Training method of welding imitation learning model and intelligent welding method

Abstract

The present application proposes a training method for a welding imitation learning model and an intelligent welding method, wherein the training method includes: constructing a welding master-slave control system, the welding master-slave control system includes a master-end device, a slave-end device and a working host, the master-end device includes a simulated welding gun and an optical motion capture camera, and the positioning tool of the optical motion capture camera is fixed on the simulated welding gun; the master-end device includes a robotic arm, a welding gun and a molten pool camera, the end of the robotic arm is fixedly connected to the welding gun, and the molten pool camera is installed on the side of the welding gun; the working host is connected to the optical motion capture camera, the molten pool camera and the industrial computer of the robotic arm, and the welding gun is controlled by the working host to run synchronously with the simulated welding gun, and a multimodal data stream in the welding process is obtained; training sample data is obtained based on the multimodal data stream; the constructed welding imitation learning model is trained based on the training sample data to obtain a trained welding imitation learning model, thereby solving the technical problem of limitations of welding robots in the prior art.

Description

Technical Field

[0001] The present application relates to the technical field of robot welding, and in particular to a training method for a welding imitation learning model and an intelligent welding method. Background Art

[0002] Traditional welding robot systems require either meticulous manual programming or extensive manual operations such as dragging the robot arm to teach the trajectory. Both solutions require significant time and manpower. This process is cumbersome and inflexible, especially when dealing with complex geometries or frequent product changes. Existing welding robots struggle to adapt to changes in workpiece geometry, material properties, and welding conditions. These changes can lead to welding defects and require manual intervention, resulting in adaptability issues. The transfer of skills required to operate existing welding robot systems is difficult. For human welders, who possess valuable expertise in welding technology and process control, controlling the welding robot using traditional methods (such as teach pendants and offline programming) is not intuitive. Excellent human welders are often not proficient in programming, and the feel of drag-and-drop teaching differs significantly from normal welding. Furthermore, traditional welding robot systems struggle to capture and replicate the skills of professional welders, making it difficult to translate the refined welding techniques of humans into efficient automated production processes.

[0003] Therefore, an intelligent welding solution tailored specifically for robotic welding applications is needed to address the above limitations of existing methods and enable robots to learn welding skills more efficiently and effectively. Summary of the Invention

[0004] To this end, the present application proposes a training method for a welding imitation learning model and an intelligent welding method to at least solve one of the technical problems in the related art to a certain extent.

[0005] The first embodiment of the present application proposes a training method for a welding imitation learning model, comprising the following steps:

[0006] Construct a welding master-slave control system, which includes a master-end device, a slave-end device, and a work host. The master-end device includes a simulated welding gun and an optical motion capture camera, and the positioning tool of the optical motion capture camera is fixed to the simulated welding gun; the slave-end device includes a robotic arm, a welding gun, and a molten pool camera. The end of the robotic arm is fixedly connected to the welding gun, and the molten pool camera is installed on the side of the welding gun; the work host is connected to the optical motion capture camera, the molten pool camera, and the industrial computer of the robotic arm, and the work host is equipped with a data acquisition system;

[0007] During the operation of the simulated welding gun for welding, the welding gun is controlled by the working host to run synchronously with the simulated welding gun, and a multimodal data stream of the welding process is obtained by the data acquisition system, wherein the multimodal data stream includes the posture change of the positioning tool acquired in real time by the optical motion capture camera, the molten pool image acquired in real time by the molten pool camera, and the process parameters uploaded by the welding machine of the welding gun and the industrial control computer of the robotic arm;

[0008] Based on the multimodal data stream, obtaining training sample data for training a welding imitation learning model;

[0009] The welding imitation learning model constructed based on the training sample data is trained to obtain a trained welding imitation learning model.

[0010] The second embodiment of the present application proposes an intelligent welding method based on a welding imitation learning model, wherein the welding imitation learning model is obtained by the training method of the welding imitation learning model described in the first aspect; the welding method comprises:

[0011] Obtaining an image sequence and a corresponding process parameter sequence captured in real time by the molten pool camera;

[0012] Inputting the image sequence and the corresponding process parameter sequence into the welding imitation learning model to obtain predicted posture changes and process parameter adjustments;

[0013] Convert the predicted pose change into the target pose in the robot base coordinate system, and then calculate the corresponding robot joint angle control instructions through the inverse kinematics solver;

[0014] The robot joint angle control instructions are sent to the robot controller of the robotic arm through the robot control interface at a fixed period to achieve intelligent welding.

