A method, device, equipment and storage medium for performing a welding task

Abstract

The present application discloses a welding task execution method, device, equipment and storage medium, which relates to the technical field of data processing and is applied to a control center. The method includes: sending welding tasks to each robot respectively according to the initial positioning information of the robot; obtaining the welding task execution results returned by each robot after executing the welding task and the current positioning information of the robot; the welding task execution results include weld images and initial position information of welding points; using a preset reinforcement learning model to correct the welding task execution results to obtain target position information; and based on the current positioning results of each robot and the target position information corresponding to the corresponding welding task, sending welding tasks to each robot again to execute the welding task. In this way, the present application can correct the position information of the welding points fed back by the robot through the reinforcement learning model, and combined with the accurate welding position, it can ensure the continuity and accuracy of welding and improve the welding efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a welding task execution method, device, equipment and storage medium. Background Art

[0002] For welding tasks in manufacturing environments (automobile manufacturing, aerospace, large steel structures, precision equipment assembly), the welding tasks are pre-distributed to robots in advance, and the robots execute the welding and feedback the corresponding welding results, that is, the positions of the welding points. In the process of determining the positions of the welding points, usually a monocular camera is used to collect weld images, and the direct method is used instead of the feature point method to establish a large-scale semi-dense map; then feature points are extracted from the images for matching to perform inter-frame estimation. Such processing has certain defects, and the change of the illumination conditions will significantly affect the performance of the monocular camera. In low-light or strong-light environments, the image quality may decline, resulting in the failure of feature point detection and matching.

[0003] It can be seen that how to accurately locate the positions of the welding points is a problem to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a welding task execution method, device, equipment and storage medium, which can correct the position information of the welding points fed back by the robot through a reinforcement learning model, and combine the accurate welding positions to ensure the continuity and accuracy of welding and improve the welding efficiency. The specific solutions are as follows:

[0005] In a first aspect, the present application provides a welding task execution method, which is applied to a control center and includes:

[0006] Distribute welding tasks to each robot according to the initial positioning information of the robot;

[0007] Obtain the welding task execution results returned by each of the robots after executing the corresponding welding tasks and the current positioning information of the robots; the welding task execution results include weld images and the initial position information of the corresponding welding points;

[0008] Use a preset reinforcement learning model to correct the welding task execution results to obtain the target position information of the corresponding welding points;

[0009] Based on the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks, re-distribute welding tasks to each of the robots so that each of the robots can execute the corresponding welding tasks.

[0010] Optionally, the weld seam image is a weld seam image containing a laser strip captured by a vision sensor after a robot projects a laser strip in a welding area after completing a corresponding welding task with a laser.

[0011] Optionally, the initial position information is the initial position information of a corresponding welding point in the coordinate system of the vision sensor calculated by the robot based on the laser triangulation method and the weld seam image;

[0012] Among them, the process of calculating the initial position information includes:

[0013] Preprocess the weld seam image, extract the feature points of the laser strip from the preprocessed weld seam image, and determine the intersection position of the laser plane and the weld plane according to the feature points;

[0014] Calculate the initial position information of the intersection position in the three-dimensional coordinate system of the vision sensor through the pinhole camera model.

[0015] Optionally, the use of a preset reinforcement learning model to correct the execution result of the welding task to obtain the target position information of the corresponding welding point includes:

[0016] Extract features from the weld seam image in the execution result of the welding task using the convolutional layer of the preset reinforcement learning model to obtain a corresponding feature vector;

[0017] Based on the feature vector, use the reinforcement learning network of the preset reinforcement learning model to adjust the coordinates of the initial position information to obtain the target position information of the corresponding welding point.

[0018] Optionally, the reissuing of the welding task to each robot based on the current positioning result of each robot and the target position information corresponding to the corresponding welding task includes:

[0019] Re-plan the welding tasks to be executed corresponding to each robot according to the current positioning result of each robot and the target position information corresponding to the corresponding welding task;

[0020] Based on the fast Fourier transform algorithm, distribute each welding task to be executed to the corresponding robot respectively.

[0021] In a second aspect, the present application provides a welding task execution method applied to a robot, including:

[0022] Obtain a welding task issued by a control center based on the initial positioning information of the robot;

[0023] Execute the welding task to obtain the corresponding welding task execution result, and send the welding task execution result and its own current positioning information to the control center, so that the control center uses a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; the welding task execution result includes a weld image and the initial position information of the corresponding welding point;

[0024] Obtain the welding task that the control center re-issues to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task, and execute the corresponding welding task.

[0025] In a third aspect, the present application provides a welding task execution device, which is applied to a control center and includes:

[0026] A first task issuing module, configured to issue welding tasks to each robot respectively according to the initial positioning information of the robot;

[0027] A data acquisition module, configured to acquire the welding task execution result and the current positioning information of the robot returned by each robot after executing the corresponding welding task; the welding task execution result includes a weld image and the initial position information of the corresponding welding point;

[0028] A correction module, configured to use a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point;

[0029] A second task issuing module, configured to re-issue welding tasks to each robot based on the current positioning result of each robot and the target position information corresponding to the corresponding welding task, so that each robot executes the corresponding welding task.

[0030] In a fourth aspect, the present application provides a welding task execution device, which is applied to a robot and includes:

[0031] A first task acquisition module, configured to acquire the welding task issued by the control center based on the initial positioning information of the robot;

[0032] A data sending module, configured to execute the welding task to obtain the corresponding welding task execution result, and send the welding task execution result and its own current positioning information to the control center, so that the control center uses a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; the welding task execution result includes a weld image and the initial position information of the corresponding welding point;

[0033] A second task acquisition module, configured to acquire the welding task reissued by the control center to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task, and execute the corresponding welding task.

