Integrated busbar latent defect detection method and related cell connection assembly detection apparatus

Abstract

The application discloses a kind of integrated busbar recessive defect detection method and related battery connection assembly detection equipment, applied to CCS detection processing equipment.Information for indicating recessive defect detection to multiple detection channels of CCS to be measured is obtained;Channel mapping information is determined according to detection task information, and the broadband impedance spectrum data of CCS to be measured is collected according to channel mapping information after being subjected to preset stress;Data validity check is carried out on broadband impedance spectrum data, impedance amplitude data and impedance phase angle data in effective impedance spectrum data are standardized and noise reduction is handled, and impedance spectrum feature atlas is formed by corresponding combination according to frequency point arrangement order;Recessive defect recognition result is determined according to impedance spectrum feature atlas and detection result is output.The application can improve the recessive defect detection capability of multiple channels of CCS and the traceability of detection result.

Description

Technical Field

[0001] This application relates to the field of battery component testing technology, and in particular to an integrated busbar latent defect detection method and related cell connection component testing equipment. Background Technology

[0002] An integrated busbar, also known as a CellsContactSystem (CCS), is a module that highly integrates electrical connections, signal acquisition, and structural support functions within a battery module. It typically includes at least one of the following: conductive copper or aluminum busbars, flexible printed circuit boards (FPCs), flexible flat cables (FFCs), printed circuit boards (PCBs), soldered connections, crimped connections, connectors, sensors, and insulating structural components. The CCS includes both power circuits for carrying large currents and signal acquisition circuits for transmitting sampled signals such as cell voltage and temperature.

[0003] During the manufacturing process, CCS may develop defects such as microcracks in the FPC circuitry, poor solder joints, conductor interface oxidation, poor crimping contact, or localized thinning of the circuitry. These defects may initially maintain electrical conductivity and not directly cause open circuits, short circuits, or insulation failure in the CCS. However, under service conditions such as temperature cycling, mechanical vibration, and current-carrying operation, they may continue to deteriorate and cause sampling anomalies, connection failures, or localized overheating. This article refers to these types of defects, which are not easily identified during routine factory testing but may cause functional failures during subsequent service, as latent defects.

[0004] Current CCS factory testing typically includes continuity testing, DC low-resistance testing, and withstand voltage testing. Continuity testing primarily determines whether an electrical path has been formed, DC low-resistance testing primarily determines whether the DC resistance of the path under test is within a predetermined range, and withstand voltage testing primarily determines whether the insulation structure has broken down. Because latent defects may only affect the impedance amplitude, phase angle, or stability of repeated measurements at a specific frequency, they may not necessarily show obvious abnormalities under DC or low-frequency testing conditions.

[0005] Metallographic sectioning can observe the structural state of some connected cross-sections, but it damages the product under test, making it difficult to use for batch product inspection. Infrared thermography typically requires applying a load to the tested path and waiting for a temperature rise, limiting its ability to identify minute defects that have not yet generated significant heat. X-ray inspection is more suitable for identifying defects that can create significant structural differences. For conductor interface oxidation or early contact defects that only cause changes in electrical interface properties without significant deformation, the identification results are easily affected by structural overlap and imaging resolution.

[0006] Broadband impedance spectroscopy can characterize the complex impedance response of a path under test at multiple frequency points. Different defects have different effects on equivalent resistance, inductance, capacitance, and interface polarization state, and therefore may manifest as variations in impedance amplitude, impedance phase angle, or local variation trends in different frequency ranges. However, CCS includes multiple power loops and signal acquisition loops, and the structural length, conductor material, connection method, and test point of different channels are different, making it difficult to make a unified judgment on multiple channels by directly using a single impedance threshold.

[0007] Furthermore, impedance changes caused by latent defects in the CCS may be on the same order of magnitude as probe contact impedance, matrix switch conduction impedance, test cable phase delay, and fixture parasitic parameters. If the test link is not calibrated and compensated, test link errors may mask defect responses or be misidentified as defects. Different CCS models also have different numbers of channels and test points, necessitating the establishment of a correspondence between product channels and actual test points.

[0008] Existing technologies lack a complete processing path suitable for CCS multi-channel detection, which integrates stress processing, channel mapping, broadband impedance spectrum acquisition, data validity verification, amplitude and phase feature construction, latent defect identification, and detection result output into an interconnected detection process.

[0009] Therefore, there is a need to provide an integrated busbar latent defect detection method to determine the latent defect identification result based on the impedance response of multiple channels under test at different frequencies without damaging the normal structure of the CCS under test. Summary of the Invention

[0010] This application provides an integrated busbar latent defect detection method to improve the detectability of CCS latent defects and establish the correspondence between the channel to be detected, broadband impedance spectrum data and detection results.

[0011] Firstly, this application provides an integrated busbar latent defect detection method, which can be executed by a CCS detection and processing device. The CCS detection and processing device acquires detection task information, which instructs multiple channels of the CCS under test to perform latent defect detection; determines channel mapping information corresponding to the multiple channels under test based on the detection task information, and acquires broadband impedance spectrum data of the CCS under test after pre-stress treatment, collected according to the channel mapping information; determines impedance spectrum characteristic information corresponding to each channel under test based on the broadband impedance spectrum data, which characterizes the impedance amplitude and impedance phase angle changes of the channel under test at different frequencies; determines the latent defect identification result of the CCS under test based on the impedance spectrum characteristic information, and outputs the detection result corresponding to the CCS under test.

[0012] In the above methods, channel mapping information establishes a correspondence between the channel to be tested and the test points on the CCS to be tested. Pre-set stress treatment can cause some defects that have not yet formed explicit failures to produce easily distinguishable impedance response changes. Wideband impedance spectrum data can simultaneously reflect the impedance amplitude and impedance phase angle changes at multiple frequency points. Therefore, the CCS detection and processing equipment can determine the latent defect identification results based on the frequency response differences of each channel to be tested.

[0013] In one possible design, the impedance spectrum feature information corresponding to each channel to be tested is determined based on broadband impedance spectrum data, including: verifying the validity of the broadband impedance spectrum data to obtain valid impedance spectrum data; standardizing and denoising the impedance amplitude data and impedance phase angle data in the valid impedance spectrum data respectively; combining the standardized and denoised impedance amplitude data and impedance phase angle data to obtain the impedance spectrum feature map corresponding to each channel to be tested, and determining the impedance spectrum feature map as the impedance spectrum feature information.

[0014] This design reduces the impact of missing frequency points, instantaneous contact jitter, and external interference on impedance spectrum characteristics, and enables impedance amplitude data and impedance phase angle data to be processed together.

[0015] In one possible design, the broadband impedance spectrum data is validated to obtain valid impedance spectrum data. This includes: checking whether the frequency point data corresponding to each channel under test is complete; determining whether to reacquire the broadband impedance spectrum data corresponding to the channel under test based on the proportion of missing frequency point data in the total frequency point data of the corresponding channel under test; determining the data distribution range based on the mean and standard deviation of the impedance amplitude data or impedance phase angle data corresponding to the same channel under test; identifying frequency point data that exceeds the data distribution range and does not appear stably in multiple repeated samplings as abnormal data; and replacing the abnormal data with frequency point data adjacent to the abnormal data to obtain valid impedance spectrum data.

[0016] This design allows for the re-acquisition of data from channels with significant data loss, and the replacement of non-repeatable local jump data, reducing the likelihood of data interruptions or transient interference being used as defect features.

[0017] In one possible design, the impedance amplitude data and impedance phase angle data in the effective impedance spectrum data are standardized and denoised separately, including: normalizing the impedance amplitude data and impedance phase angle data corresponding to the same channel to be detected; performing wavelet decomposition and reconstruction on the normalized impedance amplitude data and impedance phase angle data to remove random noise and baseline drift; smoothing the wavelet decomposition and reconstruction of the impedance amplitude data and impedance phase angle data to retain the characteristic change points in the impedance amplitude data and impedance phase angle data; and combining the smoothed impedance amplitude data and impedance phase angle data according to the order of frequency points to form an impedance spectrum feature map.

[0018] This design allows different test channels of the same product model to be processed on a uniform data scale, while suppressing noise and preserving the local variation trend in the impedance spectrum.

[0019] In one possible design, the latent defect identification result of the CCS under test is determined based on impedance spectrum feature information, including: inputting the impedance spectrum feature map corresponding to each channel to be detected into a pre-trained feature extraction model, performing convolution and pooling processing on the impedance spectrum feature map to obtain the defect feature vector of the corresponding channel to be detected; performing dimensionality reduction processing on the defect feature vector corresponding to each channel to be detected to obtain the corresponding low-dimensional defect feature vector; inputting the low-dimensional defect feature vector into a pre-trained defect classification model to determine the defect type corresponding to each channel to be detected; identifying the channels to be detected whose defect type is not "no defect" as defect channels, and forming the latent defect identification result based on the defect type corresponding to each channel to be detected.

[0020] Through this design, the feature extraction model can extract local variation features and frequency band response features from the impedance spectrum feature map, the dimensionality reduction process can reduce duplicate features, and the defect classification model can determine the defect type of the corresponding channel to be detected based on the low-dimensional defect feature vector.

[0021] In one possible design, the defect feature vectors corresponding to each channel to be tested are dimensionality-reduced to obtain corresponding low-dimensional defect feature vectors. This includes: obtaining the product model corresponding to the CCS to be tested, and obtaining the training mean vector and principal component projection matrix determined during model training and associated with the product model; decentralizing the defect feature vectors corresponding to each channel to be tested based on the training mean vector; and projecting the decentralized defect feature vectors onto the low-dimensional feature space based on the principal component projection matrix to obtain low-dimensional defect feature vectors. Defect types may include no defects, microcracks in FPC circuits, poor solder joints, conductor interface oxidation, poor crimping contact, and local thinning of circuits.

[0022] With this design, both the training samples and the products to be tested use the same training mean vector and principal component projection matrix, which enables the training data and the detection data to be located in the same low-dimensional feature space, avoiding the change of feature coordinate system caused by recalculating the principal components for each CCS to be tested.

[0023] In one possible design, determining the latent defect identification result of the CCS under test based on impedance spectrum feature information further includes: determining location feature information based on channel mapping information corresponding to the defect channel, whereby the location feature information is used to characterize the channel region or test sub-segment corresponding to the defect channel; inputting the low-dimensional defect feature vector corresponding to the defect channel, the defect channel identifier, and the location feature information into a pre-trained defect localization model to obtain the location of the latent defect in the CCS under test; determining the target defect quantitative model from multiple defect quantitative models based on the defect type, and inputting the low-dimensional defect feature vector corresponding to the defect channel into the target defect quantitative model to obtain the severity of the latent defect; and correlating the defect type, the location of the latent defect, and the severity to form the latent defect identification result. The severity is defined as the microcrack length, the proportion of cold solder joint area, the increase in contact impedance, or the proportion of line thinning corresponding to the defect type.

