An intelligent welding workstation

Abstract

The application discloses an intelligent welding workstation, and belongs to the technical field of automatic welding. The intelligent welding workstation comprises a welding robot, a welding power supply, a laser profile sensor and a control device. The control device is configured to: control the robot to perform a non-welding rehearsal motion before welding, and generate a rehearsal topography point cloud; divide a weld into multiple independent sections according to structural features such as wall thickness mutation, reinforcing rib distribution and heat dissipation boundary represented by the point cloud, and match a heat deformation weight coefficient; pre-construct a lightweight proxy model with welding current and voltage as input and weld point offset vector as output, and the output amplitude is modulated and corrected by the section weight coefficient; and during welding, the welding gun trajectory is corrected in real time as the sum of the rehearsal point cloud coordinates and the offset vector after modulation. The application realizes differentiated gradient compensation of different sections in the same weld, avoids the local welding deviation problem caused by traditional global unified compensation, and significantly improves the weld trajectory accuracy and deposition consistency.

Description

Technical Field

[0001] This application relates to the field of automated welding, and in particular to an intelligent welding workstation. Background Technology

[0002] In the field of automated welding of large structural components such as automotive axles, frames, and battery trays, welding robot workstations have been widely adopted. To compensate for workpiece thermal deformation caused by welding heat input, existing technologies mainly employ two types of solutions: one is an open-loop compensation method based on offline static deformation calibration, which involves pre-measuring the welding deformation patterns of workpieces of the same model and pre-setting a fixed compensation amount in the welding trajectory during programming; the other is a closed-loop compensation method based on online vision sensing, which involves using laser vision sensors to detect the weld position in real time during the welding process and dynamically correcting the robot trajectory. The former is cumbersome to operate and cannot adapt to deviations caused by random factors on-site, such as fluctuations in clamping consistency and changes in ambient temperature; the latter faces severe interference from welding arc light, fumes, and spatter, making it difficult for sensors to stably acquire the true morphological data of the area near the molten pool during the welding process, resulting in inherent defects in measurement robustness.

[0003] However, a more prominent problem lies in the fact that the existing compensation methods mentioned above, whether open-loop calibration or closed-loop detection, treat a weld as a uniformly deformed whole, using a single global compensation vector for trajectory correction. In actual automotive structural component welding, different sections of the same long weld often exhibit significant differences in wall thickness, uneven distribution of reinforcing ribs or bending structures, and vastly different heat dissipation boundary conditions, leading to drastically different heat accumulation and deformation patterns in each section during the welding process. For example, the initial section of the weld dissipates heat quickly with less heat accumulation and smaller deformation; the middle section reaches its peak heat accumulation with larger deformation; and the heat dissipation conditions change again at the end, causing deformation to decrease again. This "one-size-fits-all" global compensation strategy inevitably leads to insufficient or excessive compensation in different sections of the weld, ultimately resulting in localized weld misalignment and uneven weld bead width, failing to meet the quality requirements of precision welding for weld consistency.

[0004] Therefore, it is necessary to propose an intelligent welding workstation that can achieve segmented differential compensation within the same weld seam, in order to solve the problem that the existing global uniform compensation strategy cannot adapt to workpieces with structural differences. Summary of the Invention

[0005] To address the aforementioned problems, this application provides an intelligent welding workstation, employing the following technical solution: An intelligent welding workstation includes: A welding robot with a welding torch and a laser profile sensor mounted at its end; A welding power source is used to stably supply welding current and welding voltage to the welding torch; The control device is communicatively connected to the welding robot, the laser profile sensor, and the welding power source; the control device is configured to: Before the welding operation begins, the welding robot is controlled to perform a no-welding pre-simulation motion along the weld seam path, and the workpiece surface topography data is collected simultaneously through the laser contour sensor to generate a pre-simulation topography point cloud. Based on the wall thickness abrupt change, stiffener distribution and heat dissipation boundary characteristics in the pre-simulated topography point cloud, the entire weld seam is automatically divided into multiple independent segments along the weld seam path, and a corresponding thermal deformation weight coefficient is matched and assigned to each segment. A lightweight proxy model for welding heat input response is pre-built and stored. The proxy model takes the welding current and welding voltage collected in real time as input and the weld point offset vector as output. The output amplitude of the weld point offset vector is modulated and corrected by the thermal deformation weight coefficient of the corresponding section of the welding position. During the actual welding process, the real-time target motion trajectory of the welding torch is corrected to the sum of the corresponding position coordinates in the pre-simulated topography point cloud and the modulated and corrected welding point offset vector, based on the welding point offset vector output in real time by the proxy model and modified by the segment thermal deformation weight coefficient.

