A welding method for large oil storage tank pontoon plate

Abstract

The present invention provides a method for welding large-scale oil storage tank pontoon plates, which belongs to the field of electrical digital data processing technology, and comprises the following steps: first, collecting three-dimensional point cloud data through a sensor group on a crawler welding robot, and simultaneously acquiring a characteristic image of the area to be welded with a camera. Based on these data, a welding coordinate system is established and the groove profile is calculated to generate an initial welding trajectory. In combination with welding process parameters and environmental conditions, a group of Euler angle timing sequence calculation equations is established to adjust the welding head posture in real time. The groove formation data during the welding process is monitored in real time by sensors. When the joint penetration or excess height does not meet the requirements, the system will automatically recalculate the Euler angle sequence and adjust the welding parameters until the process standards are met, thereby achieving high-quality automated welding. This solves the technical problem that the existing technology is difficult to achieve in measuring the welding process and automatically adjusting the welding parameters and the motion trajectory of the welding robot according to the real-time perception of the welding status.

Description

Technical Field

[0001] The invention belongs to the technical field of electric digital data processing, and in particular relates to a welding method for a large-scale oil storage tank pontoon plate. Background Art

[0002] As vital equipment in the energy industry, the manufacturing and maintenance of large oil storage tanks has always been a hot topic within the industry. Welding is a core process in the manufacturing process. Because the tank shells are constructed of thick steel plates and feature complex and varied structures, the welding quality must meet stringent safety standards, presenting significant challenges.

[0003] To address this problem, the prior art mainly adopts the following welding methods:

[0004] 1. Manual welding: This method relies on the welder's experience and proficiency. It is inexpensive but has poor reliability, difficulty in ensuring weld quality, and low efficiency, making it difficult to adapt to the welding needs of large structural parts.

[0005] 2. Automatic welding: Using welding robots to automatically complete the entire welding process can improve efficiency and quality consistency. However, for complex shapes and changing welding environments, robots find it difficult to adapt autonomously and require a lot of offline programming and debugging work, which limits efficiency.

[0006] 3. Semi-automatic welding: Using robots to assist with manual operation, and using machine vision, force sensing and other technologies to enhance the intelligence level of manual welding, it can adapt to complex environments to a certain extent, but it still requires a lot of manual participation, and it is difficult to fully realize adaptive welding.

[0007] The above-mentioned existing technologies generally have the following problems when facing the welding requirements of large and complex structural parts:

[0008] 1. It is difficult to perceive and adapt to the complex welding environment in real time, which makes it difficult to control the quality of the weld and ensure the consistency and stability of the welding quality.

[0009] 2. Lack of intelligent welding parameter optimization and trajectory planning capabilities, unable to autonomously adjust the welding process according to real-time changes in welding status, limiting the adaptive welding capability.

[0010] 3. Due to the limitations of the robot's own kinematic and dynamic performance, it is difficult to effectively compensate for various dynamic errors in the welding process, which affects the reliability of the weld quality.

[0011] Therefore, the existing technology has the technical problem of being difficult to realize the technical problem of automatically adjusting the welding parameters and the motion trajectory of the welding robot according to the real-time perception of the welding status during the welding process. Summary of the Invention

[0012] In view of this, the present invention provides a large oil storage tank floating plate welding method, which can solve the technical problem that the existing technology is difficult to achieve in measuring the welding process and automatically adjusting the welding parameters and the motion trajectory of the welding robot according to the real-time perception of the welding status.

[0013] The present invention is achieved in that:

[0014] The present invention provides a large-scale oil storage tank pontoon plate welding method, comprising the following steps:

[0015] S10, collecting three-dimensional point cloud data of the area to be welded of the floating plate by a sensor group provided on the crawler welding robot; the sensor group includes a laser line scanning sensor, a structured light sensor, an inertial measurement unit and a torque sensor;

[0016] S20, using a camera on the welding robot to obtain an image of the area to be welded, and preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded;

[0017] S30, establishing a welding coordinate system based on the three-dimensional point cloud data and the characteristic image of the area to be welded, and calculating the three-dimensional contour of the groove;

[0018] S40, calculating the initial Euler angle of the welding head according to the three-dimensional contour of the groove, and generating a motion trajectory of the welding head;

[0019] S50, using the welding current, voltage, wire feed speed, shielding gas flow rate and welding speed in the welding process parameter database, combined with the initial Euler angle, groove geometry, plate thickness and ambient temperature, to establish an Euler angle timing sequence calculation equation group;

[0020] S60, obtaining a posture adjustment sequence of the welding head during the welding process by calculating the Euler angle timing sequence equation group, and generating a welding trajectory compensation amount according to the posture adjustment sequence;

[0021] S70, adjusting the motion trajectory of the welding head in real time based on the welding trajectory compensation amount, and collecting groove formation data during the welding process through the sensor group;

[0022] S80, calculating the joint penetration and the reinforcement using the groove forming data. When the joint penetration is less than 90% of the plate thickness or the reinforcement is greater than a preset threshold, returning to step S50 to recalculate the Euler angle timing sequence until the joint penetration and the reinforcement meet the process requirements.

