Method and system for intelligently evaluating and optimizing welding spot quality of lithium battery protection plate

Abstract

The invention relates to an intelligent evaluation and optimization method and system for the welding spot quality of a lithium battery protection board in the technical field of intelligent manufacturing, and the method comprises the steps: constructing a welding spot stereoscopic model through employing a geometric modeling algorithm based on a three-dimensional point cloud data set, and determining a welding spot tin material climbing angle and a contact interface shape feature; on the basis of the bridging defect occurrence probability distribution, a mode recognition algorithm is adopted to distinguish normal welding spots from various defect categories, and a defect classification result is obtained; generating a welding spot quality evaluation report in combination with the defect classification result and the three-dimensional point cloud data set, and determining an overall welding reliability index; if the overall welding reliability index is lower than a preset threshold value range, an alarm mechanism is triggered, and position coordinate information of related welding spots on a production line is obtained; and based on the welding spot position coordinate information, processing subsequent scanning data by adopting a data optimization algorithm, and judging whether the optimized data meet a preset precision standard or not.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, and particularly relates to a welding point quality intelligent evaluation and optimization method and system for a lithium battery protection plate. BACKGROUND

[0002] Lithium battery protection plate manufacturing is a core link of the new energy industry, and its quality directly affects the safety and service life of the battery. As a key step of connecting chips, resistors, capacitors and other components with the circuit board, the reliability of the welding process determines the electrical performance and long-term stability of the product. With the wide application of lithium batteries in electric vehicles, energy storage systems and other fields, higher requirements are put forward for the precision and efficiency of welding quality detection. Efficient and accurate detection technology not only guarantees the improvement of product quality, but also is an important support for promoting intelligent manufacturing and industrial upgrading.

[0003] However, the current welding quality detection method has significant limitations. Traditional manual visual inspection relies on operator experience, is low in efficiency, and is prone to missed detection or misjudgment due to fatigue. Although automatic optical detection improves speed, it is mainly based on two-dimensional image analysis and is difficult to capture subtle changes in the height and three-dimensional shape of the welding point. For example, a virtual weld may have normal surface morphology but poor internal contact, which is difficult for traditional methods to effectively identify. This limitation leads to instability in quality control during the production process, increasing the risk of product hazards. The core technical difficulty lies in how to accurately obtain the three-dimensional shape information of the welding point and realize intelligent recognition of complex defects on this basis. The three-dimensional shape of the welding point, such as height, volume, and tin climbing angle, directly reflects the reliability of its electrical connection, but existing technologies are difficult to efficiently reconstruct these features with micron-level precision in an industrial environment. More complex is that even if three-dimensional data is obtained, how to quickly distinguish normal welding points from various defect types (such as virtual welding and bridging) from these high-dimensional and heterogeneous information, and avoid misjudgment caused by complex rules set by humans, becomes a difficult problem to be solved. The two are complementary: accurate three-dimensional shape acquisition is the foundation, while intelligent defect recognition is the key to converting data into reliable judgments.

[0004] Therefore, how to obtain the micron-level three-dimensional shape of the welding point in real time on a high-speed production line and realize automatic and accurate classification of multiple defects based on these complex data has become a key problem to improve the efficiency and reliability of lithium battery protection plate welding quality detection. SUMMARY

[0005] The present application provides a welding point quality intelligent evaluation and optimization method for a lithium battery protection plate, comprising the following steps:

[0006] The non-contact data acquisition method is used to scan the surface of the welding point, the height distribution information and the volume related features of the welding point are extracted from the scanning data, and the three-dimensional point cloud data set of the welding point is generated; based on the three-dimensional point cloud data set, the geometric modeling algorithm is applied to construct the three-dimensional model of the welding point, and the tin material climbing angle and the contact interface shape features of the welding point are determined; the machine learning algorithm is used to process the three-dimensional model of the welding point, the height deviation features and the volume abnormal indicators are extracted from the model, and the potential false welding risk level of the welding point is judged; if the height deviation features exceed the preset threshold range, the edge analysis algorithm is used to analyze the change of the tin material climbing angle, and the bridge defect occurrence probability distribution is calculated; based on the bridge defect occurrence probability distribution, the pattern recognition algorithm is used to distinguish the normal welding point from various defect categories, and the defect classification result is obtained; the welding quality evaluation report is generated by combining the defect classification result and the three-dimensional point cloud data set, and the overall welding reliability index is determined; if the overall welding reliability index is lower than the preset threshold range, the alarm mechanism is triggered, and the related welding point position coordinate information on the production line is obtained; based on the welding point position coordinate information, the data optimization algorithm is used to process the subsequent scanning data, and it is judged whether the optimized data meets the preset precision standard.

[0007] The application provides a welding point quality intelligent evaluation and optimization system for a lithium battery protection plate, which mainly comprises: a data acquisition and scanning module, which is used to scan the surface of the welding point by using a non-contact data acquisition method, extract the height distribution information and volume related features of the welding point from the scanning data, and generate a three-dimensional point cloud data set of the welding point; a three-dimensional modeling and feature extraction module, which is used to construct a three-dimensional model of the welding point based on the three-dimensional point cloud data set, apply a geometric modeling algorithm, determine the tin material climbing angle and the contact interface shape features of the welding point; a machine learning analysis module, which is used to process the three-dimensional model of the welding point by using a machine learning algorithm, extract the height deviation features and volume abnormal indicators from the model, and judge the potential false welding risk level of the welding point; a threshold judgment and edge analysis module, which is used to analyze the change of the tin material climbing angle by using an edge analysis algorithm if the height deviation features exceed the preset threshold range, and calculate the bridge defect occurrence probability distribution; a bridge defect probability calculation module, which is used to distinguish the normal welding point from various defect categories by using a pattern recognition algorithm based on the bridge defect occurrence probability distribution, and obtain a defect classification result; a defect classification and identification module, which is used to generate a welding quality evaluation report by combining the defect classification result and the three-dimensional point cloud data set, and determine the overall welding reliability index; a quality evaluation and report generation module, which is used to trigger an alarm mechanism if the overall welding reliability index is lower than the preset threshold range, and obtain the related welding point position coordinate information on the production line; an alarm triggering and data optimization module, which is used to process the subsequent scanning data by using a data optimization algorithm based on the welding point position coordinate information, and judge whether the optimized data meets the preset precision standard. The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0008] The application discloses a kind of welding point quality intelligent evaluation and optimization method of lithium battery protection board, for the business scene problem of welding point quality detection and defect identification in electronic manufacturing, the problem that traditional method is difficult to efficiently, accurately evaluate welding point reliability and optimize production process is solved.The application generates welding point three-dimensional point cloud data set by non-contact scanning, applies geometric modeling algorithm to construct solid model, extracts height distribution, volume characteristics and tin material climbing angle Key parameters such as angle.Judge false welding risk by using machine learning algorithm to analyze height deviation and volume anomaly, and calculate the probability distribution of bridging defect by edge analysis algorithm, realize defect classification in combination with pattern recognition algorithm, generate quality evaluation report.When reliability index is lower than threshold value, trigger alarm and obtain welding point position coordinates, adopt data optimization algorithm to process subsequent scanning data, ensure that accuracy meets standard.The application significantly improves the automation and accuracy of welding point quality detection, optimizes production process, reduces defect rate, and guarantees welding reliability. BRIEF DESCRIPTION OF DRAWINGS

[0009] Fig. 1 For the flow chart of the welding point quality intelligent evaluation and optimization method of the lithium battery protection board of the application.

