Intelligent dual-gun welding apparatus, system, and welding method

Abstract

The application relates to the technical field of welding equipment, and particularly discloses an intelligent double-gun welding equipment, a system and a welding method, the welding equipment comprises automatic welding for angle iron supports, and comprises a machine body and a control module; a double-gun welding unit, a workpiece positioning device, a self-adaptive adjusting mechanism and a multi-sensor array are arranged on the machine body; the double-gun welding unit comprises two symmetrical welding guns with adjustable spacing; the welding guns are connected with the self-adaptive adjusting mechanism through mounting pieces, so that the posture of the welding guns can be adjusted in a three-dimensional space. The intelligent double-gun welding equipment, the system and the welding method of the embodiment of the application realize parallelization of the welding process through synchronous operation of the double guns, significantly shorten the welding time of single workpieces, and improve the overall production efficiency; the three-dimensionally adjustable welding gun posture adjusting mechanism cooperates with the multi-sensor array, so that accurate control of welding positions and parameters is realized.

Description

Technical Field

[0001] This invention relates to the field of welding equipment technology, and in particular to an intelligent dual-gun welding device, system and welding method. Background Technology

[0002] In the field of angle iron bracket welding, the traditional process has long adopted a serial operation mode of "positioning-single-sided welding-manual flipping-secondary welding". Each process is completed independently, which is not only time-consuming and labor-intensive, but also introduces human error through manual flipping and other operations, resulting in fluctuations in welding quality.

[0003] With the large-scale development of industries such as building steel structures, engineering machinery, and shipbuilding and repair, the demand for mass production of support components is increasing. Traditional manual welding methods can no longer meet capacity and quality requirements, and the industry urgently needs automated and intelligent welding equipment to improve efficiency and stability. Existing welding equipment has the following shortcomings:

[0004] The welding torch has insufficient control precision and low degree of automation, making it difficult to achieve high-precision synchronous welding.

[0005] The lack of real-time monitoring and adaptive adjustment mechanisms makes it impossible to dynamically adjust the welding process based on workpiece parameters. Summary of the Invention

[0006] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to provide an intelligent dual-gun welding device, system, and welding method to improve welding efficiency.

[0007] To achieve the above objectives, the present invention proposes an intelligent dual-gun welding equipment for automated welding of angle iron brackets, comprising a body and a control module, wherein the body is equipped with a dual-gun welding unit, a workpiece positioning device, an adaptive adjustment mechanism and a multi-sensor array.

[0008] The dual-gun welding unit includes two welding guns arranged symmetrically with adjustable spacing. The welding guns are connected to the adaptive adjustment mechanism through mounting components to achieve posture adjustment of the welding guns in three-dimensional space.

[0009] The workpiece positioning device includes a clamp for positioning and clamping the angle iron and the triangular plate;

[0010] The multi-sensor array includes a binocular camera and an infrared temperature sensor, used to acquire images of the welding area and the temperature distribution of the molten pool.

[0011] The control module includes a processor and a memory, in which a reinforcement learning model is stored.

[0012] In some embodiments of the present invention, the adaptive adjustment mechanism includes cylinder one, a bidirectional cylinder, and cylinder two;

[0013] The output end of cylinder one extends horizontally, and the bidirectional cylinder is installed on the output end of cylinder one, with the extension direction of the output end of the bidirectional cylinder being perpendicular to the extension direction of the output end of cylinder one.

[0014] Both output ends of the bidirectional cylinder are movably connected to the second cylinder via connectors. The output end of the second cylinder is equipped with the mounting component to achieve position adjustment of the welding torch.

[0015] In some embodiments of the present invention, the mounting component includes a fixed plate and a movable plate, the distance between the fixed plate and the movable plate is adjusted by screws and nuts, the welding torch is limited and constrained between the fixed plate and the movable plate, and a pneumatic switch is also installed at the fixed plate, the pneumatic switch receiving instructions from the control module to control the opening and closing of the welding torch.

[0016] In some embodiments of the present invention, the clamp includes a base plate, a connecting frame and an electromagnet. The base plate is fixed on the machine body, the connecting frame has an L-shaped structure, the connecting frame is disposed on the base plate, and an electromagnet is disposed at the end of the connecting frame.

[0017] In some embodiments of the present invention, the processor of the control module is used to perform the following calculations:

[0018] Surface roughness reference value ,in, It is the local roughness index. These are the weighting coefficients. This represents the total number of units in the local area.

[0019] Reference values ​​for temperature distribution differences ,in, For the first Local temperature differences in each region This represents the total number of regions.

[0020] To achieve the above objectives, the present invention also proposes an intelligent dual-gun welding system, applied to the aforementioned welding equipment, comprising:

[0021] The parameter optimization module is used to construct the state space. and action space ;

[0022] The reward function calculation module is used to calculate... ,in, , , , For single-piece welding time;

[0023] The reinforcement learning module uses the DDPG algorithm to generate welding torch adjustment actions and optimizes parameters through the Actor-Critic network.

