An automatic welding path compensation method based on machine vision feature feedback

Abstract

The present application relates to the technical field of welding, and discloses an automatic welding path compensation method based on machine vision feature feedback, which comprises the following steps: collecting molten pool images and extracting liquid-solid phase interface profile curvature asymmetry difference, establishing a thermal capacity boundary distortion model and generating space and thermal enthalpy combined correction parameters according to the asymmetry difference; driving a welding torch to adjust central coordinates, prolonging the residence time of an electric arc at a groove boundary according to the heat deficiency side indicated by the difference, and attenuating the base current when swinging back to the heat excess side; the present application uses asymmetric redistribution of electric arc residence time and energy density to in-situ counteract the base material boundary thermal capacity distortion caused by workpiece thermal deformation, drives the molten pool phase interface to converge to a symmetric equilibrium state, blocks the side wall incomplete fusion and local undercutting from the physical mechanism level, and improves the metallurgical density of the welded joint.

Description

Technical Field

[0001] This invention belongs to the field of welding technology, and in particular relates to an automatic welding path compensation method based on machine vision feature feedback. Background Technology

[0002] Currently, automated welding of medium and heavy plate components is a core process. Due to the large heat input generated during the welding process, the base material faces dynamic nonlinear thermal deformation, which causes the preset physical trajectory to deviate from the actual weld groove center, affecting the metallurgical properties of the welded joint. At present, the main approach is to use trajectory tracking technology based on machine vision, which collects groove features through vision sensors and drives the robot to move according to geometric deviations to compensate for the path.

[0003] Simply relying on mechanical axis spatial coordinate alignment while ignoring the dynamic evolution of thermodynamic boundary conditions leads to a decoupling between path compensation effects and actual fusion quality. For example, Chinese invention patent CN105033419B discloses a mobile welding robot device based on weld seam tracking in a molten pool image. It uses a curvature extremum algorithm to extract the spatial deviation between the welding wire and the leading edge of the molten pool, driving a slider mechanism to correct the deviation. In complex conditions such as deep penetration welding of thick plates, the workpiece's thermal deformation induces asymmetric heat capacity in the base material on both sides of the bevel. While rigid spatial translation logic is used to achieve geometric alignment, the mismatch between arc energy distribution and heat capacity distortion boundaries results in incomplete fusion defects on the heat-deficient side and localized undercut on the heat-excessive side. This disconnect between geometric feedback and enthalpy control limits the complexity of the welding process. The forming quality of welded joints under dynamic conditions; however, under deep penetration welding conditions, the thermal deformation of the base material causes the heat capacity boundary on both sides of the bevel to be in an asymmetrical state. If only rigid translation of spatial coordinates is used, the distribution of arc energy on both sides of the bevel will be unbalanced. The arc moving to the deviation side cannot provide sufficient latent heat of fusion, while heat accumulation occurs on the opposite side, inducing sidewall incomplete fusion or one-sided undercut. In addition, there are interferences such as strong arc light shielding and smoke and dust obstruction at the welding site, which leads to the risk of high-frequency frame loss in visual feature extraction. The inherent inertia of heavy actuators causes physical lag in system response. When the compensation command takes effect, the welding torch position has often deviated from the target compensation point. This phase misalignment between spatial compensation and physical response limits the automated forming quality of complex curved surface components.

[0004] Therefore, the technical problem to be solved by this invention is how to transform visual geometric feedback into asymmetric thermal flow field control that coordinates spatial and thermal input under thermal deformation conditions, and how to eliminate the compensation bias caused by interference and mechanical hysteresis. Summary of the Invention The present invention aims to solve the problem of difficulty in coordinating the correction of geometric trajectory deviation and molten pool thermal balance instability caused by thermal deformation in the automated welding of medium and thick plate components.

[0005] In this technical solution, an automatic welding path compensation method based on machine vision feature feedback includes the following steps: Step S1: Acquire a sequence of images of the molten pool in the welding area and extract the liquid-solid interface profile at the molten pool front. Step S2: Calculate the curvature distribution on the section perpendicular to the reference welding direction based on the liquid-solid interface profile, and determine the curvature asymmetry difference between the liquid-solid interfaces on both sides of the bevel. Step S3: Based on the curvature asymmetry difference, establish a thermal capacity boundary distortion model of the parent material on both sides of the bevel, and generate spatial-enthalpy joint correction parameters including lateral spatial translation, swing timing adjustment, and current waveform correction. Step S4: Drive the welding actuator to adjust the center coordinates of the welding torch according to the lateral spatial translation amount, in order to offset the spatial trajectory deviation caused by the dynamic thermal deformation of the workpiece. Step S5: When the welding torch swings to the heat-deficient side bevel boundary indicated by the curvature asymmetry difference, the residence time of the arc at the heat-deficient side bevel boundary is extended according to the swing timing adjustment amount. Step S6: During the stroke of the welding torch as it swings back from the edge of the groove with insufficient heat to the edge of the groove with excessive heat, the amplitude of the base current output by the welding power source is reduced according to the current waveform correction. The melting energy on both sides of the groove is balanced by the extension of the dwell time and the reduction of the base current amplitude, thereby correcting the penetration depth deviation.