[0006] Preferably, when the control device automatically divides the entire weld into multiple independent sections along the weld path, it extracts the wall thickness change points, stiffener distribution positions, and heat dissipation boundary features from the pre-simulated topography point cloud as the basis for dividing the independent sections of the weld.

[0007] Preferably, the control device is configured as follows: Based on the average wall thickness, local structural stiffness, and real-time heat conduction conditions of each weld section, the thermal deformation weight coefficient of the corresponding section is determined through pre-calibration experiments; among them, the larger the workpiece wall thickness and the stronger the local structural stiffness, the smaller the corresponding thermal deformation weight coefficient value.

[0008] Preferably, the lightweight proxy model for welding heat input response is a parameterized response model established based on historical experimental data. It uses welding current and welding voltage as core input variables and employs data tables and bilinear interpolation to output the corresponding weld point offset vector in real time.

[0009] Preferably, the control device is further configured to: after a single weld seam is completed and the workpiece is sufficiently cooled, control the welding robot again to replicate the scanning path consistent with the pre-welding simulation motion, and collect the workpiece cooling topography point cloud; generate a weld seam solidification difference map by performing a three-dimensional difference operation between the cooled topography point cloud and the pre-simulated topography point cloud.

[0010] Preferably, during the pre-welding motion stage without welding rehearsal, the laser contour sensor continuously collects workpiece contour data at a preset acquisition frequency, while the welding power supply maintains a standby state with zero current and voltage output to avoid interference from welding arc light and spatter on the shape acquisition.

[0011] Preferably, the control device communicates with the welding power source at high speed via an industrial real-time bus, continuously acquiring the current and voltage values ​​during the welding process in real time at a sampling frequency of not less than 1 kHz.

[0012] Preferably, it also includes a welding fume purification and heat recovery module integrated inside the workstation; the control device is signal-connected to the welding fume purification and heat recovery module, and can adaptively adjust the purification operation intensity and heat recovery efficiency according to the welding start / stop status and welding power.

[0013] Preferably, the control device uses an industrial PC or an embedded dedicated welding controller, which communicates bidirectionally with the welding robot controller via real-time industrial Ethernet to quickly exchange trajectory correction commands and motion status data.

[0014] In summary, this application includes at least one of the following beneficial technical effects: 1. By performing a pre-scan in a clean optical environment free from welding interference before welding, a high-precision pre-scan topographic point cloud is established as a geometric reference, which fundamentally avoids the interference of welding arc light, smoke, and spatter on topographic acquisition and solves the inherent defect of insufficient measurement robustness of traditional online visual sensing solutions during welding.

[0015] 2. By extracting the locations of abrupt changes in wall thickness, the distribution of reinforcing ribs, and the characteristics of heat dissipation boundaries based on the pre-simulated topographic point cloud, the entire weld seam is automatically divided into multiple independent sections with relatively consistent structural features. A thermal deformation weight coefficient is quantitatively matched for each section, so that the compensation strategy evolves from the traditional global uniformity to regional differentiation, thus avoiding the problem of insufficient or excessive local compensation caused by the unevenness of the workpiece structure.

[0016] 3. By constructing a lightweight proxy model of welding heat input response with real-time welding current and voltage as inputs, and modifying the model output through segment weight coefficient modulation, gradient compensation matching the thermal deformation sensitivity of each segment is automatically executed in different segments of the same weld, which significantly improves the trajectory accuracy and deposition consistency of the weld.

