A method and system for detecting lithium battery tab laser welds

Abstract

This invention discloses a method and system for detecting laser solder joints on lithium battery tabs. By constructing an acquisition system consisting of a multispectral camera, an infrared thermal imaging camera, and a photoacoustic imaging module, the system simultaneously acquires surface optical, internal thermal distribution, and microstructural features of the solder joint, and reconstructs the three-dimensional distribution of internal defects at a preset depth. A feature correlation matrix is ​​constructed, and the correlation between features and defects is verified by combining the three-dimensional distribution. A fusion feature containing three-dimensional distribution parameters is output through an attention-weighted fusion network. Visual parameters are dynamically adjusted based on operating conditions. Defects are detected by combining the fusion features with a causal database, and a prediction module outputs process adjustment instructions in advance. Microscopic defects are segmented and classified using a 3D U-Net. This method covers defects across all dimensions, improves the false defect elimination rate, reduces the defect rate, and balances detection accuracy and production efficiency, meeting the high reliability detection requirements of lithium battery tabs.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and machine vision inspection technology for lithium batteries, and in particular to a method and system for detecting laser solder joints on lithium battery tabs. Background Technology

[0002] In intelligent manufacturing of lithium batteries, the quality of laser solder joints on the tabs directly affects battery performance and safety. With the development of the new energy industry, tabs are becoming ultra-thin (0.05-0.1mm) and solder joints are becoming miniaturized (3-5mm). Traditional visual inspection technology faces three major pain points that restrict the industry's development.

[0003] First, there is a lack of detection for internal latent defects. Current technologies rely on multispectral or three-dimensional topographic scanning, which can only identify surface defects and cannot detect latent defects such as microscopic pores and stress concentrations with a depth of 0.1 mm. These defects can expand during battery cycling and lead to thermal runaway. Although X-ray solutions have been attempted to address this, the equipment is expensive, poses radiation risks, and has a detection rate of less than 60% for microscopic pores smaller than 0.05 mm, making it unsuitable for mass production.

[0004] Secondly, it exhibits poor robustness under complex working conditions. Fluctuations in ambient light in the workshop cause grayscale drift in multispectral images; weld spot temperature (150-300℃) interferes with infrared signals, increasing the temperature difference detection error to ±2℃; the reflectivity of the electrode materials (copper, aluminum, composite) varies greatly, and fixing visual parameters results in a missed detection rate of 8%-12% for aluminum electrodes and a misclassification rate of over 5% for copper electrodes.

[0005] Third, there is a disconnect between detection and process. The existing system only outputs defect results and cannot correlate them with welding parameters (power, speed, etc.), making adjustments dependent on experience. After detecting a poor weld, engineers need to make adjustments through multiple trials, and there may be continuous defects during this period, making closed-loop verification impossible. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides a detection method and system for laser welding points on lithium battery tabs. Through multimodal acquisition, feature fusion, adaptive operating conditions, and process linkage, it achieves full-dimensional detection from the surface to the interior to the microscopic level, thereby improving robustness and optimization efficiency.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a method for detecting laser solder joints on lithium battery tabs, comprising:

[0008] Step 1: Build an acquisition system consisting of a multispectral camera, an infrared thermal imaging camera, and a photoacoustic imaging module. Achieve microsecond-level time synchronization through a real-time bus, acquire the optical features of the solder joint surface, internal thermal distribution features, and microstructural features, and reconstruct the three-dimensional distribution of internal defects at a preset depth through the photoacoustic imaging module.

[0009] Step 2: Construct a feature association matrix, associating the three types of features from Step 1 with surface defects, internal defects, and microscopic defects one by one; introduce the three-dimensional distribution of internal defects as the basis for association verification to confirm whether the defects associated with the features have material anomalies corresponding to spatial locations and depths; train the feature weights using an attention-weighted fusion network, introduce a feature consistency verification algorithm, and combine the three-dimensional distribution of internal defects to confirm whether there are corresponding depth defects, eliminate false defects and correct the results, and output cross-physics field fusion features containing the three-dimensional distribution parameters of internal defects;