[0024] In some embodiments of the present invention, the reinforcement learning module includes an experience replay buffer for storing state transition data. And update network parameters using the TD algorithm, when the buffer data volume Start batch training when the condition is met.

[0025] In some embodiments of the present invention, a process knowledge base module is also included, which is used to perform cluster analysis on the optimal parameters of angle irons of similar specifications, generate a specification-parameter mapping table, and improve the parameter optimization speed of new specification workpieces through meta-learning.

[0026] To achieve the above objectives, the present invention also proposes an intelligent dual-gun welding method based on the above system, comprising the following steps:

[0027] Workpiece specification recognition: Extract the geometric features of the workpiece through a binocular camera, and match them with the specification library to generate the initial welding torch spacing and current parameters;

[0028] Self-evolutionary algorithm initialization: Search the historical database. If a record of angle iron of the same specification exists, call the optimal parameters; otherwise, use the default parameters and start the exploration mechanism.

[0029] Real-time decision-making during welding process: The state space S is collected every 50ms, and the action space A is generated through the DDPG algorithm to adjust the welding torch position and current parameters. The dual torch actions are coordinated and compensated through an attention mechanism.

[0030] In some embodiments of the present invention, welding quality feedback and model update steps are also included:

[0031] After welding is completed, the reward R is calculated based on the test results, and the data is stored in the experience playback buffer.

[0032] The knowledge distillation algorithm is used to compress deep learning models, making them suitable for industrial control computer computing resources;

[0033] Increase sampling weights for abnormal operating conditions such as cold welding to improve model robustness.

[0034] In some embodiments of the present invention, during the reinforcement learning process:

[0035] state space The temperature field in the image is converted into a 16×16 pixel matrix by an infrared sensor, and the surface roughness is... Through local gradient magnitude Calculation, where The pixel grayscale value;

[0036] Action space The welding torch position adjustment accuracy is ±1mm, and the current adjustment step is ±5A.

[0037] reward function In The threshold is 20°C; if it is exceeded, parameter correction is triggered.

[0038] This invention discloses an intelligent dual-gun welding equipment, system, and welding method. By achieving parallel welding process through simultaneous operation of two guns, the welding time of a single workpiece is significantly shortened, and the overall production efficiency is improved. The modular fixture and adaptive adjustment mechanism support the rapid changeover of angle irons of various specifications, meeting the needs of flexible production. The three-dimensional adjustable welding gun posture adjustment mechanism, in conjunction with a multi-sensor array, enables precise control of welding position and parameters. Finally, by monitoring the image and temperature distribution of the welding area in real time, the welding quality is quantitatively evaluated, improving the consistency of the weld. Attached Figure Description

[0039] Figure 1 This is a perspective view of the welding equipment in this invention;

[0040] Figure 2 This is a perspective view of the mounting component in this invention;

[0041] Figure 3 This is a perspective view of the fixture in this invention;

[0042] Figure 4 This is a three-dimensional view of the workpiece in this invention;

[0043] Figure 5 This is a schematic diagram of the welding system in this invention;

[0044] Figure 6 This is a schematic diagram of the welding method in this invention.

[0045] In the diagram: 1. Body; 2. Cylinder 1; 3. Double-acting cylinder; 4. Connecting piece; 5. Cylinder 2; 6. Mounting piece; 61. Fixing plate; 62. Movable plate; 63. Screw; 64. Nut; 65. Pneumatic switch; 7. Clamp; 71. Base plate; 72. Connecting frame; 73. Electromagnet; 8. Binocular camera; 9. Triangle plate; 91. Angle iron. Detailed Implementation

[0046] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0047] The following description, with reference to the accompanying drawings, describes an intelligent dual-gun welding device, system, and welding method according to embodiments of the present invention.

[0048] Figure 1 This is a schematic diagram of the structure of an intelligent dual-gun welding device according to an embodiment of the present invention.

[0049] Example 1:

[0050] like Figure 1 , Figure 2 , Figure 3 , Figure 4 As shown, an intelligent dual-gun welding device is specially designed for the automated welding of angle iron brackets (composed of a triangular plate 9 and an angle iron 91).

[0051] The equipment mainly consists of a basic body 1, a central control module, a dual-gun welding unit, a workpiece positioning device for precise positioning, an adaptive adjustment mechanism for adjusting the welding gun posture, and a multi-sensor array for real-time monitoring.

[0052] The dual-gun welding unit consists of two symmetrical welding guns whose spacing can be adjusted as needed; the multi-sensor array integrates a binocular camera (for acquiring three-dimensional image information of the welding area) and an infrared temperature sensor (for monitoring the temperature distribution of the weld pool); the core of the control module is a processor and a memory, in which a pre-trained reinforcement learning model is stored to optimize the welding process.

[0053] In use, the operator or automated feeding system places the triangular plate 9 and angle iron 91 to be welded onto the workpiece positioning device, which is a specially designed clamp 7 used to accurately position and firmly clamp the triangular plate 9 and angle iron 91 during the welding process.