[0006] Preferably, step S2 includes the following sub-steps: step S21, monitoring welding current fluctuation data and extracting current envelope features; step S22, when the molten pool image sequence in step S1 is missing features due to arc light shading, inferring the current physical deviation based on the changing gradient of the current envelope features and correcting the curvature asymmetry difference.

[0007] Preferably, step S4 further includes: obtaining the inherent mechanical response lag time parameter of the welding actuator; constructing an autoregressive model based on the historical compensation vector sequence to predict the expected feature offset of the next control cycle; and using the expected feature offset to correct the lateral spatial translation to neutralize the path tracking lag caused by mechanical inertia.

[0008] Preferably, before extracting the liquid-solid interface contour in step S1, the method further includes: denoising the molten pool image sequence using a median filtering algorithm; identifying the multi-layer, multi-pass weld boundary inside the weld bevel; and extracting the overlapping area features between adjacent weld passes to assist in calibrating the liquid-solid interface contour.

[0009] Preferably, the space-enthalpy joint correction parameters further include: adjusting the arc energy density according to the depth characteristics of the groove to be welded; reducing the oscillation frequency of the welding torch and increasing the peak current output of the welding power source when the groove depth increases; wherein the adjustment range of the oscillation frequency is 0.5Hz to 2.0Hz.

[0010] Preferably, the method further includes the following steps: Step S7, recording the lateral spatial translation amount and space-enthalpy joint correction parameters during the welding process of the root pass to form a reference defect topology sequence; Step S8, during the welding process of the fill layer, the reference defect topology sequence is attenuated according to the preset interlayer smoothing function to generate a pre-calibrated trajectory instruction, and it is superimposed as feedforward data into the real-time compensation loop of the fill layer.

[0011] Preferably, the molten pool image sequence is acquired by a vision sensor equipped with a narrowband filter, the center wavelength of which is 650nm and the bandwidth is 10nm.

[0012] Preferably, the action of attenuating the base current amplitude in step S6 includes: calculating the transient heat accumulation value during the torch swing stroke; and proportionally reducing the output current duty cycle of the welding power source when the transient heat accumulation value exceeds the preset safe penetration depth threshold.

[0013] Preferably, the method further includes the following steps: Step S9, calculating the cumulative vector sum of the lateral spatial translation during the arc termination phase; Step S10, calibrating the zero-point coordinates of the welding actuator based on the cumulative vector sum, so that the remaining positioning error of the welding actuator is controlled within 0.1mm.

[0014] Compared with existing technologies, the automatic welding path compensation method based on machine vision feature feedback of the present invention has the following advantages: 1. In automatic welding path compensation, a reverse hedging mechanism between the thermodynamic boundary and geometric deviation of the molten pool is established to eliminate the hidden danger of thermal instability. In the process of multi-layer and multi-pass welding of thick plates, continuous heat input induces dynamic nonlinear deformation of the base material, which leads to distortion of the heat capacity boundary on both sides of the bevel. This method extracts the asymmetric curvature characteristics of the liquid-solid interface at the front of the molten pool, converts the apparent geometric path deviation into an unbalanced parameter of the enthalpy distribution of the molten pool, and uses this to drive the reconstruction of the arc oscillation sequence and the dynamic correction of the welding current waveform. This increases the residence time of the arc on the heat-deficient side and attenuates the base current amplitude during the swing back to the heat-excess side. By utilizing the asymmetric redistribution of arc energy in the spatial and temporal dimensions, the influence of uneven heat capacity of the base material is hedging in situ, driving the molten pool phase interface to reconverge to a symmetrical equilibrium state, blocking the evolution path of metallurgical defects such as sidewall incomplete fusion and local undercut.

[0015] 2. Constructing a cross-scale fusion perception chain of visual features and current envelope to ensure control continuity in extreme environments. Welding sites present complex conditions such as strong arc light obstruction, wide-range spatter interference, and smoke and dust obstruction. Single visual sensors face the risk of high-frequency frame loss or edge distortion during feature extraction. This method introduces the feature fluctuation envelope of real-time welding current as an auxiliary signal source. Utilizing the physical mapping relationship between arc length and bevel sidewall distance, when the visual perception unit experiences transient blindness or feature loss, the current physical deviation is inferred based on the changing gradient of the current envelope. This synergistic coupling of low-frequency global image features and high-frequency transient electrical signal features ensures the logical continuity of compensation commands in different frequency domains, avoiding arc interruption or severe oscillation of the compensation path due to signal interruption, and improving the operational stability of the welding system in harsh industrial environments.