[0017] 4. By adopting a lightweight proxy model to replace the finite element full physical simulation, the calculation time for a single compensation is controlled within milliseconds, enabling the partitioned differential compensation to run in real time on ordinary industrial controllers at a frequency of no less than 1kHz, which meets the stringent requirements of the welding process for control real-time performance, while avoiding the high investment in computing hardware.

[0018] 5. By relying solely on stable and measurable electrical signals such as welding current and voltage as inputs through the proxy model, and without relying on any optical or acoustic sensors during the welding process for online navigation, the entire trajectory correction process is naturally immune to harsh working conditions such as welding arc light, fumes, and spatter. The system's robustness is significantly better than existing online sensing solutions.

[0019] 6. By integrating welding fume purification and heat recovery modules into the workstation, and adaptively adjusting the purification intensity and heat recovery efficiency according to the welding start / stop status and power level, the effective recovery and utilization of welding waste heat is achieved while ensuring environmental compliance, thereby reducing the overall operating energy consumption of the workstation. Attached Figure Description

[0020] Figure 1 This is a system block diagram of an intelligent welding workstation according to an embodiment of this application; Figure 2 This is a flowchart illustrating the operation method of an intelligent welding workstation according to an embodiment of this application.

[0021] Explanation of reference numerals in the attached drawings: 1. Welding robot; 11. Welding torch; 12. Laser profile sensor; 2. Welding power source; 3. Control device. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1 and Figure 2 This application will be described in further detail.

[0023] This application discloses an intelligent welding workstation. This workstation is particularly suitable for automated welding of large structural components such as automotive axles, frames, and battery trays, where the workpiece structure has long weld seams and unevenness along the weld seam path.

[0024] Reference Figure 1 An intelligent welding workstation includes: a welding robot 1, a welding power source 2, a laser profile sensor 12, and a control device 3.

[0025] The welding robot 1 is a six-axis industrial robot, with a welding torch 11 and a laser profile sensor 12 mounted on its end effector via a flange. The welding torch performs the welding operation, and the laser profile sensor collects surface topography data of the workpiece. The laser profile sensor is preferably a line laser profiler, with a preset sampling frequency selectable from 100Hz to 500Hz, adjustable according to welding speed and accuracy requirements. In this embodiment, the sampling frequency is configured at 200Hz. Under the condition of a robot movement speed of 20mm / s, a point cloud sampling interval of 0.1mm can be obtained along the weld path direction, and the height measurement accuracy is 0.02mm.

[0026] Welding power source 2 is an inverter-type digital welding power source used to stably supply welding current and welding voltage to the welding torch. The welding power source supports external communication control and can receive target current and voltage values ​​issued by the control device through an industrial real-time bus, and provide real-time feedback on the current and voltage values ​​being actually output.

[0027] Control device 3 employs an industrial PC, communicating bidirectionally with the welding robot controller via real-time industrial Ethernet to rapidly exchange trajectory correction commands and motion status data. Simultaneously, the control device communicates at high speed with the welding power source via an industrial real-time bus, continuously acquiring current and voltage values ​​during the welding process at a sampling frequency of 1kHz. The control device also communicates with a laser contour sensor to receive its collected shape data.

[0028] The control unit, as the core computing and scheduling unit of the workstation, is configured to perform the following four stages of operations: pre-welding pre-scanning and point cloud generation, weld structure zoning and weight coefficient allocation, pre-construction and storage of the surrogate model, and real-time trajectory correction during the welding process. (Refer to...) Figure 2 The following will explain each point in detail.

[0029] S1. Pre-welding pre-scanning and point cloud generation: Before welding begins, the workpiece to be welded is first clamped and fixed in the workstation. The control device then controls the welding robot to perform a pre-set weld path without welding. During this stage, the welding power supply remains in standby mode with zero current and voltage output, the main circuit does not output any welding power, and there is no welding arc or spatter interference in the environment.