[0023] On the basis of the above technical solution, the welding method of a large oil storage tank pontoon plate of the present invention can also be improved as follows:

[0024] The Euler angle timing sequence calculation equation group includes a posture prediction equation, a trajectory planning equation, a dynamic compensation equation, and a welding process mapping equation.

[0025] Furthermore, the posture prediction equation is used to predict the posture change trend of the welding head during the movement process. The input includes the initial Euler angle, the three-dimensional contour data of the groove and the movement speed of the welding head. The output is the predicted posture sequence of the welding head.

[0026] The trajectory planning equation is used to generate a smooth and continuous welding trajectory. The input includes the predicted posture sequence, groove geometry and welding process parameters, and the output is the reference motion trajectory of the welding head.

[0027] The dynamic compensation equation is used to compensate for the inertia effect of the welding robot during movement. The input includes the reference motion trajectory, the welding robot mass parameters and the motion speed, and the output is the compensated motion trajectory.

[0028] The welding process mapping equation is used to establish a mapping relationship between welding process parameters and molten pool behavior. The input includes welding current, voltage, wire feed speed, shielding gas flow rate and plate thickness, and the output is the expected penetration depth and excess height.

[0029] Furthermore, the preset threshold is specifically that when the thickness of the plate is not greater than 25 mm, the excess height is 3 mm, and when the thickness of the plate is greater than 25 mm, the excess height is 4 mm.

[0030] Furthermore, the step of preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded includes an image preprocessing step and an image feature extraction step.

[0031] Furthermore, the image preprocessing step specifically includes:

[0032] S21, using a Gaussian filter to perform noise reduction processing on the image of the welding area;

[0033] S22, grayscale processing is performed on the denoised image;

[0034] S23, using a histogram equalization method to improve image contrast;

[0035] S24, using morphological operations to perform edge enhancement on the image;

[0036] S25. Obtain a binary image of the area to be welded through adaptive binarization processing.

[0037] Furthermore, the feature extraction step is specifically performed using a preset welding image feature extraction model, wherein the welding image feature extraction model includes an initial feature extraction module, a mathematical model module, and an accurate feature extraction module.

[0038] Furthermore, the initial feature extraction module is used to extract the initial contour features of the groove area from the binary image. The input is the preprocessed binary image and the preset groove type parameters. The output is the initial contour feature vector of the area to be welded and the grayscale distribution matrix of the groove area. A deep convolutional neural network structure based on region growing is adopted.

[0039] The mathematical model module is a matrix equation group, including a contour optimization matrix equation, a groove characteristic matrix equation and a geometric constraint matrix equation;

[0040] The precise feature extraction module is used to generate the final feature data of the area to be welded. The input is the optimized contour feature vector, the geometric parameters of the groove and the rationality judgment result. The output is the precise feature description data of the area to be welded; the residual attention network structure is adopted.

[0041] Furthermore, the profile optimization matrix equation is used to correct and optimize the initial profile features, the input is the initial profile feature vector and the preset groove standard parameters, and the output is the optimized profile feature vector;

[0042] The groove characteristic matrix equation is used to calculate the key geometric parameters of the groove. The input is the optimized contour feature vector and the grayscale distribution matrix of the groove area. The output is the root gap, groove angle and groove depth data of the groove.

[0043] The geometric constraint matrix equation is used to verify the validity of the groove feature. The input is the root gap, groove angle and groove depth data of the groove, and the output is the rationality judgment result of the groove feature.

[0044] Furthermore, the step of establishing a training data set for the welding image feature extraction model specifically includes:

[0045] Step 1: Collect 5,000 sets of floating plate welding images under different working conditions, including different plate thicknesses, different welding currents, different welding voltages, different ambient temperatures, and different welding speeds;

[0046] Step 2: annotate the pontoon plate welding image, wherein the annotation contents include the groove contour line, the groove root position, the weld edge position and the molten pool boundary position;

[0047] Step 3: Expand the training samples using data enhancement methods, including image rotation, image scaling, brightness adjustment, contrast adjustment, and noise addition;

[0048] Step 4: Divide the training samples into a training set, a validation set, and a test set in a ratio of 8:1:1;

[0049] Step 5: Normalize the images in the training set, the validation set, and the test set, and adjust the image size to 1024 by 1024 pixels.

[0050] The step of training the welding image feature extraction model specifically includes:

[0051] Step 11: Initialize the network parameters of the deep convolutional neural network and the residual attention network;

[0052] Step 12: The deep convolutional neural network is trained using batch stochastic gradient descent, with a learning rate of 0.001, a number of training rounds of 200 rounds, and a batch size of 32.