[0024] In one possible design, the location of a latent defect includes at least the channel region corresponding to the defect channel. The channel definition file may also include intermediate test points located between the starting and ending test points, and divide a channel into multiple test sub-segments; the CCS detection and processing equipment uses the impedance spectrum characteristics and segment location characteristics of different test sub-segments to train the defect location model, so that the location output matches the actual acquired spatial information.

[0025] In one possible design, multiple quantitative defect models correspond to different defect types. The microcrack length model, the cold solder joint area model, the contact impedance increment model, and the circuit thinning ratio model are all obtained by training samples with corresponding true severity labels, thereby avoiding the use of the same regression output to represent severity parameters with different dimensions and physical meanings.

[0026] In one possible design, before acquiring broadband impedance spectrum data, the method further includes: acquiring product identification information corresponding to the CCS under test, and determining the product model corresponding to the CCS under test based on the product identification information; acquiring corresponding stress treatment parameters based on the product model; controlling the pretreatment device to perform temperature cycling and vibration treatment on the CCS under test according to the stress treatment parameters, so that the latent defects in the CCS under test produce corresponding impedance response changes; acquiring the operating status information of the pretreatment device and the connection status information of the CCS under test during the temperature cycling and vibration treatment; and, if the operating status information and connection status information meet the preset processing conditions, identifying the CCS under test that has completed the temperature cycling and vibration treatment as the CCS under test that has undergone preset stress treatment, and associating the product identification information with the corresponding stress treatment record.

[0027] Preset stress treatment can be performed using offline batch screening mode or rapid screening mode. The stress treatment parameters are determined by the safe stress window corresponding to the product model. The safe stress window refers to the stress range that has been verified to not cause new explicit damage to defect-free CCS, but can increase the impedance response difference of existing latent defects.

[0028] In one possible design, channel mapping information corresponding to multiple channels to be tested is determined based on the detection task information, and broadband impedance spectrum data collected according to the channel mapping information is acquired. This includes: determining the start and end test points corresponding to each channel to be tested based on the channel definition file corresponding to the CCS under test, forming channel mapping information; generating test paths corresponding to test probes based on the start and end test points corresponding to each channel to be tested; acquiring the image positioning results of the CCS under test, correcting the test paths based on the image positioning results, and controlling the test probes to connect to the corresponding test points according to the corrected test paths; switching multiple channels to be tested according to the channel mapping information, and collecting the voltage amplitude, current amplitude, and phase difference corresponding to each channel to be tested at multiple frequency points; determining the impedance amplitude and phase angle corresponding to each channel to be tested at multiple frequency points based on the voltage amplitude, current amplitude, and phase difference, forming broadband impedance spectrum data.

[0029] In one possible design, before acquiring broadband impedance spectrum data, at least two of the following calibration methods—open-circuit calibration, short-circuit calibration, and standard load calibration—are used to determine the test link errors corresponding to the test cables, matrix switches, test probes, and fixtures, and the acquired data is compensated based on these test link errors. For low-impedance power circuits, a four-wire test connection can be used; when the scanning frequency reaches the frequency band where the test link needs to be processed according to distributed parameters, high-frequency calibration can be performed using port extension, fixture de-embedding, or transmission parameter conversion.

[0030] In one possible design, the output of the test result for the CCS under test includes: acquiring the product identification information corresponding to the CCS under test; determining the reliability level of the CCS under test based on the latent defect identification result and preset grading rules, whereby the reliability level includes a pass level, a re-inspection level, and a fail level; associating and storing the reliability level, latent defect identification result, and broadband impedance spectrum data with the product identification information; generating a test report for the CCS under test based on the reliability level and the latent defect identification result; generating a sorting control instruction based on the reliability level, and sending the sorting control instruction to the production line sorting device so that the production line sorting device can perform the corresponding sorting operation on the CCS under test according to the sorting control instruction.

[0031] In one possible design, pre-defined grading rules are associated with the product model. If the severity of any defect reaches the corresponding non-conformance threshold, it is classified as a non-conformance; if it does not reach the non-conformance threshold but reaches the re-inspection threshold, or if the defect classification confidence level is lower than the pre-set confidence threshold, it is classified as a re-inspection; otherwise, it is classified as a pass / fail. When multiple defects exist simultaneously, the final reliability level can be determined according to the defect with the highest risk.

[0032] The broadband impedance spectroscopy testing in this application is a non-destructive electrical contact testing performed under controlled low-energy electrical excitation conditions. Pre-defined stress treatment parameters and AC excitation parameters can be verified in advance according to the product model to avoid structural or electrical damage to the CCS under test that meets quality requirements during the testing process.

[0033] The method provided in this application can establish a correlation between product identification information, channel mapping information, broadband impedance spectrum data, latent defect identification results, and reliability level, which facilitates the traceability of test results and provides data basis for subsequent product process adjustments and failure analysis.

[0034] The steps described in this application can be executed centrally by a single CCS testing and processing device, or they can be executed collaboratively by an industrial control computer, impedance test controller, motion controller, and model inference server. Any implementation of the CCS testing and processing device described in this application that can perform the corresponding data acquisition, processing, and output functions falls under this application's definition of a CCS testing and processing device. Attached Figure Description

[0035] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0036] Figure 1 This is a schematic diagram of an application scenario for a CCS detection system provided in an embodiment of this application; Figure 2A schematic diagram of the hardware architecture of a CCS detection system provided in this application embodiment; Figure 3 This application provides a schematic diagram of the software functional architecture of a CCS detection and processing device. Figure 4 A flowchart illustrating an integrated busbar latent defect detection method provided in this application embodiment; Figure 5 A flowchart of a preset stress treatment process provided in this application embodiment; Figure 6 A schematic diagram of channel mapping and test probe path provided in an embodiment of this application; Figure 7 A flowchart of test link calibration and broadband impedance spectrum acquisition provided for embodiments of this application; Figure 8 A flowchart for impedance spectrum data validity verification and preprocessing is provided for embodiments of this application; Figure 9 A flowchart illustrating model training, defect classification, location determination, and quantitative evaluation is provided for embodiments of this application. Figure 10 This application provides a flowchart for reliability grading, report generation, and production line sorting in an embodiment of the present application. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0038] In the description of the embodiments in this application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to indicate a relationship between associated objects that can exist individually or in combination. The terms "first" and "second" are used only to distinguish identical or similar technical features and are not used to limit relative importance.

[0039] In this application, the term "CCS detection processing equipment" is used uniformly to refer to a processing entity that performs at least one of the following functions: detection task configuration, channel mapping, data acquisition, data processing, model inference, and result output. CCS detection processing equipment can be an industrial control computer, a server, an edge computing device, or a combination of multiple processing devices.

[0040] In this application, product identification information is used to represent information that can distinguish the CCS under test. Product identification information can be obtained through barcodes, QR codes, RFID tags, manufacturing execution system numbers, or manual input.

[0041] In this application, channel mapping information is used uniformly to represent the correspondence between the channels to be tested and the test points. Channel mapping information can be stored in tables, matrices, key-value pairs, or graph structures. The channel mapping matrix is ​​a specific data form of channel mapping information.

[0042] This application uses impedance amplitude data and impedance phase angle data to uniformly describe the two fundamental components of the broadband impedance spectrum. The phase difference between the voltage signal and the current signal is the impedance phase angle at the corresponding frequency point.

[0043] In this application, broadband impedance spectrum data refers to impedance amplitude and impedance phase angle data obtained by measuring multiple discrete frequency points distributed across multiple frequency ranges within a preset scanning frequency range, either directly output by impedance testing equipment or obtained after test link calibration and compensation, with measurement errors meeting the testing requirements of the corresponding product model. Impedance spectrum characteristic information is information obtained by processing the broadband impedance spectrum data and used for defect identification. An impedance spectrum characteristic map is one form of impedance spectrum characteristic information, which may also include at least one of the following: impedance amplitude curve slope, impedance phase angle curve slope, local peak value, frequency band integral value, and repeatability dispersion.

[0044] In this application, the pre-stress treatment refers to temperature cycling treatment, vibration treatment, or a combination of temperature cycling treatment and vibration treatment, used to induce a detectable change in impedance response of existing latent defects without causing new explicit damage to the CCS under test that meets the quality requirements; the latent defect identification result includes at least the defect type, and in some embodiments, it also includes the defect channel, defect area, two-dimensional location, severity parameter, or severity level.

[0045] In this application, the test results are the final output information provided to operators, upper-level systems, or production line sorting devices, and may include at least one of the following: latent defect identification results, reliability level, test report, and sorting control instructions; the severity is a continuous physical quantity or discrete level corresponding to the defect type, and the continuous physical quantity may be the length of microcracks, the proportion of cold solder joint area, the increase in contact impedance, or the proportion of line thinning.

[0046] Microcracks in FPC circuits refer to incompletely disconnected cracks in the FPC conductor circuit; poor solder joints refer to the effective soldering area between the solder pad and the connecting conductor being lower than the process requirements; conductor interface oxidation refers to the formation of an oxide layer at the conductor connection interface that alters the contact electrical properties; poor crimping contact refers to the contact state of the crimped part not meeting the predetermined requirements; local thinning of the circuit refers to the local cross-sectional area of ​​the conductor circuit being smaller than the designed cross-sectional area.

[0047] Let's combine the following... Figures 1 to 3 Explain the implementation environment and software functional architecture of the CCS inspection system, and then combine it with Figures 4 to 10 The overall testing process, including preset stress treatment, channel mapping, test link calibration and broadband impedance spectrum acquisition, data preprocessing, model training, defect identification, and result output, are explained separately.

[0048] Figure 1 This is a schematic diagram illustrating an application scenario of a CCS (Computer-Assisted Separation) testing system. The system can be installed in the factory inspection area, independent quality inspection area, or reliability screening area of ​​a CCS production line. The CCS samples to be tested enter the inspection area via a conveyor or manually, where product identification, pre-set stress treatment, impedance spectrum acquisition, defect identification, and result output are completed.

[0049] Figure 2 This is a schematic diagram of the hardware architecture of a CCS inspection system. The system may include CCS inspection and processing equipment, a preset stress treatment subsystem, a broadband impedance spectrum testing subsystem, a motion control and positioning subsystem, and a production line docking and sorting subsystem.

[0050] The pre-set stress treatment subsystem may include a high and low temperature impact test chamber, a random vibration table, a CCS-specific fixture, and an equipment status monitoring unit. The high and low temperature impact test chamber is used to perform temperature cycling treatment, the random vibration table is used to perform vibration treatment in at least one direction, the CCS-specific fixture is used to fix the CCS under test, and the equipment status monitoring unit is used to acquire temperature, vibration acceleration, number of cycles, and connection status.