[0034] In a fifth aspect, the present application provides an electronic device, including:

[0035] A memory, configured to store a computer program;

[0036] A processor, configured to execute the computer program to implement the welding task execution method as described above.

[0037] In a sixth aspect, the present application provides a computer-readable storage medium, configured to store a computer program, where the computer program, when executed by a processor, implements the welding task execution method as described above.

[0038] It can be seen that in the present application, the control center can issue welding tasks to each robot respectively according to the initial positioning information of the robot; then acquire the welding task execution results returned by each robot after executing the corresponding welding tasks and the current positioning information of the robot; the welding task execution results include weld images and the initial position information of the corresponding welding points; then use a preset reinforcement learning model to correct and process the welding task execution results to obtain the target position information of the corresponding welding points; then, based on the current positioning results of each robot and the target position information corresponding to the corresponding welding tasks, reissue welding tasks to each robot so that each robot can execute the corresponding welding tasks. In this way, the present application can control multiple robots to execute welding tasks collaboratively, and correct the position information of the welding points fed back by the robots through the reinforcement learning model to obtain a more accurate welding position; further, it can re-plan the welding tasks to be executed according to the welding tasks completed by each robot and the position information of the robots, and issue them to each robot for execution; combined with the accurate welding position, it can ensure the continuity and accuracy of welding and improve the welding efficiency. Description of the Drawings

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0040] Figure 1 It is a flowchart of a welding task execution method disclosed in the present application;

[0041] Figure 2Schematic diagram of the incident angle of three pairs of antennas disclosed in this application;

[0042] Figure 3 Schematic diagram of the timing sequence of a specific transmission phase disclosed in this application;

[0043] Figure 4 Another flowchart of a specific welding task execution method disclosed in this application;

[0044] Figure 5 Another flowchart of a specific welding task execution method disclosed in this application;

[0045] Figure 6 Schematic diagram of the structure of a welding task execution device disclosed in this application;

[0046] Figure 7 Another schematic diagram of the structure of a welding task execution device disclosed in this application;

[0047] Figure 8 Schematic diagram of the structure of an electronic device disclosed in this application. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] See Figure 1 As shown, an embodiment of the present invention discloses a welding task execution method, which is applied to a control center and includes:

[0050] Step S11: Send welding tasks to each robot respectively according to the initial positioning information of the robot.

[0051] In the embodiments of the present application, the control center can manage multiple robots simultaneously. According to the initial positioning information of the multiple robots and the overall welding work to be completed, sub-welding tasks corresponding to each robot are divided, and the welding tasks are sent to the robots respectively, so that the sub-welding tasks can be executed simultaneously by multiple robots. In a specific embodiment, each robot can be equipped with an ultra-wideband sensor and a 4-antenna array for mutual communication. The RTA (Reconfigurable Transmission Array) array of UWB (Ultra Wide Band) can be utilized, and the phase difference of arrival (PDoA, a positioning algorithm based on phase difference) between different antenna pairs is compared to determine the positioning information of the robot.

[0052] Step S12: Obtain the execution results of the welding tasks returned by each of the robots after executing the corresponding welding tasks and the current positioning information of the robots; the execution results of the welding tasks include weld images and the initial position information of the corresponding welding points.

[0053] In this embodiment, when each robot executes a welding task, it can obtain the initial position information of the welding points corresponding to the welding task and the corresponding weld seam image. Further, the robot can share the execution result of its own welding task and its current positioning information with the control center. In a specific embodiment, the weld seam image is a weld seam image containing a laser stripe captured by a vision sensor after the robot projects a laser stripe in the welding area after the corresponding welding task is completed. It can be understood that the robot can locate the weld seam by projecting a laser stripe and capture the weld seam image through a vision sensor to achieve the preliminary positioning of the weld seam. Further, the initial position information is the initial position information of the corresponding welding point in the coordinate system of the vision sensor calculated by the robot based on the laser triangulation method and the weld seam image. The process of calculating the initial position information includes: preprocessing the weld seam image, extracting the feature points of the laser stripe from the preprocessed weld seam image, and determining the intersection position of the laser plane and the weld plane according to the feature points; calculating the initial position information of the intersection position in the three-dimensional coordinate system of the vision sensor through the pinhole camera model. Specifically, the robot can obtain the depth information of the weld seam through the laser triangulation method, and combine the weld seam image captured by the vision sensor to achieve the preliminary positioning of the weld seam. Specifically, a laser stripe is projected in the welding area by a laser to form an intersection with the weld surface. It can be understood that the projection angle and position of the laser stripe can be accurately calibrated to ensure the accuracy of the measurement. Correspondingly, a vision sensor (such as an industrial camera) can capture a weld seam image containing the laser stripe; since the laser stripe presents a feature line with high contrast in the image, the corresponding feature points can be further extracted subsequently. Further, the captured weld seam image needs to be preprocessed, which may include Gaussian blur and threshold segmentation, to remove noise and highlight the feature points of the laser stripe. The preprocessed image is convenient for more accurately extracting the edges and feature points of the laser stripe. Then, the feature points of the laser stripe are extracted from the preprocessed weld seam image to determine the intersection position of the laser plane and the weld surface. These intersection positions provide the depth information of the weld seam in space, and the initial position of the welding point can be further determined. The following equation can be used in combination with the pinhole model to calculate the coordinates of the intersection point P in the camera coordinates.

