Parameter adaptive welding method and system based on dynamic calibration of cladding rate and feedback of weld formation

Abstract

The application discloses a parameter self-adaptive welding method and system based on dynamic calibration of cladding rate and feedback of weld forming, and relates to the technical field of intelligent manufacturing and welding. The method first completes dynamic calibration of the cladding rate before welding and constructs a dynamic cladding rate prediction model; laser is used to collect and filter and denoise the bevel in real time during the welding process, and multiple parameters such as current, welding speed and swing are optimized layer by layer in cooperation, and then the welding is performed after amplitude limiting and smoothing; after welding, the weld is scanned to quantitatively calculate multiple forming coefficients to judge the quality, the subsequent weld parameters are adaptively corrected according to the forming result, the optimal process is accumulated through intelligent closed-loop learning, and data is shared globally for iterative optimization through multi-device cluster networking. The system adopts a six-module hardware architecture, and realizes full-closed-loop intelligent control through an industrial real-time bus. The application solves the pain points in the prior art, such as cladding rate drift, parameter mismatch of variable bevel, feedback lag of multi-layer welding, and inability of process to learn independently, and greatly improves the consistency and production efficiency of multi-layer multi-pass welding of thick plates.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and welding technology, specifically to a parameter adaptive welding method and system based on dynamic calibration of cladding rate and weld formation feedback. It is mainly applied to thick plate, multi-layer, multi-pass automated welding scenarios in industries such as steel bridge manufacturing, shipbuilding, and engineering machinery. It can achieve dynamic adaptive optimization of welding parameters for workpieces with variable bevels, ensuring welding quality and production efficiency. Background Technology

[0002] Automated welding of thick plates, involving multiple layers and passes, is a core technology in manufacturing fields such as steel bridge construction, engineering machinery, and shipbuilding. However, due to plate processing deformation, workpiece assembly errors, and thermal deformation caused by the welding process itself, the geometric dimensions of butt weld grooves (such as width and depth) and the formation of interlayer welds in actual production often exhibit significant linear or nonlinear variations along the weld length. Traditional and existing adaptive welding technologies suffer from systemic architectural and methodological deficiencies when addressing the welding requirements of variable grooves, multiple layers, multiple passes, and high quality, mainly manifested in the following aspects:

[0003] 1. Traditional static segmented control method: the inherent contradiction between efficiency and quality;

[0004] To address the issue of weld geometry variations, traditional arc welding robots often employ a "segmented welding" strategy. This involves dividing a long weld into segments of fixed distance, such as 300 millimeters, and pre-setting a set of welding parameters for each segment. This method has two fatal flaws:

[0005] The "point-to-point" approximation of parameter matching: Each weld segment is matched only at preset parameter points, while the welding parameters for most other segments are approximate values. When the assembly error itself is large or the segment distance is set unreasonable, it is very easy to cause a serious mismatch between the welding parameters and the actual requirements of the bevel, resulting in serious quality problems such as incomplete penetration, burn-through, undercut, or uneven weld formation.

[0006] Frequent manual intervention and low efficiency: Frequent parameter lookup and switching processes severely disrupt the welding process, greatly reduce the efficiency of automated welding, and are highly dependent on the operator's experience.

[0007] 2. Existing adaptive welding technology: lacks single-variable adjustment and system coupling;

[0008] Recently, some patent solutions have proposed the concept of adaptive welding. However, these existing technologies mostly focus on simple mapping based on a single variable such as bevel width, for example, only involving adjustments to welding current or oscillation amplitude, while ignoring the complex coupling relationships within the welding system, specifically manifested in:

[0009] 1) The influence of core process parameters themselves is overlooked.

[0010] Welding quality depends not only on the bevel size but also profoundly on the core process characteristic of the welding system itself: the "cladding rate" (i.e., the volume of metal deposited into the weld per unit time). The cladding rate is a dynamic variable, influenced by multiple factors such as welding machine characteristics, welding wire brand and batch, contact tip wear, and wire feeding stability. When these underlying factors change, using preset theoretical parameters will lead to a mismatch between the actual filler metal amount and the required filler amount for the bevel cross-sectional area. This severely affects the weld appearance, causing mismatches in welding parameters for subsequent weld passes, and ultimately resulting in internal weld quality defects.

[0011] 2) Lack of parameter co-optimization

[0012] Current technologies lack a mechanism to integrate real-time bevel morphology with the system cladding rate model and to collaboratively optimize welding current, welding speed, and welding torch oscillation parameters (amplitude, frequency, and center trajectory). To ensure a uniform distribution of heat input across the bevel cross-section, fine-grained control of the oscillation parameters is required, but existing solutions address this very little or remain only at the conceptual level (e.g., the functional relationship A = f(W, K) is vague). The "geometrically constrained dynamic oscillation model" and its specific implementation method represent a gap in current technology.

[0013] 3. Lack of deep feedback loops and intelligent optimization mechanisms;

[0014] A deeper problem is that the existing technical architecture lacks a complete information feedback and decision-making optimization loop, especially in the typical scenario of multi-layer, multi-pass welding. In this scenario, its main shortcomings are as follows:

[0015] 1) Lack of post-weld evaluation and real-time adjustment mechanisms:

[0016] After one or more layers of welding are completed, the bevel morphology becomes extremely complex (composed of the original bevel of the workpiece and the outline of the filled weld bead). Existing automatic welding technology can only perform welding based on preset welding specification parameters, lacking real-time quantitative judgment of welding results, and even more so lacking an executable mechanism to adaptively adjust the welding specification parameters of subsequent weld beads according to the weld appearance quality.

[0017] 2) Lack of systematic closed-loop learning ability:

[0018] Existing solutions fail to establish a complete data chain from welding environment factors, groove morphology, welding parameters, and weld formation quality, and therefore cannot establish a database and system optimization decision-making mechanism based on this data using artificial intelligence technology. They lack the ability to model the complex relationships between welding quality factors (materials, processes), groove morphology factors, welding parameter factors, and weld formation factors, and thus cannot achieve adaptive adjustment of welding parameters, continuous iterative optimization, and intelligent accumulation and reuse of optimal parameters.

