Leakage source location system

By combining a six-sided microphone array module with a machine learning positioning model, the problem of inaccurate location of leak holes in new energy power battery packs has been solved, enabling rapid and accurate leak detection and automated production.

CN123115703APending Publication Date: 2026-11-17SHANGHAI JIAOTONG UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611612203.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technology cannot accurately locate the leakage hole in the new energy power battery pack in three dimensions, resulting in long repair time and inconsistent test results, which cannot meet the fast cycle requirements of automated production lines.

Method used

By employing a six-sided microphone array module combined with a multi-channel synchronous acquisition unit and a sound source imaging processing unit, and through a finite element simulation database and a machine learning positioning model, regional-level coarse positioning and coordinate-level fine positioning of the leak point can be achieved.

Benefits of technology

It achieves precise three-dimensional positioning of battery pack leakage holes, shortens inspection time, improves inspection consistency and the cycle time of automated production lines, and avoids physical contact damage to the battery pack.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115703A_ABST
    Figure CN123115703A_ABST
Patent Text Reader

Abstract

This invention discloses a leakage source localization system, including a six-sided microphone array module, a multi-channel synchronous acquisition unit, and a sound source imaging processing unit. The multi-channel synchronous acquisition unit controls the six-sided microphone array module to perform synchronous sampling at the same time. The sound source imaging processing unit includes a finite element simulation database and a machine learning localization model. The finite element simulation database stores the mapping relationship between multiple grid cells and the theoretical sound field characteristics of each grid cell when it is used as a sound source. The machine learning localization model is used to compare and match the measured sound field characteristics obtained after feature extraction of the acoustic signals synchronously acquired by the six-sided microphone array with the theoretical sound field characteristics in the finite element simulation database. First, it outputs the area identifier code of the grid cell to which the leakage point belongs to achieve coarse localization at the area level. Then, it outputs the precise relative coordinates of the leakage point within the grid cell range to achieve fine localization at the coordinate level, and finally maps them to three-dimensional spatial coordinates.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of detection equipment technology, and in particular to a leak source location system. Background Technology

[0002] In the off-line inspection and repair process of new energy power battery packs, existing airtightness testing technologies suffer from the technical problem of not being able to accurately locate the spatial position of leak holes in three dimensions. Current pressure drop methods and helium mass spectrometry leak detection methods can only determine whether the overall leakage rate of the battery pack exceeds the standard, but cannot provide the precise three-dimensional coordinates of the leak holes on the six-sided surface of the battery pack. When a leak is detected, repair personnel can only rely on experience to conduct large-area inspections of the six weld seams or repeatedly apply soapy water to test for leaks, resulting in repair times of tens of minutes or even hours, severely restricting production line cycle time.

[0003] Handheld acoustic imaging devices require operators to scan and inspect each side (top, bottom, front, back, left, and right) of the battery pack being tested. A single-piece inspection typically takes 3 to 5 minutes, which cannot meet the requirements of automated production lines for fast-paced single-piece inspections. Furthermore, handheld inspection is highly dependent on the operator's experience and technique, resulting in poor consistency of inspection results between different personnel, making standardized quality control difficult.

[0004] Existing fixed single-sided microphone arrays can only acquire two-dimensional azimuth information of the sound source relative to the array plane, and cannot simultaneously calculate the distance and depth of the sound source. When the sound source is near the array normal direction, leakage points at different distances may produce almost the same time delay response, forming "cone-face positioning ambiguity", which makes it impossible to distinguish between leakage on the front panel and rear panel of the enclosure, or to determine whether the leakage point is located on the surface or inner layer of the weld.

[0005] Traditional contact-based detection methods (such as soapy water application or probe-type helium gas) require physical contact with the enclosure surface, which may leave scratches or residues on the sealing surface or coating, potentially causing secondary damage to the insulation and sealing performance of the high-voltage battery pack. Therefore, there is a need to develop equipment that can accurately detect the spatial location of leak holes in the battery pack. Summary of the Invention

[0006] The purpose of this invention is to provide a leakage source location system to solve the problems of the prior art.

