Automatic laser welding method and system for aluminum alloy battery shell

Abstract

The invention provides an aluminum alloy battery shell automatic laser welding method and system.The method comprises the steps that firstly, three-dimensional geometric data and contact position coordinates of the surface of a shell are obtained through a laser scanner, a digital model is constructed, then contact material characteristics are extracted, and initial welding parameters are determined; a real-time visual system and an ultrasonic sensor are used for correcting the position of a welding head, stable contact is established, energy distribution is optimized according to the heat conductivity and the geometrical shape of a material, preheating treatment is executed, in the welding process, temperature changes are monitored through infrared thermal imaging, laser parameters are adjusted, stable penetration depth is kept, and finally the quality of a welding seam is evaluated through air tightness detection. By means of multi-sensor fusion and self-adaptive control, high-precision and high-quality laser welding of the battery shell with the complex shape is achieved, and welding reliability and production efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to an automatic laser welding method and system for aluminum alloy battery shells. Background Art

[0002] As a crucial technology in the new energy sector, automated battery casing welding is directly related to battery safety, reliability, and service life. Its research and optimization are crucial for advancing energy storage technology. With the widespread application of battery technology in electric vehicles, energy storage systems, and other applications, casing welding quality has become a key factor influencing product performance. Traditional welding methods often exhibit significant shortcomings when dealing with complex casing structures and diverse materials.

[0003] Existing solutions mostly rely on manual experience or automated equipment with fixed parameters, which are difficult to adapt to changes in material properties and geometric shapes at the joints, resulting in unstable weld quality and even leaking or over-welding. These limitations are particularly prominent in modern battery manufacturing, which requires high precision and high consistency, and urgently need technological breakthroughs. Although automated welding technology has made certain progress, the key issue of preventing joints from leaking has not been completely solved. Current methods have defects in handling precise control at the joints, especially when positioning accuracy is insufficient and energy regulation is inflexible, making it difficult to ensure the uniformity and density of the welds.

[0004] The core challenges lie in the following technical factors: first, controlling the relative position of the weld head and the joint; second, adaptively adjusting welding parameters; and third, optimizing the joint design. Failure to effectively coordinate these factors can lead to fluctuations in weld quality and even lead to technical difficulties such as reduced leak-proofing performance. For example, positioning deviations can cause the weld to deviate from the intended area, while improper energy control can damage the integrity of the joint material, compromising the sealing of the housing.

[0005] Therefore, achieving high-precision positioning and adaptive energy control at the joints in automated battery casing welding is a key issue in improving weld quality and leak-proofing. Solving this problem requires optimizing the entire process, from joint design and positioning systems to energy regulation, to ensure the stability and reliability of the welding process. Summary of the Invention

[0006] In order to solve the technical problems raised in the above background technology, the first aspect of the present invention provides an automatic laser welding method for aluminum alloy battery shells, the method comprising: S1, obtains the three-dimensional geometric data of the battery shell surface and the coordinates of the contact positions, collects high-resolution point cloud information through a laser scanner, builds a digital model of the complex shape of the shell, and obtains the accurate spatial distribution of the contact area; S2, extracting material property data at the joint from the digital model, matching thermal conductivity and melting point parameters with a preset material property database, determining the welding difficulty coefficient of the joint area, and determining the initial welding parameter range; S3, within the initial welding parameter range, the real-time vision system is used to detect the relative position deviation between the welding head and the joint. If the relative position deviation value exceeds the relative position deviation threshold, the servo motor is used to adjust the welding head posture to obtain the corrected position coordinates; S4, based on the corrected position coordinates, an ultrasonic sensor is used to measure the distance change trend between the joint surface and the welding head. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state; S5, after obtaining a stable contact state, extract thermal conductivity information from the material property data and calculate the energy input requirement based on the contact geometry. If the energy requirement exceeds the upper limit of the current energy parameter, the laser power adaptive adjustment module allocates energy to obtain the optimized energy distribution; S6, using the optimized energy distribution, drives the laser welding equipment to perform preheating of the joint area, adjusts the preheating time and intensity according to the changing trend of material properties, determines the uniformity of the initial weld formation, and determines the temperature field after the preheating is completed; S7, based on the temperature field after preheating, uses an infrared thermal imager to monitor the real-time temperature changes at the joint. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data; S8, extracting weld width and density indicators from the stable weld penetration data, combining them with the joint design parameter indicators to compare with the quality threshold. If the design parameter indicators are lower than the quality threshold, the welding path is optimized through the secondary energy compensation algorithm to obtain the final weld quality parameters; S9. After obtaining the final weld quality parameters, use airtightness testing equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

[0007] Optionally, step S1, obtaining three-dimensional geometric data of the battery housing surface and the coordinates of the contact positions, collecting high-resolution point cloud information using a laser scanner, constructing a digital model for the complex shape of the housing, and obtaining the precise spatial distribution of the contact area, includes: Step S11, collecting point cloud information of the battery housing surface by a laser scanner to obtain high-resolution three-dimensional data; Step S12: pre-processing the collected three-dimensional data, using Gaussian filtering to remove noise and obtain smooth surface geometric information; Step S13, extracting the joint positions from the smooth surface geometry information using curvature analysis to determine the precise coordinates; Step S14, constructing a digital model using NURBS surface modeling based on the precise coordinates to obtain a mathematical expression of the complex shape; Step S15, performing Delaunay triangulation on the mathematical expression to obtain the spatial distribution of the contact area; Step S16: If there are outliers in the spatial distribution, they are identified and removed using DBSCAN clustering, and the completeness of the regional analysis is determined based on a preset distribution threshold; Step S17: Obtain the final three-dimensional geometric data of the battery housing through the spatial distribution after integrity judgment.

[0008] Optionally, in step S15, Delaunay triangulation is performed on the mathematical expression to obtain the spatial distribution of the contact area, including: a side length of a triangular mesh near the contact is not greater than 0.5 mm.

[0009] Optionally, in step S16, if there are outliers in the spatial distribution, DBSCAN clustering is used to identify and eliminate them, and the completeness of the regional analysis is judged based on a preset distribution threshold, including: Step S161, extracting data from the spatial distribution and using DBSCAN clustering to identify an outlier coordinate set; Step S162: Delete the data points at the corresponding spatial positions according to the abnormal point coordinate set, and generate an updated spatial distribution matrix; Step S163, based on the updated spatial distribution matrix, calculating the coordinates of the convex hull polygon vertices as the boundary range; Step S164: If the convex hull area exceeds the distribution threshold, the DBSCAN neighborhood radius is reduced to regenerate the outlier set, and the elimination process is repeated until the convex hull area meets the distribution threshold; Step S165: using a morphological dilation operation to determine the boundary continuity of the final boundary coordinate set, and outputting a connected component label map; Step S166, extracting the three regions with the largest areas from the connected component label graph, calculating their centroid coordinates using K-means clustering, and generating a set of key region polygon vertices; Step S167: outputting spatial distribution result data with area identification according to the key area polygon vertex set.

[0010] Optionally, the step S161 extracts data from the spatial distribution and uses DBSCAN clustering to identify a set of abnormal point coordinates, including: setting a neighborhood radius of 0.5 mm and a minimum number of samples of 5.

[0011] Optionally, in step S164, if the convex hull area exceeds the distribution threshold of 1000 square meters, the DBSCAN neighborhood radius is reduced to regenerate the outlier point set, and the elimination process is repeated until the convex hull area meets the distribution threshold, including: reducing the DBSCAN neighborhood radius to 0.3 mm.

[0012] Optionally, step S2 extracts material property data at the joint from the digital model, matches thermal conductivity and melting point parameters in combination with a preset material property database, determines the welding difficulty coefficient of the joint area, and determines the initial welding parameter range, including: Step S21, extracting temperature field data at the junction from the digital model using a finite element analysis tool, and calculating thermal conductivity and melting point parameters; Step S22, matching thermal conductivity and melting point parameters from a preset property database to generate a material property distribution map of the contact area; Step S23, based on the material property distribution diagram, using the heat conduction equation to calculate the heat flux density difference value as a welding difficulty indicator; Step S24, adjusting the matching range of thermal conductivity and melting point parameters based on the heat flux density difference value to determine the difficulty coefficient; Step S25, using the random forest algorithm of scikit-learn, inputting historical welding parameters and heat flux density difference values, and predicting the optimized thermal conductivity range; Step S26: Verify the uniformity of the temperature field distribution within the optimization range through the ANSYS simulation tool. If the maximum temperature difference exceeds the temperature difference threshold, narrow the thermal conductivity range and finally determine the initial welding parameter range.

[0013] Optionally, step S25 uses the random forest algorithm of scikit-learn to input historical welding parameters and heat flux density difference values ​​to predict the optimized thermal conductivity range, including: Step S251, using a preset database to obtain historical welding parameters and heat flux density distribution, and obtaining an input parameter set through data cleaning; Step S252 , processing the input parameter set using the random forest algorithm of scikit-learn, and combining the heat flux density difference to determine the prediction result data; Step S253, calculating the thermal conductivity interval value according to the prediction result data, and adjusting the parameter matching range if the thermal conductivity interval value exceeds a preset thermal conductivity interval value threshold; Step S254, obtaining the adjusted parameter matching range, and updating the welding condition data using the linear regression algorithm of scikit-learn; Step S255: Generate heat flux density distribution based on welding condition data using ANSYS tools and determine the distribution uniformity; Step S256, extracting and optimizing thermal conductivity through ANSYS tools to determine the final thermal conductivity interval value; Step S257: Update the historical data sample according to the final thermal conductivity interval value to obtain an optimized input parameter set.