[0019] In summary, existing technologies face a series of architectural bottlenecks when dealing with complex variable bevel and multi-layer, multi-pass welding. These bottlenecks range from static segmentation to single-variable adaptive welding, to the lack of coupled system calibration, and ultimately to the deficiency of deep closed-loop intelligent optimization. These bottlenecks hinder the consistency of welding quality and further improvement of production efficiency. This invention aims to systematically overcome these bottlenecks and construct a completely new and comprehensive intelligent welding solution. Summary of the Invention

[0020] To address the aforementioned shortcomings in the existing technology, this invention provides a parameter-adaptive welding method and system based on dynamic calibration of cladding rate and weld formation feedback.

[0021] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0022] The parameter-adaptive welding method based on dynamic cladding rate calibration and weld formation feedback is particularly suitable for automated welding scenarios involving thick plates, multiple layers, and multiple passes. It includes the following steps:

[0023] S1. Dynamic calibration and modeling of system cladding rate: Before formal welding, the cladding rate is dynamically calibrated for the current welding system configuration, and a dynamic cladding rate prediction model is established or updated based on the calibration data to reflect the mapping relationship between welding parameters and actual cladding rate and weld formation coefficient under the current system configuration.

[0024] S2. Online perception and feature extraction of bevel morphology: During formal welding, the laser scanning device at the front end of the welding torch continuously scans the bevel to be welded in front, acquires the point cloud data of the bevel to be welded in front in real time, and performs smoothing filtering and noise reduction on the bevel width time series data. Based on the filtered contour, the key geometric feature parameters of the bevel to be welded in front are extracted.

[0025] S3. Layered feedback welding parameter collaborative optimization generation: Based on the key geometric feature parameters of the weld bevel acquired in real time and the dynamic cladding rate prediction model, a set of optimal welding parameters is dynamically generated through a collaborative optimization algorithm.

[0026] S4. Parameter smoothing execution and welding: The dead zone and change rate gradient limit of the generated optimal welding parameters are adjusted, and after smoothing and filtering, they are synchronously sent to the welding power supply and robot controller to achieve millisecond-level closed-loop adaptive welding.

[0027] S5. Post-weld morphology comparison and quantitative evaluation of weld formation: The completed weld cross-section is scanned in real time by the laser scanning device at the rear end of the welding torch. The cross-sectional data before and after welding are compared to obtain the real-time cladding rate. At the same time, the forming coefficient of the multi-dimensional weld is quantitatively calculated to complete the quantitative evaluation of the weld formation quality.

[0028] S6. Adaptive Correction and Intelligent Closed-Loop Learning of Welding Parameters: Compare the forming coefficient measured in step S5 with the preset standard threshold, and correct the welding parameters of the next weld pass according to the process rules; associate and store data of material, groove features, welding parameters, and forming coefficient in all dimensions, establish a process knowledge base, and prioritize calling the historical best parameters for similar working conditions and continuously iterate the model.

[0029] S7. Cluster Network Collaborative Learning: Each independent welding system is connected to a network composed of central computers. All welding process parameters are uploaded to the computing center, where the central computers analyze all data, continuously optimize and summarize the optimal process and adjustment strategy, and send it to the welding systems that need it, realizing system cluster network learning and process optimization.

[0030] Preferably, in the above-described parameter adaptive welding method based on dynamic cladding rate calibration and weld formation feedback, in step S1, the dynamic cladding rate calibration is accomplished by welding test welds using at least one set of different specification parameters and scanning and measuring their actual cladding metal volume and weld formation dimensions. The calibration formula is:

[0031] , ;

[0032] in, This is the correction factor for the cladding rate. Here, N is the weld formation correction factor, and N is the number of calibration test groups. , The first The actual cladding metal volume and the theoretical cladding metal volume of the group of tests , The first The actual weld width and theoretical weld width of the group test.

[0033] Preferably, in the above-mentioned parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, in step S2, the key geometric feature parameters include the bevel centerline, root gap, and sidewall angle, and the bevel width time series data is smoothed and denoised using a moving average filtering expression:

[0034] ;

[0035] in, This is the width of the bevel after filtering. For the first The original bevel width at that moment, This represents the number of sliding sampling points on one side.

[0036] Preferably, in the above-mentioned parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, in step S3, the process of dynamically generating optimal welding parameters by the collaborative optimization algorithm includes a root pass stage and a fill stage. In the root pass stage, the initial weld bevel features are extracted, and the root pass is welded using the set basic process to complete the root pass tracking and fusion. In the fill stage, the dynamic cladding rate prediction model and the real-time remaining cross-sectional area of ​​the bevel are combined to collaboratively calculate the optimal welding current, welding speed, and welding torch oscillation parameters, and constrain the uniformity of cross-sectional heat input.

[0037] Preferably, in the above-mentioned parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, the parameter co-optimization in the filling stage includes four sets of core calculation formulas:

[0038] Target cladding rate: ;

[0039] Optimal welding current: ;

[0040] Welding torch oscillation amplitude: ;

[0041] Heat input uniformity constraint: ;

[0042] in, This represents the real-time target cross-sectional area of ​​the bevel. For welding speed, This is the correction factor for the cladding rate. Δ is the conversion coefficient of the wire feeding system, and Δ is the amount of reserved fusion edge on one side. The heat input uniformity constraint is balanced by adjusting the welding speed, oscillation frequency, and sidewall dwell time to equalize the heat input of the cross section.

[0043] Preferably, in the above-mentioned parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, in step S4, the adjustment dead zone and change rate gradient limit are: parameter update is performed only when the absolute value of a single adjustment of welding current is >10A or the change in groove width is >1mm; wherein, the maximum current adjustment rate per unit time is ≤50A / s; and the overall closed-loop control cycle of the system is less than 100ms.