[0010] Fig. 2 For the schematic diagram of the welding point quality intelligent evaluation and optimization method of the lithium battery protection board of the application.

[0011] Fig. 3 For another schematic diagram of the welding point quality intelligent evaluation and optimization method of the lithium battery protection board of the application. DETAILED DESCRIPTION

[0012] For further understanding of the content of the application, the application is described in detail in combination with drawings and examples.It can be understood that the specific embodiments described herein are only used to explain the related application, and not limited to the application.In addition, it should be noted that, for the convenience of description, only the parts related to the application are shown in the drawings.

[0013] As Figs. 1-3 , the welding point quality intelligent evaluation and optimization method of the lithium battery protection board of the embodiment can specifically include:

[0014] Step S101, non-contact data acquisition method is used to scan welding point surface, and welding point height distribution information and volume related characteristics are extracted from scanning data, to generate welding point three-dimensional point cloud data set.

[0015] The surface of the solder joint is scanned by a non-contact method to obtain scanning data, and height distribution information is extracted from the scanning data to obtain the height distribution information of the solder joint. According to the height distribution information, a volume-related feature extraction method is used to obtain the volume-related feature of the solder joint, and the volume-related feature is determined. Through the fusion processing of the volume-related feature and the height distribution information, a solder joint three-dimensional point cloud data set is generated, and the solder joint three-dimensional point cloud data set is obtained. For the solder joint three-dimensional point cloud data set, data calibration processing is performed to obtain calibrated three-dimensional point cloud data, and the calibrated three-dimensional point cloud data is determined. If the calibrated three-dimensional point cloud data meets the preset accuracy threshold, the calibrated three-dimensional point cloud data is output as the final solder joint three-dimensional model data.

[0016] Exemplarily, in an embodiment, a non-contact data acquisition method is used to scan the surface of the solder joint.

[0017] Specifically, a laser triangulation instrument is used as the data acquisition device to scan the solder joint on the electronic circuit board. This method obtains surface information by emitting a laser beam and capturing reflected light, avoiding damage caused by physical contact. The scanning process includes positioning the device above the solder joint, scanning line by line at a fixed distance and angle, and storing the collected data in the form of point coordinates.

[0018] It should be noted that this non-contact method is suitable for various welding scenarios, such as solder joint detection in surface mount technology, ensuring the accuracy and efficiency of data acquisition. In this way, the original scanning data of the solder joint surface is obtained, providing a basis for subsequent extraction. The height distribution information of the solder joint is extracted from the scanning data.

[0019] In a possible implementation, the original scanning data is first preprocessed, including noise filtering and coordinate calibration.

[0020] Specifically, a Gaussian filter is used to remove random noise in the scan, and then the height value of the reference plane calibration point is calibrated. The height distribution information is obtained by counting the Z-axis coordinates of each point, such as calculating the histogram distribution of the height, identifying the peak height and uniformity of the solder joint. The key to this extraction process is to identify the edge of the solder joint to avoid interference from the surrounding substrate.

[0021] Exemplarily, in electronic component welding, this height distribution can be used to evaluate the flatness of the solder joint, and if the distribution is uneven, it may indicate a welding defect. By analyzing the height data in detail, a height distribution map of the solder joint is formed, further supporting quality evaluation. Further, volume-related features are extracted.

[0022] Preferably, the volume features of the solder joint are calculated from the preprocessed data, including volume calculation and shape factor.

[0023] Specifically, the volume-related features are obtained by integrating volume elements in the point cloud data, for example, converting the point cloud to a mesh using voxelization method, and then accumulating the volume values of each voxel. This feature extraction takes into account the three-dimensional structure of the solder joint, such as protruding parts and recessed areas.

[0024] In one embodiment, for ball grid array solder joints, the volume feature can reflect the uniformity of the distribution of the solder material, and if the volume deviation exceeds a threshold value, it indicates a potential void defect. This method enhances the reliability of the features, ensuring that the extracted information accurately reflects the physical properties of the solder joint. A three-dimensional point cloud dataset of the solder joint is generated.

[0025] For example, the extracted height distribution information and volume-related features are integrated into the point cloud data structure.

[0026] Specifically, each point cloud point contains X, Y, Z coordinates and additional feature labels such as height values and volume contributions. These data are processed through point cloud library tools to form a complete dataset. In one implementation, this dataset can be used for subsequent machine learning model training, for example, in solder quality classification applications.

[0027] It should be noted that the generation process ensures the standardization of the dataset, such as uniform coordinate system and resolution, to support universality in different scenarios. In another embodiment, for scanning of high-density solder joint arrays, a structured light scanning method is used as a non-contact data acquisition method. This method projects a stripe pattern onto the solder joint surface and captures the deformed pattern through a camera to reconstruct the three-dimensional shape. When extracting the height distribution from the scanning data, a surface fitting algorithm is introduced to fit the curvature of the solder joint surface, thereby more accurately distributing the height information. The volume feature extraction combines Monte Carlo integration approximation to estimate the volume of irregular shapes. This implementation demonstrates the adaptability of the technical solution in complex soldering environments, such as integrated circuit board production lines. Further expansion, the extraction of volume-related features can include surface area integral calculation.

[0028] Specifically, the solder joint surface is divided into triangular meshes, and then the contribution volume of each triangular face is calculated. This detailed process helps to identify subtle solder deformations, such as volume changes under thermal stress. In this way, the generated point cloud dataset not only contains basic coordinates but also carries feature vectors, enhancing the analysis value of the dataset.

[0029] For example, in a manual soldering scenario, a portable 3D scanner can be used for non-contact scanning. After extracting the height distribution, the volume feature is obtained through a boundary detection algorithm, ensuring that the dataset is suitable for small electronic device repair. This diverse embodiment highlights the universality of the solution without going beyond the field of solder joint detection.

[0030] In one embodiment, the generated three-dimensional point cloud dataset can be further optimized in resolution.

[0031] Specifically, the number of points is reduced through downsampling techniques while preserving key features. This optimization enables efficiency in data storage and, in practical applications, enables fast transmission of the dataset for remote quality monitoring.

[0032] It can be understood that the above embodiments achieve the accuracy of the solder joint detection through objective steps, for example, when extracting the volume feature, subjective assumptions are avoided, ensuring calculations based on actual scanning data. In these embodiments, the technical solution can achieve reliable evaluation of the solder joint quality, for example, through the combination of height distribution and volume feature, early detection of welding defects, and improvement of production efficiency.

[0033] In step S102, based on the three-dimensional point cloud dataset, a geometric modeling algorithm is applied to construct a solder joint three-dimensional model to determine the solder paste climbing angle and the contact interface shape feature.

[0034] A three-dimensional point cloud dataset is obtained, and a specific algorithm is used to generate an isosurface from the point cloud data to construct an initial solder joint three-dimensional model and obtain the surface mesh of the model. According to the surface mesh of the model, surface normal vectors are extracted for the solder paste distribution area, a climbing vector is calculated as the angle between the normal vector and the base plane to determine the solder paste climbing angle. Through the solder paste climbing angle, an edge detection method is used to segment the boundary curve from the contact interface area, and the contact interface shape feature is determined according to the curvature analysis of the boundary curve. Key information is extracted from the contact interface shape feature, and the comprehensive analysis result of the solder paste climbing angle and the contact interface shape feature is obtained by fusing the solder paste climbing angle. For the comprehensive analysis result, the parameters of the model surface mesh are adjusted to optimize the precision of the solder joint three-dimensional model.