[0051] The broadband impedance spectroscopy testing subsystem may include a broadband impedance analyzer, a multi-channel matrix switch module, an electromagnetically shielded test chamber, low-loss test cables, and a high-precision probe test module. The broadband impedance analyzer generates AC excitation signals and measures voltage, current, and phase information; the multi-channel matrix switch module switches the channels under test according to channel mapping information; the electromagnetically shielded test chamber reduces external electromagnetic interference; and the test cables and probe test module establish the electrical connection between the impedance analyzer and the test points.

[0052] For low-impedance power circuits, a four-wire connection can be used. A four-wire connection includes a pair of current excitation terminals and a pair of voltage detection terminals. The voltage detection terminals essentially do not carry the excitation current, thus reducing the impact of the test cable and probe contact impedance on the low-impedance measurement results. For signal acquisition circuits, a two-wire or four-wire connection can be used depending on the line impedance range.

[0053] The motion control and positioning subsystem may include a three-axis servo motion platform, an automatic loading and unloading robot, an image acquisition unit, and a tooling quick-change unit. The image acquisition unit can be a CCD camera or a CMOS camera, used to acquire the positioning marks, pad edges, or conductor bus outlines of the CCS under test; the three-axis servo motion platform is used to drive the probe test module to move according to the corrected test path; the tooling quick-change unit is used to adapt to different models of CCS.

[0054] The production line docking and sorting subsystem may include a conveyor belt module, a product identification unit, a non-conforming product sorting unit, and a production line signal interaction unit. The product identification unit can read barcodes or RFID tags; the production line signal interaction unit can interact with upstream and downstream equipment through Profinet, EtherCAT, or other industrial communication protocols.

[0055] CCS inspection and processing equipment may include a processor, memory, communication interface, and human-machine interface. The memory can store product model configuration, channel definition file, stress processing parameters, impedance scanning parameters, calibration parameters, data preprocessing parameters, feature extraction model, principal component projection parameters, defect classification model, defect location model, defect quantification model, and preset grading rules.

[0056] The channel definition files, stress processing parameters, and model parameters corresponding to different product models can be saved separately. After the CCS inspection and processing equipment determines the product model based on the product identification information, it loads the corresponding configuration, thereby avoiding the direct use of the channel structure or model parameters of one product model for another.

[0057] Figure 3 This is a schematic diagram of the software functional architecture of a CCS inspection and processing device. The software functions may include a device driver and control layer, a data preprocessing and management layer, a core algorithm layer, and a human-computer interaction and application layer. The layers interact with each other according to the device status, raw data, processed data, and inspection results. The core algorithm layer includes at least the functions of impedance spectrum feature extraction, defect classification, defect location determination, quantitative defect assessment, and reliability grading.

[0058] Figure 4 This is a flowchart illustrating an integrated busbar latent defect detection method provided in an embodiment of this application. The method may include steps S101 to S104.

[0059] S101, Obtain detection task information.

[0060] The inspection task information is used to instruct multiple channels of the CCS under test to perform latent defect inspection. The inspection task information may include at least one of the following: product identification information, product model identifier, inspection scheme identifier, channel identifier, inspection time, and inspection batch.

[0061] In one embodiment, the product identification unit reads the product identification information of the CCS under test and sends it to the CCS inspection and processing equipment. The CCS inspection and processing equipment queries the product model from the manufacturing execution system or a local database based on the product identification information and loads the inspection plan according to the product model. In another embodiment, the operator selects the product model and inspection plan through a human-machine interface.

[0062] The channels under test may include power loop channels and signal acquisition loop channels. Each channel under test is defined by an electrical connection path between at least two test points. The test task information may instruct testing of all channels or testing of a predetermined set of channels.

[0063] S102, determine the channel mapping information corresponding to multiple channels to be tested according to the detection task information, and obtain the broadband impedance spectrum data of the CCS under test after preset stress processing and according to the channel mapping information.

[0064] The CCS testing and processing equipment determines the corresponding channel definition file based on the product model identifier in the testing task information, reads the start and end test points from the channel definition file based on the identifier of the channel to be tested, and forms channel mapping information.

[0065] Before acquiring broadband impedance spectrum data, the CCS under test is subjected to preset stress treatment according to the stress treatment parameters corresponding to the product model. The preset stress treatment is used to cause a detectable change in the contact state or equivalent electrical parameters of existing latent defects without causing new explicit damage to the CCS under test that meets the quality requirements.

[0066] After completing the preset stress treatment, the channels to be tested are connected according to the channel mapping information, and voltage and current signals are collected at multiple frequency points to obtain broadband impedance spectrum data corresponding to each channel to be tested.

[0067] S103, determine the impedance spectrum characteristic information corresponding to each channel to be detected based on broadband impedance spectrum data.

[0068] The CCS detection and processing equipment performs integrity verification, outlier identification, standardization, and noise reduction on broadband impedance spectrum data. It also combines the impedance amplitude and impedance phase angle corresponding to the same frequency point in frequency order to obtain the impedance spectrum feature map.

[0069] S104: Determine the latent defect identification result of the CCS under test based on the impedance spectrum characteristic information, and output the corresponding detection result of the CCS under test.

[0070] CCS inspection and processing equipment can extract defect feature vectors using a pre-trained feature extraction model, obtain low-dimensional defect feature vectors using principal component projection parameters, determine defect types using a defect classification model, and further determine defect location and severity using a defect localization model and a defect quantification model.

[0071] The CCS inspection and processing equipment determines the reliability level based on the results of hidden defect identification, generates an inspection report, and can send sorting control instructions to the production line sorting device.

[0072] In this embodiment, there is a sequential correlation between the detection task information, channel mapping information, broadband impedance spectrum data, impedance spectrum feature information, and latent defect identification results, which enables each detection result to be traced back to the corresponding product and the channel to be tested.

[0073] Figure 5 This document provides a flowchart of a preset stress treatment process according to an embodiment of the present application. The preset stress treatment may include steps S201 to S205.

[0074] S201, Obtain the product identification information corresponding to the CCS under test, and determine the product model based on the product identification information.

[0075] S202. Obtain the corresponding stress treatment parameters according to the product model. The stress treatment parameters may include at least one of the following: lower temperature limit, upper temperature limit, temperature change rate, high temperature holding time, low temperature holding time, number of cycles, vibration direction, vibration frequency range, vibration acceleration spectral density, and vibration duration.

[0076] Stress treatment parameters can be pre-calibrated using defect-free samples of the same model and samples with known latent defects. Specifically, candidate stress parameters are applied to multiple defect-free samples to confirm that the structure, conductivity, and insulation properties of the defect-free samples still meet product requirements after treatment. The same candidate stress parameters are applied to multiple latent defect samples, and the impedance spectrum differences between the defective and defect-free samples before and after treatment are compared. When a candidate stress parameter does not cause obvious damage to a defect-free sample and can increase the distinguishable difference between at least one latent defect and a defect-free sample, the candidate stress parameter is determined as a parameter within the safe stress window.

[0077] S203: Fix the CCS to be tested onto a special tooling fixture, and perform temperature cycling and vibration treatment according to the stress treatment parameters. Temperature cycling and vibration treatment can be performed sequentially or in combination if the equipment allows.

[0078] In one offline batch screening embodiment, the temperature cycling range can be -40℃ to 125℃, the temperature change rate can be 15℃ / min, the high-temperature end and the low-temperature end are each held for 30 minutes, and the number of cycles can be 50; the vibration treatment can use random vibration in three directions, the frequency range can be 10Hz to 2000Hz, and the vibration time can be 2 hours. The above values ​​are used to illustrate a reliability screening method and do not mean that all product models use the same parameters.

[0079] In one rapid screening embodiment, temperature variation ranges and vibration frequency bands that are sensitive to latent defects in the target can be selected from offline calibration data to reduce the number of cycles or vibration duration, and processing efficiency can be improved through batch processing or multi-station parallel processing. Specific parameters are determined by the safety stress window corresponding to the product model.

[0080] S204: During the temperature cycling and vibration processing, acquire the operating status information of the pretreatment device and the connection status information of the CCS under test. The operating status information may include the current temperature, temperature change rate, vibration acceleration, and number of cycles completed; the connection status information may include the interface connection status, channel conduction status, and tooling fixation status.

[0081] Preset processing conditions may include: the current temperature is within the allowable temperature range, the vibration acceleration is within the allowable acceleration range, the equipment has not malfunctioned, the connection between the CCS under test and the tooling remains stable, and the CCS under test has no open circuit, short circuit or visible structural damage.

[0082] When the operating status information and connection status information meet the preset processing conditions, the preset stress processing continues. When the operating status information or connection status information does not meet the preset processing conditions, the preset stress processing is paused and an error message is output. For errors that can be eliminated by re-fixing the tooling or restoring the equipment status, processing can continue after the error is eliminated; for cases where the CCS under test has experienced connection interruption or obvious damage, it can be directly determined that the CCS under test does not meet the factory requirements.

[0083] S205: When the preset stress treatment is completed and no obvious damage is found in the CCS under test, the CCS under test is identified as the CCS under test that has undergone preset stress treatment, and the product identification information, stress treatment parameters, start time, end time and treatment status are associated and stored.

[0084] To verify that the preset stress treatment will not damage products that meet quality requirements, continuity, insulation, and reference impedance spectroscopy tests can be performed on extracted, defect-free reference parts before and after treatment. When the results before and after treatment are both within the product's allowable range, the current stress treatment parameters are confirmed to be suitable for the corresponding product model.

[0085] Figure 6 This is a schematic diagram of a channel mapping and test probe path. Figure 7 This is a flowchart of a test link calibration and broadband impedance spectrum acquisition process.

[0086] like Figure 6 As shown, the CCS under test can include multiple test channels with different structures. Channel 1, channel 2, and channel 3 are used as examples in the figure. Channel 1 is defined by a starting test point, an intermediate test point, and an ending test point, and can be divided into test sub-segment A and test sub-segment B according to the conductive path between adjacent test points. Channels 2 and 3 are used to illustrate other test channels with different line directions and test point distributions.

[0087] Figure 6 The test points can be pads, busbar ends, connector terminals, via connection points, or other locations that can establish an electrical connection with the test probe. The probe test module moves to the target test point according to a planned path generated based on standard product coordinates.

[0088] The image acquisition unit acquires the positioning image of the CCS under test and identifies the actual position of the target test point. The CCS detection and processing equipment compares the actual position with the theoretical position recorded in the channel definition file, determines the translational deviation, rotational deviation, or scale deviation, and corrects the planned path accordingly to obtain the corrected test path. The probe test module connects to the corresponding test point according to the corrected test path.

[0089] S301. Based on the channel definition file corresponding to the CCS under test, determine the start and end test points for each channel to be tested, and form channel mapping information.