[0010] Step 3: Deploy ambient light and temperature sensors and material recognition cameras to collect operating parameters in real time; establish a mapping model between operating parameters and visual parameters, and dynamically adjust the visual parameters based on the real-time operating parameters to obtain optimized visual data;

[0011] Step 4: Collect multiple sets of welding parameters and their corresponding fusion features output from Step 2. Determine the causal relationship between defects and processes using a causal inference algorithm and construct a database. Detect defects based on the fusion features and the visual data optimized in Step 3. After defect detection, calculate parameter adjustment amounts using a hybrid control algorithm. The prediction module, based on the fusion feature changes of a continuously preset number of weld points and the three-dimensional distribution trend of defects within Step 1, combined with the correspondence between defects and processes in the causal database, outputs welding process parameter adjustment instructions for a preset duration. Re-verify the adjusted defects; if they do not meet the standards, perform secondary optimization.

[0012] Step 5: Input the fused features from Step 2 into the 3D U-Net network, combine the three-dimensional distribution of internal defects to segment micro-defects, determine the severity based on the volume, depth, and location of the defects, and output the defect type, quantification parameters, and process suggestions.

[0013] The beneficial effects of this invention are: addressing the pain points of traditional detection methods, such as "primarily surface defects with missed internal defects," "significant interference from operating conditions leading to numerous misjudgments of features," and "lagging process adjustments resulting in recurring defects," this method achieves three core advantages through five synergistic steps:

[0014] Firstly, the combination of multimodal acquisition and 3D reconstruction covers defects in all dimensions from "surface to interior to micro", solving the problem of missed detection of hidden defects such as micropores at the 0.05mm level and improving the detection rate of internal defects.

[0015] Secondly, the feature association verification + working condition adaptive adjustment introduces three-dimensional distribution to verify the authenticity of feature association, which improves the false defect elimination rate. At the same time, it dynamically adapts to changes in ambient light, temperature and material, which improves the stability of visual features.

[0016] Thirdly, the causal linkage and predictive optimization approach outputs adjustment instructions in advance based on the causal relationship between defects and processes, preventing defects from worsening, ensuring stable defect rate control, and balancing detection accuracy and production efficiency, thus meeting the high reliability testing requirements of lithium battery tabs.

[0017] Optionally, step 1 includes the following sub-steps:

[0018] Sub-step 1.1: The multispectral camera collects surface optical features including grayscale change rate and band reflectivity; the infrared thermal imaging camera collects internal thermal distribution features including local temperature difference and thermal diffusivity; and the photoacoustic imaging module collects microstructural features including sound pressure signal.

[0019] Sub-step 1.2: The real-time bus adopts the EtherCAT protocol, with a timing synchronization deviation of ≤1μs, to ensure spatiotemporal alignment of the three types of features;

[0020] Sub-step 1.3: The photoacoustic imaging module uses the filtered back projection method with a preset depth of 0.1-0.2mm to reconstruct the three-dimensional distribution of internal defects.

[0021] As can be seen from the above description, the EtherCAT protocol and timing synchronization deviation are limited to ≤1μs to ensure spatiotemporal alignment of data from multiple devices and eliminate defect location errors caused by data misalignment. In addition, the filter back projection method and the preset depth of 0.1-0.2mm are clearly defined to standardize the three-dimensional reconstruction process, resulting in small internal defect depth detection errors and providing accurate spatial data for subsequent verification.

[0022] Optionally, step 2 includes the following sub-steps:

[0023] Sub-step 2.1: The attention-weighted fusion network consists of an input layer, a weight training layer, a fusion layer, and an output layer. The input layer receives three types of features, and the output layer outputs 128-dimensional cross-physical field fused features.

[0024] Sub-step 2.2: The feature consistency verification algorithm calculates the overlap of a single feature with the defect regions of the other two types of features. If the overlap is ≤10% and the corresponding feature has no defect signal, it is determined to be a false defect.

[0025] Sub-step 2.3: After correcting the pseudo-defects, update the fusion features and output them as the core data input for steps 4 and 5. At the same time, generate preliminary defect labels based on the defect type.

