Self-adaptive intelligent welding system based on dynamic coefficient model

Abstract

The invention relates to the technical field of automatic welding, in particular to a self-adaptive intelligent welding system based on a dynamic coefficient model. The system comprises an information acquisition module used for acquiring point cloud data of a workpiece to be welded and extracting geometric and physical characteristics; the welding seam recognition module is used for judging the optimal welding seam type; the process parameter generation module is used for activating the single-channel fillet weld, multi-layer and multi-channel or groove welding sub-module through a process calculation agent and generating initial process parameters by utilizing a built-in dynamic coefficient model; the welding execution and feedback module is used for issuing a process instruction and performing real-time fine adjustment on voltage, current or speed based on a sensing signal linkage process coefficient; and the incremental learning module is used for continuously updating the model coefficient by analyzing the bias of the positive sample data packet. The full-closed-loop intelligent welding method solves the problems that traditional welding depends on a static expert database and cannot cope with assembly errors, complex heat accumulation and the like, and full-closed-loop intelligent welding from geometric perception to parameter generation and real-time correction is achieved.

Description

An Adaptive Intelligent Welding System Based on a Dynamic Coefficient Model Technical Field

[0001] This invention relates to the field of automated welding technology, and in particular to an adaptive intelligent welding system based on a dynamic coefficient model. Background Technology

[0002] Currently, welding robot systems generally operate using a pre-established welding process library. Their workflow typically involves: capturing basic information such as weld joint type, plate thickness, and gap using vision or other sensors; then matching this information with a vast database of pre-stored process parameters, or allowing the user to manually select parameters based on experience.

[0003] This traditional model has the following inherent drawbacks: 1. The process library is large and bloated: In order to cover various possible welding scenarios, such as different plate thickness combinations, joint types, materials, bevel angles, etc., it is necessary to establish and maintain an extremely large process parameter database, which is complex and has high maintenance costs.

[0004] 2. Poor applicability and rigidity: The dimensions of actual welds are continuously changing, such as the weld leg of a fillet weld and the actual width of the bevel, while the pre-stored process parameters are discrete. The system cannot accurately match all actual working conditions, often resulting in situations where data exists but is unusable, leading to unstable welding quality.

[0005] 3. Lack of true self-adaptive capability: Traditional systems cannot make dynamic, closed-loop process adjustments based on real-time physical phenomena during the welding process, resulting in a low level of intelligence.

[0006] 4. System fragmentation and integration difficulties: The generation modules for advanced processes such as multi-layer multi-pass welding and complex bevel welding are often independent, and data and logic cannot be communicated, making it impossible to form a unified and intelligent process decision-making core. Summary of the Invention

[0007] To overcome the above shortcomings, this invention provides an adaptive intelligent welding system based on a dynamic coefficient model. The system aims to quantify the welding process principle into calculable and adjustable fitting coefficients and process coefficients through a core dynamic coefficient model, thereby achieving intelligent, rapid, and adaptive generation of welding processes.

[0008] In a first aspect, the present invention provides the following technical solution: an adaptive intelligent welding system based on a dynamic coefficient model, comprising: an information acquisition module, used to scan the workpiece to be welded through a visual sensing device, acquire three-dimensional point cloud data, and extract the geometric and physical features of the workpiece to be welded; a weld identification module, used to automatically determine the optimal weld type of the workpiece to be welded based on the geometric and physical features; a process parameter generation module, used to activate the corresponding calculation sub-module according to the optimal weld type through a process calculation intelligence agent, and call the built-in dynamic coefficient model and benchmark algorithm to generate welding process parameters; the calculation sub-module includes a single-pass fillet weld intelligent generation sub-module, a multi-layer multi-pass fillet weld formula sub-module, and a bevel weld programmable sub-module; a welding execution and feedback module, used to send the welding process parameters to the welding execution equipment for operation, and fine-tune the parameters in conjunction with the process coefficient based on the real-time physical phenomena feedback during the welding process; and an incremental learning module, used to record and collect welding success data, and continuously optimize and update the dynamic coefficient model through an incremental update algorithm.

[0009] Preferably, the three-dimensional point cloud data is acquired through a binocular vision or depth vision sensor; the geometric and physical characteristics of the workpiece to be welded include, but are not limited to, joint type, plate thickness, gap, bevel angle, and bevel width.