[0044] Preferably, in the above-described parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, in step S5, the forming coefficient of the multi-dimensional weld bead includes four weld bead forming coefficients, and the calculation expression is as follows:

[0045] Weld appearance form factor: ;

[0046] Form factor on the left side of the weld: ;

[0047] Weld bead right-side forming factor: ;

[0048] Weld offset coefficient: , , ;

[0049] in, The length of CT is the line connecting the weld surface and the line AB at its maximum perpendicular distance. The distance between the two ends AB of the weld bead. The vertical distance between the highest point C1 on the weld surface and its left endpoint A is given. The horizontal distance between the highest point C1 on the weld surface and its left endpoint A is given. The vertical distance between the highest point C1 on the weld surface and its right endpoint B is given. The horizontal distance between the highest point C1 on the weld surface and its right endpoint B is given. This is the weld bead offset coefficient, representing the weld bead arrangement from left to right. This is the weld bead offset coefficient, which represents the weld bead arrangement from right to left. This is the horizontal distance between the left end point of the weld bead and the right end point of the previous weld bead. It is the horizontal distance between the right end point of the weld bead and the right end point of the previous weld bead.

[0050] Preferably, in the above-mentioned parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, the preset standard threshold value of the forming coefficient when the weld bead is arranged from left to right is set as follows:

[0051] ≤0.25, CX1≤0.5, CX2≤0.5, 0.3≤Y1≤2;

[0052] The modification of welding parameters for the next weld pass according to process rules includes:

[0053] If detected If the value exceeds the standard value, the welding current or voltage will be automatically adjusted proportionally according to the preset rules.

[0054] If detected Qualified but If the value exceeds the standard value, calculate the weld offset coefficient. Is it too large or too small? If it is too small, then slightly adjust the welding torch swing reference line to the left.

[0055] If detected Qualified but If the value exceeds the standard value, calculate the weld offset coefficient. Is it too large or too small? If it is too large, then slightly adjust the welding torch swing reference line to the right.

[0056] like Qualified but and If all values ​​exceed the standard value, the swing amplitude will increase.

[0057] Preferably, in the above-described parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, in step S5, the intelligent closed-loop learning includes the following steps:

[0058] The material, bevel characteristics, optimized welding parameters, and forming coefficient are fed into the intelligent learning module.

[0059] Incremental training is performed using a dynamic cladding rate prediction model to ensure that the model continuously reflects the latest state of the current system.

[0060] Will , When all welding parameters meet the standards, the excellent forming results of that group are stored as an "optimal parameter record" in the system's distributed process database.

[0061] An adaptive welding system for implementing the above method includes a cladding rate calibration module, an online bevel sensing module, a collaborative optimization decision-making module, a motion control and execution module, a data feedback and learning module, and a human-machine interaction and cluster monitoring interface, all interconnected and synchronized via a real-time industrial network.

[0062] The cladding rate calibration module includes a test welding unit, a weld scanning unit, and a model building unit. It executes steps S1 and S5 to complete the dynamic calibration of the welding system before initialization and during the welding process, and to establish and maintain the cladding rate prediction model.

[0063] The bevel online sensing module includes a welding torch front laser scanning device, a welding torch rear laser scanning device, a point cloud processing unit, and a feature extraction unit. It executes the scanning tasks in steps S2 and S5, acquires high-precision geometric data of the bevel before welding and the post-weld forming in real time, and completes time-series data filtering and geometric feature extraction.

[0064] The collaborative optimization decision module is the intelligent hub of the system. It has a built-in dynamic cladding rate prediction model, multi-parameter collaborative optimization algorithm, and parameter smoothing and limiting filter. It executes steps S3 and S4 to obtain the key geometric feature parameters of the groove to be welded in real time. Combined with the dynamic cladding rate prediction model, it generates and smooths the optimal welding parameter instruction set.

[0065] The motion control and execution module includes a welding power source, a welding robot, a smoothing filter, and a robot controller. It executes step S4, receives instructions to optimize welding parameters, and precisely controls the welding torch trajectory, oscillation, and welding energy output.

[0066] The data feedback and learning module includes a forming coefficient calculation unit, a process rule library, an AI learning engine, and a distributed database. It executes steps S5 and S6 to quantitatively evaluate the weld bead forming quality, corrects the welding parameters of the next weld bead according to the process rules, and drives the iterative update of the dynamic cladding rate prediction model and the distributed database.

[0067] The human-computer interaction and cluster monitoring interface executes step S7, which is used for preset process parameters, management of dynamic cladding rate prediction models, real-time monitoring of the welding process, alarms, and cross-robot cluster data visualization and collaborative task scheduling.

[0068] The beneficial effects of this invention are as follows:

[0069] 1. High precision at the source and strong adaptability: By dynamically calibrating the cladding rate before welding and between layers, the system errors caused by equipment and materials are fundamentally compensated, so that parameter optimization is based on the actual process capability. Combined with real-time bevel perception and collaborative optimization algorithm, it achieves precise adaptive response to millisecond-level response of variable bevel welds.

[0070] 2. Breaking through the feedback bottleneck of multi-layer welding, achieving good quality consistency: The pioneering weld formation evaluation and "weld formation feedback" control mechanism uses cross-sectional comparison before and after welding to quickly obtain the weld cross-section. It proposes weld formation coefficient and offset coefficient based on the principle of weld appearance smoothness, forming a closed-loop process optimization for multi-layer and multi-pass welding, and improving the weld pass rate.

[0071] 3. From single-machine automation to cluster intelligence: Through artificial intelligence-based process optimization and AI self-learning system, the continuous accumulation and iterative optimization of welding process parameters and quality knowledge have been realized. The system has the ability to evolve autonomously, reducing the dependence on expert experience and promoting the evolution of welding production from "automation" to "intelligence".

