Multi-point intelligent welding system and method

Abstract

The invention provides a multi-point intelligent welding system and method. The method comprises the steps that process state feedback information between the tail end of each welding gun and a workpiece is obtained; dynamically predicting the thermal deformation amount of the workpiece, and compensating the initial expected pose of each welding gun according to the thermal deformation amount to generate a thermal compensation global pose; performing kinematics calculation according to a predefined welding seam path, the thermal compensation global attitude and a space cooperative constraint relationship among the welding guns, and generating a theoretical tail end pose of the tail end of each welding gun; for each welding gun, inputting the process state feedback information of the welding gun into the process self-adaptive adjustment model to obtain a local compensation pose used for offsetting the process deviation, and further obtaining a final target pose; and finally, according to the final target poses of the welding guns, the joint angles of the welding guns are determined, joint motors are driven to track the joint angles, and cooperative control over the poses and the process states of the welding guns is achieved. According to the scheme, high-precision cooperative pose control of the multiple welding guns in the thermal deformation environment can be achieved.

Description

Technical Field

[0001] This application relates to the field of welding technology, and more specifically, to a multi-point intelligent welding system and method. Background Technology

[0002] Welding, a key process in modern manufacturing, is widely used in aerospace, automotive manufacturing, heavy machinery, and other fields. With the improvement of industrial automation, multi-torch collaborative welding systems have gradually become the mainstream method for welding complex structural parts due to their advantages such as high efficiency and good consistency. However, during multi-torch collaborative operation, factors such as uneven heating of the workpiece, differences in the thermophysical properties of materials, and dynamic thermal deformation during the welding process cause deviations between the actual position and theoretical path of the welding torch, which seriously affects the welding quality and forming accuracy.

[0003] Traditional welding systems often rely on pre-programmed paths for control, lacking the ability to perceive and compensate for the real-time state of the workpiece (such as temperature field and thermal deformation). This is particularly pronounced in multi-welding-torch collaborative scenarios, where spatial interference and overlapping heat-affected zones between the torches become more significant. While some existing research attempts to achieve localized adjustments through visual or force feedback, a systematic multi-sensor information fusion and global-local collaborative compensation mechanism are still lacking. This fails to comprehensively suppress thermal deformation and process deviations, leading to unstable welding quality and poor process adaptability. Therefore, achieving high-precision collaborative pose control of multiple welding torches under thermal deformation conditions has become a major challenge for the industry. Summary of the Invention

[0004] This application provides a multi-point intelligent welding system and method, which can realize high-precision collaborative posture control of multiple welding torches under thermal deformation environment.

[0005] In a first aspect, this application provides a multi-point intelligent welding method, comprising the following steps: Obtain process status feedback information between the tip of each welding torch and the workpiece; Based on real-time monitoring of workpiece temperature field changes and workpiece material properties, the thermal deformation of the workpiece is dynamically predicted, and the initial desired pose of each welding torch is compensated according to the thermal deformation to generate the thermal compensation global pose of each welding torch. Based on the predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch, kinematic calculations are performed to generate the theoretical end pose of each welding torch end. For each welding torch, its process status feedback information is input into the process adaptive adjustment model to calculate the local compensation pose used to offset process deviations; the theoretical end pose of the welding torch is fused with the local compensation pose to obtain the final target pose of the welding torch. Based on the final target pose of each welding torch, the joint angle of each welding torch is determined, and the joint motor is driven to track the joint angle, thereby realizing the coordinated control of the pose and process state of each welding torch.

[0006] In some embodiments, obtaining process status feedback information between the tip of each welding torch and the workpiece specifically includes: Multimodal sensors are deployed at the end of each welding torch. These multimodal sensors include vision sensors, force sensors, and temperature sensors. The visual sensor captures weld images in real time and extracts molten pool morphology features. The force sensor is used to monitor the contact force between the welding torch and the workpiece, and the temperature distribution in the weld zone is collected by the temperature sensor. The molten pool morphology, the contact force between the welding torch and the workpiece, and the temperature distribution in the weld zone are integrated to form the process status feedback information of each welding torch.

[0007] In some embodiments, based on real-time monitoring of workpiece temperature field changes and workpiece material properties, the thermal deformation of the workpiece is dynamically predicted, and the initial desired pose of each welding torch is compensated according to the thermal deformation to generate a global thermal compensation pose for each welding torch. Specifically, this includes: Real-time monitoring of workpiece temperature field changes, construction of dynamic temperature distribution model, and output of real-time temperature field data of each region of the workpiece changing over time through dynamic temperature distribution model. The material properties of the workpiece, including the coefficient of thermal expansion, thermal conductivity and elastic modulus, are input into the finite element analysis model along with the real-time temperature field data of each region over time to predict the amount of thermal deformation. The predicted thermal deformation is converted into a pose compensation matrix. The initial desired pose of each welding torch is then transformed using the pose compensation matrix to generate a thermally compensated global pose.