[0007] To address the aforementioned technical problems, embodiments of the present invention provide a leak source location system, comprising: A six-sided microphone array module includes a top array, a bottom array, and four side arrays. The top array and the bottom array are positioned above and below the device under test, respectively, and the four side arrays are positioned around the four sides of the device under test. A multi-channel synchronous acquisition unit is electrically connected to the six-sided microphone array module, and controls the six-sided microphone array module to synchronously sample the acoustic signal generated by the leakage of the test component at the same time. The sound source imaging processing unit, electrically connected to the multi-channel synchronous acquisition unit, includes: The finite element simulation database stores the mapping relationship between multiple grid cells that discretize the space around the test piece according to a preset size and the theoretical sound field characteristics when each grid cell is used as a sound source; The machine learning localization model is used to compare and match the measured sound field features obtained by feature extraction of the acoustic signals synchronously collected by the six-sided microphone array with the theoretical sound field features of the finite element simulation database. First, it outputs the area identifier code of the grid cell to which the leakage point belongs to achieve coarse localization at the regional level. Then, it outputs the precise relative coordinates of the leakage point within the grid cell to which the leakage point belongs to achieve fine localization at the coordinate level, and finally maps it to three-dimensional spatial coordinates.

[0008] In one embodiment, the six-sided microphone array module includes: A floor, located below the test piece; the bottom surface array is embedded below the floor plane and the top surface of the bottom surface array is flush with the floor; A lifting mechanism is connected to the top surface array drive and is used to control the top surface array to rise and fall above the test piece; A horizontal guide rail is driven and connected to the four side arrays to control the movement of the four side arrays to the four sides of the test piece.

[0009] In one embodiment, the multi-channel synchronous acquisition unit includes: The main acquisition card has a built-in clock source and is electrically connected to one of the top array, the bottom array, and the four side arrays. Five acquisition cards are electrically connected to the remaining five of the top array, the bottom array, and the four side arrays; The sampling clock and trigger signal output by the main acquisition card are synchronously distributed to each of the slave acquisition cards via a clock distributor, so that the sampling time deviation of the top array, the bottom array and the four side arrays is less than 1 nanosecond.

[0010] In one embodiment, the finite element simulation database construction method includes: A three-dimensional finite element acoustic model is established based on the test device and the space between the test device and the six-sided microphone array module. The test device and the space are divided into multiple grid units, and each grid unit is assigned a unique area identification code. Based on the physical laws of sound wave propagation in air, the theoretical arrival time, theoretical sound intensity, and theoretical frequency change of the sound wave in the three-dimensional finite element acoustic model corresponding to the observation point of the six-sided microphone array module are calculated for each mesh cell under the assumption of leakage. The region identifier code of each mesh cell is associated with and stored with the corresponding theoretical sound field feature to form the finite element simulation database.

[0011] In one embodiment, the machine learning localization model includes: The region identification sub-model is used to compare and match the measured sound field features with the theoretical sound field features in the finite element simulation database, and output the region identification code of the grid cell to which the leakage point belongs; The precise positioning sub-model is used to output the precise relative coordinates of the leakage point within the grid cell corresponding to the area identification code, combining the measured sound field characteristics and the theoretical sound field characteristics.

[0012] In one embodiment, the top surface array is arranged in a sparse spiral array with a central void.

[0013] In one embodiment, the side array is arranged in a cross shape.

[0014] In one embodiment, the bottom surface array is arranged in a rectangular grid.

[0015] In one embodiment, the sound source imaging processing unit further includes a differential evolution parameter optimization module, wherein the differential evolution parameter optimization module optimizes in the following way: The hyperparameters of the machine learning localization model are used as optimization variables for the differential evolution algorithm. Multiple sets of parameter vectors are generated through initialization, and mutation, crossover, and selection operations are performed sequentially for iterative search. After each generation of parameter vector updates, the model is predicted and evaluated in both laboratory standard environment and real production line environment, and the comprehensive positioning error in the two scenarios is calculated as the fitness. When the fitness converges, the optimal parameter calibration result is output and solidified into the machine learning localization model.

[0016] In one embodiment, the sound source imaging processing unit further includes a feature extraction module, which is used to preprocess and extract features from the acoustic signal, extracting the time-domain energy statistical features, spectral distribution features, and signal correlation features between the top array, the bottom array, and the four side arrays of the acoustic signal; The spectral distribution characteristics are energy characteristics within the 20kHz to 40kHz ultrasonic frequency band. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the bottom surface array according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the top surface array according to an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of a side array according to an embodiment of the present invention.

[0020] Figure 4 This is an implementation step of a battery pack leakage source location system according to an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram of a leakage source location system according to an embodiment of the present invention.

[0022] Reference numerals: 100, Leakage source locating system; 1, Top array; 2, Bottom array; 3, Side array; 4, Test piece; 5, Lifting mechanism; 6, Laboratory; 7, Sound-absorbing structure; 8, Column. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the various embodiments of this invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this invention to facilitate a better understanding of this application. However, the technical solutions claimed in the claims of this application can be implemented even without these technical details and with various variations and modifications based on the following embodiments.