[0072] 4. High engineering practicality and robustness: The engineering design of the solution, such as moving average filtering, parameter adjustment dead zone and gradient limitation, simplifies the difficulty of feature point extraction, effectively resists interference, and ensures long-term stable operation in complex industrial sites. All core steps have been quantified and visualized, with a clear engineering implementation path, and can be directly applied to multi-layer and multi-pass automated welding production lines for thick plates such as steel bridges, ships and engineering machinery. Attached Figure Description

[0073] Figure 1 This is the main flowchart of the method described in this invention;

[0074] Figure 2 This is a schematic diagram illustrating the principle of hierarchical feedback-based collaborative optimization of welding parameters in the method described in this invention.

[0075] Figure 3 A schematic diagram for measuring and calculating the volume of cladding metal and the width of the weld;

[0076] Figure 4 A schematic diagram for measuring and calculating the actual cladding cross-sectional area of ​​the non-root pass weld;

[0077] Figure 5 Weld bead appearance forming factor Schematic diagram of measurement calculation;

[0078] Figure 6 Formation coefficient on the left side of the weld bead Schematic diagram of measurement calculation;

[0079] Figure 7 Formation coefficient on the right side of the weld bead Schematic diagram of measurement calculation;

[0080] Figure 8 Weld offset coefficient Schematic diagram of measurement calculation;

[0081] Figure 9 This is a block diagram of the overall structure of the system described in this invention. Detailed Implementation

[0082] To provide a further understanding of the present invention, the invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0083] In the description of this invention, it should be noted that the terms "vertical," "upper," "lower," and "horizontal," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0084] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or a connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0085] Example 1:

[0086] Embodiment 1 of this invention proposes a parameter-adaptive welding method based on dynamic cladding rate calibration and weld formation feedback. It constructs a complete intelligent closed loop of "perception-decision-execution-evaluation-optimization" based on dynamic cladding rate calibration, quantitative evaluation of weld formation, adaptive parameter adjustment, and AI autonomous learning. This provides a complete solution for achieving high-quality welding of complex welds such as thick plates, variable bevels, and multi-layer / multi-pass welds, and is particularly suitable for automated welding scenarios involving thick plates and multi-layer / multi-pass welds. For details, please refer to... Figure 1 As shown, the method includes the following steps:

[0087] S1. Dynamic calibration and modeling of system cladding rate: Before formal welding, the cladding rate is dynamically calibrated for the current welding system configuration, and a dynamic cladding rate prediction model is established or updated based on the calibration data to reflect the mapping relationship between welding parameters and actual cladding rate and weld formation coefficient under the current system configuration.

[0088] S2. Online perception and feature extraction of bevel morphology: During formal welding, the laser scanning device at the front end of the welding torch continuously scans the bevel to be welded in front, acquires the point cloud data of the bevel to be welded in front in real time, and performs smoothing filtering and noise reduction on the bevel width time series data. Based on the filtered contour, the key geometric feature parameters of the bevel to be welded in front are extracted.

[0089] S3. Layered feedback welding parameter collaborative optimization generation: Based on the key geometric feature parameters of the weld bevel acquired in real time and the dynamic cladding rate prediction model, a set of optimal welding parameters is dynamically generated through a collaborative optimization algorithm.

[0090] S4. Parameter smoothing execution and welding: The dead zone and change rate gradient limit of the generated optimal welding parameters are adjusted, and after smoothing and filtering, they are synchronously sent to the welding power supply and robot controller to achieve millisecond-level closed-loop adaptive welding.

[0091] S5. Post-weld morphology comparison and quantitative evaluation of weld formation: The completed weld cross-section is scanned in real time by the laser scanning device at the rear end of the welding torch. The cross-sectional data before and after welding are compared to obtain the real-time cladding rate. At the same time, the forming coefficient of the multi-dimensional weld is quantitatively calculated to complete the quantitative evaluation of the weld formation quality.

[0092] S6. Adaptive Correction and Intelligent Closed-Loop Learning of Welding Parameters: Compare the forming coefficient measured in step S5 with the preset standard threshold, and correct the welding parameters of the next weld pass according to the process rules; associate and store data of material, groove features, welding parameters, and forming coefficient in all dimensions, establish a process knowledge base, and prioritize calling the historical best parameters for similar working conditions and continuously iterate the model.

[0093] S7. Cluster Network Collaborative Learning: Each independent welding system is connected to a network composed of central computers. All welding process parameters are uploaded to the computing center, where the central computers analyze all data, continuously optimize and summarize the optimal process and adjustment strategy, and send it to the welding systems that need it, realizing system cluster network learning and process optimization.

[0094] Furthermore, in a preferred embodiment of the present invention, in step S1, the dynamic calibration of the cladding rate is accomplished by welding test welds using at least one set of different specification parameters and scanning and measuring the actual cladding metal volume and weld formation dimensions, wherein the calibration formula is:

[0095] , ;

[0096] in, This is the correction factor for the cladding rate. Here, N is the weld formation correction factor, and N is the number of calibrated test groups. , The first The actual cladding metal volume and the theoretical cladding metal volume of the group of tests , The first The actual weld width and theoretical weld width of the group test.

[0097] Among them, the volume of cladding metal weld width Measurement and calculation opinions Figure 3 As shown, the cladding rate and weld width represent the weld cladding capability and forming capability, respectively.

[0098] In some preferred embodiments, for a given welding system configuration (welding machine, welding wire brand and batch, etc.), it is preferable to use three sets of different specification parameters (e.g., current 240A, 260A, 280A, fixed voltage and welding speed) to weld short welds on test plates of the same material, with the short welds having varying lengths. 100mm, discarding sections where the arc initiation and termination are unstable.