[0015] The beneficial effects of the training method of the welding imitation learning model and the intelligent welding method provided in this application are as follows:

[0016] This solution uses data collected by motion capture cameras and melt pool cameras to train an imitation learning model, and combines the melt pool images collected by the melt pool camera to achieve precise control of the robot's end welding gun posture, so as to solve the problems of traditional welding robots such as high human intervention, low environmental adaptability and difficulty in skill transfer.

[0017] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0019] Figure 1 A flowchart of a training method for a welding imitation learning model provided in an embodiment of the present application;

[0020] Figure 2 A flow chart of an intelligent welding method based on a welding imitation learning model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0022] The following describes the training method of the welding imitation learning model and the intelligent welding method of the embodiment of the present application with reference to the accompanying drawings.

[0023] Figure 1 This is a flow chart of a training method for a welding imitation learning model provided in an embodiment of the present application. Figure 1 As shown, the training method of the welding imitation learning model includes the following steps:

[0024] Step S101, construct a welding master-slave control system, which includes a master-end device, a slave-end device and a working host. The master-end device includes a simulated welding gun and an optical motion capture camera, and the positioning tool of the optical motion capture camera is fixed on the simulated welding gun; the slave-end device includes a robotic arm, a welding gun and a molten pool camera, the end of the robotic arm is fixedly connected to the welding gun, and the molten pool camera is installed on the side of the welding gun; the working host is connected to the optical motion capture camera, the molten pool camera and the industrial computer of the robotic arm, and the working host is equipped with a data acquisition system.

[0025] This step builds a welding master-slave control system so that the robot can learn welding skills by observing and imitating the welding operations of human welding experts, thereby realizing intelligent welding technology.

[0026] As an example, the simulated welding gun and the slave device's welding gun have identical dimensions, weight, and grip (surface texture, center of gravity). The master device's simulated welding gun can also be replaced with a real welding gun. The optical motion capture camera's positioning tool consists of four reflective spheres, forming a rigid coordinate system. This positioning tool is fixed 20 cm from the end of the simulated welding gun and secured with a rigid bracket (made of aluminum alloy) to ensure a fixed relative position between the positioning tool and the simulated welding gun (calibration error <0.3 mm). The optical motion capture camera captures the positioning tool's 6-degree-of-freedom (6DoF) pose changes in real time (Δpi = (x, y, z, r, p, y), where x / y / z are translations (unit: mm) and r / p / y are Euler angles (unit: degrees) around the X / Y / Z axes). The robotic arm is a UR5 collaborative robotic arm (6 degrees of freedom). The end flange of the robotic arm fixes the actual welding gun (such as a MIG welding gun) with a customized fixture. The molten pool camera is installed on the side of the welding gun at a 45° angle, 200mm away from the molten pool (according to optical simulation, this position can cover the entire field of view of the molten pool without obstruction). It is used to collect molten pool images in real time to obtain the dynamic characteristics of the molten pool (melt width, melt depth, hump). The working host is the system control center, deployed in the main end operation area, and is connected to the optical motion capture camera, molten pool camera and the industrial computer of the robotic arm via a data cable (USB cable + network cable). It is used to display the molten pool image in real time (30Hz), assist the welder in observing the welding status, and synchronously collect the posture change Δpi output by the optical motion capture camera and the joint angle of the robotic arm. And the end position, molten pool image Ii, process parameters Si (welding current, welding voltage and welding gun movement speed).

[0027] In this embodiment, the master device and the slave device are physically isolated by a black shading plate. The height of the black shading plate is 2m and the shading rate is >99%, so as to ensure that the optical motion capture camera field of view of the master device is not interfered with by the welding arc, while protecting the operator from arc damage.

[0028] In this embodiment, after building the welding master-slave control system, in order to ensure control accuracy, the relevant coordinate systems need to be clearly defined and calibrated. It is necessary to obtain the calibration relationship between the motion capture camera coordinate system and the world coordinate system, the calibration relationship between the positioning tool coordinate system and the motion capture camera coordinate system, the calibration relationship between the robot base coordinate system and the world coordinate system, the calibration relationship between the welding gun tool coordinate system and the robot end flange coordinate system, and the calibration relationship between the molten pool camera coordinate system and the robot end flange coordinate system. Among them, the world coordinate system is the system reference datum, the motion capture camera coordinate system is the optical motion capture system coordinate system, the positioning tool coordinate system is the positioning tool coordinate system fixed on the simulated welding gun, the robot base coordinate system is the robot base coordinate system, the robot end flange coordinate system is the robot end interface coordinate system, and the welding gun tool coordinate system is the real welding gun end point (TCP) coordinate system.