[0035] For example, in one embodiment, a three-dimensional point cloud dataset is first obtained, which is obtained by scanning the solder joint area with a laser scanner or a structured light camera and contains coordinate point information of the solder joint surface. These point cloud data reflect the geometric distribution of the solder joint in three-dimensional space and are used in the subsequent model construction process.

[0036] Specifically, the collection of the three-dimensional point cloud dataset needs to be performed immediately after welding to capture the actual shape of the solder paste and avoid data deviation caused by environmental interference. In this way, the accuracy and integrity of the dataset are ensured, providing a reliable basis for geometric modeling. Further, a geometric modeling algorithm is applied to construct a solder joint three-dimensional model. This algorithm is based on the surface fitting principle of point cloud data, such as using the least squares method to fit the surface, converting discrete point cloud into a continuous three-dimensional model. In this process, the point cloud is first denoised to remove abnormal points, and then the model surface is generated through a triangular meshing method.

[0037] Exemplarily, in the electronic circuit board soldering scenario, the model can accurately reproduce the connection pattern of the solder joint and the pin, helping to identify potential defects.

[0038] It should be noted that the core of the geometric modeling algorithm is iterative optimization, which ensures that the model is highly consistent with the actual solder joint geometric features, thereby supporting the accuracy requirements of subsequent feature extraction. The detailed implementation of this step includes initial mesh construction and surface smoothing. The initial mesh is generated by the Delaunay triangulation algorithm, and the surface smoothing adjusts the grid vertex position using the Laplace smoothing method to reduce surface roughness. Through these sub-steps, the three-dimensional model can accurately represent the three-dimensional structure of the solder joint, laying the foundation for angle and shape analysis. The technical goal of this process is to improve the automation level of solder joint quality evaluation. In practical applications, it can effectively reduce the error of manual detection and improve production efficiency.

[0039] Preferably, determining the solder paste climb angle is achieved by analyzing the surface slope of the three-dimensional model.

[0040] Specifically, the interface between the solder paste and the substrate is extracted from the model, and the inclination angle of the solder paste surface relative to the substrate plane is calculated.

[0041] In one embodiment, the vector method is used to calculate the angle: the vector of the solder paste edge point and the center point is selected, and the included angle between the vector and the horizontal plane is calculated to obtain the climb angle value.

[0042] For example, in chip soldering, if the climb angle is too small, it may indicate that the solder paste wettability is insufficient, and this feature helps to adjust the soldering parameters in a timely manner. In another possible implementation, the determination of the contact interface shape feature involves contour extraction and shape descriptor calculation. The contact interface region is segmented from the three-dimensional model, and then shape factors such as roundness or convexity are applied to quantify its features.

[0043] It can be understood that the roundness is obtained by comparing the ratio of the interface area to the surrounding circle area, and the convexity evaluates the convexity and concavity of the interface. These features are used in solder quality control to determine whether the interface is uniform to avoid false soldering or bridging problems. Further.

[0044] In one embodiment, the above steps are applied to the surface mount technology soldering scenario. First, the model is constructed, and then the climb angle and interface shape are calculated in sequence to ensure that the entire process is logically coherent. Through the test of multiple groups of point cloud data, this method can adapt to different solder joint sizes and provide consistent feature analysis results.

[0045] For example, in another embodiment, for the processing of ball grid array solder joints, the geometry modeling algorithm can adjust the mesh density to adapt to the spherical tin morphology, thereby accurately determining the climbing angle and the elliptical contact interface features. This diversified application demonstrates the versatility of the technical solution in the field of welding, which can realize efficient defect detection function.

[0046] In step S103, a machine learning algorithm is used to process the solder joint solid model to extract height deviation features and volume anomaly indicators from the model to determine the potential risk level of the solder joint false welding.

[0047] The height data and volume data are obtained from the solder joint solid model, the difference between the height data and the preset standard height is calculated to obtain the height deviation feature, and the ratio of the volume data to the preset standard volume is evaluated to determine the volume anomaly indicator. A convolutional neural network is used to construct a network structure, the height deviation feature and the volume anomaly indicator are input, and spatial patterns are extracted through convolution operation to obtain solder joint anomaly patterns. If the solder joint anomaly pattern exceeds the preset threshold, the anomaly pattern is classified and processed to determine the potential risk level of the solder joint false welding. According to the classification result of the solder joint anomaly pattern, a corresponding risk assessment report is generated. For high-risk solder joints in the risk assessment report, their spatial feature data are extracted and combined with the height deviation feature for secondary verification to determine the final risk level. Through the final risk level, targeted detection suggestions are output.

[0048] For example, in one embodiment, the processing of the solder joint solid model is first constructed by collecting three-dimensional data during the welding process.

[0049] Specifically, the solder joint solid model refers to the three-dimensional point cloud data of the solder joint obtained by laser scanning or optical imaging equipment, forming a digital model representing the geometric shape of the solder joint. This model contains height, width and volume information of the solder joint for subsequent analysis.

[0050] It should be noted that the machine learning algorithm is used here to automatically process these model data, avoiding the subjectivity of manual judgment. By inputting the solder joint solid model, the algorithm can learn the normal and abnormal patterns of the solder joint to extract key features. Further, the process of extracting height deviation features involves comparing the difference between the actual height of the solder joint and the standard height.

[0051] For example, in the context of welding electronic components, first, surface fitting is performed on the solid model to calculate the vertical distance from the top of the solder joint to the base, and then compared with the preset standard height threshold. If the deviation exceeds a certain range, such as plus or minus 5%, it is marked as a height deviation feature. This feature reflects the possible collapse or bulge problem of the solder joint, which directly affects the risk of false welding.

[0052] In one embodiment, a convolutional neural network is employed as part of the machine learning algorithm to perform convolution operations on the point cloud data of the model, automatically identifying abnormal regions of height distribution, thereby quantifying the deviation value. This process ensures the accuracy and efficiency of feature extraction.

[0053] Preferably, the extraction of the volume anomaly indicator is based on the overall volume calculation of the solder joint solid model.

[0054] Specifically, the volume anomaly indicator is obtained by integrating the spatial volume within the three-dimensional boundary of the solder joint, which is compared with the ideal solder joint volume.

[0055] For example, in the field of circuit board welding, the algorithm first performs meshing processing on the model, converting the point cloud into a voxel grid, and then accumulates the voxel volume to obtain the total value. If the calculated volume deviates from the standard value by more than 10%, it is considered an abnormal indicator. This indicator helps detect the risk of virtual welding caused by internal cavities or insufficient material in the solder joint.

[0056] It should be noted that the machine learning algorithm integrates a supervised learning method, using labeled solder joint data to train the model and learn the correlation between volume anomalies and virtual welding, thereby improving the accuracy of extraction.

[0057] In one possible implementation, the judgment of the potential virtual welding risk level of the solder joint combines the extracted height deviation features and volume anomaly indicators. Through a machine learning algorithm, such as a random forest classifier, these features are input into the model, and the risk probability score is output.

