Detection method and system for lithium battery tab laser welding spot

Abstract

The invention discloses a lithium battery tab laser welding spot detection method and system, and the method comprises the steps: building a collection system of a multispectral camera, an infrared thermal imaging camera and a photoacoustic imaging module, synchronously collecting the surface optics, internal heat distribution and microstructure characteristics of a welding spot, and reconstructing the internal defect three-dimensional distribution of a preset depth; constructing a feature association matrix, verifying association between features and defects in combination with three-dimensional distribution, and outputting fusion features containing three-dimensional distribution parameters through an attention weighted fusion network; dynamically adjusting visual parameters based on the working condition parameters; detecting defects in combination with fusion features and a causal database, and outputting a process adjustment instruction in advance by a pre-judgment module; and through 3D U-Net, micro defects are segmented and graded. The method covers the full-dimension defects, improves the false defect elimination rate, reduces the defect rate, gives consideration to the detection precision and the production efficiency, and meets the high-reliability detection requirements of the lithium battery tabs.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium battery intelligent manufacturing and machine vision detection, and particularly relates to a detection method and system for a lithium battery tab laser welding spot. BACKGROUND

[0002] In the intelligent manufacturing of lithium batteries, the quality of the tab laser welding spot directly affects the performance and safety of the battery. With the development of the new energy industry, the tab is becoming thinner and thinner, from 0.05-0.1 mm, and the welding spot is becoming smaller and smaller, from 3-5 mm. The traditional visual detection technology is facing three major core pain points, which is restricting the development of the industry.

[0003] Firstly, the internal hidden defect detection is missing. The existing technology relies on multi-spectral or three-dimensional topography scanning, which can only identify surface defects and cannot detect internal 0.1 mm deep microscopic holes, stress concentration and other hidden defects. Such defects will expand during the cycle use of the battery, causing thermal runaway. Although X-ray solutions are used to solve this problem, the equipment cost is high, there is a risk of radiation, and the recognition rate for microscopic holes below 0.05 mm is less than 60%, which is not suitable for mass production.

[0004] Secondly, the robustness is poor under complex working conditions. The fluctuation of the environmental light in the workshop causes the gray scale of the multi-spectral image to drift. The temperature of the welding spot (150-300℃) interferes with the infrared signal, and the temperature difference detection error expands to ±2℃. The reflectivity of the tab material (copper, aluminum, composite) differs greatly, and the fixed visual parameters cause the aluminum tab to miss the detection rate in the range of 8%-12%, and the copper tab misclassification rate exceeds 5%.

[0005] Thirdly, the detection is disconnected from the process. The existing system only outputs defect results and cannot associate with welding parameters (power, speed, etc.), and the adjustment relies on experience. After detecting a false weld, engineers need to adjust it several times through trial and error, and there may be continuous defects during this period, which cannot be verified in a closed loop. SUMMARY

[0006] To solve the above problems of the prior art, the present application provides a detection method and system for a lithium battery tab laser welding spot, which realizes "surface-internal-microscopic" full-dimensional detection through multi-modal acquisition, feature fusion, working condition self-adaptation and process linkage, and improves the robustness and optimization efficiency.

[0007] To achieve the above purpose, in a first aspect, the present application provides a detection method for a lithium battery tab laser welding spot, comprising:

[0008] Step 1: Build a collection system of multi-spectral cameras, infrared thermal imaging cameras and photoacoustic imaging modules, realize microsecond-level time sequence synchronization through real-time bus, collect welding spot surface optical features, internal thermal distribution features and microscopic structure features, and reconstruct the three-dimensional distribution of internal defects of a preset depth through the photoacoustic imaging module;

[0009] Step 2: build a feature association matrix, respectively associate the three types of features in step 1 with surface defects, internal defects and micro defects; introduce the three-dimensional distribution of internal defects as a verification basis for association, confirm whether the corresponding spatial position and depth of the defect associated with the feature exist material abnormalities; adopt attention weighted fusion network to train feature weight, introduce feature consistency verification algorithm, and combine the three-dimensional distribution of internal defects to confirm whether the corresponding depth defects exist, exclude false defects and correct the results, output cross-physical field fusion features containing three-dimensional distribution parameters of internal defects;