[0030] The laser contour sensor carried by the welding robot continuously collects workpiece surface contour data at a preset acquisition frequency of 200Hz as the robot moves along the weld seam path. Each frame of contour data contains hundreds of three-dimensional spatial coordinate points of the weld seam and the heat-affected zones on both sides. After coordinate transformation and stitching, all frames of contour data generate a high-density three-dimensional point cloud reflecting the original geometry of the workpiece before welding, called the "pre-modeling topography point cloud". Each data point in the pre-modeling topography point cloud contains three-dimensional spatial coordinates (X, Y, Z).

[0031] Because the scanning process is completed in a clean optical environment without welding, the pre-modeled topographic point cloud has extremely high accuracy and integrity, providing an accurate geometric reference for all subsequent steps.

[0032] S2. Weld structure zoning and weight coefficient allocation After obtaining the pre-simulated topographic point cloud, the control device automatically divides the entire weld into multiple independent segments along the weld path based on the workpiece structural features represented by the point cloud, and assigns a corresponding thermal deformation weight coefficient to each segment.

[0033] Specifically, the control device first extracts structural features from the pre-simulated topographic point cloud. The extraction objects include: abrupt changes in wall thickness, distribution of reinforcing ribs, and heat dissipation boundary features. These three features are used as the basis for dividing the weld seam into independent sections.

[0034] The method for extracting wall thickness abrupt change points is as follows: The workpiece surfaces on both sides of the weld in the point cloud are fitted with curved surfaces to obtain the upper and lower surface surfaces; the vertical distance between the upper and lower surfaces is calculated point by point along the weld path as the local wall thickness at that point; when the wall thickness change rate between adjacent sampling points exceeds a preset threshold (e.g., 20%), the location is marked as a wall thickness abrupt change point. The preset threshold is a configurable parameter set by the operator during equipment debugging based on the workpiece material and thickness range; for common automotive structural steel, this threshold is typically set to 15% to 25%, and in this embodiment, it is set to 20%. These points usually appear at the boundary where the workpiece transitions from a thin plate area to a cast thick-walled area.

[0035] The method for extracting the distribution location of reinforcing ribs is as follows: Local curvature analysis is performed on the pre-simulated topographic point cloud. When the surface curvature of a certain region exceeds a preset threshold and exhibits a convex shape, that region is identified as a reinforcing rib structure, and its start and end points along the weld path are recorded. The curvature calculation uses the rate of change of the normal vector of the local neighborhood of the point cloud. The preset threshold is set by the operator based on the typical geometric characteristics of the welded workpiece; in this embodiment, the curvature threshold is set to 0.05 mm. -1 .

[0036] The method for extracting heat dissipation boundary features is as follows: edge detection is performed on the point cloud to identify the geometric boundaries, heat dissipation holes, openings, and other structures of the workpiece. These features indicate the boundaries of areas where heat is easily dissipated.

[0037] When multiple segmentation criteria (wall thickness change points, stiffener start / end points, heat dissipation boundary start points) are less than 5mm apart from each other along the weld path, the control device merges them into one segmentation point and takes the position closest to the weld start point as the final segmentation point to avoid producing excessively short and meaningless segments.

[0038] Based on the spatial distribution of the above three types of structural features, the control device determines the dividing points along the weld path, dividing the entire weld into multiple sections. Each section has relatively consistent wall thickness, structural stiffness, and heat dissipation conditions, while there is at least one significant difference in structural features between adjacent sections.

[0039] The following example illustrates this using a specific vehicle axle weld. A fillet weld with a total length of 300mm has been found on a certain vehicle axle workpiece. After structural feature extraction, the following was discovered: The weld seam from 0mm to 80mm is located on a thin sheet stamping part with an average wall thickness of 2.5mm, no reinforcing ribs, and is situated in a closed area in the middle of the workpiece. A reinforcing rib is inserted in the 80mm to 180mm section of the weld, which significantly increases the local stiffness; The section of the weld from 180mm to 260mm is located in the thick-walled area of ​​the casting, with an average wall thickness of 6mm. The 260mm to 300mm section of the weld extends to the edge of the workpiece, close to the heat dissipation boundary.