[0024] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0025] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0026] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0027] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0028] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0029] This invention relates to a leak source location system 100, such as... Figure 5 As shown, it includes a six-sided microphone array module, a multi-channel synchronous acquisition unit, and a sound source imaging processing unit.

[0030] The six-sided microphone array module includes a top array 1, a bottom array 2, and four side arrays 3 (front, rear, left, and right). The top array 1 and the bottom array 2 are positioned above and below the device under test 4, respectively, and the four side arrays 3 are arranged around the four sides of the device under test 4. The device under test 4 described below uses a battery pack as an example. It should be understood that the leak source locating system 100 of this invention can also detect the airtightness of other products.

[0031] A six-sided microphone array module is arranged to completely surround the battery pack under test, forming a closed three-dimensional acoustic detection network. The multi-channel synchronous acquisition unit includes six distributed acquisition cards and a master-slave synchronization clock distributor, each connected to one of the six microphone array modules to achieve synchronous sampling of the six sound fields. The sound source imaging processing unit is an industrial computer with a built-in finite element simulation database, machine learning localization model, differential evolution parameter optimization module, and feature extraction module. It connects to the multi-channel synchronous acquisition unit via a communication interface to achieve precise localization.

[0032] The principle of this invention is to construct a six-sided acoustic observation network, or a six-sided microphone array module, around the battery pack under test. This allows the ultrasonic signal generated at the leakage point to be captured simultaneously by the array from six different spatial perspectives, providing sufficient multi-view acoustic field features for subsequent machine learning localization models.

[0033] Laboratory 6 is a semi-anechoic environment, with metal wedge-shaped sound-absorbing structures 7 covering the surrounding walls to ensure testing accuracy.

[0034] In one embodiment, such as Figure 1As shown, the bottom array 2 refers to multiple microphones arranged in a specific pattern within a base plate. The base plate and microphones are embedded below the floor plane, with their top surfaces flush with the floor. The microphones in the bottom array 2 are arranged in a high-density rectangular grid, and the spacing between the array elements meets the spatial sampling requirements of the ultrasonic frequency band. The battery pack under test is placed on top of four pillars 8, with a testing station set in the central area. The battery pack bearing area is enclosed by the four supporting pillars 8, and the battery pack under test is placed horizontally on top of the pillars 8. The bottom surface of the battery pack under test maintains a near-field detection distance from the bottom array 2, utilizing the high signal-to-noise ratio characteristics of the near field to capture the weak ultrasonic signal from the bottom weld leakage.

[0035] like Figure 2 As shown, the top array 1 refers to multiple microphones arranged in a specific pattern and mounted on a top plate. The top plate is suspended from the ceiling by a lifting mechanism 5, which is connected to the ceiling of the laboratory 6. The lifting mechanism 5 drives the top array 1 to rise and fall. During testing, the top array 1 descends to above the battery pack being tested, maintaining a preset distance from it. During loading, it rises to a position close to the ceiling to avoid obstructing movement. The lifting mechanism 5 can be a motor with gears and racks or a cylinder, etc., and the specific implementation of the lifting mechanism 5 is not limited.

[0036] The multiple microphones of the top array 1 are arranged in a sparse spiral array with a central gap to avoid interference from the hoisting fixture.

[0037] like Figure 3 As shown, the four side arrays 3 refer to multiple microphones arranged in a preset pattern and mounted on four side plates. The four side plates are respectively mounted on four sets of horizontal guide rails. The four sets of horizontal guide rails drive the four side arrays 3 to move to the four sides of the battery pack under test, maintaining a certain distance from the battery pack. During testing, the four side arrays 3 move along the horizontal guide rails to a preset distance from the side of the battery pack under test. During loading and unloading, they retract to avoid obstructing movement, leaving a passage for the automated guided vehicle or personnel to enter and exit.

[0038] The multiple microphones of the four side arrays 3 are arranged in a cross shape.

[0039] The three-dimensional coordinates of each microphone element in the bottom array 2, top array 1, and four side arrays 3 are accurately measured using a total station. A three-dimensional coordinate system is established with the geometric center of the battery pack under test as the origin, and the coordinates of each microphone element are entered into the sound source imaging processing unit.

[0040] The multi-channel synchronous acquisition unit is connected to the six-sided microphone array module, which controls the six-sided microphone array module to synchronously sample the acoustic signal generated by the leakage of the battery pack under test at the same time.

[0041] The acoustic observations of the six-sided microphone array module are strictly aligned in time to ensure that the machine localization learning model receives a complete sound field snapshot of the same leak point at the same moment from six spatial perspectives, rather than aliased data due to temporal misalignment.