[0053] Step 13: Use cross-validation to evaluate the performance of the deep convolutional neural network, and perform early stopping when the accuracy of the validation set does not improve for five consecutive rounds;

[0054] Step 14: Using the output of the trained deep convolutional neural network as the input of the residual attention network;

[0055] Step 15: Adopting the adaptive learning rate optimization algorithm to train the residual attention network, the initial learning rate is set to 0.0005, and the number of training rounds is set to 100 rounds;

[0056] Step 16: Evaluate the performance of the trained model on the test set. When the groove feature recognition accuracy reaches above 95% and the average recognition time is less than 100 milliseconds, the model training is completed.

[0057] Step 17: Save the trained parameters of the deep convolutional neural network and the residual attention network as the welding image feature extraction model.

[0058] The relevant equations are described in detail below:

[0059] 1. Contour optimization matrix equation:

[0060] The contour optimization matrix equation is specifically expressed as follows:

[0061] ;

[0062] Where, is the optimized contour feature vector ; is the initial contour feature vector ; To optimize the weight matrix ; is the contour correction matrix ; is the contour error vector ; is the standard contour vector ; is the adjustment coefficient, ; is the feature vector dimension; is the error vector dimension.

[0063] Parameter acquisition method:

[0064] 1. Obtained through the output of the initial feature extraction module;

[0065] 2. Calculate by the following steps:

[0066] ;

[0067] Where, is a feature point and The Euclidean distance between is the Gaussian kernel parameter, the default value is 0.1;

[0068] 3. and Calculated by comparing with the standard contour template:

[0069] ;

[0070] Where, is the coordinate of the current contour point; are the coordinates of the corresponding standard template points.

[0071] The equation adopts the form of weighted combination and is mainly based on the following principles:

[0072] (1) Weight matrix In exponential form ( ) is because it is necessary to reflect the rapid decay characteristics of spatial correlation. The farther the distance, the smaller the impact, which is consistent with the characteristics of the Gaussian kernel.

[0073] (2) Error correction term The linear combination form is used to maintain the controllability and stability of the correction;

[0074] (3) Standard items As a benchmark constraint, it ensures that the optimization results do not deviate too much from the standard contour;

[0075] Compared with the existing technology, this equation introduces adaptive weights and multi-level constraints, which improves the robustness of contour recognition.

[0076] 2. Groove characteristic matrix equation:

[0077] The groove characteristic matrix equation is specifically expressed as follows:

[0078] ;

[0079] Where, For the root gap; is the groove angle; is the groove depth; is the feature map matrix ; is the optimized contour feature vector ; is the grayscale influence matrix ; is the grayscale distribution matrix ; is the grayscale weight vector ; is the error vector .

[0080] Parameter acquisition method:

[0081] 1. Obtained by least squares fitting:

[0082] ;

[0083] Where, is the measured groove parameter matrix.

[0084] 2. and Obtained through grayscale gradient analysis:

[0085] ;

[0086] Where, For the characteristic parameters; For the Grayscale features.

[0087] The equation takes the form of a linear mapping plus a nonlinear correction:

[0088] (1) Linear term It reflects the basic correspondence between contour features and geometric parameters;

[0089] (2) Grayscale impact item The matrix multiplication form is adopted to reflect the modulation effect of grayscale distribution on feature extraction;

[0090] (3) Through partial derivatives Constructing the grayscale influence matrix to reflect the sensitivity relationship between parameters;

[0091] This equation innovatively introduces image grayscale information into the feature extraction process, improving the accuracy of parameter recognition.

[0092] 3. Geometric constraint matrix equation:

[0093] The geometric constraint matrix equation is specifically expressed as follows:

[0094] ;

[0095] Where, is the rationality judgment result vector ; is the constraint coefficient matrix ; is the groove parameter vector ; is the dynamic constraint matrix ; is the correction vector .

[0096] Parameter acquisition method:

[0097] and It is obtained through experimental calibration, the steps are as follows:

[0098] 1. Use standard specimens for welding tests;

[0099] 2. Record the welding quality under different parameter combinations;

[0100] 3. Establish constraint relationships through multiple regression analysis.

[0101] This equation is constructed based on the theory of dynamic systems:

[0102] (1) Static constraint items Reflects the steady-state relationship between parameters;

[0103] (2) Dynamic constraints Introducing time derivatives to consider the impact of parameter change rate;

[0104] (3) Correction vector Used to compensate for system errors and nonlinear effects;

[0105] This equation breaks through the limitations of traditional static constraints and can better adapt to the dynamic welding process.

[0106] 4.Euler angle timing sequence calculation equations:

[0107] The posture prediction equation is specifically expressed as follows:

[0108] ;

[0109] Where, is the current Euler angle vector; is the welding head velocity vector; is the speed influence matrix; is the acceleration influence matrix; is a random disturbance term.

[0110] The posture prediction equation adopts the idea of ​​Taylor expansion:

[0111] (1) Speed ​​item Reflects the first-order change trend;

[0112] (2) Acceleration term Consider the second-order dynamic characteristics;

[0113] (3) Random disturbance term Used to model system noise and uncertainty.

[0114] The trajectory planning equation is specifically expressed as follows:

[0115] ;

[0116] Where, To plan the trajectory; is the control point; for Sub-B-spline basis function; is the smoothing term; is the smoothing coefficient.