[0099] After the test welds were completed, calibration calculations were performed. A high-precision laser scanner was used to acquire the cross-sectional point cloud of each test weld group, and the actual molten metal volume of each weld was calculated. Compared with the actual weld width Then according to the formula and formula Calculate the cladding rate correction coefficient and weld formation correction coefficient under three different sets of specification parameters, and then take the average value as the initial correction coefficient for the current system configuration. For example, if the cladding rate correction coefficients under the three different specification parameters are 0.93, 0.95, and 0.92 respectively, then... ,Right now .

[0100] The process of establishing the dynamic cladding rate prediction model is as follows: Based on the calibration data mentioned above, a Gaussian process regression (GPR) model trained on historical data is adopted, with the inputs being welding current (I), voltage (U), and welding speed (…). The model outputs the predicted cladding rate and weld width. This prediction model is used to correct or compensate for the theoretical parameters in predicting weld filling and weld formation, and will serve as the basis for subsequent co-optimization.

[0101] Furthermore, in a preferred embodiment of the present invention, in step S2, the key geometric feature parameters include the bevel centerline, root gap, and sidewall angle, and the bevel width time-series data is smoothed and denoised using a moving average filtering expression:

[0102] ;

[0103] in, This is the width of the bevel after filtering. For the first The original bevel width at that moment, This represents the number of sliding sampling points on one side.

[0104] Among them, the center line of the groove is used to correct the welding path, the root gap is the width of the lowest point of the groove to be welded in front, which is used to determine whether the current needs to be increased to prevent incomplete penetration, and the side wall angle is only used to determine whether the groove to be welded in front is "steep" or "gentle", which is used to determine the oscillation and dwell time of the welding gun.

[0105] Specifically, during the actual welding process, a laser scanning device installed 20-30mm in front of the welding torch continuously scans the bevel to be welded at a frequency of 100Hz, acquiring point cloud data of the bevel in real time. Then, the acquired point cloud data undergoes moving average filtering; taking the current section as the center, the average value is calculated from data from two sections before and after the current section to obtain a stable bevel width. Simultaneously, algorithms (such as Hough transform or least squares fitting) are used to extract the center line, root gap, and sidewall angle features of the groove to be welded in real time.

[0106] In a preferred embodiment of the present invention, M is set to 3 in the moving average filtering expression, that is, the average of the data from the previous and next 3 scan points is calculated.

[0107] Furthermore, in a preferred embodiment of the present invention, in step S3, as follows: Figure 2 As shown, the collaborative optimization algorithm dynamically generates optimal welding parameters through a root pass stage and a fill pass stage. In the root pass stage, initial weld bevel features are extracted, and the root pass is welded using a pre-defined basic process, completing root pass tracking and fusion. During welding, the bevel shape is scanned in real-time using a laser scanning device at the rear of the welding torch, and the remaining fill pass area is calculated as the basis for real-time adjustment of welding parameters and planning of subsequent weld passes. In the fill pass stage, subsequent weld passes (layers) are welded based on the scanned bevel morphology. Combining the dynamic cladding rate prediction model and the real-time remaining cross-sectional area of ​​the bevel, the optimal welding current, welding speed, and welding torch oscillation parameters are collaboratively calculated, while constraining the uniformity of cross-sectional heat input.

[0108] Specifically, in embodiments of the present invention, the parameter co-optimization of the filling phase includes four sets of core computational formulas:

[0109] Target cladding rate: ;

[0110] Optimal welding current: ;

[0111] Welding torch oscillation amplitude: ;

[0112] Heat input uniformity constraint: ;

[0113] in, This represents the real-time target cross-sectional area of ​​the bevel. For welding speed, This is the correction factor for the cladding rate. Δ is the conversion coefficient of the wire feeding system, and Δ is the amount of reserved fusion edge on one side. The heat input uniformity constraint is balanced by adjusting the welding speed, oscillation frequency, and sidewall dwell time to equalize the heat input of the cross section.

[0114] In embodiments of the present invention, the single-sided reserved fusion edge amount Δ is generally taken as 1 to 2 mm. When it is taken as 1.5 mm and the bevel width is 12 mm, the calculated value is... = 12 - 3 = 9mm.

[0115] Calculate the target cladding rate based on cladding volume balance. The current is dynamically optimized based on the calibration correction coefficient determined in step S1. For example, when the bevel cross-sectional area increases, the target cladding rate... The system automatically adjusts the optimal welding current.

[0116] The collaborative optimization algorithm aims to ensure uniform heat input distribution per unit area of ​​the bevel cross-section and collaboratively adjusts the welding speed. The welding torch oscillation frequency F(t) follows the following constraint relationship: This is to ensure that heat is distributed reasonably.

[0117] Furthermore, in step S4, to ensure the continuous and stable welding process, it is also necessary to optimize the optimal welding parameters generated in step S3. The welding parameters are smoothed by filtering F(t), which involves setting dead zones and gradient limits for parameter adjustment. For example, welding parameters are only updated when the calculated welding current adjustment exceeds 10A or the bevel width changes by more than 1mm, and the current adjustment rate per unit time is limited to ≤50A / s. The processed optimal welding parameters are then synchronously sent to the welding power supply and robot controller at the terminal, achieving a real-time adaptive closed loop of "scanning, calculating, and welding simultaneously." The overall closed-loop control cycle of the system should be less than 100ms.

[0118] Furthermore, in step S5, after a section of weld is completed, the laser scanning device at the rear end of the welding torch installed behind the welding torch immediately scans the weld cross section and compares the post-weld morphology with the pre-weld morphology. This step achieves two key functions: real-time cladding rate calculation and weld bead forming coefficient calculation.