[0090] The channel definition file may include product model, channel identifier, channel type, coordinates of the starting test point, coordinates of the ending test point, line outline or line area, and allowable impedance range. Each channel record corresponds to one channel to be tested.

[0091] In one implementation, the channel mapping information can be represented as M={m_i|i=1,2,…,N}, where m_i includes at least the channel identifier of the i-th channel to be tested, the coordinates of the starting test point, and the coordinates of the ending test point, and N is the total number of channels to be tested. The channel mapping information may also include the coordinates of intermediate test points and multiple test sub-segments.

[0092] S302, generate the test path corresponding to the test probe based on the starting and ending test points. The image acquisition unit acquires an image of the CCS under test, and the CCS detection and processing equipment identifies the positioning marks or structural edges from the image and determines the translation, rotation angle, and scale change of the actual position relative to the standard position.

[0093] The CCS testing and processing equipment corrects the test path based on the aforementioned positional deviations, aligning the test probe with the corresponding test point. After alignment, the contact force sensor or probe displacement can be used to determine whether the test probe has reached the predetermined contact state.

[0094] S303 performs test link calibration before formal data acquisition. Test link calibration may include at least two of the following: short-circuit calibration, open-circuit calibration, and standard load calibration. Short-circuit calibration is used to obtain the series residual impedance of the test cables, matrix switches, and probes; open-circuit calibration is used to obtain the effects of parallel leakage and parasitic capacitance in the test link; standard load calibration is used to determine the gain and phase error of the test link at different frequency points.

[0095] In one optional implementation, before formal data acquisition, calibration measurement results of each channel under test are obtained at multiple frequency points under short-circuit, open-circuit, and standard load conditions.

[0096] Short-circuit calibration results are used to determine the series residual impedance introduced by the test cable, matrix switch, and test probe at the corresponding frequency point; standard load calibration results are used to determine the amplitude gain correction parameter and phase correction parameter at the corresponding frequency point based on the difference between the known complex impedance and the measured complex impedance of the standard load; open-circuit calibration results are used to determine or judge the effects of parallel leakage and parasitic capacitance in the test link.

[0097] During the formal data acquisition process, the CCS testing and processing equipment calls upon the calibration parameters associated with and saved by the test cables, matrix switch channels, test probes, and tooling fixtures to compensate for the acquired complex impedance. When the open-circuit calibration result, short-circuit calibration result, or standard load calibration result exceeds the corresponding allowable range, the corresponding frequency point is marked as invalid, or the test link calibration is re-executed.

[0098] When the scanning frequency reaches the frequency band that the test link needs to be processed according to the distributed parameter network, calibration compensation can be completed by using the port extension, fixture de-embedding, or transmission parameter conversion functions built into the impedance analysis equipment or vector network analysis equipment.

[0099] S304 sequentially switches multiple channels to be tested according to the channel mapping information, and collects the voltage signal amplitude, current signal amplitude, voltage signal phase and current signal phase corresponding to each channel to be tested at multiple frequency points.

[0100] In one embodiment, the scanning frequency range can be from 100Hz to 100MHz, with multiple frequency points set using a logarithmic sweep method. The excitation voltage can be 100mV, the integration time for each frequency point can be 10ms, and each channel can be sampled three times. For devices calibrated via test links and employing port extension, fixture de-embedding, or transmission parameter conversion, the upper limit of the scan can be extended to higher frequencies. The above parameters are used to illustrate one implementation method; the actual parameters are determined based on the channel impedance range, the target defect sensitive frequency band, and the detection cycle.

[0101] When the scanning frequency reaches a frequency band that the test cable, fixture, or channel under test can no longer be processed by the lumped parameter circuit, the CCS testing and processing equipment can obtain the scattering parameters using vector network analysis or impedance analysis equipment, eliminate the influence of cable electrical length by extending the port, eliminate the influence of parasitic parameters of probes and tooling by removing the fixture, and convert the calibrated scattering parameters into the complex impedance of the channel under test.

[0102] To improve detection efficiency, a rapid scan can be performed first on a set of defect-sensitive frequency points determined based on training samples; when the rapid scan results meet the abnormal triggering conditions, a full broadband scan can then be performed on the corresponding channel to be tested. Alternatively, multiple impedance testing units can be used to acquire data from different channels in parallel.

[0103] In one alternative implementation, for each frequency point of each channel to be tested, a broadband impedance analyzer acquires the voltage signal amplitude, current signal amplitude, voltage signal phase, and current signal phase, respectively.

[0104] The CCS detection and processing equipment determines the corresponding impedance amplitude based on the ratio of the voltage signal amplitude to the current signal amplitude, and determines the impedance phase angle based on the phase difference between the voltage signal phase and the current signal phase. The impedance amplitude and the impedance phase angle form the corresponding complex impedance.

[0105] When sampling the same channel and frequency point multiple times, the real and imaginary parts of the complex impedance obtained from each sampling can be averaged, and the average impedance amplitude and average impedance phase angle at that frequency point can be determined based on the averaged real and imaginary parts.

[0106] Impedance phase angle can be expressed in degrees or radians, but the same phase unit is used in both the model training phase and the CCS detection phase, and the phase unit used is saved in the model configuration parameters of the corresponding product model.

[0107] After calibration and compensation, the impedance amplitude data and impedance phase angle data corresponding to each channel to be tested are stored according to the channel identifier, frequency point and sampling time to form the original broadband impedance spectrum dataset.

[0108] S305 compensates the collected data based on the calibration parameters obtained during the test link calibration process, and obtains the impedance amplitude and impedance phase angle corresponding to each channel to be tested, forming broadband impedance spectrum data.

[0109] Specifically, the CCS testing and processing equipment uses short-circuit calibration results to correct the series residual impedance introduced by the test cables, matrix switches, and test probes; it uses standard load calibration results to correct the amplitude gain error and phase error corresponding to different frequency points; and it uses open-circuit calibration results to determine or correct the effects of parallel leakage and parasitic capacitance in the test link. The compensated impedance amplitude data and impedance phase angle data are associated and stored according to channel identifier, frequency point, and sampling time to form broadband impedance spectrum data.

[0110] Figure 8 This is a flowchart of an impedance spectrum data validity verification and preprocessing process.

[0111] S401, check whether the frequency point data corresponding to each channel to be tested is complete.

[0112] For each channel to be tested, count the number of frequency points that actually did not obtain valid data, and determine the proportion of the number of missing frequency points in the total number of frequency points that should be collected. This proportion is then determined as the data missing percentage of the channel to be tested.

[0113] When the percentage of missing data exceeds a preset missing threshold, the corresponding channel to be tested is identified as an abnormal channel, and the broadband impedance spectrum data corresponding to that channel is reacquired. In one embodiment, the preset missing threshold can be 5%.

[0114] S402, before determining the data distribution range of the impedance phase angle data, perform phase expansion on the impedance phase angle data according to the arrangement order of frequency points.

[0115] The impedance phase angle output by impedance testing equipment is typically located within a preset phase representation range. When the actual impedance phase angle change at adjacent frequency points crosses the boundary of this phase representation range, discontinuous changes may occur in the impedance phase angle data, such as abruptly changing from a positive value to a negative value or vice versa. This discontinuous change is due to phase entanglement caused by the phase representation method and does not necessarily indicate a sudden change in the impedance response of the channel under test.

[0116] Therefore, the CCS detection and processing equipment compares the impedance phase angles of adjacent frequency points sequentially, either from low to high or from high to low scanning frequency. When the phase difference between adjacent frequency points exceeds a preset phase jump threshold, the impedance phase angle between adjacent frequency points is restored to continuous change by adding or subtracting an integer phase from the current frequency point and subsequent impedance phase angles, thus obtaining the phase-expanded impedance phase angle data.

[0117] In the embodiment where the impedance phase angle is expressed in radians, the integer phase is 2π, and the preset phase jump threshold can be determined based on π; in the embodiment where the impedance phase angle is expressed in degrees, the integer phase is 360°, and the preset phase jump threshold can be determined based on 180°.

[0118] During phase expansion, only the representation range of the impedance phase angle is adjusted, without changing the actual impedance phase angle variation relationship reflected between adjacent frequency points. The same phase representation unit, frequency point arrangement order, and phase expansion method are used in both the model training phase and the CCS detection phase to ensure that the training samples and the data to be detected have a consistent data format.

[0119] For local phase abrupt changes that occur between adjacent frequency points, if the local phase abrupt change does not conform to the characteristics of an integer phase jump, it will not be eliminated by phase unrolling, but will be retained in the subsequent abnormal data judgment step to avoid mistaking the real local impedance phase angle change caused by the latent defect as phase entanglement.

[0120] S403, determine the data distribution range based on the mean and standard deviation of the impedance amplitude data and impedance phase angle data corresponding to the same channel to be tested, and determine candidate abnormal data based on the data distribution range.

[0121] Specifically, for each channel to be tested, the impedance amplitude sequence and the impedance phase angle sequence after phase expansion are obtained at multiple frequency points.

[0122] For the impedance amplitude sequence, the mean and standard deviation of the impedance amplitude sequence are determined respectively, and the data distribution range corresponding to the impedance amplitude data is determined based on the mean and standard deviation of the impedance amplitude sequence.

[0123] For the impedance phase angle sequence, the mean and standard deviation of the impedance phase angle sequence are determined respectively, and the data distribution range corresponding to the impedance phase angle data is determined based on the mean and standard deviation of the impedance phase angle sequence.

[0124] In one implementation, the range defined by three standard deviations above and below the mean can be used to determine the data distribution range of the corresponding data. In other implementations, the data distribution range can also be adjusted based on the statistical results of defect-free reference parts of the same product model, the measurement error of the impedance testing equipment, the repeatability error of the test link, and the minimum identifiable change corresponding to the target latent defect.

[0125] When the impedance amplitude corresponding to a frequency point exceeds the data distribution range corresponding to the impedance amplitude data, or the impedance phase angle corresponding to the frequency point exceeds the data distribution range corresponding to the impedance phase angle data, at least one of the impedance amplitude data and impedance phase angle data corresponding to the frequency point is determined as candidate abnormal data.

[0126] For impedance spectrum data in different frequency ranges within the same channel under test, the data distribution range of the corresponding frequency range can be determined separately. For example, the mean and standard deviation can be calculated separately for the low-frequency range, mid-frequency range, and high-frequency range to reduce the impact of the overall frequency variation trend of the impedance spectrum on the judgment of local abnormal data.

[0127] Candidate anomaly data only indicates that the data at the corresponding frequency point deviates from the current data distribution range, and cannot be directly determined to be random interference. These candidate anomaly data may be caused by probe contact jitter, switching transients, external electromagnetic interference, or equipment sampling errors, or they may be the actual impedance response generated by latent defects within a specific frequency range. Therefore, further determination of whether candidate anomaly data should be retained is needed through subsequent resampling stability assessments.