[0026] As can be seen from the above description, clarifying the hierarchical structure of the attention network and the 128-dimensional fusion feature output ensures the standardization of feature fusion and improves the defect representation ability of the fusion features. In addition, quantifying the overlap threshold of feature consistency verification avoids subjective judgment errors and reduces the false defect misjudgment rate. Furthermore, the addition of preliminary defect label generation provides a basis for subsequent final label integration and reduces the label processing time in step 5.

[0027] Optionally, step 3 includes the following sub-steps:

[0028] Sub-step 3.1: The sensor collects ambient light intensity, solder spot temperature, and electrode material;

[0029] Sub-step 3.2: The mapping model is constructed through experiments, and the visual parameters are dynamically adjusted when the ambient light intensity, solder spot temperature and tab material change;

[0030] Sub-step 3.3: Optimize visual data using image enhancement algorithms.

[0031] As can be seen from the above description, by dynamically adjusting visual parameters and combining them with image enhancement algorithms, the signal-to-noise ratio of visual data is improved, ensuring the accuracy of feature recognition under complex working conditions and adapting to the production needs of multiple scenarios.

[0032] Optionally, step 4 includes the following sub-steps:

[0033] Sub-step 4.1: Welding parameters include laser power, welding speed, and defocusing amount. Causal inference is performed using the Do-Calculus algorithm. A causal database is constructed when the causal coefficient is ≥0.8.

[0034] Sub-step 4.2: The hybrid control algorithm is a PID-causal hybrid control algorithm. When the deviation is ≤10%, the historical adjustment amount is called; when the deviation is >10%, the adjustment amount is calculated in real time.

[0035] Sub-step 4.3: The prediction module outputs instructions based on the fusion feature changes of 5 consecutive weld points. After secondary optimization, the defect rate must be ≤0.1%.

[0036] As can be seen from the above description, specifying the Do-Calculus algorithm and the causal coefficient ≥0.8 ensures the reliability of causal relationship construction and improves the accuracy of process matching. In addition, distinguishing the adjustment strategies for deviations ≤10% and >10% avoids blind adjustment. Furthermore, limiting the number of consecutive 5 solder joints and the defect rate to ≤0.1% standardizes the prediction and optimization criteria, reducing the defect recurrence rate after secondary optimization.

[0037] Optionally, step 5 includes the following sub-steps:

[0038] Sub-step 5.1: The 3D segmentation network uses 3D U-Net to segment micro-defects at the 0.05mm level and outputs 3D coordinates and volume;

[0039] Sub-step 5.2: Determine the severity based on the defect volume, location, and shape, and integrate them with the preliminary defect labels generated in step 2 to form the final defect labels;

[0040] Sub-step 5.3: Output results including defect type, quantitative parameters and process suggestions, and generate a traceable inspection report.

[0041] As can be seen from the above description, combining the preliminary defect labels to generate the final labels ensures the integrity of the label information and avoids the loss of information from a single label; the traceable inspection report facilitates production traceability and process review, improves the efficiency of tracing problematic solder joints, and meets the needs of quality control.

[0042] Optionally, step 6 is also included:

[0043] Step 6.1: Collect new "fusion feature - welding parameters - final defect label" data for every 1000 weld points inspected;

[0044] Step 6.2: Based on the collected "fusion features - welding parameters - final defect labels" data, update the parameters of the fusion network in Step 2 and the segmentation network in Step 5 using the gradient descent method;

[0045] Step 6.3: Test the updated fusion network and the 3D U-Net three-dimensional segmentation network. If the accuracy is ≥98%, save; otherwise, roll back.

[0046] As can be seen from the above description, standardizing the incremental data collection cycle (1000 solder joints) and data dimensions ensures the representativeness of the updated data and improves the model's ability to adapt to new working conditions. In addition, clarifying the gradient descent method and the test standard of accuracy ≥98% avoids performance degradation after model updates, ensures the feature adaptability of the fusion network and the defect recognition rate of the segmentation network after updates, and extends the technical life cycle of the method.

[0047] Optionally, step 1 may also include sub-step 1.4:

[0048] Sub-step 1.4: Adjust the camera aperture and field of view, focusing on the solder joint and the surrounding 2mm area.