[0029] Step S102: During the welding process of the simulated welding gun, the welding gun is controlled by the working host to run synchronously with the simulated welding gun, that is, the master-slave remote operation mode is realized; and the multimodal data stream of the welding process is obtained through the data acquisition system. The multimodal data stream includes the posture change of the positioning tool collected in real time by the optical motion capture camera, the molten pool image collected in real time by the molten pool camera, and the process parameters uploaded by the welding gun's electric welder and the industrial computer of the robotic arm.

[0030] As an implementation method, this solution involves having a skilled welder hold a simulated welding torch and activate the data acquisition system. The welder observes the weld pool image on the workstation's screen and manually adjusts the simulated welding torch's position (simulating actual welding movements). An optical motion capture camera captures the positioning tool's 6DoF pose change Δpi (the increment relative to the initial pose) at 100Hz. The workstation converts the captured Δpi into joint angle commands for the robotic arm through inverse kinematics, controlling the synchronized movement of the slave welding torch and the simulated torch until welding is complete. During this process, the weld pool camera captures weld pool images (224×224 resolution) at 30Hz. The controllers of the welding machine and robotic arm, to which the welding torch is connected, report process parameters Si at 100Hz.

[0031] Step S103: Based on the multimodal data stream, obtain training sample data for training the welding imitation learning model.

[0032] As an implementation method, a method for obtaining training sample data for training a welding imitation learning model includes: aligning the timestamps of the posture changes, molten pool images, welding voltage, welding current and movement speed in a multimodal data stream to generate a time-aligned multimodal data packet; and obtaining training sample data based on the time-aligned multimodal data packet.

[0033] As an example, the time synchronization module of the robot operating system (ROS) can be used to timestamp-align Δpi, Ii, and Si, generate a 30Hz time-aligned multimodal data packet (Ii, Δpi, Si), and store it in a preset path. Afterwards, the data in the multimodal data packet is divided into training sample data and model verification data.

[0034] Step S104: training the constructed welding imitation learning model based on the training sample data to obtain a trained welding imitation learning model.

[0035] The welding imitation learning model constructed in this embodiment includes an input layer, a feature extraction layer, a feature fusion layer, a time series modeling layer and an output layer:

[0036] The input layer is used to obtain the melt pool image sequence and the corresponding process parameter sequence from the training sample data using a sliding window. The melt pool image sequence includes n consecutive frames of melt pool images.

[0037] The feature extraction layer includes a visual feature extraction module and a process parameter embedding module. The visual feature extraction module is used to extract the spatial features of n frames of melt pool images through multiple convolutional neural networks and perform temporal splicing to obtain a temporal visual feature vector. The process parameter embedding module is used to construct the process parameter sequence into an n×3-dimensional temporal sequence, and then map the n×3-dimensional temporal sequence to an n×64-dimensional space through linear transformation to obtain a parameter feature vector.

[0038] The feature fusion layer is used to reduce the dimension of the temporal visual feature vector through linear transformation to obtain the reduced temporal visual feature vector; the reduced temporal visual feature vector is then used as the key and value, and the parameter feature vector is used as the query, and feature fusion is performed through cross-modal attention to obtain the fused feature vector;

[0039] The temporal modeling layer is used to capture the long-range temporal dependencies in the fused feature vector through a multi-layer Transformer Encoder network structure, resulting in a long-range feature vector. The multi-layer Transformer Encoder network structure includes multi-dimensional hidden layers, multi-head attention, and a feedforward network.

[0040] The output layer is used to pool the long-range feature vectors, and then map the pooled long-range feature vectors to predicted control instructions through the fully connected layer; the predicted control instructions include posture change, current adjustment, voltage adjustment and speed adjustment.

[0041] The model training goal of the welding imitation learning model in this embodiment is: input 5 consecutive frames of molten pool images (time series window) and corresponding process parameters, output the 6DoF pose change Δpi and process parameter adjustment ΔS at the current moment, and realize the end-to-end mapping of "molten pool state → welding gun action".