[0008] In some embodiments, kinematic calculations are performed based on a predefined weld path, the thermally compensated global attitude, and the spatial cooperative constraint relationship between each welding torch to generate the theoretical end pose of each welding torch tip. Specifically, this includes: Obtain predefined weld path data, which includes path geometry and velocity planning; Based on the spatial cooperative constraint relationship between each welding torch, kinematic modeling of the multi-welding torch system is performed. Based on the weld path data and the thermal compensation global attitude, the theoretical end pose of each welding torch tip is generated by inverse kinematics calculation using the established kinematic model.

[0009] In some embodiments, for each welding torch, its process state feedback information is input into a process adaptive adjustment model to calculate a local compensation pose for offsetting process deviations. The theoretical end pose of the welding torch is then fused with the local compensation pose to obtain the final target pose of the welding torch. Specifically, this includes: A process adaptive adjustment model is constructed, which is used to learn the mapping relationship between process status feedback information and deviation compensation; The process status feedback information is input into the process adaptive adjustment model, and the local compensation pose is output. The final target pose is obtained by adding or multiplying the theoretical end pose with the local compensated pose using a weighted fusion algorithm.

[0010] In some embodiments, determining the joint angle of each welding torch based on its final target pose and driving the joint motor to track the joint angle to achieve coordinated control of the pose and process state of each welding torch specifically includes: Perform inverse kinematics calculations on the final target pose of each welding torch to solve for the joint angle sequence; The joint angle sequence is input into the proportional-integral-derivative controller to generate the drive signal for the joint motor. Real-time monitoring of joint motor feedback enables closed-loop tracking, ensuring consistency between posture and process status.

[0011] In some embodiments, the method further includes: During the welding process, collaborative execution data from multiple welding torches is collected, and the parameters of the process adaptive adjustment model are updated to improve the accuracy of the next welding operation. If an abnormal process status feedback is detected, a safety interruption mechanism is triggered, welding is suspended, and an alarm message is output.

[0012] Secondly, this application provides a multi-point intelligent welding system, comprising: The data acquisition module is used to obtain process status feedback information between the end of each welding torch and the workpiece; The processing module is used to dynamically predict the thermal deformation of the workpiece based on the real-time monitoring of the workpiece temperature field changes and the material properties of the workpiece, and to compensate the initial expected pose of each welding torch according to the thermal deformation, thereby generating the thermal compensation global pose of each welding torch. The processing module is also used to perform kinematic calculations based on the predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch, to generate the theoretical end pose of each welding torch end. The processing module is also used to input the process status feedback information of each welding torch into the process adaptive adjustment model, calculate the local compensation pose used to offset the process deviation, and fuse the theoretical end pose of the welding torch with the local compensation pose to obtain the final target pose of the welding torch. The execution module is used to determine the joint angle of each welding torch based on the final target pose of each welding torch, and drive the joint motor to track the joint angle, so as to realize the coordinated control of the pose and process state of each welding torch.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described multi-point intelligent welding method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-point intelligent welding method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In the multi-point intelligent welding system and method of this application, firstly, by real-time monitoring of the workpiece temperature field changes and combining material properties, the amount of thermal deformation is dynamically predicted, and thermal compensation is performed on the initial desired pose of the welding torch to generate a thermally compensated global pose. This effectively overcomes the workpiece deformation problem caused by heat accumulation during welding and improves the accuracy of the welding path. Secondly, kinematic calculations are performed through predefined weld paths, thermally compensated global poses, and spatial collaborative constraints between welding torches to generate theoretical end poses, ensuring the geometric consistency of multi-welding torch collaborative operation. Furthermore, through a process adaptive adjustment model, local compensation poses are calculated based on the process state feedback information of each welding torch. The theoretical end pose and the local compensation pose are fused to obtain the final target pose, realizing real-time cancellation of process deviations during welding and significantly improving the stability and consistency of welding quality. Finally, the joint motor is driven by inverse kinematics calculation to track the target pose, realizing high-precision collaborative pose control of multiple welding torches under thermal deformation environment. Attached Figure Description

[0016] Figure 1 This is an exemplary flowchart of a multi-point intelligent welding method according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the generation of thermally compensated global attitude based on thermal deformation, according to some embodiments of this application. Figure 3 This is a schematic diagram of the structure of a multi-point intelligent welding system according to some embodiments of this application; Figure 4This is a schematic diagram of the structure of a computer device for implementing a multi-point intelligent welding method according to some embodiments of this application. Detailed Implementation

[0017] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is an exemplary flowchart of a multi-point intelligent welding method according to some embodiments of this application. The method mainly includes the following steps: In step 101, process status feedback information between the end of each welding torch and the workpiece is obtained.

