Self-adaptive welding control system for special-shaped curved surface steel structure

Abstract

The invention relates to the technical field of automatic welding, and discloses a special-shaped curved surface steel structure self-adaptive welding control system which comprises a three-dimensional visual scanning unit, a welding robot, an acoustic feedforward sensing unit, a molten pool state diagnosis unit and a central control and calculation unit. The acoustic feed-forward sensing unit collects sound wave field data of a to-be-welded area in front of a welding gun, the molten pool state diagnosis unit obtains real-time temperature and component information of a molten pool through spectral analysis, a model correction module in the central control and calculation unit dynamically corrects an acoustic elastic coefficient according to the real-time temperature and component information, and a stress field reconstruction module calculates a stress field of the to-be-welded area based on the coefficient and the wave field data. The method comprises the steps that a feedforward stress field distribution diagram is reconstructed, an instruction generation module analyzes the gradient and amplitude of a stress field, a welding path avoiding a high stress gradient area and a compensation stress amplitude and a parameter regulation and control instruction are generated, and feedforward self-adaptive control is achieved. The welding joint quality is improved.

Description

Technical Field

[0001] This invention relates to the field of automated welding technology, specifically to an adaptive welding control system for irregular curved steel structures. Background Technology

[0002] In bridge engineering, shipbuilding, and aerospace, large, irregularly shaped curved steel structures are critical load-bearing components. To ensure consistent production efficiency and welding quality, the use of automated welding robots has become the mainstream technical solution. However, these large steel structural components undergo multiple processing steps such as rolling, stamping, and bending before forming, inevitably generating and accumulating complex residual stresses within them. During the welding process, these initially present and unevenly distributed residual stresses couple with the instantaneous thermal stresses caused by the welding process, making the final stress state and deformation trend of the weldment extremely complex and unpredictable, easily leading to serious quality defects such as welding cracks and excessive deformation.

[0003] To address this issue, advanced welding systems in the present technology typically employ sensor-based real-time monitoring and feedback control strategies. These systems use technologies such as visual sensing or laser scanning to monitor the morphology of the weld pool or the formed weld in real time. When deviations such as fluctuations in the weld pool size or poor weld formation are detected, welding parameters are then adjusted with a lag. The inherent drawback of this control method is its lag in response; that is, the control command is always issued later than the emergence of welding defects. By the time the system identifies a problem, the weld pool has already moved to a new position, and defects in the solidified area cannot be fundamentally eliminated. Therefore, it is difficult to proactively avoid welding deformation or cracking that is about to occur due to stress concentration ahead.

[0004] To achieve more precise control, some cutting-edge technologies attempt to introduce methods such as acoustic detection to assess the stress state of weldments. However, these methods face challenges in model accuracy during application. The key physical parameters upon which their stress calculation models rely, such as the acoustoelastic coefficient, are extremely sensitive to the temperature and composition of the material. Under the intense action of the welding arc, the material temperature and molten pool composition in the welding area undergo drastic and dynamic changes. If a preset, constant physical parameter is used in the calculation, the calculated stress field will deviate significantly from the actual situation, severely affecting the accuracy of the control system and making it difficult to adapt to complex and changing actual welding conditions.

[0005] Furthermore, existing welding control systems lack robustness and intelligence in responding to unexpected anomalies during the process. When sensors detect abnormal fluctuations in stress signals, the system typically cannot pinpoint the root cause of the anomaly. It cannot effectively distinguish whether the anomaly is caused by internal material inhomogeneity or external factors such as loosening of the workpiece clamping. This lack of diagnostic capability forces the system to execute uniform, non-differentiated handling logic, such as universal parameter adjustments or direct shutdown. This not only reduces control accuracy but also affects production continuity and efficiency. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an adaptive welding control system for irregular curved steel structures, which solves the problem that existing automated welding systems lack the ability to accurately perceive and dynamically correct complex stress fields before welding, thus failing to proactively avoid welding deformation and defects.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an adaptive welding control system for irregularly shaped curved steel structures, comprising the following steps:

[0008] The 3D vision scanning unit is used to scan the steel structure workpiece to be welded and generate a 3D digital model.

[0009] A welding robot, as a welding execution unit, includes a robotic arm and a welding torch fixed to the end of the robotic arm;

[0010] An acoustic feedforward sensing unit, integrated at the front end of the welding torch, is used to acquire reference wave field data of the area to be welded before welding and to acquire real-time wave field data during welding.