[0010] Preferably, determining the optimal weld type of the workpiece to be welded includes: performing cross-sectional slicing processing on the three-dimensional point cloud data, calculating the cross-sectional features of the weld area, wherein the cross-sectional features include gradient variation, width and depth features; inputting the cross-sectional features into a pre-trained classification model to automatically determine that the workpiece to be welded belongs to any one of single-pass fillet weld, multi-layer multi-pass fillet weld or bevel weld.

[0011] Preferably, the execution steps of the single-pass fillet weld intelligent generation submodule include: mapping the target weld leg size to preliminary current and voltage based on the basic welding process formula, and introducing a transition interval compensation coefficient to smooth the process parameters of the weld transition interval; fine-tuning the voltage, wire feed speed or welding speed by linking the arc stability correction coefficient and the penetration and forming control coefficient, and automatically allocating the welding speed and oscillation parameters under the guidance of welding quality and efficiency targets according to the weight of the quality-efficiency balance coefficient.

[0012] Preferably, the execution steps of the multi-layer multi-pass fillet weld formula submodule include: automatically allocating the cross-sectional area and welding sequence of each weld pass according to the total deposited metal amount, and using the heat input distribution coefficient to provide differentiated settings for welding parameters of different layers and passes to control heat accumulation; adjusting the geometric ratio of each weld pass through the slump control coefficient to prevent metal from flowing down, optimizing the fusion between the cap weld pass and the base material using the joint control coefficient, and automatically calculating the stacking position of each weld pass in combination with the weld pass distribution ratio coefficient.

[0013] Preferably, the execution steps of the bevel weld procedural submodule include: using the cross-sectional shape fitting coefficient to convert the irregular bevel cross-section obtained by scanning into a standardized geometric shape to calculate the total amount of filler metal; after determining the root pass process, automatically calculating the number of filler weld passes and process parameters according to the preset cross-sectional area of ​​each weld and the width of the bevel surface, and generating a welding process tree in combination with the joint control coefficient.

[0014] Preferably, the parameter fine-tuning step of the linkage process coefficient includes: acquiring the sensing signal of the welding arc or molten pool in real time and converting it into a state parameter reflecting the welding stability; calculating the deviation between the state parameter and the theoretical process model, and calling the corresponding process coefficient as the adjustment step size operator; and superimposing the calculated deviation increment into the currently executed welding parameters to achieve dynamic correction of voltage, current or wire feed speed.

[0015] Preferably, the step of continuous optimization and updating through incremental update algorithm includes: recording the geometric and physical feature data, welding process parameter data, and parameter fine-tuning data during the welding process for each welding task; collecting weld formation quality data after welding is completed, and marking welding tasks that are deemed qualified as positive sample data packets; using an incremental learning algorithm to analyze the correction bias of the parameter fine-tuning data in the positive sample data packets on the initial process parameters, and calculating the update gradient of each coefficient in the dynamic coefficient model; and automatically updating the initial default value of each coefficient in the dynamic coefficient model according to the update gradient.

[0016] Secondly, the present invention provides the following technical solution: an adaptive intelligent welding method based on a dynamic coefficient model, comprising: scanning the workpiece to be welded using a visual sensing device to acquire three-dimensional point cloud data and extracting the geometric and physical features of the workpiece; automatically determining the optimal weld type of the workpiece based on the geometric and physical features; activating the corresponding calculation submodule according to the optimal weld type through a process calculation intelligence agent, and calling the built-in dynamic coefficient model and benchmark algorithm to generate welding process parameters; the calculation submodule includes a single-pass fillet weld intelligent generation submodule, a multi-layer multi-pass fillet weld formulaization submodule, and a bevel weld programming submodule; sending the welding process parameters to the welding execution equipment for operation, and fine-tuning the parameters in conjunction with the process coefficients based on real-time physical phenomena feedback during the welding process; recording and collecting successful welding data, and continuously optimizing and updating the dynamic coefficient model through an incremental update algorithm.

[0017] The present invention has the following beneficial effects: 1. De-stocking and intelligentization: It completely gets rid of the dependence on a large and rigid process library, and elevates the welding process from data matching to a new level of intelligent computing.

[0018] 2. Highly adaptive: Through a dynamic coefficient model, the system can accurately adapt to various continuously changing actual joint dimensions and can dynamically adjust according to real-time feedback from the welding process.