[0054] The clamp 7 includes a base plate 71, which is securely mounted on the equipment body 1 in a suitable position. One or more connecting brackets 72 are provided on the base plate 71. These connecting brackets 72 are designed in an L-shape to accommodate the right-angle shape of the angle iron and provide a preliminary positioning reference. Electromagnets 73 are provided at the ends of the connecting brackets 72 (or other key positioning / clamping positions).

[0055] The L-shaped connecting frame 72 here plays the main positioning role (forming a positioning surface), while the electromagnet 73 is the main clamping element, installed on the connecting frame 72 or the base plate 71 near the workpiece. There may also be additional adjustable positioning blocks used in conjunction with the electromagnet 73.

[0056] Before welding, the control module supplies power to the electromagnet 73, generating a strong magnetic field. Since the triangular plate 9 and angle iron 91 are typically made of ferromagnetic materials (such as carbon steel), they are firmly attracted by the electromagnet 73 and pressed against the positioning surface, thus achieving precise positioning and reliable clamping. Throughout the welding process, the electromagnet 73 remains continuously energized, maintaining the clamping force on the workpiece and preventing it from moving due to welding heat deformation or external disturbances, ensuring welding accuracy.

[0057] After the workpiece is fixed, the binocular camera 8 acquires 3D image information of the current workpiece and transmits it to the control module. The control module uses image processing algorithms to identify the position and shape of the weld to be welded. Combining the reinforcement learning model in memory and the CAD data of the workpiece, the processor plans the optimal dual-gun welding path, welding gun posture, welding speed, current, voltage, and the cooperative working mode of the two welding guns (as an example, simultaneously welding different welds, or relay welding of long welds). At the same time, the welding gun spacing is set to adapt to the specific welding task.

[0058] Subsequently, the control module sends commands to the adaptive adjustment mechanism (including cylinder 2, bidirectional cylinder 3, and cylinder 5) according to the planned path and posture. This mechanism drives the connected mounting piece 6, moving the two welding torches precisely to the starting welding position and adjusting the angle and height of the welding torches relative to the workpiece. The spacing adjustment mechanism (screw 63 and nut 64) on the mounting piece 6 can be used to fine-tune the distance between the two welding torches.

[0059] Next, the control module controls the two welding torches to start welding in a predetermined sequence or simultaneously via pneumatic switch 65. During welding, the binocular camera 8 continuously monitors the weld seam tracking, and the infrared temperature sensor monitors the molten pool temperature in real time. This real-time data is fed back to the control module. The reinforcement learning model within the control module compares the real-time sensor data (image deviation, temperature changes, etc.) with preset targets, dynamically adjusting the movement of the adaptive adjustment mechanism (fine-tuning the welding torch position and posture to accurately track the weld seam), welding parameters (such as current, voltage, and wire feed speed, achieved by controlling the welding power supply), and the start and stop of the pneumatic switch 65 to ensure welding quality, such as consistent weld penetration and width, and to avoid defects such as burn-through and incomplete penetration.

[0060] After welding is completed, the control module shuts off the welding torch, drives the adaptive adjustment mechanism to move the welding torch away, releases the workpiece positioning device, and the angle iron bracket that has been welded can be removed.

[0061] This welding equipment operates with two welding torches working simultaneously or in coordination, significantly improving welding efficiency compared to single-torch welding. It is particularly suitable for workpieces with multiple weld seams, such as angle iron brackets. Automated control and sensor feedback closed-loop adjustment reduce the uncertainty and fatigue errors of manual operation, ensuring precise tracking of the welding path and stable execution of welding parameters, thus improving the consistency of welding quality. The application of reinforcement learning models enables the equipment to adaptively optimize based on actual welding conditions, continuously learning and improving welding strategies, resulting in better adaptability to minor differences in incoming workpiece materials or environmental changes. Automated welding replaces heavy and relatively harsh manual welding work, improving the working environment, reducing reliance on skilled welders, and allowing sensor data and control parameters to be recorded, facilitating quality traceability and process analysis.

[0062] Example 2:

[0063] The adaptive adjustment mechanism in this embodiment is the core component for achieving precise three-dimensional movement of the welding torch. It consists of cylinder 2, bidirectional cylinder 3, and at least one (usually two symmetrical cylinders, one for each welding torch or sharing a drive connection) cylinder 5.

[0064] Cylinder 2 is typically mounted horizontally on a frame of the machine body 1, with its piston rod (output end) extending and retracting in a horizontal direction (e.g., the X-axis). A two-way cylinder 3 is mounted at the end of the piston rod of cylinder 2, with its output end designed to extend in a "cross-perpendicular" direction to the extension direction of cylinder 2. This means that the two-way cylinder 3 can drive its connecting components to move in a vertical direction (e.g., the Z-axis) and / or another horizontal direction (e.g., the Y-axis). The term "two-way cylinder 3" refers to its ability to drive movement in two directions.