[0011] The molten pool condition diagnostic unit is used to collect spectral data of the welding molten pool in real time during the operation of the welding torch;

[0012] Central control and computing unit, the central control and computing unit comprising:

[0013] Path planning module: used to plan the initial welding path based on the three-dimensional digital model generated by the three-dimensional vision scanning unit;

[0014] Model correction module: used to dynamically correct the preset acoustic and stress theory models based on the spectral data collected in real time by the molten pool state diagnosis unit;

[0015] Stress field reconstruction module: Based on the modified acoustic and stress theory model, and using the reference wave field data and the real-time wave field data obtained by the acoustic feedforward sensing unit, the module reconstructs the feedforward stress field distribution map of the area in front of the welding torch through a tomographic inversion algorithm.

[0016] Command generation and sending module: used to generate control commands based on the feedforward stress field distribution map, and send the control commands to the welding robot to control the welding operation.

[0017] Preferably, the acoustic feedforward sensing unit includes an ultrasonic phased array transmitting module and an ultrasonic phased array receiving module.

[0018] Preferably, the model correction module operates as follows: by analyzing the spectral data, it calculates the elemental composition and temperature information of the weld pool. Since the acoustoelastic coefficient of a material is a function of elemental composition and temperature, the model correction module dynamically adjusts the acoustoelastic coefficient in the acoustic and stress theory model based on the calculated elemental composition and temperature information. The relationship between the speed of sound propagation in a stress field and the speed of sound in a stress-free state can be described by acoustoelastic theory. The model correction module establishes a functional relationship between the acoustoelastic coefficient K of the material and temperature T and elemental composition C, K = f(T, C), and updates the K value in real time.

[0019] Preferably, the stress field reconstruction module operates as follows: based on the corrected acoustic and stress theory model, the change in sound wave flight time between the real-time wavefield data and the reference wavefield data is calculated into the feedforward stress field distribution map using the tomographic inversion algorithm. The calculation process is based on the acoustoelastic effect. For any one of the multiple sound wave propagation paths constructed between the transmitting and receiving modules, the change in sound wave flight time and the stress field distribution along the path satisfy the following linear integral relationship:

[0020]

[0021] in:

[0022] Δt ij This represents the change in sound wave flight time along the ij-th path;

[0023] L ij Let represent the geometric path of the i-th sound wave propagation;

[0024] σ(x,y) represents the stress component at position (x,y);

[0025] K represents the acoustoelastic coefficient of the material;

[0026] v0 represents the speed of sound of the material under stress-free conditions;

[0027] Represents acoustic time and stress sensitivity coefficient;

[0028] Indicates the stress field along path L ij The line integral.

[0029] By measuring Δt for all paths ij This forms a system of linear equations. The tomographic inversion algorithm (e.g., the algebraic reconstruction algorithm ART) is used to solve this system of equations to obtain the stress values ​​σ(x,y) on the discrete grid, and finally generate the feedforward stress field distribution map.

[0030] Preferably, the instruction generation and sending module operates by analyzing the stress gradient distribution in the feedforward stress field distribution map. A welding path adjustment command is generated. By analyzing the stress amplitude (|σ|) in the feedforward stress field distribution map, a welding parameter adjustment command is generated. The welding path adjustment command and the welding parameter adjustment command are combined to generate the control command and send it to the welding robot.

[0031] Preferably, the welding parameters are selected from the group consisting of welding current, welding voltage, and welding speed.

[0032] Preferably, the central control and computing unit further includes a cross-validation and diagnostic module, used to cross-validate the real-time wave field data acquired by the acoustic feedforward sensing unit and the spectral data collected by the molten pool state diagnostic unit, and to diagnose the source of stress anomalies.

[0033] Preferably, when the source of the stress anomaly is diagnosed as material inhomogeneity, the cross-validation and diagnosis module instructs the instruction generation and sending module to adjust the adjustment range of the welding path adjustment instruction and the welding parameter adjustment instruction when generating the control instruction in combination.

[0034] Preferably, the cross-validation and diagnostic module is further used for:

[0035] When the source of the stress anomaly is diagnosed as an external clamping factor, an alarm message and a stop command to halt the welding operation are generated.

[0036] Preferably, the central control and computing unit further includes an initial instruction generation module, which combines the initial welding path generated by the path planning module with a set of corresponding initial welding parameters to form an initial instruction for starting the welding operation.