[0019] 3. Integration and Modularization: It unifies three major scenarios—single-pass, multi-layer multi-pass, and bevel welds—under a single core algorithm, solving the problem of system fragmentation. Simultaneously, as an independent module, it can be easily integrated into existing vision welding robots.

[0020] 4. User-friendly and efficient: It greatly simplifies user operation, transforming a massive selection into one-click generation, significantly improving process programming efficiency and reducing reliance on operator experience.

[0021] 5. Evolvability: The coefficient system reserves a standard interface for future integration of more advanced sensing technologies and control algorithms. The system can continuously optimize the coefficient values ​​through machine learning and achieve self-evolution. Attached Figure Description

[0022] Figure 1 is a framework diagram of an adaptive intelligent welding system based on a dynamic coefficient model provided in an embodiment of the present invention; Figure 2 is a schematic diagram of the parameter fine-tuning steps of the linkage process coefficient provided in an embodiment of the present invention; Figure 3 is a schematic diagram of the continuous optimization and updating steps of the incremental update algorithm provided in an embodiment of the present invention; Figure 4 is a flowchart of an adaptive intelligent welding method based on a dynamic coefficient model provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Embodiments

[0024] In the first embodiment of the present invention, the present invention provides an adaptive intelligent welding system based on a dynamic coefficient model, as shown in Figure 1, comprising: an information acquisition module, used to scan the workpiece to be welded through a visual sensing device, acquire three-dimensional point cloud data and extract the geometric and physical features of the workpiece to be welded; preferably, the three-dimensional point cloud data is acquired through a binocular vision or depth vision sensor; the geometric and physical features of the workpiece to be welded include, but are not limited to, joint type, plate thickness, gap, bevel angle, and bevel surface width.

[0025] Specifically, a binocular vision sensor or a laser structured light depth vision sensor is deployed on the end effector or fixed support of the welding robot. The sensor emits structured light onto the workpiece surface or uses the principle of binocular parallax to capture the original image of the weld area. Through an internal triangulation algorithm, the image coordinates are converted into three-dimensional spatial coordinates, generating three-dimensional point cloud data containing high-precision spatial information. Due to the presence of metal reflections, spatter residues, or workpiece rust in the welding environment, the original point cloud needs to be preprocessed, including using median filtering or Gaussian filtering to remove isolated noise points, and converting the point cloud data in the sensor coordinate system to the robot tool coordinate system to ensure that the extracted geometric dimensions are consistent with the robot's motion trajectory.

[0026] The system utilizes a built-in cross-sectional feature extraction algorithm to perform real-time slice analysis on the processed point cloud: Joint type identification (Type): By analyzing the cross-sectional topology of the point cloud in the direction perpendicular to the weld, the system determines whether the joint is butt joint, corner joint, or lap joint; Plate thickness extraction (T): The system identifies the height difference between the two base materials on both sides of the weld and calculates their respective plate thickness values; Gap extraction (G): The system accurately locates the edge points at the roots of the two base materials and calculates their shortest straight-line distance; Bevel angle extraction (α): For bevel welds, the system calculates the included angle between the two bevel surfaces by fitting the point set of the bevel sidewalls; Bevel width (W) and blunt edge extraction (P): The system identifies the inflection point position at the bevel opening to obtain the width; and obtains the blunt edge by locating the vertical height of the bevel root.

[0027] The extracted features are not output in a scattered manner, but are encapsulated into structured data packets by the system. For example, for a bevel welding task, the generated data packet format is as follows: {Type:V-Groove;T:12.0mm;G:2.1mm;α:60°;P:1.5mm;W:18.5mm}.

[0028] Through the above implementation methods, the accurate conversion from 3D point cloud data to structured geometric features is achieved, providing continuous and accurate input parameters for subsequent dynamic coefficient models.

[0029] A weld seam recognition module is used to automatically determine the optimal weld seam type of the workpiece to be welded based on the geometric and physical features. Preferably, determining the optimal weld seam type of the workpiece to be welded includes: performing cross-sectional slicing processing on the three-dimensional point cloud data, calculating the cross-sectional features of the weld seam region, wherein the cross-sectional features include gradient variation, width and depth features; inputting the cross-sectional features into a pre-trained classification model to automatically determine that the workpiece to be welded belongs to any one of a single-pass fillet weld, a multi-layer multi-pass fillet weld, or a bevel weld.