[0016] 3. Implement feedforward timing misalignment control based on mechanical hysteresis constant to achieve accurate fitting of physical trajectory. Heavy welding actuators have inherent mechanical response hysteresis due to large joint inertia, which often results in a phase difference between the actual bevel position and the activation time of conventional real-time compensation commands, causing periodic serpentine twisting of the weld. This method integrates the mechanical response hysteresis time constant of the actuator in the processing unit and constructs an autoregressive model on the historical compensation vector sequence to predict the expected value of the bias in the next clock cycle. This mechanism converts the spatial compensation command into a predictive adjustment action one physical hysteresis cycle in advance, ensuring that when the issued motion intervention command reaches the physical end, it is precisely aligned with the transient position of the workpiece on the time axis, eliminating the negative impact of mechanical inertia on path tracking accuracy and achieving high-fidelity reproduction of the welding trajectory. Attached Figure Description

[0017] Figure 1 This is a flowchart of the automatic welding path space-enthalpy collaborative compensation process of the present invention; Figure 2 This is a block diagram of the multi-layer, multi-pass weld topology memory and feedforward control logic of the present invention; Figure 3 This is a diagram of the closed-loop interactive system architecture of the visual power actuator of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0020] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood in conjunction with the specific circumstances.

[0021] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0022] An automatic welding path compensation method based on machine vision feature feedback includes the following steps: Step S1: Acquire a sequence of images of the molten pool in the welding area and extract the liquid-solid interface profile at the molten pool front. Step S2: Calculate the curvature distribution on the section perpendicular to the reference welding direction based on the liquid-solid interface profile, and determine the curvature asymmetry difference between the liquid-solid interfaces on both sides of the bevel. Step S3: Based on the curvature asymmetry difference, establish a thermal capacity boundary distortion model of the parent material on both sides of the bevel, and generate spatial-enthalpy joint correction parameters including lateral spatial translation, swing timing adjustment, and current waveform correction. Step S4: Drive the welding actuator to adjust the center coordinates of the welding torch according to the lateral spatial translation amount, in order to offset the spatial trajectory deviation caused by the dynamic thermal deformation of the workpiece. Step S5: When the welding torch swings to the heat-deficient side bevel boundary indicated by the curvature asymmetry difference, the residence time of the arc at the heat-deficient side bevel boundary is extended according to the swing timing adjustment amount. Step S6: During the stroke of the welding torch as it swings back from the edge of the groove with insufficient heat to the edge of the groove with excessive heat, the amplitude of the base current output by the welding power source is reduced according to the current waveform correction. The melting energy on both sides of the groove is balanced by the extension of the dwell time and the reduction of the base current amplitude, thereby correcting the penetration depth deviation.

[0023] Preferably, step S2 includes the following sub-steps: step S21, monitoring welding current fluctuation data and extracting current envelope features; step S22, when the molten pool image sequence in step S1 is missing features due to arc light shading, inferring the current physical deviation based on the changing gradient of the current envelope features and correcting the curvature asymmetry difference.

[0024] Preferably, step S4 further includes: obtaining the inherent mechanical response lag time parameter of the welding actuator; constructing an autoregressive model based on the historical compensation vector sequence to predict the expected feature offset of the next control cycle; and using the expected feature offset to correct the lateral spatial translation to neutralize the path tracking lag caused by mechanical inertia.

[0025] Preferably, before extracting the liquid-solid interface contour in step S1, the method further includes: denoising the molten pool image sequence using a median filtering algorithm; identifying the multi-layer, multi-pass weld boundary inside the weld bevel; and extracting the overlapping area features between adjacent weld passes to assist in calibrating the liquid-solid interface contour.

[0026] Preferably, the space-enthalpy joint correction parameters further include: adjusting the arc energy density according to the depth characteristics of the groove to be welded; reducing the oscillation frequency of the welding torch and increasing the peak current output of the welding power source when the groove depth increases; wherein the adjustment range of the oscillation frequency is 0.5Hz to 2.0Hz.

[0027] Preferably, the method further includes the following steps: Step S7, recording the lateral spatial translation amount and space-enthalpy joint correction parameters during the welding process of the root pass to form a reference defect topology sequence; Step S8, during the welding process of the fill layer, the reference defect topology sequence is attenuated according to the preset interlayer smoothing function to generate a pre-calibrated trajectory instruction, and it is superimposed as feedforward data into the real-time compensation loop of the fill layer.

[0028] Preferably, the molten pool image sequence is acquired by a vision sensor equipped with a narrowband filter, the center wavelength of which is 650nm and the bandwidth is 10nm.

[0029] Preferably, the action of attenuating the base current amplitude in step S6 includes: calculating the transient heat accumulation value during the torch swing stroke; and proportionally reducing the output current duty cycle of the welding power source when the transient heat accumulation value exceeds the preset safe penetration depth threshold.