[0054] ;

[0055] where, (x c , y c , z c ) is the position of P in the camera coordinates; f is the focal length; ρ w and ρ h are the width and height of the image (in pixels); u and v represent the coordinates of the target point on the image plane; c z and c yAre calibration parameters when converting from pixel coordinates to the camera coordinate system; (u0, v0) are the coordinates of the pixel located at the center of the image. a, b, c, and d are parameters of the laser plane equation.

[0056] Step S13: Use a preset reinforcement learning model to correct the execution result of the welding task to obtain the target position information of the corresponding welding point.

[0057] In the embodiment of the present application, through the above steps, the preprocessed weld image obtained by the robot when executing the welding task and the initial position information of the corresponding welding point can be obtained; this initial position information is three-dimensional coordinate information and may have certain errors and drifts. At this time, the reinforcement learning model can be further used to correct the initial position information to obtain more accurate target position information of the welding point. In a specific embodiment, the use of a preset reinforcement learning model to correct the execution result of the welding task to obtain the target position information of the corresponding welding point may include: using the convolutional layer of the preset reinforcement learning model to extract features from the weld image in the execution result of the welding task to obtain the corresponding feature vector; based on the feature vector, using the reinforcement learning network of the preset reinforcement learning model to adjust the coordinates of the initial position information to obtain the target position information of the corresponding welding point. Specifically, the preset reinforcement learning model includes convolutional filters and a deep Q network corresponding to the corresponding reinforcement learning algorithm (combining deep learning and reinforcement learning); the convolutional layer is used to extract features from the preprocessed weld image to generate the corresponding high-dimensional feature vector; these features capture the key patterns and details in the image and can provide rich information for subsequent decision-making. Further, subsequent action decisions are made based on the reinforcement learning algorithm. Based on the extracted feature vector, the Q value of each possible action (such as the direction and amplitude of moving the tracking frame) can be calculated through the deep Q network; then the action with the highest Q value can be selected according to the greedy strategy to determine how to adjust the tracking position to reduce the error from the actual weld feature points. Further, according to the decision of the deep Q network, the initial position can be adjusted, for example, moving up one pixel; then the corrected coordinates are used for the next step of tracking and error calculation to gradually approach the actual weld position, and finally the accurate weld position information, that is, the target position information, is obtained. It can be understood that the corrected three-dimensional coordinates have higher accuracy and stability and can ensure that the welding path is consistent with the actual weld. Moreover, the action instructions output by the deep Q network can be used to guide the corresponding robot to adjust the position of the welding point tracking frame to further optimize the tracking accuracy.

[0058] Step S14: Based on the current positioning results of each robot and the target position information corresponding to the corresponding welding task, re-issue the welding task to each robot so that each robot can execute the corresponding welding task.

[0059] In this embodiment, the position correction of the welding points can be achieved through the above steps. The control center can re-plan the welding tasks of each robot according to the current positioning results of each robot and the target position information obtained by correcting the corresponding welding points of the corresponding welding tasks, so that each robot can continue to cooperate to execute the relevant welding tasks. In a specific embodiment, a time division multiplexing mechanism can be adopted. By quickly switching time slots, a robot can share data with the control center and / or other robots in different time slots; combining a specific time window to achieve data transmission and sharing can avoid data conflicts and data loss.

[0060] In a specific embodiment, the re-issuing of the welding tasks to each of the robots based on the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks may include: re-planning the welding tasks to be executed corresponding to each of the robots according to the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks; and distributing each of the welding tasks to be executed to the corresponding robot based on the fast Fourier transform algorithm. Specifically, during the process of data transmission between the control center and each robot, the fast Fourier transform can be used to improve the calculation speed, and the formula is as follows:

[0061] ;

[0062] where x[n] is the input signal. N is the signal length. j is the imaginary unit. is the rotation factor.

[0063] Correspondingly, in a multi-robot scenario, the tasks can be distributed among multiple robots through parallel computing:

[0064] ;

[0065] where x1, x2,... x m are the input signals between different robots.

[0066] In a specific embodiment, the control center may include a communication module, a data processing module, a task scheduling module, and a user interface; wherein, the communication module can achieve wireless communication with each robot to ensure efficient data transmission and synchronization; the data processing module can use the optimized phase difference estimation algorithm to calculate the relative positions of each robot in real time. The task scheduling module can dynamically allocate and adjust the welding tasks of each robot according to the real-time positioning information of the robot. The user interface is used for the operator to monitor the system status in real time and for manual intervention. The main functions of the control center include real-time collection of the position information of each robot, coordination of phase difference measurement requests, and integration and analysis of positioning data. It can be understood that considering a group of robots Move freely in a given environment without obstacles. Let the robot have its three-dimensional position defined as . Each robot is equipped with an ultra-wideband sensor and a 4-antenna array for communication with each other. The goal of each robot is to use this UWB system to obtain an estimate of the relative position of its neighboring robots with respect to its own body frame , and , where R is a three-dimensional space, 3 and represents the rotation of the robot performing relative positioning sensing from the global to the body frame, which can more accurately calibrate the UWB ranging data. B j is the body coordinate system of robot j, and R j is the rotation matrix of robot j. Combining the estimation algorithm that allows robot 𝑗 to sense the relative position with respect to robot 𝑖 when robot 𝑖 initiates a ranging request can minimize the relative positioning error as much as possible. The robot uses a set of ultra-wideband sensors to determine the distance and direction of the received communication signal. The robot may include an antenna array for sending and receiving data, a basic platform consisting of four UWB modules, capable of calculating the arrival phase of the input signal, and connected to a processing subsystem that performs all calculations. Further, to effectively process the phase differences of multiple antenna pairs, the phase difference is used to estimate the relative position, and the phase difference estimation algorithm is as follows:

[0067] ;

[0068] where A is the phase difference matrix, and A + is the pseudo-inverse matrix containing the phase difference information between multiple antenna pairs. b is the observed signal vector. To determine the direction of the signal source, the RTA array of UWB can be utilized, and the arrival phase differences between different antenna pairs can be compared. This operation can be performed each time the robot initiates a relative position detection request to obtain the positioning of the signal source. The ToA (Time of Arrival) of the radio signal between two nodes is calculated using asynchronous two-way ranging between the nodes, which is already well-established and known to provide centimeter-level accuracy. For PDoA measurements, ideally, the distance between all antenna pairs is set to half of the carrier wavelength λ / 2 to avoid ambiguity (the estimated phase difference exceeds π or -π, thus wrapping to the opposite sign), and the carrier frequencies between all UWB transceivers should be synchronized.

[0069] Assume that the arrival phases of all 4 antennas in the RTA array are:

[0070] ;

[0071] Furthermore, the phase differences of all 6 antenna pairs can be calculated using the following formula:

[0072] ;

[0073] where, n and o in represent any two of the 4 antennas numbered 1, 2, 3, and 4. n and o are not equal, and there are 6 combinations, namely 6 antenna pairs; is the arrival phase of the nth antenna of robot i and robot j, is the arrival phase of the oth antenna of robot i and j, represents the phase deviation between the nth and oth antennas of robot j. o and n are taken from different values within the same range. The phase difference calculated for a source perpendicular to the antenna pair may not be zero. The bias cancellation term can be used to make up for this. By first measuring the phase angles of all possible angle pairs without any bias to calibrate each antenna pair, the bias compensation can be found. These are compared with the true angles they should output according to their geometry and bias factor to minimize the least squares error. After that, the formula for calculating the incident angles of all 6 pairs of antennas is as follows:

[0074] ;

[0075] where, is the incident angle of the antenna pair n and o of robot i and j relative to the signal source. is the measured phase difference between the antenna pair n and o of robot i and j. The incident angles of 3 pairs of antennas are as shown in Figure 2 , showing the relative position relationship between the three pairs of antennas and the signal source. By calculating the incident angles between different antenna pairs, the precise positioning of the signal source can be achieved.

[0076] Using the 6 PDoA measurement values and the specific geometry of the antenna array, by solving the redundant equations, the unit vector pointing to the source is obtained;

[0077] ;

[0078] where, is the component of the unit direction vector from the antenna pair of robot i and robot j to the signal source in the x direction; is the component of the unit direction vector from the antenna pair of robot i and robot j to the signal source in the y direction; is the component of the unit direction vector from the antenna pair of robot i and robot j to the signal source in the z direction; In this case, n and o represent any two of the four antennas numbered 1, 2, 3, and 4, and n and o are not equal; x, y, and z are coordinate axes. Considering the characteristics of the UWB module, there is a direct correlation between noise and amplitude When the amplitude is low, the estimation error of the incident angle is small; however, when the incident angle exceeds 70°, the estimation error increases significantly. Therefore, for a specific set of received phase differences, the output antenna pairs with very large phase differences are removed from the above equations. The experimental results show that when the threshold is set to = 165°, the effect is good.

[0079] Next, based on the remaining pairs with valid phase difference values in the equation set, the pseudo-inverse of A (phase difference matrix) can be used, and the following equation below the equation can be approximately solved to obtain the normalized direction estimate:

[0080] ;

[0081] where, represents the normalized direction vector in the m-th measurement. represents the unnormalized direction vector in the m-th measurement. represents the sum of the squares of the components on the three coordinate axes x, y, and z. It can be understood that the complete AoA estimation relative positioning process can be implemented by Algorithm 1. The input of Algorithm 1 is the phase arriving at each antenna, and the output is the adjacent unit vector direction; with the above formula, Algorithm 1 can calculate the phase difference through bias compensation, calculate the incident angle of each antenna pair, the original estimate of the unit azimuth vector , and normalize the unit vector to a length of 1. Further, using the distance measurement value obtained by TWR and the direction obtained by Algorithm 1, the estimated relative positioning obtained is .

[0082] In a specific embodiment, a specific information transfer protocol is used between the robot and the RTA; such as Figure 3As shown, the figure shows the timing relationship among the information transfer phase, the ranging phase, and the orientation strategy phase. The information phase is shown in blue, the ranging phase is shown in red, and the orientation phase is shown in green. When the device acts as a transmitter, only one antenna A4 is used. All steps start from the initialization information request, and the subsequent data transmission is divided into three phases: information transfer, ranging, and orientation measurement. In addition, the control center (such as a server) manages the execution of the welding tasks of each robot and dynamically allocates new welding tasks to improve the coordination and efficiency of the overall system. Further, in the information transfer phase, the information to be transmitted will be correctly formatted into 802.15.4A data packets and sent. In a typical 802.15.4A transmission, the maximum payload is 127 bytes. Therefore, the original information can be split into data packets of 120 bytes each for transmission, and an additional 7 bytes are used for Cyclic Redundancy Check (CRC). This phase ensures that the necessary operation data, such as task status, position coordinates, and sensor data, can be efficiently and securely exchanged among the robots. In the ranging phase, the above-mentioned Two-Way Ranging (TWR) protocol can be used to achieve the distance measurement of the source. In this phase, the receiver only uses its A4 receiver. Accurate distance measurement helps the relative positioning among the robots, ensuring that collisions can be accurately avoided and the working path can be optimized during collaborative operations. In the orientation measurement phase, that is, the phase of measuring the orientation between the transmitter and the receiver, in this phase, all four antennas on the receiver side will be enabled. The orientation measurement data is used for further precise positioning to enhance the collaboration accuracy among the robots.

