Intelligent tracking method and system for free curve welding seam by mechanical arm

Abstract

The invention relates to the technical field of intelligent tracking of welding seams, in particular to an intelligent tracking method and system for a free curve welding seam through a mechanical arm. The method comprises the following steps that a laser contourgraph is deployed to scan a welding seam on a mechanical arm, geometric feature data of the welding seam are generated, and surface state mapping is conducted; thirdly, analyzing the distribution trend of pores of the welding seam, estimating the welding compactness loss, and further evaluating the fracture risk of the welding seam; according to the fracture risk data, welding heat output is regulated and controlled so as to optimize the welding quality; and finally, constructing an intelligent welding tracking model based on a random forest algorithm, and uploading the intelligent welding tracking model to a cloud platform to realize intelligent tracking welding control of the mechanical arm on the free curve welding seam. According to the invention, the intelligent tracking technology of the welding seam is optimized, so that the intelligent tracking technology of the welding seam is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent tracking of welds, and in particular to an intelligent tracking method and system for a free-curve weld by a robot arm. Background Art

[0002] The robot arm has high precision and high repeatability, and can effectively control the welding trajectory and posture when performing welding tasks. However, due to the complexity of free-curve welds, the robot arm faces the challenge of dynamic changes when tracking these irregular paths. Therefore, it is particularly important to develop a welding tracking method based on intelligent technology. This method not only requires high-precision sensors to monitor the geometric characteristics of the weld in real time, but also requires the use of data processing technology to analyze the state changes of the weld in order to dynamically adjust the welding parameters, such as welding speed, heat input, etc.; through accurate real-time data feedback, the robot arm can automatically correct the welding trajectory according to the changes in the weld, thereby maintaining the welding quality. However, the traditional method of intelligent tracking of free-curve welds by a robot arm has a large error in the intelligent tracking welding control of free-curve welds, resulting in uneven welding quality between different welds. Summary of the invention

[0003] Based on this, it is necessary to provide a method and system for intelligently tracking free-curve welds by a robotic arm to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for intelligently tracking a free-curve weld by a robotic arm is provided, the method comprising the following steps: Step S1: deploying a laser profiler on the robot arm, and performing a curved weld profile scan on the free-curve weld to obtain curved weld profile data; performing weld surface state analysis on the curved weld profile data to obtain weld surface state data; performing geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; Step S2: analyzing the weld porosity distribution trend of the weld surface state mapping data to obtain weld porosity distribution trend data; simulating and estimating the weld density loss of the weld direction distribution based on the weld porosity distribution trend data to obtain weld density loss estimation data; calculating the weld fracture risk based on the weld density loss estimation data to obtain weld fracture risk data; Step S3: matching the welding heat output constraint control of the robot arm between different weld surface states with the weld surface state mapping data according to the weld fracture risk data to obtain welding heat output constraint control data; Step S4: Based on the random forest algorithm, the free curve weld intelligent welding tracking model is constructed for the welding heat output constraint control data to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0005] Preferably, step S1 comprises the following steps: Step S11: deploying a laser profiler on the robot arm, and scanning the free-curve weld profile to obtain the curve weld profile data; Step S12: cleaning the curved weld profile data to obtain the curved weld profile cleaning data; Step S13: performing a curve weld geometric feature analysis on the curve weld profile cleaning data to generate curve weld geometric data; Step S14: performing weld surface state analysis on the curved weld profile cleaning data to obtain weld surface state data; Step S15: performing geometric trend weld surface state mapping processing on the weld surface state data according to the curved weld geometry data to obtain weld trend surface state mapping data.

[0006] Preferably, step S2 comprises the following steps: Step S21: analyzing the weld pore distribution trend on the weld surface state mapping data to obtain weld pore distribution trend data; Step S22: analyzing the trend change of the weld width / height on the curved weld geometry data to obtain the trend change data of the weld width / height; Step S23: performing a simulation estimation of the weld density loss of the weld direction distribution based on the weld porosity distribution direction data and the weld width / height trend change data to obtain weld density loss estimation data; Step S24: calculating the weld distribution brittleness increment index for the weld direction density loss estimation data to obtain the weld distribution brittleness increment index; Step S25: Calculate the weld fracture risk on the weld trend surface state mapping data according to the weld trend density loss estimation data and the weld distribution brittleness increment index to obtain the weld fracture risk data.

[0007] Preferably, step S23 includes the following steps: Step S231: performing direction distribution density identification on the weld pore distribution direction data to obtain the weld pore distribution density; performing pore chain linear distribution identification on the weld pore distribution direction data based on the weld pore distribution density to obtain pore chain linear distribution data; Step S232: calculating the effective bearing volume mean difference of the weld width / height trend change data to obtain the effective bearing volume mean difference of the weld; Step S233: performing pore connectivity analysis on the pore chain linear distribution data to obtain pore distribution pore connectivity data; Step S234: performing deviation regression pore structure normal distribution sampling on the pore chain linear distribution data according to the mean difference of the effective bearing volume of the weld and the pore connectivity data of the pore distribution to obtain deviation regression pore structure normal sampling data; Step S235: Based on the least squares estimation algorithm, the deviation regression pore structure normal sampling data is used to simulate and estimate the welding density loss of the weld trend distribution to obtain the weld trend density loss estimation data.

[0008] Preferably, step S24 includes the following steps: Step S241: performing stress instability concentration identification on the weld direction density loss estimation data to obtain the density loss stress instability concentration; Step S242: performing nonlinear simulation calculation of thermal plastic strain attenuation based on the density loss stress instability concentration and the weld direction density loss estimation data to obtain thermal plastic strain attenuation fitting data; Step S243: performing attenuation recursive median absolute deviation calculation on the thermal plastic strain attenuation fitting data to obtain the plastic recursive attenuation median absolute deviation; Step S244: performing attenuation decomposition constraint analysis on the thermal energy plastic strain attenuation fitting data according to the plastic recursive attenuation median absolute deviation and divide-and-conquer algorithm to generate thermal energy plastic strain attenuation decomposition data; Step S245: Calculate the weld distributed brittleness increment index according to the thermal plastic strain attenuation decomposition data to obtain the weld distributed brittleness increment index.

[0009] Preferably, step S3 comprises the following steps: Step S31: normalizing the weld fracture risk data to obtain weld fracture risk normalized data; Step S32: matching the welding heat output constraint control of the robot arm between different weld direction surface states with the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data to obtain the welding heat output constraint control data.

[0010] Preferably, step S32 includes the following steps: Step S321: analyzing the welding heat energy limit values ​​between different weld direction surface states on the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data, and obtaining the welding heat energy limit values ​​between different weld direction surface states; Step S322: performing robot arm contact radius matching based on welding heat energy limit values ​​between different weld seam surface states to obtain robot arm contact radius matching data; Step S323: performing robot arm welding force matching according to the welding heat energy limit value and the robot arm contact radius matching data to obtain robot arm welding force matching data; Step S324: Based on the robot arm contact radius matching data, the robot arm welding force matching data and the welding heat energy limit value, the robot arm welding heat output constraint control matching between different weld direction surface states is performed to obtain the welding heat output constraint control data.

[0011] Preferably, step S4 comprises the following steps: Step S41: performing logic learning on the welding heat output constraint control data to obtain welding heat output constraint learning data; Step S42: constructing a free curve weld intelligent welding tracking model for the welding heat output constraint learning data based on a random forest algorithm to obtain a weld intelligent welding tracking model; Step S43: Send the weld intelligent welding tracking model to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0012] Preferably, the present invention further provides an intelligent tracking system for a free-curve weld by a robot arm, which is used to execute the intelligent tracking method for a free-curve weld by a robot arm as described above, and the intelligent tracking system for a free-curve weld by a robot arm comprises: The weld surface state mapping module is used to deploy a laser profiler on the robot arm, and perform a curved weld profile scan on the free-curve weld to obtain curved weld profile data; perform weld surface state analysis on the curved weld profile data to obtain weld surface state data; perform geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; The weld fracture risk calculation module is used to analyze the weld porosity distribution trend based on the weld surface state mapping data to obtain the weld porosity distribution trend data; simulate and estimate the weld density loss based on the weld porosity distribution trend data to obtain the weld density loss estimation data; perform weld fracture risk calculation based on the weld density loss estimation data to obtain the weld fracture risk data; A welding heat output constraint control module is used to match the welding heat output constraint control of the robot arm between different weld direction surface states according to the weld fracture risk data and the weld direction surface state mapping data, so as to obtain the welding heat output constraint control data; The intelligent welding tracking model building module is used to build a free curve weld intelligent welding tracking model based on the welding heat output constraint control data based on the random forest algorithm to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0013] The beneficial effect of the present invention is that by deploying a laser profiler on a robotic arm, accurate scanning of free-curve welds can be achieved. This process obtains high-precision contour data of the weld, thereby providing basic data for subsequent geometric feature analysis. By analyzing the geometric features of the weld contour data, the detailed morphology and structural features of the weld can be extracted, including key information such as the direction, curvature, and width of the weld. Using these geometric data, surface state mapping processing is further performed to obtain weld surface state mapping data. Through this series of precise measurements and analyses, the robotic arm can clearly identify the geometric morphology and surface state of the weld, providing an accurate basis for subsequent welding trajectory control and heat input adjustment. Therefore, step S1 not only improves the accuracy of the welding trajectory, but also lays a solid foundation for the optimization of the entire intelligent welding process. The weld trend surface state mapping data mainly includes the distribution of weld surface defects, such as pores, cracks, slag inclusions, surface unevenness, etc. The distribution, size, and quantity of these defects directly reflect the quality problems in the welding process and affect the strength and durability of the welded joint. By further analyzing the mapping data of the weld direction and surface state, especially the analysis of the distribution direction of weld porosity, potential defects in the weld can be effectively identified. Porosity is one of the common defects in the welding process and has a direct impact on the welding quality. By analyzing the distribution direction of weld porosity, the density of the weld can be evaluated in real time, and the quality problems of the weld joint can be accurately predicted based on the simulated estimation of the loss of welding density. Then, based on this estimated data, the weld fracture risk is calculated to obtain the weld fracture risk data. By predicting and assessing the risks of these defects in advance, quality problems in the welding process can be effectively avoided, and necessary data support can be provided for subsequent welding control. This step ensures the high quality and reliability of the welding process and effectively reduces production losses caused by defects. According to the weld fracture risk data, the robot arm performs heat output constraint regulation and matching between different weld direction surface states. This process ensures that the robot arm can accurately adjust the welding heat input according to different weld states, thereby optimizing the welding effect. Different weld directions and surface states require different welding parameters and heat input to achieve the best welding quality. Therefore, precise heat output control can effectively avoid welding defects caused by uneven heat input, such as weld offset and thermal cracks. Through this process, the welding quality is further improved, and it is ensured that each weld can be welded under the most suitable thermal conditions, avoiding joint weakening and performance degradation caused by improper heat input. By analyzing the welding heat output constraint control data based on the random forest algorithm, an intelligent welding tracking model for free curve welds was constructed. By integrating the results of multiple decision trees, the random forest algorithm can effectively process complex multidimensional data and make accurate predictions. In the welding tracking model, this algorithm optimizes the heat output control data during the welding process, further improving the welding accuracy.The constructed intelligent welding tracking model can adapt to the changes in weld morphology and surface state during welding in real time, intelligently adjust the welding trajectory and heat input, and ensure the stability and accuracy of the welding process. After sending the model to the cloud platform, the robot arm can perform intelligent tracking welding control of free curve welds according to real-time data, thereby realizing a fully automatic and seamless welding process. This innovative intelligent control system not only improves welding efficiency, but also greatly improves welding quality and consistency, and promotes the development of the welding industry towards intelligence and automation. Therefore, the present invention is an optimization treatment of a traditional method for intelligent tracking of free curve welds by a robot arm, which solves the problem that a traditional method for intelligent tracking of free curve welds by a robot arm has a large error in the intelligent tracking welding control of free curve welds, thereby causing uneven welding quality between different welds, reducing the error in the intelligent tracking welding control of free curve welds, and improving the welding quality between different welds. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of the steps of a method for intelligently tracking a free-curve weld by a robotic arm; Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart; Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG. DETAILED DESCRIPTION