[0030] Specifically, after acquiring the 3D point cloud data, the system performs equidistant slicing along the normal direction of the weld centerline to extract the 2D cross-sectional contour point cloud. By calculating the gradient change between adjacent point clouds, the system locates the turning points in the cross-section and identifies the distance between the leftmost and rightmost effective points in the cross-section as the width feature; the system calculates the vertical displacement between the lowest point and the horizontal planes of the base material on both sides as the depth feature.

[0031] The extracted gradient, width, and depth are encapsulated into feature vectors and input into a pre-trained classification model, such as a Support Vector Machine (SVM), Random Forest, or Lightweight Convolutional Neural Network. If the cross-sectional features show a typical "L" or "T" shaped topology and the target size is within the preset single-pass deposition limit, it is determined to be a single-pass fillet weld. If the joint type is a corner joint, but the target weld leg size or plate thickness exceeds the single-pass welding limit, it is determined to be a multi-layer, multi-pass fillet weld. If the cross-sectional features show obvious "V" or "U" shaped beveling features and the depth features are significant, it is determined to be a bevel weld.

[0032] Through the above implementation methods, the system can autonomously identify the workpiece configuration, solving the rigid problem of traditional welding robots requiring manual selection of process numbers.

[0033] The process parameter generation module is used to activate the corresponding calculation sub-module according to the optimal weld type through the process calculation intelligence agent, and call the built-in dynamic coefficient model and benchmark algorithm to generate welding process parameters. The calculation sub-module includes a single-pass fillet weld intelligent generation sub-module, a multi-layer multi-pass fillet weld formulaization sub-module, and a groove weld programming sub-module. Preferably, the execution steps of the single-pass fillet weld intelligent generation sub-module include: mapping the target weld leg size to the initial current and voltage based on the basic welding process formula, and introducing a transition interval compensation coefficient to smooth the process parameters of the weld transition interval; fine-tuning the voltage, wire feed speed or welding speed by linking the arc stability correction coefficient and the penetration and forming control coefficient, and automatically allocating the welding speed and oscillation parameters under the guidance of welding quality and efficiency targets according to the weight of the quality-efficiency balance coefficient.

[0034] First, the thickness of the base material is received by the vision system. And the connector type, combined with the target solder joint size selected by the user through the interactive interface. The system is based on fundamental welding process formulas, such as... Preliminary calculations are performed. Based on the cross-sectional area requirement of the target weld and the heat sink characteristics of the base metal, the reference current required to maintain a stable molten pool is calculated. With reference voltage To address quality fluctuations caused by thermal imbalance at the weld initiation and termination points and joints, a transition range compensation coefficient is introduced into the system. In actual implementation, the system utilizes... Process parameters are smoothed and compensated for in the nonlinear transition range of the arc initiation and termination segments. For example, during the arc initiation stage, As a heat input increment operator, it can compensate for the heat loss of the cold base material in the initial stage and prevent the formation of incomplete fusion defects at the beginning; in the arc termination stage, it can... By controlling the linear decay of the current, the crater is fully filled, thereby achieving a high degree of consistency in the physical formation of the entire weld.

[0035] After generating the initial process flow, the system dynamically fine-tunes the voltage, wire feed speed, or welding speed by using a linked process coefficient. For process stability correction, the system calls upon the arc stability correction coefficient in real time. The system corrects voltage fluctuations based on feedback signals to ensure stable arc combustion under various dry-elongation conditions. For penetration and forming control, the system calls upon penetration and forming control coefficients. The ratio of wire feed speed to welding speed is constrained to ensure the root penetration of the weld while optimizing the wettability of the weld pool surface, resulting in a smooth transition at the weld edges. Finally, based on the quality-efficiency balance coefficient... The weights are automatically allocated to execution parameters under the guidance of preset goals. Users can fine-tune these parameters through the human-computer interaction interface. At this time, the system will recalculate the weighting of welding speed and oscillation parameters. If efficiency is prioritized, the system will utilize [resources] within the quality threshold. Increase the welding speed and adapt to the high-frequency oscillation command; if a better shape is desired, reduce the speed and increase the oscillation amplitude.

[0036] Through the above implementation methods, the system transforms the static target size into a dynamic mathematical calculation process, and uses compensation coefficients and balance operators to replace manual trial and error, thereby achieving adaptive and smooth parameter transition of single-pass fillet welds under complex start-up and shutdown conditions.