[0042] Each of the bottom array 2, top array 1, and four side arrays 3 is equipped with an independent multi-channel acquisition card. The six acquisition cards adopt a master-slave synchronous architecture, with one card as the master card and a built-in clock source. The other five cards are slave cards. The sampling clock and trigger signal output by the master card are synchronously distributed to the five slave cards via a clock distributor to ensure that the sampling time deviation of the six arrays is less than 1 nanosecond.

[0043] During testing, the main acquisition card receives an external trigger command and outputs a synchronization pulse. The six acquisition cards sample in parallel, converting the acoustic signals of each microphone in the six-sided array into digital sampling streams, which are then transmitted to the sound source imaging processing unit through the communication interface.

[0044] The sound source imaging processing unit includes a finite element simulation database and a machine learning localization model.

[0045] The finite element simulation database stores the mapping relationship between multiple mesh elements discretized according to preset dimensions around the battery pack under test and the theoretical sound field characteristics of each mesh element when used as a sound source. Finite element simulation modeling process: A three-dimensional finite element acoustic model is established for the battery pack under test and its surrounding near-field space. The dimensions of the battery pack under test are 1.0m × 1.5m × 0.25m, and the surrounding near-field space is a range extending outwards from the battery pack under test by 0.5m. The surrounding near-field space is the space between the simulated battery pack under test and the six-sided microphone array.

[0046] The six surfaces and near-field space of the battery pack under test are divided into three-dimensional grid units according to a preset size. In this embodiment, the grid units are divided into 50mm×50mm×50mm units. Each grid unit is assigned a unique area identification code, such as bottom surface B-01 to bottom surface B-60, front surface F-01 to front surface F-48, etc. Based on the propagation equation of sound waves in air, the theoretical sound pressure amplitude, arrival time delay, and spectral attenuation characteristics of sound waves propagating to each microphone element on the six sides are simulated and calculated when each grid unit is used as a sound source. A database mapping area identification codes and theoretical sound field characteristics is established.

[0047] The essence of this region-labeled database is to utilize the physical laws of sound wave propagation in discretized space to transform a continuous three-dimensional coordinate space into a structured set of discrete regions, which significantly reduces the search dimensionality and positioning ambiguity of subsequent machine learning localization models.

[0048] The feature extraction module is used to preprocess and extract features from the multi-channel acoustic signals, extracting the time-domain energy statistics, spectral distribution features, and signal correlation features between different array surfaces of each channel signal, and constructing a multi-channel sound field feature vector. Preferably, the spectral distribution features are the dominant frequency energy features in the 20kHz to 40kHz ultrasonic frequency band.

[0049] The machine learning localization model is used to compare and match the measured sound field features obtained by feature extraction from the acoustic signals synchronously collected by the six-sided microphone array with the theoretical sound field features of the finite element simulation database. First, it outputs the area identifier code of the grid cell to which the leakage point belongs to achieve coarse localization at the regional level. Then, it outputs the precise relative coordinates of the leakage point within the grid cell to which the leakage point belongs to achieve fine localization at the coordinate level, and finally maps it to three-dimensional spatial coordinates.

[0050] Before machine learning localization, finite element simulation mesh generation is introduced to discretize and label the detection space. The continuous sound field space around the tested battery pack is discretized using the finite element method, compressing the infinite three-dimensional space into a finite set of discrete regions. This allows the machine learning localization model to first perform coarse localization at the region level, and then perform fine localization at the coordinate level within the defined region.

[0051] The machine learning localization model includes a region identification sub-model and a fine localization sub-model. The region identification sub-model is used to compare and match the measured sound field features with the theoretical sound field features in the finite element simulation database, and output the region identification code of the grid cell to which the leakage point belongs, thus compressing the global three-dimensional search space to the scale of a single grid cell.

[0052] The precise positioning sub-model is used to output the precise relative coordinates of the leakage point within the grid cell corresponding to the area identification code, combining the measured sound field characteristics and the theoretical sound field characteristics.

[0053] The sound source imaging processing unit preprocesses and extracts features from the multi-channel acoustic signals synchronously acquired by the six-sided microphone array module. Utilizing the natural laws of sound wave propagation, it converts the raw sampling data into theoretical feature vectors that characterize the spatial distribution of the sound field.

[0054] The extracted theoretical feature vectors are input into a machine learning region identification model. This model compares and matches the measured sound field features with the theoretical sound field features in the finite element simulation database, and outputs the region identifier code to which the leakage point belongs, thus achieving coarse-level region-level localization.