[0040] Accordingly, the control device divides the weld into four independent sections: Section 1: 0-80mm, thin-walled zone; Section 2: 80-180mm, reinforcing rib area; Section 3: 180-260mm, thick-walled zone; Section 4: 260-300mm, edge heat dissipation area.

[0041] After segmenting, the control device quantitatively calculates the corresponding thermal deformation weighting coefficient for each segment. This weighting coefficient is not calculated using analytical formulas, but rather determined through pre-conducted process calibration experiments. The calibration experiments include: performing welding tests on workpieces in typical segments, measuring the actual thermal deformation, and then deriving the weighting coefficient that ensures accurate trajectory correction. The weighting coefficient ranges from 0 to 1; the larger the wall thickness, the stronger the local structural stiffness, and the worse the heat dissipation conditions, the smaller the weighting coefficient, and vice versa. As a feasible example, for ordinary carbon steel welding, a value of 0.7-0.9 is used for thin-walled, unreinforced areas, and 0.3-0.5 for thick-walled, reinforced areas. Specific values ​​can be calibrated through no more than 20 orthogonal experiments.

[0042] Taking the aforementioned axle weld as an example, the weight coefficients for each section are calculated as follows: Section 1 (thin-walled area) weight coefficient 0.8; Section 2 (reinforcing rib area) weight coefficient 0.3; Section 3 (thick-walled area) weight coefficient 0.4; Section 4 (edge ​​heat dissipation area) weight coefficient 0.6. These coefficients will be stored in the control device for subsequent real-time trajectory correction.

[0043] By dividing the sections based on abrupt changes in wall thickness, distribution of reinforcing ribs, and characteristics of heat dissipation boundaries, and assigning differentiated weight coefficients, the expected compensation amount of each section is matched with its own thermal deformation sensitivity in terms of physical laws, thus avoiding the local compensation failure caused by traditional global unified compensation.

[0044] S3, Pre-construction and storage of proxy models Before the welding operation begins, the control device also needs to build and store a lightweight proxy model of the welding heat input response in advance.

[0045] A surrogate model is a highly simplified mathematical model of the complex thermo-mechanical coupling physical processes in welding. Its computational load is far less than that of finite element simulation (which typically takes several hours to complete a single calculation), and it can run in real time on ordinary industrial controllers at a frequency of no less than 1 kHz. The term "lightweight" refers to the model's low computation time and small memory footprint, representing an order of magnitude difference compared to finite element simulation.

[0046] In this embodiment, the proxy model is specifically a two-dimensional data table interpolation model, pre-constructed under the condition that the welding speed remains constant (e.g., 20 mm / s). The first step involves conducting systematic welding process experiments for different structural types of sections (e.g., thin-walled sections, thick-walled sections, and reinforcing rib sections). In the experiments, welding is performed on workpiece samples corresponding to each typical section using different combinations of welding current and welding voltage, with the welding speed fixed at 20 mm / s. The experimental range of welding current is, for example, 150A to 300A, with a step size of 25A; the experimental range of welding voltage is, for example, 22V to 32V, with a step size of 2V.

[0047] The second step involves using a laser tracker to measure the coordinates of the weld centerline after cooling to room temperature for each set of parameters. The difference between these coordinates and the pre-simulated desired trajectory coordinates is then calculated to obtain the weld point offset vector (Δx, Δy, Δz). Before measurement, the laser tracker coordinate system and the robot base coordinate system are registered and aligned using at least three non-collinear common reference points to ensure that the measurement data are in the same coordinate system.

[0048] The third step involves constructing a two-dimensional data table with current and voltage as the horizontal and vertical axes, and the offset vector as the table entry value. For parameter combinations not directly measured in the table, bilinear interpolation is used to calculate the offset vector in real time. This data table is the core of the surrogate model.