[0030] Preferably, the method further includes the following steps: Step S9, calculating the cumulative vector sum of the lateral spatial translation during the arc termination phase; Step S10, calibrating the zero-point coordinates of the welding actuator based on the cumulative vector sum, so that the remaining positioning error of the welding actuator is controlled within 0.1mm.

[0031] Example 1: In multi-layer, multi-pass automated welding of heavy ship components, the workpiece undergoes dynamic thermal deformation due to the welding heat input, causing a spatial offset between the preset welding trajectory and the actual weld groove center. Conventional path compensation methods only align the deviation position by lateral translation of the mechanical axis coordinates, without considering the thermal capacity boundary distortion of the base material on both sides of the groove caused by thermal deformation. This results in an imbalance in heat distribution when the arc moves towards the deviation side, leading to sidewall incomplete fusion or local undercut defects during the molten pool cooling and solidification cycle. The automatic welding path compensation method based on machine vision feature feedback uses a vision sensor equipped with a narrow-band filter to acquire image sequences of the molten pool area and extract the liquid-solid interface contour at the molten pool front. The processing unit then calculates... The curvature distribution of the liquid-solid interface profile on the cross section perpendicular to the reference welding direction is used to determine the curvature asymmetry difference between the liquid-solid interfaces on both sides of the bevel. The system determines the thermal capacity boundary distortion state of the base material on both sides of the bevel based on the curvature asymmetry difference, and generates a space-enthalpy joint correction parameter including lateral spatial translation, oscillation timing adjustment, and current waveform correction. The underlying logic of this derivation is that the physical form of the liquid metal at the solidification front is directly controlled by the heat absorption capacity of the surrounding solid medium. When the welding torch is biased, causing the volume of the base material on one side to increase relatively, the local heat dissipation rate of the metal on that side increases sharply, causing the liquid metal to shrink and solidify inward at an extremely high cooling rate, which macroscopically presents as the asymmetric flattening characteristic of the phase interface curvature.

[0032] Therefore, the processing unit utilizes this physical phenomenon and, based on the Fourier heat conduction equation, directly and equivalently reverses the apparent curvature difference into an index of the implicit thermodynamic endothermic potential energy difference on both sides of the bevel. While driving the welding actuator to adjust the torch center coordinates according to the lateral spatial translation, when the torch is detected to be at the heat-deficient bevel boundary determined by the curvature asymmetry difference, the arc dwell time at the heat-deficient bevel boundary is extended according to the swing timing adjustment. During the torch's swing back from the heat-deficient bevel boundary to the heat-excess bevel boundary, the base current amplitude of the welding power supply is reduced according to the current waveform correction. The synergistic compensation mechanism composed of spatial geometric features and arc thermodynamic state uses the asymmetric redistribution of arc dwell time and energy density to compensate for the thermal capacity distortion of the base material boundary caused by workpiece thermal deformation, driving the molten pool liquid-solid phase interface to converge towards a symmetrical equilibrium state. This converts the geometric constraint deviation into an internal heat flow field distribution control process, enabling the welded joint to maintain the consistency of metallurgical density and mechanical properties under dynamic working conditions.

[0033] Example 2: Verifying the stability of the automatic welding path compensation method under complex working conditions. An experimental platform was built using a six-axis welding robot with a repeatability of 0.05 mm, an inverter welding power supply with a sampling frequency of 100 kHz, and a CMOS industrial camera integrated with a narrowband filter with a center wavelength of 650 nm. The data collected by this experimental platform came from the raw molten pool image stream acquired by a vision sensor at a frame rate of 60 fps. The effective pixel resolution of the vision sensor was 1920 × 1080, used to capture microscopic deformation features of the liquid-solid interface at the 0.01 mm level. In the experimental design, the sampling period... The setting is limited by the welding oscillation frequency and the real-time throughput of the image processing unit. The technical trade-off lies in maintaining the dynamic response speed of the control loop while avoiding computational overflow. When the welding oscillation frequency is in the range of 2.0Hz to 5.0Hz, in order to obtain a sufficient number of contour feature points to satisfy the Nyquist sampling criterion and suppress aliasing interference caused by arc flicker, the sampling period... Tendency towards its lower limit, this experiment sets it to 16.7 ms to maintain the continuity of spatial compensation at a welding speed of 1.5 m / min. The experiment uses 16.0 mm thick Q355B low-alloy high-strength structural steel as the base material. By pre-setting non-uniform constraints, the dynamic thermal deformation of heavy components under high heat input is simulated, causing a non-linear spatial offset of 2.15 mm to 4.88 mm at the bevel center within a 500.0 mm length. To simulate a real industrial electromagnetic environment, Gaussian white noise with a signal-to-noise ratio of 20 dB is actively superimposed on the video signal acquisition link, and 50 Hz power frequency harmonic interference is introduced. Three control groups are set up: control group A only performs geometric translation compensation of the mechanical axis coordinates; control group B, as a partially missing control group, only performs lateral translation and oscillation timing adjustment without introducing current waveform correction; and control group C, as an out-of-range control group, adjusts the oscillation timing amount... The value is set to 500.0ms, which deviates from the working window of 50.0ms to 200.0ms. The sample group of this invention implements the complete space-enthalpy joint correction process, and verifies the system's ability to suppress the distortion of the heat capacity boundary of the parent material through multi-dimensional data comparison.