[0015] See also Figures 1 to 3 , an intelligent tracking method for a free-curve weld by a robot arm, the method comprising the following steps: Step S1: deploying a laser profiler on the robot arm, and performing a curved weld profile scan on the free-curve weld to obtain curved weld profile data; performing weld surface state analysis on the curved weld profile data to obtain weld surface state data; performing geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; Step S2: analyzing the weld porosity distribution trend of the weld surface state mapping data to obtain weld porosity distribution trend data; simulating and estimating the weld density loss of the weld direction distribution based on the weld porosity distribution trend data to obtain weld density loss estimation data; calculating the weld fracture risk based on the weld density loss estimation data to obtain weld fracture risk data; Step S3: matching the welding heat output constraint control of the robot arm between different weld surface states with the weld surface state mapping data according to the weld fracture risk data to obtain welding heat output constraint control data; Step S4: Based on the random forest algorithm, the free curve weld intelligent welding tracking model is constructed for the welding heat output constraint control data to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0016] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for intelligently tracking a free-curve weld by a robot arm of the present invention. In this example, the method for intelligently tracking a free-curve weld by a robot arm includes the following steps: Step S1: deploying a laser profiler on the robot arm, and performing a curved weld profile scan on the free-curve weld to obtain curved weld profile data; performing weld surface state analysis on the curved weld profile data to obtain weld surface state data; performing geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; In the embodiment of the present invention, when deploying a laser profiler on a robotic arm, first select a profile sensor with high-precision laser scanning capability, such as KEYENCE The LJ-V7000 series laser profiler is fixed on the end effector of the robot arm, and the laser emission direction is perpendicular to the weld surface. During the scanning process, the robot arm moves at a constant speed according to the preset trajectory. The laser profiler collects the height change data of the weld surface in real time, and the data is stored in the controller in the form of point cloud. Subsequently, MATLAB is used to reduce the noise of the collected point cloud data, and the mean filter algorithm is used to remove high-frequency noise. Then, the free curve contour data of the weld is reconstructed by the B-spline curve fitting method. The data is stored in the SQL database. In the weld surface state analysis stage, the point cloud data is reconstructed in three dimensions by OpenCV, and the weld edge feature points are extracted by the Canny edge detection algorithm. Then, the K-means clustering algorithm is used to classify the weld surface defects, including pores, slag inclusions and cracks. The data is stored in JSON format. Then, the geometric trend weld surface state mapping processing is performed, and the Bezier curve interpolation method is used to generate the weld geometric trend. Based on the surface state data, the principal component analysis (PCA) method is used to extract the key feature points of the weld surface, and the weld trend surface state mapping data is generated. The data is stored in the HDF5 format file.

[0017] In another embodiment, a laser profiler is installed at the end of the robot arm. The laser profiler has the characteristics of a lateral resolution of 10 μm and a scanning frequency of 4 kHz, which can meet the needs of high-precision real-time contour acquisition of weld curves with large curvature changes. The robot arm adopts a six-degree-of-freedom structure, model ABB IRB 2600, with a repeatability accuracy of ±0.02 mm. By synchronously controlling the path planning of the robot arm and the scanning of the laser profiler, the robot arm is driven to scan along the preset initial path of the weld. The laser profiler continuously collects three-dimensional contour data of the weld, and the data is transmitted to the industrial computing platform in real time through the EtherCAT bus for processing. The amount of raw data of the weld contour obtained is about 12,000 point cloud data per meter of weld. Subsequently, the point cloud data is subjected to abnormal point elimination processing based on density clustering to filter out outliers caused by reflection and jitter, and the outliers are re-constructed based on the B-spline curve. An algorithm is built to continuously fit the weld contour and construct a weld curve geometric model with an accuracy of 0.02mm. The surface grayscale reflectivity change data is combined to extract the weld surface state information, which includes penetration pits, surface uneven oxidation spot areas, spatter attachment areas, etc. The weld surface state data vector is extracted through texture gradient direction recognition combined with grayscale co-occurrence matrix calculation. The vector is used to reduce the dimension through principal component analysis (PCA) method and spatially align with the weld geometric model to form a mapping data matrix between the weld geometric direction and the surface state, which serves as the basic input for subsequent defect analysis.

[0018] Step S2: analyzing the weld porosity distribution trend of the weld surface state mapping data to obtain weld porosity distribution trend data; simulating and estimating the weld density loss of the weld direction distribution based on the weld porosity distribution trend data to obtain weld density loss estimation data; calculating the weld fracture risk based on the weld density loss estimation data to obtain weld fracture risk data; In an embodiment of the present invention, in the stage of analyzing the distribution trend of weld pores, a deep learning method is used to analyze the surface state mapping data of the weld trend. First, the porosity defects are marked using the YOLOv5 target detection algorithm, and the main direction of the pore distribution is calculated according to the weld trend information. The main direction is calculated using PCA principal axis analysis to obtain the weld pore distribution trend data, and the data format is stored in CSV. Subsequently, based on the weld pore distribution trend data, a simulation estimation of the weld density loss of the weld trend distribution is performed, and the shrinkage stress of the weld metal is calculated according to the welding energy input distribution using the finite element analysis (FEA) method. A weld density loss prediction model is established in combination with the pore distribution characteristics, and the welding density loss is calculated using a custom numerical solution algorithm written in Python. The result is stored in an XML format file, and based on the weld trend density loss estimation data, a weld fracture risk calculation is performed, and a fracture mechanics analysis method is used to calculate the crack propagation rate caused by the weld porosity, and the crack propagation life is evaluated using the Paris formula.

[0019] In another embodiment, the weld seam pore distribution trend analysis is performed on the weld seam trend surface state mapping data, the density distribution of weld seam pores in different weld seam trend areas is calculated, the distribution probability of pores is calculated using the histogram statistical method, and the concentration degree of pores in different weld seam areas is determined. For the spatial distribution characteristics of weld seam pores, the Voronoi diagram analysis method is used to calculate the relative distance between pores, and the connectivity and chain distribution characteristics of pores are identified. Based on the weld seam pore distribution trend data and the geometric characteristics of the weld seam surface, a welding density loss estimation model is constructed. First, the effective bearing volume of the weld is calculated, and the volume segmentation method is used to perform layered calculations on the weld seam area, and the effective volume of each layer of the weld is calculated. The finite element analysis method is used to simulate the distribution of heat input during welding, the welding heat input is set to 350 J / mm, and the heat affected zone range of the weld is calculated. During the simulation process, the cooling rate is set to 8 ℃ / s, the phase change during the weld cooling process is calculated, and the influence of weld seam pores on welding density during the cooling process is analyzed. For the estimation of weld density loss, the regression analysis method was used to calculate the weld sealing loss, and the mapping relationship between weld sealing loss and pore distribution was established. According to the weld density loss estimation data, the fracture mechanics method was used to calculate the weld fracture risk, and the critical stress of weld crack propagation was calculated based on the Griffith energy criterion. The weld stress distribution model was established using the finite element method, the Young's modulus of the weld material was set to 210 GPa, the Poisson's ratio was set to 0.3, and the stress concentration coefficient of the weld area was calculated. For the weld fracture risk calculation, the maximum principal stress on the weld surface was calculated and compared with the fracture toughness of the weld material to determine whether the weld has a fracture risk. Finally, the weld fracture risk data was obtained, which provided basic data for subsequent weld tracking control.

[0020] Step S3: matching the welding heat output constraint control of the robot arm between different weld surface states with the weld surface state mapping data according to the weld fracture risk data to obtain welding heat output constraint control data; In an embodiment of the present invention, according to the weld fracture risk data, the robot arm welding heat output constraint control matching between the surface states of different weld directions is performed. First, a welding heat input mathematical model is established. Based on the weld geometry, material properties and welding process parameters, the temperature distribution of the weld is calculated using the heat conduction equation. The welding heat conduction simulation is performed using COMSOL Multiphysics software. Combined with the weld fracture risk data, the welding heat input of different weld directions is constrained and optimized. The optimization method adopts the particle swarm optimization (PSO) algorithm, with the minimization of welding heat input as the objective function, to ensure that the weld joint has uniform metallurgical structure and mechanical properties. The optimized welding heat output parameters are stored as a JSON file, and finally uploaded to the welding control system of the robot arm to achieve intelligent matching of welding heat output.

[0021] In another embodiment, the weld fracture risk data is input into the welding heat output control module. First, the fracture risk data is normalized to the maximum and minimum values ​​so that all risk values ​​are mapped to the [0, 1] interval. Then, the state of each weld segment is classified in combination with the weld geometric trend information. The K-means clustering algorithm is used to divide the weld trend surface state mapping data into five state segments, including high curvature high risk area, low curvature medium risk area, wide weld low risk area, etc. For each state segment, the upper and lower limits of the heat energy input are calculated according to its normalized fracture risk value and the weld width-to-height ratio. The fuzzy control method is used to set the heat output control rules to generate the corresponding welding heat output power control value. The control logic is that the higher the risk level, the lower the heat input. At the same time, the welding gun movement speed and angle are adjusted to ensure that the heat affected zone is minimized. Finally, a welding heat output constraint control data table is formed, which records the control power value, welding gun angle, welding speed and other parameters for each 1 mm weld segment.