[0049] The proxy model takes real-time collected welding current and welding voltage as inputs and weld point offset vector as output. In this invention, the weld point offset vector output by the proxy model undergoes a crucial processing step: its output amplitude is modulated and corrected by the thermal deformation weighting coefficient of the corresponding segment of the welding position. The specific modulation method is as follows: ΔP_modulated = W_i × ΔP_raw Wherein, ΔP_raw is the original weld point offset vector output by the surrogate model based on the current current and voltage values ​​through table lookup or fitting; W_i is the thermal deformation weight coefficient of the segment where the current welding position is located; and ΔP_modulated is the effective weld point offset vector after modulation correction by the segment weight coefficient. Since the weight coefficient W_i is between 0 and 1, the effect of modulation correction is to scale the original offset vector proportionally, so that the compensation amount of each segment matches its own thermal deformation sensitivity.

[0050] The aforementioned proxy model is constructed and stored in the memory of the control device before welding begins, and can be called in real time during the welding process.

[0051] This lightweight proxy model responds to real-time electrical signal input with millisecond-level computation, ensuring the real-time control requirements of welding while avoiding the shortcomings of full physical simulation calculations that are too time-consuming to be applied online.

[0052] S4. Real-time trajectory correction during welding process After completing the above pre-welding preparations, the workstation begins the formal welding operation. The welding power source outputs welding current and voltage, and the welding torch generates a welding arc. During this process, the control device continuously acquires the current and voltage values ​​output by the welding power source in real time via an industrial real-time bus at a sampling frequency of 1kHz.

[0053] Within each control cycle (1 millisecond), the control device performs the following operations: First, the currently collected welding current and voltage values ​​are input into the proxy model to obtain the original weld point offset vector output by the proxy model in real time. Then, based on the current position of the welding robot, the weld section where the welding torch is currently located is determined, and the corresponding thermal deformation weight coefficient for that section is queried. The original weld point offset vector is multiplied by this weight coefficient to obtain the effective weld point offset vector after section weight coefficient modulation and correction.

[0054] Subsequently, the control device calculates the real-time target motion trajectory of the welding torch. The specific calculation formula is as follows: P_target = P_ref + ΔP_modulated Where P_ref is the coordinate value corresponding to the current welding position in the pre-simulated topographic point cloud, and ΔP_modulated is the weld point offset vector after modulation correction. This addition operation is a component addition of three-dimensional spatial vectors, that is: X_target = X_ref + Δx_modulated Y_target = Y_ref + Δy_modulated Z_target = Z_ref + Δz_modulated The control device sends the calculated dynamic target position to the welding robot controller via real-time industrial Ethernet. The robot controller then adjusts the spatial position of the welding torch tip in real time to achieve continuous dynamic correction of the welding trajectory.

[0055] Because different sections are assigned different thermal deformation weight coefficients, the original offset vector output by the surrogate model is modulated to varying degrees in different sections. Taking the aforementioned axle weld as an example: in section one (thin-walled region, weight coefficient 0.8), the offset vector is larger, achieving strong compensation; in section two (reinforcing rib region, weight coefficient 0.3), the offset vector is significantly reduced, achieving weak compensation; sections three and four also receive differentiated compensation amounts according to their weight coefficients. Thus, different sections of the same weld obtain gradient compensation matching their respective structural characteristics, avoiding the problems of insufficient or excessive local compensation caused by traditional global uniform compensation, and significantly improving the trajectory accuracy and deposition consistency of the entire weld.

[0056] As a further optimized implementation, the workstation in this embodiment also has a post-weld quality risk prediction function.

[0057] Specifically, after a single weld seam is completed and the workpiece has cooled sufficiently to room temperature, the control device once again controls the welding robot to replicate the scanning path that is completely consistent with the pre-welding simulation motion. The same laser contour sensor is used to collect the surface topography point cloud of the workpiece after cooling, which is called the "cooled topography point cloud".