[0037] This invention provides an adaptive welding control system for irregularly shaped curved steel structures. It offers the following advantages:

[0038] 1. This invention, by setting up an acoustic feedforward sensing unit and a stress field reconstruction module, obtains the stress field distribution of the area to be welded before the weld pool arrives, and adjusts the welding path and welding parameters in advance based on this distribution map, transforming the traditional hysteresis feedback control into feedforward predictive control, thereby actively avoiding welding defects before they form, and improving the forming quality and structural reliability of irregular curved surface welding.

[0039] 2. This invention sets up a molten pool state diagnosis unit and a model correction module, uses real-time acquired spectral data to calculate the temperature and elemental composition information of the molten pool, and dynamically corrects the key parameters of the acoustic and stress theory models accordingly. This ensures that the reconstruction accuracy of the feedforward stress field distribution map is not affected by changes in the physical properties of the material during the welding process, and enhances the system's adaptability and control accuracy to complex working conditions.

[0040] 3. By setting up a cross-validation and diagnostic module, the present invention performs fusion analysis of acoustic data and spectral data to diagnose the source of stress anomalies, and performs differentiated treatments such as adjusting the control amplitude or alarm shutdown for different causes such as material inhomogeneity and external clamping factors, so that the system has fault diagnosis and decision-making capabilities, and improves the robustness and intelligence level of the entire welding control process. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the overall process of an adaptive welding control method for irregular curved steel structures according to an embodiment of the present invention.

[0042] Figure 2 This is a flowchart of the system job initialization phase according to an embodiment of the present invention;

[0043] Figure 3 This is a flowchart of the closed-loop stage of the system feedforward adaptive control according to an embodiment of the present invention.

[0044] Figure 4 This is a flowchart of the system cross-validation and anomaly handling phase according to an embodiment of the present invention.

[0045] Among them, 10 is the 3D vision scanning unit; 20 is the welding robot; 30 is the acoustic feedforward sensing unit; 40 is the molten pool state diagnosis unit; 50 is the central control and calculation unit; 51 is the path planning module; 52 is the model correction module; 53 is the stress field reconstruction module; 54 is the instruction generation and sending module; 55 is the cross-validation and diagnosis module; and 56 is the initial instruction generation module. Detailed Implementation

[0046] 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 some embodiments of the present invention, and not all embodiments. 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.

[0047] Reference Appendix Figure 1 , Figure 1 This is a schematic diagram of the overall process of an adaptive welding control method for irregular curved steel structures according to an embodiment of the present invention. The present invention provides an adaptive welding control system for irregular curved steel structures, which may include: a three-dimensional vision scanning unit 10, a welding robot 20, an acoustic feedforward sensing unit 30, a molten pool state diagnosis unit 40, and a central control and computing unit 50.

[0048] A 3D vision scanning unit 10, such as a structured light scanner or a laser scanner, is deployed in the welding operation area to scan the entire or partial surface of the steel structure workpiece to be welded before the welding operation begins, generating 3D digital model data characterizing the 3D morphology of the workpiece. The 3D vision scanning unit 10 establishes a data communication connection with the central control and computing unit 50 to send the generated 3D digital model data to the central control and computing unit 50.

[0049] The welding robot 20 is the welding operation execution unit of this system, including a multi-degree-of-freedom robotic arm and a welding torch fixed to the end effector interface of the robotic arm. The welding robot 20 is connected to the central control and computing unit 50, receives and executes the instructions sent by the central control and computing unit 50, drives the welding torch to move along the specified path and perform welding operations.

[0050] Both the acoustic feedforward sensing unit 30 and the weld pool condition diagnosis unit 40 are integrated into the front end of the welding torch. Specifically, the acoustic feedforward sensing unit 30 is physically positioned in front of the wire exit end of the welding torch, ensuring that its working area is always the area to be welded. The physical position of the weld pool condition diagnosis unit 40 is aligned with the weld pool area formed during welding torch operation. This integrated configuration allows the wave field data acquired by the acoustic feedforward sensing unit and the spectral data acquired by the weld pool condition diagnosis unit to be synchronously acquired as the welding torch moves.

[0051] The central control and computing unit 50 is the data processing and control decision center of this system, and its hardware form is an industrial control computer. The data input terminals of the central control and computing unit 50 are connected to the three-dimensional vision scanning unit 10, the acoustic feedforward sensing unit 30, and the molten pool state diagnosis unit 40, respectively; its data output terminal is connected to the controller of the welding robot 20.