[0042] As an example, the model input includes a sequence of melt pool images and the corresponding process parameter sequences for the current moment and several past time steps (for example, a total of five frames). The visual feature extraction module feeds each of the five melt pool images (224×224×3×5) into a convolutional neural network (such as ResNet-34) to extract spatial features. Each convolutional neural network outputs a 512-dimensional vector, which, after time-sequential concatenation, yields a 5×512-dimensional temporal visual feature vector that captures the dynamics of the melt pool. The process parameters at these five moments form a 5×3-dimensional temporal sequence. The process parameter embedding module maps this 5×3-dimensional temporal sequence to a 5×64-dimensional space using a linear transformation (fully connected layer with a 3×64 weight matrix) to produce a parameter feature vector (5×64) to preserve temporal correlations. The feature fusion layer uses a cross-modal attention mechanism, taking the parameter feature vector as the query (abbreviated as Q) and the temporal visual feature vector as the key (abbreviated as K) and value (abbreviated as V). This setting allows the model to dynamically focus on the most relevant areas or features in the molten pool image according to the current process parameters, thereby enhancing the understanding of the correlation between the molten pool dynamics and the welding gun action; where the key / value is the temporal visual feature vector (5×512→5×64, reduced in dimension by linear transformation); the dot product attention of the attention mechanism ( , where d k The output layer (where is the dimension of the key) outputs a 5×64-dimensional fused feature vector to enhance the dynamic focus of process parameters on weld pool characteristics, for example, focusing on changes in weld width as current increases. The temporal modeling layer feeds the resulting fused feature vector into a 6-layer TransformerEncoder with a hidden layer dimension of 64, a multi-head attention (9 heads), and a feedforward network (64→256→64) to capture long-range temporal dependencies. For example, a hump in the weld pool in the previous three frames indicates that the welding gun angle needs to be adjusted. The output layer maps the output of the temporal modeling layer (e.g., 5×64 dimensions) to a predicted control command through a fully connected network (64→128→9), outputting pose changes and process parameter adjustments (3D: current adjustment, voltage adjustment, and movement speed adjustment). The model predicts the pose change (Δppred) and possible process parameter adjustments (ΔSpred) at the next moment of the 6-DoF welding gun tip. The Transformer output sequence can be aggregated using methods such as self-attention weighted averaging to generate a final single control command.

[0043] As an implementation method, a method for obtaining a trained welding imitation learning model is provided by training a constructed welding imitation learning model based on training sample data; the method includes: performing data enhancement processing on the training sample data; using the molten pool image and process parameters in the training sample data as feature data, and using the posture change as label data, and using the loss function to train the welding imitation learning model with the goal of minimizing the mean square error between the posture change predicted by the model and the actual value of the posture change, thereby obtaining a trained welding imitation learning model.

[0044] The data augmentation method for training sample data includes adding Gaussian noise and random brightness offset to the melt pool image and adding random perturbations to the process parameters. For example, Gaussian noise (σ = 0.01) and random brightness offset (±10%) are added to the melt pool image, and a ±5% random perturbation is added to the process parameters to simulate actual working conditions.

[0045] As an example, when training the model, the training objective can be to minimize the difference between the predicted Δppred and the collected Δpi. The prediction of process parameters is handled similarly. For example, the loss function is as follows:

[0046] L = α・Lpose + β・Lparam

[0047] Among them, Lpose is the mean square error of the pose change (weight α=0.8), and Lparam is the mean square error of the process parameters (weight β=0.2).

[0048] During training, the optimizer can be used: Adam (learning rate 1e-4, weight decay 1e-5); the training platform is NVIDIA A100 GPU, the batch size is 32, and the training cycle is 50 rounds (about 48 hours).

[0049] In one embodiment, after obtaining the trained welding imitation learning model, the method includes: performing lightweight processing on the trained welding imitation learning model to obtain a lightweight welding imitation learning model; and deploying the lightweight welding imitation learning model to the ROS control node of the working host.