[0058] Subsequently, the control device spatially aligns the cooled topographic point cloud with the previously stored pre-simulated topographic point cloud. The alignment algorithm can employ the Iterative Closest Point (ICP) algorithm. After alignment, the Euclidean distance between corresponding points in the two point clouds is calculated, generating a three-dimensional deviation distribution map, which is the "weld solidification difference map". The difference map is presented in the form of a color cloud map or a grayscale image, with different colors or grayscale values ​​representing different deviation amplitudes, intuitively reflecting the distribution of solidification shrinkage and residual deformation of the weld and heat-affected zone during welding and cooling processes.

[0059] The control device internally stores association rules between difference map features and welding quality risk types. These rules were established through historical experiments: for different types of workpieces and weld types, a large number of welding samples were collected, and a difference map was generated for each sample. The actual welding quality was then determined using methods such as X-ray inspection or metallographic analysis. Common feature patterns and their corresponding risk types were then statistically analyzed. For example: When a local depression depth exceeding 0.3 mm appears in a certain section of the difference map and the depression is concentrated on the center line of the weld, it is determined that there is a risk of solidification cracking. When the deviation gradient (the amount of deviation change per unit length) in a certain region of the difference map exceeds 0.05 mm / mm, it is determined that there is a risk of porosity or incomplete fusion. When an asymmetrical deviation distribution appears at the end of the weld (the difference between the left and right sides exceeds 0.2 mm), it is determined that there is a risk of weld deviation.

[0060] Once a new difference map is generated, the control device extracts its feature vectors (including indentation depth, indentation area, magnitude and location of gradient abrupt changes, and deviation asymmetry) and compares them with the aforementioned association rules. If the features match any risk rule, the control device outputs the corresponding welding quality risk prediction result, prompting the operator to pay close attention to the weld or perform repair welding.

[0061] For example, in the difference map generated after a certain welding is completed, a local depression with a maximum depth of 0.4 mm is detected in the 180 mm to 200 mm section of the weld, and the depression is concentrated at the center line of the weld. This feature matches the "solidification crack risk" rule, and the control device outputs the prediction result that "there is a solidification crack risk in the 180-200 mm section of the weld".

[0062] This post-weld morphology differential analysis method fully reuses the hardware and path data of the pre-weld scanning, without adding extra detection equipment, but can perform preliminary screening of welding quality risks without damaging the workpiece, significantly reducing the frequency and cost of subsequent flaw detection sampling.

[0063] The workstation in this embodiment also includes a welding fume purification and heat recovery module integrated inside the workstation. This module includes a plasma purification unit, a high-efficiency filtration unit, and a heat exchange unit.

[0064] The control device is connected to the welding fume purification and heat recovery module via a signal connection. The control device can adaptively adjust the purification operation intensity and heat recovery efficiency based on the welding start / stop status and welding power. The specific adjustment rules are as follows: When the real-time welding power P = U × I > 6 kW (high-power welding), the speed of the purification fan is adjusted to 100% of the rated speed, and the flow rate of the heat recovery circulation pump is adjusted to the maximum. When 2 kW ≤ P ≤ 6 kW (medium power), the speed of the purification fan should be adjusted to 60% of the rated speed, and the flow rate of the heat recovery circulation pump should be adjusted to 60% of the rated flow rate. When P < 2 kW or welding is in standby mode, the speed of the purification fan is adjusted to 20% of the rated speed to maintain low power consumption, and the heat recovery circulation pump stops working.

[0065] Through the above-mentioned quantitative adjustment rules, the effective recovery and utilization of welding waste heat is achieved while ensuring environmental compliance, thereby reducing the overall operating energy consumption of the workstation.

[0066] In this embodiment, 10 sets of comparative tests were conducted on the aforementioned axle welds. The average weld centerline deviation of the traditional single global compensation scheme was 0.18 mm, with a maximum deviation of 0.35 mm; the average weld centerline deviation of the partitioned gradient compensation scheme in this embodiment was 0.04 mm, with a maximum deviation of 0.08 mm. The deviation consistency index CPK value increased from 1.12 to 1.95. Experimental results show that the scheme in this embodiment significantly improves the weld trajectory accuracy and deposition consistency.