[0034] During welding, the vision sensor captured an irregular, jagged outline of the original molten pool affected by thermal interference. The liquid-solid interface outline exhibited extremely high pixel grayscale fluctuations before processing. The processing unit, by calling a liquid-solid interface outline extraction algorithm, extracted the liquid-solid interface outline curves on the left and right sides of the bevel after eliminating 20dB noise interference. To eliminate the loss of depth information caused by the dimensionality reduction observation of the monocular camera, a standard three-dimensional geometric prior mesh model matching the thickness of the component to be welded and the bevel angle was pre-loaded into the system memory. After extracting the apparent two-dimensional pixel coordinate flow, the processing unit used the offline calibrated camera intrinsic and extrinsic parameter matrix, coupled with the transient molten pool estimated based on the current welding current and voltage. The axial depth value is obtained by inversely projecting the two-dimensional plane curve onto the corresponding three-dimensional cross-sectional coordinate system through epipolar geometric constraints, thereby accurately reconstructing a real spatial contour with absolute physical depth. When the robot moves to the section where thermal deformation intensifies, the system monitors the change in the curvature distribution of the liquid-solid interface contour on the cross-section perpendicular to the welding direction, including the real-time curvature on the heat-deficient side. for Real-time curvature of the side with excess heat for The curvature asymmetry difference was calculated. for Based on the curvature asymmetry difference The thermal capacity boundary distortion model is driven to generate a 3.22mm lateral spatial translation and a 125.0ms oscillation timing adjustment. and 18.5A current waveform correction The compensation instruction.

[0035] Experimental observation data show that, after geometric alignment, control group A still had a 1.25mm wide unfused strip at the root of the bevel on the heat-deficient side, and a 0.45mm deep undercut defect appeared on the heat-excess side. This confirms that single geometric compensation cannot solve the problem of asymmetric heat capacity distribution caused by workpiece thermal deformation. Although control group B improved the melting state on the deficient side by extending the dwell time by 125.0ms, the undercut depth on the excess side was only reduced to 0.32mm due to the lack of real-time attenuation of the swing stroke energy by the current waveform correction amount ΔI, failing to eliminate the defect. After excessively extending the dwell time to 500.0ms, the heat accumulation in group C exceeded the bearing limit of the base material, causing irreversible collapse of the liquid-solid interface of the molten pool. The surface tension of the molten pool could not maintain metal equilibrium. After implementing the space-enthalpy joint correction parameters, the curvature of the liquid-solid interface profile on both sides of the molten pool tended to be consistent, and the curvature asymmetry difference was reduced. Depend on convergence to The molten pool liquid-solid interface returns to a symmetrical equilibrium state; as the thermal deformation offset increases from 2.15mm to 4.88mm, the system output oscillation timing adjustment amount... The offset exhibits a definite positive correlation gradient change, demonstrating the ability to quantitatively perceive and dynamically respond to different deviation intensities. The final metallographic test results show that the weld sidewall is well fused, and the metallurgical density of the joint is increased from 92.5% in the control group A to 99.2%, solving the coupling problem of geometric deviation and heat distribution imbalance under dynamic thermal deformation conditions.

[0036] Example 3: This example combines Figures 1 to 3 This paper describes an automatic welding path compensation method based on machine vision feature feedback, such as... Figure 1 As shown, steps S1 to S6 are as follows: Specifically, the system executes step S1, acquiring a sequence of molten pool images of the welding area and extracting the liquid-solid interface profile at the molten pool front; then proceeding to step S2, calculating the curvature distribution on the cross section perpendicular to the reference welding direction based on the liquid-solid interface profile to determine the curvature asymmetry difference between the liquid-solid interfaces on both sides of the bevel; executing step S3, establishing a thermal capacity boundary distortion model of the base material on both sides of the bevel based on the curvature asymmetry difference, and generating a space-enthalpy joint correction parameter including lateral spatial translation, oscillation timing adjustment, and current waveform correction; then executing step S4, driving the welding process. The mechanism adjusts the center coordinates of the welding torch according to the lateral spatial translation to counteract the spatial trajectory deviation caused by the dynamic thermal deformation of the workpiece. In step S5, when the welding torch swings to the heat-deficient bevel boundary indicated by the curvature asymmetry difference, the dwell time of the arc at the heat-deficient bevel is extended according to the swing timing adjustment. Finally, step S6 is executed. During the stroke of the welding torch swinging back from the heat-deficient bevel boundary to the heat-excess bevel boundary, the amplitude of the base current output of the welding power source is reduced according to the current waveform correction. The melting energy on both sides of the bevel is balanced and the penetration depth deviation is corrected by extending the dwell time and reducing the base current amplitude.