[0037] Preferably, the execution steps of the multi-layer multi-pass fillet weld formula submodule include: automatically allocating the cross-sectional area and welding sequence of each weld pass according to the total deposited metal amount, and using the heat input distribution coefficient to provide differentiated settings for welding parameters of different layers and passes to control heat accumulation; adjusting the geometric ratio of each weld pass through the slump control coefficient to prevent metal from flowing down, optimizing the fusion between the cap weld pass and the base material using the joint control coefficient, and automatically calculating the stacking position of each weld pass in combination with the weld pass distribution ratio coefficient.

[0038] Specifically, the total amount of deposited metal is first calculated based on the spatial dimensions obtained visually, and then the heat input distribution coefficient is used. Automatically allocate the theoretical cross-sectional area and welding sequence of each weld pass. The coefficient determines the thickness ratio between the base coat, filler, and top coat layers. The system is based on... A cross-sectional distribution model is constructed to ensure that the amount of weld deposited in each weld bead matches the support requirements of subsequent weld beads, thereby establishing a stable interlayer physical structure.

[0039] During continuous welding, the system utilizes the heat input distribution coefficient Differentiated dynamic settings are implemented for welding parameters of different layers and passes. As the number of layers increases, the overall temperature of the workpiece rises, and the system calls... The correction factor automatically reduces the current intensity of the capping layer or simultaneously increases the welding speed. By adjusting the slump control coefficient, the system can counteract problems such as excessively large molten pools or coarse grains caused by heat accumulation, ensuring that the mechanical properties of multi-layer welding remain consistent with those of single-pass welding. To address the instability of large molten pools under gravity, the system utilizes a slump control coefficient... By adjusting the aspect ratio of each weld bead and optimizing the voltage and current matching balance, the surface tension of the molten pool is increased, preventing weld sagging or shape distortion caused by molten metal flow. At the junction of the cap weld and the base metal, the system activates the joint control coefficient. Adjusting the oscillation parameters or heat input increment ensures a smooth transition between the capping layer and the base material, eliminating undercut defects. The system automatically calculates the spatial stacking coordinates of each weld bead by combining the weld bead distribution ratio coefficient. Using this ratio coefficient, along with the actual surface height and lateral span of the previous weld, the system dynamically generates the trajectory offset for the current weld bead.

[0040] Through the above implementation methods, a heat input distribution coefficient is introduced for global cross-sectional planning, and the following methods are utilized: The problems of heat accumulation, metal collapse and boundary fusion were solved respectively, realizing the leap from "experience stacking" to "parametric precise filling" for large-size welds.

[0041] Preferably, the execution steps of the bevel weld procedural submodule include: using the cross-sectional shape fitting coefficient to convert the irregular bevel cross-section obtained by scanning into a standardized geometric shape to calculate the total amount of filler metal; after determining the root pass process, automatically calculating the number of filler weld passes and process parameters according to the preset cross-sectional area of ​​each weld and the width of the bevel surface, and generating a welding process tree in combination with the joint control coefficient.

[0042] Specifically, due to limitations in workpiece machining accuracy, the raw 3D point cloud data acquired through scanning often exhibits asymmetrical or irregular cross-sectional shapes. The system utilizes cross-sectional shape fitting coefficients... The system performs an equivalent transformation on the irregular bevel section obtained from the scan. The coefficients automatically identify the root gap, bevel angle, and sidewall straightness of the bevel, fit them into a standardized geometric model, and smooth out noise points and minor bumps at the edges, thereby accurately calculating the total amount of filler metal required for the irregular area.

[0043] After determining the root pass process, the system automatically calculates the required number of filler passes based on the preset target cross-sectional area of ​​each weld and the width of the bevel surface of the current layer. For example, when the bevel width exceeds the coverage range of a single pass, the system will automatically switch to a parallel multi-pass strategy and adjust the current and voltage references in real time according to the depth position of each layer.

[0044] When generating complex fill instruction streams, the system introduces a cohesion control coefficient. ,use The dwell time and oscillation amplitude of the electric arc at the bevel sidewall are dynamically adjusted to optimize the bonding quality between the filler layer and the bevel sidewall. Combined with... geometric output and Due to process constraints, the system ultimately synthesizes a complete welding process tree. This process tree includes the logical topological relationships of each weld from root pass, fill pass to cap pass, such as arc initiation point offset, energy input gradient, and interpass cooling requirements, ensuring the structural integrity of thick plate large bevel welding.