[0014] Optionally, in step S9, after obtaining the final weld quality parameters, using airtightness testing equipment to measure the sealing performance data at the shell joints, judging the leak-proof effect based on the test results, and determining the reliability of the welding process, including: Step S91, collecting weld quality parameter data through sensors and storing them as an initial data set; Step S92: Using airtightness testing equipment to perform pressure testing on the shell contacts to obtain sealing performance data; Step S93: If the sealing performance data exceeds the sealing performance threshold, outliers are removed using a box plot method to obtain a corrected performance data set; Step S94, calculating the leakage prevention effect index using the standard deviation method based on the corrected performance data set to determine the leakage prevention status; Step S95: Analyze the correlation between the welding process and the leakage prevention effect using a random forest algorithm to obtain a ranking of influencing factors. Step S96, after obtaining the ranking of the influencing factors, predict the reliability of the welding process using a logistic regression algorithm to determine the reliability level; Step S97: extract key features from the reliability levels, use K-means clustering to divide the welding process optimization direction, and determine the improved parameter set.

[0015] A second aspect of the present invention provides an automatic laser welding system for aluminum alloy battery casings, which uses the above-mentioned method to automatically laser weld the aluminum alloy battery casings. The system comprises: The 3D geometric data acquisition module is used to obtain the 3D geometric data of the battery shell surface and the coordinates of the contact points. The laser scanner collects high-resolution point cloud information, builds a digital model of the complex shape of the shell, and obtains the precise spatial distribution of the contact area. The material property extraction module is used to extract the material property data of the joint from the digital model, match the thermal conductivity and melting point parameters with the preset material property database, determine the welding difficulty coefficient of the joint area, and determine the initial welding parameter range; The welding head posture correction module is used to detect the relative position deviation between the welding head and the joint using a real-time vision system within the initial welding parameter range. If the relative position deviation value exceeds the relative position deviation threshold, the servo motor is used to adjust the welding head posture to obtain the corrected position coordinates; The distance dynamic adjustment module is used to measure the distance change trend between the joint surface and the welding head using an ultrasonic sensor based on the corrected position coordinates. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state; The energy optimization distribution module is used to extract thermal conductivity information from material property data after obtaining a stable contact state, and calculate the energy input requirement based on the contact geometry. If the energy demand exceeds the current energy parameter upper limit, the laser power adaptive adjustment module allocates energy to obtain the optimized energy distribution; The preheating module is used to drive the laser welding equipment to perform preheating of the joint area through optimized energy distribution, adjust the preheating time and intensity according to the changing trend of material properties, determine the uniformity of the initial weld formation, and determine the temperature field after the preheating is completed; The temperature monitoring module is used to monitor the real-time temperature changes at the joint using an infrared thermal imager based on the temperature field after preheating. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data; The weld quality optimization module is used to extract weld width and density indicators from stable weld penetration data, and compare them with the quality threshold in combination with the joint design parameter indicators. If the design parameter indicators are lower than the quality threshold, the welding path is optimized through the secondary energy compensation algorithm to obtain the final weld quality parameters; The sealing performance detection module is used to obtain the final weld quality parameters, and then use airtightness detection equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

[0016] The technical solution provided by the embodiment of the present invention has the following beneficial effects: The present invention provides an automatic laser welding method and system for aluminum alloy battery shells. First, a laser scanner is used to obtain the three-dimensional geometric data and joint position coordinates of the shell surface, and a digital model is constructed. Then, the material properties of the joints are extracted and the initial welding parameters are determined. The welding head position is corrected using a real-time vision system and an ultrasonic sensor to establish stable contact. The energy distribution is optimized according to the thermal conductivity and geometric shape of the material, and a preheating treatment is performed. During the welding process, temperature changes are monitored by infrared thermal imaging, and the laser parameters are adjusted to maintain a stable penetration depth. Finally, an airtightness test is used to evaluate the weld quality.

[0017] The present invention achieves high-precision, high-quality laser welding of battery shells with complex shapes through multi-sensor fusion and adaptive control, thereby improving welding reliability and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of an automatic laser welding method for aluminum alloy battery shells of the present invention.

[0019] Figure 2 The figure is a schematic structural diagram of an automatic laser welding system for aluminum alloy battery shells according to the present invention. DETAILED DESCRIPTION

[0020] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0021] like Figure 1 As shown, the first aspect of the present invention provides an automatic laser welding method for aluminum alloy battery casing, the method comprising: S1, obtains the three-dimensional geometric data of the battery shell surface and the coordinates of the contact position, collects high-resolution point cloud information through a laser scanner, builds a digital model of the complex shape of the shell, and obtains the accurate spatial distribution of the contact area.

[0022] Optionally, this step also includes: Step S11, using a laser scanner to collect point cloud information on the surface of the battery shell to obtain high-resolution three-dimensional data; Step S12, preprocessing the collected three-dimensional data, using Gaussian filtering to remove noise, and obtaining smooth surface geometric information; Step S13, using curvature analysis to extract the connection position from the smooth surface geometric information and determine the precise coordinates; Step S14, using NURBS surface modeling to construct a digital model based on the precise coordinates to obtain a mathematical expression of the complex shape; Step S15, performing Delaunay triangulation on the mathematical expression to obtain the spatial distribution of the connection area; Step S16, if there are abnormal points in the spatial distribution, using DBSCAN clustering to identify and eliminate them, and judging the integrity of the regional analysis based on a preset distribution threshold; Step S17, obtaining the final three-dimensional geometric data of the battery shell through the spatial distribution after integrity judgment.

[0023] Specifically, by collecting point cloud information on the surface of the battery casing using a laser scanner, high-resolution three-dimensional data can be obtained.

[0024] In one possible implementation, a laser scanner scans the battery casing surface at a rate of 100,000 points per second with an accuracy of 0.01 mm. This high-precision data can fully record the casing's tiny geometric features, such as edges and grooves, laying the foundation for subsequent analysis.

[0025] It should be noted that the advantage of laser scanning is its non-contact measurement, which avoids potential damage to the shell caused by physical contact and improves acquisition efficiency. The collected 3D data is pre-processed using Gaussian filtering to remove noise and obtain smooth surface geometry information.

[0026] Specifically, Gaussian filtering can effectively eliminate random noise introduced by ambient light or device jitter by performing weighted averaging on the neighborhood of each point.

[0027] For example, suppose a point cloud contains outliers due to reflections. Gaussian filtering smooths out the abruptness of these points, making the surface data closer to the true geometry. This step improves data quality and provides reliable input for subsequent precise analysis. Curvature analysis is used to extract the joint locations from the smoothed surface geometry and determine precise coordinates.

[0028] In one embodiment, curvature analysis identifies corners or junctions of the shell surface by calculating the local curvature value at each point.

[0029] For example, the curvature of the welding point on the battery casing is usually significantly higher than that of the smooth area. After analysis, its coordinates can be accurately located, such as x=15.3, y=22.7, z=8.1 mm.

[0030] Preferably, this method can quickly identify key feature points, which helps to improve the accuracy of subsequent modeling. Based on the precise coordinates, NURBS surface modeling is used to build a digital model to obtain a mathematical expression of the complex shape.

[0031] It can be understood that NURBS (Non-Uniform Rational B-Splines) defines smooth surfaces through control points and weights, and is particularly suitable for describing the complex shape of the battery casing.

[0032] For example, a curved shell surface can be smoothly represented mathematically using only 10 control points. This approach maintains geometric accuracy while facilitating digital storage and modification. Delaunay triangulation is then performed on the mathematical representation to determine the spatial distribution of the contact areas.

[0033] Specifically, Delaunay triangulation divides the surface into multiple triangular meshes to ensure clear spatial relationships in the joint areas.

[0034] For example, the side length of the triangular mesh near a connection point is controlled within 0.5 mm to ensure uniform distribution. This step provides structured spatial data for anomaly detection. If outliers exist in the spatial distribution, DBSCAN clustering is used to identify and remove them, and the completeness of the regional analysis is determined based on a preset distribution threshold.

[0035] In one possible implementation, DBSCAN identifies and removes outliers that deviate from the normal distribution by setting a neighborhood radius of 0.2 mm and a minimum number of points of 5.

[0036] For example, noise points generated by scanning reflections in a certain area were successfully removed, ensuring distribution integrity. This approach improves data reliability and prevents anomalies from interfering with the final results. The final 3D geometric data of the battery casing is obtained through the spatial distribution after integrity assessment.

[0037] For example, after the above processing, the error of the 3D model of the shell is controlled within 0.02 mm, accurately reflecting the true shape. The application value of this high-precision geometric data lies in supporting subsequent production verification and quality inspection.

[0038] For example, it can be used to detect whether the housing meets the design tolerance, improving product quality and consistency.