[0049] As can be seen from the above description, focusing on the solder joint and the surrounding 2mm range reduces redundant data acquisition, lowers data transmission volume, and saves storage costs. In addition, it avoids interference from non-target area data, increases the proportion of core area feature acquisition, shortens the data processing time in subsequent fusion stages, and improves overall detection efficiency.

[0050] Optionally, step 5 may also include sub-step 5.4:

[0051] Sub-step 5.4: The detection interface marks the defect location in real time and colors it according to the level, generates a defect statistics report, and uploads it to the production system.

[0052] As described above, real-time annotation and hierarchical coloring shorten the time required for operators to identify defects and avoid omissions in manual interpretation; generating statistical reports and uploading them to the production system facilitates real-time monitoring of defect trends in the workshop, improves the response speed to production anomalies, and helps with quality control in mass production.

[0053] To achieve the above objectives, in a second aspect, the present invention provides a detection system for laser solder joints on lithium battery tabs, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described detection method for laser solder joints on lithium battery tabs.

[0054] The technical effects of the detection system for laser solder joints on lithium battery tabs provided in the second aspect are described in reference to the relevant description of the detection method for laser solder joints on lithium battery tabs provided in the first aspect. Attached Figure Description

[0055] Figure 1 This is a flowchart of a method for detecting laser solder joints on lithium battery tabs according to the present invention;

[0056] Figure 2 This is a schematic diagram of a detection system for laser welding points on lithium battery tabs according to the present invention;

[0057] Explanation of reference numerals in the attached figures

[0058] 1. A detection system for laser welding points on lithium battery tabs; 2. A memory; 3. A processor. Detailed Implementation

[0059] To better explain and facilitate understanding of the present invention, it is described in detail below with reference to the accompanying drawings and specific embodiments. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a clearer and more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0060] Example 1

[0061] Please refer to Figure 1 As shown, a method for detecting laser solder joints on lithium battery tabs includes:

[0062] A method for detecting laser solder joints on lithium battery tabs, comprising:

[0063] Step 1: Build an acquisition system consisting of a multispectral camera, an infrared thermal imaging camera, and a photoacoustic imaging module. Achieve microsecond-level time synchronization through a real-time bus, acquire the optical features of the solder joint surface, internal thermal distribution features, and microstructural features, and reconstruct the three-dimensional distribution of internal defects at a preset depth through the photoacoustic imaging module.

[0064] Step 1 includes the following sub-steps:

[0065] Sub-step 1.1: The multispectral camera acquires 256-dimensional surface optical features including grayscale change rate and band reflectivity; the infrared thermal imaging camera acquires 64-dimensional internal thermal distribution features including local temperature difference and thermal diffusivity; and the photoacoustic imaging module acquires 128-dimensional microstructure features including sound pressure signal (sound pressure amplitude and signal mutation rate).

[0066] Sub-step 1.2: The real-time bus adopts the EtherCAT protocol, with a timing synchronization deviation of ≤1μs, to ensure spatiotemporal alignment of the three types of features;

[0067] Sub-step 1.3: The photoacoustic imaging module uses the filtered back projection method with a preset depth of 0.1-0.2mm to reconstruct the three-dimensional distribution of internal defects;

[0068] Step 1 also includes sub-step 1.4:

[0069] Sub-step 1.4: Adjust the camera aperture and field of view, focusing on the solder joint and the surrounding 2mm area.

[0070] Step 2: Construct a feature association matrix, associating the three types of features from Step 1 with surface defects, internal defects, and microscopic defects one by one. That is, construct a feature association matrix of "multispectral-surface defects, infrared-internal defects, photoacoustic-microscopic defects", and bind the 256-dimensional optical features, 64-dimensional thermal distribution features, and 128-dimensional microscopic features collected in Step 1 with the corresponding defect types.

[0071] The three-dimensional distribution of internal defects is introduced as the basis for association verification to confirm whether there are material anomalies with corresponding spatial locations and depths in the defects associated with the features. An attention-weighted fusion network is used to train the feature weights, and a feature consistency verification algorithm is introduced. The three-dimensional distribution of internal defects is combined to confirm whether there are defects with corresponding depths, eliminate false defects and correct the results. Finally, a verified 128-dimensional cross-physics field fusion feature containing the three-dimensional distribution parameters of internal defects is output as the core feature input for subsequent defect detection.