[0128] S404: Repeat sampling the frequency points corresponding to the candidate abnormal data multiple times, and determine whether to retain the candidate abnormal response based on the stability of the multiple repeated sampling.

[0129] Specifically, the CCS detection and processing equipment controls the impedance testing equipment to perform multiple repeated samplings on the frequency points corresponding to the candidate abnormal data while keeping the channel to be tested, the connection status of the test probe, the AC excitation parameters and environmental conditions unchanged, so as to obtain multiple repeated sampling results.

[0130] For candidate outlier data of impedance amplitude, the deviations between impedance amplitudes obtained from multiple repeated samplings are compared; for candidate outlier data of impedance phase angle, the deviations between impedance phase angles obtained from multiple repeated samplings are compared. The deviations can be represented by the difference between the maximum and minimum values, the maximum deviation relative to the mean of multiple samplings, or other parameters that characterize the degree of dispersion in the repeated sampling.

[0131] When a candidate abnormal response is determined to have repeatability stability, it is determined that the candidate abnormal response has repeatability stability when the number of times it appears stably in multiple repeated samplings reaches a preset number of repetitions and the maximum deviation between the results of multiple repeated samplings is not greater than a preset repeatability deviation threshold.

[0132] Candidate anomalous responses with repeatability stability may be caused by local impedance changes due to microcracks in the FPC circuit, poor solder joints, conductor interface oxidation, poor crimping contact, or local thinning of the circuit. Therefore, these candidate anomalous responses are retained and not used as anomalous data to be replaced.

[0133] When the number of times a candidate abnormal response appears stably in multiple repeated samplings is less than the preset number of repetitions, or the maximum deviation between multiple repeated sampling results is greater than the preset repeatability deviation threshold, it is determined that the candidate abnormal response has not appeared stably, and the corresponding frequency point data is determined as abnormal data.

[0134] In one implementation, the continuity between candidate anomalous data and adjacent frequency point data can also be used for verification. When candidate anomalous data forms continuous local peaks, local valleys, local inflection points, or continuous frequency band changes with the preceding and following adjacent frequency points, the priority of retaining the candidate anomalous response can be increased; when candidate anomalous data exhibits isolated single-point jumps and cannot be stably reproduced in repeated sampling, it can be identified as anomalous data.

[0135] Candidate anomaly data can also be compared with the frequency band response of a defect-free reference component of the same product model. When the local variation corresponding to the candidate anomaly data does not appear in the defect-free reference component but appears stably in repeated sampling of the same channel to be tested, it is retained as a possible defect feature.

[0136] Therefore, it is possible to distinguish between repeatable true impedance responses and non-repeatable random anomalous data, reducing the possibility of erroneously deleting local features corresponding to latent defects.

[0137] S405, Replace the abnormal data with frequency point data adjacent to the abnormal data to obtain effective impedance spectrum data.

[0138] For a single abnormal data point located between two valid frequency points, the frequency distance between the frequency point corresponding to the abnormal data point and the previous and next valid frequency points is determined respectively. The data corresponding to the previous and next valid frequency points are then linearly weighted according to the frequency distance to obtain the replacement data corresponding to the abnormal data.

[0139] Among them, effective frequency points that are closer to the abnormal frequency points have a larger interpolation weight, while effective frequency points that are farther from the abnormal frequency points have a smaller interpolation weight.

[0140] When the abnormal data is impedance amplitude data, the replacement impedance amplitude data is determined based on the impedance amplitude data corresponding to the adjacent effective frequency points before and after; when the abnormal data is impedance phase angle data, the replacement impedance phase angle data is determined based on the impedance phase angle data corresponding to the adjacent effective frequency points before and after phase expansion.

[0141] When using an equal-interval linear frequency sweep method for the broadband impedance spectrum, interpolation of outlier data can be performed on the actual frequency coordinates; when using a logarithmic frequency sweep method for the broadband impedance spectrum, interpolation of outlier data can be performed on the logarithmic frequency coordinates. The same frequency sweep method and interpolation method are preferably used during both the model training phase and the CCS detection phase.

[0142] When multiple consecutive adjacent frequency points are identified as abnormal data, the decision to perform piecewise interpolation or re-acquire broadband impedance spectrum data of the corresponding channel to be tested can be made based on the number of consecutive abnormal frequency points and the proportion of consecutive abnormal frequency points in all frequency points.

[0143] When the number of consecutive abnormal frequency points is less than the preset threshold for the number of consecutive abnormal points, and there are valid frequency points on both sides of the consecutive abnormal frequency points, the consecutive abnormal data can be interpolated based on the data change trend of the valid frequency points on both sides.

[0144] When the number of consecutive abnormal frequency points reaches the preset threshold for the number of consecutive abnormalities, or when the consecutive abnormal frequency points cover the main characteristic frequency band corresponding to the target defect, it is preferable to reacquire the broadband impedance spectrum data of the corresponding channel to be detected in order to avoid interpolation processing to eliminate possible defect responses.

[0145] For abnormal data located at the beginning or end of the scanning frequency band, since there is no effective frequency point on one side of the abnormal data, it is preferable to reacquire the broadband impedance spectrum data corresponding to the channel to be detected, rather than using a one-sided extrapolation method to generate replacement data.

[0146] After replacing the abnormal data, the integrity of the data at each frequency point and the continuity between the replaced data and adjacent frequency points are checked again. The impedance amplitude data and impedance phase angle data that pass the integrity and continuity checks are determined as valid impedance spectrum data.

[0147] It should be noted that S405 only replaces abnormal data that are identified in S404 and do not have repeatability stability; data that appears stably in multiple repeated samplings and is identified as a possible defective response are retained and no interpolation replacement is performed.

[0148] S406 performs standardization and noise reduction processing on the impedance amplitude data and impedance phase angle data in the effective impedance spectrum data, respectively.

[0149] Specifically, the standardization and noise reduction processes may include normalization, baseline trend determination and subtraction, wavelet decomposition and reconstruction, and smoothing.

[0150] First, the impedance amplitude data and impedance phase angle data corresponding to the same channel under test are normalized respectively.

[0151] When normalizing impedance amplitude data, the CCS detection processing device calls the lower limit and upper limit of the impedance amplitude of the training samples, which are associated with the current product model and the identifier or type of the channel to be tested. Based on the lower limit and upper limit of the impedance amplitude, the impedance amplitude data corresponding to the current channel to be tested is mapped to a preset value range.

[0152] When normalizing the impedance phase angle data, the CCS detection processing device calls the lower limit and upper limit of the impedance phase angle of the training sample, which are associated with the current product model and the identifier or type of the channel to be tested. Based on the lower limit and upper limit of the impedance phase angle, the impedance phase angle data corresponding to the current channel to be tested is mapped to a preset value range.

[0153] The lower limit of the impedance amplitude of the training sample, the upper limit of the impedance amplitude of the training sample, the lower limit of the impedance phase angle of the training sample, and the upper limit of the impedance phase angle of the training sample are determined according to the training sample during the model training stage and are associated with and saved with the product model, channel identifier or channel type and model version.

[0154] The normalization parameters saved during the model training phase are directly called during the detection phase of the CCS to be tested, without re-determining the normalization parameters for each CCS to be tested, so that the training samples and the data to be tested are on the same data scale.

[0155] When the upper limit of the training samples is the same as the lower limit of the training samples, or when the difference between the two is less than the preset stability threshold, the normalization operation can be stabilized by using the preset minimum positive number to avoid numerical anomalies.

[0156] In another embodiment, impedance amplitude data and impedance phase angle data for each channel to be tested can be normalized individually. To avoid weakening the overall impedance amplitude increase characteristics caused by local thinning of the line, poor crimping contact, or poor soldering of the pads due to single-channel normalization, at least one of the following can be retained as an auxiliary absolute feature: low-frequency impedance before normalization, average impedance of the frequency band, impedance variation range, or impedance integral value. This feature can then be combined with the normalized impedance spectrum features in subsequent feature extraction processes.

[0157] Secondly, the corresponding baseline trends are determined based on the impedance amplitude data and impedance phase angle data after normalization, and the baseline trends are subtracted from the normalized impedance amplitude data and impedance phase angle data.

[0158] The baseline trend is used to characterize the slowly varying components formed by the line length, conductor material, line routing, connection structure, and parasitic parameters of the test link of the channel under test.

[0159] In one implementation, a low-order polynomial fitting method can be used to fit the normalized impedance amplitude data and impedance phase angle data, and the slowly changing curve obtained by fitting can be determined as the baseline trend.

[0160] In another embodiment, the baseline trend can be determined by using a sliding window average, a sliding window midpoint, or a wavelet low-frequency approximation component.

[0161] The parameters for determining the baseline trend can be predetermined based on the product model, the type of channel to be tested, the number of frequency points, and the impedance spectrum variation law of the defect-free reference part, and then associated with and saved with the corresponding product model.

[0162] By subtracting the corresponding baseline trend from the normalized impedance amplitude and impedance phase angle data, local responses that may be related to latent defects, such as local peaks, local valleys, local inflection points, and frequency band changes, can be highlighted.

[0163] Next, wavelet decomposition and reconstruction are performed on the impedance amplitude data and impedance phase angle data after removing the baseline trend, respectively, to remove random noise.

[0164] Wavelet decomposition decomposes impedance amplitude data and impedance phase angle data into low-frequency approximate components and high-frequency detail components corresponding to multiple scales, respectively. Random noise is usually mainly distributed in the high-frequency detail components. Therefore, thresholding can be performed on the high-frequency detail components, and wavelet reconstruction can be performed based on the processed high-frequency detail components and low-frequency approximate components.

[0165] In one implementation, the db4 wavelet can be used to perform multi-level decomposition of the impedance amplitude data and impedance phase angle data, and a soft thresholding method can be used to process the high-frequency detail components. The wavelet type, number of decomposition levels, and thresholding parameters are determined based on the number of frequency points, the sampling noise level, and the minimum frequency band feature width corresponding to the target defect.

[0166] When determining the number of wavelet decomposition layers and threshold processing parameters, a defect-free reference and training samples with known defects can be used for verification, so that the denoised data can reduce random spikes while retaining the local impedance response caused by latent defects.

[0167] Finally, the impedance amplitude data and impedance phase angle data after wavelet decomposition and reconstruction are smoothed to preserve the corresponding characteristic change points.

[0168] The smoothing process can employ Savitzky-Golay filtering, moving average filtering, or other smoothing methods that can reduce local random fluctuations.

[0169] In one implementation, Savitzky-Golay filtering is used to process the reconstructed impedance amplitude and impedance phase angle data. The filter window length and polynomial order are determined based on the number of frequency points and the minimum feature width corresponding to the target latent defect.