[0052] The overall data flow of this system is as follows: During the job initialization phase, the three-dimensional digital model data collected by the three-dimensional vision scanning unit 10 is sent to the central control and computing unit 50 for initial path planning.

[0053] During the welding operation, this welding control system generates a real-time control data stream. The acoustic feedforward sensing unit 30 collects reference and real-time wavefield data of the area to be welded in front of the welding torch, while the molten pool state diagnosis unit 40 simultaneously collects the spectral data of the weld pool. The data is continuously transmitted to the central control and computing unit 50. Upon receiving the data, the central control and computing unit 50 performs calculations, generates control commands containing path and parameter adjustment information, and sends these commands to the welding robot 20 to achieve real-time adaptive control of the welding process.

[0054] The central control and computing unit 50 is physically an industrial control computer, and logically functions by executing program instructions stored in non-volatile memory. Functionally, the central control and computing unit 50 is divided into multiple cooperating software modules, including:

[0055] Path planning module 51. Path planning module 51 receives three-dimensional digital model data generated by three-dimensional vision scanning unit 10. Path planning module 51 first performs surface reconstruction and feature recognition on the three-dimensional digital model data to accurately locate the geometric boundary of the weld to be welded. Subsequently, path planning module 51 applies a path search algorithm (such as A* algorithm or Dijkstra algorithm) to calculate the initial motion trajectory of the welding torch in three-dimensional space, the welding torch attitude angle, and the reference initial welding speed under the geometric constraints of the weld. These data together constitute the initial welding path.

[0056] Model correction module 52. Model correction module 52 receives spectral data acquired in real time by the molten pool condition diagnostic unit 40. By analyzing the intensity, width, and center wavelength shift of specific element spectral lines in the spectral data, model correction module 52 calculates the temperature T of the weld pool using, for example, the Boltzmann plane method, and calculates the composition information C of key alloying elements through spectral line comparison. Since the acoustoelastic coefficient K of the material is a function of temperature and composition, model correction module 52 calculates and updates the acoustoelastic coefficient K value of the material in the acoustic and stress theory models in real time according to the pre-calibrated functional relationship K = f(T, C).

[0057] Stress field reconstruction module 53. Stress field reconstruction module 53 receives reference wavefield data (acquired before welding) and real-time wavefield data (acquired during welding) from acoustic feedforward sensing unit 30, and calls upon the acoustoelastic coefficient K value updated in real time by model correction module 52. Stress field reconstruction module 53 first calculates the change in acoustic wave flight time Δt for the two paths. ijSubsequently, a large sparse linear equation system is constructed based on the linear integral relationship formula of acoustoelasticity. The stress field reconstruction module 53 executes a tomographic inversion algorithm, such as the algebraic reconstruction algorithm (ART), to iteratively solve the linear equation system to calculate the stress values ​​σ(x,y) of all nodes in the area to be welded in front of the welding torch on the preset discrete grid, and finally generates a data-driven feedforward stress field distribution map.

[0058] Command generation and transmission module 54. Command generation and transmission module 54 receives the feedforward stress field distribution map generated by stress field reconstruction module 53. This module performs specific control law calculations: by calculating the stress gradient in the distribution map. The distribution map generates welding path adjustment instructions (e.g., trajectory offsets in the X and Y planes) to avoid high stress gradient regions. By analyzing the stress amplitude (|σ|) in the distribution map, welding parameter adjustment instructions (e.g., increases and decreases in welding current and welding speed) are generated to compensate for the stress amplitude. The instruction generation and transmission module 54 then combines and encapsulates these two adjustment instructions to generate a structured final control instruction conforming to the controller interface protocol of the welding robot 20, and sends the final control instruction to the welding robot 20 via the data bus.

[0059] Cross-validation and diagnostic module 55. Functionally, cross-validation and diagnostic module 55 is configured to simultaneously receive real-time wavefield data from acoustic feedforward sensing unit 30 and spectral data from molten pool state diagnostic unit 40. This module performs data consistency verification. When a severe fluctuation exceeding a preset threshold is detected in the stress field data, this module analyzes the temporal and amplitude correlation between the volatility of the spectral data and the volatility of the acoustic data, and applies a pre-trained decision tree or support vector machine model to diagnose the source of the stress anomaly. If the diagnosis result is material inhomogeneity, cross-validation and diagnostic module 55 sends a specific instruction to instruction generation and sending module 54 to adjust the adjustment amplitude; if the diagnosis result is external clamping factors (e.g., loose clamps), cross-validation and diagnostic module 55 generates alarm information and a stop instruction to halt the welding operation.