[0065] In this structure, the output ends on both sides of the bidirectional cylinder 3 are movably connected to the second cylinder 5 via connecting parts 4 (such as connecting rods, sliders, etc.). The second cylinder 5 is usually installed vertically or at an angle, and its piston rod is equipped with a mounting part 6. The mounting part 6 is used to finally fix the welding torch, and the extension and retraction of the second cylinder 5 realizes the fine adjustment of the welding torch, especially the precise positioning in the vertical direction or at a specific angle.

[0066] like Figure 2 As shown, the mounting component 6 itself includes a fixed plate 61 and a movable plate 62. The distance between the two plates can be adjusted by screws 63 and nuts 64, thereby adjusting the distance between the two welding torches. The fixed plate 61 also integrates a pneumatic switch 65, which is used to control the start and stop of the welding torches according to the instructions of the control module. This is usually achieved by triggering the switch on the welding torch handle or by directly controlling the wire feed / power supply.

[0067] The adaptive adjustment mechanism achieves translation and possible attitude adjustment of the welding torch in three-dimensional space through the combined movement of three (or more) cylinders;

[0068] The control module sends instructions to the corresponding solenoid valves to control the compressed air to enter or exit different chambers of cylinder 2, causing its piston rod to extend or retract, thereby driving the entire subsequent mechanism (bidirectional cylinder 3, cylinder 2 5, welding torch) to move along the horizontal X-axis.

[0069] Similarly, the control module controls the movement of the bidirectional cylinder 3. Based on its specific design, the descriptions of "cross-vertical" and "both sides output" suggest that it provides Y-axis translation and / or Z-axis translation, or rotation along a certain axis.

[0070] As an example, a double-rod cylinder is mounted on the X-axis slider, and the mechanism connected to its two ends enables movement along the Y-axis; or a regular double-acting cylinder is mounted vertically to achieve movement along the Z-axis. This description needs to be understood in conjunction with the accompanying drawings, but the core concept is that cylinder 3 achieves movement in the direction perpendicular to the X-axis.

[0071] Subsequently, the control module controls the action of cylinder 2 5, whose piston rod extends and retracts to drive the mounting part 6, thereby making the final fine adjustment of the welding gun position, usually the precise feed or height adjustment in the Z-axis direction, to adapt to the slight undulations on the workpiece surface or to achieve the precise distance between the welding gun and the workpiece.

[0072] The screws 63 and nuts 64 on the mounting part 6 provide a structure for limiting the angle and position of the welding torch, while the pneumatic switch 65, after receiving the electrical signal from the control module, drives the internal mechanism to act, turning on or off the trigger signal of the welding torch, thereby controlling the start and end of the welding process.

[0073] In some embodiments of the present invention, the processor of the control module is used to perform the following calculations:

[0074] 1. Surface roughness reference value:

[0075] ,

[0076] in, : Surface roughness reference value, which quantifies the degree of unevenness of the weld joint surface. The smaller the value, the smoother the surface and the better the welding quality.

[0077] The total number of local areas divided on the surface of the weld joint is dynamically adjusted according to the size of the weld joint. For example, the weld joint of L50×L50 angle iron is divided into 16 local areas.

[0078] :No. The roughness index of a local region is calculated using the local gradient magnitude.

[0079] :No. The weighting coefficients for each local region are set according to the degree of influence of that region on the welding quality, such as the weighting of the weld center region. Edge area .

[0080] The calculation steps include:

[0081] Step 1: Image preprocessing using Gaussian filtering (kernel size) Standard deviation Remove noise from the images captured by the binocular camera (8) and enhance the edge features of the solder joints;

[0082] Step 2: Calculate the local gradient magnitude for each pixel. Calculate its local neighborhood (e.g.) Gradient magnitude within the window :

[0083]

[0084] in, For pixel grayscale values, and The horizontal and vertical gradients are calculated using the Sobel operator, respectively.

[0085] Step 3: Local roughness index Calculation for each local region Calculate its average gradient magnitude: ,in For the region The number of pixels within;

[0086] Step 4: Weighted summation according to preset weight coefficients For each region Weighted summation yields the overall surface roughness reference value. .

[0087] 2. Reference values ​​for temperature distribution differences:

[0088]

[0089] in, : Reference value for temperature distribution differences around solder joints, quantifying the non-uniformity of the temperature field; the smaller the value, the more uniform the temperature distribution.

[0090] The total number of regions into which the temperature field is divided is determined by the resolution of the infrared temperature sensor, for example... The pixel matrix is ​​divided into One region;

[0091] : No. The local temperature difference in each region is calculated by comparing it with the temperature of adjacent regions.

[0092] The calculation steps include:

[0093] Step 1: Temperature field data acquisition using an infrared temperature sensor The temperature of the solder joint and surrounding area is collected at a secondary frequency to generate... Temperature matrix of pixels Temperature measurement range precision ;

[0094] Step 2: Region Division. The temperature matrix is ​​uniformly divided into regions. Square sub-regions (e.g.) Each sub-region is [size missing] Pixel;

[0095] Step 3: Local temperature differences Calculate for each sub-region Calculate the sum of the absolute values ​​of the temperature differences between it and its adjacent regions:

[0096]

[0097] in, The temperature values ​​are the corresponding locations in adjacent areas;

[0098] Step 4: Global difference integration for all sub-regions Take the average value to get .