[0039] Optionally, step S16, if there are outliers in the spatial distribution, uses DBSCAN clustering to identify and remove them, and determines the completeness of the regional analysis based on a preset distribution threshold, further comprising: Step S161: Extract data from the spatial distribution and use DBSCAN clustering (set neighborhood radius to 0.5 mm and minimum sample size to 5) to identify a set of outlier coordinates. Step S162: Delete the data points corresponding to the outlier coordinates based on the set of outlier coordinates to generate an updated spatial distribution matrix. Step S163: Calculate the coordinates of the convex hull polygon vertices as the boundary range based on the updated spatial distribution matrix. Step S164: If the convex hull area exceeds the distribution threshold of 1000 square meters, reduce the DBSCAN neighborhood radius to 0.3 mm and regenerate the outlier set. Repeat the elimination process until the convex hull area meets the distribution threshold. Step S165: Use a morphological dilation operation (structuring element radius of 2) to determine the boundary continuity of the final boundary coordinate set and output a connected domain label map. Step S166: Extract the three largest regions from the connected domain label map and use K-means clustering (number of clusters 3) to calculate their centroid coordinates to generate a set of key region polygon vertices. Step S167: Output the spatial distribution result data with region identifiers based on the key region polygon vertex set.

[0040] Specifically, when extracting data from spatial distribution, DBSCAN clustering can be used to identify the coordinate set of outliers.

[0041] For example, in the point cloud data of the battery casing surface, the neighborhood radius is set to 0.5 mm and the minimum number of samples is set to 5. By analyzing the distance between points, isolated points or points that deviate from dense areas are found.

[0042] For example, if a small number of outliers are generated in a certain area due to jitter in the scanning device, and the distance between these points and the surrounding normal data is significantly beyond the set radius, DBSCAN can mark them as anomalies. The advantage of this method is that it can quickly separate outliers while maintaining the integrity of the main distribution.

[0043] In a possible implementation, corresponding data points are deleted according to the outlier point coordinate set to generate an updated spatial distribution matrix.

[0044] Specifically, assuming the original data contains 100,000 points, 200 outliers are identified and removed. The remaining points are then reorganized into a matrix that preserves the primary geometric features of the shell surface. This cleanup process ensures that subsequent analysis is based on cleaner data, free of noise. Based on the updated spatial distribution matrix, the vertex coordinates of the convex hull polygon are calculated as the boundary range.

[0045] It is understandable that the convex hull algorithm connects the outermost points to form a closed boundary.

[0046] For example, after cleaning the point cloud data, a convex hull containing the shell's edges might be generated, with vertex coordinates such as x = 10.2, y = 15.6, and z = 5.8 mm. This boundary definition helps clarify the scope of the analysis area. If the convex hull area exceeds the distribution threshold of 1000 square meters, adjust the DBSCAN parameters and reprocess.

[0047] Preferably, reducing the DBSCAN neighborhood radius from 0.5 mm to 0.3 mm and re-running the clustering may identify more subtle outliers.

[0048] For example, previously unmarked, slightly offset points are removed under the new parameters, reducing the convex hull area to 800 square meters, meeting the requirements. This iterative adjustment gradually optimizes the data range. For the final set of boundary coordinates, a morphological dilation operation is used to determine boundary continuity and output a connected component labeling map.

[0049] In one embodiment, the boundary points are expanded with a structure element radius of 2 mm to fill small-scale fractures.

[0050] For example, a boundary with a 1mm gap due to missing data can be connected after expansion to form a complete region. This method helps confirm the integrity of the boundary. The three regions with the largest areas are extracted from the connected domain label graph, and the centroid coordinates are calculated using K-means clustering to generate the set of polygon vertices in the key region.

[0051] Specifically, suppose the labeled graph contains five connected domains. The first three with areas of 500, 300, and 200 square millimeters are selected and classified into three clusters using K-means. The centroid coordinates might be x = 12.5, y = 18.9, and z = 6.3 millimeters, respectively. This clustering method highlights key characteristic regions. Based on the set of polygon vertices in the key regions, the resulting spatial distribution data is output with region identifiers.

[0052] For example, three areas are labeled A, B, and C, and the vertex coordinates and range of each area are recorded, such as the vertex of area A includes x=11.0, y=17.2, z=5.5 mm, etc. This structured output facilitates subsequent targeted analysis and application.

[0053] S2, extracts the material property data at the joint from the digital model, matches the thermal conductivity and melting point parameters with the preset material property database, determines the welding difficulty coefficient of the joint area, and determines the initial welding parameter range.

[0054] Optionally, this step also includes: In step S21, the temperature field data at the joint is extracted from the digital model using a finite element analysis tool, and the thermal conductivity and melting point parameters are calculated. In step S22, the thermal conductivity and melting point parameters are matched from a preset attribute database to generate a material property distribution map of the joint area. In step S23, based on the material property distribution map, the heat conduction equation is used to calculate the heat flux density difference as a welding difficulty indicator. In step S24, the matching range of the thermal conductivity and melting point parameters is adjusted based on the heat flux density difference to determine the difficulty coefficient. In step S25, the random forest algorithm of scikit-learn is used to input the historical welding parameters and the heat flux density difference to predict the optimized thermal conductivity range. In step S26, the uniformity of the temperature field distribution within the optimized range is verified using the ANSYS simulation tool. If the maximum temperature difference exceeds the temperature difference threshold of 5%, the thermal conductivity range is narrowed, and the initial welding parameter range is finally determined.

[0055] Specifically, the temperature field data at the joints are extracted from the digital model using finite element analysis tools, which is the basis for subsequent analysis.

[0056] For example, in the joint area of ​​the battery housing, temperature field data can be generated by simulating the welding process.

[0057] For example, if a weld joint is subject to localized high temperatures, the finite element tool can capture the temperature distribution pattern as it gradually decreases from the center to the edge, such as reaching 600°C at the center and dropping to 200°C at the edge. This data extraction method fully reflects the actual heat transfer situation. Generating a material property distribution map for the joint region by matching thermal conductivity and melting point parameters from a pre-defined property database requires a rich material information library.

[0058] Specifically, the database may store thermal conductivity and melting point data of materials such as aluminum alloy and copper. For example, the thermal conductivity of aluminum alloy is about 200W / m·K, and the melting point is 660 degrees Celsius.

[0059] In one possible implementation, the system automatically matches the corresponding material properties based on the joint temperature field data and generates a visual distribution map that clearly shows the material differences between different areas. This step facilitates rapid identification of material properties. Based on the material property distribution map, the heat conduction equation is used to calculate the heat flux density difference as an indicator of welding difficulty. This process emphasizes the analysis of heat flow.

[0060] Preferably, the heat flux density can be derived from the temperature gradient in the junction area.

[0061] For example, the heat flux density in the center area may reach 50W / m², while that at the edge is only 10W / m². The difference between the two reflects the uneven heat distribution during welding.

[0062] It's important to note that this metric can intuitively quantify welding challenges and provide a basis for subsequent optimization. Adjusting the matching range of thermal conductivity and melting point parameters based on heat flux differences and determining the difficulty coefficient reflects the dynamic correction of initial parameters.

[0063] In one embodiment, if the heat flux difference is too large, such as exceeding 30 W / m², the thermal conductivity matching range is narrowed from 200-220 W / m·K to 205-215 W / m·K. This adjustment method gradually converges to a more suitable material parameter range, ensuring controllability of the welding process. Using scikit-learn's random forest algorithm, historical welding parameters and heat flux difference values ​​are input to predict the optimized thermal conductivity range, demonstrating data-driven intelligent analysis.

[0064] For example, historical data may contain 100 welding records involving variables such as heat flux density, temperature, and parameters.

[0065] For example, by analyzing this data, a random forest algorithm predicted the optimal thermal conductivity range to be 208-212 W / m·K. This method leverages historical experience to improve prediction accuracy. ANSYS simulation tools were used to verify the uniformity of the temperature field distribution within the optimized range, ensuring the reliability of the results.

[0066] It can be understood that in the simulation, the thermal conductivity was set in the range of 208-212 W / m·K to observe the temperature distribution in the junction area.

[0067] For example, simulation results show that the maximum temperature difference is 4%, which is lower than the 5% temperature difference threshold, so the range is feasible; if the temperature difference reaches 6%, it is further reduced to 210-212W / m·K.

[0068] In one possible implementation, this verification process can effectively screen initial welding parameters, avoiding repeated adjustments later. This repeated verification method improves the scientific nature of parameter selection and provides solid support for production.

[0069] Optionally, step S25 uses the random forest algorithm of scikit-learn to input historical welding parameters and heat flux density differences to predict the optimized thermal conductivity range, and further includes: In step S251, a preset database is used to obtain historical welding parameters and heat flux density distribution, and an input parameter set is obtained through data cleaning. In step S252, the input parameter set is processed through the random forest algorithm of scikit-learn, and the prediction result data is determined in combination with the heat flux density difference. In step S253, the thermal conductivity interval value is calculated according to the prediction result data. If the thermal conductivity interval value exceeds the preset thermal conductivity interval value threshold, the parameter matching range is adjusted. In step S254, the adjusted parameter matching range is obtained, and the welding condition data is updated through the linear regression algorithm of scikit-learn. In step S255, the heat flux density distribution is generated for the welding condition data using the ANSYS tool to determine the distribution uniformity. In step S256, the optimized thermal conductivity is extracted through the ANSYS tool to determine the final thermal conductivity interval value. In step S257, the historical data sample is updated according to the final thermal conductivity interval value to obtain the optimized input parameter set.

[0070] Specifically, when a preset database is used to obtain historical welding parameters and heat flux density distribution, it can be understood as extracting data from a database containing a large number of welding cases.

[0071] For example, the database may store the parameters of the past 100 welding tasks, including power, time and corresponding heat flux values. Data cleaning is a critical first step.

[0072] For example, the input can be optimized by removing outliers.