[0037] like Figure 2 The diagram illustrates the specific implementation flow of the multi-layer, multi-pass weld topology memory and feedforward control logic in this invention, corresponding to steps S7 and S8 in the method. In step S7, during the root pass welding process, the system records in real-time the lateral spatial translation and space-enthalpy joint correction parameters generated in that pass, mapping them to form a reference defect topology sequence. In step S8, when entering the fill layer welding process, the system calls a preset inter-layer smoothing function to attenuate the amplitude of the reference defect topology sequence, generating a pre-calibration trajectory instruction. This pre-calibration trajectory instruction, as feedforward data, is superimposed in real-time onto the real-time compensation loop of the current fill layer, through... Figure 2 The feedforward control chain shown transforms the root deformation history data into upper-level trajectory prediction, effectively suppressing the cumulative error in the multi-stage welding process.

[0038] like Figure 3 As shown, the automatic welding path compensation system achieves closed-loop regulation through the interaction of a vision sensor, a welding power source, and a welding actuator. The vision sensor is responsible for acquiring a sequence of images of the molten pool and extracting the liquid-solid interface contour. The logic processing unit calculates and determines the curvature asymmetry difference based on the extracted liquid-solid interface contour, and then establishes a thermal capacity boundary distortion model based on the curvature asymmetry difference. Based on this, it generates spatial-enthalpy joint correction parameters. The generated spatial-enthalpy joint correction parameters contain two-dimensional control components: on the one hand, it outputs a lateral spatial translation command to the welding actuator to adjust the center coordinates of the welding torch; on the other hand, it outputs an adjustment command to the welding power source to balance the melting energy on both sides of the bevel, thereby achieving closed-loop coordinated control of spatial position and heat input distribution.

[0039] Example 4: In the automatic welding process of 2.0mm thick TC4 titanium alloy sheet, because the thermal conductivity of this material is less than 10.0W / (m²), K) and is sensitive to heat accumulation. The transient heat diffusion resistance generated during welding induces asymmetric thermal capacity boundary distortion on both sides of the bevel. At this time, the molten pool image acquired by the visual sensor is interfered with by the arc plasma glow, and the pixel gradient difference of the liquid-solid interface contour decreases, causing the bevel sidewall melting boundary identified by the system to deviate from the physical boundary. The path compensation command exhibits logical oscillation. To solve the above image frame blinding problem, the system synchronously acquires welding current fluctuation data at high frequency and extracts its outer envelope. Since the medium-thick plate to be welded has inherent V-shaped or U-shaped bevel geometric constraints, when the welding torch deviates from the bevel centerline, the lateral slippage will inevitably cause the arc to move along the bevel centerline. The rising or falling of the inclined sidewall causes a rigid change in the three-dimensional arc length, which is directly reflected as a proportional step change in the current amplitude. The system utilizes this geometric trigonometric function correlation mechanism to convert the currently extracted single-dimensional current envelope change gradient, combined with a given bevel angle, into a two-dimensional lateral offset vector along the sidewall. This vector serves as an equivalent substitute input during visual current interruption to continuously correct curvature asymmetry differences. The processing unit limits the pixel search range by defining a rectangular region of interest centered on the welding torch center, segments the region of interest using a dynamic gradient thresholding method based on local mean, and calculates the average grayscale value of the current frame image. And accumulate dynamic offset factors A binarization threshold is determined, and morphological closing operations are used to fill discrete holes in the generated binarized mask. The liquid-solid interface contour at the molten pool front is extracted using single-pixel edge tracking for subsequent curvature asymmetry interpolation. The calculation.

[0040] The system will calculate the curvature asymmetry difference. This is converted into a low-level execution operator that controls the distribution of welding enthalpy, where the oscillation timing adjustment amount... Satisfy linear compensation logic ,in, For the pre-calibrated thermal response sensitivity coefficient, the processing unit outputs the remaining two joint correction parameters based on the physical coupling relationship between the spatial trajectory and the internal thermal flow field. The evolution of spatial deviation and the curvature distortion of the molten pool phase interface are directly geometrically correlated, using the formula... Determine the lateral spatial translation amount The energy accumulation within a specific time series is evaluated based on the simplified heat transfer principle of a moving point heat source, according to the formula. Calculate the corresponding current waveform correction amount. The proportional parameter in the aforementioned calculation process The geometric amplification effect under a specific bevel angle limits the peak displacement of the translation mechanism, and the integral gain constant. Characterizing the arc enthalpy compensation required within a unit curvature distortion span ensures smooth and continuous output of the underlying power supply; acquisition step size parameters are used. Defining a complete data evaluation time window filters out transient noise caused by high-frequency oscillations in the molten pool. In actual engineering configurations, the processing unit is based on the actual assembly bevel angle. Tangent function relationship for calibration scaling parameters To prevent translational overshoot, it is rigidly limited to the range of 1.5 to 3.0; integral gain constant The equivalent slope is obtained by conducting a unilateral thermal flow field melting test on a controlled test plate beforehand, and is preset to 25.0. This provides a definite initial gain for the subsequent control loop, and the thermal response sensitivity coefficient. The determination was based on an experimental calibration procedure using the thermal diffusivity of the base material. This involved gradually increasing the bevel offset under the same linear energy and measuring the recovery time of the liquid-solid interface on both sides of the molten pool until the curvature asymmetry difference was reached. convergence to The stable interval within which the curvature asymmetry difference is measured; The measured value exceeded [the specified value] within 10 consecutive sampling periods. When the threshold is reached, the system calls the linear compensation logic to generate a timing adjustment that includes the oscillation. The control commands drive the welding actuator to increase the energy input time at the bevel boundary on the heat-deficient side.