[0117] The trajectory planning equation is based on B-spline theory:

[0118] (1) Basis functions Ensure the continuity and smoothness of the trajectory;

[0119] (2) Smoothness term Used to suppress trajectory oscillation.

[0120] The dynamic compensation equation is specifically expressed as follows:

[0121] ;

[0122] Where, is the compensation amount; is the inertia matrix; is the joint angle vector; is the Coriolis moment matrix; is the gravity matrix.

[0123] The dynamic compensation equation is based on the robot dynamic model:

[0124] (1) Inertia term Reflects the inertial force generated by accelerated motion;

[0125] (2) Coriolis force term Consider joint coupling effects;

[0126] (3) Gravity term Compensate for gravity effects.

[0127] The welding process mapping equation is specifically expressed as follows:

[0128] ;

[0129] Where, is the penetration depth; For Yu Gao; is the welding current; is the voltage; is the wire feeding speed; is the gas flow rate; is the plate thickness; is the process coefficient; is the error term.

[0130] This equation uses matrix form to express the multi-parameter coupling relationship:

[0131] (1) Process coefficient matrix Reflects the influence weight of different parameters on penetration depth and reinforcement height;

[0132] (2) Error term Account for model errors and random disturbances.

[0133] 5. Real-time adjustment of trajectory compensation:

[0134] The real-time trajectory adjustment equation is specifically expressed as follows:

[0135] ;

[0136] Where, is the trajectory after compensation; is the original trajectory; is the compensation amount; is the compensation coefficient.

[0137] This equation is built based on the PID control principle:

[0138] (1) Proportional term Ensure rapid response;

[0139] (2) Integral item Eliminate steady-state errors;

[0140] (3) Differential term Provides predictive capabilities.

[0141] 6. Calculation of joint quality:

[0142] The penetration calculation equation is specifically expressed as follows:

[0143] ;

[0144] Where, is the penetration distribution; is the metal density distribution; is the depth weight function; is the plate thickness.

[0145] The penetration depth is calculated in integral form:

[0146] (1) Density distribution Reflects the melting state of the material;

[0147] (2) Weight function Consider the energy attenuation in the depth direction.

[0148] The coheight calculation equation is specifically expressed as follows:

[0149] ;

[0150] Where, is the residual height distribution; is the weld surface height; is the surface height of the parent material; is the weld width.

[0151] The residual height is calculated using the maximum value operation:

[0152] (1) Pass Calculate and extract the weld bulge height;

[0153] (2) Subtract the base height Get the actual coheight.

[0154] Compared to existing technologies, the present invention provides a method for welding large oil storage tank pontoon plates. This method first installs multiple sensors, including a laser scanner, structured light sensor, and inertial measurement unit, on a welding robot to collect real-time 3D point cloud data and image features of the weld area. Based on this high-precision 3D and 2D information, a welding coordinate system is established, and key geometric features, such as the groove profile and shape, are captured.

[0155] Next, an adaptive trajectory planning and welding parameter optimization algorithm is used, combining the groove geometry and welding process parameters to predict the posture changes of the welding head during movement and generate a smooth and continuous welding trajectory. Furthermore, a dynamic compensation algorithm is used to eliminate trajectory deviations caused by factors such as inertia, taking into account the dynamic characteristics of the welding robot itself.

[0156] During the welding process, the sensor system monitors the weld formation in real time, including characteristic parameters such as penetration depth and excess height. If a deviation in weld quality is detected, the method automatically adjusts process parameters such as welding current, voltage, and wire feed speed, and corrects the welding trajectory in real time to ensure that the weld quality meets the requirements.

[0157] Compared with the prior art, the adaptive welding method of the present invention has the following advantages:

[0158] 1. It integrates multiple advanced technologies such as 3D modeling, image recognition, and adaptive trajectory planning, and can perceive and adapt to complex welding environments in real time, greatly improving the consistency and stability of welding quality.

[0159] 2. The adaptive welding parameter optimization algorithm based on the Euler angle timing sequence can dynamically adjust the welding process according to the real-time changing welding status, significantly enhancing the adaptive welding capability.

[0160] 3. The robot dynamic compensation model is integrated to effectively eliminate various dynamic errors in the welding process and ensure the reliability of weld quality.

[0161] In summary, the adaptive welding method proposed in the present invention solves the technical problem that the existing technology is difficult to achieve in automatically adjusting the welding parameters and the motion trajectory of the welding robot according to the real-time perception of the welding status during the welding process through the integrated application of technologies such as multi-sensor fusion, adaptive parameter optimization and dynamic trajectory compensation. BRIEF DESCRIPTION OF THE DRAWINGS

[0162] Figure 1 A flow chart of the method provided by the present invention;

[0163] Figure 2 Flowchart of image preprocessing steps. DETAILED DESCRIPTION

[0164] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0165] like Figure 1 FIG. 1 is a flow chart of a large-scale oil storage tank pontoon plate welding method provided by the present invention, and the method comprises the following steps:

[0166] S10, collecting three-dimensional point cloud data of the area to be welded of the floating plate using a sensor group provided on the crawler welding robot; the sensor group includes a laser line scanning sensor, a structured light sensor, an inertial measurement unit, and a torque sensor;

[0167] S20, using a camera on the welding robot to obtain an image of the area to be welded, and preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded;

[0168] S30, establishing a welding coordinate system based on the three-dimensional point cloud data and the characteristic image of the area to be welded, and calculating the three-dimensional contour of the groove;

[0169] S40, calculating the initial Euler angle of the welding head according to the three-dimensional contour of the groove, and generating a motion trajectory of the welding head;

[0170] S50, using the welding current, voltage, wire feed speed, shielding gas flow rate and welding speed in the welding process parameter database, combined with the initial Euler angle, groove geometry, plate thickness and ambient temperature, to establish an Euler angle timing sequence calculation equation group;

[0171] S60, obtaining a posture adjustment sequence of the welding head during the welding process by calculating the equation group of the Euler angle timing sequence, and generating a welding trajectory compensation amount according to the posture adjustment sequence;

[0172] S70, adjusting the motion trajectory of the welding head in real time based on the welding trajectory compensation amount, and collecting groove formation data during the welding process through the sensor group;

[0173] S80. Calculate the joint penetration and reinforcement using the groove forming data. When the joint penetration is less than 90% of the plate thickness or the reinforcement is greater than a preset threshold, return to step S50 to recalculate the Euler angle timing sequence until the joint penetration and reinforcement meet the process requirements.

[0174] The preset threshold is specifically that when the thickness of the plate is not greater than 25 mm, the excess height is 3 mm; when the thickness of the plate is greater than 25 mm, the excess height is 4 mm.

[0175] Furthermore, the step of preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded includes an image preprocessing step and an image feature extraction step. Figure 2 As shown, the image preprocessing steps specifically include:

[0176] S21, using a Gaussian filter to perform noise reduction processing on the image of the welding area;

[0177] S22, grayscale processing is performed on the denoised image;

[0178] S23, using a histogram equalization method to improve image contrast;

[0179] S24, using morphological operations to perform edge enhancement on the image;

[0180] S25. Obtain a binary image of the area to be welded through adaptive binarization processing.

[0181] The specific implementation methods of the steps of the present invention are described in detail below:

[0182] The specific implementation of step S10 is to first collect 3D point cloud data of the area to be welded using a set of sensors installed on the welding robot. These sensors include laser line scanning sensors, structured light sensors, and inertial measurement units. The laser line scanning and structured light sensors can obtain high-precision 3D data of the area to be welded, while the inertial measurement unit provides information on the motion state of the welding equipment itself. By processing and analyzing this 3D point cloud data, the shape and contour characteristics of the area to be welded can be accurately determined, laying the foundation for subsequent welding path planning and welding parameter optimization.

[0183] The specific implementation of step S20 is to first obtain a two-dimensional image of the area to be welded using a camera installed on the welding robot. These images are then preprocessed, including Gaussian filtering for noise reduction, grayscale processing, histogram equalization for contrast enhancement, and morphological edge enhancement, ultimately obtaining a binary image of the area to be welded. These preprocessing steps aim to enhance the feature information of the area to be welded, paving the way for subsequent feature extraction.

[0184] The specific implementation of step S30 involves first establishing a welding coordinate system based on the 3D point cloud data acquired in step S10 and the characteristic image of the area to be welded obtained in step S20. This coordinate system has the geometric center of the area to be welded as its origin, the X-axis running along the edge to be welded, the Y-axis perpendicular to the X-axis, and the Z-axis perpendicular to the workpiece surface. The 3D contour information of the groove, including its shape and dimensions, is then calculated based on the 3D point cloud data. This information provides a crucial basis for subsequent welding head trajectory planning and welding parameter optimization.

[0185] The specific implementation of step S40 is to first calculate the initial Euler angle of the welding head based on the three-dimensional contour information of the groove obtained in step S30. The Euler angle is the three rotation angles that describe the posture of the welding head, including the pitch angle, yaw angle and roll angle. These initial Euler angles provide a starting point for the subsequent generation of the welding head motion trajectory. Then, combined with the geometric characteristics of the groove, a trajectory planning algorithm is used to generate a smooth and continuous welding head motion trajectory. The trajectory planning equation based on B-spline is used here ,in To plan the trajectory, is the control point, is the B-spline basis function, is the smoothing term, is the smoothing coefficient. This equation can generate a smooth welding trajectory and meet the requirements of continuity and controllability.

[0186] The specific implementation of step S50 is to first obtain the welding current from the welding process parameter database ,Voltage , wire feeding speed , shielding gas flow and welding speed Then, the initial Euler angles calculated in step S40 are combined , groove geometry, plate thickness As well as the ambient temperature, a Euler angle timing sequence calculation equation group is established. This equation group includes the following 4 equations:

[0187] 1. Posture prediction equation: , used to predict the posture change trend of the welding head during movement. and is the influence matrix of velocity and acceleration, is a random disturbance term.