[0055] This step compresses the global 3D search space into a single grid cell, eliminating the distance and angle coupling ambiguity phenomenon of single-face arrays in large-scale searches.

[0056] Within the defined area identifier corresponding to the grid cell, combining the actual measurement characteristics of the six-sided microphone array module with the theoretical sound field characteristics of the area, the precise relative coordinates of the leakage point within that area are output. Finally, the area reference coordinates and the relative offset coordinates are superimposed to map to global three-dimensional coordinates, and then converted into a description of the specific weld location on the surface of the tested battery pack.

[0057] The differential evolution parameter optimization module uses the differential evolution algorithm to globally optimize the parameters of the machine learning positioning model, so as to ensure that the model has good generalization ability between the laboratory 6 standard environment and the real battery pack production line environment.

[0058] The hyperparameters of the machine learning localization model are used as optimization variables in the differential evolution algorithm. These hyperparameters include, but are not limited to, feature weight coefficients, the number of neural network layers, activation function type, and learning rate. The algorithm generates multiple sets of parameter vectors through initialization and sequentially performs mutation, crossover, and selection operations to iteratively search for the parameter combination that optimizes the overall performance of the model.

[0059] Scene generalization evaluation and fitness calculation: After each generation of parameter vector updates, the trained model is evaluated and predicted in two separate scenarios. In the laboratory environment 6, predictive tests were conducted using a standard perforated plate within a standard semi-anechoic laboratory 6. In a real battery pack environment, predictive testing is performed on a real battery pack under test using actual weld microleakage or standard micropores.

[0060] Calculate the fitness function, such as the inverse weighted average of the localization errors in the two scenarios. If the fitness does not reach the convergence threshold, the differential evolution algorithm continues to iteratively update the parameter vector. If the fitness converges, the current parameter vector is output as the optimal parameter calibration result and fixed into the machine learning localization model.

[0061] This optimization process utilizes the global optimization capability of the differential evolution algorithm to solve the technical problem that traditional gradient descent methods are prone to getting trapped in local optima, resulting in poor generalization ability of the model on the complex structure of real battery packs.

[0062] This invention also relates to a method for locating a leakage source, such as... Figure 4 As shown, the steps include: S1: Microphone arrays are arranged in the six directions of the test piece 4: top, bottom, front, back, left, and right, forming a six-sided microphone array module.

[0063] S2: Simultaneously acquire multi-channel acoustic signals generated by leakage in the tested component 4 through a six-sided microphone array module.

[0064] S3: Preprocess and extract features from the acquired multi-channel acoustic signals, extracting the time-domain energy statistics, spectral distribution features, and signal correlation features of the six-sided microphone array module to construct a multi-channel sound field feature vector. Preferably, the spectral distribution features are the dominant frequency energy features within the 20kHz to 40kHz ultrasonic frequency band.

[0065] S4: Input the multi-channel sound field feature vector into the pre-trained machine learning localization model, and the machine learning localization model outputs the three-dimensional spatial coordinates of the leakage point.

[0066] Preferably, the process of the machine learning localization model outputting the three-dimensional spatial coordinates of the leak point includes the following steps: Based on the acoustic signals synchronously acquired by the six-sided microphone array module, the measured sound field features are obtained after feature extraction. The measured sound field features are compared and matched with the theoretical sound field features in the finite element simulation database, and the area identification code of the mesh element to which the leakage point belongs is output. Within the grid cell range corresponding to the area identifier code, output the precise relative coordinates of the leak point.

[0067] Preferably, the machine learning localization model is pre-trained through the following steps: A three-dimensional finite element acoustic model is established for the test component 4 and the space between the test component 4 and the six-sided microphone array module. The test component 4 and the space are divided into multiple grid units according to a preset size, and each grid unit is assigned a unique area identification code. In the three-dimensional finite element acoustic model, each of the grid cells is assumed to be a leakage point. Based on the physical laws of sound wave propagation in air, the theoretical sound field characteristics of each grid cell when leakage occurs are calculated. When the sound wave propagates to each observation point corresponding to the position of each microphone element in the six-sided microphone array module, a mapping database between the area identification code of the grid cell and the theoretical sound field characteristics is established. Based on the grid cell division method and the area identification code of each grid cell, standard leak holes are set one by one and sequentially at corresponding positions on the surface of the test piece 4 corresponding to each grid cell. Each standard leak hole corresponds one-to-one with the area identification code of a grid cell, and only one standard leak hole is in the open leakage state at any time. Acoustic signals are synchronously collected through the six-sided microphone array module, the measured sound field features are extracted and paired with the area identification code and three-dimensional coordinates corresponding to the position to construct a training dataset. The machine learning localization model is trained using the training dataset to learn the mapping relationship between the measured sound field features and the region identifier code of the grid cell to achieve region-level coarse localization, and to learn the mapping relationship between the measured sound field features and the precise coordinates within the region to achieve coordinate-level fine localization.