[0022] Step S4: Based on the random forest algorithm, the free curve weld intelligent welding tracking model is constructed for the welding heat output constraint control data to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0023] In an embodiment of the present invention, based on the random forest algorithm, a free curve weld intelligent welding tracking model is constructed for welding heat output constraint control data. First, different weld geometries, welding heat input and weld quality data are collected to construct a weld feature data set in a CSV format. Subsequently, the Scikit-learn library of Python is used to train the random forest model. 100 decision trees are set, and the maximum depth of each tree is 10. Five-fold cross validation is used during the training process, and mean square error (MSE) is used as an evaluation index. After the model training is completed, the random forest model is tuned by the XGBoost algorithm, and finally a weld intelligent welding tracking model is obtained. The model is stored in ONNX format. In the model deployment stage, the weld intelligent welding tracking model is uploaded to the AWS cloud platform, and communicated with the robotic arm control system through the MQTT protocol to realize remote intelligent tracking welding control.

[0024] In another embodiment, the welding heat output constraint control data is input into the intelligent tracking model construction module, and the model is constructed using a random forest algorithm with a tree structure depth of 10. The training data includes a six-dimensional feature combination of weld direction, surface state characteristics, pore distribution parameters, density loss rate, fracture risk level and heat output parameters. The number of training samples is 100,000. The training uses the Gini coefficient as the splitting criterion, and the forest structure is optimized through 5-fold cross validation to finally generate a weld intelligent welding tracking model. The model takes the weld geometric direction and surface state as input, and outputs the corresponding heat output power, welding speed and path correction angle. After the model training is completed, the model file is uploaded to the cloud platform management system through the MQTT protocol, and the cloud platform automatically sends it to the robotic arm controller. During the actual welding process, the controller reads the weld state data fed back by the laser profiler in real time, and performs path prediction and heat output control by calling the model to realize dynamic intelligent tracking welding control of free curve welds.

[0025] Step S1 includes the following steps: Step S11: deploying a laser profiler on the robot arm, and scanning the free-curve weld profile to obtain the curve weld profile data; Step S12: cleaning the curved weld profile data to obtain the curved weld profile cleaning data; Step S13: performing a curve weld geometric feature analysis on the curve weld profile cleaning data to generate curve weld geometric data; Step S14: performing weld surface state analysis on the curved weld profile cleaning data to obtain weld surface state data; Step S15: performing geometric trend weld surface state mapping processing on the weld surface state data according to the curved weld geometry data to obtain weld trend surface state mapping data.

[0026] In the embodiment of the present invention, when a laser profiler is deployed on a robotic arm, a KEYENCE LJ-V7000 series laser profiler is selected and fixed on the end effector of the robotic arm. A customized aluminum alloy bracket is used to ensure the installation stability of the laser profiler. A six-axis force sensor is used to calibrate the installation angle so that the laser beam is projected vertically onto the weld surface. The robotic arm moves at a constant speed along a set path. The laser profiler emits a high-frequency laser beam, receives a light signal reflected from the weld surface, and calculates the three-dimensional profile data of the weld surface through triangulation. The data point interval is controlled within 0.05 mm. The collected point cloud data is transmitted to the industrial computer through the EtherCAT protocol and stored as a PLY format file. When cleaning the curved weld profile data, the radius filtering algorithm is first used to remove outliers in the point cloud. The radius threshold is set to 0.1 mm, and only the data points that meet the density conditions in the local neighborhood are retained. Subsequently, the statistical filtering method is applied to calculate the average distance of each point in the point cloud, and abnormal data points that exceed 3 times the standard deviation are removed. The MLS (moving least squares) surface reconstruction method is used to smooth the curved weld profile to eliminate high-frequency noise in the acquisition process. The cleaned curved weld profile data is stored in PCL (Point Cloud Library) format and data backup is performed.

[0027] When analyzing the geometric features of curved welds on the curved weld contour cleaning data, the RANSAC (random sampling consistency) algorithm is used to perform plane fitting on the point cloud data. The weld centerline is used as the reference coordinate system to calculate the weld width, height and groove angle. The DBSCAN (density-based spatial clustering) method is used to cluster the weld boundary points and calibrate the weld boundary line. The discontinuous feature points of the weld are extracted by combining the second-order derivative change rate, and the weld curvature mutation area is identified. The least squares method is used to fit the B-spline curve to characterize the geometric direction of the weld. Finally, the weld geometry data is generated, the data is stored in JSON format, and uploaded to the welding control system. When analyzing the weld surface status of the curved weld profile cleaning data, the point cloud data is first gridded and the weld surface is divided into 0.1mm×0.1mm unit grids. The Canny edge detection algorithm of OpenCV is used to extract the main feature contours of the weld surface. The gray-level co-occurrence matrix (GLCM) is used to calculate the weld surface texture features, distinguish the molten pool area, heat-affected zone and parent material area, and the ResNet-50 deep learning model is used to classify the defects of the weld surface. The defects are divided into three categories: pores, slag inclusions and undercuts. The K-means clustering algorithm is used to analyze the surface roughness data and calculate the root mean square value (Rq) of the surface roughness. The weld surface status data is stored in HDF5 format and synchronized to the welding quality monitoring system.

[0028] When the weld surface state data is mapped to the geometric trend of the weld surface state according to the curved weld geometry data, the Bezier curve interpolation method is used to construct the weld trend curve, and the PCA (principal component analysis) method is combined to reduce the dimension of the weld surface state data, extract the key feature points of the weld surface, and use the Poisson surface reconstruction algorithm to generate a three-dimensional weld surface state model. The weld geometry information is fused with the surface state data, and a mapping relationship is established. Finally, the weld trend surface state mapping data is generated, and the data is stored as an SQLite database and imported into the weld intelligent tracking control module.

[0029] In another embodiment, a laser profiler is customized and installed, and the installation angle of the laser profiler is controlled to be 30 degrees with the normal of the weld surface, ensuring that the laser beam irradiates the weld surface in an inclined manner to enhance the reflection contrast of the concave and convex features of the weld. When the profiler is working, a line scan is performed at a frequency of 4kHz through a built-in blue laser source, the scanning line width is set to 25mm, and the resolution is set to 10μm. The robotic arm moves along the free curve weld trajectory to be welded at a speed of 100mm / s according to a preset path. The path planning adopts an offline trajectory point set generated based on the CAD model, and a trajectory control point is set every 10mm. The profiler transmits the scanned two-dimensional profile data to the industrial computing unit in the control cabinet in real time through the EtherCAT high-speed communication interface. The scanning frequency and the robotic arm speed are coordinated to control the three-dimensional point cloud reconstruction of the weld, and a weld contour data set consisting of approximately 12,000 contour lines per meter of weld is obtained. The data set includes spatial information such as weld cross-sectional height, weld boundary position, excess height, groove shape, etc.

[0030] After obtaining the weld contour data, the original point cloud data needs to be cleaned to eliminate outliers and noise points caused by surface reflection or impurities during the scanning process. The DBSCAN density clustering algorithm is used in the cleaning process to cluster and identify the point cloud on each contour line. The clustering radius is set to 0.05 mm and the minimum number of samples is 4. After clustering, isolated points and small low-density noise clusters are removed, and only the main contour point set is retained. At the same time, the point set is locally smoothed, and each contour line is smoothed using the Gaussian weighted sliding window method. The window size is set to 5 and the weight standard deviation is 1.2. The height mutations between the contour lines are further checked. If the height difference between the corresponding points of adjacent contour lines exceeds 0.2 mm, it is judged as an abnormal jump. The cubic spline interpolation method is used for transition reconstruction, and finally the three-dimensional weld contour cleaning data with good continuity and abnormal point removal is generated. The data retention rate after cleaning is about 92%. When performing geometric feature analysis on the cleaned weld contour data, the geometric parameters of each cross-sectional contour are first extracted, including weld excess height, weld width, groove angle, weld center offset, etc. The outermost point pair of the contour line is used as the width boundary, and its lateral distance is calculated as the weld width. The distance from the highest point in the vertical direction to the plate reference plane is defined as the excess height. The groove angle is obtained by tangent calculation by fitting the inclined segment of the contour edge. The weld center offset is the horizontal offset distance between the contour center of gravity and the preset center line. The collected geometric parameters are stored in units of 10 mm welds to form a curved weld geometry data sequence. At the same time, the weld trend is analyzed, and the weld center line of each sampling point is fitted using the least squares method. The curvature change of the fitting curve is extracted as the weld geometry trend parameter, and finally a geometric feature data set including weld excess height, weld width, groove angle, center offset and trend curvature is formed. When analyzing the weld surface state of the weld contour cleaning data, the grayscale reflection measurement function of the contour meter is used to obtain the reflection intensity value of each contour point. The image texture analysis technology is combined to identify the surface defect area of ​​the weld. First, the grayscale co-occurrence matrix is ​​used to extract texture features, including energy, contrast, uniformity and correlation. The window size is set to 5×5 and the step size is 1. The sliding calculation is performed on the entire weld surface, and the area with abnormal texture features is identified as the surface state abnormal area. At the same time, the weld surface is analyzed to extract the weld surface roughness Ra value, which is obtained by calculating the average value of the height difference between adjacent points. Each weld section is analyzed with a length of 100 mm as an analysis unit. The state characteristics of the surface penetration depression area, spatter attachment area, oxidation discoloration area, etc. are counted to form a weld surface state data set containing grayscale reflection value, texture parameters, surface roughness and surface abnormality type. This data set corresponds to the geometric feature data one by one for subsequent mapping processing.

[0031] According to the extracted weld geometry data, the weld surface state data is mapped to the weld surface state in the geometric direction. The surface state data is remapped to the corresponding position of the geometric model in three-dimensional space by the spatial relocation method. First, the spatial direction reference coordinate system is constructed by the weld geometry centerline. The weld direction is taken as the X-axis and the normal is taken as the Z-axis. The surface state data points are converted from the scanning coordinate system to the weld direction coordinate system by local coordinate transformation. Then, the surface state data are bound according to the weld geometry segments to establish a point-to-point data mapping matrix. Each data point contains the direction curvature value, width, residual height, groove angle and its corresponding surface state parameters of the corresponding segment. Finally, the weld direction surface state mapping data is generated. The data storage format is a multidimensional array structure. The structure contains the spatial position index, geometric parameter subset, surface state subset and mapping index relationship. The mapping data is used as the basic input for the subsequent weld structure defect assessment and welding thermal control model construction.