[0060] The central control and computing unit 50 also includes an initial instruction generation module 56. The initial instruction generation module 56 is used for job preparation before the formal start of the welding operation. The initial instruction generation module 56 calls the initial welding path generated by the path planning module 51 and retrieves a set of initial welding parameters (such as a specific set of initial welding current and initial welding voltage) corresponding to the workpiece material grade and thickness from the parameter database. The initial instruction generation module 56 combines the initial welding path and the initial welding parameters to generate initial instructions for starting the welding operation and sends the initial instructions to the welding robot 20.

[0061] The technical principles and implementation methods of the key modules of this invention are further detailed below:

[0062] The acoustic feedforward sensing unit 30 includes an ultrasonic phased array transmitting module and an ultrasonic phased array receiving module. The transmitting and receiving modules are symmetrically fixed on both sides of the wire exit end of the welding torch, with their acoustic working surfaces facing the surface of the steel structure workpiece to be welded. Each module contains an array consisting of multiple independently controllable piezoelectric transducer elements arranged linearly. Driven by a control signal, each element of the transmitting module can generate an ultrasonic beam with controllable deflection angle and focusing depth by setting different excitation delay times.

[0063] During operation, the transmitting module rapidly emits a series of ultrasonic beams with different deflection angles via electronic scanning. These beams propagate within the welding area of ​​the workpiece and are received by an array of receiving modules. In this way, a welding zone is created between the transmitting and receiving modules, covering the welding area in front of the welding torch, and consisting of multiple geometrical paths L... ij The acoustic wave propagation path network is formed. Before the welding operation begins, the acoustic feedforward sensing unit 30 acquires wavefield data of the region as reference wavefield data, which includes all paths L. ij The reference sound wave flight time t 0,ij During the welding process, this unit continuously acquires real-time wavefield data, i.e., the real-time flight time L of the acoustic waves along each path. ij These two sets of data are then sent to the stress field reconstruction module 53.

[0064] The model correction module 52 is used to ensure the accuracy of the physical parameters in the subsequent stress field calculation. The spectral data collected by the molten pool condition diagnosis unit 40 is transmitted to the model correction module 52. The model correction module 52 first applies the Boltzmann plane method to calculate the temperature of the welding arc plasma by analyzing the relative intensity of multiple atomic emission lines of iron (Fe) in the spectral data. This temperature has a definite mapping relationship with the temperature T of the welding molten pool.

[0065] Simultaneously, the model correction module 52 compares the real-time acquired spectra with a preset standard spectral database. By identifying the presence and intensity changes of characteristic spectral lines of specific alloying elements (such as chromium (Cr), nickel (Ni), and manganese (Mn), it calculates the composition information C of key elements in the molten pool. The acoustoelastic coefficient K of the material is not a constant value; it is highly sensitive to changes in temperature and material composition.

[0066] In one specific implementation of the present invention, the functional relationship K = f(T, C) is pre-calibrated experimentally and stored in the central control and calculation unit 50. The model correction module 52 substitutes the calculated temperature T and composition information C into this functional relationship to calculate the most accurate acoustoelastic coefficient K value of the material under the current working conditions, and provides this dynamically updated K value to the stress field reconstruction module 53 for use.

[0067] The stress field reconstruction module 53 is the core computational unit for implementing feedforward control. The stress field reconstruction module 53 receives reference wavefield data and real-time wavefield data from the acoustic feedforward sensing unit 30, and obtains the dynamically updated acoustoelastic coefficient K from the model correction module 52. The first operation of the stress field reconstruction module 53 is to calculate the change in sound wave flight time, i.e., for each sound wave propagation path L constructed between the transmitting and receiving modules. ij Its real-time flight time t ij With reference flight time t 0,ij Subtract them to obtain the time change Δt ij .