[0068] Preferably, the method further includes an optimization step: The hyperparameters of the machine learning localization model are used as optimization variables for the differential evolution algorithm. Multiple sets of parameter vectors are generated by initialization, and mutation, crossover, and selection operations are performed sequentially for iterative search. After each generation of parameter vector updates, the model is predicted and evaluated in both the standard laboratory environment and the real production line environment, and the comprehensive positioning error in the two scenarios is calculated as the fitness. When the fitness converges, the optimal parameter calibration result is output and solidified into the machine learning localization model.

[0069] In one specific embodiment, the implementation environment is a semi-anechoic testing laboratory 6, with indoor dimensions of 8.4 meters long, 6.2 meters wide, and 4.8 meters high, and the surrounding walls are covered with metal wedge sound-absorbing structures 7.

[0070] The tested object is a new energy power battery pack, with external dimensions of approximately 1.0 m × 1.5 m × 0.25 m, and is made of aluminum alloy.

[0071] Step 1, System Layout. Specifically, four supporting columns 8 are fixed at the center of laboratory 6, forming a battery pack support area. The bottom plate of the bottom array 2 is embedded in the floor groove, with its upper surface flush with the floor. The side plates of the four-sided array are mounted on guide rail slides. The top plate of the top array 1 is mounted on the lifting mechanism 5. The three-dimensional coordinates of each microphone array element are calibrated using a total station, establishing a coordinate system with the geometric center of the battery pack as the origin.

[0072] Step 2, Finite Element Simulation and Region Marking. Specifically, in the sound source imaging processing unit, a finite element acoustic model is established for the tested battery pack and the surrounding 0.5m near-field space. The six surfaces of the enclosure are divided into two-dimensional mesh elements, and each element is assigned a unique region identifier. The theoretical propagation delay and amplitude attenuation of the sound wave to each microphone element on the six sides are calculated by simulation when each element is used as a sound source, and a region and theoretical sound field mapping database is established.

[0073] Step 3: Data Acquisition and Model Training. Specifically, standard leak holes with a diameter of 0.2 mm are sequentially arranged at the center of each unit in the battery pack under test. Dry compressed air is injected into the chamber through a high-pressure gas cylinder via a pressure reducing valve. The inflation pressure is set to a commonly used industrial testing pressure within the range of 20 kPa to 50 kPa. Under stable leakage conditions at each calibration point, a six-sided array synchronously acquires 2 seconds of sound field data. The temporal energy characteristics, spectral energy characteristics, and inter-surface cross-correlation characteristics of all channels are extracted and paired with the region's identification code and precise three-dimensional coordinates to construct a training dataset.

[0074] Step 4, Differential Evolution Parameter Optimization. Specifically, the hyperparameters of the machine learning localization model are used as optimization variables for the differential evolution algorithm, generating 50 sets of parameter vectors. The model trained with each set of parameters is tested in both a standard leak scenario in the laboratory and a real battery pack micro-leakage scenario. The combined localization error in both scenarios is calculated as the fitness. After 200 iterations, when the fitness converges to a stable value, the optimal parameter calibration result is output and solidified into the machine learning localization model. The optimized model achieves an average region recognition accuracy greater than 98% on the validation set, with a precise localization error within the region of less than 5 mm.

[0075] Step 5: Mechanical Positioning. Specifically, the automated guided vehicle transports the battery pack to be tested to the top of column 8 and places it there. The four-sided array plate moves along the guide rail to a position 40cm from the side of the battery pack. The top array plate 1 descends to a position 40cm from the top of the battery pack. At this point, the bottom of the battery pack is 215cm from the bottom array plate.

[0076] Step 6: Inflate and Stabilize Air. Specifically, dry compressed air is injected into the battery pack under test through a high-pressure gas cylinder via a pressure reducing valve. The inflation pressure is set to a commonly used industrial testing pressure within the range of 20 kPa to 50 kPa, and the pressure is maintained for 10 seconds.

[0077] Step 7: Synchronous Acquisition and Feature Extraction. Specifically, six acquisition cards receive the same trigger signal and synchronously acquire the acoustic signals from the six-sided microphone array. The sound source imaging processing unit performs bandpass filtering on the raw time-domain data, extracts the 20kHz to 40kHz ultrasonic frequency band, calculates the time-domain energy of each channel, the spectral main frequency energy, and the signal cross-correlation values ​​between different array surfaces, and constructs a multi-channel feature vector.