[0010] Step 3: deploy ambient light, temperature sensor and material identification camera to collect real-time working condition parameters; establish a mapping model of working condition parameters and visual parameters, dynamically adjust the visual parameters based on real-time working condition parameters, and obtain optimized visual data;

[0011] Step 4: collect multiple groups of welding parameters and their corresponding fusion features output in step 2, determine the causal relationship between defects and processes through causal inference algorithm and build a database; based on the fusion features and the optimized visual data in step 3, detect defects, calculate parameter adjustment amount using mixed control algorithm after defect detection, and based on the change of fusion features of a continuous preset number of welding points and the change trend of three-dimensional distribution of internal defects in step 1, combine the corresponding relationship between defects and processes in the causal database, output welding process parameter adjustment instructions in advance for a preset period of time; re-verify the adjusted defects, and if not up to standard, optimize again.

[0012] Step 5: input the fusion features in step 2 into the 3D U-Net network, segment the micro defects combined with the three-dimensional distribution of internal defects, determine the severity based on the volume, depth and position of the defects, and output the defect type, quantitative parameters and process suggestions.

[0013] The beneficial effects of the present application are: for the pain points of traditional detection "surface defects are mainly detected, internal defects are missed" "working condition interference is large, feature misjudgment is more" "process adjustment is lagging, defects appear repeatedly", the present method realizes the three core advantages through the cooperation of the five steps:

[0014] Firstly, multi-modal acquisition + three-dimensional reconstruction combination covers "surface-internal-micro" full-dimensional defects, solves the problem of missing hidden defects such as 0.05mm micro holes, and improves the internal defect detection rate;

[0015] Secondly, feature association verification + working condition self-adaptive adjustment introduces three-dimensional distribution verification feature association authenticity, improves the false defect exclusion rate, and dynamically adapts to changes in ambient light, temperature and material, improving the stability of visual features;

[0016] Thirdly, cause-effect linkage + pre-judgment optimization, based on defect-process cause-effect relationship, output adjustment instruction in advance, avoid defect deterioration, stable defect rate control, balance detection accuracy and production efficiency, meet the high reliability detection demand of lithium battery tab.

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

[0018] Sub-step 1.1: The multispectral camera collects surface optical features containing gray scale change rate and band reflectivity, the infrared thermal imaging camera collects internal thermal distribution features containing local temperature difference and thermal diffusivity, and the photoacoustic imaging module collects microstructure features containing sound pressure signals;

[0019] Sub-step 1.2: Real-time bus adopts EtherCAT protocol, time sequence synchronization deviation ≤1μs, ensuring spatio-temporal alignment of three types of features;

[0020] Sub-step 1.3: The photoacoustic imaging module adopts filtered back-projection method, with preset depth of 0.1-0.2mm, reconstructing three-dimensional distribution of internal defects.

[0021] From the above description, it can be seen that the EtherCAT protocol and the time sequence synchronization deviation ≤1μs are limited to ensure spatio-temporal alignment of multi-device data, eliminating defect positioning errors caused by data misplacement; in addition, the filtered back-projection method and the preset depth of 0.1-0.2mm are specified to standardize the three-dimensional reconstruction process, with small internal defect depth detection error, providing accurate spatial data for subsequent verification.

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

[0023] Sub-step 2.1: The attention weighted fusion network contains 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 fusion features;

[0024] Sub-step 2.2: The feature consistency verification algorithm calculates the defect area overlap degree of a single feature and the other two types of features, and when the overlap degree ≤10% and the corresponding feature has no defect signal, it is determined as a pseudo-defect;

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

[0026] From the above description, it can be seen that the hierarchical structure of the attention network and the 128-dimensional fusion feature output are specified to ensure the standardization of feature fusion and improve the defect representation ability of the fusion features; in addition, the overlap threshold of feature consistency verification is quantified to avoid subjective judgment error and reduce the pseudo-defect misjudgment rate; furthermore, the preliminary defect label generation is added to provide a basis for subsequent final label integration and reduce the label processing time of step 5.

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

[0028] Sub-step 3.1: The sensor collects the ambient light brightness, the welding spot temperature, and the tab material;

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

[0030] Sub-step 3.3: The visual data is optimized through an image enhancement algorithm.

[0031] From the above description, by dynamically adjusting the visual parameters and combining the image enhancement algorithm, the signal-to-noise ratio of the visual data is improved, ensuring the accuracy of feature recognition under complex working conditions and adapting to multiple scene production requirements.