[0032] Step S2 includes the following steps: Step S21: analyzing the weld pore distribution trend on the weld surface state mapping data to obtain weld pore distribution trend data; Step S22: analyzing the trend change of the weld width / height on the curved weld geometry data to obtain the trend change data of the weld width / height; Step S23: performing a simulation estimation of the weld density loss of the weld direction distribution based on the weld porosity distribution direction data and the weld width / height trend change data to obtain weld density loss estimation data; Step S24: calculating the weld distribution brittleness increment index for the weld direction density loss estimation data to obtain the weld distribution brittleness increment index; Step S25: Calculate the weld fracture risk on the weld trend surface state mapping data according to the weld trend density loss estimation data and the weld distribution brittleness increment index to obtain the weld fracture risk data.

[0033] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: analyzing the weld pore distribution trend on the weld surface state mapping data to obtain weld pore distribution trend data; In an embodiment of the present invention, when analyzing the weld porosity distribution trend of the weld trend surface state mapping data, the weld trend surface state mapping data stored in the SQLite database is first called, the data format is parsed using the Python pandas library, the coordinates of the weld surface feature points and the pore annotation information are extracted, the three-dimensional point cloud data is voxelized using the Open3D tool, the weld surface is divided into 0.1 mm × 0.1 mm grid units, and the distribution density of the pores in each unit is calculated, the DBSCAN density clustering algorithm is used to identify the main direction of the pore distribution, and its direction angle is calculated, and the main direction of the overall pore distribution of the weld is obtained in combination with the PCA principal axis analysis method, and the weld pore distribution trend data is generated, the data is stored in the HDF5 format, and synchronized to the weld quality monitoring system.

[0034] In another embodiment, when analyzing the distribution trend of weld pores on the weld trend surface state mapping data, a three-dimensional weld model is first constructed based on the weld surface point cloud data, and the deep learning target detection algorithm YOLOv5 is used to identify the pore defects. The grayscale value of the pixel points in the pore area is at least 15% lower than the average grayscale value of the surrounding weld metal. The pore edges are refined using morphological operations, the center coordinates and size information of the pores are identified through Hough transform, the center coordinates of all pores are mapped to the weld geometry curve, the DBSCAN density clustering algorithm is used to classify the pores, and the distribution direction of the pores on the weld is calculated based on the clustering results. The principal component analysis (PCA) method is used to extract the main distribution axis of the pores, the weld pore distribution trend data is generated, and the data is stored in JSON format.

[0035] Step S22: analyzing the trend change of the weld width / height on the curved weld geometry data to obtain the trend change data of the weld width / height; In an embodiment of the present invention, when analyzing the trend change of weld width / height for curved weld geometry data, firstly, the curved weld geometry data is read, the matplotlib library is called to draw the weld cross-sectional profile, the change trend of weld width and height is calculated, the Sobel edge detection algorithm is used to extract weld boundary points, the weld width and height curves are fitted using the least squares method, the geometric change rate of the weld under different welding trends is calculated, the second-order derivative analysis method is used to calculate the fluctuation degree of weld width / height, and the Fourier transform method is used to analyze its frequency characteristics, and the weld width / height trend change data is generated, the data is stored in JSON format, and imported into the welding process monitoring system.

[0036] In another embodiment, when analyzing the trend change of the weld width / height of the curved weld geometry data, the least squares method is used to fit the weld centerline based on the weld point cloud data, and the width and height of the weld along the centerline direction are calculated. The moving window method is used to analyze the width / height change trend of the weld. The window length is set to 2 mm, and the sliding step size is 0.5 mm. The mean and standard deviation of the weld width and height in each window are calculated. The wavelet transform is used to perform frequency domain analysis on the change signals of the weld width and height, and the high-frequency components are extracted to detect the local mutation area of ​​the weld. The Fourier transform is used to calculate the periodic change trend of the weld width and height along the weld direction, and the weld width / height trend change data is stored in the SQLite database and synchronized to the welding process control system.

[0037] Step S23: performing a simulation estimation of the weld density loss of the weld direction distribution based on the weld porosity distribution direction data and the weld width / height trend change data to obtain weld density loss estimation data; In the embodiment of the present invention, in the process of simulating and estimating the loss of welding density in the distribution of weld strike, it is first necessary to establish a finite element analysis model for the weld area, select appropriate material parameters, such as the thermal conductivity, specific heat capacity and density of the weld base material, to ensure that the model can accurately reflect the heat conduction and phase change behavior during welding. Subsequently, based on the pore distribution data on the weld surface, pore defects of different scales are implanted in the finite element grid, the morphological characteristics of the pores (such as elliptical, circular or irregular shapes) are defined, and a lower thermal conductivity is given in the pore area to reflect the shielding effect of the pores on welding heat transfer. Steady-state and transient heat conduction coupling analysis is adopted, a welding heat source model (such as a Gaussian distribution heat source) is applied, the heat input power density range is set to 350–420 J / mm, and the moving heat source method is used to simulate the temperature field evolution process in the weld area. In the cooling stage, the dynamic deformation process of the weld geometric boundary is defined in combination with the strike change data of the weld width and height, and the shrinkage stress distribution in the weld area is calculated based on the thermal stress analysis method. Through material microstructure simulation, the density change of the weld area is calculated to quantify the loss of weld density caused by pores and weld geometry inhomogeneity. Using the effective density ratio of the weld area (i.e. the ratio of the actual density of the weld area after welding to the theoretical density), combined with the density loss rate calculation formula, the estimated data of welding density loss in different areas are obtained, and finally the distribution data of density loss along the weld direction is formed, providing basic data support for the subsequent weld fracture risk assessment.

[0038] In another embodiment, when simulating and estimating the welding density loss along the weld direction distribution based on the weld pore distribution direction data and the weld width / height direction change data, firstly, a weld micro-metallurgical structure model is established, and the phase field method is used to simulate the solidification process of the weld metal. The pore distribution information and the weld width / height change data are input into the finite element simulation software Abaqus, and the stress distribution inside the weld is calculated by thermal-structural coupling analysis. According to the thermal cycle curve inside the weld and the metallographic structure characteristics of the weld joint, the hardness gradient of the weld heat affected zone (HAZ) is calculated, and the density loss ratio of the weld along the direction is calculated by the numerical integration method. The weld direction density loss estimation data is generated, the data is stored in MAT format, and imported into the weld quality analysis system to simulate and estimate the welding density loss caused by the weld direction defect distribution.

[0039] Step S24: calculating the weld distribution brittleness increment index for the weld direction density loss estimation data to obtain the weld distribution brittleness increment index; In an embodiment of the present invention, when calculating the weld distribution brittleness increment index for the weld trend density loss estimation data, it is first necessary to call the weld trend density loss estimation data, which is stored in a database, and the metal density distribution of the weld area is stored in HDF5 format. The h5py library of Python is used to read the density loss values ​​of each weld area, and the data is standardized and converted into an input format that can be parsed by ABAQUS after removing abnormal values. In the ABAQUS finite element software, a finite element model of the weld area is first established, and the C3D8R (8-node reduced integration unit) unit type is used for meshing. The minimum unit size of the mesh is set to 0.1 mm to ensure the accuracy of the simulation results, and the welding material parameters are imported, wherein the weld metal uses ER70S-6 low-carbon steel filler metal, and the base material uses Q235 steel, and the elastic modulus E=210GPa and Poisson's ratio ν=0.3 are respectively assigned to them, and the Johnson-Cook constitutive model is defined to describe the strain hardening effect of the weld area, and the bilinear strengthening model is used to describe the weld. The plastic deformation characteristics under high temperature welding state are set. The initial yield strength of the material is set to y=350MPa, and the maximum tensile strength is set to u=500MPa. In the residual stress simulation analysis process, the heat source movement path during welding is set. The Goldak double ellipsoid heat source model is used to simulate the heat input during welding. The heat source power is set to P=6kW, the welding speed is set to v=5mm / s, and the heat source semi-axis parameters are set to af=3mm, ar=6mm, and b=4mm. The thermal cycle curve of the weld metal during welding is calculated, and the thermal-structural coupling analysis method is used to calculate the evolution of the temperature field and stress field inside the weld. Thermal boundary conditions are applied, the convection heat transfer coefficient is set to h=50W / (m²·K), and the ambient temperature is set to 293K. Finally, the residual stress distribution of the weld is solved, and the equivalent stress value of each node is derived. The obtained weld residual stress data is processed by Python's SciPy library, and the stress gradient is calculated using the np.gradient function, and then the stress concentration factor SCF (Stress Concentration Factor), the stress distribution of the weld area is classified by K-means clustering method, the K value is set to 3, the weld area is divided into three areas: high brittleness, medium brittleness, and low brittleness. The fracture mechanics theory is used to calculate the brittleness increment index of each area. The specific calculation is based on the Griffith energy release rate criterion. Calculate the brittle increment of each region, where is the maximum principal stress in the stress concentration area, The calculated brittleness incremental index is stored in the MAT file format and synchronized to the welding quality control system for subsequent welding fracture risk assessment.

[0040] Step S25: Calculate the weld fracture risk on the weld trend surface state mapping data according to the weld trend density loss estimation data and the weld distribution brittleness increment index to obtain the weld fracture risk data.