[0019] The process status feedback information is a set of parameters reflecting the real-time status of the welding process, such as the molten pool morphology, contact force, and temperature distribution. This is not specifically limited here. In some embodiments, the process status feedback information between the welding torch tip and the workpiece can be obtained in the following ways: Multimodal sensors are deployed at the end of each welding torch. These multimodal sensors include vision sensors, force sensors, and temperature sensors. The visual sensor captures weld images in real time and extracts molten pool morphological features, which may include morphological features such as weld pool width, depth and symmetry. In specific implementation, the weld image contains redundant visual information (such as background workpiece texture, spatter interference, etc.). The image can be analyzed by image processing algorithms to remove interference and extract quantitative parameters that reflect the essential morphological state of the weld pool. This will not be elaborated here. The force sensor is used to monitor the contact force between the welding torch and the workpiece. Specifically, for example, the force sensor can be used to collect three-dimensional torque data at the end of the welding torch. In addition, the temperature distribution of the weld zone collected by the temperature sensor is used to integrate the molten pool morphology, the contact force between the welding torch and the workpiece, and the temperature distribution of the weld zone to form the process status feedback information of each welding torch. In specific implementation, for example, the above multi-source data can be converted into a unified vector representation. In practice, multi-source data fusion algorithms, such as particle filtering algorithms or weighted average algorithms, can be used. No specific limitation is made here.

[0020] It should be noted that the multimodal sensor in this application can be integrated into the end effector of the welding torch to ensure real-time performance and accuracy, which will not be elaborated here.

[0021] In step 102, based on the real-time monitoring of the workpiece temperature field change and the workpiece material properties, the thermal deformation of the workpiece is dynamically predicted, and the initial expected pose of each welding torch is compensated according to the thermal deformation to generate the thermal compensation global pose of each welding torch.

[0022] Here, thermally compensated global attitude refers to the globally adjusted attitude after thermal deformation compensation of the initial desired pose, while the initial desired pose refers to the target position and attitude of the welding torch preset based on the weld path. In some embodiments, reference... Figure 2 The figure is an exemplary flowchart illustrating the generation of thermally compensated global attitude based on thermal deformation according to some embodiments of this application, specifically including: In step 1021, the temperature field changes of the workpiece are monitored in real time, and a dynamic temperature distribution model is constructed. The real-time temperature field data of each region of the workpiece changing over time is output through the dynamic temperature distribution model. For example, temperature data can be collected by an infrared thermal imager or other temperature acquisition equipment, and the dynamic temperature distribution model is constructed by simulating the temperature field evolution using the finite difference method. The core principle of the finite difference method is to discretize the continuous workpiece temperature field into a finite number of grid nodes, and use the difference between the nodes to approximate the differential of the continuously changing temperature field. This transforms the continuous heat conduction partial differential equation, which is difficult to solve directly, into a system of algebraic equations that can be solved by numerical calculation, ultimately realizing the simulation of the temperature field evolution process over time. Specifically, the following method can be used: The three-dimensional geometric model of the workpiece to be welded is divided into a uniform spatial grid (e.g., a cubic grid). Each grid intersection is a temperature calculation node. The coordinates and initial temperature of each node are recorded to realize the conversion from a continuous temperature field to a discrete node group.

[0023] Based on the thermal conductivity of the workpiece material, a partial differential equation for heat conduction describing the temperature change with time and space is established; at the same time, boundary conditions are defined (such as the temperature input of the welding heat source, the heat loss between the workpiece surface and the air, the heat exchange between the workpiece and the fixture, etc.), and the temperature change rules of the mesh boundary nodes are clarified.

[0024] Using the collected initial temperature data of the workpiece, the collected temperature values ​​are assigned to the discrete grid nodes as the initial state for temperature field evolution calculation.

[0025] For each grid node at a fixed time step, the differential terms in the heat conduction equation are approximated using finite difference formulas (such as central difference or forward difference) to establish the algebraic relationship between the temperature of that node and the temperatures of adjacent nodes and the time step. The algebraic equation system is solved to obtain the new temperature value of each node at the current time step, thus completing one temperature field evolution simulation.

[0026] The system acquires new workpiece temperature data in real time through temperature acquisition equipment, periodically corrects the temperature values ​​of grid nodes, and continuously iterates the calculation to achieve real-time updates of the dynamic temperature distribution model, which will not be elaborated here. In step 1022, the material property parameters of the workpiece, including the coefficient of thermal expansion, thermal conductivity, and elastic modulus, are input into the finite element analysis model along with real-time temperature field data of each region of the workpiece over time to predict the thermal deformation. Specifically, the finite element analysis model predicts the thermal deformation by first calculating the temperature field, then calculating the thermal stress, and finally obtaining the deformation. That is, the three-dimensional geometric model of the workpiece to be analyzed is first discretized into several independent micro-computational units (e.g., tetrahedral units, hexahedral units). Then, combined with the input material thermal conductivity and real-time temperature field data, the real-time temperature distribution of each discrete unit at different time points is determined through thermal conduction numerical calculation (i.e., the specific temperature value of each unit is determined), providing a basis for temperature change for subsequent thermal strain calculation.