[0188] 2. Trajectory planning equation: , used to generate smooth and continuous welding trajectories. is the control point, is the B-spline basis function, is the smoothing term, is the smoothing coefficient.

[0189] 3. Dynamic compensation equation: , used to compensate for the inertia effect of the welding robot during movement. is the inertia matrix, is the Coriolis force matrix, is the gravity matrix, is the joint angle vector.

[0190] 4. Welding process mapping equation: , used to establish the mapping relationship between welding process parameters and molten pool behavior. is the penetration depth, For Yu Gao, is the process coefficient, and is the error term.

[0191] This Euler angle timing sequence calculation equation group can comprehensively consider the welding head posture prediction, trajectory planning, dynamic compensation, and the relationship between welding process parameters and molten pool behavior, providing a basis for subsequent real-time adjustment of the welding trajectory.

[0192] The specific implementation of step S60 is to first use the Euler angle timing sequence calculation equations established in step S50 to obtain the posture adjustment sequence of the welding head during the welding process through a numerical solution method. This posture adjustment sequence describes the Euler angle changes that the welding head needs to make during the movement. Then, based on this posture adjustment sequence, the compensation amount of the welding trajectory is calculated. , the specific expression is:

[0193] ;

[0194] in, is the trajectory after compensation, is the original trajectory, The compensation value can adjust the motion trajectory of the welding head in real time, so that it can better adapt to the actual welding process.

[0195] The specific implementation of step S70 is that during the welding process, the sensor group in step S10 collects the groove forming data in real time, including the weld surface height and the weld surface height. and metal density distribution Then use this data to calculate the penetration depth and Yu Gao , the specific calculation formula is:

[0196] Depth of penetration calculation: ;

[0197] Calculation of residual height: ;

[0198] in, is the depth weight function, is the surface height of the parent material, By real-time monitoring of the joint penetration and reinforcement data, deviations in welding quality can be discovered in a timely manner, providing a basis for the next step of process optimization.

[0199] The specific implementation of step S80 is to first determine whether the current weld penetration and reinforcement meet the requirements. If the penetration is less than 90% of the plate thickness or the reinforcement exceeds a preset threshold (3mm for plates under 25mm and 4mm for plates over 25mm), the process returns to step S50 and recalculates the Euler angle sequence to optimize the welding parameters. This iterative process continues until the penetration and reinforcement meet the process requirements.

[0200] The following is an example of a specific application scenario of the present invention: A large petrochemical company is constructing a new 2 million cubic meter floating roof storage tank, its largest tank to date. The tank is constructed using 30 mm thick Q390D high-strength steel plate, with a pontoon plate structure for the outer shell. The company's technical team decided to adopt the adaptive welding method proposed in this invention. First, they integrated a variety of sensor devices, including laser scanning sensors, structured light sensors, inertial measurement units, and high-definition cameras, into a welding robot. This sensor system is capable of fully sensing the 3D geometric features and 2D image information of the weld area.

[0201] During the welding preparation phase, the operator moves the welding robot to the edge of the pontoon plate to be welded. The sensor system then activates and begins collecting 3D point cloud data and 2D images. By processing and analyzing this data, a welding coordinate system with the geometric center of the pontoon plate as the origin is established, and key parameters of the welded area are extracted, as shown in Table 1.

[0202] Table 1 Geometric parameters of the area to be welded

[0203] parameter Numerical Groove root gap 4.2mm Bevel angle 30° Groove depth 8.5mm Plate thickness 30mm

[0204] From the process parameter database, the operator retrieved the welding parameters suitable for 30 mm thick Q390D steel plate, as shown in Table 2:

[0205] Table 2 Welding process parameters

[0206] parameter Numerical Welding current 320A Welding voltage 28V Wire feeding speed 8m / min Shielding gas flow 18L / min Welding speed 30cm / min

[0207] With these key information, the welding control system begins to establish the Euler angle timing sequence calculation equation group. First, through the posture prediction equation , predict the posture change trend of the welding head during the movement. and They are the influence matrices of velocity and acceleration, which are calibrated by experiments and taken as and ; is Gaussian white noise with a mean of 0 and a variance of 0.01.

[0208] Then, based on the B-spline curve theory, the trajectory planning equation was established By optimizing the control points and smoothing coefficient , a smooth and continuous welding trajectory was generated.

[0209] In order to compensate for the dynamic characteristics of the welding robot itself, a dynamic compensation equation was constructed. .in, for Inertia matrix, for Coriolis force matrix, for Gravity matrix, obtained through calibration test:

[0210] kg m ;

[0211] N m s / rad N ;

[0212] Finally, in order to establish the mapping relationship between welding process parameters and molten pool behavior, the welding process mapping equation was designed. .here, is the penetration depth, For Yu Gao, is the welding current, is the welding voltage, is the wire feeding speed, For the shielding gas flow, is the plate thickness, is the process coefficient, is a random error.