[0119] Real-time cladding rate calculation: By comparing the cross-sectional point cloud (Y-coordinate change) in the same coordinate system before and after welding, the actual cladding cross-sectional area of ​​the weld bead is accurately obtained (not applicable to root pass welds), thereby quickly calculating the real-time cladding rate for model verification and calibration. The calculation principle is detailed in [link to relevant documentation]. Figure 4 As shown;

[0120] Weld bead forming coefficient calculation: Based on the principle of "smooth weld surface", four quantitative forming coefficients are automatically calculated, namely the forming coefficients of the multi-dimensional weld bead. The specific calculation expression is as follows:

[0121] Weld appearance form factor: ;

[0122] Form factor on the left side of the weld: ;

[0123] Weld bead right-side forming factor: ;

[0124] Weld offset coefficient: , , ;

[0125] in, The length of CT is the line connecting the weld surface and the line AB at its maximum perpendicular distance. The distance between the two ends AB of the weld bead is given in the figure. For the specific calculation principle, please refer to [link / reference needed]. Figure 5 As shown, the weld bead appearance forming factor Measures the overall convexity of the weld bead. Weld bead left-side profile coefficient. Weld bead right side forming factor These characterize the filler contours on the left and right sides of the weld, respectively. Specifically, in the weld bead forming factor CX1 on the left side... This is the absolute value of the vertical distance between the highest point C1 on the weld surface and its left endpoint A, i.e., the difference in the Y-coordinate. This is the horizontal distance between the highest point C1 on the weld surface and its left endpoint A, i.e., the absolute value of the difference in the X-coordinate. See [link to specific calculation principle] for details. Figure 6 As shown. The forming factor on the right side of the weld bead. , This is the absolute value of the vertical distance between the highest point C1 on the weld surface and its right endpoint B, i.e., the difference in the Y-coordinate. This is the horizontal distance between the highest point C1 on the weld surface and its right endpoint B, i.e., the absolute value of the difference in the X-coordinate. See [link to calculation principle] for details. Figure 7 As shown. In the weld offset coefficient, This is the weld bead offset coefficient, which represents the weld bead arrangement from left to right. This is the weld bead offset coefficient, which measures the offset of a weld bead relative to the previous weld bead when the weld beads are arranged from right to left in a multi-layer, multi-pass weld. , It is the horizontal distance between the left end point of the weld bead and the right end point of the previous weld bead, i.e., the absolute value of the difference in the X coordinate; Y2 represents the horizontal distance between the right endpoint of the weld bead and the right endpoint of the previous weld bead, i.e., the absolute value of the difference in the X-coordinate. Similarly, when weld beads are arranged sequentially from right to left, we can obtain Y2 = The specific calculation principle of the weld offset coefficient can be found in [link to calculation]. Figure 8 As shown.

[0126] Furthermore, in a preferred embodiment of the present invention, the preset standard threshold value of the forming coefficient when the weld beads are arranged from left to right is set as follows:

[0127] ≤0.25, CX1≤0.5, CX2≤0.5, 0.3≤Y1≤2;

[0128] The modification of welding parameters for the next weld pass according to process rules includes:

[0129] If detected If the value exceeds the standard value, the welding current or voltage will be automatically adjusted proportionally according to the preset rules.

[0130] If detected Qualified but If the value exceeds the standard value (unsmooth transition on the left side), calculate the weld offset coefficient. Is it too large or too small? If it is too small, it means that the weld bead is too far to the right. In this case, adjust the welding torch swing reference line slightly to the left.

[0131] If detected Qualified but If the value exceeds the standard value (unsmooth transition on the left side), calculate the weld offset coefficient. Is it too large or too small? If it is too large, it means that the weld bead is too far to the left. In this case, adjust the welding torch swing reference line slightly to the right.

[0132] like Qualified but and If all values ​​exceed the standard value, the swing amplitude will increase.

[0133] Furthermore, in a preferred embodiment of the present invention, in step S5, the intelligent closed-loop learning includes the following steps:

[0134] The material, bevel characteristics, optimized welding parameters, and forming coefficient are fed into the intelligent learning module.

[0135] Incremental training is performed using a dynamic cladding rate prediction model to ensure that the model continuously reflects the latest state of the current system.

[0136] Will , When all welding parameters meet the standards, the excellent forming results of that group are stored as an "optimal parameter record" in the system's distributed process database.

[0137] Specifically, taking the second layer of a multi-layer, multi-pass weld as an example, the control is carried out according to the following steps:

[0138] 1) Record the welding quality elements (material, thickness, groove type) and groove morphology elements (groove centerline, root gap, and sidewall angle) of the weld.

[0139] 2) Based on the remaining filler amount at the bevel, perform welding according to the preset parameters given by the system, and record the parameters;

[0140] 3) Determine the weld formation elements. If any element is unqualified, adjust the parameters according to the preset rules and record the correction actions.

[0141] 4) Determine the weld formation elements after parameter adjustment ( , If the result improves, the parameter is recorded as the historical best parameter, and the adjustment action is recorded as a beneficial adjustment; if the result still does not meet the requirements, the adjustment continues and the judgment continues, the optimal parameter is continuously updated according to the judgment result, and the beneficial adjustment action is recorded.

[0142] 5) When encountering the same or similar welding quality factors and weld formation factors again, the system will prioritize calling the historical best parameters and continue to judge the formation factors. If they are not ideal, further optimization will be carried out on this basis, while updating the best parameters and recording the beneficial adjustment actions.

[0143] 6) Through the above cycle, the knowledge of "the more you solder, the better" is accumulated and reused.

[0144] During the first layer of root pass welding, the system automatically switches the perception reference of "weld forming elements" to "groove morphology elements" to achieve adaptive adjustment of the root pass between layers.

[0145] For the second layer and subsequent weld passes, the system automatically switches the perception reference of "pre-weld bevel morphology" to the surface morphology after the previous weld pass is completed. Steps S2 to S6 are repeated to achieve continuous and adaptive filling between layers, completely solving the "blind welding" problem of multi-layer welding and realizing continuous and adaptive multi-layer multi-pass welding.

[0146] Furthermore, in embodiments of the present invention, both the laser scanning device at the front end of the welding torch and the laser scanning device at the rear end of the welding torch preferably use a blue laser light source with a narrow-band filter to prevent interference.

[0147] In step S7, by networking multiple welding robots, the "experience" data of all welding robot parameter optimization is shared with each other, which greatly improves the learning efficiency of welding robots and enables rapid improvement in welding quality and efficiency.