[0041] In the embodiment of the present invention, when performing weld fracture risk calculation, the weld trend density loss estimation data and weld distribution brittleness incremental index data stored in the database are first read. The weld trend density loss estimation data is stored in HDF5 format, including the density distribution information of each area of ​​the weld. The weld distribution brittleness incremental index data is in MAT format, covering the brittleness incremental index of different areas of the weld. All data are imported and processed through Python's h5py library and scipy library. Based on these data, the extended finite element method (XFEM) is then used to predict the weld crack propagation path. The XFEM method can effectively capture the crack tip singularity and the dynamic behavior of crack propagation. The ABAQUS finite element software is used to simulate the crack propagation in the weld area. For the division of the weld area, a high-precision grid is used, the minimum unit size is set to 0.1mm, and the C3D8R unit type is used for division. Considering the different material properties of the weld, heat-affected zone and parent material, the material of the weld area is selected as ER70S-6 low-carbon steel, and the parent material is selected as Q235 steel. The fracture toughness parameters of the welding material are input, including KIC (critical stress intensity factor) and GIC (critical energy release rate), which are KIC=80MPa√m and GIC=0.1kJ / m² respectively. Then, the Cohesive Zone Model (CZM) is used to simulate the expansion of cracks in different areas of the weld. CZM describes the transition zone from no crack to fracture of the material by introducing the behavior of the bonding zone, and considers the stress and strain distribution characteristics at the crack tip and the inhomogeneity of the welding area. The maximum strength of the bonding zone is defined as 70MPa and the maximum energy release rate is 0.2kJ / m² in the model. For the fatigue crack growth behavior of the weld, the Paris-Erdogan crack growth equation is used to calculate the crack growth rate of the weld under fatigue load. The form of the equation is da / dN=C(ΔK)^m, where da / dN is the crack growth rate, C=1.0×10^-11 (unit: mm / cycle), m=3.0 is the empirical coefficient of the material, ΔK is the change in the stress intensity factor, and the stress amplitude and load frequency are input based on the actual welding condition during the calculation process. The fatigue load frequency of the welding condition is set to f=1Hz, and the load amplitude is set to ΔP=200N. The crack growth life of the weld under fatigue load is calculated by these parameters. After simulating the crack growth path of the weld, the fracture risk of each area is further calculated, and the probability of weld fracture is evaluated based on the time history of crack growth. Combined with the weld trend density loss data, weld distribution brittleness increment index data and fatigue crack growth data, the Monte Carlo method is used to perform random simulation of fracture risk, and finally the weld fracture risk data is obtained.The data includes the probability of weld fracture, crack propagation time, regional fracture risk, etc. All calculation results are stored in CSV format, including the fracture risk values ​​of different weld regions and the corresponding crack propagation life information. Finally, all weld fracture risk data are uploaded to the welding quality prediction system through the API interface. The welding quality prediction system further analyzes and processes the data and generates a corresponding quality assessment report for subsequent welding process optimization and risk control.

[0042] Step S23 includes the following steps: Step S231: performing direction distribution density identification on the weld pore distribution direction data to obtain the weld pore distribution density; performing pore chain linear distribution identification on the weld pore distribution direction data based on the weld pore distribution density to obtain pore chain linear distribution data; Step S232: calculating the effective bearing volume mean difference of the weld width / height trend change data to obtain the effective bearing volume mean difference of the weld; Step S233: performing pore connectivity analysis on the pore chain linear distribution data to obtain pore distribution pore connectivity data; Step S234: performing deviation regression pore structure normal distribution sampling on the pore chain linear distribution data according to the mean difference of the effective bearing volume of the weld and the pore connectivity data of the pore distribution to obtain deviation regression pore structure normal sampling data; Step S235: Based on the least squares estimation algorithm, the deviation regression pore structure normal sampling data is used to simulate and estimate the welding density loss of the weld trend distribution to obtain the weld trend density loss estimation data.

[0043] In an embodiment of the present invention, the spatial density of the pore distribution trend data on the weld surface is first calculated, and the weld surface area is divided into a plurality of equally spaced grid units through rasterization processing. The number of pores is counted in each grid unit, and the pore density value per unit area is calculated. Then, the kernel density estimation algorithm is used to perform continuous processing on the pore distribution density on the weld surface, so that the density data can more accurately reflect the distribution characteristics of the pores in the weld area. After the pore distribution density of the weld is obtained, the connected region analysis method is used to cluster and identify the spatial arrangement of the pores, and the pore areas with continuous distribution and small spacing are divided into chain structures. The minimum spanning tree algorithm is used to calculate the shortest connection path between the pores, identify the linearly arranged pore chain structure, and extract its parameters such as length, direction angle and average spacing between pores to form pore chain linear distribution data.

[0044] The geometric characteristics of the weld cross section were analyzed to extract the spatial variation data of the weld width and height. The curve fitting method was used to smooth the changes in the width and height of the weld along the direction to eliminate the interference of local noise on data analysis. Subsequently, based on the weld cross-sectional area calculation method, the bearing volume of the weld at different positions was determined, and the mean difference of the weld bearing volume was calculated using the statistical analysis method. The calculation of the mean difference of the bearing volume involves the volume deviation of different sections in the weld area. The discrete integral method can be used to numerically calculate the trend of the weld volume change to obtain the mean difference of the effective bearing volume of the weld, which provides basic data for the subsequent simulation of welding density loss. The connectivity of the pore chain linear distribution data obtained in the previous step was evaluated by the topological analysis method. First, based on the three-dimensional reconstruction method, the pore structure on the weld surface was mapped to the three-dimensional space to construct the pore distribution topological network. The Dijkstra algorithm was used to calculate the shortest path between the pores to determine the connectivity between different pores, and the permeability of the pores was evaluated by calculating the length of the connected path and the channel impedance. Based on the fluid dynamics simulation method, the flow characteristics of the pore connection channel are analyzed, the effective permeability coefficient between the pores is calculated, and the pore distribution pore connectivity data is formed in combination with the density parameters of the weld material. Based on the statistical regression method, the deviation characteristics of the weld pore structure are modeled and analyzed.

[0045] Firstly, the least squares regression algorithm was used to establish a nonlinear regression model between the mean difference of the effective bearing volume of the weld and the pore connectivity data of the pore distribution, and the regression coefficient of the pore structure deviation of the weld was calculated. Then, based on the error distribution of the regression model, the normal distribution sampling method was used to randomly sample the deviation of the pore structure of the weld, and the pore structure deviation data that conformed to the normal distribution characteristics were generated. This data can be used for the simulation calculation of the welding density loss to ensure the consistency between the pore distribution characteristics of the weld and the welding quality assessment model. The least squares estimation algorithm was used to simulate the welding density loss of the deviation regression pore structure normal sampling data obtained in the previous stage. Firstly, based on the finite element analysis method, a heat conduction model of the weld area was established, and the thermal and mechanical parameters of the weld material were defined, including the thermal conductivity, specific heat capacity and density of the weld base material. Then, the moving heat source model was used to simulate the heat input of the weld area, and the welding heat input power range was set to 380–420 J / mm, and the phase change behavior during the cooling process of the weld was calculated. Combined with the pore structure data, the density of the weld area is evaluated. The material volume fraction analysis method is used to calculate the proportion of weld density loss caused by pore distribution, and the least squares method is used to fit the distribution curve of weld density loss. Finally, the estimated data of weld density loss is obtained, which provides data support for the risk assessment of weld fracture. Because the pores themselves are hollow defects in the weld, they directly destroy the continuity and density of the weld metal. These pores provide channels for gas or liquid. Even if pressure is applied to the weld surface or in an environment with sealing requirements, the medium can penetrate through the pores, causing the weld to lose density. In addition, due to the irregular shape around the pores, stress concentration is prone to occur, which will cause microcracks. These microcracks will further expand and connect with each other, further deteriorating the density of the weld.

[0046] In another embodiment, when the distribution density of the weld pore distribution trend data is identified, the weld trend path is first divided into equally spaced analysis units, and the length of each unit is set to 10 mm. The center points of the pores identified in each unit are counted, and the number of pores per unit length is calculated as the local pore distribution density. The density threshold is set to 0.3 / mm. When the density value exceeds the threshold, it is marked as a high-density section. At the same time, the spatial distribution direction of the pores in each section is linearly fitted, and the least squares straight line fitting method is used to obtain the main axis direction of the pore distribution, and the angle between the main axis and the main direction of the weld trend is calculated. If the angle is less than 15 degrees and the fitting residual is less than 0.05 mm, it is identified as a pore chain linear distribution area, and the parameters such as the start and end positions, fitting direction, average density, chain length, and linear fitting residual of the area are recorded as pore chain linear distribution data. The data is organized in an array structure, including the spatial index and geometric statistical characteristics of each chain area, which is used for subsequent connectivity and density modeling.

[0047] When calculating the effective bearing volume mean difference of the weld width and height trend change data, first extract the cross-sectional profile of each weld section in the weld three-dimensional model, and set the profile extraction interval to 10 mm. Calculate the weld cross-sectional area of ​​each cross-sectional profile as the local bearing volume of the section. Use the triangulation method to discretize the closed area of ​​the profile and calculate the integral area. Then, perform statistical processing on the bearing volume data of all weld sections, calculate the overall average bearing volume, and calculate the mean difference for each section. The mean difference is defined as the absolute value of the difference between the volume of the section and the overall mean. The obtained effective bearing volume mean difference of the weld is used to measure the consistency and stability of the weld structure. The mean difference value of each section is bound to its corresponding spatial index and output as the effective bearing volume mean difference data of the weld. This data provides a geometric constraint basis for subsequent pore sampling deviation modeling. When analyzing the pore connectivity of the pore chain linear distribution data, the three-dimensional voxelization method is used to divide the weld space into cubic units with a side length of 0.1 mm. The pore center points contained in each pore chain area are voxel mapped, and the voxels where the pores are located are marked as 1, and the non-pore voxels are marked as 0. Then, the three-dimensional connected area recognition algorithm is used to perform a 26-neighborhood search on the voxel group marked as 1 to identify the interconnected pore structures. The volume, surface area, maximum connected path length and connectivity coefficient of each connected area are calculated. The connectivity coefficient is defined as the product of the connected path length and the volume divided by the surface area. The larger the value, the easier it is for the pores to form connected channels. All identified connected areas are numbered and their positions in the weld space are recorded. The output pore distribution pore connectivity data includes the connected area number, start and end positions, connected path length, volume, connectivity and chain structure number, which is used for subsequent pore structure sampling and modeling.

[0048] When the deviation regression pore structure normal distribution sampling is performed on the pore chain linear distribution data according to the mean difference of the effective bearing volume of the weld and the pore connectivity data of the pore distribution, the pore parameter distribution model is first constructed, the connectivity coefficient of each chain pore area is taken as the main variable, and the mean difference of the bearing volume of the corresponding area is taken as the regression offset. The normal distribution sampling model with deviation term is used to simulate the sampling of pore size. The average pore diameter is set to 0.2mm and the standard deviation is set to 0.05mm. The regression offset is used to adjust and correct the sampling mean. If the mean difference of the bearing volume in a certain area is large, the corresponding sampling mean is offset upward to increase the pore size. At the same time, the pore sampling density is increased in the area with high connectivity. 100 samples are taken in each chain structure, and the pore position, diameter, volume and distance to the adjacent pores of each sample are recorded. The sampling results are output in the form of a structure array, named as deviation regression pore structure normal sampling data, which is used for subsequent simulation and estimation of density loss. When simulating and estimating the density loss of weld trend distribution for the normal sampling data of the deviation regression pore structure based on the least squares estimation algorithm, the sampling data is mapped to the three-dimensional model of the weld. The ratio of the total volume of all sampled pores in each segment to the local bearing volume of the weld in this segment is calculated based on the weld trend segment. This ratio is defined as the density loss ratio. The least squares objective function is constructed with the pore size, pore number and connected path length as independent variables. The least squares regression fitting is performed to obtain the density loss prediction value of each weld segment. Regularization terms are added during the fitting process to prevent overfitting. In the fitting results, each weld segment contains an estimated density value and confidence interval. The output weld trend density loss estimation data is organized in the form of a one-dimensional array. Each element contains the spatial position index, density loss rate, fitting residual and the lower and upper limits of the confidence interval, which serves as the data input for subsequent brittle increment calculation and fracture risk analysis.