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

[0033] Sub-step 4.1: The welding parameters include laser power, welding speed, and defocusing amount. The Do-Calculus algorithm is used for causal inference, and the causal coefficient is ≥0.8 to construct a causal database;

[0034] Sub-step 4.2: The mixed control algorithm is a PID-causal mixed control algorithm. When the deviation is ≤10%, the historical adjustment amount is called, and 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 the last 5 welding points, and the defect rate after secondary optimization needs to be ≤0.1%.

[0036] From the above description, the Do-Calculus algorithm and the causal coefficient ≥0.8 are clear, ensuring the reliability of the causal relationship construction and improving the process matching accuracy. In addition, the adjustment strategies are differentiated for deviations ≤10% and >10%, avoiding blind adjustment. Furthermore, the last 5 welding points and the defect rate ≤0.1% are limited, standardizing the prediction and optimization standards, and reducing the defect recurrence rate after secondary optimization.

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

[0038] Sub-step 5.1: The three-dimensional segmentation network uses 3D U-Net to segment 0.05mm-level microscopic defects and output three-dimensional coordinates and volume;

[0039] Sub-step 5.2: The severity is determined based on the defect volume, position, and morphology, and the preliminary defect label generated in step 2 is combined to form the final defect label;

[0040] Sub-step 5.3: The output results include defect type, quantitative parameters, and process suggestions, and a traceable detection report is generated.

[0041] From the above description, the final label is generated in combination with the preliminary defect label, ensuring the integrity of the label information and avoiding the loss of single label information. Through the traceable detection report, production traceability and process review are facilitated, the problem welding point traceability efficiency is improved, and the quality control requirements are met.

[0042] Optionally, it also includes step 6:

[0043] Step 6.1: Collect new "fusion feature-welding parameter-final defect label" data every 1000 welds;

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

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

[0046] From the above description, the incremental data collection period (1000 welds) and data dimension are specified to ensure the representativeness of the updated data and improve the model's ability to adapt to new working conditions. In addition, the gradient descent method and the test standard of accuracy ≥98% are specified to avoid performance degradation after model updating, ensuring the feature adaptability of the updated fusion network and the defect recognition rate of the segmentation network, and prolonging the technical life cycle of the method.

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

[0048] Sub-step 1.4: Adjust the camera aperture and field of view to focus on the weld and the surrounding 2mm range.

[0049] From the above description, focusing on the weld and the surrounding 2mm range reduces redundant data collection, reduces data transmission volume, and saves storage costs. In addition, it avoids interference from non-target area data, improves the proportion of core area feature collection, shortens the data processing time in the subsequent fusion link, and improves the overall detection efficiency.

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

[0051] Sub-step 5.4: Detect the real-time labeling of defect positions and color them according to the classification, generate a defect statistical report, and upload it to the production system.

[0052] From the above description, real-time labeling + classification coloring shortens the time spent by operators in defect identification and avoids missing manual interpretation. Generating a statistical report and uploading it to the production system facilitates real-time monitoring of defect trends in the workshop, improves the response speed to production abnormalities, and assists in quality control of mass production.

[0053] To achieve the above object, the second aspect, the present application provides a kind of detection system for lithium battery tab laser welding spot, including memory, processor and computer program stored in memory and can be run on processor, the processor executes the computer program when realizing above-mentioned detection method for lithium battery tab laser welding spot.

[0054] Wherein, the technical effect corresponding to the detection system for lithium battery tab laser welding spot provided in the second aspect, refer to the related description of the detection method for lithium battery tab laser welding spot provided in the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The flow chart of the detection method for lithium battery tab laser welding spot of the present application is shown in the figure;

[0056] Figure 2 The structure diagram of the detection system for lithium battery tab laser welding spot of the present application is shown in the figure;

[0057] REFERENCE NUMERALS

[0058] 1, a detection system for lithium battery tab laser welding spot;2, memory;3, processor. DETAILED DESCRIPTION

[0059] In order to better explain the present application, so as to be understood, the present application is described in detail by specific embodiments, combined with the drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer, more thorough understanding of the present application, and to convey the scope of the present application to those skilled in the art.