[0027] Based on the real-time temperature distribution of the units obtained above, the temperature change of each unit is first calculated (i.e., the difference between the real-time temperature of the unit and the initial temperature); then, combined with the input material thermal expansion coefficient (i.e., the expansion and contraction ratio of the material under unit temperature change), the thermal strain of each unit due to temperature change (i.e., the expansion or deformation trend of the unit) is calculated through the thermal strain formula.

[0028] In addition, by combining the input material elastic modulus (which reflects the material's ability to resist deformation; the larger the elastic modulus, the greater the stress generated under the same thermal strain), the thermal stress generated by thermal strain in each element is solved through the structural mechanics equilibrium equation. Subsequently, based on the element correlation relationship of the finite element model, the deformation of all discrete elements is superimposed to calculate the overall thermal deformation of the workpiece (including the translational deviation of the workpiece along the X / Y / Z axes, as well as the rotational deviation such as warping and deflection caused by thermal stress), which will not be elaborated here. In step 1023, the predicted thermal deformation is converted into a pose compensation matrix. The initial desired pose of each welding torch is transformed using the pose compensation matrix to generate a global thermal compensation pose. Specifically, the pose compensation matrix can be represented as a 4x4 homogeneous transformation matrix to achieve translation and rotation compensation. The constructed pose compensation matrix is ​​multiplied by the homogeneous coordinates of the initial desired pose (4×4 matrix × 4×1 vector) to obtain a new 4×1 homogeneous coordinate. The first three elements of this coordinate are the target position after compensation. Combined with the pose corresponding to the rotation matrix, they together constitute the global thermal compensation pose.

[0029] It should be noted that, in a preferred embodiment of this application, the attitude compensation matrix is ​​a 4×4 homogeneous transformation matrix constructed based on the workpiece thermal deformation. Its function is to quantify the workpiece translational deviation (e.g., the workpiece shifts along the X / Y / Z axes due to heat) and rotational deviation (e.g., the workpiece warps or deflects due to thermal stress) caused by thermal deformation into a mathematical matrix, which is used to accurately correct the initial desired pose of the welding torch (i.e., the preset target position and attitude), and finally output a thermally compensated global attitude that can adapt to the state of the workpiece after thermal deformation.

[0030] In step 103, kinematic calculations are performed based on the predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch to generate the theoretical end pose of each welding torch end.

[0031] Among them, the spatial cooperative constraint relationship refers to the constraint relationship between the position, attitude and motion of multiple welding torches, and the theoretical end pose refers to the target position and attitude of the welding torch end based on the kinematic model.

[0032] In some embodiments, kinematic calculations are performed based on a predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch to generate the theoretical end pose of each welding torch. Specifically, this can be achieved in the following manner: Obtain predefined weld path data, which includes path geometry and velocity planning; Based on the spatial cooperative constraints between welding torches, kinematic modeling of the multi-welding torch system is performed. It should be noted that the kinematic modeling based on the spatial cooperative constraints of the multi-welding torches in this application involves transforming the spatial cooperative requirements between the multiple welding torches (e.g., anti-interference, attitude synchronization) into mathematical constraints, and combining this with the kinematic model of a single welding torch (e.g., the degrees of freedom of the robotic arm and joint parameters of a single welding torch) to construct the overall multi-body kinematic model of the multi-welding torch system. Specifically, for example: first, a basic kinematic model of a single welding torch is built, that is, for each welding torch, the number of joints of the welding torch is determined (e.g., 6 movable joints). The process involves several steps: first, determining the parameters of each welding torch joint's rotation angle, the length of the connecting rods between each joint, and the direction of joint rotation; second, establishing the correspondence between the joint rotation angle and the position and posture of the welding torch end; and third, clarifying the rules governing the movement of a single welding torch. Finally, the spatial coordination requirements of multiple welding torches are transformed into specific rules. Based on the need for multiple welding torches to avoid collisions and to synchronize their postures during welding, specific coordination rules are determined, namely, anti-collision rules and posture synchronization rules. The anti-collision rule stipulates that a safe distance must be maintained between the ends of any two welding torches to prevent them from colliding during movement. The posture synchronization rule stipulates that the posture of all welding torch ends (e.g., the angle of tilt) must be consistent, and the deviation cannot exceed a set range. Finally, an overall motion model of the multi-welding torch system is integrated. This model is based on the motion model of a single welding torch, with defined cooperative rules added as global requirements to build the motion model of the entire multi-welding torch system. Specifically, the joint angles of all welding torches are used as calculation variables, and the positions and postures that each welding torch end needs to achieve are used as targets. This forms a set of calculation logic that includes the motion laws of a single welding torch plus cooperative rules, allowing the motion of multiple welding torches to be correlated, ultimately resulting in the overall motion model of the multi-welding torch system.