[0049] Step S24 includes the following steps: Step S241: performing stress instability concentration identification on the weld direction density loss estimation data to obtain the density loss stress instability concentration; Step S242: performing nonlinear simulation calculation of thermal plastic strain attenuation based on the density loss stress instability concentration and the weld direction density loss estimation data to obtain thermal plastic strain attenuation fitting data; Step S243: performing attenuation recursive median absolute deviation calculation on the thermal plastic strain attenuation fitting data to obtain the plastic recursive attenuation median absolute deviation; Step S244: performing attenuation decomposition constraint analysis on the thermal energy plastic strain attenuation fitting data according to the plastic recursive attenuation median absolute deviation and divide-and-conquer algorithm to generate thermal energy plastic strain attenuation decomposition data; Step S245: Calculate the weld distributed brittleness increment index according to the thermal plastic strain attenuation decomposition data to obtain the weld distributed brittleness increment index.

[0050] In an embodiment of the present invention, stress instability concentration is identified for the estimated data of density loss in the direction of the weld. First, it is necessary to construct a local stress gradient change function based on the stress distribution of the weld cross section, use the finite element method to numerically solve the stress state of different regions inside the weld, use the Von Mises criterion to calculate the equivalent stress distribution, select the maximum stress point in the weld area as the starting point for stress concentration calculation, set multiple monitoring nodes inside the weld, and each monitoring node is arranged along the normal direction of the weld curve. By calculating the local stress gradient change rate of each monitoring point, the range of the stress concentration area is judged, and for the area where the stress gradient changes drastically, the instability risk index is calculated, and the Gaussian mixture model (GMM) is used to classify and cluster the unstable areas, and the high stress concentration points on the direction of the weld are calibrated with different categories of stress instability areas, so as to obtain density loss stress instability concentration data. Based on the density loss stress instability concentration and the density loss estimation data of the weld direction, a nonlinear simulation calculation of thermal energy plastic strain attenuation is carried out. First, the temperature field data in the welding heat affected zone (HAZ) is extracted, and the temperature gradient change model is established using the heat conduction equation. Combined with the thermal expansion coefficient of the weld material, the thermal expansion strain is calculated, and the thermoplastic constitutive relationship is introduced to simulate the plastic strain evolution of the material under high temperature environment. The Johnson-Cook model is used to calculate the material yield strength at different temperatures. According to the nonlinear change law of the yield strength, the thermal energy plastic strain attenuation function is established. The finite element iterative solution method is used to fit the thermal energy plastic strain attenuation. The fitting residual is calculated using the least squares method, and the model parameters are adjusted to make the thermal energy plastic strain attenuation data converge to within the error range of the experimental measurement value, thereby obtaining the thermal energy plastic strain attenuation fitting data. The attenuation recursive median absolute deviation of the thermal plastic strain attenuation fitting data is extrapolated. First, the time series X(t) is defined to represent the plastic strain attenuation value at different time steps. The median M of the time series is calculated, the absolute deviations of all observation points to the median are calculated, and the median of these deviations is obtained. The median of the deviation is used as a measure of nonlinear attenuation. The recursive regression analysis method is further used to iteratively predict the changing trend of the median of the deviation in the time series. The prediction error is calculated each iteration, and the weight parameters of the recursive model are updated to minimize the prediction error. Finally, the median absolute deviation data of plastic recursive attenuation is obtained. According to the median absolute deviation of plastic recursive attenuation and the divide-and-conquer algorithm, the attenuation decomposition constraint analysis of the thermal energy plastic strain attenuation fitting data is performed. First, the thermal energy plastic strain attenuation data is Fourier transformed to extract the main attenuation components in the frequency domain, and the attenuation curve is decomposed into multiple independent attenuation modes. The divide-and-conquer algorithm is used to group different attenuation modes, and the attenuation contribution of each group is calculated. The boundary conditions of the attenuation grouping are adjusted based on the constrained optimization method to minimize the overall error of the attenuation grouping, and finally the thermal energy plastic strain attenuation decomposition data is generated.The weld distributed brittle incremental index is calculated according to the thermal plastic strain decay decomposition data. Firstly, the weld brittle incremental index is defined. Its calculation is based on the thermal strain rate, yield strength change rate and stress concentration in different areas of the weld. The principal component analysis (PCA) method is used to optimize the weights of different influencing factors. The three most important influencing factors are selected as the core variables for the brittle incremental calculation. The brittle incremental index calculation model is constructed. The gradient descent method is used to optimize the parameters of the calculation model to obtain the weld distributed brittle incremental index.

[0051] In another embodiment, first, the density loss values ​​of different areas of the weld are extracted from the density loss estimation data of the weld direction, and a grid data structure is established according to the coordinate distribution of the weld path, and a finite element grid of the weld area is constructed in a triangulated manner. The Von Mises yield criterion is used to calculate the equivalent stress in the grid unit, and the equivalent stress of each grid unit is converted into a plastic strain increment according to the stress-strain relationship curve of the welding material. The stress concentration position of the weld area is screened by a fracture instability judgment method based on the maximum principal stress criterion, and the instability degree of the stress concentration area is calculated. Finally, the Kriging interpolation method is used to interpolate and fit the stress instability data to generate the weld density loss stress instability concentration, and the result is stored as a structured data file. The density loss stress instability concentration and the estimated data of density loss along the weld direction are taken as input variables. The Johnson-Cook constitutive model is used to calculate the thermoplastic deformation behavior of welding materials under high temperature. The strain rate effect formula in continuous medium mechanics is used to calculate the attenuation trend of plastic strain with time. A nonlinear plastic strain attenuation function based on time and temperature is constructed. The weld area is divided into multiple calculation units according to the temperature gradient. The plastic strain attenuation function is numerically solved in each calculation unit. The Runge-Kutta method is used for integral calculation. Finally, the thermal energy plastic strain attenuation fitting data is generated, and the calculation results are stored as a data table in matrix format. The thermal energy plastic strain decay fitting data is screened, and after the abnormal data points are removed, the median absolute deviation (MAD) method is used to calculate the deviation distribution of the plastic strain decay data. First, the median of the data sample is calculated, and then the absolute deviation value between each data point and the median is calculated, and the median of these absolute deviation values ​​is calculated again to obtain the median absolute deviation of plastic recursive decay. Subsequently, the plastic decay data is recursively fitted by piecewise linear interpolation to generate a plastic recursive decay data set with smooth characteristics and store it in the data analysis platform. Based on the median absolute deviation of plastic recursive decay, the divide-and-conquer algorithm is used to perform regional analysis on the thermal energy plastic strain decay fitting data, and the weld area is divided into multiple independent calculation subdomains. Independent plastic strain decay calculations are performed in each subdomain. The thermal energy plastic strain decay curve is constructed using the Lagrange interpolation method, and the plastic strain decay gradients in different areas of the weld are calculated by the optimal substructure recursive solution method. Finally, the thermal energy plastic strain decay decomposition data is generated and stored as a hierarchical data structure for subsequent calculations.According to the thermal plastic strain attenuation decomposition data, the microstructure evolution model of the welding material is used to calculate the brittleness increment index of different areas of the weld. First, the grain size of the weld microstructure is statistically analyzed, and the Hall-Petch relationship is used to calculate the influence of grain size on the brittleness of the material. Then, the brittleness increment of the weld under thermal cycling is calculated in combination with the plastic deformation cumulative damage theory. The brittleness data of the weld area is reduced in dimension by the principal component analysis (PCA) method, and the least squares regression method is used to construct the weld brittleness increment prediction model. The weld distribution brittleness increment index is calculated, and the calculation formula is as follows: ,in Indicates the weld distribution brittleness increment index, is the number of mesh elements in the weld area, is the index, For the The yield strength of a grid cell, is the initial yield strength, For the The estimated grain size of each grid cell is It is the initial grain size estimate and the final calculated weld distribution brittleness increment index data.

[0052] Step S25 includes the following steps: Step S251: estimating the welding heat energy input loss of the weld direction surface state mapping data according to the weld direction density loss estimation data, and obtaining the weld direction heat energy input loss data; Step S252: performing internal residual stress analysis on the weld direction based on the heat energy input loss data on the weld direction to obtain internal residual stress data on the weld direction; Step S253: performing weld interface layer strain gradient differential calculation on weld surface state mapping data according to weld internal residual stress data and weld distribution brittleness increment index to obtain interface layer strain gradient differential data; Step S254: Deducing the critical extreme value interval of the weld according to the interface layer strain gradient difference data and the weld distribution brittleness increment index to obtain the critical extreme value interval of the weld; Step S255: Calculate the weld fracture risk based on the weld critical extreme value interval on the weld surface state mapping data to obtain weld fracture risk data.