[0041] After implementing the space-enthalpy joint correction parameters, the molten pool solid-phase interface during the welding process of TC4 titanium alloy thin plates returns to a highly symmetrical equilibrium state, and the thermal response sensitivity coefficient is improved. 850.0ms Under the condition of mm, curvature asymmetry difference Depend on convergence to The thermal capacity boundary distortion model inference is transformed into a deterministic thermal response compensation operator, which maintains deep penetration under a geometric deviation of 0.8 mm, and reduces the fish scale pattern spacing deviation on the joint weld surface to 0.15 mm.

[0042] Example 5: When the system is in the sensor optical axis angle deflection calibration state, the mapping relationship between pixel space and physical space is determined using a standard geometric sample within the field of view of a vision sensor equipped with a narrow-band filter with a center wavelength of 650nm. The processing unit calculates the pixel length of the standard sample edge at a resolution of 1920×1080 pixels and obtains the physical size conversion coefficient. Simultaneously, background grayscale distribution under different exposure levels was collected while the simulated electric arc fluctuated within the range of 150A to 250A. The dynamic offset factor was determined based on the background grayscale distribution. The reference deviation at different melt pool depths is used to complete the transformation from the pixel coordinate system to the physical coordinate system of the liquid-solid interface profile; among which... This is the physical size conversion factor. This is the dynamic offset factor.

[0043] In the process parameter calibration procedure for a base material with specific thermophysical properties, the system determines an offline compensation reference matrix based on the thermophysical property benchmark determined by the thermal conductivity and thermal diffusivity of the base material, and calibrates the oscillation timing adjustment amount by testing the unilateral thermal resistance under a controlled experimental environment. Difference with curvature asymmetry The response weights between the parameters are used by the processing unit as thermal response sensitivity coefficients. The current waveform correction amount is stored in the control register of the heat capacity boundary distortion model and is determined synchronously at a welding speed of 1.5 m / min. The regulation of the recovery time of the molten pool in thermal equilibrium state was determined to achieve the thermal response sensitivity coefficient. Synchronous configuration with current regulation parameters.

[0044] Example 6: In a newly deployed automated welding unit, the spatial coordinate systems of the vision sensor and the robot are in a state of unalignment. The system uses a standard grid calibration plate with a 2.0mm pitch to calibrate the mapping relationship between pixel space and physical space within the field of view of the vision sensor equipped with a narrow-band filter with a center wavelength of 650nm. The processing unit acquires 100 frames of calibration images and extracts the average pixel coordinates of the grid feature points. The physical size conversion coefficient is determined based on the pixel distance of the grid feature points and the actual physical distance. To linearly compensate for spatial stepping errors, the background grayscale distribution of the molten pool under different exposure levels is collected simultaneously when the simulated arc current fluctuates within the range of 150A to 250A. The dynamic offset factor is determined based on the background grayscale distribution at different depth positions of the molten pool. The baseline deviation component is used to determine the alignment parameters between the liquid-solid interface contour extraction algorithm and the current visual hardware environment; where This is the physical size conversion factor. This is the dynamic offset factor.

[0045] In the process parameter initialization procedure for a new batch of Q355B low-alloy high-strength structural steel base material, the system presets a gradient spatial offset of 1.0mm to 5.0mm on the controlled test plate. The processing unit records the curvature asymmetry difference of the liquid-solid interface profile induced by the difference in thermal resistance of the bevel sidewalls under each offset. By adjusting the swing timing Starting from 50.0 ms, the time increments are gradually increased in steps of 25.0 ms, observing the profile of the molten pool front from an asymmetric equilibrium state to... The response law of residual interval convergence within a certain range is used to process the collected curvature asymmetry difference using the processing unit. With oscillation timing adjustment amount Linear regression calculations are performed on discrete data points, and the resulting slope is used as the thermal response sensitivity coefficient. The control law register stored in the thermal capacity boundary distortion model is used to simultaneously verify the current waveform correction at a welding speed of 1.5 m / min. The sensitivity of droplet transition frequency regulation is used to determine the adaptation parameters of arc dwell time and energy distribution logic to the thermal properties of specific base materials.