[0073] For example, if the heat flux density of a welding operation exceeds three standard deviations of the mean, it is considered an anomaly and is eliminated, resulting in an input parameter set containing 50 sets of valid data. When the input parameter set is processed using the random forest algorithm of scikit-learn.

[0074] In one possible implementation, power, time, and heat flux density differences can be used as feature inputs to train the model to predict the thermal conductivity range.

[0075] Specifically, assuming 10 decision trees participate in the voting, each tree determines the thermal conductivity range based on historical data, ultimately outputting a predicted thermal conductivity value, such as 0.15-0.20 W / m·K. The advantage of this method is that it integrates multidimensional data and reduces the impact of single parameter deviations. When calculating the thermal conductivity range value based on the predicted result data, if it exceeds the preset thermal conductivity range threshold, such as the upper limit of 0.25 W / m·K, the parameter matching range needs to be adjusted.

[0076] Preferably, the parameters in the database can be re-matched by narrowing the time range, for example, from 5-10 seconds to 6-8 seconds. This adjustment effectively controls thermal conductivity fluctuations and ensures more stable subsequent welding conditions. After obtaining the adjusted parameter matching range, the welding condition data is updated using the scikit-learn linear regression algorithm.

[0077] For example, a straight line relationship can be fitted using heat flux difference and time as independent variables and power as the dependent variable.

[0078] In one embodiment, assuming the regression results show that the power needs to be reduced by 10W for every 1 second increase in time, this update can provide more accurate input for subsequent simulations. When using ANSYS tools to generate heat flux density distribution based on welding condition data, the heat flux changes in a joint area can be simulated.

[0079] It should be noted that if the simulation results show a heat flux density of 500W / m² at the edge and 450W / m² at the center, the distribution is relatively uniform. This uniformity assessment can help optimize welding quality and avoid local overheating. When using ANSYS tools to extract and optimize thermal conductivity.

[0080] For example, a simulation might show that thermal conductivity stabilizes at 0.18 W / m·K, with the final range determined to be 0.17-0.19 W / m·K.

[0081] In one embodiment, multiple iterations of simulations can be performed to gradually narrow the range, ensuring that the results are closer to actual needs. This precise thermal conductivity range is of great significance to subsequent process design.

[0082] When updating the historical data sample according to the final thermal conductivity interval value, for example, 0.17-0.19 W / m·K and corresponding parameters can be added to the database to replace some old data.

[0083] For example, the original sample with a high thermal conductivity of 0.30 W / m·K was removed, and new data was added to form 50 sets of optimized input parameters. This update improves the applicability of the database and provides a more reliable foundation for future predictions.

[0084] S3, for the initial welding parameter range, uses the real-time vision system to detect the relative position deviation between the welding head and the joint. If the relative position deviation value exceeds the relative position deviation threshold, the servo motor adjusts the welding head posture to obtain the corrected position coordinates.

[0085] Optionally, this step also includes: Step S31, obtain real-time image data between the welding head and the joint through the visual system, and use the image processing algorithm to extract the relative deviation value between the welding head and the joint. Step S32, compare the relative position deviation value with the relative position deviation threshold. If the relative position deviation value exceeds the relative position deviation threshold, determine the adjustment requirement. Step S33, according to the adjustment requirement, drive the welding head through the servo motor to perform posture correction, calculate the correction amount and obtain the corrected posture data. Step S34, use the corrected posture data to update the position coordinates of the welding head to obtain the corrected coordinate value. Step S35, use the visual system to again detect the relative position deviation between the corrected coordinate value and the joint position to determine whether the relative position deviation meets the requirements. Step S36, if the relative position deviation still does not meet the requirements, adjust the servo motor parameters through the gradient descent method to obtain the optimized posture data. Step S37, update the position coordinates according to the optimized posture data to obtain the final correction result.

[0086] Specifically, obtaining real-time image data between the welding head and the joint through the visual system is an important step in ensuring welding accuracy.

[0087] In one possible implementation, a high-resolution industrial camera can be used to capture dynamic images of the welding joint and joints. The image resolution is set to 1920×1080 pixels to ensure detail clarity. Next, image processing algorithms such as edge detection or feature point matching are used to extract relative deviation values.

[0088] For example, if the horizontal deviation between the center of the weld head and the center of the joint is detected to be 2 mm and the vertical deviation is 1.5 mm, these data will serve as the basis for subsequent adjustments. The advantage of this method is that it can reflect the spatial state of the weld head in real time, providing a basis for precise correction.

[0089] Regarding the comparison between the relative position deviation value and the relative position deviation threshold, it should be noted that the relative position deviation threshold is usually pre-set according to welding process requirements.

[0090] In one embodiment, if the process requires a relative position deviation of less than 1 mm, and the actual horizontal deviation is 2 mm, which obviously exceeds the relative position deviation threshold, the system will mark it as "adjustment required".

[0091] Specifically, this can be achieved through simple threshold comparison logic. For example, a deviation range table is set, and an adjustment signal is triggered when the deviation value exceeds the upper limit of the range. The benefit of this approach is that it can quickly determine the corrective action, reduce unnecessary correction operations, and improve efficiency. Driving the servo motor to perform posture correction according to the adjustment requirements is a key step.

[0092] In a possible implementation, after receiving the adjustment signal, the servo motor calculates the rotation angle and the displacement according to the relative position deviation value.

[0093] For example, a horizontal deviation of 2 mm might correspond to a motor rotation of 3 degrees, while a vertical deviation of 1.5 mm might correspond to a displacement of 0.5 mm. Corrected posture data, such as rotation angle and displacement, is recorded. The advantage of this approach is that precise mechanical control enables spatial adjustment of the welding head, ensuring positional accuracy. Using this corrected posture data to update the welding head's position coordinates forms the basis for subsequent optimization.

[0094] For example, the initial coordinates are (100, 50, 20), which may be updated to (98, 48.5, 20) after calibration, reflecting the correction of the relative position deviation.

[0095] Preferably, these coordinate values ​​are transmitted to the control system in real time to guide the next movement of the welding head.

[0096] It is understood that this update process can keep the system tracking the position of the welding joint dynamically. The deviation between the corrected coordinates and the joint position is re-detected by the visual system to verify the correction effect.

[0097] Specifically, if the relative position deviation detected after correction is reduced to 0.8 mm, meeting the 1 mm relative position deviation threshold requirement, the correction is successful. Conversely, if the relative position deviation remains at 1.2 mm, further optimization is required. This secondary inspection ensures that the correction results meet process requirements and avoids degradation of welding quality due to residual deviation. If the relative position deviation still does not meet the requirements, using the gradient descent method to adjust the servo motor parameters is an intelligent approach.

[0098] In one embodiment, assuming an initial motor speed of 50 rpm and slow convergence of the relative position deviation, the speed can be iteratively adjusted to 60 rpm, gradually reducing the relative position deviation to 0.5 mm. This approach offers the advantage of achieving higher accuracy through parameter optimization and adapting to complex welding scenarios. Updating the position coordinates based on the optimized posture data and obtaining the final correction result completes the entire process.

[0099] For example, after adjustment, the coordinates may stabilize at (97.5, 48.8, 20), and the relative position deviation is reduced to 0.3 mm, which fully meets the requirements.

[0100] In one embodiment, these data will be archived for reference in subsequent welding tasks. The beneficial effect of this approach is to provide stable and reliable initial conditions for the welding process, improving overall process consistency.

[0101] S4, based on the corrected position coordinates, uses an ultrasonic sensor to measure the distance change trend between the joint surface and the welding head. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state.

[0102] Optionally, this step also includes: In step S41, the distance change data between the joint surface and the welding head is obtained by an ultrasonic sensor to generate measurement data; in step S42, the distance change trend is calculated using the differential method based on the measurement data; in step S43, if the distance change trend exceeds the distance change threshold, it is determined that there is a distance fluctuation; in step S44, the feed speed adjustment amount is calculated using the PID algorithm for the distance fluctuation; in step S45, the feed speed adjustment amount is calculated based on the current distance error, error integral and error differential to obtain a dynamic adjustment parameter; in step S46, the feed speed of the welding head is adjusted by the dynamic adjustment parameter. The specific method is to add the dynamic adjustment parameter to the current feed speed to obtain a new feed speed and obtain a stable operating state; in step S47, the distance change data under the operating state is obtained, and the standard deviation method is used to determine whether a stable state is reached; in step S48, if the stable state is reached, the contact state between the contact surface and the welding head is determined by using the mean method through the measurement data; in step S49, the sampling frequency of the ultrasonic sensor is adjusted according to the contact state. The specific method is: if the contact state is close contact, the sampling frequency is reduced; if the contact state is loose contact, the sampling frequency is increased to obtain optimized measurement data.

[0103] Specifically, obtaining distance change data between the joint surface and the welding head through an ultrasonic sensor is a key step in ensuring accurate measurement.

[0104] For example, an ultrasonic sensor can be installed near the welding head, transmit ultrasonic waves to the joint surface and receive reflected signals, and calculate the distance based on the propagation time.

[0105] For example, if the initial distance is 10 mm and changes to 12 mm after a period of time, the sensor will record these changes and generate continuous measurement data.

[0106] In one possible implementation, the sensor's sampling frequency is initially set to 100 Hz to capture rapidly changing distance information. This approach has the advantage of being able to monitor dynamic processes in real time, providing a reliable basis for subsequent analysis.

[0107] When using the difference method to calculate the distance change trend based on the measurement data, it should be noted that the difference method determines the change direction by comparing the distance difference between two adjacent measurements.