[0045] Through the above implementation methods, the following is introduced To achieve standardized equivalence for irregular cross-sections, and in conjunction with The sidewall fusion problem was solved, enabling the programming of complex beveling tasks.

[0046] The welding execution and feedback module is used to send the welding process parameters to the welding execution equipment for operation, and to fine-tune the parameters in conjunction with the process coefficient based on the real-time physical phenomena feedback during the welding process. Preferably, the parameter fine-tuning step in conjunction with the process coefficient, as shown in Figure 2, includes: acquiring the sensing signal of the welding arc or molten pool in real time and converting it into a state parameter reflecting the welding stability; calculating the deviation between the state parameter and the theoretical process model, and calling the corresponding process coefficient as the adjustment step size operator; and adding the calculated deviation increment to the currently executed welding parameters to achieve dynamic correction of voltage, current, or wire feed speed.

[0047] Specifically, during the welding process, the system utilizes Hall current sensors, voltage divider sampling circuits, or visual molten pool monitoring cameras to acquire real-time sensing signals of the welding arc or molten pool. The acquired raw electrical or image signals are then transformed into state parameters reflecting welding stability through feature extraction algorithms. For example, by analyzing the probability density distribution of the voltage waveform or the fluctuation rate of the molten pool area, feature values ​​characterizing arc stability or thermal equilibrium state are extracted.

[0048] The extracted state parameters are compared with the preset theoretical process model in real time, and the deviation between the two is calculated. Based on the type of deviation, the corresponding process coefficient is invoked, such as... This is used as an adjustment step size operator. For example, when the system detects that the arc characteristic value deviates from the stable range, the arc stability correction coefficient is invoked. This coefficient defines the sensitivity and step size for correcting this bias, i.e., the correction step size. .

[0049] The system will calculate the deviation increment This adjustment is superimposed in real-time onto the currently executing welding command, affecting voltage, current, or wire feed speed. For example, if increased wire elongation is detected, causing a decrease in welding current, the system will... The coefficient calculates the compensation increment of the wire feeding speed and adds this increment to the wire feeding motor command in the next control cycle, thereby instantly restoring current stability.

[0050] Through the above implementation methods, the system transforms the static process model into a dynamic model with sensing capabilities, and uses process coefficients as adjustment operators to achieve precise correction of environmental disturbances during the welding process.

[0051] The incremental learning module is used to record and collect successful welding data, and continuously optimize and update the dynamic coefficient model through an incremental update algorithm.

[0052] Preferably, the step of continuous optimization and updating through incremental update algorithm, as shown in Figure 3, includes: recording the geometric and physical feature data, welding process parameter data, and parameter fine-tuning data during the welding process for each welding task; collecting weld formation quality data after welding is completed, and marking welding tasks that are deemed qualified as positive sample data packets; using an incremental learning algorithm to analyze the correction bias of the parameter fine-tuning data in the positive sample data packets on the initial process parameters, and calculating the update gradient of each coefficient in the dynamic coefficient model; and automatically updating the initial default value of each coefficient in the dynamic coefficient model according to the update gradient.

[0053] Specifically, a task-aware database is established to synchronously record the following three types of key data during each welding operation: Input feature data: including plate thickness extracted by the vision system. Solder leg size and bevel geometry features; execution process data: records the preset values ​​of initial current, voltage, speed, etc. generated by the system; real-time fine-tuning data: records all real-time deviation corrections triggered by the linkage process coefficient during the welding process. .

[0054] After the welding task is completed, the system collects weld formation quality data, including penetration state, surface reinforcement, and wetting angle, through a weld laser scanner or manual quality inspection. The system only packages and marks the data related to qualified welding tasks as positive sample data packages. This step ensures that the model learns in the direction of optimization and effectively filters out abnormal interference data.

[0055] The system invokes built-in incremental learning algorithms, such as stochastic gradient descent or online learning, to perform in-depth analysis of positive sample data packets. The algorithm focuses on analyzing the correction bias of fine-tuned data on initial process parameters. For example, if the system consistently performs the same correction bias under multiple similar operating conditions... If a steady state is reached only after increasing the wire feed speed by 5%, it indicates that the initial default value has a fixed negative bias. Based on the magnitude and frequency of this bias, the update gradient of each coefficient in the dynamic coefficient model is calculated. This gradient represents the direction and magnitude of the coefficient's evolution towards its optimal value.