[0046] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. An automatic welding path compensation method based on machine vision feature feedback, characterized in that, Includes the following steps: Step S1: Acquire a sequence of images of the molten pool in the welding area and extract the liquid-solid interface profile at the leading edge of the molten pool. Step S2: Calculate the curvature distribution on the section perpendicular to the reference welding direction based on the liquid-solid interface profile, and determine the curvature asymmetry difference between the liquid-solid interfaces on both sides of the bevel. Step S3, based on curvature asymmetry difference A thermal capacity boundary distortion model of the parent material on both sides of the bevel is established, and a space-enthalpy joint correction parameter is generated, including lateral spatial translation, oscillation timing adjustment, and current waveform correction. The oscillation timing adjustment... Satisfying linear compensation logic ,in, For the pre-calibrated thermal response sensitivity coefficient, the processing unit outputs the remaining two joint correction parameters based on the physical coupling relationship between the spatial trajectory and the internal thermal flow field. The evolution of spatial deviation and the curvature distortion of the molten pool phase interface are directly geometrically correlated, using the formula... Determine the lateral spatial translation amount The energy accumulation within a specific time series is evaluated based on the simplified heat transfer principle of a moving point heat source, according to the formula. Calculate the corresponding current waveform correction amount. Among them, the proportional parameter The geometric amplification effect under a specific bevel angle limits the peak displacement of the translation mechanism, and the integral gain constant. Characterizes the amount of arc enthalpy compensation required within a unit curvature distortion span, and collects step size parameters. Define the time window for complete data evaluation; Step S4: Drive the welding actuator to adjust the center coordinates of the welding torch according to the lateral spatial translation amount, in order to offset the spatial trajectory deviation caused by the dynamic thermal deformation of the workpiece. Step S5: When the welding torch swings to the heat-deficient side bevel boundary indicated by the curvature asymmetry difference, the residence time of the arc at the heat-deficient side bevel boundary is extended according to the swing timing adjustment amount. Step S6: During the stroke of the welding torch as it swings back from the edge of the groove with insufficient heat to the edge of the groove with excessive heat, the amplitude of the base current output by the welding power source is reduced according to the current waveform correction. The melting energy on both sides of the groove is balanced by the extension of the dwell time and the reduction of the base current amplitude, thereby correcting the penetration depth deviation.

2. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21, monitor welding current fluctuation data and extract current envelope features; Step S22: When the molten pool image sequence acquired in step S1 is missing features due to arc shading, the current physical deviation is deduced based on the changing gradient of the current envelope features, and the curvature asymmetry difference is corrected.

3. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, Step S4 also includes: obtaining the inherent mechanical response lag time parameters of the welding actuator; constructing an autoregressive model based on the historical compensation vector sequence to predict the expected characteristic offset of the next control cycle; and using the expected characteristic offset to correct the lateral spatial translation to neutralize the path tracking lag caused by mechanical inertia.

4. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, Before extracting the liquid-solid interface contour in step S1, the following steps are also included: denoising the molten pool image sequence using a median filtering algorithm; identifying the multi-layer, multi-pass weld boundary inside the weld bevel; and extracting the overlapping area features between adjacent weld passes to assist in calibrating the liquid-solid interface contour.

5. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, The space-enthalpy joint correction parameters also include: adjusting the arc energy density according to the depth characteristics of the groove to be welded; reducing the oscillation frequency of the welding torch and increasing the peak current output of the welding power source when the groove depth increases; wherein the adjustment range of the oscillation frequency is 0.5Hz to 2.0Hz.

6. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, The method also includes the following steps: Step S7, record the lateral spatial translation amount and space-enthalpy joint correction parameters during the welding process of the root pass to form a reference defect topology sequence; Step S8, during the welding process of the fill layer, perform amplitude attenuation on the reference defect topology sequence according to the preset interlayer smoothing function, generate a pre-calibrated trajectory instruction, and superimpose it as feedforward data into the real-time compensation loop of the fill layer.

7. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, The sequence of molten pool images was acquired by a vision sensor equipped with a narrowband filter, which has a center wavelength of 650 nm and a bandwidth of 10 nm.

8. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, The action of attenuating the base current amplitude in step S6 includes: calculating the transient heat accumulation value during the torch swing stroke; and proportionally reducing the output current duty cycle of the welding power source when the transient heat accumulation value exceeds the preset safe penetration threshold.

9. The automatic welding path compensation method based on machine vision feature feedback according to claim 1, characterized in that, The method also includes the following steps: Step S9, calculating the cumulative vector sum of the lateral spatial translation during the arc termination phase; Step S10, calibrating the zero-point coordinates of the welding actuator based on the cumulative vector sum, so that the remaining positioning error of the welding actuator is controlled within 0.1mm.

Images