[0108] For example, if the previous distance was 10 mm and the current distance is 11 mm, the distance change trend is positive 1 mm / time.

[0109] Preferably, a sequence of difference values ​​from multiple measurements, such as 10 mm, 11 mm, and 12 mm, can be used to calculate a distance trend that shows a continuous increase. This approach has the advantage of quickly reflecting the dynamic characteristics of distance, providing an intuitive basis for subsequent judgment. If the distance change trend exceeds the distance change threshold, it is determined that distance fluctuation exists.

[0110] Specifically, assuming that the process requires the distance change threshold to be ±0.5 mm, and the actual trend shows two consecutive changes of 0.8 mm and 0.7 mm respectively, which is obviously out of range, the system will mark it as a fluctuating state.

[0111] In one embodiment, such fluctuation may be caused by uneven joint surface or welding head vibration, and timely identification is helpful for subsequent adjustment. When the PID algorithm is used to calculate the adjustment amount of feed speed for distance fluctuation.

[0112] It can be understood that the PID algorithm combines the three aspects of proportion, integration and differentiation for dynamic adjustment.

[0113] For example, if the current distance error is 2 mm, the historical error accumulation is 5 mm, and the error change rate is 0.5 mm / s, the system will calculate the adjustment amount based on these parameters.

[0114] In one possible implementation, the adjustment amount may be to increase the feed rate by 0.3 mm / s to reduce the distance error. The benefit of this approach is to balance fast response and stability. When adjusting the feed rate of the welding head according to the dynamic adjustment parameters.

[0115] For example, if the current feed rate is 5 mm / s and the calculated adjustment is 0.3 mm / s, the new rate is 5.3 mm / s.

[0116] In one embodiment, the adjusted speed is immediately applied to the welding head motion control to ensure that the distance change tends to be stable. This dynamic adjustment can effectively cope with real-time changes.

[0117] When obtaining the distance change data under the running state and using the standard deviation method to judge the stable state, specifically, 10 distance data can be collected, such as 10.1 mm, 10.2 mm, 10.0 mm, etc. If the calculated standard deviation is less than 0.1 mm, it is considered stable.

[0118] For example, this energy-quantified fluctuation level directly reflects the operating status.

[0119] When the contact state is determined by using the mean method through measurement data, preferably, if the average distance of 10 measurements is 10.05 mm and the variation range is extremely small, it can be determined as close contact; if the mean is 10.5 mm and the fluctuation is large, it is loose contact.

[0120] In one embodiment, this judgment provides a basis for subsequent optimization.

[0121] The ultrasonic sensor's sampling frequency is adjusted based on the contact state. For example, during close contact, the frequency can be reduced from 100 Hz to 50 Hz to conserve resources, while during loose contact, the frequency can be increased to 150 Hz to enhance monitoring. This adaptive adjustment improves the efficiency and pertinence of data collection.

[0122] S5, after obtaining a stable contact state, extracts thermal conductivity information from the material property data and calculates the energy input requirement based on the contact geometry. If the energy requirement exceeds the upper limit of the current energy parameter, the energy is distributed through the laser power adaptive adjustment module to obtain the optimized energy distribution.

[0123] Optionally, this step also includes: Step S51, obtain sensor data to determine the contact state, extract the contact geometry information through the geometric modeling tool according to the contact state, and obtain the contact shape parameters; Step S52, retrieve material properties from the material database, combine the contact shape parameters and thermal conductivity information, use ANSYS to perform finite element analysis to calculate the energy input demand, and obtain a preliminary energy value; Step S53, if the preliminary energy value exceeds the upper limit of the energy parameter, determine whether to trigger the adjustment through the energy value threshold, and obtain an adjustment signal; Step S54, activate the laser power adaptive adjustment module according to the adjustment signal, use linear interpolation to distribute energy, and obtain an adjusted power distribution; Step S55, update the energy input through the adjusted power distribution, recalculate the energy distribution based on the contact geometry, and obtain the optimized energy distribution; Step S56, obtain the optimized energy distribution, use the MonteCarlo library in MATLAB to verify the distribution uniformity, and obtain the final energy solution; Step S57, generate control instructions through the final energy solution, output them to the laser equipment to execute energy adjustment, and complete the business process.

[0124] Specifically, when obtaining sensor data to determine the contact state, it can be understood that the ultrasonic sensor measures the distance by transmitting and receiving reflected signals, and then determines the contact state between the contact point and the welding head.

[0125] For example, if the sensor detects that the distance between the joint and the welding head is continuously around 10 mm with minimal changes, it can be considered that the contact is tight; if the distance fluctuates between 10 and 12 mm, it may be loose contact.

[0126] When extracting the contact geometry information using a geometric modeling tool based on the contact state, in one possible implementation, CAD software may be used to perform three-dimensional reconstruction of the sensor data.

[0127] For example, in tight contact, the reconstructed contact surface is a smooth circle with a diameter of approximately 5 mm, while in loose contact, it may appear as an irregular ellipse. This method can intuitively reflect the contact shape parameters, such as diameter and curvature, providing a basis for subsequent analysis.

[0128] When retrieving material properties from the material database, specifically, based on the contact material, such as aluminum alloy, it can be found that its thermal conductivity is approximately 237 W / m·K and its density is 2700 kg / m³.

[0129] When finite element analysis is performed using ANSYS in combination with the contact shape parameters and thermal conductivity information, in one embodiment, the contact is modeled as a cylinder with a diameter of 5 mm, and an initial heat flow is applied to simulate the energy input process.

[0130] For example, a preliminary energy value might be calculated as 500 joules. This approach effectively maps changes in energy requirements.

[0131] Preferably, if the preliminary energy value exceeds the upper limit of the energy parameter, for example, the process upper limit is 450 joules, and the calculated value is 500 joules, a preset energy value threshold such as ±10% can be used to determine whether to trigger adjustment.

[0132] It should be noted that if the energy value exceeds the threshold range, an adjustment signal is generated to prompt the system to optimize the energy input.

[0133] When the laser power adaptive adjustment module is activated according to the adjustment signal, in a possible implementation, a linear interpolation method is used to redistribute energy.

[0134] For example, if the initial power is 1000 watts, after adjustment, 800 watts may be allocated in the first half and 1200 watts in the second half. This allocation can balance the uniformity of energy distribution.

[0135] When updating the energy input with the adjusted power distribution, specifically, the new power values ​​are substituted into the model and recalculated in conjunction with the joint geometry.

[0136] For example, a 5 mm diameter joint might require the total energy to be reduced to 420 joules. This iteration optimizes the energy distribution.

[0137] For example, after obtaining the optimized energy distribution, the Monte Carlo library in MATLAB is used to verify uniformity. The energy distribution is simulated 1,000 times. If more than 90% of the energy distribution falls within the target range, such as 400-430 joules, the distribution is considered uniform. This verification can improve the reliability of the solution.

[0138] For example, when generating control instructions through the final energy solution, the optimized 420 joules of energy are converted into laser device parameters, such as power of 900 watts and duration of 0.47 seconds.

[0139] In one embodiment, these instructions are directly output to the equipment for execution, ensuring precise energy regulation. This process can effectively improve the adaptability of the process.

[0140] S6, through the optimized energy distribution, drives the laser welding equipment to perform preheating treatment of the joint area, adjusts the preheating time and intensity according to the changing trend of material properties, judges the uniformity of the initial weld formation, and determines the temperature field after preheating is completed.

[0141] Optionally, the steps further include: Step S61, obtain the energy distribution data of the joint area through the infrared sensor of the laser welding equipment, use the bicubic interpolation algorithm to generate a two-dimensional thermal field map, and determine the preheating intensity benchmark according to the peak position of the thermal field map. Step S62, extract the reflectivity curve of the material at different temperatures from the thermal field map, and obtain the trend of the change of the thermal conductivity of the material by least squares fitting. Step S63, when the change in the thermal conductivity of the material exceeds the threshold value of the change in the thermal conductivity of 0.15W / mK, adjust the preheating time in a linear proportion, and each 0.1W / mK change corresponds to a 5ms time increment; Step S64, write the adjusted preheating parameters into the welding robot control instructions, including a power range of 800-1500W and a scanning speed of 0.2-0.5m / s; Step S65, after executing the preheating plan, use a 500fps high-speed camera to capture the initial image of the weld, use OpenCV to perform median filtering and histogram equalization processing, and generate 8-bit grayscale image; Step S66, invert the temperature field based on the grayscale image pixel brightness value, and calculate the lateral temperature standard deviation as a uniformity indicator; Step S67, if the temperature standard deviation exceeds the temperature standard deviation threshold of 30°C, use the pre-trained ResNet18 network to process the temperature gradient map, and output the power adjustment coefficient and scanning speed compensation value. Step S68, write the power adjustment coefficient ΔP and scanning speed compensation value ΔV output by the network into the PLC controller, update the welding parameters and re-collect the temperature field data until the temperature standard deviation drops below the temperature standard deviation threshold.

[0142] Specifically, when obtaining energy distribution data of the joint area through the infrared sensor of the laser welding equipment, it can be understood that the infrared sensor generates high-resolution temperature distribution information by capturing the thermal radiation signal of the joint surface.

[0143] For example, on an aluminum alloy joint, the sensor may record a 5 mm x 5 mm area where the temperature gradually decays from 200°C in the center to 150°C towards the edge.