[0078] Step 8: Machine Learning Localization. Specifically, the multi-channel feature vector from Step 7 is input into the trained and differentially evolved machine learning localization model. The model first outputs the region identifier code to which the leak point belongs, such as bottom surface B-12, to achieve coarse region localization, and then outputs the precise relative coordinates within that region, which are superimposed and mapped to global three-dimensional coordinates.

[0079] Step 9: Rework Instructions. Specifically, the system maps the three-dimensional coordinates to the specific weld locations on the surface of the enclosure, such as the bottom weld, 35 cm from the front end and 42 cm from the left side, and outputs the data to the rework station display screen.

[0080] Step 10: Reset. Specifically, the top array 1 rises, the four-sided array retreats, and the automated guided vehicle or personnel remove the battery pack.

[0081] Using the above embodiments, the standard perforated plate was continuously tested and verified 100 times. The area identification accuracy rate was 98.5%, the average error of the bottom surface precision positioning was about 3 mm, the average error of the side and top surface precision positioning was about 4 mm to 5 mm, the missed detection rate was 0%, and the single-piece testing cycle was about 45 seconds.

[0082] The present invention arranges microphone arrays in six directions around the battery pack under test: top, bottom, front, back, left, and right. Through the top lifting mechanism 5, the side guide rail moving mechanism, and the bottom embedded installation, it realizes full-coverage acoustic observation of the weld seams on all six sides of the battery pack, eliminating the detection blind spots of single-sided arrays.

[0083] The six-sided microphone array achieves nanosecond-level synchronous sampling through a master-slave synchronous clock architecture, ensuring that the six sets of acoustic observations are strictly aligned in time, providing multi-view data of the same sound field event for subsequent positioning.

[0084] Multi-channel features such as temporal energy, spectral distribution, and inter-surface correlation are extracted from the synchronous acquisition of the six-sided array, input into a pre-trained machine learning model, and the three-dimensional spatial coordinates of the leakage point are directly output, transforming the traditional physical model-driven localization into a data-driven regression problem.

[0085] A single battery pack can complete the full coverage inspection of the six weld seams with just one setup, so there is no need to move and scan each side like a handheld device, nor is there a need for long-term vacuuming and tracer gas replacement like helium testing. The single-piece inspection cycle is shortened from several minutes to less than 60 seconds, meeting the cycle requirements of automated production lines.

[0086] Furthermore, this invention utilizes a six-sided microphone array to simultaneously observe the same leak point from different spatial perspectives. Each array provides a conical positioning constraint, and the six conical surfaces intersect in three-dimensional space to form a unique solution. This eliminates the depth ambiguity of single-sided observation, controls the leak point positioning error to within 5mm, and can directly guide repair personnel to accurately repair the weld.

[0087] This invention utilizes a bottom-embedded array, a horizontally guided moving side array 3, and a lifting mechanism 5 to drive the top array 1 to rise and fall, forming a fully enclosed detection network. This achieves seamless coverage of the weld seams on all six sides of the tested battery pack, effectively detecting the bottom weld seam, a high-risk area for leaks, with a near-zero missed detection rate. Furthermore, the detection process requires no manual intervention, allowing for direct integration into automated production lines. This enables fully automated processes from feeding, inflation, and testing of the battery pack to unmanned operation, significantly improving detection consistency. In addition, a machine learning model directly outputs the three-dimensional coordinates of the leak point and maps them to the specific weld seam location. This eliminates the need for extensive inspections or repeated leak testing by repair personnel, allowing them to pinpoint the specific weld seam to centimeter-level location, reducing repair time from tens of minutes to just a few minutes.

[0088] This invention employs acoustic non-contact measurement, maintaining an air gap of 30cm-50cm between the array and the housing to avoid secondary damage such as scratches or residues to the sealing surface or insulating coating of the battery pack being tested, thus ensuring the subsequent sealing performance of the housing.

[0089] The preferred embodiments of the present invention have been described in detail above, but it should be understood that, if necessary, aspects of the embodiments can be modified to utilize aspects, features, and concepts from various patents, applications, and publications to provide other embodiments.

[0090] In light of the detailed description above, these and other changes can be made to the embodiments. Generally, the terminology used in the claims should not be considered limited to the specific embodiments disclosed in the specification and claims, but should be understood to include all possible embodiments together with the full scope of equivalents enjoyed by these claims.

[0091] Those skilled in the art will understand that the above embodiments are specific examples of implementing the present invention, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of the present invention.