[0083] In another specific embodiment, to achieve efficient collaboration among multiple robots and ensure that each robot can receive new welding tasks in a timely manner and execute them accurately, the control center is used to issue welding tasks to the robots and integrate data. The control center is responsible for dynamically allocating new welding tasks according to production requirements and the current status of each robot (such as task completion status, position, load, etc.). By optimizing task allocation, the control center can improve the overall welding efficiency and shorten the production cycle. According to the urgency and priority of tasks, the control center can reasonably arrange the execution order of tasks to ensure that critical tasks are completed first. The control center can monitor the working status of each robot in real time, including position, task progress, welding quality, etc., to ensure the overall coordination of the system. Exception handling: When a robot fails or has an abnormal task execution, the control center can detect it in time and take corresponding emergency measures, such as reallocating tasks or notifying maintenance personnel. The control center can integrate sensor data and task execution information from all robots to form a comprehensive view of the system status. Performance optimization: By analyzing the collected data, the control center can identify system bottlenecks, optimize welding paths and task allocation strategies, and improve the overall system performance. Correspondingly, during the information interaction process between the control center and the robots, after a robot starts or completes the current task, it sends an initialization information request to the control center, reports its own status, and requests a new welding task; the control center allocates new welding tasks according to the status and task requirements of each robot, and sends the task information to the corresponding robot through the 802.15.4A protocol; after receiving the task, the robot starts to perform the welding operation and exchanges data with other robots and the control center during the information transmission, ranging, and azimuth measurement stages to ensure the smooth progress of the task; after the robot completes the task, it sends the status information of task completion to the control center, including completion time, welding quality assessment, etc. After receiving the feedback, the control center updates the task status and allocates new tasks or adjusts the existing task plan as needed; if an exception occurs during the task execution, the robot can report the fault information to the control center. The control center takes corresponding measures according to the situation, such as reallocating tasks or taking other emergency measures, to ensure that the production process is not affected.

[0084] In another specific embodiment, the relative positioning process of robot A is specifically as follows: Start relative position detection request: After completing the current welding task, robot A sends a relative position detection request to the control center to apply for relative position detection. Control center coordinates measurement: After receiving the relative position detection request, the control center sends a synchronous measurement signal to other robots (such as robots B, C, and D) to ensure that the antenna pairs of all relevant robots perform phase difference measurements at the same time. Phase difference measurement and data collection: Under the instruction of the control center, robot A and other robots use the RTA array of UWB to perform PDoA measurement and record the phase difference of each antenna pair. Data transmission and processing: All measurement data are uploaded through the control center, and the control center uses the optimized phase difference formula to calculate the relative position of robot A. Position update and feedback: After the calculation is completed, the control center updates the position information of robot A to the system database (of the control center), and feeds back the relevant position information to robot A and other robots for subsequent coordinated operations.

[0085] It can be seen that the present application can simultaneously control multiple robots to collaboratively perform welding tasks, and through the reinforcement learning model, the position information of the welding point fed back by the robot can be corrected to obtain a more accurate welding position, combined with deep reinforcement learning to improve welding accuracy, and use visual sensors and deep learning technology to track the weld in real time to ensure high-precision operation during welding. This can reduce the need for manual monitoring and shorten the production cycle; by integrating high-precision positioning sensors and real-time visual feedback systems, the relative position between the robot and the workpiece can be monitored in real time during the welding process, and the path and action of the robot can be dynamically adjusted; the time multiplexing mechanism is adopted to allow multiple robots to efficiently share positioning and welding information in a limited space, so that the robot can synchronously process information transmission and task allocation, optimize the overall welding process, and ensure the efficient cooperation of multiple robots in complex environments. Further, relying only on the robot's own sensors to achieve collaboration greatly reduces costs and complexity, and adapts to various complex construction environments; this can also ensure the accuracy of welding, and avoid positioning errors and physical deviations in traditional methods, greatly improve the intelligence level of the production line, reduce operating costs, and enhance production efficiency. Furthermore, the present application can re-plan the welding tasks that need to be performed based on the welding tasks completed by each robot and the position information of the robot, and send them to each robot for execution; combined with the precise welding position, the continuity and accuracy of the welding can be guaranteed, which can improve the welding efficiency.

[0086] like Figure 4 As shown, the embodiment of the present application discloses a welding task execution method, which is applied to a robot, comprising:

[0087] Step S21, obtaining the welding task issued by the control center based on the initial positioning information of the robot.

[0088] In the embodiments of the present application, the robot initially processes the waiting state. After receiving the initialization message sent by the control center, it can perform state initialization, that is, prepare to receive welding tasks. The control center sends initial welding tasks to each robot based on the initial positioning information of several robots and the welding tasks to be executed. It should be noted that when different robots execute welding tasks, they can feedback the task execution situation and their own position information to other robots to coordinate the execution of welding tasks. During the process, when two robots perform data transmission, the control center can also be responsible for coordinating this communication process to ensure that only this pair of robots communicate within a specific time window to comply with the two-way communication limit of the UWB sensor.

[0089] Step S22: Execute the welding task to obtain the corresponding welding task execution result, and send the welding task execution result and the current positioning information of the robot itself to the control center, so that the control center uses a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; the welding task execution result includes the weld image and the initial position information of the corresponding welding point.