[0058] Specifically, the algorithm first fuses the feature vector, for example, normalizes the height deviation value and the volume anomaly value into a multi-dimensional vector, and then calculates the posterior probability through the classifier. If the probability exceeds 0.7, it is judged as a high risk level. This method is universal in welding quality control, for example, applied to the welding scene of mobile phone mainboards, ensuring that defects in mass production are identified in a timely manner. Further, to enhance the flexibility of the scheme, in another embodiment, additional temperature data can be introduced to assist in judgment.

[0059] Exemplarily, combined with the solder joint solid model and the temperature distribution during welding, the machine learning algorithm expands the feature set, calculates the influence of temperature on height deviation, and thereby more accurately assesses the virtual welding risk. This implementation demonstrates the multiple application scenarios of the technical solution in the same welding field.

[0060] It can be understood that through the above steps, the system can realize automatic analysis of the solder joint model. In actual operation, for example, in an automotive electronics welding production line, after extracting the features, the risk judgment result can be output as a visual report to guide subsequent repair. This processing method improves the objectivity and speed of detection, avoiding the limitations of traditional visual inspection.

[0061] In an embodiment, the training process of the machine learning algorithm includes a data augmentation step.

[0062] Specifically, more training samples are generated using synthetic solder joint models, simulating different biases and abnormal situations, to optimize the adaptability of the model to real welding scenarios. This training ensures the robustness of the algorithm in feature extraction.

[0063] Preferably, the calculation of the volume anomaly indicator can use the Monte Carlo method to approximate the volume value. In solder defect detection, this method estimates the abnormal proportion by randomly sampling points inside the model, combined with the height bias feature, to further refine the risk level judgment. In an embodiment, the entire process is integrated into an embedded system to realize real-time processing of solder joint solid models.

[0064] For example, in precision instrument welding, the algorithm directly obtains data from the scanning device, quickly outputs risk assessment results, and supports the application of continuous production lines.

[0065] Step S104, if the height bias feature exceeds the preset threshold range, an edge analysis algorithm is used to analyze the change of the solder climbing angle, and the bridge defect occurrence probability distribution is calculated.

[0066] The height bias feature is obtained from the welding image, and it is determined whether the height bias feature exceeds the preset threshold range to obtain an out-of-limit signal. If the out-of-limit signal indicates that the threshold range is exceeded, the solder climbing angle is extracted from the welding image according to the out-of-limit signal. An edge analysis algorithm is used to detect the edge contour of the solder climbing angle, and the slope change of the edge contour is obtained. The bridge defect change trend is determined by analyzing the slope change. For the bridge defect change trend, the occurrence frequency of the bridge defect change trend is obtained. The bridge defect occurrence probability distribution is calculated by matching the occurrence frequency with a preset probability model, and a defect risk assessment result is obtained.

[0067] Exemplarily, in one implementation, when a height deviation feature of the solder material is detected beyond a preset threshold range, the system initiates an edge analysis algorithm to process the relevant data. The height deviation feature refers to the difference between the actual height of the solder material on the solder joint and the standard height, for example, calculated from 3D image data collected by an optical sensor during the electronic circuit board soldering process. If this deviation exceeds the threshold, such as a 10% deviation from the standard height, it indicates that there may be a climbing anomaly that needs further analysis to assess the risk of defects. This threshold range is set based on historical soldering data to ensure the sensitivity of the detection. Further, the edge analysis algorithm is used to analyze the change in the solder material climbing angle. The solder material climbing angle refers to the angle at which the solder material extends upward along the edge of the pad after being heated and melted, which affects the uniformity of the distribution of the solder material in soldering. The algorithm first extracts the contour data of the solder material edge, such as identifying the edge pixel points through image processing techniques, and then calculates the angle change curve.

[0068] Specifically, the algorithm divides the edge into multiple paragraphs, measures the climbing slope for each paragraph, and compares the difference between consecutive time points to quantify the change amplitude. This analysis helps identify potential problems caused by uneven climbing.

[0069] Exemplarily.

[0070] In one possible implementation, the edge analysis algorithm uses gradient calculation methods to quantify the angle change. Gradient calculation involves differential processing of the edge curve, such as taking the coordinate difference of the edge points to obtain the slope value. If the angle change exceeds normal fluctuations, such as a sharp rise from 30 degrees to 60 degrees, it is marked as a high-risk area. This method is widely used in electronic component soldering scenarios and can adapt to different solder materials, such as lead-free solder or lead-containing solder, ensuring the accuracy of the analysis.

[0071] It should be noted that the calculation of the probability distribution of bridge defect occurrence is based on the above analysis results. Bridge defects refer to the connection of adjacent solder joints caused by excessive climbing of the solder material, forming a short circuit. The probability distribution is calculated through a statistical model, such as using a Gaussian distribution to fit the angle change data to estimate the likelihood of defect occurrence. The specific process includes collecting angle data from multiple solder samples, calculating the mean and variance, and then generating a probability density function. If the probability exceeds 50%, an alarm is triggered. This distribution calculation provides a quantitative basis to support the optimization of the soldering process.

[0072] In one embodiment, considering the application in different soldering environments, such as in surface mount technology, the system can be integrated into an automatic optical detection device.

[0073] Preferably, the tin material image is pre-processed, such as noise filtering, before applying the edge analysis algorithm to analyze the climb angle. When calculating the probability, temperature factors can be introduced, for example, if the welding temperature is higher than the standard value, the probability distribution will shift towards high risk. This implementation demonstrates the flexibility of the technical solution in the field of electronic manufacturing, which can handle variations in mass production.

[0074] Specifically, the implementation of the edge analysis algorithm can be divided into two modules: data acquisition and feature extraction. The data acquisition module obtains the three-dimensional model of the tin material through a laser scanner and extracts height and angle data. The feature extraction module calculates the deviation and variation, such as Fourier transform of the climb angle sequence to identify periodic patterns and predict the bridging risk. This modular design facilitates integration into existing welding production lines and improves detection efficiency. Further, when calculating the probability distribution, the Monte Carlo simulation method can be used to enhance accuracy.

[0075] For example, simulate various angle variation scenarios, generate thousands of random samples, and then count the frequency of bridging occurrence to form a distribution curve. This method is particularly useful in high-precision electronic product welding and can reduce human errors.

[0076] It can be understood that the effect of the technical solution is to identify potential defects early, for example, when the probability distribution is found to be biased towards high values in the calculation, the welding parameters such as heating time can be adjusted to reduce the incidence of bridging. This objective evaluation supports the technical features of the claims and provides a reliable defect prevention mechanism. In another implementation, for the scenario of fine pitch solder joints, the edge analysis algorithm can be optimized for real-time mode. The system continuously monitors the climb angle, and if the deviation exceeds the threshold, it immediately calculates the probability distribution and outputs a visual chart. This method is suitable for smartphone circuit board welding and ensures the universality of the technical solution.

[0077] Step S105, based on the bridging defect probability distribution, using a pattern recognition algorithm to distinguish normal solder joints from various defect categories, obtaining a defect classification result.

