature map; Perform feature enhancement on the third fused feature map to obtain the third enhanced feature map; Convolutional downsampling is performed on the third enhanced feature map to obtain the second convolutional feature map; The second convolutional feature map and the initial low-level feature map are concatenated and fused to obtain the fourth fused feature map; Feature enhancement is performed on the fourth fused feature map to obtain the fourth enhanced feature map.

9. A battery pack internal fault identification system, characterized in that, It includes an electrical vision data acquisition module, a thermal feature data extraction module, a visual feature data acquisition module, and a multi-dimensional fault identification module, specifically: The electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge and discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test. The thermal feature data extraction module is used to extract the temperature difference and three-dimensional hot spot morphology map of the internal detection area based on the internal detection area and the spatial distribution of preset detection points of the battery pack to be tested. The visual feature data acquisition module is used to extract features and fuse multi-scale features from the initial internal visual image of the battery pack based on a preset improved YOLOv8 model, to obtain a first visual feature map, a second visual feature map and a third visual feature map, and to perform feature mapping transformation on the first visual feature map, the second visual feature map and the third visual feature map to obtain visual feature data. The multi-dimensional fault identification module is used to perform multi-dimensional identification of standard impedance signal, signal residual value, standard charge and discharge quality signal, internal detection area temperature difference, three-dimensional hot spot morphology map and visual feature data based on a preset fault diagnosis model, so as to obtain the internal fault results of the battery pack.

10. A battery pack internal fault identification system according to claim 9, characterized in that, The electrical vision data acquisition module is used to acquire the standard impedance signal of the battery pack under test, the signal residual value of the battery pack under test, the standard charge / discharge quality signal of the battery pack under test, and the initial internal visual image of the battery pack under test, including: The original impedance signal of the battery pack to be tested is obtained, and the original impedance signal is low-pass filtered to obtain the purified impedance signal. The fitted impedance signal and signal residual value are obtained based on the preset exponential fitting model and the purified impedance signal. The fitted impedance signal is phase-corrected to obtain a standard impedance signal; Acquire standard charge / discharge quality signals and initial internal visual images of the battery pack under test.

Images