[0067] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. An intelligent welding workstation, characterized in that, include: A welding robot with a welding torch and a laser profile sensor mounted at its end; A welding power source is used to stably supply welding current and welding voltage to the welding torch; The control device is communicatively connected to the welding robot, the laser profile sensor, and the welding power source; the control device is configured to: Before the welding operation begins, the welding robot is controlled to perform a no-welding pre-simulation motion along the weld seam path, and the workpiece surface topography data is collected simultaneously through the laser contour sensor to generate a pre-simulation topography point cloud. Based on the wall thickness abrupt change, stiffener distribution and heat dissipation boundary characteristics in the pre-simulated topography point cloud, the entire weld seam is automatically divided into multiple independent segments along the weld seam path, and a corresponding thermal deformation weight coefficient is matched and assigned to each segment. A lightweight proxy model for welding heat input response is pre-built and stored. The proxy model takes the welding current and welding voltage collected in real time as input and the weld point offset vector as output. The output amplitude of the weld point offset vector is modulated and corrected by the thermal deformation weight coefficient of the corresponding section of the welding position. During the actual welding process, the real-time target motion trajectory of the welding torch is corrected to the sum of the corresponding position coordinates in the pre-simulated topography point cloud and the modulated and corrected welding point offset vector, based on the welding point offset vector output in real time by the proxy model and modified by the segment thermal deformation weight coefficient.

2. The intelligent welding workstation according to claim 1, characterized in that, When the control device automatically divides the entire weld into multiple independent sections along the weld path, it extracts the wall thickness change points, stiffener distribution positions, and heat dissipation boundary features from the pre-simulated topography point cloud as the basis for dividing the independent sections of the weld.

3. The intelligent welding workstation according to claim 2, characterized in that, The control device is configured as follows: Based on the average wall thickness, local structural stiffness, and real-time heat conduction conditions of each weld section, the thermal deformation weight coefficient of the corresponding section is determined through pre-calibration experiments; among them, the larger the workpiece wall thickness and the stronger the local structural stiffness, the smaller the corresponding thermal deformation weight coefficient value.

4. The intelligent welding workstation according to claim 1, characterized in that, The lightweight proxy model for welding heat input response is a parameterized response model established based on historical experimental data. It uses welding current and welding voltage as core input variables and employs data tables and bilinear interpolation to output the corresponding weld point offset vector in real time.

5. The intelligent welding workstation according to claim 1, characterized in that, The control device is further configured to: after a single weld seam is completed and the workpiece is fully cooled, control the welding robot again to replicate the scanning path consistent with the pre-weld simulation motion, and collect the workpiece cooling topography point cloud; generate a weld seam solidification difference map by performing a three-dimensional difference operation between the cooled topography point cloud and the pre-simulation topography point cloud.

6. The intelligent welding workstation according to claim 1, characterized in that, During the pre-welding simulation phase, the laser contour sensor continuously collects workpiece contour data at a preset acquisition frequency. Meanwhile, the welding power supply maintains a standby state with zero current and voltage output to avoid interference from welding arc light and spatter on the shape acquisition.

7. The intelligent welding workstation according to claim 1, characterized in that, The control device communicates with the welding power source at high speed via an industrial real-time bus, continuously acquiring the current and voltage values ​​during the welding process at a sampling frequency of not less than 1 kHz.

8. The intelligent welding workstation according to claim 1, characterized in that, It also includes a welding fume purification and heat recovery module integrated inside the workstation; the control device is signal-connected to the welding fume purification and heat recovery module, and can adaptively adjust the purification operation intensity and heat recovery efficiency according to the welding start / stop status and welding power.

9. The intelligent welding workstation according to claim 1, characterized in that, The control device uses an industrial PC or an embedded dedicated welding controller, and communicates bidirectionally with the welding robot controller via real-time industrial Ethernet to quickly exchange trajectory correction commands and motion status data.

Images