[0060] Example one

[0061] Please refer to Figure 1 The detection method for lithium battery tab laser welding spot, comprising:

[0062] The detection method for lithium battery tab laser welding spot, comprising:

[0063] Step 1: build multispectral camera, infrared thermal imaging camera and photoacoustic imaging module acquisition system, realize microsecond level timing synchronization through real-time bus, collect welding spot surface optical characteristics, internal thermal distribution characteristics and microstructure characteristics, and reconstruct the three-dimensional distribution of internal defects of preset depth through photoacoustic imaging module;

[0064] Wherein, step 1 includes the following substeps:

[0065] Sub-step 1.1: The multispectral camera collects 256-dimensional surface optical features containing gray scale change rate and band reflectivity, the infrared thermal imaging camera collects 64-dimensional internal thermal distribution features containing local temperature difference and thermal diffusivity, and the photoacoustic imaging module collects 128-dimensional microscopic structure features containing sound pressure signals (sound pressure amplitude and signal mutation rate);

[0066] Sub-step 1.2: The real-time bus uses the EtherCAT protocol, and the timing synchronization deviation is less than or equal to 1 μs, ensuring the spatiotemporal alignment of the three types of features;

[0067] Sub-step 1.3: The photoacoustic imaging module uses the filtered back-projection method, and the preset depth is 0.1-0.2 mm, reconstructing the three-dimensional distribution of internal defects;

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

[0069] Sub-step 1.4: Adjust the camera aperture and field of view, and focus on the welding spot and the surrounding 2 mm range.

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

[0071] Introduce the three-dimensional distribution of internal defects as a correlation verification basis to confirm whether the features associated with defects exist corresponding spatial positions and depth abnormalities; use an attention weighted fusion network to train feature weights, introduce a feature consistency verification algorithm, and combine the three-dimensional distribution of internal defects to confirm whether there are corresponding depth defects, exclude false defects, and correct the results, finally output the 128-dimensional cross-physical field fusion features containing the three-dimensional distribution parameters of internal defects verified, 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 includes an input layer (receiving 448-dimensional original features: 256+64+128), a weight training layer, a fusion layer, and an output layer. The input layer receives the three types of features, the weighted fusion calculation uses a linear superposition formula: fusion features = (optical features x optical weight) + (thermal imaging features x thermal imaging weight) + (photoacoustic features x photoacoustic weight), and the output layer outputs 128-dimensional cross-physical field fusion features, of which the first 64 dimensions correspond to the surface-internal feature fusion results, and the last 64 dimensions correspond to the internal-microscopic feature fusion results;

[0074] Sub-step 2.2: The feature consistency verification algorithm determines the presence of a pseudo-defect by calculating the overlap of a single feature with the defect area of the other two types of features. If the overlap is less than or equal to 10% and there is no defect signal (such as a photoacoustic signal mutation rate < 15%) corresponding to the feature, it is determined to be a pseudo-defect. For example, during feature consistency verification, if a multi-spectral feature determines the presence of a false weld (gray scale change rate ≤ 0.1), the local temperature difference of the infrared thermal imaging feature (no internal stress concentration) and the photoacoustic feature signal mutation rate (no micro interface separation) need to be checked. If the above conditions are met, it is determined to be a surface oxidation pseudo-defect, and the "false weld correlation dimension" value in the fusion feature is modified to 0.05-0.08 (oxidation feature threshold). The pseudo-defect exclusion rate is greater than or equal to 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 generate preliminary defect labels based on the defect types;

[0076] Sub-step 2.4: Apply principal component analysis (PCA) to the 128-dimensional fusion features across physical fields to reduce the dimensionality, retaining principal components with a variance contribution rate of greater than or equal to 95% (dimension ≤ 64). This reduces the inference time of the 3D U-Net network in step 5.

[0077] Step 3: Deploy ambient light sensors, temperature sensors, and material identification cameras to collect real-time working condition parameters. Establish a mapping model between working condition parameters and visual parameters, and dynamically adjust the visual parameters based on real-time working condition parameters to obtain optimized visual data.

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

[0079] Sub-step 3.1: The sensors collect ambient light (500-2000 lux), weld spot temperature (150-300°C), and tab material (copper / aluminum / composite tab).

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

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

[0082] Step 4: Collect multiple sets of welding parameters and their corresponding fusion features from step 2 output. Determine the causal relationship between defects and processes through causal inference algorithms and construct a database.