[0148] Example 2:

[0149] On the other hand, Embodiment 2 of the present invention also proposes an adaptive welding system for implementing the method described in Embodiment 1 above, such as... Figure 9 As shown, the system includes a cladding rate calibration module that is interconnected and synchronizes data via a real-time industrial network, an online bevel sensing module, a collaborative optimization decision-making module, a motion control and execution module, a data feedback and learning module, and a human-machine interaction and cluster monitoring interface.

[0150] The cladding rate calibration module includes a test welding unit, a weld scanning unit, and a model building unit. It executes steps S1 and S5 to complete the dynamic calibration of the welding system before initialization and during the welding process, and to establish and maintain the cladding rate prediction model.

[0151] The bevel online sensing module includes a welding torch front laser scanning device, a welding torch rear laser scanning device, a point cloud processing unit, and a feature extraction unit. It executes the scanning tasks in steps S2 and S5, acquires high-precision geometric data of the bevel before welding and the post-weld forming in real time, and completes time-series data filtering and geometric feature extraction.

[0152] The collaborative optimization decision module is the intelligent hub of the system. It has a built-in dynamic cladding rate prediction model, multi-parameter collaborative optimization algorithm, and parameter smoothing and limiting filter. It executes steps S3 and S4 to obtain the key geometric feature parameters of the groove to be welded in real time. Combined with the dynamic cladding rate prediction model, it generates and smooths the optimal welding parameter instruction set.

[0153] The motion control and execution module includes a welding power source, a welding robot, a smoothing filter, and a robot controller. It executes step S4, receives instructions to optimize welding parameters, and precisely controls the welding torch trajectory, oscillation, and welding energy output.

[0154] The data feedback and learning module includes a forming coefficient calculation unit, a process rule library, an AI learning engine, and a distributed database. It executes steps S5 and S6 to quantitatively evaluate the weld bead forming quality, corrects the welding parameters of the next weld bead according to the process rules, and drives the iterative update of the dynamic cladding rate prediction model and the distributed database.

[0155] The human-computer interaction and cluster monitoring interface executes step S7, which is used for preset process parameters, management of dynamic cladding rate prediction models, real-time monitoring of the welding process, alarms, and cross-robot cluster data visualization and collaborative task scheduling.

[0156] Each module interacts with data through a real-time industrial network (such as EtherCAT bus) and an IoT platform to form a centralized data lake, providing support for cluster-level AI learning.

[0157] The core innovation of this invention lies in the construction of a complete intelligent closed loop based on dynamic calibration of cladding rate, quantitative evaluation of weld formation, adaptive parameter adjustment and AI autonomous learning, which provides a complete solution for achieving high-quality welding of complex welds such as thick plates, variable bevels and multi-layer multi-pass welds.

[0158] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A parameter-adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback, particularly suitable for automated welding scenarios involving thick plates, multiple layers, and multiple passes, is characterized by: Including the following steps: S1. Dynamic calibration and modeling of system cladding rate: Before formal welding, the cladding rate is dynamically calibrated for the current welding system configuration, and a dynamic cladding rate prediction model is established or updated based on the calibration data to reflect the mapping relationship between welding parameters and actual cladding rate and weld formation coefficient under the current system configuration. S2. Online perception and feature extraction of bevel morphology: During formal welding, the laser scanning device at the front end of the welding torch continuously scans the bevel to be welded in front, acquires the point cloud data of the bevel to be welded in front in real time, and performs smoothing filtering and noise reduction on the bevel width time series data. Based on the filtered contour, the key geometric feature parameters of the bevel to be welded in front are extracted. S3. Layered feedback parameter collaborative optimization generation: Based on the key geometric feature parameters of the weld bevel acquired in real time and the dynamic cladding rate prediction model, a set of optimal welding parameters is dynamically generated through a collaborative optimization algorithm. S4. Parameter smoothing execution and welding: The dead zone and change rate gradient limit of the generated optimal welding parameters are adjusted, and after smoothing and filtering, they are synchronously sent to the welding power supply and robot controller to achieve millisecond-level closed-loop adaptive welding. S5. Post-weld morphology comparison and quantitative evaluation of weld formation: The completed weld cross-section is scanned in real time by the laser scanning device at the rear end of the welding torch. The cross-sectional data before and after welding are compared to obtain the real-time cladding rate. At the same time, the forming coefficient of the multi-dimensional weld is quantitatively calculated to complete the quantitative evaluation of the weld formation quality. S6. Adaptive Correction and Intelligent Closed-Loop Learning of Welding Parameters: Compare the forming coefficient measured in step S5 with the preset standard threshold, and correct the welding parameters of the next weld pass according to the process rules; associate and store data of material, groove features, welding parameters, and forming coefficient in all dimensions, establish a process knowledge base, and prioritize calling the historical best parameters for similar working conditions and continuously iterate the model. S7. Cluster Network Collaborative Learning: Each independent welding system is connected to a network composed of central computers. All welding process parameters are uploaded to the computing center, where the central computers analyze all data, continuously optimize and summarize the optimal process and adjustment strategy, and send it to the welding systems that need it, realizing system cluster network learning and process optimization.

2. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S1, the dynamic calibration of the cladding rate is accomplished by welding test welds using at least one set of different specification parameters and scanning and measuring the actual cladding metal volume and weld formation dimensions. The calibration formula is as follows: ， ； in, This is the correction factor for the cladding rate. Here, N is the weld formation correction factor, and N is the number of calibrated test groups. , The first The actual cladding metal volume and the theoretical cladding metal volume of the group of tests , The first The actual weld width and theoretical weld width of the group test.

3. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S2, the key geometric feature parameters include the bevel centerline, root gap, and sidewall angle. The bevel width time-series data is smoothed and denoised using a moving average filter expression: ； in, This is the width of the bevel after filtering. For the first The original bevel width at that moment, This represents the number of sliding sampling points on one side.

4. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S3, the process of dynamically generating optimal welding parameters by the collaborative optimization algorithm includes a root pass stage and a fill stage. In the root pass stage, the initial weld bevel features are extracted, and the root pass is welded using the set basic process to complete the root pass tracking and fusion. In the fill stage, the dynamic cladding rate prediction model and the real-time remaining cross-sectional area of ​​the bevel are combined to collaboratively calculate the optimal welding current, welding speed, and welding torch oscillation parameters, and constrain the uniformity of cross-sectional heat input.

5. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 4, characterized in that, The parameter co-optimization in the filling phase includes four sets of core computational formulas: Target cladding rate: ; Optimal welding current: ; Welding torch oscillation amplitude: ; Heat input uniformity constraint: ; in, This represents the real-time target cross-sectional area of ​​the bevel. For welding speed, This is the correction factor for the cladding rate. Δ is the conversion coefficient of the wire feeding system, and Δ is the amount of reserved fusion edge on one side. The heat input uniformity constraint is balanced by adjusting the welding speed, oscillation frequency, and sidewall dwell time to equalize the heat input of the cross section.

6. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S4, the adjustment dead zone and change rate gradient limit are as follows: parameter updates are only performed when the absolute value of a single adjustment of the welding current is >10A or the change in groove width is >1mm; wherein, the maximum current adjustment rate per unit time is ≤50A / s; and the overall closed-loop control cycle of the system is less than 100ms.

7. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S5, the forming factor of the multi-dimensional weld bead includes four weld bead forming factors, and the calculation expression is as follows: Weld appearance form factor: ; Form factor on the left side of the weld: ; Weld bead right-side forming factor: ; Weld offset coefficient: , , ; in, The length of CT is the line connecting the weld surface and the line AB at its maximum perpendicular distance. The distance between the two ends AB of the weld bead. The vertical distance between the highest point C1 on the weld surface and its left endpoint A is given. The horizontal distance between the highest point C1 on the weld surface and its left endpoint A. The vertical distance between the highest point C1 on the weld surface and its right endpoint B is given. The horizontal distance between the highest point C1 on the weld surface and its right endpoint B is given. This is the weld bead offset coefficient, which represents the weld bead arrangement from left to right. This is the weld bead offset coefficient, which represents the weld bead arrangement from right to left. This is the horizontal distance between the left end point of the weld bead and the right end point of the previous weld bead. It is the horizontal distance between the right end point of the weld bead and the right end point of the previous weld bead.

8. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 7, characterized in that, The preset standard threshold value for the forming coefficient is set as follows when the weld beads are arranged from left to right: ≤0.25、CX1≤0.5、CX2≤0.5、0.3≤Y1≤2; The modification of welding parameters for the next weld pass according to process rules includes: If detected If the value exceeds the standard value, the welding current or voltage will be automatically adjusted proportionally according to the preset rules. If detected Qualified but If the value exceeds the standard value, calculate the weld offset coefficient. Is it too large or too small? If it is too small, then slightly adjust the welding torch swing reference line to the left. If detected Qualified but If the value exceeds the standard value, calculate the weld offset coefficient. Is it too large or too small? If it is too large, then slightly adjust the welding torch swing reference line to the right. like Qualified but and If all values ​​exceed the standard value, the swing amplitude will increase.

9. The parameter adaptive welding method based on dynamic calibration of cladding rate and weld formation feedback according to claim 1, characterized in that, In step S5, the intelligent closed-loop learning includes the following steps: The material, bevel characteristics, optimized welding parameters, and forming coefficient are fed into the intelligent learning module. Incremental training is performed using a dynamic cladding rate prediction model to ensure that the model continuously reflects the latest state of the current system. Will , When all welding parameters meet the standards, the excellent forming results of that group are stored as an "optimal parameter record" in the system's distributed process database.

10. An adaptive welding system for implementing the method according to any one of claims 1-9, characterized in that, It includes a cladding rate calibration module, a bevel online sensing module, a collaborative optimization decision-making module, a motion control and execution module, a data feedback and learning module, and a human-machine interaction and cluster monitoring interface, all connected and synchronized via real-time industrial networks. The cladding rate calibration module includes a test welding unit, a weld scanning unit, and a model building unit. It executes steps S1 and S5 to complete the dynamic calibration of the welding system before initialization and during the welding process, and to establish and maintain the cladding rate prediction model. The bevel online sensing module includes a welding torch front laser scanning device, a welding torch rear laser scanning device, a point cloud processing unit, and a feature extraction unit. It executes the scanning tasks in steps S2 and S5, acquires high-precision geometric data of the bevel before welding and the post-weld forming in real time, and completes time-series data filtering and geometric feature extraction. The collaborative optimization decision module is the intelligent hub of the system. It has a built-in dynamic cladding rate prediction model, multi-parameter collaborative optimization algorithm, and parameter smoothing and limiting filter. It executes steps S3 and S4 to obtain the key geometric feature parameters of the groove to be welded in real time. Combined with the dynamic cladding rate prediction model, it generates and smooths the optimal welding parameter instruction set. The motion control and execution module includes a welding power source, a welding robot, a smoothing filter, and a robot controller. It executes step S4, receives instructions to optimize welding parameters, and precisely controls the welding torch trajectory, oscillation, and welding energy output. The data feedback and learning module includes a forming coefficient calculation unit, a process rule library, an AI learning engine, and a distributed database. It executes steps S5 and S6 to quantitatively evaluate the weld bead forming quality, corrects the welding parameters of the next weld bead according to the process rules, and drives the iterative update of the dynamic cladding rate prediction model and the distributed database. The human-computer interaction and cluster monitoring interface executes step S7, which is used for preset process parameters, management of dynamic cladding rate prediction models, real-time monitoring of the welding process, alarms, and cross-robot cluster data visualization and collaborative task scheduling.

Images