[0033] Based on the weld path data and the thermal compensation global posture, inverse kinematics calculations are performed using the established kinematic model to generate the theoretical end pose of each welding torch. Specifically, the basic target pose can be determined by integrating the weld path and thermal compensation. First, predefined weld path data is extracted to clarify the trajectory (e.g., the direction of a straight weld, the curvature of a circular weld) and speed that each welding torch needs to follow during welding. Then, the thermal compensation global posture (which has already offset the overall posture adjustment of the welding torch due to workpiece thermal deformation) is superimposed, adjusting the original end pose corresponding to the weld path to the thermally compensated basic target pose. This represents the position and posture that the welding torch end needs to reach, conforming to both the weld path trajectory and the state of the workpiece after thermal deformation. Finally, the aforementioned basic target pose is substituted into the previously established multi-welding torch motion. The model (which includes spatial cooperative constraints such as collision avoidance and attitude synchronization) is used to check whether the basic target poses of all welding torches meet the cooperative rules: for example, whether a safe distance is maintained between the basic target poses of any two welding torches, and whether their attitudes are synchronized; if there is a conflict, the basic target pose of a single welding torch can be fine-tuned (for example, slightly adjusting the attitude angle without deviating from the weld path) until the target poses of all welding torches meet the cooperative requirements and a compliant target pose is obtained; finally, for the compliant target pose of each welding torch, inverse kinematics calculation is performed: that is, the angles that each joint of the welding torch needs to rotate are calculated by deriving from the compliant target pose that the end of the welding torch needs to reach. This process relies on the motion law of a single welding torch in the model to ensure that the derived joint parameters can accurately drive the end of the welding torch to reach the compliant target pose. Based on the joint parameters obtained from the inverse kinematics solution, the end-effector pose is back-calculated using the forward kinematics solution (calculating the end-effector pose from the joint parameters). The end-effector pose is then checked to see if it matches the compliant target pose. If it matches, the compliant target pose is the final theoretical end-effector pose for each welding torch. If the deviation exceeds the limit, the inverse kinematics parameters are readjusted until the requirements are met. This will not be elaborated further here.

[0034] It should be noted that the compliant target pose in this application is the welding torch end target pose without spatial conflict and in accordance with the cooperative rules. The kinematic model can take into account the welding torch's degrees of freedom (e.g., a 6-DOF robotic arm) to ensure the real-time performance and uniqueness of the solution, which will not be elaborated here.

[0035] In step 104, for each welding torch, its process status feedback information is input into the process adaptive adjustment model to calculate the local compensation pose used to offset the process deviation; the theoretical end pose of the welding torch is fused with the local compensation pose to obtain the final target pose of the welding torch.

[0036] Among them, the local compensation pose refers to the fine adjustment amount for the predetermined process deviation, the final target pose refers to the fused welding torch control target, and the process adaptive adjustment model is a model used to learn the mapping relationship between process state feedback information and local deviation compensation.

[0037] In some embodiments, for each welding torch, its process state feedback information is input into a process adaptive adjustment model to calculate a local compensation pose for offsetting process deviations. The theoretical end pose of the welding torch is then fused with the local compensation pose to obtain the final target pose of the welding torch. Specifically, this can be achieved in the following manner: To construct an adaptive adjustment model for the process, the following approach can be used: First, determine the input and output dimensions of the model. The input consists of process state feedback information, namely, the morphological characteristics of the weld pool (width, depth, symmetry), the contact force between the welding torch and the workpiece (three-dimensional torque data), and the temperature distribution in the weld zone (temperature field quantification data). Integrate these multi-source data into a unified input vector. The output consists of key parameters for local compensation pose, namely the position offset (such as a small amount of movement in the X / Y / Z axis direction) and attitude correction amount (such as a small rotation angle around the X / Y / Z axis) that need to be finely adjusted at the end of the welding torch, forming an output vector.

[0038] Secondly, by collecting and preprocessing training data, multiple welding experiments can be conducted to record process status feedback information, actual deviations, and effective compensation pose data under different welding conditions. The data collected can be recorded using multimodal sensors (vision, force, temperature sensors) to record the process status, and high-precision measuring equipment (such as laser displacement sensors) to record the actual deviation of the welding torch. The effective compensation pose that can offset the deviation can be determined manually or through algorithms. Preprocessing of data can include removing outliers, normalizing the data (e.g., unifying data of different magnitudes such as temperature and contact force into the same range), and dividing the data into a training set (used for model learning) and a validation set (used to verify model accuracy).