[0053] In the embodiment of the present invention, the welding heat input loss is simulated and estimated based on the weld direction surface state mapping data according to the weld direction density loss estimation data. First, based on the heat source model of the welding process, a three-dimensional welding heat transfer finite element model is constructed, and the welding current, voltage and welding speed parameters are input into the model to calculate the heat input distribution under standard welding conditions. Based on the weld direction density loss estimation data, the position and distribution characteristics of the weld defect area are extracted, the geometric morphological parameters of the weld defect are input into the finite element model, the material thermophysical property parameter changes in the defect area are defined, the local absorption and thermal resistance effect of the defect on the welding heat input are simulated, the temperature field evolution along the direction of the weld is calculated by the non-steady-state heat conduction equation, and the local loss of welding heat energy is calculated based on the heat flux conservation principle. Adaptive meshing strategy is adopted to encrypt the mesh of the high temperature gradient area of ​​the weld defect area to improve the calculation accuracy, and the temperature field is solved by implicit time integration method to obtain the heat loss distribution data of each time step in the welding process. For weld defects of different orientations, the spatial variation of heat input loss rate is calculated, and the global heat energy loss data of the weld is fitted by interpolation method. A regression model of heat energy input loss in weld orientation is established. By comparing the loss data with the standard welding heat input, the overall heat energy input loss percentage of the weld is calculated, and finally the heat energy input loss data of the weld orientation is obtained. Based on the heat energy input loss data of the weld orientation, the internal residual stress of the weld orientation is analyzed. The welding thermoelastic-plastic finite element analysis method is used to numerically simulate the residual stress inside the weld, define the welding residual stress distribution function, calculate the stress gradient of each depth layer inside the weld, and evaluate the stress concentration in the high stress area of ​​the heat affected zone and the fusion zone. The inverse solution method is used to analyze the distribution trend of residual stress in the weld orientation. Through the fitting error analysis of the residual stress distribution curve, the boundary conditions are adjusted to obtain the internal residual stress data of the weld orientation. According to the internal residual stress data of the weld and the weld distributed brittle increment index, the weld interface layer strain gradient difference calculation is performed on the weld surface state mapping data. Firstly, the strain gradient distribution model of the weld interface layer is defined, the local strain monitoring points of the interface layer are set, the residual stress data are extracted, and the elastic strain distribution of the weld interface layer is calculated based on the Hooke's law. The nonlinear change characteristics of the plastic strain are calculated in combination with the weld distributed brittle increment index. The central difference method is used to calculate the strain gradient of the weld interface layer along the weld direction. The high-order difference method is used to calculate the rate of change of the strain gradient. The area where the strain gradient changes dramatically is identified, and finally the strain gradient difference data of the interface layer is obtained.The critical extreme value interval of the weld is deduced based on the interface layer strain gradient difference data and the weld distributed brittleness increment index. First, a critical strain model for weld fracture is constructed, and the interface layer strain gradient difference data is input into the critical strain model. The local plastic limit at different positions of the weld is calculated, and the high-risk area of ​​weld fracture is determined by the distribution of critical plastic strain. Combined with the weld distributed brittleness increment index, the brittleness evolution trend of different areas of the weld is analyzed. The strain distribution along the weld direction is partitioned by the segmented regression method, and the upper and lower boundaries of the weld critical extreme value interval are defined. The boundary parameters are adjusted by the numerical iteration method so that the critical condition of weld fracture meets the safety margin range of the material, and finally the weld critical extreme value interval is obtained. The weld fracture risk calculation is performed on the surface state mapping data of the weld trend based on the critical extreme value interval of the weld. Firstly, the weld fracture probability model is established, the fracture toughness parameters of the weld material are defined, the strain and stress data of the critical extreme value interval of the weld are extracted, and the fracture safety factor at different positions of the weld is calculated by the fracture mechanics method. For the high-risk points in the critical extreme value interval of the weld, the fracture failure probability is calculated, and the Monte Carlo method is used to numerically simulate the weld fracture probability, generate the weld fracture risk distribution curve, and calculate the overall weld fracture risk index by the extreme value statistical method, finally obtaining the weld fracture risk data.

[0054] Step S3 includes the following steps: Step S31: normalizing the weld fracture risk data to obtain weld fracture risk normalized data; Step S32: matching the welding heat output constraint control of the robot arm between different weld direction surface states with the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data to obtain the welding heat output constraint control data.

[0055] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: normalizing the weld fracture risk data to obtain weld fracture risk normalized data; In an embodiment of the present invention, the weld fracture risk data is normalized. First, according to the numerical distribution of the weld fracture risk data, the maximum value and the minimum value are selected as the normalization boundaries, and the minimum-maximum normalization method is used to convert the weld fracture risk data, and all the data are mapped to the interval 0, 1. For the fracture risk data of different weld directions, the mean and standard deviation are calculated respectively, and the Z-score standardization method is used to adjust the data scale so that the mean of all data is zero and the standard deviation is one, thereby eliminating the numerical deviation between different weld directions and improving the comparability of the data. In order to prevent outliers from affecting the normalization results, the box plot analysis method is used to detect outliers, calculate the quartile range, eliminate abnormal fracture risk data points that exceed the normal distribution range, and recalculate the normalization parameters to ensure that the normalized data is evenly distributed, and finally obtain the normalized data of weld fracture risk.

[0056] Step S32: matching the welding heat output constraint control of the robot arm between different weld direction surface states with the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data to obtain the welding heat output constraint control data.

[0057] In the embodiment of the present invention, the welding heat output constraint control matching of the manipulator between different weld direction surface states is performed on the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data. First, a welding heat output control database is constructed, which includes the standard heat input range under different weld directions, and the heat energy compensation value required for different weld areas is calculated in combination with the weld direction density loss estimation data. According to the weld fracture risk normalization data, the risk weight of each weld section is calculated, and the welding heat input target value of each weld direction is calculated using the weighted average method. The dynamic constraint optimization algorithm is used to match and control the welding heat input parameters of the manipulator, and the objective function is constructed. The minimization of the deviation of the welding heat input is taken as the optimization target, and the weld surface state mapping data is introduced as the constraint condition to calculate the heat input adjustment coefficient between different weld directions. The adaptive step size adjustment strategy is adopted to dynamically adjust the welding current, voltage and moving speed of the manipulator, so that the welding heat input gradually approaches the target value, and the heat input power is adjusted in combination with the real-time temperature monitoring feedback, and finally the welding heat output constraint control data is obtained.

[0058] Step S32 includes the following steps: Step S321: analyzing the welding heat energy limit values ​​between different weld direction surface states on the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data, and obtaining the welding heat energy limit values ​​between different weld direction surface states; Step S322: performing robot arm contact radius matching based on welding heat energy limit values ​​between different weld seam surface states to obtain robot arm contact radius matching data; Step S323: performing robot arm welding force matching according to the welding heat energy limit value and the robot arm contact radius matching data to obtain robot arm welding force matching data; Step S324: Based on the robot arm contact radius matching data, the robot arm welding force matching data and the welding heat energy limit value, the robot arm welding heat output constraint control matching between different weld direction surface states is performed to obtain the welding heat output constraint control data.

[0059] In an embodiment of the present invention, according to the normalized data of weld fracture risk and the estimated data of weld strike density loss, the weld strike surface state mapping data is analyzed for the welding heat energy limit value between different weld strike surface states. First, the normalized data of weld fracture risk is partitioned and calculated, and multiple sub-intervals are divided based on the weld strike as the classification standard, and the mean and variance of the normalized data of weld fracture risk in each sub-interval are calculated. Combined with the estimated data of weld strike density loss, the Lagrange interpolation method is used to calculate the density loss change trend of the weld along the strike direction, and the numerical integration method is used to calculate the cumulative density loss of each weld area. Based on the density loss data of different weld areas and the normalized data of weld fracture risk, a welding heat energy input demand equation is constructed, the welding penetration threshold is set, and the welding heat energy limit value is calculated in combination with the heat input-penetration response curve to obtain the welding heat energy limit value between different weld strike surface states. The contact radius of the manipulator is matched based on the welding heat energy limit between the surface states of different weld directions. First, the maximum welding temperature rise of the weld area is calculated according to the welding heat energy limit, and the heat diffusion radius of the weld area is calculated using the Fourier heat conduction law. The thermal field of the weld area is numerically simulated using the heat conduction simulation software to determine the spatial range of heat diffusion. Combined with the geometric dimensions and welding angle of the welding gun at the end of the manipulator, the contact area of ​​the welding gun at the end of the manipulator is calculated in different weld directions, and the optimal contact radius of the manipulator is calculated according to the expansion radius of the welding molten pool. The contact radius of the manipulator is iteratively optimized using the gradient descent method to match the welding area determined by the welding heat energy limit, and finally the contact radius matching data of the manipulator is obtained. The welding force matching of the manipulator is carried out according to the welding heat energy limit and the contact radius matching data of the manipulator. First, the surface tension of the molten metal in the welding area is calculated using the welding heat energy limit, and the welding pressure distribution applied by the manipulator is determined in combination with the contact radius matching data of the manipulator. The force condition of the welding contact area is simulated using the finite element method, the force uniformity of different areas of the weld is calculated, and the welding pressure parameters of the manipulator are optimized using the stress distribution analysis method. Combined with the curvature change of the weld direction, the torque of the robot arm during welding is calculated, and the Newton-Raphson method is used to optimize the force matching parameters of the robot arm to meet the stability requirements of the welding pool, and finally the robot arm welding force matching data is obtained. Based on the robot arm contact radius matching data, the robot arm welding force matching data and the welding heat energy limit value, the robot arm welding heat output constraint control matching between different weld direction surface states is carried out. First, the welding heat output control matrix is ​​constructed, with the welding heat energy limit value as the target variable, the robot arm contact radius matching data and the robot arm welding force matching data as the constraint conditions, and the quadratic programming optimization algorithm is used to calculate the welding heat input optimization coefficient of different weld directions.Combined with the real-time temperature monitoring data during the welding process, the Kalman filtering method is used to dynamically adjust the welding heat input, and a feedback control system is introduced to optimize the welding current, voltage and moving speed of the robot arm in real time, so that the welding heat input is in line with the control range of the welding heat energy limit value, and finally the welding heat output constraint control data is obtained.