[0090] In this embodiment, after the robot executes the welding task and obtains the corresponding execution result, it can send the corresponding execution result and the current positioning information of the robot itself to the control center. The control center synchronously coordinates the relevant information of other robots to ensure a stable working environment and consistent data. Moreover, the control center corrects the position information of the welding point based on the reinforcement learning model to obtain more accurate target position information. This process can refer to the content of the above embodiments and will not be repeated here.

[0091] Step S23: Obtain the welding task that the control center re-sends to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task, and execute the corresponding welding task.

[0092] Furthermore, after the control center corrects the position information of the welding point, it can re-divide the welding tasks to be executed based on the completion situation of each welding task and re-allocate the welding tasks to each robot to ensure the welding quality and finally complete the welding task.

[0093] In a specific embodiment, as Figure 5 shown, to achieve the collaborative execution of welding tasks by multiple robots, it is first necessary to perform hardware configuration on the robots to facilitate three-dimensional positioning of the robots and weld positioning during subsequent execution of welding tasks. Combining the corresponding weld positioning and tracking algorithms to implement the execution of welding tasks and feedback relevant data to the control center, and the control center manages multiple robots to achieve collaborative welding of multiple robots.

[0094] It can be seen that in this application, multiple robots can cooperate to perform welding tasks, and the position information of the welding points fed back by the robots is corrected by the reinforcement learning model of the control center, so that a more accurate welding position can be obtained. Further, the control center can re-plan the welding tasks to be performed according to the welding tasks completed by each robot and the position information of the robots, and send them to each robot for execution. Combined with the accurate welding position, the continuity and accuracy of welding can be guaranteed, and the welding efficiency can be improved.

[0095] The following embodiments will specifically introduce the training process of the reinforcement learning model for correcting the initial position information of the welding points, which specifically includes:

[0096] To improve the accuracy of the weld position, deep reinforcement learning can be used to optimize the weld tracking process; specifically, the weld position can be obtained through a convolutional filter, and the tracking position can be corrected using a deep Q-network to ensure welding accuracy. This step continuously improves the weld tracking ability of the welding robot through the reinforcement learning model. It can be understood that due to the inevitable drift of the convolution-based filter tracker, there may be an error between the actual welding feature points and the position of the convolution filter tracker, which directly affects the quality of the welding process. To improve the accuracy of the weld feature points, reinforcement learning technology can be used here to improve the tracking position accuracy provided by the convolution filter tracker (CFT).

[0097] To train a deep reinforcement learning model, a virtual environment needs to be established. First, data preparation is carried out. The data comes from the videos recorded by the vision sensor when the robot is moving and the welding torch is open, and the feature points of the weld seam are extracted. Then, a convolutional filter tracker is implemented to track the laser line feature. Finally, the captured images (i.e., video frames) and the tracker position information are stored in the database. After the data preparation is completed, data processing steps are performed, including image preprocessing, labeling, and data augmentation, to prepare the dataset for the training process of the agent. In a specific scenario, the images can be cropped to an appropriate size of 200×200 pixels and resized to 100×100 pixels using the bilinear interpolation method with coefficients. Secondly, the collected data is labeled using the VGG Image Annotator (VIA). In addition, data augmentation is also carried out to enrich the environmental input and prevent overfitting when training the agent. The training dataset is augmented by random horizontal flipping and random rotation. In addition, considering the dangerous situations such as weld seam bursting that may occur during the welding process, random brightness contrast and Gaussian noise techniques are used to improve the robustness and performance of the model. Therefore, a welding image library and manually annotated ground truth feature points are constructed from more than 14,000 images containing position information. Subsequently, the dataset is divided into a training set and a test set, where the training data includes 10,000 images and the test set consists of the remaining images. It should be noted that all the images are from butt welds and involve different welding voltages and currents.

[0098] After completing the dataset and successfully creating the virtual environment and the deep Q - learning model, the robot needs to be trained using the action - reward mechanism and the gradient descent optimization method to maximize the reward function and minimize the loss. In the application of this specific welding task, the reward function can be expressed as:

[0099] ;

[0100] where d1 and d2 respectively represent the linear distances between the tracking points and the actual feature points before and after each action, is the reward function. After each step, the return of each action is calculated as , where, is the discount factor, and T - t + 1 represents the remaining number of steps from the current time t to the termination time T. Then, the parameter update rule is based on the gradient descent method:

[0101] ;

[0102] where, and are the old and new neural network weights respectively; represents the learning rate, that is, the hyperparameter that controls the step size when updating the parameters; is the probability of executing the action, is its gradient, and respectively represent the state at time t and the action to be executed, represents the neural network weights of the (i - 1)-th round of training. After each round, the neural network weights are updated . The change of the parameter is proportional to the product of the action return G t , and the probability gradient of the selected action, which is subdivided into the probability of executing the action. The sign and magnitude of G t determine the direction and magnitude of the parameter change. The following loss function can be mathematically transformed and optimized through the update rule in the following formula:

[0103] ;

[0104] The mathematical expression of the update rule is as follows:

[0105] ;

[0106] where, is the learning rate, is the loss function with respect to the parameter (neural network weights); r is the reward obtained by the robot after executing each action in the welding process; is the discount factor, which is used to weigh the influence of the current reward and future rewards. is the Q function represented by the old network, s and represent the current state and the current action to be executed; is the Q function represented by the new network, and respectively represent the next state and the possible actions in the next step. The deep learning model learned the policy network parameters through the above dataset. The state is randomly input, and the batch size is 32. Each round has at most 10 steps, or ends when the tracking error is less than 3 pixels (about 0.45 mm in actual coordinates in the visual system's field of view). After completing a round, a reward is returned to optimize the weights of the robot's policy network. Each training cycle contains 1000 rounds. After 16,000 cycles, the reward gradually converges, and the round length decreases. The finally trained model can be used to correct the position of the weld image and the position information of the corresponding welding points fed back by the robot, and obtain the corrected target position information.