[0144] In one possible implementation, when a bicubic interpolation algorithm is used to generate a two-dimensional thermal field map, discrete sensor data points are converted into a continuous surface.

[0145] For example, if the center peak temperature is 200°C and the edge is 150°C, interpolation can generate a smooth thermal field map with the peak position clearly pointing to the center of the joint. This thermal field map can intuitively reflect the energy concentration area and provide a basis for subsequent analysis.

[0146] Specifically, when extracting the material reflectivity curve from the thermal field map, the reflectivity data of the aluminum alloy between 100°C and 300°C can be recorded according to the temperature change.

[0147] For example, the reflectivity is 0.8 at 150°C, while it drops to 0.65 at 250°C.

[0148] In one embodiment, using the least squares method to fit the thermal conductivity of a material can yield a smooth curve showing that thermal conductivity decreases with increasing temperature. This trend can help predict the energy transfer behavior of a material at high temperatures.

[0149] It should be noted that when the change in thermal conductivity exceeds the thermal conductivity change threshold of 0.15 W / mK, this indicates that the thermal properties of the material have changed significantly.

[0150] For example, the initial thermal conductivity of aluminum alloy is 237W / mK, which may drop to 236.8W / mK at high temperature, a change of 0.2W / mK.

[0151] Preferably, when the preheating time is adjusted in a linear ratio, each 0.1 W / mK corresponds to an increase of 5 ms.

[0152] For example, for a change of 0.2W / mK, the warm-up time increases from 50ms to 60ms. This adjustment ensures sufficient heat penetration.

[0153] When preheating parameters are written into the control instructions, for example, the power is set to 1200W and the scanning speed is set to 0.3m / s, which is suitable for 5mm thick joints. This parameter can balance efficiency and quality.

[0154] In one possible implementation, after preheating, when a high-speed camera captures weld images, a frame rate of 500 fps can capture the moment when the molten pool is formed.

[0155] For example, an image shows a weld seam approximately 2 mm wide. When processed using OpenCV, median filtering removes noise, and histogram equalization enhances contrast, producing a clear 8-bit grayscale image. This processing emphasizes the weld seam details.

[0156] Specifically, when inverting the temperature field based on the grayscale image, the pixel brightness value can be mapped to temperature.

[0157] For example, a brightness of 200 corresponds to 250°C, and a brightness of 150 corresponds to 200°C. When calculating the horizontal temperature standard deviation, if the result is 40°C, it indicates that the distribution is not uniform.

[0158] For example, if the temperature standard deviation exceeds the temperature standard deviation threshold of 30°C, when the ResNet18 network processes the temperature gradient map, it inputs a gradient map of a 5mm weld seam, outputs a power adjustment factor of 1.1, and a speed compensation value of 0.05m / s. This intelligent adjustment optimizes heat input.

[0159] For example, when writing network output to a PLC controller, the power is adjusted from 1200W to 1320W and the speed is increased from 0.3m / s to 0.35m / s. After re-collecting the temperature field, the temperature standard deviation may drop to 25°C. This iteration effectively improves uniformity and ensures welding quality.

[0160] S7, based on the temperature field after preheating, uses an infrared thermal imager to monitor the real-time temperature changes at the joint. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data.

[0161] Optionally, this step also includes: Step S71, collect the temperature field change data at the joint through the infrared thermal imager to obtain the real-time temperature distribution; Step S72, if the real-time temperature exceeds the preset temperature threshold, it is judged as a temperature abnormality and an abnormal signal is generated; Step S73, activate the control system according to the abnormal signal, calculate the adjustment value of the laser pulse frequency; Step S74, perform the welding operation using the adjusted laser pulse frequency to obtain the weld penetration data; Step S75, compare the weld penetration data with the preset standard through the monitoring system to determine the data stability; Step S76, if the data stability does not meet the standard, optimize the frequency adjustment parameter using the support vector machine algorithm. number, and obtain a stable output; step S77, update the control system according to the optimized parameters, adjust the laser pulse frequency, and obtain optimized weld penetration data; step S78, in the control system, use a PID controller to adjust the laser pulse frequency, and the parameters of the PID controller are dynamically adjusted according to the real-time temperature distribution and weld penetration data; step S79, use the support vector machine algorithm to optimize the parameters of the PID controller, specifically using the radial basis function kernel, and determine the optimal parameters through cross-validation; step S710, the optimized parameters are used to update the control system, adjust the laser pulse frequency, and obtain the final weld penetration data.

[0162] Specifically, when collecting temperature field change data at the joint using an infrared thermal imager.

[0163] It can be understood that the infrared thermal imager generates a two-dimensional temperature distribution map in real time by sensing the infrared radiation emitted from the surface of the joint.

[0164] For example, a thermal imager might show a 10 mm x 10 mm area on a stainless steel joint with a center temperature of 280°C and an edge temperature of 220°C. This real-time data can provide a basis for subsequent assessments. If the real-time temperature exceeds a preset threshold, it is identified as abnormal and a signal is generated.

[0165] Specifically, assuming the temperature threshold is set to 260°C, when the core temperature reaches 280°C, the system will trigger an abnormal signal.

[0166] In one possible implementation, a signal is sent to the control unit in digital form, indicating that the temperature exceeds a certain limit.

[0167] It should be noted that when the control system is activated according to the abnormal signal and the laser pulse frequency adjustment value is calculated, the control system determines the frequency change according to the degree of temperature deviation.

[0168] For example, if the temperature exceeds 20°C, the pulse frequency may be reduced from 1000Hz to 950Hz. This adjustment is intended to control the amount of heat input.

[0169] In one embodiment, when performing a welding operation and acquiring weld penetration data using the adjusted laser pulse frequency, after the frequency is adjusted to 950 Hz, a 3 mm thick stainless steel plate is welded, and the penetration data may be displayed as 1.8 mm.

[0170] Preferably, an ultrasonic detector is used to measure the penetration depth to ensure the accuracy of the data.

[0171] For example, when the monitoring system compares the weld penetration data with the preset standard, if the preset standard penetration is 2 mm and the actual measurement is 1.8 mm, the weld penetration deviation is 0.2 mm.

[0172] For example, the system will record multiple measurements, such as 1.8 mm, 1.9 mm, and 1.7 mm, to determine their stability. If the data stability does not meet the standard, the support vector machine algorithm is used to optimize the frequency adjustment parameters.

[0173] In one embodiment, after inputting multiple penetration depth data and temperature profiles, the algorithm may output that the frequency needs to be further adjusted to 930 Hz. This optimization can improve penetration consistency.

[0174] For example, when the control system is updated and the laser pulse frequency is adjusted according to the optimized parameters, after the frequency is adjusted from 950 Hz to 930 Hz, welding is performed again and the penetration depth data may stabilize at 1.95 mm.

[0175] It can be understood that this iterative adjustment can gradually approach the target value.

[0176] Specifically, when a PID controller is used in the control system to adjust the laser pulse frequency, the PID is dynamically adjusted according to the real-time temperature and the melting depth deviation.

[0177] For example, if the temperature exceeds the limit by 10°C and the penetration depth is 0.1mm less, the proportional term may increase the frequency by 5Hz, while the integral term eliminates the accumulated error. This method can quickly respond to changes.

[0178] In one possible implementation, the support vector machine algorithm uses a radial basis function kernel to optimize the PID parameters. Through cross-validation of multiple experimental data, it is determined that the proportional coefficient is increased to 1.2 and the integration time is reduced to 0.5 seconds.

[0179] Preferably, this optimization allows for more precise control.

[0180] For example, when the optimized parameters update the control system and adjust the laser pulse frequency, the frequency is stabilized at 935 Hz, the penetration depth data after welding reaches 2 mm, and the deviation of multiple measurements is less than 0.05 mm.

[0181] For example, stable penetration data indicates improved process reliability.

[0182] S8, extracts weld width and density indicators from stable weld penetration data, combines them with joint design parameter indicators to compare with the quality threshold, and if the design parameter indicators are lower than the quality threshold, optimizes the welding path through the secondary energy compensation algorithm to obtain the final weld quality parameters.

[0183] Optionally, this step also includes: Step S81, obtain weld penetration data from the welding sensor, and use OpenCV's Canny edge detection algorithm to extract the weld contour; Step S82, calculate the maximum width value according to the contour coordinates, and calculate the density index by combining the penetration data with the preset weld cross-sectional area formula; Step S83, input the density index and the allowable stress value in the joint design parameters into the polynomial regression model, and output the welding quality characteristic value; Step S84, when the characteristic value is lower than the quality threshold, use the least squares method to fit the linear relationship between penetration and energy, calculate the energy compensation amount and generate the path coordinate correction coefficient; Step S85, adjust the path planning coordinates of the welding robot according to the correction coefficient, re-collect the weld data and calculate the new density index; Step S86, if the new index still does not meet the standard, use the penetration fluctuation value, compensation amount, and path offset as input features, and use the random forest algorithm to output the final path coordinate correction parameter.

[0184] Specifically, after obtaining the weld penetration data from the welding sensor, the Canny edge detection algorithm of OpenCV is used to extract the weld contour.

[0185] It can be understood that the Canny algorithm generates an accurate outline of the weld by identifying significant boundaries of grayscale changes in the image.

[0186] For example, in a weld cross-section image, the algorithm may detect a weld outline with a width of 8 mm, with clear edges, which is convenient for subsequent analysis.