[0056] The system automatically updates the initial default values ​​of each coefficient in the dynamic coefficient model based on the calculated update gradient. When the updated model encounters the same or similar geometric features again, the generated initial process parameters will directly include the previous corrections. This allows the system to become increasingly adapted to the equipment characteristics and material batches of a specific factory as operating time increases.

[0057] Through the above implementation methods, the system establishes a self-driven growth mechanism that continuously corrects the coefficient model using positive sample feedback, thus solving the problem of traditional process parameters becoming inaccurate over time. Example

[0058] For corner joints of Q355B steel with a thickness of 30mm, traditional welding methods often result in excessively high workpiece temperatures due to multi-layer continuous operations, leading to weld metal runoff or coarse grain structure. To address these issues, this invention provides an adaptive intelligent welding method based on a dynamic coefficient model, the execution flow of which is shown in Figure 4. The specific implementation process of this method is as follows: The system performs a full-length scan of the bevel using a visual sensing device to acquire high-precision three-dimensional point cloud data. By analyzing the point cloud model, the bevel angle, root gap, and base plate thickness are extracted (…). Geometric features such as Q355B are also identified. Simultaneously, the system identifies the base material as Q355B and retrieves its thermophysical performance parameters as initial background data.

[0059] Based on the extracted features, the task was identified as a "large-size asymmetric corner joint," automatically activating the multi-layer, multi-pass fillet weld formula submodule. The system then uses fitting coefficients to determine the total deposited metal requirement. The welding space is divided into 12 weld seams, including root pass, fill pass, and cover pass, and a preset welding sequence logic is generated.

[0060] The process calculation agent calls the dynamic coefficient model to generate baseline parameters for each pass. To address the temperature rise issue caused by continuous welding, the system primarily utilizes the heat input distribution coefficient. In the base section, utilize It provides high energy density to ensure root penetration; as the number of layers increases, the system adjusts its energy density based on the predicted interlayer temperature. Automatically decreasing current reference or increasing welding speed effectively prevents coarse grain formation caused by heat saturation of the base material.

[0061] During welding, the feedback module monitors the arc status in real time. To address the metal sagging issue that easily occurs in side-welded welds, the following process coefficients are linked: the system monitors the edge morphology of the molten pool in real time and controls the slump coefficient accordingly. The dynamic tightening of the voltage-to-wire-speed ratio enhances arc stiffness and molten pool surface tension, preventing molten metal from flowing downwards due to gravity. At the junction of the cap weld and the bevel sidewall, a joint control coefficient is used. Optimize the swing side stop time to ensure sufficient sidewall fusion and eliminate undercut defects.

[0062] After the welding task is completed, the system records the initial parameters and parameters of each pass. The system generates real-time correction increments. After confirming the weld appearance and internal quality are up to standard through online inspection, the system marks the data of these 12 welds as positive sample data packets. Incremental learning algorithm analysis reveals that in such thick plate scenarios, the current bias of the end filler pass is consistent. Subsequently, the system calculates the update gradient and automatically corrects the initial default values ​​of the correlation coefficients in the dynamic coefficient model, making the initial parameters of the next welding task of the same specification closer to the physical optimum.

[0063] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive intelligent welding system based on a dynamic coefficient model, characterized in that, include: The information acquisition module is used to scan the workpiece to be welded through a visual sensing device, acquire three-dimensional point cloud data, and extract the geometric and physical features of the workpiece to be welded. The weld seam recognition module is used to automatically determine the optimal weld seam type of the workpiece to be welded based on the geometric and physical characteristics. The process parameter generation module is used to activate the corresponding calculation sub-module according to the optimal weld type through the process calculation intelligence agent, and call the built-in dynamic coefficient model and benchmark algorithm to generate welding process parameters; the calculation sub-module includes a single-pass fillet weld intelligent generation sub-module, a multi-layer multi-pass fillet weld formula sub-module, and a groove weld programming sub-module; The welding execution and feedback module is used to send the welding process parameters to the welding execution equipment for operation, and to fine-tune the parameters in conjunction with the process coefficient based on the real-time physical phenomena feedback during the welding process. The incremental learning module is used to record and collect successful welding data, and continuously optimize and update the dynamic coefficient model through an incremental update algorithm.

2. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The three-dimensional point cloud data is acquired through binocular vision or depth vision sensors; the geometric and physical characteristics of the workpiece to be welded include, but are not limited to, joint type, plate thickness, gap, bevel angle, and bevel width.

3. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The determination of the optimal weld type for the workpiece to be welded includes: performing cross-sectional slicing processing on the three-dimensional point cloud data, calculating the cross-sectional features of the weld area, the cross-sectional features including gradient variation, width and depth features; inputting the cross-sectional features into a pre-trained classification model to automatically determine that the workpiece to be welded belongs to any one of single-pass fillet weld, multi-layer multi-pass fillet weld or bevel weld.

4. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The execution steps of the intelligent generation submodule for single-pass fillet welds include: mapping the target weld leg size to preliminary current and voltage based on the basic welding process formula, and introducing a transition interval compensation coefficient to smooth the process parameters of the weld transition interval; fine-tuning the voltage, wire feed speed or welding speed by linking the arc stability correction coefficient and the penetration and forming control coefficient, and automatically allocating the welding speed and oscillation parameters under the guidance of welding quality and efficiency targets according to the weight of the quality-efficiency balance coefficient.

5. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The execution steps of the multi-layer, multi-pass fillet weld formula submodule include: automatically allocating the cross-sectional area and welding sequence of each weld pass according to the total deposited metal amount, and using the heat input distribution coefficient to provide differentiated settings for welding parameters of different layers and passes to control heat accumulation; adjusting the geometric ratio of each weld pass through the slump control coefficient to prevent metal from flowing down, optimizing the fusion between the cap weld pass and the base material using the joint control coefficient, and automatically calculating the stacking position of each weld pass in combination with the weld pass distribution ratio coefficient.

6. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The execution steps of the groove weld procedural submodule include: using the cross-sectional shape fitting coefficient to convert the irregular groove cross-section obtained by scanning into a standardized geometric shape to calculate the total amount of filler metal; after determining the root pass process, automatically calculating the number of filler weld passes and process parameters based on the preset cross-sectional area of ​​each weld and the groove surface width, and generating a welding process tree in combination with the joint control coefficient.

7. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The parameter fine-tuning step of the linkage process coefficient includes: acquiring the sensing signal of the welding arc or molten pool in real time and converting it into a state parameter reflecting the welding stability; calculating the deviation between the state parameter and the theoretical process model, and calling the corresponding process coefficient as the adjustment step size operator; and superimposing the calculated deviation increment into the currently executed welding parameters to achieve dynamic correction of voltage, current or wire feed speed.

8. The adaptive intelligent welding system based on a dynamic coefficient model according to claim 1, characterized in that, The step of continuous optimization and updating through incremental update algorithm includes: recording the geometric and physical feature data, welding process parameter data, and parameter fine-tuning data of each welding task; collecting weld formation quality data after welding is completed, and marking welding tasks that are deemed qualified as positive sample data packets; using incremental learning algorithm to analyze the correction bias of parameter fine-tuning data in the positive sample data packets on the initial process parameters, and calculating the update gradient of each coefficient in the dynamic coefficient model; and automatically updating the initial default value of each coefficient in the dynamic coefficient model according to the update gradient.

9. An adaptive intelligent welding method based on a dynamic coefficient model, characterized in that, include: The workpiece to be welded is scanned using a visual sensing device to obtain three-dimensional point cloud data and extract the geometric and physical features of the workpiece. The optimal weld type for the workpiece to be welded is automatically determined based on the aforementioned geometric and physical characteristics. The process calculation agent activates the corresponding calculation submodule according to the optimal weld type and calls the built-in dynamic coefficient model and benchmark algorithm to generate welding process parameters. The calculation submodule includes a single-pass fillet weld intelligent generation submodule, a multi-layer multi-pass fillet weld formulaization submodule, and a groove weld programming submodule. The welding process parameters are sent to the welding execution equipment for operation, and the parameters are fine-tuned in conjunction with the process coefficients based on the real-time physical phenomena feedback during the welding process. Successful welding data is recorded and collected, and the dynamic coefficient model is continuously optimized and updated using an incremental update algorithm.

Images