[0107] It can be seen that by combining the laser triangulation method and the deep reinforcement learning technology, the control center can accurately locate the weld position, so as to timely adjust the welding trajectory and ensure the continuity and accuracy of welding. This automatic weld tracking greatly reduces the need for manual monitoring. At the same time, multiple robots work in parallel, speeding up the welding speed and shortening the production cycle.

[0108] As Figure 6 shown, an embodiment of the present application discloses a welding task execution device, which is applied to a control center and includes:

[0109] A first task distribution module 11, configured to distribute welding tasks to each robot respectively according to the initial positioning information of the robot;

[0110] A data acquisition module 12, configured to acquire the welding task execution results returned by each of the robots after executing the corresponding welding tasks and the current positioning information of the robots; the welding task execution results include weld images and the initial position information of the corresponding welding points;

[0111] A correction module 13, configured to perform correction processing on the welding task execution results by using a preset reinforcement learning model to obtain the target position information of the corresponding welding points;

[0112] A second task distribution module 14, configured to re-distribute welding tasks to each of the robots based on the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks, so that each of the robots executes the corresponding welding tasks.

[0113] It can be seen that the present application can simultaneously control multiple robots to cooperate in executing welding tasks, and correct the position information of the welding points fed back by the robots through the reinforcement learning model to obtain a more accurate welding position; further, the welding tasks that need to be executed can be re-planned according to the welding tasks completed by each robot and the position information of the robots, and distributed to each robot for execution; combined with the accurate welding position, the continuity and accuracy of welding can be guaranteed, and the welding efficiency can be improved.

[0114] In a specific embodiment, the correction module 13 may include:

[0115] A feature extraction unit, configured to extract features from the weld image in the welding task execution results by using the convolutional layer of a preset reinforcement learning model to obtain corresponding feature vectors;

[0116] A correction unit, configured to perform coordinate adjustment on the initial position information based on the feature vectors by using the reinforcement learning network of the preset reinforcement learning model to obtain the target position information of the corresponding welding points.

[0117] In a specific embodiment, the second task distribution module 14 may include:

[0118] A task planning unit, configured to re-plan the welding tasks to be executed corresponding to each robot according to the current positioning results of each robot and the target position information corresponding to the corresponding welding tasks;

[0119] A task distribution unit, configured to respectively distribute each welding task to be executed to the corresponding robot based on the fast Fourier transform algorithm.

[0120] As Figure 7 shown, an embodiment of the present application discloses a welding task execution device, which is applied to a robot and includes:

[0121] A first task acquisition module 21, configured to acquire the welding tasks issued by the control center based on the initial positioning information of the robot;

[0122] A data sending module 22, configured to execute the welding tasks to obtain the corresponding welding task execution results, and send the welding task execution results and its own current positioning information to the control center, so that the control center uses a preset reinforcement learning model to correct the welding task execution results to obtain the target position information of the corresponding welding points; the welding task execution results include the weld seam image and the initial position information of the corresponding welding points;

[0123] A second task acquisition module 23, configured to acquire the welding tasks re-issued by the control center to the robot based on the current positioning results of the robot and the target position information corresponding to the corresponding welding tasks, and execute the corresponding welding tasks.

[0124] It can be seen that in the present application, multiple robots can cooperate to execute welding tasks, and the position information of the welding points fed back by the robots is corrected by the reinforcement learning model of the control center, so as to obtain a more accurate welding position; further, the control center can re-plan the welding tasks to be executed according to the welding tasks completed by each robot and the position information of the robot, and distribute them to each robot for execution; combined with the accurate welding position, the continuity and accuracy of welding can be ensured, and the welding efficiency can be improved.

[0125] Furthermore, an embodiment of the present application also discloses an electronic device, Figure 8 which is a structural diagram of an electronic device 30 shown according to an exemplary embodiment, and the content in the figure should not be considered as any limitation to the scope of use of the present application.

[0126] Figure 8Schematic diagram of the structure of an electronic device 30 provided by an embodiment of the present application. The electronic device 30 may specifically include: at least one processor 31, at least one memory 32, a power supply 33, a communication interface 34, an input / output interface 35, and a communication bus 36. Among them, the memory 32 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the welding task execution method disclosed in any of the foregoing embodiments. In addition, the electronic device 30 in this embodiment may specifically be an electronic computer.

[0127] In this embodiment, the power supply 33 is used to provide operating voltage for each hardware device on the electronic device 30; the communication interface 34 can create a data transmission channel between the electronic device 30 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of the present application, and no specific limitation is imposed on it here; the input / output interface 35 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and no specific limitation is made here.

[0128] In addition, as a carrier for resource storage, the memory 32 may be a read-only memory, a random access memory, a disk, or an optical disc, etc., and the resources stored thereon may include an operating system 321, a computer program 322, etc., and the storage method may be temporary storage or permanent storage.

[0129] Among them, the operating system 321 is used to manage and control each hardware device and the computer program 322 on the electronic device 30, and it may be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the welding task execution method executed by the electronic device 30 disclosed in any of the foregoing embodiments, the computer program 322 may further include a computer program that can be used to complete other specific tasks.