[0039] Next, select a model structure and train it offline. Choose a model structure that can handle dynamic time-series data and learn nonlinear mapping relationships, such as a neural network model. The model structure adopts a combination of CNN and RNN (where CNN extracts spatial features such as molten pool images and RNN processes time-series change features such as temperature and contact force). In specific implementation, a simplified deep learning model can also be used. No specific model is limited here. Offline training involves inputting training set data into the model and iteratively optimizing the model parameters with the goal of minimizing the deviation between the model's output compensated pose and the actual effective compensated pose, until the model's compensation accuracy on the validation set reaches the preset requirements.

[0040] In practice, the process status feedback information is input into the process adaptive adjustment model, which outputs a local compensation pose to offset process deviations. Since the adaptive adjustment model is a pre-trained model, meaning it has learned the correspondence between process status feedback information and local deviation compensation through historical welding data, when real-time process status feedback information is input, the adaptive adjustment model will quickly match the deviation type and degree corresponding to the current process status, determine the direction and range to be corrected, and the local compensation pose finally output by the adaptive adjustment model is a small adjustment parameter for a single welding torch, including two parts: one is position compensation (e.g., millimeter-level small movement along the X / Y / Z axes to correct the position deviation between the welding torch and the weld); the other is attitude compensation (e.g., small-angle rotation around the X / Y / Z axes to correct the welding torch tilt angle deviation). This compensation pose only applies to the local process deviation of a single welding torch and does not affect the global collaborative relationship of multiple welding torches.

[0041] In addition, in some embodiments, the theoretical end pose of the welding torch is fused with the local compensation pose to obtain the final target pose of the welding torch. This is achieved by a weighted fusion algorithm, which adds or multiplies the theoretical end pose and the local compensation pose to obtain the final target pose. The addition is a fusion based on the position dimension. Since both the theoretical end pose and the local compensation pose contain position information, the position of the theoretical end pose is a global target position calculated based on the predefined weld path, the global thermal compensation pose, and the spatial collaborative constraints of multiple welding torches. This ensures that the welding torch conforms to the weld trajectory, adapts to the thermal deformation of the workpiece, and does not interfere with other welding torches. The position of the locally compensated pose is a tiny positional adjustment calculated by the process adaptive adjustment model for local process deviations of a single welding torch. Adding the two poses involves superimposing their corresponding position coordinates to correct local deviations based on the global position, resulting in a final position that balances compliance and accuracy. Matrix multiplication, on the other hand, is a fusion of attitude dimensions. Since the attitude information of the theoretical end pose and the locally compensated pose (e.g., the rotation angle of the welding torch around the X, Y, and Z axes) can be represented by a homogeneous transformation matrix (a 4×4 matrix containing rotation and translation information), the attitude matrix of the theoretical end pose reflects the attitude adjustment result reflecting global coordination requirements (e.g., the overall tilt angle adapting to workpiece thermal deformation). The attitude matrix of the locally compensated pose is a tiny attitude correction matrix calculated for local attitude deviations of a single welding torch. Matrix multiplication, through mathematical operations on the two attitude matrices, fuses the rotation information of the global attitude with the fine-tuned rotation information of the local attitude (i.e., matrix operations can accurately transmit rotation logic and avoid attitude conflicts), resulting in a final attitude that conforms to global coordination of multiple welding torches while correcting local attitude deviations. This will not be elaborated further here.

[0042] In step 105, the joint angle of each welding torch is determined according to the final target pose of each welding torch, and the joint motor is driven to track the joint angle to achieve coordinated control of the pose and process state of each welding torch.

[0043] Among them, joint angle refers to the rotation angle of each joint of the welding torch robotic arm, and collaborative control refers to the real-time matching of posture adjustment and process status.

[0044] In some embodiments, determining the joint angle of each welding torch based on its final target pose and driving the joint motor to track the joint angle to achieve coordinated control of the pose and process state of each welding torch specifically includes: Inverse kinematics calculations are performed on the final target pose of each welding torch to solve the joint angle sequence. For example, multiple solutions can be calculated using geometric methods or numerical optimization algorithms, and the optimal solution can be selected. No specific limitations are made here. In the actual implementation, the kinematic model of the welding torch has established the forward kinematics relationship between the joint angles and the end pose. Inverse kinematics calculation is the reverse solution. Inverse kinematics calculation may obtain multiple sets of joint angles that satisfy the target pose (i.e., "multiple solutions"). The optimal solution needs to be selected in combination with the actual motion constraints of the welding torch (such as the maximum rotation angle of the joint and the motion stability requirements). For example, the angle combination with the smallest joint motion amplitude, no risk of exceeding the limit, and no interference between the welding torch and the workpiece / other welding torches can be selected. Then, the angles are arranged according to the time step of the welding process to form a joint angle sequence. The joint angle sequence is input into a proportional-derivative-integral (PID) controller to generate drive signals for the joint motors. In practice, the joint angle sequence represents the target angles that each joint of the welding torch needs to achieve at different time steps. The actual operating status of the joint motors is then monitored in real time (e.g., the actual angle of the current joint is obtained through a motor encoder). This forms a closed-loop comparison between the target angle and the actual angle. The PID controller calculates the deviation between the target angle and the actual angle through three components: proportional (P), integral (I), and derivative (D), and outputs a control quantity to correct the deviation. Finally, the PID controller converts the calculated corrected control quantity into a drive signal that the joint motors can recognize. This will not be elaborated further here. Real-time monitoring of joint motor feedback enables closed-loop tracking, ensuring consistency between posture and process status. This involves a continuous cycle of monitoring, comparison, and correction to ensure the joint motor always tracks the target angle, thereby ensuring the welding torch end posture accurately matches the final target posture and maintains consistency with welding process status (such as molten pool, contact force, and temperature). In practice, deviations can be corrected through encoder feedback, which will not be elaborated here.