[0099] As an example, for angle irons of different specifications (such as L40×L40 to L100×L100), the weight coefficients can be automatically optimized through a reinforcement learning model. :

[0100] For L50×L50 angle iron, the weight of the weld center area Set to 0.12, with an edge region of 0.08;

[0101] For L100×L100 angle iron, due to the larger heat-affected zone, the weight of the central area is adjusted to 0.15, and the weight of the edge area is 0.05;

[0102] Weight update formula:

[0103]

[0104] in, The learning rate is 0.01. This is the gradient of the reward function with respect to the weights.

[0105] As an example, evaluate the welding quality of L60×L60 angle iron:

[0106] First, data acquisition: Infrared temperature sensors collected the temperature field of the molten pool, divided it into 16 regions, and calculated... The mean is Therefore (Not exceeding the threshold) The binocular camera extracts images of the solder joints, divides them into 16 local regions, and calculates... The mean is After weighting ;

[0107] Then the decision output: because and Within acceptable range (threshold) The algorithm maintains the current parameters (current). Welding torch spacing );

[0108] Finally, self-evolutionary learning: This set of data is stored in an experience replay buffer. When 500 similar data points accumulate, the model updates the weight coefficients, and the weight of the central region of the L60×L60 angle iron is adjusted from... Adjusted to To focus more on the quality of the core welding area.

[0109] To achieve the above objectives, such as Figure 5 As shown, an intelligent dual-gun welding system is also proposed here, applied to the aforementioned welding equipment, including:

[0110] 1. Parameter optimization module:

[0111] 1.1 State Space The construction, Angle iron specifications, welding torch coordinates, current parameters, temperature field. ;

[0112] Angle iron specifications: expressed in terms of side length (e.g., L50×L50). The edge features of the angle iron are identified by a binocular camera (8) and matched with the specification library with an accuracy of ±1mm.

[0113] Welding torch coordinates: three-dimensional spatial coordinates The displacement feedback of cylinder 1 (2), bidirectional cylinder (3) and cylinder 2 (5) determines the repeatability positioning accuracy to ±0.5mm.

[0114] Current parameters: Dual-gun welding current value (unit: A), range 150-300A, adjusted in real time by the control module;

[0115] Temperature field: 16×16 pixel temperature matrix acquired by infrared temperature sensor, temperature measurement range 800-1200°C, accuracy ±15°C;

[0116] Surface roughness reference value The unevenness of the solder joint surface is quantified by local gradient amplitude weighted calculation, with a threshold of 15 (the smaller the value, the better the quality).

[0117] 1.2 Action Space Definition, Welding torch position adjustment, current adjustment, welding speed correction;

[0118] Welding torch position adjustment: Three-dimensional adjustment range ±1mm, driven by an adaptive adjustment mechanism;

[0119] Current adjustment: single-step adjustment ±5A, dual-gun synchronization error ≤1.5%;

[0120] Welding speed correction: Adjustment range 10-40mm / s, step size 1mm / s.

[0121] 2. Reward Function Calculation Module:

[0122] 2.1 Analysis of the Reward Function Formula ;

[0123] Weighting coefficient definition:

[0124] The influence of surface roughness should be prioritized, with welding quality taking precedence.

[0125] The influence weight of temperature distribution differences is used to ensure uniform heat input.

[0126] The impact of welding time on weighting, balancing efficiency and quality;

[0127] Single piece welding time : Total time from workpiece positioning to welding completion (unit: s), the initial value is preset according to the specifications (e.g., 240s for L50×L50);

[0128] 2.2 Technical Logic of the Reward Mechanism

[0129] when and At that time, reward value If positive, the system enhances the current parameter;

[0130] If defects such as poor soldering occur, If the value is negative, trigger parameter correction (e.g., increase current by 5-10A).

[0131] 3. Reinforcement Learning Module:

[0132] 3.1 DDPG Algorithm Implementation Framework

[0133] Actor Network: Input State Output continuous motion The structure is a 3-layer fully connected network (256-128-64 neurons), with ReLU as the activation function;

[0134] Critic Network: Evaluating the Value of Actions Input state and action, output Q value, network structure is symmetrical with Actor;

[0135] Target network: The parameters of the main network are copied periodically to reduce training fluctuations, with an update cycle of 100 iterations.

[0136] 3.2 Experience Replay Mechanism

[0137] Buffer storage: state transition data Stored in chronological order, with a capacity of 10,000 records;

[0138] Batch training: When the data volume is ≥500 records, 64 records are randomly sampled to form a mini-batch, and the parameters are updated using the TD algorithm.

[0139]

[0140] in, As a discount factor, and For the target network.