[0170] The length of the filtering window is preferably smaller than the number of frequency points contained in the characteristic frequency band corresponding to the target defect, so as to avoid excessive smoothing of local peaks, local valleys or local inflection points caused by latent defects.

[0171] After smoothing, impedance amplitude data and impedance phase angle data that have undergone normalization and noise reduction are obtained. This data is then used in step S407 to combine the data according to the frequency points in sequence, forming the impedance spectrum feature map corresponding to each channel to be detected.

[0172] S407, according to the frequency point arrangement order, combines the impedance amplitude data and impedance phase angle data of the same channel to be tested to form an impedance spectrum feature map Xi. Xi can be represented as an M-row, 2-column data structure, where the first column is the impedance amplitude feature and the second column is the impedance phase angle feature.

[0173] The impedance spectrum feature maps corresponding to all channels to be detected constitute a standardized feature dataset X={Xi|i=1,2,…,N}. This dataset serves as the input to the feature extraction model.

[0174] Figure 9 This is a flowchart illustrating the process of model training, defect classification, location determination, and quantitative evaluation. This example explains how the feature extraction model, principal component projection parameters, defect classification model, defect location model, and defect quantification model are obtained and used for detection.

[0175] Model training is conducted for specific product models. Training samples can include defect-free samples, FPC circuit microcrack samples, solder pad cold solder joint samples, conductor interface oxidation samples, crimp contact poor samples, and circuit local thinning samples.

[0176] Defect-free samples can be selected from CCSs that have passed manufacturing inspection and have been confirmed to be free of abnormalities by microscopic observation, low-resistivity testing, and functional testing. Defective samples can be obtained from defective parts that are naturally formed during the manufacturing process, or they can be prepared through controlled methods.

[0177] Microcrack samples of FPC circuits can be formed with microcracks of different lengths under conditions of incomplete conductor cutting, and the crack length can be measured by microscopy or industrial microscopic imaging; solder pad cold solder joint samples can be formed with different effective solder areas by controlling the solder coverage area or soldering energy, and the proportion of cold solder joint area can be obtained by cross-sectional measurement or image measurement; conductor interface oxide samples can be formed in a controlled temperature and humidity environment, and the presence of oxide layer at the connection interface can be confirmed by microscopic imaging, surface composition detection or cross-sectional observation, and the contact impedance increment can be obtained by four-wire measurement; poor crimp contact samples can be formed by controlling crimping force, crimping height or contact area, and the corresponding contact impedance can be recorded; local thinning samples of circuits can be formed by controlled etching or mechanical thinning, and the thinning ratio can be obtained by cross-sectional measurement.

[0178] To enable the model to identify different levels of severity, multiple severity levels can be set for each defect type. For example, multiple crack lengths can be set for microcracks in FPC circuits, multiple effective solder area percentages can be set for cold solder pads, and multiple thinning percentages can be set for localized thinning of circuits. The specific levels are determined based on the product design dimensions and quality standards.

[0179] Each training sample undergoes the same pre-defined stress treatment, test link calibration, channel mapping, impedance scanning, and data preprocessing procedures as actual tests to generate an impedance spectrum feature map. Training samples must include at least the product model, channel identifier, defect type, and true severity label. When location training is required, defect channel, defect region, or two-dimensional coordinate labels are also included.

[0180] Training samples are preferably divided into training, validation, and test sets according to CCS product entities. This ensures that data from different channels of the same entity does not simultaneously enter the training and test sets, reducing data leakage. The training set is used to learn model parameters, the validation set is used to select model structure and hyperparameters, and the test set is used to evaluate the final model.

[0181] In one alternative implementation, the feature extraction model can employ a convolutional neural network. The impedance spectrum feature map is arranged according to frequency points, and the impedance amplitude feature and impedance phase angle feature are used as corresponding input features.

[0182] The convolutional layer extracts local features of impedance amplitude and impedance phase angle changes at adjacent frequency points along the frequency point arrangement direction to obtain local inflection points, local peaks, frequency band change trends, and joint change features between amplitude and phase.

[0183] Pooling layers compress the convolutional outputs within adjacent frequency ranges, preserving significant responses while reducing feature dimensionality and the impact of random perturbations. After convolution and pooling, the corresponding features are flattened and fully connected mapped to obtain the defect feature vectors for the corresponding training samples or channels to be detected.

[0184] The number of convolutional layers, kernel size, pooling window, pooling stride, number of fully connected layers, and dimension of defect feature vectors are determined based on the number of training samples, the number of frequency points, and the evaluation results of the validation set.

[0185] During the principal component analysis training phase, defect feature vectors corresponding to each training sample in the training set are collected, and the training mean vector is determined based on the defect feature vectors.

[0186] The training mean vector is subtracted from the defect feature vector of each training sample to obtain the decentralized training feature vector, and the training covariance matrix is ​​determined based on the decentralized training feature vector.

[0187] The training covariance matrix is ​​decomposed into eigenvalues. Multiple eigenvectors whose cumulative contribution rate satisfies the preset contribution condition are selected in descending order of eigenvalues ​​to form the principal component projection matrix.

[0188] The training mean vector and principal component projection matrix are associated and saved with the product model and model version. During the CCS detection phase, the defect feature vector is first decentered using the training mean vector, and then the decentered defect feature vector is projected onto a low-dimensional feature space using the principal component projection matrix to obtain a low-dimensional defect feature vector.

[0189] Therefore, the training samples and the CCS to be tested use the same feature center and principal component coordinate system, avoiding changes in the feature space caused by recalculating the principal components for each CCS to be tested.

[0190] The defect classification model can employ support vector machines and a one-to-one multi-classification strategy. For the six defect categories—no defects, microcracks in FPC circuits, poor solder joints, conductor interface oxidation, poor crimping contact, and local thinning of circuits—binary sub-classifiers can be trained for any two categories, resulting in a total of fifteen binary sub-classifiers.

[0191] For linearly inseparable low-dimensional defect features, a radial basis function kernel can be used to map these features to a high-dimensional feature space. The penalty parameters and kernel function parameters of the support vector machine are determined based on the evaluation results of the validation set.

[0192] During detection, each binary sub-classifier outputs its classification result, and the final defect type is determined by voting. A validation set can also be used to calibrate the decision results of each binary sub-classifier to obtain the classification probability corresponding to each defect category, and the highest classification probability is determined as the classification confidence level.

[0193] Defect localization and defect quantification models can employ support vector regression (SVR) models. SVR models use ε-insensitive loss, meaning that when the deviation between the predicted result and the true continuous value is within a preset ε-insensitive interval, no penalty is imposed on the deviation; when the prediction deviation exceeds the ε-insensitive interval, the regression loss is determined based on the magnitude of the excess.

[0194] During model training, model complexity is constrained by regularization, and the relationship between model complexity and training bias is adjusted by regression penalty coefficients. The regression penalty coefficient, the width of the ε-insensitive interval, and the kernel function parameters are determined based on the validation set evaluation results.

[0195] For determining the two-dimensional location, regression models for the horizontal and vertical axes can be trained separately; for determining the severity, corresponding quantitative defect models can be trained for the microcrack length, the proportion of incomplete solder area, the increase in contact impedance, and the proportion of line thinning.

[0196] After the model training is completed, the location prediction error and severity prediction error are evaluated using an independent test set that was not involved in the training. The model is only used for CCS detection when the corresponding evaluation results meet the model acceptance conditions for the product model.

[0197] After model training, the defect classification confusion, channel or region localization accuracy, coordinate localization error, and severity regression error are evaluated on an independent test set. Model acceptance criteria are determined by the detection accuracy requirements corresponding to the product model. Only when the evaluation results on the independent test set meet these criteria will the feature extraction model, training mean vector, principal component projection matrix, defect classification model, localization model, quantitative model, and standardized parameters be compiled into a model configuration package and saved in association with the product model and version number.

[0198] During model training and testing Figure 9 The feature extraction model, training mean vector, principal component projection matrix, classification model, localization model, and categorized quantitative model shown are used in conjunction with the same product model and model version. The test set samples are from CCS entities that did not participate in model training and cover all defect types and multiple severity levels. For classification results, the number of correctly identified items and the destination of incorrectly identified items are recorded; for localization and quantitative results, the location error and severity prediction error are recorded. The verification records are saved in association with the version information of the model configuration package.

[0199] S501 inputs the impedance spectrum feature map corresponding to each channel to be tested into the feature extraction model corresponding to the product model, performs convolution and pooling processing to obtain the defect feature vector.

[0200] S502, Read the training mean vector and principal component projection matrix corresponding to the product model, and then follow the formula... The low-dimensional defect feature vector is obtained.

[0201] S503 inputs the low-dimensional defect feature vector into the defect classification model. Fifteen binary sub-classifiers make judgments, and the model determines whether there are no defects, microcracks in the FPC circuit, poor solder joints, conductor interface oxidation, poor crimping contact, or local thinning of the circuit through voting.

[0202] Channels not classified as "defect-free" are identified as defective channels. The defect classification results include channel identifier, defect type, classification confidence level, and model version.

[0203] S504, determine the location of latent defects.

[0204] In the basic localization implementation, the corresponding channel mapping information is queried based on the defect channel identifier to obtain the starting test point, the ending test point, and the line area, and the line area is determined as the location feature information; the localization model is then input. Input the defect localization model, or directly determine the channel region based on the channel mapping information. In this case, the location of the latent defect indicates the channel or channel region where the defect is to be detected.

[0205] In the refined positioning implementation, the channel definition file also includes multiple intermediate test points, dividing a line into multiple test sub-segments. The CCS detection and processing equipment acquires impedance spectrum data between adjacent test points and between combinations of multiple overlapping test points, and uses the segment identifier, center coordinates, start and end coordinates or line outline of each test sub-segment as position features, which are then input into the defect location model along with the corresponding impedance spectrum features.

[0206] The target test section can be determined based on the amplitude and phase deviation of each test section within the defect-sensitive frequency band. For example, for the s-th test section, the section deviation index Ds can be determined: ; In the formula, Ds represents the deviation index of the s-th test sub-segment; Ω represents the set of sensitive frequency points corresponding to the target defect; and These represent the standardized impedance magnitude and impedance phase angle of the test sub-section at the k-th frequency point, respectively. and These represent the corresponding data for a defect-free reference part of the same model; and These represent the amplitude and phase weights, respectively; s represents the test sub-segment number; and k represents the frequency point number. The weights can be determined and normalized based on the inter-class differences of the training samples at the corresponding frequency points. The same weights can be used when no prior weights are set.