[0083] Wherein, the cause-effect relationship of data in the cause-effect database is represented as: process parameter abnormality (cause) → feature change (intermediate representation) → defect generation (effect), the cause-effect database contains 8 groups of core cause-effect relationships: ① insufficient laser power → optical feature gray level change rate ≤0.1 → virtual welding; ② too fast welding speed → local temperature difference of infrared feature ≥3℃ → micro-crack; ③ too large defocusing amount → optical feature acoustic pressure mutation rate ≥20% → micro-hole; ④ insufficient protective gas flow → optical feature blue light gray level ≥200 → pinhole; ⑤ too narrow laser pulse width → infrared feature heat diffusion rate ≥5℃ / ms → stress concentration; ⑥ too large tab butt gap → optical feature delamination signal ≥0.5V → interface virtual welding; ⑦ too high laser energy → optical feature red light gray level ≤50 → welding bump; ⑧ too slow cooling speed → infrared feature temperature difference dissipation time ≥100ms → coarse grain;

[0084] Detect defects based on the fusion features and the visual data optimized in step 3, calculate the parameter adjustment amount by using a hybrid control algorithm after defect detection, and output the welding process parameter adjustment instruction in advance based on the fusion feature changes of a continuous preset number of welding spots and the three-dimensional distribution change trend of internal defects in step 1, in combination with the corresponding relationship between defects and processes in the cause-effect database, and re-verify the adjusted defects, and perform secondary optimization if the adjusted defects do not meet the standard;

[0085] Wherein, step 4 includes the following sub-steps:

[0086] Sub-step 4.1: the welding parameters include laser power, welding speed and defocusing amount, the cause-effect inference adopts the Do-Calculus algorithm, and the cause-effect coefficient is ≥0.8 to construct the cause-effect database;

[0087] Sub-step 4.2: the hybrid control algorithm is a PID-cause-effect hybrid control algorithm, the historical adjustment amount is called when the deviation is ≤10%, and the adjustment amount is calculated in real time when the deviation is >10%; for example, when the defect is virtual welding (fusion feature gray level change rate deviation 40%), the historical adjustment amount of “power +10%” in the database is called, and the PID calculation adjustment amount =1.0×40%+15ms×40% integral+8ms×40% differential, and the final adjustment amount is the average of the two;

[0088] Sub-step 4.3: the prediction module outputs the instruction based on the fusion feature changes of the continuous 5 welding spots, and the defect rate after secondary optimization needs to be ≤0.1%;

[0089] Wherein, the prediction module is a linear regression prediction module, the slope of the fusion feature changes of the continuous 5 welding spots is input, and the adjustment instruction is output in advance 50ms; the features of the adjusted welding spots are re-collected and defects are detected, and secondary optimization is performed (the adjustment amount is 50% of the first adjustment amount) if the defect rate is >0.1%;

[0090] For example, the linear regression model of the pre-judgment module inputs the fusion feature deviation of the last 5 welding points (such as 40%, 35%, 30%, 25%, and 20%), calculates the change slope =-5% / point, predicts that the next welding point deviation is 15%, and outputs the “power+8%” instruction 50 ms in advance; if the defect rate is still >0.1% after secondary optimization, trigger tertiary optimization, the adjustment amount is 30% of the secondary adjustment amount, until the defect rate ≤0.1%, the optimization iteration number ≤3 times.

[0091] Step 5: input the fusion feature of step 2 into the 3D U-Net network, segment the micro defects based on the internal defect three-dimensional distribution, determine the severity based on the volume, depth, and position of the defect, and output the defect type, quantification parameter, and process suggestion;

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

[0093] Sub-step 5.1: the three-dimensional segmentation network adopts 3D U-Net, the 3D U-Net network contains 4 encoding stages and 4 decoding stages, the encoding stage adopts 3×3×3 convolution kernel (step length 2), the decoding stage adopts 2×2×2 deconvolution kernel, the loss function is Dice loss (weight 0.6)+L1 loss (weight 0.4), and the IoU (intersection over union) of network training is ≥0.92;

[0094] Segment 0.05 mm level micro defects and output three-dimensional coordinates and volume, combine “defect volume (≥0.001 mm³ for rework), defect position (high risk in the center area of the lug), and defect morphology (high risk of irregularity)” multi-parameter to determine the severity of the defect; finally output the defect type (such as internal micro hole), quantification parameter (such as volume 0.0012 mm³), and process suggestion (such as defocusing amount reduction 0.2 mm);