[0068] The reconstruction algorithm of stress field reconstruction module 53 is based on the acoustoelastic effect. This effect describes the influence of the internal stress field of a material on the propagation speed of sound waves. For a two-dimensional planar stress field, the change in sound wave flight time and the stress field distribution along the path satisfy the following linear integral relationship:

[0069]

[0070] in:

[0071] Δt ij This represents the change in sound wave flight time along the ij-th path, and is a known input of the module;

[0072] L ij The geometric path of the i-th sound wave propagation is determined by the geometric configuration of the acoustic feedforward sensing unit.

[0073] σ(x,y) represents the stress component at position (x,y) within the area to be welded, which is an unknown quantity that the module needs to solve for;

[0074] K represents the acoustoelastic coefficient of the material, which is determined and provided by the model correction module 52 based on real-time operating conditions.

[0075] v0 represents the speed of sound of a material in a stress-free state, which is a known constant obtained through pre-calibration;

[0076] Represents acoustic time and stress sensitivity coefficient;

[0077] This represents the unknown stress field σ(x,y) along a known path L. ij The line integral.

[0078] To solve this integral equation, the stress field reconstruction module 53 first divides the area to be welded in front of the welding torch into a discrete mesh consisting of M pixel units. It is assumed that the stress value within each pixel unit is a constant σ. m (where m = 1, 2, ..., M). Thus, for each sound wave propagation path L ij The above integral relationship is then transformed into a linear algebraic equation. By measuring the sound wave propagation path network consisting of N different paths, N such linear equations can be obtained, which together form a large system of linear equations. Here, N is the total dimension of the measurement data, which directly determines the total number of equations used to solve the unknown stress field.

[0079] The stress field reconstruction module 53 applies a tomographic inversion algorithm to solve this system of linear equations. In one specific embodiment, the algebraic reconstruction algorithm (ART) is used. The algebraic reconstruction algorithm calculates the optimal stress value σ for each pixel unit through iterative calculation. m The stress value set of all pixel units {σ1,σ2,...,σ...} M The combined data forms a two-dimensional data matrix. This matrix is ​​the feedforward stress field distribution map, which digitally represents the stress distribution in the area to be welded in front of the welding torch. Finally, the stress field reconstruction module 53 outputs the generated feedforward stress field distribution map to the instruction generation and sending module 54.

[0080] Reference Figure 1 , Figure 1 This is a schematic diagram of the overall process of an adaptive welding control method for irregular curved steel structures according to an embodiment of the present invention. The overall workflow of the system of the present invention mainly includes the operation initialization stage, the feedforward adaptive control closed-loop stage, and the parallel execution cross-validation and anomaly handling stage. The specific process of each stage will be described in detail below.

[0081] Reference Figure 2 , Figure 2 This is a flowchart illustrating the system job initialization phase according to an embodiment of the present invention. The overall workflow and control method of the system of the present invention includes the following steps in the job initialization phase.

[0082] Before the welding operation officially begins, the operator first places and fixes the steel structure workpiece to be welded within the designated work area. Subsequently, the three-dimensional vision scanning unit 10 is activated to scan the area of ​​the workpiece to be welded, generating three-dimensional digital model data containing high-density three-dimensional coordinate points of the area, and sending this data to the central control and computing unit 50.

[0083] The path planning module 51 within the central control and computing unit 50 receives the three-dimensional digital model data. The path planning module 51 first processes the data to identify the precise three-dimensional geometric path of the weld to be welded, and then, based on this path and preset welding torch attitude constraints, calculates and generates an initial welding path consisting of a series of spatial coordinate points, attitude angles, and reference speeds.

[0084] Simultaneously, the initial instruction generation module 56 is activated. Based on the workpiece material grade and plate thickness information pre-input by the operator, this module queries an internally stored welding process parameter database and retrieves a set of matching initial welding parameters. These initial welding parameters specifically include the initial welding current value, the initial welding voltage value, and the initial wire feed speed value.

[0085] The initial instruction generation module 56 receives the initial welding path generated by the path planning module 51 and retrieves the initial welding parameters. Subsequently, the initial instruction generation module 56 performs a combination and generation operation: it combines the initial welding path data and the initial welding parameter data, and encapsulates them according to the communication protocol and data format required by the welding robot 20 controller to generate a structured initial instruction for starting the welding operation.

[0086] Finally, the initial instruction generation module 56 sends the generated initial instruction to the controller of the welding robot 20 via the data interface. Upon receiving this initial instruction, the welding robot 20 drives its robotic arm and welding torch to the welding start point and begins welding operations using the initial welding parameters set in the instruction. After this step is completed, the system's operation initialization phase ends, and it transitions to the feedforward adaptive control closed-loop phase.