[0207] The test sub-segment with the largest deviation index is identified as the candidate target segment. When the deviation index of the candidate target segment is greater than the sum of the mean deviation index of the adjacent test sub-segments and a preset multiple of the standard deviation, the candidate target segment is identified as the target test sub-segment; otherwise, the location result is marked as a re-inspection. The center coordinates of the target test sub-segment can be used as the initial position, or multiple segment deviations, low-dimensional defect feature vectors, and segment location features can be input into a location regression model to obtain a two-dimensional position.

[0208] S505 determines the severity of latent defects. The CCS detection and processing equipment determines the target defect quantitative model from multiple defect quantitative models based on the defect type.

[0209] When the defect type is FPC line microcrack, the microcrack length model is called; when the defect type is solder pad cold solder joint, the cold solder joint area model is called; when the defect type is conductor interface oxidation or poor crimp contact, the contact impedance increment model is called; when the defect type is line local thinning, the thinning ratio model is called.

[0210] The low-dimensional defect feature vector corresponding to the defect channel is input into the target defect quantitative model to obtain the microcrack length, cold solder joint area ratio, contact impedance increment, or line thinning ratio corresponding to the defect type. For product models that do not have sufficiently fine training labels, only severity levels such as minor, moderate, and severe can be output.

[0211] S506 associates defect type, defect channel, defect location, severity parameter, classification confidence level and model version to form the hidden defect identification result.

[0212] When there are multiple defect channels in the same CCS under test, the latent defect identification results of each defect channel are generated, and a defect set of the CCS under test is formed.

[0213] Figure 10 This is a flowchart for reliability grading, report generation, and production line sorting.

[0214] S601 determines the reliability level of the CCS under test based on the latent defect identification results and preset grading rules. The preset grading rules are associated with the product model, defect type, and product technical specifications.

[0215] When all channels under inspection are found to be defect-free and the classification confidence level meets the preset requirements, the CCS under inspection is determined to be of the acceptable level. When at least one channel under inspection is identified as having a latent defect, but the corresponding severity is below the non-acceptance threshold, or the classification confidence level is below the preset confidence threshold, the CCS under inspection is determined to be of the re-inspection level. When the severity of at least one latent defect reaches the corresponding non-acceptance threshold, the CCS under inspection is determined to be of the non-acceptance level. When multiple defects exist simultaneously, the final reliability level is determined according to the highest risk level.

[0216] For product models that do not output continuous severity parameters, a grading rule can be established based on defect type, defect channel importance, and severity level. When multiple defects exist simultaneously, the final reliability level of the CCS under test can be determined according to the highest risk level.

[0217] In a non-limited example, microcracks shorter than the first re-inspection threshold can be considered for re-inspection, while those reaching the first non-conformance threshold are considered non-conformance. The percentage of cold solder joints, the increase in contact impedance, and the proportion of thinned circuitry are each assigned a re-inspection threshold and a non-conformance threshold corresponding to the product model. Specific thresholds are determined by product technical specifications and sample failure verification, and standards unrelated to CCS are not used as the sole grading basis.

[0218] S602, obtain the product identification information corresponding to the CCS under test, and associate and store the reliability level, latent defect identification results, broadband impedance spectrum data, preset stress treatment records, calibration records, impedance scan parameters and model version with the product identification information.

[0219] Raw broadband impedance spectroscopy data can be stored in a time-series database or a file storage system; product identification information, testing parameters, defect identification results, reliability levels, and equipment logs can be stored in a relational database. Data from the entire testing process can be retrieved using product identification information.

[0220] S603 generates an inspection report based on the reliability level and latent defect identification results. The inspection report may include product model, product identification information, inspection time, stress treatment parameters, impedance scanning parameters, impedance spectrum, defect path, defect type, defect location, severity, classification confidence level, reliability level, and model version.

[0221] S604 generates sorting control instructions based on the reliability level and sends these instructions to the production line sorting device. The production line sorting device then transports the CCS samples to be tested to the qualified product area, re-inspection area, or unqualified product area according to the sorting control instructions.

[0222] For re-inspection levels, the test report can provide the channel identification, test points, and defect types that require re-inspection, enabling operators to perform microscopic observation, local X-ray, four-wire retesting, or other re-inspections in the target area.

[0223] The following example uses a CCS under test, including a power loop and an FPC signal acquisition loop, to illustrate the testing process. This example is used to illustrate the data relationships between the steps and is not intended to limit the scanning parameters or model structure.

[0224] The CCS testing and processing equipment reads the product identification information, determines the product model to be Model A, and loads the corresponding channel definition file, stress processing parameters, standardization parameters, and model configuration package for Model A. The channel definition file records the starting test points, ending test points, and at least some intermediate test points for the channels to be tested.

[0225] The CCS under test was fixed in a dedicated fixture, and temperature cycling and three-directional random vibration treatments were performed according to the safety stress window corresponding to model A. Temperature, vibration acceleration, and connection status were continuously recorded during the treatment. After treatment, the CCS under test showed no open circuits, short circuits, or visible structural damage.

[0226] Before impedance spectroscopy acquisition, short-circuit calibration and standard load calibration are performed on the test link to obtain the calibration parameters corresponding to each matrix switch channel. The image acquisition unit acquires the image of the CCS under test and corrects the probe test path.

[0227] The channels to be tested are switched sequentially according to the channel mapping information. Repeated sampling is performed at multiple predetermined frequency points to calculate the average complex impedance and obtain the impedance amplitude and impedance phase angle. Channels with a missing frequency point ratio exceeding a threshold are re-sampled.

[0228] The effective impedance spectrum data is normalized using standardized parameters determined by the training samples of Model A, and wavelet denoising and smoothing are performed to form the impedance spectrum feature map corresponding to each channel.

[0229] Input the impedance spectrum feature map into the feature extraction model corresponding to model A to obtain the defect feature vector; obtain the low-dimensional defect feature vector based on the training mean vector and principal component projection matrix corresponding to model A; and input it into the defect classification model.

[0230] Assuming channel 37 is classified as a solder pad cold solder joint, and other channels are classified as defect-free, then channel 37 is determined as a defective channel. The CCS inspection and processing equipment obtains the location features of the solder pad connection area based on the channel mapping information corresponding to channel 37. It inputs the low-dimensional defect feature vector, channel identifier, and location features into the defect localization model to obtain the corresponding solder pad area. Furthermore, based on the solder pad cold solder joint defect type, it calls the cold solder joint area model to obtain the predicted value of the cold solder joint area percentage.

[0231] The CCS inspection and processing equipment determines the reliability level based on the cold solder joint classification threshold corresponding to model A, generates an inspection report including the 37th channel, solder pad area, predicted value of cold solder joint area ratio, and impedance spectrum, and sends corresponding sorting control instructions to the production line sorting device.

[0232] In another example, when the defect classification model gives close decision distances for two defect categories in a channel and the classification confidence level is lower than the preset confidence threshold, the CCS to be tested can be determined as a re-inspection level even if the severity prediction value does not reach the non-compliance threshold, so as to reduce the risk that uncertain identification results will directly enter the qualified product area.

[0233] In another specific identification implementation, multiple impedance spectrum feature templates can be established based on defect-free reference parts of the same model and samples of different defect types. The impedance amplitude and impedance phase angle features of the channel to be detected at defect-sensitive frequency points are extracted, and the feature distances between the channel to be detected and the multiple impedance spectrum feature templates are calculated. When the minimum feature distance is less than the identification threshold of the corresponding template, the defect type corresponding to that template is determined as the defect type of the channel to be detected. The identification threshold is determined based on verification samples.

[0234] In implementations requiring deep feature recognition, one-dimensional or two-dimensional convolutional networks can be used to extract impedance spectrum features, and support vector machines can be used to complete defect classification. Template matching and convolutional feature extraction methods constitute specific implementation paths for determining the latent defect recognition results based on impedance spectrum feature information.

[0235] The preset stress treatment in this application is not limited to a fixed combination of parameters. Temperature cycling and vibration treatment are the key implementation methods disclosed in this specification; under the condition that it can be verified through the safety stress window corresponding to the product model, does not cause obvious damage to qualified products, and can cause changes in impedance response of existing latent defects, the temperature range, vibration frequency band and treatment sequence can be adjusted.

[0236] The defect location described in this application can be at least a defect channel or a channel region. Only when the channel definition file contains intermediate test points, obtains multi-segment impedance spectra, and has location-labeled training samples, will the test segment or two-dimensional coordinates be further output, thereby ensuring that the granularity of the location output matches the actual acquired information.

[0237] The severity of defects described in this application can be a continuous physical quantity or a discrete level. Continuous physical quantities are output by a quantitative model for the corresponding defect type; when training labels are insufficient or product requirements only require grading, discrete levels such as minor, moderate, and severe can be output.

[0238] It should be noted that this application also includes a related battery cell connection component testing device, including a processor and a memory, wherein the memory stores a testing processing program, and when the testing processing program is executed by the processor, the processor: Obtain detection task information, which is used to instruct multiple channels of the CCS under test to perform latent defect detection. Based on the detection task information, determine the channel mapping information corresponding to the multiple channels to be detected, and obtain the broadband impedance spectrum data of the CCS under test after preset stress processing, collected according to the channel mapping information; Based on the broadband impedance spectrum data, the impedance spectrum feature information corresponding to each channel to be tested is determined. The impedance spectrum feature information is used to characterize the impedance amplitude change and impedance phase angle change of the channel to be tested at different frequencies. Based on the impedance spectrum characteristic information, the latent defect identification result of the CCS under test is determined, and the detection result corresponding to the CCS under test is output.

[0239] The various embodiments of this application can be combined with each other without conflict. Equivalent substitutions or conventional adjustments made by those skilled in the art based on this application's description should all fall within the protection scope of this application.