[0078] Bridge defect distribution data is obtained from the solder joint image, gray value and edge contour data of the solder joint image are extracted, and a feature vector is constructed. For the feature vector, the deviation degree from the preset normal distribution is calculated to obtain a defect probability threshold. The feature vector is input into a support vector machine algorithm using the defect probability threshold, different categories are separated by constructing a hyperplane, and a defect category classification result is determined. According to the defect category classification result, the matching degree of the labeled sample in the pre-established training data set and the classification result is obtained, and a classification result is generated. For the classification result, the relevance of the feature vector and the matching degree is analyzed, the defect classification output is optimized, and a final defect classification result is obtained. Through the defect classification result, the labeled sample in the training data set is updated, and a continuously optimized classification model is formed.

[0079] For example, in an embodiment, a bridge defect refers to an unnecessary conductive connection between adjacent solder joints during the welding process, which causes a short circuit in the circuit. The probability distribution of this defect can be constructed by statistically analyzing historical welding data, such as collecting multiple batches of electronic board welding samples and calculating the occurrence frequency of bridge defects under different welding parameters to form a probability distribution model. This model helps to predict potential defect risks and provides a basis for subsequent classification. Further, based on the bridge defect probability distribution, a pattern recognition algorithm is used for defect classification.

[0080] Specifically, a pattern recognition algorithm is a computational method that recognizes patterns by learning sample features. In solder defect detection, algorithms such as support vector machines or neural networks can be used. First, features such as solder joint shape, size, and brightness value are extracted from solder joint images or sensor data, which are combined with probability distributions to form input vectors.

[0081] For example, in an electronic component welding scenario, a set of solder joint image data is collected. The probability distribution is used to evaluate the likelihood of a bridge defect, and if the probability is higher than a threshold, it is marked as a potential defect area. Then, the pattern recognition algorithm learns the differences between normal solder joints and defective solder joints through a training phase. The training process includes inputting labeled samples into the algorithm and adjusting parameters to maximize classification accuracy. In practical applications, this algorithm can distinguish between normal solder joints, bridge defects, fake solder defects, and multiple tin defects, among other categories.

[0082] In a possible implementation, the implementation of the pattern recognition algorithm can use a k-nearest neighbor algorithm. The working principle of this algorithm is to calculate the distance between the features of a new solder joint and known samples, and select the majority class of the nearest k samples as the classification result.

[0083] For example, for a suspected bridging solder joint, the algorithm compares its probability distribution values and feature vectors with the similarity of the training set, thereby outputting a classification label. This method is universal in the welding production line and can adapt to the detection of electronic boards of different batches.

[0084] It should be noted that the construction process of the bridging defect occurrence probability distribution involves data statistical analysis.

[0085] Specifically, parameters such as temperature, pressure, and time are extracted from the welding equipment log, and the incidence of bridging defects under each parameter combination is calculated.

[0086] For example, if the temperature is too high, the bridging probability rises to 0.3, and the distribution model will record this association. The distribution is updated by the Bayesian method to ensure that the model dynamically adapts to new data. The key to this process is data cleaning and normalization to avoid noise interference with the accuracy of the distribution, thereby providing reliable input for pattern recognition. In another embodiment, for high-density solder joint boards, the pattern recognition algorithm can integrate a clustering step. First, unsupervised clustering is performed on the solder joints to group similar features, and then supervised classification of defect categories is performed based on probability distribution.

[0087] For example, after clustering, a bridging group is identified, and a decision tree algorithm is used to refine the distinction between normal and defective. This integrated approach improves the robustness of classification and is suitable for the production and detection of mobile phone circuit boards or computer motherboards in the same welding field.

[0088] Specifically, after obtaining the defect classification result, it can be output through a visualization interface, such as marking the defect location and category label. This helps operators to quickly intervene and reduce the rate of defective products. In welding quality control, this technical solution can achieve accurate differentiation of multiple defects, thereby supporting the optimization of automated production lines. Further, in one embodiment, if multiple modal data such as images are combined with temperature sensors, the pattern recognition algorithm can fuse these inputs. The fusion process includes feature-level fusion, which concatenates the probability distribution vector with the image features to form a comprehensive input, thereby improving classification accuracy.

[0089] For example, in automotive electronics welding, this fusion helps to distinguish between bridging and thermal damage defects.

[0090] It can be understood that the scalability of this method is reflected in parameter adjustment.

[0091] Preferably, by changing the algorithm threshold or distribution update frequency, different welding environments can be adapted without changing the core classification logic.

[0092] For example, in batch welding testing, after applying this solution, the defect detection accuracy can reach more than 95%, proving its effectiveness in actual business.

[0093] In step S106, the welding spot quality evaluation report is generated by combining the defect classification result and the three-dimensional point cloud data set, and the overall welding reliability index is determined.

[0094] The welding spot defect type is obtained from the three-dimensional point cloud data set, the defect position coordinates are determined, and the point cloud density distribution is obtained. According to the point cloud density distribution, a preset threshold is used to judge the welding surface morphology. If the point cloud density distribution is lower than the preset threshold, the defect position coordinates are fused to generate a quality evaluation report. Through the quality evaluation report, the welding spot defect type is obtained, and the reliability index calculation result is determined, wherein the reliability index calculation result is obtained by the ratio of the welding spot defect type to the point cloud density distribution. According to the reliability index calculation result, a report generation template is obtained from the data fusion processing, wherein the data fusion processing is obtained by merging the defect classification result and the three-dimensional point cloud data set. The report generation template is used to determine the overall welding reliability index in combination with the reliability index calculation result. Through the overall welding reliability index, a final welding quality analysis report is generated.

[0095] For example, in an embodiment, the welding spot quality evaluation system first obtains the defect classification result and the three-dimensional point cloud data set. These defect classification results are obtained by analyzing the image data of the welding area.

[0096] For example, the system uses a convolutional neural network model to extract features and classify welding spot images, and identifies defect types such as cracks, pores, or incomplete fusion.

[0097] Specifically, the defect classification process includes image preprocessing, feature map generation, and softmax classification output, ensuring that the classification accuracy reaches a preset threshold. In this way, the system can quantitatively represent the defect type, providing basic data for subsequent evaluation. Further, the three-dimensional point cloud data set is collected by scanning the welding spot using a laser scanner to generate three-dimensional coordinate data containing tens of thousands of points.

[0098] It should be noted that the three-dimensional point cloud data set reflects the geometry and surface morphology of the welding spot.

[0099] For example, each point in the point cloud data contains x, y, z coordinates and intensity values. The system performs registration processing on the point cloud to align multiple scanning views into a unified coordinate system, avoiding data drift.

[0100] In a possible implementation, the point cloud registration uses the iterative closest point algorithm to achieve accurate alignment by minimizing the distance between points. The key to this registration process is to handle noise points.

[0101] For example, isolated points are removed by statistical filtering to ensure the purity of point cloud data. Subsequently, the system extracts geometric features of the point cloud, such as curvature and normal vector, which are used to quantify the flatness and deformation of the weld.

[0102] For example, in weld evaluation, if the point cloud shows abnormal curvature, it may correspond to the crack type in defect classification, thereby realizing cross-validation of defect results and point cloud data. Based on the above defect classification results and three-dimensional point cloud data set, the system generates a weld quality evaluation report.

[0103] Preferably, the report includes a defect statistics section, a point cloud analysis section, and a comprehensive evaluation section.

[0104] Specifically, the defect statistics section lists the number and location of each type of defect.

[0105] For example, the report shows that the porosity defect accounts for 30% of the total defects and is marked on the weld coordinate map. The point cloud analysis section describes geometric indicators such as the average surface roughness, which is obtained by calculating the distance variance of adjacent points in the point cloud. This report generation process ensures data visualization and structured output, facilitating user understanding of welding quality.