[0130] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the welding task execution method disclosed above is implemented. For the specific steps of this method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details are not repeated here.

[0131] In this specification, the various embodiments are described in a progressive manner, and the key points of each embodiment are the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0132] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0133] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well known in the technical field.

[0134] Finally, it should also be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0135] The technical solutions provided in this application have been introduced in detail above. Specific examples have been used herein to illustrate the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for performing a welding task, characterized in that, Applied to a control center, including: Issuing welding tasks to each robot respectively according to the initial positioning information of the robot; Obtaining the welding task execution results returned by each of the robots after executing the corresponding welding tasks and the current positioning information of the robots; the welding task execution results include weld images and the initial position information of the corresponding welding points, and the initial position information is the position information of the welding points calculated based on the weld images; Using a preset reinforcement learning model to correct the welding task execution results to obtain the target position information of the corresponding welding points; Based on the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks, re-issuing welding tasks to each of the robots so that each of the robots executes the corresponding welding tasks; Among them, the re-issuing welding tasks to each of the robots based on the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks includes: Re-planning the welding tasks to be executed corresponding to each of the robots according to the current positioning results of each of the robots and the target position information corresponding to the corresponding welding tasks; Based on the fast Fourier transform algorithm, issuing each of the welding tasks to be executed to the corresponding robot.

2. The welding task execution method according to claim 1, characterized in that The weld image is a weld image containing a laser stripe captured by a vision sensor after a robot projects a laser stripe in a welding area after completing a corresponding welding task.

3. The welding task execution method according to claim 2, wherein, The initial position information is the initial position information of the corresponding welding points in the coordinate system of the vision sensor calculated by the robot based on the laser triangulation method and the weld image; Among them, the process of calculating the initial position information includes: Preprocessing the weld image, extracting the feature points of the laser stripe from the preprocessed weld image, and determining the intersection position of the laser plane and the weld plane according to the feature points; Calculating the initial position information of the intersection position in the three-dimensional coordinate system of the vision sensor through a pinhole camera model.

4. The welding task execution method according to claim 1, characterized in that The using a preset reinforcement learning model to correct the welding task execution results to obtain the target position information of the corresponding welding points includes: Using the convolutional layer of the preset reinforcement learning model to extract features from the weld image in the welding task execution results to obtain corresponding feature vectors; Based on the feature vectors, using the reinforcement learning network of the preset reinforcement learning model to adjust the coordinates of the initial position information to obtain the target position information of the corresponding welding points.

5. A method for performing a welding task, characterized in that, Applied to a robot, including: Obtaining the welding tasks issued by the control center based on the initial positioning information of the robot; Execute the welding task to obtain the corresponding welding task execution result, and send the welding task execution result and its own current positioning information to the control center, so that the control center can use a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; the welding task execution result includes a weld image and the initial position information of the corresponding welding point, and the initial position information is the position information of the welding point calculated based on the weld image; Obtain the welding task that the control center re-issues to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task, and execute the corresponding welding task; Among them, the obtaining of the welding task that the control center re-issues to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task includes: Send its own current positioning result and the target position information corresponding to the corresponding welding task to the control center, so that the control center can re-plan the welding tasks to be executed corresponding to each robot; Obtain the welding task to be executed that the control center issues to itself based on the fast Fourier transform algorithm.

6. A welding task execution device, characterized in that, Applied to the control center, it includes: The first task issuing module is used to issue welding tasks to each robot respectively according to the initial positioning information of the robot; The data acquisition module is used to acquire the welding task execution result returned by each robot after executing the corresponding welding task and the current positioning information of the robot; the welding task execution result includes a weld image and the initial position information of the corresponding welding point, and the initial position information is the position information of the welding point calculated based on the weld image; The correction module is used to use a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; The second task issuing module is used to re-issue welding tasks to each robot based on the current positioning result of each robot and the target position information corresponding to the corresponding welding task, so that each robot can execute the corresponding welding task; Among them, the second task issuing module includes: The task planning unit is used to re-plan the welding tasks to be executed corresponding to each robot according to the current positioning result of each robot and the target position information corresponding to the corresponding welding task; The task issuing unit is used to issue each welding task to be executed to the corresponding robot respectively based on the fast Fourier transform algorithm.

7. A welding task execution device, characterized in that, Applied to the robot, it includes: The first task acquisition module is used to acquire the welding task issued by the control center based on the initial positioning information of the robot; A data sending module, configured to execute the welding task, obtain a corresponding welding task execution result, and send the welding task execution result and its own current positioning information to the control center, so that the control center uses a preset reinforcement learning model to correct the welding task execution result to obtain the target position information of the corresponding welding point; the welding task execution result includes a weld seam image and the initial position information of the corresponding welding point, and the initial position information is the position information of the welding point calculated based on the weld seam image; A second task acquisition module, configured to acquire the welding task re-issued by the control center to the robot based on the current positioning result of the robot and the target position information corresponding to the corresponding welding task, and execute the corresponding welding task; Wherein, the process of acquiring the welding task by the second task acquisition module includes: sending its own current positioning result and the target position information corresponding to the corresponding welding task to the control center, so that the control center re-plans to obtain the to-be-executed welding tasks corresponding to each robot; acquiring the to-be-executed welding task sent by the control center to itself based on the fast Fourier transform algorithm.

8. An electronic device, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to execute the computer program to implement the welding task execution method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that, For storing a computer program, which when executed by a processor implements the welding task execution method according to any one of claims 1 to 5.

Images