[0213] With the aforementioned algorithm models in place, the welding control system begins executing the adaptive welding process. First, based on the trajectory planning algorithm in step S40, the system generates an initial trajectory for the welding head. Then, during the welding process, the sensor system collects real-time weld formation data, including parameters such as penetration depth and excess height.

[0214] The control system monitors the weld quality in real time by using the calculation formula for penetration and residual height in step S70:

[0215] Depth of penetration calculation: ;

[0216] Calculation of residual height: ;

[0217] in, is the metal density distribution, is the depth weight function, is the weld surface height distribution, mm is the height of the parent material surface.

[0218] By real-time analysis of weld quality data, once it is found that the penetration depth is less than 27mm (90% of the plate thickness) or the excess height is greater than 4mm, the control system will automatically trigger the process optimization process of step S80.

[0219] First, the system will re-optimize the welding process parameters based on the current welding parameters through the Euler angle timing sequence calculation equation group in step S50. The adjusted parameters are shown in Table 3:

[0220] Table 3 Optimized welding process parameters

[0221] parameter Numerical Welding current 340A Welding voltage 30V Wire feeding speed 9m / min Shielding gas flow 20L / min Welding speed 25cm / min

[0222] Then, based on the optimized parameters, the control system recalculates the motion trajectory of the welding head. Through the trajectory compensation algorithm in step S60, the posture and position of the welding head are adjusted in real time to adapt to the current welding state.

[0223] After this round of adaptive optimization, the sensor system monitored the weld quality again and found that the penetration depth reached 28.5mm and the residual height was controlled within 3.8mm. Thus, the adaptive welding task of this pontoon plate section was successfully completed.

[0224] The entire welding process took 12.5 hours and involved three adaptive optimization cycles. Compared to traditional manual welding, this method significantly improved welding efficiency and weld quality consistency and stability.

[0225] Specifically, the principle of the present invention is to use multi-source perception information to build a high-precision welding geometry model, and based on this, establish an adaptive welding parameter optimization and trajectory planning algorithm, and ultimately achieve intelligent control of welding quality.

[0226] First, by integrating laser scanning sensors, structured light sensors, and inertial measurement units into the welding robot, high-precision three-dimensional point cloud data of the area to be welded can be obtained. This three-dimensional information lays the foundation for the subsequent establishment of the welding coordinate system and extraction of the groove contour.

[0227] The camera mounted on the welding robot also captures 2D images of the area to be welded. These images undergo preprocessing, including filtering, grayscale conversion, histogram equalization, and edge enhancement, ultimately resulting in a binary image rich in the characteristics of the welded area. This 2D feature information is then integrated with the 3D point cloud data to provide support for the precise description of the groove geometry.

[0228] Based on the acquired 3D point cloud data and 2D feature images, the present invention first establishes a welding coordinate system with the geometric center of the welded area as its origin. Then, by analyzing the 3D point cloud data, key geometric features such as the groove shape and dimensions are extracted. This information lays the foundation for subsequent welding head trajectory planning and welding parameter optimization.

[0229] Next, the present invention uses an adaptive Euler angle time series calculation equation system, combined with welding current, voltage, wire feed speed, and other data obtained from a process parameter database, to predict the posture change trend of the welding head during motion and generate a smooth and continuous welding trajectory. This equation system includes modules such as posture prediction, trajectory planning, dynamic compensation, and welding process mapping, and can fully consider multiple factors such as the kinematic characteristics of the welding head and the impact of welding process parameters on the behavior of the weld pool.

[0230] To eliminate trajectory deviations caused by the welding robot's inherent dynamic characteristics, the present invention incorporates a compensation algorithm based on the robot's dynamic model. This ensures high welding trajectory accuracy by compensating for the effects of inertia, Coriolis force, and gravity in real time.

[0231] During the welding process, a sensor system monitors weld formation in real time, including characteristic parameters such as penetration depth and reinforcement height. If penetration falls below 90% of the plate thickness or reinforcement height exceeds a preset threshold, the method automatically returns to the process parameter optimization step and recalculates the Euler angle timing sequence until the weld quality meets the requirements. This closed-loop control ensures reliable welding quality.

[0232] It should be noted that the explanation of relevant variables is shown in Table 4 below:

[0233] Table 4 Variable explanation table

[0234]