[0050] As an example, the trained model is deployed to the ROS control node of the working host. The model can be lightweighted according to actual conditions: the model size is compressed to <15MB (inference delay <10ms) through pruning (removing redundant channels) and quantization (FP32→INT8). When the model is applied, the molten pool image captured by the molten pool camera is transmitted to the GPU for model inference. Based on the real-time input molten pool image sequence and process parameter sequence, the model outputs the predicted welding gun pose change Δppred and process parameter adjustment Δspred. The pose change Δppred is converted into the target pose in the base coordinate system of the robot system of the manipulator. The corresponding manipulator joint angle command is then calculated by the inverse kinematics solver. The joint angle command is sent to the manipulator controller through the manipulator control interface (such as the UR5 SDK) at a fixed period of 30Hz to achieve closed-loop motion control.

[0051] The training method of the welding imitation learning model of the embodiment of the present application constructs a master-slave control system to realize the master-slave remote control mode, collects the fine operation data of the human welder through the motion capture equipment, and combines the real-time visual feedback of the molten pool camera to train the welding imitation learning model to realize the precise closed-loop control of the welding gun posture at the end of the robot arm, so as to solve the problems of high manual intervention, low environmental adaptability and difficulty in skill transfer of traditional welding robots. By constructing a master-slave control system to realize data collection during the master-slave remote operation, welders do not need to perform complex programming or direct teaching in harsh environments, which significantly reduces labor costs and operation difficulty and improves data collection efficiency. Combining the high-precision posture measurement of optical motion capture and the process information of the molten pool camera, multimodal data fusion provides a richer and more reliable data foundation for model training. This solution collects real-time molten pool images through the molten pool camera, allowing welders to perceive the welding status online and dynamically adjust the welding gun posture and process parameters, so as to better adapt to the geometric differences of the workpiece, assembly errors and disturbances in the welding process, effectively improve welding quality and stability, reduce defect rates, and enhance environmental and task adaptability. This solution uses imitation learning to directly replicate the precise control strategies and subconscious compensation movements of human experts during actual operations. This overcomes the limitations of traditional methods in skill transfer fidelity, enabling the digital and automated transmission of expert welding skills, thereby improving the reliability of collected data. This solution also effectively prevents the effects of welding glare, high temperatures, and electromagnetic interference on motion capture cameras and operators through physical isolation between the master and slave terminals.

[0052] On the basis of any of the above embodiments, the embodiment of the present application further provides an intelligent welding method based on a welding imitation learning model, such as Figure 2 As shown, the intelligent welding method based on the welding imitation learning model includes the following steps:

[0053] Step S201: Acquire an image sequence and a corresponding process parameter sequence captured in real time by a molten pool camera.

[0054] Step S202: input the image sequence and the corresponding process parameter sequence into the welding imitation learning model to obtain the predicted posture change and process parameter adjustment.

[0055] In step S203, the predicted posture change is converted into a target posture in the robot base coordinate system, and then the corresponding robot joint angle control instructions are calculated by an inverse kinematics solver.

[0056] Step S204: sending the robot joint angle control instructions to the robot controller of the robotic arm through the robot control interface at a fixed period to realize intelligent welding.

[0057] The intelligent welding method based on the welding imitation learning model in the embodiment of the present application is based on the collection of fine operation data of human welders through motion capture equipment, combined with the real-time visual feedback of the molten pool camera, and the trained welding imitation learning model realizes precise closed-loop control of the position of the welding gun at the end of the robot arm, thereby realizing intelligent welding with strong adaptability and less human intervention.