[0060] Step S4 includes the following steps: Step S41: performing logic learning on the welding heat output constraint control data to obtain welding heat output constraint learning data; Step S42: constructing a free curve weld intelligent welding tracking model for the welding heat output constraint learning data based on a random forest algorithm to obtain a weld intelligent welding tracking model; Step S43: Send the weld intelligent welding tracking model to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0061] In the embodiment of the present invention, during the welding process, the welding heat input directly affects the forming quality and microstructure of the weld, so it is necessary to perform logical learning on the welding heat output constraint control data to optimize the heat input parameters. First, the welding heat input, arc voltage, current, welding speed, wire feeding speed and other parameters are extracted from the welding data acquisition system, and the sampling frequency is set to 1000 Hz to ensure the continuity and high resolution of the data. Secondly, based on the welding process constraints, the data is normalized so that the numerical range of each parameter is standardized to the [0,1] interval for subsequent modeling. Then, the time series analysis method is used to calculate the correlation between different welding parameters, and the parameter pairs with a Pearson correlation coefficient greater than 0.7 are selected as the main research objects. For example, it is found in the experiment that the correlation between welding current and welding heat input reaches 0.85, so it is necessary to focus on analyzing the influence of current fluctuation on heat input stability. Then, the welding heat input constraints are learned using a logistic regression model, and the loss function is set to the cross entropy loss to minimize the prediction error. Through iterative optimization, the optimal heat input control data under different welding parameters are obtained. Finally, the output welding heat output constraint learning data includes welding process parameter range, welding heat input optimization value and welding stability evaluation index, which provides basic data for the construction of subsequent welding tracking model. Based on the welding heat output constraint learning data, the random forest algorithm is used to construct the free curve weld intelligent welding tracking model. First, the training data set is constructed using the welding heat input, arc voltage, current, welding speed, wire feeding speed and other data extracted in the previous step. The data set size is set to 50,000 groups to ensure the generalization ability of the model. All data are divided into 80% training and 20% testing, and the data are normalized so that all eigenvalues ​​are normalized to the [0,1] interval to improve the training efficiency and stability of the model. Then, a random forest regression model is constructed, the number of decision trees is set to 100, the maximum depth is 10 to prevent overfitting, and the mean square error (MSE) is used as the loss function. The model prediction accuracy is optimized through the bagging ensemble learning method. During the training process, the model calculated the influence of welding heat input, current, welding speed and other parameters on the quality of weld formation through multiple iterations, and established the mapping relationship between input parameters and weld formation based on historical welding data. After the training was completed, the model performance was evaluated using the test data set, and the root mean square error (RMSE) was calculated as the model accuracy evaluation index. The RMSE obtained by experimental testing was 0.015 mm, indicating that the model can accurately predict the influence of welding parameters on weld quality. In the model optimization stage, the grid search method was used to adjust the hyperparameters, and the prediction accuracy of the model was further improved by tuning the parameters such as the number of decision trees, the maximum depth, and the minimum number of leaf node samples.In the final model testing stage, the welding process parameters are input and the weld trajectory tracking adjustment data is output to ensure that the model can predict the offset of the weld trajectory, welding speed adjustment value and other key control parameters in real time during the welding process, and provide high-precision path planning data for the subsequent intelligent tracking control of the weld. After the weld intelligent welding tracking model is built, it needs to be deployed to the cloud platform to realize the intelligent tracking welding control of the free curve weld by the robot arm. First, the trained model is converted into the TensorFlow Serving format to support real-time inference on the cloud. Then, the MQTT protocol is used to establish a communication channel between the model and the robot control system, and the message transmission rate is set to 50 Hz to ensure the real-time data transmission. Then, the RESTful API service is deployed on the cloud to support remote access to the welding path prediction data. The cloud server uses NVIDIAA100 GPU for inference calculation, and the single inference time is controlled within 10 ms to meet the real-time control requirements of the welding process. On the robot end, the welding control system requests the cloud prediction data once a second, and dynamically adjusts the motion trajectory of the welding gun according to the parameters such as the weld offset and welding speed adjustment value output by the model. Finally, the welding path correction based on the intelligent tracking model is realized, so that the robot arm can accurately track the welding trajectory of the free curve weld, improving the welding quality and weld consistency.

[0062] The present invention also provides an intelligent tracking system for a free-curve weld by a robot arm, which is used to execute the intelligent tracking method for a free-curve weld by a robot arm as described above. The intelligent tracking system for a free-curve weld by a robot arm comprises: The weld surface state mapping module is used to deploy a laser profiler on the robot arm, and perform a curved weld profile scan on the free-curve weld to obtain curved weld profile data; perform weld surface state analysis on the curved weld profile data to obtain weld surface state data; perform geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; The weld fracture risk calculation module is used to analyze the weld porosity distribution trend based on the weld surface state mapping data to obtain the weld porosity distribution trend data; simulate and estimate the weld density loss based on the weld porosity distribution trend data to obtain the weld density loss estimation data; perform weld fracture risk calculation based on the weld density loss estimation data to obtain the weld fracture risk data; A welding heat output constraint control module is used to match the welding heat output constraint control of the robot arm between different weld direction surface states according to the weld fracture risk data and the weld direction surface state mapping data, so as to obtain the welding heat output constraint control data; The intelligent welding tracking model building module is used to build a free curve weld intelligent welding tracking model based on the welding heat output constraint control data based on the random forest algorithm to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

[0063] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent tracking method for a free-curve weld by a robotic arm, characterized in that: The following steps are involved: Step S1: deploying a laser profiler on the robot arm, and scanning the free-curve weld profile to obtain the curve weld profile data; Perform weld surface state analysis on the curved weld profile data to obtain weld surface state data; perform geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; Step S2: analyzing the weld porosity distribution trend of the weld surface state mapping data to obtain weld porosity distribution trend data; simulating and estimating the weld density loss of the weld direction distribution based on the weld porosity distribution trend data to obtain weld density loss estimation data; calculating the weld fracture risk based on the weld density loss estimation data to obtain weld fracture risk data; Step S3: matching the welding heat output constraint control of the robot arm between different weld surface states with the weld surface state mapping data according to the weld fracture risk data to obtain welding heat output constraint control data; Step S4: Based on the random forest algorithm, the free curve weld intelligent welding tracking model is constructed for the welding heat output constraint control data to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

2. The intelligent tracking method for free curve welds by a robot arm according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: deploying a laser profiler on the robot arm, and scanning the free-curve weld profile to obtain the curve weld profile data; Step S12: cleaning the curved weld profile data to obtain the curved weld profile cleaning data; Step S13: performing a curve weld geometric feature analysis on the curve weld profile cleaning data to generate curve weld geometric data; Step S14: performing weld surface state analysis on the curved weld profile cleaning data to obtain weld surface state data; Step S15: performing geometric trend weld surface state mapping processing on the weld surface state data according to the curved weld geometry data to obtain weld trend surface state mapping data.

3. The intelligent tracking method for free curve welds by a robot arm according to claim 2 is characterized in that: Step S2 includes the following steps: Step S21: analyzing the weld pore distribution trend on the weld surface state mapping data to obtain weld pore distribution trend data; Step S22: analyzing the trend change of the weld width / height on the curved weld geometry data to obtain the trend change data of the weld width / height; Step S23: performing a simulation estimation of the weld density loss of the weld direction distribution based on the weld porosity distribution direction data and the weld width / height trend change data to obtain weld density loss estimation data; Step S24: calculating the weld distribution brittleness increment index for the weld direction density loss estimation data to obtain the weld distribution brittleness increment index; Step S25: Calculate the weld fracture risk on the weld trend surface state mapping data according to the weld trend density loss estimation data and the weld distribution brittleness increment index to obtain the weld fracture risk data.

4. The intelligent tracking method for free curve welds by a robot arm according to claim 3 is characterized in that: Step S23 includes the following steps: Step S231: performing direction distribution density identification on the weld pore distribution direction data to obtain the weld pore distribution density; performing pore chain linear distribution identification on the weld pore distribution direction data based on the weld pore distribution density to obtain pore chain linear distribution data; Step S232: calculating the effective bearing volume mean difference of the weld width / height trend change data to obtain the effective bearing volume mean difference of the weld; Step S233: performing pore connectivity analysis on the pore chain linear distribution data to obtain pore distribution pore connectivity data; Step S234: performing deviation regression pore structure normal distribution sampling on the pore chain linear distribution data according to the mean difference of the effective bearing volume of the weld and the pore connectivity data of the pore distribution to obtain deviation regression pore structure normal sampling data; Step S235: Based on the least squares estimation algorithm, the deviation regression pore structure normal sampling data is used to simulate and estimate the welding density loss of the weld trend distribution to obtain the weld trend density loss estimation data.

5. The intelligent tracking method for free curve welds by a robot arm according to claim 3 is characterized in that: Step S24 includes the following steps: Step S241: performing stress instability concentration identification on the weld direction density loss estimation data to obtain the density loss stress instability concentration; Step S242: performing nonlinear simulation calculation of thermal plastic strain attenuation based on the density loss stress instability concentration and the weld direction density loss estimation data to obtain thermal plastic strain attenuation fitting data; Step S243: performing attenuation recursive median absolute deviation calculation on the thermal plastic strain attenuation fitting data to obtain the plastic recursive attenuation median absolute deviation; Step S244: performing attenuation decomposition constraint analysis on the thermal energy plastic strain attenuation fitting data according to the plastic recursive attenuation median absolute deviation and divide-and-conquer algorithm to generate thermal energy plastic strain attenuation decomposition data; Step S245: Calculate the weld distributed brittleness increment index according to the thermal plastic strain attenuation decomposition data to obtain the weld distributed brittleness increment index.

6. The intelligent tracking method for free curve welds by a robot arm according to claim 3 is characterized in that: Step S3 includes the following steps: Step S31: normalizing the weld fracture risk data to obtain weld fracture risk normalized data; Step S32: matching the welding heat output constraint control of the robot arm between different weld direction surface states with the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data to obtain the welding heat output constraint control data.

7. The intelligent tracking method for free-curve welds by a robot arm according to claim 6, characterized in that: Step S32 includes the following steps: Step S321: analyzing the welding heat energy limit values ​​between different weld direction surface states on the weld direction surface state mapping data according to the weld fracture risk normalization data and the weld direction density loss estimation data, and obtaining the welding heat energy limit values ​​between different weld direction surface states; Step S322: performing robot arm contact radius matching based on welding heat energy limit values ​​between different weld seam surface states to obtain robot arm contact radius matching data; Step S323: performing robot arm welding force matching according to the welding heat energy limit value and the robot arm contact radius matching data to obtain robot arm welding force matching data; Step S324: Based on the robot arm contact radius matching data, the robot arm welding force matching data and the welding heat energy limit value, the robot arm welding heat output constraint control matching between different weld direction surface states is performed to obtain the welding heat output constraint control data.

8. The intelligent tracking method for free curve welds by a robot arm according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing logic learning on the welding heat output constraint control data to obtain welding heat output constraint learning data; Step S42: constructing a free curve weld intelligent welding tracking model for the welding heat output constraint learning data based on a random forest algorithm to obtain a weld intelligent welding tracking model; Step S43: Send the weld intelligent welding tracking model to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

9. An intelligent tracking system for a free-curve weld by a robotic arm, characterized in that: The method for intelligently tracking a free-curve weld by a robot arm according to claim 1 comprises: The weld surface state mapping module is used to deploy a laser profiler on the robot arm, and perform a curved weld profile scan on the free-curve weld to obtain curved weld profile data; perform weld surface state analysis on the curved weld profile data to obtain weld surface state data; perform geometric trend weld surface state mapping processing on the weld surface state data to obtain weld trend surface state mapping data; The weld fracture risk calculation module is used to analyze the weld porosity distribution trend based on the weld surface state mapping data to obtain the weld porosity distribution trend data; simulate and estimate the weld density loss based on the weld porosity distribution trend data to obtain the weld density loss estimation data; perform weld fracture risk calculation based on the weld density loss estimation data to obtain the weld fracture risk data; A welding heat output constraint control module is used to match the welding heat output constraint control of the robot arm between different weld direction surface states according to the weld fracture risk data and the weld direction surface state mapping data, so as to obtain the welding heat output constraint control data; The intelligent welding tracking model building module is used to build a free curve weld intelligent welding tracking model based on the welding heat output constraint control data based on the random forest algorithm to obtain the weld intelligent welding tracking model; the weld intelligent welding tracking model is sent to the cloud platform to execute the intelligent tracking welding control of the free curve weld by the robot arm.

Images