[0106] In one embodiment, the determination of the overall weld reliability index uses a weighted fusion model. First, the system converts the defect classification results into a defect severity score.

[0107] For example, a crack defect is given a higher weight because it has a greater impact on structural strength.

[0108] Specifically, the severity score calculation process involves a defect type weight table, where the crack weight is 0.5, the porosity is 0.3, and the incomplete fusion is 0.2, with a total normalization of 1. Then, combined with the three-dimensional point cloud data set, the system calculates the geometric reliability score, such as the completeness of the weld by the volume deviation of the point cloud. The volume deviation is calculated based on the difference between the ideal weld model and the actual point cloud.

[0109] For example, if the actual point cloud volume is less than 95% of the ideal value, the corresponding score is deducted. The key to this fusion process is multi-factor weighting.

[0110] For example, the reliability index R = (1-D) * G, where D is the total score of defect severity and G is the geometric score. In this way, the system outputs the overall reliability index.

[0111] For example, an R value greater than 0.8 indicates high reliability. This index determination process not only quantifies the welding quality but also provides data support for optimizing welding parameters, which can reduce the rate of defective products in actual production.

[0112] Exemplarily, in a car part welding scenario, the system is applied to body joint evaluation. The defect classification result shows a slight porosity at a welding joint, and the three-dimensional point cloud dataset reveals a slight surface depression. After report generation, the reliability index is calculated as 0.75, indicating that the welding current needs to be adjusted. This scenario demonstrates the versatility of the technical solution in the same field. Further, the system can introduce an optional threshold adjustment mechanism. In one embodiment, the user sets a defect severity threshold.

[0113] For example, when the severity exceeds 0.4, the report is automatically marked as a warning level. This mechanism enhances the flexibility of the evaluation without changing the core fusion logic.

[0114] It can be understood that the feature extraction module of the three-dimensional point cloud dataset further refines the processing process.

[0115] Specifically, feature extraction includes point cloud segmentation and clustering.

[0116] For example, using Euclidean distance clustering, the point cloud is divided into welding joint body and edge area. After clustering, the density of each area is calculated.

[0117] For example, the density is obtained by dividing the number of points by the volume, and if the density is lower than the average value, it indicates a potential void. This detailed extraction process supports the accurate calculation of the reliability index, plays a key role in welding quality control, and can achieve early detection of minor defects, thereby improving overall production efficiency. In another embodiment, for electronic component welding, the system processes small welding joints. The defect classification identifies an incomplete fusion defect, and the point cloud data shows uneven thickness. The reliability index is calculated as 0.85 by adjusting the weight, and the report suggests optimizing the laser power. This embodiment covers different scale scenarios in the same field, demonstrating the adaptability of the solution.

[0118] Preferably, the report generation module includes a data visualization function.

[0119] For example, a three-dimensional point cloud rendering image is generated with defect labels superimposed. This visualization displays defect severity through color coding, facilitating intuitive evaluation. Finally, when determining the overall welding reliability index, the system considers environmental factors for correction.

[0120] For example, the geometry score is adjusted based on temperature data. This correction process ensures the accuracy of the index in the actual welding environment.

[0121] Step S107, if the overall welding reliability index is below the preset threshold range, an alarm mechanism is triggered to obtain the coordinates of the relevant welding joint position on the production line.

[0122] The current value is obtained from the reliability index, compared with the preset threshold value, if the current value is lower than the preset threshold value range, the notification module is activated through the alarm mechanism, and the trigger signal is obtained. According to the trigger signal, the related welding point coordinates are extracted from the production line information, the position details are determined by fusing the defect detection results, and the coordinate data set is obtained. For the coordinate data set, combined with quality evaluation, a maintenance scheduling record is generated, the welding monitoring demand is judged, and the overall response scheme is obtained. Through the overall response scheme, the production line monitoring parameter is updated, and the adjustment instruction is generated. According to the adjustment instruction, the production line information is corrected in real time, and the optimized monitoring strategy is determined. For the optimized monitoring strategy, the reliability index data is continuously collected, and the comparison step is repeatedly executed to form a closed-loop control process.

[0123] Illustratively, in one embodiment, the welding quality evaluation system performs threshold checking on the calculated overall welding reliability index.

[0124] Specifically, the preset threshold value range can be set to 0.7 to 1.0, and a value lower than 0.7 indicates a potential risk. This checking process is realized by a comparison algorithm.

[0125] For example, the reliability index R is compared with the lower threshold value, and if R is less than the threshold value, the system is marked as an abnormal state.

[0126] It should be noted that the setting of the threshold value range is based on the statistical analysis of historical welding data to ensure the production needs of different welding materials. In this way, the system can timely identify the welding points with insufficient reliability and avoid continuous defects in the production chain. Further, when the overall welding reliability index is lower than the preset threshold value range, the system triggers the alarm mechanism.

[0127] Preferably, the alarm mechanism includes a sound alarm and a visual notification.

[0128] For example, activate the buzzer on the production line and display a red warning icon on the monitoring screen.

[0129] Specifically, after the alarm is triggered, the system sends a message to the operator containing the welding point number and the abnormal description. The design of this mechanism aims to respond quickly and ensure the interruption and correction of the welding process.

[0130] In one possible implementation, the alarm is also linked to the production line control system to automatically suspend the related welding equipment and prevent further defects. This linkage process is realized through an interface protocol, which emphasizes the real-time nature of the system in welding quality control.

[0131] Illustratively, in the automobile body welding scenario, if the reliability index is 0.65 lower than the threshold value, the system triggers an alarm and obtains the welding point position coordinate information.

[0132] Specifically, coordinate information is extracted from sensors on the production line, including x, y, and z axis data.

[0133] For example, a solder joint is located at coordinates (150, 200, 50) millimeters. The system associates these coordinates with the production layout diagram to generate an annotation map, facilitating positioning by maintenance personnel. This acquisition process involves database queries to ensure coordinate accuracy. In another embodiment, for electronic circuit board soldering, the coordinates acquired after an alarm is triggered are used for automatic positioning of the robotic arm, enabling rapid re-inspection. This scenario demonstrates the application of the technical solution in precision soldering.

[0134] Understandably, the process of obtaining the coordinates of relevant weld points on the production line further includes a data verification step.

[0135] For example, the system compares the coordinates with the point cloud dataset to ensure there is no offset. This verification enhances the reliability of the coordinate information, supporting subsequent adjustments to welding parameters.

[0136] In one embodiment, if multiple weld joints malfunction simultaneously, the system prioritizes retrieving a list of coordinates and sorting them by severity. This approach facilitates batch management and improves the overall efficiency of welding production.

[0137] Step S108: Based on the solder joint position coordinate information, use a data optimization algorithm to process the subsequent scanning data and determine whether the optimized data meets the preset accuracy standard.

[0138] Initial position data of solder joints is acquired through coordinate acquisition. The initial position data is then fitted using the least squares method to obtain optimized coordinate values, with the goal of minimizing the sum of squared errors. Subsequent scan data is compared with these optimized coordinate values; if the deviation between the optimized coordinate values ​​and the scan data is less than a preset threshold, the data optimization is deemed effective. The optimized scan data is then acquired and grouped using a clustering algorithm to obtain cluster structure data. The optimized data is then assessed to determine if it meets a preset accuracy standard; if it does, the scan data grouping is deemed effective. The optimized coordinate value database is updated using the grouped effective scan data for subsequent solder joint position calibration.