[0087] Reference Figure 3 , Figure 3 This is a flowchart illustrating the workflow of the feedforward adaptive control closed-loop stage of a system according to an embodiment of the present invention. After the welding robot 20 executes the initial instructions and establishes a stable welding arc, the system of the present invention immediately enters the feedforward adaptive control closed-loop stage. This stage is a continuous, cyclically executed real-time control process.

[0088] Within each cycle of this control loop, as the welding torch moves along the welding path, the acoustic feedforward sensing unit 30 continuously acquires real-time wave field data of the area to be welded in front of the welding torch, while the molten pool condition diagnosis unit 40 simultaneously acquires real-time spectral data of the weld pool. These two sets of real-time data are continuously sent to the central control and computing unit 50.

[0089] Upon receiving the data, the model correction module 52 and stress field reconstruction module 53 within the central control and computing unit 50 work together to calculate, as previously described, a dynamically updated feedforward stress field distribution map characterizing the stress distribution in the area in front of the welding torch. This distribution map is then transmitted to the instruction generation and sending module 54.

[0090] The instruction generation and transmission module 54 receives the feedforward stress field distribution map and analyzes it to generate adjustment instructions. For welding path adjustments, the instruction generation and transmission module 54 calculates the stress gradient at each point in the distribution map. When a region with a stress gradient exceeding a preset threshold is detected on the path ahead, the instruction generation and sending module 54 generates a welding path adjustment instruction. This instruction contains a set of spatial coordinate offsets to guide the welding torch trajectory to smoothly bypass the core area with a high stress gradient.

[0091] For adjusting welding parameters, the instruction generation and sending module 54 analyzes the stress amplitude |σ| at each point along the predetermined path in the distribution map. When the stress amplitude on the forward path is detected to exceed a preset threshold, the instruction generation and sending module 54 generates a welding parameter adjustment instruction. The welding parameters are selected from a group consisting of welding current, welding voltage, and welding speed. For example, to compensate for higher tensile stress, the instruction may include an amount of instruction to increase the welding current or an amount of instruction to decrease the welding speed to increase heat input.

[0092] Subsequently, the instruction generation and sending module 54 combines the calculated welding path adjustment instruction (coordinate offset) with the welding parameter adjustment instruction (parameter increase / decrease) to generate a unified, structured control instruction. The data format of this control instruction is compatible with the controller communication protocol of the welding robot 20.

[0093] Finally, the instruction generation and transmission module 54 sends this control instruction to the controller of the welding robot 20. Upon receiving the instruction, the controller precisely adjusts the motion trajectory of the robotic arm and the welding parameters of the welding torch in real time. This complete process, from data acquisition, stress field reconstruction, instruction generation and transmission to final execution and adjustment, is continuously cycled until the entire welding operation is completed, thereby achieving continuous feedforward adaptive control of the welding process.

[0094] Reference Figure 4 , Figure 4 This is a flowchart illustrating the system cross-validation and anomaly handling phase according to an embodiment of the present invention. The cross-validation and anomaly handling phase is executed in parallel with the feedforward adaptive control closed-loop phase and is handled by the cross-validation and diagnostic module 55.

[0095] During system operation, the cross-validation and diagnostic module 55 continuously and simultaneously receives real-time wave field data from the acoustic feedforward sensing unit 30 and spectral data from the molten pool state diagnostic unit 40. When the rate of change or amplitude of stress value in the feedforward stress field distribution map output by the stress field reconstruction module 53 exceeds a preset fluctuation threshold, the cross-validation and diagnostic module 55 is triggered to perform source diagnosis of stress anomalies.

[0096] The cross-validation and diagnostic module 55 diagnoses the source of anomalies by analyzing the correlation between real-time wavefield data and spectral data over time series. It takes the wave characteristics (e.g., signal amplitude, frequency, and phase) of the two data streams as input and applies a pre-trained classification model (e.g., a decision tree model) for discrimination.

[0097] When the classification model determines that the source of the stress anomaly is material inhomogeneity, for example, when highly correlated, localized fluctuations are detected in acoustic and spectral data at the same time point, the cross-validation and diagnostic module 55 generates a control command. This command is sent to the command generation and transmission module 54, which multiplies the adjustment range of the welding path adjustment command and the welding parameter adjustment command by a preset scaling factor when subsequently combining and generating control commands, thereby adjusting the sensitivity of the system control.