[0141] In some embodiments of the present invention, a process knowledge base module is also included, which is used to perform cluster analysis on the optimal parameters of angle irons of similar specifications, generate a specification-parameter mapping table, and improve the parameter optimization speed of new specification workpieces through meta-learning, including the following:

[0142] Clustering analysis algorithm: K-means is used to cluster angle irons of similar specifications using optimal parameters. The distance metric formula is as follows:

[0143]

[0144] in, Let the side length of the angle iron be... This refers to the welding current.

[0145] Meta-learning optimization: By using the MAML (Model Independent Meta-Learning) algorithm, parameter transfer patterns of angle irons of different specifications are extracted, and the optimization speed of the first piece parameters of new specification workpieces is improved by 60%;

[0146] The specification-parameter mapping table is shown in the table below:

[0147] L40×L40 48mm 180A 25mm / s L50×L50 60mm 210A 22mm / s

[0148] In some embodiments of the present invention, the system further includes a dual-gun collaborative attention mechanism, wherein:

[0149] Action association weight calculation:

[0150] When temperature difference hour, The synchronization compensation for dual-gun actions is ≥80%;

[0151] Example of coordinated compensation: When the left welding torch moves up by 0.5mm, the right welding torch adjusts its weight accordingly. Automatic execution Mirror compensation ensures balanced heat input.

[0152] As an example, here is an optimized welding process for L60×L60 angle iron:

[0153] Initial state:

[0154]

[0155] initial action ;

[0156] Process optimization:

[0157] During the 5th welding, Reward function output Trigger parameter adjustment:

[0158] The welding current was increased by 10A for each of the two guns, and the welding speed was reduced to 18mm / s.

[0159] Cylinder 2 (5) drives the welding torch to move upward by 0.5mm, expanding the heat-affected zone;

[0160] After the 10th welding, the amount of data in the experience playback buffer reached 500, batch training was started, the Actor network parameters were updated, and the welding torch spacing was optimized to 75mm.

[0161] Final result:

[0162] The time per item has been reduced from 280 seconds to 240 seconds;

[0163] The process knowledge base is automatically updated with the optimal parameters for L60×L60, providing a basis for transfer learning for L65×L65.

[0164] This system dynamically optimizes welding parameters based on a reinforcement learning algorithm framework, improving welding quality without manual adjustments. Furthermore, the accumulation of experience data and model iteration continuously enhance the system's adaptability to new workpiece specifications. By comprehensively evaluating welding quality and efficiency through a quantified reward function, the system achieves dynamic parameter correction and supports mass production.

[0165] To achieve the above objectives, such as Figure 6 As shown, an intelligent dual-gun welding method based on the above system is also proposed, including the following steps:

[0166] 1. Workpiece specification recognition: Extract the geometric features of the workpiece through a binocular camera (8), match the specification library to generate the initial welding torch spacing and current parameters, specifically including:

[0167] 1.1 The binocular visual images of the angle iron bracket are acquired by a binocular camera (8), and point cloud data is generated by a binocular stereo matching algorithm to extract geometric features:

[0168] Angle iron side length The formula for calculating the 3D coordinate difference of point cloud edge points is:

[0169]

[0170] Recognition accuracy ±1mm;

[0171] Angle iron thickness Extract the plane equations of the point cloud from the two sides of the angle iron:

[0172]

[0173] The formula for calculating the interplanar spacing as thickness is as follows: Accuracy ±0.5mm.

[0174] 1.2 Specification Library Matching Logic:

[0175] The specification library pre-stores typical angle iron parameters (such as L40×40×4, L50×50×5, etc., where "L" is the side length and "×" is the thickness), and matches them using the K-Nearest Neighbor (KNN) algorithm:

[0176] Calculate the Euclidean distance between the current angle iron feature and the specification library.

[0177]

[0178] Select The smallest specification is used as the matching result;

[0179] If the distance exceeds the threshold (e.g., 10mm), it is determined to be a new specification, triggering the "new specification exploration mechanism".

[0180] 2. Self-evolutionary algorithm initialization: Search the historical database. If a record of the same specification angle iron exists, the optimal parameters are used; otherwise, the default parameters are used and the exploration mechanism is initiated. This includes:

[0181] 2.1 Historical Database Structure:

[0182] The database stores fields including: angle iron specifications ( Welding parameters (welding torch spacing) Current ,Voltage ,speed Quality indicators (surface roughness) Temperature differences (Defect type).

[0183] 2.2 Optimal Parameter Calling Rules: Multi-objective optimization retrieval is adopted, simultaneously satisfying:

[0184] Surface roughness (Industry-leading quality standards);

[0185] Temperature difference (Requirement for uniformity of heat input);

[0186] Welding efficiency (speed) ≥ industry average level (e.g., 20mm / s).

[0187] 2.3 Default Parameters and Exploration Mechanism:

[0188] Default parameters: based on the side length of the angle iron Calculate the welding torch spacing. (Empirical coefficient), current (Unit: A) Unit: mm);

[0189] Exploration Mechanism: An ε-greedy strategy is adopted, with an initial exploration rate ε = 0.8 (i.e., 80% probability of trying new parameters), which decays exponentially with the number of iterations (ε = 0.8 × 10⁻⁶). (where n is the number of iterations).