[0139] For example, in one implementation, based on the solder joint position coordinate information, the three-dimensional coordinate data of the solder joint on the circuit board is first obtained, and this data is acquired by a laser scanning device.

[0140] Specifically, the welding point position coordinate information includes x, y, and z axis values, which are used to locate the precise position of the welding point on the board surface. In this way, the basis for subsequent data processing is ensured to be accurate. Further, a data optimization algorithm is used to process the subsequent scanning data. The data optimization algorithm can be a fitting algorithm based on the least squares method, which is used to smooth the noise in the scanning data.

[0141] Illustratively, in the welding quality detection scenario, the subsequent scanning data refers to multiple laser scanning results of the same welding point area, which may contain deviations caused by device jitter or environmental interference. The algorithm process includes taking the coordinate information as a reference point, calculating the error of the scanning data, and then iteratively adjusting the data point position to approach the ideal curve. Through this processing, data fluctuations are reduced and overall consistency is improved.

[0142] It should be noted that the specific implementation of the data optimization algorithm can include a filtering step.

[0143] For example, a Gaussian filter is first applied to remove high-frequency noise, and then a coordinate system calibration is performed based on the welding point coordinate information to ensure that the optimized data is aligned with the initial coordinates. This method is commonly used in circuit board welding detection and is suitable for different batches of board production.

[0144] In one possible implementation, it is determined whether the optimized data meets a predetermined accuracy standard. The predetermined accuracy standard is defined as a deviation of the optimized data from the reference coordinates being less than 0.05 millimeters. The determination process involves calculating the Euclidean distance and summing the distances of each data point from the welding point coordinates. If the average deviation is below the threshold, it is determined that the standard is met. For example, on an actual welding line, if the optimized data meets the standard, the production process continues, otherwise an alarm is triggered to re-scan. This determination ensures the reliability of the welding quality.

[0145] Preferably, in another embodiment, multiple welding point scenarios are considered. For the position coordinate information of multiple welding points on a circuit board, the data optimization algorithm is extended to a batch processing mode. The specific process is to first group the welding point coordinates, and then apply the optimization algorithm to process the respective subsequent scanning data in parallel. When judging the accuracy, an overall board surface accuracy indicator is introduced, such as calculating the root mean square value of the deviations of all welding points. This way shows the flexibility of the technical solution in complex circuit board detection, which can handle high-density welding point arrangements.

[0146] It can be understood that the technical effect of this technical solution is to improve the accuracy of data processing. In the field of welding detection, through the optimization algorithm and accuracy judgment, human errors are reduced and production efficiency is improved.

[0147] For example, on an electronic component assembly line, when the optimized data meets the standard, it can be directly used for automatic welding adjustment, avoiding defect accumulation.

[0148] Specifically, further expand the embodiment, introduce adaptive optimization algorithm. Based on the solder joint coordinate information, the algorithm adjusts parameters according to the real-time variability of the scanning data, such as dynamically setting the filter window size. This adaptive mechanism is particularly suitable in the welding workshop with variable environment, ensuring the stability of the judgment result.

[0149] For example, in a specific scenario, the solder joint coordinate information is derived from the initial scanning, and the subsequent scanning data comes from continuous production monitoring. The optimization algorithm eliminates the offset through coordinate matching and compares the consistency of the data before and after the judgment. If it meets the preset accuracy, it confirms that the solder joint quality is qualified. This example highlights the practicality of the scheme.