[0187] For example, for a stainless steel plate, the sensor indicates a 2mm penetration depth. The Canny algorithm then further delineates the boundary curves at the top and bottom of the weld, providing basic data for width and cross-section calculations. This method effectively separates the weld from the surrounding area.

[0188] It should be noted that when calculating the maximum width value based on the contour coordinates, the width is usually taken as the horizontal distance of the widest point in the contour.

[0189] In one possible implementation, when calculating the density index using a pre-set cross-sectional area formula based on penetration data, assuming a penetration depth of 2 mm and a width of 8 mm, the formula might approximate the cross-sectional area based on a triangle, yielding a value such as 8 square millimeters. The density index is then determined by the ratio of the actual cross-sectional area to the ideal value.

[0190] For example, if the ideal cross-sectional area is 10 square millimeters, the density is 0.8. This index reflects the degree of filling of the weld.

[0191] For example, when the density index and the allowable stress value in the joint design parameters are input into the polynomial regression model, the allowable stress may be set to 300 MPa and the density to 0.8. The model is trained based on historical data and outputs a characteristic value of welding quality, such as 85 points out of 100 points.

[0192] For example, if the characteristic value is lower than the quality threshold of 90 points, the welding quality is considered to be insufficient.

[0193] In one embodiment, the least squares method is used to fit the linear relationship between penetration and energy. Assuming that 2 mm penetration corresponds to 100 joules and 2.2 mm penetration corresponds to 110 joules, the fitting calculates the required energy compensation of 5 joules and generates a path correction factor, such as 1.05. This method quantifies the energy adjustment requirement. After adjusting the path planning coordinates of the welding robot based on the correction factor, the weld data is recollected.

[0194] For example, after adjustment, the penetration depth may increase to 2.1 mm and the density index becomes 0.85.

[0195] Preferably, this iterative verification can gradually improve the density index.

[0196] In one embodiment, if the new density index still does not meet the standard, a penetration fluctuation value (e.g., 0.1 mm), a compensation value of 5 joules, and a path offset of 0.5 mm are input into the random forest algorithm. The algorithm analyzes multiple decision trees and outputs the final path correction parameters, such as adjusting the offset to 0.6 mm.

[0197] For example, this approach can integrate multi-dimensional features to ensure more reasonable path optimization.

[0198] It is understandable that random forest improves prediction accuracy through ensemble learning.

[0199] For example, over multiple welding experiments, the algorithm may discover that offset is more strongly correlated with penetration fluctuations and prioritize adjusting that parameter.

[0200] It should be noted that this method is suitable for situations with a large number of data samples and can extract patterns from complex relationships.

[0201] In one embodiment, the adjusted path coordinates stabilized the penetration depth at 2.2 mm and the density index reached above 0.9. This optimization can significantly improve the consistency of the weld and provide a reliable basis for subsequent quality control.

[0202] S9. After obtaining the final weld quality parameters, use airtightness testing equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

[0203] Optionally, this step also includes: Step S91, collecting weld quality parameter data through sensors and storing them as an initial data set; Step S92, using airtightness detection equipment to perform pressure testing on the shell contacts to obtain sealing performance data; Step S93, if the sealing performance data exceeds the sealing performance threshold, removing outliers through the box plot method to obtain a corrected performance data set; Step S94, based on the corrected performance data set, using the standard deviation method to calculate the leakage prevention effect index and determine the leakage prevention status; Step S95, for the leakage prevention status, using the random forest algorithm to analyze the correlation between the welding process and the leakage prevention effect, and obtain the influencing factor ranking; Step S96, after obtaining the influencing factor ranking, predicting the welding process reliability through the logistic regression algorithm, and judging the reliability level; Step S97, extracting key features from the reliability level, using K-means clustering to divide the welding process optimization direction, and determining the improved parameter set.

[0204] Specifically, the weld quality parameter data is collected by sensors and stored as an initial data set. This process is the basis of welding quality analysis.

[0205] In one possible implementation, a high-precision laser sensor can be used to monitor the weld area in real time. The collected data, including parameters such as penetration depth, weld width, and weld surface smoothness, is stored in a time series format for subsequent analysis. Pressure testing of the shell joints using airtightness testing equipment is used to obtain sealing performance data, which is often used to verify the actual functionality of the weld.

[0206] Specifically, the housing is placed in a sealed environment, a constant pressure of 0.5 MPa is applied for 10 minutes, and the pressure decay is recorded. If the decay value is less than 0.01 MPa, the seal is initially considered good. This method directly reflects the leak-proof ability of the weld. If the sealing performance data exceeds the preset sealing performance threshold, outliers are removed using the boxplot method to obtain a corrected data set.

[0207] It is understandable that the box plot method can effectively eliminate abnormal points caused by equipment jitter or environmental interference by identifying the upper and lower quartiles of the data.

[0208] For example, assuming the pressure decay value set is 0.005, 0.006, 0.007, and 0.05 MPa, 0.05 is identified as an outlier and removed. The retained data is more representative, which helps improve the accuracy of subsequent analysis. Based on the corrected performance data set, the standard deviation method is used to calculate the leakage prevention effect index to determine the leakage prevention status.

[0209] In one embodiment, the standard deviation of the attenuation values ​​from multiple pressure tests is calculated. If the result is less than 0.002, the leak-proof state is considered stable. This method quantifies the degree of data dispersion and reflects the consistency of the weld. A random forest algorithm is used to analyze the correlation between the welding process and leak-proof performance and rank the influencing factors.

[0210] For example, the input features include welding speed, energy input, and ambient humidity. The output shows that energy input accounts for 45% of the leak prevention effect. This result can guide the priority of process adjustments. After obtaining the ranking of influencing factors, a logistic regression algorithm is used to predict the reliability of the welding process and determine the reliability level.

[0211] Preferably, the impact factors can be divided into three categories: high, medium, and low, and the model can be trained with historical data to predict the reliability probability.

[0212] For example, if the predicted reliability probability is greater than 0.9, it is considered highly reliable. This prediction helps identify potential risks in advance. Key features are extracted from the reliability levels, and K-means clustering is used to identify optimization directions for the welding process and determine the set of improved parameters.

[0213] In one possible implementation, features such as energy input and path deviation are input into the clustering algorithm and divided into three optimization directions: energy saving, stability enhancement, and efficiency priority.

[0214] For example, an energy-saving parameter set might suggest reducing energy input by 10% and fine-tuning the process path. This categorization provides flexible solutions for different needs and improves process adaptability.

[0215] like Figure 2 As shown, the second aspect of the present invention provides an automatic laser welding system for aluminum alloy battery shells, which uses the above-mentioned method to automatically laser weld aluminum alloy battery shells, and the system includes: The 3D geometric data acquisition module is used to obtain the 3D geometric data of the battery shell surface and the coordinates of the contact points. The laser scanner collects high-resolution point cloud information, builds a digital model of the complex shape of the shell, and obtains the precise spatial distribution of the contact area. The material property extraction module is used to extract the material property data of the joint from the digital model, match the thermal conductivity and melting point parameters with the preset material property database, determine the welding difficulty coefficient of the joint area, and determine the initial welding parameter range; The welding head posture correction module is used to detect the relative position deviation between the welding head and the joint using a real-time vision system within the initial welding parameter range. If the relative position deviation exceeds the relative position deviation threshold, the servo motor is used to adjust the welding head posture to obtain the corrected position coordinates. The distance dynamic adjustment module is used to measure the distance change trend between the joint surface and the welding head using an ultrasonic sensor based on the corrected position coordinates. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state; The energy optimization distribution module is used to extract thermal conductivity information from material property data after obtaining a stable contact state, and calculate the energy input requirement based on the contact geometry. If the energy demand exceeds the current parameter upper limit, the laser power adaptive adjustment module allocates energy to obtain the optimized energy distribution; The preheating module is used to drive the laser welding equipment to perform preheating of the joint area through optimized energy distribution, adjust the preheating time and intensity according to the changing trend of material properties, determine the uniformity of the initial weld formation, and determine the temperature field after the preheating is completed; The temperature monitoring module is used to monitor the real-time temperature changes at the joint using an infrared thermal imager based on the temperature field after preheating. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data; The weld quality optimization module is used to extract weld width and density indicators from stable weld penetration data, and compare them with the quality threshold in combination with the joint design parameter indicators. If the design parameter indicators are lower than the quality threshold, the welding path is optimized through the secondary energy compensation algorithm to obtain the final weld quality parameters; The sealing performance detection module is used to obtain the final weld quality parameters, and then use airtightness detection equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

[0216] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An automatic laser welding method for aluminum alloy battery shell, characterized in that: The method comprises: S1, obtains the three-dimensional geometric data of the battery shell surface and the coordinates of the contact positions, collects high-resolution point cloud information through a laser scanner, builds a digital model of the complex shape of the shell, and obtains the accurate spatial distribution of the contact area; S2, extracting material property data at the joint from the digital model, matching thermal conductivity and melting point parameters with a preset material property database, determining the welding difficulty coefficient of the joint area, and determining the initial welding parameter range; S3, within the initial welding parameter range, the real-time vision system is used to detect the relative position deviation between the welding head and the joint. If the relative position deviation value exceeds the relative position deviation threshold, the servo motor is used to adjust the welding head posture to obtain the corrected position coordinates; S4, based on the corrected position coordinates, an ultrasonic sensor is used to measure the distance change trend between the joint surface and the welding head. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state; S5, after obtaining a stable contact state, extract thermal conductivity information from the material property data and calculate the energy input requirement based on the contact geometry. If the energy requirement exceeds the upper limit of the current energy parameter, the laser power adaptive adjustment module allocates energy to obtain the optimized energy distribution; S6, using the optimized energy distribution, drives the laser welding equipment to perform preheating of the joint area, adjusts the preheating time and intensity according to the changing trend of material properties, determines the uniformity of the initial weld formation, and determines the temperature field after the preheating is completed; S7, based on the temperature field after preheating, uses an infrared thermal imager to monitor the real-time temperature changes at the joint. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data; S8, extracting weld width and density indicators from the stable weld penetration data, combining them with the joint design parameter indicators to compare with the quality threshold. If the design parameter indicators are lower than the quality threshold, the welding path is optimized through the secondary energy compensation algorithm to obtain the final weld quality parameters; S9. After obtaining the final weld quality parameters, use airtightness testing equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