[0072] Step 2 includes the following sub-steps:

[0073] Sub-step 2.1: The attention-weighted fusion network consists of an input layer (receiving 448 original features: 256+64+128), a weight training layer, a fusion layer, and an output layer. The input layer receives three types of features, and the weighted fusion calculation uses a linear superposition formula: fusion feature = (optical feature × optical weight) + (thermal imaging feature × thermal imaging weight) + (photoacoustic feature × photoacoustic weight). The output layer outputs 128-dimensional cross-physical field fusion features, where the first 64 dimensions correspond to the surface-internal feature fusion result, and the last 64 dimensions correspond to the internal-microscopic feature fusion result.

[0074] Sub-step 2.2: The feature consistency verification algorithm calculates the overlap of the defect regions of a single feature with the other two types of features. If the overlap is ≤10% and the corresponding feature has no defect signal (e.g., photoacoustic signal mutation rate <15%), it is judged as a false defect. For example, during feature consistency verification, if the multispectral feature is determined to have a false weld (grayscale change rate ≤0.1), it is necessary to verify whether the local temperature difference of the infrared thermal imaging feature is <2℃ (no internal stress concentration) and whether the signal mutation rate of the photoacoustic feature is <15% (no microscopic interface separation). If these conditions are met, it is judged as a surface oxidation false defect, and the value of "false weld association dimension" in the fused feature is corrected to 0.05-0.08 (oxidation feature threshold). The false defect elimination rate is ≥99%.

[0075] Sub-step 2.3: After correcting the pseudo-defects, update the fusion features and output them as the core data input for steps 4 and 5, and at the same time generate preliminary defect labels based on the defect type;

[0076] Sub-step 2.4: Principal component analysis (PCA) is used to reduce the dimensionality of the 128-dimensional cross-physics field fusion features, retaining principal components with a variance contribution rate ≥95% (dimension ≤64), thus shortening the inference time of the 3D U-Net network in step 5;

[0077] Step 3: Deploy ambient light sensors, temperature sensors, and material recognition cameras to collect operating parameters in real time; establish a mapping model between operating parameters and visual parameters, and dynamically adjust the visual parameters based on real-time operating parameters to obtain optimized visual data;

[0078] Step 3 includes the following sub-steps:

[0079] Sub-step 3.1: Sensor acquisition. The sensor acquires ambient light (500-2000 lux), solder spot temperature (150-300℃), and electrode material (copper / aluminum / composite composite electrode).

[0080] Sub-step 3.2: The mapping model is constructed through experiments. When the ambient light intensity, solder spot temperature, and tab material change, the visual parameters are dynamically adjusted. For example, for every 500 lux increase in ambient light, the exposure of the multispectral camera is reduced by 10% and the power of the light source is increased by 15%; for every 50°C increase in solder spot temperature, the integration time of the infrared thermal imaging camera is increased by 20%; and the gain of the multispectral 850nm band is increased by 30% when detecting aluminum tabs. The above visual parameters are dynamically adjusted based on real-time operating conditions to obtain optimized visual data.

[0081] Sub-step 3.3: Optimize visual data using image enhancement algorithms.

[0082] Step 4: Collect multiple sets of welding parameters and their corresponding fusion features output from Step 2, determine the causal relationship between defects and processes using a causal inference algorithm, and build a database.

[0083] The causal relationships in the causal database are represented as follows: abnormal process parameters (cause) → characteristic changes (intermediate characterization) → defect generation (effect). The causal database contains 8 core causal relationships: ① Insufficient laser power → optical characteristic grayscale change rate ≤ 0.1 → cold weld; ② Excessive welding speed → infrared characteristic local temperature difference ≥ 3℃ → microcrack; ③ Excessive defocusing amount → photoacoustic characteristic sound pressure change rate ≥ 20% → micropore; ④ Insufficient shielding gas flow rate → optical characteristic blue light grayscale ≥ 200 → pinhole; ⑤ Excessively narrow laser pulse width → infrared characteristic thermal diffusion rate ≥ 5℃ / ms → stress concentration; ⑥ Excessively large electrode butt gap → photoacoustic characteristic delamination signal ≥ 0.5V → interface cold weld; ⑦ Excessively high laser energy → optical characteristic red light grayscale ≤ 50 → weld bead; ⑧ Excessively slow cooling rate → infrared characteristic temperature difference dissipation time ≥ 100ms → coarse grains.