[0190] 3. Real-time decision-making during welding: The state space S is collected every 50ms, and the action space A is generated through the DDPG algorithm. The welding torch position and current parameters are adjusted. The dual-torch actions are coordinated and compensated through an attention mechanism, specifically including:

[0191] 3.1 State Space Precise definition:

[0192]

[0193] in, Angle iron side length and thickness (specification identification results);

[0194] Real-time three-dimensional coordinates of dual welding guns (feedback from cylinder displacement sensor, accuracy ±0.1mm);

[0195] Real-time welding current of dual welding guns (acquired by current sensor, accuracy ±1A);

[0196] Temperature field matrix (16×16 pixels, acquired by infrared sensor, temperature measurement range 800-1200°C, accuracy ±10°C).

[0197] Surface roughness (calculated by local gradient magnitude).

[0198] 3.2 Technical Implementation of the DDPG Algorithm:

[0199] Actor network: Input state S, output action A (welding gun position, current adjustment), the network structure is a 3-layer fully connected layer (input layer → hidden layer 1 → hidden layer 2 → output layer), and the activation function is ReLU;

[0200] Critic network: evaluates action value Q(S, A), has a structure symmetric to Actor, and outputs Q value to guide parameter updates;

[0201] Experience replay buffer: Stores (S, A, R, S') data (R is the reward, S' is the next state), with a capacity of 10,000 records and a sampling batch of 64 records / time.

[0202] 3.3 Cooperative compensation of attention mechanisms:

[0203] Define the dual-gun synergy coefficient α, and calculate it as follows:

[0204] α=

[0205] Where k=0.1 (adjustment coefficient, which makes α change linearly with temperature difference).

[0206] when At >20°C, α>0.86, and the movement of both guns is synchronized (e.g., if the left gun moves up 1mm, the right gun moves up α×1mm).

[0207] In some embodiments of the present invention, welding quality feedback and model update steps are also included:

[0208] 1. After welding is completed, calculate the reward R based on the test results and store the data in the experience playback buffer;

[0209] Complete calculation of the reward function R:

[0210]

[0211] λ1=0.6 (surface roughness weight, quality priority);

[0212] λ2=0.3 (temperature difference weight, thermal uniformity requirement);

[0213] λ3 = 0.1 (efficiency weight, balancing production cycle time);

[0214] t: Single piece welding time (unit: s, the time from workpiece positioning to welding completion).

[0215] 2. The deep learning model is compressed using a knowledge distillation algorithm to adapt to the computing resources of industrial control computers;

[0216] 3. Increase sampling weights for abnormal working conditions such as cold welding to improve model robustness.

[0217] As an example, defect data such as cold solder joints and undercut are marked as "abnormal labels," and the sampling weights are... (Normal data weight) The model's ability to identify and correct defects is improved through weighted experience replay.

[0218] In some embodiments of the present invention, during reinforcement learning:

[0219] state space The temperature field in the image is converted into a 16×16 pixel matrix by an infrared sensor, and the surface roughness is... Through local gradient magnitude Calculation, where The pixel grayscale value;

[0220] Action space The welding torch position adjustment accuracy is ±1mm, and the current adjustment step is ±5A; reward function In The threshold is 20°C; if it is exceeded, parameter correction is triggered.

[0221] As an example, the welding process for L50×50×5 angle iron is shown:

[0222] Workpiece specification recognition: Images are acquired by binocular cameras, and point computing yields L=50.2mm, d=4.8mm, matching the specification library as L50×50×5;

[0223] Initial parameters: welding torch spacing (1.2×50), current (20×50);

[0224] Self-evolutionary algorithm initialization: The optimal parameters for L50×50×5 are retrieved from the historical database: , Call this parameter;

[0225] Exploration rate (The first welding attempt leaves room for exploration);

[0226] Real-time decision-making during welding process: The status S is collected every 50ms, and the DDPG algorithm outputs the action: welding torch spacing is fine-tuned by -1mm (to 57mm), and the current is increased by 5A (to 985A).

[0227] Due to temperature differences Attention Coordination Coefficient Dual-gun synchronous compensation of 0.9mm;

[0228] Quality Feedback and Model Updates: Rewards (Requires optimization);

[0229] Data is stored in a buffer, and after knowledge distillation, the student model's reasoning speed is increased to 20ms / time.

[0230] The weight of the faulty solder joint data was set to 2.0 to enhance model learning.

[0231] This welding method automatically identifies workpiece specifications and matches optimal welding parameters, simplifying the production preparation process. The real-time decision-making mechanism dynamically adjusts the process based on the welding status, ensuring process stability. By quickly responding to abnormal situations during the welding process, it reduces the defect rate. Learning from historical data enhances the system's ability to identify and correct various defects. Furthermore, it supports rapid switching between multiple workpiece specifications to adapt to diverse production needs. The continuous accumulation and optimization of process knowledge drives the iterative upgrading of the welding method.