[0095] Sub-step 5.2: determine the severity based on the defect volume, position, and morphology (which can be set to 1 to 4 levels, level 1 being no impact; level 2 being low risk, needing to be observed, there is potential defect, and it does not affect short-term use; level 3 being medium-high risk, affecting use reliability, and must be repaired; level 4 being scrap), combine the preliminary defect label generated in step 2 (such as “surface oxidation-pseudo defect” and “internal micro hole-real defect”), and integrate into the final defect label (format such as: “defect type-severity”, such as “internal micro hole-3 level” and “surface virtual welding-2 level”);

[0096] Sub-step 5.3: the output result contains the defect type, quantification parameter (volume, coordinates, and morphology coefficient), and process suggestion (corresponding to the reverse parameter adjustment of the cause-and-effect database of step 4), generates a traceable detection report (containing welding point ID, detection time, and feature spectrum), and uploads it to the MES production system in real time;

[0097] Step 5 further comprises sub-step 5.4:

[0098] Sub-step 5.4: Detecting interface real-time labeling defect position and grading coloring, generating defect statistical report, uploading to production system.

[0099] Step 6:

[0100] Step 6.1: Collect new "fusion features-welding parameters-final defect label" data every 1000 detected welding points, form incremental data set (sample size ≥ 50 groups);

[0101] Step 6.2: Based on the collected "fusion features-welding parameters-final defect label" data, the gradient descent method is used to update the fusion network of step 2 and the segmentation network parameters of step 5;

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

[0103] Example two

[0104] Please refer to Figure 2 A detection system 1 for lithium battery tab laser welding points, comprising a memory 2, a processor 3 and a computer program stored on the memory 2 and executable on the processor 3, wherein the processor 3 implements the steps in example one when executing the computer program.

[0105] Because the system / device described in the above-mentioned embodiments of the present application is the system / device used for implementing the method of the above-mentioned embodiments of the present application, the specific structure and modification of the system / device can be understood by those skilled in the art based on the method described in the above-mentioned embodiments of the present application, and thus will not be described here. Any system / device used in the method of the above-mentioned embodiments of the present application belongs to the scope of the present application.

[0106] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, device or computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented 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 is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

[0108] Although preferred embodiments of the application have been described herein, it will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

[0109] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for detecting a laser welding spot of a lithium battery tab, characterized in that, Comprise: Step 1: build a multispectral camera, infrared thermal imaging camera and photoacoustic imaging module acquisition system, realize microsecond level time sequence synchronization through real-time bus, collect surface optical features, internal thermal distribution features and microstructure features, reconstruct internal defect three-dimensional distribution of preset depth through photoacoustic imaging module; Step 2: build a feature association matrix, associate the three types of features in step 1 with surface defects, internal defects and micro defects one by one; Introduce internal defect three-dimensional distribution as a basis for association verification to confirm whether the feature associated defects exist corresponding spatial position and depth material anomaly; Adopt attention weighted fusion network to train feature weight, introduce feature consistency verification algorithm, and combine internal defect three-dimensional distribution to confirm whether corresponding depth defects exist, exclude false defects and correct results, output cross-physical field fusion features containing internal defect three-dimensional distribution parameters; Step 3: deploy ambient light, temperature sensor and material identification camera to collect real-time working condition parameters; Establish a mapping model of working condition parameters and visual parameters, dynamically adjust the visual parameters based on real-time working condition parameters, and get the optimized visual data; Step 4: Collect multiple groups of welding parameters and their corresponding fusion features output by step 2, determine the causal relationship between defects and process through causal inference algorithm and build a database; Based on the fusion features and the optimized visual data in step 3, detect defects, calculate parameter adjustment amount using mixed control algorithm after defect detection, and output welding process parameter adjustment instruction in advance based on the change of fusion features of a continuous preset number of welding spots and the change trend of internal defect three-dimensional distribution in step 1, combined with the corresponding relationship between defects and process in the causal database; Reverify the adjusted defects, and optimize them again if they do not meet the standards; Step 5: input the fusion features of step 2 into 3D U-Net network, segment micro defects combined with internal defect three-dimensional distribution, determine severity based on volume, depth and location of defects, and output defect type, quantitative parameters and process suggestion.