[0084] Based on the fusion features and the visual data optimized in step 3, defects are detected. After defect detection, the hybrid control algorithm is used to calculate the parameter adjustment amount. The prediction module, based on the fusion feature changes of a continuously preset number of weld points and the three-dimensional distribution change trend of internal defects in step 1, combined with the correspondence between defects and processes in the causal database, outputs welding process parameter adjustment instructions for a preset time in advance. The adjusted defects are re-verified, and if they do not meet the standards, a second optimization is performed.

[0085] Step 4 includes the following sub-steps:

[0086] Sub-step 4.1: Welding parameters include laser power, welding speed, and defocusing amount. Causal inference is performed using the Do-Calculus algorithm. A causal database is constructed when the causal coefficient is ≥0.8.

[0087] Sub-step 4.2: The hybrid control algorithm is a PID-causal hybrid control algorithm. When the deviation is ≤10%, the historical adjustment amount is called; when the deviation is >10%, the adjustment amount is calculated in real time. For example, when the defect is a cold solder joint (the gray scale change rate deviation of the fusion feature is 40%), the historical adjustment amount of "power +10%" in the database is called. At the same time, the PID calculates the adjustment amount = 1.0 × 40% + 15ms × 40% integral + 8ms × 40% derivative, and finally takes the average of the two as the final adjustment amount.

[0088] Sub-step 4.3: The prediction module outputs instructions based on the fusion characteristic changes of 5 consecutive weld points. After secondary optimization, the defect rate must be ≤0.1%.

[0089] Specifically, the prediction module is a linear regression prediction module. It takes the slope of the fusion feature change of 5 consecutive solder joints as input and outputs the adjustment command 50ms in advance. The features of the adjusted solder joints are re-acquired and defects are detected. If the defect rate is >0.1%, a second optimization is performed (the adjustment amount is 50% of the first adjustment amount).

[0090] For example, the linear regression model of the prediction module takes into account the fusion feature deviation of 5 consecutive solder joints (e.g., 40%, 35%, 30%, 25%, 20%), calculates the change slope = -5% / point, predicts the deviation of the next solder joint by 15%, and outputs the "power +8%" command 50ms in advance; if the defect rate is still >0.1% after the second optimization, the third-level optimization is triggered, and the adjustment amount is 30% of the second adjustment amount, until the defect rate is ≤0.1% and the number of optimization iterations is ≤3.

[0091] Step 5: Input the fused features from Step 2 into the 3D U-Net network, segment the micro-defects by combining the three-dimensional distribution of internal defects, determine the severity based on the volume, depth, and location of the defects, and output the defect type, quantification parameters, and process suggestions.

[0092] Step 5 includes the following sub-steps:

[0093] Sub-step 5.1: The 3D segmentation network adopts 3D U-Net. The 3D U-Net network contains 4 encoding stages and 4 decoding stages. The encoding stage uses 3×3×3 convolutional kernels (stride 2), and the decoding stage uses 2×2×2 deconvolutional kernels. The loss function is Dice loss (weight 0.6) + L1 loss (weight 0.4). The IoU (Intersection over Union) of the network training is ≥0.92.

[0094] The system segments micro-defects at the 0.05mm level and outputs their three-dimensional coordinates and volume. Combining multiple parameters such as "defect volume (≥0.001mm³ requires rework), defect location (center area of ​​the tab is high risk), and defect morphology (irregularity is high risk)," the severity of the defect is determined. Finally, the system outputs the defect type (e.g., internal micropores), quantitative parameters (e.g., volume 0.0012mm³), and process recommendations (e.g., reducing defocus by 0.2mm).

[0095] Sub-step 5.2: Determine the severity based on the defect volume, location, and morphology (which can be set to levels 1 to 4, where level 1 is no impact; level 2 is low risk, requires observation, has potential defects, but does not affect short-term use; level 3 is medium to high risk, affects reliability, and must be repaired; level 4 is scrap). Combine this with the preliminary defect labels generated in step 2 (such as "surface oxidation - pseudo defect" and "internal micropores - real defect") to form the final defect label (format: "defect type - severity", such as "internal micropores - level 3" and "surface cold solder joints - level 2").