[0058] In the descriptions of the foregoing embodiments, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0059] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0060] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A training method for a welding imitation learning model, characterized in that: The following steps are involved: Construct a welding master-slave control system, which includes a master-end device, a slave-end device, and a work host. The master-end device includes a simulated welding gun and an optical motion capture camera, and the positioning tool of the optical motion capture camera is fixed to the simulated welding gun; the slave-end device includes a robotic arm, a welding gun, and a molten pool camera. The end of the robotic arm is fixedly connected to the welding gun, and the molten pool camera is installed on the side of the welding gun; the work host is connected to the optical motion capture camera, the molten pool camera, and the industrial computer of the robotic arm, and the work host is equipped with a data acquisition system; During the operation of the simulated welding gun for welding, the welding gun is controlled by the working host to run synchronously with the simulated welding gun, and a multimodal data stream of the welding process is obtained by the data acquisition system, wherein the multimodal data stream includes the posture change of the positioning tool acquired in real time by the optical motion capture camera, the molten pool image acquired in real time by the molten pool camera, and the process parameters uploaded by the welding machine of the welding gun and the industrial control computer of the robotic arm; Based on the multimodal data stream, obtaining training sample data for training a welding imitation learning model; The welding imitation learning model constructed based on the training sample data is trained to obtain a trained welding imitation learning model; wherein, The welding imitation learning model includes: An input layer, configured to acquire a molten pool image sequence and a corresponding process parameter sequence from the training sample data using a sliding window, wherein the molten pool image sequence includes n consecutive frames of molten pool images; A feature extraction layer, comprising a visual feature extraction module and a process parameter embedding module. The visual feature extraction module is configured to extract spatial features of the n frames of melt pool images using multiple convolutional neural networks and perform temporal splicing to obtain a temporal visual feature vector. The process parameter embedding module is configured to construct the process parameter sequence into an n×3-dimensional temporal sequence, and then map the n×3-dimensional temporal sequence to an n×64-dimensional space through a linear transformation to obtain a parameter feature vector. A feature fusion layer is used to reduce the dimension of the temporal visual feature vector by linear transformation to obtain a reduced-dimensional temporal visual feature vector; then, the reduced-dimensional temporal visual feature vector is used as the key and value, and the parameter feature vector is used as the query, and feature fusion is performed through cross-modal attention to obtain a fused feature vector; A temporal modeling layer, configured to capture the long-range temporal dependencies in the fused feature vector through a multi-layer Transformer Encoder network structure comprising a multi-dimensional hidden layer, a multi-head attention network, and a feedforward network to obtain a long-range feature vector; The output layer is used to pool the long-range feature vectors and then map the pooled long-range feature vectors to predicted control instructions through a fully connected layer; the predicted control instructions include posture change, current adjustment, voltage adjustment and speed adjustment.

2. The method according to claim 1, characterized in that The process parameters include welding voltage, welding current and the moving speed of the welding gun.

3. The method according to claim 2, characterized in that The method of acquiring training sample data for training a welding imitation learning model based on the multimodal data stream includes: Performing time stamp alignment on the posture change, the molten pool image, the welding voltage, the welding current, and the moving speed in the multimodal data stream to generate a time-aligned multimodal data packet; Based on the time-aligned multimodal data packets, training sample data is obtained.

4. The method according to claim 1, wherein The welding imitation learning model constructed based on the training sample data is trained to obtain a trained welding imitation learning model; comprising: Performing data enhancement processing on the training sample data; The molten pool image and process parameters in the training sample data are used as feature data, and the posture change is used as label data. The loss function is used to train the welding imitation learning model with the goal of minimizing the mean square error between the posture change predicted by the model and the actual value of the posture change, thereby obtaining a trained welding imitation learning model.

5. The method according to claim 4, characterized in that Performing data enhancement processing on the training sample data includes: Adding Gaussian noise and random brightness offset to the melt pool image; Random perturbations are added to the process parameters.

6. The method according to claim 1, characterized in that After obtaining the trained welding imitation learning model, it includes: Performing lightweight processing on the trained welding imitation learning model to obtain a lightweight welding imitation learning model; The lightweight welding imitation learning model is deployed to the ROS control node of the working host.

7. The method according to claim 1, characterized in that The robotic arm is fixedly connected to the welding gun via an end flange, and the method further comprises: Obtain the calibration relationship between the motion capture camera coordinate system and the world coordinate system, the calibration relationship between the positioning tool coordinate system and the motion capture camera coordinate system, the calibration relationship between the robot arm base coordinate system and the world coordinate system, the calibration relationship between the welding gun tool coordinate system and the robot arm end flange coordinate system, and the calibration relationship between the molten pool camera coordinate system and the robot arm end flange coordinate system.

8. The method according to claim 1, characterized in that The master-end device and the slave-end device are physically isolated by a black shading plate.

9. An intelligent welding method based on welding imitation learning model, characterized in that: The welding imitation learning model is obtained by the training method of the welding imitation learning model according to any one of claims 1 to 8; The welding method comprises: Obtaining an image sequence and a corresponding process parameter sequence captured in real time by the molten pool camera; Inputting the image sequence and the corresponding process parameter sequence into the welding imitation learning model to obtain predicted posture changes and process parameter adjustments; Convert the predicted pose change into the target pose in the robot base coordinate system, and then calculate the corresponding robot joint angle control instructions through the inverse kinematics solver; The robot joint angle control instructions are sent to the robot controller of the robotic arm through the robot control interface at a fixed period to achieve intelligent welding.

Images