[0045] In some embodiments, the method further includes: During the welding process, collaborative execution data from multiple welding torches is collected, and the parameters of the adaptive process adjustment model are updated to improve the accuracy of subsequent welding operations. Collecting collaborative execution data involves recording key execution data from each welding torch in real time during the welding process. This includes process status feedback information for each torch (molten pool morphology, contact force, weld zone temperature), actual end-effector pose (deviation from theoretical pose), joint motor operating parameters (such as actual joint angles), and the local compensation pose output by the adaptive process adjustment model, along with the corresponding compensation effect (such as whether the deviation is reduced after compensation). Updating the adaptive process adjustment model parameters involves organizing the collected collaborative execution data into sample pairs of process status and compensation effect. These new samples are used to fine-tune the model parameters (such as the weights and biases of the neural network). For example, if the original compensation amount is found to be too small under a certain type of molten pool deviation, the algorithm corrects the mapping relationship corresponding to that type of process status in the model, making the model more closely match the actual welding conditions. Clearly, the updated model can more accurately identify process deviations in subsequent welding operations, output more suitable local compensation poses, reduce the deviation between the welding torch end-effector pose and the ideal state, and thus improve the forming accuracy and quality consistency of subsequent welding operations.

[0046] In addition, if abnormal process status feedback information is detected, a safety interruption mechanism can be triggered to suspend welding and output alarm information, which will not be elaborated here.

[0047] In another aspect, in some embodiments, this application provides a multi-point intelligent welding system, referring to... Figure 3The figure is a schematic diagram of the structure of a multi-point intelligent welding system according to some embodiments of this application. The multi-point intelligent welding system 400 includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 is used to acquire process status feedback information between the end of each welding torch and the workpiece; The processing module 402 is used to dynamically predict the thermal deformation of the workpiece based on the real-time monitoring of the workpiece temperature field change and the material properties of the workpiece, and to compensate the initial expected pose of each welding torch according to the thermal deformation, thereby generating the thermal compensation global pose of each welding torch. The processing module 402 is also used to perform kinematic calculations based on the predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch, to generate the theoretical end pose of each welding torch end. The processing module 402 is also used to input the process status feedback information of each welding torch into the process adaptive adjustment model, calculate the local compensation pose for offsetting process deviations, and fuse the theoretical end pose of the welding torch with the local compensation pose to obtain the final target pose of the welding torch. The execution module 403 is used to determine the joint angle of each welding torch according to the final target pose of each welding torch, and drive the joint motor to track the joint angle, so as to realize the coordinated control of the pose and process state of each welding torch.

[0048] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described multi-point intelligent welding method.

[0049] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing a multi-point intelligent welding method according to some embodiments of this application. The method in the above embodiments can be implemented through... Figure 4 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0050] The processor 501 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more for controlling the execution of the multi-point intelligent welding method in this application.

[0051] The communication bus 502 may include a path for transmitting information between the aforementioned components.

[0052] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0053] The memory 503 stores program code for executing the solution of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. In the above embodiments, the execution of the multi-point intelligent welding method can be achieved through the processor 501 and one or more software modules in the program code in the memory 503.

[0054] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0055] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0056] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0057] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-point intelligent welding method.

[0058] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0059] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A multi-point intelligent welding method, characterized in that, Includes the following steps: Obtain process status feedback information between the tip of each welding torch and the workpiece; Based on real-time monitoring of workpiece temperature field changes and workpiece material properties, the thermal deformation of the workpiece is dynamically predicted. The initial desired pose of each welding torch is compensated according to the thermal deformation, generating a global thermal compensation pose for each welding torch. Kinematic calculations are performed based on a predefined weld path, the global thermal compensation pose, and the spatial cooperative constraint relationship between the welding torches to generate the theoretical end-effector pose of each welding torch. For each welding torch, its process state feedback information is input into a process adaptive adjustment model to calculate a local compensation pose used to offset process deviations. The theoretical end-effector pose of the welding torch is fused with the local compensation pose to obtain the final target pose of the welding torch. Based on the final target pose of each welding torch, the joint angle of each welding torch is determined, and the joint motor is driven to track the joint angle, thereby realizing the coordinated control of the pose and process state of each welding torch.