Claims

1. An integrated method for detecting latent defects in busbars, characterized in that, Applied to CCS detection and processing equipment, the method includes: Obtain detection task information, which is used to instruct multiple channels of the CCS under test to perform latent defect detection. Based on the detection task information, determine the channel mapping information corresponding to the multiple channels to be detected. Before acquiring broadband impedance spectrum data, the CCS under test is subjected to preset stress treatment according to the stress treatment parameters corresponding to the product model of the CCS under test. The preset stress treatment is used to cause a detectable change in the contact state or equivalent electrical parameters of existing latent defects without causing new explicit damage to the CCS under test that meets the quality requirements. The stress treatment parameters are determined by the safe stress window corresponding to the product model. The safe stress window refers to the stress range that has been verified not to cause new explicit damage to defect-free CCS, but can increase the impedance response difference of existing latent defects. The determination of the safe stress window includes: applying candidate stress parameters to multiple defect-free samples to confirm that the structure, conductivity and insulation performance of the defect-free samples still meet the product requirements after treatment; applying the candidate stress parameters to multiple latent defect samples and comparing the impedance spectrum difference between the latent defect samples and the defect-free samples before and after treatment; when the candidate stress parameters do not cause explicit damage to the defect-free samples and can increase the distinguishable difference between at least one latent defect and the defect-free samples, the candidate stress parameters are determined as parameters within the safe stress window. Before acquiring the broadband impedance spectrum data, calibration measurement results are obtained for each of the multiple channels under test in short-circuit, open-circuit, and standard load states at multiple frequency points. Specifically, the calibration measurement results in the short-circuit state are used to determine the series residual impedance introduced by the test cable, matrix switch, and test probe at the corresponding frequency points; the calibration measurement results in the standard load state are used to determine the amplitude gain correction parameter and phase correction parameter at the corresponding frequency points based on the difference between the known complex impedance and the measured complex impedance of the standard load; and the calibration measurement results in the open-circuit state are used to determine or judge the effects of parallel leakage and parasitic capacitance in the test link. The broadband impedance spectrum data of the CCS under test is acquired after the preset stress treatment and collected according to the channel mapping information. During the acquisition of the broadband impedance spectrum data, the calibration parameters associated with the test cable, matrix switch channel, test probe and tooling fixture are called to compensate the acquired complex impedance. When the calibration measurement results under the open circuit state, the short circuit state or the standard load state exceed the corresponding allowable range, the corresponding frequency point is marked as invalid, or the test link calibration is re-executed. The broadband impedance spectrum data is validated to obtain valid impedance spectrum data; The impedance amplitude data and impedance phase angle data in the effective impedance spectrum data are respectively standardized and denoised. According to the arrangement order of the frequency points, the impedance amplitude data and impedance phase angle data that have undergone the standardization and noise reduction processes are combined accordingly to obtain the impedance spectrum feature map corresponding to each channel to be detected. The impedance spectrum feature map is determined as the impedance spectrum feature information, which is used to characterize the impedance amplitude change and impedance phase angle change of the channel to be detected at different frequencies. Based on the impedance spectrum characteristic information, the latent defect identification result of the CCS under test is determined, and the detection result corresponding to the CCS under test is output.

2. The method as described in claim 1, characterized in that, The step of validating the broadband impedance spectrum data to obtain valid impedance spectrum data includes: Check whether the frequency point data corresponding to each channel to be tested is complete. Based on the proportion of missing frequency point data in the total frequency point data of the corresponding channel to be tested, determine whether to reacquire the broadband impedance spectrum data corresponding to the channel to be tested. The data distribution range is determined based on the mean and standard deviation of the impedance amplitude data or impedance phase angle data corresponding to the same channel to be tested. Frequency points that exceed the data distribution range and do not appear stably in multiple repeated samplings are identified as abnormal data. The abnormal data is replaced by frequency point data adjacent to the abnormal data to obtain the effective impedance spectrum data.

3. The method as described in claim 2, characterized in that, The standardization and noise reduction processes performed on the impedance amplitude data and impedance phase angle data in the effective impedance spectrum data include: The impedance amplitude data and impedance phase angle data corresponding to the same channel under test are normalized respectively; Based on the impedance amplitude data and impedance phase angle data that have undergone the normalization process, a baseline trend is determined and the baseline trend is subtracted. Wavelet decomposition and reconstruction are then performed on the impedance amplitude data and impedance phase angle data after subtracting the baseline trend to remove random noise. The impedance amplitude data and impedance phase angle data after wavelet decomposition and reconstruction are smoothed to retain the corresponding characteristic change points.

4. The method according to any one of claims 2-3, characterized in that, The step of determining the latent defect identification result of the CCS under test based on the impedance spectrum characteristic information includes: The impedance spectrum feature map corresponding to each channel to be detected is input into a pre-trained feature extraction model. The impedance spectrum feature map is then subjected to convolution and pooling to obtain the defect feature vector of the corresponding channel to be detected. The defect feature vectors corresponding to each channel to be detected are reduced in dimensionality to obtain the corresponding low-dimensional defect feature vectors. The low-dimensional defect feature vector is input into a pre-trained defect classification model to determine the defect type corresponding to each channel to be detected. The channels to be tested that are not of the defect-free type are identified as defective channels, and the latent defect identification results are formed according to the defect type corresponding to each channel to be tested.

5. The method as described in claim 4, characterized in that, The step of reducing the dimensionality of the defect feature vectors corresponding to each channel to be detected to obtain the corresponding low-dimensional defect feature vectors includes: Obtain the product model corresponding to the CCS to be tested, and obtain the training mean vector and principal component projection matrix that are determined during the model training process and associated with the product model. The defect feature vectors corresponding to each channel to be detected are decentered based on the training mean vector. Based on the principal component projection matrix, the defect feature vector that has undergone the decentralization process is projected onto a low-dimensional feature space to obtain the low-dimensional defect feature vector. The defect types include no defects, microcracks in FPC circuits, poor soldering of pads, oxidation of conductor interfaces, poor crimping contact, and local thinning of circuits.

6. The method as described in claim 5, characterized in that, The step of determining the latent defect identification result of the CCS under test based on the impedance spectrum characteristic information further includes: The location feature information is determined based on the channel mapping information corresponding to the defective channel, and the location feature information is used to characterize the channel region or test sub-segment corresponding to the defective channel. The low-dimensional defect feature vector corresponding to the defect channel, the defect channel identifier, and the location feature information are input into a pre-trained defect localization model to obtain the location of the latent defect in the CCS to be tested. Based on the defect type, a target defect quantitative model is determined from multiple defect quantitative models, and the low-dimensional defect feature vector corresponding to the defect channel is input into the target defect quantitative model to obtain the severity of the latent defect. The latent defect identification result is formed by associating the defect type, the location of the latent defect, and the severity. The severity level is defined as the length of the microcrack, the percentage of the area of ​​the poor solder joint, the increase in contact impedance, or the percentage of the thinning of the circuit, which are all related to the defect type.

7. The method as described in claim 1, characterized in that, Before acquiring the broadband impedance spectrum data, the method further includes: Obtain the product identification information corresponding to the CCS under test, and determine the product model corresponding to the CCS under test based on the product identification information; Obtain the corresponding stress treatment parameters based on the product model; According to the stress treatment parameters, the pretreatment device is controlled to perform temperature cycling and vibration treatment on the CCS under test, so that the latent defects in the CCS under test produce corresponding impedance response changes. During the temperature cycling process and the vibration process, the operating status information of the pretreatment device and the connection status information of the CCS under test are acquired. If the operating status information and the connection status information meet the preset processing conditions, the CCS to be tested that has completed the temperature cycle processing and the vibration processing is identified as the CCS to be tested that has undergone the preset stress processing, and the product identification information is associated with the corresponding stress processing record.

8. The method as described in claim 1, characterized in that, The step of determining the channel mapping information corresponding to the plurality of channels to be detected based on the detection task information, and acquiring broadband impedance spectrum data collected according to the channel mapping information, includes: Based on the channel definition file corresponding to the CCS under test, determine the start test point and end test point corresponding to each channel to be tested, and form the channel mapping information. Generate the test path corresponding to the test probe based on the start test point and end test point of each channel to be tested; The image localization result of the CCS under test is obtained, the test path is corrected according to the image localization result, and the test probe is controlled to connect to the corresponding test point according to the corrected test path. According to the channel mapping information, the multiple channels to be detected are switched sequentially, and the voltage signal amplitude, current signal amplitude, and phase difference between the voltage signal and the current signal are collected at multiple frequency points for each channel to be detected. The impedance amplitude of each channel to be detected at the multiple frequency points is determined based on the ratio of the voltage signal amplitude to the current signal amplitude, and the phase difference is determined as the impedance phase angle to form the broadband impedance spectrum data.

9. A battery cell connection assembly testing device, used to implement the method of claim 1, characterized in that, Includes a processor and a memory, wherein the memory stores a detection processing program, and when the detection processing program is executed by the processor, the processor: Obtain detection task information, which is used to instruct multiple channels of the CCS under test to perform latent defect detection. Determine the channel mapping information corresponding to the plurality of channels to be detected based on the detection task information; Before acquiring broadband impedance spectrum data, the CCS under test is subjected to preset stress treatment according to the stress treatment parameters corresponding to the product model of the CCS under test. The preset stress treatment is used to cause a detectable change in the contact state or equivalent electrical parameters of existing latent defects without causing new explicit damage to the CCS under test that meets the quality requirements. The stress treatment parameters are determined by the safe stress window corresponding to the product model. The safe stress window refers to the stress range that has been verified not to cause new explicit damage to defect-free CCS, but can increase the impedance response difference of existing latent defects. The determination of the safe stress window includes: applying candidate stress parameters to multiple defect-free samples to confirm that the structure, conductivity and insulation performance of the defect-free samples still meet the product requirements after treatment; applying the candidate stress parameters to multiple latent defect samples and comparing the impedance spectrum difference between the latent defect samples and the defect-free samples before and after treatment; when the candidate stress parameters do not cause explicit damage to the defect-free samples and can increase the distinguishable difference between at least one latent defect and the defect-free samples, the candidate stress parameters are determined as parameters within the safe stress window. Before acquiring the broadband impedance spectrum data, calibration measurement results are obtained for each of the multiple channels under test in short-circuit, open-circuit, and standard load states at multiple frequency points. Specifically, the calibration measurement results in the short-circuit state are used to determine the series residual impedance introduced by the test cable, matrix switch, and test probe at the corresponding frequency points; the calibration measurement results in the standard load state are used to determine the amplitude gain correction parameter and phase correction parameter at the corresponding frequency points based on the difference between the known complex impedance and the measured complex impedance of the standard load; and the calibration measurement results in the open-circuit state are used to determine or judge the effects of parallel leakage and parasitic capacitance in the test link. The broadband impedance spectrum data of the CCS under test is acquired after the preset stress treatment and collected according to the channel mapping information. During the acquisition of the broadband impedance spectrum data, the calibration parameters associated with the test cable, matrix switch channel, test probe and tooling fixture are called to compensate the acquired complex impedance. When the calibration measurement results under the open circuit state, the short circuit state or the standard load state exceed the corresponding allowable range, the corresponding frequency point is marked as invalid, or the test link calibration is re-executed. The broadband impedance spectrum data is validated to obtain valid impedance spectrum data; The impedance amplitude data and impedance phase angle data in the effective impedance spectrum data are respectively standardized and denoised. According to the arrangement order of the frequency points, the impedance amplitude data and impedance phase angle data that have undergone the standardization and noise reduction processes are combined accordingly to obtain the impedance spectrum feature map corresponding to each channel to be detected. The impedance spectrum feature map is determined as the impedance spectrum feature information, which is used to characterize the impedance amplitude change and impedance phase angle change of the channel to be detected at different frequencies. Based on the impedance spectrum characteristic information, the latent defect identification result of the CCS under test is determined, and the detection result corresponding to the CCS under test is output.

Images