[0098] When the classification model determines that the source of the stress anomaly is an external clamping factor, such as detecting a large-scale, continuous, and severe fluctuation in acoustic data while the spectral data shows no corresponding change, the cross-validation and diagnostic module 55 performs two independent operations. First, it generates an alarm message with a specific code and sends this message to the system's user monitoring interface. Second, it generates a stop command with the highest execution priority and sends this stop command directly to the controller of the welding robot 20 to immediately stop the welding operation.

Claims

1. A self-adaptive welding control system for a special-shaped curved steel structure, characterized in that, The method comprises the following steps: a three-dimensional visual scanning unit for scanning a steel structure workpiece to be welded to generate a three-dimensional digital model; a welding robot as a welding execution unit, the welding robot comprising a mechanical arm and a welding torch fixed to the end of the mechanical arm; an acoustic feedforward sensing unit integrated at the front end of the welding torch for acquiring reference wave field data of the welding area before welding and acquiring real-time wave field data during welding; a molten pool state diagnosis unit for collecting spectral data of the welding molten pool in real time when the welding torch is operating; a central control and calculation unit, the central control and calculation unit comprising: a path planning module for planning an initial welding path according to the three-dimensional digital model generated by the three-dimensional visual scanning unit; a model correction module for dynamically correcting a preset acoustic and stress theoretical model according to the spectral data collected by the molten pool state diagnosis unit in real time; a stress field reconstruction module for reconstructing a feedforward stress field distribution map of the area in front of the welding torch by tomographic inversion algorithm according to the corrected acoustic and stress theoretical model and using the reference wave field data and the real-time wave field data acquired by the acoustic feedforward sensing unit; an instruction generation and sending module for generating a control instruction according to the feedforward stress field distribution map and sending the control instruction to the welding robot to control the welding operation.

2. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The acoustic feedforward sensing unit comprises an ultrasonic phased array emission module and an ultrasonic phased array receiving module.

3. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The model correction module specifically comprises: solving welding molten pool element composition information and temperature information by analyzing the spectral data; dynamically adjusting the acoustic-elastic coefficient in the acoustic and stress theoretical model according to the element composition information and the temperature information.

4. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The central control and calculation unit uses the reference wave field data and the real-time wave field data specifically by: calculating the acoustic wave flight time variation of the real-time wave field data relative to the reference wave field data according to the corrected acoustic and stress theoretical model, and generating the feedforward stress field distribution map by solving the acoustic wave flight time variation through the tomographic inversion algorithm.

5. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The central control and calculation unit generates a control instruction according to the feedforward stress field distribution map specifically by: generating a welding path adjustment instruction by analyzing the stress gradient distribution in the feedforward stress field distribution map; generating a welding parameter adjustment instruction by analyzing the stress amplitude in the feedforward stress field distribution map; and combining the welding path adjustment instruction and the welding parameter adjustment instruction to generate the control instruction and send it to the welding robot.

6. The self-adapting welding control system for a special-shaped curved steel structure according to claim 5, characterized in that, The welding parameters are selected from the group consisting of welding current, welding voltage and welding speed.

7. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The central control and calculation unit further comprises a cross-validation and diagnosis module for cross-validating the real-time wave field data acquired by the acoustic feedforward sensing unit and the spectral data collected by the molten pool state diagnosis unit, and diagnosing the source of stress abnormality.

8. The self-adapting welding control system for a special-shaped curved steel structure according to claim 7, characterized in that, When the source of the stress anomaly is diagnosed as material inhomogeneity, the cross-validation and diagnosis module instructs the instruction generation and sending module to adjust the adjustment amplitude of the welding path adjustment instruction and the welding parameter adjustment instruction when combining to generate the control instruction.

9. The self-adapting welding control system for a special-shaped curved steel structure according to claim 7, characterized in that, The cross-validation and diagnosis module is further used for: When the source of the stress anomaly is diagnosed as external clamping factors, generating an alarm information and a stop instruction for stopping the welding operation.

10. The self-adapting welding control system for a special-shaped curved steel structure according to claim 1, characterized in that, The central control and computing unit further comprises an initial instruction generation module for combining the initial welding path generated by the path planning module with a corresponding set of initial welding parameters to form an initial instruction for starting the welding operation.

Images