2. The multi-point intelligent welding method according to claim 1, characterized in that, The specific information obtained regarding the process status between the tip of each welding torch and the workpiece includes: Multimodal sensors are deployed at the end of each welding torch. These multimodal sensors include vision sensors, force sensors, and temperature sensors. The visual sensor captures weld images in real time and extracts molten pool morphology features. The force sensor is used to monitor the contact force between the welding torch and the workpiece, and the temperature distribution in the weld zone is collected by the temperature sensor. The molten pool morphology, the contact force between the welding torch and the workpiece, and the temperature distribution in the weld zone are integrated to form the process status feedback information of each welding torch.

3. The multi-point intelligent welding method according to claim 1, characterized in that, Based on real-time monitoring of workpiece temperature field changes and workpiece material properties, the thermal deformation of the workpiece is dynamically predicted. The initial desired pose of each welding torch is compensated according to the thermal deformation, generating a global thermal compensation pose for each welding torch. Specifically, this includes: Real-time monitoring of workpiece temperature field changes, construction of dynamic temperature distribution model, and output of real-time temperature field data of each region of the workpiece changing over time through dynamic temperature distribution model. The material properties of the workpiece, including the coefficient of thermal expansion, thermal conductivity and elastic modulus, are input into the finite element analysis model along with the real-time temperature field data of each region over time to predict the amount of thermal deformation. The predicted thermal deformation is converted into a pose compensation matrix. The initial desired pose of each welding torch is then transformed using the pose compensation matrix to generate a thermally compensated global pose.

4. The multi-point intelligent welding method according to claim 1, characterized in that, Based on the predefined weld path, the thermally compensated global attitude, and the spatial cooperative constraint relationship between each welding torch, kinematic calculations are performed to generate the theoretical end pose of each welding torch tip, specifically including: Obtain predefined weld path data, which includes path geometry and velocity planning; perform kinematic modeling on the multi-welding-torch system based on the spatial cooperative constraint relationship between each welding torch; and generate the theoretical end pose of each welding torch end by performing inverse kinematic calculation based on the weld path data and the thermally compensated global attitude through the established kinematic model.

5. The multi-point intelligent welding method according to claim 1, characterized in that, For each welding torch, its process status feedback information is input into the process adaptive adjustment model to calculate the local compensation pose used to offset process deviations. The theoretical end pose of the welding torch is then fused with the local compensation pose to obtain the final target pose of the welding torch. Specifically, this includes: A process adaptive adjustment model is constructed, which is used to learn the mapping relationship between process status feedback information and deviation compensation; The process status feedback information is input into the process adaptive adjustment model, and the local compensation pose is output. The final target pose is obtained by adding or multiplying the theoretical end pose with the local compensated pose using a weighted fusion algorithm.

6. The multi-point intelligent welding method according to claim 1, characterized in that, Based on the final target pose of each welding torch, the joint angle of each welding torch is determined, and the joint motor is driven to track the joint angle, thereby achieving coordinated control of the pose and process state of each welding torch. Specifically, this includes: Perform inverse kinematics calculations on the final target pose of each welding torch to solve for the joint angle sequence; The joint angle sequence is input into the proportional-integral-derivative controller to generate the drive signal for the joint motor. Real-time monitoring of joint motor feedback enables closed-loop tracking, ensuring consistency between posture and process status.

7. The multi-point intelligent welding method according to claim 1, characterized in that, The method further includes: During the welding process, collaborative execution data from multiple welding torches is collected, and the parameters of the process adaptive adjustment model are updated to improve the accuracy of the next welding operation. If an abnormal process status feedback is detected, a safety interruption mechanism is triggered, welding is suspended, and an alarm message is output.

8. A multi-point intelligent welding system, characterized in that, include: The data acquisition module is used to obtain process status feedback information between the end of each welding torch and the workpiece; The processing module is used to dynamically predict the thermal deformation of the workpiece based on the real-time monitoring of the workpiece temperature field changes and the material properties of the workpiece, and to compensate the initial expected pose of each welding torch according to the thermal deformation, thereby generating the thermal compensation global pose of each welding torch. The processing module is also used to perform kinematic calculations based on the predefined weld path, the thermal compensation global attitude, and the spatial cooperative constraint relationship between each welding torch, to generate the theoretical end pose of each welding torch end. The processing module is also used to input the process status feedback information of each welding torch into the process adaptive adjustment model, calculate the local compensation pose used to offset the process deviation, and fuse the theoretical end pose of the welding torch with the local compensation pose to obtain the final target pose of the welding torch. The execution module is used to determine the joint angle of each welding torch based on the final target pose of each welding torch, and drive the joint motor to track the joint angle, so as to realize the coordinated control of the pose and process state of each welding torch.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the multi-point intelligent welding method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-point intelligent welding method as described in any one of claims 1 to 7.

Images