2. The method for detecting a laser weld of a lithium battery tab according to claim 1, wherein, Step 1 includes the following sub-steps: Sub-step 1.1: multispectral camera collects surface optical features containing gray scale change rate and waveband reflectivity, infrared thermal imaging camera collects internal thermal distribution features containing local temperature difference and thermal diffusivity, and photoacoustic imaging module collects microstructure features containing sound pressure signal; Sub-step 1.2: real-time bus uses EtherCAT protocol, time sequence synchronization deviation is less than or equal to 1us, ensuring that the three types of features are spatio-temporally aligned; Sub-step 1.3: photoacoustic imaging module uses filtered back-projection method, preset depth is 0.1-0.2mm, and internal defect three-dimensional distribution is reconstructed.

3. The method for detecting a laser weld of a lithium battery tab according to claim 1, wherein, Step 2 includes the following sub-steps: Sub-step 2.1: attention weighted fusion network contains input layer, weight training layer, fusion layer and output layer, input layer receives three types of features, and output layer outputs 128-dimensional cross-physical field fusion features; Sub-step 2.2: feature consistency verification algorithm calculates the overlap degree of single feature and other two types of features, and when the overlap degree is less than or equal to 10% and there is no defect signal in the corresponding feature, it is determined as a false defect; Sub-step 2.3: After the pseudo-defects are corrected, update the fusion features and output them as the core data input for steps 4 and 5, and generate preliminary defect labels based on the defect types.

4. The method for detecting a laser weld of a lithium battery tab according to claim 1, wherein, Step 3 includes the following sub-steps: Sub-step 3.1: The sensor collects the ambient light brightness, weld spot temperature, and tab material; Sub-step 3.2: The mapping model is constructed through experiments, and when the ambient light brightness, weld spot temperature, and tab material change, the visual parameters are dynamically adjusted; Sub-step 3.3: The visual data is optimized through image enhancement algorithms.

5. The method for detecting a laser weld of a lithium battery tab according to claim 1, wherein, Step 4 includes the following sub-steps: Sub-step 4.1: The welding parameters include laser power, welding speed, and defocusing amount. The causal inference uses the Do-Calculus algorithm, and when the causal coefficient is greater than or equal to 0.8, a causal database is constructed; Sub-step 4.2: The mixed control algorithm is a PID-cause mixed control algorithm. When the deviation is less than or equal to 10%, the historical adjustment amount is called, and when the deviation is greater than 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 the last 5 weld points, and the defect rate after secondary optimization needs to be less than or equal to 0.1%.

6. The method for detecting a laser weld of a lithium battery tab according to claim 3, wherein, Step 5 includes the following sub-steps: Sub-step 5.1: The three-dimensional segmentation network uses 3D U-Net to segment 0.05mm-level micro-defects and output three-dimensional coordinates and volume; Sub-step 5.2: Based on the defect volume, position, and shape, the severity is determined, and the preliminary defect labels generated in step 2 are integrated into the final defect labels; Sub-step 5.3: The output results include defect types, quantitative parameters, and process suggestions, and a traceable detection report is generated.

7. The method for detecting a laser weld of a lithium battery tab according to claim 6, wherein, Also includes step 6: Step 6.1: Every 1000 weld points are detected, new "fusion feature-welding parameter-final defect label" data is collected; Step 6.2: Based on the collected "fusion feature-welding parameter-final defect label" data, the gradient descent method is used to update the fusion network of step 2 and the segmentation network parameters of step 5; Step 6.3: Test the updated fusion network and 3D U-Net three-dimensional segmentation network. If the accuracy is greater than or equal to 98%, save it, otherwise roll back. 8.The method for detecting a laser welding spot of a lithium battery tab according to claim 2, wherein, Step 1 also includes sub-step 1.4: Sub-step 1.4: Adjust the camera aperture and field of view, focus on the weld point and the surrounding 2mm range. 9.The method for detecting a laser welding spot of a lithium battery tab according to claim 6, wherein, Step 5 also includes sub-step 5.4: Sub-step 5.4: The detection interface real-time labels the defect position and colors it according to the classification, generates a defect statistics report, and uploads it to the production system.

10. A system for detecting lithium battery tab laser welds, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The processor executes the computer program to realize the detection method for lithium battery tab laser welds according to any one of claims 1-9. The processor executes the computer program to realize the detection method for lithium battery tab laser welds according to any one of claims 1-9.

Images