[0096] Sub-step 5.3: Output results including defect type, quantitative parameters (volume, coordinates, morphology coefficients) and process suggestions (corresponding to the reverse parameter adjustment of the causal database in step 4), generate a traceable inspection report (including weld ID, inspection time, and feature map), and upload it to the MES production system in real time;

[0097] Step 5 also includes sub-step 5.4:

[0098] Sub-step 5.4: The detection interface marks the defect location in real time and colors it according to the level, generates a defect statistics report, and uploads it to the production system.

[0099] Step 6:

[0100] Step 6.1: For every 1000 weld points inspected, new "fusion features - welding parameters - final defect labels" data are collected to form an incremental dataset (sample size ≥ 50 groups).

[0101] Step 6.2: Based on the collected "fusion features - welding parameters - final defect labels" data, update the parameters of the fusion network in Step 2 and the segmentation network in Step 5 using the gradient descent method;

[0102] Step 6.3: Test the updated fusion network and 3D U-Net 3D segmentation network. If the accuracy is ≥98%, save; otherwise, roll back. Ensure that the model accuracy loss of the fusion network and 3D U-Net 3D segmentation network is ≤0.5%.

[0103] Example 2

[0104] Please refer to Figure 2 A detection system 1 for laser welding joints of lithium battery tabs includes a memory 2, a processor 3, and a computer program stored in the memory 2 and run on the processor 3. When the processor 3 executes the computer program, it implements the steps in Embodiment 1.

[0105] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0106] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] It should be noted that, in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0108] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0109] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for detecting a laser welding spot of a lithium battery tab, characterized in that, include: Step 1: Build an acquisition system consisting of a multispectral camera, an infrared thermal imaging camera, and a photoacoustic imaging module. Achieve microsecond-level time synchronization through a real-time bus, acquire the optical features of the solder joint surface, internal thermal distribution features, and microstructural features, and reconstruct the three-dimensional distribution of internal defects at a preset depth through the photoacoustic imaging module. Step 2: Construct a feature association matrix, associating the three types of features from Step 1 with surface defects, internal defects, and microscopic defects one by one; introduce the three-dimensional distribution of internal defects as the basis for association verification to confirm whether the defects associated with the features have material anomalies corresponding to spatial locations and depths; train the feature weights using an attention-weighted fusion network, introduce a feature consistency verification algorithm, and combine the three-dimensional distribution of internal defects to confirm whether there are corresponding depth defects, eliminate false defects and correct the results, and output cross-physics field fusion features containing the three-dimensional distribution parameters of internal defects; Step 3: Deploy ambient light and temperature sensors and material recognition cameras to collect operating parameters in real time; establish a mapping model between operating parameters and visual parameters, and dynamically adjust the visual parameters based on the real-time operating parameters to obtain optimized visual data; Step 4: Collect multiple sets of welding parameters and their corresponding fusion features output from Step 2. Determine the causal relationship between defects and processes using a causal inference algorithm and construct a database. Detect defects based on the fusion features and the visual data optimized in Step 3. After defect detection, calculate parameter adjustment amounts using a hybrid control algorithm. The prediction module, based on the fusion feature changes of a continuously preset number of weld points and the three-dimensional distribution change trend of defects within Step 1, combined with the correspondence between defects and processes in the causal database, outputs welding process parameter adjustment instructions at a preset time. Re-verify the adjusted defects; if they do not meet the standards, perform a second optimization. Step 5: Input the fused features from Step 2 into the 3D U-Net network, combine the three-dimensional distribution of internal defects to segment micro-defects, determine the severity based on the volume, depth, and location of the defects, and output the defect type, quantification parameters, and process suggestions.