[0232] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, solder, propagate, or transmit a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, the computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0233] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0234] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0235] 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0236] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent dual-gun welding device for automated welding of angle iron brackets, characterized in that, It includes a body (1) and a control module. The body (1) is equipped with a dual-gun welding unit, a workpiece positioning device, an adaptive adjustment mechanism and a multi-sensor array. The dual-gun welding unit includes two welding guns arranged symmetrically with adjustable spacing. The welding guns are connected to the adaptive adjustment mechanism through the mounting component (6) to realize the attitude adjustment of the welding guns in three-dimensional space. The workpiece positioning device includes a clamp (7) for positioning and clamping the angle iron (9) and the triangular plate (91); The multi-sensor array includes a binocular camera (8) and an infrared temperature sensor, used to acquire images of the welding area and the temperature distribution of the molten pool; The control module includes a processor and a memory, and the memory stores a reinforcement learning model. The adaptive adjustment mechanism includes cylinder one (2), bidirectional cylinder (3), and cylinder two (5); The output end of the cylinder (2) extends horizontally, and the bidirectional cylinder (3) is installed on the output end of the cylinder (2), and the extension direction of the output end of the bidirectional cylinder (3) is perpendicular to the extension direction of the output end of the cylinder (2). The output ends on both sides of the bidirectional cylinder (3) are movably connected to the cylinder two (5) through the connector (4). The output end of the cylinder two (5) is provided with the mounting part (6) to realize the position adjustment of the welding gun. The mounting component (6) includes a fixed plate (61) and a movable plate (62). The distance between the fixed plate (61) and the movable plate (62) is adjusted by screws (63) and nuts (64). The welding torch is limited and constrained between the fixed plate (61) and the movable plate (62). A pneumatic switch (65) is also installed at the fixed plate (61). The pneumatic switch (65) receives instructions from the control module and then controls the opening and closing of the welding torch. The processor of the control module is used to perform the following calculations: Surface roughness reference value ,in, It is the local roughness index. These are the weighting coefficients. This represents the total number of units in the local area. Reference values ​​for temperature distribution differences ,in, For the first Local temperature differences in each region Total number of regions; The reinforcement learning model within the control module compares real-time sensor data with a preset target to dynamically adjust the motion and welding parameters of the adaptive adjustment mechanism, as well as the start and stop of the pneumatic switch (65).

2. The welding equipment according to claim 1, characterized in that, The clamp (7) includes a base plate (71), a connecting frame (72) and an electromagnet (73). The base plate (71) is fixed on the machine body (1). The connecting frame (72) has an L-shaped structure and is set on the base plate (71). An electromagnet (73) is provided at the end of the connecting frame (72).

3. An intelligent dual-gun welding system, applied to the welding equipment according to any one of claims 1-2, characterized in that, include: The parameter optimization module is used to construct the state space. and action space ; The reward function calculation module is used to calculate... ,in, , , , For single-piece welding time; The reinforcement learning module uses the DDPG algorithm to generate welding torch adjustment actions and optimizes parameters through the Actor-Critic network.

4. The welding system according to claim 3, characterized in that, The reinforcement learning module includes an experience replay buffer for storing state transition data. And update network parameters using the TD algorithm, when the buffer data volume Start batch training when the condition is met.

5. The welding system according to claim 3, characterized in that, It also includes a process knowledge base module, which is used to perform cluster analysis on the optimal parameters of angle irons of similar specifications, generate a specification-parameter mapping table, and improve the parameter optimization speed of new specification workpieces through meta-learning.

6. An intelligent dual-gun welding method based on the system described in any one of claims 3-5, characterized in that, Includes the following steps: workpiece Specification recognition: Extract the geometric features of the workpiece through a binocular camera (8), and generate the initial welding gun spacing and current parameters by matching the specification library; Self-evolutionary algorithm initialization: Search the historical database. If a record of angle iron of the same specification exists, call the optimal parameters; otherwise, use the default parameters and start the exploration mechanism. Real-time decision-making during welding process: The state space S is collected every 50ms, and the action space A is generated through the DDPG algorithm to adjust the welding torch position and current parameters. The dual torch actions are coordinated and compensated through an attention mechanism.

7. The welding method according to claim 6, characterized in that, It also includes welding quality feedback and model update steps: After welding is completed, the reward R is calculated based on the test results, and the data is stored in the experience playback buffer. The knowledge distillation algorithm is used to compress deep learning models, making them suitable for industrial control computer computing resources; Increase the sampling weight for abnormal welding conditions to improve the robustness of the model.

8. The welding method according to claim 6, characterized in that, During the reinforcement learning process: state space The temperature field in the image is converted into a 16×16 pixel matrix by an infrared sensor, and the surface roughness is... Through local gradient magnitude Calculation, where The pixel grayscale value; Action space The welding torch position adjustment accuracy is ±1mm, and the current adjustment step is ±5A. reward function In The threshold is 20°C; if it is exceeded, parameter correction is triggered.

Images