2. The method according to claim 1, characterized in that Step S1, obtaining the three-dimensional geometric data of the battery shell surface and the coordinates of the contact positions, collecting high-resolution point cloud information through a laser scanner, building a digital model for the complex shape of the shell, and obtaining the accurate spatial distribution of the contact area, includes: Step S11, collecting point cloud information of the battery housing surface by a laser scanner to obtain high-resolution three-dimensional data; Step S12: pre-processing the collected three-dimensional data, using Gaussian filtering to remove noise and obtain smooth surface geometric information; Step S13, extracting the joint positions from the smooth surface geometry information using curvature analysis to determine the precise coordinates; Step S14, constructing a digital model using NURBS surface modeling based on the precise coordinates to obtain a mathematical expression of the complex shape; Step S15, performing Delaunay triangulation on the mathematical expression to obtain the spatial distribution of the contact area; Step S16: If there are outliers in the spatial distribution, they are identified and removed using DBSCAN clustering, and the completeness of the regional analysis is determined based on a preset distribution threshold; Step S17: Obtain the final three-dimensional geometric data of the battery housing through the spatial distribution after integrity judgment.

3. The method according to claim 2, characterized in that The step S15, performing Delaunay triangulation on the mathematical expression to obtain the spatial distribution of the contact area, includes: a triangular mesh side length near the contact is not greater than 0.5 mm.

4. The method according to claim 2, characterized in that In step S16, if there are outliers in the spatial distribution, DBSCAN clustering is used to identify and remove them, and the integrity of the regional analysis is judged based on a preset distribution threshold, including: Step S161, extracting data from the spatial distribution and using DBSCAN clustering to identify an outlier coordinate set; Step S162: Delete the data points at the corresponding spatial positions according to the abnormal point coordinate set, and generate an updated spatial distribution matrix; Step S163, based on the updated spatial distribution matrix, calculating the coordinates of the convex hull polygon vertices as the boundary range; Step S164: If the convex hull area exceeds the distribution threshold, the DBSCAN neighborhood radius is reduced to regenerate the outlier set, and the elimination process is repeated until the convex hull area meets the distribution threshold; Step S165: using a morphological dilation operation to determine the boundary continuity of the final boundary coordinate set, and outputting a connected component label map; Step S166, extracting the three regions with the largest areas from the connected component label graph, calculating their centroid coordinates using K-means clustering, and generating a set of key region polygon vertices; Step S167: outputting spatial distribution result data with area identification according to the key area polygon vertex set.

5. The method according to claim 4, characterized in that The step S161 extracts data from the spatial distribution and uses DBSCAN clustering to identify an outlier coordinate set, including setting a neighborhood radius of 0.5 mm and a minimum number of samples of 5.

6. The method according to claim 5, characterized in that Step S164 , if the convex hull area exceeds the distribution threshold, reducing the DBSCAN neighborhood radius to regenerate the outlier set, and repeating the elimination process until the convex hull area meets the distribution threshold, including: reducing the DBSCAN neighborhood radius to 0.3 mm.

7. The method according to claim 1, characterized in that The step S2 extracts the material property data at the joint from the digital model, matches the thermal conductivity and melting point parameters in combination with a preset material property database, determines the welding difficulty coefficient of the joint area, and determines the initial welding parameter range, including: Step S21, extracting temperature field data at the junction from the digital model using a finite element analysis tool, and calculating thermal conductivity and melting point parameters; Step S22, matching thermal conductivity and melting point parameters from a preset property database to generate a material property distribution map of the contact area; Step S23, based on the material property distribution diagram, using the heat conduction equation to calculate the heat flux density difference value as a welding difficulty indicator; Step S24, adjusting the matching range of thermal conductivity and melting point parameters based on the heat flux density difference value to determine the difficulty coefficient; Step S25, using the random forest algorithm of scikit-learn, inputting historical welding parameters and heat flux density difference values, and predicting the optimized thermal conductivity range; Step S26: Verify the uniformity of the temperature field distribution within the optimization range through the ANSYS simulation tool. If the maximum temperature difference exceeds the temperature difference threshold, narrow the thermal conductivity range and finally determine the initial welding parameter range.

8. The method according to claim 7, characterized in that Step S25 uses the random forest algorithm of scikit-learn to input historical welding parameters and heat flux density difference values ​​to predict the optimized thermal conductivity range, including: Step S251, using a preset database to obtain historical welding parameters and heat flux density distribution, and obtaining an input parameter set through data cleaning; Step S252 , processing the input parameter set using the random forest algorithm of scikit-learn, and combining the heat flux density difference to determine the prediction result data; Step S253, calculating the thermal conductivity interval value according to the prediction result data, and adjusting the parameter matching range if the thermal conductivity interval value exceeds a preset thermal conductivity interval value threshold; Step S254, obtaining the adjusted parameter matching range, and updating the welding condition data using the linear regression algorithm of scikit-learn; Step S255: Generate heat flux density distribution based on welding condition data using ANSYS tools and determine the distribution uniformity; Step S256, extracting and optimizing thermal conductivity through ANSYS tools to determine the final thermal conductivity interval value; Step S257: Update the historical data sample according to the final thermal conductivity interval value to obtain an optimized input parameter set.

9. The method according to claim 1, characterized in that The step S9, after obtaining the final weld quality parameters, uses airtightness testing equipment to measure the sealing performance data at the shell joints, judges the leak-proof effect based on the test results, and determines the reliability of the welding process, including: Step S91, collecting weld quality parameter data through sensors and storing them as an initial data set; Step S92: Using airtightness testing equipment to perform pressure testing on the shell contacts to obtain sealing performance data; Step S93: If the sealing performance data exceeds the sealing performance threshold, outliers are removed using a box plot method to obtain a corrected performance data set; Step S94, calculating the leakage prevention effect index using the standard deviation method based on the corrected performance data set to determine the leakage prevention status; Step S95: Analyze the correlation between the welding process and the leakage prevention effect using a random forest algorithm to obtain a ranking of influencing factors. Step S96, after obtaining the ranking of the influencing factors, predict the reliability of the welding process using a logistic regression algorithm to determine the reliability level; Step S97: extract key features from the reliability levels, use K-means clustering to divide the welding process optimization direction, and determine the improved parameter set.

10. An automatic laser welding system for aluminum alloy battery shells, characterized in that: The aluminum alloy battery case is automatically laser welded using the method according to any one of claims 1 to 9, the system comprising: The 3D geometric data acquisition module is used to obtain the 3D geometric data of the battery shell surface and the coordinates of the contact points. The laser scanner collects high-resolution point cloud information, builds a digital model of the complex shape of the shell, and obtains the precise spatial distribution of the contact area. The material property extraction module is used to extract the material property data of the joint from the digital model, match the thermal conductivity and melting point parameters with the preset material property database, determine the welding difficulty coefficient of the joint area, and determine the initial welding parameter range; The welding head posture correction module is used to detect the relative position deviation between the welding head and the joint using a real-time vision system within the initial welding parameter range. If the relative position deviation value exceeds the relative position deviation threshold, the servo motor is used to adjust the welding head posture to obtain the corrected position coordinates; The distance dynamic adjustment module is used to measure the distance change trend between the joint surface and the welding head using an ultrasonic sensor based on the corrected position coordinates. When the distance fluctuates, the feed speed of the welding head is dynamically adjusted through the PID algorithm to determine the stable contact state; The energy optimization distribution module is used to extract thermal conductivity information from material property data after obtaining a stable contact state, and calculate the energy input requirement based on the contact geometry. If the energy demand exceeds the current energy parameter upper limit, the laser power adaptive adjustment module allocates energy to obtain the optimized energy distribution; The preheating module is used to drive the laser welding equipment to perform preheating of the joint area through optimized energy distribution, adjust the preheating time and intensity according to the changing trend of material properties, determine the uniformity of the initial weld formation, and determine the temperature field after the preheating is completed; The temperature monitoring module is used to monitor the real-time temperature changes at the joint using an infrared thermal imager based on the temperature field after preheating. When the temperature is abnormal, the laser pulse frequency is adjusted through the feedback control system to obtain stable weld penetration data; The weld quality optimization module is used to extract weld width and density indicators from stable weld penetration data, and compare them with the quality threshold in combination with the joint design parameter indicators. If the design parameter indicators are lower than the quality threshold, the welding path is optimized through the secondary energy compensation algorithm to obtain the final weld quality parameters; The sealing performance detection module is used to obtain the final weld quality parameters, and then use airtightness detection equipment to measure the sealing performance data at the shell joints, judge the leak-proof effect based on the test results, and determine the reliability of the welding process.

Images