2. The method for detecting laser solder joints on lithium battery tabs according to claim 1, characterized in that, Step 1 includes the following sub-steps: Sub-step 1.1: The multispectral camera collects surface optical features including grayscale change rate and band reflectivity; the infrared thermal imaging camera collects internal thermal distribution features including local temperature difference and thermal diffusivity; and the photoacoustic imaging module collects microstructural features including sound pressure signal. Sub-step 1.2: The real-time bus adopts the EtherCAT protocol, with a timing synchronization deviation of ≤1μs, to ensure spatiotemporal alignment of the three types of features; Sub-step 1.3: The photoacoustic imaging module uses the filtered back projection method with a preset depth of 0.1-0.2mm to reconstruct the three-dimensional distribution of internal defects.

3. The method for detecting laser solder joints on lithium battery tabs according to claim 1, characterized in that, Step 2 includes the following sub-steps: Sub-step 2.1: The attention-weighted fusion network consists of an input layer, a weight training layer, a fusion layer, and an output layer. The input layer receives three types of features, and the output layer outputs 128-dimensional cross-physical field fused features. Sub-step 2.2: The feature consistency verification algorithm calculates the overlap of a single feature with the defect regions of the other two types of features. If the overlap is ≤10% and the corresponding feature has no defect signal, it is determined to be a false defect. Sub-step 2.3: After correcting the pseudo-defects, update the fusion features and output them as the core data input for steps 4 and 5. At the same time, generate preliminary defect labels based on the defect type.

4. The method for detecting laser solder joints on lithium battery tabs according to claim 1, characterized in that, Step 3 includes the following sub-steps: Sub-step 3.1: The sensor collects ambient light intensity, solder spot temperature, and electrode material; Sub-step 3.2: The mapping model is constructed through experiments, and the visual parameters are dynamically adjusted when the ambient light intensity, solder spot temperature and tab material change; Sub-step 3.3: Optimize visual data using image enhancement algorithms.

5. The method for detecting laser solder joints on lithium battery tabs according to claim 1, characterized in that, Step 4 includes the following sub-steps: Sub-step 4.1: Welding parameters include laser power, welding speed, and defocusing amount. Causal inference is performed using the Do-Calculus algorithm. A causal database is constructed when the causal coefficient is ≥0.

8. Sub-step 4.2: The hybrid control algorithm is a PID-causal hybrid control algorithm. When the deviation is ≤10%, the historical adjustment amount is called; when the deviation is >10%, the adjustment amount is calculated in real time. Sub-step 4.3: The prediction module outputs instructions based on the fusion feature changes of 5 consecutive weld points. After secondary optimization, the defect rate must be ≤0.1%.

6. The method for detecting laser solder joints on lithium battery tabs according to claim 3, characterized in that, Step 5 includes the following sub-steps: Sub-step 5.1: The 3D segmentation network uses 3D U-Net to segment micro-defects at the 0.05mm level and outputs 3D coordinates and volume; Sub-step 5.2: Determine the severity based on the defect volume, location, and shape, and integrate them with the preliminary defect labels generated in step 2 to form the final defect labels; Sub-step 5.3: Output results including defect type, quantitative parameters and process suggestions, and generate a traceable inspection report.

7. The method for detecting laser solder joints on lithium battery tabs according to claim 6, characterized in that, It also includes step 6: Step 6.1: Collect new "fusion feature - welding parameters - final defect label" data for every 1000 weld points inspected; Step 6.2: Based on the collected "fusion features - welding parameters - final defect labels" data, the gradient descent method is used to update the parameters of the fusion network in step 2 and the segmentation network in step 5; Step 6.3: Test the updated fusion network and the 3D U-Net three-dimensional segmentation network. If the accuracy is ≥98%, save; otherwise, roll back.

8. The method for detecting laser solder joints on lithium battery tabs according to claim 2, characterized in that, Step 1 also includes sub-step 1.4: Sub-step 1.4: Adjust the camera aperture and field of view, focusing on the solder joint and the surrounding 2mm area.

9. The method for detecting laser solder joints on lithium battery tabs according to claim 6, characterized in that, Step 5 also includes sub-step 5.4: Sub-step 5.4: The detection interface marks the defect location in real time and colors it according to the level, generates a defect statistics report, and uploads it to the production system.

10. A detection system for laser solder joints on lithium battery tabs, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the detection method for laser solder joints of lithium battery tabs as described in any one of claims 1 to 9.

Images