[0150] The present application provides a lithium battery protection plate solder joint quality intelligent evaluation and optimization system, mainly including: data acquisition and scanning module, for using non-contact data acquisition method to scan the solder joint surface, extracting solder joint height distribution information and volume related features from scanning data, generating solder joint three-dimensional point cloud data set; three-dimensional modeling and feature extraction module, for constructing solder joint three-dimensional model based on the three-dimensional point cloud data set, determining solder joint tin material climbing angle and contact interface shape feature by applying geometric modeling algorithm; machine learning analysis module, for processing the solder joint three-dimensional model by using machine learning algorithm, extracting height deviation feature and volume anomaly index from the model, judging the potential risk level of solder joint false welding; threshold judgment and edge analysis module, for if the height deviation feature exceeds the preset threshold range, using edge analysis algorithm to analyze the change of the tin material climbing angle, calculating the probability distribution of bridge defect occurrence; bridge defect probability calculation module, for based on the bridge defect probability distribution, using pattern recognition algorithm to distinguish normal solder joint and various defect categories, obtaining defect classification result; defect classification and identification module, for combining the defect classification result and the three-dimensional point cloud data set, generating solder joint quality evaluation report, determining overall welding reliability index; quality evaluation and report generation module, for if the overall welding reliability index is lower than the preset threshold range, triggering alarm mechanism, obtaining relevant solder joint position coordinate information on the production line; alarm triggering and data optimization module, for based on the solder joint position coordinate information, using data optimization algorithm to process subsequent scanning data, judging whether the optimized data meets the preset accuracy standard. The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for intelligent evaluation and optimization of the welding quality of lithium battery protection plates, characterized in that, The method comprises the following steps: Step S101, using a non-contact data acquisition method to scan the surface of the solder joint, extracting solder joint height distribution information and volume-related features from the scanning data, and generating a solder joint three-dimensional point cloud data set; Step S102, based on the three-dimensional point cloud data set, applying a geometric modeling algorithm to construct a solder joint three-dimensional model, determining the solder joint tin material climbing angle and the contact interface shape feature; Step S103, using a machine learning algorithm to process the solder joint three-dimensional model, extracting height deviation features and volume anomaly indicators from the model, and determining the potential false solder risk level of the solder joint; The step S103 further comprises: Obtain height data and volume data from the solder joint three-dimensional model, calculate the difference between the height data and the preset standard height to obtain the height deviation feature, and evaluate the ratio of the volume data to the preset standard volume to determine the volume anomaly indicator; Using a convolutional neural network to construct a network structure, inputting the height deviation feature and the volume anomaly indicator, and extracting spatial patterns through convolution operation to obtain solder joint anomaly patterns; If the solder joint anomaly pattern exceeds the preset threshold, the anomaly pattern is classified and processed to determine the potential false solder risk level of the solder joint; According to the classification result of the solder joint anomaly pattern, a corresponding risk assessment report is generated; For high-risk solder joints in the risk assessment report, extract their spatial feature data and combine the height deviation feature for secondary verification to determine the final risk level; Through the final risk level, output a targeted detection suggestion; Step S104, if the height deviation feature exceeds the preset threshold range, use an edge analysis algorithm to analyze the change of the tin material climbing angle, and calculate the bridge defect occurrence probability distribution; Step S105, based on the bridge defect occurrence probability distribution, use a pattern recognition algorithm to distinguish normal solder joints from various defect categories to obtain a defect classification result; The step S105 further comprises: Obtain bridge defect distribution data from the solder joint image, extract the gray value and edge contour data of the solder joint image, and construct a feature vector; For the feature vector, calculate the deviation degree from the preset normal distribution to obtain a defect probability threshold; Using the defect probability threshold, input the feature vector into a support vector machine algorithm, separate different categories by constructing a hyperplane, and determine the defect category classification result; According to the defect category classification result, obtain the matching degree of the labeled sample and the classification result from the pre-established training data set to generate a classification result; For the classification result, analyze the relevance of the feature vector and the matching degree, optimize the defect classification output, and obtain the final defect classification result; The labeled samples in the training data set are updated through the defect classification result to form a continuously optimized classification model; step S106, a welding spot quality evaluation report is generated by combining the defect classification result and the three-dimensional point cloud data set to determine an overall welding reliability index; step S107, if the overall welding reliability index is lower than a preset threshold range, an alarm mechanism is triggered to obtain relevant welding spot position coordinate information on a production line; step S108, based on the welding spot position coordinate information, a data optimization algorithm is used to process subsequent scanning data to determine whether the optimized data meets a preset precision standard. 2.The method of claim 1, wherein The step S101 comprises: a welding spot surface is scanned in a non-contact manner to obtain scanning data, height distribution information is extracted from the scanning data to obtain height distribution information of the welding spot; based on the height distribution information, a volume-related feature extraction method is used to obtain welding spot volume-related features to determine the volume-related features; through fusion processing of the volume-related features and the height distribution information, a welding spot three-dimensional point cloud data set is generated to obtain the welding spot three-dimensional point cloud data set; data calibration processing is performed on the welding spot three-dimensional point cloud data set to obtain calibrated three-dimensional point cloud data to determine the calibrated three-dimensional point cloud data; if the calibrated three-dimensional point cloud data meets a preset precision threshold, the calibrated three-dimensional point cloud data is output as the final welding spot three-dimensional model data. 3.The method of claim 1, wherein The step S102 comprises: a three-dimensional point cloud data set is obtained, and a specific algorithm is used to generate an isosurface from the point cloud data to construct an initial welding spot solid model to obtain a surface mesh of the model; based on the model surface mesh, a surface normal vector is extracted for a solder distribution area, a climbing vector is calculated as an included angle between the normal vector and a base plane to determine a solder climbing angle; through the solder climbing angle, an edge detection method is used to segment a boundary curve from a contact interface area, and a contact interface shape feature is determined according to curvature analysis of the boundary curve; key information is extracted from the contact interface shape feature, and a comprehensive analysis result of the solder climbing angle and the contact interface shape feature is obtained by fusing the solder climbing angle; for the comprehensive analysis result, the parameters of the model surface mesh are adjusted to optimize the precision of the welding spot solid model. 4.The method of claim 1, wherein The step S104 comprises: height deviation features are obtained from a welding image to determine whether the height deviation features exceed a preset threshold range to obtain an out-of-limit signal; if the out-of-limit signal indicates that the threshold range is exceeded, a solder climbing angle is extracted from the welding image according to the out-of-limit signal; an edge analysis algorithm is used to detect an edge profile of the solder climbing angle to obtain a slope change of the edge profile; a bridging defect change trend is determined by analyzing the slope change; for the bridging defect change trend, a frequency of occurrence of the bridging defect change trend is obtained; a bridging defect occurrence probability distribution is calculated by matching the frequency with a preset probability model to obtain a defect risk evaluation result. 5.The method of claim 1, wherein The step S106 comprises: Obtaining a welding defect type from a three-dimensional point cloud dataset, determining a defect position coordinate, and obtaining a point cloud density distribution; According to the point cloud density distribution, a preset threshold is used to judge the welding surface morphology, and if the point cloud density distribution is lower than the preset threshold, the defect position coordinate is fused to generate a quality evaluation report; Through the quality evaluation report, the welding defect type is obtained, and a reliability index calculation result is determined, wherein the reliability index calculation result is obtained by the ratio of the welding defect type to the point cloud density distribution; According to the reliability index calculation result, a report generation template is obtained from data fusion processing, wherein the data fusion processing is obtained by merging the defect classification result and the three-dimensional point cloud dataset; Using the report generation template, the reliability index calculation result is combined to determine the overall welding reliability index; Through the overall welding reliability index, a final welding quality analysis report is generated. 6.The method of claim 1, wherein The step S107 includes: Obtaining a current value from the reliability index, comparing it with a preset threshold, and if the current value is lower than the preset threshold range, activating a notification module through an alarm mechanism to obtain a trigger signal; According to the trigger signal, relevant welding point coordinates are extracted from the production line information, and position details are determined by fusing defect detection results to obtain a coordinate dataset; For the coordinate dataset, a maintenance scheduling record is generated in combination with quality evaluation, welding monitoring requirements are judged, and an overall response scheme is obtained; Through the overall response scheme, the production line monitoring parameters are updated, and an adjustment instruction is generated; According to the adjustment instruction, the production line information is corrected in real time, and an optimized monitoring strategy is determined; For the optimized monitoring strategy, reliability index data is continuously collected, the comparison step is repeatedly executed, and a closed-loop control process is formed. 7.The method of claim 1, wherein The step S108 includes: Obtaining welding initial position data, which is formed by coordinate collection; Using the least square method to fit the initial position data, obtaining optimized coordinate values, and the optimized coordinate values take the minimum sum of squared errors as the target; According to the optimized coordinate values, subsequent scanning data is compared, and if the deviation between the optimized coordinate values and the scanning data is less than a preset threshold, it is determined that the data optimization is effective; Obtaining scanning data with effective data optimization, grouping the scanning data through a clustering algorithm to obtain cluster structure data; According to the cluster structure data, it is judged whether the optimized data meets the preset precision standard, and if it meets, it is determined that the scanning data grouping is effective; Updating the optimized coordinate value database through the grouping effective scanning data for subsequent welding point position calibration.

8. A system for intelligent evaluation and optimization of the quality of welds of lithium battery protection plates, characterized by, The system is used for implementing the welding point quality intelligent evaluation and optimization method of the lithium battery protection plate according to any one of claims 1-7, and the system comprises: a data acquisition and scanning module, which is used for scanning the welding point surface by using a non-contact data acquisition method, extracting welding point height distribution information and volume related features from the scanning data, and generating a welding point three-dimensional point cloud data set; a three-dimensional modeling and feature extraction module, which is used for constructing a welding point three-dimensional model by applying a geometric modeling algorithm based on the three-dimensional point cloud data set, and determining a solder material climbing angle and a contact interface shape feature of the welding point; a machine learning analysis module, which is used for processing the welding point three-dimensional model by using a machine learning algorithm, extracting height deviation features and volume anomaly indicators from the model, and judging a potential false welding risk level of the welding point; a threshold value judgment and edge analysis module, which is used for analyzing a change of the solder material climbing angle by using an edge analysis algorithm if the height deviation features exceed a preset threshold value range, and calculating a bridge defect occurrence probability distribution; a bridge defect probability calculation module, which is used for distinguishing normal welding points and various defect categories by using a pattern recognition algorithm based on the bridge defect occurrence probability distribution, and obtaining a defect classification result; a defect classification and identification module, which is used for generating a welding point quality evaluation report and determining an overall welding reliability index by combining the defect classification result and the three-dimensional point cloud data set; a quality evaluation and report generation module, which is used for triggering an alarm mechanism and obtaining relevant welding point position coordinate information on a production line if the overall welding reliability index is lower than a preset threshold value range; and an alarm triggering and data optimization module, which is used for processing subsequent scanning data by using a data optimization algorithm based on the welding point position coordinate information, and judging whether the optimized data meets a preset precision standard.

Images