Claims

1. A leak source positioning system, characterized by, include: A six-sided microphone array module includes a top array, a bottom array, and four side arrays. The top array and the bottom array are positioned above and below the device under test, respectively, and the four side arrays are positioned around the four sides of the device under test. A multi-channel synchronous acquisition unit is electrically connected to the six-sided microphone array module, and controls the six-sided microphone array module to synchronously sample the acoustic signal generated by the leakage of the test component at the same time. The sound source imaging processing unit, electrically connected to the multi-channel synchronous acquisition unit, includes: The finite element simulation database stores the mapping relationship between multiple grid cells that discretize the space around the test piece according to a preset size and the theoretical sound field characteristics when each grid cell is used as a sound source; The machine learning localization model is used to compare and match the measured sound field features obtained by feature extraction of the acoustic signals synchronously collected by the six-sided microphone array with the theoretical sound field features of the finite element simulation database. First, it outputs the area identifier code of the grid cell to which the leakage point belongs to achieve coarse localization at the regional level. Then, it outputs the precise relative coordinates of the leakage point within the grid cell to which the leakage point belongs to achieve fine localization at the coordinate level, and finally maps it to three-dimensional spatial coordinates.

2. The leak source localization system of claim 1, wherein, The six-sided microphone array module includes: A floor, located below the test piece; the bottom surface array is embedded below the floor plane and the top surface of the bottom surface array is flush with the floor; A lifting mechanism is connected to the top surface array drive and is used to control the top surface array to rise and fall above the test piece; A horizontal guide rail is driven and connected to the four side arrays to control the movement of the four side arrays to the four sides of the test piece.

3. The leakage source locating system according to claim 1, characterized in that, The multi-channel synchronous acquisition unit includes: The main acquisition card has a built-in clock source and is electrically connected to one of the top array, the bottom array, and the four side arrays. Five acquisition cards are electrically connected to the remaining five of the top array, the bottom array, and the four side arrays; The sampling clock and trigger signal output by the main acquisition card are synchronously distributed to each of the slave acquisition cards via a clock distributor, so that the sampling time deviation of the top array, the bottom array and the four side arrays is less than 1 nanosecond.

4. The leakage source locating system according to claim 1, characterized in that, The finite element simulation database construction methods include: A three-dimensional finite element acoustic model is established based on the test device and the space between the test device and the six-sided microphone array module. The test device and the space are divided into multiple grid units, and each grid unit is assigned a unique area identification code. Based on the physical laws of sound wave propagation in air, the theoretical arrival time, theoretical sound intensity, and theoretical frequency change of the sound wave in the three-dimensional finite element acoustic model corresponding to the observation point of the six-sided microphone array module are calculated for each mesh cell under the assumption of leakage. The region identifier code of each grid cell is associated with and stored with the corresponding theoretical sound field feature to form the finite element simulation database.

5. The leakage source locating system according to claim 1, characterized in that, The machine learning localization model includes: The region identification sub-model is used to compare and match the measured sound field features with the theoretical sound field features in the finite element simulation database, and output the region identification code of the grid cell to which the leakage point belongs; The precise positioning sub-model is used to output the precise relative coordinates of the leakage point within the grid cell corresponding to the area identification code, combining the measured sound field characteristics and the theoretical sound field characteristics.

6. The leakage source locating system according to claim 1, characterized in that, The top array is arranged in a sparse spiral array with a central void.

7. The leak source locating system according to claim 1, characterized in that, The side array is arranged in a cross shape.

8. The leakage source locating system according to claim 1, characterized in that, The bottom array is arranged in a rectangular grid.

9. The leakage source locating system according to claim 1, characterized in that, The sound source imaging processing unit further includes a differential evolution parameter optimization module, and the optimization method of the differential evolution parameter optimization module is as follows: The hyperparameters of the machine learning localization model are used as optimization variables for the differential evolution algorithm. Multiple sets of parameter vectors are generated through initialization, and mutation, crossover, and selection operations are performed sequentially for iterative search. After each generation of parameter vector updates, the model is predicted and evaluated in both laboratory standard environment and real production line environment, and the comprehensive positioning error in the two scenarios is calculated as the fitness. When the fitness converges, the optimal parameter calibration result is output and solidified into the machine learning localization model.

10. The leak source locating system according to claim 1, characterized in that, The sound source imaging processing unit further includes a feature extraction module, which is used to preprocess and extract features from the acoustic signal, extracting the time-domain energy statistical features, spectral distribution features, and signal correlation features between the top array, the bottom array, and the four side arrays of the acoustic signal; The spectral distribution characteristics are energy characteristics within the 20kHz to 40kHz ultrasonic frequency band.