[0235] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A large oil storage tank pontoon plate welding method, characterized in that: The following steps are involved: S10, collecting three-dimensional point cloud data of the area to be welded of the floating plate by a sensor group provided on the crawler welding robot; the sensor group includes a laser line scanning sensor, a structured light sensor, an inertial measurement unit and a torque sensor; S20, using a camera on the welding robot to obtain an image of the area to be welded, and preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded; S30, establishing a welding coordinate system based on the three-dimensional point cloud data and the characteristic image of the area to be welded, and calculating the three-dimensional contour of the groove; S40, calculating the initial Euler angle of the welding head according to the three-dimensional contour of the groove, and generating a motion trajectory of the welding head; S50, using the welding current, voltage, wire feed speed, shielding gas flow rate and welding speed in the welding process parameter database, combined with the initial Euler angle, groove geometry, plate thickness and ambient temperature, to establish an Euler angle timing sequence calculation equation group; S60, obtaining a posture adjustment sequence of the welding head during the welding process by calculating the Euler angle timing sequence equation group, and generating a welding trajectory compensation amount according to the posture adjustment sequence; S70, adjusting the motion trajectory of the welding head in real time based on the welding trajectory compensation amount, and collecting groove formation data during the welding process through the sensor group; S80, calculating the joint penetration and the reinforcement using the groove forming data. When the joint penetration is less than 90% of the plate thickness or the reinforcement is greater than a preset threshold, returning to step S50 to recalculate the Euler angle timing sequence until the joint penetration and the reinforcement meet the process requirements; The Euler angle timing sequence calculation equation group includes a posture prediction equation, a trajectory planning equation, a dynamic compensation equation, and a welding process mapping equation; The posture prediction equation is used to predict the posture change trend of the welding head during movement. The input includes the initial Euler angle, the three-dimensional contour data of the groove and the movement speed of the welding head. The output is the predicted posture sequence of the welding head. The trajectory planning equation is used to generate a smooth and continuous welding trajectory. The input includes the predicted posture sequence, groove geometry and welding process parameters, and the output is the reference motion trajectory of the welding head. The dynamic compensation equation is used to compensate for the inertia effect of the welding robot during movement. The input includes the reference motion trajectory, the welding robot mass parameters and the motion speed, and the output is the compensated motion trajectory. The welding process mapping equation is used to establish a mapping relationship between welding process parameters and molten pool behavior. The input includes welding current, voltage, wire feed speed, shielding gas flow rate and plate thickness, and the output is the expected penetration depth and excess height.

2. A large oil storage tank pontoon plate welding method according to claim 1, characterized in that: The preset threshold is specifically that when the thickness of the plate is not greater than 25 mm, the excess height is 3 mm; when the thickness of the plate is greater than 25 mm, the excess height is 4 mm.

3. A large oil storage tank pontoon plate welding method according to claim 2, characterized in that: The step of preprocessing the image of the area to be welded to obtain a characteristic image of the area to be welded includes an image preprocessing step and an image feature extraction step.

4. A large oil storage tank pontoon plate welding method according to claim 3, characterized in that: The image preprocessing step specifically includes: S21, using a Gaussian filter to perform noise reduction processing on the image of the welding area; S22, grayscale processing is performed on the denoised image; S23, using a histogram equalization method to improve image contrast; S24, using morphological operations to perform edge enhancement on the image; S25. Obtain a binary image of the area to be welded through adaptive binarization processing.

5. A large oil storage tank pontoon plate welding method according to claim 4, characterized in that: The feature extraction step is specifically to extract the features using a preset welding image feature extraction model, wherein the welding image feature extraction model includes an initial feature extraction module, a mathematical model module, and an accurate feature extraction module.

6. A large oil storage tank pontoon plate welding method according to claim 5, characterized in that: The initial feature extraction module is used to extract the initial contour features of the groove area from the binary image. The input is the preprocessed binary image and the preset groove type parameters. The output is the initial contour feature vector of the area to be welded and the grayscale distribution matrix of the groove area. A deep convolutional neural network structure based on region growing is adopted. The mathematical model module is a matrix equation group, including a contour optimization matrix equation, a groove characteristic matrix equation and a geometric constraint matrix equation; The precise feature extraction module is used to generate the final feature data of the area to be welded. The input is the optimized contour feature vector, the geometric parameters of the groove and the rationality judgment result. The output is the precise feature description data of the area to be welded; the residual attention network structure is adopted.

7. A large oil storage tank pontoon plate welding method according to claim 6, characterized in that: The profile optimization matrix equation is used to correct and optimize the initial profile features, the input is the initial profile feature vector and the preset groove standard parameters, and the output is the optimized profile feature vector; The groove characteristic matrix equation is used to calculate the key geometric parameters of the groove. The input is the optimized contour feature vector and the grayscale distribution matrix of the groove area. The output is the root gap, groove angle and groove depth data of the groove. The geometric constraint matrix equation is used to verify the validity of the groove feature. The input is the root gap, groove angle and groove depth data of the groove, and the output is the rationality judgment result of the groove feature.

8. A large oil storage tank pontoon plate welding method according to claim 7, characterized in that: The steps of establishing a training data set for the welding image feature extraction model specifically include: Step 1: Collect 5,000 sets of floating plate welding images under different working conditions, including different plate thicknesses, different welding currents, different welding voltages, different ambient temperatures, and different welding speeds; Step 2: annotate the pontoon plate welding image, wherein the annotation contents include the groove contour line, the groove root position, the weld edge position and the molten pool boundary position; Step 3: Expand the training samples using data enhancement methods, including image rotation, image scaling, brightness adjustment, contrast adjustment, and noise addition; Step 4: Divide the training samples into a training set, a validation set, and a test set in a ratio of 8:1:1; Step 5: Normalize the images in the training set, the validation set, and the test set, and adjust the image size to 1024 by 1024 pixels.

Images